Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 1 | /* |
| 2 | * kernel/stop_machine.c |
| 3 | * |
| 4 | * Copyright (C) 2008, 2005 IBM Corporation. |
| 5 | * Copyright (C) 2008, 2005 Rusty Russell rusty@rustcorp.com.au |
| 6 | * Copyright (C) 2010 SUSE Linux Products GmbH |
| 7 | * Copyright (C) 2010 Tejun Heo <tj@kernel.org> |
| 8 | * |
| 9 | * This file is released under the GPLv2 and any later version. |
Rusty Russell | e5582ca | 2006-09-29 02:01:35 -0700 | [diff] [blame] | 10 | */ |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 11 | #include <linux/completion.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 12 | #include <linux/cpu.h> |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 13 | #include <linux/init.h> |
Prarit Bhargava | ee527cd | 2007-05-08 00:25:08 -0700 | [diff] [blame] | 14 | #include <linux/kthread.h> |
Paul Gortmaker | 9984de1 | 2011-05-23 14:51:41 -0400 | [diff] [blame] | 15 | #include <linux/export.h> |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 16 | #include <linux/percpu.h> |
Prarit Bhargava | ee527cd | 2007-05-08 00:25:08 -0700 | [diff] [blame] | 17 | #include <linux/sched.h> |
| 18 | #include <linux/stop_machine.h> |
Benjamin Herrenschmidt | a12bb44 | 2007-05-10 22:22:47 -0700 | [diff] [blame] | 19 | #include <linux/interrupt.h> |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 20 | #include <linux/kallsyms.h> |
Thomas Gleixner | 14e568e | 2013-01-31 12:11:14 +0000 | [diff] [blame] | 21 | #include <linux/smpboot.h> |
Arun Sharma | 60063497 | 2011-07-26 16:09:06 -0700 | [diff] [blame] | 22 | #include <linux/atomic.h> |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 23 | |
| 24 | /* |
| 25 | * Structure to determine completion condition and record errors. May |
| 26 | * be shared by works on different cpus. |
| 27 | */ |
| 28 | struct cpu_stop_done { |
| 29 | atomic_t nr_todo; /* nr left to execute */ |
| 30 | bool executed; /* actually executed? */ |
| 31 | int ret; /* collected return value */ |
| 32 | struct completion completion; /* fired if nr_todo reaches 0 */ |
| 33 | }; |
| 34 | |
| 35 | /* the actual stopper, one per every possible cpu, enabled on online cpus */ |
| 36 | struct cpu_stopper { |
| 37 | spinlock_t lock; |
Richard Kennedy | 878ae12 | 2010-08-09 17:20:34 -0700 | [diff] [blame] | 38 | bool enabled; /* is this stopper enabled? */ |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 39 | struct list_head works; /* list of pending works */ |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 40 | }; |
| 41 | |
| 42 | static DEFINE_PER_CPU(struct cpu_stopper, cpu_stopper); |
Thomas Gleixner | 860a0ff | 2013-01-31 12:11:13 +0000 | [diff] [blame] | 43 | static DEFINE_PER_CPU(struct task_struct *, cpu_stopper_task); |
Jeremy Fitzhardinge | f445027 | 2011-10-31 17:11:15 -0700 | [diff] [blame] | 44 | static bool stop_machine_initialized = false; |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 45 | |
| 46 | static void cpu_stop_init_done(struct cpu_stop_done *done, unsigned int nr_todo) |
| 47 | { |
| 48 | memset(done, 0, sizeof(*done)); |
| 49 | atomic_set(&done->nr_todo, nr_todo); |
| 50 | init_completion(&done->completion); |
| 51 | } |
| 52 | |
| 53 | /* signal completion unless @done is NULL */ |
| 54 | static void cpu_stop_signal_done(struct cpu_stop_done *done, bool executed) |
| 55 | { |
| 56 | if (done) { |
| 57 | if (executed) |
| 58 | done->executed = true; |
| 59 | if (atomic_dec_and_test(&done->nr_todo)) |
| 60 | complete(&done->completion); |
| 61 | } |
| 62 | } |
| 63 | |
| 64 | /* queue @work to @stopper. if offline, @work is completed immediately */ |
Thomas Gleixner | 860a0ff | 2013-01-31 12:11:13 +0000 | [diff] [blame] | 65 | static void cpu_stop_queue_work(unsigned int cpu, struct cpu_stop_work *work) |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 66 | { |
Thomas Gleixner | 860a0ff | 2013-01-31 12:11:13 +0000 | [diff] [blame] | 67 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); |
| 68 | struct task_struct *p = per_cpu(cpu_stopper_task, cpu); |
| 69 | |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 70 | unsigned long flags; |
| 71 | |
| 72 | spin_lock_irqsave(&stopper->lock, flags); |
| 73 | |
| 74 | if (stopper->enabled) { |
| 75 | list_add_tail(&work->list, &stopper->works); |
Thomas Gleixner | 860a0ff | 2013-01-31 12:11:13 +0000 | [diff] [blame] | 76 | wake_up_process(p); |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 77 | } else |
| 78 | cpu_stop_signal_done(work->done, false); |
| 79 | |
| 80 | spin_unlock_irqrestore(&stopper->lock, flags); |
| 81 | } |
| 82 | |
| 83 | /** |
| 84 | * stop_one_cpu - stop a cpu |
| 85 | * @cpu: cpu to stop |
| 86 | * @fn: function to execute |
| 87 | * @arg: argument to @fn |
| 88 | * |
| 89 | * Execute @fn(@arg) on @cpu. @fn is run in a process context with |
| 90 | * the highest priority preempting any task on the cpu and |
| 91 | * monopolizing it. This function returns after the execution is |
| 92 | * complete. |
| 93 | * |
| 94 | * This function doesn't guarantee @cpu stays online till @fn |
| 95 | * completes. If @cpu goes down in the middle, execution may happen |
| 96 | * partially or fully on different cpus. @fn should either be ready |
| 97 | * for that or the caller should ensure that @cpu stays online until |
| 98 | * this function completes. |
| 99 | * |
| 100 | * CONTEXT: |
| 101 | * Might sleep. |
| 102 | * |
| 103 | * RETURNS: |
| 104 | * -ENOENT if @fn(@arg) was not executed because @cpu was offline; |
| 105 | * otherwise, the return value of @fn. |
| 106 | */ |
| 107 | int stop_one_cpu(unsigned int cpu, cpu_stop_fn_t fn, void *arg) |
| 108 | { |
| 109 | struct cpu_stop_done done; |
| 110 | struct cpu_stop_work work = { .fn = fn, .arg = arg, .done = &done }; |
| 111 | |
| 112 | cpu_stop_init_done(&done, 1); |
Thomas Gleixner | 860a0ff | 2013-01-31 12:11:13 +0000 | [diff] [blame] | 113 | cpu_stop_queue_work(cpu, &work); |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 114 | wait_for_completion(&done.completion); |
| 115 | return done.executed ? done.ret : -ENOENT; |
| 116 | } |
| 117 | |
Peter Zijlstra | 1be0bd77 | 2013-10-07 11:29:15 +0100 | [diff] [blame] | 118 | /* This controls the threads on each CPU. */ |
| 119 | enum multi_stop_state { |
| 120 | /* Dummy starting state for thread. */ |
| 121 | MULTI_STOP_NONE, |
| 122 | /* Awaiting everyone to be scheduled. */ |
| 123 | MULTI_STOP_PREPARE, |
| 124 | /* Disable interrupts. */ |
| 125 | MULTI_STOP_DISABLE_IRQ, |
| 126 | /* Run the function */ |
| 127 | MULTI_STOP_RUN, |
| 128 | /* Exit */ |
| 129 | MULTI_STOP_EXIT, |
| 130 | }; |
| 131 | |
| 132 | struct multi_stop_data { |
| 133 | int (*fn)(void *); |
| 134 | void *data; |
| 135 | /* Like num_online_cpus(), but hotplug cpu uses us, so we need this. */ |
| 136 | unsigned int num_threads; |
| 137 | const struct cpumask *active_cpus; |
| 138 | |
| 139 | enum multi_stop_state state; |
| 140 | atomic_t thread_ack; |
| 141 | }; |
| 142 | |
| 143 | static void set_state(struct multi_stop_data *msdata, |
| 144 | enum multi_stop_state newstate) |
| 145 | { |
| 146 | /* Reset ack counter. */ |
| 147 | atomic_set(&msdata->thread_ack, msdata->num_threads); |
| 148 | smp_wmb(); |
| 149 | msdata->state = newstate; |
| 150 | } |
| 151 | |
| 152 | /* Last one to ack a state moves to the next state. */ |
| 153 | static void ack_state(struct multi_stop_data *msdata) |
| 154 | { |
| 155 | if (atomic_dec_and_test(&msdata->thread_ack)) |
| 156 | set_state(msdata, msdata->state + 1); |
| 157 | } |
| 158 | |
| 159 | /* This is the cpu_stop function which stops the CPU. */ |
| 160 | static int multi_cpu_stop(void *data) |
| 161 | { |
| 162 | struct multi_stop_data *msdata = data; |
| 163 | enum multi_stop_state curstate = MULTI_STOP_NONE; |
| 164 | int cpu = smp_processor_id(), err = 0; |
| 165 | unsigned long flags; |
| 166 | bool is_active; |
| 167 | |
| 168 | /* |
| 169 | * When called from stop_machine_from_inactive_cpu(), irq might |
| 170 | * already be disabled. Save the state and restore it on exit. |
| 171 | */ |
| 172 | local_save_flags(flags); |
| 173 | |
| 174 | if (!msdata->active_cpus) |
| 175 | is_active = cpu == cpumask_first(cpu_online_mask); |
| 176 | else |
| 177 | is_active = cpumask_test_cpu(cpu, msdata->active_cpus); |
| 178 | |
| 179 | /* Simple state machine */ |
| 180 | do { |
| 181 | /* Chill out and ensure we re-read multi_stop_state. */ |
| 182 | cpu_relax(); |
| 183 | if (msdata->state != curstate) { |
| 184 | curstate = msdata->state; |
| 185 | switch (curstate) { |
| 186 | case MULTI_STOP_DISABLE_IRQ: |
| 187 | local_irq_disable(); |
| 188 | hard_irq_disable(); |
| 189 | break; |
| 190 | case MULTI_STOP_RUN: |
| 191 | if (is_active) |
| 192 | err = msdata->fn(msdata->data); |
| 193 | break; |
| 194 | default: |
| 195 | break; |
| 196 | } |
| 197 | ack_state(msdata); |
| 198 | } |
| 199 | } while (curstate != MULTI_STOP_EXIT); |
| 200 | |
| 201 | local_irq_restore(flags); |
| 202 | return err; |
| 203 | } |
| 204 | |
| 205 | struct irq_cpu_stop_queue_work_info { |
| 206 | int cpu1; |
| 207 | int cpu2; |
| 208 | struct cpu_stop_work *work1; |
| 209 | struct cpu_stop_work *work2; |
| 210 | }; |
| 211 | |
| 212 | /* |
| 213 | * This function is always run with irqs and preemption disabled. |
| 214 | * This guarantees that both work1 and work2 get queued, before |
| 215 | * our local migrate thread gets the chance to preempt us. |
| 216 | */ |
| 217 | static void irq_cpu_stop_queue_work(void *arg) |
| 218 | { |
| 219 | struct irq_cpu_stop_queue_work_info *info = arg; |
| 220 | cpu_stop_queue_work(info->cpu1, info->work1); |
| 221 | cpu_stop_queue_work(info->cpu2, info->work2); |
| 222 | } |
| 223 | |
| 224 | /** |
| 225 | * stop_two_cpus - stops two cpus |
| 226 | * @cpu1: the cpu to stop |
| 227 | * @cpu2: the other cpu to stop |
| 228 | * @fn: function to execute |
| 229 | * @arg: argument to @fn |
| 230 | * |
| 231 | * Stops both the current and specified CPU and runs @fn on one of them. |
| 232 | * |
| 233 | * returns when both are completed. |
| 234 | */ |
| 235 | int stop_two_cpus(unsigned int cpu1, unsigned int cpu2, cpu_stop_fn_t fn, void *arg) |
| 236 | { |
Peter Zijlstra | 1be0bd77 | 2013-10-07 11:29:15 +0100 | [diff] [blame] | 237 | struct cpu_stop_done done; |
| 238 | struct cpu_stop_work work1, work2; |
| 239 | struct irq_cpu_stop_queue_work_info call_args; |
Peter Zijlstra | 6acce3e | 2013-10-11 14:38:20 +0200 | [diff] [blame^] | 240 | struct multi_stop_data msdata; |
| 241 | |
| 242 | preempt_disable(); |
| 243 | msdata = (struct multi_stop_data){ |
Peter Zijlstra | 1be0bd77 | 2013-10-07 11:29:15 +0100 | [diff] [blame] | 244 | .fn = fn, |
| 245 | .data = arg, |
| 246 | .num_threads = 2, |
| 247 | .active_cpus = cpumask_of(cpu1), |
| 248 | }; |
| 249 | |
| 250 | work1 = work2 = (struct cpu_stop_work){ |
| 251 | .fn = multi_cpu_stop, |
| 252 | .arg = &msdata, |
| 253 | .done = &done |
| 254 | }; |
| 255 | |
| 256 | call_args = (struct irq_cpu_stop_queue_work_info){ |
| 257 | .cpu1 = cpu1, |
| 258 | .cpu2 = cpu2, |
| 259 | .work1 = &work1, |
| 260 | .work2 = &work2, |
| 261 | }; |
| 262 | |
| 263 | cpu_stop_init_done(&done, 2); |
| 264 | set_state(&msdata, MULTI_STOP_PREPARE); |
| 265 | |
| 266 | /* |
Peter Zijlstra | 6acce3e | 2013-10-11 14:38:20 +0200 | [diff] [blame^] | 267 | * If we observe both CPUs active we know _cpu_down() cannot yet have |
| 268 | * queued its stop_machine works and therefore ours will get executed |
| 269 | * first. Or its not either one of our CPUs that's getting unplugged, |
| 270 | * in which case we don't care. |
| 271 | * |
| 272 | * This relies on the stopper workqueues to be FIFO. |
| 273 | */ |
| 274 | if (!cpu_active(cpu1) || !cpu_active(cpu2)) { |
| 275 | preempt_enable(); |
| 276 | return -ENOENT; |
| 277 | } |
| 278 | |
| 279 | /* |
Peter Zijlstra | 1be0bd77 | 2013-10-07 11:29:15 +0100 | [diff] [blame] | 280 | * Queuing needs to be done by the lowest numbered CPU, to ensure |
| 281 | * that works are always queued in the same order on every CPU. |
| 282 | * This prevents deadlocks. |
| 283 | */ |
Peter Zijlstra | 6acce3e | 2013-10-11 14:38:20 +0200 | [diff] [blame^] | 284 | smp_call_function_single(min(cpu1, cpu2), |
| 285 | &irq_cpu_stop_queue_work, |
Peter Zijlstra | 1be0bd77 | 2013-10-07 11:29:15 +0100 | [diff] [blame] | 286 | &call_args, 0); |
Peter Zijlstra | 6acce3e | 2013-10-11 14:38:20 +0200 | [diff] [blame^] | 287 | preempt_enable(); |
Peter Zijlstra | 1be0bd77 | 2013-10-07 11:29:15 +0100 | [diff] [blame] | 288 | |
| 289 | wait_for_completion(&done.completion); |
Peter Zijlstra | 6acce3e | 2013-10-11 14:38:20 +0200 | [diff] [blame^] | 290 | |
Peter Zijlstra | 1be0bd77 | 2013-10-07 11:29:15 +0100 | [diff] [blame] | 291 | return done.executed ? done.ret : -ENOENT; |
| 292 | } |
| 293 | |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 294 | /** |
| 295 | * stop_one_cpu_nowait - stop a cpu but don't wait for completion |
| 296 | * @cpu: cpu to stop |
| 297 | * @fn: function to execute |
| 298 | * @arg: argument to @fn |
| 299 | * |
| 300 | * Similar to stop_one_cpu() but doesn't wait for completion. The |
| 301 | * caller is responsible for ensuring @work_buf is currently unused |
| 302 | * and will remain untouched until stopper starts executing @fn. |
| 303 | * |
| 304 | * CONTEXT: |
| 305 | * Don't care. |
| 306 | */ |
| 307 | void stop_one_cpu_nowait(unsigned int cpu, cpu_stop_fn_t fn, void *arg, |
| 308 | struct cpu_stop_work *work_buf) |
| 309 | { |
| 310 | *work_buf = (struct cpu_stop_work){ .fn = fn, .arg = arg, }; |
Thomas Gleixner | 860a0ff | 2013-01-31 12:11:13 +0000 | [diff] [blame] | 311 | cpu_stop_queue_work(cpu, work_buf); |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 312 | } |
| 313 | |
| 314 | /* static data for stop_cpus */ |
Suresh Siddha | 192d885 | 2011-06-23 11:19:29 -0700 | [diff] [blame] | 315 | static DEFINE_MUTEX(stop_cpus_mutex); |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 316 | static DEFINE_PER_CPU(struct cpu_stop_work, stop_cpus_work); |
| 317 | |
Tejun Heo | fd7355b | 2011-06-23 11:19:27 -0700 | [diff] [blame] | 318 | static void queue_stop_cpus_work(const struct cpumask *cpumask, |
| 319 | cpu_stop_fn_t fn, void *arg, |
| 320 | struct cpu_stop_done *done) |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 321 | { |
| 322 | struct cpu_stop_work *work; |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 323 | unsigned int cpu; |
| 324 | |
| 325 | /* initialize works and done */ |
| 326 | for_each_cpu(cpu, cpumask) { |
| 327 | work = &per_cpu(stop_cpus_work, cpu); |
| 328 | work->fn = fn; |
| 329 | work->arg = arg; |
Tejun Heo | fd7355b | 2011-06-23 11:19:27 -0700 | [diff] [blame] | 330 | work->done = done; |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 331 | } |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 332 | |
| 333 | /* |
| 334 | * Disable preemption while queueing to avoid getting |
| 335 | * preempted by a stopper which might wait for other stoppers |
| 336 | * to enter @fn which can lead to deadlock. |
| 337 | */ |
| 338 | preempt_disable(); |
| 339 | for_each_cpu(cpu, cpumask) |
Thomas Gleixner | 860a0ff | 2013-01-31 12:11:13 +0000 | [diff] [blame] | 340 | cpu_stop_queue_work(cpu, &per_cpu(stop_cpus_work, cpu)); |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 341 | preempt_enable(); |
Tejun Heo | fd7355b | 2011-06-23 11:19:27 -0700 | [diff] [blame] | 342 | } |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 343 | |
Tejun Heo | fd7355b | 2011-06-23 11:19:27 -0700 | [diff] [blame] | 344 | static int __stop_cpus(const struct cpumask *cpumask, |
| 345 | cpu_stop_fn_t fn, void *arg) |
| 346 | { |
| 347 | struct cpu_stop_done done; |
| 348 | |
| 349 | cpu_stop_init_done(&done, cpumask_weight(cpumask)); |
| 350 | queue_stop_cpus_work(cpumask, fn, arg, &done); |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 351 | wait_for_completion(&done.completion); |
| 352 | return done.executed ? done.ret : -ENOENT; |
| 353 | } |
| 354 | |
| 355 | /** |
| 356 | * stop_cpus - stop multiple cpus |
| 357 | * @cpumask: cpus to stop |
| 358 | * @fn: function to execute |
| 359 | * @arg: argument to @fn |
| 360 | * |
| 361 | * Execute @fn(@arg) on online cpus in @cpumask. On each target cpu, |
| 362 | * @fn is run in a process context with the highest priority |
| 363 | * preempting any task on the cpu and monopolizing it. This function |
| 364 | * returns after all executions are complete. |
| 365 | * |
| 366 | * This function doesn't guarantee the cpus in @cpumask stay online |
| 367 | * till @fn completes. If some cpus go down in the middle, execution |
| 368 | * on the cpu may happen partially or fully on different cpus. @fn |
| 369 | * should either be ready for that or the caller should ensure that |
| 370 | * the cpus stay online until this function completes. |
| 371 | * |
| 372 | * All stop_cpus() calls are serialized making it safe for @fn to wait |
| 373 | * for all cpus to start executing it. |
| 374 | * |
| 375 | * CONTEXT: |
| 376 | * Might sleep. |
| 377 | * |
| 378 | * RETURNS: |
| 379 | * -ENOENT if @fn(@arg) was not executed at all because all cpus in |
| 380 | * @cpumask were offline; otherwise, 0 if all executions of @fn |
| 381 | * returned 0, any non zero return value if any returned non zero. |
| 382 | */ |
| 383 | int stop_cpus(const struct cpumask *cpumask, cpu_stop_fn_t fn, void *arg) |
| 384 | { |
| 385 | int ret; |
| 386 | |
| 387 | /* static works are used, process one request at a time */ |
| 388 | mutex_lock(&stop_cpus_mutex); |
| 389 | ret = __stop_cpus(cpumask, fn, arg); |
| 390 | mutex_unlock(&stop_cpus_mutex); |
| 391 | return ret; |
| 392 | } |
| 393 | |
| 394 | /** |
| 395 | * try_stop_cpus - try to stop multiple cpus |
| 396 | * @cpumask: cpus to stop |
| 397 | * @fn: function to execute |
| 398 | * @arg: argument to @fn |
| 399 | * |
| 400 | * Identical to stop_cpus() except that it fails with -EAGAIN if |
| 401 | * someone else is already using the facility. |
| 402 | * |
| 403 | * CONTEXT: |
| 404 | * Might sleep. |
| 405 | * |
| 406 | * RETURNS: |
| 407 | * -EAGAIN if someone else is already stopping cpus, -ENOENT if |
| 408 | * @fn(@arg) was not executed at all because all cpus in @cpumask were |
| 409 | * offline; otherwise, 0 if all executions of @fn returned 0, any non |
| 410 | * zero return value if any returned non zero. |
| 411 | */ |
| 412 | int try_stop_cpus(const struct cpumask *cpumask, cpu_stop_fn_t fn, void *arg) |
| 413 | { |
| 414 | int ret; |
| 415 | |
| 416 | /* static works are used, process one request at a time */ |
| 417 | if (!mutex_trylock(&stop_cpus_mutex)) |
| 418 | return -EAGAIN; |
| 419 | ret = __stop_cpus(cpumask, fn, arg); |
| 420 | mutex_unlock(&stop_cpus_mutex); |
| 421 | return ret; |
| 422 | } |
| 423 | |
Thomas Gleixner | 14e568e | 2013-01-31 12:11:14 +0000 | [diff] [blame] | 424 | static int cpu_stop_should_run(unsigned int cpu) |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 425 | { |
Thomas Gleixner | 14e568e | 2013-01-31 12:11:14 +0000 | [diff] [blame] | 426 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); |
| 427 | unsigned long flags; |
| 428 | int run; |
| 429 | |
| 430 | spin_lock_irqsave(&stopper->lock, flags); |
| 431 | run = !list_empty(&stopper->works); |
| 432 | spin_unlock_irqrestore(&stopper->lock, flags); |
| 433 | return run; |
| 434 | } |
| 435 | |
| 436 | static void cpu_stopper_thread(unsigned int cpu) |
| 437 | { |
| 438 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 439 | struct cpu_stop_work *work; |
| 440 | int ret; |
| 441 | |
| 442 | repeat: |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 443 | work = NULL; |
| 444 | spin_lock_irq(&stopper->lock); |
| 445 | if (!list_empty(&stopper->works)) { |
| 446 | work = list_first_entry(&stopper->works, |
| 447 | struct cpu_stop_work, list); |
| 448 | list_del_init(&work->list); |
| 449 | } |
| 450 | spin_unlock_irq(&stopper->lock); |
| 451 | |
| 452 | if (work) { |
| 453 | cpu_stop_fn_t fn = work->fn; |
| 454 | void *arg = work->arg; |
| 455 | struct cpu_stop_done *done = work->done; |
Rakib Mullick | ca51c5a | 2010-10-26 14:22:44 -0700 | [diff] [blame] | 456 | char ksym_buf[KSYM_NAME_LEN] __maybe_unused; |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 457 | |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 458 | /* cpu stop callbacks are not allowed to sleep */ |
| 459 | preempt_disable(); |
| 460 | |
| 461 | ret = fn(arg); |
| 462 | if (ret) |
| 463 | done->ret = ret; |
| 464 | |
| 465 | /* restore preemption and check it's still balanced */ |
| 466 | preempt_enable(); |
| 467 | WARN_ONCE(preempt_count(), |
| 468 | "cpu_stop: %s(%p) leaked preempt count\n", |
| 469 | kallsyms_lookup((unsigned long)fn, NULL, NULL, NULL, |
| 470 | ksym_buf), arg); |
| 471 | |
| 472 | cpu_stop_signal_done(done, true); |
Thomas Gleixner | 14e568e | 2013-01-31 12:11:14 +0000 | [diff] [blame] | 473 | goto repeat; |
| 474 | } |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 475 | } |
| 476 | |
Peter Zijlstra | 34f971f | 2010-09-22 13:53:15 +0200 | [diff] [blame] | 477 | extern void sched_set_stop_task(int cpu, struct task_struct *stop); |
| 478 | |
Thomas Gleixner | 14e568e | 2013-01-31 12:11:14 +0000 | [diff] [blame] | 479 | static void cpu_stop_create(unsigned int cpu) |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 480 | { |
Thomas Gleixner | 14e568e | 2013-01-31 12:11:14 +0000 | [diff] [blame] | 481 | sched_set_stop_task(cpu, per_cpu(cpu_stopper_task, cpu)); |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 482 | } |
| 483 | |
Thomas Gleixner | 14e568e | 2013-01-31 12:11:14 +0000 | [diff] [blame] | 484 | static void cpu_stop_park(unsigned int cpu) |
| 485 | { |
| 486 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); |
| 487 | struct cpu_stop_work *work; |
| 488 | unsigned long flags; |
| 489 | |
| 490 | /* drain remaining works */ |
| 491 | spin_lock_irqsave(&stopper->lock, flags); |
| 492 | list_for_each_entry(work, &stopper->works, list) |
| 493 | cpu_stop_signal_done(work->done, false); |
| 494 | stopper->enabled = false; |
| 495 | spin_unlock_irqrestore(&stopper->lock, flags); |
| 496 | } |
| 497 | |
| 498 | static void cpu_stop_unpark(unsigned int cpu) |
| 499 | { |
| 500 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); |
| 501 | |
| 502 | spin_lock_irq(&stopper->lock); |
| 503 | stopper->enabled = true; |
| 504 | spin_unlock_irq(&stopper->lock); |
| 505 | } |
| 506 | |
| 507 | static struct smp_hotplug_thread cpu_stop_threads = { |
| 508 | .store = &cpu_stopper_task, |
| 509 | .thread_should_run = cpu_stop_should_run, |
| 510 | .thread_fn = cpu_stopper_thread, |
| 511 | .thread_comm = "migration/%u", |
| 512 | .create = cpu_stop_create, |
| 513 | .setup = cpu_stop_unpark, |
| 514 | .park = cpu_stop_park, |
Thomas Gleixner | 46c498c | 2013-02-26 18:44:33 +0100 | [diff] [blame] | 515 | .pre_unpark = cpu_stop_unpark, |
Thomas Gleixner | 14e568e | 2013-01-31 12:11:14 +0000 | [diff] [blame] | 516 | .selfparking = true, |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 517 | }; |
| 518 | |
| 519 | static int __init cpu_stop_init(void) |
| 520 | { |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 521 | unsigned int cpu; |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 522 | |
| 523 | for_each_possible_cpu(cpu) { |
| 524 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); |
| 525 | |
| 526 | spin_lock_init(&stopper->lock); |
| 527 | INIT_LIST_HEAD(&stopper->works); |
| 528 | } |
| 529 | |
Thomas Gleixner | 14e568e | 2013-01-31 12:11:14 +0000 | [diff] [blame] | 530 | BUG_ON(smpboot_register_percpu_thread(&cpu_stop_threads)); |
Jeremy Fitzhardinge | f445027 | 2011-10-31 17:11:15 -0700 | [diff] [blame] | 531 | stop_machine_initialized = true; |
Tejun Heo | 1142d81 | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 532 | return 0; |
| 533 | } |
| 534 | early_initcall(cpu_stop_init); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 535 | |
Tejun Heo | bbf1bb3 | 2010-05-08 16:20:53 +0200 | [diff] [blame] | 536 | #ifdef CONFIG_STOP_MACHINE |
| 537 | |
Rusty Russell | 41c7bb9 | 2009-01-01 10:12:28 +1030 | [diff] [blame] | 538 | int __stop_machine(int (*fn)(void *), void *data, const struct cpumask *cpus) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 539 | { |
Peter Zijlstra | 1be0bd77 | 2013-10-07 11:29:15 +0100 | [diff] [blame] | 540 | struct multi_stop_data msdata = { |
| 541 | .fn = fn, |
| 542 | .data = data, |
| 543 | .num_threads = num_online_cpus(), |
| 544 | .active_cpus = cpus, |
| 545 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 546 | |
Jeremy Fitzhardinge | f445027 | 2011-10-31 17:11:15 -0700 | [diff] [blame] | 547 | if (!stop_machine_initialized) { |
| 548 | /* |
| 549 | * Handle the case where stop_machine() is called |
| 550 | * early in boot before stop_machine() has been |
| 551 | * initialized. |
| 552 | */ |
| 553 | unsigned long flags; |
| 554 | int ret; |
| 555 | |
Peter Zijlstra | 1be0bd77 | 2013-10-07 11:29:15 +0100 | [diff] [blame] | 556 | WARN_ON_ONCE(msdata.num_threads != 1); |
Jeremy Fitzhardinge | f445027 | 2011-10-31 17:11:15 -0700 | [diff] [blame] | 557 | |
| 558 | local_irq_save(flags); |
| 559 | hard_irq_disable(); |
| 560 | ret = (*fn)(data); |
| 561 | local_irq_restore(flags); |
| 562 | |
| 563 | return ret; |
| 564 | } |
| 565 | |
Tejun Heo | 3fc1f1e | 2010-05-06 18:49:20 +0200 | [diff] [blame] | 566 | /* Set the initial state and stop all online cpus. */ |
Peter Zijlstra | 1be0bd77 | 2013-10-07 11:29:15 +0100 | [diff] [blame] | 567 | set_state(&msdata, MULTI_STOP_PREPARE); |
| 568 | return stop_cpus(cpu_online_mask, multi_cpu_stop, &msdata); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 569 | } |
| 570 | |
Rusty Russell | 41c7bb9 | 2009-01-01 10:12:28 +1030 | [diff] [blame] | 571 | int stop_machine(int (*fn)(void *), void *data, const struct cpumask *cpus) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 572 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 573 | int ret; |
| 574 | |
| 575 | /* No CPUs can come up or down during this. */ |
Gautham R Shenoy | 86ef5c9 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 576 | get_online_cpus(); |
Rusty Russell | eeec4fa | 2008-07-28 12:16:30 -0500 | [diff] [blame] | 577 | ret = __stop_machine(fn, data, cpus); |
Gautham R Shenoy | 86ef5c9 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 578 | put_online_cpus(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 579 | return ret; |
| 580 | } |
Rusty Russell | eeec4fa | 2008-07-28 12:16:30 -0500 | [diff] [blame] | 581 | EXPORT_SYMBOL_GPL(stop_machine); |
Tejun Heo | bbf1bb3 | 2010-05-08 16:20:53 +0200 | [diff] [blame] | 582 | |
Tejun Heo | f740e6cd | 2011-06-23 11:19:28 -0700 | [diff] [blame] | 583 | /** |
| 584 | * stop_machine_from_inactive_cpu - stop_machine() from inactive CPU |
| 585 | * @fn: the function to run |
| 586 | * @data: the data ptr for the @fn() |
| 587 | * @cpus: the cpus to run the @fn() on (NULL = any online cpu) |
| 588 | * |
| 589 | * This is identical to stop_machine() but can be called from a CPU which |
| 590 | * is not active. The local CPU is in the process of hotplug (so no other |
| 591 | * CPU hotplug can start) and not marked active and doesn't have enough |
| 592 | * context to sleep. |
| 593 | * |
| 594 | * This function provides stop_machine() functionality for such state by |
| 595 | * using busy-wait for synchronization and executing @fn directly for local |
| 596 | * CPU. |
| 597 | * |
| 598 | * CONTEXT: |
| 599 | * Local CPU is inactive. Temporarily stops all active CPUs. |
| 600 | * |
| 601 | * RETURNS: |
| 602 | * 0 if all executions of @fn returned 0, any non zero return value if any |
| 603 | * returned non zero. |
| 604 | */ |
| 605 | int stop_machine_from_inactive_cpu(int (*fn)(void *), void *data, |
| 606 | const struct cpumask *cpus) |
| 607 | { |
Peter Zijlstra | 1be0bd77 | 2013-10-07 11:29:15 +0100 | [diff] [blame] | 608 | struct multi_stop_data msdata = { .fn = fn, .data = data, |
Tejun Heo | f740e6cd | 2011-06-23 11:19:28 -0700 | [diff] [blame] | 609 | .active_cpus = cpus }; |
| 610 | struct cpu_stop_done done; |
| 611 | int ret; |
| 612 | |
| 613 | /* Local CPU must be inactive and CPU hotplug in progress. */ |
| 614 | BUG_ON(cpu_active(raw_smp_processor_id())); |
Peter Zijlstra | 1be0bd77 | 2013-10-07 11:29:15 +0100 | [diff] [blame] | 615 | msdata.num_threads = num_active_cpus() + 1; /* +1 for local */ |
Tejun Heo | f740e6cd | 2011-06-23 11:19:28 -0700 | [diff] [blame] | 616 | |
| 617 | /* No proper task established and can't sleep - busy wait for lock. */ |
| 618 | while (!mutex_trylock(&stop_cpus_mutex)) |
| 619 | cpu_relax(); |
| 620 | |
| 621 | /* Schedule work on other CPUs and execute directly for local CPU */ |
Peter Zijlstra | 1be0bd77 | 2013-10-07 11:29:15 +0100 | [diff] [blame] | 622 | set_state(&msdata, MULTI_STOP_PREPARE); |
Tejun Heo | f740e6cd | 2011-06-23 11:19:28 -0700 | [diff] [blame] | 623 | cpu_stop_init_done(&done, num_active_cpus()); |
Peter Zijlstra | 1be0bd77 | 2013-10-07 11:29:15 +0100 | [diff] [blame] | 624 | queue_stop_cpus_work(cpu_active_mask, multi_cpu_stop, &msdata, |
Tejun Heo | f740e6cd | 2011-06-23 11:19:28 -0700 | [diff] [blame] | 625 | &done); |
Peter Zijlstra | 1be0bd77 | 2013-10-07 11:29:15 +0100 | [diff] [blame] | 626 | ret = multi_cpu_stop(&msdata); |
Tejun Heo | f740e6cd | 2011-06-23 11:19:28 -0700 | [diff] [blame] | 627 | |
| 628 | /* Busy wait for completion. */ |
| 629 | while (!completion_done(&done.completion)) |
| 630 | cpu_relax(); |
| 631 | |
| 632 | mutex_unlock(&stop_cpus_mutex); |
| 633 | return ret ?: done.ret; |
| 634 | } |
| 635 | |
Tejun Heo | bbf1bb3 | 2010-05-08 16:20:53 +0200 | [diff] [blame] | 636 | #endif /* CONFIG_STOP_MACHINE */ |