git.oblomov.eu Git - linux-2.6/blob - kernel/workqueue.c

   1 /*
   2  * linux/kernel/workqueue.c
   3  *
   4  * Generic mechanism for defining kernel helper threads for running
   5  * arbitrary tasks in process context.
   6  *
   7  * Started by Ingo Molnar, Copyright (C) 2002
   8  *
   9  * Derived from the taskqueue/keventd code by:
  10  *
  11  *   David Woodhouse <dwmw2@infradead.org>
  12  *   Andrew Morton <andrewm@uow.edu.au>
  13  *   Kai Petzke <wpp@marie.physik.tu-berlin.de>
  14  *   Theodore Ts'o <tytso@mit.edu>
  15  *
  16  * Made to use alloc_percpu by Christoph Lameter <clameter@sgi.com>.
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/kernel.h>
  21 #include <linux/sched.h>
  22 #include <linux/init.h>
  23 #include <linux/signal.h>
  24 #include <linux/completion.h>
  25 #include <linux/workqueue.h>
  26 #include <linux/slab.h>
  27 #include <linux/cpu.h>
  28 #include <linux/notifier.h>
  29 #include <linux/kthread.h>
  30 #include <linux/hardirq.h>
  31 #include <linux/mempolicy.h>
  32 #include <linux/freezer.h>
  33 #include <linux/kallsyms.h>
  34 #include <linux/debug_locks.h>
  35
  36 /*
  37  * The per-CPU workqueue (if single thread, we always use the first
  38  * possible cpu).
  39  *
  40  * The sequence counters are for flush_scheduled_work().  It wants to wait
  41  * until all currently-scheduled works are completed, but it doesn't
  42  * want to be livelocked by new, incoming ones.  So it waits until
  43  * remove_sequence is >= the insert_sequence which pertained when
  44  * flush_scheduled_work() was called.
  45  */
  46 struct cpu_workqueue_struct {
  47
  48         spinlock_t lock;
  49
  50         long remove_sequence;   /* Least-recently added (next to run) */
  51         long insert_sequence;   /* Next to add */
  52
  53         struct list_head worklist;
  54         wait_queue_head_t more_work;
  55         wait_queue_head_t work_done;
  56
  57         struct workqueue_struct *wq;
  58         struct task_struct *thread;
  59
  60         int run_depth;          /* Detect run_workqueue() recursion depth */
  61
  62         int freezeable;         /* Freeze the thread during suspend */
  63 } ____cacheline_aligned;
  64
  65 /*
  66  * The externally visible workqueue abstraction is an array of
  67  * per-CPU workqueues:
  68  */
  69 struct workqueue_struct {
  70         struct cpu_workqueue_struct *cpu_wq;
  71         const char *name;
  72         struct list_head list;  /* Empty if single thread */
  73 };
  74
  75 /* All the per-cpu workqueues on the system, for hotplug cpu to add/remove
  76    threads to each one as cpus come/go. */
  77 static DEFINE_MUTEX(workqueue_mutex);
  78 static LIST_HEAD(workqueues);
  79
  80 static int singlethread_cpu;
  81
  82 /* If it's single threaded, it isn't in the list of workqueues. */
  83 static inline int is_single_threaded(struct workqueue_struct *wq)
  84 {
  85         return list_empty(&wq->list);
  86 }
  87
  88 /*
  89  * Set the workqueue on which a work item is to be run
  90  * - Must *only* be called if the pending flag is set
  91  */
  92 static inline void set_wq_data(struct work_struct *work, void *wq)
  93 {
  94         unsigned long new;
  95
  96         BUG_ON(!work_pending(work));
  97
  98         new = (unsigned long) wq | (1UL << WORK_STRUCT_PENDING);
  99         new |= WORK_STRUCT_FLAG_MASK & *work_data_bits(work);
 100         atomic_long_set(&work->data, new);
 101 }
 102
 103 static inline void *get_wq_data(struct work_struct *work)
 104 {
 105         return (void *) (atomic_long_read(&work->data) & WORK_STRUCT_WQ_DATA_MASK);
 106 }
 107
 108 static int __run_work(struct cpu_workqueue_struct *cwq, struct work_struct *work)
 109 {
 110         int ret = 0;
 111         unsigned long flags;
 112
 113         spin_lock_irqsave(&cwq->lock, flags);
 114         /*
 115          * We need to re-validate the work info after we've gotten
 116          * the cpu_workqueue lock. We can run the work now iff:
 117          *
 118          *  - the wq_data still matches the cpu_workqueue_struct
 119          *  - AND the work is still marked pending
 120          *  - AND the work is still on a list (which will be this
 121          *    workqueue_struct list)
 122          *
 123          * All these conditions are important, because we
 124          * need to protect against the work being run right
 125          * now on another CPU (all but the last one might be
 126          * true if it's currently running and has not been
 127          * released yet, for example).
 128          */
 129         if (get_wq_data(work) == cwq
 130             && work_pending(work)
 131             && !list_empty(&work->entry)) {
 132                 work_func_t f = work->func;
 133                 list_del_init(&work->entry);
 134                 spin_unlock_irqrestore(&cwq->lock, flags);
 135
 136                 if (!test_bit(WORK_STRUCT_NOAUTOREL, work_data_bits(work)))
 137                         work_release(work);
 138                 f(work);
 139
 140                 spin_lock_irqsave(&cwq->lock, flags);
 141                 cwq->remove_sequence++;
 142                 wake_up(&cwq->work_done);
 143                 ret = 1;
 144         }
 145         spin_unlock_irqrestore(&cwq->lock, flags);
 146         return ret;
 147 }
 148
 149 /**
 150  * run_scheduled_work - run scheduled work synchronously
 151  * @work: work to run
 152  *
 153  * This checks if the work was pending, and runs it
 154  * synchronously if so. It returns a boolean to indicate
 155  * whether it had any scheduled work to run or not.
 156  *
 157  * NOTE! This _only_ works for normal work_structs. You
 158  * CANNOT use this for delayed work, because the wq data
 159  * for delayed work will not point properly to the per-
 160  * CPU workqueue struct, but will change!
 161  */
 162 int fastcall run_scheduled_work(struct work_struct *work)
 163 {
 164         for (;;) {
 165                 struct cpu_workqueue_struct *cwq;
 166
 167                 if (!work_pending(work))
 168                         return 0;
 169                 if (list_empty(&work->entry))
 170                         return 0;
 171                 /* NOTE! This depends intimately on __queue_work! */
 172                 cwq = get_wq_data(work);
 173                 if (!cwq)
 174                         return 0;
 175                 if (__run_work(cwq, work))
 176                         return 1;
 177         }
 178 }
 179 EXPORT_SYMBOL(run_scheduled_work);
 180
 181 /* Preempt must be disabled. */
 182 static void __queue_work(struct cpu_workqueue_struct *cwq,
 183                          struct work_struct *work)
 184 {
 185         unsigned long flags;
 186
 187         spin_lock_irqsave(&cwq->lock, flags);
 188         set_wq_data(work, cwq);
 189         list_add_tail(&work->entry, &cwq->worklist);
 190         cwq->insert_sequence++;
 191         wake_up(&cwq->more_work);
 192         spin_unlock_irqrestore(&cwq->lock, flags);
 193 }
 194
 195 /**
 196  * queue_work - queue work on a workqueue
 197  * @wq: workqueue to use
 198  * @work: work to queue
 199  *
 200  * Returns 0 if @work was already on a queue, non-zero otherwise.
 201  *
 202  * We queue the work to the CPU it was submitted, but there is no
 203  * guarantee that it will be processed by that CPU.
 204  */
 205 int fastcall queue_work(struct workqueue_struct *wq, struct work_struct *work)
 206 {
 207         int ret = 0, cpu = get_cpu();
 208
 209         if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) {
 210                 if (unlikely(is_single_threaded(wq)))
 211                         cpu = singlethread_cpu;
 212                 BUG_ON(!list_empty(&work->entry));
 213                 __queue_work(per_cpu_ptr(wq->cpu_wq, cpu), work);
 214                 ret = 1;
 215         }
 216         put_cpu();
 217         return ret;
 218 }
 219 EXPORT_SYMBOL_GPL(queue_work);
 220
 221 static void delayed_work_timer_fn(unsigned long __data)
 222 {
 223         struct delayed_work *dwork = (struct delayed_work *)__data;
 224         struct workqueue_struct *wq = get_wq_data(&dwork->work);
 225         int cpu = smp_processor_id();
 226
 227         if (unlikely(is_single_threaded(wq)))
 228                 cpu = singlethread_cpu;
 229
 230         __queue_work(per_cpu_ptr(wq->cpu_wq, cpu), &dwork->work);
 231 }
 232
 233 /**
 234  * queue_delayed_work - queue work on a workqueue after delay
 235  * @wq: workqueue to use
 236  * @dwork: delayable work to queue
 237  * @delay: number of jiffies to wait before queueing
 238  *
 239  * Returns 0 if @work was already on a queue, non-zero otherwise.
 240  */
 241 int fastcall queue_delayed_work(struct workqueue_struct *wq,
 242                         struct delayed_work *dwork, unsigned long delay)
 243 {
 244         int ret = 0;
 245         struct timer_list *timer = &dwork->timer;
 246         struct work_struct *work = &dwork->work;
 247
 248         if (delay == 0)
 249                 return queue_work(wq, work);
 250
 251         if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) {
 252                 BUG_ON(timer_pending(timer));
 253                 BUG_ON(!list_empty(&work->entry));
 254
 255                 /* This stores wq for the moment, for the timer_fn */
 256                 set_wq_data(work, wq);
 257                 timer->expires = jiffies + delay;
 258                 timer->data = (unsigned long)dwork;
 259                 timer->function = delayed_work_timer_fn;
 260                 add_timer(timer);
 261                 ret = 1;
 262         }
 263         return ret;
 264 }
 265 EXPORT_SYMBOL_GPL(queue_delayed_work);
 266
 267 /**
 268  * queue_delayed_work_on - queue work on specific CPU after delay
 269  * @cpu: CPU number to execute work on
 270  * @wq: workqueue to use
 271  * @dwork: work to queue
 272  * @delay: number of jiffies to wait before queueing
 273  *
 274  * Returns 0 if @work was already on a queue, non-zero otherwise.
 275  */
 276 int queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
 277                         struct delayed_work *dwork, unsigned long delay)
 278 {
 279         int ret = 0;
 280         struct timer_list *timer = &dwork->timer;
 281         struct work_struct *work = &dwork->work;
 282
 283         if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) {
 284                 BUG_ON(timer_pending(timer));
 285                 BUG_ON(!list_empty(&work->entry));
 286
 287                 /* This stores wq for the moment, for the timer_fn */
 288                 set_wq_data(work, wq);
 289                 timer->expires = jiffies + delay;
 290                 timer->data = (unsigned long)dwork;
 291                 timer->function = delayed_work_timer_fn;
 292                 add_timer_on(timer, cpu);
 293                 ret = 1;
 294         }
 295         return ret;
 296 }
 297 EXPORT_SYMBOL_GPL(queue_delayed_work_on);
 298
 299 static void run_workqueue(struct cpu_workqueue_struct *cwq)
 300 {
 301         unsigned long flags;
 302
 303         /*
 304          * Keep taking off work from the queue until
 305          * done.
 306          */
 307         spin_lock_irqsave(&cwq->lock, flags);
 308         cwq->run_depth++;
 309         if (cwq->run_depth > 3) {
 310                 /* morton gets to eat his hat */
 311                 printk("%s: recursion depth exceeded: %d\n",
 312                         __FUNCTION__, cwq->run_depth);
 313                 dump_stack();
 314         }
 315         while (!list_empty(&cwq->worklist)) {
 316                 struct work_struct *work = list_entry(cwq->worklist.next,
 317                                                 struct work_struct, entry);
 318                 work_func_t f = work->func;
 319
 320                 list_del_init(cwq->worklist.next);
 321                 spin_unlock_irqrestore(&cwq->lock, flags);
 322
 323                 BUG_ON(get_wq_data(work) != cwq);
 324                 if (!test_bit(WORK_STRUCT_NOAUTOREL, work_data_bits(work)))
 325                         work_release(work);
 326                 f(work);
 327
 328                 if (unlikely(in_atomic() || lockdep_depth(current) > 0)) {
 329                         printk(KERN_ERR "BUG: workqueue leaked lock or atomic: "
 330                                         "%s/0x%08x/%d\n",
 331                                         current->comm, preempt_count(),
 332                                         current->pid);
 333                         printk(KERN_ERR "    last function: ");
 334                         print_symbol("%s\n", (unsigned long)f);
 335                         debug_show_held_locks(current);
 336                         dump_stack();
 337                 }
 338
 339                 spin_lock_irqsave(&cwq->lock, flags);
 340                 cwq->remove_sequence++;
 341                 wake_up(&cwq->work_done);
 342         }
 343         cwq->run_depth--;
 344         spin_unlock_irqrestore(&cwq->lock, flags);
 345 }
 346
 347 static int worker_thread(void *__cwq)
 348 {
 349         struct cpu_workqueue_struct *cwq = __cwq;
 350         DECLARE_WAITQUEUE(wait, current);
 351         struct k_sigaction sa;
 352         sigset_t blocked;
 353
 354         if (!cwq->freezeable)
 355                 current->flags |= PF_NOFREEZE;
 356
 357         set_user_nice(current, -5);
 358
 359         /* Block and flush all signals */
 360         sigfillset(&blocked);
 361         sigprocmask(SIG_BLOCK, &blocked, NULL);
 362         flush_signals(current);
 363
 364         /*
 365          * We inherited MPOL_INTERLEAVE from the booting kernel.
 366          * Set MPOL_DEFAULT to insure node local allocations.
 367          */
 368         numa_default_policy();
 369
 370         /* SIG_IGN makes children autoreap: see do_notify_parent(). */
 371         sa.sa.sa_handler = SIG_IGN;
 372         sa.sa.sa_flags = 0;
 373         siginitset(&sa.sa.sa_mask, sigmask(SIGCHLD));
 374         do_sigaction(SIGCHLD, &sa, (struct k_sigaction *)0);
 375
 376         set_current_state(TASK_INTERRUPTIBLE);
 377         while (!kthread_should_stop()) {
 378                 if (cwq->freezeable)
 379                         try_to_freeze();
 380
 381                 add_wait_queue(&cwq->more_work, &wait);
 382                 if (list_empty(&cwq->worklist))
 383                         schedule();
 384                 else
 385                         __set_current_state(TASK_RUNNING);
 386                 remove_wait_queue(&cwq->more_work, &wait);
 387
 388                 if (!list_empty(&cwq->worklist))
 389                         run_workqueue(cwq);
 390                 set_current_state(TASK_INTERRUPTIBLE);
 391         }
 392         __set_current_state(TASK_RUNNING);
 393         return 0;
 394 }
 395
 396 static void flush_cpu_workqueue(struct cpu_workqueue_struct *cwq)
 397 {
 398         if (cwq->thread == current) {
 399                 /*
 400                  * Probably keventd trying to flush its own queue. So simply run
 401                  * it by hand rather than deadlocking.
 402                  */
 403                 run_workqueue(cwq);
 404         } else {
 405                 DEFINE_WAIT(wait);
 406                 long sequence_needed;
 407
 408                 spin_lock_irq(&cwq->lock);
 409                 sequence_needed = cwq->insert_sequence;
 410
 411                 while (sequence_needed - cwq->remove_sequence > 0) {
 412                         prepare_to_wait(&cwq->work_done, &wait,
 413                                         TASK_UNINTERRUPTIBLE);
 414                         spin_unlock_irq(&cwq->lock);
 415                         schedule();
 416                         spin_lock_irq(&cwq->lock);
 417                 }
 418                 finish_wait(&cwq->work_done, &wait);
 419                 spin_unlock_irq(&cwq->lock);
 420         }
 421 }
 422
 423 /**
 424  * flush_workqueue - ensure that any scheduled work has run to completion.
 425  * @wq: workqueue to flush
 426  *
 427  * Forces execution of the workqueue and blocks until its completion.
 428  * This is typically used in driver shutdown handlers.
 429  *
 430  * This function will sample each workqueue's current insert_sequence number and
 431  * will sleep until the head sequence is greater than or equal to that.  This
 432  * means that we sleep until all works which were queued on entry have been
 433  * handled, but we are not livelocked by new incoming ones.
 434  *
 435  * This function used to run the workqueues itself.  Now we just wait for the
 436  * helper threads to do it.
 437  */
 438 void fastcall flush_workqueue(struct workqueue_struct *wq)
 439 {
 440         might_sleep();
 441
 442         if (is_single_threaded(wq)) {
 443                 /* Always use first cpu's area. */
 444                 flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, singlethread_cpu));
 445         } else {
 446                 int cpu;
 447
 448                 mutex_lock(&workqueue_mutex);
 449                 for_each_online_cpu(cpu)
 450                         flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, cpu));
 451                 mutex_unlock(&workqueue_mutex);
 452         }
 453 }
 454 EXPORT_SYMBOL_GPL(flush_workqueue);
 455
 456 static struct task_struct *create_workqueue_thread(struct workqueue_struct *wq,
 457                                                    int cpu, int freezeable)
 458 {
 459         struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
 460         struct task_struct *p;
 461
 462         spin_lock_init(&cwq->lock);
 463         cwq->wq = wq;
 464         cwq->thread = NULL;
 465         cwq->insert_sequence = 0;
 466         cwq->remove_sequence = 0;
 467         cwq->freezeable = freezeable;
 468         INIT_LIST_HEAD(&cwq->worklist);
 469         init_waitqueue_head(&cwq->more_work);
 470         init_waitqueue_head(&cwq->work_done);
 471
 472         if (is_single_threaded(wq))
 473                 p = kthread_create(worker_thread, cwq, "%s", wq->name);
 474         else
 475                 p = kthread_create(worker_thread, cwq, "%s/%d", wq->name, cpu);
 476         if (IS_ERR(p))
 477                 return NULL;
 478         cwq->thread = p;
 479         return p;
 480 }
 481
 482 struct workqueue_struct *__create_workqueue(const char *name,
 483                                             int singlethread, int freezeable)
 484 {
 485         int cpu, destroy = 0;
 486         struct workqueue_struct *wq;
 487         struct task_struct *p;
 488
 489         wq = kzalloc(sizeof(*wq), GFP_KERNEL);
 490         if (!wq)
 491                 return NULL;
 492
 493         wq->cpu_wq = alloc_percpu(struct cpu_workqueue_struct);
 494         if (!wq->cpu_wq) {
 495                 kfree(wq);
 496                 return NULL;
 497         }
 498
 499         wq->name = name;
 500         mutex_lock(&workqueue_mutex);
 501         if (singlethread) {
 502                 INIT_LIST_HEAD(&wq->list);
 503                 p = create_workqueue_thread(wq, singlethread_cpu, freezeable);
 504                 if (!p)
 505                         destroy = 1;
 506                 else
 507                         wake_up_process(p);
 508         } else {
 509                 list_add(&wq->list, &workqueues);
 510                 for_each_online_cpu(cpu) {
 511                         p = create_workqueue_thread(wq, cpu, freezeable);
 512                         if (p) {
 513                                 kthread_bind(p, cpu);
 514                                 wake_up_process(p);
 515                         } else
 516                                 destroy = 1;
 517                 }
 518         }
 519         mutex_unlock(&workqueue_mutex);
 520
 521         /*
 522          * Was there any error during startup? If yes then clean up:
 523          */
 524         if (destroy) {
 525                 destroy_workqueue(wq);
 526                 wq = NULL;
 527         }
 528         return wq;
 529 }
 530 EXPORT_SYMBOL_GPL(__create_workqueue);
 531
 532 static void cleanup_workqueue_thread(struct workqueue_struct *wq, int cpu)
 533 {
 534         struct cpu_workqueue_struct *cwq;
 535         unsigned long flags;
 536         struct task_struct *p;
 537
 538         cwq = per_cpu_ptr(wq->cpu_wq, cpu);
 539         spin_lock_irqsave(&cwq->lock, flags);
 540         p = cwq->thread;
 541         cwq->thread = NULL;
 542         spin_unlock_irqrestore(&cwq->lock, flags);
 543         if (p)
 544                 kthread_stop(p);
 545 }
 546
 547 /**
 548  * destroy_workqueue - safely terminate a workqueue
 549  * @wq: target workqueue
 550  *
 551  * Safely destroy a workqueue. All work currently pending will be done first.
 552  */
 553 void destroy_workqueue(struct workqueue_struct *wq)
 554 {
 555         int cpu;
 556
 557         flush_workqueue(wq);
 558
 559         /* We don't need the distraction of CPUs appearing and vanishing. */
 560         mutex_lock(&workqueue_mutex);
 561         if (is_single_threaded(wq))
 562                 cleanup_workqueue_thread(wq, singlethread_cpu);
 563         else {
 564                 for_each_online_cpu(cpu)
 565                         cleanup_workqueue_thread(wq, cpu);
 566                 list_del(&wq->list);
 567         }
 568         mutex_unlock(&workqueue_mutex);
 569         free_percpu(wq->cpu_wq);
 570         kfree(wq);
 571 }
 572 EXPORT_SYMBOL_GPL(destroy_workqueue);
 573
 574 static struct workqueue_struct *keventd_wq;
 575
 576 /**
 577  * schedule_work - put work task in global workqueue
 578  * @work: job to be done
 579  *
 580  * This puts a job in the kernel-global workqueue.
 581  */
 582 int fastcall schedule_work(struct work_struct *work)
 583 {
 584         return queue_work(keventd_wq, work);
 585 }
 586 EXPORT_SYMBOL(schedule_work);
 587
 588 /**
 589  * schedule_delayed_work - put work task in global workqueue after delay
 590  * @dwork: job to be done
 591  * @delay: number of jiffies to wait or 0 for immediate execution
 592  *
 593  * After waiting for a given time this puts a job in the kernel-global
 594  * workqueue.
 595  */
 596 int fastcall schedule_delayed_work(struct delayed_work *dwork, unsigned long delay)
 597 {
 598         return queue_delayed_work(keventd_wq, dwork, delay);
 599 }
 600 EXPORT_SYMBOL(schedule_delayed_work);
 601
 602 /**
 603  * schedule_delayed_work_on - queue work in global workqueue on CPU after delay
 604  * @cpu: cpu to use
 605  * @dwork: job to be done
 606  * @delay: number of jiffies to wait
 607  *
 608  * After waiting for a given time this puts a job in the kernel-global
 609  * workqueue on the specified CPU.
 610  */
 611 int schedule_delayed_work_on(int cpu,
 612                         struct delayed_work *dwork, unsigned long delay)
 613 {
 614         return queue_delayed_work_on(cpu, keventd_wq, dwork, delay);
 615 }
 616 EXPORT_SYMBOL(schedule_delayed_work_on);
 617
 618 /**
 619  * schedule_on_each_cpu - call a function on each online CPU from keventd
 620  * @func: the function to call
 621  *
 622  * Returns zero on success.
 623  * Returns -ve errno on failure.
 624  *
 625  * Appears to be racy against CPU hotplug.
 626  *
 627  * schedule_on_each_cpu() is very slow.
 628  */
 629 int schedule_on_each_cpu(work_func_t func)
 630 {
 631         int cpu;
 632         struct work_struct *works;
 633
 634         works = alloc_percpu(struct work_struct);
 635         if (!works)
 636                 return -ENOMEM;
 637
 638         mutex_lock(&workqueue_mutex);
 639         for_each_online_cpu(cpu) {
 640                 struct work_struct *work = per_cpu_ptr(works, cpu);
 641
 642                 INIT_WORK(work, func);
 643                 set_bit(WORK_STRUCT_PENDING, work_data_bits(work));
 644                 __queue_work(per_cpu_ptr(keventd_wq->cpu_wq, cpu), work);
 645         }
 646         mutex_unlock(&workqueue_mutex);
 647         flush_workqueue(keventd_wq);
 648         free_percpu(works);
 649         return 0;
 650 }
 651
 652 void flush_scheduled_work(void)
 653 {
 654         flush_workqueue(keventd_wq);
 655 }
 656 EXPORT_SYMBOL(flush_scheduled_work);
 657
 658 /**
 659  * cancel_rearming_delayed_workqueue - reliably kill off a delayed work whose handler rearms the delayed work.
 660  * @wq:   the controlling workqueue structure
 661  * @dwork: the delayed work struct
 662  */
 663 void cancel_rearming_delayed_workqueue(struct workqueue_struct *wq,
 664                                        struct delayed_work *dwork)
 665 {
 666         while (!cancel_delayed_work(dwork))
 667                 flush_workqueue(wq);
 668 }
 669 EXPORT_SYMBOL(cancel_rearming_delayed_workqueue);
 670
 671 /**
 672  * cancel_rearming_delayed_work - reliably kill off a delayed keventd work whose handler rearms the delayed work.
 673  * @dwork: the delayed work struct
 674  */
 675 void cancel_rearming_delayed_work(struct delayed_work *dwork)
 676 {
 677         cancel_rearming_delayed_workqueue(keventd_wq, dwork);
 678 }
 679 EXPORT_SYMBOL(cancel_rearming_delayed_work);
 680
 681 /**
 682  * execute_in_process_context - reliably execute the routine with user context
 683  * @fn:         the function to execute
 684  * @ew:         guaranteed storage for the execute work structure (must
 685  *              be available when the work executes)
 686  *
 687  * Executes the function immediately if process context is available,
 688  * otherwise schedules the function for delayed execution.
 689  *
 690  * Returns:     0 - function was executed
 691  *              1 - function was scheduled for execution
 692  */
 693 int execute_in_process_context(work_func_t fn, struct execute_work *ew)
 694 {
 695         if (!in_interrupt()) {
 696                 fn(&ew->work);
 697                 return 0;
 698         }
 699
 700         INIT_WORK(&ew->work, fn);
 701         schedule_work(&ew->work);
 702
 703         return 1;
 704 }
 705 EXPORT_SYMBOL_GPL(execute_in_process_context);
 706
 707 int keventd_up(void)
 708 {
 709         return keventd_wq != NULL;
 710 }
 711
 712 int current_is_keventd(void)
 713 {
 714         struct cpu_workqueue_struct *cwq;
 715         int cpu = smp_processor_id();   /* preempt-safe: keventd is per-cpu */
 716         int ret = 0;
 717
 718         BUG_ON(!keventd_wq);
 719
 720         cwq = per_cpu_ptr(keventd_wq->cpu_wq, cpu);
 721         if (current == cwq->thread)
 722                 ret = 1;
 723
 724         return ret;
 725
 726 }
 727
 728 /* Take the work from this (downed) CPU. */
 729 static void take_over_work(struct workqueue_struct *wq, unsigned int cpu)
 730 {
 731         struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
 732         struct list_head list;
 733         struct work_struct *work;
 734
 735         spin_lock_irq(&cwq->lock);
 736         list_replace_init(&cwq->worklist, &list);
 737
 738         while (!list_empty(&list)) {
 739                 printk("Taking work for %s\n", wq->name);
 740                 work = list_entry(list.next,struct work_struct,entry);
 741                 list_del(&work->entry);
 742                 __queue_work(per_cpu_ptr(wq->cpu_wq, smp_processor_id()), work);
 743         }
 744         spin_unlock_irq(&cwq->lock);
 745 }
 746
 747 /* We're holding the cpucontrol mutex here */
 748 static int __devinit workqueue_cpu_callback(struct notifier_block *nfb,
 749                                   unsigned long action,
 750                                   void *hcpu)
 751 {
 752         unsigned int hotcpu = (unsigned long)hcpu;
 753         struct workqueue_struct *wq;
 754
 755         switch (action) {
 756         case CPU_UP_PREPARE:
 757                 mutex_lock(&workqueue_mutex);
 758                 /* Create a new workqueue thread for it. */
 759                 list_for_each_entry(wq, &workqueues, list) {
 760                         if (!create_workqueue_thread(wq, hotcpu, 0)) {
 761                                 printk("workqueue for %i failed\n", hotcpu);
 762                                 return NOTIFY_BAD;
 763                         }
 764                 }
 765                 break;
 766
 767         case CPU_ONLINE:
 768                 /* Kick off worker threads. */
 769                 list_for_each_entry(wq, &workqueues, list) {
 770                         struct cpu_workqueue_struct *cwq;
 771
 772                         cwq = per_cpu_ptr(wq->cpu_wq, hotcpu);
 773                         kthread_bind(cwq->thread, hotcpu);
 774                         wake_up_process(cwq->thread);
 775                 }
 776                 mutex_unlock(&workqueue_mutex);
 777                 break;
 778
 779         case CPU_UP_CANCELED:
 780                 list_for_each_entry(wq, &workqueues, list) {
 781                         if (!per_cpu_ptr(wq->cpu_wq, hotcpu)->thread)
 782                                 continue;
 783                         /* Unbind so it can run. */
 784                         kthread_bind(per_cpu_ptr(wq->cpu_wq, hotcpu)->thread,
 785                                      any_online_cpu(cpu_online_map));
 786                         cleanup_workqueue_thread(wq, hotcpu);
 787                 }
 788                 mutex_unlock(&workqueue_mutex);
 789                 break;
 790
 791         case CPU_DOWN_PREPARE:
 792                 mutex_lock(&workqueue_mutex);
 793                 break;
 794
 795         case CPU_DOWN_FAILED:
 796                 mutex_unlock(&workqueue_mutex);
 797                 break;
 798
 799         case CPU_DEAD:
 800                 list_for_each_entry(wq, &workqueues, list)
 801                         cleanup_workqueue_thread(wq, hotcpu);
 802                 list_for_each_entry(wq, &workqueues, list)
 803                         take_over_work(wq, hotcpu);
 804                 mutex_unlock(&workqueue_mutex);
 805                 break;
 806         }
 807
 808         return NOTIFY_OK;
 809 }
 810
 811 void init_workqueues(void)
 812 {
 813         singlethread_cpu = first_cpu(cpu_possible_map);
 814         hotcpu_notifier(workqueue_cpu_callback, 0);
 815         keventd_wq = create_workqueue("events");
 816         BUG_ON(!keventd_wq);
 817 }
 818