[PATCH] workqueue: Fix possible unexpectedly worker wakeup before started

[PATCH] workqueue: Fix possible unexpectedly worker wakeup before started

[Lai Jiangshan]

If a worker is wokenup unexpectedly, it will start to work incorretly.
Although it hardly happen, we should catch it and wait for being started
if it does happen.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
---
 kernel/workqueue.c |    6 ++++++
 1 files changed, 6 insertions(+), 0 deletions(-)

diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 82ef9f3..bee5fe1 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -2284,6 +2284,12 @@ static int worker_thread(void *__worker)
 	struct worker *worker = __worker;
 	struct worker_pool *pool = worker->pool;
 
+	if (WARN_ON_ONCE(!(worker->flags & WORKER_STARTED))) {
+		/* The worker is wokenup unexpectedly before started */
+		mutex_lock(&pool->manager_mutex);
+		mutex_unlock(&pool->manager_mutex);
+	}
+
 	/* tell the scheduler that this is a workqueue worker */
 	worker->task->flags |= PF_WQ_WORKER;
 woke_up:
-- 
1.7.4.4

Re: [PATCH] workqueue: Fix possible unexpectedly worker wakeup before started

[Tejun Heo]

Hello, Lai.

On Wed, Feb 19, 2014 at 11:47:58AM +0800, Lai Jiangshan wrote:
> If a worker is wokenup unexpectedly, it will start to work incorretly.
> Although it hardly happen, we should catch it and wait for being started
> if it does happen.

Can this actually happen?  If so, how?

> Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
> ---
>  kernel/workqueue.c |    6 ++++++
>  1 files changed, 6 insertions(+), 0 deletions(-)
> 
> diff --git a/kernel/workqueue.c b/kernel/workqueue.c
> index 82ef9f3..bee5fe1 100644
> --- a/kernel/workqueue.c
> +++ b/kernel/workqueue.c
> @@ -2284,6 +2284,12 @@ static int worker_thread(void *__worker)
>  	struct worker *worker = __worker;
>  	struct worker_pool *pool = worker->pool;
>  
> +	if (WARN_ON_ONCE(!(worker->flags & WORKER_STARTED))) {

And if this is something which can legitimately happen, why are we
triggering WARN on it?

> +		/* The worker is wokenup unexpectedly before started */
> +		mutex_lock(&pool->manager_mutex);
> +		mutex_unlock(&pool->manager_mutex);

And what does these mutex cycling achieve (they need comment)?

Thanks.

-- 
tejun

Re: [PATCH] workqueue: Fix possible unexpectedly worker wakeup before started

[Lai Jiangshan]

On 02/20/2014 08:11 AM, Tejun Heo wrote:
> Hello, Lai.
> 
> On Wed, Feb 19, 2014 at 11:47:58AM +0800, Lai Jiangshan wrote:
>> If a worker is wokenup unexpectedly, it will start to work incorretly.
>> Although it hardly happen, we should catch it and wait for being started
>> if it does happen.
> 
> Can this actually happen?  If so, how?

I don't think it can happen.
It depends on the system outside of workqueue.

I'm afraid someone see the task and wake up it.
workqueue protect itself.

> 
>> Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
>> ---
>>  kernel/workqueue.c |    6 ++++++
>>  1 files changed, 6 insertions(+), 0 deletions(-)
>>
>> diff --git a/kernel/workqueue.c b/kernel/workqueue.c
>> index 82ef9f3..bee5fe1 100644
>> --- a/kernel/workqueue.c
>> +++ b/kernel/workqueue.c
>> @@ -2284,6 +2284,12 @@ static int worker_thread(void *__worker)
>>  	struct worker *worker = __worker;
>>  	struct worker_pool *pool = worker->pool;
>>  
>> +	if (WARN_ON_ONCE(!(worker->flags & WORKER_STARTED))) {
> 
> And if this is something which can legitimately happen, why are we
> triggering WARN on it?

If it happens, it means there is something wrong in the system.

> 
>> +		/* The worker is wokenup unexpectedly before started */
>> +		mutex_lock(&pool->manager_mutex);
>> +		mutex_unlock(&pool->manager_mutex);
> 
> And what does these mutex cycling achieve (they need comment)?

Synchronize the manager to finish.

> 
> Thanks.
> 

Re: [PATCH] workqueue: Fix possible unexpectedly worker wakeup before started

[Lai Jiangshan]

On 02/20/2014 09:50 AM, Lai Jiangshan wrote:
> On 02/20/2014 08:11 AM, Tejun Heo wrote:
>> Hello, Lai.
>>
>> On Wed, Feb 19, 2014 at 11:47:58AM +0800, Lai Jiangshan wrote:
>>> If a worker is wokenup unexpectedly, it will start to work incorretly.
>>> Although it hardly happen, we should catch it and wait for being started
>>> if it does happen.
>>
>> Can this actually happen?  If so, how?
> 
> I don't think it can happen.
> It depends on the system outside of workqueue.
> 
> I'm afraid someone see the task and wake up it.
> workqueue protect itself.

I was extremely nervously, please drop the patch.
I also don't like too much protective code in workqueue.

> 
>>
>>> Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
>>> ---
>>>  kernel/workqueue.c |    6 ++++++
>>>  1 files changed, 6 insertions(+), 0 deletions(-)
>>>
>>> diff --git a/kernel/workqueue.c b/kernel/workqueue.c
>>> index 82ef9f3..bee5fe1 100644
>>> --- a/kernel/workqueue.c
>>> +++ b/kernel/workqueue.c
>>> @@ -2284,6 +2284,12 @@ static int worker_thread(void *__worker)
>>>  	struct worker *worker = __worker;
>>>  	struct worker_pool *pool = worker->pool;
>>>  
>>> +	if (WARN_ON_ONCE(!(worker->flags & WORKER_STARTED))) {
>>
>> And if this is something which can legitimately happen, why are we
>> triggering WARN on it?
> 
> If it happens, it means there is something wrong in the system.
> 
>>
>>> +		/* The worker is wokenup unexpectedly before started */
>>> +		mutex_lock(&pool->manager_mutex);
>>> +		mutex_unlock(&pool->manager_mutex);
>>
>> And what does these mutex cycling achieve (they need comment)?
> 
> Synchronize the manager to finish.
> 
>>
>> Thanks.
>>
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>