[Intel-gfx] mm/huge_memory: do not clobber swp_entry_t during THP split

[Mel Gorman]

On Tue, Oct 25, 2022 at 08:26:06AM -0700, Hugh Dickins wrote:
> > 
> > > If so I
> > > can temporarily put it in until it arrives via the next rc - assuming that
> > > would be the flow from upstream pov?
> 
> The right thing for now is for GregKH to drop Mel's from 6.0.4:
> I've just sent a mail asking for that (I would have asked yesterday,
> but mistook that GregKH was not in Cc).
> 

Thanks for catching that, I only saw the mail this morning that it had been
picked up as a stable candidate and was internally screaming "no no no"
until I saw your mail :P. I added the warning thinking "we have almost a
full rc cycle to catch any additional fallout".

> Of course Mel's fix is much more important than the harmless
> (unless panic on warn) warning, but let's delay it a few more days,
> it just flowed into stable too quickly.
> 
> Thanks Mel: I never knowingly hit the THP_SWAP issue which your patch
> is fixing, but it now looks like it was also responsible for mysterious
> occasional OOM kills that I had been chasing for weeks.
> 

I'm glad it helped! I worried that the additional warning would trigger an
excessive number of new bugs but it served its intended purpose -- catch
fallout from clobbering page->private causing subtle bugs later that are
hard to debug.

-- 
Mel Gorman
SUSE Labs