Re: mcelog stalls on 2.6.39-rc5

[Paul E. McKenney]

On Thu, Apr 28, 2011 at 10:06:57AM -0600, Jim Schutt wrote:
> Jim Schutt wrote:
> >Paul E. McKenney wrote:
> >>On Wed, Apr 27, 2011 at 01:59:01PM -0600, Jim Schutt wrote:
> >>>Hi,
> >>>
> >>>Testing 2.6.39-rc5 is giving me the following stall:
> >>>
> >>>    [ 5767.731001] INFO: rcu_sched_state detected stall on CPU
> >>>1 (t=60001 jiffies)
> >>>    [ 5767.732001] INFO: rcu_sched_state detected stalls on
> >>>CPUs/tasks: { 1} (detected by 0, t=60002 jiffies)
> >>>    [ 5947.763001] INFO: rcu_sched_state detected stall on CPU
> >>>1 (t=240032 jiffies)
> >>>    [ 5947.764001] INFO: rcu_sched_state detected stalls on
> >>>CPUs/tasks: { 1} (detected by 0, t=240034 jiffies)
> >>>    [ 6024.489362] libceph: mon0 172.17.40.34:6789 socket closed
> >>>    [ 6121.281139] INFO: task mcelog:6513 blocked for more
> >>>than 120 seconds.
> >>>    [ 6121.287575] "echo 0 >
> >>>/proc/sys/kernel/hung_task_timeout_secs" disables this
> >>>message.
> >>>    [ 6121.295397]  ffff880177aefce8 0000000000000082
> >>>ffffffff810339b6 ffff880226d35a40
> >>>    [ 6121.302840]  ffff88018226c3b0 ffff88018226c3b0
> >>>0000000000011e80 ffff880226d35a40
> >>>    [ 6121.310284]  ffff88018226c760 ffff880177aefe80
> >>>ffff880177aefd18 ffffffff813af047
> >>>    [ 6121.317725] Call Trace:
> >>>    [ 6121.320176]  [<ffffffff810339b6>] ?
> >>>calc_load_account_idle+0xe/0x1d
> >>>    [ 6121.326437]  [<ffffffff813af047>] schedule+0x159/0x193
> >>>    [ 6121.331569]  [<ffffffff813af449>] schedule_timeout+0x36/0xe2
> >>>    [ 6121.337223]  [<ffffffff810ad9eb>] ? trace_hardirqs_on+0x9/0x20
> >>>    [ 6121.343047]  [<ffffffff813aee2b>] do_wait_for_common+0x97/0xe3
> >>>    [ 6121.348967]  [<ffffffff8103e8fa>] ? try_to_wake_up+0x200/0x200
> >>>    [ 6121.354794]  [<ffffffff8107a4bf>] ? __raw_spin_lock_irq+0x17/0x2f
> >>>    [ 6121.360878]  [<ffffffff813af2a9>] wait_for_common+0x36/0x4d
> >>>    [ 6121.366441]  [<ffffffff813af378>] wait_for_completion+0x1d/0x1f
> >>>    [ 6121.372356]  [<ffffffff8109ae6d>] synchronize_sched+0x40/0x49
> >>>    [ 6121.378096]  [<ffffffff810635b8>] ? find_get_pid+0x1b/0x1b
> >>>    [ 6121.383574]  [<ffffffff81015fe5>] mce_read+0x17f/0x25d
> >>>    [ 6121.388707]  [<ffffffff81111af5>] ? rw_verify_area+0xac/0xdb
> >>>    [ 6121.394358]  [<ffffffff811121f1>] vfs_read+0xa9/0xe1
> >>>    [ 6121.399317]  [<ffffffff8111248d>] sys_read+0x4c/0x70
> >>>    [ 6121.404278]  [<ffffffff813b6deb>] system_call_fastpath+0x16/0x1b
> >>>    [ 6127.795001] INFO: rcu_sched_state detected stall on CPU
> >>>1 (t=420064 jiffies)
> >>>    [ 6127.796001] INFO: rcu_sched_state detected stalls on
> >>>CPUs/tasks: { 1} (detected by 0, t=420066 jiffies)
> >>>    [ 6241.410171] INFO: task mcelog:6513 blocked for more
> >>>than 120 seconds.
> >>>
> >>>Reverting commit a4dd99250dc makes the stalls go away:
> >>>
> >>>    rcu: create new rcu_access_index() and use in mce
> >>>
> >>>    The MCE subsystem needs to sample an RCU-protected index outside of
> >>>    any protection for that index.  If this was a pointer, we would use
> >>>    rcu_access_pointer(), but there is no corresponding
> >>>rcu_access_index().
> >>>    This commit therefore creates an rcu_access_index() and applies it
> >>>    to MCE.
> >>>
> >>>    Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
> >>>    Tested-by: Zdenek Kabelac <zkabelac@xxxxxxxxxx>
> >>
> >>Wow!  This is just instructions, just wrapping the access in
> >>ACCESS_ONCE().
> >>
> >>Was the original RCU CPU stall repeatable?
> >
> >Yes.  I saw it on two different machines.
> >Both were running as Ceph clients, doing sustained
> >streaming writes, if that helps any.
> >
> >I can attempt to repeat with any extra debugging
> >you'd like me to try.
> 
> Evidently I cannot reproduce this after all, so my revert
> test was incorrect.
> 
> If I can learn how to reproduce this I'll post with details;
> in the meantime ignore this, and sorry for the noise.

OK -- I must confess that I could not understand how this change
could cause the problem!  ;-)

						Thanx, Paul
--
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/