Re: [NFS] 2.6.23-rc1-mm2

[Oleg Nesterov]

On 08/06, Johannes Berg wrote:
>
> On Fri, 2007-08-03 at 21:21 +0400, Oleg Nesterov wrote:
> 
> > To avoid a possible confusion: it is still OK if work->func() flushes
> > its own workqueue, so strictly speaking this trace is false positive,
> > but it would be very nice if we can get rid of this practice.
> 
> However, the question whether we should allow flush_workqueue from
> within a struct work is mainly an API policy issue; it doesn't hurt to
> flush a workqueue from within a work,

I am not sure, but currently I hope we can forbid this eventually, so I
personally think it is good that your patch complains.

> --- wireless-dev.orig/kernel/workqueue.c	2007-08-06 08:11:23.297846657 +0200
> +++ wireless-dev/kernel/workqueue.c	2007-08-06 08:19:54.727846657 +0200
> @@ -272,7 +272,7 @@ static void run_workqueue(struct cpu_wor
>  
>  		BUG_ON(get_wq_data(work) != cwq);
>  		work_clear_pending(work);
> -		lock_acquire(&cwq->wq->lockdep_map, 0, 0, 0, 2, _THIS_IP_);
> +		lock_acquire(&cwq->wq->lockdep_map, 0, 0, 1, 2, _THIS_IP_);
>  		lock_acquire(&lockdep_map, 0, 0, 0, 2, _THIS_IP_);
>  		f(work);
>  		lock_release(&lockdep_map, 1, _THIS_IP_);
> @@ -395,7 +395,7 @@ void fastcall flush_workqueue(struct wor
>  	int cpu;
>  
>  	might_sleep();
> -	lock_acquire(&wq->lockdep_map, 0, 0, 0, 2, _THIS_IP_);
> +	lock_acquire(&wq->lockdep_map, 0, 0, 1, 2, _THIS_IP_);
>  	lock_release(&wq->lockdep_map, 1, _THIS_IP_);
>  	for_each_cpu_mask(cpu, *cpu_map)
>  		flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, cpu));
> @@ -779,7 +779,7 @@ static void cleanup_workqueue_thread(str
>  	if (cwq->thread == NULL)
>  		return;
>  
> -	lock_acquire(&cwq->wq->lockdep_map, 0, 0, 0, 2, _THIS_IP_);
> +	lock_acquire(&cwq->wq->lockdep_map, 0, 0, 1, 2, _THIS_IP_);
>  	lock_release(&cwq->wq->lockdep_map, 1, _THIS_IP_);
>  
>  	flush_cpu_workqueue(cwq);

But this makes ->lockdep_map meaningless? We always take wq->lockdep_map
for reading, now we can't detect deadlocks.

	read_lock(A);
	lock(B);

vs
	lock(B);
	read_lock(A);

is valid, kernel/lockdep.c should not complain.

No?

Oleg.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/