Re: [PATCH 5/9] perf/x86: implement cross-HT corruption bug workaround

[Stephane Eranian]

On Thu, Jun 5, 2014 at 4:21 PM, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> On Thu, Jun 05, 2014 at 04:15:17PM +0200, Stephane Eranian wrote:
>> On Thu, Jun 5, 2014 at 4:04 PM, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>> > On Wed, Jun 04, 2014 at 11:34:14PM +0200, Stephane Eranian wrote:
>> >> +
>> >> +     /*
>> >> +      * Modify static constraint with current dynamic
>> >> +      * state of thread
>> >> +      *
>> >> +      * EXCLUSIVE: sibling counter measuring exclusive event
>> >> +      * SHARED   : sibling counter measuring non-exclusive event
>> >> +      * UNUSED   : sibling counter unused
>> >> +      */
>> >> +     for_each_set_bit(i, cx->idxmsk, X86_PMC_IDX_MAX) {
>> >> +             /*
>> >> +              * exclusive event in sibling counter
>> >> +              * our corresponding counter cannot be used
>> >> +              * regardless of our event
>> >> +              */
>> >> +             if (xl->state[i] == INTEL_EXCL_EXCLUSIVE)
>> >> +                     __clear_bit(i, cx->idxmsk);
>> >> +             /*
>> >> +              * if measuring an exclusive event, sibling
>> >> +              * measuring non-exclusive, then counter cannot
>> >> +              * be used
>> >> +              */
>> >> +             if (is_excl && xl->state[i] == INTEL_EXCL_SHARED)
>> >> +                     __clear_bit(i, cx->idxmsk);
>> >> +     }
>> >> +
>> >> +     /*
>> >> +      * recompute actual bit weight for scheduling algorithm
>> >> +      */
>> >> +     cx->weight = hweight64(cx->idxmsk64);
>> >
>> > So I think we talked about this a bit; what happens if CPU0 (taking your
>> > 4 core HSW-client) is first to program its counters and takes all 4 in
>> > exclusive mode?
>> >
>> > Then there's none left for CPU4.
>> >
>> > Did I miss where we avoid that problem, or is that an actual issue?
>>
>> Yes, this patch series does not address this problem yet. It will be
>> in a second series.
>> Don't have a good solution yet.
>
> We could limit each cpu to num_counters/2 exclusive slots. That'll still
> be painful with some constrained events I imagine, but in general that
> should 'work' I suppose.

Ok, tried this. It avoids the problems and just requires a 3 line patch.
So we could go with that for now. It would avoid total starvation.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/