Re: [PATCH 1/2] workqueue: Catch more locking problems with flush_work()

[Stephen Boyd <sboyd@codeaurora.org>]

On 4/20/2012 12:18 AM, Yong Zhang wrote:
> On Thu, Apr 19, 2012 at 11:26:47PM -0700, Stephen Boyd wrote:
>> complain in the case where the work is not queued. That case is not a
>> false positive. We will get a lockdep warning if the work is running
> IIRC, flush_work() is just a nop when a work is not queued nor running.

Agreed, but it's better to always print a lockdep warning instead of
only when the deadlock is going to occur.

>
>> (when start_flush_work() returns true) solely with the
>> lock_map_acquire() on cwq->wq->lockdep_map.
> Yeah, that is the point we use lockdep to detect deadlock for workqueue.
>
> But when looking at start_flush_work(), for some case
> !(cwq->wq->saved_max_active == 1 || cwq->wq->flags & WQ_RESCUER),
> lock_map_acquire_read() is called, but recursive read is not added to
> the chain list. So when lock_map_acquire_read(&cwq->wq->lockdep_map)
> is called, deadlock will not be detected. I hope you don't hit that
> special case.

Hmm. Originally I had what you suggested in my patch but I left it out
because I wasn't sure if it would cause false positives. Do you see any
possibility for false positives? I'll add it back in if not.

-- 
Sent by an employee of the Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.