Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
---
kernel/events/core.c | 51 +++++++++++++++++++++++----------------------------
1 file changed, 23 insertions(+), 28 deletions(-)
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -4090,7 +4090,7 @@ perf_adjust_freq_unthr_context(struct pe
if (!(ctx->nr_freq || unthrottle))
return;
- raw_spin_lock(&ctx->lock);
+ guard(raw_spinlock)(&ctx->lock);
list_for_each_entry_rcu(event, &ctx->event_list, event_entry) {
if (event->state != PERF_EVENT_STATE_ACTIVE)
@@ -4100,7 +4100,7 @@ perf_adjust_freq_unthr_context(struct pe
if (!event_filter_match(event))
continue;
- perf_pmu_disable(event->pmu);
+ guard(perf_pmu_disable)(event->pmu);
hwc = &event->hw;
@@ -4110,34 +4110,29 @@ perf_adjust_freq_unthr_context(struct pe
event->pmu->start(event, 0);
}
- if (!event->attr.freq || !event->attr.sample_freq)
- goto next;
+ if (event->attr.freq && event->attr.sample_freq) {
+ /*
+ * stop the event and update event->count
+ */
+ event->pmu->stop(event, PERF_EF_UPDATE);
+
+ now = local64_read(&event->count);
+ delta = now - hwc->freq_count_stamp;
+ hwc->freq_count_stamp = now;
+
+ /*
+ * restart the event
+ * reload only if value has changed
+ * we have stopped the event so tell that
+ * to perf_adjust_period() to avoid stopping it
+ * twice.
+ */
+ if (delta > 0)
+ perf_adjust_period(event, period, delta, false);
- /*
- * stop the event and update event->count
- */
- event->pmu->stop(event, PERF_EF_UPDATE);
-
- now = local64_read(&event->count);
- delta = now - hwc->freq_count_stamp;
- hwc->freq_count_stamp = now;
-
- /*
- * restart the event
- * reload only if value has changed
- * we have stopped the event so tell that
- * to perf_adjust_period() to avoid stopping it
- * twice.
- */
- if (delta > 0)
- perf_adjust_period(event, period, delta, false);
-
- event->pmu->start(event, delta > 0 ? PERF_EF_RELOAD : 0);
- next:
- perf_pmu_enable(event->pmu);
+ event->pmu->start(event, delta > 0 ? PERF_EF_RELOAD : 0);
+ }
}
-
- raw_spin_unlock(&ctx->lock);
}
/*
On Thu, Nov 2, 2023 at 8:32 AM Peter Zijlstra <[email protected]> wrote:
>
>
> Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
> ---
> kernel/events/core.c | 51 +++++++++++++++++++++++----------------------------
> 1 file changed, 23 insertions(+), 28 deletions(-)
>
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -4090,7 +4090,7 @@ perf_adjust_freq_unthr_context(struct pe
> if (!(ctx->nr_freq || unthrottle))
> return;
>
> - raw_spin_lock(&ctx->lock);
> + guard(raw_spinlock)(&ctx->lock);
>
> list_for_each_entry_rcu(event, &ctx->event_list, event_entry) {
> if (event->state != PERF_EVENT_STATE_ACTIVE)
> @@ -4100,7 +4100,7 @@ perf_adjust_freq_unthr_context(struct pe
> if (!event_filter_match(event))
> continue;
>
> - perf_pmu_disable(event->pmu);
> + guard(perf_pmu_disable)(event->pmu);
>
> hwc = &event->hw;
>
> @@ -4110,34 +4110,29 @@ perf_adjust_freq_unthr_context(struct pe
> event->pmu->start(event, 0);
> }
>
> - if (!event->attr.freq || !event->attr.sample_freq)
> - goto next;
> + if (event->attr.freq && event->attr.sample_freq) {
Any reason for this change? I think we can just change the
'goto next' to 'continue', no?
Also I think this code needs changes to optimize the access.
A similar reason for the cgroup switch, it accesses all the
pmu/events in the context even before checking the sampling
frequency. This is bad for uncore PMUs (and KVM too).
But this is a different issue..
Thanks,
Namhyung
> + /*
> + * stop the event and update event->count
> + */
> + event->pmu->stop(event, PERF_EF_UPDATE);
> +
> + now = local64_read(&event->count);
> + delta = now - hwc->freq_count_stamp;
> + hwc->freq_count_stamp = now;
> +
> + /*
> + * restart the event
> + * reload only if value has changed
> + * we have stopped the event so tell that
> + * to perf_adjust_period() to avoid stopping it
> + * twice.
> + */
> + if (delta > 0)
> + perf_adjust_period(event, period, delta, false);
>
> - /*
> - * stop the event and update event->count
> - */
> - event->pmu->stop(event, PERF_EF_UPDATE);
> -
> - now = local64_read(&event->count);
> - delta = now - hwc->freq_count_stamp;
> - hwc->freq_count_stamp = now;
> -
> - /*
> - * restart the event
> - * reload only if value has changed
> - * we have stopped the event so tell that
> - * to perf_adjust_period() to avoid stopping it
> - * twice.
> - */
> - if (delta > 0)
> - perf_adjust_period(event, period, delta, false);
> -
> - event->pmu->start(event, delta > 0 ? PERF_EF_RELOAD : 0);
> - next:
> - perf_pmu_enable(event->pmu);
> + event->pmu->start(event, delta > 0 ? PERF_EF_RELOAD : 0);
> + }
> }
> -
> - raw_spin_unlock(&ctx->lock);
> }
>
> /*
>
>
On Fri, Nov 03, 2023 at 06:14:32PM -0700, Namhyung Kim wrote:
> On Thu, Nov 2, 2023 at 8:32 AM Peter Zijlstra <[email protected]> wrote:
> >
> >
> > Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
> > ---
> > kernel/events/core.c | 51 +++++++++++++++++++++++----------------------------
> > 1 file changed, 23 insertions(+), 28 deletions(-)
> >
> > --- a/kernel/events/core.c
> > +++ b/kernel/events/core.c
> > @@ -4090,7 +4090,7 @@ perf_adjust_freq_unthr_context(struct pe
> > if (!(ctx->nr_freq || unthrottle))
> > return;
> >
> > - raw_spin_lock(&ctx->lock);
> > + guard(raw_spinlock)(&ctx->lock);
> >
> > list_for_each_entry_rcu(event, &ctx->event_list, event_entry) {
> > if (event->state != PERF_EVENT_STATE_ACTIVE)
> > @@ -4100,7 +4100,7 @@ perf_adjust_freq_unthr_context(struct pe
> > if (!event_filter_match(event))
> > continue;
> >
> > - perf_pmu_disable(event->pmu);
> > + guard(perf_pmu_disable)(event->pmu);
> >
> > hwc = &event->hw;
> >
> > @@ -4110,34 +4110,29 @@ perf_adjust_freq_unthr_context(struct pe
> > event->pmu->start(event, 0);
> > }
> >
> > - if (!event->attr.freq || !event->attr.sample_freq)
> > - goto next;
> > + if (event->attr.freq && event->attr.sample_freq) {
>
> Any reason for this change? I think we can just change the
> 'goto next' to 'continue', no?
Linus initially got confused about the life-time of for-loop scopes, but
yeah, this could be continue just fine.
> Also I think this code needs changes to optimize the access.
> A similar reason for the cgroup switch, it accesses all the
> pmu/events in the context even before checking the sampling
> frequency. This is bad for uncore PMUs (and KVM too).
>
> But this is a different issue..
Right, lets do that in another patch. Also, there seems to be a problem
with the cgroup thing :/