]> err.no Git - linux-2.6/blobdiff - kernel/workqueue.c
workqueue: introduce cpu_singlethread_map
[linux-2.6] / kernel / workqueue.c
index db49886bfae17271ee19709a984c57c208c5065b..6308a4bc6a821f654bedbab4fefb1999d67cfa5e 100644 (file)
 /*
  * The per-CPU workqueue (if single thread, we always use the first
  * possible cpu).
- *
- * The sequence counters are for flush_scheduled_work().  It wants to wait
- * until all currently-scheduled works are completed, but it doesn't
- * want to be livelocked by new, incoming ones.  So it waits until
- * remove_sequence is >= the insert_sequence which pertained when
- * flush_scheduled_work() was called.
  */
 struct cpu_workqueue_struct {
 
        spinlock_t lock;
 
-       long remove_sequence;   /* Least-recently added (next to run) */
-       long insert_sequence;   /* Next to add */
-
        struct list_head worklist;
        wait_queue_head_t more_work;
-       wait_queue_head_t work_done;
+       struct work_struct *current_work;
 
        struct workqueue_struct *wq;
        struct task_struct *thread;
+       int should_stop;
 
        int run_depth;          /* Detect run_workqueue() recursion depth */
-
-       int freezeable;         /* Freeze the thread during suspend */
 } ____cacheline_aligned;
 
 /*
@@ -70,6 +60,7 @@ struct workqueue_struct {
        struct cpu_workqueue_struct *cpu_wq;
        const char *name;
        struct list_head list;  /* Empty if single thread */
+       int freezeable;         /* Freeze threads during suspend */
 };
 
 /* All the per-cpu workqueues on the system, for hotplug cpu to add/remove
@@ -77,7 +68,10 @@ struct workqueue_struct {
 static DEFINE_MUTEX(workqueue_mutex);
 static LIST_HEAD(workqueues);
 
-static int singlethread_cpu;
+static int singlethread_cpu __read_mostly;
+static cpumask_t cpu_singlethread_map __read_mostly;
+/* optimization, we could use cpu_possible_map */
+static cpumask_t cpu_populated_map __read_mostly;
 
 /* If it's single threaded, it isn't in the list of workqueues. */
 static inline int is_single_threaded(struct workqueue_struct *wq)
@@ -85,6 +79,12 @@ static inline int is_single_threaded(struct workqueue_struct *wq)
        return list_empty(&wq->list);
 }
 
+static const cpumask_t *wq_cpu_map(struct workqueue_struct *wq)
+{
+       return is_single_threaded(wq)
+               ? &cpu_singlethread_map : &cpu_populated_map;
+}
+
 /*
  * Set the workqueue on which a work item is to be run
  * - Must *only* be called if the pending flag is set
@@ -96,87 +96,25 @@ static inline void set_wq_data(struct work_struct *work, void *wq)
        BUG_ON(!work_pending(work));
 
        new = (unsigned long) wq | (1UL << WORK_STRUCT_PENDING);
-       new |= work->management & WORK_STRUCT_FLAG_MASK;
-       work->management = new;
+       new |= WORK_STRUCT_FLAG_MASK & *work_data_bits(work);
+       atomic_long_set(&work->data, new);
 }
 
 static inline void *get_wq_data(struct work_struct *work)
 {
-       return (void *) (work->management & WORK_STRUCT_WQ_DATA_MASK);
+       return (void *) (atomic_long_read(&work->data) & WORK_STRUCT_WQ_DATA_MASK);
 }
 
-static int __run_work(struct cpu_workqueue_struct *cwq, struct work_struct *work)
+static void insert_work(struct cpu_workqueue_struct *cwq,
+                               struct work_struct *work, int tail)
 {
-       int ret = 0;
-       unsigned long flags;
-
-       spin_lock_irqsave(&cwq->lock, flags);
-       /*
-        * We need to re-validate the work info after we've gotten
-        * the cpu_workqueue lock. We can run the work now iff:
-        *
-        *  - the wq_data still matches the cpu_workqueue_struct
-        *  - AND the work is still marked pending
-        *  - AND the work is still on a list (which will be this
-        *    workqueue_struct list)
-        *
-        * All these conditions are important, because we
-        * need to protect against the work being run right
-        * now on another CPU (all but the last one might be
-        * true if it's currently running and has not been
-        * released yet, for example).
-        */
-       if (get_wq_data(work) == cwq
-           && work_pending(work)
-           && !list_empty(&work->entry)) {
-               work_func_t f = work->func;
-               list_del_init(&work->entry);
-               spin_unlock_irqrestore(&cwq->lock, flags);
-
-               if (!test_bit(WORK_STRUCT_NOAUTOREL, &work->management))
-                       work_release(work);
-               f(work);
-
-               spin_lock_irqsave(&cwq->lock, flags);
-               cwq->remove_sequence++;
-               wake_up(&cwq->work_done);
-               ret = 1;
-       }
-       spin_unlock_irqrestore(&cwq->lock, flags);
-       return ret;
-}
-
-/**
- * run_scheduled_work - run scheduled work synchronously
- * @work: work to run
- *
- * This checks if the work was pending, and runs it
- * synchronously if so. It returns a boolean to indicate
- * whether it had any scheduled work to run or not.
- *
- * NOTE! This _only_ works for normal work_structs. You
- * CANNOT use this for delayed work, because the wq data
- * for delayed work will not point properly to the per-
- * CPU workqueue struct, but will change!
- */
-int fastcall run_scheduled_work(struct work_struct *work)
-{
-       for (;;) {
-               struct cpu_workqueue_struct *cwq;
-
-               if (!work_pending(work))
-                       return 0;
-               if (list_empty(&work->entry))
-                       return 0;
-               /* NOTE! This depends intimately on __queue_work! */
-               cwq = get_wq_data(work);
-               if (!cwq)
-                       return 0;
-               if (__run_work(cwq, work))
-                       return 1;
-       }
+       set_wq_data(work, cwq);
+       if (tail)
+               list_add_tail(&work->entry, &cwq->worklist);
+       else
+               list_add(&work->entry, &cwq->worklist);
+       wake_up(&cwq->more_work);
 }
-EXPORT_SYMBOL(run_scheduled_work);
 
 /* Preempt must be disabled. */
 static void __queue_work(struct cpu_workqueue_struct *cwq,
@@ -185,10 +123,7 @@ static void __queue_work(struct cpu_workqueue_struct *cwq,
        unsigned long flags;
 
        spin_lock_irqsave(&cwq->lock, flags);
-       set_wq_data(work, cwq);
-       list_add_tail(&work->entry, &cwq->worklist);
-       cwq->insert_sequence++;
-       wake_up(&cwq->more_work);
+       insert_work(cwq, work, 1);
        spin_unlock_irqrestore(&cwq->lock, flags);
 }
 
@@ -206,7 +141,7 @@ int fastcall queue_work(struct workqueue_struct *wq, struct work_struct *work)
 {
        int ret = 0, cpu = get_cpu();
 
-       if (!test_and_set_bit(WORK_STRUCT_PENDING, &work->management)) {
+       if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) {
                if (unlikely(is_single_threaded(wq)))
                        cpu = singlethread_cpu;
                BUG_ON(!list_empty(&work->entry));
@@ -218,7 +153,7 @@ int fastcall queue_work(struct workqueue_struct *wq, struct work_struct *work)
 }
 EXPORT_SYMBOL_GPL(queue_work);
 
-static void delayed_work_timer_fn(unsigned long __data)
+void delayed_work_timer_fn(unsigned long __data)
 {
        struct delayed_work *dwork = (struct delayed_work *)__data;
        struct workqueue_struct *wq = get_wq_data(&dwork->work);
@@ -233,7 +168,7 @@ static void delayed_work_timer_fn(unsigned long __data)
 /**
  * queue_delayed_work - queue work on a workqueue after delay
  * @wq: workqueue to use
- * @work: delayable work to queue
+ * @dwork: delayable work to queue
  * @delay: number of jiffies to wait before queueing
  *
  * Returns 0 if @work was already on a queue, non-zero otherwise.
@@ -245,10 +180,11 @@ int fastcall queue_delayed_work(struct workqueue_struct *wq,
        struct timer_list *timer = &dwork->timer;
        struct work_struct *work = &dwork->work;
 
+       timer_stats_timer_set_start_info(timer);
        if (delay == 0)
                return queue_work(wq, work);
 
-       if (!test_and_set_bit(WORK_STRUCT_PENDING, &work->management)) {
+       if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) {
                BUG_ON(timer_pending(timer));
                BUG_ON(!list_empty(&work->entry));
 
@@ -268,7 +204,7 @@ EXPORT_SYMBOL_GPL(queue_delayed_work);
  * queue_delayed_work_on - queue work on specific CPU after delay
  * @cpu: CPU number to execute work on
  * @wq: workqueue to use
- * @work: work to queue
+ * @dwork: work to queue
  * @delay: number of jiffies to wait before queueing
  *
  * Returns 0 if @work was already on a queue, non-zero otherwise.
@@ -280,7 +216,7 @@ int queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
        struct timer_list *timer = &dwork->timer;
        struct work_struct *work = &dwork->work;
 
-       if (!test_and_set_bit(WORK_STRUCT_PENDING, &work->management)) {
+       if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) {
                BUG_ON(timer_pending(timer));
                BUG_ON(!list_empty(&work->entry));
 
@@ -298,13 +234,7 @@ EXPORT_SYMBOL_GPL(queue_delayed_work_on);
 
 static void run_workqueue(struct cpu_workqueue_struct *cwq)
 {
-       unsigned long flags;
-
-       /*
-        * Keep taking off work from the queue until
-        * done.
-        */
-       spin_lock_irqsave(&cwq->lock, flags);
+       spin_lock_irq(&cwq->lock);
        cwq->run_depth++;
        if (cwq->run_depth > 3) {
                /* morton gets to eat his hat */
@@ -317,11 +247,12 @@ static void run_workqueue(struct cpu_workqueue_struct *cwq)
                                                struct work_struct, entry);
                work_func_t f = work->func;
 
+               cwq->current_work = work;
                list_del_init(cwq->worklist.next);
-               spin_unlock_irqrestore(&cwq->lock, flags);
+               spin_unlock_irq(&cwq->lock);
 
                BUG_ON(get_wq_data(work) != cwq);
-               if (!test_bit(WORK_STRUCT_NOAUTOREL, &work->management))
+               if (!test_bit(WORK_STRUCT_NOAUTOREL, work_data_bits(work)))
                        work_release(work);
                f(work);
 
@@ -336,22 +267,39 @@ static void run_workqueue(struct cpu_workqueue_struct *cwq)
                        dump_stack();
                }
 
-               spin_lock_irqsave(&cwq->lock, flags);
-               cwq->remove_sequence++;
-               wake_up(&cwq->work_done);
+               spin_lock_irq(&cwq->lock);
+               cwq->current_work = NULL;
        }
        cwq->run_depth--;
-       spin_unlock_irqrestore(&cwq->lock, flags);
+       spin_unlock_irq(&cwq->lock);
+}
+
+/*
+ * NOTE: the caller must not touch *cwq if this func returns true
+ */
+static int cwq_should_stop(struct cpu_workqueue_struct *cwq)
+{
+       int should_stop = cwq->should_stop;
+
+       if (unlikely(should_stop)) {
+               spin_lock_irq(&cwq->lock);
+               should_stop = cwq->should_stop && list_empty(&cwq->worklist);
+               if (should_stop)
+                       cwq->thread = NULL;
+               spin_unlock_irq(&cwq->lock);
+       }
+
+       return should_stop;
 }
 
 static int worker_thread(void *__cwq)
 {
        struct cpu_workqueue_struct *cwq = __cwq;
-       DECLARE_WAITQUEUE(wait, current);
+       DEFINE_WAIT(wait);
        struct k_sigaction sa;
        sigset_t blocked;
 
-       if (!cwq->freezeable)
+       if (!cwq->wq->freezeable)
                current->flags |= PF_NOFREEZE;
 
        set_user_nice(current, -5);
@@ -373,26 +321,46 @@ static int worker_thread(void *__cwq)
        siginitset(&sa.sa.sa_mask, sigmask(SIGCHLD));
        do_sigaction(SIGCHLD, &sa, (struct k_sigaction *)0);
 
-       set_current_state(TASK_INTERRUPTIBLE);
-       while (!kthread_should_stop()) {
-               if (cwq->freezeable)
+       for (;;) {
+               if (cwq->wq->freezeable)
                        try_to_freeze();
 
-               add_wait_queue(&cwq->more_work, &wait);
-               if (list_empty(&cwq->worklist))
+               prepare_to_wait(&cwq->more_work, &wait, TASK_INTERRUPTIBLE);
+               if (!cwq->should_stop && list_empty(&cwq->worklist))
                        schedule();
-               else
-                       __set_current_state(TASK_RUNNING);
-               remove_wait_queue(&cwq->more_work, &wait);
+               finish_wait(&cwq->more_work, &wait);
+
+               if (cwq_should_stop(cwq))
+                       break;
 
-               if (!list_empty(&cwq->worklist))
-                       run_workqueue(cwq);
-               set_current_state(TASK_INTERRUPTIBLE);
+               run_workqueue(cwq);
        }
-       __set_current_state(TASK_RUNNING);
+
        return 0;
 }
 
+struct wq_barrier {
+       struct work_struct      work;
+       struct completion       done;
+};
+
+static void wq_barrier_func(struct work_struct *work)
+{
+       struct wq_barrier *barr = container_of(work, struct wq_barrier, work);
+       complete(&barr->done);
+}
+
+static void insert_wq_barrier(struct cpu_workqueue_struct *cwq,
+                                       struct wq_barrier *barr, int tail)
+{
+       INIT_WORK(&barr->work, wq_barrier_func);
+       __set_bit(WORK_STRUCT_PENDING, work_data_bits(&barr->work));
+
+       init_completion(&barr->done);
+
+       insert_work(cwq, &barr->work, tail);
+}
+
 static void flush_cpu_workqueue(struct cpu_workqueue_struct *cwq)
 {
        if (cwq->thread == current) {
@@ -402,21 +370,18 @@ static void flush_cpu_workqueue(struct cpu_workqueue_struct *cwq)
                 */
                run_workqueue(cwq);
        } else {
-               DEFINE_WAIT(wait);
-               long sequence_needed;
+               struct wq_barrier barr;
+               int active = 0;
 
                spin_lock_irq(&cwq->lock);
-               sequence_needed = cwq->insert_sequence;
-
-               while (sequence_needed - cwq->remove_sequence > 0) {
-                       prepare_to_wait(&cwq->work_done, &wait,
-                                       TASK_UNINTERRUPTIBLE);
-                       spin_unlock_irq(&cwq->lock);
-                       schedule();
-                       spin_lock_irq(&cwq->lock);
+               if (!list_empty(&cwq->worklist) || cwq->current_work != NULL) {
+                       insert_wq_barrier(cwq, &barr, 1);
+                       active = 1;
                }
-               finish_wait(&cwq->work_done, &wait);
                spin_unlock_irq(&cwq->lock);
+
+               if (active)
+                       wait_for_completion(&barr.done);
        }
 }
 
@@ -427,149 +392,81 @@ static void flush_cpu_workqueue(struct cpu_workqueue_struct *cwq)
  * Forces execution of the workqueue and blocks until its completion.
  * This is typically used in driver shutdown handlers.
  *
- * This function will sample each workqueue's current insert_sequence number and
- * will sleep until the head sequence is greater than or equal to that.  This
- * means that we sleep until all works which were queued on entry have been
- * handled, but we are not livelocked by new incoming ones.
+ * We sleep until all works which were queued on entry have been handled,
+ * but we are not livelocked by new incoming ones.
  *
  * This function used to run the workqueues itself.  Now we just wait for the
  * helper threads to do it.
  */
 void fastcall flush_workqueue(struct workqueue_struct *wq)
 {
-       might_sleep();
+       const cpumask_t *cpu_map = wq_cpu_map(wq);
+       int cpu
 
-       if (is_single_threaded(wq)) {
-               /* Always use first cpu's area. */
-               flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, singlethread_cpu));
-       } else {
-               int cpu;
-
-               mutex_lock(&workqueue_mutex);
-               for_each_online_cpu(cpu)
-                       flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, cpu));
-               mutex_unlock(&workqueue_mutex);
-       }
+       might_sleep();
+       for_each_cpu_mask(cpu, *cpu_map)
+               flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, cpu));
 }
 EXPORT_SYMBOL_GPL(flush_workqueue);
 
-static struct task_struct *create_workqueue_thread(struct workqueue_struct *wq,
-                                                  int cpu, int freezeable)
+static void wait_on_work(struct cpu_workqueue_struct *cwq,
+                               struct work_struct *work)
 {
-       struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
-       struct task_struct *p;
-
-       spin_lock_init(&cwq->lock);
-       cwq->wq = wq;
-       cwq->thread = NULL;
-       cwq->insert_sequence = 0;
-       cwq->remove_sequence = 0;
-       cwq->freezeable = freezeable;
-       INIT_LIST_HEAD(&cwq->worklist);
-       init_waitqueue_head(&cwq->more_work);
-       init_waitqueue_head(&cwq->work_done);
-
-       if (is_single_threaded(wq))
-               p = kthread_create(worker_thread, cwq, "%s", wq->name);
-       else
-               p = kthread_create(worker_thread, cwq, "%s/%d", wq->name, cpu);
-       if (IS_ERR(p))
-               return NULL;
-       cwq->thread = p;
-       return p;
-}
-
-struct workqueue_struct *__create_workqueue(const char *name,
-                                           int singlethread, int freezeable)
-{
-       int cpu, destroy = 0;
-       struct workqueue_struct *wq;
-       struct task_struct *p;
+       struct wq_barrier barr;
+       int running = 0;
 
-       wq = kzalloc(sizeof(*wq), GFP_KERNEL);
-       if (!wq)
-               return NULL;
-
-       wq->cpu_wq = alloc_percpu(struct cpu_workqueue_struct);
-       if (!wq->cpu_wq) {
-               kfree(wq);
-               return NULL;
-       }
-
-       wq->name = name;
-       mutex_lock(&workqueue_mutex);
-       if (singlethread) {
-               INIT_LIST_HEAD(&wq->list);
-               p = create_workqueue_thread(wq, singlethread_cpu, freezeable);
-               if (!p)
-                       destroy = 1;
-               else
-                       wake_up_process(p);
-       } else {
-               list_add(&wq->list, &workqueues);
-               for_each_online_cpu(cpu) {
-                       p = create_workqueue_thread(wq, cpu, freezeable);
-                       if (p) {
-                               kthread_bind(p, cpu);
-                               wake_up_process(p);
-                       } else
-                               destroy = 1;
-               }
-       }
-       mutex_unlock(&workqueue_mutex);
-
-       /*
-        * Was there any error during startup? If yes then clean up:
-        */
-       if (destroy) {
-               destroy_workqueue(wq);
-               wq = NULL;
+       spin_lock_irq(&cwq->lock);
+       if (unlikely(cwq->current_work == work)) {
+               insert_wq_barrier(cwq, &barr, 0);
+               running = 1;
        }
-       return wq;
-}
-EXPORT_SYMBOL_GPL(__create_workqueue);
-
-static void cleanup_workqueue_thread(struct workqueue_struct *wq, int cpu)
-{
-       struct cpu_workqueue_struct *cwq;
-       unsigned long flags;
-       struct task_struct *p;
+       spin_unlock_irq(&cwq->lock);
 
-       cwq = per_cpu_ptr(wq->cpu_wq, cpu);
-       spin_lock_irqsave(&cwq->lock, flags);
-       p = cwq->thread;
-       cwq->thread = NULL;
-       spin_unlock_irqrestore(&cwq->lock, flags);
-       if (p)
-               kthread_stop(p);
+       if (unlikely(running))
+               wait_for_completion(&barr.done);
 }
 
 /**
- * destroy_workqueue - safely terminate a workqueue
- * @wq: target workqueue
+ * flush_work - block until a work_struct's callback has terminated
+ * @wq: the workqueue on which the work is queued
+ * @work: the work which is to be flushed
  *
- * Safely destroy a workqueue. All work currently pending will be done first.
+ * flush_work() will attempt to cancel the work if it is queued.  If the work's
+ * callback appears to be running, flush_work() will block until it has
+ * completed.
+ *
+ * flush_work() is designed to be used when the caller is tearing down data
+ * structures which the callback function operates upon.  It is expected that,
+ * prior to calling flush_work(), the caller has arranged for the work to not
+ * be requeued.
  */
-void destroy_workqueue(struct workqueue_struct *wq)
+void flush_work(struct workqueue_struct *wq, struct work_struct *work)
 {
+       const cpumask_t *cpu_map = wq_cpu_map(wq);
+       struct cpu_workqueue_struct *cwq;
        int cpu;
 
-       flush_workqueue(wq);
+       might_sleep();
 
-       /* We don't need the distraction of CPUs appearing and vanishing. */
-       mutex_lock(&workqueue_mutex);
-       if (is_single_threaded(wq))
-               cleanup_workqueue_thread(wq, singlethread_cpu);
-       else {
-               for_each_online_cpu(cpu)
-                       cleanup_workqueue_thread(wq, cpu);
-               list_del(&wq->list);
-       }
-       mutex_unlock(&workqueue_mutex);
-       free_percpu(wq->cpu_wq);
-       kfree(wq);
+       cwq = get_wq_data(work);
+       /* Was it ever queued ? */
+       if (!cwq)
+               return;
+
+       /*
+        * This work can't be re-queued, no need to re-check that
+        * get_wq_data() is still the same when we take cwq->lock.
+        */
+       spin_lock_irq(&cwq->lock);
+       list_del_init(&work->entry);
+       work_release(work);
+       spin_unlock_irq(&cwq->lock);
+
+       for_each_cpu_mask(cpu, *cpu_map)
+               wait_on_work(per_cpu_ptr(wq->cpu_wq, cpu), work);
 }
-EXPORT_SYMBOL_GPL(destroy_workqueue);
+EXPORT_SYMBOL_GPL(flush_work);
+
 
 static struct workqueue_struct *keventd_wq;
 
@@ -593,8 +490,10 @@ EXPORT_SYMBOL(schedule_work);
  * After waiting for a given time this puts a job in the kernel-global
  * workqueue.
  */
-int fastcall schedule_delayed_work(struct delayed_work *dwork, unsigned long delay)
+int fastcall schedule_delayed_work(struct delayed_work *dwork,
+                                       unsigned long delay)
 {
+       timer_stats_timer_set_start_info(&dwork->timer);
        return queue_delayed_work(keventd_wq, dwork, delay);
 }
 EXPORT_SYMBOL(schedule_delayed_work);
@@ -635,13 +534,15 @@ int schedule_on_each_cpu(work_func_t func)
        if (!works)
                return -ENOMEM;
 
-       mutex_lock(&workqueue_mutex);
+       preempt_disable();              /* CPU hotplug */
        for_each_online_cpu(cpu) {
-               INIT_WORK(per_cpu_ptr(works, cpu), func);
-               __queue_work(per_cpu_ptr(keventd_wq->cpu_wq, cpu),
-                               per_cpu_ptr(works, cpu));
+               struct work_struct *work = per_cpu_ptr(works, cpu);
+
+               INIT_WORK(work, func);
+               set_bit(WORK_STRUCT_PENDING, work_data_bits(work));
+               __queue_work(per_cpu_ptr(keventd_wq->cpu_wq, cpu), work);
        }
-       mutex_unlock(&workqueue_mutex);
+       preempt_enable();
        flush_workqueue(keventd_wq);
        free_percpu(works);
        return 0;
@@ -653,23 +554,31 @@ void flush_scheduled_work(void)
 }
 EXPORT_SYMBOL(flush_scheduled_work);
 
+void flush_work_keventd(struct work_struct *work)
+{
+       flush_work(keventd_wq, work);
+}
+EXPORT_SYMBOL(flush_work_keventd);
+
 /**
- * cancel_rearming_delayed_workqueue - reliably kill off a delayed
- *                     work whose handler rearms the delayed work.
+ * cancel_rearming_delayed_workqueue - reliably kill off a delayed work whose handler rearms the delayed work.
  * @wq:   the controlling workqueue structure
  * @dwork: the delayed work struct
  */
 void cancel_rearming_delayed_workqueue(struct workqueue_struct *wq,
                                       struct delayed_work *dwork)
 {
+       /* Was it ever queued ? */
+       if (!get_wq_data(&dwork->work))
+               return;
+
        while (!cancel_delayed_work(dwork))
                flush_workqueue(wq);
 }
 EXPORT_SYMBOL(cancel_rearming_delayed_workqueue);
 
 /**
- * cancel_rearming_delayed_work - reliably kill off a delayed keventd
- *                     work whose handler rearms the delayed work.
+ * cancel_rearming_delayed_work - reliably kill off a delayed keventd work whose handler rearms the delayed work.
  * @dwork: the delayed work struct
  */
 void cancel_rearming_delayed_work(struct delayed_work *dwork)
@@ -725,84 +634,188 @@ int current_is_keventd(void)
 
 }
 
-/* Take the work from this (downed) CPU. */
-static void take_over_work(struct workqueue_struct *wq, unsigned int cpu)
+static struct cpu_workqueue_struct *
+init_cpu_workqueue(struct workqueue_struct *wq, int cpu)
 {
        struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
-       struct list_head list;
-       struct work_struct *work;
 
-       spin_lock_irq(&cwq->lock);
-       list_replace_init(&cwq->worklist, &list);
+       cwq->wq = wq;
+       spin_lock_init(&cwq->lock);
+       INIT_LIST_HEAD(&cwq->worklist);
+       init_waitqueue_head(&cwq->more_work);
 
-       while (!list_empty(&list)) {
-               printk("Taking work for %s\n", wq->name);
-               work = list_entry(list.next,struct work_struct,entry);
-               list_del(&work->entry);
-               __queue_work(per_cpu_ptr(wq->cpu_wq, smp_processor_id()), work);
-       }
-       spin_unlock_irq(&cwq->lock);
+       return cwq;
 }
 
-/* We're holding the cpucontrol mutex here */
-static int __devinit workqueue_cpu_callback(struct notifier_block *nfb,
-                                 unsigned long action,
-                                 void *hcpu)
+static int create_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu)
+{
+       struct workqueue_struct *wq = cwq->wq;
+       const char *fmt = is_single_threaded(wq) ? "%s" : "%s/%d";
+       struct task_struct *p;
+
+       p = kthread_create(worker_thread, cwq, fmt, wq->name, cpu);
+       /*
+        * Nobody can add the work_struct to this cwq,
+        *      if (caller is __create_workqueue)
+        *              nobody should see this wq
+        *      else // caller is CPU_UP_PREPARE
+        *              cpu is not on cpu_online_map
+        * so we can abort safely.
+        */
+       if (IS_ERR(p))
+               return PTR_ERR(p);
+
+       cwq->thread = p;
+       cwq->should_stop = 0;
+       if (!is_single_threaded(wq))
+               kthread_bind(p, cpu);
+
+       if (is_single_threaded(wq) || cpu_online(cpu))
+               wake_up_process(p);
+
+       return 0;
+}
+
+struct workqueue_struct *__create_workqueue(const char *name,
+                                           int singlethread, int freezeable)
 {
-       unsigned int hotcpu = (unsigned long)hcpu;
        struct workqueue_struct *wq;
+       struct cpu_workqueue_struct *cwq;
+       int err = 0, cpu;
 
-       switch (action) {
-       case CPU_UP_PREPARE:
-               mutex_lock(&workqueue_mutex);
-               /* Create a new workqueue thread for it. */
-               list_for_each_entry(wq, &workqueues, list) {
-                       if (!create_workqueue_thread(wq, hotcpu, 0)) {
-                               printk("workqueue for %i failed\n", hotcpu);
-                               return NOTIFY_BAD;
-                       }
-               }
-               break;
+       wq = kzalloc(sizeof(*wq), GFP_KERNEL);
+       if (!wq)
+               return NULL;
+
+       wq->cpu_wq = alloc_percpu(struct cpu_workqueue_struct);
+       if (!wq->cpu_wq) {
+               kfree(wq);
+               return NULL;
+       }
 
-       case CPU_ONLINE:
-               /* Kick off worker threads. */
-               list_for_each_entry(wq, &workqueues, list) {
-                       struct cpu_workqueue_struct *cwq;
+       wq->name = name;
+       wq->freezeable = freezeable;
 
-                       cwq = per_cpu_ptr(wq->cpu_wq, hotcpu);
-                       kthread_bind(cwq->thread, hotcpu);
-                       wake_up_process(cwq->thread);
-               }
-               mutex_unlock(&workqueue_mutex);
-               break;
+       if (singlethread) {
+               INIT_LIST_HEAD(&wq->list);
+               cwq = init_cpu_workqueue(wq, singlethread_cpu);
+               err = create_workqueue_thread(cwq, singlethread_cpu);
+       } else {
+               mutex_lock(&workqueue_mutex);
+               list_add(&wq->list, &workqueues);
 
-       case CPU_UP_CANCELED:
-               list_for_each_entry(wq, &workqueues, list) {
-                       if (!per_cpu_ptr(wq->cpu_wq, hotcpu)->thread)
+               for_each_possible_cpu(cpu) {
+                       cwq = init_cpu_workqueue(wq, cpu);
+                       if (err || !cpu_online(cpu))
                                continue;
-                       /* Unbind so it can run. */
-                       kthread_bind(per_cpu_ptr(wq->cpu_wq, hotcpu)->thread,
-                                    any_online_cpu(cpu_online_map));
-                       cleanup_workqueue_thread(wq, hotcpu);
+                       err = create_workqueue_thread(cwq, cpu);
                }
                mutex_unlock(&workqueue_mutex);
-               break;
+       }
+
+       if (err) {
+               destroy_workqueue(wq);
+               wq = NULL;
+       }
+       return wq;
+}
+EXPORT_SYMBOL_GPL(__create_workqueue);
+
+static void cleanup_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu)
+{
+       struct wq_barrier barr;
+       int alive = 0;
+
+       spin_lock_irq(&cwq->lock);
+       if (cwq->thread != NULL) {
+               insert_wq_barrier(cwq, &barr, 1);
+               cwq->should_stop = 1;
+               alive = 1;
+       }
+       spin_unlock_irq(&cwq->lock);
+
+       if (alive) {
+               wait_for_completion(&barr.done);
+
+               while (unlikely(cwq->thread != NULL))
+                       cpu_relax();
+               /*
+                * Wait until cwq->thread unlocks cwq->lock,
+                * it won't touch *cwq after that.
+                */
+               smp_rmb();
+               spin_unlock_wait(&cwq->lock);
+       }
+}
+
+/**
+ * destroy_workqueue - safely terminate a workqueue
+ * @wq: target workqueue
+ *
+ * Safely destroy a workqueue. All work currently pending will be done first.
+ */
+void destroy_workqueue(struct workqueue_struct *wq)
+{
+       const cpumask_t *cpu_map = wq_cpu_map(wq);
+       struct cpu_workqueue_struct *cwq;
+       int cpu;
 
-       case CPU_DOWN_PREPARE:
+       mutex_lock(&workqueue_mutex);
+       list_del(&wq->list);
+       mutex_unlock(&workqueue_mutex);
+
+       for_each_cpu_mask(cpu, *cpu_map) {
+               cwq = per_cpu_ptr(wq->cpu_wq, cpu);
+               cleanup_workqueue_thread(cwq, cpu);
+       }
+
+       free_percpu(wq->cpu_wq);
+       kfree(wq);
+}
+EXPORT_SYMBOL_GPL(destroy_workqueue);
+
+static int __devinit workqueue_cpu_callback(struct notifier_block *nfb,
+                                               unsigned long action,
+                                               void *hcpu)
+{
+       unsigned int cpu = (unsigned long)hcpu;
+       struct cpu_workqueue_struct *cwq;
+       struct workqueue_struct *wq;
+
+       switch (action) {
+       case CPU_LOCK_ACQUIRE:
                mutex_lock(&workqueue_mutex);
-               break;
+               return NOTIFY_OK;
 
-       case CPU_DOWN_FAILED:
+       case CPU_LOCK_RELEASE:
                mutex_unlock(&workqueue_mutex);
-               break;
+               return NOTIFY_OK;
 
-       case CPU_DEAD:
-               list_for_each_entry(wq, &workqueues, list)
-                       cleanup_workqueue_thread(wq, hotcpu);
-               list_for_each_entry(wq, &workqueues, list)
-                       take_over_work(wq, hotcpu);
-               mutex_unlock(&workqueue_mutex);
-               break;
+       case CPU_UP_PREPARE:
+               cpu_set(cpu, cpu_populated_map);
+       }
+
+       list_for_each_entry(wq, &workqueues, list) {
+               cwq = per_cpu_ptr(wq->cpu_wq, cpu);
+
+               switch (action) {
+               case CPU_UP_PREPARE:
+                       if (!create_workqueue_thread(cwq, cpu))
+                               break;
+                       printk(KERN_ERR "workqueue for %i failed\n", cpu);
+                       return NOTIFY_BAD;
+
+               case CPU_ONLINE:
+                       wake_up_process(cwq->thread);
+                       break;
+
+               case CPU_UP_CANCELED:
+                       if (cwq->thread)
+                               wake_up_process(cwq->thread);
+               case CPU_DEAD:
+                       cleanup_workqueue_thread(cwq, cpu);
+                       break;
+               }
        }
 
        return NOTIFY_OK;
@@ -810,9 +823,10 @@ static int __devinit workqueue_cpu_callback(struct notifier_block *nfb,
 
 void init_workqueues(void)
 {
+       cpu_populated_map = cpu_online_map;
        singlethread_cpu = first_cpu(cpu_possible_map);
+       cpu_singlethread_map = cpumask_of_cpu(singlethread_cpu);
        hotcpu_notifier(workqueue_cpu_callback, 0);
        keventd_wq = create_workqueue("events");
        BUG_ON(!keventd_wq);
 }
-