[PATCH] workqueue: lock cwq access in drain_workqueue

[PATCH] workqueue: lock cwq access in drain_workqueue

[Thomas Tuttle]

Take cwq->gcwq->lock to avoid racing between drain_workqueue checking
to make sure the workqueues are empty and cwq_dec_nr_in_flight
decrementing and then incrementing nr_active when it activates a
delayed work.

We discovered this when a corner case in one of our drivers resulted in
us trying to destroy a workqueue in which the remaining work would
always requeue itself again in the same workqueue.  We would hit this
race condition and trip the BUG_ON on workqueue.c:3080.

Patch is against HEAD as of Fri Sep  9 15:16:09 UTC 2011
(e4e436e0bd480668834fe6849a52c5397b7be4fb).

Signed-off-by: Thomas Tuttle <ttuttle@chromium.org>
---
 kernel/workqueue.c |    8 +++++++-
 1 files changed, 7 insertions(+), 1 deletions(-)

diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 25fb1b0..d610ced 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -2412,8 +2412,14 @@ reflush:
 
 	for_each_cwq_cpu(cpu, wq) {
 		struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
+		int cwq_flushed;
 
-		if (!cwq->nr_active && list_empty(&cwq->delayed_works))
+		spin_lock_irq(&cwq->gcwq->lock);
+		cwq_flushed = !cwq->nr_active
+			      && list_empty(&cwq->delayed_works);
+		spin_unlock_irq(&cwq->gcwq->lock);
+
+		if (cwq_flushed)
 			continue;
 
 		if (++flush_cnt == 10 ||
-- 
1.7.3.1

[PATCH v2] workqueue: lock cwq access in drain_workqueue

[Thomas Tuttle]

Take cwq->gcwq->lock to avoid racing between drain_workqueue checking
to make sure the workqueues are empty and cwq_dec_nr_in_flight
decrementing and then incrementing nr_active when it activates a
delayed work.

We discovered this when a corner case in one of our drivers resulted in
us trying to destroy a workqueue in which the remaining work would
always requeue itself again in the same workqueue.  We would hit this
race condition and trip the BUG_ON on workqueue.c:3080.

Signed-off-by: Thomas Tuttle <ttuttle@chromium.org>
---
Updated to use bool instead of int (d'oh), and CCed maintainer.

 kernel/workqueue.c |    8 +++++++-
 1 files changed, 7 insertions(+), 1 deletions(-)

diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 25fb1b0..0c2e585 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -2412,8 +2412,14 @@ reflush:
 
 	for_each_cwq_cpu(cpu, wq) {
 		struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
+		bool cwq_flushed;
 
-		if (!cwq->nr_active && list_empty(&cwq->delayed_works))
+		spin_lock_irq(&cwq->gcwq->lock);
+		cwq_flushed = !cwq->nr_active
+			      && list_empty(&cwq->delayed_works);
+		spin_unlock_irq(&cwq->gcwq->lock);
+
+		if (cwq_flushed)
 			continue;
 
 		if (++flush_cnt == 10 ||
-- 
1.7.3.1

Re: [PATCH v2] workqueue: lock cwq access in drain_workqueue

[Tejun Heo]

Hello,

On Fri, Sep 09, 2011 at 07:00:53PM -0400, Thomas Tuttle wrote:
> Take cwq->gcwq->lock to avoid racing between drain_workqueue checking
> to make sure the workqueues are empty and cwq_dec_nr_in_flight
> decrementing and then incrementing nr_active when it activates a
> delayed work.

Nice catch.  Just few minor nits below.

> We discovered this when a corner case in one of our drivers resulted in
> us trying to destroy a workqueue in which the remaining work would
> always requeue itself again in the same workqueue.  We would hit this
> race condition and trip the BUG_ON on workqueue.c:3080.
> 
> Signed-off-by: Thomas Tuttle <ttuttle@chromium.org>
> ---
> Updated to use bool instead of int (d'oh), and CCed maintainer.
> 
>  kernel/workqueue.c |    8 +++++++-
>  1 files changed, 7 insertions(+), 1 deletions(-)
> 
> diff --git a/kernel/workqueue.c b/kernel/workqueue.c
> index 25fb1b0..0c2e585 100644
> --- a/kernel/workqueue.c
> +++ b/kernel/workqueue.c
> @@ -2412,8 +2412,14 @@ reflush:
>  
>  	for_each_cwq_cpu(cpu, wq) {
>  		struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
> +		bool cwq_flushed;

Maybe "drained" would be better?

> -		if (!cwq->nr_active && list_empty(&cwq->delayed_works))
> +		spin_lock_irq(&cwq->gcwq->lock);
> +		cwq_flushed = !cwq->nr_active
> +			      && list_empty(&cwq->delayed_works);

and then this should fit inside 80 column, right?

Thanks.

-- 
tejun

[PATCH v3] workqueue: lock cwq access in drain_workqueue

[Thomas Tuttle]

Take cwq->gcwq->lock to avoid racing between drain_workqueue checking
to make sure the workqueues are empty and cwq_dec_nr_in_flight
decrementing and then incrementing nr_active when it activates a
delayed work.

We discovered this when a corner case in one of our drivers resulted in
us trying to destroy a workqueue in which the remaining work would
always requeue itself again in the same workqueue.  We would hit this
race condition and trip the BUG_ON on workqueue.c:3080.

Signed-off-by: Thomas Tuttle <ttuttle@chromium.org>
---
Renamed "cwq_flushed" to "drained" as requested and rebased against
current HEAD (d0a77454c70d0449a5f87087deb8f0cb15145e90).

 kernel/workqueue.c |    7 ++++++-
 1 files changed, 6 insertions(+), 1 deletions(-)

diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 25fb1b0..1783aab 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -2412,8 +2412,13 @@ reflush:
 
 	for_each_cwq_cpu(cpu, wq) {
 		struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
+		bool drained;
 
-		if (!cwq->nr_active && list_empty(&cwq->delayed_works))
+		spin_lock_irq(&cwq->gcwq->lock);
+		drained = !cwq->nr_active && list_empty(&cwq->delayed_works);
+		spin_unlock_irq(&cwq->gcwq->lock);
+
+		if (drained)
 			continue;
 
 		if (++flush_cnt == 10 ||
-- 
1.7.3.1

Re: [PATCH v3] workqueue: lock cwq access in drain_workqueue

[Tejun Heo]

Hello,

On Sat, Sep 10, 2011 at 11:26:41PM -0400, Thomas Tuttle wrote:
> Take cwq->gcwq->lock to avoid racing between drain_workqueue checking
> to make sure the workqueues are empty and cwq_dec_nr_in_flight
> decrementing and then incrementing nr_active when it activates a
> delayed work.
> 
> We discovered this when a corner case in one of our drivers resulted in
> us trying to destroy a workqueue in which the remaining work would
> always requeue itself again in the same workqueue.  We would hit this
> race condition and trip the BUG_ON on workqueue.c:3080.
> 
> Signed-off-by: Thomas Tuttle <ttuttle@chromium.org>

  Acked-by: Tejun Heo <tj@kernel.org>

Can you please also add "Cc: stable@kernel.org" and send it to Andrew
Morton <akpm@linux-foundation.org>?

Thank you very much.

-- 
tejun

[PATCH v3] workqueue: lock cwq access in drain_workqueue

[Thomas Tuttle]

Take cwq->gcwq->lock to avoid racing between drain_workqueue checking
to make sure the workqueues are empty and cwq_dec_nr_in_flight
decrementing and then incrementing nr_active when it activates a
delayed work.

We discovered this when a corner case in one of our drivers resulted in
us trying to destroy a workqueue in which the remaining work would
always requeue itself again in the same workqueue.  We would hit this
race condition and trip the BUG_ON on workqueue.c:3080.

Signed-off-by: Thomas Tuttle <ttuttle@chromium.org>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: stable@kernel.org
---
 kernel/workqueue.c |    7 ++++++-
 1 files changed, 6 insertions(+), 1 deletions(-)

diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 25fb1b0..1783aab 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -2412,8 +2412,13 @@ reflush:
 
 	for_each_cwq_cpu(cpu, wq) {
 		struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq);
+		bool drained;
 
-		if (!cwq->nr_active && list_empty(&cwq->delayed_works))
+		spin_lock_irq(&cwq->gcwq->lock);
+		drained = !cwq->nr_active && list_empty(&cwq->delayed_works);
+		spin_unlock_irq(&cwq->gcwq->lock);
+
+		if (drained)
 			continue;
 
 		if (++flush_cnt == 10 ||
-- 
1.7.3.1


----- End forwarded message -----

Re: [PATCH v3] workqueue: lock cwq access in drain_workqueue

[Tejun Heo]

On Sat, Sep 10, 2011 at 11:39:53PM -0400, Thomas Tuttle wrote:
> Take cwq->gcwq->lock to avoid racing between drain_workqueue checking
> to make sure the workqueues are empty and cwq_dec_nr_in_flight
> decrementing and then incrementing nr_active when it activates a
> delayed work.
> 
> We discovered this when a corner case in one of our drivers resulted in
> us trying to destroy a workqueue in which the remaining work would
> always requeue itself again in the same workqueue.  We would hit this
> race condition and trip the BUG_ON on workqueue.c:3080.
> 
> Signed-off-by: Thomas Tuttle <ttuttle@chromium.org>
> Acked-by: Tejun Heo <tj@kernel.org>
> Cc: stable@kernel.org

Andrew, can you please route this patch through -mm?  korg is still
down and wq is unlikely to receive many more patches in this cycle.

Thank you.

-- 
tejun