Re: Found the commit that causes the OOMs

[Mel Gorman <mel@csn.ul.ie>]

On Tue, Jun 30, 2009 at 01:07:41PM +0900, Minchan Kim wrote:
> On Mon, 29 Jun 2009 17:07:25 +0100
> Mel Gorman <mel@csn.ul.ie> wrote:
> 
> > On Tue, Jun 30, 2009 at 12:00:26AM +0900, Minchan Kim wrote:
> > > On Mon, Jun 29, 2009 at 11:21 PM, David Howells<dhowells@redhat.com> wrote:
> > > > Wu Fengguang <fengguang.wu@intel.com> wrote:
> > > >
> > > >> Sorry! This one compiles OK:
> > > >
> > > > Sadly that doesn't seem to work either:
> > > >
> > > > msgctl11 invoked oom-killer: gfp_mask=0x200da, order=0, oom_adj=0
> > > > msgctl11 cpuset=/ mems_allowed=0
> > > > Pid: 30858, comm: msgctl11 Not tainted 2.6.31-rc1-cachefs #146
> > > > Call Trace:
> > > >  [<ffffffff8107207e>] ? oom_kill_process.clone.0+0xa9/0x245
> > > >  [<ffffffff81072345>] ? __out_of_memory+0x12b/0x142
> > > >  [<ffffffff810723c6>] ? out_of_memory+0x6a/0x94
> > > >  [<ffffffff81074a90>] ? __alloc_pages_nodemask+0x42e/0x51d
> > > >  [<ffffffff81080843>] ? do_wp_page+0x2c6/0x5f5
> > > >  [<ffffffff810820c1>] ? handle_mm_fault+0x5dd/0x62f
> > > >  [<ffffffff81022c32>] ? do_page_fault+0x1f8/0x20d
> > > >  [<ffffffff812e069f>] ? page_fault+0x1f/0x30
> > > > Mem-Info:
> > > > DMA per-cpu:
> > > > CPU    0: hi:    0, btch:   1 usd:   0
> > > > CPU    1: hi:    0, btch:   1 usd:   0
> > > > DMA32 per-cpu:
> > > > CPU    0: hi:  186, btch:  31 usd:  38
> > > > CPU    1: hi:  186, btch:  31 usd: 106
> > > > Active_anon:75040 active_file:0 inactive_anon:2031
> > > >  inactive_file:0 unevictable:0 dirty:0 writeback:0 unstable:0
> > > >  free:1951 slab:41499 mapped:301 pagetables:60674 bounce:0
> > > > DMA free:3932kB min:60kB low:72kB high:88kB active_anon:2868kB inactive_anon:384kB active_file:0kB inactive_file:0kB unevictable:0kB present:15364kB pages_scanned:0 all_unreclaimable? no
> > > > lowmem_reserve[]: 0 968 968 968
> > > > DMA32 free:3872kB min:3948kB low:4932kB high:5920kB active_anon:297292kB inactive_anon:7740kB active_file:0kB inactive_file:0kB unevictable:0kB present:992032kB pages_scanned:0 all_unreclaimable? no
> > > > lowmem_reserve[]: 0 0 0 0
> > > > DMA: 7*4kB 0*8kB 0*16kB 0*32kB 1*64kB 0*128kB 1*256kB 1*512kB 1*1024kB 1*2048kB 0*4096kB = 3932kB
> > > > DMA32: 500*4kB 2*8kB 0*16kB 0*32kB 1*64kB 0*128kB 1*256kB 1*512kB 1*1024kB 0*2048kB 0*4096kB = 3872kB
> > > > 1928 total pagecache pages
> > > > 0 pages in swap cache
> > > > Swap cache stats: add 0, delete 0, find 0/0
> > > > Free swap  = 0kB
> > > > Total swap = 0kB
> > > > 255744 pages RAM
> > > > 5589 pages reserved
> > > > 238251 pages shared
> > > > 216210 pages non-shared
> > > > Out of memory: kill process 25221 (msgctl11) score 130560 or a child
> > > > Killed process 26379 (msgctl11)
> > > 
> > > Totally, I can't understand this situation.
> > > Now, this page allocation is order zero and It is just likely GFP_HIGHUSER.
> > > So it's unlikely interrupt context.
> > 
> > The GFP flags that are set are
> > 
> > #define __GFP_HIGHMEM	(0x02)
> > #define __GFP_MOVABLE	(0x08)  /* Page is movable */
> > #define __GFP_WAIT	(0x10)	/* Can wait and reschedule? */
> > #define __GFP_IO	(0x40)	/* Can start physical IO? */
> > #define __GFP_FS	(0x80)	/* Can call down to low-level FS? */
> > #define __GFP_HARDWALL   (0x20000) /* Enforce hardwall cpuset memory allocs */
> > 
> > which are fairly permissive in terms of what action can be taken.
> > 
> > > Buddy already has enough fallback DMA32, I think.
> > 
> > It doesn't really. We are below the minimum watermark. It wouldn't be
> > able to grant the allocation until a few pages had been freed.
> 
> Yes. I missed that. 
> 
> > > Why kernel can't allocate page for order 0 ?
> > > Is it allocator bug ?
> > > 
> > 
> > If it is, it is not because the allocation failed as the watermarks were not
> > being met. For this situation to be occuring, it has to be scanning the LRU
> > lists and making no forward progress. Odd things to note;
> > 
> > o active_anon is very large in comparison to inactive_anon. Is this
> >   because there is no swap and they are no longer being rotated?
> 
> Yes. My patch's intention was that. 
> 
>        commit 69c854817566db82c362797b4a6521d0b00fe1d8
>        Author: MinChan Kim <minchan.kim@gmail.com>
>        Date:   Tue Jun 16 15:32:44 2009 -0700
> 
> > o Slab and pagetables are very large. Is slab genuinely unshrinkable?
> >
> > I think this system might be genuinely OOM. It can't reclaim memory and
> > we are below the minimum watermarks.
> > 
> > Is it possible there are pages that are counted as active_anon that in
> > fact are reclaimable because they are on the wrong LRU list? If that was
> > the case, the lack of rotation to inactive list would prevent them
> > getting discovered.
> 
> I agree. 
> One of them is that "[BUGFIX][PATCH] fix lumpy reclaim lru handiling at
> isolate_lru_pages v2" as Kosaki already said. 
> 
> Unfortunately, David said it's not. 
> But I think your guessing make sense. 
> 
> David. Doesn't it happen OOM if you revert my patch, still?
> 

In the event the OOM does not happen with the patch reverted, I suggest
you put together a debugging patch that prints out details of all pages
on the active_anon LRU list in the event of an OOM. The intention is to
figure out what pages are on the active_anon list that shouldn't be.

-- 
Mel Gorman
Part-time Phd Student                          Linux Technology Center
University of Limerick                         IBM Dublin Software Lab