[Intel-gfx] [PATCH 4/4] drm/i915: Invalidate our pages under memory pressure

[Chris Wilson]

On Fri, Apr 11, 2014 at 10:38:25AM +0200, Daniel Vetter wrote:
> On Fri, Apr 11, 2014 at 09:30:20AM +0100, Chris Wilson wrote:
> > On Tue, Mar 25, 2014 at 01:23:06PM +0000, Chris Wilson wrote:
> > > Try to flush out dirty pages into the swapcache (and from there into the
> > > swapfile) when under memory pressure and forced to drop GEM objects from
> > > memory. In effect, this should just allow us to discard unused pages for
> > > memory reclaim and to start writeback earlier.
> > > 
> > > v2: Hugh Dickins warned that explicitly starting writeback from
> > > shrink_slab was prone to deadlocks within shmemfs.
> > > 
> > > Cc: Hugh Dickins <hughd at google.com>
> > > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> > 
> > Good news! QA have declared that this series really does prevent the
> > random OOM where we have completely unused swap. So all it needs is
> > someone brave enough to review.
> 
> Awesome work.
> 
> Do we need the additional patch you've just posted to improve the
> writeback stalling after calling shrinkers too, or is that not required?

That looks to be required (or at least I hope it provokes the mm gods
into doing something sensible) for a different problem. However, that
test is still behaving strangely (inactive_anon =~ 2x shmem, it should
be almost equal) as it appears that there is severe overallocation, or a
leak.

But that we can generate massive amounts of writeback from our shrinker
which may not be cleared in time for the allocation to succeed is a
problem (addressed by that mm patch).
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre