Re: [PATCH v1 02/25] mm/swap: Don't abuse the seqcount latching API

[Peter Zijlstra]

On Wed, May 20, 2020 at 03:22:15PM +0300, Konstantin Khlebnikov wrote:
> On 20/05/2020 00.45, Ahmed S. Darwish wrote:
> > Commit eef1a429f234 ("mm/swap.c: piggyback lru_add_drain_all() calls")
> > implemented an optimization mechanism to exit the to-be-started LRU
> > drain operation (name it A) if another drain operation *started and
> > finished* while (A) was blocked on the LRU draining mutex.

That commit is horrible...

> Well, I thought it fits perfectly =)
> 
> Maybe it's worth to add helpers with appropriate semantics?
> This is pretty common pattern.

Where's more sites?

> > @@ -725,21 +735,48 @@ void lru_add_drain_all(void)
> >   	if (WARN_ON(!mm_percpu_wq))
> >   		return;
> > -	seq = raw_read_seqcount_latch(&seqcount);
> >   	mutex_lock(&lock);
> >   	/*
> > -	 * Piggyback on drain started and finished while we waited for lock:
> > -	 * all pages pended at the time of our enter were drained from vectors.
> >   	 */
> > -	if (__read_seqcount_retry(&seqcount, seq))
> >   		goto done;

Since there is no ordering in raw_read_seqcount_latch(), and
mutex_lock() is an ACQUIRE, there's no guarantee the read actually
happens before the mutex is acquired.

> > -	raw_write_seqcount_latch(&seqcount);
> >   	cpumask_clear(&has_work);