Re: CMA broken in next-20120926

[Mel Gorman <mgorman@suse.de>]

On Tue, Oct 09, 2012 at 01:08:30PM +0200, Bartlomiej Zolnierkiewicz wrote:
> On Tuesday 09 October 2012 12:11:43 Mel Gorman wrote:
> > On Tue, Oct 09, 2012 at 10:40:10AM +0200, Bartlomiej Zolnierkiewicz wrote:
> > > I also need following patch to make CONFIG_CMA=y && CONFIG_COMPACTION=y case
> > > work:
> > > 
> > > From: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
> > > Subject: [PATCH] mm: compaction: cache if a pageblock was scanned and no pages were isolated - cma fix
> > > 
> > > Patch "mm: compaction: cache if a pageblock was scanned and no pages
> > > were isolated" needs a following fix to successfully boot next-20121002
> > > kernel (same with next-20121008) with CONFIG_CMA=y and CONFIG_COMPACTION=y
> > > (with applied -fix1, -fix2, -fix3 patches from Mel Gorman and also with
> > > cmatest module from Thierry Reding compiled in).
> > > 
> > 
> > Why is it needed to make it boot? CMA should not care about the
> 
> It boots without Thierry's cmatest module but then fails on CMA
> allocation attempt (I used out-of-tree /dev/cma_test interface to
> generate CMA allocation request from user-space).
> 

I see.

> > PG_migrate_skip hint being set because it should always ignore it in
> > alloc_contig_range() due to cc->ignore_skip_hint. It's not obvious to
> > me why this fixes a boot failure and I wonder if it's papering over some
> > underlying problem. Can you provide more details please?
> 
> I just compared CONFIG_COMPACTION=n and =y cases initially, figured
> out the difference and did the change.  However on a closer look it
> seems that {get,clear,set}_pageblock_skip() use incorrect bit ranges
> (please compare to bit ranges used by {get,set}_pageblock_flags()
> used for migration types) and can overwrite pageblock migratetype of
> the next pageblock in the bitmap

You're right.

> (I wonder how could this code ever
> worked before?).
> 

Because it would corrupt the adjacent block and muck up the migratetype.
This would hurt fragmentation avoidance but it would not be obvious for
a long time. When I was checking this, I only checked that the values of
the block being set were correct, I missed that I was corrupting the
adjacent blocks.

> > > Signed-off-by: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
> > > Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
> > > ---
> > >  mm/compaction.c |    3 ++-
> > >  1 file changed, 2 insertions(+), 1 deletion(-)
> > > 
> > > Index: b/mm/compaction.c
> > > ===================================================================
> > > --- a/mm/compaction.c	2012-10-08 18:10:53.491679716 +0200
> > > +++ b/mm/compaction.c	2012-10-08 18:11:33.615679713 +0200
> > > @@ -117,7 +117,8 @@ static void update_pageblock_skip(struct
> > >  			bool migrate_scanner)
> > >  {
> > >  	struct zone *zone = cc->zone;
> > > -	if (!page)
> > > +
> > > +	if (!page || cc->ignore_skip_hint)
> > >  		return;
> > >  
> > >  	if (!nr_isolated) {
> 
> The patch below also fixes the issue for me:
> 
> From: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
> Subject: [PATCH] mm: compaction: fix bit ranges in {get,clear,set}_pageblock_skip() 
> 
> {get,clear,set}_pageblock_skip() use incorrect bit ranges (please compare
> to bit ranges used by {get,set}_pageblock_flags() used for migration types)
> and can overwrite pageblock migratetype of the next pageblock in the bitmap.
> 
> Signed-off-by: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
> Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>

Acked-by: Mel Gorman <mgorman@suse.de>

Can you resend this patch on its own to Andrew asking for it to be picked
up please? This thread is quite long and could easily get lost in the
noise. Thanks

-- 
Mel Gorman
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>