Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
Peter Zijlstra | 391e43d | 2011-11-15 17:14:39 +0100 | [diff] [blame] | 2 | * kernel/sched/core.c |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3 | * |
| 4 | * Kernel scheduler and related syscalls |
| 5 | * |
| 6 | * Copyright (C) 1991-2002 Linus Torvalds |
| 7 | * |
| 8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and |
| 9 | * make semaphores SMP safe |
| 10 | * 1998-11-19 Implemented schedule_timeout() and related stuff |
| 11 | * by Andrea Arcangeli |
| 12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: |
| 13 | * hybrid priority-list and round-robin design with |
| 14 | * an array-switch method of distributing timeslices |
| 15 | * and per-CPU runqueues. Cleanups and useful suggestions |
| 16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. |
| 17 | * 2003-09-03 Interactivity tuning by Con Kolivas. |
| 18 | * 2004-04-02 Scheduler domains code by Nick Piggin |
Ingo Molnar | c31f2e8 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
| 20 | * fair scheduling design by Con Kolivas. |
| 21 | * 2007-05-05 Load balancing (smp-nice) and other improvements |
| 22 | * by Peter Williams |
| 23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith |
| 24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri |
Ingo Molnar | b913176 | 2008-01-25 21:08:19 +0100 | [diff] [blame] | 25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
| 26 | * Thomas Gleixner, Mike Kravetz |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 | */ |
| 28 | |
Mark Rutland | e1b77c9 | 2016-03-09 14:08:18 -0800 | [diff] [blame] | 29 | #include <linux/kasan.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | #include <linux/mm.h> |
| 31 | #include <linux/module.h> |
| 32 | #include <linux/nmi.h> |
| 33 | #include <linux/init.h> |
Ingo Molnar | dff06c1 | 2007-07-09 18:52:00 +0200 | [diff] [blame] | 34 | #include <linux/uaccess.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 35 | #include <linux/highmem.h> |
Andy Lutomirski | f98db60 | 2016-04-26 09:39:06 -0700 | [diff] [blame] | 36 | #include <linux/mmu_context.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 37 | #include <linux/interrupt.h> |
Randy.Dunlap | c59ede7 | 2006-01-11 12:17:46 -0800 | [diff] [blame] | 38 | #include <linux/capability.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 39 | #include <linux/completion.h> |
| 40 | #include <linux/kernel_stat.h> |
Ingo Molnar | 9a11b49a | 2006-07-03 00:24:33 -0700 | [diff] [blame] | 41 | #include <linux/debug_locks.h> |
Ingo Molnar | cdd6c48 | 2009-09-21 12:02:48 +0200 | [diff] [blame] | 42 | #include <linux/perf_event.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | #include <linux/security.h> |
| 44 | #include <linux/notifier.h> |
| 45 | #include <linux/profile.h> |
Nigel Cunningham | 7dfb710 | 2006-12-06 20:34:23 -0800 | [diff] [blame] | 46 | #include <linux/freezer.h> |
akpm@osdl.org | 198e2f1 | 2006-01-12 01:05:30 -0800 | [diff] [blame] | 47 | #include <linux/vmalloc.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 48 | #include <linux/blkdev.h> |
| 49 | #include <linux/delay.h> |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 50 | #include <linux/pid_namespace.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 51 | #include <linux/smp.h> |
| 52 | #include <linux/threads.h> |
| 53 | #include <linux/timer.h> |
| 54 | #include <linux/rcupdate.h> |
| 55 | #include <linux/cpu.h> |
| 56 | #include <linux/cpuset.h> |
| 57 | #include <linux/percpu.h> |
Alexey Dobriyan | b5aadf7 | 2008-10-06 13:23:43 +0400 | [diff] [blame] | 58 | #include <linux/proc_fs.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 59 | #include <linux/seq_file.h> |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 60 | #include <linux/sysctl.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 61 | #include <linux/syscalls.h> |
| 62 | #include <linux/times.h> |
Jay Lan | 8f0ab51 | 2006-09-30 23:28:59 -0700 | [diff] [blame] | 63 | #include <linux/tsacct_kern.h> |
bibo mao | c6fd91f | 2006-03-26 01:38:20 -0800 | [diff] [blame] | 64 | #include <linux/kprobes.h> |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 65 | #include <linux/delayacct.h> |
Ingo Molnar | dff06c1 | 2007-07-09 18:52:00 +0200 | [diff] [blame] | 66 | #include <linux/unistd.h> |
Jens Axboe | f5ff842 | 2007-09-21 09:19:54 +0200 | [diff] [blame] | 67 | #include <linux/pagemap.h> |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 68 | #include <linux/hrtimer.h> |
Reynes Philippe | 30914a5 | 2008-03-17 16:19:05 -0700 | [diff] [blame] | 69 | #include <linux/tick.h> |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 70 | #include <linux/ctype.h> |
Steven Rostedt | 6cd8a4b | 2008-05-12 21:20:42 +0200 | [diff] [blame] | 71 | #include <linux/ftrace.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 72 | #include <linux/slab.h> |
Carsten Emde | f1c6f1a | 2011-10-26 23:14:16 +0200 | [diff] [blame] | 73 | #include <linux/init_task.h> |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 74 | #include <linux/context_tracking.h> |
Gideon Israel Dsouza | 52f5684c | 2014-04-07 15:39:20 -0700 | [diff] [blame] | 75 | #include <linux/compiler.h> |
Josh Poimboeuf | 8e05e96 | 2016-02-28 22:22:38 -0600 | [diff] [blame] | 76 | #include <linux/frame.h> |
Giovanni Gherdovich | 6075620 | 2016-08-05 10:21:56 +0200 | [diff] [blame] | 77 | #include <linux/prefetch.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 78 | |
David Howells | 96f951e | 2012-03-28 18:30:03 +0100 | [diff] [blame] | 79 | #include <asm/switch_to.h> |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 80 | #include <asm/tlb.h> |
Satyam Sharma | 838225b | 2007-10-24 18:23:50 +0200 | [diff] [blame] | 81 | #include <asm/irq_regs.h> |
Christian Borntraeger | db7e527 | 2012-01-11 08:58:16 +0100 | [diff] [blame] | 82 | #include <asm/mutex.h> |
Glauber Costa | e6e6685 | 2011-07-11 15:28:17 -0400 | [diff] [blame] | 83 | #ifdef CONFIG_PARAVIRT |
| 84 | #include <asm/paravirt.h> |
| 85 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 86 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 87 | #include "sched.h" |
Tejun Heo | ea13844 | 2013-01-18 14:05:55 -0800 | [diff] [blame] | 88 | #include "../workqueue_internal.h" |
Thomas Gleixner | 29d5e04 | 2012-04-20 13:05:45 +0000 | [diff] [blame] | 89 | #include "../smpboot.h" |
Gregory Haskins | 6e0534f | 2008-05-12 21:21:01 +0200 | [diff] [blame] | 90 | |
Steven Rostedt | a8d154b | 2009-04-10 09:36:00 -0400 | [diff] [blame] | 91 | #define CREATE_TRACE_POINTS |
Steven Rostedt | ad8d75f | 2009-04-14 19:39:12 -0400 | [diff] [blame] | 92 | #include <trace/events/sched.h> |
Steven Rostedt | a8d154b | 2009-04-10 09:36:00 -0400 | [diff] [blame] | 93 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 94 | DEFINE_MUTEX(sched_domains_mutex); |
| 95 | DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
Peter Zijlstra | dc61b1d | 2010-06-08 11:40:42 +0200 | [diff] [blame] | 96 | |
Peter Zijlstra | fe44d62 | 2010-12-09 14:15:34 +0100 | [diff] [blame] | 97 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
Venkatesh Pallipadi | 305e683 | 2010-10-04 17:03:21 -0700 | [diff] [blame] | 98 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 99 | void update_rq_clock(struct rq *rq) |
Peter Zijlstra | 3e51f33 | 2008-05-03 18:29:28 +0200 | [diff] [blame] | 100 | { |
Peter Zijlstra | fe44d62 | 2010-12-09 14:15:34 +0100 | [diff] [blame] | 101 | s64 delta; |
Venkatesh Pallipadi | 305e683 | 2010-10-04 17:03:21 -0700 | [diff] [blame] | 102 | |
Peter Zijlstra | 9edfbfe | 2015-01-05 11:18:11 +0100 | [diff] [blame] | 103 | lockdep_assert_held(&rq->lock); |
| 104 | |
| 105 | if (rq->clock_skip_update & RQCF_ACT_SKIP) |
Mike Galbraith | f26f9af | 2010-12-08 11:05:42 +0100 | [diff] [blame] | 106 | return; |
Venkatesh Pallipadi | aa48380 | 2010-10-04 17:03:22 -0700 | [diff] [blame] | 107 | |
Peter Zijlstra | fe44d62 | 2010-12-09 14:15:34 +0100 | [diff] [blame] | 108 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
Mike Galbraith | 4036ac1 | 2014-06-24 07:49:40 +0200 | [diff] [blame] | 109 | if (delta < 0) |
| 110 | return; |
Peter Zijlstra | fe44d62 | 2010-12-09 14:15:34 +0100 | [diff] [blame] | 111 | rq->clock += delta; |
| 112 | update_rq_clock_task(rq, delta); |
Peter Zijlstra | 3e51f33 | 2008-05-03 18:29:28 +0200 | [diff] [blame] | 113 | } |
| 114 | |
Ingo Molnar | e436d80 | 2007-07-19 21:28:35 +0200 | [diff] [blame] | 115 | /* |
Ingo Molnar | bf5c91b | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 116 | * Debugging: various feature bits |
| 117 | */ |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 118 | |
| 119 | #define SCHED_FEAT(name, enabled) \ |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 120 | (1UL << __SCHED_FEAT_##name) * enabled | |
| 121 | |
| 122 | const_debug unsigned int sysctl_sched_features = |
Peter Zijlstra | 391e43d | 2011-11-15 17:14:39 +0100 | [diff] [blame] | 123 | #include "features.h" |
Peter Zijlstra | f00b45c | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 124 | 0; |
| 125 | |
| 126 | #undef SCHED_FEAT |
| 127 | |
Ingo Molnar | bf5c91b | 2007-10-15 17:00:04 +0200 | [diff] [blame] | 128 | /* |
Peter Zijlstra | b82d9fd | 2007-11-09 22:39:39 +0100 | [diff] [blame] | 129 | * Number of tasks to iterate in a single balance run. |
| 130 | * Limited because this is done with IRQs disabled. |
| 131 | */ |
| 132 | const_debug unsigned int sysctl_sched_nr_migrate = 32; |
| 133 | |
| 134 | /* |
Peter Zijlstra | e9e9250 | 2009-09-01 10:34:37 +0200 | [diff] [blame] | 135 | * period over which we average the RT time consumption, measured |
| 136 | * in ms. |
| 137 | * |
| 138 | * default: 1s |
| 139 | */ |
| 140 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; |
| 141 | |
| 142 | /* |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 143 | * period over which we measure -rt task cpu usage in us. |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 144 | * default: 1s |
| 145 | */ |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 146 | unsigned int sysctl_sched_rt_period = 1000000; |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 147 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 148 | __read_mostly int scheduler_running; |
Ingo Molnar | 6892b75 | 2008-02-13 14:02:36 +0100 | [diff] [blame] | 149 | |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 150 | /* |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 151 | * part of the period that we allow rt tasks to run in us. |
| 152 | * default: 0.95s |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 153 | */ |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 154 | int sysctl_sched_rt_runtime = 950000; |
| 155 | |
Rik van Riel | 3fa0818 | 2015-03-09 12:12:07 -0400 | [diff] [blame] | 156 | /* cpus with isolated domains */ |
| 157 | cpumask_var_t cpu_isolated_map; |
| 158 | |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 159 | /* |
Robert P. J. Day | cc2a73b | 2006-12-10 02:20:00 -0800 | [diff] [blame] | 160 | * this_rq_lock - lock this runqueue and disable interrupts. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 161 | */ |
Alexey Dobriyan | a995744 | 2007-10-15 17:00:13 +0200 | [diff] [blame] | 162 | static struct rq *this_rq_lock(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 163 | __acquires(rq->lock) |
| 164 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 165 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 166 | |
| 167 | local_irq_disable(); |
| 168 | rq = this_rq(); |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 169 | raw_spin_lock(&rq->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 170 | |
| 171 | return rq; |
| 172 | } |
| 173 | |
Peter Zijlstra | 3e71a46 | 2016-04-28 16:16:33 +0200 | [diff] [blame] | 174 | /* |
| 175 | * __task_rq_lock - lock the rq @p resides on. |
| 176 | */ |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 177 | struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
Peter Zijlstra | 3e71a46 | 2016-04-28 16:16:33 +0200 | [diff] [blame] | 178 | __acquires(rq->lock) |
| 179 | { |
| 180 | struct rq *rq; |
| 181 | |
| 182 | lockdep_assert_held(&p->pi_lock); |
| 183 | |
| 184 | for (;;) { |
| 185 | rq = task_rq(p); |
| 186 | raw_spin_lock(&rq->lock); |
| 187 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 188 | rf->cookie = lockdep_pin_lock(&rq->lock); |
Peter Zijlstra | 3e71a46 | 2016-04-28 16:16:33 +0200 | [diff] [blame] | 189 | return rq; |
| 190 | } |
| 191 | raw_spin_unlock(&rq->lock); |
| 192 | |
| 193 | while (unlikely(task_on_rq_migrating(p))) |
| 194 | cpu_relax(); |
| 195 | } |
| 196 | } |
| 197 | |
| 198 | /* |
| 199 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
| 200 | */ |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 201 | struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) |
Peter Zijlstra | 3e71a46 | 2016-04-28 16:16:33 +0200 | [diff] [blame] | 202 | __acquires(p->pi_lock) |
| 203 | __acquires(rq->lock) |
| 204 | { |
| 205 | struct rq *rq; |
| 206 | |
| 207 | for (;;) { |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 208 | raw_spin_lock_irqsave(&p->pi_lock, rf->flags); |
Peter Zijlstra | 3e71a46 | 2016-04-28 16:16:33 +0200 | [diff] [blame] | 209 | rq = task_rq(p); |
| 210 | raw_spin_lock(&rq->lock); |
| 211 | /* |
| 212 | * move_queued_task() task_rq_lock() |
| 213 | * |
| 214 | * ACQUIRE (rq->lock) |
| 215 | * [S] ->on_rq = MIGRATING [L] rq = task_rq() |
| 216 | * WMB (__set_task_cpu()) ACQUIRE (rq->lock); |
| 217 | * [S] ->cpu = new_cpu [L] task_rq() |
| 218 | * [L] ->on_rq |
| 219 | * RELEASE (rq->lock) |
| 220 | * |
| 221 | * If we observe the old cpu in task_rq_lock, the acquire of |
| 222 | * the old rq->lock will fully serialize against the stores. |
| 223 | * |
| 224 | * If we observe the new cpu in task_rq_lock, the acquire will |
| 225 | * pair with the WMB to ensure we must then also see migrating. |
| 226 | */ |
| 227 | if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) { |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 228 | rf->cookie = lockdep_pin_lock(&rq->lock); |
Peter Zijlstra | 3e71a46 | 2016-04-28 16:16:33 +0200 | [diff] [blame] | 229 | return rq; |
| 230 | } |
| 231 | raw_spin_unlock(&rq->lock); |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 232 | raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); |
Peter Zijlstra | 3e71a46 | 2016-04-28 16:16:33 +0200 | [diff] [blame] | 233 | |
| 234 | while (unlikely(task_on_rq_migrating(p))) |
| 235 | cpu_relax(); |
| 236 | } |
| 237 | } |
| 238 | |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 239 | #ifdef CONFIG_SCHED_HRTICK |
| 240 | /* |
| 241 | * Use HR-timers to deliver accurate preemption points. |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 242 | */ |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 243 | |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 244 | static void hrtick_clear(struct rq *rq) |
| 245 | { |
| 246 | if (hrtimer_active(&rq->hrtick_timer)) |
| 247 | hrtimer_cancel(&rq->hrtick_timer); |
| 248 | } |
| 249 | |
| 250 | /* |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 251 | * High-resolution timer tick. |
| 252 | * Runs from hardirq context with interrupts disabled. |
| 253 | */ |
| 254 | static enum hrtimer_restart hrtick(struct hrtimer *timer) |
| 255 | { |
| 256 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); |
| 257 | |
| 258 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); |
| 259 | |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 260 | raw_spin_lock(&rq->lock); |
Peter Zijlstra | 3e51f33 | 2008-05-03 18:29:28 +0200 | [diff] [blame] | 261 | update_rq_clock(rq); |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 262 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 263 | raw_spin_unlock(&rq->lock); |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 264 | |
| 265 | return HRTIMER_NORESTART; |
| 266 | } |
| 267 | |
Rabin Vincent | 95e904c | 2008-05-11 05:55:33 +0530 | [diff] [blame] | 268 | #ifdef CONFIG_SMP |
Peter Zijlstra | 971ee28 | 2013-06-28 11:18:53 +0200 | [diff] [blame] | 269 | |
Thomas Gleixner | 4961b6e | 2015-04-14 21:09:05 +0000 | [diff] [blame] | 270 | static void __hrtick_restart(struct rq *rq) |
Peter Zijlstra | 971ee28 | 2013-06-28 11:18:53 +0200 | [diff] [blame] | 271 | { |
| 272 | struct hrtimer *timer = &rq->hrtick_timer; |
Peter Zijlstra | 971ee28 | 2013-06-28 11:18:53 +0200 | [diff] [blame] | 273 | |
Thomas Gleixner | 4961b6e | 2015-04-14 21:09:05 +0000 | [diff] [blame] | 274 | hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED); |
Peter Zijlstra | 971ee28 | 2013-06-28 11:18:53 +0200 | [diff] [blame] | 275 | } |
| 276 | |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 277 | /* |
| 278 | * called from hardirq (IPI) context |
| 279 | */ |
| 280 | static void __hrtick_start(void *arg) |
Peter Zijlstra | b328ca1 | 2008-04-29 10:02:46 +0200 | [diff] [blame] | 281 | { |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 282 | struct rq *rq = arg; |
Peter Zijlstra | b328ca1 | 2008-04-29 10:02:46 +0200 | [diff] [blame] | 283 | |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 284 | raw_spin_lock(&rq->lock); |
Peter Zijlstra | 971ee28 | 2013-06-28 11:18:53 +0200 | [diff] [blame] | 285 | __hrtick_restart(rq); |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 286 | rq->hrtick_csd_pending = 0; |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 287 | raw_spin_unlock(&rq->lock); |
Peter Zijlstra | b328ca1 | 2008-04-29 10:02:46 +0200 | [diff] [blame] | 288 | } |
| 289 | |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 290 | /* |
| 291 | * Called to set the hrtick timer state. |
| 292 | * |
| 293 | * called with rq->lock held and irqs disabled |
| 294 | */ |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 295 | void hrtick_start(struct rq *rq, u64 delay) |
Peter Zijlstra | b328ca1 | 2008-04-29 10:02:46 +0200 | [diff] [blame] | 296 | { |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 297 | struct hrtimer *timer = &rq->hrtick_timer; |
xiaofeng.yan | 177ef2a | 2014-08-26 03:15:41 +0000 | [diff] [blame] | 298 | ktime_t time; |
| 299 | s64 delta; |
| 300 | |
| 301 | /* |
| 302 | * Don't schedule slices shorter than 10000ns, that just |
| 303 | * doesn't make sense and can cause timer DoS. |
| 304 | */ |
| 305 | delta = max_t(s64, delay, 10000LL); |
| 306 | time = ktime_add_ns(timer->base->get_time(), delta); |
Peter Zijlstra | b328ca1 | 2008-04-29 10:02:46 +0200 | [diff] [blame] | 307 | |
Arjan van de Ven | cc584b2 | 2008-09-01 15:02:30 -0700 | [diff] [blame] | 308 | hrtimer_set_expires(timer, time); |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 309 | |
| 310 | if (rq == this_rq()) { |
Peter Zijlstra | 971ee28 | 2013-06-28 11:18:53 +0200 | [diff] [blame] | 311 | __hrtick_restart(rq); |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 312 | } else if (!rq->hrtick_csd_pending) { |
Frederic Weisbecker | c46fff2 | 2014-02-24 16:40:02 +0100 | [diff] [blame] | 313 | smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd); |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 314 | rq->hrtick_csd_pending = 1; |
| 315 | } |
Peter Zijlstra | b328ca1 | 2008-04-29 10:02:46 +0200 | [diff] [blame] | 316 | } |
| 317 | |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 318 | #else |
| 319 | /* |
| 320 | * Called to set the hrtick timer state. |
| 321 | * |
| 322 | * called with rq->lock held and irqs disabled |
| 323 | */ |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 324 | void hrtick_start(struct rq *rq, u64 delay) |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 325 | { |
Wanpeng Li | 8689333 | 2014-11-26 08:44:06 +0800 | [diff] [blame] | 326 | /* |
| 327 | * Don't schedule slices shorter than 10000ns, that just |
| 328 | * doesn't make sense. Rely on vruntime for fairness. |
| 329 | */ |
| 330 | delay = max_t(u64, delay, 10000LL); |
Thomas Gleixner | 4961b6e | 2015-04-14 21:09:05 +0000 | [diff] [blame] | 331 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), |
| 332 | HRTIMER_MODE_REL_PINNED); |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 333 | } |
Rabin Vincent | 95e904c | 2008-05-11 05:55:33 +0530 | [diff] [blame] | 334 | #endif /* CONFIG_SMP */ |
Peter Zijlstra | b328ca1 | 2008-04-29 10:02:46 +0200 | [diff] [blame] | 335 | |
| 336 | static void init_rq_hrtick(struct rq *rq) |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 337 | { |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 338 | #ifdef CONFIG_SMP |
| 339 | rq->hrtick_csd_pending = 0; |
| 340 | |
| 341 | rq->hrtick_csd.flags = 0; |
| 342 | rq->hrtick_csd.func = __hrtick_start; |
| 343 | rq->hrtick_csd.info = rq; |
| 344 | #endif |
| 345 | |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 346 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
| 347 | rq->hrtick_timer.function = hrtick; |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 348 | } |
Andrew Morton | 006c75f | 2008-09-22 14:55:46 -0700 | [diff] [blame] | 349 | #else /* CONFIG_SCHED_HRTICK */ |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 350 | static inline void hrtick_clear(struct rq *rq) |
| 351 | { |
| 352 | } |
| 353 | |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 354 | static inline void init_rq_hrtick(struct rq *rq) |
| 355 | { |
| 356 | } |
Andrew Morton | 006c75f | 2008-09-22 14:55:46 -0700 | [diff] [blame] | 357 | #endif /* CONFIG_SCHED_HRTICK */ |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 358 | |
Frederic Weisbecker | 5529578 | 2016-03-24 15:38:01 +0100 | [diff] [blame] | 359 | /* |
| 360 | * cmpxchg based fetch_or, macro so it works for different integer types |
| 361 | */ |
| 362 | #define fetch_or(ptr, mask) \ |
| 363 | ({ \ |
| 364 | typeof(ptr) _ptr = (ptr); \ |
| 365 | typeof(mask) _mask = (mask); \ |
| 366 | typeof(*_ptr) _old, _val = *_ptr; \ |
| 367 | \ |
| 368 | for (;;) { \ |
| 369 | _old = cmpxchg(_ptr, _val, _val | _mask); \ |
| 370 | if (_old == _val) \ |
| 371 | break; \ |
| 372 | _val = _old; \ |
| 373 | } \ |
| 374 | _old; \ |
| 375 | }) |
| 376 | |
Peter Zijlstra | e3baac4 | 2014-06-04 10:31:18 -0700 | [diff] [blame] | 377 | #if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG) |
Peter Zijlstra | fd99f91 | 2014-04-09 15:35:08 +0200 | [diff] [blame] | 378 | /* |
| 379 | * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG, |
| 380 | * this avoids any races wrt polling state changes and thereby avoids |
| 381 | * spurious IPIs. |
| 382 | */ |
| 383 | static bool set_nr_and_not_polling(struct task_struct *p) |
| 384 | { |
| 385 | struct thread_info *ti = task_thread_info(p); |
| 386 | return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG); |
| 387 | } |
Peter Zijlstra | e3baac4 | 2014-06-04 10:31:18 -0700 | [diff] [blame] | 388 | |
| 389 | /* |
| 390 | * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set. |
| 391 | * |
| 392 | * If this returns true, then the idle task promises to call |
| 393 | * sched_ttwu_pending() and reschedule soon. |
| 394 | */ |
| 395 | static bool set_nr_if_polling(struct task_struct *p) |
| 396 | { |
| 397 | struct thread_info *ti = task_thread_info(p); |
Jason Low | 316c1608d | 2015-04-28 13:00:20 -0700 | [diff] [blame] | 398 | typeof(ti->flags) old, val = READ_ONCE(ti->flags); |
Peter Zijlstra | e3baac4 | 2014-06-04 10:31:18 -0700 | [diff] [blame] | 399 | |
| 400 | for (;;) { |
| 401 | if (!(val & _TIF_POLLING_NRFLAG)) |
| 402 | return false; |
| 403 | if (val & _TIF_NEED_RESCHED) |
| 404 | return true; |
| 405 | old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED); |
| 406 | if (old == val) |
| 407 | break; |
| 408 | val = old; |
| 409 | } |
| 410 | return true; |
| 411 | } |
| 412 | |
Peter Zijlstra | fd99f91 | 2014-04-09 15:35:08 +0200 | [diff] [blame] | 413 | #else |
| 414 | static bool set_nr_and_not_polling(struct task_struct *p) |
| 415 | { |
| 416 | set_tsk_need_resched(p); |
| 417 | return true; |
| 418 | } |
Peter Zijlstra | e3baac4 | 2014-06-04 10:31:18 -0700 | [diff] [blame] | 419 | |
| 420 | #ifdef CONFIG_SMP |
| 421 | static bool set_nr_if_polling(struct task_struct *p) |
| 422 | { |
| 423 | return false; |
| 424 | } |
| 425 | #endif |
Peter Zijlstra | fd99f91 | 2014-04-09 15:35:08 +0200 | [diff] [blame] | 426 | #endif |
| 427 | |
Peter Zijlstra | 7675104 | 2015-05-01 08:27:50 -0700 | [diff] [blame] | 428 | void wake_q_add(struct wake_q_head *head, struct task_struct *task) |
| 429 | { |
| 430 | struct wake_q_node *node = &task->wake_q; |
| 431 | |
| 432 | /* |
| 433 | * Atomically grab the task, if ->wake_q is !nil already it means |
| 434 | * its already queued (either by us or someone else) and will get the |
| 435 | * wakeup due to that. |
| 436 | * |
| 437 | * This cmpxchg() implies a full barrier, which pairs with the write |
Davidlohr Bueso | 58fe9c4 | 2016-05-08 20:58:10 -0700 | [diff] [blame] | 438 | * barrier implied by the wakeup in wake_up_q(). |
Peter Zijlstra | 7675104 | 2015-05-01 08:27:50 -0700 | [diff] [blame] | 439 | */ |
| 440 | if (cmpxchg(&node->next, NULL, WAKE_Q_TAIL)) |
| 441 | return; |
| 442 | |
| 443 | get_task_struct(task); |
| 444 | |
| 445 | /* |
| 446 | * The head is context local, there can be no concurrency. |
| 447 | */ |
| 448 | *head->lastp = node; |
| 449 | head->lastp = &node->next; |
| 450 | } |
| 451 | |
| 452 | void wake_up_q(struct wake_q_head *head) |
| 453 | { |
| 454 | struct wake_q_node *node = head->first; |
| 455 | |
| 456 | while (node != WAKE_Q_TAIL) { |
| 457 | struct task_struct *task; |
| 458 | |
| 459 | task = container_of(node, struct task_struct, wake_q); |
| 460 | BUG_ON(!task); |
| 461 | /* task can safely be re-inserted now */ |
| 462 | node = node->next; |
| 463 | task->wake_q.next = NULL; |
| 464 | |
| 465 | /* |
| 466 | * wake_up_process() implies a wmb() to pair with the queueing |
| 467 | * in wake_q_add() so as not to miss wakeups. |
| 468 | */ |
| 469 | wake_up_process(task); |
| 470 | put_task_struct(task); |
| 471 | } |
| 472 | } |
| 473 | |
Peter Zijlstra | fd99f91 | 2014-04-09 15:35:08 +0200 | [diff] [blame] | 474 | /* |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 475 | * resched_curr - mark rq's current task 'to be rescheduled now'. |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 476 | * |
| 477 | * On UP this means the setting of the need_resched flag, on SMP it |
| 478 | * might also involve a cross-CPU call to trigger the scheduler on |
| 479 | * the target CPU. |
| 480 | */ |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 481 | void resched_curr(struct rq *rq) |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 482 | { |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 483 | struct task_struct *curr = rq->curr; |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 484 | int cpu; |
| 485 | |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 486 | lockdep_assert_held(&rq->lock); |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 487 | |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 488 | if (test_tsk_need_resched(curr)) |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 489 | return; |
| 490 | |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 491 | cpu = cpu_of(rq); |
Peter Zijlstra | fd99f91 | 2014-04-09 15:35:08 +0200 | [diff] [blame] | 492 | |
Peter Zijlstra | f27dde8 | 2013-08-14 14:55:31 +0200 | [diff] [blame] | 493 | if (cpu == smp_processor_id()) { |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 494 | set_tsk_need_resched(curr); |
Peter Zijlstra | f27dde8 | 2013-08-14 14:55:31 +0200 | [diff] [blame] | 495 | set_preempt_need_resched(); |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 496 | return; |
Peter Zijlstra | f27dde8 | 2013-08-14 14:55:31 +0200 | [diff] [blame] | 497 | } |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 498 | |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 499 | if (set_nr_and_not_polling(curr)) |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 500 | smp_send_reschedule(cpu); |
Andy Lutomirski | dfc68f2 | 2014-06-04 10:31:15 -0700 | [diff] [blame] | 501 | else |
| 502 | trace_sched_wake_idle_without_ipi(cpu); |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 503 | } |
| 504 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 505 | void resched_cpu(int cpu) |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 506 | { |
| 507 | struct rq *rq = cpu_rq(cpu); |
| 508 | unsigned long flags; |
| 509 | |
Paul E. McKenney | fb8bd56 | 2017-09-18 08:54:40 -0700 | [diff] [blame] | 510 | raw_spin_lock_irqsave(&rq->lock, flags); |
Paul E. McKenney | cce2b93 | 2017-10-13 16:24:28 -0700 | [diff] [blame] | 511 | if (cpu_online(cpu) || cpu == smp_processor_id()) |
| 512 | resched_curr(rq); |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 513 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 514 | } |
Thomas Gleixner | 06d8308 | 2008-03-22 09:20:24 +0100 | [diff] [blame] | 515 | |
Peter Zijlstra | b021fe3 | 2013-09-17 09:30:55 +0200 | [diff] [blame] | 516 | #ifdef CONFIG_SMP |
Frederic Weisbecker | 3451d02 | 2011-08-10 23:21:01 +0200 | [diff] [blame] | 517 | #ifdef CONFIG_NO_HZ_COMMON |
Thomas Gleixner | 06d8308 | 2008-03-22 09:20:24 +0100 | [diff] [blame] | 518 | /* |
Venkatesh Pallipadi | 83cd4fe | 2010-05-21 17:09:41 -0700 | [diff] [blame] | 519 | * In the semi idle case, use the nearest busy cpu for migrating timers |
| 520 | * from an idle cpu. This is good for power-savings. |
| 521 | * |
| 522 | * We don't do similar optimization for completely idle system, as |
| 523 | * selecting an idle cpu will add more delays to the timers than intended |
| 524 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). |
| 525 | */ |
Thomas Gleixner | bc7a34b | 2015-05-26 22:50:33 +0000 | [diff] [blame] | 526 | int get_nohz_timer_target(void) |
Venkatesh Pallipadi | 83cd4fe | 2010-05-21 17:09:41 -0700 | [diff] [blame] | 527 | { |
Thomas Gleixner | bc7a34b | 2015-05-26 22:50:33 +0000 | [diff] [blame] | 528 | int i, cpu = smp_processor_id(); |
Venkatesh Pallipadi | 83cd4fe | 2010-05-21 17:09:41 -0700 | [diff] [blame] | 529 | struct sched_domain *sd; |
| 530 | |
Vatika Harlalka | 9642d18 | 2015-09-01 16:50:59 +0200 | [diff] [blame] | 531 | if (!idle_cpu(cpu) && is_housekeeping_cpu(cpu)) |
Viresh Kumar | 6201b4d | 2014-03-18 16:26:07 +0530 | [diff] [blame] | 532 | return cpu; |
| 533 | |
Peter Zijlstra | 057f3fa | 2011-04-18 11:24:34 +0200 | [diff] [blame] | 534 | rcu_read_lock(); |
Venkatesh Pallipadi | 83cd4fe | 2010-05-21 17:09:41 -0700 | [diff] [blame] | 535 | for_each_domain(cpu, sd) { |
Peter Zijlstra | 057f3fa | 2011-04-18 11:24:34 +0200 | [diff] [blame] | 536 | for_each_cpu(i, sched_domain_span(sd)) { |
Wanpeng Li | 4449692 | 2016-05-04 14:45:34 +0800 | [diff] [blame] | 537 | if (cpu == i) |
| 538 | continue; |
| 539 | |
| 540 | if (!idle_cpu(i) && is_housekeeping_cpu(i)) { |
Peter Zijlstra | 057f3fa | 2011-04-18 11:24:34 +0200 | [diff] [blame] | 541 | cpu = i; |
| 542 | goto unlock; |
| 543 | } |
| 544 | } |
Venkatesh Pallipadi | 83cd4fe | 2010-05-21 17:09:41 -0700 | [diff] [blame] | 545 | } |
Vatika Harlalka | 9642d18 | 2015-09-01 16:50:59 +0200 | [diff] [blame] | 546 | |
| 547 | if (!is_housekeeping_cpu(cpu)) |
| 548 | cpu = housekeeping_any_cpu(); |
Peter Zijlstra | 057f3fa | 2011-04-18 11:24:34 +0200 | [diff] [blame] | 549 | unlock: |
| 550 | rcu_read_unlock(); |
Venkatesh Pallipadi | 83cd4fe | 2010-05-21 17:09:41 -0700 | [diff] [blame] | 551 | return cpu; |
| 552 | } |
| 553 | /* |
Thomas Gleixner | 06d8308 | 2008-03-22 09:20:24 +0100 | [diff] [blame] | 554 | * When add_timer_on() enqueues a timer into the timer wheel of an |
| 555 | * idle CPU then this timer might expire before the next timer event |
| 556 | * which is scheduled to wake up that CPU. In case of a completely |
| 557 | * idle system the next event might even be infinite time into the |
| 558 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and |
| 559 | * leaves the inner idle loop so the newly added timer is taken into |
| 560 | * account when the CPU goes back to idle and evaluates the timer |
| 561 | * wheel for the next timer event. |
| 562 | */ |
Frederic Weisbecker | 1c20091 | 2011-08-10 23:21:01 +0200 | [diff] [blame] | 563 | static void wake_up_idle_cpu(int cpu) |
Thomas Gleixner | 06d8308 | 2008-03-22 09:20:24 +0100 | [diff] [blame] | 564 | { |
| 565 | struct rq *rq = cpu_rq(cpu); |
| 566 | |
| 567 | if (cpu == smp_processor_id()) |
| 568 | return; |
| 569 | |
Andy Lutomirski | 67b9ca7 | 2014-06-04 10:31:17 -0700 | [diff] [blame] | 570 | if (set_nr_and_not_polling(rq->idle)) |
Thomas Gleixner | 06d8308 | 2008-03-22 09:20:24 +0100 | [diff] [blame] | 571 | smp_send_reschedule(cpu); |
Andy Lutomirski | dfc68f2 | 2014-06-04 10:31:15 -0700 | [diff] [blame] | 572 | else |
| 573 | trace_sched_wake_idle_without_ipi(cpu); |
Thomas Gleixner | 06d8308 | 2008-03-22 09:20:24 +0100 | [diff] [blame] | 574 | } |
Mike Galbraith | 39c0cbe | 2010-03-11 17:17:13 +0100 | [diff] [blame] | 575 | |
Frederic Weisbecker | c5bfece | 2013-04-12 16:45:34 +0200 | [diff] [blame] | 576 | static bool wake_up_full_nohz_cpu(int cpu) |
Frederic Weisbecker | 1c20091 | 2011-08-10 23:21:01 +0200 | [diff] [blame] | 577 | { |
Frederic Weisbecker | 53c5fa1 | 2014-06-04 16:20:21 +0200 | [diff] [blame] | 578 | /* |
| 579 | * We just need the target to call irq_exit() and re-evaluate |
| 580 | * the next tick. The nohz full kick at least implies that. |
| 581 | * If needed we can still optimize that later with an |
| 582 | * empty IRQ. |
| 583 | */ |
Paul E. McKenney | 379d9ec | 2016-06-30 10:37:20 -0700 | [diff] [blame] | 584 | if (cpu_is_offline(cpu)) |
| 585 | return true; /* Don't try to wake offline CPUs. */ |
Frederic Weisbecker | c5bfece | 2013-04-12 16:45:34 +0200 | [diff] [blame] | 586 | if (tick_nohz_full_cpu(cpu)) { |
Frederic Weisbecker | 1c20091 | 2011-08-10 23:21:01 +0200 | [diff] [blame] | 587 | if (cpu != smp_processor_id() || |
| 588 | tick_nohz_tick_stopped()) |
Frederic Weisbecker | 53c5fa1 | 2014-06-04 16:20:21 +0200 | [diff] [blame] | 589 | tick_nohz_full_kick_cpu(cpu); |
Frederic Weisbecker | 1c20091 | 2011-08-10 23:21:01 +0200 | [diff] [blame] | 590 | return true; |
| 591 | } |
| 592 | |
| 593 | return false; |
| 594 | } |
| 595 | |
Paul E. McKenney | 379d9ec | 2016-06-30 10:37:20 -0700 | [diff] [blame] | 596 | /* |
| 597 | * Wake up the specified CPU. If the CPU is going offline, it is the |
| 598 | * caller's responsibility to deal with the lost wakeup, for example, |
| 599 | * by hooking into the CPU_DEAD notifier like timers and hrtimers do. |
| 600 | */ |
Frederic Weisbecker | 1c20091 | 2011-08-10 23:21:01 +0200 | [diff] [blame] | 601 | void wake_up_nohz_cpu(int cpu) |
| 602 | { |
Frederic Weisbecker | c5bfece | 2013-04-12 16:45:34 +0200 | [diff] [blame] | 603 | if (!wake_up_full_nohz_cpu(cpu)) |
Frederic Weisbecker | 1c20091 | 2011-08-10 23:21:01 +0200 | [diff] [blame] | 604 | wake_up_idle_cpu(cpu); |
| 605 | } |
| 606 | |
Suresh Siddha | ca38062 | 2011-10-03 15:09:00 -0700 | [diff] [blame] | 607 | static inline bool got_nohz_idle_kick(void) |
| 608 | { |
Suresh Siddha | 1c792db | 2011-12-01 17:07:32 -0800 | [diff] [blame] | 609 | int cpu = smp_processor_id(); |
Vincent Guittot | 873b4c6 | 2013-06-05 10:13:11 +0200 | [diff] [blame] | 610 | |
| 611 | if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu))) |
| 612 | return false; |
| 613 | |
| 614 | if (idle_cpu(cpu) && !need_resched()) |
| 615 | return true; |
| 616 | |
| 617 | /* |
| 618 | * We can't run Idle Load Balance on this CPU for this time so we |
| 619 | * cancel it and clear NOHZ_BALANCE_KICK |
| 620 | */ |
| 621 | clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)); |
| 622 | return false; |
Suresh Siddha | ca38062 | 2011-10-03 15:09:00 -0700 | [diff] [blame] | 623 | } |
| 624 | |
Frederic Weisbecker | 3451d02 | 2011-08-10 23:21:01 +0200 | [diff] [blame] | 625 | #else /* CONFIG_NO_HZ_COMMON */ |
Suresh Siddha | ca38062 | 2011-10-03 15:09:00 -0700 | [diff] [blame] | 626 | |
| 627 | static inline bool got_nohz_idle_kick(void) |
| 628 | { |
| 629 | return false; |
| 630 | } |
| 631 | |
Frederic Weisbecker | 3451d02 | 2011-08-10 23:21:01 +0200 | [diff] [blame] | 632 | #endif /* CONFIG_NO_HZ_COMMON */ |
Thomas Gleixner | 06d8308 | 2008-03-22 09:20:24 +0100 | [diff] [blame] | 633 | |
Frederic Weisbecker | ce831b3 | 2013-04-20 15:15:35 +0200 | [diff] [blame] | 634 | #ifdef CONFIG_NO_HZ_FULL |
Frederic Weisbecker | 76d92ac | 2015-07-17 22:25:49 +0200 | [diff] [blame] | 635 | bool sched_can_stop_tick(struct rq *rq) |
Frederic Weisbecker | ce831b3 | 2013-04-20 15:15:35 +0200 | [diff] [blame] | 636 | { |
Frederic Weisbecker | 76d92ac | 2015-07-17 22:25:49 +0200 | [diff] [blame] | 637 | int fifo_nr_running; |
| 638 | |
| 639 | /* Deadline tasks, even if single, need the tick */ |
| 640 | if (rq->dl.dl_nr_running) |
| 641 | return false; |
| 642 | |
Frederic Weisbecker | 3882ec6 | 2014-03-18 22:54:04 +0100 | [diff] [blame] | 643 | /* |
Peter Zijlstra | 2548d54 | 2016-04-21 18:03:15 +0200 | [diff] [blame] | 644 | * If there are more than one RR tasks, we need the tick to effect the |
| 645 | * actual RR behaviour. |
Rik van Riel | 1e78cdb | 2015-02-16 15:23:49 -0500 | [diff] [blame] | 646 | */ |
Frederic Weisbecker | 76d92ac | 2015-07-17 22:25:49 +0200 | [diff] [blame] | 647 | if (rq->rt.rr_nr_running) { |
| 648 | if (rq->rt.rr_nr_running == 1) |
| 649 | return true; |
| 650 | else |
| 651 | return false; |
Rik van Riel | 1e78cdb | 2015-02-16 15:23:49 -0500 | [diff] [blame] | 652 | } |
| 653 | |
Peter Zijlstra | 2548d54 | 2016-04-21 18:03:15 +0200 | [diff] [blame] | 654 | /* |
| 655 | * If there's no RR tasks, but FIFO tasks, we can skip the tick, no |
| 656 | * forced preemption between FIFO tasks. |
| 657 | */ |
| 658 | fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running; |
| 659 | if (fifo_nr_running) |
| 660 | return true; |
| 661 | |
| 662 | /* |
| 663 | * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left; |
| 664 | * if there's more than one we need the tick for involuntary |
| 665 | * preemption. |
| 666 | */ |
| 667 | if (rq->nr_running > 1) |
Viresh Kumar | 541b826 | 2014-06-24 14:04:12 +0530 | [diff] [blame] | 668 | return false; |
Frederic Weisbecker | ce831b3 | 2013-04-20 15:15:35 +0200 | [diff] [blame] | 669 | |
Viresh Kumar | 541b826 | 2014-06-24 14:04:12 +0530 | [diff] [blame] | 670 | return true; |
Frederic Weisbecker | ce831b3 | 2013-04-20 15:15:35 +0200 | [diff] [blame] | 671 | } |
| 672 | #endif /* CONFIG_NO_HZ_FULL */ |
Ingo Molnar | 45bf76d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 673 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 674 | void sched_avg_update(struct rq *rq) |
Peter Zijlstra | e9e9250 | 2009-09-01 10:34:37 +0200 | [diff] [blame] | 675 | { |
| 676 | s64 period = sched_avg_period(); |
| 677 | |
Frederic Weisbecker | 78becc2 | 2013-04-12 01:51:02 +0200 | [diff] [blame] | 678 | while ((s64)(rq_clock(rq) - rq->age_stamp) > period) { |
Will Deacon | 0d98bb2 | 2010-05-24 12:11:43 -0700 | [diff] [blame] | 679 | /* |
| 680 | * Inline assembly required to prevent the compiler |
| 681 | * optimising this loop into a divmod call. |
| 682 | * See __iter_div_u64_rem() for another example of this. |
| 683 | */ |
| 684 | asm("" : "+rm" (rq->age_stamp)); |
Peter Zijlstra | e9e9250 | 2009-09-01 10:34:37 +0200 | [diff] [blame] | 685 | rq->age_stamp += period; |
| 686 | rq->rt_avg /= 2; |
| 687 | } |
| 688 | } |
| 689 | |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 690 | #endif /* CONFIG_SMP */ |
Ingo Molnar | 45bf76d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 691 | |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 692 | #if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \ |
| 693 | (defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH))) |
Peter Zijlstra | eb75580 | 2008-08-19 12:33:05 +0200 | [diff] [blame] | 694 | /* |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 695 | * Iterate task_group tree rooted at *from, calling @down when first entering a |
| 696 | * node and @up when leaving it for the final time. |
| 697 | * |
| 698 | * Caller must hold rcu_lock or sufficient equivalent. |
Peter Zijlstra | eb75580 | 2008-08-19 12:33:05 +0200 | [diff] [blame] | 699 | */ |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 700 | int walk_tg_tree_from(struct task_group *from, |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 701 | tg_visitor down, tg_visitor up, void *data) |
Peter Zijlstra | eb75580 | 2008-08-19 12:33:05 +0200 | [diff] [blame] | 702 | { |
| 703 | struct task_group *parent, *child; |
| 704 | int ret; |
| 705 | |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 706 | parent = from; |
| 707 | |
Peter Zijlstra | eb75580 | 2008-08-19 12:33:05 +0200 | [diff] [blame] | 708 | down: |
| 709 | ret = (*down)(parent, data); |
| 710 | if (ret) |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 711 | goto out; |
Peter Zijlstra | eb75580 | 2008-08-19 12:33:05 +0200 | [diff] [blame] | 712 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
| 713 | parent = child; |
| 714 | goto down; |
| 715 | |
| 716 | up: |
| 717 | continue; |
| 718 | } |
| 719 | ret = (*up)(parent, data); |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 720 | if (ret || parent == from) |
| 721 | goto out; |
Peter Zijlstra | eb75580 | 2008-08-19 12:33:05 +0200 | [diff] [blame] | 722 | |
| 723 | child = parent; |
| 724 | parent = parent->parent; |
| 725 | if (parent) |
| 726 | goto up; |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 727 | out: |
Peter Zijlstra | eb75580 | 2008-08-19 12:33:05 +0200 | [diff] [blame] | 728 | return ret; |
| 729 | } |
| 730 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 731 | int tg_nop(struct task_group *tg, void *data) |
Peter Zijlstra | eb75580 | 2008-08-19 12:33:05 +0200 | [diff] [blame] | 732 | { |
| 733 | return 0; |
| 734 | } |
| 735 | #endif |
| 736 | |
Ingo Molnar | 45bf76d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 737 | static void set_load_weight(struct task_struct *p) |
| 738 | { |
Nikhil Rao | f05998d | 2011-05-18 10:09:38 -0700 | [diff] [blame] | 739 | int prio = p->static_prio - MAX_RT_PRIO; |
| 740 | struct load_weight *load = &p->se.load; |
| 741 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 742 | /* |
| 743 | * SCHED_IDLE tasks get minimal weight: |
| 744 | */ |
Henrik Austad | 20f9cd2 | 2015-09-09 17:00:41 +0200 | [diff] [blame] | 745 | if (idle_policy(p->policy)) { |
Nikhil Rao | c8b2811 | 2011-05-18 14:37:48 -0700 | [diff] [blame] | 746 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
Nikhil Rao | f05998d | 2011-05-18 10:09:38 -0700 | [diff] [blame] | 747 | load->inv_weight = WMULT_IDLEPRIO; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 748 | return; |
| 749 | } |
| 750 | |
Andi Kleen | ed82b8a | 2015-11-29 20:59:43 -0800 | [diff] [blame] | 751 | load->weight = scale_load(sched_prio_to_weight[prio]); |
| 752 | load->inv_weight = sched_prio_to_wmult[prio]; |
Ingo Molnar | 45bf76d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 753 | } |
| 754 | |
Peter Zijlstra | 1de6444 | 2015-09-30 17:44:13 +0200 | [diff] [blame] | 755 | static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
Gregory Haskins | 2087a1a | 2008-06-27 14:30:00 -0600 | [diff] [blame] | 756 | { |
Mike Galbraith | a64692a | 2010-03-11 17:16:20 +0100 | [diff] [blame] | 757 | update_rq_clock(rq); |
Peter Zijlstra | 1de6444 | 2015-09-30 17:44:13 +0200 | [diff] [blame] | 758 | if (!(flags & ENQUEUE_RESTORE)) |
| 759 | sched_info_queued(rq, p); |
Peter Zijlstra | 371fd7e | 2010-03-24 16:38:48 +0100 | [diff] [blame] | 760 | p->sched_class->enqueue_task(rq, p, flags); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 761 | } |
| 762 | |
Peter Zijlstra | 1de6444 | 2015-09-30 17:44:13 +0200 | [diff] [blame] | 763 | static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 764 | { |
Mike Galbraith | a64692a | 2010-03-11 17:16:20 +0100 | [diff] [blame] | 765 | update_rq_clock(rq); |
Peter Zijlstra | 1de6444 | 2015-09-30 17:44:13 +0200 | [diff] [blame] | 766 | if (!(flags & DEQUEUE_SAVE)) |
| 767 | sched_info_dequeued(rq, p); |
Peter Zijlstra | 371fd7e | 2010-03-24 16:38:48 +0100 | [diff] [blame] | 768 | p->sched_class->dequeue_task(rq, p, flags); |
Ingo Molnar | 71f8bd4 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 769 | } |
| 770 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 771 | void activate_task(struct rq *rq, struct task_struct *p, int flags) |
Peter Zijlstra | 1e3c88b | 2009-12-17 17:00:43 +0100 | [diff] [blame] | 772 | { |
| 773 | if (task_contributes_to_load(p)) |
| 774 | rq->nr_uninterruptible--; |
| 775 | |
Peter Zijlstra | 371fd7e | 2010-03-24 16:38:48 +0100 | [diff] [blame] | 776 | enqueue_task(rq, p, flags); |
Peter Zijlstra | 1e3c88b | 2009-12-17 17:00:43 +0100 | [diff] [blame] | 777 | } |
| 778 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 779 | void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
Peter Zijlstra | 1e3c88b | 2009-12-17 17:00:43 +0100 | [diff] [blame] | 780 | { |
| 781 | if (task_contributes_to_load(p)) |
| 782 | rq->nr_uninterruptible++; |
| 783 | |
Peter Zijlstra | 371fd7e | 2010-03-24 16:38:48 +0100 | [diff] [blame] | 784 | dequeue_task(rq, p, flags); |
Peter Zijlstra | 1e3c88b | 2009-12-17 17:00:43 +0100 | [diff] [blame] | 785 | } |
| 786 | |
Peter Zijlstra | fe44d62 | 2010-12-09 14:15:34 +0100 | [diff] [blame] | 787 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
Venkatesh Pallipadi | aa48380 | 2010-10-04 17:03:22 -0700 | [diff] [blame] | 788 | { |
Glauber Costa | 095c0aa | 2011-07-11 15:28:18 -0400 | [diff] [blame] | 789 | /* |
| 790 | * In theory, the compile should just see 0 here, and optimize out the call |
| 791 | * to sched_rt_avg_update. But I don't trust it... |
| 792 | */ |
| 793 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
| 794 | s64 steal = 0, irq_delta = 0; |
| 795 | #endif |
| 796 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
Peter Zijlstra | 8e92c20 | 2010-12-09 14:15:34 +0100 | [diff] [blame] | 797 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
Peter Zijlstra | fe44d62 | 2010-12-09 14:15:34 +0100 | [diff] [blame] | 798 | |
| 799 | /* |
| 800 | * Since irq_time is only updated on {soft,}irq_exit, we might run into |
| 801 | * this case when a previous update_rq_clock() happened inside a |
| 802 | * {soft,}irq region. |
| 803 | * |
| 804 | * When this happens, we stop ->clock_task and only update the |
| 805 | * prev_irq_time stamp to account for the part that fit, so that a next |
| 806 | * update will consume the rest. This ensures ->clock_task is |
| 807 | * monotonic. |
| 808 | * |
| 809 | * It does however cause some slight miss-attribution of {soft,}irq |
| 810 | * time, a more accurate solution would be to update the irq_time using |
| 811 | * the current rq->clock timestamp, except that would require using |
| 812 | * atomic ops. |
| 813 | */ |
| 814 | if (irq_delta > delta) |
| 815 | irq_delta = delta; |
| 816 | |
| 817 | rq->prev_irq_time += irq_delta; |
| 818 | delta -= irq_delta; |
Glauber Costa | 095c0aa | 2011-07-11 15:28:18 -0400 | [diff] [blame] | 819 | #endif |
| 820 | #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING |
Ingo Molnar | c5905af | 2012-02-24 08:31:31 +0100 | [diff] [blame] | 821 | if (static_key_false((¶virt_steal_rq_enabled))) { |
Glauber Costa | 095c0aa | 2011-07-11 15:28:18 -0400 | [diff] [blame] | 822 | steal = paravirt_steal_clock(cpu_of(rq)); |
| 823 | steal -= rq->prev_steal_time_rq; |
| 824 | |
| 825 | if (unlikely(steal > delta)) |
| 826 | steal = delta; |
| 827 | |
Glauber Costa | 095c0aa | 2011-07-11 15:28:18 -0400 | [diff] [blame] | 828 | rq->prev_steal_time_rq += steal; |
Glauber Costa | 095c0aa | 2011-07-11 15:28:18 -0400 | [diff] [blame] | 829 | delta -= steal; |
| 830 | } |
| 831 | #endif |
| 832 | |
Peter Zijlstra | fe44d62 | 2010-12-09 14:15:34 +0100 | [diff] [blame] | 833 | rq->clock_task += delta; |
| 834 | |
Glauber Costa | 095c0aa | 2011-07-11 15:28:18 -0400 | [diff] [blame] | 835 | #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) |
Nicolas Pitre | 5d4dfdd | 2014-05-27 13:50:41 -0400 | [diff] [blame] | 836 | if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY)) |
Glauber Costa | 095c0aa | 2011-07-11 15:28:18 -0400 | [diff] [blame] | 837 | sched_rt_avg_update(rq, irq_delta + steal); |
| 838 | #endif |
Venkatesh Pallipadi | aa48380 | 2010-10-04 17:03:22 -0700 | [diff] [blame] | 839 | } |
| 840 | |
Peter Zijlstra | 34f971f | 2010-09-22 13:53:15 +0200 | [diff] [blame] | 841 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
| 842 | { |
| 843 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; |
| 844 | struct task_struct *old_stop = cpu_rq(cpu)->stop; |
| 845 | |
| 846 | if (stop) { |
| 847 | /* |
| 848 | * Make it appear like a SCHED_FIFO task, its something |
| 849 | * userspace knows about and won't get confused about. |
| 850 | * |
| 851 | * Also, it will make PI more or less work without too |
| 852 | * much confusion -- but then, stop work should not |
| 853 | * rely on PI working anyway. |
| 854 | */ |
| 855 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); |
| 856 | |
| 857 | stop->sched_class = &stop_sched_class; |
| 858 | } |
| 859 | |
| 860 | cpu_rq(cpu)->stop = stop; |
| 861 | |
| 862 | if (old_stop) { |
| 863 | /* |
| 864 | * Reset it back to a normal scheduling class so that |
| 865 | * it can die in pieces. |
| 866 | */ |
| 867 | old_stop->sched_class = &rt_sched_class; |
| 868 | } |
| 869 | } |
| 870 | |
Peter Zijlstra | 1e3c88b | 2009-12-17 17:00:43 +0100 | [diff] [blame] | 871 | /* |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 872 | * __normal_prio - return the priority that is based on the static prio |
Ingo Molnar | 71f8bd4 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 873 | */ |
Ingo Molnar | 1453118 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 874 | static inline int __normal_prio(struct task_struct *p) |
| 875 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 876 | return p->static_prio; |
Ingo Molnar | 1453118 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 877 | } |
| 878 | |
| 879 | /* |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 880 | * Calculate the expected normal priority: i.e. priority |
| 881 | * without taking RT-inheritance into account. Might be |
| 882 | * boosted by interactivity modifiers. Changes upon fork, |
| 883 | * setprio syscalls, and whenever the interactivity |
| 884 | * estimator recalculates. |
| 885 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 886 | static inline int normal_prio(struct task_struct *p) |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 887 | { |
| 888 | int prio; |
| 889 | |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 890 | if (task_has_dl_policy(p)) |
| 891 | prio = MAX_DL_PRIO-1; |
| 892 | else if (task_has_rt_policy(p)) |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 893 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
| 894 | else |
| 895 | prio = __normal_prio(p); |
| 896 | return prio; |
| 897 | } |
| 898 | |
| 899 | /* |
| 900 | * Calculate the current priority, i.e. the priority |
| 901 | * taken into account by the scheduler. This value might |
| 902 | * be boosted by RT tasks, or might be boosted by |
| 903 | * interactivity modifiers. Will be RT if the task got |
| 904 | * RT-boosted. If not then it returns p->normal_prio. |
| 905 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 906 | static int effective_prio(struct task_struct *p) |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 907 | { |
| 908 | p->normal_prio = normal_prio(p); |
| 909 | /* |
| 910 | * If we are RT tasks or we were boosted to RT priority, |
| 911 | * keep the priority unchanged. Otherwise, update priority |
| 912 | * to the normal priority: |
| 913 | */ |
| 914 | if (!rt_prio(p->prio)) |
| 915 | return p->normal_prio; |
| 916 | return p->prio; |
| 917 | } |
| 918 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 919 | /** |
| 920 | * task_curr - is this task currently executing on a CPU? |
| 921 | * @p: the task in question. |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 922 | * |
| 923 | * Return: 1 if the task is currently executing. 0 otherwise. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 924 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 925 | inline int task_curr(const struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 926 | { |
| 927 | return cpu_curr(task_cpu(p)) == p; |
| 928 | } |
| 929 | |
Kirill Tkhai | 67dfa1b | 2014-10-27 17:40:52 +0300 | [diff] [blame] | 930 | /* |
Peter Zijlstra | 4c9a4bc | 2015-06-11 14:46:39 +0200 | [diff] [blame] | 931 | * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock, |
| 932 | * use the balance_callback list if you want balancing. |
| 933 | * |
| 934 | * this means any call to check_class_changed() must be followed by a call to |
| 935 | * balance_callback(). |
Kirill Tkhai | 67dfa1b | 2014-10-27 17:40:52 +0300 | [diff] [blame] | 936 | */ |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 937 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
| 938 | const struct sched_class *prev_class, |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 939 | int oldprio) |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 940 | { |
| 941 | if (prev_class != p->sched_class) { |
| 942 | if (prev_class->switched_from) |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 943 | prev_class->switched_from(rq, p); |
Peter Zijlstra | 4c9a4bc | 2015-06-11 14:46:39 +0200 | [diff] [blame] | 944 | |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 945 | p->sched_class->switched_to(rq, p); |
Dario Faggioli | 2d3d891 | 2013-11-07 14:43:44 +0100 | [diff] [blame] | 946 | } else if (oldprio != p->prio || dl_task(p)) |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 947 | p->sched_class->prio_changed(rq, p, oldprio); |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 948 | } |
| 949 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 950 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
Peter Zijlstra | 1e5a740 | 2010-10-31 12:37:04 +0100 | [diff] [blame] | 951 | { |
| 952 | const struct sched_class *class; |
| 953 | |
| 954 | if (p->sched_class == rq->curr->sched_class) { |
| 955 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
| 956 | } else { |
| 957 | for_each_class(class) { |
| 958 | if (class == rq->curr->sched_class) |
| 959 | break; |
| 960 | if (class == p->sched_class) { |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 961 | resched_curr(rq); |
Peter Zijlstra | 1e5a740 | 2010-10-31 12:37:04 +0100 | [diff] [blame] | 962 | break; |
| 963 | } |
| 964 | } |
| 965 | } |
| 966 | |
| 967 | /* |
| 968 | * A queue event has occurred, and we're going to schedule. In |
| 969 | * this case, we can save a useless back to back clock update. |
| 970 | */ |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 971 | if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr)) |
Peter Zijlstra | 9edfbfe | 2015-01-05 11:18:11 +0100 | [diff] [blame] | 972 | rq_clock_skip_update(rq, true); |
Peter Zijlstra | 1e5a740 | 2010-10-31 12:37:04 +0100 | [diff] [blame] | 973 | } |
| 974 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 975 | #ifdef CONFIG_SMP |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 976 | /* |
| 977 | * This is how migration works: |
| 978 | * |
| 979 | * 1) we invoke migration_cpu_stop() on the target CPU using |
| 980 | * stop_one_cpu(). |
| 981 | * 2) stopper starts to run (implicitly forcing the migrated thread |
| 982 | * off the CPU) |
| 983 | * 3) it checks whether the migrated task is still in the wrong runqueue. |
| 984 | * 4) if it's in the wrong runqueue then the migration thread removes |
| 985 | * it and puts it into the right queue. |
| 986 | * 5) stopper completes and stop_one_cpu() returns and the migration |
| 987 | * is done. |
| 988 | */ |
| 989 | |
| 990 | /* |
| 991 | * move_queued_task - move a queued task to new rq. |
| 992 | * |
| 993 | * Returns (locked) new rq. Old rq's lock is released. |
| 994 | */ |
Peter Zijlstra | 5e16bbc | 2015-06-11 14:46:51 +0200 | [diff] [blame] | 995 | static struct rq *move_queued_task(struct rq *rq, struct task_struct *p, int new_cpu) |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 996 | { |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 997 | lockdep_assert_held(&rq->lock); |
| 998 | |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 999 | p->on_rq = TASK_ON_RQ_MIGRATING; |
Joonwoo Park | 3ea94de | 2015-11-12 19:38:54 -0800 | [diff] [blame] | 1000 | dequeue_task(rq, p, 0); |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1001 | set_task_cpu(p, new_cpu); |
| 1002 | raw_spin_unlock(&rq->lock); |
| 1003 | |
| 1004 | rq = cpu_rq(new_cpu); |
| 1005 | |
| 1006 | raw_spin_lock(&rq->lock); |
| 1007 | BUG_ON(task_cpu(p) != new_cpu); |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1008 | enqueue_task(rq, p, 0); |
Joonwoo Park | 3ea94de | 2015-11-12 19:38:54 -0800 | [diff] [blame] | 1009 | p->on_rq = TASK_ON_RQ_QUEUED; |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1010 | check_preempt_curr(rq, p, 0); |
| 1011 | |
| 1012 | return rq; |
| 1013 | } |
| 1014 | |
| 1015 | struct migration_arg { |
| 1016 | struct task_struct *task; |
| 1017 | int dest_cpu; |
| 1018 | }; |
| 1019 | |
| 1020 | /* |
| 1021 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
| 1022 | * this because either it can't run here any more (set_cpus_allowed() |
| 1023 | * away from this CPU, or CPU going down), or because we're |
| 1024 | * attempting to rebalance this task on exec (sched_exec). |
| 1025 | * |
| 1026 | * So we race with normal scheduler movements, but that's OK, as long |
| 1027 | * as the task is no longer on this CPU. |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1028 | */ |
Peter Zijlstra | 5e16bbc | 2015-06-11 14:46:51 +0200 | [diff] [blame] | 1029 | static struct rq *__migrate_task(struct rq *rq, struct task_struct *p, int dest_cpu) |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1030 | { |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1031 | if (unlikely(!cpu_active(dest_cpu))) |
Peter Zijlstra | 5e16bbc | 2015-06-11 14:46:51 +0200 | [diff] [blame] | 1032 | return rq; |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1033 | |
| 1034 | /* Affinity changed (again). */ |
| 1035 | if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
Peter Zijlstra | 5e16bbc | 2015-06-11 14:46:51 +0200 | [diff] [blame] | 1036 | return rq; |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1037 | |
Peter Zijlstra | 5e16bbc | 2015-06-11 14:46:51 +0200 | [diff] [blame] | 1038 | rq = move_queued_task(rq, p, dest_cpu); |
| 1039 | |
| 1040 | return rq; |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1041 | } |
| 1042 | |
| 1043 | /* |
| 1044 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
| 1045 | * and performs thread migration by bumping thread off CPU then |
| 1046 | * 'pushing' onto another runqueue. |
| 1047 | */ |
| 1048 | static int migration_cpu_stop(void *data) |
| 1049 | { |
| 1050 | struct migration_arg *arg = data; |
Peter Zijlstra | 5e16bbc | 2015-06-11 14:46:51 +0200 | [diff] [blame] | 1051 | struct task_struct *p = arg->task; |
| 1052 | struct rq *rq = this_rq(); |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1053 | |
| 1054 | /* |
| 1055 | * The original target cpu might have gone down and we might |
| 1056 | * be on another cpu but it doesn't matter. |
| 1057 | */ |
| 1058 | local_irq_disable(); |
| 1059 | /* |
| 1060 | * We need to explicitly wake pending tasks before running |
| 1061 | * __migrate_task() such that we will not miss enforcing cpus_allowed |
| 1062 | * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test. |
| 1063 | */ |
| 1064 | sched_ttwu_pending(); |
Peter Zijlstra | 5e16bbc | 2015-06-11 14:46:51 +0200 | [diff] [blame] | 1065 | |
| 1066 | raw_spin_lock(&p->pi_lock); |
| 1067 | raw_spin_lock(&rq->lock); |
| 1068 | /* |
| 1069 | * If task_rq(p) != rq, it cannot be migrated here, because we're |
| 1070 | * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because |
| 1071 | * we're holding p->pi_lock. |
| 1072 | */ |
Cheng Chao | bf89a30 | 2016-09-14 10:01:50 +0800 | [diff] [blame] | 1073 | if (task_rq(p) == rq) { |
| 1074 | if (task_on_rq_queued(p)) |
| 1075 | rq = __migrate_task(rq, p, arg->dest_cpu); |
| 1076 | else |
| 1077 | p->wake_cpu = arg->dest_cpu; |
| 1078 | } |
Peter Zijlstra | 5e16bbc | 2015-06-11 14:46:51 +0200 | [diff] [blame] | 1079 | raw_spin_unlock(&rq->lock); |
| 1080 | raw_spin_unlock(&p->pi_lock); |
| 1081 | |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1082 | local_irq_enable(); |
| 1083 | return 0; |
| 1084 | } |
| 1085 | |
Peter Zijlstra | c5b2803 | 2015-05-15 17:43:35 +0200 | [diff] [blame] | 1086 | /* |
| 1087 | * sched_class::set_cpus_allowed must do the below, but is not required to |
| 1088 | * actually call this function. |
| 1089 | */ |
| 1090 | void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask) |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1091 | { |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1092 | cpumask_copy(&p->cpus_allowed, new_mask); |
| 1093 | p->nr_cpus_allowed = cpumask_weight(new_mask); |
| 1094 | } |
| 1095 | |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1096 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
| 1097 | { |
Peter Zijlstra | 6c37067 | 2015-05-15 17:43:36 +0200 | [diff] [blame] | 1098 | struct rq *rq = task_rq(p); |
| 1099 | bool queued, running; |
| 1100 | |
Peter Zijlstra | 25834c7 | 2015-05-15 17:43:34 +0200 | [diff] [blame] | 1101 | lockdep_assert_held(&p->pi_lock); |
Peter Zijlstra | 6c37067 | 2015-05-15 17:43:36 +0200 | [diff] [blame] | 1102 | |
| 1103 | queued = task_on_rq_queued(p); |
| 1104 | running = task_current(rq, p); |
| 1105 | |
| 1106 | if (queued) { |
| 1107 | /* |
| 1108 | * Because __kthread_bind() calls this on blocked tasks without |
| 1109 | * holding rq->lock. |
| 1110 | */ |
| 1111 | lockdep_assert_held(&rq->lock); |
Peter Zijlstra | 1de6444 | 2015-09-30 17:44:13 +0200 | [diff] [blame] | 1112 | dequeue_task(rq, p, DEQUEUE_SAVE); |
Peter Zijlstra | 6c37067 | 2015-05-15 17:43:36 +0200 | [diff] [blame] | 1113 | } |
| 1114 | if (running) |
| 1115 | put_prev_task(rq, p); |
| 1116 | |
Peter Zijlstra | c5b2803 | 2015-05-15 17:43:35 +0200 | [diff] [blame] | 1117 | p->sched_class->set_cpus_allowed(p, new_mask); |
Peter Zijlstra | 6c37067 | 2015-05-15 17:43:36 +0200 | [diff] [blame] | 1118 | |
Peter Zijlstra | 6c37067 | 2015-05-15 17:43:36 +0200 | [diff] [blame] | 1119 | if (queued) |
Peter Zijlstra | 1de6444 | 2015-09-30 17:44:13 +0200 | [diff] [blame] | 1120 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
Vincent Guittot | a399d23 | 2016-09-12 09:47:52 +0200 | [diff] [blame] | 1121 | if (running) |
Peter Zijlstra | b2bf6c3 | 2016-09-20 22:00:38 +0200 | [diff] [blame] | 1122 | set_curr_task(rq, p); |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1123 | } |
| 1124 | |
| 1125 | /* |
| 1126 | * Change a given task's CPU affinity. Migrate the thread to a |
| 1127 | * proper CPU and schedule it away if the CPU it's executing on |
| 1128 | * is removed from the allowed bitmask. |
| 1129 | * |
| 1130 | * NOTE: the caller must have a valid reference to the task, the |
| 1131 | * task must not exit() & deallocate itself prematurely. The |
| 1132 | * call is not atomic; no spinlocks may be held. |
| 1133 | */ |
Peter Zijlstra | 25834c7 | 2015-05-15 17:43:34 +0200 | [diff] [blame] | 1134 | static int __set_cpus_allowed_ptr(struct task_struct *p, |
| 1135 | const struct cpumask *new_mask, bool check) |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1136 | { |
Peter Zijlstra (Intel) | e9d867a | 2016-03-10 12:54:08 +0100 | [diff] [blame] | 1137 | const struct cpumask *cpu_valid_mask = cpu_active_mask; |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1138 | unsigned int dest_cpu; |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 1139 | struct rq_flags rf; |
| 1140 | struct rq *rq; |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1141 | int ret = 0; |
| 1142 | |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 1143 | rq = task_rq_lock(p, &rf); |
Wanpeng Li | ab3d531 | 2017-02-21 23:52:55 -0800 | [diff] [blame] | 1144 | update_rq_clock(rq); |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1145 | |
Peter Zijlstra (Intel) | e9d867a | 2016-03-10 12:54:08 +0100 | [diff] [blame] | 1146 | if (p->flags & PF_KTHREAD) { |
| 1147 | /* |
| 1148 | * Kernel threads are allowed on online && !active CPUs |
| 1149 | */ |
| 1150 | cpu_valid_mask = cpu_online_mask; |
| 1151 | } |
| 1152 | |
Peter Zijlstra | 25834c7 | 2015-05-15 17:43:34 +0200 | [diff] [blame] | 1153 | /* |
| 1154 | * Must re-check here, to close a race against __kthread_bind(), |
| 1155 | * sched_setaffinity() is not guaranteed to observe the flag. |
| 1156 | */ |
| 1157 | if (check && (p->flags & PF_NO_SETAFFINITY)) { |
| 1158 | ret = -EINVAL; |
| 1159 | goto out; |
| 1160 | } |
| 1161 | |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1162 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
| 1163 | goto out; |
| 1164 | |
Peter Zijlstra (Intel) | e9d867a | 2016-03-10 12:54:08 +0100 | [diff] [blame] | 1165 | if (!cpumask_intersects(new_mask, cpu_valid_mask)) { |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1166 | ret = -EINVAL; |
| 1167 | goto out; |
| 1168 | } |
| 1169 | |
| 1170 | do_set_cpus_allowed(p, new_mask); |
| 1171 | |
Peter Zijlstra (Intel) | e9d867a | 2016-03-10 12:54:08 +0100 | [diff] [blame] | 1172 | if (p->flags & PF_KTHREAD) { |
| 1173 | /* |
| 1174 | * For kernel threads that do indeed end up on online && |
| 1175 | * !active we want to ensure they are strict per-cpu threads. |
| 1176 | */ |
| 1177 | WARN_ON(cpumask_intersects(new_mask, cpu_online_mask) && |
| 1178 | !cpumask_intersects(new_mask, cpu_active_mask) && |
| 1179 | p->nr_cpus_allowed != 1); |
| 1180 | } |
| 1181 | |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1182 | /* Can the task run on the task's current CPU? If so, we're done */ |
| 1183 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
| 1184 | goto out; |
| 1185 | |
Peter Zijlstra (Intel) | e9d867a | 2016-03-10 12:54:08 +0100 | [diff] [blame] | 1186 | dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask); |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1187 | if (task_running(rq, p) || p->state == TASK_WAKING) { |
| 1188 | struct migration_arg arg = { p, dest_cpu }; |
| 1189 | /* Need help from migration thread: drop lock and wait. */ |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 1190 | task_rq_unlock(rq, p, &rf); |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1191 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
| 1192 | tlb_migrate_finish(p->mm); |
| 1193 | return 0; |
Peter Zijlstra | cbce1a6 | 2015-06-11 14:46:54 +0200 | [diff] [blame] | 1194 | } else if (task_on_rq_queued(p)) { |
| 1195 | /* |
| 1196 | * OK, since we're going to drop the lock immediately |
| 1197 | * afterwards anyway. |
| 1198 | */ |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 1199 | lockdep_unpin_lock(&rq->lock, rf.cookie); |
Peter Zijlstra | 5e16bbc | 2015-06-11 14:46:51 +0200 | [diff] [blame] | 1200 | rq = move_queued_task(rq, p, dest_cpu); |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 1201 | lockdep_repin_lock(&rq->lock, rf.cookie); |
Peter Zijlstra | cbce1a6 | 2015-06-11 14:46:54 +0200 | [diff] [blame] | 1202 | } |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1203 | out: |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 1204 | task_rq_unlock(rq, p, &rf); |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1205 | |
| 1206 | return ret; |
| 1207 | } |
Peter Zijlstra | 25834c7 | 2015-05-15 17:43:34 +0200 | [diff] [blame] | 1208 | |
| 1209 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
| 1210 | { |
| 1211 | return __set_cpus_allowed_ptr(p, new_mask, false); |
| 1212 | } |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1213 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
| 1214 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1215 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
Ingo Molnar | c65cc87 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1216 | { |
Peter Zijlstra | e291200 | 2009-12-16 18:04:36 +0100 | [diff] [blame] | 1217 | #ifdef CONFIG_SCHED_DEBUG |
| 1218 | /* |
| 1219 | * We should never call set_task_cpu() on a blocked task, |
| 1220 | * ttwu() will sort out the placement. |
| 1221 | */ |
Peter Zijlstra | 077614e | 2009-12-17 13:16:31 +0100 | [diff] [blame] | 1222 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
Oleg Nesterov | e2336f6 | 2014-10-08 20:33:48 +0200 | [diff] [blame] | 1223 | !p->on_rq); |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 1224 | |
Joonwoo Park | 3ea94de | 2015-11-12 19:38:54 -0800 | [diff] [blame] | 1225 | /* |
| 1226 | * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING, |
| 1227 | * because schedstat_wait_{start,end} rebase migrating task's wait_start |
| 1228 | * time relying on p->on_rq. |
| 1229 | */ |
| 1230 | WARN_ON_ONCE(p->state == TASK_RUNNING && |
| 1231 | p->sched_class == &fair_sched_class && |
| 1232 | (p->on_rq && !task_on_rq_migrating(p))); |
| 1233 | |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 1234 | #ifdef CONFIG_LOCKDEP |
Peter Zijlstra | 6c6c54e | 2011-06-03 17:37:07 +0200 | [diff] [blame] | 1235 | /* |
| 1236 | * The caller should hold either p->pi_lock or rq->lock, when changing |
| 1237 | * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks. |
| 1238 | * |
| 1239 | * sched_move_task() holds both and thus holding either pins the cgroup, |
Peter Zijlstra | 8323f26 | 2012-06-22 13:36:05 +0200 | [diff] [blame] | 1240 | * see task_group(). |
Peter Zijlstra | 6c6c54e | 2011-06-03 17:37:07 +0200 | [diff] [blame] | 1241 | * |
| 1242 | * Furthermore, all task_rq users should acquire both locks, see |
| 1243 | * task_rq_lock(). |
| 1244 | */ |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 1245 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || |
| 1246 | lockdep_is_held(&task_rq(p)->lock))); |
| 1247 | #endif |
Peter Zijlstra | e291200 | 2009-12-16 18:04:36 +0100 | [diff] [blame] | 1248 | #endif |
| 1249 | |
Mathieu Desnoyers | de1d728 | 2009-05-05 16:49:59 +0800 | [diff] [blame] | 1250 | trace_sched_migrate_task(p, new_cpu); |
Peter Zijlstra | cbc34ed | 2008-12-10 08:08:22 +0100 | [diff] [blame] | 1251 | |
Peter Zijlstra | 0c69774 | 2009-12-22 15:43:19 +0100 | [diff] [blame] | 1252 | if (task_cpu(p) != new_cpu) { |
Paul Turner | 0a74bef | 2012-10-04 13:18:30 +0200 | [diff] [blame] | 1253 | if (p->sched_class->migrate_task_rq) |
xiaofeng.yan | 5a4fd03 | 2015-09-23 14:55:59 +0800 | [diff] [blame] | 1254 | p->sched_class->migrate_task_rq(p); |
Peter Zijlstra | 0c69774 | 2009-12-22 15:43:19 +0100 | [diff] [blame] | 1255 | p->se.nr_migrations++; |
Peter Zijlstra | ff303e6 | 2015-04-17 20:05:30 +0200 | [diff] [blame] | 1256 | perf_event_task_migrate(p); |
Peter Zijlstra | 0c69774 | 2009-12-22 15:43:19 +0100 | [diff] [blame] | 1257 | } |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1258 | |
| 1259 | __set_task_cpu(p, new_cpu); |
Ingo Molnar | c65cc87 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1260 | } |
| 1261 | |
Peter Zijlstra | ac66f54 | 2013-10-07 11:29:16 +0100 | [diff] [blame] | 1262 | static void __migrate_swap_task(struct task_struct *p, int cpu) |
| 1263 | { |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 1264 | if (task_on_rq_queued(p)) { |
Peter Zijlstra | ac66f54 | 2013-10-07 11:29:16 +0100 | [diff] [blame] | 1265 | struct rq *src_rq, *dst_rq; |
| 1266 | |
| 1267 | src_rq = task_rq(p); |
| 1268 | dst_rq = cpu_rq(cpu); |
| 1269 | |
Joonwoo Park | 3ea94de | 2015-11-12 19:38:54 -0800 | [diff] [blame] | 1270 | p->on_rq = TASK_ON_RQ_MIGRATING; |
Peter Zijlstra | ac66f54 | 2013-10-07 11:29:16 +0100 | [diff] [blame] | 1271 | deactivate_task(src_rq, p, 0); |
| 1272 | set_task_cpu(p, cpu); |
| 1273 | activate_task(dst_rq, p, 0); |
Joonwoo Park | 3ea94de | 2015-11-12 19:38:54 -0800 | [diff] [blame] | 1274 | p->on_rq = TASK_ON_RQ_QUEUED; |
Peter Zijlstra | ac66f54 | 2013-10-07 11:29:16 +0100 | [diff] [blame] | 1275 | check_preempt_curr(dst_rq, p, 0); |
| 1276 | } else { |
| 1277 | /* |
| 1278 | * Task isn't running anymore; make it appear like we migrated |
| 1279 | * it before it went to sleep. This means on wakeup we make the |
Leo Yan | a1fd465 | 2016-08-05 14:32:38 +0800 | [diff] [blame] | 1280 | * previous cpu our target instead of where it really is. |
Peter Zijlstra | ac66f54 | 2013-10-07 11:29:16 +0100 | [diff] [blame] | 1281 | */ |
| 1282 | p->wake_cpu = cpu; |
| 1283 | } |
| 1284 | } |
| 1285 | |
| 1286 | struct migration_swap_arg { |
| 1287 | struct task_struct *src_task, *dst_task; |
| 1288 | int src_cpu, dst_cpu; |
| 1289 | }; |
| 1290 | |
| 1291 | static int migrate_swap_stop(void *data) |
| 1292 | { |
| 1293 | struct migration_swap_arg *arg = data; |
| 1294 | struct rq *src_rq, *dst_rq; |
| 1295 | int ret = -EAGAIN; |
| 1296 | |
Peter Zijlstra | 62694cd | 2015-10-09 18:36:29 +0200 | [diff] [blame] | 1297 | if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu)) |
| 1298 | return -EAGAIN; |
| 1299 | |
Peter Zijlstra | ac66f54 | 2013-10-07 11:29:16 +0100 | [diff] [blame] | 1300 | src_rq = cpu_rq(arg->src_cpu); |
| 1301 | dst_rq = cpu_rq(arg->dst_cpu); |
| 1302 | |
Peter Zijlstra | 7460231 | 2013-10-10 20:17:22 +0200 | [diff] [blame] | 1303 | double_raw_lock(&arg->src_task->pi_lock, |
| 1304 | &arg->dst_task->pi_lock); |
Peter Zijlstra | ac66f54 | 2013-10-07 11:29:16 +0100 | [diff] [blame] | 1305 | double_rq_lock(src_rq, dst_rq); |
Peter Zijlstra | 62694cd | 2015-10-09 18:36:29 +0200 | [diff] [blame] | 1306 | |
Peter Zijlstra | ac66f54 | 2013-10-07 11:29:16 +0100 | [diff] [blame] | 1307 | if (task_cpu(arg->dst_task) != arg->dst_cpu) |
| 1308 | goto unlock; |
| 1309 | |
| 1310 | if (task_cpu(arg->src_task) != arg->src_cpu) |
| 1311 | goto unlock; |
| 1312 | |
| 1313 | if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task))) |
| 1314 | goto unlock; |
| 1315 | |
| 1316 | if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task))) |
| 1317 | goto unlock; |
| 1318 | |
| 1319 | __migrate_swap_task(arg->src_task, arg->dst_cpu); |
| 1320 | __migrate_swap_task(arg->dst_task, arg->src_cpu); |
| 1321 | |
| 1322 | ret = 0; |
| 1323 | |
| 1324 | unlock: |
| 1325 | double_rq_unlock(src_rq, dst_rq); |
Peter Zijlstra | 7460231 | 2013-10-10 20:17:22 +0200 | [diff] [blame] | 1326 | raw_spin_unlock(&arg->dst_task->pi_lock); |
| 1327 | raw_spin_unlock(&arg->src_task->pi_lock); |
Peter Zijlstra | ac66f54 | 2013-10-07 11:29:16 +0100 | [diff] [blame] | 1328 | |
| 1329 | return ret; |
| 1330 | } |
| 1331 | |
| 1332 | /* |
| 1333 | * Cross migrate two tasks |
| 1334 | */ |
| 1335 | int migrate_swap(struct task_struct *cur, struct task_struct *p) |
| 1336 | { |
| 1337 | struct migration_swap_arg arg; |
| 1338 | int ret = -EINVAL; |
| 1339 | |
Peter Zijlstra | ac66f54 | 2013-10-07 11:29:16 +0100 | [diff] [blame] | 1340 | arg = (struct migration_swap_arg){ |
| 1341 | .src_task = cur, |
| 1342 | .src_cpu = task_cpu(cur), |
| 1343 | .dst_task = p, |
| 1344 | .dst_cpu = task_cpu(p), |
| 1345 | }; |
| 1346 | |
| 1347 | if (arg.src_cpu == arg.dst_cpu) |
| 1348 | goto out; |
| 1349 | |
Peter Zijlstra | 6acce3e | 2013-10-11 14:38:20 +0200 | [diff] [blame] | 1350 | /* |
| 1351 | * These three tests are all lockless; this is OK since all of them |
| 1352 | * will be re-checked with proper locks held further down the line. |
| 1353 | */ |
Peter Zijlstra | ac66f54 | 2013-10-07 11:29:16 +0100 | [diff] [blame] | 1354 | if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu)) |
| 1355 | goto out; |
| 1356 | |
| 1357 | if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task))) |
| 1358 | goto out; |
| 1359 | |
| 1360 | if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task))) |
| 1361 | goto out; |
| 1362 | |
Mel Gorman | 286549d | 2014-01-21 15:51:03 -0800 | [diff] [blame] | 1363 | trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu); |
Peter Zijlstra | ac66f54 | 2013-10-07 11:29:16 +0100 | [diff] [blame] | 1364 | ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg); |
| 1365 | |
| 1366 | out: |
Peter Zijlstra | ac66f54 | 2013-10-07 11:29:16 +0100 | [diff] [blame] | 1367 | return ret; |
| 1368 | } |
| 1369 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1370 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1371 | * wait_task_inactive - wait for a thread to unschedule. |
| 1372 | * |
Roland McGrath | 85ba2d8 | 2008-07-25 19:45:58 -0700 | [diff] [blame] | 1373 | * If @match_state is nonzero, it's the @p->state value just checked and |
| 1374 | * not expected to change. If it changes, i.e. @p might have woken up, |
| 1375 | * then return zero. When we succeed in waiting for @p to be off its CPU, |
| 1376 | * we return a positive number (its total switch count). If a second call |
| 1377 | * a short while later returns the same number, the caller can be sure that |
| 1378 | * @p has remained unscheduled the whole time. |
| 1379 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1380 | * The caller must ensure that the task *will* unschedule sometime soon, |
| 1381 | * else this function might spin for a *long* time. This function can't |
| 1382 | * be called with interrupts off, or it may introduce deadlock with |
| 1383 | * smp_call_function() if an IPI is sent by the same process we are |
| 1384 | * waiting to become inactive. |
| 1385 | */ |
Roland McGrath | 85ba2d8 | 2008-07-25 19:45:58 -0700 | [diff] [blame] | 1386 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1387 | { |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 1388 | int running, queued; |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 1389 | struct rq_flags rf; |
Roland McGrath | 85ba2d8 | 2008-07-25 19:45:58 -0700 | [diff] [blame] | 1390 | unsigned long ncsw; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 1391 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1392 | |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 1393 | for (;;) { |
| 1394 | /* |
| 1395 | * We do the initial early heuristics without holding |
| 1396 | * any task-queue locks at all. We'll only try to get |
| 1397 | * the runqueue lock when things look like they will |
| 1398 | * work out! |
| 1399 | */ |
| 1400 | rq = task_rq(p); |
Linus Torvalds | fa490cf | 2007-06-18 09:34:40 -0700 | [diff] [blame] | 1401 | |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 1402 | /* |
| 1403 | * If the task is actively running on another CPU |
| 1404 | * still, just relax and busy-wait without holding |
| 1405 | * any locks. |
| 1406 | * |
| 1407 | * NOTE! Since we don't hold any locks, it's not |
| 1408 | * even sure that "rq" stays as the right runqueue! |
| 1409 | * But we don't care, since "task_running()" will |
| 1410 | * return false if the runqueue has changed and p |
| 1411 | * is actually now running somewhere else! |
| 1412 | */ |
Roland McGrath | 85ba2d8 | 2008-07-25 19:45:58 -0700 | [diff] [blame] | 1413 | while (task_running(rq, p)) { |
| 1414 | if (match_state && unlikely(p->state != match_state)) |
| 1415 | return 0; |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 1416 | cpu_relax(); |
Roland McGrath | 85ba2d8 | 2008-07-25 19:45:58 -0700 | [diff] [blame] | 1417 | } |
Linus Torvalds | fa490cf | 2007-06-18 09:34:40 -0700 | [diff] [blame] | 1418 | |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 1419 | /* |
| 1420 | * Ok, time to look more closely! We need the rq |
| 1421 | * lock now, to be *sure*. If we're wrong, we'll |
| 1422 | * just go back and repeat. |
| 1423 | */ |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 1424 | rq = task_rq_lock(p, &rf); |
Peter Zijlstra | 27a9da6 | 2010-05-04 20:36:56 +0200 | [diff] [blame] | 1425 | trace_sched_wait_task(p); |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 1426 | running = task_running(rq, p); |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 1427 | queued = task_on_rq_queued(p); |
Roland McGrath | 85ba2d8 | 2008-07-25 19:45:58 -0700 | [diff] [blame] | 1428 | ncsw = 0; |
Oleg Nesterov | f31e11d | 2008-08-20 16:54:44 -0700 | [diff] [blame] | 1429 | if (!match_state || p->state == match_state) |
Oleg Nesterov | 93dcf55 | 2008-08-20 16:54:44 -0700 | [diff] [blame] | 1430 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 1431 | task_rq_unlock(rq, p, &rf); |
Linus Torvalds | fa490cf | 2007-06-18 09:34:40 -0700 | [diff] [blame] | 1432 | |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 1433 | /* |
Roland McGrath | 85ba2d8 | 2008-07-25 19:45:58 -0700 | [diff] [blame] | 1434 | * If it changed from the expected state, bail out now. |
| 1435 | */ |
| 1436 | if (unlikely(!ncsw)) |
| 1437 | break; |
| 1438 | |
| 1439 | /* |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 1440 | * Was it really running after all now that we |
| 1441 | * checked with the proper locks actually held? |
| 1442 | * |
| 1443 | * Oops. Go back and try again.. |
| 1444 | */ |
| 1445 | if (unlikely(running)) { |
| 1446 | cpu_relax(); |
| 1447 | continue; |
| 1448 | } |
| 1449 | |
| 1450 | /* |
| 1451 | * It's not enough that it's not actively running, |
| 1452 | * it must be off the runqueue _entirely_, and not |
| 1453 | * preempted! |
| 1454 | * |
Luis Henriques | 80dd99b | 2009-03-16 19:58:09 +0000 | [diff] [blame] | 1455 | * So if it was still runnable (but just not actively |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 1456 | * running right now), it's preempted, and we should |
| 1457 | * yield - it could be a while. |
| 1458 | */ |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 1459 | if (unlikely(queued)) { |
Thomas Gleixner | 8eb90c3 | 2011-02-23 23:52:21 +0000 | [diff] [blame] | 1460 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
| 1461 | |
| 1462 | set_current_state(TASK_UNINTERRUPTIBLE); |
| 1463 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 1464 | continue; |
| 1465 | } |
| 1466 | |
| 1467 | /* |
| 1468 | * Ahh, all good. It wasn't running, and it wasn't |
| 1469 | * runnable, which means that it will never become |
| 1470 | * running in the future either. We're all done! |
| 1471 | */ |
| 1472 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1473 | } |
Roland McGrath | 85ba2d8 | 2008-07-25 19:45:58 -0700 | [diff] [blame] | 1474 | |
| 1475 | return ncsw; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1476 | } |
| 1477 | |
| 1478 | /*** |
| 1479 | * kick_process - kick a running thread to enter/exit the kernel |
| 1480 | * @p: the to-be-kicked thread |
| 1481 | * |
| 1482 | * Cause a process which is running on another CPU to enter |
| 1483 | * kernel-mode, without any delay. (to get signals handled.) |
| 1484 | * |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 1485 | * NOTE: this function doesn't have to take the runqueue lock, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1486 | * because all it wants to ensure is that the remote task enters |
| 1487 | * the kernel. If the IPI races and the task has been migrated |
| 1488 | * to another CPU then no harm is done and the purpose has been |
| 1489 | * achieved as well. |
| 1490 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 1491 | void kick_process(struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1492 | { |
| 1493 | int cpu; |
| 1494 | |
| 1495 | preempt_disable(); |
| 1496 | cpu = task_cpu(p); |
| 1497 | if ((cpu != smp_processor_id()) && task_curr(p)) |
| 1498 | smp_send_reschedule(cpu); |
| 1499 | preempt_enable(); |
| 1500 | } |
Rusty Russell | b43e352 | 2009-06-12 22:27:00 -0600 | [diff] [blame] | 1501 | EXPORT_SYMBOL_GPL(kick_process); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1502 | |
Oleg Nesterov | 30da688 | 2010-03-15 10:10:19 +0100 | [diff] [blame] | 1503 | /* |
Peter Zijlstra | 013fdb8 | 2011-04-05 17:23:45 +0200 | [diff] [blame] | 1504 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
Peter Zijlstra (Intel) | e9d867a | 2016-03-10 12:54:08 +0100 | [diff] [blame] | 1505 | * |
| 1506 | * A few notes on cpu_active vs cpu_online: |
| 1507 | * |
| 1508 | * - cpu_active must be a subset of cpu_online |
| 1509 | * |
| 1510 | * - on cpu-up we allow per-cpu kthreads on the online && !active cpu, |
| 1511 | * see __set_cpus_allowed_ptr(). At this point the newly online |
| 1512 | * cpu isn't yet part of the sched domains, and balancing will not |
| 1513 | * see it. |
| 1514 | * |
| 1515 | * - on cpu-down we clear cpu_active() to mask the sched domains and |
| 1516 | * avoid the load balancer to place new tasks on the to be removed |
| 1517 | * cpu. Existing tasks will remain running there and will be taken |
| 1518 | * off. |
| 1519 | * |
| 1520 | * This means that fallback selection must not select !active CPUs. |
| 1521 | * And can assume that any active CPU must be online. Conversely |
| 1522 | * select_task_rq() below may allow selection of !active CPUs in order |
| 1523 | * to satisfy the above rules. |
Oleg Nesterov | 30da688 | 2010-03-15 10:10:19 +0100 | [diff] [blame] | 1524 | */ |
Peter Zijlstra | 5da9a0f | 2009-12-16 18:04:38 +0100 | [diff] [blame] | 1525 | static int select_fallback_rq(int cpu, struct task_struct *p) |
| 1526 | { |
Tang Chen | aa00d89 | 2013-02-22 16:33:33 -0800 | [diff] [blame] | 1527 | int nid = cpu_to_node(cpu); |
| 1528 | const struct cpumask *nodemask = NULL; |
Peter Zijlstra | 2baab4e | 2012-03-20 15:57:01 +0100 | [diff] [blame] | 1529 | enum { cpuset, possible, fail } state = cpuset; |
| 1530 | int dest_cpu; |
Peter Zijlstra | 5da9a0f | 2009-12-16 18:04:38 +0100 | [diff] [blame] | 1531 | |
Tang Chen | aa00d89 | 2013-02-22 16:33:33 -0800 | [diff] [blame] | 1532 | /* |
| 1533 | * If the node that the cpu is on has been offlined, cpu_to_node() |
| 1534 | * will return -1. There is no cpu on the node, and we should |
| 1535 | * select the cpu on the other node. |
| 1536 | */ |
| 1537 | if (nid != -1) { |
| 1538 | nodemask = cpumask_of_node(nid); |
| 1539 | |
| 1540 | /* Look for allowed, online CPU in same node. */ |
| 1541 | for_each_cpu(dest_cpu, nodemask) { |
Tang Chen | aa00d89 | 2013-02-22 16:33:33 -0800 | [diff] [blame] | 1542 | if (!cpu_active(dest_cpu)) |
| 1543 | continue; |
| 1544 | if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p))) |
| 1545 | return dest_cpu; |
| 1546 | } |
Peter Zijlstra | 2baab4e | 2012-03-20 15:57:01 +0100 | [diff] [blame] | 1547 | } |
Peter Zijlstra | 5da9a0f | 2009-12-16 18:04:38 +0100 | [diff] [blame] | 1548 | |
Peter Zijlstra | 2baab4e | 2012-03-20 15:57:01 +0100 | [diff] [blame] | 1549 | for (;;) { |
| 1550 | /* Any allowed, online CPU? */ |
Srivatsa S. Bhat | e3831ed | 2012-03-30 19:40:28 +0530 | [diff] [blame] | 1551 | for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) { |
Tejun Heo | feb245e | 2016-06-16 15:35:04 -0400 | [diff] [blame] | 1552 | if (!(p->flags & PF_KTHREAD) && !cpu_active(dest_cpu)) |
| 1553 | continue; |
| 1554 | if (!cpu_online(dest_cpu)) |
Peter Zijlstra | 2baab4e | 2012-03-20 15:57:01 +0100 | [diff] [blame] | 1555 | continue; |
| 1556 | goto out; |
| 1557 | } |
Peter Zijlstra | 5da9a0f | 2009-12-16 18:04:38 +0100 | [diff] [blame] | 1558 | |
Oleg Nesterov | e73e85f | 2015-10-10 20:53:15 +0200 | [diff] [blame] | 1559 | /* No more Mr. Nice Guy. */ |
Peter Zijlstra | 2baab4e | 2012-03-20 15:57:01 +0100 | [diff] [blame] | 1560 | switch (state) { |
| 1561 | case cpuset: |
Oleg Nesterov | e73e85f | 2015-10-10 20:53:15 +0200 | [diff] [blame] | 1562 | if (IS_ENABLED(CONFIG_CPUSETS)) { |
| 1563 | cpuset_cpus_allowed_fallback(p); |
| 1564 | state = possible; |
| 1565 | break; |
| 1566 | } |
| 1567 | /* fall-through */ |
Peter Zijlstra | 2baab4e | 2012-03-20 15:57:01 +0100 | [diff] [blame] | 1568 | case possible: |
| 1569 | do_set_cpus_allowed(p, cpu_possible_mask); |
| 1570 | state = fail; |
| 1571 | break; |
| 1572 | |
| 1573 | case fail: |
| 1574 | BUG(); |
| 1575 | break; |
| 1576 | } |
| 1577 | } |
| 1578 | |
| 1579 | out: |
| 1580 | if (state != cpuset) { |
| 1581 | /* |
| 1582 | * Don't tell them about moving exiting tasks or |
| 1583 | * kernel threads (both mm NULL), since they never |
| 1584 | * leave kernel. |
| 1585 | */ |
| 1586 | if (p->mm && printk_ratelimit()) { |
John Stultz | aac74dc | 2014-06-04 16:11:40 -0700 | [diff] [blame] | 1587 | printk_deferred("process %d (%s) no longer affine to cpu%d\n", |
Peter Zijlstra | 2baab4e | 2012-03-20 15:57:01 +0100 | [diff] [blame] | 1588 | task_pid_nr(p), p->comm, cpu); |
| 1589 | } |
Peter Zijlstra | 5da9a0f | 2009-12-16 18:04:38 +0100 | [diff] [blame] | 1590 | } |
| 1591 | |
| 1592 | return dest_cpu; |
| 1593 | } |
| 1594 | |
Peter Zijlstra | e291200 | 2009-12-16 18:04:36 +0100 | [diff] [blame] | 1595 | /* |
Peter Zijlstra | 013fdb8 | 2011-04-05 17:23:45 +0200 | [diff] [blame] | 1596 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
Peter Zijlstra | e291200 | 2009-12-16 18:04:36 +0100 | [diff] [blame] | 1597 | */ |
Peter Zijlstra | 970b13b | 2009-11-25 13:31:39 +0100 | [diff] [blame] | 1598 | static inline |
Peter Zijlstra | ac66f54 | 2013-10-07 11:29:16 +0100 | [diff] [blame] | 1599 | int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags) |
Peter Zijlstra | 970b13b | 2009-11-25 13:31:39 +0100 | [diff] [blame] | 1600 | { |
Peter Zijlstra | cbce1a6 | 2015-06-11 14:46:54 +0200 | [diff] [blame] | 1601 | lockdep_assert_held(&p->pi_lock); |
| 1602 | |
Thomas Gleixner | 50605ff | 2016-05-11 14:23:31 +0200 | [diff] [blame] | 1603 | if (tsk_nr_cpus_allowed(p) > 1) |
Wanpeng Li | 6c1d941 | 2014-11-05 09:14:37 +0800 | [diff] [blame] | 1604 | cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags); |
Peter Zijlstra (Intel) | e9d867a | 2016-03-10 12:54:08 +0100 | [diff] [blame] | 1605 | else |
| 1606 | cpu = cpumask_any(tsk_cpus_allowed(p)); |
Peter Zijlstra | e291200 | 2009-12-16 18:04:36 +0100 | [diff] [blame] | 1607 | |
| 1608 | /* |
| 1609 | * In order not to call set_task_cpu() on a blocking task we need |
| 1610 | * to rely on ttwu() to place the task on a valid ->cpus_allowed |
| 1611 | * cpu. |
| 1612 | * |
| 1613 | * Since this is common to all placement strategies, this lives here. |
| 1614 | * |
| 1615 | * [ this allows ->select_task() to simply return task_cpu(p) and |
| 1616 | * not worry about this generic constraint ] |
| 1617 | */ |
Peter Zijlstra | fa17b50 | 2011-06-16 12:23:22 +0200 | [diff] [blame] | 1618 | if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) || |
Peter Zijlstra | 70f1120 | 2009-12-20 17:36:27 +0100 | [diff] [blame] | 1619 | !cpu_online(cpu))) |
Peter Zijlstra | 5da9a0f | 2009-12-16 18:04:38 +0100 | [diff] [blame] | 1620 | cpu = select_fallback_rq(task_cpu(p), p); |
Peter Zijlstra | e291200 | 2009-12-16 18:04:36 +0100 | [diff] [blame] | 1621 | |
| 1622 | return cpu; |
Peter Zijlstra | 970b13b | 2009-11-25 13:31:39 +0100 | [diff] [blame] | 1623 | } |
Mike Galbraith | 09a40af | 2010-04-15 07:29:59 +0200 | [diff] [blame] | 1624 | |
| 1625 | static void update_avg(u64 *avg, u64 sample) |
| 1626 | { |
| 1627 | s64 diff = sample - *avg; |
| 1628 | *avg += diff >> 3; |
| 1629 | } |
Peter Zijlstra | 25834c7 | 2015-05-15 17:43:34 +0200 | [diff] [blame] | 1630 | |
| 1631 | #else |
| 1632 | |
| 1633 | static inline int __set_cpus_allowed_ptr(struct task_struct *p, |
| 1634 | const struct cpumask *new_mask, bool check) |
| 1635 | { |
| 1636 | return set_cpus_allowed_ptr(p, new_mask); |
| 1637 | } |
| 1638 | |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 1639 | #endif /* CONFIG_SMP */ |
Peter Zijlstra | 970b13b | 2009-11-25 13:31:39 +0100 | [diff] [blame] | 1640 | |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 1641 | static void |
Peter Zijlstra | b84cb5d | 2011-04-05 17:23:55 +0200 | [diff] [blame] | 1642 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1643 | { |
Josh Poimboeuf | 4fa8d299 | 2016-06-17 12:43:26 -0500 | [diff] [blame] | 1644 | struct rq *rq; |
| 1645 | |
| 1646 | if (!schedstat_enabled()) |
| 1647 | return; |
| 1648 | |
| 1649 | rq = this_rq(); |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1650 | |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 1651 | #ifdef CONFIG_SMP |
Josh Poimboeuf | 4fa8d299 | 2016-06-17 12:43:26 -0500 | [diff] [blame] | 1652 | if (cpu == rq->cpu) { |
Josh Poimboeuf | ae92882 | 2016-06-17 12:43:24 -0500 | [diff] [blame] | 1653 | schedstat_inc(rq->ttwu_local); |
| 1654 | schedstat_inc(p->se.statistics.nr_wakeups_local); |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 1655 | } else { |
| 1656 | struct sched_domain *sd; |
| 1657 | |
Josh Poimboeuf | ae92882 | 2016-06-17 12:43:24 -0500 | [diff] [blame] | 1658 | schedstat_inc(p->se.statistics.nr_wakeups_remote); |
Peter Zijlstra | 057f3fa | 2011-04-18 11:24:34 +0200 | [diff] [blame] | 1659 | rcu_read_lock(); |
Josh Poimboeuf | 4fa8d299 | 2016-06-17 12:43:26 -0500 | [diff] [blame] | 1660 | for_each_domain(rq->cpu, sd) { |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 1661 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
Josh Poimboeuf | ae92882 | 2016-06-17 12:43:24 -0500 | [diff] [blame] | 1662 | schedstat_inc(sd->ttwu_wake_remote); |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 1663 | break; |
| 1664 | } |
| 1665 | } |
Peter Zijlstra | 057f3fa | 2011-04-18 11:24:34 +0200 | [diff] [blame] | 1666 | rcu_read_unlock(); |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 1667 | } |
Peter Zijlstra | f339b9d | 2011-05-31 10:49:20 +0200 | [diff] [blame] | 1668 | |
| 1669 | if (wake_flags & WF_MIGRATED) |
Josh Poimboeuf | ae92882 | 2016-06-17 12:43:24 -0500 | [diff] [blame] | 1670 | schedstat_inc(p->se.statistics.nr_wakeups_migrate); |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 1671 | #endif /* CONFIG_SMP */ |
| 1672 | |
Josh Poimboeuf | ae92882 | 2016-06-17 12:43:24 -0500 | [diff] [blame] | 1673 | schedstat_inc(rq->ttwu_count); |
| 1674 | schedstat_inc(p->se.statistics.nr_wakeups); |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 1675 | |
| 1676 | if (wake_flags & WF_SYNC) |
Josh Poimboeuf | ae92882 | 2016-06-17 12:43:24 -0500 | [diff] [blame] | 1677 | schedstat_inc(p->se.statistics.nr_wakeups_sync); |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1678 | } |
| 1679 | |
Peter Zijlstra | 1de6444 | 2015-09-30 17:44:13 +0200 | [diff] [blame] | 1680 | static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1681 | { |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1682 | activate_task(rq, p, en_flags); |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 1683 | p->on_rq = TASK_ON_RQ_QUEUED; |
Peter Zijlstra | c2f7115 | 2011-04-13 13:28:56 +0200 | [diff] [blame] | 1684 | |
| 1685 | /* if a worker is waking up, notify workqueue */ |
| 1686 | if (p->flags & PF_WQ_WORKER) |
| 1687 | wq_worker_waking_up(p, cpu_of(rq)); |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1688 | } |
| 1689 | |
Peter Zijlstra | 23f41ee | 2011-04-05 17:23:56 +0200 | [diff] [blame] | 1690 | /* |
| 1691 | * Mark the task runnable and perform wakeup-preemption. |
| 1692 | */ |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 1693 | static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags, |
| 1694 | struct pin_cookie cookie) |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1695 | { |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1696 | check_preempt_curr(rq, p, wake_flags); |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1697 | p->state = TASK_RUNNING; |
Peter Zijlstra | fbd705a | 2015-06-09 11:13:36 +0200 | [diff] [blame] | 1698 | trace_sched_wakeup(p); |
| 1699 | |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1700 | #ifdef CONFIG_SMP |
Peter Zijlstra | 4c9a4bc | 2015-06-11 14:46:39 +0200 | [diff] [blame] | 1701 | if (p->sched_class->task_woken) { |
| 1702 | /* |
Peter Zijlstra | cbce1a6 | 2015-06-11 14:46:54 +0200 | [diff] [blame] | 1703 | * Our task @p is fully woken up and running; so its safe to |
| 1704 | * drop the rq->lock, hereafter rq is only used for statistics. |
Peter Zijlstra | 4c9a4bc | 2015-06-11 14:46:39 +0200 | [diff] [blame] | 1705 | */ |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 1706 | lockdep_unpin_lock(&rq->lock, cookie); |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1707 | p->sched_class->task_woken(rq, p); |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 1708 | lockdep_repin_lock(&rq->lock, cookie); |
Peter Zijlstra | 4c9a4bc | 2015-06-11 14:46:39 +0200 | [diff] [blame] | 1709 | } |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1710 | |
Steven Rostedt | e69c634 | 2010-12-06 17:10:31 -0500 | [diff] [blame] | 1711 | if (rq->idle_stamp) { |
Frederic Weisbecker | 78becc2 | 2013-04-12 01:51:02 +0200 | [diff] [blame] | 1712 | u64 delta = rq_clock(rq) - rq->idle_stamp; |
Jason Low | 9bd721c | 2013-09-13 11:26:52 -0700 | [diff] [blame] | 1713 | u64 max = 2*rq->max_idle_balance_cost; |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1714 | |
Jason Low | abfafa5 | 2013-09-13 11:26:51 -0700 | [diff] [blame] | 1715 | update_avg(&rq->avg_idle, delta); |
| 1716 | |
| 1717 | if (rq->avg_idle > max) |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1718 | rq->avg_idle = max; |
Jason Low | abfafa5 | 2013-09-13 11:26:51 -0700 | [diff] [blame] | 1719 | |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1720 | rq->idle_stamp = 0; |
| 1721 | } |
| 1722 | #endif |
| 1723 | } |
| 1724 | |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1725 | static void |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 1726 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags, |
| 1727 | struct pin_cookie cookie) |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1728 | { |
Peter Zijlstra | b5179ac | 2016-05-11 16:10:34 +0200 | [diff] [blame] | 1729 | int en_flags = ENQUEUE_WAKEUP; |
| 1730 | |
Peter Zijlstra | cbce1a6 | 2015-06-11 14:46:54 +0200 | [diff] [blame] | 1731 | lockdep_assert_held(&rq->lock); |
| 1732 | |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1733 | #ifdef CONFIG_SMP |
| 1734 | if (p->sched_contributes_to_load) |
| 1735 | rq->nr_uninterruptible--; |
Peter Zijlstra | b5179ac | 2016-05-11 16:10:34 +0200 | [diff] [blame] | 1736 | |
Peter Zijlstra | b5179ac | 2016-05-11 16:10:34 +0200 | [diff] [blame] | 1737 | if (wake_flags & WF_MIGRATED) |
Peter Zijlstra | 59efa0b | 2016-05-10 18:24:37 +0200 | [diff] [blame] | 1738 | en_flags |= ENQUEUE_MIGRATED; |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1739 | #endif |
| 1740 | |
Peter Zijlstra | b5179ac | 2016-05-11 16:10:34 +0200 | [diff] [blame] | 1741 | ttwu_activate(rq, p, en_flags); |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 1742 | ttwu_do_wakeup(rq, p, wake_flags, cookie); |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1743 | } |
| 1744 | |
| 1745 | /* |
| 1746 | * Called in case the task @p isn't fully descheduled from its runqueue, |
| 1747 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, |
| 1748 | * since all we need to do is flip p->state to TASK_RUNNING, since |
| 1749 | * the task is still ->on_rq. |
| 1750 | */ |
| 1751 | static int ttwu_remote(struct task_struct *p, int wake_flags) |
| 1752 | { |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 1753 | struct rq_flags rf; |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1754 | struct rq *rq; |
| 1755 | int ret = 0; |
| 1756 | |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 1757 | rq = __task_rq_lock(p, &rf); |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 1758 | if (task_on_rq_queued(p)) { |
Frederic Weisbecker | 1ad4ec0 | 2013-04-12 01:51:00 +0200 | [diff] [blame] | 1759 | /* check_preempt_curr() may use rq clock */ |
| 1760 | update_rq_clock(rq); |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 1761 | ttwu_do_wakeup(rq, p, wake_flags, rf.cookie); |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1762 | ret = 1; |
| 1763 | } |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 1764 | __task_rq_unlock(rq, &rf); |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1765 | |
| 1766 | return ret; |
| 1767 | } |
| 1768 | |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 1769 | #ifdef CONFIG_SMP |
Peter Zijlstra | e3baac4 | 2014-06-04 10:31:18 -0700 | [diff] [blame] | 1770 | void sched_ttwu_pending(void) |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 1771 | { |
| 1772 | struct rq *rq = this_rq(); |
Peter Zijlstra | fa14ff4 | 2011-09-12 13:06:17 +0200 | [diff] [blame] | 1773 | struct llist_node *llist = llist_del_all(&rq->wake_list); |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 1774 | struct pin_cookie cookie; |
Peter Zijlstra | fa14ff4 | 2011-09-12 13:06:17 +0200 | [diff] [blame] | 1775 | struct task_struct *p; |
Peter Zijlstra | e3baac4 | 2014-06-04 10:31:18 -0700 | [diff] [blame] | 1776 | unsigned long flags; |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 1777 | |
Peter Zijlstra | e3baac4 | 2014-06-04 10:31:18 -0700 | [diff] [blame] | 1778 | if (!llist) |
| 1779 | return; |
| 1780 | |
| 1781 | raw_spin_lock_irqsave(&rq->lock, flags); |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 1782 | cookie = lockdep_pin_lock(&rq->lock); |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 1783 | |
Peter Zijlstra | fa14ff4 | 2011-09-12 13:06:17 +0200 | [diff] [blame] | 1784 | while (llist) { |
Peter Zijlstra | b7e7ade | 2016-05-23 11:19:07 +0200 | [diff] [blame] | 1785 | int wake_flags = 0; |
| 1786 | |
Peter Zijlstra | fa14ff4 | 2011-09-12 13:06:17 +0200 | [diff] [blame] | 1787 | p = llist_entry(llist, struct task_struct, wake_entry); |
| 1788 | llist = llist_next(llist); |
Peter Zijlstra | b7e7ade | 2016-05-23 11:19:07 +0200 | [diff] [blame] | 1789 | |
| 1790 | if (p->sched_remote_wakeup) |
| 1791 | wake_flags = WF_MIGRATED; |
| 1792 | |
| 1793 | ttwu_do_activate(rq, p, wake_flags, cookie); |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 1794 | } |
| 1795 | |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 1796 | lockdep_unpin_lock(&rq->lock, cookie); |
Peter Zijlstra | e3baac4 | 2014-06-04 10:31:18 -0700 | [diff] [blame] | 1797 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 1798 | } |
| 1799 | |
| 1800 | void scheduler_ipi(void) |
| 1801 | { |
Peter Zijlstra | f27dde8 | 2013-08-14 14:55:31 +0200 | [diff] [blame] | 1802 | /* |
| 1803 | * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting |
| 1804 | * TIF_NEED_RESCHED remotely (for the first time) will also send |
| 1805 | * this IPI. |
| 1806 | */ |
Peter Zijlstra | 8cb75e0 | 2013-11-20 12:22:37 +0100 | [diff] [blame] | 1807 | preempt_fold_need_resched(); |
Peter Zijlstra | f27dde8 | 2013-08-14 14:55:31 +0200 | [diff] [blame] | 1808 | |
Frederic Weisbecker | fd2ac4f | 2014-03-18 21:12:53 +0100 | [diff] [blame] | 1809 | if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick()) |
Peter Zijlstra | c5d753a | 2011-07-19 15:07:25 -0700 | [diff] [blame] | 1810 | return; |
| 1811 | |
| 1812 | /* |
| 1813 | * Not all reschedule IPI handlers call irq_enter/irq_exit, since |
| 1814 | * traditionally all their work was done from the interrupt return |
| 1815 | * path. Now that we actually do some work, we need to make sure |
| 1816 | * we do call them. |
| 1817 | * |
| 1818 | * Some archs already do call them, luckily irq_enter/exit nest |
| 1819 | * properly. |
| 1820 | * |
| 1821 | * Arguably we should visit all archs and update all handlers, |
| 1822 | * however a fair share of IPIs are still resched only so this would |
| 1823 | * somewhat pessimize the simple resched case. |
| 1824 | */ |
| 1825 | irq_enter(); |
Peter Zijlstra | fa14ff4 | 2011-09-12 13:06:17 +0200 | [diff] [blame] | 1826 | sched_ttwu_pending(); |
Suresh Siddha | ca38062 | 2011-10-03 15:09:00 -0700 | [diff] [blame] | 1827 | |
| 1828 | /* |
| 1829 | * Check if someone kicked us for doing the nohz idle load balance. |
| 1830 | */ |
Vincent Guittot | 873b4c6 | 2013-06-05 10:13:11 +0200 | [diff] [blame] | 1831 | if (unlikely(got_nohz_idle_kick())) { |
Suresh Siddha | 6eb57e0 | 2011-10-03 15:09:01 -0700 | [diff] [blame] | 1832 | this_rq()->idle_balance = 1; |
Suresh Siddha | ca38062 | 2011-10-03 15:09:00 -0700 | [diff] [blame] | 1833 | raise_softirq_irqoff(SCHED_SOFTIRQ); |
Suresh Siddha | 6eb57e0 | 2011-10-03 15:09:01 -0700 | [diff] [blame] | 1834 | } |
Peter Zijlstra | c5d753a | 2011-07-19 15:07:25 -0700 | [diff] [blame] | 1835 | irq_exit(); |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 1836 | } |
| 1837 | |
Peter Zijlstra | b7e7ade | 2016-05-23 11:19:07 +0200 | [diff] [blame] | 1838 | static void ttwu_queue_remote(struct task_struct *p, int cpu, int wake_flags) |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 1839 | { |
Peter Zijlstra | e3baac4 | 2014-06-04 10:31:18 -0700 | [diff] [blame] | 1840 | struct rq *rq = cpu_rq(cpu); |
| 1841 | |
Peter Zijlstra | b7e7ade | 2016-05-23 11:19:07 +0200 | [diff] [blame] | 1842 | p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED); |
| 1843 | |
Peter Zijlstra | e3baac4 | 2014-06-04 10:31:18 -0700 | [diff] [blame] | 1844 | if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) { |
| 1845 | if (!set_nr_if_polling(rq->idle)) |
| 1846 | smp_send_reschedule(cpu); |
| 1847 | else |
| 1848 | trace_sched_wake_idle_without_ipi(cpu); |
| 1849 | } |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 1850 | } |
Peter Zijlstra | d6aa8f8 | 2011-05-26 14:21:33 +0200 | [diff] [blame] | 1851 | |
Chuansheng Liu | f6be8af | 2014-09-04 15:17:53 +0800 | [diff] [blame] | 1852 | void wake_up_if_idle(int cpu) |
| 1853 | { |
| 1854 | struct rq *rq = cpu_rq(cpu); |
| 1855 | unsigned long flags; |
| 1856 | |
Andy Lutomirski | fd7de1e | 2014-11-29 08:13:51 -0800 | [diff] [blame] | 1857 | rcu_read_lock(); |
| 1858 | |
| 1859 | if (!is_idle_task(rcu_dereference(rq->curr))) |
| 1860 | goto out; |
Chuansheng Liu | f6be8af | 2014-09-04 15:17:53 +0800 | [diff] [blame] | 1861 | |
| 1862 | if (set_nr_if_polling(rq->idle)) { |
| 1863 | trace_sched_wake_idle_without_ipi(cpu); |
| 1864 | } else { |
| 1865 | raw_spin_lock_irqsave(&rq->lock, flags); |
| 1866 | if (is_idle_task(rq->curr)) |
| 1867 | smp_send_reschedule(cpu); |
| 1868 | /* Else cpu is not in idle, do nothing here */ |
| 1869 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
| 1870 | } |
Andy Lutomirski | fd7de1e | 2014-11-29 08:13:51 -0800 | [diff] [blame] | 1871 | |
| 1872 | out: |
| 1873 | rcu_read_unlock(); |
Chuansheng Liu | f6be8af | 2014-09-04 15:17:53 +0800 | [diff] [blame] | 1874 | } |
| 1875 | |
Peter Zijlstra | 39be350 | 2012-01-26 12:44:34 +0100 | [diff] [blame] | 1876 | bool cpus_share_cache(int this_cpu, int that_cpu) |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 1877 | { |
| 1878 | return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); |
| 1879 | } |
Peter Zijlstra | d6aa8f8 | 2011-05-26 14:21:33 +0200 | [diff] [blame] | 1880 | #endif /* CONFIG_SMP */ |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 1881 | |
Peter Zijlstra | b5179ac | 2016-05-11 16:10:34 +0200 | [diff] [blame] | 1882 | static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags) |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1883 | { |
| 1884 | struct rq *rq = cpu_rq(cpu); |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 1885 | struct pin_cookie cookie; |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1886 | |
Daniel Hellstrom | 17d9f31 | 2011-05-20 04:01:10 +0000 | [diff] [blame] | 1887 | #if defined(CONFIG_SMP) |
Peter Zijlstra | 39be350 | 2012-01-26 12:44:34 +0100 | [diff] [blame] | 1888 | if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) { |
Peter Zijlstra | f01114c | 2011-05-31 12:26:55 +0200 | [diff] [blame] | 1889 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
Peter Zijlstra | b7e7ade | 2016-05-23 11:19:07 +0200 | [diff] [blame] | 1890 | ttwu_queue_remote(p, cpu, wake_flags); |
Peter Zijlstra | 317f394 | 2011-04-05 17:23:58 +0200 | [diff] [blame] | 1891 | return; |
| 1892 | } |
| 1893 | #endif |
| 1894 | |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1895 | raw_spin_lock(&rq->lock); |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 1896 | cookie = lockdep_pin_lock(&rq->lock); |
Peter Zijlstra | b5179ac | 2016-05-11 16:10:34 +0200 | [diff] [blame] | 1897 | ttwu_do_activate(rq, p, wake_flags, cookie); |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 1898 | lockdep_unpin_lock(&rq->lock, cookie); |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 1899 | raw_spin_unlock(&rq->lock); |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1900 | } |
| 1901 | |
Peter Zijlstra | 8643cda | 2015-11-17 19:01:11 +0100 | [diff] [blame] | 1902 | /* |
| 1903 | * Notes on Program-Order guarantees on SMP systems. |
| 1904 | * |
| 1905 | * MIGRATION |
| 1906 | * |
| 1907 | * The basic program-order guarantee on SMP systems is that when a task [t] |
| 1908 | * migrates, all its activity on its old cpu [c0] happens-before any subsequent |
| 1909 | * execution on its new cpu [c1]. |
| 1910 | * |
| 1911 | * For migration (of runnable tasks) this is provided by the following means: |
| 1912 | * |
| 1913 | * A) UNLOCK of the rq(c0)->lock scheduling out task t |
| 1914 | * B) migration for t is required to synchronize *both* rq(c0)->lock and |
| 1915 | * rq(c1)->lock (if not at the same time, then in that order). |
| 1916 | * C) LOCK of the rq(c1)->lock scheduling in task |
| 1917 | * |
| 1918 | * Transitivity guarantees that B happens after A and C after B. |
| 1919 | * Note: we only require RCpc transitivity. |
| 1920 | * Note: the cpu doing B need not be c0 or c1 |
| 1921 | * |
| 1922 | * Example: |
| 1923 | * |
| 1924 | * CPU0 CPU1 CPU2 |
| 1925 | * |
| 1926 | * LOCK rq(0)->lock |
| 1927 | * sched-out X |
| 1928 | * sched-in Y |
| 1929 | * UNLOCK rq(0)->lock |
| 1930 | * |
| 1931 | * LOCK rq(0)->lock // orders against CPU0 |
| 1932 | * dequeue X |
| 1933 | * UNLOCK rq(0)->lock |
| 1934 | * |
| 1935 | * LOCK rq(1)->lock |
| 1936 | * enqueue X |
| 1937 | * UNLOCK rq(1)->lock |
| 1938 | * |
| 1939 | * LOCK rq(1)->lock // orders against CPU2 |
| 1940 | * sched-out Z |
| 1941 | * sched-in X |
| 1942 | * UNLOCK rq(1)->lock |
| 1943 | * |
| 1944 | * |
| 1945 | * BLOCKING -- aka. SLEEP + WAKEUP |
| 1946 | * |
| 1947 | * For blocking we (obviously) need to provide the same guarantee as for |
| 1948 | * migration. However the means are completely different as there is no lock |
| 1949 | * chain to provide order. Instead we do: |
| 1950 | * |
| 1951 | * 1) smp_store_release(X->on_cpu, 0) |
Peter Zijlstra | 1f03e8d | 2016-04-04 10:57:12 +0200 | [diff] [blame] | 1952 | * 2) smp_cond_load_acquire(!X->on_cpu) |
Peter Zijlstra | 8643cda | 2015-11-17 19:01:11 +0100 | [diff] [blame] | 1953 | * |
| 1954 | * Example: |
| 1955 | * |
| 1956 | * CPU0 (schedule) CPU1 (try_to_wake_up) CPU2 (schedule) |
| 1957 | * |
| 1958 | * LOCK rq(0)->lock LOCK X->pi_lock |
| 1959 | * dequeue X |
| 1960 | * sched-out X |
| 1961 | * smp_store_release(X->on_cpu, 0); |
| 1962 | * |
Peter Zijlstra | 1f03e8d | 2016-04-04 10:57:12 +0200 | [diff] [blame] | 1963 | * smp_cond_load_acquire(&X->on_cpu, !VAL); |
Peter Zijlstra | 8643cda | 2015-11-17 19:01:11 +0100 | [diff] [blame] | 1964 | * X->state = WAKING |
| 1965 | * set_task_cpu(X,2) |
| 1966 | * |
| 1967 | * LOCK rq(2)->lock |
| 1968 | * enqueue X |
| 1969 | * X->state = RUNNING |
| 1970 | * UNLOCK rq(2)->lock |
| 1971 | * |
| 1972 | * LOCK rq(2)->lock // orders against CPU1 |
| 1973 | * sched-out Z |
| 1974 | * sched-in X |
| 1975 | * UNLOCK rq(2)->lock |
| 1976 | * |
| 1977 | * UNLOCK X->pi_lock |
| 1978 | * UNLOCK rq(0)->lock |
| 1979 | * |
| 1980 | * |
| 1981 | * However; for wakeups there is a second guarantee we must provide, namely we |
| 1982 | * must observe the state that lead to our wakeup. That is, not only must our |
| 1983 | * task observe its own prior state, it must also observe the stores prior to |
| 1984 | * its wakeup. |
| 1985 | * |
| 1986 | * This means that any means of doing remote wakeups must order the CPU doing |
| 1987 | * the wakeup against the CPU the task is going to end up running on. This, |
| 1988 | * however, is already required for the regular Program-Order guarantee above, |
Peter Zijlstra | 1f03e8d | 2016-04-04 10:57:12 +0200 | [diff] [blame] | 1989 | * since the waking CPU is the one issueing the ACQUIRE (smp_cond_load_acquire). |
Peter Zijlstra | 8643cda | 2015-11-17 19:01:11 +0100 | [diff] [blame] | 1990 | * |
| 1991 | */ |
| 1992 | |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1993 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1994 | * try_to_wake_up - wake up a thread |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1995 | * @p: the thread to be awakened |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1996 | * @state: the mask of task states that can be woken |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 1997 | * @wake_flags: wake modifier flags (WF_*) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1998 | * |
| 1999 | * Put it on the run-queue if it's not already there. The "current" |
| 2000 | * thread is always on the run-queue (except when the actual |
| 2001 | * re-schedule is in progress), and as such you're allowed to do |
| 2002 | * the simpler "current->state = TASK_RUNNING" to mark yourself |
| 2003 | * runnable without the overhead of this. |
| 2004 | * |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 2005 | * Return: %true if @p was woken up, %false if it was already running. |
Tejun Heo | 9ed3811 | 2009-12-03 15:08:03 +0900 | [diff] [blame] | 2006 | * or @state didn't match @p's state. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2007 | */ |
Peter Zijlstra | e4a52bc | 2011-04-05 17:23:54 +0200 | [diff] [blame] | 2008 | static int |
| 2009 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2010 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2011 | unsigned long flags; |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 2012 | int cpu, success = 0; |
Peter Zijlstra | 2398f2c | 2008-06-27 13:41:35 +0200 | [diff] [blame] | 2013 | |
Oleg Nesterov | e0acd0a | 2013-08-12 18:14:00 +0200 | [diff] [blame] | 2014 | /* |
| 2015 | * If we are going to wake up a thread waiting for CONDITION we |
| 2016 | * need to ensure that CONDITION=1 done by the caller can not be |
| 2017 | * reordered with p->state check below. This pairs with mb() in |
| 2018 | * set_current_state() the waiting thread does. |
| 2019 | */ |
| 2020 | smp_mb__before_spinlock(); |
Peter Zijlstra | 013fdb8 | 2011-04-05 17:23:45 +0200 | [diff] [blame] | 2021 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
Peter Zijlstra | e9c8431 | 2009-09-15 14:43:03 +0200 | [diff] [blame] | 2022 | if (!(p->state & state)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2023 | goto out; |
| 2024 | |
Peter Zijlstra | fbd705a | 2015-06-09 11:13:36 +0200 | [diff] [blame] | 2025 | trace_sched_waking(p); |
| 2026 | |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 2027 | success = 1; /* we're going to change ->state */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2028 | cpu = task_cpu(p); |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 2029 | |
Balbir Singh | 135e8c9 | 2016-09-05 13:16:40 +1000 | [diff] [blame] | 2030 | /* |
| 2031 | * Ensure we load p->on_rq _after_ p->state, otherwise it would |
| 2032 | * be possible to, falsely, observe p->on_rq == 0 and get stuck |
| 2033 | * in smp_cond_load_acquire() below. |
| 2034 | * |
| 2035 | * sched_ttwu_pending() try_to_wake_up() |
| 2036 | * [S] p->on_rq = 1; [L] P->state |
| 2037 | * UNLOCK rq->lock -----. |
| 2038 | * \ |
| 2039 | * +--- RMB |
| 2040 | * schedule() / |
| 2041 | * LOCK rq->lock -----' |
| 2042 | * UNLOCK rq->lock |
| 2043 | * |
| 2044 | * [task p] |
| 2045 | * [S] p->state = UNINTERRUPTIBLE [L] p->on_rq |
| 2046 | * |
| 2047 | * Pairs with the UNLOCK+LOCK on rq->lock from the |
| 2048 | * last wakeup of our task and the schedule that got our task |
| 2049 | * current. |
| 2050 | */ |
| 2051 | smp_rmb(); |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 2052 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
| 2053 | goto stat; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2054 | |
| 2055 | #ifdef CONFIG_SMP |
Peter Zijlstra | e9c8431 | 2009-09-15 14:43:03 +0200 | [diff] [blame] | 2056 | /* |
Peter Zijlstra | ecf7d01 | 2015-10-07 14:14:13 +0200 | [diff] [blame] | 2057 | * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be |
| 2058 | * possible to, falsely, observe p->on_cpu == 0. |
| 2059 | * |
| 2060 | * One must be running (->on_cpu == 1) in order to remove oneself |
| 2061 | * from the runqueue. |
| 2062 | * |
| 2063 | * [S] ->on_cpu = 1; [L] ->on_rq |
| 2064 | * UNLOCK rq->lock |
| 2065 | * RMB |
| 2066 | * LOCK rq->lock |
| 2067 | * [S] ->on_rq = 0; [L] ->on_cpu |
| 2068 | * |
| 2069 | * Pairs with the full barrier implied in the UNLOCK+LOCK on rq->lock |
| 2070 | * from the consecutive calls to schedule(); the first switching to our |
| 2071 | * task, the second putting it to sleep. |
| 2072 | */ |
| 2073 | smp_rmb(); |
| 2074 | |
| 2075 | /* |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 2076 | * If the owning (remote) cpu is still in the middle of schedule() with |
| 2077 | * this task as prev, wait until its done referencing the task. |
Peter Zijlstra | b75a225 | 2015-10-06 14:36:17 +0200 | [diff] [blame] | 2078 | * |
| 2079 | * Pairs with the smp_store_release() in finish_lock_switch(). |
| 2080 | * |
| 2081 | * This ensures that tasks getting woken will be fully ordered against |
| 2082 | * their previous state and preserve Program Order. |
Peter Zijlstra | e4a52bc | 2011-04-05 17:23:54 +0200 | [diff] [blame] | 2083 | */ |
Peter Zijlstra | 1f03e8d | 2016-04-04 10:57:12 +0200 | [diff] [blame] | 2084 | smp_cond_load_acquire(&p->on_cpu, !VAL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2085 | |
Peter Zijlstra | a8e4f2e | 2011-04-05 17:23:49 +0200 | [diff] [blame] | 2086 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
Peter Zijlstra | e9c8431 | 2009-09-15 14:43:03 +0200 | [diff] [blame] | 2087 | p->state = TASK_WAKING; |
Peter Zijlstra | efbbd05 | 2009-12-16 18:04:40 +0100 | [diff] [blame] | 2088 | |
Peter Zijlstra | ac66f54 | 2013-10-07 11:29:16 +0100 | [diff] [blame] | 2089 | cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags); |
Peter Zijlstra | f339b9d | 2011-05-31 10:49:20 +0200 | [diff] [blame] | 2090 | if (task_cpu(p) != cpu) { |
| 2091 | wake_flags |= WF_MIGRATED; |
Mike Galbraith | 055a008 | 2009-11-12 11:07:44 +0100 | [diff] [blame] | 2092 | set_task_cpu(p, cpu); |
Peter Zijlstra | f339b9d | 2011-05-31 10:49:20 +0200 | [diff] [blame] | 2093 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2094 | #endif /* CONFIG_SMP */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2095 | |
Peter Zijlstra | b5179ac | 2016-05-11 16:10:34 +0200 | [diff] [blame] | 2096 | ttwu_queue(p, cpu, wake_flags); |
Peter Zijlstra | c05fbaf | 2011-04-05 17:23:57 +0200 | [diff] [blame] | 2097 | stat: |
Josh Poimboeuf | 4fa8d299 | 2016-06-17 12:43:26 -0500 | [diff] [blame] | 2098 | ttwu_stat(p, cpu, wake_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2099 | out: |
Peter Zijlstra | 013fdb8 | 2011-04-05 17:23:45 +0200 | [diff] [blame] | 2100 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2101 | |
| 2102 | return success; |
| 2103 | } |
| 2104 | |
David Howells | 50fa610 | 2009-04-28 15:01:38 +0100 | [diff] [blame] | 2105 | /** |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 2106 | * try_to_wake_up_local - try to wake up a local task with rq lock held |
| 2107 | * @p: the thread to be awakened |
Luis de Bethencourt | 9279e0d | 2016-07-10 15:00:26 +0100 | [diff] [blame] | 2108 | * @cookie: context's cookie for pinning |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 2109 | * |
Peter Zijlstra | 2acca55 | 2011-04-05 17:23:50 +0200 | [diff] [blame] | 2110 | * Put @p on the run-queue if it's not already there. The caller must |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 2111 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
Peter Zijlstra | 2acca55 | 2011-04-05 17:23:50 +0200 | [diff] [blame] | 2112 | * the current task. |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 2113 | */ |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 2114 | static void try_to_wake_up_local(struct task_struct *p, struct pin_cookie cookie) |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 2115 | { |
| 2116 | struct rq *rq = task_rq(p); |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 2117 | |
Tejun Heo | 383efcd | 2013-03-18 12:22:34 -0700 | [diff] [blame] | 2118 | if (WARN_ON_ONCE(rq != this_rq()) || |
| 2119 | WARN_ON_ONCE(p == current)) |
| 2120 | return; |
| 2121 | |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 2122 | lockdep_assert_held(&rq->lock); |
| 2123 | |
Peter Zijlstra | 2acca55 | 2011-04-05 17:23:50 +0200 | [diff] [blame] | 2124 | if (!raw_spin_trylock(&p->pi_lock)) { |
Peter Zijlstra | cbce1a6 | 2015-06-11 14:46:54 +0200 | [diff] [blame] | 2125 | /* |
| 2126 | * This is OK, because current is on_cpu, which avoids it being |
| 2127 | * picked for load-balance and preemption/IRQs are still |
| 2128 | * disabled avoiding further scheduler activity on it and we've |
| 2129 | * not yet picked a replacement task. |
| 2130 | */ |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 2131 | lockdep_unpin_lock(&rq->lock, cookie); |
Peter Zijlstra | 2acca55 | 2011-04-05 17:23:50 +0200 | [diff] [blame] | 2132 | raw_spin_unlock(&rq->lock); |
| 2133 | raw_spin_lock(&p->pi_lock); |
| 2134 | raw_spin_lock(&rq->lock); |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 2135 | lockdep_repin_lock(&rq->lock, cookie); |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 2136 | } |
Peter Zijlstra | 2acca55 | 2011-04-05 17:23:50 +0200 | [diff] [blame] | 2137 | |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 2138 | if (!(p->state & TASK_NORMAL)) |
Peter Zijlstra | 2acca55 | 2011-04-05 17:23:50 +0200 | [diff] [blame] | 2139 | goto out; |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 2140 | |
Peter Zijlstra | fbd705a | 2015-06-09 11:13:36 +0200 | [diff] [blame] | 2141 | trace_sched_waking(p); |
| 2142 | |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 2143 | if (!task_on_rq_queued(p)) |
Peter Zijlstra | d7c01d2 | 2011-04-05 17:23:43 +0200 | [diff] [blame] | 2144 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
| 2145 | |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 2146 | ttwu_do_wakeup(rq, p, 0, cookie); |
Josh Poimboeuf | 4fa8d299 | 2016-06-17 12:43:26 -0500 | [diff] [blame] | 2147 | ttwu_stat(p, smp_processor_id(), 0); |
Peter Zijlstra | 2acca55 | 2011-04-05 17:23:50 +0200 | [diff] [blame] | 2148 | out: |
| 2149 | raw_spin_unlock(&p->pi_lock); |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 2150 | } |
| 2151 | |
| 2152 | /** |
David Howells | 50fa610 | 2009-04-28 15:01:38 +0100 | [diff] [blame] | 2153 | * wake_up_process - Wake up a specific process |
| 2154 | * @p: The process to be woken up. |
| 2155 | * |
| 2156 | * Attempt to wake up the nominated process and move it to the set of runnable |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 2157 | * processes. |
| 2158 | * |
| 2159 | * Return: 1 if the process was woken up, 0 if it was already running. |
David Howells | 50fa610 | 2009-04-28 15:01:38 +0100 | [diff] [blame] | 2160 | * |
| 2161 | * It may be assumed that this function implies a write memory barrier before |
| 2162 | * changing the task state if and only if any tasks are woken up. |
| 2163 | */ |
Harvey Harrison | 7ad5b3a | 2008-02-08 04:19:53 -0800 | [diff] [blame] | 2164 | int wake_up_process(struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2165 | { |
Oleg Nesterov | 9067ac8 | 2013-01-21 20:48:17 +0100 | [diff] [blame] | 2166 | return try_to_wake_up(p, TASK_NORMAL, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2167 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2168 | EXPORT_SYMBOL(wake_up_process); |
| 2169 | |
Harvey Harrison | 7ad5b3a | 2008-02-08 04:19:53 -0800 | [diff] [blame] | 2170 | int wake_up_state(struct task_struct *p, unsigned int state) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2171 | { |
| 2172 | return try_to_wake_up(p, state, 0); |
| 2173 | } |
| 2174 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2175 | /* |
Juri Lelli | a5e7be3 | 2014-09-19 10:22:39 +0100 | [diff] [blame] | 2176 | * This function clears the sched_dl_entity static params. |
| 2177 | */ |
| 2178 | void __dl_clear_params(struct task_struct *p) |
| 2179 | { |
| 2180 | struct sched_dl_entity *dl_se = &p->dl; |
| 2181 | |
| 2182 | dl_se->dl_runtime = 0; |
| 2183 | dl_se->dl_deadline = 0; |
| 2184 | dl_se->dl_period = 0; |
| 2185 | dl_se->flags = 0; |
| 2186 | dl_se->dl_bw = 0; |
Daniel Bristot de Oliveira | 0559ea3 | 2017-05-29 16:24:03 +0200 | [diff] [blame] | 2187 | dl_se->dl_density = 0; |
Peter Zijlstra | 40767b0 | 2015-01-28 15:08:03 +0100 | [diff] [blame] | 2188 | |
| 2189 | dl_se->dl_throttled = 0; |
Peter Zijlstra | 40767b0 | 2015-01-28 15:08:03 +0100 | [diff] [blame] | 2190 | dl_se->dl_yielded = 0; |
Juri Lelli | a5e7be3 | 2014-09-19 10:22:39 +0100 | [diff] [blame] | 2191 | } |
| 2192 | |
| 2193 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2194 | * Perform scheduler related setup for a newly forked process p. |
| 2195 | * p is forked by current. |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2196 | * |
| 2197 | * __sched_fork() is basic setup used by init_idle() too: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2198 | */ |
Rik van Riel | 5e1576e | 2013-10-07 11:29:26 +0100 | [diff] [blame] | 2199 | static void __sched_fork(unsigned long clone_flags, struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2200 | { |
Peter Zijlstra | fd2f441 | 2011-04-05 17:23:44 +0200 | [diff] [blame] | 2201 | p->on_rq = 0; |
| 2202 | |
| 2203 | p->se.on_rq = 0; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2204 | p->se.exec_start = 0; |
| 2205 | p->se.sum_exec_runtime = 0; |
Ingo Molnar | f6cf891 | 2007-08-28 12:53:24 +0200 | [diff] [blame] | 2206 | p->se.prev_sum_exec_runtime = 0; |
Ingo Molnar | 6c594c2 | 2008-12-14 12:34:15 +0100 | [diff] [blame] | 2207 | p->se.nr_migrations = 0; |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 2208 | p->se.vruntime = 0; |
Peter Zijlstra | fd2f441 | 2011-04-05 17:23:44 +0200 | [diff] [blame] | 2209 | INIT_LIST_HEAD(&p->se.group_node); |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 2210 | |
Byungchul Park | ad936d8 | 2015-10-24 01:16:19 +0900 | [diff] [blame] | 2211 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 2212 | p->se.cfs_rq = NULL; |
| 2213 | #endif |
| 2214 | |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 2215 | #ifdef CONFIG_SCHEDSTATS |
Mel Gorman | cb25176 | 2016-02-05 09:08:36 +0000 | [diff] [blame] | 2216 | /* Even if schedstat is disabled, there should not be garbage */ |
Lucas De Marchi | 41acab8 | 2010-03-10 23:37:45 -0300 | [diff] [blame] | 2217 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 2218 | #endif |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 2219 | |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 2220 | RB_CLEAR_NODE(&p->dl.rb_node); |
Peter Zijlstra | 40767b0 | 2015-01-28 15:08:03 +0100 | [diff] [blame] | 2221 | init_dl_task_timer(&p->dl); |
Juri Lelli | a5e7be3 | 2014-09-19 10:22:39 +0100 | [diff] [blame] | 2222 | __dl_clear_params(p); |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 2223 | |
Peter Zijlstra | fa71706 | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 2224 | INIT_LIST_HEAD(&p->rt.run_list); |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 2225 | p->rt.timeout = 0; |
| 2226 | p->rt.time_slice = sched_rr_timeslice; |
| 2227 | p->rt.on_rq = 0; |
| 2228 | p->rt.on_list = 0; |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 2229 | |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2230 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
| 2231 | INIT_HLIST_HEAD(&p->preempt_notifiers); |
| 2232 | #endif |
Peter Zijlstra | cbee9f8 | 2012-10-25 14:16:43 +0200 | [diff] [blame] | 2233 | |
| 2234 | #ifdef CONFIG_NUMA_BALANCING |
| 2235 | if (p->mm && atomic_read(&p->mm->mm_users) == 1) { |
Mel Gorman | 7e8d16b | 2013-10-07 11:28:54 +0100 | [diff] [blame] | 2236 | p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay); |
Peter Zijlstra | cbee9f8 | 2012-10-25 14:16:43 +0200 | [diff] [blame] | 2237 | p->mm->numa_scan_seq = 0; |
| 2238 | } |
| 2239 | |
Rik van Riel | 5e1576e | 2013-10-07 11:29:26 +0100 | [diff] [blame] | 2240 | if (clone_flags & CLONE_VM) |
| 2241 | p->numa_preferred_nid = current->numa_preferred_nid; |
| 2242 | else |
| 2243 | p->numa_preferred_nid = -1; |
| 2244 | |
Peter Zijlstra | cbee9f8 | 2012-10-25 14:16:43 +0200 | [diff] [blame] | 2245 | p->node_stamp = 0ULL; |
| 2246 | p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0; |
Peter Zijlstra | 4b96a29b | 2012-10-25 14:16:47 +0200 | [diff] [blame] | 2247 | p->numa_scan_period = sysctl_numa_balancing_scan_delay; |
Peter Zijlstra | cbee9f8 | 2012-10-25 14:16:43 +0200 | [diff] [blame] | 2248 | p->numa_work.next = &p->numa_work; |
Iulia Manda | 44dba3d | 2014-10-31 02:13:31 +0200 | [diff] [blame] | 2249 | p->numa_faults = NULL; |
Rik van Riel | 7e2703e | 2014-01-27 17:03:45 -0500 | [diff] [blame] | 2250 | p->last_task_numa_placement = 0; |
| 2251 | p->last_sum_exec_runtime = 0; |
Peter Zijlstra | 8c8a743 | 2013-10-07 11:29:21 +0100 | [diff] [blame] | 2252 | |
Peter Zijlstra | 8c8a743 | 2013-10-07 11:29:21 +0100 | [diff] [blame] | 2253 | p->numa_group = NULL; |
Peter Zijlstra | cbee9f8 | 2012-10-25 14:16:43 +0200 | [diff] [blame] | 2254 | #endif /* CONFIG_NUMA_BALANCING */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2255 | } |
| 2256 | |
Srikar Dronamraju | 2a59572 | 2015-08-11 21:54:21 +0530 | [diff] [blame] | 2257 | DEFINE_STATIC_KEY_FALSE(sched_numa_balancing); |
| 2258 | |
Mel Gorman | 1a687c2 | 2012-11-22 11:16:36 +0000 | [diff] [blame] | 2259 | #ifdef CONFIG_NUMA_BALANCING |
Mel Gorman | 3105b86 | 2012-11-23 11:23:49 +0000 | [diff] [blame] | 2260 | |
| 2261 | void set_numabalancing_state(bool enabled) |
| 2262 | { |
Srikar Dronamraju | 2a59572 | 2015-08-11 21:54:21 +0530 | [diff] [blame] | 2263 | if (enabled) |
| 2264 | static_branch_enable(&sched_numa_balancing); |
| 2265 | else |
| 2266 | static_branch_disable(&sched_numa_balancing); |
Mel Gorman | 3105b86 | 2012-11-23 11:23:49 +0000 | [diff] [blame] | 2267 | } |
Andi Kleen | 54a43d5 | 2014-01-23 15:53:13 -0800 | [diff] [blame] | 2268 | |
| 2269 | #ifdef CONFIG_PROC_SYSCTL |
| 2270 | int sysctl_numa_balancing(struct ctl_table *table, int write, |
| 2271 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 2272 | { |
| 2273 | struct ctl_table t; |
| 2274 | int err; |
Srikar Dronamraju | 2a59572 | 2015-08-11 21:54:21 +0530 | [diff] [blame] | 2275 | int state = static_branch_likely(&sched_numa_balancing); |
Andi Kleen | 54a43d5 | 2014-01-23 15:53:13 -0800 | [diff] [blame] | 2276 | |
| 2277 | if (write && !capable(CAP_SYS_ADMIN)) |
| 2278 | return -EPERM; |
| 2279 | |
| 2280 | t = *table; |
| 2281 | t.data = &state; |
| 2282 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); |
| 2283 | if (err < 0) |
| 2284 | return err; |
| 2285 | if (write) |
| 2286 | set_numabalancing_state(state); |
| 2287 | return err; |
| 2288 | } |
| 2289 | #endif |
| 2290 | #endif |
Mel Gorman | 1a687c2 | 2012-11-22 11:16:36 +0000 | [diff] [blame] | 2291 | |
Mel Gorman | cb25176 | 2016-02-05 09:08:36 +0000 | [diff] [blame] | 2292 | #ifdef CONFIG_SCHEDSTATS |
Josh Poimboeuf | 4698f88 | 2016-06-07 14:43:16 -0500 | [diff] [blame] | 2293 | |
| 2294 | DEFINE_STATIC_KEY_FALSE(sched_schedstats); |
| 2295 | static bool __initdata __sched_schedstats = false; |
| 2296 | |
Mel Gorman | cb25176 | 2016-02-05 09:08:36 +0000 | [diff] [blame] | 2297 | static void set_schedstats(bool enabled) |
| 2298 | { |
| 2299 | if (enabled) |
| 2300 | static_branch_enable(&sched_schedstats); |
| 2301 | else |
| 2302 | static_branch_disable(&sched_schedstats); |
| 2303 | } |
| 2304 | |
| 2305 | void force_schedstat_enabled(void) |
| 2306 | { |
| 2307 | if (!schedstat_enabled()) { |
| 2308 | pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n"); |
| 2309 | static_branch_enable(&sched_schedstats); |
| 2310 | } |
| 2311 | } |
| 2312 | |
| 2313 | static int __init setup_schedstats(char *str) |
| 2314 | { |
| 2315 | int ret = 0; |
| 2316 | if (!str) |
| 2317 | goto out; |
| 2318 | |
Josh Poimboeuf | 4698f88 | 2016-06-07 14:43:16 -0500 | [diff] [blame] | 2319 | /* |
| 2320 | * This code is called before jump labels have been set up, so we can't |
| 2321 | * change the static branch directly just yet. Instead set a temporary |
| 2322 | * variable so init_schedstats() can do it later. |
| 2323 | */ |
Mel Gorman | cb25176 | 2016-02-05 09:08:36 +0000 | [diff] [blame] | 2324 | if (!strcmp(str, "enable")) { |
Josh Poimboeuf | 4698f88 | 2016-06-07 14:43:16 -0500 | [diff] [blame] | 2325 | __sched_schedstats = true; |
Mel Gorman | cb25176 | 2016-02-05 09:08:36 +0000 | [diff] [blame] | 2326 | ret = 1; |
| 2327 | } else if (!strcmp(str, "disable")) { |
Josh Poimboeuf | 4698f88 | 2016-06-07 14:43:16 -0500 | [diff] [blame] | 2328 | __sched_schedstats = false; |
Mel Gorman | cb25176 | 2016-02-05 09:08:36 +0000 | [diff] [blame] | 2329 | ret = 1; |
| 2330 | } |
| 2331 | out: |
| 2332 | if (!ret) |
| 2333 | pr_warn("Unable to parse schedstats=\n"); |
| 2334 | |
| 2335 | return ret; |
| 2336 | } |
| 2337 | __setup("schedstats=", setup_schedstats); |
| 2338 | |
Josh Poimboeuf | 4698f88 | 2016-06-07 14:43:16 -0500 | [diff] [blame] | 2339 | static void __init init_schedstats(void) |
| 2340 | { |
| 2341 | set_schedstats(__sched_schedstats); |
| 2342 | } |
| 2343 | |
Mel Gorman | cb25176 | 2016-02-05 09:08:36 +0000 | [diff] [blame] | 2344 | #ifdef CONFIG_PROC_SYSCTL |
| 2345 | int sysctl_schedstats(struct ctl_table *table, int write, |
| 2346 | void __user *buffer, size_t *lenp, loff_t *ppos) |
| 2347 | { |
| 2348 | struct ctl_table t; |
| 2349 | int err; |
| 2350 | int state = static_branch_likely(&sched_schedstats); |
| 2351 | |
| 2352 | if (write && !capable(CAP_SYS_ADMIN)) |
| 2353 | return -EPERM; |
| 2354 | |
| 2355 | t = *table; |
| 2356 | t.data = &state; |
| 2357 | err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos); |
| 2358 | if (err < 0) |
| 2359 | return err; |
| 2360 | if (write) |
| 2361 | set_schedstats(state); |
| 2362 | return err; |
| 2363 | } |
Josh Poimboeuf | 4698f88 | 2016-06-07 14:43:16 -0500 | [diff] [blame] | 2364 | #endif /* CONFIG_PROC_SYSCTL */ |
| 2365 | #else /* !CONFIG_SCHEDSTATS */ |
| 2366 | static inline void init_schedstats(void) {} |
| 2367 | #endif /* CONFIG_SCHEDSTATS */ |
Mel Gorman | cb25176 | 2016-02-05 09:08:36 +0000 | [diff] [blame] | 2368 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2369 | /* |
| 2370 | * fork()/clone()-time setup: |
| 2371 | */ |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 2372 | int sched_fork(unsigned long clone_flags, struct task_struct *p) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2373 | { |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 2374 | unsigned long flags; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2375 | int cpu = get_cpu(); |
| 2376 | |
Rik van Riel | 5e1576e | 2013-10-07 11:29:26 +0100 | [diff] [blame] | 2377 | __sched_fork(clone_flags, p); |
Peter Zijlstra | 06b83b5 | 2009-12-16 18:04:35 +0100 | [diff] [blame] | 2378 | /* |
Peter Zijlstra | 7dc603c | 2016-06-16 13:29:28 +0200 | [diff] [blame] | 2379 | * We mark the process as NEW here. This guarantees that |
Peter Zijlstra | 06b83b5 | 2009-12-16 18:04:35 +0100 | [diff] [blame] | 2380 | * nobody will actually run it, and a signal or other external |
| 2381 | * event cannot wake it up and insert it on the runqueue either. |
| 2382 | */ |
Peter Zijlstra | 7dc603c | 2016-06-16 13:29:28 +0200 | [diff] [blame] | 2383 | p->state = TASK_NEW; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2384 | |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 2385 | /* |
Mike Galbraith | c350a04 | 2011-07-27 17:14:55 +0200 | [diff] [blame] | 2386 | * Make sure we do not leak PI boosting priority to the child. |
| 2387 | */ |
| 2388 | p->prio = current->normal_prio; |
| 2389 | |
| 2390 | /* |
Mike Galbraith | b9dc29e | 2009-06-17 10:46:01 +0200 | [diff] [blame] | 2391 | * Revert to default priority/policy on fork if requested. |
| 2392 | */ |
| 2393 | if (unlikely(p->sched_reset_on_fork)) { |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 2394 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
Mike Galbraith | b9dc29e | 2009-06-17 10:46:01 +0200 | [diff] [blame] | 2395 | p->policy = SCHED_NORMAL; |
Mike Galbraith | 6c697bd | 2009-06-17 10:48:02 +0200 | [diff] [blame] | 2396 | p->static_prio = NICE_TO_PRIO(0); |
Mike Galbraith | c350a04 | 2011-07-27 17:14:55 +0200 | [diff] [blame] | 2397 | p->rt_priority = 0; |
| 2398 | } else if (PRIO_TO_NICE(p->static_prio) < 0) |
| 2399 | p->static_prio = NICE_TO_PRIO(0); |
| 2400 | |
| 2401 | p->prio = p->normal_prio = __normal_prio(p); |
| 2402 | set_load_weight(p); |
Mike Galbraith | 6c697bd | 2009-06-17 10:48:02 +0200 | [diff] [blame] | 2403 | |
Mike Galbraith | b9dc29e | 2009-06-17 10:46:01 +0200 | [diff] [blame] | 2404 | /* |
| 2405 | * We don't need the reset flag anymore after the fork. It has |
| 2406 | * fulfilled its duty: |
| 2407 | */ |
| 2408 | p->sched_reset_on_fork = 0; |
| 2409 | } |
Lennart Poettering | ca94c44 | 2009-06-15 17:17:47 +0200 | [diff] [blame] | 2410 | |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 2411 | if (dl_prio(p->prio)) { |
| 2412 | put_cpu(); |
| 2413 | return -EAGAIN; |
| 2414 | } else if (rt_prio(p->prio)) { |
| 2415 | p->sched_class = &rt_sched_class; |
| 2416 | } else { |
Hiroshi Shimamoto | 2ddbf95 | 2007-10-15 17:00:11 +0200 | [diff] [blame] | 2417 | p->sched_class = &fair_sched_class; |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 2418 | } |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 2419 | |
Peter Zijlstra | 7dc603c | 2016-06-16 13:29:28 +0200 | [diff] [blame] | 2420 | init_entity_runnable_average(&p->se); |
Peter Zijlstra | cd29fe6 | 2009-11-27 17:32:46 +0100 | [diff] [blame] | 2421 | |
Peter Zijlstra | 8695159 | 2010-06-22 11:44:53 +0200 | [diff] [blame] | 2422 | /* |
| 2423 | * The child is not yet in the pid-hash so no cgroup attach races, |
| 2424 | * and the cgroup is pinned to this child due to cgroup_fork() |
| 2425 | * is ran before sched_fork(). |
| 2426 | * |
| 2427 | * Silence PROVE_RCU. |
| 2428 | */ |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 2429 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
Peter Zijlstra | e210bff | 2016-06-16 18:51:48 +0200 | [diff] [blame] | 2430 | /* |
| 2431 | * We're setting the cpu for the first time, we don't migrate, |
| 2432 | * so use __set_task_cpu(). |
| 2433 | */ |
| 2434 | __set_task_cpu(p, cpu); |
| 2435 | if (p->sched_class->task_fork) |
| 2436 | p->sched_class->task_fork(p); |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 2437 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
Peter Zijlstra | 5f3edc1 | 2009-09-10 13:42:00 +0200 | [diff] [blame] | 2438 | |
Naveen N. Rao | f6db834 | 2015-06-25 23:53:37 +0530 | [diff] [blame] | 2439 | #ifdef CONFIG_SCHED_INFO |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2440 | if (likely(sched_info_on())) |
Chandra Seetharaman | 52f17b6 | 2006-07-14 00:24:38 -0700 | [diff] [blame] | 2441 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2442 | #endif |
Peter Zijlstra | 3ca7a44 | 2011-04-05 17:23:40 +0200 | [diff] [blame] | 2443 | #if defined(CONFIG_SMP) |
| 2444 | p->on_cpu = 0; |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 2445 | #endif |
Peter Zijlstra | 0102874 | 2013-08-14 14:55:46 +0200 | [diff] [blame] | 2446 | init_task_preempt_count(p); |
Dario Faggioli | 806c09a | 2010-11-30 19:51:33 +0100 | [diff] [blame] | 2447 | #ifdef CONFIG_SMP |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 2448 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
Juri Lelli | 1baca4c | 2013-11-07 14:43:38 +0100 | [diff] [blame] | 2449 | RB_CLEAR_NODE(&p->pushable_dl_tasks); |
Dario Faggioli | 806c09a | 2010-11-30 19:51:33 +0100 | [diff] [blame] | 2450 | #endif |
Gregory Haskins | 917b627 | 2008-12-29 09:39:53 -0500 | [diff] [blame] | 2451 | |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 2452 | put_cpu(); |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 2453 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2454 | } |
| 2455 | |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 2456 | unsigned long to_ratio(u64 period, u64 runtime) |
| 2457 | { |
| 2458 | if (runtime == RUNTIME_INF) |
| 2459 | return 1ULL << 20; |
| 2460 | |
| 2461 | /* |
| 2462 | * Doing this here saves a lot of checks in all |
| 2463 | * the calling paths, and returning zero seems |
| 2464 | * safe for them anyway. |
| 2465 | */ |
| 2466 | if (period == 0) |
| 2467 | return 0; |
| 2468 | |
| 2469 | return div64_u64(runtime << 20, period); |
| 2470 | } |
| 2471 | |
| 2472 | #ifdef CONFIG_SMP |
| 2473 | inline struct dl_bw *dl_bw_of(int i) |
| 2474 | { |
Paul E. McKenney | f78f5b9 | 2015-06-18 15:50:02 -0700 | [diff] [blame] | 2475 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
| 2476 | "sched RCU must be held"); |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 2477 | return &cpu_rq(i)->rd->dl_bw; |
| 2478 | } |
| 2479 | |
Peter Zijlstra | de212f1 | 2013-12-19 11:54:45 +0100 | [diff] [blame] | 2480 | static inline int dl_bw_cpus(int i) |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 2481 | { |
Peter Zijlstra | de212f1 | 2013-12-19 11:54:45 +0100 | [diff] [blame] | 2482 | struct root_domain *rd = cpu_rq(i)->rd; |
| 2483 | int cpus = 0; |
| 2484 | |
Paul E. McKenney | f78f5b9 | 2015-06-18 15:50:02 -0700 | [diff] [blame] | 2485 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), |
| 2486 | "sched RCU must be held"); |
Peter Zijlstra | de212f1 | 2013-12-19 11:54:45 +0100 | [diff] [blame] | 2487 | for_each_cpu_and(i, rd->span, cpu_active_mask) |
| 2488 | cpus++; |
| 2489 | |
| 2490 | return cpus; |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 2491 | } |
| 2492 | #else |
| 2493 | inline struct dl_bw *dl_bw_of(int i) |
| 2494 | { |
| 2495 | return &cpu_rq(i)->dl.dl_bw; |
| 2496 | } |
| 2497 | |
Peter Zijlstra | de212f1 | 2013-12-19 11:54:45 +0100 | [diff] [blame] | 2498 | static inline int dl_bw_cpus(int i) |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 2499 | { |
| 2500 | return 1; |
| 2501 | } |
| 2502 | #endif |
| 2503 | |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 2504 | /* |
| 2505 | * We must be sure that accepting a new task (or allowing changing the |
| 2506 | * parameters of an existing one) is consistent with the bandwidth |
| 2507 | * constraints. If yes, this function also accordingly updates the currently |
| 2508 | * allocated bandwidth to reflect the new situation. |
| 2509 | * |
| 2510 | * This function is called while holding p's rq->lock. |
Peter Zijlstra | 40767b0 | 2015-01-28 15:08:03 +0100 | [diff] [blame] | 2511 | * |
| 2512 | * XXX we should delay bw change until the task's 0-lag point, see |
| 2513 | * __setparam_dl(). |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 2514 | */ |
| 2515 | static int dl_overflow(struct task_struct *p, int policy, |
| 2516 | const struct sched_attr *attr) |
| 2517 | { |
| 2518 | |
| 2519 | struct dl_bw *dl_b = dl_bw_of(task_cpu(p)); |
Steven Rostedt | 4df1638 | 2014-02-19 13:53:35 -0500 | [diff] [blame] | 2520 | u64 period = attr->sched_period ?: attr->sched_deadline; |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 2521 | u64 runtime = attr->sched_runtime; |
| 2522 | u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0; |
Peter Zijlstra | de212f1 | 2013-12-19 11:54:45 +0100 | [diff] [blame] | 2523 | int cpus, err = -1; |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 2524 | |
Xunlei Pang | fec148c | 2016-04-14 20:19:28 +0800 | [diff] [blame] | 2525 | /* !deadline task may carry old deadline bandwidth */ |
| 2526 | if (new_bw == p->dl.dl_bw && task_has_dl_policy(p)) |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 2527 | return 0; |
| 2528 | |
| 2529 | /* |
| 2530 | * Either if a task, enters, leave, or stays -deadline but changes |
| 2531 | * its parameters, we may need to update accordingly the total |
| 2532 | * allocated bandwidth of the container. |
| 2533 | */ |
| 2534 | raw_spin_lock(&dl_b->lock); |
Peter Zijlstra | de212f1 | 2013-12-19 11:54:45 +0100 | [diff] [blame] | 2535 | cpus = dl_bw_cpus(task_cpu(p)); |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 2536 | if (dl_policy(policy) && !task_has_dl_policy(p) && |
| 2537 | !__dl_overflow(dl_b, cpus, 0, new_bw)) { |
| 2538 | __dl_add(dl_b, new_bw); |
| 2539 | err = 0; |
| 2540 | } else if (dl_policy(policy) && task_has_dl_policy(p) && |
| 2541 | !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) { |
| 2542 | __dl_clear(dl_b, p->dl.dl_bw); |
| 2543 | __dl_add(dl_b, new_bw); |
| 2544 | err = 0; |
| 2545 | } else if (!dl_policy(policy) && task_has_dl_policy(p)) { |
| 2546 | __dl_clear(dl_b, p->dl.dl_bw); |
| 2547 | err = 0; |
| 2548 | } |
| 2549 | raw_spin_unlock(&dl_b->lock); |
| 2550 | |
| 2551 | return err; |
| 2552 | } |
| 2553 | |
| 2554 | extern void init_dl_bw(struct dl_bw *dl_b); |
| 2555 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2556 | /* |
| 2557 | * wake_up_new_task - wake up a newly created task for the first time. |
| 2558 | * |
| 2559 | * This function will do some initial scheduler statistics housekeeping |
| 2560 | * that must be done for every newly created context, then puts the task |
| 2561 | * on the runqueue and wakes it. |
| 2562 | */ |
Samir Bellabes | 3e51e3e | 2011-05-11 18:18:05 +0200 | [diff] [blame] | 2563 | void wake_up_new_task(struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2564 | { |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 2565 | struct rq_flags rf; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2566 | struct rq *rq; |
Peter Zijlstra | fabf318 | 2010-01-21 21:04:57 +0100 | [diff] [blame] | 2567 | |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 2568 | raw_spin_lock_irqsave(&p->pi_lock, rf.flags); |
Peter Zijlstra | 7dc603c | 2016-06-16 13:29:28 +0200 | [diff] [blame] | 2569 | p->state = TASK_RUNNING; |
Peter Zijlstra | fabf318 | 2010-01-21 21:04:57 +0100 | [diff] [blame] | 2570 | #ifdef CONFIG_SMP |
| 2571 | /* |
| 2572 | * Fork balancing, do it here and not earlier because: |
| 2573 | * - cpus_allowed can change in the fork path |
| 2574 | * - any previously selected cpu might disappear through hotplug |
Peter Zijlstra | e210bff | 2016-06-16 18:51:48 +0200 | [diff] [blame] | 2575 | * |
| 2576 | * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq, |
| 2577 | * as we're not fully set-up yet. |
Peter Zijlstra | fabf318 | 2010-01-21 21:04:57 +0100 | [diff] [blame] | 2578 | */ |
Peter Zijlstra | e210bff | 2016-06-16 18:51:48 +0200 | [diff] [blame] | 2579 | __set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0)); |
Peter Zijlstra | fabf318 | 2010-01-21 21:04:57 +0100 | [diff] [blame] | 2580 | #endif |
Peter Zijlstra | b7fa30c | 2016-06-09 15:07:50 +0200 | [diff] [blame] | 2581 | rq = __task_rq_lock(p, &rf); |
Yuyang Du | 2b8c41d | 2016-03-30 04:30:56 +0800 | [diff] [blame] | 2582 | post_init_entity_util_avg(&p->se); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2583 | |
Peter Zijlstra | cd29fe6 | 2009-11-27 17:32:46 +0100 | [diff] [blame] | 2584 | activate_task(rq, p, 0); |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 2585 | p->on_rq = TASK_ON_RQ_QUEUED; |
Peter Zijlstra | fbd705a | 2015-06-09 11:13:36 +0200 | [diff] [blame] | 2586 | trace_sched_wakeup_new(p); |
Peter Zijlstra | a7558e0 | 2009-09-14 20:02:34 +0200 | [diff] [blame] | 2587 | check_preempt_curr(rq, p, WF_FORK); |
Steven Rostedt | 9a897c5 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2588 | #ifdef CONFIG_SMP |
Peter Zijlstra | 0aaafaa | 2015-10-23 11:50:08 +0200 | [diff] [blame] | 2589 | if (p->sched_class->task_woken) { |
| 2590 | /* |
| 2591 | * Nothing relies on rq->lock after this, so its fine to |
| 2592 | * drop it. |
| 2593 | */ |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 2594 | lockdep_unpin_lock(&rq->lock, rf.cookie); |
Peter Zijlstra | efbbd05 | 2009-12-16 18:04:40 +0100 | [diff] [blame] | 2595 | p->sched_class->task_woken(rq, p); |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 2596 | lockdep_repin_lock(&rq->lock, rf.cookie); |
Peter Zijlstra | 0aaafaa | 2015-10-23 11:50:08 +0200 | [diff] [blame] | 2597 | } |
Steven Rostedt | 9a897c5 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 2598 | #endif |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 2599 | task_rq_unlock(rq, p, &rf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2600 | } |
| 2601 | |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2602 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
| 2603 | |
Peter Zijlstra | 1cde293 | 2015-06-08 16:00:30 +0200 | [diff] [blame] | 2604 | static struct static_key preempt_notifier_key = STATIC_KEY_INIT_FALSE; |
| 2605 | |
Peter Zijlstra | 2ecd9d2 | 2015-07-03 18:53:58 +0200 | [diff] [blame] | 2606 | void preempt_notifier_inc(void) |
| 2607 | { |
| 2608 | static_key_slow_inc(&preempt_notifier_key); |
| 2609 | } |
| 2610 | EXPORT_SYMBOL_GPL(preempt_notifier_inc); |
| 2611 | |
| 2612 | void preempt_notifier_dec(void) |
| 2613 | { |
| 2614 | static_key_slow_dec(&preempt_notifier_key); |
| 2615 | } |
| 2616 | EXPORT_SYMBOL_GPL(preempt_notifier_dec); |
| 2617 | |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2618 | /** |
Luis Henriques | 80dd99b | 2009-03-16 19:58:09 +0000 | [diff] [blame] | 2619 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
Randy Dunlap | 421cee2 | 2007-07-31 00:37:50 -0700 | [diff] [blame] | 2620 | * @notifier: notifier struct to register |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2621 | */ |
| 2622 | void preempt_notifier_register(struct preempt_notifier *notifier) |
| 2623 | { |
Peter Zijlstra | 2ecd9d2 | 2015-07-03 18:53:58 +0200 | [diff] [blame] | 2624 | if (!static_key_false(&preempt_notifier_key)) |
| 2625 | WARN(1, "registering preempt_notifier while notifiers disabled\n"); |
| 2626 | |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2627 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
| 2628 | } |
| 2629 | EXPORT_SYMBOL_GPL(preempt_notifier_register); |
| 2630 | |
| 2631 | /** |
| 2632 | * preempt_notifier_unregister - no longer interested in preemption notifications |
Randy Dunlap | 421cee2 | 2007-07-31 00:37:50 -0700 | [diff] [blame] | 2633 | * @notifier: notifier struct to unregister |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2634 | * |
Mathieu Desnoyers | d84525a | 2015-05-17 12:53:10 -0400 | [diff] [blame] | 2635 | * This is *not* safe to call from within a preemption notifier. |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2636 | */ |
| 2637 | void preempt_notifier_unregister(struct preempt_notifier *notifier) |
| 2638 | { |
| 2639 | hlist_del(¬ifier->link); |
| 2640 | } |
| 2641 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); |
| 2642 | |
Peter Zijlstra | 1cde293 | 2015-06-08 16:00:30 +0200 | [diff] [blame] | 2643 | static void __fire_sched_in_preempt_notifiers(struct task_struct *curr) |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2644 | { |
| 2645 | struct preempt_notifier *notifier; |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2646 | |
Sasha Levin | b67bfe0 | 2013-02-27 17:06:00 -0800 | [diff] [blame] | 2647 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2648 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
| 2649 | } |
| 2650 | |
Peter Zijlstra | 1cde293 | 2015-06-08 16:00:30 +0200 | [diff] [blame] | 2651 | static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
| 2652 | { |
| 2653 | if (static_key_false(&preempt_notifier_key)) |
| 2654 | __fire_sched_in_preempt_notifiers(curr); |
| 2655 | } |
| 2656 | |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2657 | static void |
Peter Zijlstra | 1cde293 | 2015-06-08 16:00:30 +0200 | [diff] [blame] | 2658 | __fire_sched_out_preempt_notifiers(struct task_struct *curr, |
| 2659 | struct task_struct *next) |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2660 | { |
| 2661 | struct preempt_notifier *notifier; |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2662 | |
Sasha Levin | b67bfe0 | 2013-02-27 17:06:00 -0800 | [diff] [blame] | 2663 | hlist_for_each_entry(notifier, &curr->preempt_notifiers, link) |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2664 | notifier->ops->sched_out(notifier, next); |
| 2665 | } |
| 2666 | |
Peter Zijlstra | 1cde293 | 2015-06-08 16:00:30 +0200 | [diff] [blame] | 2667 | static __always_inline void |
| 2668 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
| 2669 | struct task_struct *next) |
| 2670 | { |
| 2671 | if (static_key_false(&preempt_notifier_key)) |
| 2672 | __fire_sched_out_preempt_notifiers(curr, next); |
| 2673 | } |
| 2674 | |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 2675 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2676 | |
Peter Zijlstra | 1cde293 | 2015-06-08 16:00:30 +0200 | [diff] [blame] | 2677 | static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2678 | { |
| 2679 | } |
| 2680 | |
Peter Zijlstra | 1cde293 | 2015-06-08 16:00:30 +0200 | [diff] [blame] | 2681 | static inline void |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2682 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
| 2683 | struct task_struct *next) |
| 2684 | { |
| 2685 | } |
| 2686 | |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 2687 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2688 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2689 | /** |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 2690 | * prepare_task_switch - prepare to switch tasks |
| 2691 | * @rq: the runqueue preparing to switch |
Randy Dunlap | 421cee2 | 2007-07-31 00:37:50 -0700 | [diff] [blame] | 2692 | * @prev: the current task that is being switched out |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 2693 | * @next: the task we are going to switch to. |
| 2694 | * |
| 2695 | * This is called with the rq lock held and interrupts off. It must |
| 2696 | * be paired with a subsequent finish_task_switch after the context |
| 2697 | * switch. |
| 2698 | * |
| 2699 | * prepare_task_switch sets up locking and calls architecture specific |
| 2700 | * hooks. |
| 2701 | */ |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2702 | static inline void |
| 2703 | prepare_task_switch(struct rq *rq, struct task_struct *prev, |
| 2704 | struct task_struct *next) |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 2705 | { |
Michael S. Tsirkin | 4314895 | 2013-09-22 17:20:54 +0300 | [diff] [blame] | 2706 | sched_info_switch(rq, prev, next); |
Peter Zijlstra | fe4b04f | 2011-02-02 13:19:09 +0100 | [diff] [blame] | 2707 | perf_event_task_sched_out(prev, next); |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2708 | fire_sched_out_preempt_notifiers(prev, next); |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 2709 | prepare_lock_switch(rq, next); |
| 2710 | prepare_arch_switch(next); |
| 2711 | } |
| 2712 | |
| 2713 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2714 | * finish_task_switch - clean up after a task-switch |
| 2715 | * @prev: the thread we just switched away from. |
| 2716 | * |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 2717 | * finish_task_switch must be called after the context switch, paired |
| 2718 | * with a prepare_task_switch call before the context switch. |
| 2719 | * finish_task_switch will reconcile locking set up by prepare_task_switch, |
| 2720 | * and do any other architecture-specific cleanup actions. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2721 | * |
| 2722 | * Note that we may have delayed dropping an mm in context_switch(). If |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 2723 | * so, we finish that here outside of the runqueue lock. (Doing it |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2724 | * with the lock held can cause deadlocks; see schedule() for |
| 2725 | * details.) |
Oleg Nesterov | dfa50b6 | 2014-10-09 21:32:32 +0200 | [diff] [blame] | 2726 | * |
| 2727 | * The context switch have flipped the stack from under us and restored the |
| 2728 | * local variables which were saved when this task called schedule() in the |
| 2729 | * past. prev == current is still correct but we need to recalculate this_rq |
| 2730 | * because prev may have moved to another CPU. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2731 | */ |
Oleg Nesterov | dfa50b6 | 2014-10-09 21:32:32 +0200 | [diff] [blame] | 2732 | static struct rq *finish_task_switch(struct task_struct *prev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2733 | __releases(rq->lock) |
| 2734 | { |
Oleg Nesterov | dfa50b6 | 2014-10-09 21:32:32 +0200 | [diff] [blame] | 2735 | struct rq *rq = this_rq(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2736 | struct mm_struct *mm = rq->prev_mm; |
Oleg Nesterov | 55a101f | 2006-09-29 02:01:10 -0700 | [diff] [blame] | 2737 | long prev_state; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2738 | |
Peter Zijlstra | 609ca06 | 2015-09-28 17:52:18 +0200 | [diff] [blame] | 2739 | /* |
| 2740 | * The previous task will have left us with a preempt_count of 2 |
| 2741 | * because it left us after: |
| 2742 | * |
| 2743 | * schedule() |
| 2744 | * preempt_disable(); // 1 |
| 2745 | * __schedule() |
| 2746 | * raw_spin_lock_irq(&rq->lock) // 2 |
| 2747 | * |
| 2748 | * Also, see FORK_PREEMPT_COUNT. |
| 2749 | */ |
Peter Zijlstra | e2bf1c4 | 2015-09-29 12:18:46 +0200 | [diff] [blame] | 2750 | if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET, |
| 2751 | "corrupted preempt_count: %s/%d/0x%x\n", |
| 2752 | current->comm, current->pid, preempt_count())) |
| 2753 | preempt_count_set(FORK_PREEMPT_COUNT); |
Peter Zijlstra | 609ca06 | 2015-09-28 17:52:18 +0200 | [diff] [blame] | 2754 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2755 | rq->prev_mm = NULL; |
| 2756 | |
| 2757 | /* |
| 2758 | * A task struct has one reference for the use as "current". |
Oleg Nesterov | c394cc9 | 2006-09-29 02:01:11 -0700 | [diff] [blame] | 2759 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
Oleg Nesterov | 55a101f | 2006-09-29 02:01:10 -0700 | [diff] [blame] | 2760 | * schedule one last time. The schedule call will never return, and |
| 2761 | * the scheduled task must drop that reference. |
Peter Zijlstra | 95913d9 | 2015-09-29 14:45:09 +0200 | [diff] [blame] | 2762 | * |
| 2763 | * We must observe prev->state before clearing prev->on_cpu (in |
| 2764 | * finish_lock_switch), otherwise a concurrent wakeup can get prev |
| 2765 | * running on another CPU and we could rave with its RUNNING -> DEAD |
| 2766 | * transition, resulting in a double drop. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2767 | */ |
Oleg Nesterov | 55a101f | 2006-09-29 02:01:10 -0700 | [diff] [blame] | 2768 | prev_state = prev->state; |
Frederic Weisbecker | bf9fae9 | 2012-09-08 15:23:11 +0200 | [diff] [blame] | 2769 | vtime_task_switch(prev); |
Stephane Eranian | a8d757e | 2011-08-25 15:58:03 +0200 | [diff] [blame] | 2770 | perf_event_task_sched_in(prev, current); |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 2771 | finish_lock_switch(rq, prev); |
Catalin Marinas | 01f23e1 | 2011-11-27 21:43:10 +0000 | [diff] [blame] | 2772 | finish_arch_post_lock_switch(); |
Steven Rostedt | e8fa136 | 2008-01-25 21:08:05 +0100 | [diff] [blame] | 2773 | |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2774 | fire_sched_in_preempt_notifiers(current); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2775 | if (mm) |
| 2776 | mmdrop(mm); |
Oleg Nesterov | c394cc9 | 2006-09-29 02:01:11 -0700 | [diff] [blame] | 2777 | if (unlikely(prev_state == TASK_DEAD)) { |
Dario Faggioli | e6c390f | 2013-11-07 14:43:35 +0100 | [diff] [blame] | 2778 | if (prev->sched_class->task_dead) |
| 2779 | prev->sched_class->task_dead(prev); |
| 2780 | |
bibo mao | c6fd91f | 2006-03-26 01:38:20 -0800 | [diff] [blame] | 2781 | /* |
| 2782 | * Remove function-return probe instances associated with this |
| 2783 | * task and put them back on the free list. |
Ingo Molnar | 9761eea | 2007-07-09 18:52:00 +0200 | [diff] [blame] | 2784 | */ |
bibo mao | c6fd91f | 2006-03-26 01:38:20 -0800 | [diff] [blame] | 2785 | kprobe_flush_task(prev); |
Andy Lutomirski | 68f24b08 | 2016-09-15 22:45:48 -0700 | [diff] [blame] | 2786 | |
| 2787 | /* Task is done with its stack. */ |
| 2788 | put_task_stack(prev); |
| 2789 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2790 | put_task_struct(prev); |
bibo mao | c6fd91f | 2006-03-26 01:38:20 -0800 | [diff] [blame] | 2791 | } |
Frederic Weisbecker | 99e5ada | 2013-04-20 17:11:50 +0200 | [diff] [blame] | 2792 | |
Frederic Weisbecker | de734f8 | 2015-06-11 18:07:12 +0200 | [diff] [blame] | 2793 | tick_nohz_task_switch(); |
Oleg Nesterov | dfa50b6 | 2014-10-09 21:32:32 +0200 | [diff] [blame] | 2794 | return rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2795 | } |
| 2796 | |
Gregory Haskins | 3f029d3 | 2009-07-29 11:08:47 -0400 | [diff] [blame] | 2797 | #ifdef CONFIG_SMP |
| 2798 | |
Gregory Haskins | 3f029d3 | 2009-07-29 11:08:47 -0400 | [diff] [blame] | 2799 | /* rq->lock is NOT held, but preemption is disabled */ |
Peter Zijlstra | e3fca9e | 2015-06-11 14:46:37 +0200 | [diff] [blame] | 2800 | static void __balance_callback(struct rq *rq) |
Gregory Haskins | 3f029d3 | 2009-07-29 11:08:47 -0400 | [diff] [blame] | 2801 | { |
Peter Zijlstra | e3fca9e | 2015-06-11 14:46:37 +0200 | [diff] [blame] | 2802 | struct callback_head *head, *next; |
| 2803 | void (*func)(struct rq *rq); |
| 2804 | unsigned long flags; |
Gregory Haskins | 3f029d3 | 2009-07-29 11:08:47 -0400 | [diff] [blame] | 2805 | |
Peter Zijlstra | e3fca9e | 2015-06-11 14:46:37 +0200 | [diff] [blame] | 2806 | raw_spin_lock_irqsave(&rq->lock, flags); |
| 2807 | head = rq->balance_callback; |
| 2808 | rq->balance_callback = NULL; |
| 2809 | while (head) { |
| 2810 | func = (void (*)(struct rq *))head->func; |
| 2811 | next = head->next; |
| 2812 | head->next = NULL; |
| 2813 | head = next; |
Gregory Haskins | 3f029d3 | 2009-07-29 11:08:47 -0400 | [diff] [blame] | 2814 | |
Peter Zijlstra | e3fca9e | 2015-06-11 14:46:37 +0200 | [diff] [blame] | 2815 | func(rq); |
Gregory Haskins | 3f029d3 | 2009-07-29 11:08:47 -0400 | [diff] [blame] | 2816 | } |
Peter Zijlstra | e3fca9e | 2015-06-11 14:46:37 +0200 | [diff] [blame] | 2817 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
| 2818 | } |
| 2819 | |
| 2820 | static inline void balance_callback(struct rq *rq) |
| 2821 | { |
| 2822 | if (unlikely(rq->balance_callback)) |
| 2823 | __balance_callback(rq); |
Gregory Haskins | 3f029d3 | 2009-07-29 11:08:47 -0400 | [diff] [blame] | 2824 | } |
| 2825 | |
| 2826 | #else |
| 2827 | |
Peter Zijlstra | e3fca9e | 2015-06-11 14:46:37 +0200 | [diff] [blame] | 2828 | static inline void balance_callback(struct rq *rq) |
Gregory Haskins | 3f029d3 | 2009-07-29 11:08:47 -0400 | [diff] [blame] | 2829 | { |
| 2830 | } |
| 2831 | |
| 2832 | #endif |
| 2833 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2834 | /** |
| 2835 | * schedule_tail - first thing a freshly forked thread must call. |
| 2836 | * @prev: the thread we just switched away from. |
| 2837 | */ |
Andi Kleen | 722a9f9 | 2014-05-02 00:44:38 +0200 | [diff] [blame] | 2838 | asmlinkage __visible void schedule_tail(struct task_struct *prev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2839 | __releases(rq->lock) |
| 2840 | { |
Oleg Nesterov | 1a43a14 | 2014-10-08 21:36:44 +0200 | [diff] [blame] | 2841 | struct rq *rq; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 2842 | |
Peter Zijlstra | 609ca06 | 2015-09-28 17:52:18 +0200 | [diff] [blame] | 2843 | /* |
| 2844 | * New tasks start with FORK_PREEMPT_COUNT, see there and |
| 2845 | * finish_task_switch() for details. |
| 2846 | * |
| 2847 | * finish_task_switch() will drop rq->lock() and lower preempt_count |
| 2848 | * and the preempt_enable() will end up enabling preemption (on |
| 2849 | * PREEMPT_COUNT kernels). |
| 2850 | */ |
| 2851 | |
Oleg Nesterov | dfa50b6 | 2014-10-09 21:32:32 +0200 | [diff] [blame] | 2852 | rq = finish_task_switch(prev); |
Peter Zijlstra | e3fca9e | 2015-06-11 14:46:37 +0200 | [diff] [blame] | 2853 | balance_callback(rq); |
Oleg Nesterov | 1a43a14 | 2014-10-08 21:36:44 +0200 | [diff] [blame] | 2854 | preempt_enable(); |
Steven Rostedt | da19ab5 | 2009-07-29 00:21:22 -0400 | [diff] [blame] | 2855 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2856 | if (current->set_child_tid) |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 2857 | put_user(task_pid_vnr(current), current->set_child_tid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2858 | } |
| 2859 | |
| 2860 | /* |
Oleg Nesterov | dfa50b6 | 2014-10-09 21:32:32 +0200 | [diff] [blame] | 2861 | * context_switch - switch to the new MM and the new thread's register state. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2862 | */ |
Josh Poimboeuf | 0493694 | 2016-02-28 22:22:39 -0600 | [diff] [blame] | 2863 | static __always_inline struct rq * |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 2864 | context_switch(struct rq *rq, struct task_struct *prev, |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 2865 | struct task_struct *next, struct pin_cookie cookie) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2866 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2867 | struct mm_struct *mm, *oldmm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2868 | |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 2869 | prepare_task_switch(rq, prev, next); |
Peter Zijlstra | fe4b04f | 2011-02-02 13:19:09 +0100 | [diff] [blame] | 2870 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2871 | mm = next->mm; |
| 2872 | oldmm = prev->active_mm; |
Zachary Amsden | 9226d12 | 2007-02-13 13:26:21 +0100 | [diff] [blame] | 2873 | /* |
| 2874 | * For paravirt, this is coupled with an exit in switch_to to |
| 2875 | * combine the page table reload and the switch backend into |
| 2876 | * one hypercall. |
| 2877 | */ |
Jeremy Fitzhardinge | 224101e | 2009-02-18 11:18:57 -0800 | [diff] [blame] | 2878 | arch_start_context_switch(prev); |
Zachary Amsden | 9226d12 | 2007-02-13 13:26:21 +0100 | [diff] [blame] | 2879 | |
Heiko Carstens | 31915ab | 2010-09-16 14:42:25 +0200 | [diff] [blame] | 2880 | if (!mm) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2881 | next->active_mm = oldmm; |
| 2882 | atomic_inc(&oldmm->mm_count); |
| 2883 | enter_lazy_tlb(oldmm, next); |
| 2884 | } else |
Andy Lutomirski | f98db60 | 2016-04-26 09:39:06 -0700 | [diff] [blame] | 2885 | switch_mm_irqs_off(oldmm, mm, next); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2886 | |
Heiko Carstens | 31915ab | 2010-09-16 14:42:25 +0200 | [diff] [blame] | 2887 | if (!prev->mm) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2888 | prev->active_mm = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2889 | rq->prev_mm = oldmm; |
| 2890 | } |
Ingo Molnar | 3a5f5e4 | 2006-07-14 00:24:27 -0700 | [diff] [blame] | 2891 | /* |
| 2892 | * Since the runqueue lock will be released by the next |
| 2893 | * task (which is an invalid locking op but in the case |
| 2894 | * of the scheduler it's an obvious special-case), so we |
| 2895 | * do an early lockdep release here: |
| 2896 | */ |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 2897 | lockdep_unpin_lock(&rq->lock, cookie); |
Ingo Molnar | 8a25d5d | 2006-07-03 00:24:54 -0700 | [diff] [blame] | 2898 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2899 | |
| 2900 | /* Here we just switch the register state and the stack. */ |
| 2901 | switch_to(prev, next, prev); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2902 | barrier(); |
Oleg Nesterov | dfa50b6 | 2014-10-09 21:32:32 +0200 | [diff] [blame] | 2903 | |
| 2904 | return finish_task_switch(prev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2905 | } |
| 2906 | |
| 2907 | /* |
Sha Zhengju | 1c3e826 | 2013-02-20 17:14:38 +0800 | [diff] [blame] | 2908 | * nr_running and nr_context_switches: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2909 | * |
| 2910 | * externally visible scheduler statistics: current number of runnable |
Sha Zhengju | 1c3e826 | 2013-02-20 17:14:38 +0800 | [diff] [blame] | 2911 | * threads, total number of context switches performed since bootup. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2912 | */ |
| 2913 | unsigned long nr_running(void) |
| 2914 | { |
| 2915 | unsigned long i, sum = 0; |
| 2916 | |
| 2917 | for_each_online_cpu(i) |
| 2918 | sum += cpu_rq(i)->nr_running; |
| 2919 | |
| 2920 | return sum; |
| 2921 | } |
| 2922 | |
Tim Chen | 2ee507c | 2014-07-31 10:29:48 -0700 | [diff] [blame] | 2923 | /* |
| 2924 | * Check if only the current task is running on the cpu. |
Dominik Dingel | 00cc163 | 2015-09-18 11:27:45 +0200 | [diff] [blame] | 2925 | * |
| 2926 | * Caution: this function does not check that the caller has disabled |
| 2927 | * preemption, thus the result might have a time-of-check-to-time-of-use |
| 2928 | * race. The caller is responsible to use it correctly, for example: |
| 2929 | * |
| 2930 | * - from a non-preemptable section (of course) |
| 2931 | * |
| 2932 | * - from a thread that is bound to a single CPU |
| 2933 | * |
| 2934 | * - in a loop with very short iterations (e.g. a polling loop) |
Tim Chen | 2ee507c | 2014-07-31 10:29:48 -0700 | [diff] [blame] | 2935 | */ |
| 2936 | bool single_task_running(void) |
| 2937 | { |
Dominik Dingel | 00cc163 | 2015-09-18 11:27:45 +0200 | [diff] [blame] | 2938 | return raw_rq()->nr_running == 1; |
Tim Chen | 2ee507c | 2014-07-31 10:29:48 -0700 | [diff] [blame] | 2939 | } |
| 2940 | EXPORT_SYMBOL(single_task_running); |
| 2941 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2942 | unsigned long long nr_context_switches(void) |
| 2943 | { |
Steven Rostedt | cc94abf | 2006-06-27 02:54:31 -0700 | [diff] [blame] | 2944 | int i; |
| 2945 | unsigned long long sum = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2946 | |
KAMEZAWA Hiroyuki | 0a94502 | 2006-03-28 01:56:37 -0800 | [diff] [blame] | 2947 | for_each_possible_cpu(i) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2948 | sum += cpu_rq(i)->nr_switches; |
| 2949 | |
| 2950 | return sum; |
| 2951 | } |
| 2952 | |
| 2953 | unsigned long nr_iowait(void) |
| 2954 | { |
| 2955 | unsigned long i, sum = 0; |
| 2956 | |
KAMEZAWA Hiroyuki | 0a94502 | 2006-03-28 01:56:37 -0800 | [diff] [blame] | 2957 | for_each_possible_cpu(i) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2958 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
| 2959 | |
| 2960 | return sum; |
| 2961 | } |
| 2962 | |
Peter Zijlstra | 8c215bd | 2010-07-01 09:07:17 +0200 | [diff] [blame] | 2963 | unsigned long nr_iowait_cpu(int cpu) |
Arjan van de Ven | 69d2587 | 2009-09-21 17:04:08 -0700 | [diff] [blame] | 2964 | { |
Peter Zijlstra | 8c215bd | 2010-07-01 09:07:17 +0200 | [diff] [blame] | 2965 | struct rq *this = cpu_rq(cpu); |
Arjan van de Ven | 69d2587 | 2009-09-21 17:04:08 -0700 | [diff] [blame] | 2966 | return atomic_read(&this->nr_iowait); |
| 2967 | } |
| 2968 | |
Mel Gorman | 372ba8c | 2014-08-06 14:19:21 +0100 | [diff] [blame] | 2969 | void get_iowait_load(unsigned long *nr_waiters, unsigned long *load) |
| 2970 | { |
Peter Zijlstra | 3289bdb | 2015-04-14 13:19:42 +0200 | [diff] [blame] | 2971 | struct rq *rq = this_rq(); |
| 2972 | *nr_waiters = atomic_read(&rq->nr_iowait); |
| 2973 | *load = rq->load.weight; |
Mel Gorman | 372ba8c | 2014-08-06 14:19:21 +0100 | [diff] [blame] | 2974 | } |
| 2975 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2976 | #ifdef CONFIG_SMP |
| 2977 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2978 | /* |
Peter Zijlstra | 3802290 | 2009-12-16 18:04:37 +0100 | [diff] [blame] | 2979 | * sched_exec - execve() is a valuable balancing opportunity, because at |
| 2980 | * this point the task has the smallest effective memory and cache footprint. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2981 | */ |
Peter Zijlstra | 3802290 | 2009-12-16 18:04:37 +0100 | [diff] [blame] | 2982 | void sched_exec(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2983 | { |
Peter Zijlstra | 3802290 | 2009-12-16 18:04:37 +0100 | [diff] [blame] | 2984 | struct task_struct *p = current; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2985 | unsigned long flags; |
Peter Zijlstra | 0017d73 | 2010-03-24 18:34:10 +0100 | [diff] [blame] | 2986 | int dest_cpu; |
Peter Zijlstra | 3802290 | 2009-12-16 18:04:37 +0100 | [diff] [blame] | 2987 | |
Peter Zijlstra | 8f42ced | 2011-04-05 17:23:53 +0200 | [diff] [blame] | 2988 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
Peter Zijlstra | ac66f54 | 2013-10-07 11:29:16 +0100 | [diff] [blame] | 2989 | dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0); |
Peter Zijlstra | 0017d73 | 2010-03-24 18:34:10 +0100 | [diff] [blame] | 2990 | if (dest_cpu == smp_processor_id()) |
| 2991 | goto unlock; |
Peter Zijlstra | 3802290 | 2009-12-16 18:04:37 +0100 | [diff] [blame] | 2992 | |
Peter Zijlstra | 8f42ced | 2011-04-05 17:23:53 +0200 | [diff] [blame] | 2993 | if (likely(cpu_active(dest_cpu))) { |
Tejun Heo | 969c792 | 2010-05-06 18:49:21 +0200 | [diff] [blame] | 2994 | struct migration_arg arg = { p, dest_cpu }; |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 2995 | |
Peter Zijlstra | 8f42ced | 2011-04-05 17:23:53 +0200 | [diff] [blame] | 2996 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
| 2997 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2998 | return; |
| 2999 | } |
Peter Zijlstra | 0017d73 | 2010-03-24 18:34:10 +0100 | [diff] [blame] | 3000 | unlock: |
Peter Zijlstra | 8f42ced | 2011-04-05 17:23:53 +0200 | [diff] [blame] | 3001 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3002 | } |
| 3003 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3004 | #endif |
| 3005 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3006 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
Glauber Costa | 3292beb | 2011-11-28 14:45:17 -0200 | [diff] [blame] | 3007 | DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3008 | |
| 3009 | EXPORT_PER_CPU_SYMBOL(kstat); |
Glauber Costa | 3292beb | 2011-11-28 14:45:17 -0200 | [diff] [blame] | 3010 | EXPORT_PER_CPU_SYMBOL(kernel_cpustat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3011 | |
| 3012 | /* |
Giovanni Gherdovich | 6075620 | 2016-08-05 10:21:56 +0200 | [diff] [blame] | 3013 | * The function fair_sched_class.update_curr accesses the struct curr |
| 3014 | * and its field curr->exec_start; when called from task_sched_runtime(), |
| 3015 | * we observe a high rate of cache misses in practice. |
| 3016 | * Prefetching this data results in improved performance. |
| 3017 | */ |
| 3018 | static inline void prefetch_curr_exec_start(struct task_struct *p) |
| 3019 | { |
| 3020 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 3021 | struct sched_entity *curr = (&p->se)->cfs_rq->curr; |
| 3022 | #else |
| 3023 | struct sched_entity *curr = (&task_rq(p)->cfs)->curr; |
| 3024 | #endif |
| 3025 | prefetch(curr); |
| 3026 | prefetch(&curr->exec_start); |
| 3027 | } |
| 3028 | |
| 3029 | /* |
Hidetoshi Seto | c5f8d99 | 2009-03-31 16:56:03 +0900 | [diff] [blame] | 3030 | * Return accounted runtime for the task. |
| 3031 | * In case the task is currently running, return the runtime plus current's |
| 3032 | * pending runtime that have not been accounted yet. |
| 3033 | */ |
| 3034 | unsigned long long task_sched_runtime(struct task_struct *p) |
| 3035 | { |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 3036 | struct rq_flags rf; |
Hidetoshi Seto | c5f8d99 | 2009-03-31 16:56:03 +0900 | [diff] [blame] | 3037 | struct rq *rq; |
Stanislaw Gruszka | 6e99891 | 2014-11-12 16:58:44 +0100 | [diff] [blame] | 3038 | u64 ns; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 3039 | |
Peter Zijlstra | 911b289 | 2013-11-11 18:21:56 +0100 | [diff] [blame] | 3040 | #if defined(CONFIG_64BIT) && defined(CONFIG_SMP) |
| 3041 | /* |
| 3042 | * 64-bit doesn't need locks to atomically read a 64bit value. |
| 3043 | * So we have a optimization chance when the task's delta_exec is 0. |
| 3044 | * Reading ->on_cpu is racy, but this is ok. |
| 3045 | * |
| 3046 | * If we race with it leaving cpu, we'll take a lock. So we're correct. |
| 3047 | * If we race with it entering cpu, unaccounted time is 0. This is |
| 3048 | * indistinguishable from the read occurring a few cycles earlier. |
Mike Galbraith | 4036ac1 | 2014-06-24 07:49:40 +0200 | [diff] [blame] | 3049 | * If we see ->on_cpu without ->on_rq, the task is leaving, and has |
| 3050 | * been accounted, so we're correct here as well. |
Peter Zijlstra | 911b289 | 2013-11-11 18:21:56 +0100 | [diff] [blame] | 3051 | */ |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 3052 | if (!p->on_cpu || !task_on_rq_queued(p)) |
Peter Zijlstra | 911b289 | 2013-11-11 18:21:56 +0100 | [diff] [blame] | 3053 | return p->se.sum_exec_runtime; |
| 3054 | #endif |
| 3055 | |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 3056 | rq = task_rq_lock(p, &rf); |
Stanislaw Gruszka | 6e99891 | 2014-11-12 16:58:44 +0100 | [diff] [blame] | 3057 | /* |
| 3058 | * Must be ->curr _and_ ->on_rq. If dequeued, we would |
| 3059 | * project cycles that may never be accounted to this |
| 3060 | * thread, breaking clock_gettime(). |
| 3061 | */ |
| 3062 | if (task_current(rq, p) && task_on_rq_queued(p)) { |
Giovanni Gherdovich | 6075620 | 2016-08-05 10:21:56 +0200 | [diff] [blame] | 3063 | prefetch_curr_exec_start(p); |
Stanislaw Gruszka | 6e99891 | 2014-11-12 16:58:44 +0100 | [diff] [blame] | 3064 | update_rq_clock(rq); |
| 3065 | p->sched_class->update_curr(rq); |
| 3066 | } |
| 3067 | ns = p->se.sum_exec_runtime; |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 3068 | task_rq_unlock(rq, p, &rf); |
Hidetoshi Seto | c5f8d99 | 2009-03-31 16:56:03 +0900 | [diff] [blame] | 3069 | |
| 3070 | return ns; |
| 3071 | } |
| 3072 | |
Balbir Singh | 4904862 | 2008-09-05 18:12:23 +0200 | [diff] [blame] | 3073 | /* |
Christoph Lameter | 7835b98 | 2006-12-10 02:20:22 -0800 | [diff] [blame] | 3074 | * This function gets called by the timer code, with HZ frequency. |
| 3075 | * We call it with interrupts disabled. |
Christoph Lameter | 7835b98 | 2006-12-10 02:20:22 -0800 | [diff] [blame] | 3076 | */ |
| 3077 | void scheduler_tick(void) |
| 3078 | { |
Christoph Lameter | 7835b98 | 2006-12-10 02:20:22 -0800 | [diff] [blame] | 3079 | int cpu = smp_processor_id(); |
| 3080 | struct rq *rq = cpu_rq(cpu); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3081 | struct task_struct *curr = rq->curr; |
Peter Zijlstra | 3e51f33 | 2008-05-03 18:29:28 +0200 | [diff] [blame] | 3082 | |
| 3083 | sched_clock_tick(); |
Christoph Lameter | 7835b98 | 2006-12-10 02:20:22 -0800 | [diff] [blame] | 3084 | |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 3085 | raw_spin_lock(&rq->lock); |
Peter Zijlstra | 3e51f33 | 2008-05-03 18:29:28 +0200 | [diff] [blame] | 3086 | update_rq_clock(rq); |
Peter Zijlstra | fa85ae2 | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 3087 | curr->sched_class->task_tick(rq, curr, 0); |
Frederic Weisbecker | cee1afc | 2016-04-13 15:56:50 +0200 | [diff] [blame] | 3088 | cpu_load_update_active(rq); |
Peter Zijlstra | 3289bdb | 2015-04-14 13:19:42 +0200 | [diff] [blame] | 3089 | calc_global_load_tick(rq); |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 3090 | raw_spin_unlock(&rq->lock); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3091 | |
Peter Zijlstra | e9d2b06 | 2010-09-17 11:28:50 +0200 | [diff] [blame] | 3092 | perf_event_task_tick(); |
Peter Zijlstra | e220d2d | 2009-05-23 18:28:55 +0200 | [diff] [blame] | 3093 | |
Christoph Lameter | e418e1c | 2006-12-10 02:20:23 -0800 | [diff] [blame] | 3094 | #ifdef CONFIG_SMP |
Suresh Siddha | 6eb57e0 | 2011-10-03 15:09:01 -0700 | [diff] [blame] | 3095 | rq->idle_balance = idle_cpu(cpu); |
Daniel Lezcano | 7caff66 | 2014-01-06 12:34:38 +0100 | [diff] [blame] | 3096 | trigger_load_balance(rq); |
Christoph Lameter | e418e1c | 2006-12-10 02:20:23 -0800 | [diff] [blame] | 3097 | #endif |
Frederic Weisbecker | 265f22a | 2013-05-03 03:39:05 +0200 | [diff] [blame] | 3098 | rq_last_tick_reset(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3099 | } |
| 3100 | |
Frederic Weisbecker | 265f22a | 2013-05-03 03:39:05 +0200 | [diff] [blame] | 3101 | #ifdef CONFIG_NO_HZ_FULL |
| 3102 | /** |
| 3103 | * scheduler_tick_max_deferment |
| 3104 | * |
| 3105 | * Keep at least one tick per second when a single |
| 3106 | * active task is running because the scheduler doesn't |
| 3107 | * yet completely support full dynticks environment. |
| 3108 | * |
| 3109 | * This makes sure that uptime, CFS vruntime, load |
| 3110 | * balancing, etc... continue to move forward, even |
| 3111 | * with a very low granularity. |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 3112 | * |
| 3113 | * Return: Maximum deferment in nanoseconds. |
Frederic Weisbecker | 265f22a | 2013-05-03 03:39:05 +0200 | [diff] [blame] | 3114 | */ |
| 3115 | u64 scheduler_tick_max_deferment(void) |
| 3116 | { |
| 3117 | struct rq *rq = this_rq(); |
Jason Low | 316c1608d | 2015-04-28 13:00:20 -0700 | [diff] [blame] | 3118 | unsigned long next, now = READ_ONCE(jiffies); |
Frederic Weisbecker | 265f22a | 2013-05-03 03:39:05 +0200 | [diff] [blame] | 3119 | |
| 3120 | next = rq->last_sched_tick + HZ; |
| 3121 | |
| 3122 | if (time_before_eq(next, now)) |
| 3123 | return 0; |
| 3124 | |
Kevin Hilman | 8fe8ff0 | 2014-01-15 14:51:38 +0100 | [diff] [blame] | 3125 | return jiffies_to_nsecs(next - now); |
Frederic Weisbecker | 265f22a | 2013-05-03 03:39:05 +0200 | [diff] [blame] | 3126 | } |
| 3127 | #endif |
| 3128 | |
Steven Rostedt | 7e49fcc | 2009-01-22 19:01:40 -0500 | [diff] [blame] | 3129 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
| 3130 | defined(CONFIG_PREEMPT_TRACER)) |
Steven Rostedt | 47252cf | 2016-03-21 11:23:39 -0400 | [diff] [blame] | 3131 | /* |
| 3132 | * If the value passed in is equal to the current preempt count |
| 3133 | * then we just disabled preemption. Start timing the latency. |
| 3134 | */ |
| 3135 | static inline void preempt_latency_start(int val) |
| 3136 | { |
| 3137 | if (preempt_count() == val) { |
| 3138 | unsigned long ip = get_lock_parent_ip(); |
| 3139 | #ifdef CONFIG_DEBUG_PREEMPT |
| 3140 | current->preempt_disable_ip = ip; |
| 3141 | #endif |
| 3142 | trace_preempt_off(CALLER_ADDR0, ip); |
| 3143 | } |
| 3144 | } |
Steven Rostedt | 7e49fcc | 2009-01-22 19:01:40 -0500 | [diff] [blame] | 3145 | |
Masami Hiramatsu | edafe3a | 2014-04-17 17:18:42 +0900 | [diff] [blame] | 3146 | void preempt_count_add(int val) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3147 | { |
Steven Rostedt | 6cd8a4b | 2008-05-12 21:20:42 +0200 | [diff] [blame] | 3148 | #ifdef CONFIG_DEBUG_PREEMPT |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3149 | /* |
| 3150 | * Underflow? |
| 3151 | */ |
Ingo Molnar | 9a11b49a | 2006-07-03 00:24:33 -0700 | [diff] [blame] | 3152 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
| 3153 | return; |
Steven Rostedt | 6cd8a4b | 2008-05-12 21:20:42 +0200 | [diff] [blame] | 3154 | #endif |
Peter Zijlstra | bdb4380 | 2013-09-10 12:15:23 +0200 | [diff] [blame] | 3155 | __preempt_count_add(val); |
Steven Rostedt | 6cd8a4b | 2008-05-12 21:20:42 +0200 | [diff] [blame] | 3156 | #ifdef CONFIG_DEBUG_PREEMPT |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3157 | /* |
| 3158 | * Spinlock count overflowing soon? |
| 3159 | */ |
Miguel Ojeda Sandonis | 33859f7 | 2006-12-10 02:20:38 -0800 | [diff] [blame] | 3160 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
| 3161 | PREEMPT_MASK - 10); |
Steven Rostedt | 6cd8a4b | 2008-05-12 21:20:42 +0200 | [diff] [blame] | 3162 | #endif |
Steven Rostedt | 47252cf | 2016-03-21 11:23:39 -0400 | [diff] [blame] | 3163 | preempt_latency_start(val); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3164 | } |
Peter Zijlstra | bdb4380 | 2013-09-10 12:15:23 +0200 | [diff] [blame] | 3165 | EXPORT_SYMBOL(preempt_count_add); |
Masami Hiramatsu | edafe3a | 2014-04-17 17:18:42 +0900 | [diff] [blame] | 3166 | NOKPROBE_SYMBOL(preempt_count_add); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3167 | |
Steven Rostedt | 47252cf | 2016-03-21 11:23:39 -0400 | [diff] [blame] | 3168 | /* |
| 3169 | * If the value passed in equals to the current preempt count |
| 3170 | * then we just enabled preemption. Stop timing the latency. |
| 3171 | */ |
| 3172 | static inline void preempt_latency_stop(int val) |
| 3173 | { |
| 3174 | if (preempt_count() == val) |
| 3175 | trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip()); |
| 3176 | } |
| 3177 | |
Masami Hiramatsu | edafe3a | 2014-04-17 17:18:42 +0900 | [diff] [blame] | 3178 | void preempt_count_sub(int val) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3179 | { |
Steven Rostedt | 6cd8a4b | 2008-05-12 21:20:42 +0200 | [diff] [blame] | 3180 | #ifdef CONFIG_DEBUG_PREEMPT |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3181 | /* |
| 3182 | * Underflow? |
| 3183 | */ |
Ingo Molnar | 01e3eb8 | 2009-01-12 13:00:50 +0100 | [diff] [blame] | 3184 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
Ingo Molnar | 9a11b49a | 2006-07-03 00:24:33 -0700 | [diff] [blame] | 3185 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3186 | /* |
| 3187 | * Is the spinlock portion underflowing? |
| 3188 | */ |
Ingo Molnar | 9a11b49a | 2006-07-03 00:24:33 -0700 | [diff] [blame] | 3189 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
| 3190 | !(preempt_count() & PREEMPT_MASK))) |
| 3191 | return; |
Steven Rostedt | 6cd8a4b | 2008-05-12 21:20:42 +0200 | [diff] [blame] | 3192 | #endif |
Ingo Molnar | 9a11b49a | 2006-07-03 00:24:33 -0700 | [diff] [blame] | 3193 | |
Steven Rostedt | 47252cf | 2016-03-21 11:23:39 -0400 | [diff] [blame] | 3194 | preempt_latency_stop(val); |
Peter Zijlstra | bdb4380 | 2013-09-10 12:15:23 +0200 | [diff] [blame] | 3195 | __preempt_count_sub(val); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3196 | } |
Peter Zijlstra | bdb4380 | 2013-09-10 12:15:23 +0200 | [diff] [blame] | 3197 | EXPORT_SYMBOL(preempt_count_sub); |
Masami Hiramatsu | edafe3a | 2014-04-17 17:18:42 +0900 | [diff] [blame] | 3198 | NOKPROBE_SYMBOL(preempt_count_sub); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3199 | |
Steven Rostedt | 47252cf | 2016-03-21 11:23:39 -0400 | [diff] [blame] | 3200 | #else |
| 3201 | static inline void preempt_latency_start(int val) { } |
| 3202 | static inline void preempt_latency_stop(int val) { } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3203 | #endif |
| 3204 | |
| 3205 | /* |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3206 | * Print scheduling while atomic bug: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3207 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3208 | static noinline void __schedule_bug(struct task_struct *prev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3209 | { |
Vegard Nossum | d1c6d14 | 2016-07-23 09:46:39 +0200 | [diff] [blame] | 3210 | /* Save this before calling printk(), since that will clobber it */ |
| 3211 | unsigned long preempt_disable_ip = get_preempt_disable_ip(current); |
| 3212 | |
Dave Jones | 664dfa6 | 2011-12-22 16:39:30 -0500 | [diff] [blame] | 3213 | if (oops_in_progress) |
| 3214 | return; |
| 3215 | |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 3216 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
| 3217 | prev->comm, prev->pid, preempt_count()); |
Satyam Sharma | 838225b | 2007-10-24 18:23:50 +0200 | [diff] [blame] | 3218 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3219 | debug_show_held_locks(prev); |
Arjan van de Ven | e21f5b1 | 2008-05-23 09:05:58 -0700 | [diff] [blame] | 3220 | print_modules(); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3221 | if (irqs_disabled()) |
| 3222 | print_irqtrace_events(prev); |
Vegard Nossum | d1c6d14 | 2016-07-23 09:46:39 +0200 | [diff] [blame] | 3223 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
| 3224 | && in_atomic_preempt_off()) { |
Thomas Gleixner | 8f47b18 | 2014-02-07 20:58:39 +0100 | [diff] [blame] | 3225 | pr_err("Preemption disabled at:"); |
Vegard Nossum | d1c6d14 | 2016-07-23 09:46:39 +0200 | [diff] [blame] | 3226 | print_ip_sym(preempt_disable_ip); |
Thomas Gleixner | 8f47b18 | 2014-02-07 20:58:39 +0100 | [diff] [blame] | 3227 | pr_cont("\n"); |
| 3228 | } |
Daniel Bristot de Oliveira | 748c720 | 2016-06-03 17:10:18 -0300 | [diff] [blame] | 3229 | if (panic_on_warn) |
| 3230 | panic("scheduling while atomic\n"); |
| 3231 | |
Stephen Boyd | 6135fc1 | 2012-03-28 17:10:47 -0700 | [diff] [blame] | 3232 | dump_stack(); |
Rusty Russell | 373d4d0 | 2013-01-21 17:17:39 +1030 | [diff] [blame] | 3233 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3234 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3235 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3236 | /* |
| 3237 | * Various schedule()-time debugging checks and statistics: |
| 3238 | */ |
| 3239 | static inline void schedule_debug(struct task_struct *prev) |
| 3240 | { |
Aaron Tomlin | 0d9e263 | 2014-09-12 14:16:19 +0100 | [diff] [blame] | 3241 | #ifdef CONFIG_SCHED_STACK_END_CHECK |
Jann Horn | 29d6455 | 2016-06-01 11:55:07 +0200 | [diff] [blame] | 3242 | if (task_stack_end_corrupted(prev)) |
| 3243 | panic("corrupted stack end detected inside scheduler\n"); |
Aaron Tomlin | 0d9e263 | 2014-09-12 14:16:19 +0100 | [diff] [blame] | 3244 | #endif |
Peter Zijlstra | b99def8 | 2015-09-28 18:02:03 +0200 | [diff] [blame] | 3245 | |
Peter Zijlstra | 1dc0fff | 2015-09-28 17:57:39 +0200 | [diff] [blame] | 3246 | if (unlikely(in_atomic_preempt_off())) { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3247 | __schedule_bug(prev); |
Peter Zijlstra | 1dc0fff | 2015-09-28 17:57:39 +0200 | [diff] [blame] | 3248 | preempt_count_set(PREEMPT_DISABLED); |
| 3249 | } |
Paul E. McKenney | b3fbab0 | 2011-05-24 08:31:09 -0700 | [diff] [blame] | 3250 | rcu_sleep_check(); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3251 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3252 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
| 3253 | |
Josh Poimboeuf | ae92882 | 2016-06-17 12:43:24 -0500 | [diff] [blame] | 3254 | schedstat_inc(this_rq()->sched_count); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3255 | } |
| 3256 | |
| 3257 | /* |
| 3258 | * Pick up the highest-prio task: |
| 3259 | */ |
| 3260 | static inline struct task_struct * |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 3261 | pick_next_task(struct rq *rq, struct task_struct *prev, struct pin_cookie cookie) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3262 | { |
Peter Zijlstra | 37e117c | 2014-02-14 12:25:08 +0100 | [diff] [blame] | 3263 | const struct sched_class *class = &fair_sched_class; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3264 | struct task_struct *p; |
| 3265 | |
| 3266 | /* |
| 3267 | * Optimization: we know that if all tasks are in |
| 3268 | * the fair class we can call that function directly: |
| 3269 | */ |
Peter Zijlstra | 37e117c | 2014-02-14 12:25:08 +0100 | [diff] [blame] | 3270 | if (likely(prev->sched_class == class && |
Peter Zijlstra | 38033c3 | 2014-01-23 20:32:21 +0100 | [diff] [blame] | 3271 | rq->nr_running == rq->cfs.h_nr_running)) { |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 3272 | p = fair_sched_class.pick_next_task(rq, prev, cookie); |
Peter Zijlstra | 6ccdc84 | 2014-04-24 12:00:47 +0200 | [diff] [blame] | 3273 | if (unlikely(p == RETRY_TASK)) |
| 3274 | goto again; |
| 3275 | |
| 3276 | /* assumes fair_sched_class->next == idle_sched_class */ |
| 3277 | if (unlikely(!p)) |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 3278 | p = idle_sched_class.pick_next_task(rq, prev, cookie); |
Peter Zijlstra | 6ccdc84 | 2014-04-24 12:00:47 +0200 | [diff] [blame] | 3279 | |
| 3280 | return p; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3281 | } |
| 3282 | |
Peter Zijlstra | 37e117c | 2014-02-14 12:25:08 +0100 | [diff] [blame] | 3283 | again: |
Peter Zijlstra | 34f971f | 2010-09-22 13:53:15 +0200 | [diff] [blame] | 3284 | for_each_class(class) { |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 3285 | p = class->pick_next_task(rq, prev, cookie); |
Peter Zijlstra | 37e117c | 2014-02-14 12:25:08 +0100 | [diff] [blame] | 3286 | if (p) { |
| 3287 | if (unlikely(p == RETRY_TASK)) |
| 3288 | goto again; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3289 | return p; |
Peter Zijlstra | 37e117c | 2014-02-14 12:25:08 +0100 | [diff] [blame] | 3290 | } |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3291 | } |
Peter Zijlstra | 34f971f | 2010-09-22 13:53:15 +0200 | [diff] [blame] | 3292 | |
| 3293 | BUG(); /* the idle class will always have a runnable task */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3294 | } |
| 3295 | |
| 3296 | /* |
Thomas Gleixner | c259e01 | 2011-06-22 19:47:00 +0200 | [diff] [blame] | 3297 | * __schedule() is the main scheduler function. |
Pekka Enberg | edde96e | 2012-08-04 11:49:47 +0300 | [diff] [blame] | 3298 | * |
| 3299 | * The main means of driving the scheduler and thus entering this function are: |
| 3300 | * |
| 3301 | * 1. Explicit blocking: mutex, semaphore, waitqueue, etc. |
| 3302 | * |
| 3303 | * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return |
| 3304 | * paths. For example, see arch/x86/entry_64.S. |
| 3305 | * |
| 3306 | * To drive preemption between tasks, the scheduler sets the flag in timer |
| 3307 | * interrupt handler scheduler_tick(). |
| 3308 | * |
| 3309 | * 3. Wakeups don't really cause entry into schedule(). They add a |
| 3310 | * task to the run-queue and that's it. |
| 3311 | * |
| 3312 | * Now, if the new task added to the run-queue preempts the current |
| 3313 | * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets |
| 3314 | * called on the nearest possible occasion: |
| 3315 | * |
| 3316 | * - If the kernel is preemptible (CONFIG_PREEMPT=y): |
| 3317 | * |
| 3318 | * - in syscall or exception context, at the next outmost |
| 3319 | * preempt_enable(). (this might be as soon as the wake_up()'s |
| 3320 | * spin_unlock()!) |
| 3321 | * |
| 3322 | * - in IRQ context, return from interrupt-handler to |
| 3323 | * preemptible context |
| 3324 | * |
| 3325 | * - If the kernel is not preemptible (CONFIG_PREEMPT is not set) |
| 3326 | * then at the next: |
| 3327 | * |
| 3328 | * - cond_resched() call |
| 3329 | * - explicit schedule() call |
| 3330 | * - return from syscall or exception to user-space |
| 3331 | * - return from interrupt-handler to user-space |
Frederic Weisbecker | bfd9b2b | 2015-01-28 01:24:09 +0100 | [diff] [blame] | 3332 | * |
Frederic Weisbecker | b30f0e3 | 2015-05-12 16:41:49 +0200 | [diff] [blame] | 3333 | * WARNING: must be called with preemption disabled! |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3334 | */ |
Peter Zijlstra | 499d795 | 2015-09-28 18:52:36 +0200 | [diff] [blame] | 3335 | static void __sched notrace __schedule(bool preempt) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3336 | { |
| 3337 | struct task_struct *prev, *next; |
Harvey Harrison | 67ca7bd | 2008-02-15 09:56:36 -0800 | [diff] [blame] | 3338 | unsigned long *switch_count; |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 3339 | struct pin_cookie cookie; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3340 | struct rq *rq; |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 3341 | int cpu; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3342 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3343 | cpu = smp_processor_id(); |
| 3344 | rq = cpu_rq(cpu); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3345 | prev = rq->curr; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3346 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3347 | schedule_debug(prev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3348 | |
Peter Zijlstra | 3165651 | 2008-07-18 18:01:23 +0200 | [diff] [blame] | 3349 | if (sched_feat(HRTICK)) |
Mike Galbraith | f333fdc | 2008-05-12 21:20:55 +0200 | [diff] [blame] | 3350 | hrtick_clear(rq); |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 3351 | |
Paul E. McKenney | 46a5d16 | 2015-10-07 09:10:48 -0700 | [diff] [blame] | 3352 | local_irq_disable(); |
| 3353 | rcu_note_context_switch(); |
| 3354 | |
Oleg Nesterov | e0acd0a | 2013-08-12 18:14:00 +0200 | [diff] [blame] | 3355 | /* |
| 3356 | * Make sure that signal_pending_state()->signal_pending() below |
| 3357 | * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE) |
| 3358 | * done by the caller to avoid the race with signal_wake_up(). |
| 3359 | */ |
| 3360 | smp_mb__before_spinlock(); |
Paul E. McKenney | 46a5d16 | 2015-10-07 09:10:48 -0700 | [diff] [blame] | 3361 | raw_spin_lock(&rq->lock); |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 3362 | cookie = lockdep_pin_lock(&rq->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3363 | |
Peter Zijlstra | 9edfbfe | 2015-01-05 11:18:11 +0100 | [diff] [blame] | 3364 | rq->clock_skip_update <<= 1; /* promote REQ to ACT */ |
| 3365 | |
Oleg Nesterov | 246d86b | 2010-05-19 14:57:11 +0200 | [diff] [blame] | 3366 | switch_count = &prev->nivcsw; |
Peter Zijlstra | fc13aeb | 2015-09-28 18:05:34 +0200 | [diff] [blame] | 3367 | if (!preempt && prev->state) { |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 3368 | if (unlikely(signal_pending_state(prev->state, prev))) { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3369 | prev->state = TASK_RUNNING; |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 3370 | } else { |
Peter Zijlstra | 2acca55 | 2011-04-05 17:23:50 +0200 | [diff] [blame] | 3371 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
| 3372 | prev->on_rq = 0; |
| 3373 | |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 3374 | /* |
Peter Zijlstra | 2acca55 | 2011-04-05 17:23:50 +0200 | [diff] [blame] | 3375 | * If a worker went to sleep, notify and ask workqueue |
| 3376 | * whether it wants to wake up a task to maintain |
| 3377 | * concurrency. |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 3378 | */ |
| 3379 | if (prev->flags & PF_WQ_WORKER) { |
| 3380 | struct task_struct *to_wakeup; |
| 3381 | |
Alexander Gordeev | 9b7f659 | 2016-03-02 12:53:31 +0100 | [diff] [blame] | 3382 | to_wakeup = wq_worker_sleeping(prev); |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 3383 | if (to_wakeup) |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 3384 | try_to_wake_up_local(to_wakeup, cookie); |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 3385 | } |
Tejun Heo | 21aa9af | 2010-06-08 21:40:37 +0200 | [diff] [blame] | 3386 | } |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3387 | switch_count = &prev->nvcsw; |
| 3388 | } |
| 3389 | |
Peter Zijlstra | 9edfbfe | 2015-01-05 11:18:11 +0100 | [diff] [blame] | 3390 | if (task_on_rq_queued(prev)) |
Peter Zijlstra | 606dba2 | 2012-02-11 06:05:00 +0100 | [diff] [blame] | 3391 | update_rq_clock(rq); |
| 3392 | |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 3393 | next = pick_next_task(rq, prev, cookie); |
Mike Galbraith | f26f9af | 2010-12-08 11:05:42 +0100 | [diff] [blame] | 3394 | clear_tsk_need_resched(prev); |
Peter Zijlstra | f27dde8 | 2013-08-14 14:55:31 +0200 | [diff] [blame] | 3395 | clear_preempt_need_resched(); |
Peter Zijlstra | 9edfbfe | 2015-01-05 11:18:11 +0100 | [diff] [blame] | 3396 | rq->clock_skip_update = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3397 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3398 | if (likely(prev != next)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3399 | rq->nr_switches++; |
| 3400 | rq->curr = next; |
| 3401 | ++*switch_count; |
| 3402 | |
Peter Zijlstra | c73464b | 2015-09-28 18:06:56 +0200 | [diff] [blame] | 3403 | trace_sched_switch(preempt, prev, next); |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 3404 | rq = context_switch(rq, prev, next, cookie); /* unlocks the rq */ |
Peter Zijlstra | cbce1a6 | 2015-06-11 14:46:54 +0200 | [diff] [blame] | 3405 | } else { |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 3406 | lockdep_unpin_lock(&rq->lock, cookie); |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 3407 | raw_spin_unlock_irq(&rq->lock); |
Peter Zijlstra | cbce1a6 | 2015-06-11 14:46:54 +0200 | [diff] [blame] | 3408 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3409 | |
Peter Zijlstra | e3fca9e | 2015-06-11 14:46:37 +0200 | [diff] [blame] | 3410 | balance_callback(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3411 | } |
Thomas Gleixner | c259e01 | 2011-06-22 19:47:00 +0200 | [diff] [blame] | 3412 | |
Peter Zijlstra | 9af6528 | 2016-09-13 18:37:29 +0200 | [diff] [blame] | 3413 | void __noreturn do_task_dead(void) |
| 3414 | { |
| 3415 | /* |
| 3416 | * The setting of TASK_RUNNING by try_to_wake_up() may be delayed |
| 3417 | * when the following two conditions become true. |
| 3418 | * - There is race condition of mmap_sem (It is acquired by |
| 3419 | * exit_mm()), and |
| 3420 | * - SMI occurs before setting TASK_RUNINNG. |
| 3421 | * (or hypervisor of virtual machine switches to other guest) |
| 3422 | * As a result, we may become TASK_RUNNING after becoming TASK_DEAD |
| 3423 | * |
| 3424 | * To avoid it, we have to wait for releasing tsk->pi_lock which |
| 3425 | * is held by try_to_wake_up() |
| 3426 | */ |
| 3427 | smp_mb(); |
| 3428 | raw_spin_unlock_wait(¤t->pi_lock); |
| 3429 | |
| 3430 | /* causes final put_task_struct in finish_task_switch(). */ |
| 3431 | __set_current_state(TASK_DEAD); |
| 3432 | current->flags |= PF_NOFREEZE; /* tell freezer to ignore us */ |
| 3433 | __schedule(false); |
| 3434 | BUG(); |
| 3435 | /* Avoid "noreturn function does return". */ |
| 3436 | for (;;) |
| 3437 | cpu_relax(); /* For when BUG is null */ |
| 3438 | } |
| 3439 | |
Thomas Gleixner | 9c40cef | 2011-06-22 19:47:01 +0200 | [diff] [blame] | 3440 | static inline void sched_submit_work(struct task_struct *tsk) |
| 3441 | { |
Thomas Gleixner | 3c7d518 | 2011-07-17 20:46:52 +0200 | [diff] [blame] | 3442 | if (!tsk->state || tsk_is_pi_blocked(tsk)) |
Thomas Gleixner | 9c40cef | 2011-06-22 19:47:01 +0200 | [diff] [blame] | 3443 | return; |
| 3444 | /* |
| 3445 | * If we are going to sleep and we have plugged IO queued, |
| 3446 | * make sure to submit it to avoid deadlocks. |
| 3447 | */ |
| 3448 | if (blk_needs_flush_plug(tsk)) |
| 3449 | blk_schedule_flush_plug(tsk); |
| 3450 | } |
| 3451 | |
Andi Kleen | 722a9f9 | 2014-05-02 00:44:38 +0200 | [diff] [blame] | 3452 | asmlinkage __visible void __sched schedule(void) |
Thomas Gleixner | c259e01 | 2011-06-22 19:47:00 +0200 | [diff] [blame] | 3453 | { |
Thomas Gleixner | 9c40cef | 2011-06-22 19:47:01 +0200 | [diff] [blame] | 3454 | struct task_struct *tsk = current; |
| 3455 | |
| 3456 | sched_submit_work(tsk); |
Frederic Weisbecker | bfd9b2b | 2015-01-28 01:24:09 +0100 | [diff] [blame] | 3457 | do { |
Frederic Weisbecker | b30f0e3 | 2015-05-12 16:41:49 +0200 | [diff] [blame] | 3458 | preempt_disable(); |
Peter Zijlstra | fc13aeb | 2015-09-28 18:05:34 +0200 | [diff] [blame] | 3459 | __schedule(false); |
Frederic Weisbecker | b30f0e3 | 2015-05-12 16:41:49 +0200 | [diff] [blame] | 3460 | sched_preempt_enable_no_resched(); |
Frederic Weisbecker | bfd9b2b | 2015-01-28 01:24:09 +0100 | [diff] [blame] | 3461 | } while (need_resched()); |
Thomas Gleixner | c259e01 | 2011-06-22 19:47:00 +0200 | [diff] [blame] | 3462 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3463 | EXPORT_SYMBOL(schedule); |
| 3464 | |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 3465 | #ifdef CONFIG_CONTEXT_TRACKING |
Andi Kleen | 722a9f9 | 2014-05-02 00:44:38 +0200 | [diff] [blame] | 3466 | asmlinkage __visible void __sched schedule_user(void) |
Frederic Weisbecker | 20ab65e3 | 2012-07-11 20:26:37 +0200 | [diff] [blame] | 3467 | { |
| 3468 | /* |
| 3469 | * If we come here after a random call to set_need_resched(), |
| 3470 | * or we have been woken up remotely but the IPI has not yet arrived, |
| 3471 | * we haven't yet exited the RCU idle mode. Do it here manually until |
| 3472 | * we find a better solution. |
Andy Lutomirski | 7cc78f8 | 2014-12-03 15:37:08 -0800 | [diff] [blame] | 3473 | * |
| 3474 | * NB: There are buggy callers of this function. Ideally we |
Frederic Weisbecker | c467ea7 | 2015-03-04 18:06:33 +0100 | [diff] [blame] | 3475 | * should warn if prev_state != CONTEXT_USER, but that will trigger |
Andy Lutomirski | 7cc78f8 | 2014-12-03 15:37:08 -0800 | [diff] [blame] | 3476 | * too frequently to make sense yet. |
Frederic Weisbecker | 20ab65e3 | 2012-07-11 20:26:37 +0200 | [diff] [blame] | 3477 | */ |
Andy Lutomirski | 7cc78f8 | 2014-12-03 15:37:08 -0800 | [diff] [blame] | 3478 | enum ctx_state prev_state = exception_enter(); |
Frederic Weisbecker | 20ab65e3 | 2012-07-11 20:26:37 +0200 | [diff] [blame] | 3479 | schedule(); |
Andy Lutomirski | 7cc78f8 | 2014-12-03 15:37:08 -0800 | [diff] [blame] | 3480 | exception_exit(prev_state); |
Frederic Weisbecker | 20ab65e3 | 2012-07-11 20:26:37 +0200 | [diff] [blame] | 3481 | } |
| 3482 | #endif |
| 3483 | |
Thomas Gleixner | c5491ea | 2011-03-21 12:09:35 +0100 | [diff] [blame] | 3484 | /** |
| 3485 | * schedule_preempt_disabled - called with preemption disabled |
| 3486 | * |
| 3487 | * Returns with preemption disabled. Note: preempt_count must be 1 |
| 3488 | */ |
| 3489 | void __sched schedule_preempt_disabled(void) |
| 3490 | { |
Thomas Gleixner | ba74c14 | 2011-03-21 13:32:17 +0100 | [diff] [blame] | 3491 | sched_preempt_enable_no_resched(); |
Thomas Gleixner | c5491ea | 2011-03-21 12:09:35 +0100 | [diff] [blame] | 3492 | schedule(); |
| 3493 | preempt_disable(); |
| 3494 | } |
| 3495 | |
Frederic Weisbecker | 06b1f80 | 2015-02-16 19:20:07 +0100 | [diff] [blame] | 3496 | static void __sched notrace preempt_schedule_common(void) |
Frederic Weisbecker | a18b5d0 | 2015-01-22 18:08:04 +0100 | [diff] [blame] | 3497 | { |
| 3498 | do { |
Steven Rostedt | 47252cf | 2016-03-21 11:23:39 -0400 | [diff] [blame] | 3499 | /* |
| 3500 | * Because the function tracer can trace preempt_count_sub() |
| 3501 | * and it also uses preempt_enable/disable_notrace(), if |
| 3502 | * NEED_RESCHED is set, the preempt_enable_notrace() called |
| 3503 | * by the function tracer will call this function again and |
| 3504 | * cause infinite recursion. |
| 3505 | * |
| 3506 | * Preemption must be disabled here before the function |
| 3507 | * tracer can trace. Break up preempt_disable() into two |
| 3508 | * calls. One to disable preemption without fear of being |
| 3509 | * traced. The other to still record the preemption latency, |
| 3510 | * which can also be traced by the function tracer. |
| 3511 | */ |
Peter Zijlstra | 499d795 | 2015-09-28 18:52:36 +0200 | [diff] [blame] | 3512 | preempt_disable_notrace(); |
Steven Rostedt | 47252cf | 2016-03-21 11:23:39 -0400 | [diff] [blame] | 3513 | preempt_latency_start(1); |
Peter Zijlstra | fc13aeb | 2015-09-28 18:05:34 +0200 | [diff] [blame] | 3514 | __schedule(true); |
Steven Rostedt | 47252cf | 2016-03-21 11:23:39 -0400 | [diff] [blame] | 3515 | preempt_latency_stop(1); |
Peter Zijlstra | 499d795 | 2015-09-28 18:52:36 +0200 | [diff] [blame] | 3516 | preempt_enable_no_resched_notrace(); |
Frederic Weisbecker | a18b5d0 | 2015-01-22 18:08:04 +0100 | [diff] [blame] | 3517 | |
| 3518 | /* |
| 3519 | * Check again in case we missed a preemption opportunity |
| 3520 | * between schedule and now. |
| 3521 | */ |
Frederic Weisbecker | a18b5d0 | 2015-01-22 18:08:04 +0100 | [diff] [blame] | 3522 | } while (need_resched()); |
| 3523 | } |
| 3524 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3525 | #ifdef CONFIG_PREEMPT |
| 3526 | /* |
Andreas Mohr | 2ed6e34 | 2006-07-10 04:43:52 -0700 | [diff] [blame] | 3527 | * this is the entry point to schedule() from in-kernel preemption |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 3528 | * off of preempt_enable. Kernel preemptions off return from interrupt |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3529 | * occur there and call schedule directly. |
| 3530 | */ |
Andi Kleen | 722a9f9 | 2014-05-02 00:44:38 +0200 | [diff] [blame] | 3531 | asmlinkage __visible void __sched notrace preempt_schedule(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3532 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3533 | /* |
| 3534 | * If there is a non-zero preempt_count or interrupts are disabled, |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 3535 | * we do not want to preempt the current task. Just return.. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3536 | */ |
Frederic Weisbecker | fbb00b5 | 2013-06-19 23:56:22 +0200 | [diff] [blame] | 3537 | if (likely(!preemptible())) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3538 | return; |
| 3539 | |
Frederic Weisbecker | a18b5d0 | 2015-01-22 18:08:04 +0100 | [diff] [blame] | 3540 | preempt_schedule_common(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3541 | } |
Masami Hiramatsu | 376e242 | 2014-04-17 17:17:05 +0900 | [diff] [blame] | 3542 | NOKPROBE_SYMBOL(preempt_schedule); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3543 | EXPORT_SYMBOL(preempt_schedule); |
Oleg Nesterov | 009f60e | 2014-10-05 22:23:22 +0200 | [diff] [blame] | 3544 | |
Oleg Nesterov | 009f60e | 2014-10-05 22:23:22 +0200 | [diff] [blame] | 3545 | /** |
Frederic Weisbecker | 4eaca0a | 2015-06-04 17:39:08 +0200 | [diff] [blame] | 3546 | * preempt_schedule_notrace - preempt_schedule called by tracing |
Oleg Nesterov | 009f60e | 2014-10-05 22:23:22 +0200 | [diff] [blame] | 3547 | * |
| 3548 | * The tracing infrastructure uses preempt_enable_notrace to prevent |
| 3549 | * recursion and tracing preempt enabling caused by the tracing |
| 3550 | * infrastructure itself. But as tracing can happen in areas coming |
| 3551 | * from userspace or just about to enter userspace, a preempt enable |
| 3552 | * can occur before user_exit() is called. This will cause the scheduler |
| 3553 | * to be called when the system is still in usermode. |
| 3554 | * |
| 3555 | * To prevent this, the preempt_enable_notrace will use this function |
| 3556 | * instead of preempt_schedule() to exit user context if needed before |
| 3557 | * calling the scheduler. |
| 3558 | */ |
Frederic Weisbecker | 4eaca0a | 2015-06-04 17:39:08 +0200 | [diff] [blame] | 3559 | asmlinkage __visible void __sched notrace preempt_schedule_notrace(void) |
Oleg Nesterov | 009f60e | 2014-10-05 22:23:22 +0200 | [diff] [blame] | 3560 | { |
| 3561 | enum ctx_state prev_ctx; |
| 3562 | |
| 3563 | if (likely(!preemptible())) |
| 3564 | return; |
| 3565 | |
| 3566 | do { |
Steven Rostedt | 47252cf | 2016-03-21 11:23:39 -0400 | [diff] [blame] | 3567 | /* |
| 3568 | * Because the function tracer can trace preempt_count_sub() |
| 3569 | * and it also uses preempt_enable/disable_notrace(), if |
| 3570 | * NEED_RESCHED is set, the preempt_enable_notrace() called |
| 3571 | * by the function tracer will call this function again and |
| 3572 | * cause infinite recursion. |
| 3573 | * |
| 3574 | * Preemption must be disabled here before the function |
| 3575 | * tracer can trace. Break up preempt_disable() into two |
| 3576 | * calls. One to disable preemption without fear of being |
| 3577 | * traced. The other to still record the preemption latency, |
| 3578 | * which can also be traced by the function tracer. |
| 3579 | */ |
Peter Zijlstra | 3d8f74d | 2015-09-28 18:09:19 +0200 | [diff] [blame] | 3580 | preempt_disable_notrace(); |
Steven Rostedt | 47252cf | 2016-03-21 11:23:39 -0400 | [diff] [blame] | 3581 | preempt_latency_start(1); |
Oleg Nesterov | 009f60e | 2014-10-05 22:23:22 +0200 | [diff] [blame] | 3582 | /* |
| 3583 | * Needs preempt disabled in case user_exit() is traced |
| 3584 | * and the tracer calls preempt_enable_notrace() causing |
| 3585 | * an infinite recursion. |
| 3586 | */ |
| 3587 | prev_ctx = exception_enter(); |
Peter Zijlstra | fc13aeb | 2015-09-28 18:05:34 +0200 | [diff] [blame] | 3588 | __schedule(true); |
Oleg Nesterov | 009f60e | 2014-10-05 22:23:22 +0200 | [diff] [blame] | 3589 | exception_exit(prev_ctx); |
| 3590 | |
Steven Rostedt | 47252cf | 2016-03-21 11:23:39 -0400 | [diff] [blame] | 3591 | preempt_latency_stop(1); |
Peter Zijlstra | 3d8f74d | 2015-09-28 18:09:19 +0200 | [diff] [blame] | 3592 | preempt_enable_no_resched_notrace(); |
Oleg Nesterov | 009f60e | 2014-10-05 22:23:22 +0200 | [diff] [blame] | 3593 | } while (need_resched()); |
| 3594 | } |
Frederic Weisbecker | 4eaca0a | 2015-06-04 17:39:08 +0200 | [diff] [blame] | 3595 | EXPORT_SYMBOL_GPL(preempt_schedule_notrace); |
Oleg Nesterov | 009f60e | 2014-10-05 22:23:22 +0200 | [diff] [blame] | 3596 | |
Thomas Gleixner | 32e475d | 2013-11-21 12:41:44 +0100 | [diff] [blame] | 3597 | #endif /* CONFIG_PREEMPT */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3598 | |
| 3599 | /* |
Andreas Mohr | 2ed6e34 | 2006-07-10 04:43:52 -0700 | [diff] [blame] | 3600 | * this is the entry point to schedule() from kernel preemption |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3601 | * off of irq context. |
| 3602 | * Note, that this is called and return with irqs disabled. This will |
| 3603 | * protect us against recursive calling from irq. |
| 3604 | */ |
Andi Kleen | 722a9f9 | 2014-05-02 00:44:38 +0200 | [diff] [blame] | 3605 | asmlinkage __visible void __sched preempt_schedule_irq(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3606 | { |
Frederic Weisbecker | b22366c | 2013-02-24 12:59:30 +0100 | [diff] [blame] | 3607 | enum ctx_state prev_state; |
Ingo Molnar | 6478d88 | 2008-01-25 21:08:33 +0100 | [diff] [blame] | 3608 | |
Andreas Mohr | 2ed6e34 | 2006-07-10 04:43:52 -0700 | [diff] [blame] | 3609 | /* Catch callers which need to be fixed */ |
Peter Zijlstra | f27dde8 | 2013-08-14 14:55:31 +0200 | [diff] [blame] | 3610 | BUG_ON(preempt_count() || !irqs_disabled()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3611 | |
Frederic Weisbecker | b22366c | 2013-02-24 12:59:30 +0100 | [diff] [blame] | 3612 | prev_state = exception_enter(); |
| 3613 | |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3614 | do { |
Peter Zijlstra | 3d8f74d | 2015-09-28 18:09:19 +0200 | [diff] [blame] | 3615 | preempt_disable(); |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3616 | local_irq_enable(); |
Peter Zijlstra | fc13aeb | 2015-09-28 18:05:34 +0200 | [diff] [blame] | 3617 | __schedule(true); |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 3618 | local_irq_disable(); |
Peter Zijlstra | 3d8f74d | 2015-09-28 18:09:19 +0200 | [diff] [blame] | 3619 | sched_preempt_enable_no_resched(); |
Lai Jiangshan | 5ed0cec | 2009-03-06 19:40:20 +0800 | [diff] [blame] | 3620 | } while (need_resched()); |
Frederic Weisbecker | b22366c | 2013-02-24 12:59:30 +0100 | [diff] [blame] | 3621 | |
| 3622 | exception_exit(prev_state); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3623 | } |
| 3624 | |
Peter Zijlstra | 63859d4 | 2009-09-15 19:14:42 +0200 | [diff] [blame] | 3625 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
Ingo Molnar | 95cdf3b | 2005-09-10 00:26:11 -0700 | [diff] [blame] | 3626 | void *key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3627 | { |
Peter Zijlstra | 63859d4 | 2009-09-15 19:14:42 +0200 | [diff] [blame] | 3628 | return try_to_wake_up(curr->private, mode, wake_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3629 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3630 | EXPORT_SYMBOL(default_wake_function); |
| 3631 | |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3632 | #ifdef CONFIG_RT_MUTEXES |
| 3633 | |
| 3634 | /* |
| 3635 | * rt_mutex_setprio - set the current priority of a task |
| 3636 | * @p: task |
| 3637 | * @prio: prio value (kernel-internal form) |
| 3638 | * |
| 3639 | * This function changes the 'effective' priority of a task. It does |
| 3640 | * not touch ->normal_prio like __setscheduler(). |
| 3641 | * |
Thomas Gleixner | c365c29 | 2014-02-07 20:58:42 +0100 | [diff] [blame] | 3642 | * Used by the rt_mutex code to implement priority inheritance |
| 3643 | * logic. Call site only calls if the priority of the task changed. |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3644 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 3645 | void rt_mutex_setprio(struct task_struct *p, int prio) |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3646 | { |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 3647 | int oldprio, queued, running, queue_flag = DEQUEUE_SAVE | DEQUEUE_MOVE; |
Thomas Gleixner | 83ab0aa | 2010-02-17 09:05:48 +0100 | [diff] [blame] | 3648 | const struct sched_class *prev_class; |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 3649 | struct rq_flags rf; |
| 3650 | struct rq *rq; |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3651 | |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 3652 | BUG_ON(prio > MAX_PRIO); |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3653 | |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 3654 | rq = __task_rq_lock(p, &rf); |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3655 | |
Thomas Gleixner | 1c4dd99 | 2011-06-06 20:07:38 +0200 | [diff] [blame] | 3656 | /* |
| 3657 | * Idle task boosting is a nono in general. There is one |
| 3658 | * exception, when PREEMPT_RT and NOHZ is active: |
| 3659 | * |
| 3660 | * The idle task calls get_next_timer_interrupt() and holds |
| 3661 | * the timer wheel base->lock on the CPU and another CPU wants |
| 3662 | * to access the timer (probably to cancel it). We can safely |
| 3663 | * ignore the boosting request, as the idle CPU runs this code |
| 3664 | * with interrupts disabled and will complete the lock |
| 3665 | * protected section without being interrupted. So there is no |
| 3666 | * real need to boost. |
| 3667 | */ |
| 3668 | if (unlikely(p == rq->idle)) { |
| 3669 | WARN_ON(p != rq->curr); |
| 3670 | WARN_ON(p->pi_blocked_on); |
| 3671 | goto out_unlock; |
| 3672 | } |
| 3673 | |
Steven Rostedt | a802707 | 2010-09-20 15:13:34 -0400 | [diff] [blame] | 3674 | trace_sched_pi_setprio(p, prio); |
Andrew Morton | d5f9f94 | 2007-05-08 20:27:06 -0700 | [diff] [blame] | 3675 | oldprio = p->prio; |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 3676 | |
| 3677 | if (oldprio == prio) |
| 3678 | queue_flag &= ~DEQUEUE_MOVE; |
| 3679 | |
Thomas Gleixner | 83ab0aa | 2010-02-17 09:05:48 +0100 | [diff] [blame] | 3680 | prev_class = p->sched_class; |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 3681 | queued = task_on_rq_queued(p); |
Dmitry Adamushko | 051a1d1 | 2007-12-18 15:21:13 +0100 | [diff] [blame] | 3682 | running = task_current(rq, p); |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 3683 | if (queued) |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 3684 | dequeue_task(rq, p, queue_flag); |
Hiroshi Shimamoto | 0e1f348 | 2008-03-10 11:01:20 -0700 | [diff] [blame] | 3685 | if (running) |
Kirill Tkhai | f3cd1c4 | 2014-09-12 17:41:40 +0400 | [diff] [blame] | 3686 | put_prev_task(rq, p); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3687 | |
Dario Faggioli | 2d3d891 | 2013-11-07 14:43:44 +0100 | [diff] [blame] | 3688 | /* |
| 3689 | * Boosting condition are: |
| 3690 | * 1. -rt task is running and holds mutex A |
| 3691 | * --> -dl task blocks on mutex A |
| 3692 | * |
| 3693 | * 2. -dl task is running and holds mutex A |
| 3694 | * --> -dl task blocks on mutex A and could preempt the |
| 3695 | * running task |
| 3696 | */ |
| 3697 | if (dl_prio(prio)) { |
Oleg Nesterov | 466af29 | 2014-06-06 18:52:06 +0200 | [diff] [blame] | 3698 | struct task_struct *pi_task = rt_mutex_get_top_task(p); |
| 3699 | if (!dl_prio(p->normal_prio) || |
| 3700 | (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) { |
Dario Faggioli | 2d3d891 | 2013-11-07 14:43:44 +0100 | [diff] [blame] | 3701 | p->dl.dl_boosted = 1; |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 3702 | queue_flag |= ENQUEUE_REPLENISH; |
Dario Faggioli | 2d3d891 | 2013-11-07 14:43:44 +0100 | [diff] [blame] | 3703 | } else |
| 3704 | p->dl.dl_boosted = 0; |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 3705 | p->sched_class = &dl_sched_class; |
Dario Faggioli | 2d3d891 | 2013-11-07 14:43:44 +0100 | [diff] [blame] | 3706 | } else if (rt_prio(prio)) { |
| 3707 | if (dl_prio(oldprio)) |
| 3708 | p->dl.dl_boosted = 0; |
| 3709 | if (oldprio < prio) |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 3710 | queue_flag |= ENQUEUE_HEAD; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3711 | p->sched_class = &rt_sched_class; |
Dario Faggioli | 2d3d891 | 2013-11-07 14:43:44 +0100 | [diff] [blame] | 3712 | } else { |
| 3713 | if (dl_prio(oldprio)) |
| 3714 | p->dl.dl_boosted = 0; |
Brian Silverman | 746db94 | 2015-02-18 16:23:56 -0800 | [diff] [blame] | 3715 | if (rt_prio(oldprio)) |
| 3716 | p->rt.timeout = 0; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3717 | p->sched_class = &fair_sched_class; |
Dario Faggioli | 2d3d891 | 2013-11-07 14:43:44 +0100 | [diff] [blame] | 3718 | } |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3719 | |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3720 | p->prio = prio; |
| 3721 | |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 3722 | if (queued) |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 3723 | enqueue_task(rq, p, queue_flag); |
Vincent Guittot | a399d23 | 2016-09-12 09:47:52 +0200 | [diff] [blame] | 3724 | if (running) |
Peter Zijlstra | b2bf6c3 | 2016-09-20 22:00:38 +0200 | [diff] [blame] | 3725 | set_curr_task(rq, p); |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 3726 | |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 3727 | check_class_changed(rq, p, prev_class, oldprio); |
Thomas Gleixner | 1c4dd99 | 2011-06-06 20:07:38 +0200 | [diff] [blame] | 3728 | out_unlock: |
Peter Zijlstra | 4c9a4bc | 2015-06-11 14:46:39 +0200 | [diff] [blame] | 3729 | preempt_disable(); /* avoid rq from going away on us */ |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 3730 | __task_rq_unlock(rq, &rf); |
Peter Zijlstra | 4c9a4bc | 2015-06-11 14:46:39 +0200 | [diff] [blame] | 3731 | |
| 3732 | balance_callback(rq); |
| 3733 | preempt_enable(); |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3734 | } |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3735 | #endif |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 3736 | |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 3737 | void set_user_nice(struct task_struct *p, long nice) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3738 | { |
Peter Zijlstra | 49bd21e | 2016-09-20 22:06:01 +0200 | [diff] [blame] | 3739 | bool queued, running; |
| 3740 | int old_prio, delta; |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 3741 | struct rq_flags rf; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 3742 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3743 | |
Dongsheng Yang | 75e45d5 | 2014-02-11 15:34:50 +0800 | [diff] [blame] | 3744 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3745 | return; |
| 3746 | /* |
| 3747 | * We have to be careful, if called from sys_setpriority(), |
| 3748 | * the task might be in the middle of scheduling on another CPU. |
| 3749 | */ |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 3750 | rq = task_rq_lock(p, &rf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3751 | /* |
| 3752 | * The RT priorities are set via sched_setscheduler(), but we still |
| 3753 | * allow the 'normal' nice value to be set - but as expected |
| 3754 | * it wont have any effect on scheduling until the task is |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 3755 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3756 | */ |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 3757 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3758 | p->static_prio = NICE_TO_PRIO(nice); |
| 3759 | goto out_unlock; |
| 3760 | } |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 3761 | queued = task_on_rq_queued(p); |
Peter Zijlstra | 49bd21e | 2016-09-20 22:06:01 +0200 | [diff] [blame] | 3762 | running = task_current(rq, p); |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 3763 | if (queued) |
Peter Zijlstra | 1de6444 | 2015-09-30 17:44:13 +0200 | [diff] [blame] | 3764 | dequeue_task(rq, p, DEQUEUE_SAVE); |
Peter Zijlstra | 49bd21e | 2016-09-20 22:06:01 +0200 | [diff] [blame] | 3765 | if (running) |
| 3766 | put_prev_task(rq, p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3767 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3768 | p->static_prio = NICE_TO_PRIO(nice); |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 3769 | set_load_weight(p); |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3770 | old_prio = p->prio; |
| 3771 | p->prio = effective_prio(p); |
| 3772 | delta = p->prio - old_prio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3773 | |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 3774 | if (queued) { |
Peter Zijlstra | 1de6444 | 2015-09-30 17:44:13 +0200 | [diff] [blame] | 3775 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3776 | /* |
Andrew Morton | d5f9f94 | 2007-05-08 20:27:06 -0700 | [diff] [blame] | 3777 | * If the task increased its priority or is running and |
| 3778 | * lowered its priority, then reschedule its CPU: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3779 | */ |
Andrew Morton | d5f9f94 | 2007-05-08 20:27:06 -0700 | [diff] [blame] | 3780 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 3781 | resched_curr(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3782 | } |
Peter Zijlstra | 49bd21e | 2016-09-20 22:06:01 +0200 | [diff] [blame] | 3783 | if (running) |
| 3784 | set_curr_task(rq, p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3785 | out_unlock: |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 3786 | task_rq_unlock(rq, p, &rf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3787 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3788 | EXPORT_SYMBOL(set_user_nice); |
| 3789 | |
Matt Mackall | e43379f | 2005-05-01 08:59:00 -0700 | [diff] [blame] | 3790 | /* |
| 3791 | * can_nice - check if a task can reduce its nice value |
| 3792 | * @p: task |
| 3793 | * @nice: nice value |
| 3794 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 3795 | int can_nice(const struct task_struct *p, const int nice) |
Matt Mackall | e43379f | 2005-05-01 08:59:00 -0700 | [diff] [blame] | 3796 | { |
Matt Mackall | 024f474 | 2005-08-18 11:24:19 -0700 | [diff] [blame] | 3797 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
Dongsheng Yang | 7aa2c01 | 2014-05-08 18:33:49 +0900 | [diff] [blame] | 3798 | int nice_rlim = nice_to_rlimit(nice); |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 3799 | |
Jiri Slaby | 78d7d40 | 2010-03-05 13:42:54 -0800 | [diff] [blame] | 3800 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
Matt Mackall | e43379f | 2005-05-01 08:59:00 -0700 | [diff] [blame] | 3801 | capable(CAP_SYS_NICE)); |
| 3802 | } |
| 3803 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3804 | #ifdef __ARCH_WANT_SYS_NICE |
| 3805 | |
| 3806 | /* |
| 3807 | * sys_nice - change the priority of the current process. |
| 3808 | * @increment: priority increment |
| 3809 | * |
| 3810 | * sys_setpriority is a more generic, but much slower function that |
| 3811 | * does similar things. |
| 3812 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 3813 | SYSCALL_DEFINE1(nice, int, increment) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3814 | { |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 3815 | long nice, retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3816 | |
| 3817 | /* |
| 3818 | * Setpriority might change our priority at the same moment. |
| 3819 | * We don't have to worry. Conceptually one call occurs first |
| 3820 | * and we have a single winner. |
| 3821 | */ |
Dongsheng Yang | a9467fa | 2014-05-08 18:35:15 +0900 | [diff] [blame] | 3822 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); |
Dongsheng Yang | d0ea026 | 2014-01-27 22:00:45 -0500 | [diff] [blame] | 3823 | nice = task_nice(current) + increment; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3824 | |
Dongsheng Yang | a9467fa | 2014-05-08 18:35:15 +0900 | [diff] [blame] | 3825 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); |
Matt Mackall | e43379f | 2005-05-01 08:59:00 -0700 | [diff] [blame] | 3826 | if (increment < 0 && !can_nice(current, nice)) |
| 3827 | return -EPERM; |
| 3828 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3829 | retval = security_task_setnice(current, nice); |
| 3830 | if (retval) |
| 3831 | return retval; |
| 3832 | |
| 3833 | set_user_nice(current, nice); |
| 3834 | return 0; |
| 3835 | } |
| 3836 | |
| 3837 | #endif |
| 3838 | |
| 3839 | /** |
| 3840 | * task_prio - return the priority value of a given task. |
| 3841 | * @p: the task in question. |
| 3842 | * |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 3843 | * Return: The priority value as seen by users in /proc. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3844 | * RT tasks are offset by -200. Normal tasks are centered |
| 3845 | * around 0, value goes from -16 to +15. |
| 3846 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 3847 | int task_prio(const struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3848 | { |
| 3849 | return p->prio - MAX_RT_PRIO; |
| 3850 | } |
| 3851 | |
| 3852 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3853 | * idle_cpu - is a given cpu idle currently? |
| 3854 | * @cpu: the processor in question. |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 3855 | * |
| 3856 | * Return: 1 if the CPU is currently idle. 0 otherwise. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3857 | */ |
| 3858 | int idle_cpu(int cpu) |
| 3859 | { |
Thomas Gleixner | 908a328 | 2011-09-15 15:32:06 +0200 | [diff] [blame] | 3860 | struct rq *rq = cpu_rq(cpu); |
| 3861 | |
| 3862 | if (rq->curr != rq->idle) |
| 3863 | return 0; |
| 3864 | |
| 3865 | if (rq->nr_running) |
| 3866 | return 0; |
| 3867 | |
| 3868 | #ifdef CONFIG_SMP |
| 3869 | if (!llist_empty(&rq->wake_list)) |
| 3870 | return 0; |
| 3871 | #endif |
| 3872 | |
| 3873 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3874 | } |
| 3875 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3876 | /** |
| 3877 | * idle_task - return the idle task for a given cpu. |
| 3878 | * @cpu: the processor in question. |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 3879 | * |
| 3880 | * Return: The idle task for the cpu @cpu. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3881 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 3882 | struct task_struct *idle_task(int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3883 | { |
| 3884 | return cpu_rq(cpu)->idle; |
| 3885 | } |
| 3886 | |
| 3887 | /** |
| 3888 | * find_process_by_pid - find a process with a matching PID value. |
| 3889 | * @pid: the pid in question. |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 3890 | * |
| 3891 | * The task of @pid, if found. %NULL otherwise. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3892 | */ |
Alexey Dobriyan | a995744 | 2007-10-15 17:00:13 +0200 | [diff] [blame] | 3893 | static struct task_struct *find_process_by_pid(pid_t pid) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3894 | { |
Pavel Emelyanov | 228ebcb | 2007-10-18 23:40:16 -0700 | [diff] [blame] | 3895 | return pid ? find_task_by_vpid(pid) : current; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3896 | } |
| 3897 | |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 3898 | /* |
| 3899 | * This function initializes the sched_dl_entity of a newly becoming |
| 3900 | * SCHED_DEADLINE task. |
| 3901 | * |
| 3902 | * Only the static values are considered here, the actual runtime and the |
| 3903 | * absolute deadline will be properly calculated when the task is enqueued |
| 3904 | * for the first time with its new policy. |
| 3905 | */ |
| 3906 | static void |
| 3907 | __setparam_dl(struct task_struct *p, const struct sched_attr *attr) |
| 3908 | { |
| 3909 | struct sched_dl_entity *dl_se = &p->dl; |
| 3910 | |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 3911 | dl_se->dl_runtime = attr->sched_runtime; |
| 3912 | dl_se->dl_deadline = attr->sched_deadline; |
Harald Gustafsson | 755378a | 2013-11-07 14:43:40 +0100 | [diff] [blame] | 3913 | dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline; |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 3914 | dl_se->flags = attr->sched_flags; |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 3915 | dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime); |
Daniel Bristot de Oliveira | 0559ea3 | 2017-05-29 16:24:03 +0200 | [diff] [blame] | 3916 | dl_se->dl_density = to_ratio(dl_se->dl_deadline, dl_se->dl_runtime); |
Peter Zijlstra | 40767b0 | 2015-01-28 15:08:03 +0100 | [diff] [blame] | 3917 | |
| 3918 | /* |
| 3919 | * Changing the parameters of a task is 'tricky' and we're not doing |
| 3920 | * the correct thing -- also see task_dead_dl() and switched_from_dl(). |
| 3921 | * |
| 3922 | * What we SHOULD do is delay the bandwidth release until the 0-lag |
| 3923 | * point. This would include retaining the task_struct until that time |
| 3924 | * and change dl_overflow() to not immediately decrement the current |
| 3925 | * amount. |
| 3926 | * |
| 3927 | * Instead we retain the current runtime/deadline and let the new |
| 3928 | * parameters take effect after the current reservation period lapses. |
| 3929 | * This is safe (albeit pessimistic) because the 0-lag point is always |
| 3930 | * before the current scheduling deadline. |
| 3931 | * |
| 3932 | * We can still have temporary overloads because we do not delay the |
| 3933 | * change in bandwidth until that time; so admission control is |
| 3934 | * not on the safe side. It does however guarantee tasks will never |
| 3935 | * consume more than promised. |
| 3936 | */ |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 3937 | } |
| 3938 | |
Steven Rostedt | c13db6b | 2014-07-23 11:28:26 -0400 | [diff] [blame] | 3939 | /* |
| 3940 | * sched_setparam() passes in -1 for its policy, to let the functions |
| 3941 | * it calls know not to change it. |
| 3942 | */ |
| 3943 | #define SETPARAM_POLICY -1 |
| 3944 | |
Thomas Gleixner | c365c29 | 2014-02-07 20:58:42 +0100 | [diff] [blame] | 3945 | static void __setscheduler_params(struct task_struct *p, |
| 3946 | const struct sched_attr *attr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3947 | { |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 3948 | int policy = attr->sched_policy; |
| 3949 | |
Steven Rostedt | c13db6b | 2014-07-23 11:28:26 -0400 | [diff] [blame] | 3950 | if (policy == SETPARAM_POLICY) |
Peter Zijlstra | 39fd8fd | 2014-01-15 16:33:20 +0100 | [diff] [blame] | 3951 | policy = p->policy; |
| 3952 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3953 | p->policy = policy; |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 3954 | |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 3955 | if (dl_policy(policy)) |
| 3956 | __setparam_dl(p, attr); |
Peter Zijlstra | 39fd8fd | 2014-01-15 16:33:20 +0100 | [diff] [blame] | 3957 | else if (fair_policy(policy)) |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 3958 | p->static_prio = NICE_TO_PRIO(attr->sched_nice); |
| 3959 | |
Peter Zijlstra | 39fd8fd | 2014-01-15 16:33:20 +0100 | [diff] [blame] | 3960 | /* |
| 3961 | * __sched_setscheduler() ensures attr->sched_priority == 0 when |
| 3962 | * !rt_policy. Always setting this ensures that things like |
| 3963 | * getparam()/getattr() don't report silly values for !rt tasks. |
| 3964 | */ |
| 3965 | p->rt_priority = attr->sched_priority; |
Steven Rostedt | 383afd0 | 2014-03-11 19:24:20 -0400 | [diff] [blame] | 3966 | p->normal_prio = normal_prio(p); |
Thomas Gleixner | c365c29 | 2014-02-07 20:58:42 +0100 | [diff] [blame] | 3967 | set_load_weight(p); |
| 3968 | } |
Peter Zijlstra | 39fd8fd | 2014-01-15 16:33:20 +0100 | [diff] [blame] | 3969 | |
Thomas Gleixner | c365c29 | 2014-02-07 20:58:42 +0100 | [diff] [blame] | 3970 | /* Actually do priority change: must hold pi & rq lock. */ |
| 3971 | static void __setscheduler(struct rq *rq, struct task_struct *p, |
Thomas Gleixner | 0782e63 | 2015-05-05 19:49:49 +0200 | [diff] [blame] | 3972 | const struct sched_attr *attr, bool keep_boost) |
Thomas Gleixner | c365c29 | 2014-02-07 20:58:42 +0100 | [diff] [blame] | 3973 | { |
| 3974 | __setscheduler_params(p, attr); |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 3975 | |
Steven Rostedt | 383afd0 | 2014-03-11 19:24:20 -0400 | [diff] [blame] | 3976 | /* |
Thomas Gleixner | 0782e63 | 2015-05-05 19:49:49 +0200 | [diff] [blame] | 3977 | * Keep a potential priority boosting if called from |
| 3978 | * sched_setscheduler(). |
Steven Rostedt | 383afd0 | 2014-03-11 19:24:20 -0400 | [diff] [blame] | 3979 | */ |
Thomas Gleixner | 0782e63 | 2015-05-05 19:49:49 +0200 | [diff] [blame] | 3980 | if (keep_boost) |
| 3981 | p->prio = rt_mutex_get_effective_prio(p, normal_prio(p)); |
| 3982 | else |
| 3983 | p->prio = normal_prio(p); |
Steven Rostedt | 383afd0 | 2014-03-11 19:24:20 -0400 | [diff] [blame] | 3984 | |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 3985 | if (dl_prio(p->prio)) |
| 3986 | p->sched_class = &dl_sched_class; |
| 3987 | else if (rt_prio(p->prio)) |
Peter Zijlstra | ffd44db | 2009-11-10 20:12:01 +0100 | [diff] [blame] | 3988 | p->sched_class = &rt_sched_class; |
| 3989 | else |
| 3990 | p->sched_class = &fair_sched_class; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3991 | } |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 3992 | |
| 3993 | static void |
| 3994 | __getparam_dl(struct task_struct *p, struct sched_attr *attr) |
| 3995 | { |
| 3996 | struct sched_dl_entity *dl_se = &p->dl; |
| 3997 | |
| 3998 | attr->sched_priority = p->rt_priority; |
| 3999 | attr->sched_runtime = dl_se->dl_runtime; |
| 4000 | attr->sched_deadline = dl_se->dl_deadline; |
Harald Gustafsson | 755378a | 2013-11-07 14:43:40 +0100 | [diff] [blame] | 4001 | attr->sched_period = dl_se->dl_period; |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 4002 | attr->sched_flags = dl_se->flags; |
| 4003 | } |
| 4004 | |
| 4005 | /* |
| 4006 | * This function validates the new parameters of a -deadline task. |
| 4007 | * We ask for the deadline not being zero, and greater or equal |
Harald Gustafsson | 755378a | 2013-11-07 14:43:40 +0100 | [diff] [blame] | 4008 | * than the runtime, as well as the period of being zero or |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 4009 | * greater than deadline. Furthermore, we have to be sure that |
Juri Lelli | b0827819 | 2014-05-13 14:11:31 +0200 | [diff] [blame] | 4010 | * user parameters are above the internal resolution of 1us (we |
| 4011 | * check sched_runtime only since it is always the smaller one) and |
| 4012 | * below 2^63 ns (we have to check both sched_deadline and |
| 4013 | * sched_period, as the latter can be zero). |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 4014 | */ |
| 4015 | static bool |
| 4016 | __checkparam_dl(const struct sched_attr *attr) |
| 4017 | { |
Juri Lelli | b0827819 | 2014-05-13 14:11:31 +0200 | [diff] [blame] | 4018 | /* deadline != 0 */ |
| 4019 | if (attr->sched_deadline == 0) |
| 4020 | return false; |
| 4021 | |
| 4022 | /* |
| 4023 | * Since we truncate DL_SCALE bits, make sure we're at least |
| 4024 | * that big. |
| 4025 | */ |
| 4026 | if (attr->sched_runtime < (1ULL << DL_SCALE)) |
| 4027 | return false; |
| 4028 | |
| 4029 | /* |
| 4030 | * Since we use the MSB for wrap-around and sign issues, make |
| 4031 | * sure it's not set (mind that period can be equal to zero). |
| 4032 | */ |
| 4033 | if (attr->sched_deadline & (1ULL << 63) || |
| 4034 | attr->sched_period & (1ULL << 63)) |
| 4035 | return false; |
| 4036 | |
| 4037 | /* runtime <= deadline <= period (if period != 0) */ |
| 4038 | if ((attr->sched_period != 0 && |
| 4039 | attr->sched_period < attr->sched_deadline) || |
| 4040 | attr->sched_deadline < attr->sched_runtime) |
| 4041 | return false; |
| 4042 | |
| 4043 | return true; |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 4044 | } |
| 4045 | |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 4046 | /* |
| 4047 | * check the target process has a UID that matches the current process's |
| 4048 | */ |
| 4049 | static bool check_same_owner(struct task_struct *p) |
| 4050 | { |
| 4051 | const struct cred *cred = current_cred(), *pcred; |
| 4052 | bool match; |
| 4053 | |
| 4054 | rcu_read_lock(); |
| 4055 | pcred = __task_cred(p); |
Eric W. Biederman | 9c806aa | 2012-02-02 18:54:02 -0800 | [diff] [blame] | 4056 | match = (uid_eq(cred->euid, pcred->euid) || |
| 4057 | uid_eq(cred->euid, pcred->uid)); |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 4058 | rcu_read_unlock(); |
| 4059 | return match; |
| 4060 | } |
| 4061 | |
Wanpeng Li | 7538160 | 2014-11-26 08:44:04 +0800 | [diff] [blame] | 4062 | static bool dl_param_changed(struct task_struct *p, |
| 4063 | const struct sched_attr *attr) |
| 4064 | { |
| 4065 | struct sched_dl_entity *dl_se = &p->dl; |
| 4066 | |
| 4067 | if (dl_se->dl_runtime != attr->sched_runtime || |
| 4068 | dl_se->dl_deadline != attr->sched_deadline || |
| 4069 | dl_se->dl_period != attr->sched_period || |
| 4070 | dl_se->flags != attr->sched_flags) |
| 4071 | return true; |
| 4072 | |
| 4073 | return false; |
| 4074 | } |
| 4075 | |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4076 | static int __sched_setscheduler(struct task_struct *p, |
| 4077 | const struct sched_attr *attr, |
Peter Zijlstra | dbc7f06 | 2015-06-11 14:46:38 +0200 | [diff] [blame] | 4078 | bool user, bool pi) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4079 | { |
Steven Rostedt | 383afd0 | 2014-03-11 19:24:20 -0400 | [diff] [blame] | 4080 | int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 : |
| 4081 | MAX_RT_PRIO - 1 - attr->sched_priority; |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 4082 | int retval, oldprio, oldpolicy = -1, queued, running; |
Thomas Gleixner | 0782e63 | 2015-05-05 19:49:49 +0200 | [diff] [blame] | 4083 | int new_effective_prio, policy = attr->sched_policy; |
Thomas Gleixner | 83ab0aa | 2010-02-17 09:05:48 +0100 | [diff] [blame] | 4084 | const struct sched_class *prev_class; |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 4085 | struct rq_flags rf; |
Lennart Poettering | ca94c44 | 2009-06-15 17:17:47 +0200 | [diff] [blame] | 4086 | int reset_on_fork; |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 4087 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE; |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 4088 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4089 | |
Steven Rostedt | 66e5393 | 2006-06-27 02:54:44 -0700 | [diff] [blame] | 4090 | /* may grab non-irq protected spin_locks */ |
| 4091 | BUG_ON(in_interrupt()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4092 | recheck: |
| 4093 | /* double check policy once rq lock held */ |
Lennart Poettering | ca94c44 | 2009-06-15 17:17:47 +0200 | [diff] [blame] | 4094 | if (policy < 0) { |
| 4095 | reset_on_fork = p->sched_reset_on_fork; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4096 | policy = oldpolicy = p->policy; |
Lennart Poettering | ca94c44 | 2009-06-15 17:17:47 +0200 | [diff] [blame] | 4097 | } else { |
Peter Zijlstra | 7479f3c | 2014-01-15 17:05:04 +0100 | [diff] [blame] | 4098 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); |
Lennart Poettering | ca94c44 | 2009-06-15 17:17:47 +0200 | [diff] [blame] | 4099 | |
Henrik Austad | 20f9cd2 | 2015-09-09 17:00:41 +0200 | [diff] [blame] | 4100 | if (!valid_policy(policy)) |
Lennart Poettering | ca94c44 | 2009-06-15 17:17:47 +0200 | [diff] [blame] | 4101 | return -EINVAL; |
| 4102 | } |
| 4103 | |
Peter Zijlstra | 7479f3c | 2014-01-15 17:05:04 +0100 | [diff] [blame] | 4104 | if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK)) |
| 4105 | return -EINVAL; |
| 4106 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4107 | /* |
| 4108 | * Valid priorities for SCHED_FIFO and SCHED_RR are |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4109 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
| 4110 | * SCHED_BATCH and SCHED_IDLE is 0. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4111 | */ |
Peter Zijlstra | 0bb040a | 2014-01-15 17:15:13 +0100 | [diff] [blame] | 4112 | if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) || |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4113 | (!p->mm && attr->sched_priority > MAX_RT_PRIO-1)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4114 | return -EINVAL; |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 4115 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || |
| 4116 | (rt_policy(policy) != (attr->sched_priority != 0))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4117 | return -EINVAL; |
| 4118 | |
Olivier Croquette | 37e4ab3 | 2005-06-25 14:57:32 -0700 | [diff] [blame] | 4119 | /* |
| 4120 | * Allow unprivileged RT tasks to decrease priority: |
| 4121 | */ |
Rusty Russell | 961ccdd | 2008-06-23 13:55:38 +1000 | [diff] [blame] | 4122 | if (user && !capable(CAP_SYS_NICE)) { |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4123 | if (fair_policy(policy)) { |
Dongsheng Yang | d0ea026 | 2014-01-27 22:00:45 -0500 | [diff] [blame] | 4124 | if (attr->sched_nice < task_nice(p) && |
Peter Zijlstra | eaad451 | 2014-01-16 17:54:25 +0100 | [diff] [blame] | 4125 | !can_nice(p, attr->sched_nice)) |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4126 | return -EPERM; |
| 4127 | } |
| 4128 | |
Ingo Molnar | e05606d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4129 | if (rt_policy(policy)) { |
Oleg Nesterov | a44702e | 2010-06-11 01:09:44 +0200 | [diff] [blame] | 4130 | unsigned long rlim_rtprio = |
| 4131 | task_rlimit(p, RLIMIT_RTPRIO); |
Oleg Nesterov | 5fe1d75 | 2006-09-29 02:00:48 -0700 | [diff] [blame] | 4132 | |
Oleg Nesterov | 8dc3e90 | 2006-09-29 02:00:50 -0700 | [diff] [blame] | 4133 | /* can't set/change the rt policy */ |
| 4134 | if (policy != p->policy && !rlim_rtprio) |
| 4135 | return -EPERM; |
| 4136 | |
| 4137 | /* can't increase priority */ |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4138 | if (attr->sched_priority > p->rt_priority && |
| 4139 | attr->sched_priority > rlim_rtprio) |
Oleg Nesterov | 8dc3e90 | 2006-09-29 02:00:50 -0700 | [diff] [blame] | 4140 | return -EPERM; |
| 4141 | } |
Darren Hart | c02aa73 | 2011-02-17 15:37:07 -0800 | [diff] [blame] | 4142 | |
Juri Lelli | d44753b | 2014-03-03 12:09:21 +0100 | [diff] [blame] | 4143 | /* |
| 4144 | * Can't set/change SCHED_DEADLINE policy at all for now |
| 4145 | * (safest behavior); in the future we would like to allow |
| 4146 | * unprivileged DL tasks to increase their relative deadline |
| 4147 | * or reduce their runtime (both ways reducing utilization) |
| 4148 | */ |
| 4149 | if (dl_policy(policy)) |
| 4150 | return -EPERM; |
| 4151 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4152 | /* |
Darren Hart | c02aa73 | 2011-02-17 15:37:07 -0800 | [diff] [blame] | 4153 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
| 4154 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4155 | */ |
Henrik Austad | 20f9cd2 | 2015-09-09 17:00:41 +0200 | [diff] [blame] | 4156 | if (idle_policy(p->policy) && !idle_policy(policy)) { |
Dongsheng Yang | d0ea026 | 2014-01-27 22:00:45 -0500 | [diff] [blame] | 4157 | if (!can_nice(p, task_nice(p))) |
Darren Hart | c02aa73 | 2011-02-17 15:37:07 -0800 | [diff] [blame] | 4158 | return -EPERM; |
| 4159 | } |
Oleg Nesterov | 8dc3e90 | 2006-09-29 02:00:50 -0700 | [diff] [blame] | 4160 | |
Olivier Croquette | 37e4ab3 | 2005-06-25 14:57:32 -0700 | [diff] [blame] | 4161 | /* can't change other user's priorities */ |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 4162 | if (!check_same_owner(p)) |
Olivier Croquette | 37e4ab3 | 2005-06-25 14:57:32 -0700 | [diff] [blame] | 4163 | return -EPERM; |
Lennart Poettering | ca94c44 | 2009-06-15 17:17:47 +0200 | [diff] [blame] | 4164 | |
| 4165 | /* Normal users shall not reset the sched_reset_on_fork flag */ |
| 4166 | if (p->sched_reset_on_fork && !reset_on_fork) |
| 4167 | return -EPERM; |
Olivier Croquette | 37e4ab3 | 2005-06-25 14:57:32 -0700 | [diff] [blame] | 4168 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4169 | |
Jeremy Fitzhardinge | 725aad2 | 2008-08-03 09:33:03 -0700 | [diff] [blame] | 4170 | if (user) { |
KOSAKI Motohiro | b0ae198 | 2010-10-15 04:21:18 +0900 | [diff] [blame] | 4171 | retval = security_task_setscheduler(p); |
Jeremy Fitzhardinge | 725aad2 | 2008-08-03 09:33:03 -0700 | [diff] [blame] | 4172 | if (retval) |
| 4173 | return retval; |
| 4174 | } |
| 4175 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4176 | /* |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 4177 | * make sure no PI-waiters arrive (or leave) while we are |
| 4178 | * changing the priority of the task: |
Peter Zijlstra | 0122ec5 | 2011-04-05 17:23:51 +0200 | [diff] [blame] | 4179 | * |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 4180 | * To be able to change p->policy safely, the appropriate |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4181 | * runqueue lock must be held. |
| 4182 | */ |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 4183 | rq = task_rq_lock(p, &rf); |
Peter Zijlstra | dc61b1d | 2010-06-08 11:40:42 +0200 | [diff] [blame] | 4184 | |
Peter Zijlstra | 34f971f | 2010-09-22 13:53:15 +0200 | [diff] [blame] | 4185 | /* |
| 4186 | * Changing the policy of the stop threads its a very bad idea |
| 4187 | */ |
| 4188 | if (p == rq->stop) { |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 4189 | task_rq_unlock(rq, p, &rf); |
Peter Zijlstra | 34f971f | 2010-09-22 13:53:15 +0200 | [diff] [blame] | 4190 | return -EINVAL; |
| 4191 | } |
| 4192 | |
Dario Faggioli | a51e919 | 2011-03-24 14:00:18 +0100 | [diff] [blame] | 4193 | /* |
Thomas Gleixner | d6b1e91 | 2014-02-07 20:58:40 +0100 | [diff] [blame] | 4194 | * If not changing anything there's no need to proceed further, |
| 4195 | * but store a possible modification of reset_on_fork. |
Dario Faggioli | a51e919 | 2011-03-24 14:00:18 +0100 | [diff] [blame] | 4196 | */ |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4197 | if (unlikely(policy == p->policy)) { |
Dongsheng Yang | d0ea026 | 2014-01-27 22:00:45 -0500 | [diff] [blame] | 4198 | if (fair_policy(policy) && attr->sched_nice != task_nice(p)) |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4199 | goto change; |
| 4200 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) |
| 4201 | goto change; |
Wanpeng Li | 7538160 | 2014-11-26 08:44:04 +0800 | [diff] [blame] | 4202 | if (dl_policy(policy) && dl_param_changed(p, attr)) |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 4203 | goto change; |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4204 | |
Thomas Gleixner | d6b1e91 | 2014-02-07 20:58:40 +0100 | [diff] [blame] | 4205 | p->sched_reset_on_fork = reset_on_fork; |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 4206 | task_rq_unlock(rq, p, &rf); |
Dario Faggioli | a51e919 | 2011-03-24 14:00:18 +0100 | [diff] [blame] | 4207 | return 0; |
| 4208 | } |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4209 | change: |
Dario Faggioli | a51e919 | 2011-03-24 14:00:18 +0100 | [diff] [blame] | 4210 | |
Peter Zijlstra | dc61b1d | 2010-06-08 11:40:42 +0200 | [diff] [blame] | 4211 | if (user) { |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 4212 | #ifdef CONFIG_RT_GROUP_SCHED |
Peter Zijlstra | dc61b1d | 2010-06-08 11:40:42 +0200 | [diff] [blame] | 4213 | /* |
| 4214 | * Do not allow realtime tasks into groups that have no runtime |
| 4215 | * assigned. |
| 4216 | */ |
| 4217 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
Mike Galbraith | f449377 | 2011-01-13 04:54:50 +0100 | [diff] [blame] | 4218 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
| 4219 | !task_group_is_autogroup(task_group(p))) { |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 4220 | task_rq_unlock(rq, p, &rf); |
Peter Zijlstra | dc61b1d | 2010-06-08 11:40:42 +0200 | [diff] [blame] | 4221 | return -EPERM; |
| 4222 | } |
Peter Zijlstra | dc61b1d | 2010-06-08 11:40:42 +0200 | [diff] [blame] | 4223 | #endif |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 4224 | #ifdef CONFIG_SMP |
| 4225 | if (dl_bandwidth_enabled() && dl_policy(policy)) { |
| 4226 | cpumask_t *span = rq->rd->span; |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 4227 | |
| 4228 | /* |
| 4229 | * Don't allow tasks with an affinity mask smaller than |
| 4230 | * the entire root_domain to become SCHED_DEADLINE. We |
| 4231 | * will also fail if there's no bandwidth available. |
| 4232 | */ |
Peter Zijlstra | e4099a5 | 2013-12-17 10:03:34 +0100 | [diff] [blame] | 4233 | if (!cpumask_subset(span, &p->cpus_allowed) || |
| 4234 | rq->rd->dl_bw.bw == 0) { |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 4235 | task_rq_unlock(rq, p, &rf); |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 4236 | return -EPERM; |
| 4237 | } |
| 4238 | } |
| 4239 | #endif |
| 4240 | } |
Peter Zijlstra | dc61b1d | 2010-06-08 11:40:42 +0200 | [diff] [blame] | 4241 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4242 | /* recheck policy now with rq lock held */ |
| 4243 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { |
| 4244 | policy = oldpolicy = -1; |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 4245 | task_rq_unlock(rq, p, &rf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4246 | goto recheck; |
| 4247 | } |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 4248 | |
| 4249 | /* |
| 4250 | * If setscheduling to SCHED_DEADLINE (or changing the parameters |
| 4251 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth |
| 4252 | * is available. |
| 4253 | */ |
Peter Zijlstra | e4099a5 | 2013-12-17 10:03:34 +0100 | [diff] [blame] | 4254 | if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) { |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 4255 | task_rq_unlock(rq, p, &rf); |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 4256 | return -EBUSY; |
| 4257 | } |
| 4258 | |
Thomas Gleixner | c365c29 | 2014-02-07 20:58:42 +0100 | [diff] [blame] | 4259 | p->sched_reset_on_fork = reset_on_fork; |
| 4260 | oldprio = p->prio; |
| 4261 | |
Peter Zijlstra | dbc7f06 | 2015-06-11 14:46:38 +0200 | [diff] [blame] | 4262 | if (pi) { |
| 4263 | /* |
| 4264 | * Take priority boosted tasks into account. If the new |
| 4265 | * effective priority is unchanged, we just store the new |
| 4266 | * normal parameters and do not touch the scheduler class and |
| 4267 | * the runqueue. This will be done when the task deboost |
| 4268 | * itself. |
| 4269 | */ |
| 4270 | new_effective_prio = rt_mutex_get_effective_prio(p, newprio); |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 4271 | if (new_effective_prio == oldprio) |
| 4272 | queue_flags &= ~DEQUEUE_MOVE; |
Thomas Gleixner | c365c29 | 2014-02-07 20:58:42 +0100 | [diff] [blame] | 4273 | } |
| 4274 | |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 4275 | queued = task_on_rq_queued(p); |
Dmitry Adamushko | 051a1d1 | 2007-12-18 15:21:13 +0100 | [diff] [blame] | 4276 | running = task_current(rq, p); |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 4277 | if (queued) |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 4278 | dequeue_task(rq, p, queue_flags); |
Hiroshi Shimamoto | 0e1f348 | 2008-03-10 11:01:20 -0700 | [diff] [blame] | 4279 | if (running) |
Kirill Tkhai | f3cd1c4 | 2014-09-12 17:41:40 +0400 | [diff] [blame] | 4280 | put_prev_task(rq, p); |
Dmitry Adamushko | f6b5320 | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 4281 | |
Thomas Gleixner | 83ab0aa | 2010-02-17 09:05:48 +0100 | [diff] [blame] | 4282 | prev_class = p->sched_class; |
Peter Zijlstra | dbc7f06 | 2015-06-11 14:46:38 +0200 | [diff] [blame] | 4283 | __setscheduler(rq, p, attr, pi); |
Dmitry Adamushko | f6b5320 | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 4284 | |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 4285 | if (queued) { |
Thomas Gleixner | 81a44c5 | 2014-02-07 20:58:41 +0100 | [diff] [blame] | 4286 | /* |
| 4287 | * We enqueue to tail when the priority of a task is |
| 4288 | * increased (user space view). |
| 4289 | */ |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 4290 | if (oldprio < p->prio) |
| 4291 | queue_flags |= ENQUEUE_HEAD; |
Peter Zijlstra | 1de6444 | 2015-09-30 17:44:13 +0200 | [diff] [blame] | 4292 | |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 4293 | enqueue_task(rq, p, queue_flags); |
Thomas Gleixner | 81a44c5 | 2014-02-07 20:58:41 +0100 | [diff] [blame] | 4294 | } |
Vincent Guittot | a399d23 | 2016-09-12 09:47:52 +0200 | [diff] [blame] | 4295 | if (running) |
Peter Zijlstra | b2bf6c3 | 2016-09-20 22:00:38 +0200 | [diff] [blame] | 4296 | set_curr_task(rq, p); |
Steven Rostedt | cb46984 | 2008-01-25 21:08:22 +0100 | [diff] [blame] | 4297 | |
Peter Zijlstra | da7a735 | 2011-01-17 17:03:27 +0100 | [diff] [blame] | 4298 | check_class_changed(rq, p, prev_class, oldprio); |
Peter Zijlstra | 4c9a4bc | 2015-06-11 14:46:39 +0200 | [diff] [blame] | 4299 | preempt_disable(); /* avoid rq from going away on us */ |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 4300 | task_rq_unlock(rq, p, &rf); |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 4301 | |
Peter Zijlstra | dbc7f06 | 2015-06-11 14:46:38 +0200 | [diff] [blame] | 4302 | if (pi) |
| 4303 | rt_mutex_adjust_pi(p); |
Thomas Gleixner | 95e02ca | 2006-06-27 02:55:02 -0700 | [diff] [blame] | 4304 | |
Peter Zijlstra | 4c9a4bc | 2015-06-11 14:46:39 +0200 | [diff] [blame] | 4305 | /* |
| 4306 | * Run balance callbacks after we've adjusted the PI chain. |
| 4307 | */ |
| 4308 | balance_callback(rq); |
| 4309 | preempt_enable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4310 | |
| 4311 | return 0; |
| 4312 | } |
Rusty Russell | 961ccdd | 2008-06-23 13:55:38 +1000 | [diff] [blame] | 4313 | |
Peter Zijlstra | 7479f3c | 2014-01-15 17:05:04 +0100 | [diff] [blame] | 4314 | static int _sched_setscheduler(struct task_struct *p, int policy, |
| 4315 | const struct sched_param *param, bool check) |
| 4316 | { |
| 4317 | struct sched_attr attr = { |
| 4318 | .sched_policy = policy, |
| 4319 | .sched_priority = param->sched_priority, |
| 4320 | .sched_nice = PRIO_TO_NICE(p->static_prio), |
| 4321 | }; |
| 4322 | |
Steven Rostedt | c13db6b | 2014-07-23 11:28:26 -0400 | [diff] [blame] | 4323 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ |
| 4324 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { |
Peter Zijlstra | 7479f3c | 2014-01-15 17:05:04 +0100 | [diff] [blame] | 4325 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
| 4326 | policy &= ~SCHED_RESET_ON_FORK; |
| 4327 | attr.sched_policy = policy; |
| 4328 | } |
| 4329 | |
Peter Zijlstra | dbc7f06 | 2015-06-11 14:46:38 +0200 | [diff] [blame] | 4330 | return __sched_setscheduler(p, &attr, check, true); |
Peter Zijlstra | 7479f3c | 2014-01-15 17:05:04 +0100 | [diff] [blame] | 4331 | } |
Rusty Russell | 961ccdd | 2008-06-23 13:55:38 +1000 | [diff] [blame] | 4332 | /** |
| 4333 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
| 4334 | * @p: the task in question. |
| 4335 | * @policy: new policy. |
| 4336 | * @param: structure containing the new RT priority. |
| 4337 | * |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 4338 | * Return: 0 on success. An error code otherwise. |
| 4339 | * |
Rusty Russell | 961ccdd | 2008-06-23 13:55:38 +1000 | [diff] [blame] | 4340 | * NOTE that the task may be already dead. |
| 4341 | */ |
| 4342 | int sched_setscheduler(struct task_struct *p, int policy, |
KOSAKI Motohiro | fe7de49 | 2010-10-20 16:01:12 -0700 | [diff] [blame] | 4343 | const struct sched_param *param) |
Rusty Russell | 961ccdd | 2008-06-23 13:55:38 +1000 | [diff] [blame] | 4344 | { |
Peter Zijlstra | 7479f3c | 2014-01-15 17:05:04 +0100 | [diff] [blame] | 4345 | return _sched_setscheduler(p, policy, param, true); |
Rusty Russell | 961ccdd | 2008-06-23 13:55:38 +1000 | [diff] [blame] | 4346 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4347 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
| 4348 | |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4349 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) |
| 4350 | { |
Peter Zijlstra | dbc7f06 | 2015-06-11 14:46:38 +0200 | [diff] [blame] | 4351 | return __sched_setscheduler(p, attr, true, true); |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4352 | } |
| 4353 | EXPORT_SYMBOL_GPL(sched_setattr); |
| 4354 | |
Rusty Russell | 961ccdd | 2008-06-23 13:55:38 +1000 | [diff] [blame] | 4355 | /** |
| 4356 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. |
| 4357 | * @p: the task in question. |
| 4358 | * @policy: new policy. |
| 4359 | * @param: structure containing the new RT priority. |
| 4360 | * |
| 4361 | * Just like sched_setscheduler, only don't bother checking if the |
| 4362 | * current context has permission. For example, this is needed in |
| 4363 | * stop_machine(): we create temporary high priority worker threads, |
| 4364 | * but our caller might not have that capability. |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 4365 | * |
| 4366 | * Return: 0 on success. An error code otherwise. |
Rusty Russell | 961ccdd | 2008-06-23 13:55:38 +1000 | [diff] [blame] | 4367 | */ |
| 4368 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, |
KOSAKI Motohiro | fe7de49 | 2010-10-20 16:01:12 -0700 | [diff] [blame] | 4369 | const struct sched_param *param) |
Rusty Russell | 961ccdd | 2008-06-23 13:55:38 +1000 | [diff] [blame] | 4370 | { |
Peter Zijlstra | 7479f3c | 2014-01-15 17:05:04 +0100 | [diff] [blame] | 4371 | return _sched_setscheduler(p, policy, param, false); |
Rusty Russell | 961ccdd | 2008-06-23 13:55:38 +1000 | [diff] [blame] | 4372 | } |
Davidlohr Bueso | 8477847 | 2015-09-02 01:28:44 -0700 | [diff] [blame] | 4373 | EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck); |
Rusty Russell | 961ccdd | 2008-06-23 13:55:38 +1000 | [diff] [blame] | 4374 | |
Ingo Molnar | 95cdf3b | 2005-09-10 00:26:11 -0700 | [diff] [blame] | 4375 | static int |
| 4376 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4377 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4378 | struct sched_param lparam; |
| 4379 | struct task_struct *p; |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4380 | int retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4381 | |
| 4382 | if (!param || pid < 0) |
| 4383 | return -EINVAL; |
| 4384 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) |
| 4385 | return -EFAULT; |
Oleg Nesterov | 5fe1d75 | 2006-09-29 02:00:48 -0700 | [diff] [blame] | 4386 | |
| 4387 | rcu_read_lock(); |
| 4388 | retval = -ESRCH; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4389 | p = find_process_by_pid(pid); |
Oleg Nesterov | 5fe1d75 | 2006-09-29 02:00:48 -0700 | [diff] [blame] | 4390 | if (p != NULL) |
| 4391 | retval = sched_setscheduler(p, policy, &lparam); |
| 4392 | rcu_read_unlock(); |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4393 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4394 | return retval; |
| 4395 | } |
| 4396 | |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4397 | /* |
| 4398 | * Mimics kernel/events/core.c perf_copy_attr(). |
| 4399 | */ |
| 4400 | static int sched_copy_attr(struct sched_attr __user *uattr, |
| 4401 | struct sched_attr *attr) |
| 4402 | { |
| 4403 | u32 size; |
| 4404 | int ret; |
| 4405 | |
| 4406 | if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0)) |
| 4407 | return -EFAULT; |
| 4408 | |
| 4409 | /* |
| 4410 | * zero the full structure, so that a short copy will be nice. |
| 4411 | */ |
| 4412 | memset(attr, 0, sizeof(*attr)); |
| 4413 | |
| 4414 | ret = get_user(size, &uattr->size); |
| 4415 | if (ret) |
| 4416 | return ret; |
| 4417 | |
| 4418 | if (size > PAGE_SIZE) /* silly large */ |
| 4419 | goto err_size; |
| 4420 | |
| 4421 | if (!size) /* abi compat */ |
| 4422 | size = SCHED_ATTR_SIZE_VER0; |
| 4423 | |
| 4424 | if (size < SCHED_ATTR_SIZE_VER0) |
| 4425 | goto err_size; |
| 4426 | |
| 4427 | /* |
| 4428 | * If we're handed a bigger struct than we know of, |
| 4429 | * ensure all the unknown bits are 0 - i.e. new |
| 4430 | * user-space does not rely on any kernel feature |
| 4431 | * extensions we dont know about yet. |
| 4432 | */ |
| 4433 | if (size > sizeof(*attr)) { |
| 4434 | unsigned char __user *addr; |
| 4435 | unsigned char __user *end; |
| 4436 | unsigned char val; |
| 4437 | |
| 4438 | addr = (void __user *)uattr + sizeof(*attr); |
| 4439 | end = (void __user *)uattr + size; |
| 4440 | |
| 4441 | for (; addr < end; addr++) { |
| 4442 | ret = get_user(val, addr); |
| 4443 | if (ret) |
| 4444 | return ret; |
| 4445 | if (val) |
| 4446 | goto err_size; |
| 4447 | } |
| 4448 | size = sizeof(*attr); |
| 4449 | } |
| 4450 | |
| 4451 | ret = copy_from_user(attr, uattr, size); |
| 4452 | if (ret) |
| 4453 | return -EFAULT; |
| 4454 | |
| 4455 | /* |
| 4456 | * XXX: do we want to be lenient like existing syscalls; or do we want |
| 4457 | * to be strict and return an error on out-of-bounds values? |
| 4458 | */ |
Dongsheng Yang | 75e45d5 | 2014-02-11 15:34:50 +0800 | [diff] [blame] | 4459 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4460 | |
Michael Kerrisk | e78c7bc | 2014-05-09 16:54:28 +0200 | [diff] [blame] | 4461 | return 0; |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4462 | |
| 4463 | err_size: |
| 4464 | put_user(sizeof(*attr), &uattr->size); |
Michael Kerrisk | e78c7bc | 2014-05-09 16:54:28 +0200 | [diff] [blame] | 4465 | return -E2BIG; |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4466 | } |
| 4467 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4468 | /** |
| 4469 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority |
| 4470 | * @pid: the pid in question. |
| 4471 | * @policy: new policy. |
| 4472 | * @param: structure containing the new RT priority. |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 4473 | * |
| 4474 | * Return: 0 on success. An error code otherwise. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4475 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 4476 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
| 4477 | struct sched_param __user *, param) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4478 | { |
Jason Baron | c21761f | 2006-01-18 17:43:03 -0800 | [diff] [blame] | 4479 | /* negative values for policy are not valid */ |
| 4480 | if (policy < 0) |
| 4481 | return -EINVAL; |
| 4482 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4483 | return do_sched_setscheduler(pid, policy, param); |
| 4484 | } |
| 4485 | |
| 4486 | /** |
| 4487 | * sys_sched_setparam - set/change the RT priority of a thread |
| 4488 | * @pid: the pid in question. |
| 4489 | * @param: structure containing the new RT priority. |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 4490 | * |
| 4491 | * Return: 0 on success. An error code otherwise. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4492 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 4493 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4494 | { |
Steven Rostedt | c13db6b | 2014-07-23 11:28:26 -0400 | [diff] [blame] | 4495 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4496 | } |
| 4497 | |
| 4498 | /** |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4499 | * sys_sched_setattr - same as above, but with extended sched_attr |
| 4500 | * @pid: the pid in question. |
Juri Lelli | 5778fcc | 2014-01-14 16:10:39 +0100 | [diff] [blame] | 4501 | * @uattr: structure containing the extended parameters. |
Masanari Iida | db66d75 | 2014-04-18 01:59:15 +0900 | [diff] [blame] | 4502 | * @flags: for future extension. |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4503 | */ |
Peter Zijlstra | 6d35ab4 | 2014-02-14 17:19:29 +0100 | [diff] [blame] | 4504 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, |
| 4505 | unsigned int, flags) |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4506 | { |
| 4507 | struct sched_attr attr; |
| 4508 | struct task_struct *p; |
| 4509 | int retval; |
| 4510 | |
Peter Zijlstra | 6d35ab4 | 2014-02-14 17:19:29 +0100 | [diff] [blame] | 4511 | if (!uattr || pid < 0 || flags) |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4512 | return -EINVAL; |
| 4513 | |
Michael Kerrisk | 143cf23 | 2014-05-09 16:54:15 +0200 | [diff] [blame] | 4514 | retval = sched_copy_attr(uattr, &attr); |
| 4515 | if (retval) |
| 4516 | return retval; |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4517 | |
Richard Weinberger | b14ed2c | 2014-06-02 22:38:34 +0200 | [diff] [blame] | 4518 | if ((int)attr.sched_policy < 0) |
Peter Zijlstra | dbdb227 | 2014-05-09 10:49:03 +0200 | [diff] [blame] | 4519 | return -EINVAL; |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4520 | |
| 4521 | rcu_read_lock(); |
| 4522 | retval = -ESRCH; |
| 4523 | p = find_process_by_pid(pid); |
| 4524 | if (p != NULL) |
| 4525 | retval = sched_setattr(p, &attr); |
| 4526 | rcu_read_unlock(); |
| 4527 | |
| 4528 | return retval; |
| 4529 | } |
| 4530 | |
| 4531 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4532 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread |
| 4533 | * @pid: the pid in question. |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 4534 | * |
| 4535 | * Return: On success, the policy of the thread. Otherwise, a negative error |
| 4536 | * code. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4537 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 4538 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4539 | { |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4540 | struct task_struct *p; |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 4541 | int retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4542 | |
| 4543 | if (pid < 0) |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 4544 | return -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4545 | |
| 4546 | retval = -ESRCH; |
Thomas Gleixner | 5fe85be | 2009-12-09 10:14:58 +0000 | [diff] [blame] | 4547 | rcu_read_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4548 | p = find_process_by_pid(pid); |
| 4549 | if (p) { |
| 4550 | retval = security_task_getscheduler(p); |
| 4551 | if (!retval) |
Lennart Poettering | ca94c44 | 2009-06-15 17:17:47 +0200 | [diff] [blame] | 4552 | retval = p->policy |
| 4553 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4554 | } |
Thomas Gleixner | 5fe85be | 2009-12-09 10:14:58 +0000 | [diff] [blame] | 4555 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4556 | return retval; |
| 4557 | } |
| 4558 | |
| 4559 | /** |
Lennart Poettering | ca94c44 | 2009-06-15 17:17:47 +0200 | [diff] [blame] | 4560 | * sys_sched_getparam - get the RT priority of a thread |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4561 | * @pid: the pid in question. |
| 4562 | * @param: structure containing the RT priority. |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 4563 | * |
| 4564 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error |
| 4565 | * code. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4566 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 4567 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4568 | { |
Peter Zijlstra | ce5f7f8 | 2014-05-12 22:50:34 +0200 | [diff] [blame] | 4569 | struct sched_param lp = { .sched_priority = 0 }; |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4570 | struct task_struct *p; |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 4571 | int retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4572 | |
| 4573 | if (!param || pid < 0) |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 4574 | return -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4575 | |
Thomas Gleixner | 5fe85be | 2009-12-09 10:14:58 +0000 | [diff] [blame] | 4576 | rcu_read_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4577 | p = find_process_by_pid(pid); |
| 4578 | retval = -ESRCH; |
| 4579 | if (!p) |
| 4580 | goto out_unlock; |
| 4581 | |
| 4582 | retval = security_task_getscheduler(p); |
| 4583 | if (retval) |
| 4584 | goto out_unlock; |
| 4585 | |
Peter Zijlstra | ce5f7f8 | 2014-05-12 22:50:34 +0200 | [diff] [blame] | 4586 | if (task_has_rt_policy(p)) |
| 4587 | lp.sched_priority = p->rt_priority; |
Thomas Gleixner | 5fe85be | 2009-12-09 10:14:58 +0000 | [diff] [blame] | 4588 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4589 | |
| 4590 | /* |
| 4591 | * This one might sleep, we cannot do it with a spinlock held ... |
| 4592 | */ |
| 4593 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; |
| 4594 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4595 | return retval; |
| 4596 | |
| 4597 | out_unlock: |
Thomas Gleixner | 5fe85be | 2009-12-09 10:14:58 +0000 | [diff] [blame] | 4598 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4599 | return retval; |
| 4600 | } |
| 4601 | |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4602 | static int sched_read_attr(struct sched_attr __user *uattr, |
| 4603 | struct sched_attr *attr, |
| 4604 | unsigned int usize) |
| 4605 | { |
| 4606 | int ret; |
| 4607 | |
| 4608 | if (!access_ok(VERIFY_WRITE, uattr, usize)) |
| 4609 | return -EFAULT; |
| 4610 | |
| 4611 | /* |
| 4612 | * If we're handed a smaller struct than we know of, |
| 4613 | * ensure all the unknown bits are 0 - i.e. old |
| 4614 | * user-space does not get uncomplete information. |
| 4615 | */ |
| 4616 | if (usize < sizeof(*attr)) { |
| 4617 | unsigned char *addr; |
| 4618 | unsigned char *end; |
| 4619 | |
| 4620 | addr = (void *)attr + usize; |
| 4621 | end = (void *)attr + sizeof(*attr); |
| 4622 | |
| 4623 | for (; addr < end; addr++) { |
| 4624 | if (*addr) |
Michael Kerrisk | 2240067 | 2014-05-09 16:54:33 +0200 | [diff] [blame] | 4625 | return -EFBIG; |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4626 | } |
| 4627 | |
| 4628 | attr->size = usize; |
| 4629 | } |
| 4630 | |
Vegard Nossum | 4efbc45 | 2014-02-16 22:24:17 +0100 | [diff] [blame] | 4631 | ret = copy_to_user(uattr, attr, attr->size); |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4632 | if (ret) |
| 4633 | return -EFAULT; |
| 4634 | |
Michael Kerrisk | 2240067 | 2014-05-09 16:54:33 +0200 | [diff] [blame] | 4635 | return 0; |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4636 | } |
| 4637 | |
| 4638 | /** |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 4639 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4640 | * @pid: the pid in question. |
Juri Lelli | 5778fcc | 2014-01-14 16:10:39 +0100 | [diff] [blame] | 4641 | * @uattr: structure containing the extended parameters. |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4642 | * @size: sizeof(attr) for fwd/bwd comp. |
Masanari Iida | db66d75 | 2014-04-18 01:59:15 +0900 | [diff] [blame] | 4643 | * @flags: for future extension. |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4644 | */ |
Peter Zijlstra | 6d35ab4 | 2014-02-14 17:19:29 +0100 | [diff] [blame] | 4645 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, |
| 4646 | unsigned int, size, unsigned int, flags) |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4647 | { |
| 4648 | struct sched_attr attr = { |
| 4649 | .size = sizeof(struct sched_attr), |
| 4650 | }; |
| 4651 | struct task_struct *p; |
| 4652 | int retval; |
| 4653 | |
| 4654 | if (!uattr || pid < 0 || size > PAGE_SIZE || |
Peter Zijlstra | 6d35ab4 | 2014-02-14 17:19:29 +0100 | [diff] [blame] | 4655 | size < SCHED_ATTR_SIZE_VER0 || flags) |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4656 | return -EINVAL; |
| 4657 | |
| 4658 | rcu_read_lock(); |
| 4659 | p = find_process_by_pid(pid); |
| 4660 | retval = -ESRCH; |
| 4661 | if (!p) |
| 4662 | goto out_unlock; |
| 4663 | |
| 4664 | retval = security_task_getscheduler(p); |
| 4665 | if (retval) |
| 4666 | goto out_unlock; |
| 4667 | |
| 4668 | attr.sched_policy = p->policy; |
Peter Zijlstra | 7479f3c | 2014-01-15 17:05:04 +0100 | [diff] [blame] | 4669 | if (p->sched_reset_on_fork) |
| 4670 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 4671 | if (task_has_dl_policy(p)) |
| 4672 | __getparam_dl(p, &attr); |
| 4673 | else if (task_has_rt_policy(p)) |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4674 | attr.sched_priority = p->rt_priority; |
| 4675 | else |
Dongsheng Yang | d0ea026 | 2014-01-27 22:00:45 -0500 | [diff] [blame] | 4676 | attr.sched_nice = task_nice(p); |
Dario Faggioli | d50dde5 | 2013-11-07 14:43:36 +0100 | [diff] [blame] | 4677 | |
| 4678 | rcu_read_unlock(); |
| 4679 | |
| 4680 | retval = sched_read_attr(uattr, &attr, size); |
| 4681 | return retval; |
| 4682 | |
| 4683 | out_unlock: |
| 4684 | rcu_read_unlock(); |
| 4685 | return retval; |
| 4686 | } |
| 4687 | |
Rusty Russell | 96f874e | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 4688 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4689 | { |
Rusty Russell | 5a16f3d | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4690 | cpumask_var_t cpus_allowed, new_mask; |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4691 | struct task_struct *p; |
| 4692 | int retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4693 | |
Thomas Gleixner | 23f5d14 | 2009-12-09 10:15:01 +0000 | [diff] [blame] | 4694 | rcu_read_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4695 | |
| 4696 | p = find_process_by_pid(pid); |
| 4697 | if (!p) { |
Thomas Gleixner | 23f5d14 | 2009-12-09 10:15:01 +0000 | [diff] [blame] | 4698 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4699 | return -ESRCH; |
| 4700 | } |
| 4701 | |
Thomas Gleixner | 23f5d14 | 2009-12-09 10:15:01 +0000 | [diff] [blame] | 4702 | /* Prevent p going away */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4703 | get_task_struct(p); |
Thomas Gleixner | 23f5d14 | 2009-12-09 10:15:01 +0000 | [diff] [blame] | 4704 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4705 | |
Tejun Heo | 14a40ff | 2013-03-19 13:45:20 -0700 | [diff] [blame] | 4706 | if (p->flags & PF_NO_SETAFFINITY) { |
| 4707 | retval = -EINVAL; |
| 4708 | goto out_put_task; |
| 4709 | } |
Rusty Russell | 5a16f3d | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4710 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
| 4711 | retval = -ENOMEM; |
| 4712 | goto out_put_task; |
| 4713 | } |
| 4714 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { |
| 4715 | retval = -ENOMEM; |
| 4716 | goto out_free_cpus_allowed; |
| 4717 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4718 | retval = -EPERM; |
Eric W. Biederman | 4c44aaa | 2012-07-26 05:05:21 -0700 | [diff] [blame] | 4719 | if (!check_same_owner(p)) { |
| 4720 | rcu_read_lock(); |
| 4721 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { |
| 4722 | rcu_read_unlock(); |
Kirill Tkhai | 16303ab | 2014-09-22 22:36:30 +0400 | [diff] [blame] | 4723 | goto out_free_new_mask; |
Eric W. Biederman | 4c44aaa | 2012-07-26 05:05:21 -0700 | [diff] [blame] | 4724 | } |
| 4725 | rcu_read_unlock(); |
| 4726 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4727 | |
KOSAKI Motohiro | b0ae198 | 2010-10-15 04:21:18 +0900 | [diff] [blame] | 4728 | retval = security_task_setscheduler(p); |
David Quigley | e7834f8 | 2006-06-23 02:03:59 -0700 | [diff] [blame] | 4729 | if (retval) |
Kirill Tkhai | 16303ab | 2014-09-22 22:36:30 +0400 | [diff] [blame] | 4730 | goto out_free_new_mask; |
David Quigley | e7834f8 | 2006-06-23 02:03:59 -0700 | [diff] [blame] | 4731 | |
Peter Zijlstra | e4099a5 | 2013-12-17 10:03:34 +0100 | [diff] [blame] | 4732 | |
| 4733 | cpuset_cpus_allowed(p, cpus_allowed); |
| 4734 | cpumask_and(new_mask, in_mask, cpus_allowed); |
| 4735 | |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 4736 | /* |
| 4737 | * Since bandwidth control happens on root_domain basis, |
| 4738 | * if admission test is enabled, we only admit -deadline |
| 4739 | * tasks allowed to run on all the CPUs in the task's |
| 4740 | * root_domain. |
| 4741 | */ |
| 4742 | #ifdef CONFIG_SMP |
Kirill Tkhai | f1e3a09 | 2014-09-22 22:36:36 +0400 | [diff] [blame] | 4743 | if (task_has_dl_policy(p) && dl_bandwidth_enabled()) { |
| 4744 | rcu_read_lock(); |
| 4745 | if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) { |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 4746 | retval = -EBUSY; |
Kirill Tkhai | f1e3a09 | 2014-09-22 22:36:36 +0400 | [diff] [blame] | 4747 | rcu_read_unlock(); |
Kirill Tkhai | 16303ab | 2014-09-22 22:36:30 +0400 | [diff] [blame] | 4748 | goto out_free_new_mask; |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 4749 | } |
Kirill Tkhai | f1e3a09 | 2014-09-22 22:36:36 +0400 | [diff] [blame] | 4750 | rcu_read_unlock(); |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 4751 | } |
| 4752 | #endif |
Peter Zijlstra | 4924627 | 2010-10-17 21:46:10 +0200 | [diff] [blame] | 4753 | again: |
Peter Zijlstra | 25834c7 | 2015-05-15 17:43:34 +0200 | [diff] [blame] | 4754 | retval = __set_cpus_allowed_ptr(p, new_mask, true); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4755 | |
Paul Menage | 8707d8b | 2007-10-18 23:40:22 -0700 | [diff] [blame] | 4756 | if (!retval) { |
Rusty Russell | 5a16f3d | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4757 | cpuset_cpus_allowed(p, cpus_allowed); |
| 4758 | if (!cpumask_subset(new_mask, cpus_allowed)) { |
Paul Menage | 8707d8b | 2007-10-18 23:40:22 -0700 | [diff] [blame] | 4759 | /* |
| 4760 | * We must have raced with a concurrent cpuset |
| 4761 | * update. Just reset the cpus_allowed to the |
| 4762 | * cpuset's cpus_allowed |
| 4763 | */ |
Rusty Russell | 5a16f3d | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4764 | cpumask_copy(new_mask, cpus_allowed); |
Paul Menage | 8707d8b | 2007-10-18 23:40:22 -0700 | [diff] [blame] | 4765 | goto again; |
| 4766 | } |
| 4767 | } |
Kirill Tkhai | 16303ab | 2014-09-22 22:36:30 +0400 | [diff] [blame] | 4768 | out_free_new_mask: |
Rusty Russell | 5a16f3d | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4769 | free_cpumask_var(new_mask); |
| 4770 | out_free_cpus_allowed: |
| 4771 | free_cpumask_var(cpus_allowed); |
| 4772 | out_put_task: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4773 | put_task_struct(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4774 | return retval; |
| 4775 | } |
| 4776 | |
| 4777 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, |
Rusty Russell | 96f874e | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 4778 | struct cpumask *new_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4779 | { |
Rusty Russell | 96f874e | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 4780 | if (len < cpumask_size()) |
| 4781 | cpumask_clear(new_mask); |
| 4782 | else if (len > cpumask_size()) |
| 4783 | len = cpumask_size(); |
| 4784 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4785 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
| 4786 | } |
| 4787 | |
| 4788 | /** |
| 4789 | * sys_sched_setaffinity - set the cpu affinity of a process |
| 4790 | * @pid: pid of the process |
| 4791 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr |
| 4792 | * @user_mask_ptr: user-space pointer to the new cpu mask |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 4793 | * |
| 4794 | * Return: 0 on success. An error code otherwise. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4795 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 4796 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
| 4797 | unsigned long __user *, user_mask_ptr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4798 | { |
Rusty Russell | 5a16f3d | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4799 | cpumask_var_t new_mask; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4800 | int retval; |
| 4801 | |
Rusty Russell | 5a16f3d | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4802 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
| 4803 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4804 | |
Rusty Russell | 5a16f3d | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4805 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
| 4806 | if (retval == 0) |
| 4807 | retval = sched_setaffinity(pid, new_mask); |
| 4808 | free_cpumask_var(new_mask); |
| 4809 | return retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4810 | } |
| 4811 | |
Rusty Russell | 96f874e | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 4812 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4813 | { |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4814 | struct task_struct *p; |
Thomas Gleixner | 3160568 | 2009-12-08 20:24:16 +0000 | [diff] [blame] | 4815 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4816 | int retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4817 | |
Thomas Gleixner | 23f5d14 | 2009-12-09 10:15:01 +0000 | [diff] [blame] | 4818 | rcu_read_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4819 | |
| 4820 | retval = -ESRCH; |
| 4821 | p = find_process_by_pid(pid); |
| 4822 | if (!p) |
| 4823 | goto out_unlock; |
| 4824 | |
David Quigley | e7834f8 | 2006-06-23 02:03:59 -0700 | [diff] [blame] | 4825 | retval = security_task_getscheduler(p); |
| 4826 | if (retval) |
| 4827 | goto out_unlock; |
| 4828 | |
Peter Zijlstra | 013fdb8 | 2011-04-05 17:23:45 +0200 | [diff] [blame] | 4829 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
Peter Zijlstra | 6acce3e | 2013-10-11 14:38:20 +0200 | [diff] [blame] | 4830 | cpumask_and(mask, &p->cpus_allowed, cpu_active_mask); |
Peter Zijlstra | 013fdb8 | 2011-04-05 17:23:45 +0200 | [diff] [blame] | 4831 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4832 | |
| 4833 | out_unlock: |
Thomas Gleixner | 23f5d14 | 2009-12-09 10:15:01 +0000 | [diff] [blame] | 4834 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4835 | |
Ulrich Drepper | 9531b62 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 4836 | return retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4837 | } |
| 4838 | |
| 4839 | /** |
| 4840 | * sys_sched_getaffinity - get the cpu affinity of a process |
| 4841 | * @pid: pid of the process |
| 4842 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr |
| 4843 | * @user_mask_ptr: user-space pointer to hold the current cpu mask |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 4844 | * |
Zev Weiss | 599b484 | 2016-06-26 16:13:23 -0500 | [diff] [blame] | 4845 | * Return: size of CPU mask copied to user_mask_ptr on success. An |
| 4846 | * error code otherwise. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4847 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 4848 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
| 4849 | unsigned long __user *, user_mask_ptr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4850 | { |
| 4851 | int ret; |
Rusty Russell | f17c860 | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4852 | cpumask_var_t mask; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4853 | |
Anton Blanchard | 84fba5e | 2010-04-06 17:02:19 +1000 | [diff] [blame] | 4854 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
KOSAKI Motohiro | cd3d803 | 2010-03-12 16:15:36 +0900 | [diff] [blame] | 4855 | return -EINVAL; |
| 4856 | if (len & (sizeof(unsigned long)-1)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4857 | return -EINVAL; |
| 4858 | |
Rusty Russell | f17c860 | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4859 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
| 4860 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4861 | |
Rusty Russell | f17c860 | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4862 | ret = sched_getaffinity(pid, mask); |
| 4863 | if (ret == 0) { |
KOSAKI Motohiro | 8bc037f | 2010-03-17 09:36:58 +0900 | [diff] [blame] | 4864 | size_t retlen = min_t(size_t, len, cpumask_size()); |
KOSAKI Motohiro | cd3d803 | 2010-03-12 16:15:36 +0900 | [diff] [blame] | 4865 | |
| 4866 | if (copy_to_user(user_mask_ptr, mask, retlen)) |
Rusty Russell | f17c860 | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4867 | ret = -EFAULT; |
| 4868 | else |
KOSAKI Motohiro | cd3d803 | 2010-03-12 16:15:36 +0900 | [diff] [blame] | 4869 | ret = retlen; |
Rusty Russell | f17c860 | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4870 | } |
| 4871 | free_cpumask_var(mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4872 | |
Rusty Russell | f17c860 | 2008-11-25 02:35:11 +1030 | [diff] [blame] | 4873 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4874 | } |
| 4875 | |
| 4876 | /** |
| 4877 | * sys_sched_yield - yield the current processor to other threads. |
| 4878 | * |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4879 | * This function yields the current CPU to other tasks. If there are no |
| 4880 | * other threads running on this CPU then this function will return. |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 4881 | * |
| 4882 | * Return: 0. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4883 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 4884 | SYSCALL_DEFINE0(sched_yield) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4885 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 4886 | struct rq *rq = this_rq_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4887 | |
Josh Poimboeuf | ae92882 | 2016-06-17 12:43:24 -0500 | [diff] [blame] | 4888 | schedstat_inc(rq->yld_count); |
Dmitry Adamushko | 4530d7a | 2007-10-15 17:00:08 +0200 | [diff] [blame] | 4889 | current->sched_class->yield_task(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4890 | |
| 4891 | /* |
| 4892 | * Since we are going to call schedule() anyway, there's |
| 4893 | * no need to preempt or enable interrupts: |
| 4894 | */ |
| 4895 | __release(rq->lock); |
Ingo Molnar | 8a25d5d | 2006-07-03 00:24:54 -0700 | [diff] [blame] | 4896 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
Thomas Gleixner | 9828ea9 | 2009-12-03 20:55:53 +0100 | [diff] [blame] | 4897 | do_raw_spin_unlock(&rq->lock); |
Thomas Gleixner | ba74c14 | 2011-03-21 13:32:17 +0100 | [diff] [blame] | 4898 | sched_preempt_enable_no_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4899 | |
| 4900 | schedule(); |
| 4901 | |
| 4902 | return 0; |
| 4903 | } |
| 4904 | |
Peter Zijlstra | 35a773a | 2016-09-19 12:57:53 +0200 | [diff] [blame] | 4905 | #ifndef CONFIG_PREEMPT |
Herbert Xu | 02b67cc3 | 2008-01-25 21:08:28 +0100 | [diff] [blame] | 4906 | int __sched _cond_resched(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4907 | { |
Konstantin Khlebnikov | fe32d3c | 2015-07-15 12:52:04 +0300 | [diff] [blame] | 4908 | if (should_resched(0)) { |
Frederic Weisbecker | a18b5d0 | 2015-01-22 18:08:04 +0100 | [diff] [blame] | 4909 | preempt_schedule_common(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4910 | return 1; |
| 4911 | } |
| 4912 | return 0; |
| 4913 | } |
Herbert Xu | 02b67cc3 | 2008-01-25 21:08:28 +0100 | [diff] [blame] | 4914 | EXPORT_SYMBOL(_cond_resched); |
Peter Zijlstra | 35a773a | 2016-09-19 12:57:53 +0200 | [diff] [blame] | 4915 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4916 | |
| 4917 | /* |
Frederic Weisbecker | 613afbf | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 4918 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4919 | * call schedule, and on return reacquire the lock. |
| 4920 | * |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 4921 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4922 | * operations here to prevent schedule() from being called twice (once via |
| 4923 | * spin_unlock(), once by hand). |
| 4924 | */ |
Frederic Weisbecker | 613afbf | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 4925 | int __cond_resched_lock(spinlock_t *lock) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4926 | { |
Konstantin Khlebnikov | fe32d3c | 2015-07-15 12:52:04 +0300 | [diff] [blame] | 4927 | int resched = should_resched(PREEMPT_LOCK_OFFSET); |
Jan Kara | 6df3cec | 2005-06-13 15:52:32 -0700 | [diff] [blame] | 4928 | int ret = 0; |
| 4929 | |
Peter Zijlstra | f607c66 | 2009-07-20 19:16:29 +0200 | [diff] [blame] | 4930 | lockdep_assert_held(lock); |
| 4931 | |
Paul E. McKenney | 4a81e83 | 2014-06-20 16:49:01 -0700 | [diff] [blame] | 4932 | if (spin_needbreak(lock) || resched) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4933 | spin_unlock(lock); |
Peter Zijlstra | d86ee48 | 2009-07-10 14:57:57 +0200 | [diff] [blame] | 4934 | if (resched) |
Frederic Weisbecker | a18b5d0 | 2015-01-22 18:08:04 +0100 | [diff] [blame] | 4935 | preempt_schedule_common(); |
Nick Piggin | 95c354f | 2008-01-30 13:31:20 +0100 | [diff] [blame] | 4936 | else |
| 4937 | cpu_relax(); |
Jan Kara | 6df3cec | 2005-06-13 15:52:32 -0700 | [diff] [blame] | 4938 | ret = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4939 | spin_lock(lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4940 | } |
Jan Kara | 6df3cec | 2005-06-13 15:52:32 -0700 | [diff] [blame] | 4941 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4942 | } |
Frederic Weisbecker | 613afbf | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 4943 | EXPORT_SYMBOL(__cond_resched_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4944 | |
Frederic Weisbecker | 613afbf | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 4945 | int __sched __cond_resched_softirq(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4946 | { |
| 4947 | BUG_ON(!in_softirq()); |
| 4948 | |
Konstantin Khlebnikov | fe32d3c | 2015-07-15 12:52:04 +0300 | [diff] [blame] | 4949 | if (should_resched(SOFTIRQ_DISABLE_OFFSET)) { |
Thomas Gleixner | 98d82567 | 2007-05-23 13:58:18 -0700 | [diff] [blame] | 4950 | local_bh_enable(); |
Frederic Weisbecker | a18b5d0 | 2015-01-22 18:08:04 +0100 | [diff] [blame] | 4951 | preempt_schedule_common(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4952 | local_bh_disable(); |
| 4953 | return 1; |
| 4954 | } |
| 4955 | return 0; |
| 4956 | } |
Frederic Weisbecker | 613afbf | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 4957 | EXPORT_SYMBOL(__cond_resched_softirq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4958 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4959 | /** |
| 4960 | * yield - yield the current processor to other threads. |
| 4961 | * |
Peter Zijlstra | 8e3fabf | 2012-03-06 18:54:26 +0100 | [diff] [blame] | 4962 | * Do not ever use this function, there's a 99% chance you're doing it wrong. |
| 4963 | * |
| 4964 | * The scheduler is at all times free to pick the calling task as the most |
| 4965 | * eligible task to run, if removing the yield() call from your code breaks |
| 4966 | * it, its already broken. |
| 4967 | * |
| 4968 | * Typical broken usage is: |
| 4969 | * |
| 4970 | * while (!event) |
| 4971 | * yield(); |
| 4972 | * |
| 4973 | * where one assumes that yield() will let 'the other' process run that will |
| 4974 | * make event true. If the current task is a SCHED_FIFO task that will never |
| 4975 | * happen. Never use yield() as a progress guarantee!! |
| 4976 | * |
| 4977 | * If you want to use yield() to wait for something, use wait_event(). |
| 4978 | * If you want to use yield() to be 'nice' for others, use cond_resched(). |
| 4979 | * If you still want to use yield(), do not! |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4980 | */ |
| 4981 | void __sched yield(void) |
| 4982 | { |
| 4983 | set_current_state(TASK_RUNNING); |
| 4984 | sys_sched_yield(); |
| 4985 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4986 | EXPORT_SYMBOL(yield); |
| 4987 | |
Mike Galbraith | d95f412 | 2011-02-01 09:50:51 -0500 | [diff] [blame] | 4988 | /** |
| 4989 | * yield_to - yield the current processor to another thread in |
| 4990 | * your thread group, or accelerate that thread toward the |
| 4991 | * processor it's on. |
Randy Dunlap | 16addf9 | 2011-03-18 09:34:53 -0700 | [diff] [blame] | 4992 | * @p: target task |
| 4993 | * @preempt: whether task preemption is allowed or not |
Mike Galbraith | d95f412 | 2011-02-01 09:50:51 -0500 | [diff] [blame] | 4994 | * |
| 4995 | * It's the caller's job to ensure that the target task struct |
| 4996 | * can't go away on us before we can do any checks. |
| 4997 | * |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 4998 | * Return: |
Peter Zijlstra | 7b270f6 | 2013-01-22 13:09:13 +0530 | [diff] [blame] | 4999 | * true (>0) if we indeed boosted the target task. |
| 5000 | * false (0) if we failed to boost the target. |
| 5001 | * -ESRCH if there's no task to yield to. |
Mike Galbraith | d95f412 | 2011-02-01 09:50:51 -0500 | [diff] [blame] | 5002 | */ |
Dan Carpenter | fa93384 | 2014-05-23 13:20:42 +0300 | [diff] [blame] | 5003 | int __sched yield_to(struct task_struct *p, bool preempt) |
Mike Galbraith | d95f412 | 2011-02-01 09:50:51 -0500 | [diff] [blame] | 5004 | { |
| 5005 | struct task_struct *curr = current; |
| 5006 | struct rq *rq, *p_rq; |
| 5007 | unsigned long flags; |
Dan Carpenter | c3c1864 | 2013-02-05 14:37:51 +0300 | [diff] [blame] | 5008 | int yielded = 0; |
Mike Galbraith | d95f412 | 2011-02-01 09:50:51 -0500 | [diff] [blame] | 5009 | |
| 5010 | local_irq_save(flags); |
| 5011 | rq = this_rq(); |
| 5012 | |
| 5013 | again: |
| 5014 | p_rq = task_rq(p); |
Peter Zijlstra | 7b270f6 | 2013-01-22 13:09:13 +0530 | [diff] [blame] | 5015 | /* |
| 5016 | * If we're the only runnable task on the rq and target rq also |
| 5017 | * has only one task, there's absolutely no point in yielding. |
| 5018 | */ |
| 5019 | if (rq->nr_running == 1 && p_rq->nr_running == 1) { |
| 5020 | yielded = -ESRCH; |
| 5021 | goto out_irq; |
| 5022 | } |
| 5023 | |
Mike Galbraith | d95f412 | 2011-02-01 09:50:51 -0500 | [diff] [blame] | 5024 | double_rq_lock(rq, p_rq); |
Shigeru Yoshida | 39e24d8f | 2013-11-23 18:38:01 +0900 | [diff] [blame] | 5025 | if (task_rq(p) != p_rq) { |
Mike Galbraith | d95f412 | 2011-02-01 09:50:51 -0500 | [diff] [blame] | 5026 | double_rq_unlock(rq, p_rq); |
| 5027 | goto again; |
| 5028 | } |
| 5029 | |
| 5030 | if (!curr->sched_class->yield_to_task) |
Peter Zijlstra | 7b270f6 | 2013-01-22 13:09:13 +0530 | [diff] [blame] | 5031 | goto out_unlock; |
Mike Galbraith | d95f412 | 2011-02-01 09:50:51 -0500 | [diff] [blame] | 5032 | |
| 5033 | if (curr->sched_class != p->sched_class) |
Peter Zijlstra | 7b270f6 | 2013-01-22 13:09:13 +0530 | [diff] [blame] | 5034 | goto out_unlock; |
Mike Galbraith | d95f412 | 2011-02-01 09:50:51 -0500 | [diff] [blame] | 5035 | |
| 5036 | if (task_running(p_rq, p) || p->state) |
Peter Zijlstra | 7b270f6 | 2013-01-22 13:09:13 +0530 | [diff] [blame] | 5037 | goto out_unlock; |
Mike Galbraith | d95f412 | 2011-02-01 09:50:51 -0500 | [diff] [blame] | 5038 | |
| 5039 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); |
Venkatesh Pallipadi | 6d1cafd | 2011-03-01 16:28:21 -0800 | [diff] [blame] | 5040 | if (yielded) { |
Josh Poimboeuf | ae92882 | 2016-06-17 12:43:24 -0500 | [diff] [blame] | 5041 | schedstat_inc(rq->yld_count); |
Venkatesh Pallipadi | 6d1cafd | 2011-03-01 16:28:21 -0800 | [diff] [blame] | 5042 | /* |
| 5043 | * Make p's CPU reschedule; pick_next_entity takes care of |
| 5044 | * fairness. |
| 5045 | */ |
| 5046 | if (preempt && rq != p_rq) |
Kirill Tkhai | 8875125 | 2014-06-29 00:03:57 +0400 | [diff] [blame] | 5047 | resched_curr(p_rq); |
Venkatesh Pallipadi | 6d1cafd | 2011-03-01 16:28:21 -0800 | [diff] [blame] | 5048 | } |
Mike Galbraith | d95f412 | 2011-02-01 09:50:51 -0500 | [diff] [blame] | 5049 | |
Peter Zijlstra | 7b270f6 | 2013-01-22 13:09:13 +0530 | [diff] [blame] | 5050 | out_unlock: |
Mike Galbraith | d95f412 | 2011-02-01 09:50:51 -0500 | [diff] [blame] | 5051 | double_rq_unlock(rq, p_rq); |
Peter Zijlstra | 7b270f6 | 2013-01-22 13:09:13 +0530 | [diff] [blame] | 5052 | out_irq: |
Mike Galbraith | d95f412 | 2011-02-01 09:50:51 -0500 | [diff] [blame] | 5053 | local_irq_restore(flags); |
| 5054 | |
Peter Zijlstra | 7b270f6 | 2013-01-22 13:09:13 +0530 | [diff] [blame] | 5055 | if (yielded > 0) |
Mike Galbraith | d95f412 | 2011-02-01 09:50:51 -0500 | [diff] [blame] | 5056 | schedule(); |
| 5057 | |
| 5058 | return yielded; |
| 5059 | } |
| 5060 | EXPORT_SYMBOL_GPL(yield_to); |
| 5061 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5062 | /* |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 5063 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5064 | * that process accounting knows that this is a task in IO wait state. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5065 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5066 | long __sched io_schedule_timeout(long timeout) |
| 5067 | { |
NeilBrown | 9cff8ad | 2015-02-13 15:49:17 +1100 | [diff] [blame] | 5068 | int old_iowait = current->in_iowait; |
| 5069 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5070 | long ret; |
| 5071 | |
Arjan van de Ven | 8f0dfc3 | 2009-07-20 11:26:58 -0700 | [diff] [blame] | 5072 | current->in_iowait = 1; |
Shaohua Li | 10d784e | 2015-05-08 10:51:29 -0700 | [diff] [blame] | 5073 | blk_schedule_flush_plug(current); |
NeilBrown | 9cff8ad | 2015-02-13 15:49:17 +1100 | [diff] [blame] | 5074 | |
| 5075 | delayacct_blkio_start(); |
| 5076 | rq = raw_rq(); |
| 5077 | atomic_inc(&rq->nr_iowait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5078 | ret = schedule_timeout(timeout); |
NeilBrown | 9cff8ad | 2015-02-13 15:49:17 +1100 | [diff] [blame] | 5079 | current->in_iowait = old_iowait; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5080 | atomic_dec(&rq->nr_iowait); |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 5081 | delayacct_blkio_end(); |
NeilBrown | 9cff8ad | 2015-02-13 15:49:17 +1100 | [diff] [blame] | 5082 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5083 | return ret; |
| 5084 | } |
NeilBrown | 9cff8ad | 2015-02-13 15:49:17 +1100 | [diff] [blame] | 5085 | EXPORT_SYMBOL(io_schedule_timeout); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5086 | |
| 5087 | /** |
| 5088 | * sys_sched_get_priority_max - return maximum RT priority. |
| 5089 | * @policy: scheduling class. |
| 5090 | * |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 5091 | * Return: On success, this syscall returns the maximum |
| 5092 | * rt_priority that can be used by a given scheduling class. |
| 5093 | * On failure, a negative error code is returned. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5094 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 5095 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5096 | { |
| 5097 | int ret = -EINVAL; |
| 5098 | |
| 5099 | switch (policy) { |
| 5100 | case SCHED_FIFO: |
| 5101 | case SCHED_RR: |
| 5102 | ret = MAX_USER_RT_PRIO-1; |
| 5103 | break; |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 5104 | case SCHED_DEADLINE: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5105 | case SCHED_NORMAL: |
Ingo Molnar | b0a9499 | 2006-01-14 13:20:41 -0800 | [diff] [blame] | 5106 | case SCHED_BATCH: |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5107 | case SCHED_IDLE: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5108 | ret = 0; |
| 5109 | break; |
| 5110 | } |
| 5111 | return ret; |
| 5112 | } |
| 5113 | |
| 5114 | /** |
| 5115 | * sys_sched_get_priority_min - return minimum RT priority. |
| 5116 | * @policy: scheduling class. |
| 5117 | * |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 5118 | * Return: On success, this syscall returns the minimum |
| 5119 | * rt_priority that can be used by a given scheduling class. |
| 5120 | * On failure, a negative error code is returned. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5121 | */ |
Heiko Carstens | 5add95d | 2009-01-14 14:14:08 +0100 | [diff] [blame] | 5122 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5123 | { |
| 5124 | int ret = -EINVAL; |
| 5125 | |
| 5126 | switch (policy) { |
| 5127 | case SCHED_FIFO: |
| 5128 | case SCHED_RR: |
| 5129 | ret = 1; |
| 5130 | break; |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 5131 | case SCHED_DEADLINE: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5132 | case SCHED_NORMAL: |
Ingo Molnar | b0a9499 | 2006-01-14 13:20:41 -0800 | [diff] [blame] | 5133 | case SCHED_BATCH: |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5134 | case SCHED_IDLE: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5135 | ret = 0; |
| 5136 | } |
| 5137 | return ret; |
| 5138 | } |
| 5139 | |
| 5140 | /** |
| 5141 | * sys_sched_rr_get_interval - return the default timeslice of a process. |
| 5142 | * @pid: pid of the process. |
| 5143 | * @interval: userspace pointer to the timeslice value. |
| 5144 | * |
| 5145 | * this syscall writes the default timeslice value of a given process |
| 5146 | * into the user-space timespec buffer. A value of '0' means infinity. |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 5147 | * |
| 5148 | * Return: On success, 0 and the timeslice is in @interval. Otherwise, |
| 5149 | * an error code. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5150 | */ |
Heiko Carstens | 17da2bd | 2009-01-14 14:14:10 +0100 | [diff] [blame] | 5151 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
Heiko Carstens | 754fe8d | 2009-01-14 14:14:09 +0100 | [diff] [blame] | 5152 | struct timespec __user *, interval) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5153 | { |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 5154 | struct task_struct *p; |
Dmitry Adamushko | a4ec24b | 2007-10-15 17:00:13 +0200 | [diff] [blame] | 5155 | unsigned int time_slice; |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 5156 | struct rq_flags rf; |
| 5157 | struct timespec t; |
Thomas Gleixner | dba091b | 2009-12-09 09:32:03 +0100 | [diff] [blame] | 5158 | struct rq *rq; |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 5159 | int retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5160 | |
| 5161 | if (pid < 0) |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 5162 | return -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5163 | |
| 5164 | retval = -ESRCH; |
Thomas Gleixner | 1a551ae | 2009-12-09 10:15:11 +0000 | [diff] [blame] | 5165 | rcu_read_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5166 | p = find_process_by_pid(pid); |
| 5167 | if (!p) |
| 5168 | goto out_unlock; |
| 5169 | |
| 5170 | retval = security_task_getscheduler(p); |
| 5171 | if (retval) |
| 5172 | goto out_unlock; |
| 5173 | |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 5174 | rq = task_rq_lock(p, &rf); |
Peter Zijlstra | a57beec | 2014-01-27 11:54:13 +0100 | [diff] [blame] | 5175 | time_slice = 0; |
| 5176 | if (p->sched_class->get_rr_interval) |
| 5177 | time_slice = p->sched_class->get_rr_interval(rq, p); |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 5178 | task_rq_unlock(rq, p, &rf); |
Dmitry Adamushko | a4ec24b | 2007-10-15 17:00:13 +0200 | [diff] [blame] | 5179 | |
Thomas Gleixner | 1a551ae | 2009-12-09 10:15:11 +0000 | [diff] [blame] | 5180 | rcu_read_unlock(); |
Dmitry Adamushko | a4ec24b | 2007-10-15 17:00:13 +0200 | [diff] [blame] | 5181 | jiffies_to_timespec(time_slice, &t); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5182 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5183 | return retval; |
Andi Kleen | 3a5c359 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 5184 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5185 | out_unlock: |
Thomas Gleixner | 1a551ae | 2009-12-09 10:15:11 +0000 | [diff] [blame] | 5186 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5187 | return retval; |
| 5188 | } |
| 5189 | |
Steven Rostedt | 7c731e0 | 2008-05-12 21:20:41 +0200 | [diff] [blame] | 5190 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 5191 | |
Ingo Molnar | 82a1fcb | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 5192 | void sched_show_task(struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5193 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5194 | unsigned long free = 0; |
Paul E. McKenney | 4e79752 | 2012-11-07 13:35:32 -0800 | [diff] [blame] | 5195 | int ppid; |
Tetsuo Handa | 1f8a763 | 2014-12-05 21:22:22 +0900 | [diff] [blame] | 5196 | unsigned long state = p->state; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5197 | |
Tetsuo Handa | 3820050 | 2016-11-02 19:50:29 +0900 | [diff] [blame] | 5198 | if (!try_get_task_stack(p)) |
| 5199 | return; |
Tetsuo Handa | 1f8a763 | 2014-12-05 21:22:22 +0900 | [diff] [blame] | 5200 | if (state) |
| 5201 | state = __ffs(state) + 1; |
Erik Gilling | 28d0686 | 2010-11-19 18:08:51 -0800 | [diff] [blame] | 5202 | printk(KERN_INFO "%-15.15s %c", p->comm, |
Andreas Mohr | 2ed6e34 | 2006-07-10 04:43:52 -0700 | [diff] [blame] | 5203 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5204 | if (state == TASK_RUNNING) |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5205 | printk(KERN_CONT " running task "); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5206 | #ifdef CONFIG_DEBUG_STACK_USAGE |
Eric Sandeen | 7c9f886 | 2008-04-22 16:38:23 -0500 | [diff] [blame] | 5207 | free = stack_not_used(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5208 | #endif |
Oleg Nesterov | a90e984 | 2014-12-10 15:45:21 -0800 | [diff] [blame] | 5209 | ppid = 0; |
Paul E. McKenney | 4e79752 | 2012-11-07 13:35:32 -0800 | [diff] [blame] | 5210 | rcu_read_lock(); |
Oleg Nesterov | a90e984 | 2014-12-10 15:45:21 -0800 | [diff] [blame] | 5211 | if (pid_alive(p)) |
| 5212 | ppid = task_pid_nr(rcu_dereference(p->real_parent)); |
Paul E. McKenney | 4e79752 | 2012-11-07 13:35:32 -0800 | [diff] [blame] | 5213 | rcu_read_unlock(); |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5214 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
Paul E. McKenney | 4e79752 | 2012-11-07 13:35:32 -0800 | [diff] [blame] | 5215 | task_pid_nr(p), ppid, |
David Rientjes | aa47b7e | 2009-05-04 01:38:05 -0700 | [diff] [blame] | 5216 | (unsigned long)task_thread_info(p)->flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5217 | |
Tejun Heo | 3d1cb20 | 2013-04-30 15:27:22 -0700 | [diff] [blame] | 5218 | print_worker_info(KERN_INFO, p); |
Nick Piggin | 5fb5e6d | 2008-01-25 21:08:34 +0100 | [diff] [blame] | 5219 | show_stack(p, NULL); |
Tetsuo Handa | 3820050 | 2016-11-02 19:50:29 +0900 | [diff] [blame] | 5220 | put_task_stack(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5221 | } |
| 5222 | |
Ingo Molnar | e59e2ae | 2006-12-06 20:35:59 -0800 | [diff] [blame] | 5223 | void show_state_filter(unsigned long state_filter) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5224 | { |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 5225 | struct task_struct *g, *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5226 | |
Ingo Molnar | 4bd7732 | 2007-07-11 21:21:47 +0200 | [diff] [blame] | 5227 | #if BITS_PER_LONG == 32 |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5228 | printk(KERN_INFO |
| 5229 | " task PC stack pid father\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5230 | #else |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5231 | printk(KERN_INFO |
| 5232 | " task PC stack pid father\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5233 | #endif |
Thomas Gleixner | 510f5ac | 2011-07-17 20:47:54 +0200 | [diff] [blame] | 5234 | rcu_read_lock(); |
Oleg Nesterov | 5d07f42 | 2014-08-13 21:19:53 +0200 | [diff] [blame] | 5235 | for_each_process_thread(g, p) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5236 | /* |
| 5237 | * reset the NMI-timeout, listing all files on a slow |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 5238 | * console might take a lot of time: |
Andrey Ryabinin | 57675cb | 2016-06-09 15:20:05 +0300 | [diff] [blame] | 5239 | * Also, reset softlockup watchdogs on all CPUs, because |
| 5240 | * another CPU might be blocked waiting for us to process |
| 5241 | * an IPI. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5242 | */ |
| 5243 | touch_nmi_watchdog(); |
Andrey Ryabinin | 57675cb | 2016-06-09 15:20:05 +0300 | [diff] [blame] | 5244 | touch_all_softlockup_watchdogs(); |
Ingo Molnar | 39bc89f | 2007-04-25 20:50:03 -0700 | [diff] [blame] | 5245 | if (!state_filter || (p->state & state_filter)) |
Ingo Molnar | 82a1fcb | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 5246 | sched_show_task(p); |
Oleg Nesterov | 5d07f42 | 2014-08-13 21:19:53 +0200 | [diff] [blame] | 5247 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5248 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5249 | #ifdef CONFIG_SCHED_DEBUG |
Rabin Vincent | fb90a6e | 2016-04-04 15:42:02 +0200 | [diff] [blame] | 5250 | if (!state_filter) |
| 5251 | sysrq_sched_debug_show(); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5252 | #endif |
Thomas Gleixner | 510f5ac | 2011-07-17 20:47:54 +0200 | [diff] [blame] | 5253 | rcu_read_unlock(); |
Ingo Molnar | e59e2ae | 2006-12-06 20:35:59 -0800 | [diff] [blame] | 5254 | /* |
| 5255 | * Only show locks if all tasks are dumped: |
| 5256 | */ |
Shmulik Ladkani | 93335a2 | 2009-11-25 15:23:41 +0200 | [diff] [blame] | 5257 | if (!state_filter) |
Ingo Molnar | e59e2ae | 2006-12-06 20:35:59 -0800 | [diff] [blame] | 5258 | debug_show_all_locks(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5259 | } |
| 5260 | |
Paul Gortmaker | 0db0628 | 2013-06-19 14:53:51 -0400 | [diff] [blame] | 5261 | void init_idle_bootup_task(struct task_struct *idle) |
Ingo Molnar | 1df2105 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 5262 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5263 | idle->sched_class = &idle_sched_class; |
Ingo Molnar | 1df2105 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 5264 | } |
| 5265 | |
Ingo Molnar | f340c0d | 2005-06-28 16:40:42 +0200 | [diff] [blame] | 5266 | /** |
| 5267 | * init_idle - set up an idle thread for a given CPU |
| 5268 | * @idle: task in question |
| 5269 | * @cpu: cpu the idle task belongs to |
| 5270 | * |
| 5271 | * NOTE: this function does not set the idle thread's NEED_RESCHED |
| 5272 | * flag, to make booting more robust. |
| 5273 | */ |
Paul Gortmaker | 0db0628 | 2013-06-19 14:53:51 -0400 | [diff] [blame] | 5274 | void init_idle(struct task_struct *idle, int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5275 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 5276 | struct rq *rq = cpu_rq(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5277 | unsigned long flags; |
| 5278 | |
Peter Zijlstra | 25834c7 | 2015-05-15 17:43:34 +0200 | [diff] [blame] | 5279 | raw_spin_lock_irqsave(&idle->pi_lock, flags); |
| 5280 | raw_spin_lock(&rq->lock); |
Ingo Molnar | 5cbd54e | 2008-11-12 20:05:50 +0100 | [diff] [blame] | 5281 | |
Rik van Riel | 5e1576e | 2013-10-07 11:29:26 +0100 | [diff] [blame] | 5282 | __sched_fork(0, idle); |
Peter Zijlstra | 06b83b5 | 2009-12-16 18:04:35 +0100 | [diff] [blame] | 5283 | idle->state = TASK_RUNNING; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5284 | idle->se.exec_start = sched_clock(); |
| 5285 | |
Mark Rutland | e1b77c9 | 2016-03-09 14:08:18 -0800 | [diff] [blame] | 5286 | kasan_unpoison_task_stack(idle); |
| 5287 | |
Peter Zijlstra | de9b8f5 | 2015-08-13 23:09:29 +0200 | [diff] [blame] | 5288 | #ifdef CONFIG_SMP |
| 5289 | /* |
| 5290 | * Its possible that init_idle() gets called multiple times on a task, |
| 5291 | * in that case do_set_cpus_allowed() will not do the right thing. |
| 5292 | * |
| 5293 | * And since this is boot we can forgo the serialization. |
| 5294 | */ |
| 5295 | set_cpus_allowed_common(idle, cpumask_of(cpu)); |
| 5296 | #endif |
Peter Zijlstra | 6506cf6c | 2010-09-16 17:50:31 +0200 | [diff] [blame] | 5297 | /* |
| 5298 | * We're having a chicken and egg problem, even though we are |
| 5299 | * holding rq->lock, the cpu isn't yet set to this cpu so the |
| 5300 | * lockdep check in task_group() will fail. |
| 5301 | * |
| 5302 | * Similar case to sched_fork(). / Alternatively we could |
| 5303 | * use task_rq_lock() here and obtain the other rq->lock. |
| 5304 | * |
| 5305 | * Silence PROVE_RCU |
| 5306 | */ |
| 5307 | rcu_read_lock(); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5308 | __set_task_cpu(idle, cpu); |
Peter Zijlstra | 6506cf6c | 2010-09-16 17:50:31 +0200 | [diff] [blame] | 5309 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5310 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5311 | rq->curr = rq->idle = idle; |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 5312 | idle->on_rq = TASK_ON_RQ_QUEUED; |
Peter Zijlstra | de9b8f5 | 2015-08-13 23:09:29 +0200 | [diff] [blame] | 5313 | #ifdef CONFIG_SMP |
Peter Zijlstra | 3ca7a44 | 2011-04-05 17:23:40 +0200 | [diff] [blame] | 5314 | idle->on_cpu = 1; |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 5315 | #endif |
Peter Zijlstra | 25834c7 | 2015-05-15 17:43:34 +0200 | [diff] [blame] | 5316 | raw_spin_unlock(&rq->lock); |
| 5317 | raw_spin_unlock_irqrestore(&idle->pi_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5318 | |
| 5319 | /* Set the preempt count _outside_ the spinlocks! */ |
Peter Zijlstra | 0102874 | 2013-08-14 14:55:46 +0200 | [diff] [blame] | 5320 | init_idle_preempt_count(idle, cpu); |
Jonathan Corbet | 625f2a3 | 2011-04-22 11:19:10 -0600 | [diff] [blame] | 5321 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5322 | /* |
| 5323 | * The idle tasks have their own, simple scheduling class: |
| 5324 | */ |
| 5325 | idle->sched_class = &idle_sched_class; |
Steven Rostedt | 868baf0 | 2011-02-10 21:26:13 -0500 | [diff] [blame] | 5326 | ftrace_graph_init_idle_task(idle, cpu); |
Frederic Weisbecker | 45eacc6 | 2013-05-15 22:16:32 +0200 | [diff] [blame] | 5327 | vtime_init_idle(idle, cpu); |
Peter Zijlstra | de9b8f5 | 2015-08-13 23:09:29 +0200 | [diff] [blame] | 5328 | #ifdef CONFIG_SMP |
Carsten Emde | f1c6f1a | 2011-10-26 23:14:16 +0200 | [diff] [blame] | 5329 | sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu); |
| 5330 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5331 | } |
| 5332 | |
Juri Lelli | f82f804 | 2014-10-07 09:52:11 +0100 | [diff] [blame] | 5333 | int cpuset_cpumask_can_shrink(const struct cpumask *cur, |
| 5334 | const struct cpumask *trial) |
| 5335 | { |
| 5336 | int ret = 1, trial_cpus; |
| 5337 | struct dl_bw *cur_dl_b; |
| 5338 | unsigned long flags; |
| 5339 | |
Mike Galbraith | bb2bc55 | 2015-01-28 04:53:55 +0100 | [diff] [blame] | 5340 | if (!cpumask_weight(cur)) |
| 5341 | return ret; |
| 5342 | |
Juri Lelli | 75e23e4 | 2014-10-28 11:54:46 +0000 | [diff] [blame] | 5343 | rcu_read_lock_sched(); |
Juri Lelli | f82f804 | 2014-10-07 09:52:11 +0100 | [diff] [blame] | 5344 | cur_dl_b = dl_bw_of(cpumask_any(cur)); |
| 5345 | trial_cpus = cpumask_weight(trial); |
| 5346 | |
| 5347 | raw_spin_lock_irqsave(&cur_dl_b->lock, flags); |
| 5348 | if (cur_dl_b->bw != -1 && |
| 5349 | cur_dl_b->bw * trial_cpus < cur_dl_b->total_bw) |
| 5350 | ret = 0; |
| 5351 | raw_spin_unlock_irqrestore(&cur_dl_b->lock, flags); |
Juri Lelli | 75e23e4 | 2014-10-28 11:54:46 +0000 | [diff] [blame] | 5352 | rcu_read_unlock_sched(); |
Juri Lelli | f82f804 | 2014-10-07 09:52:11 +0100 | [diff] [blame] | 5353 | |
| 5354 | return ret; |
| 5355 | } |
| 5356 | |
Juri Lelli | 7f51412 | 2014-09-19 10:22:40 +0100 | [diff] [blame] | 5357 | int task_can_attach(struct task_struct *p, |
| 5358 | const struct cpumask *cs_cpus_allowed) |
| 5359 | { |
| 5360 | int ret = 0; |
| 5361 | |
| 5362 | /* |
| 5363 | * Kthreads which disallow setaffinity shouldn't be moved |
| 5364 | * to a new cpuset; we don't want to change their cpu |
| 5365 | * affinity and isolating such threads by their set of |
| 5366 | * allowed nodes is unnecessary. Thus, cpusets are not |
| 5367 | * applicable for such threads. This prevents checking for |
| 5368 | * success of set_cpus_allowed_ptr() on all attached tasks |
| 5369 | * before cpus_allowed may be changed. |
| 5370 | */ |
| 5371 | if (p->flags & PF_NO_SETAFFINITY) { |
| 5372 | ret = -EINVAL; |
| 5373 | goto out; |
| 5374 | } |
| 5375 | |
| 5376 | #ifdef CONFIG_SMP |
| 5377 | if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span, |
| 5378 | cs_cpus_allowed)) { |
| 5379 | unsigned int dest_cpu = cpumask_any_and(cpu_active_mask, |
| 5380 | cs_cpus_allowed); |
Juri Lelli | 75e23e4 | 2014-10-28 11:54:46 +0000 | [diff] [blame] | 5381 | struct dl_bw *dl_b; |
Juri Lelli | 7f51412 | 2014-09-19 10:22:40 +0100 | [diff] [blame] | 5382 | bool overflow; |
| 5383 | int cpus; |
| 5384 | unsigned long flags; |
| 5385 | |
Juri Lelli | 75e23e4 | 2014-10-28 11:54:46 +0000 | [diff] [blame] | 5386 | rcu_read_lock_sched(); |
| 5387 | dl_b = dl_bw_of(dest_cpu); |
Juri Lelli | 7f51412 | 2014-09-19 10:22:40 +0100 | [diff] [blame] | 5388 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
| 5389 | cpus = dl_bw_cpus(dest_cpu); |
| 5390 | overflow = __dl_overflow(dl_b, cpus, 0, p->dl.dl_bw); |
| 5391 | if (overflow) |
| 5392 | ret = -EBUSY; |
| 5393 | else { |
| 5394 | /* |
| 5395 | * We reserve space for this task in the destination |
| 5396 | * root_domain, as we can't fail after this point. |
| 5397 | * We will free resources in the source root_domain |
| 5398 | * later on (see set_cpus_allowed_dl()). |
| 5399 | */ |
| 5400 | __dl_add(dl_b, p->dl.dl_bw); |
| 5401 | } |
| 5402 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
Juri Lelli | 75e23e4 | 2014-10-28 11:54:46 +0000 | [diff] [blame] | 5403 | rcu_read_unlock_sched(); |
Juri Lelli | 7f51412 | 2014-09-19 10:22:40 +0100 | [diff] [blame] | 5404 | |
| 5405 | } |
| 5406 | #endif |
| 5407 | out: |
| 5408 | return ret; |
| 5409 | } |
| 5410 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5411 | #ifdef CONFIG_SMP |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5412 | |
Thomas Gleixner | e26fbff | 2016-03-10 12:54:10 +0100 | [diff] [blame] | 5413 | static bool sched_smp_initialized __read_mostly; |
| 5414 | |
Mel Gorman | e6628d5 | 2013-10-07 11:29:02 +0100 | [diff] [blame] | 5415 | #ifdef CONFIG_NUMA_BALANCING |
| 5416 | /* Migrate current task p to target_cpu */ |
| 5417 | int migrate_task_to(struct task_struct *p, int target_cpu) |
| 5418 | { |
| 5419 | struct migration_arg arg = { p, target_cpu }; |
| 5420 | int curr_cpu = task_cpu(p); |
| 5421 | |
| 5422 | if (curr_cpu == target_cpu) |
| 5423 | return 0; |
| 5424 | |
| 5425 | if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p))) |
| 5426 | return -EINVAL; |
| 5427 | |
| 5428 | /* TODO: This is not properly updating schedstats */ |
| 5429 | |
Mel Gorman | 286549d | 2014-01-21 15:51:03 -0800 | [diff] [blame] | 5430 | trace_sched_move_numa(p, curr_cpu, target_cpu); |
Mel Gorman | e6628d5 | 2013-10-07 11:29:02 +0100 | [diff] [blame] | 5431 | return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg); |
| 5432 | } |
Peter Zijlstra | 0ec8aa0 | 2013-10-07 11:29:33 +0100 | [diff] [blame] | 5433 | |
| 5434 | /* |
| 5435 | * Requeue a task on a given node and accurately track the number of NUMA |
| 5436 | * tasks on the runqueues |
| 5437 | */ |
| 5438 | void sched_setnuma(struct task_struct *p, int nid) |
| 5439 | { |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 5440 | bool queued, running; |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 5441 | struct rq_flags rf; |
| 5442 | struct rq *rq; |
Peter Zijlstra | 0ec8aa0 | 2013-10-07 11:29:33 +0100 | [diff] [blame] | 5443 | |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 5444 | rq = task_rq_lock(p, &rf); |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 5445 | queued = task_on_rq_queued(p); |
Peter Zijlstra | 0ec8aa0 | 2013-10-07 11:29:33 +0100 | [diff] [blame] | 5446 | running = task_current(rq, p); |
| 5447 | |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 5448 | if (queued) |
Peter Zijlstra | 1de6444 | 2015-09-30 17:44:13 +0200 | [diff] [blame] | 5449 | dequeue_task(rq, p, DEQUEUE_SAVE); |
Peter Zijlstra | 0ec8aa0 | 2013-10-07 11:29:33 +0100 | [diff] [blame] | 5450 | if (running) |
Kirill Tkhai | f3cd1c4 | 2014-09-12 17:41:40 +0400 | [diff] [blame] | 5451 | put_prev_task(rq, p); |
Peter Zijlstra | 0ec8aa0 | 2013-10-07 11:29:33 +0100 | [diff] [blame] | 5452 | |
| 5453 | p->numa_preferred_nid = nid; |
Peter Zijlstra | 0ec8aa0 | 2013-10-07 11:29:33 +0100 | [diff] [blame] | 5454 | |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 5455 | if (queued) |
Peter Zijlstra | 1de6444 | 2015-09-30 17:44:13 +0200 | [diff] [blame] | 5456 | enqueue_task(rq, p, ENQUEUE_RESTORE); |
Vincent Guittot | a399d23 | 2016-09-12 09:47:52 +0200 | [diff] [blame] | 5457 | if (running) |
Peter Zijlstra | b2bf6c3 | 2016-09-20 22:00:38 +0200 | [diff] [blame] | 5458 | set_curr_task(rq, p); |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 5459 | task_rq_unlock(rq, p, &rf); |
Peter Zijlstra | 0ec8aa0 | 2013-10-07 11:29:33 +0100 | [diff] [blame] | 5460 | } |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 5461 | #endif /* CONFIG_NUMA_BALANCING */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5462 | |
| 5463 | #ifdef CONFIG_HOTPLUG_CPU |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5464 | /* |
| 5465 | * Ensures that the idle task is using init_mm right before its cpu goes |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5466 | * offline. |
| 5467 | */ |
| 5468 | void idle_task_exit(void) |
| 5469 | { |
| 5470 | struct mm_struct *mm = current->active_mm; |
| 5471 | |
| 5472 | BUG_ON(cpu_online(smp_processor_id())); |
| 5473 | |
Martin Schwidefsky | a53efe5 | 2012-10-26 17:17:44 +0200 | [diff] [blame] | 5474 | if (mm != &init_mm) { |
Andy Lutomirski | 8a48b7e | 2017-06-09 11:49:15 -0700 | [diff] [blame] | 5475 | switch_mm(mm, &init_mm, current); |
Martin Schwidefsky | a53efe5 | 2012-10-26 17:17:44 +0200 | [diff] [blame] | 5476 | finish_arch_post_lock_switch(); |
| 5477 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5478 | mmdrop(mm); |
| 5479 | } |
| 5480 | |
Peter Zijlstra | 48c5cca | 2010-11-13 19:32:29 +0100 | [diff] [blame] | 5481 | /* |
Peter Zijlstra | 5d18023 | 2012-08-20 11:26:57 +0200 | [diff] [blame] | 5482 | * Since this CPU is going 'away' for a while, fold any nr_active delta |
| 5483 | * we might have. Assumes we're called after migrate_tasks() so that the |
Thomas Gleixner | d60585c | 2016-07-12 18:33:56 +0200 | [diff] [blame] | 5484 | * nr_active count is stable. We need to take the teardown thread which |
| 5485 | * is calling this into account, so we hand in adjust = 1 to the load |
| 5486 | * calculation. |
Peter Zijlstra | 5d18023 | 2012-08-20 11:26:57 +0200 | [diff] [blame] | 5487 | * |
| 5488 | * Also see the comment "Global load-average calculations". |
Peter Zijlstra | 48c5cca | 2010-11-13 19:32:29 +0100 | [diff] [blame] | 5489 | */ |
Peter Zijlstra | 5d18023 | 2012-08-20 11:26:57 +0200 | [diff] [blame] | 5490 | static void calc_load_migrate(struct rq *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5491 | { |
Thomas Gleixner | d60585c | 2016-07-12 18:33:56 +0200 | [diff] [blame] | 5492 | long delta = calc_load_fold_active(rq, 1); |
Peter Zijlstra | 5d18023 | 2012-08-20 11:26:57 +0200 | [diff] [blame] | 5493 | if (delta) |
| 5494 | atomic_long_add(delta, &calc_load_tasks); |
Thomas Gleixner | dce48a8 | 2009-04-11 10:43:41 +0200 | [diff] [blame] | 5495 | } |
Peter Zijlstra | 48c5cca | 2010-11-13 19:32:29 +0100 | [diff] [blame] | 5496 | |
Peter Zijlstra | 3f1d2a3 | 2014-02-12 10:49:30 +0100 | [diff] [blame] | 5497 | static void put_prev_task_fake(struct rq *rq, struct task_struct *prev) |
| 5498 | { |
| 5499 | } |
| 5500 | |
| 5501 | static const struct sched_class fake_sched_class = { |
| 5502 | .put_prev_task = put_prev_task_fake, |
| 5503 | }; |
| 5504 | |
| 5505 | static struct task_struct fake_task = { |
| 5506 | /* |
| 5507 | * Avoid pull_{rt,dl}_task() |
| 5508 | */ |
| 5509 | .prio = MAX_PRIO + 1, |
| 5510 | .sched_class = &fake_sched_class, |
| 5511 | }; |
| 5512 | |
Peter Zijlstra | 48c5cca | 2010-11-13 19:32:29 +0100 | [diff] [blame] | 5513 | /* |
| 5514 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
| 5515 | * try_to_wake_up()->select_task_rq(). |
| 5516 | * |
| 5517 | * Called with rq->lock held even though we'er in stop_machine() and |
| 5518 | * there's no concurrency possible, we hold the required locks anyway |
| 5519 | * because of lock validation efforts. |
| 5520 | */ |
Peter Zijlstra | 5e16bbc | 2015-06-11 14:46:51 +0200 | [diff] [blame] | 5521 | static void migrate_tasks(struct rq *dead_rq) |
Peter Zijlstra | 48c5cca | 2010-11-13 19:32:29 +0100 | [diff] [blame] | 5522 | { |
Peter Zijlstra | 5e16bbc | 2015-06-11 14:46:51 +0200 | [diff] [blame] | 5523 | struct rq *rq = dead_rq; |
Peter Zijlstra | 48c5cca | 2010-11-13 19:32:29 +0100 | [diff] [blame] | 5524 | struct task_struct *next, *stop = rq->stop; |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 5525 | struct pin_cookie cookie; |
Peter Zijlstra | 48c5cca | 2010-11-13 19:32:29 +0100 | [diff] [blame] | 5526 | int dest_cpu; |
| 5527 | |
| 5528 | /* |
| 5529 | * Fudge the rq selection such that the below task selection loop |
| 5530 | * doesn't get stuck on the currently eligible stop task. |
| 5531 | * |
| 5532 | * We're currently inside stop_machine() and the rq is either stuck |
| 5533 | * in the stop_machine_cpu_stop() loop, or we're executing this code, |
| 5534 | * either way we should never end up calling schedule() until we're |
| 5535 | * done here. |
| 5536 | */ |
| 5537 | rq->stop = NULL; |
| 5538 | |
Frederic Weisbecker | 77bd397 | 2013-04-12 01:50:58 +0200 | [diff] [blame] | 5539 | /* |
| 5540 | * put_prev_task() and pick_next_task() sched |
| 5541 | * class method both need to have an up-to-date |
| 5542 | * value of rq->clock[_task] |
| 5543 | */ |
| 5544 | update_rq_clock(rq); |
| 5545 | |
Peter Zijlstra | 5e16bbc | 2015-06-11 14:46:51 +0200 | [diff] [blame] | 5546 | for (;;) { |
Peter Zijlstra | 48c5cca | 2010-11-13 19:32:29 +0100 | [diff] [blame] | 5547 | /* |
| 5548 | * There's this thread running, bail when that's the only |
| 5549 | * remaining thread. |
| 5550 | */ |
| 5551 | if (rq->nr_running == 1) |
| 5552 | break; |
| 5553 | |
Peter Zijlstra | cbce1a6 | 2015-06-11 14:46:54 +0200 | [diff] [blame] | 5554 | /* |
Wanpeng Li | 5473e0cc | 2015-08-28 14:55:56 +0800 | [diff] [blame] | 5555 | * pick_next_task assumes pinned rq->lock. |
Peter Zijlstra | cbce1a6 | 2015-06-11 14:46:54 +0200 | [diff] [blame] | 5556 | */ |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 5557 | cookie = lockdep_pin_lock(&rq->lock); |
| 5558 | next = pick_next_task(rq, &fake_task, cookie); |
Peter Zijlstra | 48c5cca | 2010-11-13 19:32:29 +0100 | [diff] [blame] | 5559 | BUG_ON(!next); |
| 5560 | next->sched_class->put_prev_task(rq, next); |
| 5561 | |
Wanpeng Li | 5473e0cc | 2015-08-28 14:55:56 +0800 | [diff] [blame] | 5562 | /* |
| 5563 | * Rules for changing task_struct::cpus_allowed are holding |
| 5564 | * both pi_lock and rq->lock, such that holding either |
| 5565 | * stabilizes the mask. |
| 5566 | * |
| 5567 | * Drop rq->lock is not quite as disastrous as it usually is |
| 5568 | * because !cpu_active at this point, which means load-balance |
| 5569 | * will not interfere. Also, stop-machine. |
| 5570 | */ |
Peter Zijlstra | e7904a2 | 2015-08-01 19:25:08 +0200 | [diff] [blame] | 5571 | lockdep_unpin_lock(&rq->lock, cookie); |
Wanpeng Li | 5473e0cc | 2015-08-28 14:55:56 +0800 | [diff] [blame] | 5572 | raw_spin_unlock(&rq->lock); |
| 5573 | raw_spin_lock(&next->pi_lock); |
| 5574 | raw_spin_lock(&rq->lock); |
| 5575 | |
| 5576 | /* |
| 5577 | * Since we're inside stop-machine, _nothing_ should have |
| 5578 | * changed the task, WARN if weird stuff happened, because in |
| 5579 | * that case the above rq->lock drop is a fail too. |
| 5580 | */ |
| 5581 | if (WARN_ON(task_rq(next) != rq || !task_on_rq_queued(next))) { |
| 5582 | raw_spin_unlock(&next->pi_lock); |
| 5583 | continue; |
| 5584 | } |
| 5585 | |
Peter Zijlstra | 48c5cca | 2010-11-13 19:32:29 +0100 | [diff] [blame] | 5586 | /* Find suitable destination for @next, with force if needed. */ |
Peter Zijlstra | 5e16bbc | 2015-06-11 14:46:51 +0200 | [diff] [blame] | 5587 | dest_cpu = select_fallback_rq(dead_rq->cpu, next); |
Peter Zijlstra | 48c5cca | 2010-11-13 19:32:29 +0100 | [diff] [blame] | 5588 | |
Peter Zijlstra | 5e16bbc | 2015-06-11 14:46:51 +0200 | [diff] [blame] | 5589 | rq = __migrate_task(rq, next, dest_cpu); |
| 5590 | if (rq != dead_rq) { |
| 5591 | raw_spin_unlock(&rq->lock); |
| 5592 | rq = dead_rq; |
| 5593 | raw_spin_lock(&rq->lock); |
| 5594 | } |
Wanpeng Li | 5473e0cc | 2015-08-28 14:55:56 +0800 | [diff] [blame] | 5595 | raw_spin_unlock(&next->pi_lock); |
Peter Zijlstra | 48c5cca | 2010-11-13 19:32:29 +0100 | [diff] [blame] | 5596 | } |
| 5597 | |
| 5598 | rq->stop = stop; |
| 5599 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5600 | #endif /* CONFIG_HOTPLUG_CPU */ |
| 5601 | |
Gregory Haskins | 1f11eb6 | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 5602 | static void set_rq_online(struct rq *rq) |
| 5603 | { |
| 5604 | if (!rq->online) { |
| 5605 | const struct sched_class *class; |
| 5606 | |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5607 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
Gregory Haskins | 1f11eb6 | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 5608 | rq->online = 1; |
| 5609 | |
| 5610 | for_each_class(class) { |
| 5611 | if (class->rq_online) |
| 5612 | class->rq_online(rq); |
| 5613 | } |
| 5614 | } |
| 5615 | } |
| 5616 | |
| 5617 | static void set_rq_offline(struct rq *rq) |
| 5618 | { |
| 5619 | if (rq->online) { |
| 5620 | const struct sched_class *class; |
| 5621 | |
| 5622 | for_each_class(class) { |
| 5623 | if (class->rq_offline) |
| 5624 | class->rq_offline(rq); |
| 5625 | } |
| 5626 | |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5627 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
Gregory Haskins | 1f11eb6 | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 5628 | rq->online = 0; |
| 5629 | } |
| 5630 | } |
| 5631 | |
Thomas Gleixner | 9cf7243 | 2016-03-10 12:54:09 +0100 | [diff] [blame] | 5632 | static void set_cpu_rq_start_time(unsigned int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5633 | { |
Tejun Heo | 969c792 | 2010-05-06 18:49:21 +0200 | [diff] [blame] | 5634 | struct rq *rq = cpu_rq(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5635 | |
Corey Minyard | a803f02 | 2014-05-08 13:47:39 -0500 | [diff] [blame] | 5636 | rq->age_stamp = sched_clock_cpu(cpu); |
| 5637 | } |
| 5638 | |
Peter Zijlstra | 4cb9883 | 2011-04-07 14:09:58 +0200 | [diff] [blame] | 5639 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
| 5640 | |
Ingo Molnar | 3e9830d | 2007-10-15 17:00:13 +0200 | [diff] [blame] | 5641 | #ifdef CONFIG_SCHED_DEBUG |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5642 | |
Peter Zijlstra | d039ac6 | 2012-05-31 21:20:16 +0200 | [diff] [blame] | 5643 | static __read_mostly int sched_debug_enabled; |
Mike Travis | f663011 | 2009-11-17 18:22:15 -0600 | [diff] [blame] | 5644 | |
Peter Zijlstra | d039ac6 | 2012-05-31 21:20:16 +0200 | [diff] [blame] | 5645 | static int __init sched_debug_setup(char *str) |
Mike Travis | f663011 | 2009-11-17 18:22:15 -0600 | [diff] [blame] | 5646 | { |
Peter Zijlstra | d039ac6 | 2012-05-31 21:20:16 +0200 | [diff] [blame] | 5647 | sched_debug_enabled = 1; |
Mike Travis | f663011 | 2009-11-17 18:22:15 -0600 | [diff] [blame] | 5648 | |
| 5649 | return 0; |
| 5650 | } |
Peter Zijlstra | d039ac6 | 2012-05-31 21:20:16 +0200 | [diff] [blame] | 5651 | early_param("sched_debug", sched_debug_setup); |
| 5652 | |
| 5653 | static inline bool sched_debug(void) |
| 5654 | { |
| 5655 | return sched_debug_enabled; |
| 5656 | } |
Mike Travis | f663011 | 2009-11-17 18:22:15 -0600 | [diff] [blame] | 5657 | |
Mike Travis | 7c16ec5 | 2008-04-04 18:11:11 -0700 | [diff] [blame] | 5658 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
Rusty Russell | 96f874e | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 5659 | struct cpumask *groupmask) |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5660 | { |
| 5661 | struct sched_group *group = sd->groups; |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5662 | |
Rusty Russell | 96f874e | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 5663 | cpumask_clear(groupmask); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5664 | |
| 5665 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); |
| 5666 | |
| 5667 | if (!(sd->flags & SD_LOAD_BALANCE)) { |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5668 | printk("does not load-balance\n"); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5669 | if (sd->parent) |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5670 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
| 5671 | " has parent"); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5672 | return -1; |
| 5673 | } |
| 5674 | |
Tejun Heo | 333470e | 2015-02-13 14:37:28 -0800 | [diff] [blame] | 5675 | printk(KERN_CONT "span %*pbl level %s\n", |
| 5676 | cpumask_pr_args(sched_domain_span(sd)), sd->name); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5677 | |
Rusty Russell | 758b2cd | 2008-11-25 02:35:04 +1030 | [diff] [blame] | 5678 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5679 | printk(KERN_ERR "ERROR: domain->span does not contain " |
| 5680 | "CPU%d\n", cpu); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5681 | } |
Rusty Russell | 758b2cd | 2008-11-25 02:35:04 +1030 | [diff] [blame] | 5682 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5683 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
| 5684 | " CPU%d\n", cpu); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5685 | } |
| 5686 | |
| 5687 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
| 5688 | do { |
| 5689 | if (!group) { |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5690 | printk("\n"); |
| 5691 | printk(KERN_ERR "ERROR: group is NULL\n"); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5692 | break; |
| 5693 | } |
| 5694 | |
Rusty Russell | 758b2cd | 2008-11-25 02:35:04 +1030 | [diff] [blame] | 5695 | if (!cpumask_weight(sched_group_cpus(group))) { |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5696 | printk(KERN_CONT "\n"); |
| 5697 | printk(KERN_ERR "ERROR: empty group\n"); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5698 | break; |
| 5699 | } |
| 5700 | |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 5701 | if (!(sd->flags & SD_OVERLAP) && |
| 5702 | cpumask_intersects(groupmask, sched_group_cpus(group))) { |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5703 | printk(KERN_CONT "\n"); |
| 5704 | printk(KERN_ERR "ERROR: repeated CPUs\n"); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5705 | break; |
| 5706 | } |
| 5707 | |
Rusty Russell | 758b2cd | 2008-11-25 02:35:04 +1030 | [diff] [blame] | 5708 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5709 | |
Tejun Heo | 333470e | 2015-02-13 14:37:28 -0800 | [diff] [blame] | 5710 | printk(KERN_CONT " %*pbl", |
| 5711 | cpumask_pr_args(sched_group_cpus(group))); |
Nicolas Pitre | ca8ce3d | 2014-05-26 18:19:39 -0400 | [diff] [blame] | 5712 | if (group->sgc->capacity != SCHED_CAPACITY_SCALE) { |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 5713 | printk(KERN_CONT " (cpu_capacity = %d)", |
| 5714 | group->sgc->capacity); |
Gautham R Shenoy | 381512c | 2009-04-14 09:09:36 +0530 | [diff] [blame] | 5715 | } |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5716 | |
| 5717 | group = group->next; |
| 5718 | } while (group != sd->groups); |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5719 | printk(KERN_CONT "\n"); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5720 | |
Rusty Russell | 758b2cd | 2008-11-25 02:35:04 +1030 | [diff] [blame] | 5721 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5722 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5723 | |
Rusty Russell | 758b2cd | 2008-11-25 02:35:04 +1030 | [diff] [blame] | 5724 | if (sd->parent && |
| 5725 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 5726 | printk(KERN_ERR "ERROR: parent span is not a superset " |
| 5727 | "of domain->span\n"); |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5728 | return 0; |
| 5729 | } |
| 5730 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5731 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
| 5732 | { |
| 5733 | int level = 0; |
| 5734 | |
Peter Zijlstra | d039ac6 | 2012-05-31 21:20:16 +0200 | [diff] [blame] | 5735 | if (!sched_debug_enabled) |
Mike Travis | f663011 | 2009-11-17 18:22:15 -0600 | [diff] [blame] | 5736 | return; |
| 5737 | |
Nick Piggin | 41c7ce9 | 2005-06-25 14:57:24 -0700 | [diff] [blame] | 5738 | if (!sd) { |
| 5739 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); |
| 5740 | return; |
| 5741 | } |
| 5742 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5743 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
| 5744 | |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5745 | for (;;) { |
Peter Zijlstra | 4cb9883 | 2011-04-07 14:09:58 +0200 | [diff] [blame] | 5746 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5747 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5748 | level++; |
| 5749 | sd = sd->parent; |
Miguel Ojeda Sandonis | 33859f7 | 2006-12-10 02:20:38 -0800 | [diff] [blame] | 5750 | if (!sd) |
Ingo Molnar | 4dcf6af | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 5751 | break; |
| 5752 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5753 | } |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 5754 | #else /* !CONFIG_SCHED_DEBUG */ |
Peter Zijlstra | a18a579 | 2016-09-20 11:05:31 +0200 | [diff] [blame] | 5755 | |
| 5756 | # define sched_debug_enabled 0 |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5757 | # define sched_domain_debug(sd, cpu) do { } while (0) |
Peter Zijlstra | d039ac6 | 2012-05-31 21:20:16 +0200 | [diff] [blame] | 5758 | static inline bool sched_debug(void) |
| 5759 | { |
| 5760 | return false; |
| 5761 | } |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 5762 | #endif /* CONFIG_SCHED_DEBUG */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5763 | |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 5764 | static int sd_degenerate(struct sched_domain *sd) |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5765 | { |
Rusty Russell | 758b2cd | 2008-11-25 02:35:04 +1030 | [diff] [blame] | 5766 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5767 | return 1; |
| 5768 | |
| 5769 | /* Following flags need at least 2 groups */ |
| 5770 | if (sd->flags & (SD_LOAD_BALANCE | |
| 5771 | SD_BALANCE_NEWIDLE | |
| 5772 | SD_BALANCE_FORK | |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 5773 | SD_BALANCE_EXEC | |
Nicolas Pitre | 5d4dfdd | 2014-05-27 13:50:41 -0400 | [diff] [blame] | 5774 | SD_SHARE_CPUCAPACITY | |
Morten Rasmussen | 1f6e6c7 | 2016-07-25 14:34:22 +0100 | [diff] [blame] | 5775 | SD_ASYM_CPUCAPACITY | |
Vincent Guittot | d77b3ed | 2014-04-11 11:44:40 +0200 | [diff] [blame] | 5776 | SD_SHARE_PKG_RESOURCES | |
| 5777 | SD_SHARE_POWERDOMAIN)) { |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5778 | if (sd->groups != sd->groups->next) |
| 5779 | return 0; |
| 5780 | } |
| 5781 | |
| 5782 | /* Following flags don't use groups */ |
Peter Zijlstra | c88d591 | 2009-09-10 13:50:02 +0200 | [diff] [blame] | 5783 | if (sd->flags & (SD_WAKE_AFFINE)) |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5784 | return 0; |
| 5785 | |
| 5786 | return 1; |
| 5787 | } |
| 5788 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5789 | static int |
| 5790 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5791 | { |
| 5792 | unsigned long cflags = sd->flags, pflags = parent->flags; |
| 5793 | |
| 5794 | if (sd_degenerate(parent)) |
| 5795 | return 1; |
| 5796 | |
Rusty Russell | 758b2cd | 2008-11-25 02:35:04 +1030 | [diff] [blame] | 5797 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5798 | return 0; |
| 5799 | |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5800 | /* Flags needing groups don't count if only 1 group in parent */ |
| 5801 | if (parent->groups == parent->groups->next) { |
| 5802 | pflags &= ~(SD_LOAD_BALANCE | |
| 5803 | SD_BALANCE_NEWIDLE | |
| 5804 | SD_BALANCE_FORK | |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 5805 | SD_BALANCE_EXEC | |
Morten Rasmussen | 1f6e6c7 | 2016-07-25 14:34:22 +0100 | [diff] [blame] | 5806 | SD_ASYM_CPUCAPACITY | |
Nicolas Pitre | 5d4dfdd | 2014-05-27 13:50:41 -0400 | [diff] [blame] | 5807 | SD_SHARE_CPUCAPACITY | |
Peter Zijlstra | 10866e62 | 2013-08-19 16:57:04 +0200 | [diff] [blame] | 5808 | SD_SHARE_PKG_RESOURCES | |
Vincent Guittot | d77b3ed | 2014-04-11 11:44:40 +0200 | [diff] [blame] | 5809 | SD_PREFER_SIBLING | |
| 5810 | SD_SHARE_POWERDOMAIN); |
Ken Chen | 5436499 | 2008-12-07 18:47:37 -0800 | [diff] [blame] | 5811 | if (nr_node_ids == 1) |
| 5812 | pflags &= ~SD_SERIALIZE; |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5813 | } |
| 5814 | if (~cflags & pflags) |
| 5815 | return 0; |
| 5816 | |
| 5817 | return 1; |
| 5818 | } |
| 5819 | |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5820 | static void free_rootdomain(struct rcu_head *rcu) |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5821 | { |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5822 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
Peter Zijlstra | 047106a | 2009-11-16 10:28:09 +0100 | [diff] [blame] | 5823 | |
Rusty Russell | 68e7456 | 2008-11-25 02:35:13 +1030 | [diff] [blame] | 5824 | cpupri_cleanup(&rd->cpupri); |
Juri Lelli | 6bfd6d7 | 2013-11-07 14:43:47 +0100 | [diff] [blame] | 5825 | cpudl_cleanup(&rd->cpudl); |
Juri Lelli | 1baca4c | 2013-11-07 14:43:38 +0100 | [diff] [blame] | 5826 | free_cpumask_var(rd->dlo_mask); |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5827 | free_cpumask_var(rd->rto_mask); |
| 5828 | free_cpumask_var(rd->online); |
| 5829 | free_cpumask_var(rd->span); |
| 5830 | kfree(rd); |
| 5831 | } |
| 5832 | |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5833 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
| 5834 | { |
Ingo Molnar | a0490fa | 2009-02-12 11:35:40 +0100 | [diff] [blame] | 5835 | struct root_domain *old_rd = NULL; |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5836 | unsigned long flags; |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5837 | |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 5838 | raw_spin_lock_irqsave(&rq->lock, flags); |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5839 | |
| 5840 | if (rq->rd) { |
Ingo Molnar | a0490fa | 2009-02-12 11:35:40 +0100 | [diff] [blame] | 5841 | old_rd = rq->rd; |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5842 | |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5843 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
Gregory Haskins | 1f11eb6 | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 5844 | set_rq_offline(rq); |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5845 | |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5846 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
Gregory Haskins | dc93852 | 2008-01-25 21:08:26 +0100 | [diff] [blame] | 5847 | |
Ingo Molnar | a0490fa | 2009-02-12 11:35:40 +0100 | [diff] [blame] | 5848 | /* |
Shigeru Yoshida | 0515973 | 2013-11-17 12:12:36 +0900 | [diff] [blame] | 5849 | * If we dont want to free the old_rd yet then |
Ingo Molnar | a0490fa | 2009-02-12 11:35:40 +0100 | [diff] [blame] | 5850 | * set old_rd to NULL to skip the freeing later |
| 5851 | * in this function: |
| 5852 | */ |
| 5853 | if (!atomic_dec_and_test(&old_rd->refcount)) |
| 5854 | old_rd = NULL; |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5855 | } |
| 5856 | |
| 5857 | atomic_inc(&rd->refcount); |
| 5858 | rq->rd = rd; |
| 5859 | |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5860 | cpumask_set_cpu(rq->cpu, rd->span); |
Gregory Haskins | 00aec93 | 2009-07-30 10:57:23 -0400 | [diff] [blame] | 5861 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
Gregory Haskins | 1f11eb6 | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 5862 | set_rq_online(rq); |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5863 | |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 5864 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
Ingo Molnar | a0490fa | 2009-02-12 11:35:40 +0100 | [diff] [blame] | 5865 | |
| 5866 | if (old_rd) |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5867 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5868 | } |
| 5869 | |
Steven Rostedt (VMware) | a384e54 | 2018-01-23 20:45:38 -0500 | [diff] [blame] | 5870 | void sched_get_rd(struct root_domain *rd) |
| 5871 | { |
| 5872 | atomic_inc(&rd->refcount); |
| 5873 | } |
| 5874 | |
| 5875 | void sched_put_rd(struct root_domain *rd) |
| 5876 | { |
| 5877 | if (!atomic_dec_and_test(&rd->refcount)) |
| 5878 | return; |
| 5879 | |
| 5880 | call_rcu_sched(&rd->rcu, free_rootdomain); |
| 5881 | } |
| 5882 | |
Pekka Enberg | 68c38fc | 2010-07-15 23:18:22 +0300 | [diff] [blame] | 5883 | static int init_rootdomain(struct root_domain *rd) |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5884 | { |
| 5885 | memset(rd, 0, sizeof(*rd)); |
| 5886 | |
Xunlei Pang | 8295c69 | 2015-12-02 19:52:59 +0800 | [diff] [blame] | 5887 | if (!zalloc_cpumask_var(&rd->span, GFP_KERNEL)) |
Li Zefan | 0c910d2 | 2009-01-06 17:39:06 +0800 | [diff] [blame] | 5888 | goto out; |
Xunlei Pang | 8295c69 | 2015-12-02 19:52:59 +0800 | [diff] [blame] | 5889 | if (!zalloc_cpumask_var(&rd->online, GFP_KERNEL)) |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5890 | goto free_span; |
Xunlei Pang | 8295c69 | 2015-12-02 19:52:59 +0800 | [diff] [blame] | 5891 | if (!zalloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL)) |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5892 | goto free_online; |
Xunlei Pang | 8295c69 | 2015-12-02 19:52:59 +0800 | [diff] [blame] | 5893 | if (!zalloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
Juri Lelli | 1baca4c | 2013-11-07 14:43:38 +0100 | [diff] [blame] | 5894 | goto free_dlo_mask; |
Gregory Haskins | 6e0534f | 2008-05-12 21:21:01 +0200 | [diff] [blame] | 5895 | |
Steven Rostedt (Red Hat) | 1c37ff7 | 2017-10-06 14:05:04 -0400 | [diff] [blame] | 5896 | #ifdef HAVE_RT_PUSH_IPI |
| 5897 | rd->rto_cpu = -1; |
| 5898 | raw_spin_lock_init(&rd->rto_lock); |
| 5899 | init_irq_work(&rd->rto_push_work, rto_push_irq_work_func); |
| 5900 | #endif |
| 5901 | |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 5902 | init_dl_bw(&rd->dl_bw); |
Juri Lelli | 6bfd6d7 | 2013-11-07 14:43:47 +0100 | [diff] [blame] | 5903 | if (cpudl_init(&rd->cpudl) != 0) |
| 5904 | goto free_dlo_mask; |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 5905 | |
Pekka Enberg | 68c38fc | 2010-07-15 23:18:22 +0300 | [diff] [blame] | 5906 | if (cpupri_init(&rd->cpupri) != 0) |
Rusty Russell | 68e7456 | 2008-11-25 02:35:13 +1030 | [diff] [blame] | 5907 | goto free_rto_mask; |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5908 | return 0; |
| 5909 | |
Rusty Russell | 68e7456 | 2008-11-25 02:35:13 +1030 | [diff] [blame] | 5910 | free_rto_mask: |
| 5911 | free_cpumask_var(rd->rto_mask); |
Juri Lelli | 1baca4c | 2013-11-07 14:43:38 +0100 | [diff] [blame] | 5912 | free_dlo_mask: |
| 5913 | free_cpumask_var(rd->dlo_mask); |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5914 | free_online: |
| 5915 | free_cpumask_var(rd->online); |
| 5916 | free_span: |
| 5917 | free_cpumask_var(rd->span); |
Li Zefan | 0c910d2 | 2009-01-06 17:39:06 +0800 | [diff] [blame] | 5918 | out: |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5919 | return -ENOMEM; |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5920 | } |
| 5921 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 5922 | /* |
| 5923 | * By default the system creates a single root-domain with all cpus as |
| 5924 | * members (mimicking the global state we have today). |
| 5925 | */ |
| 5926 | struct root_domain def_root_domain; |
| 5927 | |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5928 | static void init_defrootdomain(void) |
| 5929 | { |
Pekka Enberg | 68c38fc | 2010-07-15 23:18:22 +0300 | [diff] [blame] | 5930 | init_rootdomain(&def_root_domain); |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5931 | |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5932 | atomic_set(&def_root_domain.refcount, 1); |
| 5933 | } |
| 5934 | |
Gregory Haskins | dc93852 | 2008-01-25 21:08:26 +0100 | [diff] [blame] | 5935 | static struct root_domain *alloc_rootdomain(void) |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5936 | { |
| 5937 | struct root_domain *rd; |
| 5938 | |
| 5939 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); |
| 5940 | if (!rd) |
| 5941 | return NULL; |
| 5942 | |
Pekka Enberg | 68c38fc | 2010-07-15 23:18:22 +0300 | [diff] [blame] | 5943 | if (init_rootdomain(rd) != 0) { |
Rusty Russell | c6c4927 | 2008-11-25 02:35:05 +1030 | [diff] [blame] | 5944 | kfree(rd); |
| 5945 | return NULL; |
| 5946 | } |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 5947 | |
| 5948 | return rd; |
| 5949 | } |
| 5950 | |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 5951 | static void free_sched_groups(struct sched_group *sg, int free_sgc) |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5952 | { |
| 5953 | struct sched_group *tmp, *first; |
| 5954 | |
| 5955 | if (!sg) |
| 5956 | return; |
| 5957 | |
| 5958 | first = sg; |
| 5959 | do { |
| 5960 | tmp = sg->next; |
| 5961 | |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 5962 | if (free_sgc && atomic_dec_and_test(&sg->sgc->ref)) |
| 5963 | kfree(sg->sgc); |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5964 | |
| 5965 | kfree(sg); |
| 5966 | sg = tmp; |
| 5967 | } while (sg != first); |
| 5968 | } |
| 5969 | |
Peter Zijlstra | 16f3ef4 | 2016-05-09 10:37:57 +0200 | [diff] [blame] | 5970 | static void destroy_sched_domain(struct sched_domain *sd) |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5971 | { |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 5972 | /* |
| 5973 | * If its an overlapping domain it has private groups, iterate and |
| 5974 | * nuke them all. |
| 5975 | */ |
| 5976 | if (sd->flags & SD_OVERLAP) { |
| 5977 | free_sched_groups(sd->groups, 1); |
| 5978 | } else if (atomic_dec_and_test(&sd->groups->ref)) { |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 5979 | kfree(sd->groups->sgc); |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5980 | kfree(sd->groups); |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 5981 | } |
Peter Zijlstra | 24fc7ed | 2016-05-09 10:37:59 +0200 | [diff] [blame] | 5982 | if (sd->shared && atomic_dec_and_test(&sd->shared->ref)) |
| 5983 | kfree(sd->shared); |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5984 | kfree(sd); |
| 5985 | } |
| 5986 | |
Peter Zijlstra | 16f3ef4 | 2016-05-09 10:37:57 +0200 | [diff] [blame] | 5987 | static void destroy_sched_domains_rcu(struct rcu_head *rcu) |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5988 | { |
Peter Zijlstra | 16f3ef4 | 2016-05-09 10:37:57 +0200 | [diff] [blame] | 5989 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); |
| 5990 | |
| 5991 | while (sd) { |
| 5992 | struct sched_domain *parent = sd->parent; |
| 5993 | destroy_sched_domain(sd); |
| 5994 | sd = parent; |
| 5995 | } |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5996 | } |
| 5997 | |
Peter Zijlstra | f39180e | 2016-05-09 10:37:54 +0200 | [diff] [blame] | 5998 | static void destroy_sched_domains(struct sched_domain *sd) |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 5999 | { |
Peter Zijlstra | 16f3ef4 | 2016-05-09 10:37:57 +0200 | [diff] [blame] | 6000 | if (sd) |
| 6001 | call_rcu(&sd->rcu, destroy_sched_domains_rcu); |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6002 | } |
| 6003 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6004 | /* |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 6005 | * Keep a special pointer to the highest sched_domain that has |
| 6006 | * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this |
| 6007 | * allows us to avoid some pointer chasing select_idle_sibling(). |
| 6008 | * |
| 6009 | * Also keep a unique ID per domain (we use the first cpu number in |
| 6010 | * the cpumask of the domain), this allows us to quickly tell if |
Peter Zijlstra | 39be350 | 2012-01-26 12:44:34 +0100 | [diff] [blame] | 6011 | * two cpus are in the same cache domain, see cpus_share_cache(). |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 6012 | */ |
| 6013 | DEFINE_PER_CPU(struct sched_domain *, sd_llc); |
Peter Zijlstra | 7d9ffa8 | 2013-07-04 12:56:46 +0800 | [diff] [blame] | 6014 | DEFINE_PER_CPU(int, sd_llc_size); |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 6015 | DEFINE_PER_CPU(int, sd_llc_id); |
Peter Zijlstra | 0e369d7 | 2016-05-09 10:38:01 +0200 | [diff] [blame] | 6016 | DEFINE_PER_CPU(struct sched_domain_shared *, sd_llc_shared); |
Mel Gorman | fb13c7e | 2013-10-07 11:29:17 +0100 | [diff] [blame] | 6017 | DEFINE_PER_CPU(struct sched_domain *, sd_numa); |
Preeti U Murthy | 37dc6b5 | 2013-10-30 08:42:52 +0530 | [diff] [blame] | 6018 | DEFINE_PER_CPU(struct sched_domain *, sd_asym); |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 6019 | |
| 6020 | static void update_top_cache_domain(int cpu) |
| 6021 | { |
Peter Zijlstra | 0e369d7 | 2016-05-09 10:38:01 +0200 | [diff] [blame] | 6022 | struct sched_domain_shared *sds = NULL; |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 6023 | struct sched_domain *sd; |
| 6024 | int id = cpu; |
Peter Zijlstra | 7d9ffa8 | 2013-07-04 12:56:46 +0800 | [diff] [blame] | 6025 | int size = 1; |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 6026 | |
| 6027 | sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES); |
Peter Zijlstra | 7d9ffa8 | 2013-07-04 12:56:46 +0800 | [diff] [blame] | 6028 | if (sd) { |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 6029 | id = cpumask_first(sched_domain_span(sd)); |
Peter Zijlstra | 7d9ffa8 | 2013-07-04 12:56:46 +0800 | [diff] [blame] | 6030 | size = cpumask_weight(sched_domain_span(sd)); |
Peter Zijlstra | 0e369d7 | 2016-05-09 10:38:01 +0200 | [diff] [blame] | 6031 | sds = sd->shared; |
Peter Zijlstra | 7d9ffa8 | 2013-07-04 12:56:46 +0800 | [diff] [blame] | 6032 | } |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 6033 | |
| 6034 | rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); |
Peter Zijlstra | 7d9ffa8 | 2013-07-04 12:56:46 +0800 | [diff] [blame] | 6035 | per_cpu(sd_llc_size, cpu) = size; |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 6036 | per_cpu(sd_llc_id, cpu) = id; |
Peter Zijlstra | 0e369d7 | 2016-05-09 10:38:01 +0200 | [diff] [blame] | 6037 | rcu_assign_pointer(per_cpu(sd_llc_shared, cpu), sds); |
Mel Gorman | fb13c7e | 2013-10-07 11:29:17 +0100 | [diff] [blame] | 6038 | |
| 6039 | sd = lowest_flag_domain(cpu, SD_NUMA); |
| 6040 | rcu_assign_pointer(per_cpu(sd_numa, cpu), sd); |
Preeti U Murthy | 37dc6b5 | 2013-10-30 08:42:52 +0530 | [diff] [blame] | 6041 | |
| 6042 | sd = highest_flag_domain(cpu, SD_ASYM_PACKING); |
| 6043 | rcu_assign_pointer(per_cpu(sd_asym, cpu), sd); |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 6044 | } |
| 6045 | |
| 6046 | /* |
Ingo Molnar | 0eab914 | 2008-01-25 21:08:19 +0100 | [diff] [blame] | 6047 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6048 | * hold the hotplug lock. |
| 6049 | */ |
Ingo Molnar | 0eab914 | 2008-01-25 21:08:19 +0100 | [diff] [blame] | 6050 | static void |
| 6051 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6052 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 6053 | struct rq *rq = cpu_rq(cpu); |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 6054 | struct sched_domain *tmp; |
| 6055 | |
| 6056 | /* Remove the sched domains which do not contribute to scheduling. */ |
Li Zefan | f29c9b1 | 2008-11-06 09:45:16 +0800 | [diff] [blame] | 6057 | for (tmp = sd; tmp; ) { |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 6058 | struct sched_domain *parent = tmp->parent; |
| 6059 | if (!parent) |
| 6060 | break; |
Li Zefan | f29c9b1 | 2008-11-06 09:45:16 +0800 | [diff] [blame] | 6061 | |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 6062 | if (sd_parent_degenerate(tmp, parent)) { |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 6063 | tmp->parent = parent->parent; |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 6064 | if (parent->parent) |
| 6065 | parent->parent->child = tmp; |
Peter Zijlstra | 10866e62 | 2013-08-19 16:57:04 +0200 | [diff] [blame] | 6066 | /* |
| 6067 | * Transfer SD_PREFER_SIBLING down in case of a |
| 6068 | * degenerate parent; the spans match for this |
| 6069 | * so the property transfers. |
| 6070 | */ |
| 6071 | if (parent->flags & SD_PREFER_SIBLING) |
| 6072 | tmp->flags |= SD_PREFER_SIBLING; |
Peter Zijlstra | f39180e | 2016-05-09 10:37:54 +0200 | [diff] [blame] | 6073 | destroy_sched_domain(parent); |
Li Zefan | f29c9b1 | 2008-11-06 09:45:16 +0800 | [diff] [blame] | 6074 | } else |
| 6075 | tmp = tmp->parent; |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 6076 | } |
| 6077 | |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 6078 | if (sd && sd_degenerate(sd)) { |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6079 | tmp = sd; |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 6080 | sd = sd->parent; |
Peter Zijlstra | f39180e | 2016-05-09 10:37:54 +0200 | [diff] [blame] | 6081 | destroy_sched_domain(tmp); |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 6082 | if (sd) |
| 6083 | sd->child = NULL; |
| 6084 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6085 | |
Peter Zijlstra | 4cb9883 | 2011-04-07 14:09:58 +0200 | [diff] [blame] | 6086 | sched_domain_debug(sd, cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6087 | |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 6088 | rq_attach_root(rq, rd); |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6089 | tmp = rq->sd; |
Nick Piggin | 674311d | 2005-06-25 14:57:27 -0700 | [diff] [blame] | 6090 | rcu_assign_pointer(rq->sd, sd); |
Peter Zijlstra | f39180e | 2016-05-09 10:37:54 +0200 | [diff] [blame] | 6091 | destroy_sched_domains(tmp); |
Peter Zijlstra | 518cd62 | 2011-12-07 15:07:31 +0100 | [diff] [blame] | 6092 | |
| 6093 | update_top_cache_domain(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6094 | } |
| 6095 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6096 | /* Setup the mask of cpus configured for isolated domains */ |
| 6097 | static int __init isolated_cpu_setup(char *str) |
| 6098 | { |
Prarit Bhargava | a6e4491 | 2016-02-04 09:38:00 -0500 | [diff] [blame] | 6099 | int ret; |
| 6100 | |
Rusty Russell | bdddd29 | 2009-12-02 14:09:16 +1030 | [diff] [blame] | 6101 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
Prarit Bhargava | a6e4491 | 2016-02-04 09:38:00 -0500 | [diff] [blame] | 6102 | ret = cpulist_parse(str, cpu_isolated_map); |
| 6103 | if (ret) { |
| 6104 | pr_err("sched: Error, all isolcpus= values must be between 0 and %d\n", nr_cpu_ids); |
| 6105 | return 0; |
| 6106 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6107 | return 1; |
| 6108 | } |
Ingo Molnar | 8927f49 | 2007-10-15 17:00:13 +0200 | [diff] [blame] | 6109 | __setup("isolcpus=", isolated_cpu_setup); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6110 | |
Andreas Herrmann | 49a02c5 | 2009-08-18 12:51:52 +0200 | [diff] [blame] | 6111 | struct s_data { |
Peter Zijlstra | 21d42cc | 2011-04-07 14:09:48 +0200 | [diff] [blame] | 6112 | struct sched_domain ** __percpu sd; |
Andreas Herrmann | 49a02c5 | 2009-08-18 12:51:52 +0200 | [diff] [blame] | 6113 | struct root_domain *rd; |
| 6114 | }; |
| 6115 | |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 6116 | enum s_alloc { |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 6117 | sa_rootdomain, |
Peter Zijlstra | 21d42cc | 2011-04-07 14:09:48 +0200 | [diff] [blame] | 6118 | sa_sd, |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6119 | sa_sd_storage, |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 6120 | sa_none, |
| 6121 | }; |
| 6122 | |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 6123 | /* |
| 6124 | * Build an iteration mask that can exclude certain CPUs from the upwards |
| 6125 | * domain traversal. |
| 6126 | * |
Peter Zijlstra | 758dc6a | 2017-04-25 14:00:49 +0200 | [diff] [blame] | 6127 | * Only CPUs that can arrive at this group should be considered to continue |
| 6128 | * balancing. |
| 6129 | * |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 6130 | * Asymmetric node setups can result in situations where the domain tree is of |
| 6131 | * unequal depth, make sure to skip domains that already cover the entire |
| 6132 | * range. |
| 6133 | * |
| 6134 | * In that case build_sched_domains() will have terminated the iteration early |
| 6135 | * and our sibling sd spans will be empty. Domains should always include the |
| 6136 | * cpu they're built on, so check that. |
| 6137 | * |
| 6138 | */ |
| 6139 | static void build_group_mask(struct sched_domain *sd, struct sched_group *sg) |
| 6140 | { |
Lauro Ramos Venancio | 3e165b2 | 2017-04-20 16:51:40 -0300 | [diff] [blame] | 6141 | const struct cpumask *sg_span = sched_group_cpus(sg); |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 6142 | struct sd_data *sdd = sd->private; |
| 6143 | struct sched_domain *sibling; |
| 6144 | int i; |
| 6145 | |
Lauro Ramos Venancio | 3e165b2 | 2017-04-20 16:51:40 -0300 | [diff] [blame] | 6146 | for_each_cpu(i, sg_span) { |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 6147 | sibling = *per_cpu_ptr(sdd->sd, i); |
Peter Zijlstra | 758dc6a | 2017-04-25 14:00:49 +0200 | [diff] [blame] | 6148 | |
| 6149 | /* |
| 6150 | * Can happen in the asymmetric case, where these siblings are |
| 6151 | * unused. The mask will not be empty because those CPUs that |
| 6152 | * do have the top domain _should_ span the domain. |
| 6153 | */ |
| 6154 | if (!sibling->child) |
| 6155 | continue; |
| 6156 | |
| 6157 | /* If we would not end up here, we can't continue from here */ |
| 6158 | if (!cpumask_equal(sg_span, sched_domain_span(sibling->child))) |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 6159 | continue; |
| 6160 | |
| 6161 | cpumask_set_cpu(i, sched_group_mask(sg)); |
| 6162 | } |
Peter Zijlstra | 758dc6a | 2017-04-25 14:00:49 +0200 | [diff] [blame] | 6163 | |
| 6164 | /* We must not have empty masks here */ |
| 6165 | WARN_ON_ONCE(cpumask_empty(sched_group_mask(sg))); |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 6166 | } |
| 6167 | |
| 6168 | /* |
| 6169 | * Return the canonical balance cpu for this group, this is the first cpu |
| 6170 | * of this group that's also in the iteration mask. |
| 6171 | */ |
| 6172 | int group_balance_cpu(struct sched_group *sg) |
| 6173 | { |
| 6174 | return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg)); |
| 6175 | } |
| 6176 | |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6177 | static int |
| 6178 | build_overlap_sched_groups(struct sched_domain *sd, int cpu) |
| 6179 | { |
| 6180 | struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg; |
| 6181 | const struct cpumask *span = sched_domain_span(sd); |
| 6182 | struct cpumask *covered = sched_domains_tmpmask; |
| 6183 | struct sd_data *sdd = sd->private; |
Zhihui Zhang | aaecac4 | 2014-08-01 21:18:03 -0400 | [diff] [blame] | 6184 | struct sched_domain *sibling; |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6185 | int i; |
| 6186 | |
| 6187 | cpumask_clear(covered); |
| 6188 | |
Peter Zijlstra | 7c3f08e | 2017-04-14 17:24:02 +0200 | [diff] [blame] | 6189 | for_each_cpu_wrap(i, span, cpu) { |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6190 | struct cpumask *sg_span; |
| 6191 | |
| 6192 | if (cpumask_test_cpu(i, covered)) |
| 6193 | continue; |
| 6194 | |
Zhihui Zhang | aaecac4 | 2014-08-01 21:18:03 -0400 | [diff] [blame] | 6195 | sibling = *per_cpu_ptr(sdd->sd, i); |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 6196 | |
| 6197 | /* See the comment near build_group_mask(). */ |
Zhihui Zhang | aaecac4 | 2014-08-01 21:18:03 -0400 | [diff] [blame] | 6198 | if (!cpumask_test_cpu(i, sched_domain_span(sibling))) |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 6199 | continue; |
| 6200 | |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6201 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
Suresh Siddha | 4d78a22 | 2011-11-18 15:03:29 -0800 | [diff] [blame] | 6202 | GFP_KERNEL, cpu_to_node(cpu)); |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6203 | |
| 6204 | if (!sg) |
| 6205 | goto fail; |
| 6206 | |
| 6207 | sg_span = sched_group_cpus(sg); |
Zhihui Zhang | aaecac4 | 2014-08-01 21:18:03 -0400 | [diff] [blame] | 6208 | if (sibling->child) |
| 6209 | cpumask_copy(sg_span, sched_domain_span(sibling->child)); |
| 6210 | else |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6211 | cpumask_set_cpu(i, sg_span); |
| 6212 | |
| 6213 | cpumask_or(covered, covered, sg_span); |
| 6214 | |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 6215 | sg->sgc = *per_cpu_ptr(sdd->sgc, i); |
| 6216 | if (atomic_inc_return(&sg->sgc->ref) == 1) |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 6217 | build_group_mask(sd, sg); |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6218 | |
Peter Zijlstra | c3decf0 | 2012-05-31 12:05:32 +0200 | [diff] [blame] | 6219 | /* |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 6220 | * Initialize sgc->capacity such that even if we mess up the |
Peter Zijlstra | c3decf0 | 2012-05-31 12:05:32 +0200 | [diff] [blame] | 6221 | * domains and no possible iteration will get us here, we won't |
| 6222 | * die on a /0 trap. |
| 6223 | */ |
Nicolas Pitre | ca8ce3d | 2014-05-26 18:19:39 -0400 | [diff] [blame] | 6224 | sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span); |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 6225 | |
| 6226 | /* |
| 6227 | * Make sure the first group of this domain contains the |
| 6228 | * canonical balance cpu. Otherwise the sched_domain iteration |
| 6229 | * breaks. See update_sg_lb_stats(). |
| 6230 | */ |
Peter Zijlstra | 74a5ce2 | 2012-05-23 18:00:43 +0200 | [diff] [blame] | 6231 | if ((!groups && cpumask_test_cpu(cpu, sg_span)) || |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 6232 | group_balance_cpu(sg) == cpu) |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6233 | groups = sg; |
| 6234 | |
| 6235 | if (!first) |
| 6236 | first = sg; |
| 6237 | if (last) |
| 6238 | last->next = sg; |
| 6239 | last = sg; |
| 6240 | last->next = first; |
| 6241 | } |
| 6242 | sd->groups = groups; |
| 6243 | |
| 6244 | return 0; |
| 6245 | |
| 6246 | fail: |
| 6247 | free_sched_groups(first, 0); |
| 6248 | |
| 6249 | return -ENOMEM; |
| 6250 | } |
| 6251 | |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6252 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6253 | { |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6254 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
| 6255 | struct sched_domain *child = sd->child; |
| 6256 | |
| 6257 | if (child) |
| 6258 | cpu = cpumask_first(sched_domain_span(child)); |
| 6259 | |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 6260 | if (sg) { |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6261 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 6262 | (*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu); |
| 6263 | atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */ |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 6264 | } |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6265 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6266 | return cpu; |
| 6267 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6268 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 6269 | /* |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6270 | * build_sched_groups will build a circular linked list of the groups |
| 6271 | * covered by the given span, and will set each group's ->cpumask correctly, |
Nicolas Pitre | ced549f | 2014-05-26 18:19:38 -0400 | [diff] [blame] | 6272 | * and ->cpu_capacity to 0. |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6273 | * |
| 6274 | * Assumes the sched_domain tree is fully constructed |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 6275 | */ |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6276 | static int |
| 6277 | build_sched_groups(struct sched_domain *sd, int cpu) |
Siddha, Suresh B | 1e9f28f | 2006-03-27 01:15:22 -0800 | [diff] [blame] | 6278 | { |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6279 | struct sched_group *first = NULL, *last = NULL; |
| 6280 | struct sd_data *sdd = sd->private; |
| 6281 | const struct cpumask *span = sched_domain_span(sd); |
Peter Zijlstra | f96225f | 2011-04-07 14:09:57 +0200 | [diff] [blame] | 6282 | struct cpumask *covered; |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6283 | int i; |
| 6284 | |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6285 | get_group(cpu, sdd, &sd->groups); |
| 6286 | atomic_inc(&sd->groups->ref); |
| 6287 | |
Viresh Kumar | 0936629 | 2013-06-11 16:32:43 +0530 | [diff] [blame] | 6288 | if (cpu != cpumask_first(span)) |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6289 | return 0; |
| 6290 | |
Peter Zijlstra | f96225f | 2011-04-07 14:09:57 +0200 | [diff] [blame] | 6291 | lockdep_assert_held(&sched_domains_mutex); |
| 6292 | covered = sched_domains_tmpmask; |
| 6293 | |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6294 | cpumask_clear(covered); |
| 6295 | |
| 6296 | for_each_cpu(i, span) { |
| 6297 | struct sched_group *sg; |
Viresh Kumar | cd08e92 | 2013-06-11 16:32:44 +0530 | [diff] [blame] | 6298 | int group, j; |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6299 | |
| 6300 | if (cpumask_test_cpu(i, covered)) |
| 6301 | continue; |
| 6302 | |
Viresh Kumar | cd08e92 | 2013-06-11 16:32:44 +0530 | [diff] [blame] | 6303 | group = get_group(i, sdd, &sg); |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 6304 | cpumask_setall(sched_group_mask(sg)); |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6305 | |
| 6306 | for_each_cpu(j, span) { |
| 6307 | if (get_group(j, sdd, NULL) != group) |
| 6308 | continue; |
| 6309 | |
| 6310 | cpumask_set_cpu(j, covered); |
| 6311 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
| 6312 | } |
| 6313 | |
| 6314 | if (!first) |
| 6315 | first = sg; |
| 6316 | if (last) |
| 6317 | last->next = sg; |
| 6318 | last = sg; |
| 6319 | } |
| 6320 | last->next = first; |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6321 | |
| 6322 | return 0; |
Siddha, Suresh B | 1e9f28f | 2006-03-27 01:15:22 -0800 | [diff] [blame] | 6323 | } |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 6324 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6325 | /* |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 6326 | * Initialize sched groups cpu_capacity. |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 6327 | * |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 6328 | * cpu_capacity indicates the capacity of sched group, which is used while |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 6329 | * distributing the load between different sched groups in a sched domain. |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 6330 | * Typically cpu_capacity for all the groups in a sched domain will be same |
| 6331 | * unless there are asymmetries in the topology. If there are asymmetries, |
| 6332 | * group having more cpu_capacity will pickup more load compared to the |
| 6333 | * group having less cpu_capacity. |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 6334 | */ |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 6335 | static void init_sched_groups_capacity(int cpu, struct sched_domain *sd) |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 6336 | { |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6337 | struct sched_group *sg = sd->groups; |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 6338 | |
Viresh Kumar | 94c95ba | 2013-06-11 16:32:45 +0530 | [diff] [blame] | 6339 | WARN_ON(!sg); |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6340 | |
| 6341 | do { |
| 6342 | sg->group_weight = cpumask_weight(sched_group_cpus(sg)); |
| 6343 | sg = sg->next; |
| 6344 | } while (sg != sd->groups); |
| 6345 | |
Peter Zijlstra | c117487 | 2012-05-31 14:47:33 +0200 | [diff] [blame] | 6346 | if (cpu != group_balance_cpu(sg)) |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 6347 | return; |
| 6348 | |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 6349 | update_group_capacity(sd, cpu); |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 6350 | } |
| 6351 | |
| 6352 | /* |
Mike Travis | 7c16ec5 | 2008-04-04 18:11:11 -0700 | [diff] [blame] | 6353 | * Initializers for schedule domains |
| 6354 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() |
| 6355 | */ |
| 6356 | |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 6357 | static int default_relax_domain_level = -1; |
Peter Zijlstra | 60495e7 | 2011-04-07 14:10:04 +0200 | [diff] [blame] | 6358 | int sched_domain_level_max; |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 6359 | |
| 6360 | static int __init setup_relax_domain_level(char *str) |
| 6361 | { |
Dimitri Sivanich | a841f8c | 2012-06-05 13:44:36 -0500 | [diff] [blame] | 6362 | if (kstrtoint(str, 0, &default_relax_domain_level)) |
| 6363 | pr_warn("Unable to set relax_domain_level\n"); |
Li Zefan | 30e0e17 | 2008-05-13 10:27:17 +0800 | [diff] [blame] | 6364 | |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 6365 | return 1; |
| 6366 | } |
| 6367 | __setup("relax_domain_level=", setup_relax_domain_level); |
| 6368 | |
| 6369 | static void set_domain_attribute(struct sched_domain *sd, |
| 6370 | struct sched_domain_attr *attr) |
| 6371 | { |
| 6372 | int request; |
| 6373 | |
| 6374 | if (!attr || attr->relax_domain_level < 0) { |
| 6375 | if (default_relax_domain_level < 0) |
| 6376 | return; |
| 6377 | else |
| 6378 | request = default_relax_domain_level; |
| 6379 | } else |
| 6380 | request = attr->relax_domain_level; |
| 6381 | if (request < sd->level) { |
| 6382 | /* turn off idle balance on this domain */ |
Peter Zijlstra | c88d591 | 2009-09-10 13:50:02 +0200 | [diff] [blame] | 6383 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 6384 | } else { |
| 6385 | /* turn on idle balance on this domain */ |
Peter Zijlstra | c88d591 | 2009-09-10 13:50:02 +0200 | [diff] [blame] | 6386 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 6387 | } |
| 6388 | } |
| 6389 | |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6390 | static void __sdt_free(const struct cpumask *cpu_map); |
| 6391 | static int __sdt_alloc(const struct cpumask *cpu_map); |
| 6392 | |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 6393 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
| 6394 | const struct cpumask *cpu_map) |
| 6395 | { |
| 6396 | switch (what) { |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 6397 | case sa_rootdomain: |
Peter Zijlstra | 822ff79 | 2011-04-07 14:09:51 +0200 | [diff] [blame] | 6398 | if (!atomic_read(&d->rd->refcount)) |
| 6399 | free_rootdomain(&d->rd->rcu); /* fall through */ |
Peter Zijlstra | 21d42cc | 2011-04-07 14:09:48 +0200 | [diff] [blame] | 6400 | case sa_sd: |
| 6401 | free_percpu(d->sd); /* fall through */ |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6402 | case sa_sd_storage: |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6403 | __sdt_free(cpu_map); /* fall through */ |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 6404 | case sa_none: |
| 6405 | break; |
| 6406 | } |
| 6407 | } |
| 6408 | |
| 6409 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
| 6410 | const struct cpumask *cpu_map) |
| 6411 | { |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6412 | memset(d, 0, sizeof(*d)); |
| 6413 | |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6414 | if (__sdt_alloc(cpu_map)) |
| 6415 | return sa_sd_storage; |
Peter Zijlstra | 21d42cc | 2011-04-07 14:09:48 +0200 | [diff] [blame] | 6416 | d->sd = alloc_percpu(struct sched_domain *); |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6417 | if (!d->sd) |
| 6418 | return sa_sd_storage; |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 6419 | d->rd = alloc_rootdomain(); |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6420 | if (!d->rd) |
Peter Zijlstra | 21d42cc | 2011-04-07 14:09:48 +0200 | [diff] [blame] | 6421 | return sa_sd; |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 6422 | return sa_rootdomain; |
| 6423 | } |
| 6424 | |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6425 | /* |
| 6426 | * NULL the sd_data elements we've used to build the sched_domain and |
| 6427 | * sched_group structure so that the subsequent __free_domain_allocs() |
| 6428 | * will not free the data we're using. |
| 6429 | */ |
| 6430 | static void claim_allocations(int cpu, struct sched_domain *sd) |
| 6431 | { |
| 6432 | struct sd_data *sdd = sd->private; |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6433 | |
| 6434 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); |
| 6435 | *per_cpu_ptr(sdd->sd, cpu) = NULL; |
| 6436 | |
Peter Zijlstra | 24fc7ed | 2016-05-09 10:37:59 +0200 | [diff] [blame] | 6437 | if (atomic_read(&(*per_cpu_ptr(sdd->sds, cpu))->ref)) |
| 6438 | *per_cpu_ptr(sdd->sds, cpu) = NULL; |
| 6439 | |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6440 | if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref)) |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6441 | *per_cpu_ptr(sdd->sg, cpu) = NULL; |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 6442 | |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 6443 | if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref)) |
| 6444 | *per_cpu_ptr(sdd->sgc, cpu) = NULL; |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 6445 | } |
| 6446 | |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6447 | #ifdef CONFIG_NUMA |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6448 | static int sched_domains_numa_levels; |
Rik van Riel | e3fe70b | 2014-10-17 03:29:50 -0400 | [diff] [blame] | 6449 | enum numa_topology_type sched_numa_topology_type; |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6450 | static int *sched_domains_numa_distance; |
Rik van Riel | 9942f79 | 2014-10-17 03:29:49 -0400 | [diff] [blame] | 6451 | int sched_max_numa_distance; |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6452 | static struct cpumask ***sched_domains_numa_masks; |
| 6453 | static int sched_domains_curr_level; |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6454 | #endif |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6455 | |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6456 | /* |
| 6457 | * SD_flags allowed in topology descriptions. |
| 6458 | * |
Peter Zijlstra | 94f438c | 2016-08-15 12:54:59 +0200 | [diff] [blame] | 6459 | * These flags are purely descriptive of the topology and do not prescribe |
| 6460 | * behaviour. Behaviour is artificial and mapped in the below sd_init() |
| 6461 | * function: |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6462 | * |
Peter Zijlstra | 94f438c | 2016-08-15 12:54:59 +0200 | [diff] [blame] | 6463 | * SD_SHARE_CPUCAPACITY - describes SMT topologies |
| 6464 | * SD_SHARE_PKG_RESOURCES - describes shared caches |
| 6465 | * SD_NUMA - describes NUMA topologies |
| 6466 | * SD_SHARE_POWERDOMAIN - describes shared power domain |
Morten Rasmussen | 1f6e6c7 | 2016-07-25 14:34:22 +0100 | [diff] [blame] | 6467 | * SD_ASYM_CPUCAPACITY - describes mixed capacity topologies |
Peter Zijlstra | 94f438c | 2016-08-15 12:54:59 +0200 | [diff] [blame] | 6468 | * |
| 6469 | * Odd one out, which beside describing the topology has a quirk also |
| 6470 | * prescribes the desired behaviour that goes along with it: |
| 6471 | * |
| 6472 | * SD_ASYM_PACKING - describes SMT quirks |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6473 | */ |
| 6474 | #define TOPOLOGY_SD_FLAGS \ |
Nicolas Pitre | 5d4dfdd | 2014-05-27 13:50:41 -0400 | [diff] [blame] | 6475 | (SD_SHARE_CPUCAPACITY | \ |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6476 | SD_SHARE_PKG_RESOURCES | \ |
| 6477 | SD_NUMA | \ |
Vincent Guittot | d77b3ed | 2014-04-11 11:44:40 +0200 | [diff] [blame] | 6478 | SD_ASYM_PACKING | \ |
Morten Rasmussen | 1f6e6c7 | 2016-07-25 14:34:22 +0100 | [diff] [blame] | 6479 | SD_ASYM_CPUCAPACITY | \ |
Vincent Guittot | d77b3ed | 2014-04-11 11:44:40 +0200 | [diff] [blame] | 6480 | SD_SHARE_POWERDOMAIN) |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6481 | |
| 6482 | static struct sched_domain * |
Morten Rasmussen | 3676b13 | 2016-07-25 14:34:23 +0100 | [diff] [blame] | 6483 | sd_init(struct sched_domain_topology_level *tl, |
Peter Zijlstra | 24fc7ed | 2016-05-09 10:37:59 +0200 | [diff] [blame] | 6484 | const struct cpumask *cpu_map, |
Morten Rasmussen | 3676b13 | 2016-07-25 14:34:23 +0100 | [diff] [blame] | 6485 | struct sched_domain *child, int cpu) |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6486 | { |
Peter Zijlstra | 24fc7ed | 2016-05-09 10:37:59 +0200 | [diff] [blame] | 6487 | struct sd_data *sdd = &tl->data; |
| 6488 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
| 6489 | int sd_id, sd_weight, sd_flags = 0; |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6490 | |
| 6491 | #ifdef CONFIG_NUMA |
| 6492 | /* |
| 6493 | * Ugly hack to pass state to sd_numa_mask()... |
| 6494 | */ |
| 6495 | sched_domains_curr_level = tl->numa_level; |
| 6496 | #endif |
| 6497 | |
| 6498 | sd_weight = cpumask_weight(tl->mask(cpu)); |
| 6499 | |
| 6500 | if (tl->sd_flags) |
| 6501 | sd_flags = (*tl->sd_flags)(); |
| 6502 | if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS, |
| 6503 | "wrong sd_flags in topology description\n")) |
| 6504 | sd_flags &= ~TOPOLOGY_SD_FLAGS; |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6505 | |
| 6506 | *sd = (struct sched_domain){ |
| 6507 | .min_interval = sd_weight, |
| 6508 | .max_interval = 2*sd_weight, |
| 6509 | .busy_factor = 32, |
Peter Zijlstra | 870a0bb | 2012-05-11 00:26:27 +0200 | [diff] [blame] | 6510 | .imbalance_pct = 125, |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6511 | |
| 6512 | .cache_nice_tries = 0, |
| 6513 | .busy_idx = 0, |
| 6514 | .idle_idx = 0, |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6515 | .newidle_idx = 0, |
| 6516 | .wake_idx = 0, |
| 6517 | .forkexec_idx = 0, |
| 6518 | |
| 6519 | .flags = 1*SD_LOAD_BALANCE |
| 6520 | | 1*SD_BALANCE_NEWIDLE |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6521 | | 1*SD_BALANCE_EXEC |
| 6522 | | 1*SD_BALANCE_FORK |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6523 | | 0*SD_BALANCE_WAKE |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6524 | | 1*SD_WAKE_AFFINE |
Nicolas Pitre | 5d4dfdd | 2014-05-27 13:50:41 -0400 | [diff] [blame] | 6525 | | 0*SD_SHARE_CPUCAPACITY |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6526 | | 0*SD_SHARE_PKG_RESOURCES |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6527 | | 0*SD_SERIALIZE |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6528 | | 0*SD_PREFER_SIBLING |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6529 | | 0*SD_NUMA |
| 6530 | | sd_flags |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6531 | , |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6532 | |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6533 | .last_balance = jiffies, |
| 6534 | .balance_interval = sd_weight, |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6535 | .smt_gain = 0, |
Jason Low | 2b4cfe6 | 2014-04-23 18:30:34 -0700 | [diff] [blame] | 6536 | .max_newidle_lb_cost = 0, |
| 6537 | .next_decay_max_lb_cost = jiffies, |
Morten Rasmussen | 3676b13 | 2016-07-25 14:34:23 +0100 | [diff] [blame] | 6538 | .child = child, |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6539 | #ifdef CONFIG_SCHED_DEBUG |
| 6540 | .name = tl->name, |
| 6541 | #endif |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6542 | }; |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6543 | |
Peter Zijlstra | 24fc7ed | 2016-05-09 10:37:59 +0200 | [diff] [blame] | 6544 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); |
| 6545 | sd_id = cpumask_first(sched_domain_span(sd)); |
| 6546 | |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6547 | /* |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6548 | * Convert topological properties into behaviour. |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6549 | */ |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6550 | |
Morten Rasmussen | 9ee1cda | 2016-07-25 14:34:24 +0100 | [diff] [blame] | 6551 | if (sd->flags & SD_ASYM_CPUCAPACITY) { |
| 6552 | struct sched_domain *t = sd; |
| 6553 | |
| 6554 | for_each_lower_domain(t) |
| 6555 | t->flags |= SD_BALANCE_WAKE; |
| 6556 | } |
| 6557 | |
Nicolas Pitre | 5d4dfdd | 2014-05-27 13:50:41 -0400 | [diff] [blame] | 6558 | if (sd->flags & SD_SHARE_CPUCAPACITY) { |
Vincent Guittot | caff37e | 2015-02-27 16:54:13 +0100 | [diff] [blame] | 6559 | sd->flags |= SD_PREFER_SIBLING; |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6560 | sd->imbalance_pct = 110; |
| 6561 | sd->smt_gain = 1178; /* ~15% */ |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6562 | |
| 6563 | } else if (sd->flags & SD_SHARE_PKG_RESOURCES) { |
| 6564 | sd->imbalance_pct = 117; |
| 6565 | sd->cache_nice_tries = 1; |
| 6566 | sd->busy_idx = 2; |
| 6567 | |
| 6568 | #ifdef CONFIG_NUMA |
| 6569 | } else if (sd->flags & SD_NUMA) { |
| 6570 | sd->cache_nice_tries = 2; |
| 6571 | sd->busy_idx = 3; |
| 6572 | sd->idle_idx = 2; |
| 6573 | |
| 6574 | sd->flags |= SD_SERIALIZE; |
| 6575 | if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) { |
| 6576 | sd->flags &= ~(SD_BALANCE_EXEC | |
| 6577 | SD_BALANCE_FORK | |
| 6578 | SD_WAKE_AFFINE); |
| 6579 | } |
| 6580 | |
| 6581 | #endif |
| 6582 | } else { |
| 6583 | sd->flags |= SD_PREFER_SIBLING; |
| 6584 | sd->cache_nice_tries = 1; |
| 6585 | sd->busy_idx = 2; |
| 6586 | sd->idle_idx = 1; |
| 6587 | } |
| 6588 | |
Peter Zijlstra | 24fc7ed | 2016-05-09 10:37:59 +0200 | [diff] [blame] | 6589 | /* |
| 6590 | * For all levels sharing cache; connect a sched_domain_shared |
| 6591 | * instance. |
| 6592 | */ |
| 6593 | if (sd->flags & SD_SHARE_PKG_RESOURCES) { |
| 6594 | sd->shared = *per_cpu_ptr(sdd->sds, sd_id); |
| 6595 | atomic_inc(&sd->shared->ref); |
Peter Zijlstra | 0e369d7 | 2016-05-09 10:38:01 +0200 | [diff] [blame] | 6596 | atomic_set(&sd->shared->nr_busy_cpus, sd_weight); |
Peter Zijlstra | 24fc7ed | 2016-05-09 10:37:59 +0200 | [diff] [blame] | 6597 | } |
| 6598 | |
| 6599 | sd->private = sdd; |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6600 | |
| 6601 | return sd; |
| 6602 | } |
| 6603 | |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6604 | /* |
| 6605 | * Topology list, bottom-up. |
| 6606 | */ |
| 6607 | static struct sched_domain_topology_level default_topology[] = { |
| 6608 | #ifdef CONFIG_SCHED_SMT |
| 6609 | { cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) }, |
| 6610 | #endif |
| 6611 | #ifdef CONFIG_SCHED_MC |
| 6612 | { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) }, |
| 6613 | #endif |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6614 | { cpu_cpu_mask, SD_INIT_NAME(DIE) }, |
| 6615 | { NULL, }, |
| 6616 | }; |
| 6617 | |
Juergen Gross | c6e1e7b | 2015-09-22 12:48:59 +0200 | [diff] [blame] | 6618 | static struct sched_domain_topology_level *sched_domain_topology = |
| 6619 | default_topology; |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6620 | |
| 6621 | #define for_each_sd_topology(tl) \ |
| 6622 | for (tl = sched_domain_topology; tl->mask; tl++) |
| 6623 | |
| 6624 | void set_sched_topology(struct sched_domain_topology_level *tl) |
| 6625 | { |
Tim Chen | 8f37961 | 2016-09-21 12:19:03 -0700 | [diff] [blame] | 6626 | if (WARN_ON_ONCE(sched_smp_initialized)) |
| 6627 | return; |
| 6628 | |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6629 | sched_domain_topology = tl; |
| 6630 | } |
| 6631 | |
| 6632 | #ifdef CONFIG_NUMA |
| 6633 | |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6634 | static const struct cpumask *sd_numa_mask(int cpu) |
| 6635 | { |
| 6636 | return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)]; |
| 6637 | } |
| 6638 | |
Peter Zijlstra | d039ac6 | 2012-05-31 21:20:16 +0200 | [diff] [blame] | 6639 | static void sched_numa_warn(const char *str) |
| 6640 | { |
| 6641 | static int done = false; |
| 6642 | int i,j; |
| 6643 | |
| 6644 | if (done) |
| 6645 | return; |
| 6646 | |
| 6647 | done = true; |
| 6648 | |
| 6649 | printk(KERN_WARNING "ERROR: %s\n\n", str); |
| 6650 | |
| 6651 | for (i = 0; i < nr_node_ids; i++) { |
| 6652 | printk(KERN_WARNING " "); |
| 6653 | for (j = 0; j < nr_node_ids; j++) |
| 6654 | printk(KERN_CONT "%02d ", node_distance(i,j)); |
| 6655 | printk(KERN_CONT "\n"); |
| 6656 | } |
| 6657 | printk(KERN_WARNING "\n"); |
| 6658 | } |
| 6659 | |
Rik van Riel | 9942f79 | 2014-10-17 03:29:49 -0400 | [diff] [blame] | 6660 | bool find_numa_distance(int distance) |
Peter Zijlstra | d039ac6 | 2012-05-31 21:20:16 +0200 | [diff] [blame] | 6661 | { |
| 6662 | int i; |
| 6663 | |
| 6664 | if (distance == node_distance(0, 0)) |
| 6665 | return true; |
| 6666 | |
| 6667 | for (i = 0; i < sched_domains_numa_levels; i++) { |
| 6668 | if (sched_domains_numa_distance[i] == distance) |
| 6669 | return true; |
| 6670 | } |
| 6671 | |
| 6672 | return false; |
| 6673 | } |
| 6674 | |
Rik van Riel | e3fe70b | 2014-10-17 03:29:50 -0400 | [diff] [blame] | 6675 | /* |
| 6676 | * A system can have three types of NUMA topology: |
| 6677 | * NUMA_DIRECT: all nodes are directly connected, or not a NUMA system |
| 6678 | * NUMA_GLUELESS_MESH: some nodes reachable through intermediary nodes |
| 6679 | * NUMA_BACKPLANE: nodes can reach other nodes through a backplane |
| 6680 | * |
| 6681 | * The difference between a glueless mesh topology and a backplane |
| 6682 | * topology lies in whether communication between not directly |
| 6683 | * connected nodes goes through intermediary nodes (where programs |
| 6684 | * could run), or through backplane controllers. This affects |
| 6685 | * placement of programs. |
| 6686 | * |
| 6687 | * The type of topology can be discerned with the following tests: |
| 6688 | * - If the maximum distance between any nodes is 1 hop, the system |
| 6689 | * is directly connected. |
| 6690 | * - If for two nodes A and B, located N > 1 hops away from each other, |
| 6691 | * there is an intermediary node C, which is < N hops away from both |
| 6692 | * nodes A and B, the system is a glueless mesh. |
| 6693 | */ |
| 6694 | static void init_numa_topology_type(void) |
| 6695 | { |
| 6696 | int a, b, c, n; |
| 6697 | |
| 6698 | n = sched_max_numa_distance; |
| 6699 | |
Aravind Gopalakrishnan | e237882 | 2015-08-10 20:20:48 -0500 | [diff] [blame] | 6700 | if (sched_domains_numa_levels <= 1) { |
Rik van Riel | e3fe70b | 2014-10-17 03:29:50 -0400 | [diff] [blame] | 6701 | sched_numa_topology_type = NUMA_DIRECT; |
Aravind Gopalakrishnan | e237882 | 2015-08-10 20:20:48 -0500 | [diff] [blame] | 6702 | return; |
| 6703 | } |
Rik van Riel | e3fe70b | 2014-10-17 03:29:50 -0400 | [diff] [blame] | 6704 | |
| 6705 | for_each_online_node(a) { |
| 6706 | for_each_online_node(b) { |
| 6707 | /* Find two nodes furthest removed from each other. */ |
| 6708 | if (node_distance(a, b) < n) |
| 6709 | continue; |
| 6710 | |
| 6711 | /* Is there an intermediary node between a and b? */ |
| 6712 | for_each_online_node(c) { |
| 6713 | if (node_distance(a, c) < n && |
| 6714 | node_distance(b, c) < n) { |
| 6715 | sched_numa_topology_type = |
| 6716 | NUMA_GLUELESS_MESH; |
| 6717 | return; |
| 6718 | } |
| 6719 | } |
| 6720 | |
| 6721 | sched_numa_topology_type = NUMA_BACKPLANE; |
| 6722 | return; |
| 6723 | } |
| 6724 | } |
| 6725 | } |
| 6726 | |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6727 | static void sched_init_numa(void) |
| 6728 | { |
| 6729 | int next_distance, curr_distance = node_distance(0, 0); |
| 6730 | struct sched_domain_topology_level *tl; |
| 6731 | int level = 0; |
| 6732 | int i, j, k; |
| 6733 | |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6734 | sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL); |
| 6735 | if (!sched_domains_numa_distance) |
| 6736 | return; |
| 6737 | |
| 6738 | /* |
| 6739 | * O(nr_nodes^2) deduplicating selection sort -- in order to find the |
| 6740 | * unique distances in the node_distance() table. |
| 6741 | * |
| 6742 | * Assumes node_distance(0,j) includes all distances in |
| 6743 | * node_distance(i,j) in order to avoid cubic time. |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6744 | */ |
| 6745 | next_distance = curr_distance; |
| 6746 | for (i = 0; i < nr_node_ids; i++) { |
| 6747 | for (j = 0; j < nr_node_ids; j++) { |
Peter Zijlstra | d039ac6 | 2012-05-31 21:20:16 +0200 | [diff] [blame] | 6748 | for (k = 0; k < nr_node_ids; k++) { |
| 6749 | int distance = node_distance(i, k); |
| 6750 | |
| 6751 | if (distance > curr_distance && |
| 6752 | (distance < next_distance || |
| 6753 | next_distance == curr_distance)) |
| 6754 | next_distance = distance; |
| 6755 | |
| 6756 | /* |
| 6757 | * While not a strong assumption it would be nice to know |
| 6758 | * about cases where if node A is connected to B, B is not |
| 6759 | * equally connected to A. |
| 6760 | */ |
| 6761 | if (sched_debug() && node_distance(k, i) != distance) |
| 6762 | sched_numa_warn("Node-distance not symmetric"); |
| 6763 | |
| 6764 | if (sched_debug() && i && !find_numa_distance(distance)) |
| 6765 | sched_numa_warn("Node-0 not representative"); |
| 6766 | } |
| 6767 | if (next_distance != curr_distance) { |
| 6768 | sched_domains_numa_distance[level++] = next_distance; |
| 6769 | sched_domains_numa_levels = level; |
| 6770 | curr_distance = next_distance; |
| 6771 | } else break; |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6772 | } |
Peter Zijlstra | d039ac6 | 2012-05-31 21:20:16 +0200 | [diff] [blame] | 6773 | |
| 6774 | /* |
| 6775 | * In case of sched_debug() we verify the above assumption. |
| 6776 | */ |
| 6777 | if (!sched_debug()) |
| 6778 | break; |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6779 | } |
Andrey Ryabinin | c123588 | 2014-11-07 17:53:40 +0300 | [diff] [blame] | 6780 | |
| 6781 | if (!level) |
| 6782 | return; |
| 6783 | |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6784 | /* |
| 6785 | * 'level' contains the number of unique distances, excluding the |
| 6786 | * identity distance node_distance(i,i). |
| 6787 | * |
Viresh Kumar | 28b4a52 | 2013-04-05 16:26:46 +0530 | [diff] [blame] | 6788 | * The sched_domains_numa_distance[] array includes the actual distance |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6789 | * numbers. |
| 6790 | */ |
| 6791 | |
Tang Chen | 5f7865f | 2012-09-25 21:12:30 +0800 | [diff] [blame] | 6792 | /* |
| 6793 | * Here, we should temporarily reset sched_domains_numa_levels to 0. |
| 6794 | * If it fails to allocate memory for array sched_domains_numa_masks[][], |
| 6795 | * the array will contain less then 'level' members. This could be |
| 6796 | * dangerous when we use it to iterate array sched_domains_numa_masks[][] |
| 6797 | * in other functions. |
| 6798 | * |
| 6799 | * We reset it to 'level' at the end of this function. |
| 6800 | */ |
| 6801 | sched_domains_numa_levels = 0; |
| 6802 | |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6803 | sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); |
| 6804 | if (!sched_domains_numa_masks) |
| 6805 | return; |
| 6806 | |
| 6807 | /* |
| 6808 | * Now for each level, construct a mask per node which contains all |
| 6809 | * cpus of nodes that are that many hops away from us. |
| 6810 | */ |
| 6811 | for (i = 0; i < level; i++) { |
| 6812 | sched_domains_numa_masks[i] = |
| 6813 | kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL); |
| 6814 | if (!sched_domains_numa_masks[i]) |
| 6815 | return; |
| 6816 | |
| 6817 | for (j = 0; j < nr_node_ids; j++) { |
Peter Zijlstra | 2ea4580 | 2012-05-25 09:26:43 +0200 | [diff] [blame] | 6818 | struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL); |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6819 | if (!mask) |
| 6820 | return; |
| 6821 | |
| 6822 | sched_domains_numa_masks[i][j] = mask; |
| 6823 | |
Raghavendra K T | 9c03ee1 | 2016-01-16 00:31:23 +0530 | [diff] [blame] | 6824 | for_each_node(k) { |
Peter Zijlstra | dd7d863 | 2012-05-11 00:56:20 +0200 | [diff] [blame] | 6825 | if (node_distance(j, k) > sched_domains_numa_distance[i]) |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6826 | continue; |
| 6827 | |
| 6828 | cpumask_or(mask, mask, cpumask_of_node(k)); |
| 6829 | } |
| 6830 | } |
| 6831 | } |
| 6832 | |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6833 | /* Compute default topology size */ |
| 6834 | for (i = 0; sched_domain_topology[i].mask; i++); |
| 6835 | |
Vincent Guittot | c515db8 | 2014-05-13 11:11:01 +0200 | [diff] [blame] | 6836 | tl = kzalloc((i + level + 1) * |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6837 | sizeof(struct sched_domain_topology_level), GFP_KERNEL); |
| 6838 | if (!tl) |
| 6839 | return; |
| 6840 | |
| 6841 | /* |
| 6842 | * Copy the default topology bits.. |
| 6843 | */ |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6844 | for (i = 0; sched_domain_topology[i].mask; i++) |
| 6845 | tl[i] = sched_domain_topology[i]; |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6846 | |
| 6847 | /* |
| 6848 | * .. and append 'j' levels of NUMA goodness. |
| 6849 | */ |
| 6850 | for (j = 0; j < level; i++, j++) { |
| 6851 | tl[i] = (struct sched_domain_topology_level){ |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6852 | .mask = sd_numa_mask, |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6853 | .sd_flags = cpu_numa_flags, |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6854 | .flags = SDTL_OVERLAP, |
| 6855 | .numa_level = j, |
Vincent Guittot | 143e1e2 | 2014-04-11 11:44:37 +0200 | [diff] [blame] | 6856 | SD_INIT_NAME(NUMA) |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6857 | }; |
| 6858 | } |
| 6859 | |
| 6860 | sched_domain_topology = tl; |
Tang Chen | 5f7865f | 2012-09-25 21:12:30 +0800 | [diff] [blame] | 6861 | |
| 6862 | sched_domains_numa_levels = level; |
Rik van Riel | 9942f79 | 2014-10-17 03:29:49 -0400 | [diff] [blame] | 6863 | sched_max_numa_distance = sched_domains_numa_distance[level - 1]; |
Rik van Riel | e3fe70b | 2014-10-17 03:29:50 -0400 | [diff] [blame] | 6864 | |
| 6865 | init_numa_topology_type(); |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6866 | } |
Tang Chen | 301a5cb | 2012-09-25 21:12:31 +0800 | [diff] [blame] | 6867 | |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 6868 | static void sched_domains_numa_masks_set(unsigned int cpu) |
Tang Chen | 301a5cb | 2012-09-25 21:12:31 +0800 | [diff] [blame] | 6869 | { |
Tang Chen | 301a5cb | 2012-09-25 21:12:31 +0800 | [diff] [blame] | 6870 | int node = cpu_to_node(cpu); |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 6871 | int i, j; |
Tang Chen | 301a5cb | 2012-09-25 21:12:31 +0800 | [diff] [blame] | 6872 | |
| 6873 | for (i = 0; i < sched_domains_numa_levels; i++) { |
| 6874 | for (j = 0; j < nr_node_ids; j++) { |
| 6875 | if (node_distance(j, node) <= sched_domains_numa_distance[i]) |
| 6876 | cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]); |
| 6877 | } |
| 6878 | } |
| 6879 | } |
| 6880 | |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 6881 | static void sched_domains_numa_masks_clear(unsigned int cpu) |
Tang Chen | 301a5cb | 2012-09-25 21:12:31 +0800 | [diff] [blame] | 6882 | { |
| 6883 | int i, j; |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 6884 | |
Tang Chen | 301a5cb | 2012-09-25 21:12:31 +0800 | [diff] [blame] | 6885 | for (i = 0; i < sched_domains_numa_levels; i++) { |
| 6886 | for (j = 0; j < nr_node_ids; j++) |
| 6887 | cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]); |
| 6888 | } |
| 6889 | } |
| 6890 | |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6891 | #else |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 6892 | static inline void sched_init_numa(void) { } |
| 6893 | static void sched_domains_numa_masks_set(unsigned int cpu) { } |
| 6894 | static void sched_domains_numa_masks_clear(unsigned int cpu) { } |
Peter Zijlstra | cb83b62 | 2012-04-17 15:49:36 +0200 | [diff] [blame] | 6895 | #endif /* CONFIG_NUMA */ |
| 6896 | |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6897 | static int __sdt_alloc(const struct cpumask *cpu_map) |
| 6898 | { |
| 6899 | struct sched_domain_topology_level *tl; |
| 6900 | int j; |
| 6901 | |
Viresh Kumar | 27723a6 | 2013-06-10 16:27:20 +0530 | [diff] [blame] | 6902 | for_each_sd_topology(tl) { |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6903 | struct sd_data *sdd = &tl->data; |
| 6904 | |
| 6905 | sdd->sd = alloc_percpu(struct sched_domain *); |
| 6906 | if (!sdd->sd) |
| 6907 | return -ENOMEM; |
| 6908 | |
Peter Zijlstra | 24fc7ed | 2016-05-09 10:37:59 +0200 | [diff] [blame] | 6909 | sdd->sds = alloc_percpu(struct sched_domain_shared *); |
| 6910 | if (!sdd->sds) |
| 6911 | return -ENOMEM; |
| 6912 | |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6913 | sdd->sg = alloc_percpu(struct sched_group *); |
| 6914 | if (!sdd->sg) |
| 6915 | return -ENOMEM; |
| 6916 | |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 6917 | sdd->sgc = alloc_percpu(struct sched_group_capacity *); |
| 6918 | if (!sdd->sgc) |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 6919 | return -ENOMEM; |
| 6920 | |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6921 | for_each_cpu(j, cpu_map) { |
| 6922 | struct sched_domain *sd; |
Peter Zijlstra | 24fc7ed | 2016-05-09 10:37:59 +0200 | [diff] [blame] | 6923 | struct sched_domain_shared *sds; |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6924 | struct sched_group *sg; |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 6925 | struct sched_group_capacity *sgc; |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6926 | |
Peter Zijlstra | 5cc389b | 2015-06-11 14:46:50 +0200 | [diff] [blame] | 6927 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6928 | GFP_KERNEL, cpu_to_node(j)); |
| 6929 | if (!sd) |
| 6930 | return -ENOMEM; |
| 6931 | |
| 6932 | *per_cpu_ptr(sdd->sd, j) = sd; |
| 6933 | |
Peter Zijlstra | 24fc7ed | 2016-05-09 10:37:59 +0200 | [diff] [blame] | 6934 | sds = kzalloc_node(sizeof(struct sched_domain_shared), |
| 6935 | GFP_KERNEL, cpu_to_node(j)); |
| 6936 | if (!sds) |
| 6937 | return -ENOMEM; |
| 6938 | |
| 6939 | *per_cpu_ptr(sdd->sds, j) = sds; |
| 6940 | |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6941 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), |
| 6942 | GFP_KERNEL, cpu_to_node(j)); |
| 6943 | if (!sg) |
| 6944 | return -ENOMEM; |
| 6945 | |
Igor Mammedov | 30b4e9e | 2012-05-09 12:38:28 +0200 | [diff] [blame] | 6946 | sg->next = sg; |
| 6947 | |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6948 | *per_cpu_ptr(sdd->sg, j) = sg; |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 6949 | |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 6950 | sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(), |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 6951 | GFP_KERNEL, cpu_to_node(j)); |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 6952 | if (!sgc) |
Peter Zijlstra | 9c3f75c | 2011-07-14 13:00:06 +0200 | [diff] [blame] | 6953 | return -ENOMEM; |
| 6954 | |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 6955 | *per_cpu_ptr(sdd->sgc, j) = sgc; |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6956 | } |
| 6957 | } |
| 6958 | |
| 6959 | return 0; |
| 6960 | } |
| 6961 | |
| 6962 | static void __sdt_free(const struct cpumask *cpu_map) |
| 6963 | { |
| 6964 | struct sched_domain_topology_level *tl; |
| 6965 | int j; |
| 6966 | |
Viresh Kumar | 27723a6 | 2013-06-10 16:27:20 +0530 | [diff] [blame] | 6967 | for_each_sd_topology(tl) { |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6968 | struct sd_data *sdd = &tl->data; |
| 6969 | |
| 6970 | for_each_cpu(j, cpu_map) { |
he, bo | fb2cf2c | 2012-04-25 19:59:21 +0800 | [diff] [blame] | 6971 | struct sched_domain *sd; |
| 6972 | |
| 6973 | if (sdd->sd) { |
| 6974 | sd = *per_cpu_ptr(sdd->sd, j); |
| 6975 | if (sd && (sd->flags & SD_OVERLAP)) |
| 6976 | free_sched_groups(sd->groups, 0); |
| 6977 | kfree(*per_cpu_ptr(sdd->sd, j)); |
| 6978 | } |
| 6979 | |
Peter Zijlstra | 24fc7ed | 2016-05-09 10:37:59 +0200 | [diff] [blame] | 6980 | if (sdd->sds) |
| 6981 | kfree(*per_cpu_ptr(sdd->sds, j)); |
he, bo | fb2cf2c | 2012-04-25 19:59:21 +0800 | [diff] [blame] | 6982 | if (sdd->sg) |
| 6983 | kfree(*per_cpu_ptr(sdd->sg, j)); |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 6984 | if (sdd->sgc) |
| 6985 | kfree(*per_cpu_ptr(sdd->sgc, j)); |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6986 | } |
| 6987 | free_percpu(sdd->sd); |
he, bo | fb2cf2c | 2012-04-25 19:59:21 +0800 | [diff] [blame] | 6988 | sdd->sd = NULL; |
Peter Zijlstra | 24fc7ed | 2016-05-09 10:37:59 +0200 | [diff] [blame] | 6989 | free_percpu(sdd->sds); |
| 6990 | sdd->sds = NULL; |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6991 | free_percpu(sdd->sg); |
he, bo | fb2cf2c | 2012-04-25 19:59:21 +0800 | [diff] [blame] | 6992 | sdd->sg = NULL; |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 6993 | free_percpu(sdd->sgc); |
| 6994 | sdd->sgc = NULL; |
Peter Zijlstra | 54ab4ff | 2011-04-07 14:10:03 +0200 | [diff] [blame] | 6995 | } |
| 6996 | } |
| 6997 | |
Peter Zijlstra | 2c402dc | 2011-04-07 14:10:01 +0200 | [diff] [blame] | 6998 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
Viresh Kumar | 4a850cb | 2013-06-04 16:12:43 +0530 | [diff] [blame] | 6999 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, |
| 7000 | struct sched_domain *child, int cpu) |
Peter Zijlstra | 2c402dc | 2011-04-07 14:10:01 +0200 | [diff] [blame] | 7001 | { |
Peter Zijlstra | 24fc7ed | 2016-05-09 10:37:59 +0200 | [diff] [blame] | 7002 | struct sched_domain *sd = sd_init(tl, cpu_map, child, cpu); |
Peter Zijlstra | 2c402dc | 2011-04-07 14:10:01 +0200 | [diff] [blame] | 7003 | |
Peter Zijlstra | 60495e7 | 2011-04-07 14:10:04 +0200 | [diff] [blame] | 7004 | if (child) { |
| 7005 | sd->level = child->level + 1; |
| 7006 | sched_domain_level_max = max(sched_domain_level_max, sd->level); |
Peter Zijlstra | d069b91 | 2011-04-07 14:10:02 +0200 | [diff] [blame] | 7007 | child->parent = sd; |
Peter Zijlstra | 6ae72df | 2014-07-22 11:47:40 +0200 | [diff] [blame] | 7008 | |
| 7009 | if (!cpumask_subset(sched_domain_span(child), |
| 7010 | sched_domain_span(sd))) { |
| 7011 | pr_err("BUG: arch topology borken\n"); |
| 7012 | #ifdef CONFIG_SCHED_DEBUG |
| 7013 | pr_err(" the %s domain not a subset of the %s domain\n", |
| 7014 | child->name, sd->name); |
| 7015 | #endif |
| 7016 | /* Fixup, ensure @sd has at least @child cpus. */ |
| 7017 | cpumask_or(sched_domain_span(sd), |
| 7018 | sched_domain_span(sd), |
| 7019 | sched_domain_span(child)); |
| 7020 | } |
| 7021 | |
Peter Zijlstra | 60495e7 | 2011-04-07 14:10:04 +0200 | [diff] [blame] | 7022 | } |
Dimitri Sivanich | a841f8c | 2012-06-05 13:44:36 -0500 | [diff] [blame] | 7023 | set_domain_attribute(sd, attr); |
Peter Zijlstra | 2c402dc | 2011-04-07 14:10:01 +0200 | [diff] [blame] | 7024 | |
| 7025 | return sd; |
| 7026 | } |
| 7027 | |
Mike Travis | 7c16ec5 | 2008-04-04 18:11:11 -0700 | [diff] [blame] | 7028 | /* |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 7029 | * Build sched domains for a given set of cpus and attach the sched domains |
| 7030 | * to the individual cpus |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7031 | */ |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 7032 | static int build_sched_domains(const struct cpumask *cpu_map, |
| 7033 | struct sched_domain_attr *attr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7034 | { |
Viresh Kumar | 1c63216 | 2013-06-10 16:27:18 +0530 | [diff] [blame] | 7035 | enum s_alloc alloc_state; |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 7036 | struct sched_domain *sd; |
Andreas Herrmann | 49a02c5 | 2009-08-18 12:51:52 +0200 | [diff] [blame] | 7037 | struct s_data d; |
Dietmar Eggemann | cd92bfd | 2016-08-01 19:53:35 +0100 | [diff] [blame] | 7038 | struct rq *rq = NULL; |
Peter Zijlstra | 822ff79 | 2011-04-07 14:09:51 +0200 | [diff] [blame] | 7039 | int i, ret = -ENOMEM; |
Rusty Russell | 3404c8d | 2008-11-25 02:35:03 +1030 | [diff] [blame] | 7040 | |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 7041 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
| 7042 | if (alloc_state != sa_rootdomain) |
| 7043 | goto error; |
Mike Travis | 7c16ec5 | 2008-04-04 18:11:11 -0700 | [diff] [blame] | 7044 | |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 7045 | /* Set up domains for cpus specified by the cpu_map. */ |
Rusty Russell | abcd083 | 2008-11-25 02:35:02 +1030 | [diff] [blame] | 7046 | for_each_cpu(i, cpu_map) { |
Peter Zijlstra | eb7a74e6 | 2011-04-07 14:10:00 +0200 | [diff] [blame] | 7047 | struct sched_domain_topology_level *tl; |
| 7048 | |
Peter Zijlstra | 3bd65a8 | 2011-04-07 14:09:54 +0200 | [diff] [blame] | 7049 | sd = NULL; |
Viresh Kumar | 27723a6 | 2013-06-10 16:27:20 +0530 | [diff] [blame] | 7050 | for_each_sd_topology(tl) { |
Viresh Kumar | 4a850cb | 2013-06-04 16:12:43 +0530 | [diff] [blame] | 7051 | sd = build_sched_domain(tl, cpu_map, attr, sd, i); |
Viresh Kumar | 22da956 | 2013-06-04 15:41:15 +0530 | [diff] [blame] | 7052 | if (tl == sched_domain_topology) |
| 7053 | *per_cpu_ptr(d.sd, i) = sd; |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 7054 | if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP)) |
| 7055 | sd->flags |= SD_OVERLAP; |
Peter Zijlstra | d110235 | 2011-07-20 18:42:57 +0200 | [diff] [blame] | 7056 | if (cpumask_equal(cpu_map, sched_domain_span(sd))) |
| 7057 | break; |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 7058 | } |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 7059 | } |
Peter Zijlstra | 21d42cc | 2011-04-07 14:09:48 +0200 | [diff] [blame] | 7060 | |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 7061 | /* Build the groups for the domains */ |
| 7062 | for_each_cpu(i, cpu_map) { |
| 7063 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
| 7064 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); |
Peter Zijlstra | e3589f6 | 2011-07-15 10:35:52 +0200 | [diff] [blame] | 7065 | if (sd->flags & SD_OVERLAP) { |
| 7066 | if (build_overlap_sched_groups(sd, i)) |
| 7067 | goto error; |
| 7068 | } else { |
| 7069 | if (build_sched_groups(sd, i)) |
| 7070 | goto error; |
| 7071 | } |
Peter Zijlstra | 1cf51902 | 2011-04-07 14:09:47 +0200 | [diff] [blame] | 7072 | } |
Peter Zijlstra | a06dadb | 2011-04-07 14:09:44 +0200 | [diff] [blame] | 7073 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7074 | |
Nicolas Pitre | ced549f | 2014-05-26 18:19:38 -0400 | [diff] [blame] | 7075 | /* Calculate CPU capacity for physical packages and nodes */ |
Peter Zijlstra | a9c9a9b | 2011-04-07 14:09:49 +0200 | [diff] [blame] | 7076 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
| 7077 | if (!cpumask_test_cpu(i, cpu_map)) |
| 7078 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7079 | |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 7080 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
| 7081 | claim_allocations(i, sd); |
Nicolas Pitre | 63b2ca3 | 2014-05-26 18:19:37 -0400 | [diff] [blame] | 7082 | init_sched_groups_capacity(i, sd); |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 7083 | } |
Siddha, Suresh B | f712c0c | 2006-07-30 03:02:59 -0700 | [diff] [blame] | 7084 | } |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 7085 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7086 | /* Attach the domains */ |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 7087 | rcu_read_lock(); |
Rusty Russell | abcd083 | 2008-11-25 02:35:02 +1030 | [diff] [blame] | 7088 | for_each_cpu(i, cpu_map) { |
Dietmar Eggemann | cd92bfd | 2016-08-01 19:53:35 +0100 | [diff] [blame] | 7089 | rq = cpu_rq(i); |
Peter Zijlstra | 21d42cc | 2011-04-07 14:09:48 +0200 | [diff] [blame] | 7090 | sd = *per_cpu_ptr(d.sd, i); |
Dietmar Eggemann | cd92bfd | 2016-08-01 19:53:35 +0100 | [diff] [blame] | 7091 | |
| 7092 | /* Use READ_ONCE()/WRITE_ONCE() to avoid load/store tearing: */ |
| 7093 | if (rq->cpu_capacity_orig > READ_ONCE(d.rd->max_cpu_capacity)) |
| 7094 | WRITE_ONCE(d.rd->max_cpu_capacity, rq->cpu_capacity_orig); |
| 7095 | |
Andreas Herrmann | 49a02c5 | 2009-08-18 12:51:52 +0200 | [diff] [blame] | 7096 | cpu_attach_domain(sd, d.rd, i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7097 | } |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 7098 | rcu_read_unlock(); |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 7099 | |
Peter Zijlstra | a18a579 | 2016-09-20 11:05:31 +0200 | [diff] [blame] | 7100 | if (rq && sched_debug_enabled) { |
Dietmar Eggemann | cd92bfd | 2016-08-01 19:53:35 +0100 | [diff] [blame] | 7101 | pr_info("span: %*pbl (max cpu_capacity = %lu)\n", |
| 7102 | cpumask_pr_args(cpu_map), rq->rd->max_cpu_capacity); |
| 7103 | } |
| 7104 | |
Peter Zijlstra | 822ff79 | 2011-04-07 14:09:51 +0200 | [diff] [blame] | 7105 | ret = 0; |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 7106 | error: |
Andreas Herrmann | 2109b99 | 2009-08-18 12:53:00 +0200 | [diff] [blame] | 7107 | __free_domain_allocs(&d, alloc_state, cpu_map); |
Peter Zijlstra | 822ff79 | 2011-04-07 14:09:51 +0200 | [diff] [blame] | 7108 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7109 | } |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7110 | |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 7111 | static cpumask_var_t *doms_cur; /* current sched domains */ |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7112 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
Ingo Molnar | 4285f594 | 2008-05-16 17:47:14 +0200 | [diff] [blame] | 7113 | static struct sched_domain_attr *dattr_cur; |
| 7114 | /* attribues of custom domains in 'doms_cur' */ |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7115 | |
| 7116 | /* |
| 7117 | * Special case: If a kmalloc of a doms_cur partition (array of |
Rusty Russell | 4212823 | 2008-11-25 02:35:12 +1030 | [diff] [blame] | 7118 | * cpumask) fails, then fallback to a single sched domain, |
| 7119 | * as determined by the single cpumask fallback_doms. |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7120 | */ |
Rusty Russell | 4212823 | 2008-11-25 02:35:12 +1030 | [diff] [blame] | 7121 | static cpumask_var_t fallback_doms; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7122 | |
Heiko Carstens | ee79d1b | 2008-12-09 18:49:50 +0100 | [diff] [blame] | 7123 | /* |
| 7124 | * arch_update_cpu_topology lets virtualized architectures update the |
| 7125 | * cpu core maps. It is supposed to return 1 if the topology changed |
| 7126 | * or 0 if it stayed the same. |
| 7127 | */ |
Gideon Israel Dsouza | 52f5684c | 2014-04-07 15:39:20 -0700 | [diff] [blame] | 7128 | int __weak arch_update_cpu_topology(void) |
Heiko Carstens | 22e52b0 | 2008-03-12 18:31:59 +0100 | [diff] [blame] | 7129 | { |
Heiko Carstens | ee79d1b | 2008-12-09 18:49:50 +0100 | [diff] [blame] | 7130 | return 0; |
Heiko Carstens | 22e52b0 | 2008-03-12 18:31:59 +0100 | [diff] [blame] | 7131 | } |
| 7132 | |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 7133 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
| 7134 | { |
| 7135 | int i; |
| 7136 | cpumask_var_t *doms; |
| 7137 | |
| 7138 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); |
| 7139 | if (!doms) |
| 7140 | return NULL; |
| 7141 | for (i = 0; i < ndoms; i++) { |
| 7142 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { |
| 7143 | free_sched_domains(doms, i); |
| 7144 | return NULL; |
| 7145 | } |
| 7146 | } |
| 7147 | return doms; |
| 7148 | } |
| 7149 | |
| 7150 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) |
| 7151 | { |
| 7152 | unsigned int i; |
| 7153 | for (i = 0; i < ndoms; i++) |
| 7154 | free_cpumask_var(doms[i]); |
| 7155 | kfree(doms); |
| 7156 | } |
| 7157 | |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 7158 | /* |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 7159 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7160 | * For now this just excludes isolated cpus, but could be used to |
| 7161 | * exclude other special cases in the future. |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 7162 | */ |
Peter Zijlstra | c4a8849 | 2011-04-07 14:09:42 +0200 | [diff] [blame] | 7163 | static int init_sched_domains(const struct cpumask *cpu_map) |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 7164 | { |
Milton Miller | 7378547 | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 7165 | int err; |
| 7166 | |
Heiko Carstens | 22e52b0 | 2008-03-12 18:31:59 +0100 | [diff] [blame] | 7167 | arch_update_cpu_topology(); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7168 | ndoms_cur = 1; |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 7169 | doms_cur = alloc_sched_domains(ndoms_cur); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7170 | if (!doms_cur) |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 7171 | doms_cur = &fallback_doms; |
| 7172 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 7173 | err = build_sched_domains(doms_cur[0], NULL); |
Milton Miller | 6382bc9 | 2007-10-15 17:00:19 +0200 | [diff] [blame] | 7174 | register_sched_domain_sysctl(); |
Milton Miller | 7378547 | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 7175 | |
| 7176 | return err; |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 7177 | } |
| 7178 | |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 7179 | /* |
| 7180 | * Detach sched domains from a group of cpus specified in cpu_map |
| 7181 | * These cpus will now be attached to the NULL domain |
| 7182 | */ |
Rusty Russell | 96f874e | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 7183 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 7184 | { |
| 7185 | int i; |
| 7186 | |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 7187 | rcu_read_lock(); |
Rusty Russell | abcd083 | 2008-11-25 02:35:02 +1030 | [diff] [blame] | 7188 | for_each_cpu(i, cpu_map) |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 7189 | cpu_attach_domain(NULL, &def_root_domain, i); |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 7190 | rcu_read_unlock(); |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 7191 | } |
| 7192 | |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 7193 | /* handle null as "default" */ |
| 7194 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, |
| 7195 | struct sched_domain_attr *new, int idx_new) |
| 7196 | { |
| 7197 | struct sched_domain_attr tmp; |
| 7198 | |
| 7199 | /* fast path */ |
| 7200 | if (!new && !cur) |
| 7201 | return 1; |
| 7202 | |
| 7203 | tmp = SD_ATTR_INIT; |
| 7204 | return !memcmp(cur ? (cur + idx_cur) : &tmp, |
| 7205 | new ? (new + idx_new) : &tmp, |
| 7206 | sizeof(struct sched_domain_attr)); |
| 7207 | } |
| 7208 | |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7209 | /* |
| 7210 | * Partition sched domains as specified by the 'ndoms_new' |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 7211 | * cpumasks in the array doms_new[] of cpumasks. This compares |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7212 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
| 7213 | * It destroys each deleted domain and builds each new domain. |
| 7214 | * |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 7215 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 7216 | * The masks don't intersect (don't overlap.) We should setup one |
| 7217 | * sched domain for each mask. CPUs not in any of the cpumasks will |
| 7218 | * not be load balanced. If the same cpumask appears both in the |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7219 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
| 7220 | * it as it is. |
| 7221 | * |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 7222 | * The passed in 'doms_new' should be allocated using |
| 7223 | * alloc_sched_domains. This routine takes ownership of it and will |
| 7224 | * free_sched_domains it when done with it. If the caller failed the |
| 7225 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, |
| 7226 | * and partition_sched_domains() will fallback to the single partition |
| 7227 | * 'fallback_doms', it also forces the domains to be rebuilt. |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7228 | * |
Rusty Russell | 96f874e | 2008-11-25 02:35:14 +1030 | [diff] [blame] | 7229 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
Li Zefan | 700018e | 2008-11-18 14:02:03 +0800 | [diff] [blame] | 7230 | * ndoms_new == 0 is a special case for destroying existing domains, |
| 7231 | * and it will not create the default domain. |
Max Krasnyansky | dfb512e | 2008-08-29 13:11:41 -0700 | [diff] [blame] | 7232 | * |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7233 | * Call with hotplug lock held |
| 7234 | */ |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 7235 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 7236 | struct sched_domain_attr *dattr_new) |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7237 | { |
Max Krasnyansky | dfb512e | 2008-08-29 13:11:41 -0700 | [diff] [blame] | 7238 | int i, j, n; |
Heiko Carstens | d65bd5e | 2008-12-09 18:49:51 +0100 | [diff] [blame] | 7239 | int new_topology; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7240 | |
Heiko Carstens | 712555e | 2008-04-28 11:33:07 +0200 | [diff] [blame] | 7241 | mutex_lock(&sched_domains_mutex); |
Srivatsa Vaddagiri | a183561 | 2008-01-25 21:08:00 +0100 | [diff] [blame] | 7242 | |
Milton Miller | 7378547 | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 7243 | /* always unregister in case we don't destroy any domains */ |
| 7244 | unregister_sched_domain_sysctl(); |
| 7245 | |
Heiko Carstens | d65bd5e | 2008-12-09 18:49:51 +0100 | [diff] [blame] | 7246 | /* Let architecture update cpu core mappings. */ |
| 7247 | new_topology = arch_update_cpu_topology(); |
| 7248 | |
Max Krasnyansky | dfb512e | 2008-08-29 13:11:41 -0700 | [diff] [blame] | 7249 | n = doms_new ? ndoms_new : 0; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7250 | |
| 7251 | /* Destroy deleted domains */ |
| 7252 | for (i = 0; i < ndoms_cur; i++) { |
Heiko Carstens | d65bd5e | 2008-12-09 18:49:51 +0100 | [diff] [blame] | 7253 | for (j = 0; j < n && !new_topology; j++) { |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 7254 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 7255 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7256 | goto match1; |
| 7257 | } |
| 7258 | /* no match - a current sched domain not in new doms_new[] */ |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 7259 | detach_destroy_domains(doms_cur[i]); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7260 | match1: |
| 7261 | ; |
| 7262 | } |
| 7263 | |
Xiaotian Feng | c8d2d47 | 2013-08-06 20:06:42 +0800 | [diff] [blame] | 7264 | n = ndoms_cur; |
Max Krasnyansky | e761b77 | 2008-07-15 04:43:49 -0700 | [diff] [blame] | 7265 | if (doms_new == NULL) { |
Xiaotian Feng | c8d2d47 | 2013-08-06 20:06:42 +0800 | [diff] [blame] | 7266 | n = 0; |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 7267 | doms_new = &fallback_doms; |
Peter Zijlstra | 6ad4c18 | 2009-11-25 13:31:39 +0100 | [diff] [blame] | 7268 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
Li Zefan | faa2f98 | 2008-11-04 16:20:23 +0800 | [diff] [blame] | 7269 | WARN_ON_ONCE(dattr_new); |
Max Krasnyansky | e761b77 | 2008-07-15 04:43:49 -0700 | [diff] [blame] | 7270 | } |
| 7271 | |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7272 | /* Build new domains */ |
| 7273 | for (i = 0; i < ndoms_new; i++) { |
Xiaotian Feng | c8d2d47 | 2013-08-06 20:06:42 +0800 | [diff] [blame] | 7274 | for (j = 0; j < n && !new_topology; j++) { |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 7275 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 7276 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7277 | goto match2; |
| 7278 | } |
| 7279 | /* no match - add a new doms_new */ |
Peter Zijlstra | dce840a | 2011-04-07 14:09:50 +0200 | [diff] [blame] | 7280 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7281 | match2: |
| 7282 | ; |
| 7283 | } |
| 7284 | |
| 7285 | /* Remember the new sched domains */ |
Rusty Russell | acc3f5d | 2009-11-03 14:53:40 +1030 | [diff] [blame] | 7286 | if (doms_cur != &fallback_doms) |
| 7287 | free_sched_domains(doms_cur, ndoms_cur); |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 7288 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7289 | doms_cur = doms_new; |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 7290 | dattr_cur = dattr_new; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7291 | ndoms_cur = ndoms_new; |
Milton Miller | 7378547 | 2007-10-24 18:23:48 +0200 | [diff] [blame] | 7292 | |
| 7293 | register_sched_domain_sysctl(); |
Srivatsa Vaddagiri | a183561 | 2008-01-25 21:08:00 +0100 | [diff] [blame] | 7294 | |
Heiko Carstens | 712555e | 2008-04-28 11:33:07 +0200 | [diff] [blame] | 7295 | mutex_unlock(&sched_domains_mutex); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7296 | } |
| 7297 | |
Srivatsa S. Bhat | d35be8b | 2012-05-24 19:46:26 +0530 | [diff] [blame] | 7298 | static int num_cpus_frozen; /* used to mark begin/end of suspend/resume */ |
| 7299 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7300 | /* |
Tejun Heo | 3a101d0 | 2010-06-08 21:40:36 +0200 | [diff] [blame] | 7301 | * Update cpusets according to cpu_active mask. If cpusets are |
| 7302 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper |
| 7303 | * around partition_sched_domains(). |
Srivatsa S. Bhat | d35be8b | 2012-05-24 19:46:26 +0530 | [diff] [blame] | 7304 | * |
| 7305 | * If we come here as part of a suspend/resume, don't touch cpusets because we |
| 7306 | * want to restore it back to its original state upon resume anyway. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7307 | */ |
Thomas Gleixner | 40190a7 | 2016-03-10 12:54:13 +0100 | [diff] [blame] | 7308 | static void cpuset_cpu_active(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7309 | { |
Thomas Gleixner | 40190a7 | 2016-03-10 12:54:13 +0100 | [diff] [blame] | 7310 | if (cpuhp_tasks_frozen) { |
Srivatsa S. Bhat | d35be8b | 2012-05-24 19:46:26 +0530 | [diff] [blame] | 7311 | /* |
| 7312 | * num_cpus_frozen tracks how many CPUs are involved in suspend |
| 7313 | * resume sequence. As long as this is not the last online |
| 7314 | * operation in the resume sequence, just build a single sched |
| 7315 | * domain, ignoring cpusets. |
| 7316 | */ |
Peter Zijlstra | ba15518 | 2017-09-07 11:13:38 +0200 | [diff] [blame] | 7317 | partition_sched_domains(1, NULL, NULL); |
| 7318 | if (--num_cpus_frozen) |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 7319 | return; |
Srivatsa S. Bhat | d35be8b | 2012-05-24 19:46:26 +0530 | [diff] [blame] | 7320 | /* |
| 7321 | * This is the last CPU online operation. So fall through and |
| 7322 | * restore the original sched domains by considering the |
| 7323 | * cpuset configurations. |
| 7324 | */ |
Peter Zijlstra | ba15518 | 2017-09-07 11:13:38 +0200 | [diff] [blame] | 7325 | cpuset_force_rebuild(); |
Max Krasnyansky | e761b77 | 2008-07-15 04:43:49 -0700 | [diff] [blame] | 7326 | } |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 7327 | cpuset_update_active_cpus(true); |
Max Krasnyansky | e761b77 | 2008-07-15 04:43:49 -0700 | [diff] [blame] | 7328 | } |
Tejun Heo | 3a101d0 | 2010-06-08 21:40:36 +0200 | [diff] [blame] | 7329 | |
Thomas Gleixner | 40190a7 | 2016-03-10 12:54:13 +0100 | [diff] [blame] | 7330 | static int cpuset_cpu_inactive(unsigned int cpu) |
Tejun Heo | 3a101d0 | 2010-06-08 21:40:36 +0200 | [diff] [blame] | 7331 | { |
Juri Lelli | 3c18d44 | 2015-03-31 09:53:37 +0100 | [diff] [blame] | 7332 | unsigned long flags; |
Juri Lelli | 3c18d44 | 2015-03-31 09:53:37 +0100 | [diff] [blame] | 7333 | struct dl_bw *dl_b; |
Omar Sandoval | 533445c | 2015-05-04 03:09:36 -0700 | [diff] [blame] | 7334 | bool overflow; |
| 7335 | int cpus; |
Juri Lelli | 3c18d44 | 2015-03-31 09:53:37 +0100 | [diff] [blame] | 7336 | |
Thomas Gleixner | 40190a7 | 2016-03-10 12:54:13 +0100 | [diff] [blame] | 7337 | if (!cpuhp_tasks_frozen) { |
Omar Sandoval | 533445c | 2015-05-04 03:09:36 -0700 | [diff] [blame] | 7338 | rcu_read_lock_sched(); |
| 7339 | dl_b = dl_bw_of(cpu); |
Juri Lelli | 3c18d44 | 2015-03-31 09:53:37 +0100 | [diff] [blame] | 7340 | |
Omar Sandoval | 533445c | 2015-05-04 03:09:36 -0700 | [diff] [blame] | 7341 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
| 7342 | cpus = dl_bw_cpus(cpu); |
| 7343 | overflow = __dl_overflow(dl_b, cpus, 0, 0); |
| 7344 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
Juri Lelli | 3c18d44 | 2015-03-31 09:53:37 +0100 | [diff] [blame] | 7345 | |
Omar Sandoval | 533445c | 2015-05-04 03:09:36 -0700 | [diff] [blame] | 7346 | rcu_read_unlock_sched(); |
Juri Lelli | 3c18d44 | 2015-03-31 09:53:37 +0100 | [diff] [blame] | 7347 | |
Omar Sandoval | 533445c | 2015-05-04 03:09:36 -0700 | [diff] [blame] | 7348 | if (overflow) |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 7349 | return -EBUSY; |
Srivatsa S. Bhat | 7ddf96b | 2012-05-24 19:46:55 +0530 | [diff] [blame] | 7350 | cpuset_update_active_cpus(false); |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 7351 | } else { |
Srivatsa S. Bhat | d35be8b | 2012-05-24 19:46:26 +0530 | [diff] [blame] | 7352 | num_cpus_frozen++; |
| 7353 | partition_sched_domains(1, NULL, NULL); |
Tejun Heo | 3a101d0 | 2010-06-08 21:40:36 +0200 | [diff] [blame] | 7354 | } |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 7355 | return 0; |
Tejun Heo | 3a101d0 | 2010-06-08 21:40:36 +0200 | [diff] [blame] | 7356 | } |
Max Krasnyansky | e761b77 | 2008-07-15 04:43:49 -0700 | [diff] [blame] | 7357 | |
Thomas Gleixner | 40190a7 | 2016-03-10 12:54:13 +0100 | [diff] [blame] | 7358 | int sched_cpu_activate(unsigned int cpu) |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 7359 | { |
Thomas Gleixner | 7d97669 | 2016-03-10 12:54:17 +0100 | [diff] [blame] | 7360 | struct rq *rq = cpu_rq(cpu); |
| 7361 | unsigned long flags; |
| 7362 | |
Thomas Gleixner | 40190a7 | 2016-03-10 12:54:13 +0100 | [diff] [blame] | 7363 | set_cpu_active(cpu, true); |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 7364 | |
Thomas Gleixner | 40190a7 | 2016-03-10 12:54:13 +0100 | [diff] [blame] | 7365 | if (sched_smp_initialized) { |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 7366 | sched_domains_numa_masks_set(cpu); |
Thomas Gleixner | 40190a7 | 2016-03-10 12:54:13 +0100 | [diff] [blame] | 7367 | cpuset_cpu_active(); |
Nick Piggin | 5c1e176 | 2006-10-03 01:14:04 -0700 | [diff] [blame] | 7368 | } |
Thomas Gleixner | 7d97669 | 2016-03-10 12:54:17 +0100 | [diff] [blame] | 7369 | |
| 7370 | /* |
| 7371 | * Put the rq online, if not already. This happens: |
| 7372 | * |
| 7373 | * 1) In the early boot process, because we build the real domains |
| 7374 | * after all cpus have been brought up. |
| 7375 | * |
| 7376 | * 2) At runtime, if cpuset_cpu_active() fails to rebuild the |
| 7377 | * domains. |
| 7378 | */ |
| 7379 | raw_spin_lock_irqsave(&rq->lock, flags); |
| 7380 | if (rq->rd) { |
| 7381 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
| 7382 | set_rq_online(rq); |
| 7383 | } |
| 7384 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
| 7385 | |
| 7386 | update_max_interval(); |
| 7387 | |
Thomas Gleixner | 40190a7 | 2016-03-10 12:54:13 +0100 | [diff] [blame] | 7388 | return 0; |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 7389 | } |
| 7390 | |
Thomas Gleixner | 40190a7 | 2016-03-10 12:54:13 +0100 | [diff] [blame] | 7391 | int sched_cpu_deactivate(unsigned int cpu) |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 7392 | { |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 7393 | int ret; |
| 7394 | |
Thomas Gleixner | 40190a7 | 2016-03-10 12:54:13 +0100 | [diff] [blame] | 7395 | set_cpu_active(cpu, false); |
Peter Zijlstra | b2454ca | 2016-03-10 12:54:14 +0100 | [diff] [blame] | 7396 | /* |
| 7397 | * We've cleared cpu_active_mask, wait for all preempt-disabled and RCU |
| 7398 | * users of this state to go away such that all new such users will |
| 7399 | * observe it. |
| 7400 | * |
| 7401 | * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might |
| 7402 | * not imply sync_sched(), so wait for both. |
| 7403 | * |
| 7404 | * Do sync before park smpboot threads to take care the rcu boost case. |
| 7405 | */ |
| 7406 | if (IS_ENABLED(CONFIG_PREEMPT)) |
| 7407 | synchronize_rcu_mult(call_rcu, call_rcu_sched); |
| 7408 | else |
| 7409 | synchronize_rcu(); |
Thomas Gleixner | 40190a7 | 2016-03-10 12:54:13 +0100 | [diff] [blame] | 7410 | |
| 7411 | if (!sched_smp_initialized) |
| 7412 | return 0; |
| 7413 | |
| 7414 | ret = cpuset_cpu_inactive(cpu); |
| 7415 | if (ret) { |
| 7416 | set_cpu_active(cpu, true); |
| 7417 | return ret; |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 7418 | } |
Thomas Gleixner | 40190a7 | 2016-03-10 12:54:13 +0100 | [diff] [blame] | 7419 | sched_domains_numa_masks_clear(cpu); |
| 7420 | return 0; |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 7421 | } |
| 7422 | |
Thomas Gleixner | 94baf7a | 2016-03-10 12:54:15 +0100 | [diff] [blame] | 7423 | static void sched_rq_cpu_starting(unsigned int cpu) |
| 7424 | { |
| 7425 | struct rq *rq = cpu_rq(cpu); |
| 7426 | |
| 7427 | rq->calc_load_update = calc_load_update; |
Thomas Gleixner | 94baf7a | 2016-03-10 12:54:15 +0100 | [diff] [blame] | 7428 | update_max_interval(); |
| 7429 | } |
| 7430 | |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 7431 | int sched_cpu_starting(unsigned int cpu) |
| 7432 | { |
| 7433 | set_cpu_rq_start_time(cpu); |
Thomas Gleixner | 94baf7a | 2016-03-10 12:54:15 +0100 | [diff] [blame] | 7434 | sched_rq_cpu_starting(cpu); |
Thomas Gleixner | 135fb3e | 2016-03-10 12:54:11 +0100 | [diff] [blame] | 7435 | return 0; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7436 | } |
| 7437 | |
Thomas Gleixner | f2785dd | 2016-03-10 12:54:18 +0100 | [diff] [blame] | 7438 | #ifdef CONFIG_HOTPLUG_CPU |
| 7439 | int sched_cpu_dying(unsigned int cpu) |
| 7440 | { |
| 7441 | struct rq *rq = cpu_rq(cpu); |
| 7442 | unsigned long flags; |
| 7443 | |
| 7444 | /* Handle pending wakeups and then migrate everything off */ |
| 7445 | sched_ttwu_pending(); |
| 7446 | raw_spin_lock_irqsave(&rq->lock, flags); |
| 7447 | if (rq->rd) { |
| 7448 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
| 7449 | set_rq_offline(rq); |
| 7450 | } |
| 7451 | migrate_tasks(rq); |
| 7452 | BUG_ON(rq->nr_running != 1); |
| 7453 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
| 7454 | calc_load_migrate(rq); |
| 7455 | update_max_interval(); |
Thomas Gleixner | 20a5c8c | 2016-03-10 12:54:20 +0100 | [diff] [blame] | 7456 | nohz_balance_exit_idle(cpu); |
Thomas Gleixner | e5ef27d | 2016-03-10 12:54:21 +0100 | [diff] [blame] | 7457 | hrtick_clear(rq); |
Thomas Gleixner | f2785dd | 2016-03-10 12:54:18 +0100 | [diff] [blame] | 7458 | return 0; |
| 7459 | } |
| 7460 | #endif |
| 7461 | |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7462 | void __init sched_init_smp(void) |
| 7463 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7464 | cpumask_var_t non_isolated_cpus; |
| 7465 | |
Christoph Lameter | 476f353 | 2007-05-06 14:48:58 -0700 | [diff] [blame] | 7466 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 7467 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 7468 | |
| 7469 | sched_init_numa(); |
| 7470 | |
Peter Zijlstra | 6acce3e | 2013-10-11 14:38:20 +0200 | [diff] [blame] | 7471 | /* |
| 7472 | * There's no userspace yet to cause hotplug operations; hence all the |
| 7473 | * cpu masks are stable and all blatant races in the below code cannot |
| 7474 | * happen. |
| 7475 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7476 | mutex_lock(&sched_domains_mutex); |
| 7477 | init_sched_domains(cpu_active_mask); |
| 7478 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
| 7479 | if (cpumask_empty(non_isolated_cpus)) |
| 7480 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); |
| 7481 | mutex_unlock(&sched_domains_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7482 | |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 7483 | /* Move init over to a non-isolated CPU */ |
| 7484 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
| 7485 | BUG(); |
| 7486 | sched_init_granularity(); |
Rusty Russell | dcc30a3 | 2008-11-25 02:35:12 +1030 | [diff] [blame] | 7487 | free_cpumask_var(non_isolated_cpus); |
Rusty Russell | 4212823 | 2008-11-25 02:35:12 +1030 | [diff] [blame] | 7488 | |
Rusty Russell | 0e3900e | 2008-11-25 02:35:13 +1030 | [diff] [blame] | 7489 | init_sched_rt_class(); |
Juri Lelli | 1baca4c | 2013-11-07 14:43:38 +0100 | [diff] [blame] | 7490 | init_sched_dl_class(); |
Thomas Gleixner | e26fbff | 2016-03-10 12:54:10 +0100 | [diff] [blame] | 7491 | sched_smp_initialized = true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7492 | } |
Thomas Gleixner | e26fbff | 2016-03-10 12:54:10 +0100 | [diff] [blame] | 7493 | |
| 7494 | static int __init migration_init(void) |
| 7495 | { |
Thomas Gleixner | 94baf7a | 2016-03-10 12:54:15 +0100 | [diff] [blame] | 7496 | sched_rq_cpu_starting(smp_processor_id()); |
Thomas Gleixner | e26fbff | 2016-03-10 12:54:10 +0100 | [diff] [blame] | 7497 | return 0; |
| 7498 | } |
| 7499 | early_initcall(migration_init); |
| 7500 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 7501 | #else |
| 7502 | void __init sched_init_smp(void) |
| 7503 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7504 | sched_init_granularity(); |
| 7505 | } |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 7506 | #endif /* CONFIG_SMP */ |
Heiko Carstens | b50f60c | 2006-07-30 03:03:52 -0700 | [diff] [blame] | 7507 | |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 7508 | int in_sched_functions(unsigned long addr) |
| 7509 | { |
| 7510 | return in_lock_functions(addr) || |
| 7511 | (addr >= (unsigned long)__sched_text_start |
Christoph Lameter | c9819f4 | 2006-12-10 02:20:25 -0800 | [diff] [blame] | 7512 | && addr < (unsigned long)__sched_text_end); |
Christoph Lameter | 476f353 | 2007-05-06 14:48:58 -0700 | [diff] [blame] | 7513 | } |
Christoph Lameter | c9819f4 | 2006-12-10 02:20:25 -0800 | [diff] [blame] | 7514 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 7515 | #ifdef CONFIG_CGROUP_SCHED |
Li Zefan | 27b4b93 | 2013-03-05 16:07:52 +0800 | [diff] [blame] | 7516 | /* |
| 7517 | * Default task group. |
| 7518 | * Every task in system belongs to this group at bootup. |
| 7519 | */ |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 7520 | struct task_group root_task_group; |
Mike Galbraith | 35cf4e5 | 2012-08-07 05:00:13 +0200 | [diff] [blame] | 7521 | LIST_HEAD(task_groups); |
Waiman Long | b036762 | 2015-12-02 13:41:49 -0500 | [diff] [blame] | 7522 | |
| 7523 | /* Cacheline aligned slab cache for task_group */ |
| 7524 | static struct kmem_cache *task_group_cache __read_mostly; |
Heiko Carstens | b50f60c | 2006-07-30 03:03:52 -0700 | [diff] [blame] | 7525 | #endif |
| 7526 | |
Joonsoo Kim | e6252c3 | 2013-04-23 17:27:41 +0900 | [diff] [blame] | 7527 | DECLARE_PER_CPU(cpumask_var_t, load_balance_mask); |
Peter Zijlstra | 10e2f1a | 2016-05-09 10:38:05 +0200 | [diff] [blame] | 7528 | DECLARE_PER_CPU(cpumask_var_t, select_idle_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7529 | |
Linus Torvalds | 9dcb8b6 | 2016-10-26 10:15:30 -0700 | [diff] [blame] | 7530 | #define WAIT_TABLE_BITS 8 |
| 7531 | #define WAIT_TABLE_SIZE (1 << WAIT_TABLE_BITS) |
| 7532 | static wait_queue_head_t bit_wait_table[WAIT_TABLE_SIZE] __cacheline_aligned; |
| 7533 | |
| 7534 | wait_queue_head_t *bit_waitqueue(void *word, int bit) |
| 7535 | { |
| 7536 | const int shift = BITS_PER_LONG == 32 ? 5 : 6; |
| 7537 | unsigned long val = (unsigned long)word << shift | bit; |
| 7538 | |
| 7539 | return bit_wait_table + hash_long(val, WAIT_TABLE_BITS); |
| 7540 | } |
| 7541 | EXPORT_SYMBOL(bit_waitqueue); |
| 7542 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7543 | void __init sched_init(void) |
| 7544 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7545 | int i, j; |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 7546 | unsigned long alloc_size = 0, ptr; |
| 7547 | |
Linus Torvalds | 9dcb8b6 | 2016-10-26 10:15:30 -0700 | [diff] [blame] | 7548 | for (i = 0; i < WAIT_TABLE_SIZE; i++) |
| 7549 | init_waitqueue_head(bit_wait_table + i); |
| 7550 | |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 7551 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 7552 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); |
| 7553 | #endif |
| 7554 | #ifdef CONFIG_RT_GROUP_SCHED |
| 7555 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); |
| 7556 | #endif |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 7557 | if (alloc_size) { |
Pekka Enberg | 36b7b6d | 2009-06-10 23:42:36 +0300 | [diff] [blame] | 7558 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 7559 | |
| 7560 | #ifdef CONFIG_FAIR_GROUP_SCHED |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 7561 | root_task_group.se = (struct sched_entity **)ptr; |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 7562 | ptr += nr_cpu_ids * sizeof(void **); |
| 7563 | |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 7564 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 7565 | ptr += nr_cpu_ids * sizeof(void **); |
Peter Zijlstra | eff766a | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 7566 | |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 7567 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 7568 | #ifdef CONFIG_RT_GROUP_SCHED |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 7569 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 7570 | ptr += nr_cpu_ids * sizeof(void **); |
| 7571 | |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 7572 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
Peter Zijlstra | eff766a | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 7573 | ptr += nr_cpu_ids * sizeof(void **); |
| 7574 | |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 7575 | #endif /* CONFIG_RT_GROUP_SCHED */ |
Mike Travis | 434d53b | 2008-04-04 18:11:04 -0700 | [diff] [blame] | 7576 | } |
Alex Thorlton | b74e627 | 2014-12-18 12:44:30 -0600 | [diff] [blame] | 7577 | #ifdef CONFIG_CPUMASK_OFFSTACK |
| 7578 | for_each_possible_cpu(i) { |
| 7579 | per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node( |
| 7580 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); |
Peter Zijlstra | 10e2f1a | 2016-05-09 10:38:05 +0200 | [diff] [blame] | 7581 | per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node( |
| 7582 | cpumask_size(), GFP_KERNEL, cpu_to_node(i)); |
Alex Thorlton | b74e627 | 2014-12-18 12:44:30 -0600 | [diff] [blame] | 7583 | } |
| 7584 | #endif /* CONFIG_CPUMASK_OFFSTACK */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7585 | |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 7586 | init_rt_bandwidth(&def_rt_bandwidth, |
| 7587 | global_rt_period(), global_rt_runtime()); |
| 7588 | init_dl_bandwidth(&def_dl_bandwidth, |
Peter Zijlstra | 1724813 | 2013-12-17 12:44:49 +0100 | [diff] [blame] | 7589 | global_rt_period(), global_rt_runtime()); |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 7590 | |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 7591 | #ifdef CONFIG_SMP |
| 7592 | init_defrootdomain(); |
| 7593 | #endif |
| 7594 | |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7595 | #ifdef CONFIG_RT_GROUP_SCHED |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 7596 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7597 | global_rt_period(), global_rt_runtime()); |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 7598 | #endif /* CONFIG_RT_GROUP_SCHED */ |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 7599 | |
Dhaval Giani | 7c94143 | 2010-01-20 13:26:18 +0100 | [diff] [blame] | 7600 | #ifdef CONFIG_CGROUP_SCHED |
Waiman Long | b036762 | 2015-12-02 13:41:49 -0500 | [diff] [blame] | 7601 | task_group_cache = KMEM_CACHE(task_group, 0); |
| 7602 | |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 7603 | list_add(&root_task_group.list, &task_groups); |
| 7604 | INIT_LIST_HEAD(&root_task_group.children); |
Glauber Costa | f4d6f6c | 2011-11-01 19:19:07 -0200 | [diff] [blame] | 7605 | INIT_LIST_HEAD(&root_task_group.siblings); |
Mike Galbraith | 5091faa | 2010-11-30 14:18:03 +0100 | [diff] [blame] | 7606 | autogroup_init(&init_task); |
Dhaval Giani | 7c94143 | 2010-01-20 13:26:18 +0100 | [diff] [blame] | 7607 | #endif /* CONFIG_CGROUP_SCHED */ |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7608 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 7609 | for_each_possible_cpu(i) { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 7610 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7611 | |
| 7612 | rq = cpu_rq(i); |
Thomas Gleixner | 05fa785 | 2009-11-17 14:28:38 +0100 | [diff] [blame] | 7613 | raw_spin_lock_init(&rq->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7614 | rq->nr_running = 0; |
Thomas Gleixner | dce48a8 | 2009-04-11 10:43:41 +0200 | [diff] [blame] | 7615 | rq->calc_load_active = 0; |
| 7616 | rq->calc_load_update = jiffies + LOAD_FREQ; |
Jan H. Schönherr | acb5a9b | 2011-07-14 18:32:43 +0200 | [diff] [blame] | 7617 | init_cfs_rq(&rq->cfs); |
Abel Vesa | 07c54f7 | 2015-03-03 13:50:27 +0200 | [diff] [blame] | 7618 | init_rt_rq(&rq->rt); |
| 7619 | init_dl_rq(&rq->dl); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7620 | #ifdef CONFIG_FAIR_GROUP_SCHED |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 7621 | root_task_group.shares = ROOT_TASK_GROUP_LOAD; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7622 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
Dhaval Giani | 354d60c | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 7623 | /* |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 7624 | * How much cpu bandwidth does root_task_group get? |
Dhaval Giani | 354d60c | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 7625 | * |
| 7626 | * In case of task-groups formed thr' the cgroup filesystem, it |
| 7627 | * gets 100% of the cpu resources in the system. This overall |
| 7628 | * system cpu resource is divided among the tasks of |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 7629 | * root_task_group and its child task-groups in a fair manner, |
Dhaval Giani | 354d60c | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 7630 | * based on each entity's (task or task-group's) weight |
| 7631 | * (se->load.weight). |
| 7632 | * |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 7633 | * In other words, if root_task_group has 10 tasks of weight |
Dhaval Giani | 354d60c | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 7634 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
| 7635 | * then A0's share of the cpu resource is: |
| 7636 | * |
Ingo Molnar | 0d905bc | 2009-05-04 19:13:30 +0200 | [diff] [blame] | 7637 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
Dhaval Giani | 354d60c | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 7638 | * |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 7639 | * We achieve this by letting root_task_group's tasks sit |
| 7640 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). |
Dhaval Giani | 354d60c | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 7641 | */ |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 7642 | init_cfs_bandwidth(&root_task_group.cfs_bandwidth); |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 7643 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
Dhaval Giani | 354d60c | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 7644 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
| 7645 | |
| 7646 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7647 | #ifdef CONFIG_RT_GROUP_SCHED |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 7648 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7649 | #endif |
Ingo Molnar | 91368d7 | 2006-03-23 03:00:54 -0800 | [diff] [blame] | 7650 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7651 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
| 7652 | rq->cpu_load[j] = 0; |
Venkatesh Pallipadi | fdf3e95 | 2010-05-17 18:14:43 -0700 | [diff] [blame] | 7653 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7654 | #ifdef CONFIG_SMP |
Peter Zijlstra | a4c410f | 2006-12-06 20:37:21 -0800 | [diff] [blame] | 7655 | rq->sd = NULL; |
Gregory Haskins | 57d885f | 2008-01-25 21:08:18 +0100 | [diff] [blame] | 7656 | rq->rd = NULL; |
Vincent Guittot | ca6d75e | 2015-02-27 16:54:09 +0100 | [diff] [blame] | 7657 | rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE; |
Peter Zijlstra | e3fca9e | 2015-06-11 14:46:37 +0200 | [diff] [blame] | 7658 | rq->balance_callback = NULL; |
Ingo Molnar | 3117df0 | 2006-12-13 00:34:43 -0800 | [diff] [blame] | 7659 | rq->active_balance = 0; |
| 7660 | rq->next_balance = jiffies; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7661 | rq->push_cpu = 0; |
| 7662 | rq->cpu = i; |
Gregory Haskins | 1f11eb6 | 2008-06-04 15:04:05 -0400 | [diff] [blame] | 7663 | rq->online = 0; |
Mike Galbraith | eae0c9d | 2009-11-10 03:50:02 +0100 | [diff] [blame] | 7664 | rq->idle_stamp = 0; |
| 7665 | rq->avg_idle = 2*sysctl_sched_migration_cost; |
Jason Low | 9bd721c | 2013-09-13 11:26:52 -0700 | [diff] [blame] | 7666 | rq->max_idle_balance_cost = sysctl_sched_migration_cost; |
Peter Zijlstra | 367456c | 2012-02-20 21:49:09 +0100 | [diff] [blame] | 7667 | |
| 7668 | INIT_LIST_HEAD(&rq->cfs_tasks); |
| 7669 | |
Gregory Haskins | dc93852 | 2008-01-25 21:08:26 +0100 | [diff] [blame] | 7670 | rq_attach_root(rq, &def_root_domain); |
Frederic Weisbecker | 3451d02 | 2011-08-10 23:21:01 +0200 | [diff] [blame] | 7671 | #ifdef CONFIG_NO_HZ_COMMON |
Frederic Weisbecker | 9fd81dd | 2016-04-19 17:36:51 +0200 | [diff] [blame] | 7672 | rq->last_load_update_tick = jiffies; |
Suresh Siddha | 1c792db | 2011-12-01 17:07:32 -0800 | [diff] [blame] | 7673 | rq->nohz_flags = 0; |
Venkatesh Pallipadi | 83cd4fe | 2010-05-21 17:09:41 -0700 | [diff] [blame] | 7674 | #endif |
Frederic Weisbecker | 265f22a | 2013-05-03 03:39:05 +0200 | [diff] [blame] | 7675 | #ifdef CONFIG_NO_HZ_FULL |
| 7676 | rq->last_sched_tick = 0; |
| 7677 | #endif |
Frederic Weisbecker | 9fd81dd | 2016-04-19 17:36:51 +0200 | [diff] [blame] | 7678 | #endif /* CONFIG_SMP */ |
Peter Zijlstra | 8f4d37e | 2008-01-25 21:08:29 +0100 | [diff] [blame] | 7679 | init_rq_hrtick(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7680 | atomic_set(&rq->nr_iowait, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7681 | } |
Ingo Molnar | a0f98a1 | 2007-06-17 18:37:45 +0200 | [diff] [blame] | 7682 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 7683 | set_load_weight(&init_task); |
| 7684 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7685 | /* |
| 7686 | * The boot idle thread does lazy MMU switching as well: |
| 7687 | */ |
| 7688 | atomic_inc(&init_mm.mm_count); |
| 7689 | enter_lazy_tlb(&init_mm, current); |
| 7690 | |
| 7691 | /* |
| 7692 | * Make us the idle thread. Technically, schedule() should not be |
| 7693 | * called from this thread, however somewhere below it might be, |
| 7694 | * but because we are the idle thread, we just pick up running again |
| 7695 | * when this runqueue becomes "idle". |
| 7696 | */ |
| 7697 | init_idle(current, smp_processor_id()); |
Thomas Gleixner | dce48a8 | 2009-04-11 10:43:41 +0200 | [diff] [blame] | 7698 | |
| 7699 | calc_load_update = jiffies + LOAD_FREQ; |
| 7700 | |
Rusty Russell | bf4d83f | 2008-11-25 09:57:51 +1030 | [diff] [blame] | 7701 | #ifdef CONFIG_SMP |
Peter Zijlstra | 4cb9883 | 2011-04-07 14:09:58 +0200 | [diff] [blame] | 7702 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
Rusty Russell | bdddd29 | 2009-12-02 14:09:16 +1030 | [diff] [blame] | 7703 | /* May be allocated at isolcpus cmdline parse time */ |
| 7704 | if (cpu_isolated_map == NULL) |
| 7705 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); |
Thomas Gleixner | 29d5e04 | 2012-04-20 13:05:45 +0000 | [diff] [blame] | 7706 | idle_thread_set_boot_cpu(); |
Thomas Gleixner | 9cf7243 | 2016-03-10 12:54:09 +0100 | [diff] [blame] | 7707 | set_cpu_rq_start_time(smp_processor_id()); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 7708 | #endif |
| 7709 | init_sched_fair_class(); |
Rusty Russell | 6a7b3dc | 2008-11-25 02:35:04 +1030 | [diff] [blame] | 7710 | |
Josh Poimboeuf | 4698f88 | 2016-06-07 14:43:16 -0500 | [diff] [blame] | 7711 | init_schedstats(); |
| 7712 | |
Ingo Molnar | 6892b75 | 2008-02-13 14:02:36 +0100 | [diff] [blame] | 7713 | scheduler_running = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7714 | } |
| 7715 | |
Frederic Weisbecker | d902db1 | 2011-06-08 19:31:56 +0200 | [diff] [blame] | 7716 | #ifdef CONFIG_DEBUG_ATOMIC_SLEEP |
Frederic Weisbecker | e4aafea | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 7717 | static inline int preempt_count_equals(int preempt_offset) |
| 7718 | { |
Peter Zijlstra | da7142e | 2015-09-28 18:11:45 +0200 | [diff] [blame] | 7719 | int nested = preempt_count() + rcu_preempt_depth(); |
Frederic Weisbecker | e4aafea | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 7720 | |
Arnd Bergmann | 4ba8216 | 2011-01-25 22:52:22 +0100 | [diff] [blame] | 7721 | return (nested == preempt_offset); |
Frederic Weisbecker | e4aafea | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 7722 | } |
| 7723 | |
Simon Kagstrom | d894837 | 2009-12-23 11:08:18 +0100 | [diff] [blame] | 7724 | void __might_sleep(const char *file, int line, int preempt_offset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7725 | { |
Peter Zijlstra | 8eb23b9 | 2014-09-24 10:18:55 +0200 | [diff] [blame] | 7726 | /* |
| 7727 | * Blocking primitives will set (and therefore destroy) current->state, |
| 7728 | * since we will exit with TASK_RUNNING make sure we enter with it, |
| 7729 | * otherwise we will destroy state. |
| 7730 | */ |
Linus Torvalds | 00845eb | 2015-02-01 12:23:32 -0800 | [diff] [blame] | 7731 | WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change, |
Peter Zijlstra | 8eb23b9 | 2014-09-24 10:18:55 +0200 | [diff] [blame] | 7732 | "do not call blocking ops when !TASK_RUNNING; " |
| 7733 | "state=%lx set at [<%p>] %pS\n", |
| 7734 | current->state, |
| 7735 | (void *)current->task_state_change, |
Linus Torvalds | 00845eb | 2015-02-01 12:23:32 -0800 | [diff] [blame] | 7736 | (void *)current->task_state_change); |
Peter Zijlstra | 8eb23b9 | 2014-09-24 10:18:55 +0200 | [diff] [blame] | 7737 | |
Peter Zijlstra | 3427445 | 2014-09-24 10:18:56 +0200 | [diff] [blame] | 7738 | ___might_sleep(file, line, preempt_offset); |
| 7739 | } |
| 7740 | EXPORT_SYMBOL(__might_sleep); |
| 7741 | |
| 7742 | void ___might_sleep(const char *file, int line, int preempt_offset) |
| 7743 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7744 | static unsigned long prev_jiffy; /* ratelimiting */ |
Vegard Nossum | d1c6d14 | 2016-07-23 09:46:39 +0200 | [diff] [blame] | 7745 | unsigned long preempt_disable_ip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7746 | |
Paul E. McKenney | b3fbab0 | 2011-05-24 08:31:09 -0700 | [diff] [blame] | 7747 | rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */ |
Thomas Gleixner | db273be | 2014-02-07 20:58:38 +0100 | [diff] [blame] | 7748 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled() && |
| 7749 | !is_idle_task(current)) || |
Frederic Weisbecker | e4aafea | 2009-07-16 15:44:29 +0200 | [diff] [blame] | 7750 | system_state != SYSTEM_RUNNING || oops_in_progress) |
Ingo Molnar | aef745f | 2008-08-28 11:34:43 +0200 | [diff] [blame] | 7751 | return; |
| 7752 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) |
| 7753 | return; |
| 7754 | prev_jiffy = jiffies; |
| 7755 | |
Vegard Nossum | d1c6d14 | 2016-07-23 09:46:39 +0200 | [diff] [blame] | 7756 | /* Save this before calling printk(), since that will clobber it */ |
| 7757 | preempt_disable_ip = get_preempt_disable_ip(current); |
| 7758 | |
Peter Zijlstra | 3df0fc5 | 2009-12-20 14:23:57 +0100 | [diff] [blame] | 7759 | printk(KERN_ERR |
| 7760 | "BUG: sleeping function called from invalid context at %s:%d\n", |
| 7761 | file, line); |
| 7762 | printk(KERN_ERR |
| 7763 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", |
| 7764 | in_atomic(), irqs_disabled(), |
| 7765 | current->pid, current->comm); |
Ingo Molnar | aef745f | 2008-08-28 11:34:43 +0200 | [diff] [blame] | 7766 | |
Eric Sandeen | a8b686b | 2014-12-16 16:25:28 -0600 | [diff] [blame] | 7767 | if (task_stack_end_corrupted(current)) |
| 7768 | printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); |
| 7769 | |
Ingo Molnar | aef745f | 2008-08-28 11:34:43 +0200 | [diff] [blame] | 7770 | debug_show_held_locks(current); |
| 7771 | if (irqs_disabled()) |
| 7772 | print_irqtrace_events(current); |
Vegard Nossum | d1c6d14 | 2016-07-23 09:46:39 +0200 | [diff] [blame] | 7773 | if (IS_ENABLED(CONFIG_DEBUG_PREEMPT) |
| 7774 | && !preempt_count_equals(preempt_offset)) { |
Thomas Gleixner | 8f47b18 | 2014-02-07 20:58:39 +0100 | [diff] [blame] | 7775 | pr_err("Preemption disabled at:"); |
Vegard Nossum | d1c6d14 | 2016-07-23 09:46:39 +0200 | [diff] [blame] | 7776 | print_ip_sym(preempt_disable_ip); |
Thomas Gleixner | 8f47b18 | 2014-02-07 20:58:39 +0100 | [diff] [blame] | 7777 | pr_cont("\n"); |
| 7778 | } |
Ingo Molnar | aef745f | 2008-08-28 11:34:43 +0200 | [diff] [blame] | 7779 | dump_stack(); |
Vegard Nossum | f0b22e3 | 2016-07-22 21:46:02 +0200 | [diff] [blame] | 7780 | add_taint(TAINT_WARN, LOCKDEP_STILL_OK); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7781 | } |
Peter Zijlstra | 3427445 | 2014-09-24 10:18:56 +0200 | [diff] [blame] | 7782 | EXPORT_SYMBOL(___might_sleep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7783 | #endif |
| 7784 | |
| 7785 | #ifdef CONFIG_MAGIC_SYSRQ |
| 7786 | void normalize_rt_tasks(void) |
| 7787 | { |
| 7788 | struct task_struct *g, *p; |
Peter Zijlstra | dbc7f06 | 2015-06-11 14:46:38 +0200 | [diff] [blame] | 7789 | struct sched_attr attr = { |
| 7790 | .sched_policy = SCHED_NORMAL, |
| 7791 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7792 | |
Oleg Nesterov | 3472eaa | 2014-09-21 21:33:38 +0200 | [diff] [blame] | 7793 | read_lock(&tasklist_lock); |
Oleg Nesterov | 5d07f42 | 2014-08-13 21:19:53 +0200 | [diff] [blame] | 7794 | for_each_process_thread(g, p) { |
Ingo Molnar | 178be79 | 2007-10-15 17:00:18 +0200 | [diff] [blame] | 7795 | /* |
| 7796 | * Only normalize user tasks: |
| 7797 | */ |
Oleg Nesterov | 3472eaa | 2014-09-21 21:33:38 +0200 | [diff] [blame] | 7798 | if (p->flags & PF_KTHREAD) |
Ingo Molnar | 178be79 | 2007-10-15 17:00:18 +0200 | [diff] [blame] | 7799 | continue; |
| 7800 | |
Josh Poimboeuf | 4fa8d299 | 2016-06-17 12:43:26 -0500 | [diff] [blame] | 7801 | p->se.exec_start = 0; |
| 7802 | schedstat_set(p->se.statistics.wait_start, 0); |
| 7803 | schedstat_set(p->se.statistics.sleep_start, 0); |
| 7804 | schedstat_set(p->se.statistics.block_start, 0); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 7805 | |
Dario Faggioli | aab03e0 | 2013-11-28 11:14:43 +0100 | [diff] [blame] | 7806 | if (!dl_task(p) && !rt_task(p)) { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 7807 | /* |
| 7808 | * Renice negative nice level userspace |
| 7809 | * tasks back to 0: |
| 7810 | */ |
Oleg Nesterov | 3472eaa | 2014-09-21 21:33:38 +0200 | [diff] [blame] | 7811 | if (task_nice(p) < 0) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 7812 | set_user_nice(p, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7813 | continue; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 7814 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7815 | |
Peter Zijlstra | dbc7f06 | 2015-06-11 14:46:38 +0200 | [diff] [blame] | 7816 | __sched_setscheduler(p, &attr, false, false); |
Oleg Nesterov | 5d07f42 | 2014-08-13 21:19:53 +0200 | [diff] [blame] | 7817 | } |
Oleg Nesterov | 3472eaa | 2014-09-21 21:33:38 +0200 | [diff] [blame] | 7818 | read_unlock(&tasklist_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7819 | } |
| 7820 | |
| 7821 | #endif /* CONFIG_MAGIC_SYSRQ */ |
Linus Torvalds | 1df5c10 | 2005-09-12 07:59:21 -0700 | [diff] [blame] | 7822 | |
Jason Wessel | 67fc4e0 | 2010-05-20 21:04:21 -0500 | [diff] [blame] | 7823 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
Linus Torvalds | 1df5c10 | 2005-09-12 07:59:21 -0700 | [diff] [blame] | 7824 | /* |
Jason Wessel | 67fc4e0 | 2010-05-20 21:04:21 -0500 | [diff] [blame] | 7825 | * These functions are only useful for the IA64 MCA handling, or kdb. |
Linus Torvalds | 1df5c10 | 2005-09-12 07:59:21 -0700 | [diff] [blame] | 7826 | * |
| 7827 | * They can only be called when the whole system has been |
| 7828 | * stopped - every CPU needs to be quiescent, and no scheduling |
| 7829 | * activity can take place. Using them for anything else would |
| 7830 | * be a serious bug, and as a result, they aren't even visible |
| 7831 | * under any other configuration. |
| 7832 | */ |
| 7833 | |
| 7834 | /** |
| 7835 | * curr_task - return the current task for a given cpu. |
| 7836 | * @cpu: the processor in question. |
| 7837 | * |
| 7838 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! |
Yacine Belkadi | e69f618 | 2013-07-12 20:45:47 +0200 | [diff] [blame] | 7839 | * |
| 7840 | * Return: The current task for @cpu. |
Linus Torvalds | 1df5c10 | 2005-09-12 07:59:21 -0700 | [diff] [blame] | 7841 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 7842 | struct task_struct *curr_task(int cpu) |
Linus Torvalds | 1df5c10 | 2005-09-12 07:59:21 -0700 | [diff] [blame] | 7843 | { |
| 7844 | return cpu_curr(cpu); |
| 7845 | } |
| 7846 | |
Jason Wessel | 67fc4e0 | 2010-05-20 21:04:21 -0500 | [diff] [blame] | 7847 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
| 7848 | |
| 7849 | #ifdef CONFIG_IA64 |
Linus Torvalds | 1df5c10 | 2005-09-12 07:59:21 -0700 | [diff] [blame] | 7850 | /** |
| 7851 | * set_curr_task - set the current task for a given cpu. |
| 7852 | * @cpu: the processor in question. |
| 7853 | * @p: the task pointer to set. |
| 7854 | * |
| 7855 | * Description: This function must only be used when non-maskable interrupts |
Ingo Molnar | 41a2d6c | 2007-12-05 15:46:09 +0100 | [diff] [blame] | 7856 | * are serviced on a separate stack. It allows the architecture to switch the |
| 7857 | * notion of the current task on a cpu in a non-blocking manner. This function |
Linus Torvalds | 1df5c10 | 2005-09-12 07:59:21 -0700 | [diff] [blame] | 7858 | * must be called with all CPU's synchronized, and interrupts disabled, the |
| 7859 | * and caller must save the original value of the current task (see |
| 7860 | * curr_task() above) and restore that value before reenabling interrupts and |
| 7861 | * re-starting the system. |
| 7862 | * |
| 7863 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! |
| 7864 | */ |
Peter Zijlstra | a458ae2 | 2016-09-20 20:29:40 +0200 | [diff] [blame] | 7865 | void ia64_set_curr_task(int cpu, struct task_struct *p) |
Linus Torvalds | 1df5c10 | 2005-09-12 07:59:21 -0700 | [diff] [blame] | 7866 | { |
| 7867 | cpu_curr(cpu) = p; |
| 7868 | } |
| 7869 | |
| 7870 | #endif |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7871 | |
Dhaval Giani | 7c94143 | 2010-01-20 13:26:18 +0100 | [diff] [blame] | 7872 | #ifdef CONFIG_CGROUP_SCHED |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 7873 | /* task_group_lock serializes the addition/removal of task groups */ |
| 7874 | static DEFINE_SPINLOCK(task_group_lock); |
| 7875 | |
Peter Zijlstra | 2f5177f | 2016-03-16 16:22:45 +0100 | [diff] [blame] | 7876 | static void sched_free_group(struct task_group *tg) |
Peter Zijlstra | bccbe08 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7877 | { |
| 7878 | free_fair_sched_group(tg); |
| 7879 | free_rt_sched_group(tg); |
Mike Galbraith | e9aa1dd | 2011-01-05 11:11:25 +0100 | [diff] [blame] | 7880 | autogroup_free(tg); |
Waiman Long | b036762 | 2015-12-02 13:41:49 -0500 | [diff] [blame] | 7881 | kmem_cache_free(task_group_cache, tg); |
Peter Zijlstra | bccbe08 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7882 | } |
| 7883 | |
| 7884 | /* allocate runqueue etc for a new task group */ |
Dhaval Giani | ec7dc8a | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 7885 | struct task_group *sched_create_group(struct task_group *parent) |
Peter Zijlstra | bccbe08 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7886 | { |
| 7887 | struct task_group *tg; |
Peter Zijlstra | bccbe08 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7888 | |
Waiman Long | b036762 | 2015-12-02 13:41:49 -0500 | [diff] [blame] | 7889 | tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO); |
Peter Zijlstra | bccbe08 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7890 | if (!tg) |
| 7891 | return ERR_PTR(-ENOMEM); |
| 7892 | |
Dhaval Giani | ec7dc8a | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 7893 | if (!alloc_fair_sched_group(tg, parent)) |
Peter Zijlstra | bccbe08 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7894 | goto err; |
| 7895 | |
Dhaval Giani | ec7dc8a | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 7896 | if (!alloc_rt_sched_group(tg, parent)) |
Peter Zijlstra | bccbe08 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 7897 | goto err; |
| 7898 | |
Li Zefan | ace783b | 2013-01-24 14:30:48 +0800 | [diff] [blame] | 7899 | return tg; |
| 7900 | |
| 7901 | err: |
Peter Zijlstra | 2f5177f | 2016-03-16 16:22:45 +0100 | [diff] [blame] | 7902 | sched_free_group(tg); |
Li Zefan | ace783b | 2013-01-24 14:30:48 +0800 | [diff] [blame] | 7903 | return ERR_PTR(-ENOMEM); |
| 7904 | } |
| 7905 | |
| 7906 | void sched_online_group(struct task_group *tg, struct task_group *parent) |
| 7907 | { |
| 7908 | unsigned long flags; |
| 7909 | |
Peter Zijlstra | 8ed3699 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7910 | spin_lock_irqsave(&task_group_lock, flags); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7911 | list_add_rcu(&tg->list, &task_groups); |
Peter Zijlstra | f473aa5 | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 7912 | |
| 7913 | WARN_ON(!parent); /* root should already exist */ |
| 7914 | |
| 7915 | tg->parent = parent; |
Peter Zijlstra | f473aa5 | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 7916 | INIT_LIST_HEAD(&tg->children); |
Zhang, Yanmin | 09f2724 | 2030-08-14 15:56:40 +0800 | [diff] [blame] | 7917 | list_add_rcu(&tg->siblings, &parent->children); |
Peter Zijlstra | 8ed3699 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7918 | spin_unlock_irqrestore(&task_group_lock, flags); |
Peter Zijlstra | 8663e24 | 2016-06-22 14:58:02 +0200 | [diff] [blame] | 7919 | |
| 7920 | online_fair_sched_group(tg); |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7921 | } |
| 7922 | |
Srivatsa Vaddagiri | 9b5b775 | 2007-10-15 17:00:09 +0200 | [diff] [blame] | 7923 | /* rcu callback to free various structures associated with a task group */ |
Peter Zijlstra | 2f5177f | 2016-03-16 16:22:45 +0100 | [diff] [blame] | 7924 | static void sched_free_group_rcu(struct rcu_head *rhp) |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7925 | { |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7926 | /* now it should be safe to free those cfs_rqs */ |
Peter Zijlstra | 2f5177f | 2016-03-16 16:22:45 +0100 | [diff] [blame] | 7927 | sched_free_group(container_of(rhp, struct task_group, rcu)); |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7928 | } |
| 7929 | |
Ingo Molnar | 4cf86d7 | 2007-10-15 17:00:14 +0200 | [diff] [blame] | 7930 | void sched_destroy_group(struct task_group *tg) |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7931 | { |
Li Zefan | ace783b | 2013-01-24 14:30:48 +0800 | [diff] [blame] | 7932 | /* wait for possible concurrent references to cfs_rqs complete */ |
Peter Zijlstra | 2f5177f | 2016-03-16 16:22:45 +0100 | [diff] [blame] | 7933 | call_rcu(&tg->rcu, sched_free_group_rcu); |
Li Zefan | ace783b | 2013-01-24 14:30:48 +0800 | [diff] [blame] | 7934 | } |
| 7935 | |
| 7936 | void sched_offline_group(struct task_group *tg) |
| 7937 | { |
Peter Zijlstra | 8ed3699 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7938 | unsigned long flags; |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7939 | |
Peter Zijlstra | 3d4b47b | 2010-11-15 15:47:01 -0800 | [diff] [blame] | 7940 | /* end participation in shares distribution */ |
Peter Zijlstra | 6fe1f34 | 2016-01-21 22:24:16 +0100 | [diff] [blame] | 7941 | unregister_fair_sched_group(tg); |
Peter Zijlstra | 3d4b47b | 2010-11-15 15:47:01 -0800 | [diff] [blame] | 7942 | |
| 7943 | spin_lock_irqsave(&task_group_lock, flags); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 7944 | list_del_rcu(&tg->list); |
Peter Zijlstra | f473aa5 | 2008-04-19 19:45:00 +0200 | [diff] [blame] | 7945 | list_del_rcu(&tg->siblings); |
Peter Zijlstra | 8ed3699 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 7946 | spin_unlock_irqrestore(&task_group_lock, flags); |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7947 | } |
| 7948 | |
Vincent Guittot | ea86cb4 | 2016-06-17 13:38:55 +0200 | [diff] [blame] | 7949 | static void sched_change_group(struct task_struct *tsk, int type) |
| 7950 | { |
| 7951 | struct task_group *tg; |
| 7952 | |
| 7953 | /* |
| 7954 | * All callers are synchronized by task_rq_lock(); we do not use RCU |
| 7955 | * which is pointless here. Thus, we pass "true" to task_css_check() |
| 7956 | * to prevent lockdep warnings. |
| 7957 | */ |
| 7958 | tg = container_of(task_css_check(tsk, cpu_cgrp_id, true), |
| 7959 | struct task_group, css); |
| 7960 | tg = autogroup_task_group(tsk, tg); |
| 7961 | tsk->sched_task_group = tg; |
| 7962 | |
| 7963 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 7964 | if (tsk->sched_class->task_change_group) |
| 7965 | tsk->sched_class->task_change_group(tsk, type); |
| 7966 | else |
| 7967 | #endif |
| 7968 | set_task_rq(tsk, task_cpu(tsk)); |
| 7969 | } |
| 7970 | |
| 7971 | /* |
| 7972 | * Change task's runqueue when it moves between groups. |
| 7973 | * |
| 7974 | * The caller of this function should have put the task in its new group by |
| 7975 | * now. This function just updates tsk->se.cfs_rq and tsk->se.parent to reflect |
| 7976 | * its new group. |
Srivatsa Vaddagiri | 9b5b775 | 2007-10-15 17:00:09 +0200 | [diff] [blame] | 7977 | */ |
| 7978 | void sched_move_task(struct task_struct *tsk) |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7979 | { |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 7980 | int queued, running; |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 7981 | struct rq_flags rf; |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7982 | struct rq *rq; |
| 7983 | |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 7984 | rq = task_rq_lock(tsk, &rf); |
Peter Zijlstra | 6da1c98 | 2017-01-23 16:05:55 +0100 | [diff] [blame] | 7985 | update_rq_clock(rq); |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7986 | |
Dmitry Adamushko | 051a1d1 | 2007-12-18 15:21:13 +0100 | [diff] [blame] | 7987 | running = task_current(rq, tsk); |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 7988 | queued = task_on_rq_queued(tsk); |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7989 | |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 7990 | if (queued) |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 7991 | dequeue_task(rq, tsk, DEQUEUE_SAVE | DEQUEUE_MOVE); |
Hiroshi Shimamoto | 0e1f348 | 2008-03-10 11:01:20 -0700 | [diff] [blame] | 7992 | if (unlikely(running)) |
Kirill Tkhai | f3cd1c4 | 2014-09-12 17:41:40 +0400 | [diff] [blame] | 7993 | put_prev_task(rq, tsk); |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 7994 | |
Vincent Guittot | ea86cb4 | 2016-06-17 13:38:55 +0200 | [diff] [blame] | 7995 | sched_change_group(tsk, TASK_MOVE_GROUP); |
Peter Zijlstra | 810b381 | 2008-02-29 15:21:01 -0500 | [diff] [blame] | 7996 | |
Kirill Tkhai | da0c1e6 | 2014-08-20 13:47:32 +0400 | [diff] [blame] | 7997 | if (queued) |
Peter Zijlstra | ff77e46 | 2016-01-18 15:27:07 +0100 | [diff] [blame] | 7998 | enqueue_task(rq, tsk, ENQUEUE_RESTORE | ENQUEUE_MOVE); |
Vincent Guittot | a399d23 | 2016-09-12 09:47:52 +0200 | [diff] [blame] | 7999 | if (unlikely(running)) |
Peter Zijlstra | b2bf6c3 | 2016-09-20 22:00:38 +0200 | [diff] [blame] | 8000 | set_curr_task(rq, tsk); |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 8001 | |
Peter Zijlstra | eb58075 | 2015-07-31 21:28:18 +0200 | [diff] [blame] | 8002 | task_rq_unlock(rq, tsk, &rf); |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 8003 | } |
Dhaval Giani | 7c94143 | 2010-01-20 13:26:18 +0100 | [diff] [blame] | 8004 | #endif /* CONFIG_CGROUP_SCHED */ |
Srivatsa Vaddagiri | 29f59db | 2007-10-15 17:00:07 +0200 | [diff] [blame] | 8005 | |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 8006 | #ifdef CONFIG_RT_GROUP_SCHED |
| 8007 | /* |
| 8008 | * Ensure that the real time constraints are schedulable. |
| 8009 | */ |
| 8010 | static DEFINE_MUTEX(rt_constraints_mutex); |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 8011 | |
Dhaval Giani | 521f1a24 | 2008-02-28 15:21:56 +0530 | [diff] [blame] | 8012 | /* Must be called with tasklist_lock held */ |
| 8013 | static inline int tg_has_rt_tasks(struct task_group *tg) |
| 8014 | { |
| 8015 | struct task_struct *g, *p; |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 8016 | |
Peter Zijlstra | 1fe89e1 | 2015-02-09 11:53:18 +0100 | [diff] [blame] | 8017 | /* |
| 8018 | * Autogroups do not have RT tasks; see autogroup_create(). |
| 8019 | */ |
| 8020 | if (task_group_is_autogroup(tg)) |
| 8021 | return 0; |
| 8022 | |
Oleg Nesterov | 5d07f42 | 2014-08-13 21:19:53 +0200 | [diff] [blame] | 8023 | for_each_process_thread(g, p) { |
Oleg Nesterov | 8651c65 | 2014-09-21 21:33:36 +0200 | [diff] [blame] | 8024 | if (rt_task(p) && task_group(p) == tg) |
Dhaval Giani | 521f1a24 | 2008-02-28 15:21:56 +0530 | [diff] [blame] | 8025 | return 1; |
Oleg Nesterov | 5d07f42 | 2014-08-13 21:19:53 +0200 | [diff] [blame] | 8026 | } |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 8027 | |
Dhaval Giani | 521f1a24 | 2008-02-28 15:21:56 +0530 | [diff] [blame] | 8028 | return 0; |
| 8029 | } |
| 8030 | |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 8031 | struct rt_schedulable_data { |
| 8032 | struct task_group *tg; |
| 8033 | u64 rt_period; |
| 8034 | u64 rt_runtime; |
| 8035 | }; |
| 8036 | |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 8037 | static int tg_rt_schedulable(struct task_group *tg, void *data) |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 8038 | { |
| 8039 | struct rt_schedulable_data *d = data; |
| 8040 | struct task_group *child; |
| 8041 | unsigned long total, sum = 0; |
| 8042 | u64 period, runtime; |
| 8043 | |
| 8044 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
| 8045 | runtime = tg->rt_bandwidth.rt_runtime; |
| 8046 | |
| 8047 | if (tg == d->tg) { |
| 8048 | period = d->rt_period; |
| 8049 | runtime = d->rt_runtime; |
| 8050 | } |
| 8051 | |
Peter Zijlstra | 4653f80 | 2008-09-23 15:33:44 +0200 | [diff] [blame] | 8052 | /* |
| 8053 | * Cannot have more runtime than the period. |
| 8054 | */ |
| 8055 | if (runtime > period && runtime != RUNTIME_INF) |
| 8056 | return -EINVAL; |
| 8057 | |
| 8058 | /* |
| 8059 | * Ensure we don't starve existing RT tasks. |
| 8060 | */ |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 8061 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
| 8062 | return -EBUSY; |
| 8063 | |
| 8064 | total = to_ratio(period, runtime); |
| 8065 | |
Peter Zijlstra | 4653f80 | 2008-09-23 15:33:44 +0200 | [diff] [blame] | 8066 | /* |
| 8067 | * Nobody can have more than the global setting allows. |
| 8068 | */ |
| 8069 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) |
| 8070 | return -EINVAL; |
| 8071 | |
| 8072 | /* |
| 8073 | * The sum of our children's runtime should not exceed our own. |
| 8074 | */ |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 8075 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
| 8076 | period = ktime_to_ns(child->rt_bandwidth.rt_period); |
| 8077 | runtime = child->rt_bandwidth.rt_runtime; |
| 8078 | |
| 8079 | if (child == d->tg) { |
| 8080 | period = d->rt_period; |
| 8081 | runtime = d->rt_runtime; |
| 8082 | } |
| 8083 | |
| 8084 | sum += to_ratio(period, runtime); |
| 8085 | } |
| 8086 | |
| 8087 | if (sum > total) |
| 8088 | return -EINVAL; |
| 8089 | |
| 8090 | return 0; |
| 8091 | } |
| 8092 | |
| 8093 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
| 8094 | { |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 8095 | int ret; |
| 8096 | |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 8097 | struct rt_schedulable_data data = { |
| 8098 | .tg = tg, |
| 8099 | .rt_period = period, |
| 8100 | .rt_runtime = runtime, |
| 8101 | }; |
| 8102 | |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 8103 | rcu_read_lock(); |
| 8104 | ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data); |
| 8105 | rcu_read_unlock(); |
| 8106 | |
| 8107 | return ret; |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 8108 | } |
| 8109 | |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8110 | static int tg_set_rt_bandwidth(struct task_group *tg, |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8111 | u64 rt_period, u64 rt_runtime) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 8112 | { |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 8113 | int i, err = 0; |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 8114 | |
Peter Zijlstra | 2636ed5 | 2015-02-09 12:23:20 +0100 | [diff] [blame] | 8115 | /* |
| 8116 | * Disallowing the root group RT runtime is BAD, it would disallow the |
| 8117 | * kernel creating (and or operating) RT threads. |
| 8118 | */ |
| 8119 | if (tg == &root_task_group && rt_runtime == 0) |
| 8120 | return -EINVAL; |
| 8121 | |
| 8122 | /* No period doesn't make any sense. */ |
| 8123 | if (rt_period == 0) |
| 8124 | return -EINVAL; |
| 8125 | |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 8126 | mutex_lock(&rt_constraints_mutex); |
Dhaval Giani | 521f1a24 | 2008-02-28 15:21:56 +0530 | [diff] [blame] | 8127 | read_lock(&tasklist_lock); |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 8128 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
| 8129 | if (err) |
Dhaval Giani | 521f1a24 | 2008-02-28 15:21:56 +0530 | [diff] [blame] | 8130 | goto unlock; |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 8131 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 8132 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8133 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
| 8134 | tg->rt_bandwidth.rt_runtime = rt_runtime; |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 8135 | |
| 8136 | for_each_possible_cpu(i) { |
| 8137 | struct rt_rq *rt_rq = tg->rt_rq[i]; |
| 8138 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 8139 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 8140 | rt_rq->rt_runtime = rt_runtime; |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 8141 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 8142 | } |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 8143 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
Peter Zijlstra | 4924627 | 2010-10-17 21:46:10 +0200 | [diff] [blame] | 8144 | unlock: |
Dhaval Giani | 521f1a24 | 2008-02-28 15:21:56 +0530 | [diff] [blame] | 8145 | read_unlock(&tasklist_lock); |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 8146 | mutex_unlock(&rt_constraints_mutex); |
| 8147 | |
| 8148 | return err; |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 8149 | } |
| 8150 | |
Li Zefan | 25cc7da | 2013-03-05 16:07:33 +0800 | [diff] [blame] | 8151 | static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8152 | { |
| 8153 | u64 rt_runtime, rt_period; |
| 8154 | |
| 8155 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
| 8156 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; |
| 8157 | if (rt_runtime_us < 0) |
| 8158 | rt_runtime = RUNTIME_INF; |
| 8159 | |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8160 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8161 | } |
| 8162 | |
Li Zefan | 25cc7da | 2013-03-05 16:07:33 +0800 | [diff] [blame] | 8163 | static long sched_group_rt_runtime(struct task_group *tg) |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 8164 | { |
| 8165 | u64 rt_runtime_us; |
| 8166 | |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8167 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 8168 | return -1; |
| 8169 | |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8170 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 8171 | do_div(rt_runtime_us, NSEC_PER_USEC); |
| 8172 | return rt_runtime_us; |
| 8173 | } |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8174 | |
Nicholas Mc Guire | ce2f5fe | 2015-05-03 10:51:56 +0200 | [diff] [blame] | 8175 | static int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us) |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8176 | { |
| 8177 | u64 rt_runtime, rt_period; |
| 8178 | |
Nicholas Mc Guire | ce2f5fe | 2015-05-03 10:51:56 +0200 | [diff] [blame] | 8179 | rt_period = rt_period_us * NSEC_PER_USEC; |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8180 | rt_runtime = tg->rt_bandwidth.rt_runtime; |
| 8181 | |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8182 | return tg_set_rt_bandwidth(tg, rt_period, rt_runtime); |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8183 | } |
| 8184 | |
Li Zefan | 25cc7da | 2013-03-05 16:07:33 +0800 | [diff] [blame] | 8185 | static long sched_group_rt_period(struct task_group *tg) |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8186 | { |
| 8187 | u64 rt_period_us; |
| 8188 | |
| 8189 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); |
| 8190 | do_div(rt_period_us, NSEC_PER_USEC); |
| 8191 | return rt_period_us; |
| 8192 | } |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 8193 | #endif /* CONFIG_RT_GROUP_SCHED */ |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8194 | |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 8195 | #ifdef CONFIG_RT_GROUP_SCHED |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8196 | static int sched_rt_global_constraints(void) |
| 8197 | { |
| 8198 | int ret = 0; |
| 8199 | |
| 8200 | mutex_lock(&rt_constraints_mutex); |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 8201 | read_lock(&tasklist_lock); |
Peter Zijlstra | 4653f80 | 2008-09-23 15:33:44 +0200 | [diff] [blame] | 8202 | ret = __rt_schedulable(NULL, 0, 0); |
Peter Zijlstra | 9a7e0b1 | 2008-08-19 12:33:06 +0200 | [diff] [blame] | 8203 | read_unlock(&tasklist_lock); |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8204 | mutex_unlock(&rt_constraints_mutex); |
| 8205 | |
| 8206 | return ret; |
| 8207 | } |
Dhaval Giani | 54e9912 | 2009-02-27 15:13:54 +0530 | [diff] [blame] | 8208 | |
Li Zefan | 25cc7da | 2013-03-05 16:07:33 +0800 | [diff] [blame] | 8209 | static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) |
Dhaval Giani | 54e9912 | 2009-02-27 15:13:54 +0530 | [diff] [blame] | 8210 | { |
| 8211 | /* Don't accept realtime tasks when there is no way for them to run */ |
| 8212 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) |
| 8213 | return 0; |
| 8214 | |
| 8215 | return 1; |
| 8216 | } |
| 8217 | |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 8218 | #else /* !CONFIG_RT_GROUP_SCHED */ |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8219 | static int sched_rt_global_constraints(void) |
| 8220 | { |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 8221 | unsigned long flags; |
Muhammad Falak R Wani | 8c5e955 | 2016-05-05 15:21:19 +0530 | [diff] [blame] | 8222 | int i; |
Hiroshi Shimamoto | ec5d498 | 2008-09-10 17:00:19 -0700 | [diff] [blame] | 8223 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 8224 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 8225 | for_each_possible_cpu(i) { |
| 8226 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; |
| 8227 | |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 8228 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 8229 | rt_rq->rt_runtime = global_rt_runtime(); |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 8230 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 8231 | } |
Thomas Gleixner | 0986b11 | 2009-11-17 15:32:06 +0100 | [diff] [blame] | 8232 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
Peter Zijlstra | ac086bc | 2008-04-19 19:44:58 +0200 | [diff] [blame] | 8233 | |
Muhammad Falak R Wani | 8c5e955 | 2016-05-05 15:21:19 +0530 | [diff] [blame] | 8234 | return 0; |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 8235 | } |
| 8236 | #endif /* CONFIG_RT_GROUP_SCHED */ |
| 8237 | |
Wanpeng Li | a1963b8 | 2015-03-17 19:15:31 +0800 | [diff] [blame] | 8238 | static int sched_dl_global_validate(void) |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 8239 | { |
Peter Zijlstra | 1724813 | 2013-12-17 12:44:49 +0100 | [diff] [blame] | 8240 | u64 runtime = global_rt_runtime(); |
| 8241 | u64 period = global_rt_period(); |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 8242 | u64 new_bw = to_ratio(period, runtime); |
Kirill Tkhai | f10e00f | 2014-09-30 12:23:37 +0400 | [diff] [blame] | 8243 | struct dl_bw *dl_b; |
Peter Zijlstra | 1724813 | 2013-12-17 12:44:49 +0100 | [diff] [blame] | 8244 | int cpu, ret = 0; |
Juri Lelli | 4951634 | 2014-02-11 09:24:27 +0100 | [diff] [blame] | 8245 | unsigned long flags; |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 8246 | |
| 8247 | /* |
| 8248 | * Here we want to check the bandwidth not being set to some |
| 8249 | * value smaller than the currently allocated bandwidth in |
| 8250 | * any of the root_domains. |
| 8251 | * |
| 8252 | * FIXME: Cycling on all the CPUs is overdoing, but simpler than |
| 8253 | * cycling on root_domains... Discussion on different/better |
| 8254 | * solutions is welcome! |
| 8255 | */ |
Peter Zijlstra | 1724813 | 2013-12-17 12:44:49 +0100 | [diff] [blame] | 8256 | for_each_possible_cpu(cpu) { |
Kirill Tkhai | f10e00f | 2014-09-30 12:23:37 +0400 | [diff] [blame] | 8257 | rcu_read_lock_sched(); |
| 8258 | dl_b = dl_bw_of(cpu); |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 8259 | |
Juri Lelli | 4951634 | 2014-02-11 09:24:27 +0100 | [diff] [blame] | 8260 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
Peter Zijlstra | 1724813 | 2013-12-17 12:44:49 +0100 | [diff] [blame] | 8261 | if (new_bw < dl_b->total_bw) |
| 8262 | ret = -EBUSY; |
Juri Lelli | 4951634 | 2014-02-11 09:24:27 +0100 | [diff] [blame] | 8263 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
Peter Zijlstra | 1724813 | 2013-12-17 12:44:49 +0100 | [diff] [blame] | 8264 | |
Kirill Tkhai | f10e00f | 2014-09-30 12:23:37 +0400 | [diff] [blame] | 8265 | rcu_read_unlock_sched(); |
| 8266 | |
Peter Zijlstra | 1724813 | 2013-12-17 12:44:49 +0100 | [diff] [blame] | 8267 | if (ret) |
| 8268 | break; |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 8269 | } |
| 8270 | |
Peter Zijlstra | 1724813 | 2013-12-17 12:44:49 +0100 | [diff] [blame] | 8271 | return ret; |
| 8272 | } |
| 8273 | |
| 8274 | static void sched_dl_do_global(void) |
| 8275 | { |
| 8276 | u64 new_bw = -1; |
Kirill Tkhai | f10e00f | 2014-09-30 12:23:37 +0400 | [diff] [blame] | 8277 | struct dl_bw *dl_b; |
Peter Zijlstra | 1724813 | 2013-12-17 12:44:49 +0100 | [diff] [blame] | 8278 | int cpu; |
Juri Lelli | 4951634 | 2014-02-11 09:24:27 +0100 | [diff] [blame] | 8279 | unsigned long flags; |
Peter Zijlstra | 1724813 | 2013-12-17 12:44:49 +0100 | [diff] [blame] | 8280 | |
| 8281 | def_dl_bandwidth.dl_period = global_rt_period(); |
| 8282 | def_dl_bandwidth.dl_runtime = global_rt_runtime(); |
| 8283 | |
| 8284 | if (global_rt_runtime() != RUNTIME_INF) |
| 8285 | new_bw = to_ratio(global_rt_period(), global_rt_runtime()); |
| 8286 | |
| 8287 | /* |
| 8288 | * FIXME: As above... |
| 8289 | */ |
| 8290 | for_each_possible_cpu(cpu) { |
Kirill Tkhai | f10e00f | 2014-09-30 12:23:37 +0400 | [diff] [blame] | 8291 | rcu_read_lock_sched(); |
| 8292 | dl_b = dl_bw_of(cpu); |
Peter Zijlstra | 1724813 | 2013-12-17 12:44:49 +0100 | [diff] [blame] | 8293 | |
Juri Lelli | 4951634 | 2014-02-11 09:24:27 +0100 | [diff] [blame] | 8294 | raw_spin_lock_irqsave(&dl_b->lock, flags); |
Peter Zijlstra | 1724813 | 2013-12-17 12:44:49 +0100 | [diff] [blame] | 8295 | dl_b->bw = new_bw; |
Juri Lelli | 4951634 | 2014-02-11 09:24:27 +0100 | [diff] [blame] | 8296 | raw_spin_unlock_irqrestore(&dl_b->lock, flags); |
Kirill Tkhai | f10e00f | 2014-09-30 12:23:37 +0400 | [diff] [blame] | 8297 | |
| 8298 | rcu_read_unlock_sched(); |
Peter Zijlstra | 1724813 | 2013-12-17 12:44:49 +0100 | [diff] [blame] | 8299 | } |
| 8300 | } |
| 8301 | |
| 8302 | static int sched_rt_global_validate(void) |
| 8303 | { |
| 8304 | if (sysctl_sched_rt_period <= 0) |
| 8305 | return -EINVAL; |
| 8306 | |
Juri Lelli | e9e7cb3 | 2014-02-11 09:24:26 +0100 | [diff] [blame] | 8307 | if ((sysctl_sched_rt_runtime != RUNTIME_INF) && |
| 8308 | (sysctl_sched_rt_runtime > sysctl_sched_rt_period)) |
Peter Zijlstra | 1724813 | 2013-12-17 12:44:49 +0100 | [diff] [blame] | 8309 | return -EINVAL; |
| 8310 | |
Dario Faggioli | 332ac17 | 2013-11-07 14:43:45 +0100 | [diff] [blame] | 8311 | return 0; |
| 8312 | } |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8313 | |
Peter Zijlstra | 1724813 | 2013-12-17 12:44:49 +0100 | [diff] [blame] | 8314 | static void sched_rt_do_global(void) |
| 8315 | { |
| 8316 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); |
| 8317 | def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period()); |
| 8318 | } |
| 8319 | |
| 8320 | int sched_rt_handler(struct ctl_table *table, int write, |
| 8321 | void __user *buffer, size_t *lenp, |
| 8322 | loff_t *ppos) |
| 8323 | { |
| 8324 | int old_period, old_runtime; |
| 8325 | static DEFINE_MUTEX(mutex); |
| 8326 | int ret; |
| 8327 | |
| 8328 | mutex_lock(&mutex); |
| 8329 | old_period = sysctl_sched_rt_period; |
| 8330 | old_runtime = sysctl_sched_rt_runtime; |
| 8331 | |
| 8332 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
| 8333 | |
| 8334 | if (!ret && write) { |
| 8335 | ret = sched_rt_global_validate(); |
| 8336 | if (ret) |
| 8337 | goto undo; |
| 8338 | |
Wanpeng Li | a1963b8 | 2015-03-17 19:15:31 +0800 | [diff] [blame] | 8339 | ret = sched_dl_global_validate(); |
Peter Zijlstra | 1724813 | 2013-12-17 12:44:49 +0100 | [diff] [blame] | 8340 | if (ret) |
| 8341 | goto undo; |
| 8342 | |
Wanpeng Li | a1963b8 | 2015-03-17 19:15:31 +0800 | [diff] [blame] | 8343 | ret = sched_rt_global_constraints(); |
Peter Zijlstra | 1724813 | 2013-12-17 12:44:49 +0100 | [diff] [blame] | 8344 | if (ret) |
| 8345 | goto undo; |
| 8346 | |
| 8347 | sched_rt_do_global(); |
| 8348 | sched_dl_do_global(); |
| 8349 | } |
| 8350 | if (0) { |
| 8351 | undo: |
| 8352 | sysctl_sched_rt_period = old_period; |
| 8353 | sysctl_sched_rt_runtime = old_runtime; |
| 8354 | } |
| 8355 | mutex_unlock(&mutex); |
| 8356 | |
| 8357 | return ret; |
| 8358 | } |
| 8359 | |
Clark Williams | ce0dbbb | 2013-02-07 09:47:04 -0600 | [diff] [blame] | 8360 | int sched_rr_handler(struct ctl_table *table, int write, |
| 8361 | void __user *buffer, size_t *lenp, |
| 8362 | loff_t *ppos) |
| 8363 | { |
| 8364 | int ret; |
| 8365 | static DEFINE_MUTEX(mutex); |
| 8366 | |
| 8367 | mutex_lock(&mutex); |
| 8368 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
| 8369 | /* make sure that internally we keep jiffies */ |
| 8370 | /* also, writing zero resets timeslice to default */ |
| 8371 | if (!ret && write) { |
| 8372 | sched_rr_timeslice = sched_rr_timeslice <= 0 ? |
| 8373 | RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice); |
| 8374 | } |
| 8375 | mutex_unlock(&mutex); |
| 8376 | return ret; |
| 8377 | } |
| 8378 | |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 8379 | #ifdef CONFIG_CGROUP_SCHED |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8380 | |
Tejun Heo | a7c6d55 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 8381 | static inline struct task_group *css_tg(struct cgroup_subsys_state *css) |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8382 | { |
Tejun Heo | a7c6d55 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 8383 | return css ? container_of(css, struct task_group, css) : NULL; |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8384 | } |
| 8385 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 8386 | static struct cgroup_subsys_state * |
| 8387 | cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8388 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 8389 | struct task_group *parent = css_tg(parent_css); |
| 8390 | struct task_group *tg; |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8391 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 8392 | if (!parent) { |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8393 | /* This is early initialization for the top cgroup */ |
Yong Zhang | 07e06b0 | 2011-01-07 15:17:36 +0800 | [diff] [blame] | 8394 | return &root_task_group.css; |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8395 | } |
| 8396 | |
Dhaval Giani | ec7dc8a | 2008-04-19 19:44:59 +0200 | [diff] [blame] | 8397 | tg = sched_create_group(parent); |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8398 | if (IS_ERR(tg)) |
| 8399 | return ERR_PTR(-ENOMEM); |
| 8400 | |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8401 | return &tg->css; |
| 8402 | } |
| 8403 | |
Konstantin Khlebnikov | 62b5776 | 2017-02-08 14:27:27 +0300 | [diff] [blame] | 8404 | /* Expose task group only after completing cgroup initialization */ |
| 8405 | static int cpu_cgroup_css_online(struct cgroup_subsys_state *css) |
| 8406 | { |
| 8407 | struct task_group *tg = css_tg(css); |
| 8408 | struct task_group *parent = css_tg(css->parent); |
| 8409 | |
| 8410 | if (parent) |
| 8411 | sched_online_group(tg, parent); |
| 8412 | return 0; |
| 8413 | } |
| 8414 | |
Peter Zijlstra | 2f5177f | 2016-03-16 16:22:45 +0100 | [diff] [blame] | 8415 | static void cpu_cgroup_css_released(struct cgroup_subsys_state *css) |
Li Zefan | ace783b | 2013-01-24 14:30:48 +0800 | [diff] [blame] | 8416 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 8417 | struct task_group *tg = css_tg(css); |
Li Zefan | ace783b | 2013-01-24 14:30:48 +0800 | [diff] [blame] | 8418 | |
Peter Zijlstra | 2f5177f | 2016-03-16 16:22:45 +0100 | [diff] [blame] | 8419 | sched_offline_group(tg); |
Li Zefan | ace783b | 2013-01-24 14:30:48 +0800 | [diff] [blame] | 8420 | } |
| 8421 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 8422 | static void cpu_cgroup_css_free(struct cgroup_subsys_state *css) |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8423 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 8424 | struct task_group *tg = css_tg(css); |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8425 | |
Peter Zijlstra | 2f5177f | 2016-03-16 16:22:45 +0100 | [diff] [blame] | 8426 | /* |
| 8427 | * Relies on the RCU grace period between css_released() and this. |
| 8428 | */ |
| 8429 | sched_free_group(tg); |
Li Zefan | ace783b | 2013-01-24 14:30:48 +0800 | [diff] [blame] | 8430 | } |
| 8431 | |
Vincent Guittot | ea86cb4 | 2016-06-17 13:38:55 +0200 | [diff] [blame] | 8432 | /* |
| 8433 | * This is called before wake_up_new_task(), therefore we really only |
| 8434 | * have to set its group bits, all the other stuff does not apply. |
| 8435 | */ |
Oleg Nesterov | b53202e | 2015-12-03 10:24:08 -0500 | [diff] [blame] | 8436 | static void cpu_cgroup_fork(struct task_struct *task) |
Kirill Tkhai | eeb61e5 | 2014-10-27 14:18:25 +0400 | [diff] [blame] | 8437 | { |
Vincent Guittot | ea86cb4 | 2016-06-17 13:38:55 +0200 | [diff] [blame] | 8438 | struct rq_flags rf; |
| 8439 | struct rq *rq; |
| 8440 | |
| 8441 | rq = task_rq_lock(task, &rf); |
| 8442 | |
| 8443 | sched_change_group(task, TASK_SET_GROUP); |
| 8444 | |
| 8445 | task_rq_unlock(rq, task, &rf); |
Kirill Tkhai | eeb61e5 | 2014-10-27 14:18:25 +0400 | [diff] [blame] | 8446 | } |
| 8447 | |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 8448 | static int cpu_cgroup_can_attach(struct cgroup_taskset *tset) |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8449 | { |
Tejun Heo | bb9d97b | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 8450 | struct task_struct *task; |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 8451 | struct cgroup_subsys_state *css; |
Peter Zijlstra | 7dc603c | 2016-06-16 13:29:28 +0200 | [diff] [blame] | 8452 | int ret = 0; |
Tejun Heo | bb9d97b | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 8453 | |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 8454 | cgroup_taskset_for_each(task, css, tset) { |
Peter Zijlstra | b68aa23 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 8455 | #ifdef CONFIG_RT_GROUP_SCHED |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 8456 | if (!sched_rt_can_attach(css_tg(css), task)) |
Tejun Heo | bb9d97b | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 8457 | return -EINVAL; |
Peter Zijlstra | b68aa23 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 8458 | #else |
Tejun Heo | bb9d97b | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 8459 | /* We don't support RT-tasks being in separate groups */ |
| 8460 | if (task->sched_class != &fair_sched_class) |
| 8461 | return -EINVAL; |
Peter Zijlstra | b68aa23 | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 8462 | #endif |
Peter Zijlstra | 7dc603c | 2016-06-16 13:29:28 +0200 | [diff] [blame] | 8463 | /* |
| 8464 | * Serialize against wake_up_new_task() such that if its |
| 8465 | * running, we're sure to observe its full state. |
| 8466 | */ |
| 8467 | raw_spin_lock_irq(&task->pi_lock); |
| 8468 | /* |
| 8469 | * Avoid calling sched_move_task() before wake_up_new_task() |
| 8470 | * has happened. This would lead to problems with PELT, due to |
| 8471 | * move wanting to detach+attach while we're not attached yet. |
| 8472 | */ |
| 8473 | if (task->state == TASK_NEW) |
| 8474 | ret = -EINVAL; |
| 8475 | raw_spin_unlock_irq(&task->pi_lock); |
| 8476 | |
| 8477 | if (ret) |
| 8478 | break; |
Tejun Heo | bb9d97b | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 8479 | } |
Peter Zijlstra | 7dc603c | 2016-06-16 13:29:28 +0200 | [diff] [blame] | 8480 | return ret; |
Ben Blum | be367d0 | 2009-09-23 15:56:31 -0700 | [diff] [blame] | 8481 | } |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8482 | |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 8483 | static void cpu_cgroup_attach(struct cgroup_taskset *tset) |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8484 | { |
Tejun Heo | bb9d97b | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 8485 | struct task_struct *task; |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 8486 | struct cgroup_subsys_state *css; |
Tejun Heo | bb9d97b | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 8487 | |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 8488 | cgroup_taskset_for_each(task, css, tset) |
Tejun Heo | bb9d97b | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 8489 | sched_move_task(task); |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8490 | } |
| 8491 | |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 8492 | #ifdef CONFIG_FAIR_GROUP_SCHED |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8493 | static int cpu_shares_write_u64(struct cgroup_subsys_state *css, |
| 8494 | struct cftype *cftype, u64 shareval) |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8495 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8496 | return sched_group_set_shares(css_tg(css), scale_load(shareval)); |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8497 | } |
| 8498 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8499 | static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css, |
| 8500 | struct cftype *cft) |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8501 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8502 | struct task_group *tg = css_tg(css); |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8503 | |
Nikhil Rao | c8b2811 | 2011-05-18 14:37:48 -0700 | [diff] [blame] | 8504 | return (u64) scale_load_down(tg->shares); |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8505 | } |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8506 | |
| 8507 | #ifdef CONFIG_CFS_BANDWIDTH |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 8508 | static DEFINE_MUTEX(cfs_constraints_mutex); |
| 8509 | |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8510 | const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */ |
| 8511 | const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */ |
| 8512 | |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 8513 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime); |
| 8514 | |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8515 | static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota) |
| 8516 | { |
Paul Turner | 56f570e | 2011-11-07 20:26:33 -0800 | [diff] [blame] | 8517 | int i, ret = 0, runtime_enabled, runtime_was_enabled; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 8518 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8519 | |
| 8520 | if (tg == &root_task_group) |
| 8521 | return -EINVAL; |
| 8522 | |
| 8523 | /* |
| 8524 | * Ensure we have at some amount of bandwidth every period. This is |
| 8525 | * to prevent reaching a state of large arrears when throttled via |
| 8526 | * entity_tick() resulting in prolonged exit starvation. |
| 8527 | */ |
| 8528 | if (quota < min_cfs_quota_period || period < min_cfs_quota_period) |
| 8529 | return -EINVAL; |
| 8530 | |
| 8531 | /* |
| 8532 | * Likewise, bound things on the otherside by preventing insane quota |
| 8533 | * periods. This also allows us to normalize in computing quota |
| 8534 | * feasibility. |
| 8535 | */ |
| 8536 | if (period > max_cfs_quota_period) |
| 8537 | return -EINVAL; |
| 8538 | |
Kirill Tkhai | 0e59bda | 2014-06-25 12:19:42 +0400 | [diff] [blame] | 8539 | /* |
| 8540 | * Prevent race between setting of cfs_rq->runtime_enabled and |
| 8541 | * unthrottle_offline_cfs_rqs(). |
| 8542 | */ |
| 8543 | get_online_cpus(); |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 8544 | mutex_lock(&cfs_constraints_mutex); |
| 8545 | ret = __cfs_schedulable(tg, period, quota); |
| 8546 | if (ret) |
| 8547 | goto out_unlock; |
| 8548 | |
Paul Turner | 58088ad | 2011-07-21 09:43:31 -0700 | [diff] [blame] | 8549 | runtime_enabled = quota != RUNTIME_INF; |
Paul Turner | 56f570e | 2011-11-07 20:26:33 -0800 | [diff] [blame] | 8550 | runtime_was_enabled = cfs_b->quota != RUNTIME_INF; |
Ben Segall | 1ee14e6 | 2013-10-16 11:16:12 -0700 | [diff] [blame] | 8551 | /* |
| 8552 | * If we need to toggle cfs_bandwidth_used, off->on must occur |
| 8553 | * before making related changes, and on->off must occur afterwards |
| 8554 | */ |
| 8555 | if (runtime_enabled && !runtime_was_enabled) |
| 8556 | cfs_bandwidth_usage_inc(); |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8557 | raw_spin_lock_irq(&cfs_b->lock); |
| 8558 | cfs_b->period = ns_to_ktime(period); |
| 8559 | cfs_b->quota = quota; |
Paul Turner | 58088ad | 2011-07-21 09:43:31 -0700 | [diff] [blame] | 8560 | |
Paul Turner | a9cf55b | 2011-07-21 09:43:32 -0700 | [diff] [blame] | 8561 | __refill_cfs_bandwidth_runtime(cfs_b); |
Paul Turner | 58088ad | 2011-07-21 09:43:31 -0700 | [diff] [blame] | 8562 | /* restart the period timer (if active) to handle new period expiry */ |
Peter Zijlstra | 77a4d1a | 2015-04-15 11:41:57 +0200 | [diff] [blame] | 8563 | if (runtime_enabled) |
| 8564 | start_cfs_bandwidth(cfs_b); |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8565 | raw_spin_unlock_irq(&cfs_b->lock); |
| 8566 | |
Kirill Tkhai | 0e59bda | 2014-06-25 12:19:42 +0400 | [diff] [blame] | 8567 | for_each_online_cpu(i) { |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8568 | struct cfs_rq *cfs_rq = tg->cfs_rq[i]; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 8569 | struct rq *rq = cfs_rq->rq; |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8570 | |
| 8571 | raw_spin_lock_irq(&rq->lock); |
Paul Turner | 58088ad | 2011-07-21 09:43:31 -0700 | [diff] [blame] | 8572 | cfs_rq->runtime_enabled = runtime_enabled; |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8573 | cfs_rq->runtime_remaining = 0; |
Paul Turner | 671fd9d | 2011-07-21 09:43:34 -0700 | [diff] [blame] | 8574 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 8575 | if (cfs_rq->throttled) |
Paul Turner | 671fd9d | 2011-07-21 09:43:34 -0700 | [diff] [blame] | 8576 | unthrottle_cfs_rq(cfs_rq); |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8577 | raw_spin_unlock_irq(&rq->lock); |
| 8578 | } |
Ben Segall | 1ee14e6 | 2013-10-16 11:16:12 -0700 | [diff] [blame] | 8579 | if (runtime_was_enabled && !runtime_enabled) |
| 8580 | cfs_bandwidth_usage_dec(); |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 8581 | out_unlock: |
| 8582 | mutex_unlock(&cfs_constraints_mutex); |
Kirill Tkhai | 0e59bda | 2014-06-25 12:19:42 +0400 | [diff] [blame] | 8583 | put_online_cpus(); |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8584 | |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 8585 | return ret; |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8586 | } |
| 8587 | |
| 8588 | int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us) |
| 8589 | { |
| 8590 | u64 quota, period; |
| 8591 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 8592 | period = ktime_to_ns(tg->cfs_bandwidth.period); |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8593 | if (cfs_quota_us < 0) |
| 8594 | quota = RUNTIME_INF; |
| 8595 | else |
| 8596 | quota = (u64)cfs_quota_us * NSEC_PER_USEC; |
| 8597 | |
| 8598 | return tg_set_cfs_bandwidth(tg, period, quota); |
| 8599 | } |
| 8600 | |
| 8601 | long tg_get_cfs_quota(struct task_group *tg) |
| 8602 | { |
| 8603 | u64 quota_us; |
| 8604 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 8605 | if (tg->cfs_bandwidth.quota == RUNTIME_INF) |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8606 | return -1; |
| 8607 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 8608 | quota_us = tg->cfs_bandwidth.quota; |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8609 | do_div(quota_us, NSEC_PER_USEC); |
| 8610 | |
| 8611 | return quota_us; |
| 8612 | } |
| 8613 | |
| 8614 | int tg_set_cfs_period(struct task_group *tg, long cfs_period_us) |
| 8615 | { |
| 8616 | u64 quota, period; |
| 8617 | |
| 8618 | period = (u64)cfs_period_us * NSEC_PER_USEC; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 8619 | quota = tg->cfs_bandwidth.quota; |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8620 | |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8621 | return tg_set_cfs_bandwidth(tg, period, quota); |
| 8622 | } |
| 8623 | |
| 8624 | long tg_get_cfs_period(struct task_group *tg) |
| 8625 | { |
| 8626 | u64 cfs_period_us; |
| 8627 | |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 8628 | cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period); |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8629 | do_div(cfs_period_us, NSEC_PER_USEC); |
| 8630 | |
| 8631 | return cfs_period_us; |
| 8632 | } |
| 8633 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8634 | static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css, |
| 8635 | struct cftype *cft) |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8636 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8637 | return tg_get_cfs_quota(css_tg(css)); |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8638 | } |
| 8639 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8640 | static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css, |
| 8641 | struct cftype *cftype, s64 cfs_quota_us) |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8642 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8643 | return tg_set_cfs_quota(css_tg(css), cfs_quota_us); |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8644 | } |
| 8645 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8646 | static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css, |
| 8647 | struct cftype *cft) |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8648 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8649 | return tg_get_cfs_period(css_tg(css)); |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8650 | } |
| 8651 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8652 | static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css, |
| 8653 | struct cftype *cftype, u64 cfs_period_us) |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8654 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8655 | return tg_set_cfs_period(css_tg(css), cfs_period_us); |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8656 | } |
| 8657 | |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 8658 | struct cfs_schedulable_data { |
| 8659 | struct task_group *tg; |
| 8660 | u64 period, quota; |
| 8661 | }; |
| 8662 | |
| 8663 | /* |
| 8664 | * normalize group quota/period to be quota/max_period |
| 8665 | * note: units are usecs |
| 8666 | */ |
| 8667 | static u64 normalize_cfs_quota(struct task_group *tg, |
| 8668 | struct cfs_schedulable_data *d) |
| 8669 | { |
| 8670 | u64 quota, period; |
| 8671 | |
| 8672 | if (tg == d->tg) { |
| 8673 | period = d->period; |
| 8674 | quota = d->quota; |
| 8675 | } else { |
| 8676 | period = tg_get_cfs_period(tg); |
| 8677 | quota = tg_get_cfs_quota(tg); |
| 8678 | } |
| 8679 | |
| 8680 | /* note: these should typically be equivalent */ |
| 8681 | if (quota == RUNTIME_INF || quota == -1) |
| 8682 | return RUNTIME_INF; |
| 8683 | |
| 8684 | return to_ratio(period, quota); |
| 8685 | } |
| 8686 | |
| 8687 | static int tg_cfs_schedulable_down(struct task_group *tg, void *data) |
| 8688 | { |
| 8689 | struct cfs_schedulable_data *d = data; |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 8690 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 8691 | s64 quota = 0, parent_quota = -1; |
| 8692 | |
| 8693 | if (!tg->parent) { |
| 8694 | quota = RUNTIME_INF; |
| 8695 | } else { |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 8696 | struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth; |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 8697 | |
| 8698 | quota = normalize_cfs_quota(tg, d); |
Zhihui Zhang | 9c58c79 | 2014-09-20 21:24:36 -0400 | [diff] [blame] | 8699 | parent_quota = parent_b->hierarchical_quota; |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 8700 | |
| 8701 | /* |
| 8702 | * ensure max(child_quota) <= parent_quota, inherit when no |
| 8703 | * limit is set |
| 8704 | */ |
| 8705 | if (quota == RUNTIME_INF) |
| 8706 | quota = parent_quota; |
| 8707 | else if (parent_quota != RUNTIME_INF && quota > parent_quota) |
| 8708 | return -EINVAL; |
| 8709 | } |
Zhihui Zhang | 9c58c79 | 2014-09-20 21:24:36 -0400 | [diff] [blame] | 8710 | cfs_b->hierarchical_quota = quota; |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 8711 | |
| 8712 | return 0; |
| 8713 | } |
| 8714 | |
| 8715 | static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota) |
| 8716 | { |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 8717 | int ret; |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 8718 | struct cfs_schedulable_data data = { |
| 8719 | .tg = tg, |
| 8720 | .period = period, |
| 8721 | .quota = quota, |
| 8722 | }; |
| 8723 | |
| 8724 | if (quota != RUNTIME_INF) { |
| 8725 | do_div(data.period, NSEC_PER_USEC); |
| 8726 | do_div(data.quota, NSEC_PER_USEC); |
| 8727 | } |
| 8728 | |
Paul Turner | 8277434 | 2011-07-21 09:43:35 -0700 | [diff] [blame] | 8729 | rcu_read_lock(); |
| 8730 | ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data); |
| 8731 | rcu_read_unlock(); |
| 8732 | |
| 8733 | return ret; |
Paul Turner | a790de9 | 2011-07-21 09:43:29 -0700 | [diff] [blame] | 8734 | } |
Nikhil Rao | e8da1b1 | 2011-07-21 09:43:40 -0700 | [diff] [blame] | 8735 | |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 8736 | static int cpu_stats_show(struct seq_file *sf, void *v) |
Nikhil Rao | e8da1b1 | 2011-07-21 09:43:40 -0700 | [diff] [blame] | 8737 | { |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 8738 | struct task_group *tg = css_tg(seq_css(sf)); |
Peter Zijlstra | 029632f | 2011-10-25 10:00:11 +0200 | [diff] [blame] | 8739 | struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth; |
Nikhil Rao | e8da1b1 | 2011-07-21 09:43:40 -0700 | [diff] [blame] | 8740 | |
Tejun Heo | 44ffc75 | 2013-12-05 12:28:01 -0500 | [diff] [blame] | 8741 | seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods); |
| 8742 | seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled); |
| 8743 | seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time); |
Nikhil Rao | e8da1b1 | 2011-07-21 09:43:40 -0700 | [diff] [blame] | 8744 | |
| 8745 | return 0; |
| 8746 | } |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8747 | #endif /* CONFIG_CFS_BANDWIDTH */ |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 8748 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8749 | |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 8750 | #ifdef CONFIG_RT_GROUP_SCHED |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8751 | static int cpu_rt_runtime_write(struct cgroup_subsys_state *css, |
| 8752 | struct cftype *cft, s64 val) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 8753 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8754 | return sched_group_set_rt_runtime(css_tg(css), val); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 8755 | } |
| 8756 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8757 | static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css, |
| 8758 | struct cftype *cft) |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 8759 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8760 | return sched_group_rt_runtime(css_tg(css)); |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 8761 | } |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8762 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8763 | static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css, |
| 8764 | struct cftype *cftype, u64 rt_period_us) |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8765 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8766 | return sched_group_set_rt_period(css_tg(css), rt_period_us); |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8767 | } |
| 8768 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8769 | static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css, |
| 8770 | struct cftype *cft) |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8771 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 8772 | return sched_group_rt_period(css_tg(css)); |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8773 | } |
Dhaval Giani | 6d6bc0a | 2008-05-30 14:23:45 +0200 | [diff] [blame] | 8774 | #endif /* CONFIG_RT_GROUP_SCHED */ |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 8775 | |
Paul Menage | fe5c7cc | 2007-10-29 21:18:11 +0100 | [diff] [blame] | 8776 | static struct cftype cpu_files[] = { |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 8777 | #ifdef CONFIG_FAIR_GROUP_SCHED |
Paul Menage | fe5c7cc | 2007-10-29 21:18:11 +0100 | [diff] [blame] | 8778 | { |
| 8779 | .name = "shares", |
Paul Menage | f4c753b | 2008-04-29 00:59:56 -0700 | [diff] [blame] | 8780 | .read_u64 = cpu_shares_read_u64, |
| 8781 | .write_u64 = cpu_shares_write_u64, |
Paul Menage | fe5c7cc | 2007-10-29 21:18:11 +0100 | [diff] [blame] | 8782 | }, |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 8783 | #endif |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8784 | #ifdef CONFIG_CFS_BANDWIDTH |
| 8785 | { |
| 8786 | .name = "cfs_quota_us", |
| 8787 | .read_s64 = cpu_cfs_quota_read_s64, |
| 8788 | .write_s64 = cpu_cfs_quota_write_s64, |
| 8789 | }, |
| 8790 | { |
| 8791 | .name = "cfs_period_us", |
| 8792 | .read_u64 = cpu_cfs_period_read_u64, |
| 8793 | .write_u64 = cpu_cfs_period_write_u64, |
| 8794 | }, |
Nikhil Rao | e8da1b1 | 2011-07-21 09:43:40 -0700 | [diff] [blame] | 8795 | { |
| 8796 | .name = "stat", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 8797 | .seq_show = cpu_stats_show, |
Nikhil Rao | e8da1b1 | 2011-07-21 09:43:40 -0700 | [diff] [blame] | 8798 | }, |
Paul Turner | ab84d31 | 2011-07-21 09:43:28 -0700 | [diff] [blame] | 8799 | #endif |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 8800 | #ifdef CONFIG_RT_GROUP_SCHED |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 8801 | { |
Peter Zijlstra | 9f0c1e5 | 2008-02-13 15:45:39 +0100 | [diff] [blame] | 8802 | .name = "rt_runtime_us", |
Paul Menage | 06ecb27 | 2008-04-29 01:00:06 -0700 | [diff] [blame] | 8803 | .read_s64 = cpu_rt_runtime_read, |
| 8804 | .write_s64 = cpu_rt_runtime_write, |
Peter Zijlstra | 6f505b1 | 2008-01-25 21:08:30 +0100 | [diff] [blame] | 8805 | }, |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8806 | { |
| 8807 | .name = "rt_period_us", |
Paul Menage | f4c753b | 2008-04-29 00:59:56 -0700 | [diff] [blame] | 8808 | .read_u64 = cpu_rt_period_read_uint, |
| 8809 | .write_u64 = cpu_rt_period_write_uint, |
Peter Zijlstra | d0b27fa | 2008-04-19 19:44:57 +0200 | [diff] [blame] | 8810 | }, |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 8811 | #endif |
Tejun Heo | 4baf6e3 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 8812 | { } /* terminate */ |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8813 | }; |
| 8814 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 8815 | struct cgroup_subsys cpu_cgrp_subsys = { |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 8816 | .css_alloc = cpu_cgroup_css_alloc, |
Konstantin Khlebnikov | 62b5776 | 2017-02-08 14:27:27 +0300 | [diff] [blame] | 8817 | .css_online = cpu_cgroup_css_online, |
Peter Zijlstra | 2f5177f | 2016-03-16 16:22:45 +0100 | [diff] [blame] | 8818 | .css_released = cpu_cgroup_css_released, |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 8819 | .css_free = cpu_cgroup_css_free, |
Kirill Tkhai | eeb61e5 | 2014-10-27 14:18:25 +0400 | [diff] [blame] | 8820 | .fork = cpu_cgroup_fork, |
Tejun Heo | bb9d97b | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 8821 | .can_attach = cpu_cgroup_can_attach, |
| 8822 | .attach = cpu_cgroup_attach, |
Tejun Heo | 5577964 | 2014-07-15 11:05:09 -0400 | [diff] [blame] | 8823 | .legacy_cftypes = cpu_files, |
Tejun Heo | b38e42e | 2016-02-23 10:00:50 -0500 | [diff] [blame] | 8824 | .early_init = true, |
Srivatsa Vaddagiri | 68318b8 | 2007-10-18 23:41:03 -0700 | [diff] [blame] | 8825 | }; |
| 8826 | |
Peter Zijlstra | 052f1dc | 2008-02-13 15:45:40 +0100 | [diff] [blame] | 8827 | #endif /* CONFIG_CGROUP_SCHED */ |
Srivatsa Vaddagiri | d842de8 | 2007-12-02 20:04:49 +0100 | [diff] [blame] | 8828 | |
Paul E. McKenney | b637a32 | 2012-09-19 16:58:38 -0700 | [diff] [blame] | 8829 | void dump_cpu_task(int cpu) |
| 8830 | { |
| 8831 | pr_info("Task dump for CPU %d:\n", cpu); |
| 8832 | sched_show_task(cpu_curr(cpu)); |
| 8833 | } |
Andi Kleen | ed82b8a | 2015-11-29 20:59:43 -0800 | [diff] [blame] | 8834 | |
| 8835 | /* |
| 8836 | * Nice levels are multiplicative, with a gentle 10% change for every |
| 8837 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to |
| 8838 | * nice 1, it will get ~10% less CPU time than another CPU-bound task |
| 8839 | * that remained on nice 0. |
| 8840 | * |
| 8841 | * The "10% effect" is relative and cumulative: from _any_ nice level, |
| 8842 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level |
| 8843 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
| 8844 | * If a task goes up by ~10% and another task goes down by ~10% then |
| 8845 | * the relative distance between them is ~25%.) |
| 8846 | */ |
| 8847 | const int sched_prio_to_weight[40] = { |
| 8848 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
| 8849 | /* -15 */ 29154, 23254, 18705, 14949, 11916, |
| 8850 | /* -10 */ 9548, 7620, 6100, 4904, 3906, |
| 8851 | /* -5 */ 3121, 2501, 1991, 1586, 1277, |
| 8852 | /* 0 */ 1024, 820, 655, 526, 423, |
| 8853 | /* 5 */ 335, 272, 215, 172, 137, |
| 8854 | /* 10 */ 110, 87, 70, 56, 45, |
| 8855 | /* 15 */ 36, 29, 23, 18, 15, |
| 8856 | }; |
| 8857 | |
| 8858 | /* |
| 8859 | * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated. |
| 8860 | * |
| 8861 | * In cases where the weight does not change often, we can use the |
| 8862 | * precalculated inverse to speed up arithmetics by turning divisions |
| 8863 | * into multiplications: |
| 8864 | */ |
| 8865 | const u32 sched_prio_to_wmult[40] = { |
| 8866 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
| 8867 | /* -15 */ 147320, 184698, 229616, 287308, 360437, |
| 8868 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, |
| 8869 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, |
| 8870 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, |
| 8871 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, |
| 8872 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, |
| 8873 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, |
| 8874 | }; |