workqueue: implement unbound workqueue
authorTejun Heo <tj@kernel.org>
Fri, 2 Jul 2010 08:03:51 +0000 (10:03 +0200)
committerTejun Heo <tj@kernel.org>
Fri, 2 Jul 2010 09:00:02 +0000 (11:00 +0200)
This patch implements unbound workqueue which can be specified with
WQ_UNBOUND flag on creation.  An unbound workqueue has the following
properties.

* It uses a dedicated gcwq with a pseudo CPU number WORK_CPU_UNBOUND.
  This gcwq is always online and disassociated.

* Workers are not bound to any CPU and not concurrency managed.  Works
  are dispatched to workers as soon as possible and the only applied
  limitation is @max_active.  IOW, all unbound workqeueues are
  implicitly high priority.

Unbound workqueues can be used as simple execution context provider.
Contexts unbound to any cpu are served as soon as possible.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: David Howells <dhowells@redhat.com>
include/linux/workqueue.h
kernel/workqueue.c

index 139069a6286cf88ca862cf0d82f9d5f67fbafc4b..67ce734747f6979aeeec03587172adbe8416efcb 100644 (file)
@@ -51,7 +51,8 @@ enum {
        WORK_NO_COLOR           = WORK_NR_COLORS,
 
        /* special cpu IDs */
-       WORK_CPU_NONE           = NR_CPUS,
+       WORK_CPU_UNBOUND        = NR_CPUS,
+       WORK_CPU_NONE           = NR_CPUS + 1,
        WORK_CPU_LAST           = WORK_CPU_NONE,
 
        /*
@@ -237,11 +238,17 @@ enum {
        WQ_RESCUER              = 1 << 3, /* has an rescue worker */
        WQ_HIGHPRI              = 1 << 4, /* high priority */
        WQ_CPU_INTENSIVE        = 1 << 5, /* cpu instensive workqueue */
+       WQ_UNBOUND              = 1 << 6, /* not bound to any cpu */
 
        WQ_MAX_ACTIVE           = 512,    /* I like 512, better ideas? */
+       WQ_MAX_UNBOUND_PER_CPU  = 4,      /* 4 * #cpus for unbound wq */
        WQ_DFL_ACTIVE           = WQ_MAX_ACTIVE / 2,
 };
 
+/* unbound wq's aren't per-cpu, scale max_active according to #cpus */
+#define WQ_UNBOUND_MAX_ACTIVE  \
+       max_t(int, WQ_MAX_ACTIVE, num_possible_cpus() * WQ_MAX_UNBOUND_PER_CPU)
+
 /*
  * System-wide workqueues which are always present.
  *
@@ -256,10 +263,16 @@ enum {
  * system_nrt_wq is non-reentrant and guarantees that any given work
  * item is never executed in parallel by multiple CPUs.  Queue
  * flushing might take relatively long.
+ *
+ * system_unbound_wq is unbound workqueue.  Workers are not bound to
+ * any specific CPU, not concurrency managed, and all queued works are
+ * executed immediately as long as max_active limit is not reached and
+ * resources are available.
  */
 extern struct workqueue_struct *system_wq;
 extern struct workqueue_struct *system_long_wq;
 extern struct workqueue_struct *system_nrt_wq;
+extern struct workqueue_struct *system_unbound_wq;
 
 extern struct workqueue_struct *
 __alloc_workqueue_key(const char *name, unsigned int flags, int max_active,
index a105ddf55f793e5aea5fe82eb22959fed29d05b2..4608563cdd63dd0e2ba6c17f1e7399bfd7548d11 100644 (file)
@@ -53,9 +53,10 @@ enum {
        WORKER_ROGUE            = 1 << 4,       /* not bound to any cpu */
        WORKER_REBIND           = 1 << 5,       /* mom is home, come back */
        WORKER_CPU_INTENSIVE    = 1 << 6,       /* cpu intensive */
+       WORKER_UNBOUND          = 1 << 7,       /* worker is unbound */
 
        WORKER_NOT_RUNNING      = WORKER_PREP | WORKER_ROGUE | WORKER_REBIND |
-                                 WORKER_CPU_INTENSIVE,
+                                 WORKER_CPU_INTENSIVE | WORKER_UNBOUND,
 
        /* gcwq->trustee_state */
        TRUSTEE_START           = 0,            /* start */
@@ -96,7 +97,7 @@ enum {
  * X: During normal operation, modification requires gcwq->lock and
  *    should be done only from local cpu.  Either disabling preemption
  *    on local cpu or grabbing gcwq->lock is enough for read access.
- *    While trustee is in charge, it's identical to L.
+ *    If GCWQ_DISASSOCIATED is set, it's identical to L.
  *
  * F: wq->flush_mutex protected.
  *
@@ -220,14 +221,52 @@ struct workqueue_struct {
 struct workqueue_struct *system_wq __read_mostly;
 struct workqueue_struct *system_long_wq __read_mostly;
 struct workqueue_struct *system_nrt_wq __read_mostly;
+struct workqueue_struct *system_unbound_wq __read_mostly;
 EXPORT_SYMBOL_GPL(system_wq);
 EXPORT_SYMBOL_GPL(system_long_wq);
 EXPORT_SYMBOL_GPL(system_nrt_wq);
+EXPORT_SYMBOL_GPL(system_unbound_wq);
 
 #define for_each_busy_worker(worker, i, pos, gcwq)                     \
        for (i = 0; i < BUSY_WORKER_HASH_SIZE; i++)                     \
                hlist_for_each_entry(worker, pos, &gcwq->busy_hash[i], hentry)
 
+static inline int __next_gcwq_cpu(int cpu, const struct cpumask *mask,
+                                 unsigned int sw)
+{
+       if (cpu < nr_cpu_ids) {
+               if (sw & 1) {
+                       cpu = cpumask_next(cpu, mask);
+                       if (cpu < nr_cpu_ids)
+                               return cpu;
+               }
+               if (sw & 2)
+                       return WORK_CPU_UNBOUND;
+       }
+       return WORK_CPU_NONE;
+}
+
+static inline int __next_wq_cpu(int cpu, const struct cpumask *mask,
+                               struct workqueue_struct *wq)
+{
+       return __next_gcwq_cpu(cpu, mask, !(wq->flags & WQ_UNBOUND) ? 1 : 2);
+}
+
+#define for_each_gcwq_cpu(cpu)                                         \
+       for ((cpu) = __next_gcwq_cpu(-1, cpu_possible_mask, 3);         \
+            (cpu) < WORK_CPU_NONE;                                     \
+            (cpu) = __next_gcwq_cpu((cpu), cpu_possible_mask, 3))
+
+#define for_each_online_gcwq_cpu(cpu)                                  \
+       for ((cpu) = __next_gcwq_cpu(-1, cpu_online_mask, 3);           \
+            (cpu) < WORK_CPU_NONE;                                     \
+            (cpu) = __next_gcwq_cpu((cpu), cpu_online_mask, 3))
+
+#define for_each_cwq_cpu(cpu, wq)                                      \
+       for ((cpu) = __next_wq_cpu(-1, cpu_possible_mask, (wq));        \
+            (cpu) < WORK_CPU_NONE;                                     \
+            (cpu) = __next_wq_cpu((cpu), cpu_possible_mask, (wq)))
+
 #ifdef CONFIG_DEBUG_OBJECTS_WORK
 
 static struct debug_obj_descr work_debug_descr;
@@ -351,26 +390,46 @@ static bool workqueue_freezing;           /* W: have wqs started freezing? */
 static DEFINE_PER_CPU(struct global_cwq, global_cwq);
 static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t, gcwq_nr_running);
 
+/*
+ * Global cpu workqueue and nr_running counter for unbound gcwq.  The
+ * gcwq is always online, has GCWQ_DISASSOCIATED set, and all its
+ * workers have WORKER_UNBOUND set.
+ */
+static struct global_cwq unbound_global_cwq;
+static atomic_t unbound_gcwq_nr_running = ATOMIC_INIT(0);      /* always 0 */
+
 static int worker_thread(void *__worker);
 
 static struct global_cwq *get_gcwq(unsigned int cpu)
 {
-       return &per_cpu(global_cwq, cpu);
+       if (cpu != WORK_CPU_UNBOUND)
+               return &per_cpu(global_cwq, cpu);
+       else
+               return &unbound_global_cwq;
 }
 
 static atomic_t *get_gcwq_nr_running(unsigned int cpu)
 {
-       return &per_cpu(gcwq_nr_running, cpu);
+       if (cpu != WORK_CPU_UNBOUND)
+               return &per_cpu(gcwq_nr_running, cpu);
+       else
+               return &unbound_gcwq_nr_running;
 }
 
 static struct cpu_workqueue_struct *get_cwq(unsigned int cpu,
                                            struct workqueue_struct *wq)
 {
-#ifndef CONFIG_SMP
-       return wq->cpu_wq.single;
+       if (!(wq->flags & WQ_UNBOUND)) {
+               if (likely(cpu < nr_cpu_ids)) {
+#ifdef CONFIG_SMP
+                       return per_cpu_ptr(wq->cpu_wq.pcpu, cpu);
 #else
-       return per_cpu_ptr(wq->cpu_wq.pcpu, cpu);
+                       return wq->cpu_wq.single;
 #endif
+               }
+       } else if (likely(cpu == WORK_CPU_UNBOUND))
+               return wq->cpu_wq.single;
+       return NULL;
 }
 
 static unsigned int work_color_to_flags(int color)
@@ -453,7 +512,7 @@ static struct global_cwq *get_work_gcwq(struct work_struct *work)
        if (cpu == WORK_CPU_NONE)
                return NULL;
 
-       BUG_ON(cpu >= nr_cpu_ids);
+       BUG_ON(cpu >= nr_cpu_ids && cpu != WORK_CPU_UNBOUND);
        return get_gcwq(cpu);
 }
 
@@ -869,11 +928,14 @@ static void __queue_work(unsigned int cpu, struct workqueue_struct *wq,
 
        debug_work_activate(work);
 
+       if (unlikely(cpu == WORK_CPU_UNBOUND))
+               cpu = raw_smp_processor_id();
+
        /*
         * Determine gcwq to use.  SINGLE_CPU is inherently
         * NON_REENTRANT, so test it first.
         */
-       if (!(wq->flags & WQ_SINGLE_CPU)) {
+       if (!(wq->flags & (WQ_SINGLE_CPU | WQ_UNBOUND))) {
                struct global_cwq *last_gcwq;
 
                /*
@@ -900,7 +962,7 @@ static void __queue_work(unsigned int cpu, struct workqueue_struct *wq,
                        }
                } else
                        spin_lock_irqsave(&gcwq->lock, flags);
-       } else {
+       } else if (!(wq->flags & WQ_UNBOUND)) {
                unsigned int req_cpu = cpu;
 
                /*
@@ -932,6 +994,9 @@ static void __queue_work(unsigned int cpu, struct workqueue_struct *wq,
                        spin_unlock_irqrestore(&gcwq->lock, flags);
                        goto retry;
                }
+       } else {
+               gcwq = get_gcwq(WORK_CPU_UNBOUND);
+               spin_lock_irqsave(&gcwq->lock, flags);
        }
 
        /* gcwq determined, get cwq and queue */
@@ -1166,7 +1231,8 @@ static bool worker_maybe_bind_and_lock(struct worker *worker)
                 * it races with cpu hotunplug operation.  Verify
                 * against GCWQ_DISASSOCIATED.
                 */
-               set_cpus_allowed_ptr(task, get_cpu_mask(gcwq->cpu));
+               if (!(gcwq->flags & GCWQ_DISASSOCIATED))
+                       set_cpus_allowed_ptr(task, get_cpu_mask(gcwq->cpu));
 
                spin_lock_irq(&gcwq->lock);
                if (gcwq->flags & GCWQ_DISASSOCIATED)
@@ -1231,8 +1297,9 @@ static struct worker *alloc_worker(void)
  */
 static struct worker *create_worker(struct global_cwq *gcwq, bool bind)
 {
-       int id = -1;
+       bool on_unbound_cpu = gcwq->cpu == WORK_CPU_UNBOUND;
        struct worker *worker = NULL;
+       int id = -1;
 
        spin_lock_irq(&gcwq->lock);
        while (ida_get_new(&gcwq->worker_ida, &id)) {
@@ -1250,8 +1317,12 @@ static struct worker *create_worker(struct global_cwq *gcwq, bool bind)
        worker->gcwq = gcwq;
        worker->id = id;
 
-       worker->task = kthread_create(worker_thread, worker, "kworker/%u:%d",
-                                     gcwq->cpu, id);
+       if (!on_unbound_cpu)
+               worker->task = kthread_create(worker_thread, worker,
+                                             "kworker/%u:%d", gcwq->cpu, id);
+       else
+               worker->task = kthread_create(worker_thread, worker,
+                                             "kworker/u:%d", id);
        if (IS_ERR(worker->task))
                goto fail;
 
@@ -1260,10 +1331,13 @@ static struct worker *create_worker(struct global_cwq *gcwq, bool bind)
         * online later on.  Make sure every worker has
         * PF_THREAD_BOUND set.
         */
-       if (bind)
+       if (bind && !on_unbound_cpu)
                kthread_bind(worker->task, gcwq->cpu);
-       else
+       else {
                worker->task->flags |= PF_THREAD_BOUND;
+               if (on_unbound_cpu)
+                       worker->flags |= WORKER_UNBOUND;
+       }
 
        return worker;
 fail:
@@ -1358,12 +1432,17 @@ static bool send_mayday(struct work_struct *work)
 {
        struct cpu_workqueue_struct *cwq = get_work_cwq(work);
        struct workqueue_struct *wq = cwq->wq;
+       unsigned int cpu;
 
        if (!(wq->flags & WQ_RESCUER))
                return false;
 
        /* mayday mayday mayday */
-       if (!cpumask_test_and_set_cpu(cwq->gcwq->cpu, wq->mayday_mask))
+       cpu = cwq->gcwq->cpu;
+       /* WORK_CPU_UNBOUND can't be set in cpumask, use cpu 0 instead */
+       if (cpu == WORK_CPU_UNBOUND)
+               cpu = 0;
+       if (!cpumask_test_and_set_cpu(cpu, wq->mayday_mask))
                wake_up_process(wq->rescuer->task);
        return true;
 }
@@ -1882,6 +1961,7 @@ static int rescuer_thread(void *__wq)
        struct workqueue_struct *wq = __wq;
        struct worker *rescuer = wq->rescuer;
        struct list_head *scheduled = &rescuer->scheduled;
+       bool is_unbound = wq->flags & WQ_UNBOUND;
        unsigned int cpu;
 
        set_user_nice(current, RESCUER_NICE_LEVEL);
@@ -1891,8 +1971,13 @@ repeat:
        if (kthread_should_stop())
                return 0;
 
+       /*
+        * See whether any cpu is asking for help.  Unbounded
+        * workqueues use cpu 0 in mayday_mask for CPU_UNBOUND.
+        */
        for_each_cpu(cpu, wq->mayday_mask) {
-               struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
+               unsigned int tcpu = is_unbound ? WORK_CPU_UNBOUND : cpu;
+               struct cpu_workqueue_struct *cwq = get_cwq(tcpu, wq);
                struct global_cwq *gcwq = cwq->gcwq;
                struct work_struct *work, *n;
 
@@ -2034,7 +2119,7 @@ static bool flush_workqueue_prep_cwqs(struct workqueue_struct *wq,
                atomic_set(&wq->nr_cwqs_to_flush, 1);
        }
 
-       for_each_possible_cpu(cpu) {
+       for_each_cwq_cpu(cpu, wq) {
                struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
                struct global_cwq *gcwq = cwq->gcwq;
 
@@ -2344,7 +2429,7 @@ static void wait_on_work(struct work_struct *work)
        lock_map_acquire(&work->lockdep_map);
        lock_map_release(&work->lockdep_map);
 
-       for_each_possible_cpu(cpu)
+       for_each_gcwq_cpu(cpu)
                wait_on_cpu_work(get_gcwq(cpu), work);
 }
 
@@ -2590,23 +2675,25 @@ static int alloc_cwqs(struct workqueue_struct *wq)
        const size_t size = sizeof(struct cpu_workqueue_struct);
        const size_t align = max_t(size_t, 1 << WORK_STRUCT_FLAG_BITS,
                                   __alignof__(unsigned long long));
-#ifndef CONFIG_SMP
-       void *ptr;
 
-       /*
-        * Allocate enough room to align cwq and put an extra pointer
-        * at the end pointing back to the originally allocated
-        * pointer which will be used for free.
-        */
-       ptr = kzalloc(size + align + sizeof(void *), GFP_KERNEL);
-       if (ptr) {
-               wq->cpu_wq.single = PTR_ALIGN(ptr, align);
-               *(void **)(wq->cpu_wq.single + 1) = ptr;
+       if (CONFIG_SMP && !(wq->flags & WQ_UNBOUND)) {
+               /* on SMP, percpu allocator can align itself */
+               wq->cpu_wq.pcpu = __alloc_percpu(size, align);
+       } else {
+               void *ptr;
+
+               /*
+                * Allocate enough room to align cwq and put an extra
+                * pointer at the end pointing back to the originally
+                * allocated pointer which will be used for free.
+                */
+               ptr = kzalloc(size + align + sizeof(void *), GFP_KERNEL);
+               if (ptr) {
+                       wq->cpu_wq.single = PTR_ALIGN(ptr, align);
+                       *(void **)(wq->cpu_wq.single + 1) = ptr;
+               }
        }
-#else
-       /* On SMP, percpu allocator can align itself */
-       wq->cpu_wq.pcpu = __alloc_percpu(size, align);
-#endif
+
        /* just in case, make sure it's actually aligned */
        BUG_ON(!IS_ALIGNED(wq->cpu_wq.v, align));
        return wq->cpu_wq.v ? 0 : -ENOMEM;
@@ -2614,23 +2701,25 @@ static int alloc_cwqs(struct workqueue_struct *wq)
 
 static void free_cwqs(struct workqueue_struct *wq)
 {
-#ifndef CONFIG_SMP
-       /* on UP, the pointer to free is stored right after the cwq */
-       if (wq->cpu_wq.single)
+       if (CONFIG_SMP && !(wq->flags & WQ_UNBOUND))
+               free_percpu(wq->cpu_wq.pcpu);
+       else if (wq->cpu_wq.single) {
+               /* the pointer to free is stored right after the cwq */
                kfree(*(void **)(wq->cpu_wq.single + 1));
-#else
-       free_percpu(wq->cpu_wq.pcpu);
-#endif
+       }
 }
 
-static int wq_clamp_max_active(int max_active, const char *name)
+static int wq_clamp_max_active(int max_active, unsigned int flags,
+                              const char *name)
 {
-       if (max_active < 1 || max_active > WQ_MAX_ACTIVE)
+       int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE;
+
+       if (max_active < 1 || max_active > lim)
                printk(KERN_WARNING "workqueue: max_active %d requested for %s "
                       "is out of range, clamping between %d and %d\n",
-                      max_active, name, 1, WQ_MAX_ACTIVE);
+                      max_active, name, 1, lim);
 
-       return clamp_val(max_active, 1, WQ_MAX_ACTIVE);
+       return clamp_val(max_active, 1, lim);
 }
 
 struct workqueue_struct *__alloc_workqueue_key(const char *name,
@@ -2642,8 +2731,15 @@ struct workqueue_struct *__alloc_workqueue_key(const char *name,
        struct workqueue_struct *wq;
        unsigned int cpu;
 
+       /*
+        * Unbound workqueues aren't concurrency managed and should be
+        * dispatched to workers immediately.
+        */
+       if (flags & WQ_UNBOUND)
+               flags |= WQ_HIGHPRI;
+
        max_active = max_active ?: WQ_DFL_ACTIVE;
-       max_active = wq_clamp_max_active(max_active, name);
+       max_active = wq_clamp_max_active(max_active, flags, name);
 
        wq = kzalloc(sizeof(*wq), GFP_KERNEL);
        if (!wq)
@@ -2664,7 +2760,7 @@ struct workqueue_struct *__alloc_workqueue_key(const char *name,
        if (alloc_cwqs(wq) < 0)
                goto err;
 
-       for_each_possible_cpu(cpu) {
+       for_each_cwq_cpu(cpu, wq) {
                struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
                struct global_cwq *gcwq = get_gcwq(cpu);
 
@@ -2703,7 +2799,7 @@ struct workqueue_struct *__alloc_workqueue_key(const char *name,
        spin_lock(&workqueue_lock);
 
        if (workqueue_freezing && wq->flags & WQ_FREEZEABLE)
-               for_each_possible_cpu(cpu)
+               for_each_cwq_cpu(cpu, wq)
                        get_cwq(cpu, wq)->max_active = 0;
 
        list_add(&wq->list, &workqueues);
@@ -2743,7 +2839,7 @@ void destroy_workqueue(struct workqueue_struct *wq)
        spin_unlock(&workqueue_lock);
 
        /* sanity check */
-       for_each_possible_cpu(cpu) {
+       for_each_cwq_cpu(cpu, wq) {
                struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
                int i;
 
@@ -2777,13 +2873,13 @@ void workqueue_set_max_active(struct workqueue_struct *wq, int max_active)
 {
        unsigned int cpu;
 
-       max_active = wq_clamp_max_active(max_active, wq->name);
+       max_active = wq_clamp_max_active(max_active, wq->flags, wq->name);
 
        spin_lock(&workqueue_lock);
 
        wq->saved_max_active = max_active;
 
-       for_each_possible_cpu(cpu) {
+       for_each_cwq_cpu(cpu, wq) {
                struct global_cwq *gcwq = get_gcwq(cpu);
 
                spin_lock_irq(&gcwq->lock);
@@ -3310,7 +3406,7 @@ void freeze_workqueues_begin(void)
        BUG_ON(workqueue_freezing);
        workqueue_freezing = true;
 
-       for_each_possible_cpu(cpu) {
+       for_each_gcwq_cpu(cpu) {
                struct global_cwq *gcwq = get_gcwq(cpu);
                struct workqueue_struct *wq;
 
@@ -3322,7 +3418,7 @@ void freeze_workqueues_begin(void)
                list_for_each_entry(wq, &workqueues, list) {
                        struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
 
-                       if (wq->flags & WQ_FREEZEABLE)
+                       if (cwq && wq->flags & WQ_FREEZEABLE)
                                cwq->max_active = 0;
                }
 
@@ -3354,7 +3450,7 @@ bool freeze_workqueues_busy(void)
 
        BUG_ON(!workqueue_freezing);
 
-       for_each_possible_cpu(cpu) {
+       for_each_gcwq_cpu(cpu) {
                struct workqueue_struct *wq;
                /*
                 * nr_active is monotonically decreasing.  It's safe
@@ -3363,7 +3459,7 @@ bool freeze_workqueues_busy(void)
                list_for_each_entry(wq, &workqueues, list) {
                        struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
 
-                       if (!(wq->flags & WQ_FREEZEABLE))
+                       if (!cwq || !(wq->flags & WQ_FREEZEABLE))
                                continue;
 
                        BUG_ON(cwq->nr_active < 0);
@@ -3396,7 +3492,7 @@ void thaw_workqueues(void)
        if (!workqueue_freezing)
                goto out_unlock;
 
-       for_each_possible_cpu(cpu) {
+       for_each_gcwq_cpu(cpu) {
                struct global_cwq *gcwq = get_gcwq(cpu);
                struct workqueue_struct *wq;
 
@@ -3408,7 +3504,7 @@ void thaw_workqueues(void)
                list_for_each_entry(wq, &workqueues, list) {
                        struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
 
-                       if (!(wq->flags & WQ_FREEZEABLE))
+                       if (!cwq || !(wq->flags & WQ_FREEZEABLE))
                                continue;
 
                        /* restore max_active and repopulate worklist */
@@ -3451,12 +3547,14 @@ void __init init_workqueues(void)
        hotcpu_notifier(workqueue_cpu_callback, CPU_PRI_WORKQUEUE);
 
        /* initialize gcwqs */
-       for_each_possible_cpu(cpu) {
+       for_each_gcwq_cpu(cpu) {
                struct global_cwq *gcwq = get_gcwq(cpu);
 
                spin_lock_init(&gcwq->lock);
                INIT_LIST_HEAD(&gcwq->worklist);
                gcwq->cpu = cpu;
+               if (cpu == WORK_CPU_UNBOUND)
+                       gcwq->flags |= GCWQ_DISASSOCIATED;
 
                INIT_LIST_HEAD(&gcwq->idle_list);
                for (i = 0; i < BUSY_WORKER_HASH_SIZE; i++)
@@ -3476,7 +3574,7 @@ void __init init_workqueues(void)
        }
 
        /* create the initial worker */
-       for_each_online_cpu(cpu) {
+       for_each_online_gcwq_cpu(cpu) {
                struct global_cwq *gcwq = get_gcwq(cpu);
                struct worker *worker;
 
@@ -3490,5 +3588,7 @@ void __init init_workqueues(void)
        system_wq = alloc_workqueue("events", 0, 0);
        system_long_wq = alloc_workqueue("events_long", 0, 0);
        system_nrt_wq = alloc_workqueue("events_nrt", WQ_NON_REENTRANT, 0);
+       system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND,
+                                           WQ_UNBOUND_MAX_ACTIVE);
        BUG_ON(!system_wq || !system_long_wq || !system_nrt_wq);
 }