Re: [RFC 1/9] RISC-V: Define a helper function to probe number of hardware counters

[Atish Patra]

On Tue, Nov 1, 2022 at 5:30 AM Andrew Jones <ajones@xxxxxxxxxxxxxxxx> wrote:
>
> On Mon, Jul 18, 2022 at 10:01:57AM -0700, Atish Patra wrote:
> > KVM module needs to know how many hardware counters the platform supports.
> > Otherwise, it will not be able to show optimal value of virtual
>                                         ^ the
> > counters to the guest.
> >
> > Signed-off-by: Atish Patra <atishp@xxxxxxxxxxxx>
> > ---
> >  drivers/perf/riscv_pmu_sbi.c   | 23 +++++++++++++++++------
> >  include/linux/perf/riscv_pmu.h |  4 ++++
> >  2 files changed, 21 insertions(+), 6 deletions(-)
> >
> > diff --git a/drivers/perf/riscv_pmu_sbi.c b/drivers/perf/riscv_pmu_sbi.c
> > index 24124546844c..1723af68ffa1 100644
> > --- a/drivers/perf/riscv_pmu_sbi.c
> > +++ b/drivers/perf/riscv_pmu_sbi.c
> > @@ -27,6 +27,7 @@
> >   */
> >  static union sbi_pmu_ctr_info *pmu_ctr_list;
> >  static unsigned int riscv_pmu_irq;
> > +static struct riscv_pmu *rvpmu;
>
> Do we really need rvpmu? From a quick scan of the series it's only used
> for num_hw_counters, which has to be added to struct riscv_pmu, and
> num_counters. How about instead creating a static global for num_counters

Yes. I added rvpmu just for future usage if any.

> and then getting num_hw_counters by iterating pmu_ctr_list. If we want

iteration is fine as we are doing that for hpm_width anyways.

> riscv_pmu_sbi_get_num_hw_ctrs() to be faster, then we can cache the value
> in a static variable in the function.
>

We have cmask now which can be cached in a static variable. We need to
retrieve the
counter width and the hardware counters for kvm. I have combined PATCH
1 & PATCH 2
to return both the values in one function.

> Thanks,
> drew



-- 
Regards,
Atish