Re: [PATCH v2 01/11] perf,x86: Fix event/group validation

[Peter Zijlstra <peterz@infradead.org>]

On Tue, May 26, 2015 at 02:24:38AM -0700, Stephane Eranian wrote:
> On Fri, May 22, 2015 at 6:40 AM, Peter Zijlstra <peterz@infradead.org> wrote:
> > On Fri, May 22, 2015 at 03:29:06PM +0200, Peter Zijlstra wrote:
> >> @@ -788,9 +788,9 @@ int x86_schedule_events(struct cpu_hw_ev
> >>               x86_pmu.start_scheduling(cpuc);
> >>
> >>       for (i = 0, wmin = X86_PMC_IDX_MAX, wmax = 0; i < n; i++) {
> >> +             cpuc->event_constraint[i] = NULL;
> >
> > ^^^ that is new, which is esp. important in light of the
> > intel_get_event_constraints() hunk below, which would happily continue
> > life with a garbage constraint.
> >
> You've moved the constraint list from event to cpuc. Yet, it is still
> an array of pointers
> to constraints. So here you are saying, that in the case validate_group() is
> preempted and there is a context switch, there is still a risk of
> overwriting the
> constraint? I don't see how because validate_group() is using a fake_cpuc.
> So yes, the cpuc->event_constraint[] array is modified but it is not the same
> as the actual cpuc used by non-validate code. Or am I still missing something?
> 
> When using dynamic constraints, we already have constraint storage in cpuc
> (to avoid calling kmalloc() in ctxsw context). Thus, I am wondering if it would
> not be easier to always use cpuc for constraint storage (no more pointers).

No; the problem here is repeated use of the cpuc (the real one). Say one
scheduling run installs a constraint pointer for event i. Then event i
gets removed and another installed in the same spot.

Then the next scheduling run will pick up the old pointer in
intel_get_event_constraints() as a base for the new one.