Re: workqueue deadlock

[Andrew Morton <akpm@osdl.org>]

On Sun, 10 Dec 2006 13:19:15 +0100
Ingo Molnar <mingo@elte.hu> wrote:

> 
> * Andrew Morton <akpm@osdl.org> wrote:
> 
> > This is actually not cpu-hotplug safe ;)  
> > 
> > > > > 	{
> > > > > 		int cpu = raw_smp_processor_id();
> > > > > 		/*
> > > > > 		 * Interrupts/softirqs are hotplug-safe:
> > > > > 		 */
> > > > > 		if (in_interrupt())
> > > > > 			return;
> > > > > 		if (current->hotplug_depth++)
> > > > > 			return;
> > 
> > <preempt, cpu hot-unplug, resume on different CPU>
> > 
> > > > > 		current->hotplug_lock = &per_cpu(hotplug_lock, cpu);
> > 
> > <use-after-free>
> > 
> > > > > 		mutex_lock(current->hotplug_lock);
> > 
> > And it sleeps, so we can't use preempt_disable().
> 
> i explained it in the other mail - this is the 'read' side. The 'write' 
> side (code actually wanting to /do/ a CPU hotplug state transition) has 
> to take /all/ these locks before it can take a CPU down.

Doesn't matter - the race is still there.

Well, not really, because we don't free the percpu data of offlined CPUs,
but we'd like to.

And it's easily fixable by using a statically-allocated array.  That would
make life easier for code which wants to take this lock early in boot too.

> so this is still a global CPU hotplug lock, but made scalable.

Scalability is not the problem.  At present, at least.