circular locking dependency on shutdown

circular locking dependency on shutdown

[Johannes Berg]

[-- Attachment #1: Type: text/plain, Size: 738 bytes --]

Hi,

Every time I shut down my machine with a 2.6.29-rc2 kernel, I get a
lockdep report that scrolls off. I tried capturing it on video, the
relevant frames are here:
http://johannes.sipsolutions.net/files/shutdown-lockdep-images.tar.bz2

Some info I transscribed:

halt is trying to acquire lock
	(&per_cpu(cpu_policy_rwsem, cpu)){---}

but task is already holding lock
	(&cpu_hotplug.lock){--..}

which already yadda yadda

lock chain:
#1	(&cpu_hotplug.lock){--..}
	...
	get_online_cpus
#0	(&per_cpu(cpu_policy_rwsem, cpu))
	...
	disable_nonboot_cpus

stacktrace:
	down_write
	lock_policy_rwsem_write
	cpufreq_cpu_callback
	notifier_call_chain
	disable_nonboot_cpus
	kernel_power_off
	...


johannes

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: circular locking dependency on shutdown

[Johannes Berg]

[-- Attachment #1: Type: text/plain, Size: 192 bytes --]

On Thu, 2009-01-29 at 19:51 +0100, Johannes Berg wrote:
> Hi,
> 
> Every time I shut down my machine with a 2.6.29-rc2 kernel,

Seems it's gone in -rc3, sorry for the noise!

johannes

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: circular locking dependency on shutdown

[Ingo Molnar]

* Johannes Berg <johannes@sipsolutions.net> wrote:

> On Thu, 2009-01-29 at 19:51 +0100, Johannes Berg wrote:
> > Hi,
> > 
> > Every time I shut down my machine with a 2.6.29-rc2 kernel,
> 
> Seems it's gone in -rc3, sorry for the noise!

yeah - there were a few fixes in this area.

	Ingo

Re: circular locking dependency on shutdown

[Rusty Russell]

On Friday 30 January 2009 05:20:49 Johannes Berg wrote:
> Hi,
> 
> Every time I shut down my machine with a 2.6.29-rc2 kernel, I get a
> lockdep report that scrolls off.
...
> halt is trying to acquire lock
> 	(&per_cpu(cpu_policy_rwsem, cpu)){---}
> 
> but task is already holding lock
> 	(&cpu_hotplug.lock){--..}

Hi Johannes,

   This looks much like the work_on_cpu()-getting-hotplug-lock problem which caused us to stop work_on_cpu doing that.  -rc3 has that fix; any chance of a re-test?

Thanks,
Rusty.