Re: [PATCH 1/2] mm, oom: introduce oom reaper

[Michal Hocko <mhocko@kernel.org>]

On Thu 24-12-15 13:44:03, Ross Zwisler wrote:
> On Thu, Dec 24, 2015 at 2:47 AM, Michal Hocko <mhocko@kernel.org> wrote:
> > On Wed 23-12-15 16:00:09, Ross Zwisler wrote:
> > [...]
> >> While running xfstests on next-20151223 I hit a pair of kernel BUGs
> >> that bisected to this commit:
> >>
> >> 1eb3a80d8239 ("mm, oom: introduce oom reaper")
> >
> > Thank you for the report and the bisection.
> >
> >> Here is a BUG produced by generic/029 when run against XFS:
> >>
> >> [  235.751723] ------------[ cut here ]------------
> >> [  235.752194] kernel BUG at mm/filemap.c:208!
> >
> > This is VM_BUG_ON_PAGE(page_mapped(page), page), right? Could you attach
> > the full kernel log? It all smells like a race when OOM reaper tears
> > down the mapping and there is a truncate still in progress. But hitting
> > the BUG_ON just because of that doesn't make much sense to me. OOM
> > reaper is essentially MADV_DONTNEED. I have to think about this some
> > more, though, but I am in a holiday mode until early next year so please
> > bear with me.
> 
> The two stack traces were gathered with next-20151223, so the line numbers
> may have moved around a bit when compared to the actual "mm, oom: introduce
> oom reaper" commit.

I was looking at the same next tree, I believe
$ git describe
next-20151223

[...]
> > There was a warning before this triggered. The full kernel log would be
> > helpful as well.
> 
> Sure, I can gather full kernel logs, but it'll probably after the new year.

OK, I will wait for the logs. It is really interesting to see what was
the timing between OOM killer invocation and this trace.

> > [...]
> >> [  609.425325] Call Trace:
> >> [  609.425797]  [<ffffffff811dc307>] invalidate_inode_pages2+0x17/0x20
> >> [  609.426971]  [<ffffffff81482167>] xfs_file_read_iter+0x297/0x300
> >> [  609.428097]  [<ffffffff81259ac9>] __vfs_read+0xc9/0x100
> >> [  609.429073]  [<ffffffff8125a319>] vfs_read+0x89/0x130
> >> [  609.430010]  [<ffffffff8125b418>] SyS_read+0x58/0xd0
> >> [  609.430943]  [<ffffffff81a527b2>] entry_SYSCALL_64_fastpath+0x12/0x76
> >> [  609.432139] Code: 85 d8 fe ff ff 01 00 00 00 f6 c4 40 0f 84 59 ff
> >> ff ff 49 8b 47 20 48 8d 78 ff a8 01 49 0f 44 ff 8b 47 48 85 c0 0f 88
> >> bd 01 00 00 <0f> 0b 4d 3b 67 08 0f 85 70 ff ff ff 49 f7 07 00 18 00 00
> >> 74 15
> > [...]
> >> My test setup is a qemu guest machine with a pair of 4 GiB PMEM
> >> ramdisk test devices, one for the xfstest test disk and one for the
> >> scratch disk.
> >
> > Is this just a plain ramdisk device or it needs a special configuration?
> 
> Just a plain PMEM ram disk with DAX turned off.  Configuration instructions
> for PMEM can be found here:
> 
> https://nvdimm.wiki.kernel.org/

Thanks I will try to reproduce early next year. But so far I think this
is just a general issue of MADV_DONTNEED vs. truncate and oom_reaper is
just lucky to trigger it. There shouldn't be anything oom_reaper
specific here. Maybe there is some additional locking missing?

-- 
Michal Hocko
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>