| 1 | /* CPU control. |
| 2 | * (C) 2001, 2002, 2003, 2004 Rusty Russell |
| 3 | * |
| 4 | * This code is licenced under the GPL. |
| 5 | */ |
| 6 | #include <linux/proc_fs.h> |
| 7 | #include <linux/smp.h> |
| 8 | #include <linux/init.h> |
| 9 | #include <linux/notifier.h> |
| 10 | #include <linux/sched.h> |
| 11 | #include <linux/unistd.h> |
| 12 | #include <linux/cpu.h> |
| 13 | #include <linux/oom.h> |
| 14 | #include <linux/rcupdate.h> |
| 15 | #include <linux/export.h> |
| 16 | #include <linux/bug.h> |
| 17 | #include <linux/kthread.h> |
| 18 | #include <linux/stop_machine.h> |
| 19 | #include <linux/mutex.h> |
| 20 | #include <linux/gfp.h> |
| 21 | #include <linux/suspend.h> |
| 22 | #include <linux/lockdep.h> |
| 23 | #include <trace/events/power.h> |
| 24 | |
| 25 | #include "smpboot.h" |
| 26 | |
| 27 | #ifdef CONFIG_SMP |
| 28 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
| 29 | static DEFINE_MUTEX(cpu_add_remove_lock); |
| 30 | |
| 31 | /* |
| 32 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
| 33 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. |
| 34 | * The APIs cpu_notifier_register_begin/done() must be used to protect CPU |
| 35 | * hotplug callback (un)registration performed using __register_cpu_notifier() |
| 36 | * or __unregister_cpu_notifier(). |
| 37 | */ |
| 38 | void cpu_maps_update_begin(void) |
| 39 | { |
| 40 | mutex_lock(&cpu_add_remove_lock); |
| 41 | } |
| 42 | EXPORT_SYMBOL(cpu_notifier_register_begin); |
| 43 | |
| 44 | void cpu_maps_update_done(void) |
| 45 | { |
| 46 | mutex_unlock(&cpu_add_remove_lock); |
| 47 | } |
| 48 | EXPORT_SYMBOL(cpu_notifier_register_done); |
| 49 | |
| 50 | static RAW_NOTIFIER_HEAD(cpu_chain); |
| 51 | |
| 52 | /* If set, cpu_up and cpu_down will return -EBUSY and do nothing. |
| 53 | * Should always be manipulated under cpu_add_remove_lock |
| 54 | */ |
| 55 | static int cpu_hotplug_disabled; |
| 56 | |
| 57 | #ifdef CONFIG_HOTPLUG_CPU |
| 58 | |
| 59 | static struct { |
| 60 | struct task_struct *active_writer; |
| 61 | /* wait queue to wake up the active_writer */ |
| 62 | wait_queue_head_t wq; |
| 63 | /* verifies that no writer will get active while readers are active */ |
| 64 | struct mutex lock; |
| 65 | /* |
| 66 | * Also blocks the new readers during |
| 67 | * an ongoing cpu hotplug operation. |
| 68 | */ |
| 69 | atomic_t refcount; |
| 70 | |
| 71 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
| 72 | struct lockdep_map dep_map; |
| 73 | #endif |
| 74 | } cpu_hotplug = { |
| 75 | .active_writer = NULL, |
| 76 | .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq), |
| 77 | .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock), |
| 78 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
| 79 | .dep_map = {.name = "cpu_hotplug.lock" }, |
| 80 | #endif |
| 81 | }; |
| 82 | |
| 83 | /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */ |
| 84 | #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map) |
| 85 | #define cpuhp_lock_acquire_tryread() \ |
| 86 | lock_map_acquire_tryread(&cpu_hotplug.dep_map) |
| 87 | #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map) |
| 88 | #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map) |
| 89 | |
| 90 | |
| 91 | void get_online_cpus(void) |
| 92 | { |
| 93 | might_sleep(); |
| 94 | if (cpu_hotplug.active_writer == current) |
| 95 | return; |
| 96 | cpuhp_lock_acquire_read(); |
| 97 | mutex_lock(&cpu_hotplug.lock); |
| 98 | atomic_inc(&cpu_hotplug.refcount); |
| 99 | mutex_unlock(&cpu_hotplug.lock); |
| 100 | } |
| 101 | EXPORT_SYMBOL_GPL(get_online_cpus); |
| 102 | |
| 103 | bool try_get_online_cpus(void) |
| 104 | { |
| 105 | if (cpu_hotplug.active_writer == current) |
| 106 | return true; |
| 107 | if (!mutex_trylock(&cpu_hotplug.lock)) |
| 108 | return false; |
| 109 | cpuhp_lock_acquire_tryread(); |
| 110 | atomic_inc(&cpu_hotplug.refcount); |
| 111 | mutex_unlock(&cpu_hotplug.lock); |
| 112 | return true; |
| 113 | } |
| 114 | EXPORT_SYMBOL_GPL(try_get_online_cpus); |
| 115 | |
| 116 | void put_online_cpus(void) |
| 117 | { |
| 118 | int refcount; |
| 119 | |
| 120 | if (cpu_hotplug.active_writer == current) |
| 121 | return; |
| 122 | |
| 123 | refcount = atomic_dec_return(&cpu_hotplug.refcount); |
| 124 | if (WARN_ON(refcount < 0)) /* try to fix things up */ |
| 125 | atomic_inc(&cpu_hotplug.refcount); |
| 126 | |
| 127 | if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq)) |
| 128 | wake_up(&cpu_hotplug.wq); |
| 129 | |
| 130 | cpuhp_lock_release(); |
| 131 | |
| 132 | } |
| 133 | EXPORT_SYMBOL_GPL(put_online_cpus); |
| 134 | |
| 135 | /* |
| 136 | * This ensures that the hotplug operation can begin only when the |
| 137 | * refcount goes to zero. |
| 138 | * |
| 139 | * Note that during a cpu-hotplug operation, the new readers, if any, |
| 140 | * will be blocked by the cpu_hotplug.lock |
| 141 | * |
| 142 | * Since cpu_hotplug_begin() is always called after invoking |
| 143 | * cpu_maps_update_begin(), we can be sure that only one writer is active. |
| 144 | * |
| 145 | * Note that theoretically, there is a possibility of a livelock: |
| 146 | * - Refcount goes to zero, last reader wakes up the sleeping |
| 147 | * writer. |
| 148 | * - Last reader unlocks the cpu_hotplug.lock. |
| 149 | * - A new reader arrives at this moment, bumps up the refcount. |
| 150 | * - The writer acquires the cpu_hotplug.lock finds the refcount |
| 151 | * non zero and goes to sleep again. |
| 152 | * |
| 153 | * However, this is very difficult to achieve in practice since |
| 154 | * get_online_cpus() not an api which is called all that often. |
| 155 | * |
| 156 | */ |
| 157 | void cpu_hotplug_begin(void) |
| 158 | { |
| 159 | DEFINE_WAIT(wait); |
| 160 | |
| 161 | cpu_hotplug.active_writer = current; |
| 162 | cpuhp_lock_acquire(); |
| 163 | |
| 164 | for (;;) { |
| 165 | mutex_lock(&cpu_hotplug.lock); |
| 166 | prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE); |
| 167 | if (likely(!atomic_read(&cpu_hotplug.refcount))) |
| 168 | break; |
| 169 | mutex_unlock(&cpu_hotplug.lock); |
| 170 | schedule(); |
| 171 | } |
| 172 | finish_wait(&cpu_hotplug.wq, &wait); |
| 173 | } |
| 174 | |
| 175 | void cpu_hotplug_done(void) |
| 176 | { |
| 177 | cpu_hotplug.active_writer = NULL; |
| 178 | mutex_unlock(&cpu_hotplug.lock); |
| 179 | cpuhp_lock_release(); |
| 180 | } |
| 181 | |
| 182 | /* |
| 183 | * Wait for currently running CPU hotplug operations to complete (if any) and |
| 184 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects |
| 185 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the |
| 186 | * hotplug path before performing hotplug operations. So acquiring that lock |
| 187 | * guarantees mutual exclusion from any currently running hotplug operations. |
| 188 | */ |
| 189 | void cpu_hotplug_disable(void) |
| 190 | { |
| 191 | cpu_maps_update_begin(); |
| 192 | cpu_hotplug_disabled = 1; |
| 193 | cpu_maps_update_done(); |
| 194 | } |
| 195 | |
| 196 | void cpu_hotplug_enable(void) |
| 197 | { |
| 198 | cpu_maps_update_begin(); |
| 199 | cpu_hotplug_disabled = 0; |
| 200 | cpu_maps_update_done(); |
| 201 | } |
| 202 | |
| 203 | #endif /* CONFIG_HOTPLUG_CPU */ |
| 204 | |
| 205 | /* Need to know about CPUs going up/down? */ |
| 206 | int __ref register_cpu_notifier(struct notifier_block *nb) |
| 207 | { |
| 208 | int ret; |
| 209 | cpu_maps_update_begin(); |
| 210 | ret = raw_notifier_chain_register(&cpu_chain, nb); |
| 211 | cpu_maps_update_done(); |
| 212 | return ret; |
| 213 | } |
| 214 | |
| 215 | int __ref __register_cpu_notifier(struct notifier_block *nb) |
| 216 | { |
| 217 | return raw_notifier_chain_register(&cpu_chain, nb); |
| 218 | } |
| 219 | |
| 220 | static int __cpu_notify(unsigned long val, void *v, int nr_to_call, |
| 221 | int *nr_calls) |
| 222 | { |
| 223 | int ret; |
| 224 | |
| 225 | ret = __raw_notifier_call_chain(&cpu_chain, val, v, nr_to_call, |
| 226 | nr_calls); |
| 227 | |
| 228 | return notifier_to_errno(ret); |
| 229 | } |
| 230 | |
| 231 | static int cpu_notify(unsigned long val, void *v) |
| 232 | { |
| 233 | return __cpu_notify(val, v, -1, NULL); |
| 234 | } |
| 235 | |
| 236 | #ifdef CONFIG_HOTPLUG_CPU |
| 237 | |
| 238 | static void cpu_notify_nofail(unsigned long val, void *v) |
| 239 | { |
| 240 | BUG_ON(cpu_notify(val, v)); |
| 241 | } |
| 242 | EXPORT_SYMBOL(register_cpu_notifier); |
| 243 | EXPORT_SYMBOL(__register_cpu_notifier); |
| 244 | |
| 245 | void __ref unregister_cpu_notifier(struct notifier_block *nb) |
| 246 | { |
| 247 | cpu_maps_update_begin(); |
| 248 | raw_notifier_chain_unregister(&cpu_chain, nb); |
| 249 | cpu_maps_update_done(); |
| 250 | } |
| 251 | EXPORT_SYMBOL(unregister_cpu_notifier); |
| 252 | |
| 253 | void __ref __unregister_cpu_notifier(struct notifier_block *nb) |
| 254 | { |
| 255 | raw_notifier_chain_unregister(&cpu_chain, nb); |
| 256 | } |
| 257 | EXPORT_SYMBOL(__unregister_cpu_notifier); |
| 258 | |
| 259 | /** |
| 260 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU |
| 261 | * @cpu: a CPU id |
| 262 | * |
| 263 | * This function walks all processes, finds a valid mm struct for each one and |
| 264 | * then clears a corresponding bit in mm's cpumask. While this all sounds |
| 265 | * trivial, there are various non-obvious corner cases, which this function |
| 266 | * tries to solve in a safe manner. |
| 267 | * |
| 268 | * Also note that the function uses a somewhat relaxed locking scheme, so it may |
| 269 | * be called only for an already offlined CPU. |
| 270 | */ |
| 271 | void clear_tasks_mm_cpumask(int cpu) |
| 272 | { |
| 273 | struct task_struct *p; |
| 274 | |
| 275 | /* |
| 276 | * This function is called after the cpu is taken down and marked |
| 277 | * offline, so its not like new tasks will ever get this cpu set in |
| 278 | * their mm mask. -- Peter Zijlstra |
| 279 | * Thus, we may use rcu_read_lock() here, instead of grabbing |
| 280 | * full-fledged tasklist_lock. |
| 281 | */ |
| 282 | WARN_ON(cpu_online(cpu)); |
| 283 | rcu_read_lock(); |
| 284 | for_each_process(p) { |
| 285 | struct task_struct *t; |
| 286 | |
| 287 | /* |
| 288 | * Main thread might exit, but other threads may still have |
| 289 | * a valid mm. Find one. |
| 290 | */ |
| 291 | t = find_lock_task_mm(p); |
| 292 | if (!t) |
| 293 | continue; |
| 294 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); |
| 295 | task_unlock(t); |
| 296 | } |
| 297 | rcu_read_unlock(); |
| 298 | } |
| 299 | |
| 300 | static inline void check_for_tasks(int dead_cpu) |
| 301 | { |
| 302 | struct task_struct *g, *p; |
| 303 | |
| 304 | read_lock_irq(&tasklist_lock); |
| 305 | do_each_thread(g, p) { |
| 306 | if (!p->on_rq) |
| 307 | continue; |
| 308 | /* |
| 309 | * We do the check with unlocked task_rq(p)->lock. |
| 310 | * Order the reading to do not warn about a task, |
| 311 | * which was running on this cpu in the past, and |
| 312 | * it's just been woken on another cpu. |
| 313 | */ |
| 314 | rmb(); |
| 315 | if (task_cpu(p) != dead_cpu) |
| 316 | continue; |
| 317 | |
| 318 | pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n", |
| 319 | p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags); |
| 320 | } while_each_thread(g, p); |
| 321 | read_unlock_irq(&tasklist_lock); |
| 322 | } |
| 323 | |
| 324 | struct take_cpu_down_param { |
| 325 | unsigned long mod; |
| 326 | void *hcpu; |
| 327 | }; |
| 328 | |
| 329 | /* Take this CPU down. */ |
| 330 | static int __ref take_cpu_down(void *_param) |
| 331 | { |
| 332 | struct take_cpu_down_param *param = _param; |
| 333 | int err; |
| 334 | |
| 335 | /* Ensure this CPU doesn't handle any more interrupts. */ |
| 336 | err = __cpu_disable(); |
| 337 | if (err < 0) |
| 338 | return err; |
| 339 | |
| 340 | cpu_notify(CPU_DYING | param->mod, param->hcpu); |
| 341 | /* Park the stopper thread */ |
| 342 | kthread_park(current); |
| 343 | return 0; |
| 344 | } |
| 345 | |
| 346 | /* Requires cpu_add_remove_lock to be held */ |
| 347 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen) |
| 348 | { |
| 349 | int err, nr_calls = 0; |
| 350 | void *hcpu = (void *)(long)cpu; |
| 351 | unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0; |
| 352 | struct take_cpu_down_param tcd_param = { |
| 353 | .mod = mod, |
| 354 | .hcpu = hcpu, |
| 355 | }; |
| 356 | |
| 357 | if (num_online_cpus() == 1) |
| 358 | return -EBUSY; |
| 359 | |
| 360 | if (!cpu_online(cpu)) |
| 361 | return -EINVAL; |
| 362 | |
| 363 | cpu_hotplug_begin(); |
| 364 | |
| 365 | err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls); |
| 366 | if (err) { |
| 367 | nr_calls--; |
| 368 | __cpu_notify(CPU_DOWN_FAILED | mod, hcpu, nr_calls, NULL); |
| 369 | pr_warn("%s: attempt to take down CPU %u failed\n", |
| 370 | __func__, cpu); |
| 371 | goto out_release; |
| 372 | } |
| 373 | |
| 374 | /* |
| 375 | * By now we've cleared cpu_active_mask, wait for all preempt-disabled |
| 376 | * and RCU users of this state to go away such that all new such users |
| 377 | * will observe it. |
| 378 | * |
| 379 | * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might |
| 380 | * not imply sync_sched(), so explicitly call both. |
| 381 | * |
| 382 | * Do sync before park smpboot threads to take care the rcu boost case. |
| 383 | */ |
| 384 | #ifdef CONFIG_PREEMPT |
| 385 | synchronize_sched(); |
| 386 | #endif |
| 387 | synchronize_rcu(); |
| 388 | |
| 389 | smpboot_park_threads(cpu); |
| 390 | |
| 391 | /* |
| 392 | * So now all preempt/rcu users must observe !cpu_active(). |
| 393 | */ |
| 394 | |
| 395 | err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu)); |
| 396 | if (err) { |
| 397 | /* CPU didn't die: tell everyone. Can't complain. */ |
| 398 | smpboot_unpark_threads(cpu); |
| 399 | cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu); |
| 400 | goto out_release; |
| 401 | } |
| 402 | BUG_ON(cpu_online(cpu)); |
| 403 | |
| 404 | /* |
| 405 | * The migration_call() CPU_DYING callback will have removed all |
| 406 | * runnable tasks from the cpu, there's only the idle task left now |
| 407 | * that the migration thread is done doing the stop_machine thing. |
| 408 | * |
| 409 | * Wait for the stop thread to go away. |
| 410 | */ |
| 411 | while (!idle_cpu(cpu)) |
| 412 | cpu_relax(); |
| 413 | |
| 414 | /* This actually kills the CPU. */ |
| 415 | __cpu_die(cpu); |
| 416 | |
| 417 | /* CPU is completely dead: tell everyone. Too late to complain. */ |
| 418 | cpu_notify_nofail(CPU_DEAD | mod, hcpu); |
| 419 | |
| 420 | check_for_tasks(cpu); |
| 421 | |
| 422 | out_release: |
| 423 | cpu_hotplug_done(); |
| 424 | if (!err) |
| 425 | cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu); |
| 426 | return err; |
| 427 | } |
| 428 | |
| 429 | int __ref cpu_down(unsigned int cpu) |
| 430 | { |
| 431 | int err; |
| 432 | |
| 433 | cpu_maps_update_begin(); |
| 434 | |
| 435 | if (cpu_hotplug_disabled) { |
| 436 | err = -EBUSY; |
| 437 | goto out; |
| 438 | } |
| 439 | |
| 440 | err = _cpu_down(cpu, 0); |
| 441 | |
| 442 | out: |
| 443 | cpu_maps_update_done(); |
| 444 | return err; |
| 445 | } |
| 446 | EXPORT_SYMBOL(cpu_down); |
| 447 | #endif /*CONFIG_HOTPLUG_CPU*/ |
| 448 | |
| 449 | /* Requires cpu_add_remove_lock to be held */ |
| 450 | static int _cpu_up(unsigned int cpu, int tasks_frozen) |
| 451 | { |
| 452 | int ret, nr_calls = 0; |
| 453 | void *hcpu = (void *)(long)cpu; |
| 454 | unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0; |
| 455 | struct task_struct *idle; |
| 456 | |
| 457 | cpu_hotplug_begin(); |
| 458 | |
| 459 | if (cpu_online(cpu) || !cpu_present(cpu)) { |
| 460 | ret = -EINVAL; |
| 461 | goto out; |
| 462 | } |
| 463 | |
| 464 | idle = idle_thread_get(cpu); |
| 465 | if (IS_ERR(idle)) { |
| 466 | ret = PTR_ERR(idle); |
| 467 | goto out; |
| 468 | } |
| 469 | |
| 470 | ret = smpboot_create_threads(cpu); |
| 471 | if (ret) |
| 472 | goto out; |
| 473 | |
| 474 | ret = __cpu_notify(CPU_UP_PREPARE | mod, hcpu, -1, &nr_calls); |
| 475 | if (ret) { |
| 476 | nr_calls--; |
| 477 | pr_warn("%s: attempt to bring up CPU %u failed\n", |
| 478 | __func__, cpu); |
| 479 | goto out_notify; |
| 480 | } |
| 481 | |
| 482 | /* Arch-specific enabling code. */ |
| 483 | ret = __cpu_up(cpu, idle); |
| 484 | if (ret != 0) |
| 485 | goto out_notify; |
| 486 | BUG_ON(!cpu_online(cpu)); |
| 487 | |
| 488 | /* Wake the per cpu threads */ |
| 489 | smpboot_unpark_threads(cpu); |
| 490 | |
| 491 | /* Now call notifier in preparation. */ |
| 492 | cpu_notify(CPU_ONLINE | mod, hcpu); |
| 493 | |
| 494 | out_notify: |
| 495 | if (ret != 0) |
| 496 | __cpu_notify(CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL); |
| 497 | out: |
| 498 | cpu_hotplug_done(); |
| 499 | |
| 500 | return ret; |
| 501 | } |
| 502 | |
| 503 | int cpu_up(unsigned int cpu) |
| 504 | { |
| 505 | int err = 0; |
| 506 | |
| 507 | if (!cpu_possible(cpu)) { |
| 508 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
| 509 | cpu); |
| 510 | #if defined(CONFIG_IA64) |
| 511 | pr_err("please check additional_cpus= boot parameter\n"); |
| 512 | #endif |
| 513 | return -EINVAL; |
| 514 | } |
| 515 | |
| 516 | err = try_online_node(cpu_to_node(cpu)); |
| 517 | if (err) |
| 518 | return err; |
| 519 | |
| 520 | cpu_maps_update_begin(); |
| 521 | |
| 522 | if (cpu_hotplug_disabled) { |
| 523 | err = -EBUSY; |
| 524 | goto out; |
| 525 | } |
| 526 | |
| 527 | err = _cpu_up(cpu, 0); |
| 528 | |
| 529 | out: |
| 530 | cpu_maps_update_done(); |
| 531 | return err; |
| 532 | } |
| 533 | EXPORT_SYMBOL_GPL(cpu_up); |
| 534 | |
| 535 | #ifdef CONFIG_PM_SLEEP_SMP |
| 536 | static cpumask_var_t frozen_cpus; |
| 537 | |
| 538 | int disable_nonboot_cpus(void) |
| 539 | { |
| 540 | int cpu, first_cpu, error = 0; |
| 541 | |
| 542 | cpu_maps_update_begin(); |
| 543 | first_cpu = cpumask_first(cpu_online_mask); |
| 544 | /* |
| 545 | * We take down all of the non-boot CPUs in one shot to avoid races |
| 546 | * with the userspace trying to use the CPU hotplug at the same time |
| 547 | */ |
| 548 | cpumask_clear(frozen_cpus); |
| 549 | |
| 550 | pr_info("Disabling non-boot CPUs ...\n"); |
| 551 | for_each_online_cpu(cpu) { |
| 552 | if (cpu == first_cpu) |
| 553 | continue; |
| 554 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
| 555 | error = _cpu_down(cpu, 1); |
| 556 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
| 557 | if (!error) |
| 558 | cpumask_set_cpu(cpu, frozen_cpus); |
| 559 | else { |
| 560 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
| 561 | break; |
| 562 | } |
| 563 | } |
| 564 | |
| 565 | if (!error) { |
| 566 | BUG_ON(num_online_cpus() > 1); |
| 567 | /* Make sure the CPUs won't be enabled by someone else */ |
| 568 | cpu_hotplug_disabled = 1; |
| 569 | } else { |
| 570 | pr_err("Non-boot CPUs are not disabled\n"); |
| 571 | } |
| 572 | cpu_maps_update_done(); |
| 573 | return error; |
| 574 | } |
| 575 | |
| 576 | void __weak arch_enable_nonboot_cpus_begin(void) |
| 577 | { |
| 578 | } |
| 579 | |
| 580 | void __weak arch_enable_nonboot_cpus_end(void) |
| 581 | { |
| 582 | } |
| 583 | |
| 584 | void __ref enable_nonboot_cpus(void) |
| 585 | { |
| 586 | int cpu, error; |
| 587 | |
| 588 | /* Allow everyone to use the CPU hotplug again */ |
| 589 | cpu_maps_update_begin(); |
| 590 | cpu_hotplug_disabled = 0; |
| 591 | if (cpumask_empty(frozen_cpus)) |
| 592 | goto out; |
| 593 | |
| 594 | pr_info("Enabling non-boot CPUs ...\n"); |
| 595 | |
| 596 | arch_enable_nonboot_cpus_begin(); |
| 597 | |
| 598 | for_each_cpu(cpu, frozen_cpus) { |
| 599 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
| 600 | error = _cpu_up(cpu, 1); |
| 601 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
| 602 | if (!error) { |
| 603 | pr_info("CPU%d is up\n", cpu); |
| 604 | continue; |
| 605 | } |
| 606 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
| 607 | } |
| 608 | |
| 609 | arch_enable_nonboot_cpus_end(); |
| 610 | |
| 611 | cpumask_clear(frozen_cpus); |
| 612 | out: |
| 613 | cpu_maps_update_done(); |
| 614 | } |
| 615 | |
| 616 | static int __init alloc_frozen_cpus(void) |
| 617 | { |
| 618 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) |
| 619 | return -ENOMEM; |
| 620 | return 0; |
| 621 | } |
| 622 | core_initcall(alloc_frozen_cpus); |
| 623 | |
| 624 | /* |
| 625 | * When callbacks for CPU hotplug notifications are being executed, we must |
| 626 | * ensure that the state of the system with respect to the tasks being frozen |
| 627 | * or not, as reported by the notification, remains unchanged *throughout the |
| 628 | * duration* of the execution of the callbacks. |
| 629 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. |
| 630 | * |
| 631 | * This synchronization is implemented by mutually excluding regular CPU |
| 632 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ |
| 633 | * Hibernate notifications. |
| 634 | */ |
| 635 | static int |
| 636 | cpu_hotplug_pm_callback(struct notifier_block *nb, |
| 637 | unsigned long action, void *ptr) |
| 638 | { |
| 639 | switch (action) { |
| 640 | |
| 641 | case PM_SUSPEND_PREPARE: |
| 642 | case PM_HIBERNATION_PREPARE: |
| 643 | cpu_hotplug_disable(); |
| 644 | break; |
| 645 | |
| 646 | case PM_POST_SUSPEND: |
| 647 | case PM_POST_HIBERNATION: |
| 648 | cpu_hotplug_enable(); |
| 649 | break; |
| 650 | |
| 651 | default: |
| 652 | return NOTIFY_DONE; |
| 653 | } |
| 654 | |
| 655 | return NOTIFY_OK; |
| 656 | } |
| 657 | |
| 658 | |
| 659 | static int __init cpu_hotplug_pm_sync_init(void) |
| 660 | { |
| 661 | /* |
| 662 | * cpu_hotplug_pm_callback has higher priority than x86 |
| 663 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback |
| 664 | * to disable cpu hotplug to avoid cpu hotplug race. |
| 665 | */ |
| 666 | pm_notifier(cpu_hotplug_pm_callback, 0); |
| 667 | return 0; |
| 668 | } |
| 669 | core_initcall(cpu_hotplug_pm_sync_init); |
| 670 | |
| 671 | #endif /* CONFIG_PM_SLEEP_SMP */ |
| 672 | |
| 673 | /** |
| 674 | * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers |
| 675 | * @cpu: cpu that just started |
| 676 | * |
| 677 | * This function calls the cpu_chain notifiers with CPU_STARTING. |
| 678 | * It must be called by the arch code on the new cpu, before the new cpu |
| 679 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). |
| 680 | */ |
| 681 | void notify_cpu_starting(unsigned int cpu) |
| 682 | { |
| 683 | unsigned long val = CPU_STARTING; |
| 684 | |
| 685 | #ifdef CONFIG_PM_SLEEP_SMP |
| 686 | if (frozen_cpus != NULL && cpumask_test_cpu(cpu, frozen_cpus)) |
| 687 | val = CPU_STARTING_FROZEN; |
| 688 | #endif /* CONFIG_PM_SLEEP_SMP */ |
| 689 | cpu_notify(val, (void *)(long)cpu); |
| 690 | } |
| 691 | |
| 692 | #endif /* CONFIG_SMP */ |
| 693 | |
| 694 | /* |
| 695 | * cpu_bit_bitmap[] is a special, "compressed" data structure that |
| 696 | * represents all NR_CPUS bits binary values of 1<<nr. |
| 697 | * |
| 698 | * It is used by cpumask_of() to get a constant address to a CPU |
| 699 | * mask value that has a single bit set only. |
| 700 | */ |
| 701 | |
| 702 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
| 703 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
| 704 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
| 705 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) |
| 706 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) |
| 707 | |
| 708 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
| 709 | |
| 710 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), |
| 711 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), |
| 712 | #if BITS_PER_LONG > 32 |
| 713 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), |
| 714 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), |
| 715 | #endif |
| 716 | }; |
| 717 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
| 718 | |
| 719 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; |
| 720 | EXPORT_SYMBOL(cpu_all_bits); |
| 721 | |
| 722 | #ifdef CONFIG_INIT_ALL_POSSIBLE |
| 723 | static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly |
| 724 | = CPU_BITS_ALL; |
| 725 | #else |
| 726 | static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly; |
| 727 | #endif |
| 728 | const struct cpumask *const cpu_possible_mask = to_cpumask(cpu_possible_bits); |
| 729 | EXPORT_SYMBOL(cpu_possible_mask); |
| 730 | |
| 731 | static DECLARE_BITMAP(cpu_online_bits, CONFIG_NR_CPUS) __read_mostly; |
| 732 | const struct cpumask *const cpu_online_mask = to_cpumask(cpu_online_bits); |
| 733 | EXPORT_SYMBOL(cpu_online_mask); |
| 734 | |
| 735 | static DECLARE_BITMAP(cpu_present_bits, CONFIG_NR_CPUS) __read_mostly; |
| 736 | const struct cpumask *const cpu_present_mask = to_cpumask(cpu_present_bits); |
| 737 | EXPORT_SYMBOL(cpu_present_mask); |
| 738 | |
| 739 | static DECLARE_BITMAP(cpu_active_bits, CONFIG_NR_CPUS) __read_mostly; |
| 740 | const struct cpumask *const cpu_active_mask = to_cpumask(cpu_active_bits); |
| 741 | EXPORT_SYMBOL(cpu_active_mask); |
| 742 | |
| 743 | void set_cpu_possible(unsigned int cpu, bool possible) |
| 744 | { |
| 745 | if (possible) |
| 746 | cpumask_set_cpu(cpu, to_cpumask(cpu_possible_bits)); |
| 747 | else |
| 748 | cpumask_clear_cpu(cpu, to_cpumask(cpu_possible_bits)); |
| 749 | } |
| 750 | |
| 751 | void set_cpu_present(unsigned int cpu, bool present) |
| 752 | { |
| 753 | if (present) |
| 754 | cpumask_set_cpu(cpu, to_cpumask(cpu_present_bits)); |
| 755 | else |
| 756 | cpumask_clear_cpu(cpu, to_cpumask(cpu_present_bits)); |
| 757 | } |
| 758 | |
| 759 | void set_cpu_online(unsigned int cpu, bool online) |
| 760 | { |
| 761 | if (online) { |
| 762 | cpumask_set_cpu(cpu, to_cpumask(cpu_online_bits)); |
| 763 | cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits)); |
| 764 | } else { |
| 765 | cpumask_clear_cpu(cpu, to_cpumask(cpu_online_bits)); |
| 766 | } |
| 767 | } |
| 768 | |
| 769 | void set_cpu_active(unsigned int cpu, bool active) |
| 770 | { |
| 771 | if (active) |
| 772 | cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits)); |
| 773 | else |
| 774 | cpumask_clear_cpu(cpu, to_cpumask(cpu_active_bits)); |
| 775 | } |
| 776 | |
| 777 | void init_cpu_present(const struct cpumask *src) |
| 778 | { |
| 779 | cpumask_copy(to_cpumask(cpu_present_bits), src); |
| 780 | } |
| 781 | |
| 782 | void init_cpu_possible(const struct cpumask *src) |
| 783 | { |
| 784 | cpumask_copy(to_cpumask(cpu_possible_bits), src); |
| 785 | } |
| 786 | |
| 787 | void init_cpu_online(const struct cpumask *src) |
| 788 | { |
| 789 | cpumask_copy(to_cpumask(cpu_online_bits), src); |
| 790 | } |