]> bbs.cooldavid.org Git - net-next-2.6.git/commitdiff
workqueue: reimplement work flushing using linked works
authorTejun Heo <tj@kernel.org>
Tue, 29 Jun 2010 08:07:12 +0000 (10:07 +0200)
committerTejun Heo <tj@kernel.org>
Tue, 29 Jun 2010 08:07:12 +0000 (10:07 +0200)
A work is linked to the next one by having WORK_STRUCT_LINKED bit set
and these links can be chained.  When a linked work is dispatched to a
worker, all linked works are dispatched to the worker's newly added
->scheduled queue and processed back-to-back.

Currently, as there's only single worker per cwq, having linked works
doesn't make any visible behavior difference.  This change is to
prepare for multiple shared workers per cpu.

Signed-off-by: Tejun Heo <tj@kernel.org>
include/linux/workqueue.h
kernel/workqueue.c

index 8762f62103d8eb832e835d000b6c0dfc70c598e8..4f4fdba722c39b3192f972340c6f293ab607013a 100644 (file)
@@ -24,8 +24,9 @@ typedef void (*work_func_t)(struct work_struct *work);
 
 enum {
        WORK_STRUCT_PENDING_BIT = 0,    /* work item is pending execution */
+       WORK_STRUCT_LINKED_BIT  = 1,    /* next work is linked to this one */
 #ifdef CONFIG_DEBUG_OBJECTS_WORK
-       WORK_STRUCT_STATIC_BIT  = 1,    /* static initializer (debugobjects) */
+       WORK_STRUCT_STATIC_BIT  = 2,    /* static initializer (debugobjects) */
        WORK_STRUCT_COLOR_SHIFT = 3,    /* color for workqueue flushing */
 #else
        WORK_STRUCT_COLOR_SHIFT = 2,    /* color for workqueue flushing */
@@ -34,6 +35,7 @@ enum {
        WORK_STRUCT_COLOR_BITS  = 4,
 
        WORK_STRUCT_PENDING     = 1 << WORK_STRUCT_PENDING_BIT,
+       WORK_STRUCT_LINKED      = 1 << WORK_STRUCT_LINKED_BIT,
 #ifdef CONFIG_DEBUG_OBJECTS_WORK
        WORK_STRUCT_STATIC      = 1 << WORK_STRUCT_STATIC_BIT,
 #else
index 600db10a4dbf3857e18038461eda41b02fc0ddf3..9953d3c7bd1077ee28f484a89f12c2c6305ccc59 100644 (file)
@@ -51,6 +51,7 @@ struct cpu_workqueue_struct;
 
 struct worker {
        struct work_struct      *current_work;  /* L: work being processed */
+       struct list_head        scheduled;      /* L: scheduled works */
        struct task_struct      *task;          /* I: worker task */
        struct cpu_workqueue_struct *cwq;       /* I: the associated cwq */
        int                     id;             /* I: worker id */
@@ -445,6 +446,8 @@ static struct worker *alloc_worker(void)
        struct worker *worker;
 
        worker = kzalloc(sizeof(*worker), GFP_KERNEL);
+       if (worker)
+               INIT_LIST_HEAD(&worker->scheduled);
        return worker;
 }
 
@@ -530,6 +533,7 @@ static void destroy_worker(struct worker *worker)
 
        /* sanity check frenzy */
        BUG_ON(worker->current_work);
+       BUG_ON(!list_empty(&worker->scheduled));
 
        kthread_stop(worker->task);
        kfree(worker);
@@ -539,6 +543,47 @@ static void destroy_worker(struct worker *worker)
        spin_unlock(&workqueue_lock);
 }
 
+/**
+ * move_linked_works - move linked works to a list
+ * @work: start of series of works to be scheduled
+ * @head: target list to append @work to
+ * @nextp: out paramter for nested worklist walking
+ *
+ * Schedule linked works starting from @work to @head.  Work series to
+ * be scheduled starts at @work and includes any consecutive work with
+ * WORK_STRUCT_LINKED set in its predecessor.
+ *
+ * If @nextp is not NULL, it's updated to point to the next work of
+ * the last scheduled work.  This allows move_linked_works() to be
+ * nested inside outer list_for_each_entry_safe().
+ *
+ * CONTEXT:
+ * spin_lock_irq(cwq->lock).
+ */
+static void move_linked_works(struct work_struct *work, struct list_head *head,
+                             struct work_struct **nextp)
+{
+       struct work_struct *n;
+
+       /*
+        * Linked worklist will always end before the end of the list,
+        * use NULL for list head.
+        */
+       list_for_each_entry_safe_from(work, n, NULL, entry) {
+               list_move_tail(&work->entry, head);
+               if (!(*work_data_bits(work) & WORK_STRUCT_LINKED))
+                       break;
+       }
+
+       /*
+        * If we're already inside safe list traversal and have moved
+        * multiple works to the scheduled queue, the next position
+        * needs to be updated.
+        */
+       if (nextp)
+               *nextp = n;
+}
+
 /**
  * cwq_dec_nr_in_flight - decrement cwq's nr_in_flight
  * @cwq: cwq of interest
@@ -639,17 +684,25 @@ static void process_one_work(struct worker *worker, struct work_struct *work)
        cwq_dec_nr_in_flight(cwq, work_color);
 }
 
-static void run_workqueue(struct worker *worker)
+/**
+ * process_scheduled_works - process scheduled works
+ * @worker: self
+ *
+ * Process all scheduled works.  Please note that the scheduled list
+ * may change while processing a work, so this function repeatedly
+ * fetches a work from the top and executes it.
+ *
+ * CONTEXT:
+ * spin_lock_irq(cwq->lock) which may be released and regrabbed
+ * multiple times.
+ */
+static void process_scheduled_works(struct worker *worker)
 {
-       struct cpu_workqueue_struct *cwq = worker->cwq;
-
-       spin_lock_irq(&cwq->lock);
-       while (!list_empty(&cwq->worklist)) {
-               struct work_struct *work = list_entry(cwq->worklist.next,
+       while (!list_empty(&worker->scheduled)) {
+               struct work_struct *work = list_first_entry(&worker->scheduled,
                                                struct work_struct, entry);
                process_one_work(worker, work);
        }
-       spin_unlock_irq(&cwq->lock);
 }
 
 /**
@@ -684,7 +737,28 @@ static int worker_thread(void *__worker)
                                            get_cpu_mask(cwq->cpu))))
                        set_cpus_allowed_ptr(worker->task,
                                             get_cpu_mask(cwq->cpu));
-               run_workqueue(worker);
+
+               spin_lock_irq(&cwq->lock);
+
+               while (!list_empty(&cwq->worklist)) {
+                       struct work_struct *work =
+                               list_first_entry(&cwq->worklist,
+                                                struct work_struct, entry);
+
+                       if (likely(!(*work_data_bits(work) &
+                                    WORK_STRUCT_LINKED))) {
+                               /* optimization path, not strictly necessary */
+                               process_one_work(worker, work);
+                               if (unlikely(!list_empty(&worker->scheduled)))
+                                       process_scheduled_works(worker);
+                       } else {
+                               move_linked_works(work, &worker->scheduled,
+                                                 NULL);
+                               process_scheduled_works(worker);
+                       }
+               }
+
+               spin_unlock_irq(&cwq->lock);
        }
 
        return 0;
@@ -705,16 +779,33 @@ static void wq_barrier_func(struct work_struct *work)
  * insert_wq_barrier - insert a barrier work
  * @cwq: cwq to insert barrier into
  * @barr: wq_barrier to insert
- * @head: insertion point
+ * @target: target work to attach @barr to
+ * @worker: worker currently executing @target, NULL if @target is not executing
  *
- * Insert barrier @barr into @cwq before @head.
+ * @barr is linked to @target such that @barr is completed only after
+ * @target finishes execution.  Please note that the ordering
+ * guarantee is observed only with respect to @target and on the local
+ * cpu.
+ *
+ * Currently, a queued barrier can't be canceled.  This is because
+ * try_to_grab_pending() can't determine whether the work to be
+ * grabbed is at the head of the queue and thus can't clear LINKED
+ * flag of the previous work while there must be a valid next work
+ * after a work with LINKED flag set.
+ *
+ * Note that when @worker is non-NULL, @target may be modified
+ * underneath us, so we can't reliably determine cwq from @target.
  *
  * CONTEXT:
  * spin_lock_irq(cwq->lock).
  */
 static void insert_wq_barrier(struct cpu_workqueue_struct *cwq,
-                       struct wq_barrier *barr, struct list_head *head)
+                             struct wq_barrier *barr,
+                             struct work_struct *target, struct worker *worker)
 {
+       struct list_head *head;
+       unsigned int linked = 0;
+
        /*
         * debugobject calls are safe here even with cwq->lock locked
         * as we know for sure that this will not trigger any of the
@@ -725,8 +816,24 @@ static void insert_wq_barrier(struct cpu_workqueue_struct *cwq,
        __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work));
        init_completion(&barr->done);
 
+       /*
+        * If @target is currently being executed, schedule the
+        * barrier to the worker; otherwise, put it after @target.
+        */
+       if (worker)
+               head = worker->scheduled.next;
+       else {
+               unsigned long *bits = work_data_bits(target);
+
+               head = target->entry.next;
+               /* there can already be other linked works, inherit and set */
+               linked = *bits & WORK_STRUCT_LINKED;
+               __set_bit(WORK_STRUCT_LINKED_BIT, bits);
+       }
+
        debug_work_activate(&barr->work);
-       insert_work(cwq, &barr->work, head, work_color_to_flags(WORK_NO_COLOR));
+       insert_work(cwq, &barr->work, head,
+                   work_color_to_flags(WORK_NO_COLOR) | linked);
 }
 
 /**
@@ -964,8 +1071,8 @@ EXPORT_SYMBOL_GPL(flush_workqueue);
  */
 int flush_work(struct work_struct *work)
 {
+       struct worker *worker = NULL;
        struct cpu_workqueue_struct *cwq;
-       struct list_head *prev;
        struct wq_barrier barr;
 
        might_sleep();
@@ -985,14 +1092,14 @@ int flush_work(struct work_struct *work)
                smp_rmb();
                if (unlikely(cwq != get_wq_data(work)))
                        goto already_gone;
-               prev = &work->entry;
        } else {
-               if (!cwq->worker || cwq->worker->current_work != work)
+               if (cwq->worker && cwq->worker->current_work == work)
+                       worker = cwq->worker;
+               if (!worker)
                        goto already_gone;
-               prev = &cwq->worklist;
        }
-       insert_wq_barrier(cwq, &barr, prev->next);
 
+       insert_wq_barrier(cwq, &barr, work, worker);
        spin_unlock_irq(&cwq->lock);
        wait_for_completion(&barr.done);
        destroy_work_on_stack(&barr.work);
@@ -1048,16 +1155,19 @@ static void wait_on_cpu_work(struct cpu_workqueue_struct *cwq,
                                struct work_struct *work)
 {
        struct wq_barrier barr;
-       int running = 0;
+       struct worker *worker;
 
        spin_lock_irq(&cwq->lock);
+
+       worker = NULL;
        if (unlikely(cwq->worker && cwq->worker->current_work == work)) {
-               insert_wq_barrier(cwq, &barr, cwq->worklist.next);
-               running = 1;
+               worker = cwq->worker;
+               insert_wq_barrier(cwq, &barr, work, worker);
        }
+
        spin_unlock_irq(&cwq->lock);
 
-       if (unlikely(running)) {
+       if (unlikely(worker)) {
                wait_for_completion(&barr.done);
                destroy_work_on_stack(&barr.work);
        }