Re: CPUFREQ-CPUHOTPLUG: Possible circular locking dependency

[Gautham R Shenoy <ego@in.ibm.com>]

On Thu, Nov 30, 2006 at 09:29:34AM +0100, Ingo Molnar wrote:
> 
> * Gautham R Shenoy <ego@in.ibm.com> wrote:
> 
> > Ok, I see that we are already doing it :(. So we can end up in a
> > deadlock.
> >
> > Here's the culprit callpath:
> 
> in general lockdep is 100% correct when it comes to "individual locks".
> The overwhelming majority of lockdep false-positives is not due to
> lockdep not getting the dependencies right, but due to the "lock class"
> not being correctly identified. That's not an issue here i think.

You're right. That's not the issue.

> 
> what lockdep does is it observes actual locking dependencies as they
> happen individually in various contexts, and then 'completes' the
> dependency graph by combining all the possible scenarios how contexts
> might preempt each other. So if lockdep sees independent dependencies
> and concludes that they are circular, there's nothing that saves us from
> the deadlock.
> 

Ah! I get it now. I had taken neither preemption nor the SMP scenario
into account before concluding that the warning might be a false
positive.

All I need to do is to run my test cases on a preemptible kernel 
or in parallel on a smp box. It'll definitely deadlock there!

> The only way for those dependencies to /never/ trigger simultaneously on
> different CPUs would be via the use of a further 'outer' exclusion
> mechanism (i.e. a lock) - but all explicit kernel-API exclusion
> mechanisms are tracked by lockdep => Q.E.D. (Open-coded exclusion might
> escape the attention of lockdep, but those are extremely rare and are
> also easily found.)

Thanks for making it clear :-)

> 
> 	Ingo

regards
gautham.
-- 
Gautham R Shenoy
Linux Technology Center
IBM India.
"Freedom comes with a price tag of responsibility, which is still a bargain,
because Freedom is priceless!"