Frederic Weisbecker | 4eacdf1 | 2013-01-16 17:16:37 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Context tracking: Probe on high level context boundaries such as kernel |
| 3 | * and userspace. This includes syscalls and exceptions entry/exit. |
| 4 | * |
| 5 | * This is used by RCU to remove its dependency on the timer tick while a CPU |
| 6 | * runs in userspace. |
| 7 | * |
| 8 | * Started by Frederic Weisbecker: |
| 9 | * |
| 10 | * Copyright (C) 2012 Red Hat, Inc., Frederic Weisbecker <fweisbec@redhat.com> |
| 11 | * |
| 12 | * Many thanks to Gilad Ben-Yossef, Paul McKenney, Ingo Molnar, Andrew Morton, |
| 13 | * Steven Rostedt, Peter Zijlstra for suggestions and improvements. |
| 14 | * |
| 15 | */ |
| 16 | |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 17 | #include <linux/context_tracking.h> |
| 18 | #include <linux/rcupdate.h> |
| 19 | #include <linux/sched.h> |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 20 | #include <linux/hardirq.h> |
Frederic Weisbecker | 6a61671 | 2012-12-16 20:00:34 +0100 | [diff] [blame] | 21 | #include <linux/export.h> |
Masami Hiramatsu | 4cdf77a | 2014-06-14 06:47:12 +0000 | [diff] [blame] | 22 | #include <linux/kprobes.h> |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 23 | |
Frederic Weisbecker | 1b6a259 | 2013-07-11 20:27:43 +0200 | [diff] [blame] | 24 | #define CREATE_TRACE_POINTS |
| 25 | #include <trace/events/context_tracking.h> |
| 26 | |
Frederic Weisbecker | 65f382f | 2013-07-11 19:12:32 +0200 | [diff] [blame] | 27 | struct static_key context_tracking_enabled = STATIC_KEY_INIT_FALSE; |
Frederic Weisbecker | 48d6a81 | 2013-07-10 02:44:35 +0200 | [diff] [blame] | 28 | EXPORT_SYMBOL_GPL(context_tracking_enabled); |
Frederic Weisbecker | 65f382f | 2013-07-11 19:12:32 +0200 | [diff] [blame] | 29 | |
| 30 | DEFINE_PER_CPU(struct context_tracking, context_tracking); |
Frederic Weisbecker | 48d6a81 | 2013-07-10 02:44:35 +0200 | [diff] [blame] | 31 | EXPORT_SYMBOL_GPL(context_tracking); |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 32 | |
Frederic Weisbecker | 2e70933 | 2013-07-10 00:55:25 +0200 | [diff] [blame] | 33 | void context_tracking_cpu_set(int cpu) |
| 34 | { |
Frederic Weisbecker | 65f382f | 2013-07-11 19:12:32 +0200 | [diff] [blame] | 35 | if (!per_cpu(context_tracking.active, cpu)) { |
| 36 | per_cpu(context_tracking.active, cpu) = true; |
| 37 | static_key_slow_inc(&context_tracking_enabled); |
| 38 | } |
Frederic Weisbecker | 2e70933 | 2013-07-10 00:55:25 +0200 | [diff] [blame] | 39 | } |
| 40 | |
Frederic Weisbecker | aed5ed4 | 2015-05-06 18:04:23 +0200 | [diff] [blame^] | 41 | static bool context_tracking_recursion_enter(void) |
| 42 | { |
| 43 | int recursion; |
| 44 | |
| 45 | recursion = __this_cpu_inc_return(context_tracking.recursion); |
| 46 | if (recursion == 1) |
| 47 | return true; |
| 48 | |
| 49 | WARN_ONCE((recursion < 1), "Invalid context tracking recursion value %d\n", recursion); |
| 50 | __this_cpu_dec(context_tracking.recursion); |
| 51 | |
| 52 | return false; |
| 53 | } |
| 54 | |
| 55 | static void context_tracking_recursion_exit(void) |
| 56 | { |
| 57 | __this_cpu_dec(context_tracking.recursion); |
| 58 | } |
| 59 | |
Frederic Weisbecker | 4eacdf1 | 2013-01-16 17:16:37 +0100 | [diff] [blame] | 60 | /** |
Rik van Riel | 3aab4f5 | 2015-02-10 15:27:50 -0500 | [diff] [blame] | 61 | * context_tracking_enter - Inform the context tracking that the CPU is going |
| 62 | * enter user or guest space mode. |
Frederic Weisbecker | 4eacdf1 | 2013-01-16 17:16:37 +0100 | [diff] [blame] | 63 | * |
| 64 | * This function must be called right before we switch from the kernel |
Rik van Riel | 3aab4f5 | 2015-02-10 15:27:50 -0500 | [diff] [blame] | 65 | * to user or guest space, when it's guaranteed the remaining kernel |
| 66 | * instructions to execute won't use any RCU read side critical section |
| 67 | * because this function sets RCU in extended quiescent state. |
Frederic Weisbecker | 4eacdf1 | 2013-01-16 17:16:37 +0100 | [diff] [blame] | 68 | */ |
Rik van Riel | 3aab4f5 | 2015-02-10 15:27:50 -0500 | [diff] [blame] | 69 | void context_tracking_enter(enum ctx_state state) |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 70 | { |
| 71 | unsigned long flags; |
| 72 | |
| 73 | /* |
Frederic Weisbecker | 0c06a5d | 2013-09-10 00:54:17 +0200 | [diff] [blame] | 74 | * Repeat the user_enter() check here because some archs may be calling |
| 75 | * this from asm and if no CPU needs context tracking, they shouldn't |
Frederic Weisbecker | 58135f5 | 2013-11-06 14:45:57 +0100 | [diff] [blame] | 76 | * go further. Repeat the check here until they support the inline static |
| 77 | * key check. |
Frederic Weisbecker | 0c06a5d | 2013-09-10 00:54:17 +0200 | [diff] [blame] | 78 | */ |
Frederic Weisbecker | 58135f5 | 2013-11-06 14:45:57 +0100 | [diff] [blame] | 79 | if (!context_tracking_is_enabled()) |
Frederic Weisbecker | 0c06a5d | 2013-09-10 00:54:17 +0200 | [diff] [blame] | 80 | return; |
| 81 | |
| 82 | /* |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 83 | * Some contexts may involve an exception occuring in an irq, |
| 84 | * leading to that nesting: |
| 85 | * rcu_irq_enter() rcu_user_exit() rcu_user_exit() rcu_irq_exit() |
| 86 | * This would mess up the dyntick_nesting count though. And rcu_irq_*() |
| 87 | * helpers are enough to protect RCU uses inside the exception. So |
| 88 | * just return immediately if we detect we are in an IRQ. |
| 89 | */ |
| 90 | if (in_interrupt()) |
| 91 | return; |
| 92 | |
Frederic Weisbecker | 4eacdf1 | 2013-01-16 17:16:37 +0100 | [diff] [blame] | 93 | /* Kernel threads aren't supposed to go to userspace */ |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 94 | WARN_ON_ONCE(!current->mm); |
| 95 | |
| 96 | local_irq_save(flags); |
Frederic Weisbecker | aed5ed4 | 2015-05-06 18:04:23 +0200 | [diff] [blame^] | 97 | if (!context_tracking_recursion_enter()) |
| 98 | goto out_irq_restore; |
| 99 | |
Rik van Riel | 3aab4f5 | 2015-02-10 15:27:50 -0500 | [diff] [blame] | 100 | if ( __this_cpu_read(context_tracking.state) != state) { |
Frederic Weisbecker | d65ec12 | 2013-07-11 23:59:33 +0200 | [diff] [blame] | 101 | if (__this_cpu_read(context_tracking.active)) { |
| 102 | /* |
| 103 | * At this stage, only low level arch entry code remains and |
| 104 | * then we'll run in userspace. We can assume there won't be |
| 105 | * any RCU read-side critical section until the next call to |
| 106 | * user_exit() or rcu_irq_enter(). Let's remove RCU's dependency |
| 107 | * on the tick. |
| 108 | */ |
Rik van Riel | 19fdd98 | 2015-02-10 15:27:52 -0500 | [diff] [blame] | 109 | if (state == CONTEXT_USER) { |
| 110 | trace_user_enter(0); |
| 111 | vtime_user_enter(current); |
| 112 | } |
Frederic Weisbecker | d65ec12 | 2013-07-11 23:59:33 +0200 | [diff] [blame] | 113 | rcu_user_enter(); |
| 114 | } |
Frederic Weisbecker | 4eacdf1 | 2013-01-16 17:16:37 +0100 | [diff] [blame] | 115 | /* |
Frederic Weisbecker | d65ec12 | 2013-07-11 23:59:33 +0200 | [diff] [blame] | 116 | * Even if context tracking is disabled on this CPU, because it's outside |
| 117 | * the full dynticks mask for example, we still have to keep track of the |
| 118 | * context transitions and states to prevent inconsistency on those of |
| 119 | * other CPUs. |
| 120 | * If a task triggers an exception in userspace, sleep on the exception |
| 121 | * handler and then migrate to another CPU, that new CPU must know where |
| 122 | * the exception returns by the time we call exception_exit(). |
| 123 | * This information can only be provided by the previous CPU when it called |
| 124 | * exception_enter(). |
| 125 | * OTOH we can spare the calls to vtime and RCU when context_tracking.active |
| 126 | * is false because we know that CPU is not tickless. |
Frederic Weisbecker | 4eacdf1 | 2013-01-16 17:16:37 +0100 | [diff] [blame] | 127 | */ |
Rik van Riel | 3aab4f5 | 2015-02-10 15:27:50 -0500 | [diff] [blame] | 128 | __this_cpu_write(context_tracking.state, state); |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 129 | } |
Frederic Weisbecker | aed5ed4 | 2015-05-06 18:04:23 +0200 | [diff] [blame^] | 130 | context_tracking_recursion_exit(); |
| 131 | out_irq_restore: |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 132 | local_irq_restore(flags); |
| 133 | } |
Rik van Riel | 3aab4f5 | 2015-02-10 15:27:50 -0500 | [diff] [blame] | 134 | NOKPROBE_SYMBOL(context_tracking_enter); |
Rik van Riel | efc1e2c | 2015-02-10 15:27:53 -0500 | [diff] [blame] | 135 | EXPORT_SYMBOL_GPL(context_tracking_enter); |
Rik van Riel | 3aab4f5 | 2015-02-10 15:27:50 -0500 | [diff] [blame] | 136 | |
| 137 | void context_tracking_user_enter(void) |
| 138 | { |
| 139 | context_tracking_enter(CONTEXT_USER); |
| 140 | } |
Masami Hiramatsu | 4cdf77a | 2014-06-14 06:47:12 +0000 | [diff] [blame] | 141 | NOKPROBE_SYMBOL(context_tracking_user_enter); |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 142 | |
Frederic Weisbecker | 4eacdf1 | 2013-01-16 17:16:37 +0100 | [diff] [blame] | 143 | /** |
Rik van Riel | 3aab4f5 | 2015-02-10 15:27:50 -0500 | [diff] [blame] | 144 | * context_tracking_exit - Inform the context tracking that the CPU is |
| 145 | * exiting user or guest mode and entering the kernel. |
Frederic Weisbecker | 4eacdf1 | 2013-01-16 17:16:37 +0100 | [diff] [blame] | 146 | * |
Rik van Riel | 3aab4f5 | 2015-02-10 15:27:50 -0500 | [diff] [blame] | 147 | * This function must be called after we entered the kernel from user or |
| 148 | * guest space before any use of RCU read side critical section. This |
| 149 | * potentially include any high level kernel code like syscalls, exceptions, |
| 150 | * signal handling, etc... |
Frederic Weisbecker | 4eacdf1 | 2013-01-16 17:16:37 +0100 | [diff] [blame] | 151 | * |
| 152 | * This call supports re-entrancy. This way it can be called from any exception |
| 153 | * handler without needing to know if we came from userspace or not. |
| 154 | */ |
Rik van Riel | 3aab4f5 | 2015-02-10 15:27:50 -0500 | [diff] [blame] | 155 | void context_tracking_exit(enum ctx_state state) |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 156 | { |
| 157 | unsigned long flags; |
| 158 | |
Frederic Weisbecker | 58135f5 | 2013-11-06 14:45:57 +0100 | [diff] [blame] | 159 | if (!context_tracking_is_enabled()) |
Frederic Weisbecker | 0c06a5d | 2013-09-10 00:54:17 +0200 | [diff] [blame] | 160 | return; |
| 161 | |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 162 | if (in_interrupt()) |
| 163 | return; |
| 164 | |
| 165 | local_irq_save(flags); |
Frederic Weisbecker | aed5ed4 | 2015-05-06 18:04:23 +0200 | [diff] [blame^] | 166 | if (!context_tracking_recursion_enter()) |
| 167 | goto out_irq_restore; |
| 168 | |
Rik van Riel | 3aab4f5 | 2015-02-10 15:27:50 -0500 | [diff] [blame] | 169 | if (__this_cpu_read(context_tracking.state) == state) { |
Frederic Weisbecker | d65ec12 | 2013-07-11 23:59:33 +0200 | [diff] [blame] | 170 | if (__this_cpu_read(context_tracking.active)) { |
| 171 | /* |
| 172 | * We are going to run code that may use RCU. Inform |
| 173 | * RCU core about that (ie: we may need the tick again). |
| 174 | */ |
| 175 | rcu_user_exit(); |
Rik van Riel | 19fdd98 | 2015-02-10 15:27:52 -0500 | [diff] [blame] | 176 | if (state == CONTEXT_USER) { |
| 177 | vtime_user_exit(current); |
| 178 | trace_user_exit(0); |
| 179 | } |
Frederic Weisbecker | d65ec12 | 2013-07-11 23:59:33 +0200 | [diff] [blame] | 180 | } |
Frederic Weisbecker | c467ea7 | 2015-03-04 18:06:33 +0100 | [diff] [blame] | 181 | __this_cpu_write(context_tracking.state, CONTEXT_KERNEL); |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 182 | } |
Frederic Weisbecker | aed5ed4 | 2015-05-06 18:04:23 +0200 | [diff] [blame^] | 183 | context_tracking_recursion_exit(); |
| 184 | out_irq_restore: |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 185 | local_irq_restore(flags); |
| 186 | } |
Rik van Riel | 3aab4f5 | 2015-02-10 15:27:50 -0500 | [diff] [blame] | 187 | NOKPROBE_SYMBOL(context_tracking_exit); |
Rik van Riel | efc1e2c | 2015-02-10 15:27:53 -0500 | [diff] [blame] | 188 | EXPORT_SYMBOL_GPL(context_tracking_exit); |
Rik van Riel | 3aab4f5 | 2015-02-10 15:27:50 -0500 | [diff] [blame] | 189 | |
| 190 | void context_tracking_user_exit(void) |
| 191 | { |
| 192 | context_tracking_exit(CONTEXT_USER); |
| 193 | } |
Masami Hiramatsu | 4cdf77a | 2014-06-14 06:47:12 +0000 | [diff] [blame] | 194 | NOKPROBE_SYMBOL(context_tracking_user_exit); |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 195 | |
Frederic Weisbecker | 4eacdf1 | 2013-01-16 17:16:37 +0100 | [diff] [blame] | 196 | /** |
Frederic Weisbecker | 73d424f | 2013-07-11 19:42:13 +0200 | [diff] [blame] | 197 | * __context_tracking_task_switch - context switch the syscall callbacks |
Frederic Weisbecker | 4eacdf1 | 2013-01-16 17:16:37 +0100 | [diff] [blame] | 198 | * @prev: the task that is being switched out |
| 199 | * @next: the task that is being switched in |
| 200 | * |
| 201 | * The context tracking uses the syscall slow path to implement its user-kernel |
| 202 | * boundaries probes on syscalls. This way it doesn't impact the syscall fast |
| 203 | * path on CPUs that don't do context tracking. |
| 204 | * |
| 205 | * But we need to clear the flag on the previous task because it may later |
| 206 | * migrate to some CPU that doesn't do the context tracking. As such the TIF |
| 207 | * flag may not be desired there. |
| 208 | */ |
Frederic Weisbecker | 73d424f | 2013-07-11 19:42:13 +0200 | [diff] [blame] | 209 | void __context_tracking_task_switch(struct task_struct *prev, |
| 210 | struct task_struct *next) |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 211 | { |
Frederic Weisbecker | d65ec12 | 2013-07-11 23:59:33 +0200 | [diff] [blame] | 212 | clear_tsk_thread_flag(prev, TIF_NOHZ); |
| 213 | set_tsk_thread_flag(next, TIF_NOHZ); |
Frederic Weisbecker | 91d1aa43 | 2012-11-27 19:33:25 +0100 | [diff] [blame] | 214 | } |
Frederic Weisbecker | 65f382f | 2013-07-11 19:12:32 +0200 | [diff] [blame] | 215 | |
| 216 | #ifdef CONFIG_CONTEXT_TRACKING_FORCE |
| 217 | void __init context_tracking_init(void) |
| 218 | { |
| 219 | int cpu; |
| 220 | |
| 221 | for_each_possible_cpu(cpu) |
| 222 | context_tracking_cpu_set(cpu); |
| 223 | } |
| 224 | #endif |