RE: [PATCH v2 3/8] ARCv2: perf: implement "event_set_period" for future use with interrupts

[Alexey Brodkin]

Hi Peter,

> -----Original Message-----
> From: Peter Zijlstra [mailto:peterz@xxxxxxxxxxxxx]
> Sent: 18 августа 2015 г. 20:55
> To: Alexey Brodkin
> Cc: linux-arch@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; Vineet.Gupta1@xxxxxxxxxxxx; arc-linux-dev@xxxxxxxxxxxx;
> arnd@xxxxxxxx; Arnaldo Carvalho de Melo
> Subject: Re: [PATCH v2 3/8] ARCv2: perf: implement "event_set_period" for future use with interrupts
> 
> On Wed, Aug 05, 2015 at 06:13:29PM +0300, Alexey Brodkin wrote:
> > +static int arc_pmu_event_set_period(struct perf_event *event)
> > +{
> > +	struct hw_perf_event *hwc = &event->hw;
> > +	s64 left = local64_read(&hwc->period_left);
> > +	s64 period = hwc->sample_period;
> > +	int idx = hwc->idx;
> > +	int overflow = 0;
> > +	u64 value;
> > +
> > +	if (unlikely(left <= -period)) {
> > +		/* left underflowed by more than period. */
> > +		left = period;
> > +		local64_set(&hwc->period_left, left);
> > +		hwc->last_period = period;
> > +		overflow = 1;
> > +	} else	if (unlikely(left <= 0)) {
> > +		/* left underflowed by less than period. */
> > +		left += period;
> > +		local64_set(&hwc->period_left, left);
> > +		hwc->last_period = period;
> > +		overflow = 1;
> > +	}
> > +
> > +	if (left > arc_pmu->max_period) {
> > +		left = arc_pmu->max_period;
> > +		local64_set(&hwc->period_left, left);
> 
> Given that you set counter_size to 32+bct_bcr.s << 4, I'm assuming these
> counters are not 64bit wide (or at least the hardware has the option of
> not being full width).

Indeed our counters could be 32/48(default)/64 bits wide.
 
> That means this local64_set() is wrong.

You mean the one used for setting "hwc->period_left"?

> The purpose here is to emulate a longer period with a short counter. So
> even though we have to take the interrupt to observe the counter width
> overflow and reprogram, we must not decrease the @left value.
> 
> Doing so will trigger one of the above two cases and result in @overflow
> == 1, even though we've not actually had hwc->sample_period counts.

My understanding was that here we're just checking if for some reason in
arc_perf_event_update() we decremented "hwc->period_left" too much
and it became either just <0 or even <(0 - period). IMHO that may happen
if not in sampling even case (where we expect interrupt to happen close
to a period being crossed) but in case of non-sampling event IMHO that
is pretty possible if frequency of checking counter value is way too low.

-Alexey
--
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/