Re: [PATCH 3/5] page allocator: Wait on both sync and async congestion after direct reclaim

[Pekka Enberg <penberg-bbCR+/B0CizivPeTLB3BmA@public.gmane.org>]

Hi Jens,

On Fri, Nov 13, 2009 at 3:32 PM, Jens Axboe <jens.axboe-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org> wrote:
>> Suggest an alternative that brings congestion_wait() more in line with
>> 2.6.30 behaviour then.
>
> I don't have a good explanation as to why the delays have changed,
> unfortunately. Are we sure that they have between .30 and .31? The
> dm-crypt case is overly complex and lots of changes could have broken
> that house of cards.

Hand-waving or not, we have end user reports stating that reverting
commit 8aa7e847d834ed937a9ad37a0f2ad5b8584c1ab0 ("Fix
congestion_wait() sync/async vs read/write confusion") fixes their
(rather serious) OOM regression. The commit in question _does_
introduce a functional change and if this was your average regression,
people would be kicking and screaming to get it reverted.

So is there a reason we shouldn't send a partial revert of the commit
(switching to BLK_RW_SYNC) to Linus until the "real" issue gets
resolved? Yes, I realize it's ugly voodoo magic but dammit, it used to
work!

                        Pekka