On Tue, Dec 07, 2021 at 07:14:29PM +0000, Robin Murphy wrote:
> On 2021-12-07 18:44, Mark Rutland wrote:
> > On Tue, Dec 07, 2021 at 06:20:41PM +0000, Robin Murphy wrote:
> > > Wire up the new DT compatibles so we can present appropriate
> > > PMU names to userspace for the latest and greatest CPUs.
> > >
> > > Signed-off-by: Robin Murphy <robin.murphy@xxxxxxx>
> > > ---
> > > arch/arm64/kernel/perf_event.c | 36 ++++++++++++++++++++++++++++++++++
> > > 1 file changed, 36 insertions(+)
> > >
> > > diff --git a/arch/arm64/kernel/perf_event.c b/arch/arm64/kernel/perf_event.c
> > > index 57720372da62..3fe4dcfc28d4 100644
> > > --- a/arch/arm64/kernel/perf_event.c
> > > +++ b/arch/arm64/kernel/perf_event.c
> > > @@ -1215,6 +1215,26 @@ static int armv8_a78_pmu_init(struct arm_pmu *cpu_pmu)
> > > return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cortex_a78", NULL);
> > > }
> > > +static int armv9_a510_pmu_init(struct arm_pmu *cpu_pmu)
> > > +{
> > > + return armv8_pmu_init_nogroups(cpu_pmu, "armv9_cortex_a510", NULL);
> > > +}
> > > +
> > > +static int armv9_a710_pmu_init(struct arm_pmu *cpu_pmu)
> > > +{
> > > + return armv8_pmu_init_nogroups(cpu_pmu, "armv9_cortex_a710", NULL);
> > > +}
> > > +
> > > +static int armv8_x1_pmu_init(struct arm_pmu *cpu_pmu)
> > > +{
> > > + return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cortex_x1", NULL);
> > > +}
> > > +
> > > +static int armv9_x2_pmu_init(struct arm_pmu *cpu_pmu)
> > > +{
> > > + return armv8_pmu_init_nogroups(cpu_pmu, "armv9_cortex_x2", NULL);
> > > +}
> >
> > I wonder if it'd be better to do something like:
> >
> > #define PMU_INIT_SIMPLE(name) \
> > static int name##_pmu_init(struct arm_pmu *cpu_pmu) \
> > {
> > return armv8_pmu_init_nogroups(cpu_pmu, #name, NULL); \
> > }
> >
> > PMU_INIT_SIMPLE(armv9_cortex_a510)
> > PMU_INIT_SIMPLE(armv9_cortex_a710)
> > PMU_INIT_SIMPLE(armv8_xortex_x1)
> > PMU_INIT_SIMPLE(armv9_xortex_x2)
> >
> > ... and fix up the armv8_pmu_of_device_ids[] table to use the longer init names
> > that results in?
>
> Indeed I did ponder doing almost exactly that, but at that point I'd rather
> try refactoring a bit deeper to convert most of the arm_pmu init business to
> pure data, so I figured I'd chuck in the simple tweak to mitigate these new
> additions with minimal churn, then have a go at the bigger change in its own
> right.
Sure; that makes sense to me, so for this as-is:
Acked-by: Mark Rutland <mark.rutland@xxxxxxx>
... and I'll leave it to Will to have the final say on whether we want the
"armv9_" prefix or whether we stick with "armv8_" for consistenct, when he
chooses to pick this.
One thing I've just realised is that for the ACPI case, we're stuck with
"armv8_pmuv3_%d" regardless, which I think is fine itself, but we might want to
call that out.
Thanks,
Mark.
