Re: [PATCH 1/3] perf: add context field to perf_event

[Avi Kivity <avi@redhat.com>]

On 07/12/2011 12:07 AM, Will Deacon wrote:
> On Mon, Jul 04, 2011 at 03:36:57PM +0100, Frederic Weisbecker wrote:
> >  On Mon, Jul 04, 2011 at 05:10:20PM +0300, Avi Kivity wrote:
> >  >  On 07/04/2011 04:58 PM, Frederic Weisbecker wrote:
> >  >  >Another thing I would like to do in the even longer term is to not use perf anymore
> >  >  >for ptrace breakpoints, because that involves a heavy dependency and few people are
> >  >  >happy with that. Instead we should just have a generic hook into the sched_switch()
> >  >  >and handle pure ptrace breakpoints there. The central breakpoint API would still be
> >  >  >there to reserve/schedule breakpoint resources between ptrace and perf.
> >  >  >
> >  >
> >  >  'struct preempt_notifier' may be the hook you're looking for.
> >
> >  Yeah looks like a perfect fit as it's per task.
>
> I had a quick look at this and I think the preempt_notifier stuff needs
> slightly extending so that we can register a notifier for a task other than
> current [e.g. the child of current on which we are installing breakpoints].
>
> If the task in question is running, it looks like this will introduce a race
> condition between notifier registration and rescheduling. For the purposes
> of ptrace this shouldn't be a problem as the child will be stopped, but
> others might also want to make use of the new functionality.
>
> Any ideas on how this could be achieved, or am I better off just restricting
> this to children that are being traced?

Maybe we need a generic "run this function in this task's context" 
mechanism instead.  Like an IPI, but targeting tasks instead of cpus.

-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.