[Pkg-xen-devel] Bug#869123: task xenconsoled blocked in state D after reboot

[Hans van Kranenburg]

Package: xen-utils-4.4
Version: 4.4.1-9+deb8u9
Severity: normal

Hi,

Since the latest security update of the xen 4.4 packages we notice that
after rebooting a dom0, there's a chance that the xenconsoled process
ends up hanging in D state.

The kernel logs show these messages repeatedly:

INFO: task xenconsoled:1135 blocked for more than 120 seconds.
      Not tainted 3.16.0-4-amd64 #1
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
xenconsoled     D ffff88017ce2c628     0  1135      1 0x00000000
 ffff88017ce2c1d0 0000000000000286 0000000000012f40 ffff8800776b3fd8
 0000000000012f40 ffff88017ce2c1d0 ffff88017ccc7848 ffff8800776b3f20
 ffff88017ccc784c ffff88017ce2c1d0 00000000ffffffff ffff88017ccc7850
Call Trace:
 [<ffffffff81517095>] ? schedule_preempt_disabled+0x25/0x70
 [<ffffffff81518b33>] ? __mutex_lock_slowpath+0xd3/0x1d0
 [<ffffffff81518c4b>] ? mutex_lock+0x1b/0x2a
 [<ffffffff811c753d>] ? __fdget_pos+0x3d/0x50
 [<ffffffff811aba4a>] ? SyS_write+0x1a/0xa0
 [<ffffffff8151a48d>] ? system_call_fast_compare_end+0x10/0x15

Also:

-# cat /proc/1135/stack
[<ffffffff811c753d>] __fdget_pos+0x3d/0x50
[<ffffffff811aba4a>] SyS_write+0x1a/0xa0
[<ffffffff8151a48d>] system_call_fast_compare_end+0x10/0x15
[<ffffffffffffffff>] 0xffffffffffffffff

I have seen this happening multiple times now, and have never seen it in
the past with Xen on Jessie. The servers used are HP Proliant dl360 gen8
and gen9 types.

I have an empty dom0 in this state currently, so I could do more
investigation right now. If there's quick feedback on this bug, I can
probably keep it blacklisted for usage until monday.

Another problem that might possibly be related is the fact that the
serial console of the dom0 stops giving any output right at '(XEN) Dom0
has maximum 4 VCPUs'. But, that problem is already present for years,
and if that happens, we usually (argh) reboot again until it stays
responding. I haven't ever seen a proper explanation or fix for this
issue in the past years. But it's also happening right now here. When
doing a reboot later, during shutdown, the serial console (here attached
to the HP ilo virtual serial console) will suddenly start printing all
text like everything has been buffered somewhere.

But, I'm not touching this thing right now or trying anything yet. Hope
we can come up with some useful debugging steps that might help finding
this xenconsoled problem.

-- 
Thanks,
Hans van Kranenburg