Re: [linux-pm] [RFC][PATCH] PM: Force GFP_NOIO during suspend/resume (was: Re: Memory allocations in .suspend became very unreliable)

[Rafael J. Wysocki]

On Monday 18 January 2010, KOSAKI Motohiro wrote:
> > Index: linux-2.6/mm/page_alloc.c
> > ===================================================================
> > --- linux-2.6.orig/mm/page_alloc.c
> > +++ linux-2.6/mm/page_alloc.c
> > @@ -1963,10 +1963,13 @@ __alloc_pages_nodemask(gfp_t gfp_mask, u
> >  	page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
> >  			zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
> >  			preferred_zone, migratetype);
> > -	if (unlikely(!page))
> > +	if (unlikely(!page)) {
> > +		mm_lock_suspend(gfp_mask);
> >  		page = __alloc_pages_slowpath(gfp_mask, order,
> >  				zonelist, high_zoneidx, nodemask,
> >  				preferred_zone, migratetype);
> > +		mm_unlock_suspend(gfp_mask);
> > +	}
> >  
> >  	trace_mm_page_alloc(page, order, gfp_mask, migratetype);
> >  	return page;
> 
> I think we don't need read side lock at all. generally, no lock might makes race.
> But in this case, changing gfp_allowed_mask and nvidia suspend method should be
> serialized higher level. Why the above two code need to run concurrently?

The changing of gfp_allowed_mask is serialized with the suspend of devices,
so there's no concurrency here.

I was concerned about another problem, though, which is what happens if the
suspend process runs in parallel with a memory allocation that started earlier
and happens to do some I/O.  I that case the suspend process doesn't know
about the I/O done by the mm subsystem and may disturb it in principle.

That said, perhaps that should be a concern for the block devices subsystem to
prevent such situations from happening.

So, perhaps I'll remove the reader-side lock altogether and go back to
something like the first version of the patch.

Rafael
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/