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
  31 /*
  32  * The per-CPU workqueue (if single thread, we always use cpu 0's).
  33  *
  34  * The sequence counters are for flush_scheduled_work().  It wants to wait
  35  * until until all currently-scheduled works are completed, but it doesn't
  36  * want to be livelocked by new, incoming ones.  So it waits until
  37  * remove_sequence is >= the insert_sequence which pertained when
  38  * flush_scheduled_work() was called.
  39  */
  40 struct cpu_workqueue_struct {
  41
  42         spinlock_t lock;
  43
  44         long remove_sequence;   /* Least-recently added (next to run) */
  45         long insert_sequence;   /* Next to add */
  46
  47         struct list_head worklist;
  48         wait_queue_head_t more_work;
  49         wait_queue_head_t work_done;
  50
  51         struct workqueue_struct *wq;
  52         task_t *thread;
  53
  54         int run_depth;          /* Detect run_workqueue() recursion depth */
  55 } ____cacheline_aligned;
  56
  57 /*
  58  * The externally visible workqueue abstraction is an array of
  59  * per-CPU workqueues:
  60  */
  61 struct workqueue_struct {
  62         struct cpu_workqueue_struct *cpu_wq;
  63         const char *name;
  64         struct list_head list;  /* Empty if single thread */
  65 };
  66
  67 /* All the per-cpu workqueues on the system, for hotplug cpu to add/remove
  68    threads to each one as cpus come/go. */
  69 static DEFINE_SPINLOCK(workqueue_lock);
  70 static LIST_HEAD(workqueues);
  71
  72 /* If it's single threaded, it isn't in the list of workqueues. */
  73 static inline int is_single_threaded(struct workqueue_struct *wq)
  74 {
  75         return list_empty(&wq->list);
  76 }
  77
  78 /* Preempt must be disabled. */
  79 static void __queue_work(struct cpu_workqueue_struct *cwq,
  80                          struct work_struct *work)
  81 {
  82         unsigned long flags;
  83
  84         spin_lock_irqsave(&cwq->lock, flags);
  85         work->wq_data = cwq;
  86         list_add_tail(&work->entry, &cwq->worklist);
  87         cwq->insert_sequence++;
  88         wake_up(&cwq->more_work);
  89         spin_unlock_irqrestore(&cwq->lock, flags);
  90 }
  91
  92 /*
  93  * Queue work on a workqueue. Return non-zero if it was successfully
  94  * added.
  95  *
  96  * We queue the work to the CPU it was submitted, but there is no
  97  * guarantee that it will be processed by that CPU.
  98  */
  99 int fastcall queue_work(struct workqueue_struct *wq, struct work_struct *work)
 100 {
 101         int ret = 0, cpu = get_cpu();
 102
 103         if (!test_and_set_bit(0, &work->pending)) {
 104                 if (unlikely(is_single_threaded(wq)))
 105                         cpu = 0;
 106                 BUG_ON(!list_empty(&work->entry));
 107                 __queue_work(per_cpu_ptr(wq->cpu_wq, cpu), work);
 108                 ret = 1;
 109         }
 110         put_cpu();
 111         return ret;
 112 }
 113
 114 static void delayed_work_timer_fn(unsigned long __data)
 115 {
 116         struct work_struct *work = (struct work_struct *)__data;
 117         struct workqueue_struct *wq = work->wq_data;
 118         int cpu = smp_processor_id();
 119
 120         if (unlikely(is_single_threaded(wq)))
 121                 cpu = 0;
 122
 123         __queue_work(per_cpu_ptr(wq->cpu_wq, cpu), work);
 124 }
 125
 126 int fastcall queue_delayed_work(struct workqueue_struct *wq,
 127                         struct work_struct *work, unsigned long delay)
 128 {
 129         int ret = 0;
 130         struct timer_list *timer = &work->timer;
 131
 132         if (!test_and_set_bit(0, &work->pending)) {
 133                 BUG_ON(timer_pending(timer));
 134                 BUG_ON(!list_empty(&work->entry));
 135
 136                 /* This stores wq for the moment, for the timer_fn */
 137                 work->wq_data = wq;
 138                 timer->expires = jiffies + delay;
 139                 timer->data = (unsigned long)work;
 140                 timer->function = delayed_work_timer_fn;
 141                 add_timer(timer);
 142                 ret = 1;
 143         }
 144         return ret;
 145 }
 146
 147 static inline void run_workqueue(struct cpu_workqueue_struct *cwq)
 148 {
 149         unsigned long flags;
 150
 151         /*
 152          * Keep taking off work from the queue until
 153          * done.
 154          */
 155         spin_lock_irqsave(&cwq->lock, flags);
 156         cwq->run_depth++;
 157         if (cwq->run_depth > 3) {
 158                 /* morton gets to eat his hat */
 159                 printk("%s: recursion depth exceeded: %d\n",
 160                         __FUNCTION__, cwq->run_depth);
 161                 dump_stack();
 162         }
 163         while (!list_empty(&cwq->worklist)) {
 164                 struct work_struct *work = list_entry(cwq->worklist.next,
 165                                                 struct work_struct, entry);
 166                 void (*f) (void *) = work->func;
 167                 void *data = work->data;
 168
 169                 list_del_init(cwq->worklist.next);
 170                 spin_unlock_irqrestore(&cwq->lock, flags);
 171
 172                 BUG_ON(work->wq_data != cwq);
 173                 clear_bit(0, &work->pending);
 174                 f(data);
 175
 176                 spin_lock_irqsave(&cwq->lock, flags);
 177                 cwq->remove_sequence++;
 178                 wake_up(&cwq->work_done);
 179         }
 180         cwq->run_depth--;
 181         spin_unlock_irqrestore(&cwq->lock, flags);
 182 }
 183
 184 static int worker_thread(void *__cwq)
 185 {
 186         struct cpu_workqueue_struct *cwq = __cwq;
 187         DECLARE_WAITQUEUE(wait, current);
 188         struct k_sigaction sa;
 189         sigset_t blocked;
 190
 191         current->flags |= PF_NOFREEZE;
 192
 193         set_user_nice(current, -5);
 194
 195         /* Block and flush all signals */
 196         sigfillset(&blocked);
 197         sigprocmask(SIG_BLOCK, &blocked, NULL);
 198         flush_signals(current);
 199
 200         /* SIG_IGN makes children autoreap: see do_notify_parent(). */
 201         sa.sa.sa_handler = SIG_IGN;
 202         sa.sa.sa_flags = 0;
 203         siginitset(&sa.sa.sa_mask, sigmask(SIGCHLD));
 204         do_sigaction(SIGCHLD, &sa, (struct k_sigaction *)0);
 205
 206         set_current_state(TASK_INTERRUPTIBLE);
 207         while (!kthread_should_stop()) {
 208                 add_wait_queue(&cwq->more_work, &wait);
 209                 if (list_empty(&cwq->worklist))
 210                         schedule();
 211                 else
 212                         __set_current_state(TASK_RUNNING);
 213                 remove_wait_queue(&cwq->more_work, &wait);
 214
 215                 if (!list_empty(&cwq->worklist))
 216                         run_workqueue(cwq);
 217                 set_current_state(TASK_INTERRUPTIBLE);
 218         }
 219         __set_current_state(TASK_RUNNING);
 220         return 0;
 221 }
 222
 223 static void flush_cpu_workqueue(struct cpu_workqueue_struct *cwq)
 224 {
 225         if (cwq->thread == current) {
 226                 /*
 227                  * Probably keventd trying to flush its own queue. So simply run
 228                  * it by hand rather than deadlocking.
 229                  */
 230                 run_workqueue(cwq);
 231         } else {
 232                 DEFINE_WAIT(wait);
 233                 long sequence_needed;
 234
 235                 spin_lock_irq(&cwq->lock);
 236                 sequence_needed = cwq->insert_sequence;
 237
 238                 while (sequence_needed - cwq->remove_sequence > 0) {
 239                         prepare_to_wait(&cwq->work_done, &wait,
 240                                         TASK_UNINTERRUPTIBLE);
 241                         spin_unlock_irq(&cwq->lock);
 242                         schedule();
 243                         spin_lock_irq(&cwq->lock);
 244                 }
 245                 finish_wait(&cwq->work_done, &wait);
 246                 spin_unlock_irq(&cwq->lock);
 247         }
 248 }
 249
 250 /*
 251  * flush_workqueue - ensure that any scheduled work has run to completion.
 252  *
 253  * Forces execution of the workqueue and blocks until its completion.
 254  * This is typically used in driver shutdown handlers.
 255  *
 256  * This function will sample each workqueue's current insert_sequence number and
 257  * will sleep until the head sequence is greater than or equal to that.  This
 258  * means that we sleep until all works which were queued on entry have been
 259  * handled, but we are not livelocked by new incoming ones.
 260  *
 261  * This function used to run the workqueues itself.  Now we just wait for the
 262  * helper threads to do it.
 263  */
 264 void fastcall flush_workqueue(struct workqueue_struct *wq)
 265 {
 266         might_sleep();
 267
 268         if (is_single_threaded(wq)) {
 269                 /* Always use cpu 0's area. */
 270                 flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, 0));
 271         } else {
 272                 int cpu;
 273
 274                 lock_cpu_hotplug();
 275                 for_each_online_cpu(cpu)
 276                         flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, cpu));
 277                 unlock_cpu_hotplug();
 278         }
 279 }
 280
 281 static struct task_struct *create_workqueue_thread(struct workqueue_struct *wq,
 282                                                    int cpu)
 283 {
 284         struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
 285         struct task_struct *p;
 286
 287         spin_lock_init(&cwq->lock);
 288         cwq->wq = wq;
 289         cwq->thread = NULL;
 290         cwq->insert_sequence = 0;
 291         cwq->remove_sequence = 0;
 292         INIT_LIST_HEAD(&cwq->worklist);
 293         init_waitqueue_head(&cwq->more_work);
 294         init_waitqueue_head(&cwq->work_done);
 295
 296         if (is_single_threaded(wq))
 297                 p = kthread_create(worker_thread, cwq, "%s", wq->name);
 298         else
 299                 p = kthread_create(worker_thread, cwq, "%s/%d", wq->name, cpu);
 300         if (IS_ERR(p))
 301                 return NULL;
 302         cwq->thread = p;
 303         return p;
 304 }
 305
 306 struct workqueue_struct *__create_workqueue(const char *name,
 307                                             int singlethread)
 308 {
 309         int cpu, destroy = 0;
 310         struct workqueue_struct *wq;
 311         struct task_struct *p;
 312
 313         wq = kzalloc(sizeof(*wq), GFP_KERNEL);
 314         if (!wq)
 315                 return NULL;
 316
 317         wq->cpu_wq = alloc_percpu(struct cpu_workqueue_struct);
 318         wq->name = name;
 319         /* We don't need the distraction of CPUs appearing and vanishing. */
 320         lock_cpu_hotplug();
 321         if (singlethread) {
 322                 INIT_LIST_HEAD(&wq->list);
 323                 p = create_workqueue_thread(wq, 0);
 324                 if (!p)
 325                         destroy = 1;
 326                 else
 327                         wake_up_process(p);
 328         } else {
 329                 spin_lock(&workqueue_lock);
 330                 list_add(&wq->list, &workqueues);
 331                 spin_unlock(&workqueue_lock);
 332                 for_each_online_cpu(cpu) {
 333                         p = create_workqueue_thread(wq, cpu);
 334                         if (p) {
 335                                 kthread_bind(p, cpu);
 336                                 wake_up_process(p);
 337                         } else
 338                                 destroy = 1;
 339                 }
 340         }
 341         unlock_cpu_hotplug();
 342
 343         /*
 344          * Was there any error during startup? If yes then clean up:
 345          */
 346         if (destroy) {
 347                 destroy_workqueue(wq);
 348                 wq = NULL;
 349         }
 350         return wq;
 351 }
 352
 353 static void cleanup_workqueue_thread(struct workqueue_struct *wq, int cpu)
 354 {
 355         struct cpu_workqueue_struct *cwq;
 356         unsigned long flags;
 357         struct task_struct *p;
 358
 359         cwq = per_cpu_ptr(wq->cpu_wq, cpu);
 360         spin_lock_irqsave(&cwq->lock, flags);
 361         p = cwq->thread;
 362         cwq->thread = NULL;
 363         spin_unlock_irqrestore(&cwq->lock, flags);
 364         if (p)
 365                 kthread_stop(p);
 366 }
 367
 368 void destroy_workqueue(struct workqueue_struct *wq)
 369 {
 370         int cpu;
 371
 372         flush_workqueue(wq);
 373
 374         /* We don't need the distraction of CPUs appearing and vanishing. */
 375         lock_cpu_hotplug();
 376         if (is_single_threaded(wq))
 377                 cleanup_workqueue_thread(wq, 0);
 378         else {
 379                 for_each_online_cpu(cpu)
 380                         cleanup_workqueue_thread(wq, cpu);
 381                 spin_lock(&workqueue_lock);
 382                 list_del(&wq->list);
 383                 spin_unlock(&workqueue_lock);
 384         }
 385         unlock_cpu_hotplug();
 386         free_percpu(wq->cpu_wq);
 387         kfree(wq);
 388 }
 389
 390 static struct workqueue_struct *keventd_wq;
 391
 392 int fastcall schedule_work(struct work_struct *work)
 393 {
 394         return queue_work(keventd_wq, work);
 395 }
 396
 397 int fastcall schedule_delayed_work(struct work_struct *work, unsigned long delay)
 398 {
 399         return queue_delayed_work(keventd_wq, work, delay);
 400 }
 401
 402 int schedule_delayed_work_on(int cpu,
 403                         struct work_struct *work, unsigned long delay)
 404 {
 405         int ret = 0;
 406         struct timer_list *timer = &work->timer;
 407
 408         if (!test_and_set_bit(0, &work->pending)) {
 409                 BUG_ON(timer_pending(timer));
 410                 BUG_ON(!list_empty(&work->entry));
 411                 /* This stores keventd_wq for the moment, for the timer_fn */
 412                 work->wq_data = keventd_wq;
 413                 timer->expires = jiffies + delay;
 414                 timer->data = (unsigned long)work;
 415                 timer->function = delayed_work_timer_fn;
 416                 add_timer_on(timer, cpu);
 417                 ret = 1;
 418         }
 419         return ret;
 420 }
 421
 422 void flush_scheduled_work(void)
 423 {
 424         flush_workqueue(keventd_wq);
 425 }
 426
 427 /**
 428  * cancel_rearming_delayed_workqueue - reliably kill off a delayed
 429  *                      work whose handler rearms the delayed work.
 430  * @wq:   the controlling workqueue structure
 431  * @work: the delayed work struct
 432  */
 433 void cancel_rearming_delayed_workqueue(struct workqueue_struct *wq,
 434                                        struct work_struct *work)
 435 {
 436         while (!cancel_delayed_work(work))
 437                 flush_workqueue(wq);
 438 }
 439 EXPORT_SYMBOL(cancel_rearming_delayed_workqueue);
 440
 441 /**
 442  * cancel_rearming_delayed_work - reliably kill off a delayed keventd
 443  *                      work whose handler rearms the delayed work.
 444  * @work: the delayed work struct
 445  */
 446 void cancel_rearming_delayed_work(struct work_struct *work)
 447 {
 448         cancel_rearming_delayed_workqueue(keventd_wq, work);
 449 }
 450 EXPORT_SYMBOL(cancel_rearming_delayed_work);
 451
 452 int keventd_up(void)
 453 {
 454         return keventd_wq != NULL;
 455 }
 456
 457 int current_is_keventd(void)
 458 {
 459         struct cpu_workqueue_struct *cwq;
 460         int cpu = smp_processor_id();   /* preempt-safe: keventd is per-cpu */
 461         int ret = 0;
 462
 463         BUG_ON(!keventd_wq);
 464
 465         cwq = per_cpu_ptr(keventd_wq->cpu_wq, cpu);
 466         if (current == cwq->thread)
 467                 ret = 1;
 468
 469         return ret;
 470
 471 }
 472
 473 #ifdef CONFIG_HOTPLUG_CPU
 474 /* Take the work from this (downed) CPU. */
 475 static void take_over_work(struct workqueue_struct *wq, unsigned int cpu)
 476 {
 477         struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
 478         LIST_HEAD(list);
 479         struct work_struct *work;
 480
 481         spin_lock_irq(&cwq->lock);
 482         list_splice_init(&cwq->worklist, &list);
 483
 484         while (!list_empty(&list)) {
 485                 printk("Taking work for %s\n", wq->name);
 486                 work = list_entry(list.next,struct work_struct,entry);
 487                 list_del(&work->entry);
 488                 __queue_work(per_cpu_ptr(wq->cpu_wq, smp_processor_id()), work);
 489         }
 490         spin_unlock_irq(&cwq->lock);
 491 }
 492
 493 /* We're holding the cpucontrol mutex here */
 494 static int __devinit workqueue_cpu_callback(struct notifier_block *nfb,
 495                                   unsigned long action,
 496                                   void *hcpu)
 497 {
 498         unsigned int hotcpu = (unsigned long)hcpu;
 499         struct workqueue_struct *wq;
 500
 501         switch (action) {
 502         case CPU_UP_PREPARE:
 503                 /* Create a new workqueue thread for it. */
 504                 list_for_each_entry(wq, &workqueues, list) {
 505                         if (!create_workqueue_thread(wq, hotcpu)) {
 506                                 printk("workqueue for %i failed\n", hotcpu);
 507                                 return NOTIFY_BAD;
 508                         }
 509                 }
 510                 break;
 511
 512         case CPU_ONLINE:
 513                 /* Kick off worker threads. */
 514                 list_for_each_entry(wq, &workqueues, list) {
 515                         struct cpu_workqueue_struct *cwq;
 516
 517                         cwq = per_cpu_ptr(wq->cpu_wq, hotcpu);
 518                         kthread_bind(cwq->thread, hotcpu);
 519                         wake_up_process(cwq->thread);
 520                 }
 521                 break;
 522
 523         case CPU_UP_CANCELED:
 524                 list_for_each_entry(wq, &workqueues, list) {
 525                         /* Unbind so it can run. */
 526                         kthread_bind(per_cpu_ptr(wq->cpu_wq, hotcpu)->thread,
 527                                      any_online_cpu(cpu_online_map));
 528                         cleanup_workqueue_thread(wq, hotcpu);
 529                 }
 530                 break;
 531
 532         case CPU_DEAD:
 533                 list_for_each_entry(wq, &workqueues, list)
 534                         cleanup_workqueue_thread(wq, hotcpu);
 535                 list_for_each_entry(wq, &workqueues, list)
 536                         take_over_work(wq, hotcpu);
 537                 break;
 538         }
 539
 540         return NOTIFY_OK;
 541 }
 542 #endif
 543
 544 void init_workqueues(void)
 545 {
 546         hotcpu_notifier(workqueue_cpu_callback, 0);
 547         keventd_wq = create_workqueue("events");
 548         BUG_ON(!keventd_wq);
 549 }
 550
 551 EXPORT_SYMBOL_GPL(__create_workqueue);
 552 EXPORT_SYMBOL_GPL(queue_work);
 553 EXPORT_SYMBOL_GPL(queue_delayed_work);
 554 EXPORT_SYMBOL_GPL(flush_workqueue);
 555 EXPORT_SYMBOL_GPL(destroy_workqueue);
 556
 557 EXPORT_SYMBOL(schedule_work);
 558 EXPORT_SYMBOL(schedule_delayed_work);
 559 EXPORT_SYMBOL(schedule_delayed_work_on);
 560 EXPORT_SYMBOL(flush_scheduled_work);