Re: 2.6.39-rc3, 2.6.39-rc4: XFS lockup - regression since 2.6.38

[Markus Trippelsdorf]

On 2011.04.29 at 11:19 +1000, Dave Chinner wrote:
> On Thu, Apr 28, 2011 at 09:45:28PM +0200, Markus Trippelsdorf wrote:
> > On 2011.04.27 at 18:26 +0200, Bruno Prémont wrote:
> > > On Wed, 27 April 2011 Dave Chinner <david@xxxxxxxxxxxxx> wrote:
> > > > On Sat, Apr 23, 2011 at 10:44:03PM +0200, Bruno Prémont wrote:
> > > > > Running 2.6.39-rc3+ and now again on 2.6.39-rc4+ (I've not tested -rc1
> > > > > or -rc2) I've hit a "dying machine" where processes writing to disk end
> > > > > up in D state.
> > > > > From occurrence with -rc3+ I don't have logs as those never hit the disk,
> > > > > for -rc4+ I have the following (sysrq+t was too big, what I have of it
> > > > > misses a dozen of kernel tasks - if needed, please ask):
> > > > > 
> > > > > The -rc4 kernel is at commit 584f79046780e10cb24367a691f8c28398a00e84
> > > > > (+ 1 patch of mine to stop disk on reboot),
> > > > > full dmesg available if needed; kernel config attached (only selected
> > > > > options). In case there is something I should do at next occurrence
> > > > > please tell. Unfortunately I have no trigger for it and it does not
> > > > > happen very often.
> > > > > 
> > > > > [32040.120055] INFO: task flush-8:0:1665 blocked for more than 120 seconds.
> > > > > [32040.120068] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
> > > > > [32040.120077] flush-8:0       D 00000000  4908  1665      2 0x00000000
> > > > > [32040.120099]  f55efb5c 00000046 00000000 00000000 00000000 00000001 e0382924 00000000
> > > > > [32040.120118]  f55efb0c f55efb5c 00000004 f629ba70 572f01a2 00001cfe f629ba70 ffffffc0
> > > > > [32040.120135]  f55efc68 f55efb30 f889d7f8 f55efb20 00000000 f55efc68 e0382900 f55efc94
> > > > > [32040.120153] Call Trace:
> > > > > [32040.120220]  [<f889d7f8>] ? xfs_bmap_search_multi_extents+0x88/0xe0 [xfs]
> > > > > [32040.120239]  [<c109ce1d>] ? kmem_cache_alloc+0x2d/0x110
> > > > > [32040.120294]  [<f88c88ca>] ? xlog_space_left+0x2a/0xc0 [xfs]
> > > > > [32040.120346]  [<f88c85cb>] xlog_wait+0x4b/0x70 [xfs]
> > > > > [32040.120359]  [<c102ca00>] ? try_to_wake_up+0xc0/0xc0
> > > > > [32040.120411]  [<f88c948b>] xlog_grant_log_space+0x8b/0x240 [xfs]
> > > > > [32040.120464]  [<f88c936e>] ? xlog_grant_push_ail+0xbe/0xf0 [xfs]
> > > > > [32040.120516]  [<f88c99db>] xfs_log_reserve+0xab/0xb0 [xfs]
> > > > > [32040.120571]  [<f88d6dc8>] xfs_trans_reserve+0x78/0x1f0 [xfs]
> > > > 
> > > > Hmmmmm. That may be caused by the conversion of the xfsaild to a
> > > > work queue. Can you post the output of "xfs_info <mntpt>" and the
> > > > mount options (/proc/mounts) used on you system?
> > 
> > I may have hit the same problem today and managed to capture some sysrq-l
> > and sysrq-w output. 
> > 
> > The system was largely unusable during this incident. I could still
> > switch between X and the console (and press the sysrq key-combination),
> > but I couldn't run any commands in the terminal.
> 
> OK, so the common elements here appears to be root filesystems
> with small log sizes, which means they are tail pushing all the
> time metadata operations are in progress. Definitely seems like a
> race in the AIL workqueue trigger mechanism. I'll see if I can
> reproduce this and cook up a patch to fix it.

Hmm, I'm wondering if this issue is somehow related to the hrtimer bug,
that Thomas Gleixner fixed yesterday:
http://git.us.kernel.org/?p=linux/kernel/git/tip/linux-2.6-tip.git;a=commit;h=ce31332d3c77532d6ea97ddcb475a2b02dd358b4
http://thread.gmane.org/gmane.linux.kernel.mm/61909/

It also looks similar to the issue that James Bottomley reported
earlier: http://thread.gmane.org/gmane.linux.kernel.mm/62185/ 

-- 
Markus
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/