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> |
| 15 | #include <linux/module.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> |
Benjamin Herrenschmidt | a12bb44 | 2007-05-10 22:22:47 -0700 | [diff] [blame] | 21 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 22 | #include <asm/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; |
| 38 | struct list_head works; /* list of pending works */ |
| 39 | struct task_struct *thread; /* stopper thread */ |
| 40 | bool enabled; /* is this stopper enabled? */ |
| 41 | }; |
| 42 | |
| 43 | static DEFINE_PER_CPU(struct cpu_stopper, cpu_stopper); |
| 44 | |
| 45 | static void cpu_stop_init_done(struct cpu_stop_done *done, unsigned int nr_todo) |
| 46 | { |
| 47 | memset(done, 0, sizeof(*done)); |
| 48 | atomic_set(&done->nr_todo, nr_todo); |
| 49 | init_completion(&done->completion); |
| 50 | } |
| 51 | |
| 52 | /* signal completion unless @done is NULL */ |
| 53 | static void cpu_stop_signal_done(struct cpu_stop_done *done, bool executed) |
| 54 | { |
| 55 | if (done) { |
| 56 | if (executed) |
| 57 | done->executed = true; |
| 58 | if (atomic_dec_and_test(&done->nr_todo)) |
| 59 | complete(&done->completion); |
| 60 | } |
| 61 | } |
| 62 | |
| 63 | /* queue @work to @stopper. if offline, @work is completed immediately */ |
| 64 | static void cpu_stop_queue_work(struct cpu_stopper *stopper, |
| 65 | struct cpu_stop_work *work) |
| 66 | { |
| 67 | unsigned long flags; |
| 68 | |
| 69 | spin_lock_irqsave(&stopper->lock, flags); |
| 70 | |
| 71 | if (stopper->enabled) { |
| 72 | list_add_tail(&work->list, &stopper->works); |
| 73 | wake_up_process(stopper->thread); |
| 74 | } else |
| 75 | cpu_stop_signal_done(work->done, false); |
| 76 | |
| 77 | spin_unlock_irqrestore(&stopper->lock, flags); |
| 78 | } |
| 79 | |
| 80 | /** |
| 81 | * stop_one_cpu - stop a cpu |
| 82 | * @cpu: cpu to stop |
| 83 | * @fn: function to execute |
| 84 | * @arg: argument to @fn |
| 85 | * |
| 86 | * Execute @fn(@arg) on @cpu. @fn is run in a process context with |
| 87 | * the highest priority preempting any task on the cpu and |
| 88 | * monopolizing it. This function returns after the execution is |
| 89 | * complete. |
| 90 | * |
| 91 | * This function doesn't guarantee @cpu stays online till @fn |
| 92 | * completes. If @cpu goes down in the middle, execution may happen |
| 93 | * partially or fully on different cpus. @fn should either be ready |
| 94 | * for that or the caller should ensure that @cpu stays online until |
| 95 | * this function completes. |
| 96 | * |
| 97 | * CONTEXT: |
| 98 | * Might sleep. |
| 99 | * |
| 100 | * RETURNS: |
| 101 | * -ENOENT if @fn(@arg) was not executed because @cpu was offline; |
| 102 | * otherwise, the return value of @fn. |
| 103 | */ |
| 104 | int stop_one_cpu(unsigned int cpu, cpu_stop_fn_t fn, void *arg) |
| 105 | { |
| 106 | struct cpu_stop_done done; |
| 107 | struct cpu_stop_work work = { .fn = fn, .arg = arg, .done = &done }; |
| 108 | |
| 109 | cpu_stop_init_done(&done, 1); |
| 110 | cpu_stop_queue_work(&per_cpu(cpu_stopper, cpu), &work); |
| 111 | wait_for_completion(&done.completion); |
| 112 | return done.executed ? done.ret : -ENOENT; |
| 113 | } |
| 114 | |
| 115 | /** |
| 116 | * stop_one_cpu_nowait - stop a cpu but don't wait for completion |
| 117 | * @cpu: cpu to stop |
| 118 | * @fn: function to execute |
| 119 | * @arg: argument to @fn |
| 120 | * |
| 121 | * Similar to stop_one_cpu() but doesn't wait for completion. The |
| 122 | * caller is responsible for ensuring @work_buf is currently unused |
| 123 | * and will remain untouched until stopper starts executing @fn. |
| 124 | * |
| 125 | * CONTEXT: |
| 126 | * Don't care. |
| 127 | */ |
| 128 | void stop_one_cpu_nowait(unsigned int cpu, cpu_stop_fn_t fn, void *arg, |
| 129 | struct cpu_stop_work *work_buf) |
| 130 | { |
| 131 | *work_buf = (struct cpu_stop_work){ .fn = fn, .arg = arg, }; |
| 132 | cpu_stop_queue_work(&per_cpu(cpu_stopper, cpu), work_buf); |
| 133 | } |
| 134 | |
| 135 | /* static data for stop_cpus */ |
| 136 | static DEFINE_MUTEX(stop_cpus_mutex); |
| 137 | static DEFINE_PER_CPU(struct cpu_stop_work, stop_cpus_work); |
| 138 | |
| 139 | int __stop_cpus(const struct cpumask *cpumask, cpu_stop_fn_t fn, void *arg) |
| 140 | { |
| 141 | struct cpu_stop_work *work; |
| 142 | struct cpu_stop_done done; |
| 143 | unsigned int cpu; |
| 144 | |
| 145 | /* initialize works and done */ |
| 146 | for_each_cpu(cpu, cpumask) { |
| 147 | work = &per_cpu(stop_cpus_work, cpu); |
| 148 | work->fn = fn; |
| 149 | work->arg = arg; |
| 150 | work->done = &done; |
| 151 | } |
| 152 | cpu_stop_init_done(&done, cpumask_weight(cpumask)); |
| 153 | |
| 154 | /* |
| 155 | * Disable preemption while queueing to avoid getting |
| 156 | * preempted by a stopper which might wait for other stoppers |
| 157 | * to enter @fn which can lead to deadlock. |
| 158 | */ |
| 159 | preempt_disable(); |
| 160 | for_each_cpu(cpu, cpumask) |
| 161 | cpu_stop_queue_work(&per_cpu(cpu_stopper, cpu), |
| 162 | &per_cpu(stop_cpus_work, cpu)); |
| 163 | preempt_enable(); |
| 164 | |
| 165 | wait_for_completion(&done.completion); |
| 166 | return done.executed ? done.ret : -ENOENT; |
| 167 | } |
| 168 | |
| 169 | /** |
| 170 | * stop_cpus - stop multiple cpus |
| 171 | * @cpumask: cpus to stop |
| 172 | * @fn: function to execute |
| 173 | * @arg: argument to @fn |
| 174 | * |
| 175 | * Execute @fn(@arg) on online cpus in @cpumask. On each target cpu, |
| 176 | * @fn is run in a process context with the highest priority |
| 177 | * preempting any task on the cpu and monopolizing it. This function |
| 178 | * returns after all executions are complete. |
| 179 | * |
| 180 | * This function doesn't guarantee the cpus in @cpumask stay online |
| 181 | * till @fn completes. If some cpus go down in the middle, execution |
| 182 | * on the cpu may happen partially or fully on different cpus. @fn |
| 183 | * should either be ready for that or the caller should ensure that |
| 184 | * the cpus stay online until this function completes. |
| 185 | * |
| 186 | * All stop_cpus() calls are serialized making it safe for @fn to wait |
| 187 | * for all cpus to start executing it. |
| 188 | * |
| 189 | * CONTEXT: |
| 190 | * Might sleep. |
| 191 | * |
| 192 | * RETURNS: |
| 193 | * -ENOENT if @fn(@arg) was not executed at all because all cpus in |
| 194 | * @cpumask were offline; otherwise, 0 if all executions of @fn |
| 195 | * returned 0, any non zero return value if any returned non zero. |
| 196 | */ |
| 197 | int stop_cpus(const struct cpumask *cpumask, cpu_stop_fn_t fn, void *arg) |
| 198 | { |
| 199 | int ret; |
| 200 | |
| 201 | /* static works are used, process one request at a time */ |
| 202 | mutex_lock(&stop_cpus_mutex); |
| 203 | ret = __stop_cpus(cpumask, fn, arg); |
| 204 | mutex_unlock(&stop_cpus_mutex); |
| 205 | return ret; |
| 206 | } |
| 207 | |
| 208 | /** |
| 209 | * try_stop_cpus - try to stop multiple cpus |
| 210 | * @cpumask: cpus to stop |
| 211 | * @fn: function to execute |
| 212 | * @arg: argument to @fn |
| 213 | * |
| 214 | * Identical to stop_cpus() except that it fails with -EAGAIN if |
| 215 | * someone else is already using the facility. |
| 216 | * |
| 217 | * CONTEXT: |
| 218 | * Might sleep. |
| 219 | * |
| 220 | * RETURNS: |
| 221 | * -EAGAIN if someone else is already stopping cpus, -ENOENT if |
| 222 | * @fn(@arg) was not executed at all because all cpus in @cpumask were |
| 223 | * offline; otherwise, 0 if all executions of @fn returned 0, any non |
| 224 | * zero return value if any returned non zero. |
| 225 | */ |
| 226 | int try_stop_cpus(const struct cpumask *cpumask, cpu_stop_fn_t fn, void *arg) |
| 227 | { |
| 228 | int ret; |
| 229 | |
| 230 | /* static works are used, process one request at a time */ |
| 231 | if (!mutex_trylock(&stop_cpus_mutex)) |
| 232 | return -EAGAIN; |
| 233 | ret = __stop_cpus(cpumask, fn, arg); |
| 234 | mutex_unlock(&stop_cpus_mutex); |
| 235 | return ret; |
| 236 | } |
| 237 | |
| 238 | static int cpu_stopper_thread(void *data) |
| 239 | { |
| 240 | struct cpu_stopper *stopper = data; |
| 241 | struct cpu_stop_work *work; |
| 242 | int ret; |
| 243 | |
| 244 | repeat: |
| 245 | set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */ |
| 246 | |
| 247 | if (kthread_should_stop()) { |
| 248 | __set_current_state(TASK_RUNNING); |
| 249 | return 0; |
| 250 | } |
| 251 | |
| 252 | work = NULL; |
| 253 | spin_lock_irq(&stopper->lock); |
| 254 | if (!list_empty(&stopper->works)) { |
| 255 | work = list_first_entry(&stopper->works, |
| 256 | struct cpu_stop_work, list); |
| 257 | list_del_init(&work->list); |
| 258 | } |
| 259 | spin_unlock_irq(&stopper->lock); |
| 260 | |
| 261 | if (work) { |
| 262 | cpu_stop_fn_t fn = work->fn; |
| 263 | void *arg = work->arg; |
| 264 | struct cpu_stop_done *done = work->done; |
| 265 | char ksym_buf[KSYM_NAME_LEN]; |
| 266 | |
| 267 | __set_current_state(TASK_RUNNING); |
| 268 | |
| 269 | /* cpu stop callbacks are not allowed to sleep */ |
| 270 | preempt_disable(); |
| 271 | |
| 272 | ret = fn(arg); |
| 273 | if (ret) |
| 274 | done->ret = ret; |
| 275 | |
| 276 | /* restore preemption and check it's still balanced */ |
| 277 | preempt_enable(); |
| 278 | WARN_ONCE(preempt_count(), |
| 279 | "cpu_stop: %s(%p) leaked preempt count\n", |
| 280 | kallsyms_lookup((unsigned long)fn, NULL, NULL, NULL, |
| 281 | ksym_buf), arg); |
| 282 | |
| 283 | cpu_stop_signal_done(done, true); |
| 284 | } else |
| 285 | schedule(); |
| 286 | |
| 287 | goto repeat; |
| 288 | } |
| 289 | |
| 290 | /* manage stopper for a cpu, mostly lifted from sched migration thread mgmt */ |
| 291 | static int __cpuinit cpu_stop_cpu_callback(struct notifier_block *nfb, |
| 292 | unsigned long action, void *hcpu) |
| 293 | { |
| 294 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; |
| 295 | unsigned int cpu = (unsigned long)hcpu; |
| 296 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); |
| 297 | struct cpu_stop_work *work; |
| 298 | struct task_struct *p; |
| 299 | |
| 300 | switch (action & ~CPU_TASKS_FROZEN) { |
| 301 | case CPU_UP_PREPARE: |
| 302 | BUG_ON(stopper->thread || stopper->enabled || |
| 303 | !list_empty(&stopper->works)); |
| 304 | p = kthread_create(cpu_stopper_thread, stopper, "stopper/%d", |
| 305 | cpu); |
| 306 | if (IS_ERR(p)) |
| 307 | return NOTIFY_BAD; |
| 308 | sched_setscheduler_nocheck(p, SCHED_FIFO, ¶m); |
| 309 | get_task_struct(p); |
| 310 | stopper->thread = p; |
| 311 | break; |
| 312 | |
| 313 | case CPU_ONLINE: |
| 314 | kthread_bind(stopper->thread, cpu); |
| 315 | /* strictly unnecessary, as first user will wake it */ |
| 316 | wake_up_process(stopper->thread); |
| 317 | /* mark enabled */ |
| 318 | spin_lock_irq(&stopper->lock); |
| 319 | stopper->enabled = true; |
| 320 | spin_unlock_irq(&stopper->lock); |
| 321 | break; |
| 322 | |
| 323 | #ifdef CONFIG_HOTPLUG_CPU |
| 324 | case CPU_UP_CANCELED: |
| 325 | case CPU_DEAD: |
| 326 | /* kill the stopper */ |
| 327 | kthread_stop(stopper->thread); |
| 328 | /* drain remaining works */ |
| 329 | spin_lock_irq(&stopper->lock); |
| 330 | list_for_each_entry(work, &stopper->works, list) |
| 331 | cpu_stop_signal_done(work->done, false); |
| 332 | stopper->enabled = false; |
| 333 | spin_unlock_irq(&stopper->lock); |
| 334 | /* release the stopper */ |
| 335 | put_task_struct(stopper->thread); |
| 336 | stopper->thread = NULL; |
| 337 | break; |
| 338 | #endif |
| 339 | } |
| 340 | |
| 341 | return NOTIFY_OK; |
| 342 | } |
| 343 | |
| 344 | /* |
| 345 | * Give it a higher priority so that cpu stopper is available to other |
| 346 | * cpu notifiers. It currently shares the same priority as sched |
| 347 | * migration_notifier. |
| 348 | */ |
| 349 | static struct notifier_block __cpuinitdata cpu_stop_cpu_notifier = { |
| 350 | .notifier_call = cpu_stop_cpu_callback, |
| 351 | .priority = 10, |
| 352 | }; |
| 353 | |
| 354 | static int __init cpu_stop_init(void) |
| 355 | { |
| 356 | void *bcpu = (void *)(long)smp_processor_id(); |
| 357 | unsigned int cpu; |
| 358 | int err; |
| 359 | |
| 360 | for_each_possible_cpu(cpu) { |
| 361 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); |
| 362 | |
| 363 | spin_lock_init(&stopper->lock); |
| 364 | INIT_LIST_HEAD(&stopper->works); |
| 365 | } |
| 366 | |
| 367 | /* start one for the boot cpu */ |
| 368 | err = cpu_stop_cpu_callback(&cpu_stop_cpu_notifier, CPU_UP_PREPARE, |
| 369 | bcpu); |
| 370 | BUG_ON(err == NOTIFY_BAD); |
| 371 | cpu_stop_cpu_callback(&cpu_stop_cpu_notifier, CPU_ONLINE, bcpu); |
| 372 | register_cpu_notifier(&cpu_stop_cpu_notifier); |
| 373 | |
| 374 | return 0; |
| 375 | } |
| 376 | early_initcall(cpu_stop_init); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 377 | |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 378 | /* This controls the threads on each CPU. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 379 | enum stopmachine_state { |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 380 | /* Dummy starting state for thread. */ |
| 381 | STOPMACHINE_NONE, |
| 382 | /* Awaiting everyone to be scheduled. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 383 | STOPMACHINE_PREPARE, |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 384 | /* Disable interrupts. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 385 | STOPMACHINE_DISABLE_IRQ, |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 386 | /* Run the function */ |
Jason Baron | 5c2aed6 | 2008-02-28 11:33:03 -0500 | [diff] [blame] | 387 | STOPMACHINE_RUN, |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 388 | /* Exit */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 389 | STOPMACHINE_EXIT, |
| 390 | }; |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 391 | static enum stopmachine_state state; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 392 | |
Jason Baron | 5c2aed6 | 2008-02-28 11:33:03 -0500 | [diff] [blame] | 393 | struct stop_machine_data { |
| 394 | int (*fn)(void *); |
| 395 | void *data; |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 396 | int fnret; |
| 397 | }; |
Jason Baron | 5c2aed6 | 2008-02-28 11:33:03 -0500 | [diff] [blame] | 398 | |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 399 | /* Like num_online_cpus(), but hotplug cpu uses us, so we need this. */ |
| 400 | static unsigned int num_threads; |
| 401 | static atomic_t thread_ack; |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 402 | static DEFINE_MUTEX(lock); |
Heiko Carstens | 9ea09af | 2008-12-22 12:36:30 +0100 | [diff] [blame] | 403 | /* setup_lock protects refcount, stop_machine_wq and stop_machine_work. */ |
| 404 | static DEFINE_MUTEX(setup_lock); |
| 405 | /* Users of stop_machine. */ |
| 406 | static int refcount; |
Heiko Carstens | c9583e5 | 2008-10-13 23:50:10 +0200 | [diff] [blame] | 407 | static struct workqueue_struct *stop_machine_wq; |
| 408 | static struct stop_machine_data active, idle; |
Rusty Russell | 612a726 | 2009-03-30 22:05:16 -0600 | [diff] [blame] | 409 | static const struct cpumask *active_cpus; |
Tejun Heo | 43cf38e | 2010-02-02 14:38:57 +0900 | [diff] [blame] | 410 | static void __percpu *stop_machine_work; |
Heiko Carstens | c9583e5 | 2008-10-13 23:50:10 +0200 | [diff] [blame] | 411 | |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 412 | static void set_state(enum stopmachine_state newstate) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 413 | { |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 414 | /* Reset ack counter. */ |
| 415 | atomic_set(&thread_ack, num_threads); |
| 416 | smp_wmb(); |
| 417 | state = newstate; |
| 418 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 419 | |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 420 | /* Last one to ack a state moves to the next state. */ |
| 421 | static void ack_state(void) |
| 422 | { |
Heiko Carstens | c9583e5 | 2008-10-13 23:50:10 +0200 | [diff] [blame] | 423 | if (atomic_dec_and_test(&thread_ack)) |
| 424 | set_state(state + 1); |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 425 | } |
Andrew Morton | d8cb7c1 | 2006-07-03 17:32:22 -0700 | [diff] [blame] | 426 | |
Heiko Carstens | c9583e5 | 2008-10-13 23:50:10 +0200 | [diff] [blame] | 427 | /* This is the actual function which stops the CPU. It runs |
| 428 | * in the context of a dedicated stopmachine workqueue. */ |
| 429 | static void stop_cpu(struct work_struct *unused) |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 430 | { |
| 431 | enum stopmachine_state curstate = STOPMACHINE_NONE; |
Heiko Carstens | c9583e5 | 2008-10-13 23:50:10 +0200 | [diff] [blame] | 432 | struct stop_machine_data *smdata = &idle; |
| 433 | int cpu = smp_processor_id(); |
Heiko Carstens | 8163bca | 2008-10-22 10:00:26 -0500 | [diff] [blame] | 434 | int err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 435 | |
Heiko Carstens | c9583e5 | 2008-10-13 23:50:10 +0200 | [diff] [blame] | 436 | if (!active_cpus) { |
Rusty Russell | 41c7bb9 | 2009-01-01 10:12:28 +1030 | [diff] [blame] | 437 | if (cpu == cpumask_first(cpu_online_mask)) |
Heiko Carstens | c9583e5 | 2008-10-13 23:50:10 +0200 | [diff] [blame] | 438 | smdata = &active; |
| 439 | } else { |
Rusty Russell | 41c7bb9 | 2009-01-01 10:12:28 +1030 | [diff] [blame] | 440 | if (cpumask_test_cpu(cpu, active_cpus)) |
Heiko Carstens | c9583e5 | 2008-10-13 23:50:10 +0200 | [diff] [blame] | 441 | smdata = &active; |
| 442 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 443 | /* Simple state machine */ |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 444 | do { |
| 445 | /* Chill out and ensure we re-read stopmachine_state. */ |
Christian Borntraeger | 3401a61e | 2008-05-08 15:20:38 +0200 | [diff] [blame] | 446 | cpu_relax(); |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 447 | if (state != curstate) { |
| 448 | curstate = state; |
| 449 | switch (curstate) { |
| 450 | case STOPMACHINE_DISABLE_IRQ: |
| 451 | local_irq_disable(); |
| 452 | hard_irq_disable(); |
| 453 | break; |
| 454 | case STOPMACHINE_RUN: |
Heiko Carstens | 8163bca | 2008-10-22 10:00:26 -0500 | [diff] [blame] | 455 | /* On multiple CPUs only a single error code |
| 456 | * is needed to tell that something failed. */ |
| 457 | err = smdata->fn(smdata->data); |
| 458 | if (err) |
| 459 | smdata->fnret = err; |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 460 | break; |
| 461 | default: |
| 462 | break; |
| 463 | } |
| 464 | ack_state(); |
| 465 | } |
| 466 | } while (curstate != STOPMACHINE_EXIT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 467 | |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 468 | local_irq_enable(); |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 469 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 470 | |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 471 | /* Callback for CPUs which aren't supposed to do anything. */ |
| 472 | static int chill(void *unused) |
| 473 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 474 | return 0; |
| 475 | } |
| 476 | |
Heiko Carstens | 9ea09af | 2008-12-22 12:36:30 +0100 | [diff] [blame] | 477 | int stop_machine_create(void) |
| 478 | { |
| 479 | mutex_lock(&setup_lock); |
| 480 | if (refcount) |
| 481 | goto done; |
| 482 | stop_machine_wq = create_rt_workqueue("kstop"); |
| 483 | if (!stop_machine_wq) |
| 484 | goto err_out; |
| 485 | stop_machine_work = alloc_percpu(struct work_struct); |
| 486 | if (!stop_machine_work) |
| 487 | goto err_out; |
| 488 | done: |
| 489 | refcount++; |
| 490 | mutex_unlock(&setup_lock); |
| 491 | return 0; |
| 492 | |
| 493 | err_out: |
| 494 | if (stop_machine_wq) |
| 495 | destroy_workqueue(stop_machine_wq); |
| 496 | mutex_unlock(&setup_lock); |
| 497 | return -ENOMEM; |
| 498 | } |
| 499 | EXPORT_SYMBOL_GPL(stop_machine_create); |
| 500 | |
| 501 | void stop_machine_destroy(void) |
| 502 | { |
| 503 | mutex_lock(&setup_lock); |
| 504 | refcount--; |
| 505 | if (refcount) |
| 506 | goto done; |
| 507 | destroy_workqueue(stop_machine_wq); |
| 508 | free_percpu(stop_machine_work); |
| 509 | done: |
| 510 | mutex_unlock(&setup_lock); |
| 511 | } |
| 512 | EXPORT_SYMBOL_GPL(stop_machine_destroy); |
| 513 | |
Rusty Russell | 41c7bb9 | 2009-01-01 10:12:28 +1030 | [diff] [blame] | 514 | int __stop_machine(int (*fn)(void *), void *data, const struct cpumask *cpus) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 515 | { |
Heiko Carstens | c9583e5 | 2008-10-13 23:50:10 +0200 | [diff] [blame] | 516 | struct work_struct *sm_work; |
Rusty Russell | e14c8bf | 2008-11-17 08:22:18 +1030 | [diff] [blame] | 517 | int i, ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 518 | |
Heiko Carstens | c9583e5 | 2008-10-13 23:50:10 +0200 | [diff] [blame] | 519 | /* Set up initial state. */ |
| 520 | mutex_lock(&lock); |
| 521 | num_threads = num_online_cpus(); |
| 522 | active_cpus = cpus; |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 523 | active.fn = fn; |
| 524 | active.data = data; |
| 525 | active.fnret = 0; |
| 526 | idle.fn = chill; |
| 527 | idle.data = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 528 | |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 529 | set_state(STOPMACHINE_PREPARE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 530 | |
Heiko Carstens | c9583e5 | 2008-10-13 23:50:10 +0200 | [diff] [blame] | 531 | /* Schedule the stop_cpu work on all cpus: hold this CPU so one |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 532 | * doesn't hit this CPU until we're ready. */ |
Rusty Russell | eeec4fa | 2008-07-28 12:16:30 -0500 | [diff] [blame] | 533 | get_cpu(); |
Heiko Carstens | c9583e5 | 2008-10-13 23:50:10 +0200 | [diff] [blame] | 534 | for_each_online_cpu(i) { |
Rusty Russell | b36128c | 2009-02-20 16:29:08 +0900 | [diff] [blame] | 535 | sm_work = per_cpu_ptr(stop_machine_work, i); |
Heiko Carstens | c9583e5 | 2008-10-13 23:50:10 +0200 | [diff] [blame] | 536 | INIT_WORK(sm_work, stop_cpu); |
| 537 | queue_work_on(i, stop_machine_wq, sm_work); |
| 538 | } |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 539 | /* This will release the thread on our CPU. */ |
| 540 | put_cpu(); |
Heiko Carstens | c9583e5 | 2008-10-13 23:50:10 +0200 | [diff] [blame] | 541 | flush_workqueue(stop_machine_wq); |
Rusty Russell | e14c8bf | 2008-11-17 08:22:18 +1030 | [diff] [blame] | 542 | ret = active.fnret; |
Rusty Russell | ffdb597 | 2008-07-28 12:16:28 -0500 | [diff] [blame] | 543 | mutex_unlock(&lock); |
Rusty Russell | e14c8bf | 2008-11-17 08:22:18 +1030 | [diff] [blame] | 544 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 545 | } |
| 546 | |
Rusty Russell | 41c7bb9 | 2009-01-01 10:12:28 +1030 | [diff] [blame] | 547 | int stop_machine(int (*fn)(void *), void *data, const struct cpumask *cpus) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 548 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 549 | int ret; |
| 550 | |
Heiko Carstens | 9ea09af | 2008-12-22 12:36:30 +0100 | [diff] [blame] | 551 | ret = stop_machine_create(); |
| 552 | if (ret) |
| 553 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 554 | /* No CPUs can come up or down during this. */ |
Gautham R Shenoy | 86ef5c9 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 555 | get_online_cpus(); |
Rusty Russell | eeec4fa | 2008-07-28 12:16:30 -0500 | [diff] [blame] | 556 | ret = __stop_machine(fn, data, cpus); |
Gautham R Shenoy | 86ef5c9 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 557 | put_online_cpus(); |
Heiko Carstens | 9ea09af | 2008-12-22 12:36:30 +0100 | [diff] [blame] | 558 | stop_machine_destroy(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 559 | return ret; |
| 560 | } |
Rusty Russell | eeec4fa | 2008-07-28 12:16:30 -0500 | [diff] [blame] | 561 | EXPORT_SYMBOL_GPL(stop_machine); |