Re: lockdep splat due to reclaim recursion detected

From	Dave Chinner <david@fromorbit.com>
Newsgroups	linux.kernel
Subject	Re: lockdep splat due to reclaim recursion detected
Date	2016-10-07 12:40 +0200
Message-ID	<spAJj-7vM-1@gated-at.bofh.it> (permalink)
References	<spyo9-5NZ-1@gated-at.bofh.it> <spyxP-5Vj-9@gated-at.bofh.it>
Organization	linux.* mail to news gateway

Show all headers | View raw

On Fri, Oct 07, 2016 at 10:13:40AM +0200, Michal Hocko wrote:
> Fix up xfs ML address
> 
> On Fri 07-10-16 10:07:39, Michal Hocko wrote:
> > Hi Dave,
> > while playing with the test case you have suggested [1], I have hit the
> > following lockdep splat. This is with mmotm git tree [2] but I didn't
> > get to retest with the current linux-next (or any other tree of your
> > preference) so there is a chance that something is broken in my tree so
> > take this as a heads up. As soon as I am done with testing of the patch
> > in the above email thread I will retest with linux-next.
....
> > [   61.878792] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Debian-1.8.2-1 04/01/2014
> > [   61.878792]  0000000000000000 ffff88001c8b3718 ffffffff81312f78 ffffffff825d99d0
> > [   61.878792]  ffff88001cd04880 ffff88001c8b3750 ffffffff811260d1 000000000000000a
> > [   61.878792]  ffff88001cd05178 ffff88001cd04880 ffffffff81095395 ffff88001cd04880
> > [   61.878792] Call Trace:
> > [   61.878792]  [<ffffffff81312f78>] dump_stack+0x68/0x92
> > [   61.878792]  [<ffffffff811260d1>] print_usage_bug.part.26+0x25b/0x26a
> > [   61.878792]  [<ffffffff81095395>] ? print_shortest_lock_dependencies+0x17f/0x17f
> > [   61.878792]  [<ffffffff81096074>] mark_lock+0x381/0x56d
> > [   61.878792]  [<ffffffff810962be>] mark_held_locks+0x5e/0x74
> > [   61.878792]  [<ffffffff8109875c>] lockdep_trace_alloc+0xaf/0xb2
> > [   61.878792]  [<ffffffff8117d0f7>] kmem_cache_alloc_trace+0x3a/0x270
> > [   61.878792]  [<ffffffff81169454>] ? vm_map_ram+0x2d2/0x4a6
> > [   61.878792]  [<ffffffff8116924b>] ? vm_map_ram+0xc9/0x4a6
> > [   61.878792]  [<ffffffff81169454>] vm_map_ram+0x2d2/0x4a6
> > [   61.878792]  [<ffffffffa0051069>] _xfs_buf_map_pages+0xae/0x10b [xfs]
> > [   61.878792]  [<ffffffffa0052cd0>] xfs_buf_get_map+0xaa/0x24f [xfs]
> > [   61.878792]  [<ffffffffa0081d10>] xfs_trans_get_buf_map+0x144/0x2ef [xfs]

Aw, come on! I explained this lockdep annotation bug a couple of
days ago.

https://www.spinics.net/lists/linux-fsdevel/msg102588.html

And this isn't the first time I've explained this either. ISTR this
same issue triggered a long whole discussion about how to move
memory allocation to task based context flags or to push more
context specific information into the shrinkers so they could decide
if the needed to avoid deadlocks or not. That was about 6 months
ago, IIRC, and there's been no followup from the mm side of
things...

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

Thread

lockdep splat due to reclaim recursion detected Michal Hocko <mhocko@kernel.org> - 2016-10-07 10:10 +0200
  Re: lockdep splat due to reclaim recursion detected Michal Hocko <mhocko@kernel.org> - 2016-10-07 10:20 +0200
    Re: lockdep splat due to reclaim recursion detected Dave Chinner <david@fromorbit.com> - 2016-10-07 12:40 +0200
      Re: lockdep splat due to reclaim recursion detected Michal Hocko <mhocko@kernel.org> - 2016-10-07 13:00 +0200

csiph-web