Re: [tip:tracing/hw-branch-tracing] tracing/hw-branch-tracing: convert bts-tracer mutex to a spinlock

[Ingo Molnar <mingo@elte.hu>]

* Metzger, Markus T <markus.t.metzger@intel.com> wrote:

> >-----Original Message-----
> >From: Ingo Molnar [mailto:mingo@elte.hu]
> >Sent: Wednesday, February 25, 2009 12:11 PM
> >To: Metzger, Markus T
> >Cc: hpa@zytor.com; mingo@redhat.com; tglx@linutronix.de; linux-kernel@vger.kernel.org; linux-tip-
> >commits@vger.kernel.org
> 
> 
> >> And I assume you like the spinlock better than the
> >> get/put_online_cpus(), as well.
> >
> >yeah - and get/put_online_cpus is sleepable too, so it doesnt
> >really help unless i'm missing something ...
> 
> I suggested to use get/put_online_cpus() instead of the lock.
> 
> The hotplug code waits until the cpu_hotplug.refcount is zero
> and it holds the cpu_hotplug.lock during hotplug operations
> (see cpu_hotplug_begin/done()).
> 
> In turn, get_online_cpus() needs to grab the cpu_hotplug.lock
> to increment the cpu_hotplug.refcount.
> 
> Thus, we will use the cpu_hotplug.lock instead of our own lock.

... which, if you use it in the exact same spots as now still 
does a potential sleep with irqs disabled => bad.

We might be able to not take the hotplug lock in the affected 
codepath, but we should really not expand on the use of that 
lock and should make this code self-sufficient.

	Ingo