Firstly, to avoid the confusion, let me repeat I think your patch is fine.
I too thought that non-exclusive waiters do not care about the bit state
and thus wake_page_function() can simply wake them all up.
But then I did "git blame", found your commit 3510ca20ece0150 and came to
conclusion there are reasons we should not do this.
On 07/25, Linus Torvalds wrote:
>
> On Sat, Jul 25, 2020 at 3:14 AM Oleg Nesterov <oleg@xxxxxxxxxx> wrote:
> >
> > In essense, this partly reverts your commit 3510ca20ece0150
> > ("Minor page waitqueue cleanups"). I mean this part:
>
> Well, no. I mean, it keeps the "always add to the fail" behavior.
Ah, sorry for confusion, this doesn't matter. I didn't mean "fairness".
What I tried to say. AFAICS before that commit we had (almost) the same
behaviour you propose now: unlock_page/etc wakes all the non-exclusive
waiters up.
No?
Or I misunderstood your reply? Quite possibly, too late for me...
Oleg.
