[PATCHSET] workqueue: implement alloc_ordered_workqueue() and flush[_delayed]_work_sync()

[PATCHSET] workqueue: implement alloc_ordered_workqueue() and flush[_delayed]_work_sync()

[Tejun Heo]

Hello,

This patchset implements new workqueue API fucntions -
alloc_ordered_workqueue() and flush[_delayed]_work_sync().

The former is rather straight-forward replacement for
alloc_singlethread/frozen_workqueue().

The latter two are similar to cancel[_delayed]_work_sync() but wait
for the work to finish execution rather than cancelling them.  These
syncing flushes make sure that works which were queued on multiple
workqueues or a reentrant workqueues have finished execution on all
CPUs.  One usage of these new flushes is replacing
flush_scheduled_works() with finer scoped flushes.

This patchset contains the following four patches.

 0001-workqueue-implement-alloc_ordered_workqueue.patch
 0002-workqueue-cleanup-flush-cancel-functions.patch
 0003-workqueue-factor-out-start_flush_work.patch
 0004-workqueue-implement-flush-_delayed-_work_sync.patch

0001 adds alloc_ordered_workqueue().  0002-0003 prepare for syncing
flushes.  0004 implements them.

Dmitry, as I wrote before, I think serio should be able to simply use
schedule_work() combined with flush_work_sync() implemented by this
patchset.  Would this work for you?

The patches are available in the following git tree.

 git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq.git review-flush_sync

and contains the following changes.

 include/linux/workqueue.h |   38 +++++-
 kernel/workqueue.c        |  265 +++++++++++++++++++++++++++++-----------------
 2 files changed, 201 insertions(+), 102 deletions(-)

Thanks.

--
tejun

[PATCH 1/4] workqueue: implement alloc_ordered_workqueue()

[Tejun Heo]

alloc_ordered_workqueue() creates a workqueue which processes each
work itemp one by one in the queued order.  This will be used to
replace create_freezeable_workqueue() and
create_singlethread_workqueue().

Signed-off-by: Tejun Heo <tj@kernel.org>
---
 include/linux/workqueue.h |   18 ++++++++++++++++++
 1 files changed, 18 insertions(+), 0 deletions(-)

diff --git a/include/linux/workqueue.h b/include/linux/workqueue.h
index 25e02c9..07c4892 100644
--- a/include/linux/workqueue.h
+++ b/include/linux/workqueue.h
@@ -306,6 +306,24 @@ __alloc_workqueue_key(const char *name, unsigned int flags, int max_active,
 	__alloc_workqueue_key((name), (flags), (max_active), NULL, NULL)
 #endif
 
+/**
+ * alloc_ordered_workqueue - allocate an ordered workqueue
+ * @name: name of the workqueue
+ * @flags: WQ_* flags (only WQ_FREEZEABLE and WQ_RESCUER are meaningful)
+ *
+ * Allocate an ordered workqueue.  An ordered workqueue executes at
+ * most one work item at any given time in the queued order.  They are
+ * implemented as unbound workqueues with @max_active of one.
+ *
+ * RETURNS:
+ * Pointer to the allocated workqueue on success, %NULL on failure.
+ */
+static inline struct workqueue_struct *
+alloc_ordered_workqueue(const char *name, unsigned int flags)
+{
+	return alloc_workqueue(name, WQ_UNBOUND | flags, 1);
+}
+
 #define create_workqueue(name)					\
 	alloc_workqueue((name), WQ_RESCUER, 1)
 #define create_freezeable_workqueue(name)			\
-- 
1.7.1

[PATCH 2/4] workqueue: cleanup flush/cancel functions

[Tejun Heo]

Make the following cleanup changes.

* Relocate flush/cancel function prototypes and definitions.

* Relocate wait_on_cpu_work() and wait_on_work() before
  try_to_grab_pending().  These will be used to implement
  flush_work_sync().

* Make all flush/cancel functions return bool instead of int.

* Update wait_on_cpu_work() and wait_on_work() to return %true if they
  actually waited.

* Add / update comments.

This patch doesn't cause any functional changes.

Signed-off-by: Tejun Heo <tj@kernel.org>
---
 include/linux/workqueue.h |   18 +++---
 kernel/workqueue.c        |  175 ++++++++++++++++++++++++---------------------
 2 files changed, 103 insertions(+), 90 deletions(-)

diff --git a/include/linux/workqueue.h b/include/linux/workqueue.h
index 07c4892..bb9b683 100644
--- a/include/linux/workqueue.h
+++ b/include/linux/workqueue.h
@@ -343,7 +343,6 @@ extern int queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
 
 extern void flush_workqueue(struct workqueue_struct *wq);
 extern void flush_scheduled_work(void);
-extern void flush_delayed_work(struct delayed_work *work);
 
 extern int schedule_work(struct work_struct *work);
 extern int schedule_work_on(int cpu, struct work_struct *work);
@@ -355,8 +354,11 @@ extern int keventd_up(void);
 
 int execute_in_process_context(work_func_t fn, struct execute_work *);
 
-extern int flush_work(struct work_struct *work);
-extern int cancel_work_sync(struct work_struct *work);
+extern bool flush_work(struct work_struct *work);
+extern bool cancel_work_sync(struct work_struct *work);
+
+extern bool flush_delayed_work(struct delayed_work *dwork);
+extern bool cancel_delayed_work_sync(struct delayed_work *dwork);
 
 extern void workqueue_set_max_active(struct workqueue_struct *wq,
 				     int max_active);
@@ -370,9 +372,9 @@ extern unsigned int work_busy(struct work_struct *work);
  * it returns 1 and the work doesn't re-arm itself. Run flush_workqueue() or
  * cancel_work_sync() to wait on it.
  */
-static inline int cancel_delayed_work(struct delayed_work *work)
+static inline bool cancel_delayed_work(struct delayed_work *work)
 {
-	int ret;
+	bool ret;
 
 	ret = del_timer_sync(&work->timer);
 	if (ret)
@@ -385,9 +387,9 @@ static inline int cancel_delayed_work(struct delayed_work *work)
  * if it returns 0 the timer function may be running and the queueing is in
  * progress.
  */
-static inline int __cancel_delayed_work(struct delayed_work *work)
+static inline bool __cancel_delayed_work(struct delayed_work *work)
 {
-	int ret;
+	bool ret;
 
 	ret = del_timer(&work->timer);
 	if (ret)
@@ -395,8 +397,6 @@ static inline int __cancel_delayed_work(struct delayed_work *work)
 	return ret;
 }
 
-extern int cancel_delayed_work_sync(struct delayed_work *work);
-
 /* Obsolete. use cancel_delayed_work_sync() */
 static inline
 void cancel_rearming_delayed_workqueue(struct workqueue_struct *wq,
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index f77afd9..1240b9d 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -2327,16 +2327,24 @@ out_unlock:
 EXPORT_SYMBOL_GPL(flush_workqueue);
 
 /**
- * flush_work - block until a work_struct's callback has terminated
- * @work: the work which is to be flushed
+ * flush_work - wait for a work to finish executing the last queueing instance
+ * @work: the work to flush
  *
- * Returns false if @work has already terminated.
+ * Wait until @work has finished execution.  This function considers
+ * only the last queueing instance of @work.  If @work has been
+ * enqueued across different CPUs on a non-reentrant workqueue or on
+ * multiple workqueues, @work might still be executing on return on
+ * some of the CPUs from earlier queueing.
  *
- * It is expected that, prior to calling flush_work(), the caller has
- * arranged for the work to not be requeued, otherwise it doesn't make
- * sense to use this function.
+ * If @work was queued only on a non-reentrant, ordered or unbound
+ * workqueue, @work is guaranteed to be idle on return if it hasn't
+ * been requeued since flush started.
+ *
+ * RETURNS:
+ * %true if flush_work() waited for the work to finish execution,
+ * %false if it was already idle.
  */
-int flush_work(struct work_struct *work)
+bool flush_work(struct work_struct *work)
 {
 	struct worker *worker = NULL;
 	struct global_cwq *gcwq;
@@ -2374,13 +2382,49 @@ int flush_work(struct work_struct *work)
 
 	wait_for_completion(&barr.done);
 	destroy_work_on_stack(&barr.work);
-	return 1;
+	return true;
 already_gone:
 	spin_unlock_irq(&gcwq->lock);
-	return 0;
+	return false;
 }
 EXPORT_SYMBOL_GPL(flush_work);
 
+static bool wait_on_cpu_work(struct global_cwq *gcwq, struct work_struct *work)
+{
+	struct wq_barrier barr;
+	struct worker *worker;
+
+	spin_lock_irq(&gcwq->lock);
+
+	worker = find_worker_executing_work(gcwq, work);
+	if (unlikely(worker))
+		insert_wq_barrier(worker->current_cwq, &barr, work, worker);
+
+	spin_unlock_irq(&gcwq->lock);
+
+	if (unlikely(worker)) {
+		wait_for_completion(&barr.done);
+		destroy_work_on_stack(&barr.work);
+		return true;
+	} else
+		return false;
+}
+
+static bool wait_on_work(struct work_struct *work)
+{
+	bool ret = false;
+	int cpu;
+
+	might_sleep();
+
+	lock_map_acquire(&work->lockdep_map);
+	lock_map_release(&work->lockdep_map);
+
+	for_each_gcwq_cpu(cpu)
+		ret |= wait_on_cpu_work(get_gcwq(cpu), work);
+	return ret;
+}
+
 /*
  * Upon a successful return (>= 0), the caller "owns" WORK_STRUCT_PENDING bit,
  * so this work can't be re-armed in any way.
@@ -2423,39 +2467,7 @@ static int try_to_grab_pending(struct work_struct *work)
 	return ret;
 }
 
-static void wait_on_cpu_work(struct global_cwq *gcwq, struct work_struct *work)
-{
-	struct wq_barrier barr;
-	struct worker *worker;
-
-	spin_lock_irq(&gcwq->lock);
-
-	worker = find_worker_executing_work(gcwq, work);
-	if (unlikely(worker))
-		insert_wq_barrier(worker->current_cwq, &barr, work, worker);
-
-	spin_unlock_irq(&gcwq->lock);
-
-	if (unlikely(worker)) {
-		wait_for_completion(&barr.done);
-		destroy_work_on_stack(&barr.work);
-	}
-}
-
-static void wait_on_work(struct work_struct *work)
-{
-	int cpu;
-
-	might_sleep();
-
-	lock_map_acquire(&work->lockdep_map);
-	lock_map_release(&work->lockdep_map);
-
-	for_each_gcwq_cpu(cpu)
-		wait_on_cpu_work(get_gcwq(cpu), work);
-}
-
-static int __cancel_work_timer(struct work_struct *work,
+static bool __cancel_work_timer(struct work_struct *work,
 				struct timer_list* timer)
 {
 	int ret;
@@ -2472,42 +2484,60 @@ static int __cancel_work_timer(struct work_struct *work,
 }
 
 /**
- * cancel_work_sync - block until a work_struct's callback has terminated
- * @work: the work which is to be flushed
- *
- * Returns true if @work was pending.
+ * cancel_work_sync - cancel a work and wait for it to finish
+ * @work: the work to cancel
  *
- * cancel_work_sync() will cancel the work if it is queued. If the work's
- * callback appears to be running, cancel_work_sync() will block until it
- * has completed.
+ * Cancel @work and wait for its execution to finish.  This function
+ * can be used even if the work re-queues itself or migrates to
+ * another workqueue.  On return from this function, @work is
+ * guaranteed to be not pending or executing on any CPU.
  *
- * It is possible to use this function if the work re-queues itself. It can
- * cancel the work even if it migrates to another workqueue, however in that
- * case it only guarantees that work->func() has completed on the last queued
- * workqueue.
- *
- * cancel_work_sync(&delayed_work->work) should be used only if ->timer is not
- * pending, otherwise it goes into a busy-wait loop until the timer expires.
+ * cancel_work_sync(&delayed_work->work) must not be used for
+ * delayed_work's.  Use cancel_delayed_work_sync() instead.
  *
- * The caller must ensure that workqueue_struct on which this work was last
+ * The caller must ensure that the workqueue on which @work was last
  * queued can't be destroyed before this function returns.
+ *
+ * RETURNS:
+ * %true if @work was pending, %false otherwise.
  */
-int cancel_work_sync(struct work_struct *work)
+bool cancel_work_sync(struct work_struct *work)
 {
 	return __cancel_work_timer(work, NULL);
 }
 EXPORT_SYMBOL_GPL(cancel_work_sync);
 
 /**
- * cancel_delayed_work_sync - reliably kill off a delayed work.
- * @dwork: the delayed work struct
+ * flush_delayed_work - wait for a dwork to finish executing the last queueing
+ * @dwork: the delayed work to flush
+ *
+ * Delayed timer is cancelled and the pending work is queued for
+ * immediate execution.  Like flush_work(), this function only
+ * considers the last queueing instance of @dwork.
+ *
+ * RETURNS:
+ * %true if flush_work() waited for the work to finish execution,
+ * %false if it was already idle.
+ */
+bool flush_delayed_work(struct delayed_work *dwork)
+{
+	if (del_timer_sync(&dwork->timer))
+		__queue_work(raw_smp_processor_id(),
+			     get_work_cwq(&dwork->work)->wq, &dwork->work);
+	return flush_work(&dwork->work);
+}
+EXPORT_SYMBOL(flush_delayed_work);
+
+/**
+ * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish
+ * @dwork: the delayed work cancel
  *
- * Returns true if @dwork was pending.
+ * This is cancel_work_sync() for delayed works.
  *
- * It is possible to use this function if @dwork rearms itself via queue_work()
- * or queue_delayed_work(). See also the comment for cancel_work_sync().
+ * RETURNS:
+ * %true if @dwork was pending, %false otherwise.
  */
-int cancel_delayed_work_sync(struct delayed_work *dwork)
+bool cancel_delayed_work_sync(struct delayed_work *dwork)
 {
 	return __cancel_work_timer(&dwork->work, &dwork->timer);
 }
@@ -2559,23 +2589,6 @@ int schedule_delayed_work(struct delayed_work *dwork,
 EXPORT_SYMBOL(schedule_delayed_work);
 
 /**
- * flush_delayed_work - block until a dwork_struct's callback has terminated
- * @dwork: the delayed work which is to be flushed
- *
- * Any timeout is cancelled, and any pending work is run immediately.
- */
-void flush_delayed_work(struct delayed_work *dwork)
-{
-	if (del_timer_sync(&dwork->timer)) {
-		__queue_work(get_cpu(), get_work_cwq(&dwork->work)->wq,
-			     &dwork->work);
-		put_cpu();
-	}
-	flush_work(&dwork->work);
-}
-EXPORT_SYMBOL(flush_delayed_work);
-
-/**
  * schedule_delayed_work_on - queue work in global workqueue on CPU after delay
  * @cpu: cpu to use
  * @dwork: job to be done
-- 
1.7.1

[PATCH 3/4] workqueue: factor out start_flush_work()

[Tejun Heo]

Factor out start_flush_work() from flush_work().  start_flush_work()
has @wait_executing argument which controls whether the barrier is
queued only if the work is pending or also if executing.  As
flush_work() needs to wait for execution too, it uses %true.

This commit doesn't cause any behavior difference.  start_flush_work()
will be used to implement flush_work_sync().

Signed-off-by: Tejun Heo <tj@kernel.org>
---
 kernel/workqueue.c |   64 ++++++++++++++++++++++++++++++----------------------
 1 files changed, 37 insertions(+), 27 deletions(-)

diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 1240b9d..33d31d7 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -2326,35 +2326,17 @@ out_unlock:
 }
 EXPORT_SYMBOL_GPL(flush_workqueue);
 
-/**
- * flush_work - wait for a work to finish executing the last queueing instance
- * @work: the work to flush
- *
- * Wait until @work has finished execution.  This function considers
- * only the last queueing instance of @work.  If @work has been
- * enqueued across different CPUs on a non-reentrant workqueue or on
- * multiple workqueues, @work might still be executing on return on
- * some of the CPUs from earlier queueing.
- *
- * If @work was queued only on a non-reentrant, ordered or unbound
- * workqueue, @work is guaranteed to be idle on return if it hasn't
- * been requeued since flush started.
- *
- * RETURNS:
- * %true if flush_work() waited for the work to finish execution,
- * %false if it was already idle.
- */
-bool flush_work(struct work_struct *work)
+static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr,
+			     bool wait_executing)
 {
 	struct worker *worker = NULL;
 	struct global_cwq *gcwq;
 	struct cpu_workqueue_struct *cwq;
-	struct wq_barrier barr;
 
 	might_sleep();
 	gcwq = get_work_gcwq(work);
 	if (!gcwq)
-		return 0;
+		return false;
 
 	spin_lock_irq(&gcwq->lock);
 	if (!list_empty(&work->entry)) {
@@ -2367,26 +2349,54 @@ bool flush_work(struct work_struct *work)
 		cwq = get_work_cwq(work);
 		if (unlikely(!cwq || gcwq != cwq->gcwq))
 			goto already_gone;
-	} else {
+	} else if (wait_executing) {
 		worker = find_worker_executing_work(gcwq, work);
 		if (!worker)
 			goto already_gone;
 		cwq = worker->current_cwq;
-	}
+	} else
+		goto already_gone;
 
-	insert_wq_barrier(cwq, &barr, work, worker);
+	insert_wq_barrier(cwq, barr, work, worker);
 	spin_unlock_irq(&gcwq->lock);
 
 	lock_map_acquire(&cwq->wq->lockdep_map);
 	lock_map_release(&cwq->wq->lockdep_map);
-
-	wait_for_completion(&barr.done);
-	destroy_work_on_stack(&barr.work);
 	return true;
 already_gone:
 	spin_unlock_irq(&gcwq->lock);
 	return false;
 }
+
+/**
+ * flush_work - wait for a work to finish executing the last queueing instance
+ * @work: the work to flush
+ *
+ * Wait until @work has finished execution.  This function considers
+ * only the last queueing instance of @work.  If @work has been
+ * enqueued across different CPUs on a non-reentrant workqueue or on
+ * multiple workqueues, @work might still be executing on return on
+ * some of the CPUs from earlier queueing.
+ *
+ * If @work was queued only on a non-reentrant, ordered or unbound
+ * workqueue, @work is guaranteed to be idle on return if it hasn't
+ * been requeued since flush started.
+ *
+ * RETURNS:
+ * %true if flush_work() waited for the work to finish execution,
+ * %false if it was already idle.
+ */
+bool flush_work(struct work_struct *work)
+{
+	struct wq_barrier barr;
+
+	if (start_flush_work(work, &barr, true)) {
+		wait_for_completion(&barr.done);
+		destroy_work_on_stack(&barr.work);
+		return true;
+	} else
+		return false;
+}
 EXPORT_SYMBOL_GPL(flush_work);
 
 static bool wait_on_cpu_work(struct global_cwq *gcwq, struct work_struct *work)
-- 
1.7.1

[PATCH 4/4] workqueue: implement flush[_delayed]_work_sync()

[Tejun Heo]

Implement flush[_delayed]_work_sync().  These are flush functions
which also make sure no CPU is still executing the target work from
earlier queueing instances.  These are similar to
cancel[_delayed]_work_sync() except that the target work item is
flushed instead of cancelled.

Signed-off-by: Tejun Heo <tj@kernel.org>
---
 include/linux/workqueue.h |    2 +
 kernel/workqueue.c        |   56 +++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 58 insertions(+), 0 deletions(-)

diff --git a/include/linux/workqueue.h b/include/linux/workqueue.h
index bb9b683..e33ff4a 100644
--- a/include/linux/workqueue.h
+++ b/include/linux/workqueue.h
@@ -355,9 +355,11 @@ extern int keventd_up(void);
 int execute_in_process_context(work_func_t fn, struct execute_work *);
 
 extern bool flush_work(struct work_struct *work);
+extern bool flush_work_sync(struct work_struct *work);
 extern bool cancel_work_sync(struct work_struct *work);
 
 extern bool flush_delayed_work(struct delayed_work *dwork);
+extern bool flush_delayed_work_sync(struct delayed_work *work);
 extern bool cancel_delayed_work_sync(struct delayed_work *dwork);
 
 extern void workqueue_set_max_active(struct workqueue_struct *wq,
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 33d31d7..19e4bc1 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -2435,6 +2435,41 @@ static bool wait_on_work(struct work_struct *work)
 	return ret;
 }
 
+/**
+ * flush_work_sync - wait until a work has finished execution
+ * @work: the work to flush
+ *
+ * Wait until @work has finished execution.  On return, it's
+ * guaranteed that all queueing instances of @work which happened
+ * before this function is called are finished.  In other words, if
+ * @work hasn't been requeued since this function was called, @work is
+ * guaranteed to be idle on return.
+ *
+ * RETURNS:
+ * %true if flush_work_sync() waited for the work to finish execution,
+ * %false if it was already idle.
+ */
+bool flush_work_sync(struct work_struct *work)
+{
+	struct wq_barrier barr;
+	bool pending, waited;
+
+	/* we'll wait for executions separately, queue barr only if pending */
+	pending = start_flush_work(work, &barr, false);
+
+	/* wait for executions to finish */
+	waited = wait_on_work(work);
+
+	/* wait for the pending one */
+	if (pending) {
+		wait_for_completion(&barr.done);
+		destroy_work_on_stack(&barr.work);
+	}
+
+	return pending || waited;
+}
+EXPORT_SYMBOL_GPL(flush_work_sync);
+
 /*
  * Upon a successful return (>= 0), the caller "owns" WORK_STRUCT_PENDING bit,
  * so this work can't be re-armed in any way.
@@ -2539,6 +2574,27 @@ bool flush_delayed_work(struct delayed_work *dwork)
 EXPORT_SYMBOL(flush_delayed_work);
 
 /**
+ * flush_delayed_work_sync - wait for a dwork to finish
+ * @dwork: the delayed work to flush
+ *
+ * Delayed timer is cancelled and the pending work is queued for
+ * execution immediately.  Other than timer handling, its behavior
+ * is identical to flush_work_sync().
+ *
+ * RETURNS:
+ * %true if flush_work_sync() waited for the work to finish execution,
+ * %false if it was already idle.
+ */
+bool flush_delayed_work_sync(struct delayed_work *dwork)
+{
+	if (del_timer_sync(&dwork->timer))
+		__queue_work(raw_smp_processor_id(),
+			     get_work_cwq(&dwork->work)->wq, &dwork->work);
+	return flush_work_sync(&dwork->work);
+}
+EXPORT_SYMBOL(flush_delayed_work_sync);
+
+/**
  * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish
  * @dwork: the delayed work cancel
  *
-- 
1.7.1