kernel/cpu.c

   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 <linux/tick.h>
  24 #include <linux/irq.h>
  25 #include <trace/events/power.h>
  26
  27 #include "smpboot.h"
  28
  29 #ifdef CONFIG_SMP
  30 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
  31 static DEFINE_MUTEX(cpu_add_remove_lock);
  32 bool cpuhp_tasks_frozen;
  33 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
  34
  35 /*
  36  * The following two APIs (cpu_maps_update_begin/done) must be used when
  37  * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
  38  * The APIs cpu_notifier_register_begin/done() must be used to protect CPU
  39  * hotplug callback (un)registration performed using __register_cpu_notifier()
  40  * or __unregister_cpu_notifier().
  41  */
  42 void cpu_maps_update_begin(void)
  43 {
  44         mutex_lock(&cpu_add_remove_lock);
  45 }
  46 EXPORT_SYMBOL(cpu_notifier_register_begin);
  47
  48 void cpu_maps_update_done(void)
  49 {
  50         mutex_unlock(&cpu_add_remove_lock);
  51 }
  52 EXPORT_SYMBOL(cpu_notifier_register_done);
  53
  54 static RAW_NOTIFIER_HEAD(cpu_chain);
  55
  56 /* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
  57  * Should always be manipulated under cpu_add_remove_lock
  58  */
  59 static int cpu_hotplug_disabled;
  60
  61 #ifdef CONFIG_HOTPLUG_CPU
  62
  63 static struct {
  64         struct task_struct *active_writer;
  65         /* wait queue to wake up the active_writer */
  66         wait_queue_head_t wq;
  67         /* verifies that no writer will get active while readers are active */
  68         struct mutex lock;
  69         /*
  70          * Also blocks the new readers during
  71          * an ongoing cpu hotplug operation.
  72          */
  73         atomic_t refcount;
  74
  75 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  76         struct lockdep_map dep_map;
  77 #endif
  78 } cpu_hotplug = {
  79         .active_writer = NULL,
  80         .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq),
  81         .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock),
  82 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  83         .dep_map = {.name = "cpu_hotplug.lock" },
  84 #endif
  85 };
  86
  87 /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */
  88 #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map)
  89 #define cpuhp_lock_acquire_tryread() \
  90                                   lock_map_acquire_tryread(&cpu_hotplug.dep_map)
  91 #define cpuhp_lock_acquire()      lock_map_acquire(&cpu_hotplug.dep_map)
  92 #define cpuhp_lock_release()      lock_map_release(&cpu_hotplug.dep_map)
  93
  94
  95 void get_online_cpus(void)
  96 {
  97         might_sleep();
  98         if (cpu_hotplug.active_writer == current)
  99                 return;
 100         cpuhp_lock_acquire_read();
 101         mutex_lock(&cpu_hotplug.lock);
 102         atomic_inc(&cpu_hotplug.refcount);
 103         mutex_unlock(&cpu_hotplug.lock);
 104 }
 105 EXPORT_SYMBOL_GPL(get_online_cpus);
 106
 107 void put_online_cpus(void)
 108 {
 109         int refcount;
 110
 111         if (cpu_hotplug.active_writer == current)
 112                 return;
 113
 114         refcount = atomic_dec_return(&cpu_hotplug.refcount);
 115         if (WARN_ON(refcount < 0)) /* try to fix things up */
 116                 atomic_inc(&cpu_hotplug.refcount);
 117
 118         if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq))
 119                 wake_up(&cpu_hotplug.wq);
 120
 121         cpuhp_lock_release();
 122
 123 }
 124 EXPORT_SYMBOL_GPL(put_online_cpus);
 125
 126 /*
 127  * This ensures that the hotplug operation can begin only when the
 128  * refcount goes to zero.
 129  *
 130  * Note that during a cpu-hotplug operation, the new readers, if any,
 131  * will be blocked by the cpu_hotplug.lock
 132  *
 133  * Since cpu_hotplug_begin() is always called after invoking
 134  * cpu_maps_update_begin(), we can be sure that only one writer is active.
 135  *
 136  * Note that theoretically, there is a possibility of a livelock:
 137  * - Refcount goes to zero, last reader wakes up the sleeping
 138  *   writer.
 139  * - Last reader unlocks the cpu_hotplug.lock.
 140  * - A new reader arrives at this moment, bumps up the refcount.
 141  * - The writer acquires the cpu_hotplug.lock finds the refcount
 142  *   non zero and goes to sleep again.
 143  *
 144  * However, this is very difficult to achieve in practice since
 145  * get_online_cpus() not an api which is called all that often.
 146  *
 147  */
 148 void cpu_hotplug_begin(void)
 149 {
 150         DEFINE_WAIT(wait);
 151
 152         cpu_hotplug.active_writer = current;
 153         cpuhp_lock_acquire();
 154
 155         for (;;) {
 156                 mutex_lock(&cpu_hotplug.lock);
 157                 prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE);
 158                 if (likely(!atomic_read(&cpu_hotplug.refcount)))
 159                                 break;
 160                 mutex_unlock(&cpu_hotplug.lock);
 161                 schedule();
 162         }
 163         finish_wait(&cpu_hotplug.wq, &wait);
 164 }
 165
 166 void cpu_hotplug_done(void)
 167 {
 168         cpu_hotplug.active_writer = NULL;
 169         mutex_unlock(&cpu_hotplug.lock);
 170         cpuhp_lock_release();
 171 }
 172
 173 /*
 174  * Wait for currently running CPU hotplug operations to complete (if any) and
 175  * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
 176  * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
 177  * hotplug path before performing hotplug operations. So acquiring that lock
 178  * guarantees mutual exclusion from any currently running hotplug operations.
 179  */
 180 void cpu_hotplug_disable(void)
 181 {
 182         cpu_maps_update_begin();
 183         cpu_hotplug_disabled++;
 184         cpu_maps_update_done();
 185 }
 186 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
 187
 188 void cpu_hotplug_enable(void)
 189 {
 190         cpu_maps_update_begin();
 191         WARN_ON(--cpu_hotplug_disabled < 0);
 192         cpu_maps_update_done();
 193 }
 194 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
 195 #endif  /* CONFIG_HOTPLUG_CPU */
 196
 197 /* Need to know about CPUs going up/down? */
 198 int register_cpu_notifier(struct notifier_block *nb)
 199 {
 200         int ret;
 201         cpu_maps_update_begin();
 202         ret = raw_notifier_chain_register(&cpu_chain, nb);
 203         cpu_maps_update_done();
 204         return ret;
 205 }
 206
 207 int __register_cpu_notifier(struct notifier_block *nb)
 208 {
 209         return raw_notifier_chain_register(&cpu_chain, nb);
 210 }
 211
 212 static int __cpu_notify(unsigned long val, unsigned int cpu, int nr_to_call,
 213                         int *nr_calls)
 214 {
 215         unsigned long mod = cpuhp_tasks_frozen ? CPU_TASKS_FROZEN : 0;
 216         void *hcpu = (void *)(long)cpu;
 217
 218         int ret;
 219
 220         ret = __raw_notifier_call_chain(&cpu_chain, val | mod, hcpu, nr_to_call,
 221                                         nr_calls);
 222
 223         return notifier_to_errno(ret);
 224 }
 225
 226 static int cpu_notify(unsigned long val, unsigned int cpu)
 227 {
 228         return __cpu_notify(val, cpu, -1, NULL);
 229 }
 230
 231 /* Notifier wrappers for transitioning to state machine */
 232 static int notify_prepare(unsigned int cpu)
 233 {
 234         int nr_calls = 0;
 235         int ret;
 236
 237         ret = __cpu_notify(CPU_UP_PREPARE, cpu, -1, &nr_calls);
 238         if (ret) {
 239                 nr_calls--;
 240                 printk(KERN_WARNING "%s: attempt to bring up CPU %u failed\n",
 241                                 __func__, cpu);
 242                 __cpu_notify(CPU_UP_CANCELED, cpu, nr_calls, NULL);
 243         }
 244         return ret;
 245 }
 246
 247 static int notify_online(unsigned int cpu)
 248 {
 249         cpu_notify(CPU_ONLINE, cpu);
 250         return 0;
 251 }
 252
 253 static int bringup_cpu(unsigned int cpu)
 254 {
 255         struct task_struct *idle = idle_thread_get(cpu);
 256         int ret;
 257
 258         /* Arch-specific enabling code. */
 259         ret = __cpu_up(cpu, idle);
 260         if (ret) {
 261                 cpu_notify(CPU_UP_CANCELED, cpu);
 262                 return ret;
 263         }
 264         BUG_ON(!cpu_online(cpu));
 265         return 0;
 266 }
 267
 268 #ifdef CONFIG_HOTPLUG_CPU
 269 EXPORT_SYMBOL(register_cpu_notifier);
 270 EXPORT_SYMBOL(__register_cpu_notifier);
 271
 272 void unregister_cpu_notifier(struct notifier_block *nb)
 273 {
 274         cpu_maps_update_begin();
 275         raw_notifier_chain_unregister(&cpu_chain, nb);
 276         cpu_maps_update_done();
 277 }
 278 EXPORT_SYMBOL(unregister_cpu_notifier);
 279
 280 void __unregister_cpu_notifier(struct notifier_block *nb)
 281 {
 282         raw_notifier_chain_unregister(&cpu_chain, nb);
 283 }
 284 EXPORT_SYMBOL(__unregister_cpu_notifier);
 285
 286 /**
 287  * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
 288  * @cpu: a CPU id
 289  *
 290  * This function walks all processes, finds a valid mm struct for each one and
 291  * then clears a corresponding bit in mm's cpumask.  While this all sounds
 292  * trivial, there are various non-obvious corner cases, which this function
 293  * tries to solve in a safe manner.
 294  *
 295  * Also note that the function uses a somewhat relaxed locking scheme, so it may
 296  * be called only for an already offlined CPU.
 297  */
 298 void clear_tasks_mm_cpumask(int cpu)
 299 {
 300         struct task_struct *p;
 301
 302         /*
 303          * This function is called after the cpu is taken down and marked
 304          * offline, so its not like new tasks will ever get this cpu set in
 305          * their mm mask. -- Peter Zijlstra
 306          * Thus, we may use rcu_read_lock() here, instead of grabbing
 307          * full-fledged tasklist_lock.
 308          */
 309         WARN_ON(cpu_online(cpu));
 310         rcu_read_lock();
 311         for_each_process(p) {
 312                 struct task_struct *t;
 313
 314                 /*
 315                  * Main thread might exit, but other threads may still have
 316                  * a valid mm. Find one.
 317                  */
 318                 t = find_lock_task_mm(p);
 319                 if (!t)
 320                         continue;
 321                 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
 322                 task_unlock(t);
 323         }
 324         rcu_read_unlock();
 325 }
 326
 327 static inline void check_for_tasks(int dead_cpu)
 328 {
 329         struct task_struct *g, *p;
 330
 331         read_lock(&tasklist_lock);
 332         for_each_process_thread(g, p) {
 333                 if (!p->on_rq)
 334                         continue;
 335                 /*
 336                  * We do the check with unlocked task_rq(p)->lock.
 337                  * Order the reading to do not warn about a task,
 338                  * which was running on this cpu in the past, and
 339                  * it's just been woken on another cpu.
 340                  */
 341                 rmb();
 342                 if (task_cpu(p) != dead_cpu)
 343                         continue;
 344
 345                 pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n",
 346                         p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags);
 347         }
 348         read_unlock(&tasklist_lock);
 349 }
 350
 351 static void cpu_notify_nofail(unsigned long val, unsigned int cpu)
 352 {
 353         BUG_ON(cpu_notify(val, cpu));
 354 }
 355
 356 static int notify_down_prepare(unsigned int cpu)
 357 {
 358         int err, nr_calls = 0;
 359
 360         err = __cpu_notify(CPU_DOWN_PREPARE, cpu, -1, &nr_calls);
 361         if (err) {
 362                 nr_calls--;
 363                 __cpu_notify(CPU_DOWN_FAILED, cpu, nr_calls, NULL);
 364                 pr_warn("%s: attempt to take down CPU %u failed\n",
 365                                 __func__, cpu);
 366         }
 367         return err;
 368 }
 369
 370 /* Take this CPU down. */
 371 static int take_cpu_down(void *_param)
 372 {
 373         int err, cpu = smp_processor_id();
 374
 375         /* Ensure this CPU doesn't handle any more interrupts. */
 376         err = __cpu_disable();
 377         if (err < 0)
 378                 return err;
 379
 380         cpu_notify(CPU_DYING, cpu);
 381         /* Give up timekeeping duties */
 382         tick_handover_do_timer();
 383         /* Park the stopper thread */
 384         stop_machine_park(cpu);
 385         return 0;
 386 }
 387
 388 static int takedown_cpu(unsigned int cpu)
 389 {
 390         int err;
 391
 392         /*
 393          * By now we've cleared cpu_active_mask, wait for all preempt-disabled
 394          * and RCU users of this state to go away such that all new such users
 395          * will observe it.
 396          *
 397          * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might
 398          * not imply sync_sched(), so wait for both.
 399          *
 400          * Do sync before park smpboot threads to take care the rcu boost case.
 401          */
 402         if (IS_ENABLED(CONFIG_PREEMPT))
 403                 synchronize_rcu_mult(call_rcu, call_rcu_sched);
 404         else
 405                 synchronize_rcu();
 406
 407         smpboot_park_threads(cpu);
 408
 409         /*
 410          * Prevent irq alloc/free while the dying cpu reorganizes the
 411          * interrupt affinities.
 412          */
 413         irq_lock_sparse();
 414
 415         /*
 416          * So now all preempt/rcu users must observe !cpu_active().
 417          */
 418         err = stop_machine(take_cpu_down, NULL, cpumask_of(cpu));
 419         if (err) {
 420                 /* CPU didn't die: tell everyone.  Can't complain. */
 421                 cpu_notify_nofail(CPU_DOWN_FAILED, cpu);
 422                 irq_unlock_sparse();
 423                 return err;
 424         }
 425         BUG_ON(cpu_online(cpu));
 426
 427         /*
 428          * The migration_call() CPU_DYING callback will have removed all
 429          * runnable tasks from the cpu, there's only the idle task left now
 430          * that the migration thread is done doing the stop_machine thing.
 431          *
 432          * Wait for the stop thread to go away.
 433          */
 434         while (!per_cpu(cpu_dead_idle, cpu))
 435                 cpu_relax();
 436         smp_mb(); /* Read from cpu_dead_idle before __cpu_die(). */
 437         per_cpu(cpu_dead_idle, cpu) = false;
 438
 439         /* Interrupts are moved away from the dying cpu, reenable alloc/free */
 440         irq_unlock_sparse();
 441
 442         hotplug_cpu__broadcast_tick_pull(cpu);
 443         /* This actually kills the CPU. */
 444         __cpu_die(cpu);
 445
 446         tick_cleanup_dead_cpu(cpu);
 447         return 0;
 448 }
 449
 450 static int notify_dead(unsigned int cpu)
 451 {
 452         cpu_notify_nofail(CPU_DEAD, cpu);
 453         check_for_tasks(cpu);
 454         return 0;
 455 }
 456
 457 /* Requires cpu_add_remove_lock to be held */
 458 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
 459 {
 460         int err;
 461
 462         if (num_online_cpus() == 1)
 463                 return -EBUSY;
 464
 465         if (!cpu_online(cpu))
 466                 return -EINVAL;
 467
 468         cpu_hotplug_begin();
 469
 470         cpuhp_tasks_frozen = tasks_frozen;
 471
 472         err = notify_down_prepare(cpu);
 473         if (err)
 474                 goto out_release;
 475         err = takedown_cpu(cpu);
 476         if (err)
 477                 goto out_release;
 478
 479         notify_dead(cpu);
 480
 481 out_release:
 482         cpu_hotplug_done();
 483         if (!err)
 484                 cpu_notify_nofail(CPU_POST_DEAD, cpu);
 485         return err;
 486 }
 487
 488 int cpu_down(unsigned int cpu)
 489 {
 490         int err;
 491
 492         cpu_maps_update_begin();
 493
 494         if (cpu_hotplug_disabled) {
 495                 err = -EBUSY;
 496                 goto out;
 497         }
 498
 499         err = _cpu_down(cpu, 0);
 500
 501 out:
 502         cpu_maps_update_done();
 503         return err;
 504 }
 505 EXPORT_SYMBOL(cpu_down);
 506 #endif /*CONFIG_HOTPLUG_CPU*/
 507
 508 /*
 509  * Unpark per-CPU smpboot kthreads at CPU-online time.
 510  */
 511 static int smpboot_thread_call(struct notifier_block *nfb,
 512                                unsigned long action, void *hcpu)
 513 {
 514         int cpu = (long)hcpu;
 515
 516         switch (action & ~CPU_TASKS_FROZEN) {
 517
 518         case CPU_DOWN_FAILED:
 519         case CPU_ONLINE:
 520                 smpboot_unpark_threads(cpu);
 521                 break;
 522
 523         default:
 524                 break;
 525         }
 526
 527         return NOTIFY_OK;
 528 }
 529
 530 static struct notifier_block smpboot_thread_notifier = {
 531         .notifier_call = smpboot_thread_call,
 532         .priority = CPU_PRI_SMPBOOT,
 533 };
 534
 535 void smpboot_thread_init(void)
 536 {
 537         register_cpu_notifier(&smpboot_thread_notifier);
 538 }
 539
 540 /* Requires cpu_add_remove_lock to be held */
 541 static int _cpu_up(unsigned int cpu, int tasks_frozen)
 542 {
 543         struct task_struct *idle;
 544         int ret;
 545
 546         cpu_hotplug_begin();
 547
 548         if (cpu_online(cpu) || !cpu_present(cpu)) {
 549                 ret = -EINVAL;
 550                 goto out;
 551         }
 552
 553         idle = idle_thread_get(cpu);
 554         if (IS_ERR(idle)) {
 555                 ret = PTR_ERR(idle);
 556                 goto out;
 557         }
 558
 559         cpuhp_tasks_frozen = tasks_frozen;
 560
 561         ret = smpboot_create_threads(cpu);
 562         if (ret)
 563                 goto out;
 564
 565         ret = notify_prepare(cpu);
 566         if (ret)
 567                 goto out;
 568
 569         ret = bringup_cpu(cpu);
 570         if (ret)
 571                 goto out;
 572
 573         notify_online(cpu);
 574 out:
 575         cpu_hotplug_done();
 576
 577         return ret;
 578 }
 579
 580 int cpu_up(unsigned int cpu)
 581 {
 582         int err = 0;
 583
 584         if (!cpu_possible(cpu)) {
 585                 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
 586                        cpu);
 587 #if defined(CONFIG_IA64)
 588                 pr_err("please check additional_cpus= boot parameter\n");
 589 #endif
 590                 return -EINVAL;
 591         }
 592
 593         err = try_online_node(cpu_to_node(cpu));
 594         if (err)
 595                 return err;
 596
 597         cpu_maps_update_begin();
 598
 599         if (cpu_hotplug_disabled) {
 600                 err = -EBUSY;
 601                 goto out;
 602         }
 603
 604         err = _cpu_up(cpu, 0);
 605
 606 out:
 607         cpu_maps_update_done();
 608         return err;
 609 }
 610 EXPORT_SYMBOL_GPL(cpu_up);
 611
 612 #ifdef CONFIG_PM_SLEEP_SMP
 613 static cpumask_var_t frozen_cpus;
 614
 615 int disable_nonboot_cpus(void)
 616 {
 617         int cpu, first_cpu, error = 0;
 618
 619         cpu_maps_update_begin();
 620         first_cpu = cpumask_first(cpu_online_mask);
 621         /*
 622          * We take down all of the non-boot CPUs in one shot to avoid races
 623          * with the userspace trying to use the CPU hotplug at the same time
 624          */
 625         cpumask_clear(frozen_cpus);
 626
 627         pr_info("Disabling non-boot CPUs ...\n");
 628         for_each_online_cpu(cpu) {
 629                 if (cpu == first_cpu)
 630                         continue;
 631                 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
 632                 error = _cpu_down(cpu, 1);
 633                 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
 634                 if (!error)
 635                         cpumask_set_cpu(cpu, frozen_cpus);
 636                 else {
 637                         pr_err("Error taking CPU%d down: %d\n", cpu, error);
 638                         break;
 639                 }
 640         }
 641
 642         if (!error)
 643                 BUG_ON(num_online_cpus() > 1);
 644         else
 645                 pr_err("Non-boot CPUs are not disabled\n");
 646
 647         /*
 648          * Make sure the CPUs won't be enabled by someone else. We need to do
 649          * this even in case of failure as all disable_nonboot_cpus() users are
 650          * supposed to do enable_nonboot_cpus() on the failure path.
 651          */
 652         cpu_hotplug_disabled++;
 653
 654         cpu_maps_update_done();
 655         return error;
 656 }
 657
 658 void __weak arch_enable_nonboot_cpus_begin(void)
 659 {
 660 }
 661
 662 void __weak arch_enable_nonboot_cpus_end(void)
 663 {
 664 }
 665
 666 void enable_nonboot_cpus(void)
 667 {
 668         int cpu, error;
 669
 670         /* Allow everyone to use the CPU hotplug again */
 671         cpu_maps_update_begin();
 672         WARN_ON(--cpu_hotplug_disabled < 0);
 673         if (cpumask_empty(frozen_cpus))
 674                 goto out;
 675
 676         pr_info("Enabling non-boot CPUs ...\n");
 677
 678         arch_enable_nonboot_cpus_begin();
 679
 680         for_each_cpu(cpu, frozen_cpus) {
 681                 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
 682                 error = _cpu_up(cpu, 1);
 683                 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
 684                 if (!error) {
 685                         pr_info("CPU%d is up\n", cpu);
 686                         continue;
 687                 }
 688                 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
 689         }
 690
 691         arch_enable_nonboot_cpus_end();
 692
 693         cpumask_clear(frozen_cpus);
 694 out:
 695         cpu_maps_update_done();
 696 }
 697
 698 static int __init alloc_frozen_cpus(void)
 699 {
 700         if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
 701                 return -ENOMEM;
 702         return 0;
 703 }
 704 core_initcall(alloc_frozen_cpus);
 705
 706 /*
 707  * When callbacks for CPU hotplug notifications are being executed, we must
 708  * ensure that the state of the system with respect to the tasks being frozen
 709  * or not, as reported by the notification, remains unchanged *throughout the
 710  * duration* of the execution of the callbacks.
 711  * Hence we need to prevent the freezer from racing with regular CPU hotplug.
 712  *
 713  * This synchronization is implemented by mutually excluding regular CPU
 714  * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
 715  * Hibernate notifications.
 716  */
 717 static int
 718 cpu_hotplug_pm_callback(struct notifier_block *nb,
 719                         unsigned long action, void *ptr)
 720 {
 721         switch (action) {
 722
 723         case PM_SUSPEND_PREPARE:
 724         case PM_HIBERNATION_PREPARE:
 725                 cpu_hotplug_disable();
 726                 break;
 727
 728         case PM_POST_SUSPEND:
 729         case PM_POST_HIBERNATION:
 730                 cpu_hotplug_enable();
 731                 break;
 732
 733         default:
 734                 return NOTIFY_DONE;
 735         }
 736
 737         return NOTIFY_OK;
 738 }
 739
 740
 741 static int __init cpu_hotplug_pm_sync_init(void)
 742 {
 743         /*
 744          * cpu_hotplug_pm_callback has higher priority than x86
 745          * bsp_pm_callback which depends on cpu_hotplug_pm_callback
 746          * to disable cpu hotplug to avoid cpu hotplug race.
 747          */
 748         pm_notifier(cpu_hotplug_pm_callback, 0);
 749         return 0;
 750 }
 751 core_initcall(cpu_hotplug_pm_sync_init);
 752
 753 #endif /* CONFIG_PM_SLEEP_SMP */
 754
 755 /**
 756  * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers
 757  * @cpu: cpu that just started
 758  *
 759  * This function calls the cpu_chain notifiers with CPU_STARTING.
 760  * It must be called by the arch code on the new cpu, before the new cpu
 761  * enables interrupts and before the "boot" cpu returns from __cpu_up().
 762  */
 763 void notify_cpu_starting(unsigned int cpu)
 764 {
 765         cpu_notify(CPU_STARTING, cpu);
 766 }
 767
 768 #endif /* CONFIG_SMP */
 769
 770 /*
 771  * cpu_bit_bitmap[] is a special, "compressed" data structure that
 772  * represents all NR_CPUS bits binary values of 1<<nr.
 773  *
 774  * It is used by cpumask_of() to get a constant address to a CPU
 775  * mask value that has a single bit set only.
 776  */
 777
 778 /* cpu_bit_bitmap[0] is empty - so we can back into it */
 779 #define MASK_DECLARE_1(x)       [x+1][0] = (1UL << (x))
 780 #define MASK_DECLARE_2(x)       MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
 781 #define MASK_DECLARE_4(x)       MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
 782 #define MASK_DECLARE_8(x)       MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
 783
 784 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
 785
 786         MASK_DECLARE_8(0),      MASK_DECLARE_8(8),
 787         MASK_DECLARE_8(16),     MASK_DECLARE_8(24),
 788 #if BITS_PER_LONG > 32
 789         MASK_DECLARE_8(32),     MASK_DECLARE_8(40),
 790         MASK_DECLARE_8(48),     MASK_DECLARE_8(56),
 791 #endif
 792 };
 793 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
 794
 795 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
 796 EXPORT_SYMBOL(cpu_all_bits);
 797
 798 #ifdef CONFIG_INIT_ALL_POSSIBLE
 799 struct cpumask __cpu_possible_mask __read_mostly
 800         = {CPU_BITS_ALL};
 801 #else
 802 struct cpumask __cpu_possible_mask __read_mostly;
 803 #endif
 804 EXPORT_SYMBOL(__cpu_possible_mask);
 805
 806 struct cpumask __cpu_online_mask __read_mostly;
 807 EXPORT_SYMBOL(__cpu_online_mask);
 808
 809 struct cpumask __cpu_present_mask __read_mostly;
 810 EXPORT_SYMBOL(__cpu_present_mask);
 811
 812 struct cpumask __cpu_active_mask __read_mostly;
 813 EXPORT_SYMBOL(__cpu_active_mask);
 814
 815 void init_cpu_present(const struct cpumask *src)
 816 {
 817         cpumask_copy(&__cpu_present_mask, src);
 818 }
 819
 820 void init_cpu_possible(const struct cpumask *src)
 821 {
 822         cpumask_copy(&__cpu_possible_mask, src);
 823 }
 824
 825 void init_cpu_online(const struct cpumask *src)
 826 {
 827         cpumask_copy(&__cpu_online_mask, src);
 828 }