Re: init_emergency_isa_pool calling mempool_create in non-sleeping context

[Mike Snitzer <snitzer@redhat.com>]

On Wed, Jun 17 2009 at  6:42pm -0400,
Mike Snitzer <snitzer@redhat.com> wrote:

> On Wed, Jun 17 2009 at  6:22pm -0400,
> NeilBrown <neilb@suse.de> wrote:
> 
> > 
> > You wanted dm-devel@redhat.com, not linux-raid@vger.kernel.org.
> > 
> > NeilBrown
> > 
> > 
> > On Thu, June 18, 2009 7:35 am, Jeremy Fitzhardinge wrote:
> > > I just saw this when booting a current linux-2.6.git kernel
> > > (3fe0344faf7fdcb158bd5c1a9aec960a8d70c8e8).
> > > .config and full bootlog attached.
> > >
> > > BUG: sleeping function called from invalid context at
> > > /home/jeremy/git/linux/mm/slab.c:3029
> > > in_atomic(): 1, irqs_disabled(): 0, pid: 388, name: lvm
> > > Pid: 388, comm: lvm Not tainted 2.6.30 #225
> > > Call Trace:
> > >  [<ffffffff81137a65>] ? __init_rwsem+0x29/0x39
> > >  [<ffffffff81035346>] __might_sleep+0x110/0x112
> > >  [<ffffffff810acf07>] kmem_cache_alloc+0x38/0x12b
> > >  [<ffffffff810a71d4>] ? mempool_alloc_pages_isa+0x0/0x10
> > >  [<ffffffff8108a5df>] mempool_create_node+0x2f/0x152
> > >  [<ffffffff8108a31c>] ? mempool_free_pages+0x0/0xb
> > >  [<ffffffff8108a70f>] mempool_create+0xd/0xf
> > >  [<ffffffff810a71ad>] init_emergency_isa_pool+0x28/0x4f
> > >  [<ffffffff81127b90>] blk_queue_bounce_limit+0x40/0x56
> > >  [<ffffffffa00f856c>] dm_table_set_restrictions+0x82/0xa6 [dm_mod]
> > >  [<ffffffffa00f6415>] dm_swap_table+0x132/0x15a [dm_mod]
> > >  [<ffffffffa00faf61>] dev_suspend+0xfd/0x177 [dm_mod]
> > >  [<ffffffffa00fb959>] dm_ctl_ioctl+0x252/0x2a2 [dm_mod]
> > >  [<ffffffffa00fae64>] ? dev_suspend+0x0/0x177 [dm_mod]
> > >  [<ffffffff810bd4c0>] vfs_ioctl+0x2a/0x77
> > >  [<ffffffff810bd9b4>] do_vfs_ioctl+0x4a7/0x4fa
> > >  [<ffffffff810bda5e>] sys_ioctl+0x57/0x7a
> > >  [<ffffffff81011042>] system_call_fastpath+0x16/0x1b
> 
> 
> This has been resolved and is waiting to be pushed along with the rest
> of the DM changes for 2.6.31.
> 
> See:
> ftp://ftp.kernel.org/pub/linux/kernel/people/agk/patches/2.6/editing/dm-table-establish-queue-limits-by-copying-table-limits.patch
> 
> Maybe this fix should go now rather than wait a day or so for the larger
> DM push?

But in order to do so this patch needs to be merged first:

ftp://ftp.kernel.org/pub/linux/kernel/people/agk/patches/2.6/editing/dm-table-replace-struct-io_restrictions-with-struct-queue_limits.patch

I'll defer to Alasdair on whether we want to push the DM changes in 2
batches in light of the current linux-2.6 DM breakage that occurred as a
side-effect of some block changes.

Mike