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 struct cpu_workqueue_struct {
  41
  42         spinlock_t lock;
  43
  44         struct list_head worklist;
  45         wait_queue_head_t more_work;
  46         struct work_struct *current_work;
  47
  48         struct workqueue_struct *wq;
  49         struct task_struct *thread;
  50         int should_stop;
  51
  52         int run_depth;          /* Detect run_workqueue() recursion depth */
  53 } ____cacheline_aligned;
  54
  55 /*
  56  * The externally visible workqueue abstraction is an array of
  57  * per-CPU workqueues:
  58  */
  59 struct workqueue_struct {
  60         struct cpu_workqueue_struct *cpu_wq;
  61         struct list_head list;
  62         const char *name;
  63         int singlethread;
  64         int freezeable;         /* Freeze threads during suspend */
  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_MUTEX(workqueue_mutex);
  70 static LIST_HEAD(workqueues);
  71
  72 static int singlethread_cpu __read_mostly;
  73 static cpumask_t cpu_singlethread_map __read_mostly;
  74 /* optimization, we could use cpu_possible_map */
  75 static cpumask_t cpu_populated_map __read_mostly;
  76
  77 /* If it's single threaded, it isn't in the list of workqueues. */
  78 static inline int is_single_threaded(struct workqueue_struct *wq)
  79 {
  80         return wq->singlethread;
  81 }
  82
  83 static const cpumask_t *wq_cpu_map(struct workqueue_struct *wq)
  84 {
  85         return is_single_threaded(wq)
  86                 ? &cpu_singlethread_map : &cpu_populated_map;
  87 }
  88
  89 static
  90 struct cpu_workqueue_struct *wq_per_cpu(struct workqueue_struct *wq, int cpu)
  91 {
  92         if (unlikely(is_single_threaded(wq)))
  93                 cpu = singlethread_cpu;
  94         return per_cpu_ptr(wq->cpu_wq, cpu);
  95 }
  96
  97 /*
  98  * Set the workqueue on which a work item is to be run
  99  * - Must *only* be called if the pending flag is set
 100  */
 101 static inline void set_wq_data(struct work_struct *work,
 102                                 struct cpu_workqueue_struct *cwq)
 103 {
 104         unsigned long new;
 105
 106         BUG_ON(!work_pending(work));
 107
 108         new = (unsigned long) cwq | (1UL << WORK_STRUCT_PENDING);
 109         new |= WORK_STRUCT_FLAG_MASK & *work_data_bits(work);
 110         atomic_long_set(&work->data, new);
 111 }
 112
 113 static inline
 114 struct cpu_workqueue_struct *get_wq_data(struct work_struct *work)
 115 {
 116         return (void *) (atomic_long_read(&work->data) & WORK_STRUCT_WQ_DATA_MASK);
 117 }
 118
 119 static void insert_work(struct cpu_workqueue_struct *cwq,
 120                                 struct work_struct *work, int tail)
 121 {
 122         set_wq_data(work, cwq);
 123         /*
 124          * Ensure that we get the right work->data if we see the
 125          * result of list_add() below, see try_to_grab_pending().
 126          */
 127         smp_wmb();
 128         if (tail)
 129                 list_add_tail(&work->entry, &cwq->worklist);
 130         else
 131                 list_add(&work->entry, &cwq->worklist);
 132         wake_up(&cwq->more_work);
 133 }
 134
 135 /* Preempt must be disabled. */
 136 static void __queue_work(struct cpu_workqueue_struct *cwq,
 137                          struct work_struct *work)
 138 {
 139         unsigned long flags;
 140
 141         spin_lock_irqsave(&cwq->lock, flags);
 142         insert_work(cwq, work, 1);
 143         spin_unlock_irqrestore(&cwq->lock, flags);
 144 }
 145
 146 /**
 147  * queue_work - queue work on a workqueue
 148  * @wq: workqueue to use
 149  * @work: work to queue
 150  *
 151  * Returns 0 if @work was already on a queue, non-zero otherwise.
 152  *
 153  * We queue the work to the CPU it was submitted, but there is no
 154  * guarantee that it will be processed by that CPU.
 155  */
 156 int fastcall queue_work(struct workqueue_struct *wq, struct work_struct *work)
 157 {
 158         int ret = 0;
 159
 160         if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) {
 161                 BUG_ON(!list_empty(&work->entry));
 162                 __queue_work(wq_per_cpu(wq, get_cpu()), work);
 163                 put_cpu();
 164                 ret = 1;
 165         }
 166         return ret;
 167 }
 168 EXPORT_SYMBOL_GPL(queue_work);
 169
 170 void delayed_work_timer_fn(unsigned long __data)
 171 {
 172         struct delayed_work *dwork = (struct delayed_work *)__data;
 173         struct cpu_workqueue_struct *cwq = get_wq_data(&dwork->work);
 174         struct workqueue_struct *wq = cwq->wq;
 175
 176         __queue_work(wq_per_cpu(wq, smp_processor_id()), &dwork->work);
 177 }
 178
 179 /**
 180  * queue_delayed_work - queue work on a workqueue after delay
 181  * @wq: workqueue to use
 182  * @dwork: delayable work to queue
 183  * @delay: number of jiffies to wait before queueing
 184  *
 185  * Returns 0 if @work was already on a queue, non-zero otherwise.
 186  */
 187 int fastcall queue_delayed_work(struct workqueue_struct *wq,
 188                         struct delayed_work *dwork, unsigned long delay)
 189 {
 190         timer_stats_timer_set_start_info(&dwork->timer);
 191         if (delay == 0)
 192                 return queue_work(wq, &dwork->work);
 193
 194         return queue_delayed_work_on(-1, wq, dwork, delay);
 195 }
 196 EXPORT_SYMBOL_GPL(queue_delayed_work);
 197
 198 /**
 199  * queue_delayed_work_on - queue work on specific CPU after delay
 200  * @cpu: CPU number to execute work on
 201  * @wq: workqueue to use
 202  * @dwork: work to queue
 203  * @delay: number of jiffies to wait before queueing
 204  *
 205  * Returns 0 if @work was already on a queue, non-zero otherwise.
 206  */
 207 int queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
 208                         struct delayed_work *dwork, unsigned long delay)
 209 {
 210         int ret = 0;
 211         struct timer_list *timer = &dwork->timer;
 212         struct work_struct *work = &dwork->work;
 213
 214         if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) {
 215                 BUG_ON(timer_pending(timer));
 216                 BUG_ON(!list_empty(&work->entry));
 217
 218                 /* This stores cwq for the moment, for the timer_fn */
 219                 set_wq_data(work, wq_per_cpu(wq, raw_smp_processor_id()));
 220                 timer->expires = jiffies + delay;
 221                 timer->data = (unsigned long)dwork;
 222                 timer->function = delayed_work_timer_fn;
 223
 224                 if (unlikely(cpu >= 0))
 225                         add_timer_on(timer, cpu);
 226                 else
 227                         add_timer(timer);
 228                 ret = 1;
 229         }
 230         return ret;
 231 }
 232 EXPORT_SYMBOL_GPL(queue_delayed_work_on);
 233
 234 static void run_workqueue(struct cpu_workqueue_struct *cwq)
 235 {
 236         spin_lock_irq(&cwq->lock);
 237         cwq->run_depth++;
 238         if (cwq->run_depth > 3) {
 239                 /* morton gets to eat his hat */
 240                 printk("%s: recursion depth exceeded: %d\n",
 241                         __FUNCTION__, cwq->run_depth);
 242                 dump_stack();
 243         }
 244         while (!list_empty(&cwq->worklist)) {
 245                 struct work_struct *work = list_entry(cwq->worklist.next,
 246                                                 struct work_struct, entry);
 247                 work_func_t f = work->func;
 248
 249                 cwq->current_work = work;
 250                 list_del_init(cwq->worklist.next);
 251                 spin_unlock_irq(&cwq->lock);
 252
 253                 BUG_ON(get_wq_data(work) != cwq);
 254                 work_clear_pending(work);
 255                 f(work);
 256
 257                 if (unlikely(in_atomic() || lockdep_depth(current) > 0)) {
 258                         printk(KERN_ERR "BUG: workqueue leaked lock or atomic: "
 259                                         "%s/0x%08x/%d\n",
 260                                         current->comm, preempt_count(),
 261                                         current->pid);
 262                         printk(KERN_ERR "    last function: ");
 263                         print_symbol("%s\n", (unsigned long)f);
 264                         debug_show_held_locks(current);
 265                         dump_stack();
 266                 }
 267
 268                 spin_lock_irq(&cwq->lock);
 269                 cwq->current_work = NULL;
 270         }
 271         cwq->run_depth--;
 272         spin_unlock_irq(&cwq->lock);
 273 }
 274
 275 /*
 276  * NOTE: the caller must not touch *cwq if this func returns true
 277  */
 278 static int cwq_should_stop(struct cpu_workqueue_struct *cwq)
 279 {
 280         int should_stop = cwq->should_stop;
 281
 282         if (unlikely(should_stop)) {
 283                 spin_lock_irq(&cwq->lock);
 284                 should_stop = cwq->should_stop && list_empty(&cwq->worklist);
 285                 if (should_stop)
 286                         cwq->thread = NULL;
 287                 spin_unlock_irq(&cwq->lock);
 288         }
 289
 290         return should_stop;
 291 }
 292
 293 static int worker_thread(void *__cwq)
 294 {
 295         struct cpu_workqueue_struct *cwq = __cwq;
 296         DEFINE_WAIT(wait);
 297
 298         if (!cwq->wq->freezeable)
 299                 current->flags |= PF_NOFREEZE;
 300
 301         set_user_nice(current, -5);
 302
 303         for (;;) {
 304                 prepare_to_wait(&cwq->more_work, &wait, TASK_INTERRUPTIBLE);
 305                 if (!freezing(current) && !cwq->should_stop
 306                     && list_empty(&cwq->worklist))
 307                         schedule();
 308                 finish_wait(&cwq->more_work, &wait);
 309
 310                 try_to_freeze();
 311
 312                 if (cwq_should_stop(cwq))
 313                         break;
 314
 315                 run_workqueue(cwq);
 316         }
 317
 318         return 0;
 319 }
 320
 321 struct wq_barrier {
 322         struct work_struct      work;
 323         struct completion       done;
 324 };
 325
 326 static void wq_barrier_func(struct work_struct *work)
 327 {
 328         struct wq_barrier *barr = container_of(work, struct wq_barrier, work);
 329         complete(&barr->done);
 330 }
 331
 332 static void insert_wq_barrier(struct cpu_workqueue_struct *cwq,
 333                                         struct wq_barrier *barr, int tail)
 334 {
 335         INIT_WORK(&barr->work, wq_barrier_func);
 336         __set_bit(WORK_STRUCT_PENDING, work_data_bits(&barr->work));
 337
 338         init_completion(&barr->done);
 339
 340         insert_work(cwq, &barr->work, tail);
 341 }
 342
 343 static void flush_cpu_workqueue(struct cpu_workqueue_struct *cwq)
 344 {
 345         if (cwq->thread == current) {
 346                 /*
 347                  * Probably keventd trying to flush its own queue. So simply run
 348                  * it by hand rather than deadlocking.
 349                  */
 350                 run_workqueue(cwq);
 351         } else {
 352                 struct wq_barrier barr;
 353                 int active = 0;
 354
 355                 spin_lock_irq(&cwq->lock);
 356                 if (!list_empty(&cwq->worklist) || cwq->current_work != NULL) {
 357                         insert_wq_barrier(cwq, &barr, 1);
 358                         active = 1;
 359                 }
 360                 spin_unlock_irq(&cwq->lock);
 361
 362                 if (active)
 363                         wait_for_completion(&barr.done);
 364         }
 365 }
 366
 367 /**
 368  * flush_workqueue - ensure that any scheduled work has run to completion.
 369  * @wq: workqueue to flush
 370  *
 371  * Forces execution of the workqueue and blocks until its completion.
 372  * This is typically used in driver shutdown handlers.
 373  *
 374  * We sleep until all works which were queued on entry have been handled,
 375  * but we are not livelocked by new incoming ones.
 376  *
 377  * This function used to run the workqueues itself.  Now we just wait for the
 378  * helper threads to do it.
 379  */
 380 void fastcall flush_workqueue(struct workqueue_struct *wq)
 381 {
 382         const cpumask_t *cpu_map = wq_cpu_map(wq);
 383         int cpu;
 384
 385         might_sleep();
 386         for_each_cpu_mask(cpu, *cpu_map)
 387                 flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, cpu));
 388 }
 389 EXPORT_SYMBOL_GPL(flush_workqueue);
 390
 391 /*
 392  * Upon a successful return, the caller "owns" WORK_STRUCT_PENDING bit,
 393  * so this work can't be re-armed in any way.
 394  */
 395 static int try_to_grab_pending(struct work_struct *work)
 396 {
 397         struct cpu_workqueue_struct *cwq;
 398         int ret = 0;
 399
 400         if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work)))
 401                 return 1;
 402
 403         /*
 404          * The queueing is in progress, or it is already queued. Try to
 405          * steal it from ->worklist without clearing WORK_STRUCT_PENDING.
 406          */
 407
 408         cwq = get_wq_data(work);
 409         if (!cwq)
 410                 return ret;
 411
 412         spin_lock_irq(&cwq->lock);
 413         if (!list_empty(&work->entry)) {
 414                 /*
 415                  * This work is queued, but perhaps we locked the wrong cwq.
 416                  * In that case we must see the new value after rmb(), see
 417                  * insert_work()->wmb().
 418                  */
 419                 smp_rmb();
 420                 if (cwq == get_wq_data(work)) {
 421                         list_del_init(&work->entry);
 422                         ret = 1;
 423                 }
 424         }
 425         spin_unlock_irq(&cwq->lock);
 426
 427         return ret;
 428 }
 429
 430 static void wait_on_cpu_work(struct cpu_workqueue_struct *cwq,
 431                                 struct work_struct *work)
 432 {
 433         struct wq_barrier barr;
 434         int running = 0;
 435
 436         spin_lock_irq(&cwq->lock);
 437         if (unlikely(cwq->current_work == work)) {
 438                 insert_wq_barrier(cwq, &barr, 0);
 439                 running = 1;
 440         }
 441         spin_unlock_irq(&cwq->lock);
 442
 443         if (unlikely(running))
 444                 wait_for_completion(&barr.done);
 445 }
 446
 447 static void wait_on_work(struct work_struct *work)
 448 {
 449         struct cpu_workqueue_struct *cwq;
 450         struct workqueue_struct *wq;
 451         const cpumask_t *cpu_map;
 452         int cpu;
 453
 454         might_sleep();
 455
 456         cwq = get_wq_data(work);
 457         if (!cwq)
 458                 return;
 459
 460         wq = cwq->wq;
 461         cpu_map = wq_cpu_map(wq);
 462
 463         for_each_cpu_mask(cpu, *cpu_map)
 464                 wait_on_cpu_work(per_cpu_ptr(wq->cpu_wq, cpu), work);
 465 }
 466
 467 /**
 468  * cancel_work_sync - block until a work_struct's callback has terminated
 469  * @work: the work which is to be flushed
 470  *
 471  * cancel_work_sync() will cancel the work if it is queued. If the work's
 472  * callback appears to be running, cancel_work_sync() will block until it
 473  * has completed.
 474  *
 475  * It is possible to use this function if the work re-queues itself. It can
 476  * cancel the work even if it migrates to another workqueue, however in that
 477  * case it only guarantees that work->func() has completed on the last queued
 478  * workqueue.
 479  *
 480  * cancel_work_sync(&delayed_work->work) should be used only if ->timer is not
 481  * pending, otherwise it goes into a busy-wait loop until the timer expires.
 482  *
 483  * The caller must ensure that workqueue_struct on which this work was last
 484  * queued can't be destroyed before this function returns.
 485  */
 486 void cancel_work_sync(struct work_struct *work)
 487 {
 488         while (!try_to_grab_pending(work))
 489                 cpu_relax();
 490         wait_on_work(work);
 491         work_clear_pending(work);
 492 }
 493 EXPORT_SYMBOL_GPL(cancel_work_sync);
 494
 495 /**
 496  * cancel_rearming_delayed_work - reliably kill off a delayed work.
 497  * @dwork: the delayed work struct
 498  *
 499  * It is possible to use this function if @dwork rearms itself via queue_work()
 500  * or queue_delayed_work(). See also the comment for cancel_work_sync().
 501  */
 502 void cancel_rearming_delayed_work(struct delayed_work *dwork)
 503 {
 504         while (!del_timer(&dwork->timer) &&
 505                !try_to_grab_pending(&dwork->work))
 506                 cpu_relax();
 507         wait_on_work(&dwork->work);
 508         work_clear_pending(&dwork->work);
 509 }
 510 EXPORT_SYMBOL(cancel_rearming_delayed_work);
 511
 512 static struct workqueue_struct *keventd_wq __read_mostly;
 513
 514 /**
 515  * schedule_work - put work task in global workqueue
 516  * @work: job to be done
 517  *
 518  * This puts a job in the kernel-global workqueue.
 519  */
 520 int fastcall schedule_work(struct work_struct *work)
 521 {
 522         return queue_work(keventd_wq, work);
 523 }
 524 EXPORT_SYMBOL(schedule_work);
 525
 526 /**
 527  * schedule_delayed_work - put work task in global workqueue after delay
 528  * @dwork: job to be done
 529  * @delay: number of jiffies to wait or 0 for immediate execution
 530  *
 531  * After waiting for a given time this puts a job in the kernel-global
 532  * workqueue.
 533  */
 534 int fastcall schedule_delayed_work(struct delayed_work *dwork,
 535                                         unsigned long delay)
 536 {
 537         timer_stats_timer_set_start_info(&dwork->timer);
 538         return queue_delayed_work(keventd_wq, dwork, delay);
 539 }
 540 EXPORT_SYMBOL(schedule_delayed_work);
 541
 542 /**
 543  * schedule_delayed_work_on - queue work in global workqueue on CPU after delay
 544  * @cpu: cpu to use
 545  * @dwork: job to be done
 546  * @delay: number of jiffies to wait
 547  *
 548  * After waiting for a given time this puts a job in the kernel-global
 549  * workqueue on the specified CPU.
 550  */
 551 int schedule_delayed_work_on(int cpu,
 552                         struct delayed_work *dwork, unsigned long delay)
 553 {
 554         return queue_delayed_work_on(cpu, keventd_wq, dwork, delay);
 555 }
 556 EXPORT_SYMBOL(schedule_delayed_work_on);
 557
 558 /**
 559  * schedule_on_each_cpu - call a function on each online CPU from keventd
 560  * @func: the function to call
 561  *
 562  * Returns zero on success.
 563  * Returns -ve errno on failure.
 564  *
 565  * Appears to be racy against CPU hotplug.
 566  *
 567  * schedule_on_each_cpu() is very slow.
 568  */
 569 int schedule_on_each_cpu(work_func_t func)
 570 {
 571         int cpu;
 572         struct work_struct *works;
 573
 574         works = alloc_percpu(struct work_struct);
 575         if (!works)
 576                 return -ENOMEM;
 577
 578         preempt_disable();              /* CPU hotplug */
 579         for_each_online_cpu(cpu) {
 580                 struct work_struct *work = per_cpu_ptr(works, cpu);
 581
 582                 INIT_WORK(work, func);
 583                 set_bit(WORK_STRUCT_PENDING, work_data_bits(work));
 584                 __queue_work(per_cpu_ptr(keventd_wq->cpu_wq, cpu), work);
 585         }
 586         preempt_enable();
 587         flush_workqueue(keventd_wq);
 588         free_percpu(works);
 589         return 0;
 590 }
 591
 592 void flush_scheduled_work(void)
 593 {
 594         flush_workqueue(keventd_wq);
 595 }
 596 EXPORT_SYMBOL(flush_scheduled_work);
 597
 598 /**
 599  * execute_in_process_context - reliably execute the routine with user context
 600  * @fn:         the function to execute
 601  * @ew:         guaranteed storage for the execute work structure (must
 602  *              be available when the work executes)
 603  *
 604  * Executes the function immediately if process context is available,
 605  * otherwise schedules the function for delayed execution.
 606  *
 607  * Returns:     0 - function was executed
 608  *              1 - function was scheduled for execution
 609  */
 610 int execute_in_process_context(work_func_t fn, struct execute_work *ew)
 611 {
 612         if (!in_interrupt()) {
 613                 fn(&ew->work);
 614                 return 0;
 615         }
 616
 617         INIT_WORK(&ew->work, fn);
 618         schedule_work(&ew->work);
 619
 620         return 1;
 621 }
 622 EXPORT_SYMBOL_GPL(execute_in_process_context);
 623
 624 int keventd_up(void)
 625 {
 626         return keventd_wq != NULL;
 627 }
 628
 629 int current_is_keventd(void)
 630 {
 631         struct cpu_workqueue_struct *cwq;
 632         int cpu = smp_processor_id();   /* preempt-safe: keventd is per-cpu */
 633         int ret = 0;
 634
 635         BUG_ON(!keventd_wq);
 636
 637         cwq = per_cpu_ptr(keventd_wq->cpu_wq, cpu);
 638         if (current == cwq->thread)
 639                 ret = 1;
 640
 641         return ret;
 642
 643 }
 644
 645 static struct cpu_workqueue_struct *
 646 init_cpu_workqueue(struct workqueue_struct *wq, int cpu)
 647 {
 648         struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
 649
 650         cwq->wq = wq;
 651         spin_lock_init(&cwq->lock);
 652         INIT_LIST_HEAD(&cwq->worklist);
 653         init_waitqueue_head(&cwq->more_work);
 654
 655         return cwq;
 656 }
 657
 658 static int create_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu)
 659 {
 660         struct workqueue_struct *wq = cwq->wq;
 661         const char *fmt = is_single_threaded(wq) ? "%s" : "%s/%d";
 662         struct task_struct *p;
 663
 664         p = kthread_create(worker_thread, cwq, fmt, wq->name, cpu);
 665         /*
 666          * Nobody can add the work_struct to this cwq,
 667          *      if (caller is __create_workqueue)
 668          *              nobody should see this wq
 669          *      else // caller is CPU_UP_PREPARE
 670          *              cpu is not on cpu_online_map
 671          * so we can abort safely.
 672          */
 673         if (IS_ERR(p))
 674                 return PTR_ERR(p);
 675
 676         cwq->thread = p;
 677         cwq->should_stop = 0;
 678
 679         return 0;
 680 }
 681
 682 static void start_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu)
 683 {
 684         struct task_struct *p = cwq->thread;
 685
 686         if (p != NULL) {
 687                 if (cpu >= 0)
 688                         kthread_bind(p, cpu);
 689                 wake_up_process(p);
 690         }
 691 }
 692
 693 struct workqueue_struct *__create_workqueue(const char *name,
 694                                             int singlethread, int freezeable)
 695 {
 696         struct workqueue_struct *wq;
 697         struct cpu_workqueue_struct *cwq;
 698         int err = 0, cpu;
 699
 700         wq = kzalloc(sizeof(*wq), GFP_KERNEL);
 701         if (!wq)
 702                 return NULL;
 703
 704         wq->cpu_wq = alloc_percpu(struct cpu_workqueue_struct);
 705         if (!wq->cpu_wq) {
 706                 kfree(wq);
 707                 return NULL;
 708         }
 709
 710         wq->name = name;
 711         wq->singlethread = singlethread;
 712         wq->freezeable = freezeable;
 713         INIT_LIST_HEAD(&wq->list);
 714
 715         if (singlethread) {
 716                 cwq = init_cpu_workqueue(wq, singlethread_cpu);
 717                 err = create_workqueue_thread(cwq, singlethread_cpu);
 718                 start_workqueue_thread(cwq, -1);
 719         } else {
 720                 mutex_lock(&workqueue_mutex);
 721                 list_add(&wq->list, &workqueues);
 722
 723                 for_each_possible_cpu(cpu) {
 724                         cwq = init_cpu_workqueue(wq, cpu);
 725                         if (err || !cpu_online(cpu))
 726                                 continue;
 727                         err = create_workqueue_thread(cwq, cpu);
 728                         start_workqueue_thread(cwq, cpu);
 729                 }
 730                 mutex_unlock(&workqueue_mutex);
 731         }
 732
 733         if (err) {
 734                 destroy_workqueue(wq);
 735                 wq = NULL;
 736         }
 737         return wq;
 738 }
 739 EXPORT_SYMBOL_GPL(__create_workqueue);
 740
 741 static void cleanup_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu)
 742 {
 743         struct wq_barrier barr;
 744         int alive = 0;
 745
 746         spin_lock_irq(&cwq->lock);
 747         if (cwq->thread != NULL) {
 748                 insert_wq_barrier(cwq, &barr, 1);
 749                 cwq->should_stop = 1;
 750                 alive = 1;
 751         }
 752         spin_unlock_irq(&cwq->lock);
 753
 754         if (alive) {
 755                 wait_for_completion(&barr.done);
 756
 757                 while (unlikely(cwq->thread != NULL))
 758                         cpu_relax();
 759                 /*
 760                  * Wait until cwq->thread unlocks cwq->lock,
 761                  * it won't touch *cwq after that.
 762                  */
 763                 smp_rmb();
 764                 spin_unlock_wait(&cwq->lock);
 765         }
 766 }
 767
 768 /**
 769  * destroy_workqueue - safely terminate a workqueue
 770  * @wq: target workqueue
 771  *
 772  * Safely destroy a workqueue. All work currently pending will be done first.
 773  */
 774 void destroy_workqueue(struct workqueue_struct *wq)
 775 {
 776         const cpumask_t *cpu_map = wq_cpu_map(wq);
 777         struct cpu_workqueue_struct *cwq;
 778         int cpu;
 779
 780         mutex_lock(&workqueue_mutex);
 781         list_del(&wq->list);
 782         mutex_unlock(&workqueue_mutex);
 783
 784         for_each_cpu_mask(cpu, *cpu_map) {
 785                 cwq = per_cpu_ptr(wq->cpu_wq, cpu);
 786                 cleanup_workqueue_thread(cwq, cpu);
 787         }
 788
 789         free_percpu(wq->cpu_wq);
 790         kfree(wq);
 791 }
 792 EXPORT_SYMBOL_GPL(destroy_workqueue);
 793
 794 static int __devinit workqueue_cpu_callback(struct notifier_block *nfb,
 795                                                 unsigned long action,
 796                                                 void *hcpu)
 797 {
 798         unsigned int cpu = (unsigned long)hcpu;
 799         struct cpu_workqueue_struct *cwq;
 800         struct workqueue_struct *wq;
 801
 802         action &= ~CPU_TASKS_FROZEN;
 803
 804         switch (action) {
 805         case CPU_LOCK_ACQUIRE:
 806                 mutex_lock(&workqueue_mutex);
 807                 return NOTIFY_OK;
 808
 809         case CPU_LOCK_RELEASE:
 810                 mutex_unlock(&workqueue_mutex);
 811                 return NOTIFY_OK;
 812
 813         case CPU_UP_PREPARE:
 814                 cpu_set(cpu, cpu_populated_map);
 815         }
 816
 817         list_for_each_entry(wq, &workqueues, list) {
 818                 cwq = per_cpu_ptr(wq->cpu_wq, cpu);
 819
 820                 switch (action) {
 821                 case CPU_UP_PREPARE:
 822                         if (!create_workqueue_thread(cwq, cpu))
 823                                 break;
 824                         printk(KERN_ERR "workqueue for %i failed\n", cpu);
 825                         return NOTIFY_BAD;
 826
 827                 case CPU_ONLINE:
 828                         start_workqueue_thread(cwq, cpu);
 829                         break;
 830
 831                 case CPU_UP_CANCELED:
 832                         start_workqueue_thread(cwq, -1);
 833                 case CPU_DEAD:
 834                         cleanup_workqueue_thread(cwq, cpu);
 835                         break;
 836                 }
 837         }
 838
 839         return NOTIFY_OK;
 840 }
 841
 842 void __init init_workqueues(void)
 843 {
 844         cpu_populated_map = cpu_online_map;
 845         singlethread_cpu = first_cpu(cpu_possible_map);
 846         cpu_singlethread_map = cpumask_of_cpu(singlethread_cpu);
 847         hotcpu_notifier(workqueue_cpu_callback, 0);
 848         keventd_wq = create_workqueue("events");
 849         BUG_ON(!keventd_wq);
 850 }