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
  33 /*
  34  * The following two APIs (cpu_maps_update_begin/done) must be used when
  35  * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
  36  * The APIs cpu_notifier_register_begin/done() must be used to protect CPU
  37  * hotplug callback (un)registration performed using __register_cpu_notifier()
  38  * or __unregister_cpu_notifier().
  39  */
  40 void cpu_maps_update_begin(void)
  41 {
  42         mutex_lock(&cpu_add_remove_lock);
  43 }
  44 EXPORT_SYMBOL(cpu_notifier_register_begin);
  45
  46 void cpu_maps_update_done(void)
  47 {
  48         mutex_unlock(&cpu_add_remove_lock);
  49 }
  50 EXPORT_SYMBOL(cpu_notifier_register_done);
  51
  52 static RAW_NOTIFIER_HEAD(cpu_chain);
  53
  54 /* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
  55  * Should always be manipulated under cpu_add_remove_lock
  56  */
  57 static int cpu_hotplug_disabled;
  58
  59 #ifdef CONFIG_HOTPLUG_CPU
  60
  61 static struct {
  62         struct task_struct *active_writer;
  63         /* wait queue to wake up the active_writer */
  64         wait_queue_head_t wq;
  65         /* verifies that no writer will get active while readers are active */
  66         struct mutex lock;
  67         /*
  68          * Also blocks the new readers during
  69          * an ongoing cpu hotplug operation.
  70          */
  71         atomic_t refcount;
  72
  73 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  74         struct lockdep_map dep_map;
  75 #endif
  76 } cpu_hotplug = {
  77         .active_writer = NULL,
  78         .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq),
  79         .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock),
  80 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  81         .dep_map = {.name = "cpu_hotplug.lock" },
  82 #endif
  83 };
  84
  85 /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */
  86 #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map)
  87 #define cpuhp_lock_acquire_tryread() \
  88                                   lock_map_acquire_tryread(&cpu_hotplug.dep_map)
  89 #define cpuhp_lock_acquire()      lock_map_acquire(&cpu_hotplug.dep_map)
  90 #define cpuhp_lock_release()      lock_map_release(&cpu_hotplug.dep_map)
  91
  92
  93 void get_online_cpus(void)
  94 {
  95         might_sleep();
  96         if (cpu_hotplug.active_writer == current)
  97                 return;
  98         cpuhp_lock_acquire_read();
  99         mutex_lock(&cpu_hotplug.lock);
 100         atomic_inc(&cpu_hotplug.refcount);
 101         mutex_unlock(&cpu_hotplug.lock);
 102 }
 103 EXPORT_SYMBOL_GPL(get_online_cpus);
 104
 105 bool try_get_online_cpus(void)
 106 {
 107         if (cpu_hotplug.active_writer == current)
 108                 return true;
 109         if (!mutex_trylock(&cpu_hotplug.lock))
 110                 return false;
 111         cpuhp_lock_acquire_tryread();
 112         atomic_inc(&cpu_hotplug.refcount);
 113         mutex_unlock(&cpu_hotplug.lock);
 114         return true;
 115 }
 116 EXPORT_SYMBOL_GPL(try_get_online_cpus);
 117
 118 void put_online_cpus(void)
 119 {
 120         int refcount;
 121
 122         if (cpu_hotplug.active_writer == current)
 123                 return;
 124
 125         refcount = atomic_dec_return(&cpu_hotplug.refcount);
 126         if (WARN_ON(refcount < 0)) /* try to fix things up */
 127                 atomic_inc(&cpu_hotplug.refcount);
 128
 129         if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq))
 130                 wake_up(&cpu_hotplug.wq);
 131
 132         cpuhp_lock_release();
 133
 134 }
 135 EXPORT_SYMBOL_GPL(put_online_cpus);
 136
 137 /*
 138  * This ensures that the hotplug operation can begin only when the
 139  * refcount goes to zero.
 140  *
 141  * Note that during a cpu-hotplug operation, the new readers, if any,
 142  * will be blocked by the cpu_hotplug.lock
 143  *
 144  * Since cpu_hotplug_begin() is always called after invoking
 145  * cpu_maps_update_begin(), we can be sure that only one writer is active.
 146  *
 147  * Note that theoretically, there is a possibility of a livelock:
 148  * - Refcount goes to zero, last reader wakes up the sleeping
 149  *   writer.
 150  * - Last reader unlocks the cpu_hotplug.lock.
 151  * - A new reader arrives at this moment, bumps up the refcount.
 152  * - The writer acquires the cpu_hotplug.lock finds the refcount
 153  *   non zero and goes to sleep again.
 154  *
 155  * However, this is very difficult to achieve in practice since
 156  * get_online_cpus() not an api which is called all that often.
 157  *
 158  */
 159 void cpu_hotplug_begin(void)
 160 {
 161         DEFINE_WAIT(wait);
 162
 163         cpu_hotplug.active_writer = current;
 164         cpuhp_lock_acquire();
 165
 166         for (;;) {
 167                 mutex_lock(&cpu_hotplug.lock);
 168                 prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE);
 169                 if (likely(!atomic_read(&cpu_hotplug.refcount)))
 170                                 break;
 171                 mutex_unlock(&cpu_hotplug.lock);
 172                 schedule();
 173         }
 174         finish_wait(&cpu_hotplug.wq, &wait);
 175 }
 176
 177 void cpu_hotplug_done(void)
 178 {
 179         cpu_hotplug.active_writer = NULL;
 180         mutex_unlock(&cpu_hotplug.lock);
 181         cpuhp_lock_release();
 182 }
 183
 184 /*
 185  * Wait for currently running CPU hotplug operations to complete (if any) and
 186  * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
 187  * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
 188  * hotplug path before performing hotplug operations. So acquiring that lock
 189  * guarantees mutual exclusion from any currently running hotplug operations.
 190  */
 191 void cpu_hotplug_disable(void)
 192 {
 193         cpu_maps_update_begin();
 194         cpu_hotplug_disabled++;
 195         cpu_maps_update_done();
 196 }
 197 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
 198
 199 void cpu_hotplug_enable(void)
 200 {
 201         cpu_maps_update_begin();
 202         WARN_ON(--cpu_hotplug_disabled < 0);
 203         cpu_maps_update_done();
 204 }
 205 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
 206 #endif  /* CONFIG_HOTPLUG_CPU */
 207
 208 /* Need to know about CPUs going up/down? */
 209 int register_cpu_notifier(struct notifier_block *nb)
 210 {
 211         int ret;
 212         cpu_maps_update_begin();
 213         ret = raw_notifier_chain_register(&cpu_chain, nb);
 214         cpu_maps_update_done();
 215         return ret;
 216 }
 217
 218 int __register_cpu_notifier(struct notifier_block *nb)
 219 {
 220         return raw_notifier_chain_register(&cpu_chain, nb);
 221 }
 222
 223 static int __cpu_notify(unsigned long val, void *v, int nr_to_call,
 224                         int *nr_calls)
 225 {
 226         int ret;
 227
 228         ret = __raw_notifier_call_chain(&cpu_chain, val, v, nr_to_call,
 229                                         nr_calls);
 230
 231         return notifier_to_errno(ret);
 232 }
 233
 234 static int cpu_notify(unsigned long val, void *v)
 235 {
 236         return __cpu_notify(val, v, -1, NULL);
 237 }
 238
 239 #ifdef CONFIG_HOTPLUG_CPU
 240
 241 static void cpu_notify_nofail(unsigned long val, void *v)
 242 {
 243         BUG_ON(cpu_notify(val, v));
 244 }
 245 EXPORT_SYMBOL(register_cpu_notifier);
 246 EXPORT_SYMBOL(__register_cpu_notifier);
 247
 248 void unregister_cpu_notifier(struct notifier_block *nb)
 249 {
 250         cpu_maps_update_begin();
 251         raw_notifier_chain_unregister(&cpu_chain, nb);
 252         cpu_maps_update_done();
 253 }
 254 EXPORT_SYMBOL(unregister_cpu_notifier);
 255
 256 void __unregister_cpu_notifier(struct notifier_block *nb)
 257 {
 258         raw_notifier_chain_unregister(&cpu_chain, nb);
 259 }
 260 EXPORT_SYMBOL(__unregister_cpu_notifier);
 261
 262 /**
 263  * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
 264  * @cpu: a CPU id
 265  *
 266  * This function walks all processes, finds a valid mm struct for each one and
 267  * then clears a corresponding bit in mm's cpumask.  While this all sounds
 268  * trivial, there are various non-obvious corner cases, which this function
 269  * tries to solve in a safe manner.
 270  *
 271  * Also note that the function uses a somewhat relaxed locking scheme, so it may
 272  * be called only for an already offlined CPU.
 273  */
 274 void clear_tasks_mm_cpumask(int cpu)
 275 {
 276         struct task_struct *p;
 277
 278         /*
 279          * This function is called after the cpu is taken down and marked
 280          * offline, so its not like new tasks will ever get this cpu set in
 281          * their mm mask. -- Peter Zijlstra
 282          * Thus, we may use rcu_read_lock() here, instead of grabbing
 283          * full-fledged tasklist_lock.
 284          */
 285         WARN_ON(cpu_online(cpu));
 286         rcu_read_lock();
 287         for_each_process(p) {
 288                 struct task_struct *t;
 289
 290                 /*
 291                  * Main thread might exit, but other threads may still have
 292                  * a valid mm. Find one.
 293                  */
 294                 t = find_lock_task_mm(p);
 295                 if (!t)
 296                         continue;
 297                 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
 298                 task_unlock(t);
 299         }
 300         rcu_read_unlock();
 301 }
 302
 303 static inline void check_for_tasks(int dead_cpu)
 304 {
 305         struct task_struct *g, *p;
 306
 307         read_lock(&tasklist_lock);
 308         for_each_process_thread(g, p) {
 309                 if (!p->on_rq)
 310                         continue;
 311                 /*
 312                  * We do the check with unlocked task_rq(p)->lock.
 313                  * Order the reading to do not warn about a task,
 314                  * which was running on this cpu in the past, and
 315                  * it's just been woken on another cpu.
 316                  */
 317                 rmb();
 318                 if (task_cpu(p) != dead_cpu)
 319                         continue;
 320
 321                 pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n",
 322                         p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags);
 323         }
 324         read_unlock(&tasklist_lock);
 325 }
 326
 327 struct take_cpu_down_param {
 328         unsigned long mod;
 329         void *hcpu;
 330 };
 331
 332 /* Take this CPU down. */
 333 static int take_cpu_down(void *_param)
 334 {
 335         struct take_cpu_down_param *param = _param;
 336         int err;
 337
 338         /* Ensure this CPU doesn't handle any more interrupts. */
 339         err = __cpu_disable();
 340         if (err < 0)
 341                 return err;
 342
 343         cpu_notify(CPU_DYING | param->mod, param->hcpu);
 344         /* Give up timekeeping duties */
 345         tick_handover_do_timer();
 346         /* Park the stopper thread */
 347         stop_machine_park((long)param->hcpu);
 348         return 0;
 349 }
 350
 351 /* Requires cpu_add_remove_lock to be held */
 352 static int _cpu_down(unsigned int cpu, int tasks_frozen)
 353 {
 354         int err, nr_calls = 0;
 355         void *hcpu = (void *)(long)cpu;
 356         unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
 357         struct take_cpu_down_param tcd_param = {
 358                 .mod = mod,
 359                 .hcpu = hcpu,
 360         };
 361
 362         if (num_online_cpus() == 1)
 363                 return -EBUSY;
 364
 365         if (!cpu_online(cpu))
 366                 return -EINVAL;
 367
 368         cpu_hotplug_begin();
 369
 370         err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls);
 371         if (err) {
 372                 nr_calls--;
 373                 __cpu_notify(CPU_DOWN_FAILED | mod, hcpu, nr_calls, NULL);
 374                 pr_warn("%s: attempt to take down CPU %u failed\n",
 375                         __func__, cpu);
 376                 goto out_release;
 377         }
 378
 379         /*
 380          * By now we've cleared cpu_active_mask, wait for all preempt-disabled
 381          * and RCU users of this state to go away such that all new such users
 382          * will observe it.
 383          *
 384          * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might
 385          * not imply sync_sched(), so wait for both.
 386          *
 387          * Do sync before park smpboot threads to take care the rcu boost case.
 388          */
 389         if (IS_ENABLED(CONFIG_PREEMPT))
 390                 synchronize_rcu_mult(call_rcu, call_rcu_sched);
 391         else
 392                 synchronize_rcu();
 393
 394         smpboot_park_threads(cpu);
 395
 396         /*
 397          * Prevent irq alloc/free while the dying cpu reorganizes the
 398          * interrupt affinities.
 399          */
 400         irq_lock_sparse();
 401
 402         /*
 403          * So now all preempt/rcu users must observe !cpu_active().
 404          */
 405         err = stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu));
 406         if (err) {
 407                 /* CPU didn't die: tell everyone.  Can't complain. */
 408                 cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu);
 409                 irq_unlock_sparse();
 410                 goto out_release;
 411         }
 412         BUG_ON(cpu_online(cpu));
 413
 414         /*
 415          * The migration_call() CPU_DYING callback will have removed all
 416          * runnable tasks from the cpu, there's only the idle task left now
 417          * that the migration thread is done doing the stop_machine thing.
 418          *
 419          * Wait for the stop thread to go away.
 420          */
 421         while (!per_cpu(cpu_dead_idle, cpu))
 422                 cpu_relax();
 423         smp_mb(); /* Read from cpu_dead_idle before __cpu_die(). */
 424         per_cpu(cpu_dead_idle, cpu) = false;
 425
 426         /* Interrupts are moved away from the dying cpu, reenable alloc/free */
 427         irq_unlock_sparse();
 428
 429         hotplug_cpu__broadcast_tick_pull(cpu);
 430         /* This actually kills the CPU. */
 431         __cpu_die(cpu);
 432
 433         /* CPU is completely dead: tell everyone.  Too late to complain. */
 434         tick_cleanup_dead_cpu(cpu);
 435         cpu_notify_nofail(CPU_DEAD | mod, hcpu);
 436
 437         check_for_tasks(cpu);
 438
 439 out_release:
 440         cpu_hotplug_done();
 441         if (!err)
 442                 cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu);
 443         return err;
 444 }
 445
 446 int cpu_down(unsigned int cpu)
 447 {
 448         int err;
 449
 450         cpu_maps_update_begin();
 451
 452         if (cpu_hotplug_disabled) {
 453                 err = -EBUSY;
 454                 goto out;
 455         }
 456
 457         err = _cpu_down(cpu, 0);
 458
 459 out:
 460         cpu_maps_update_done();
 461         return err;
 462 }
 463 EXPORT_SYMBOL(cpu_down);
 464 #endif /*CONFIG_HOTPLUG_CPU*/
 465
 466 /*
 467  * Unpark per-CPU smpboot kthreads at CPU-online time.
 468  */
 469 static int smpboot_thread_call(struct notifier_block *nfb,
 470                                unsigned long action, void *hcpu)
 471 {
 472         int cpu = (long)hcpu;
 473
 474         switch (action & ~CPU_TASKS_FROZEN) {
 475
 476         case CPU_DOWN_FAILED:
 477         case CPU_ONLINE:
 478                 stop_machine_unpark(cpu);
 479                 smpboot_unpark_threads(cpu);
 480                 break;
 481
 482         default:
 483                 break;
 484         }
 485
 486         return NOTIFY_OK;
 487 }
 488
 489 static struct notifier_block smpboot_thread_notifier = {
 490         .notifier_call = smpboot_thread_call,
 491         .priority = CPU_PRI_SMPBOOT,
 492 };
 493
 494 void smpboot_thread_init(void)
 495 {
 496         register_cpu_notifier(&smpboot_thread_notifier);
 497 }
 498
 499 /* Requires cpu_add_remove_lock to be held */
 500 static int _cpu_up(unsigned int cpu, int tasks_frozen)
 501 {
 502         int ret, nr_calls = 0;
 503         void *hcpu = (void *)(long)cpu;
 504         unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
 505         struct task_struct *idle;
 506
 507         cpu_hotplug_begin();
 508
 509         if (cpu_online(cpu) || !cpu_present(cpu)) {
 510                 ret = -EINVAL;
 511                 goto out;
 512         }
 513
 514         idle = idle_thread_get(cpu);
 515         if (IS_ERR(idle)) {
 516                 ret = PTR_ERR(idle);
 517                 goto out;
 518         }
 519
 520         ret = smpboot_create_threads(cpu);
 521         if (ret)
 522                 goto out;
 523
 524         ret = __cpu_notify(CPU_UP_PREPARE | mod, hcpu, -1, &nr_calls);
 525         if (ret) {
 526                 nr_calls--;
 527                 pr_warn("%s: attempt to bring up CPU %u failed\n",
 528                         __func__, cpu);
 529                 goto out_notify;
 530         }
 531
 532         /* Arch-specific enabling code. */
 533         ret = __cpu_up(cpu, idle);
 534
 535         if (ret != 0)
 536                 goto out_notify;
 537         BUG_ON(!cpu_online(cpu));
 538
 539         /* Now call notifier in preparation. */
 540         cpu_notify(CPU_ONLINE | mod, hcpu);
 541
 542 out_notify:
 543         if (ret != 0)
 544                 __cpu_notify(CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL);
 545 out:
 546         cpu_hotplug_done();
 547
 548         return ret;
 549 }
 550
 551 int cpu_up(unsigned int cpu)
 552 {
 553         int err = 0;
 554
 555         if (!cpu_possible(cpu)) {
 556                 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
 557                        cpu);
 558 #if defined(CONFIG_IA64)
 559                 pr_err("please check additional_cpus= boot parameter\n");
 560 #endif
 561                 return -EINVAL;
 562         }
 563
 564         err = try_online_node(cpu_to_node(cpu));
 565         if (err)
 566                 return err;
 567
 568         cpu_maps_update_begin();
 569
 570         if (cpu_hotplug_disabled) {
 571                 err = -EBUSY;
 572                 goto out;
 573         }
 574
 575         err = _cpu_up(cpu, 0);
 576
 577 out:
 578         cpu_maps_update_done();
 579         return err;
 580 }
 581 EXPORT_SYMBOL_GPL(cpu_up);
 582
 583 #ifdef CONFIG_PM_SLEEP_SMP
 584 static cpumask_var_t frozen_cpus;
 585
 586 int disable_nonboot_cpus(void)
 587 {
 588         int cpu, first_cpu, error = 0;
 589
 590         cpu_maps_update_begin();
 591         first_cpu = cpumask_first(cpu_online_mask);
 592         /*
 593          * We take down all of the non-boot CPUs in one shot to avoid races
 594          * with the userspace trying to use the CPU hotplug at the same time
 595          */
 596         cpumask_clear(frozen_cpus);
 597
 598         pr_info("Disabling non-boot CPUs ...\n");
 599         for_each_online_cpu(cpu) {
 600                 if (cpu == first_cpu)
 601                         continue;
 602                 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
 603                 error = _cpu_down(cpu, 1);
 604                 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
 605                 if (!error)
 606                         cpumask_set_cpu(cpu, frozen_cpus);
 607                 else {
 608                         pr_err("Error taking CPU%d down: %d\n", cpu, error);
 609                         break;
 610                 }
 611         }
 612
 613         if (!error)
 614                 BUG_ON(num_online_cpus() > 1);
 615         else
 616                 pr_err("Non-boot CPUs are not disabled\n");
 617
 618         /*
 619          * Make sure the CPUs won't be enabled by someone else. We need to do
 620          * this even in case of failure as all disable_nonboot_cpus() users are
 621          * supposed to do enable_nonboot_cpus() on the failure path.
 622          */
 623         cpu_hotplug_disabled++;
 624
 625         cpu_maps_update_done();
 626         return error;
 627 }
 628
 629 void __weak arch_enable_nonboot_cpus_begin(void)
 630 {
 631 }
 632
 633 void __weak arch_enable_nonboot_cpus_end(void)
 634 {
 635 }
 636
 637 void enable_nonboot_cpus(void)
 638 {
 639         int cpu, error;
 640
 641         /* Allow everyone to use the CPU hotplug again */
 642         cpu_maps_update_begin();
 643         WARN_ON(--cpu_hotplug_disabled < 0);
 644         if (cpumask_empty(frozen_cpus))
 645                 goto out;
 646
 647         pr_info("Enabling non-boot CPUs ...\n");
 648
 649         arch_enable_nonboot_cpus_begin();
 650
 651         for_each_cpu(cpu, frozen_cpus) {
 652                 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
 653                 error = _cpu_up(cpu, 1);
 654                 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
 655                 if (!error) {
 656                         pr_info("CPU%d is up\n", cpu);
 657                         continue;
 658                 }
 659                 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
 660         }
 661
 662         arch_enable_nonboot_cpus_end();
 663
 664         cpumask_clear(frozen_cpus);
 665 out:
 666         cpu_maps_update_done();
 667 }
 668
 669 static int __init alloc_frozen_cpus(void)
 670 {
 671         if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
 672                 return -ENOMEM;
 673         return 0;
 674 }
 675 core_initcall(alloc_frozen_cpus);
 676
 677 /*
 678  * When callbacks for CPU hotplug notifications are being executed, we must
 679  * ensure that the state of the system with respect to the tasks being frozen
 680  * or not, as reported by the notification, remains unchanged *throughout the
 681  * duration* of the execution of the callbacks.
 682  * Hence we need to prevent the freezer from racing with regular CPU hotplug.
 683  *
 684  * This synchronization is implemented by mutually excluding regular CPU
 685  * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
 686  * Hibernate notifications.
 687  */
 688 static int
 689 cpu_hotplug_pm_callback(struct notifier_block *nb,
 690                         unsigned long action, void *ptr)
 691 {
 692         switch (action) {
 693
 694         case PM_SUSPEND_PREPARE:
 695         case PM_HIBERNATION_PREPARE:
 696                 cpu_hotplug_disable();
 697                 break;
 698
 699         case PM_POST_SUSPEND:
 700         case PM_POST_HIBERNATION:
 701                 cpu_hotplug_enable();
 702                 break;
 703
 704         default:
 705                 return NOTIFY_DONE;
 706         }
 707
 708         return NOTIFY_OK;
 709 }
 710
 711
 712 static int __init cpu_hotplug_pm_sync_init(void)
 713 {
 714         /*
 715          * cpu_hotplug_pm_callback has higher priority than x86
 716          * bsp_pm_callback which depends on cpu_hotplug_pm_callback
 717          * to disable cpu hotplug to avoid cpu hotplug race.
 718          */
 719         pm_notifier(cpu_hotplug_pm_callback, 0);
 720         return 0;
 721 }
 722 core_initcall(cpu_hotplug_pm_sync_init);
 723
 724 #endif /* CONFIG_PM_SLEEP_SMP */
 725
 726 /**
 727  * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers
 728  * @cpu: cpu that just started
 729  *
 730  * This function calls the cpu_chain notifiers with CPU_STARTING.
 731  * It must be called by the arch code on the new cpu, before the new cpu
 732  * enables interrupts and before the "boot" cpu returns from __cpu_up().
 733  */
 734 void notify_cpu_starting(unsigned int cpu)
 735 {
 736         unsigned long val = CPU_STARTING;
 737
 738 #ifdef CONFIG_PM_SLEEP_SMP
 739         if (frozen_cpus != NULL && cpumask_test_cpu(cpu, frozen_cpus))
 740                 val = CPU_STARTING_FROZEN;
 741 #endif /* CONFIG_PM_SLEEP_SMP */
 742         cpu_notify(val, (void *)(long)cpu);
 743 }
 744
 745 #endif /* CONFIG_SMP */
 746
 747 /*
 748  * cpu_bit_bitmap[] is a special, "compressed" data structure that
 749  * represents all NR_CPUS bits binary values of 1<<nr.
 750  *
 751  * It is used by cpumask_of() to get a constant address to a CPU
 752  * mask value that has a single bit set only.
 753  */
 754
 755 /* cpu_bit_bitmap[0] is empty - so we can back into it */
 756 #define MASK_DECLARE_1(x)       [x+1][0] = (1UL << (x))
 757 #define MASK_DECLARE_2(x)       MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
 758 #define MASK_DECLARE_4(x)       MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
 759 #define MASK_DECLARE_8(x)       MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
 760
 761 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
 762
 763         MASK_DECLARE_8(0),      MASK_DECLARE_8(8),
 764         MASK_DECLARE_8(16),     MASK_DECLARE_8(24),
 765 #if BITS_PER_LONG > 32
 766         MASK_DECLARE_8(32),     MASK_DECLARE_8(40),
 767         MASK_DECLARE_8(48),     MASK_DECLARE_8(56),
 768 #endif
 769 };
 770 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
 771
 772 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
 773 EXPORT_SYMBOL(cpu_all_bits);
 774
 775 #ifdef CONFIG_INIT_ALL_POSSIBLE
 776 static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly
 777         = CPU_BITS_ALL;
 778 #else
 779 static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly;
 780 #endif
 781 const struct cpumask *const cpu_possible_mask = to_cpumask(cpu_possible_bits);
 782 EXPORT_SYMBOL(cpu_possible_mask);
 783
 784 static DECLARE_BITMAP(cpu_online_bits, CONFIG_NR_CPUS) __read_mostly;
 785 const struct cpumask *const cpu_online_mask = to_cpumask(cpu_online_bits);
 786 EXPORT_SYMBOL(cpu_online_mask);
 787
 788 static DECLARE_BITMAP(cpu_present_bits, CONFIG_NR_CPUS) __read_mostly;
 789 const struct cpumask *const cpu_present_mask = to_cpumask(cpu_present_bits);
 790 EXPORT_SYMBOL(cpu_present_mask);
 791
 792 static DECLARE_BITMAP(cpu_active_bits, CONFIG_NR_CPUS) __read_mostly;
 793 const struct cpumask *const cpu_active_mask = to_cpumask(cpu_active_bits);
 794 EXPORT_SYMBOL(cpu_active_mask);
 795
 796 void set_cpu_possible(unsigned int cpu, bool possible)
 797 {
 798         if (possible)
 799                 cpumask_set_cpu(cpu, to_cpumask(cpu_possible_bits));
 800         else
 801                 cpumask_clear_cpu(cpu, to_cpumask(cpu_possible_bits));
 802 }
 803
 804 void set_cpu_present(unsigned int cpu, bool present)
 805 {
 806         if (present)
 807                 cpumask_set_cpu(cpu, to_cpumask(cpu_present_bits));
 808         else
 809                 cpumask_clear_cpu(cpu, to_cpumask(cpu_present_bits));
 810 }
 811
 812 void set_cpu_online(unsigned int cpu, bool online)
 813 {
 814         if (online) {
 815                 cpumask_set_cpu(cpu, to_cpumask(cpu_online_bits));
 816                 cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits));
 817         } else {
 818                 cpumask_clear_cpu(cpu, to_cpumask(cpu_online_bits));
 819         }
 820 }
 821
 822 void set_cpu_active(unsigned int cpu, bool active)
 823 {
 824         if (active)
 825                 cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits));
 826         else
 827                 cpumask_clear_cpu(cpu, to_cpumask(cpu_active_bits));
 828 }
 829
 830 void init_cpu_present(const struct cpumask *src)
 831 {
 832         cpumask_copy(to_cpumask(cpu_present_bits), src);
 833 }
 834
 835 void init_cpu_possible(const struct cpumask *src)
 836 {
 837         cpumask_copy(to_cpumask(cpu_possible_bits), src);
 838 }
 839
 840 void init_cpu_online(const struct cpumask *src)
 841 {
 842         cpumask_copy(to_cpumask(cpu_online_bits), src);
 843 }