Re: [PATCH 4/5] nds32: Fix perf multiple events map to same counter.

From: Nick Hu
Date: Mon Oct 22 2018 - 06:21:49 EST


Hi Mark,

On Thu, Oct 18, 2018 at 10:29:05PM +0800, Mark Rutland wrote:
> On Thu, Oct 18, 2018 at 04:43:16PM +0800, Nickhu wrote:
> > When there are multiple events map to the same counter, the counter
> > counts inaccurately. This is because each counter only counts one event
> > in the same time.
> > So when there are multiple events map to same counter, they have to take
> > turns in each context.
> >
> > There are two solution:
> > 1. Print the error message when multiple events map to the same counter.
> > But print the error message would let the program hang in loop. The ltp
> > (linux test program) would be failed when the program hang in loop.
> >
> > 2. Don't print the error message, the ltp would pass. But the user need to
> > have the knowledge that don't count the events which map to the same
> > counter, or the user will get the inaccurate results.
> >
> > We choose method 2 for the solution
>
> This is the correct solution. Perf exposes the active/enabled time in
> the perf event, so the user can determine that the event wasn't enabled
> all of the time.
>
> This should be folded into the commit adding perf support.
>
> Thanks,
> Mark.
>
> >
> > Signed-off-by: Nickhu <nickhu@xxxxxxxxxxxxx>
> > ---
> > arch/nds32/include/asm/pmu.h | 1 +
> > arch/nds32/kernel/perf_event_cpu.c | 30 ++++++++++++++++++++----------
> > 2 files changed, 21 insertions(+), 10 deletions(-)
> >
> > diff --git a/arch/nds32/include/asm/pmu.h b/arch/nds32/include/asm/pmu.h
> > index 3fbbe97c2d42..e75ec34af5f6 100644
> > --- a/arch/nds32/include/asm/pmu.h
> > +++ b/arch/nds32/include/asm/pmu.h
> > @@ -55,6 +55,7 @@ enum { PFMC0, PFMC1, PFMC2, MAX_COUNTERS };
> > */
> > #define NDS32_IDX_CYCLE_COUNTER 0
> > #define NDS32_IDX_COUNTER0 1
> > +#define NDS32_IDX_COUNTER1 2
> > #define NDS32_IDX_COUNTER_LAST(cpu_pmu) \
> > (NDS32_IDX_CYCLE_COUNTER + (cpu_pmu)->num_events - 1)
> >
> > diff --git a/arch/nds32/kernel/perf_event_cpu.c b/arch/nds32/kernel/perf_event_cpu.c
> > index 7bb4ebb87b5c..e9a0d8bb2bc1 100644
> > --- a/arch/nds32/kernel/perf_event_cpu.c
> > +++ b/arch/nds32/kernel/perf_event_cpu.c
> > @@ -566,16 +566,26 @@ static int nds32_pmu_get_event_idx(struct pmu_hw_events *cpuc,
> > /*
> > * Try to get the counter for correpsonding event
> > */
> > - if (!test_and_set_bit(idx, cpuc->used_mask))
> > - return idx;
> > -
> > - /*
> > - * The counter is in use.
> > - * The system will hang in the loop.
> > - */
> > - pr_err
> > - ("Multiple events map to one counter, the behavior is undefined.\n");
> > - return -EPERM;
> > + if (evtype == SPAV3_0_SEL_TOTAL_CYCLES) {
> > + if (!test_and_set_bit(idx, cpuc->used_mask))
> > + return idx;
> > + if (!test_and_set_bit(NDS32_IDX_COUNTER0, cpuc->used_mask))
> > + return NDS32_IDX_COUNTER0;
> > + if (!test_and_set_bit(NDS32_IDX_COUNTER1, cpuc->used_mask))
> > + return NDS32_IDX_COUNTER1;
> > + } else if (evtype == SPAV3_1_SEL_COMPLETED_INSTRUCTION) {
> > + if (!test_and_set_bit(idx, cpuc->used_mask))
> > + return idx;
> > + else if (!test_and_set_bit(NDS32_IDX_COUNTER1, cpuc->used_mask))
> > + return NDS32_IDX_COUNTER1;
> > + else if (!test_and_set_bit
> > + (NDS32_IDX_CYCLE_COUNTER, cpuc->used_mask))
> > + return NDS32_IDX_CYCLE_COUNTER;
> > + } else {
> > + if (!test_and_set_bit(idx, cpuc->used_mask))
> > + return idx;
> > + }
> > + return -EAGAIN;
> > }
> >
> > static void nds32_pmu_start(struct nds32_pmu *cpu_pmu)
> > --
> > 2.17.0
> >
Thanks for the comment.
I will folded it into the commit adding perf support.