Re: rcu_sched_state detected stalls on Alpha with generic config

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Wed, Dec 07, 2011 at 07:51:23PM +1300, Michael Cree wrote:
> I am seeing "rcu_sched_state detected stall on CPU" messages on Alpha
> architecture with a generic SMP config.  Interactive tasks are seen to
> lock up, with "INFO: task X blocked for more than 120 seconds" in the
> kernel logs, and eventual kernel oops and panic, on latest 3.2-rc4 and
> traceable back to 3.0.  Bisection between 2.6.39 and 3.0 leads to commit:
> 
> 09223371deac67d08ca0b70bd18787920284c967
> rcu: Use softirq to address performance regression
> 
> as the first bad commit.
> 
> Tested on an Alpha ES45 (Titan) with three 1.25 GHz CPUs and 4 GByte
> memory.  Testing procedure is to build git software and run its test
> suite with -j4 in the make command argument.
> 
> The CPU stall messages and eventually system lockup is only seen with a
> generic Alpha config, never with a Titan machine specific config.
> 
> An example of kernel logs is (this one probably produced when I tried to
> shutdown the system when it is falling over):
> 
> [45360.930876] INFO: rcu_sched_state detected stall on CPU 1 (t=798848
> jiffies)
> [45360.931853] INFO: rcu_sched_state detected stalls on CPUs/tasks: { 1}
> (detected by 0, t=798850 jiffies)
> [45489.080225] INFO: task umount:17371 blocked for more than 120 seconds.
> [45489.158350] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs"
> disables this message.
> [45489.252100] umount          D fffffc00013461ac     0 17371  17368
> 0x00000000
> [45489.336084] fffffc00fdd53db8 fffffc00fdd97bb8 fffffc000108ca1c
> fffffc00dcc9e800
> [45489.422998]        fffffc00dcc9e810 fffffc00013b3a5d fffffc000106289c
> fffffc00ff0dfda8
> [45489.519678]        0000000000000000 fffffc000108c81c fffffc0001cd73f0
> 0000000000000001
> [45489.615381]        fffffc00010627f0 0000000000000000 fffffc00dcc9e920
> fffffc00ff0bf780
> [45489.712060]        fffffc00010111b8 fffffc00ff0dfda8 fffffc00ff0dfde8
> fffffc0001cdaa58
> [45489.808740]        0000000000000000 0000000000000000 fffffc0000000000
> fffffc0000000000
> [45489.907373] Trace:
> [45489.930810] [<fffffc000108ca1c>] watchdog+0x200/0x27c
> [45489.991357] [<fffffc000106289c>] kthread+0xac/0xc4
> [45490.048974] [<fffffc000108c81c>] watchdog+0x0/0x27c
> [45490.107568] [<fffffc00010627f0>] kthread+0x0/0xc4
> [45490.164209] [<fffffc00010111b8>] kernel_thread+0x28/0x90
> [45490.227685]
> 
> Let me know if any other information is needed to narrow down the problem.

Are there any specific differences between the Titan config and the
generic Alpha config that might help me work out what is going on?
Odd -- it looks like the watchdog is complaining at the time of the hang.
Is it possible to use sysrq to get the state of the other tasks?

							Thanx, Paul