Re: [PATCH v3] mm/hwpoison: do not lock page again when me_huge_page() successfully recovers

["HORIGUCHI NAOYA(堀口 直也)" <naoya.horiguchi@nec.com>]

On Mon, Jun 14, 2021 at 07:35:28PM -0700, Andrew Morton wrote:
> On Fri, 11 Jun 2021 00:23:29 +0000 HORIGUCHI NAOYA(堀口　直也) <naoya.horiguchi@nec.com> wrote:
> 
> > > 
> > > --- mm/memory-failure.c
> > > +++ mm/memory-failure.c
> > > @@ -1782,6 +1796,8 @@ int memory_failure(unsigned long pfn, int flags)
> > >  
> > >  identify_page_state:
> > >  	res = identify_page_state(pfn, p, page_flags);
> > > +	mutex_unlock(&mf_mutex);
> > > +	return res;
> > >  unlock_page:
> > >  	unlock_page(p);
> > >  unlock_mutex:
> > > 
> > > and...  That mutex_unlock() looks odd.  The patch adds no matching
> > > mutex_lock?
> > 
> > Yes, memory_failure() already has one mutex_lock (introduced by
> > mm-memory-failure-use-a-mutex-to-avoid-memory_failure-races.patch,
> > sorry for not clarifying that), and the change introduces a separate
> > return path.  But I now think that I should have used "goto unlock_mutex"
> > to use existing return path.
> 
> But mm-memory-failure-use-a-mutex-to-avoid-memory_failure-races.patch
> is part of Tony's three patch series which is not marked for -stable. 
> So it isn't appropriate that this patch be based on top of that three
> patch series.
> 
> Should Tony's patchset also be targeted to -stable?  If so then OK.

Yes, that's fine. And I think that the first two patches
(mm-memory-failure-use-a-mutex-to-avoid-memory_failure-races.pathc and
mmhwpoison-return-ehwpoison-to-denote-that-the-page-has-already-been-poisoned.patch)
can be marked to stable, but 3/3 (mmhwpoison-send-sigbus-with-error-virutal-address.patch)
may not because it's a little too large and not the main part of the fixes.

Thanks,
Naoya Horiguchi

> 
> If not then please let's prepare your -stable patch against current
> mainline, as it is higher priority than the 5.14-rc1 material in
> linux-next.
>