Re: remove cpu hotplug bustification in cpufreq.

[Arjan van de Ven]

On Tue, 2006-07-25 at 20:54 +0200, Ingo Molnar wrote:
> * Linus Torvalds <torvalds@xxxxxxxx> wrote:
> 
> > The current -git tree will complain about some of the more obvious 
> > problems. If you see a "Lukewarm IQ" message, it's a sign of somebody 
> > re-taking a cpu lock that is already held.
> 
> testing on my latest-rawhide laptop (kernel-2.6.17-1.2445.fc6 and later 
> rpms have this change) seems to have pushed the problem over to another 
> lock:
> 
>   S06cpuspeed/1580 is trying to acquire lock:
>    (&policy->lock){--..}, at: [<c06075f9>] mutex_lock+0x21/0x24
> 
>   but task is already holding lock:
>    (cpu_bitmask_lock){--..}, at: [<c06075f9>] mutex_lock+0x21/0x24
> 
>   which lock already depends on the new lock.


so cpufreq_set_policy() takes policy->lock, and then calls into the
userspace governer code
(__cpufreq_set_policy->cpufreq_governor->cpufreq_governor_userspace)
which calls __cpufreq_driver_target... which does lock_cpu_hotplug().


now on the other side:
store_scaling_governor() has the following code:

        lock_cpu_hotplug();

        /* Do not use cpufreq_set_policy here or the user_policy.max
           will be wrongly overridden */
        mutex_lock(&policy->lock);

so that's the entirely opposite lock order, and a classic AB-BA
deadlock.

Greetings,
     Arjan -- who's just cleaned Linus' wall to prepare it for more head
banging


-
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/