Re: [PATCH v6 02/10] arm64: perf: Enable PMU counter direct access for perf event

[Rob Herring]

On Wed, Mar 31, 2021 at 10:38 AM Will Deacon <will@xxxxxxxxxx> wrote:
>
> On Tue, Mar 30, 2021 at 04:08:11PM -0500, Rob Herring wrote:
> > On Tue, Mar 30, 2021 at 12:09 PM Rob Herring <robh@xxxxxxxxxx> wrote:
> > > On Tue, Mar 30, 2021 at 10:31 AM Will Deacon <will@xxxxxxxxxx> wrote:
> > > > The logic here feels like it
> > > > could with a bit of untangling.
> > >
> > > Yes, I don't love it, but couldn't come up with anything better. It is
> > > complicated by the fact that flags have to be set before we assign the
> > > counter and can't set/change them when we assign the counter. It would
> > > take a lot of refactoring with armpmu code to fix that.
> >
> > How's this instead?:
> >
> > if (armv8pmu_event_want_user_access(event) || !armv8pmu_event_is_64bit(event))
> >         event->hw.flags |= ARMPMU_EL0_RD_CNTR;
> >
> > /*
> > * At this point, the counter is not assigned. If a 64-bit counter is
> > * requested, we must make sure the h/w has 64-bit counters if we set
> > * the event size to 64-bit because chaining is not supported with
> > * userspace access. This may still fail later on if the CPU cycle
> > * counter is in use.
> > */
> > if (armv8pmu_event_is_64bit(event) &&
> >     (!armv8pmu_event_want_user_access(event) ||
> >      armv8pmu_has_long_event(cpu_pmu) || (hw_event_id ==
> > ARMV8_PMUV3_PERFCTR_CPU_CYCLES)))
> >         event->hw.flags |= ARMPMU_EVT_64BIT;
>
> I thought there were some cases where we could assign cycles event to an
> event counter; does that not happen anymore?

Yes, but if we hit that scenario when the user has asked for 64-bit
user access, then we return an error later when assigning the counter.
I think we can assume if users have gone to the trouble of requesting
64-bit counters, then they can deal with ensuring they don't have
multiple users.

Otherwise, the only way I see to simplify this is we only support
64-bit counters in userspace when we have v8.5 PMU.

Rob