[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <f632d5c8-93db-402c-82ec-9f3b79906458@arm.com>
Date: Fri, 6 Sep 2024 10:28:29 +0530
From: Anshuman Khandual <anshuman.khandual@....com>
To: Ilkka Koskinen <ilkka@...amperecomputing.com>,
Will Deacon <will@...nel.org>, Mark Rutland <mark.rutland@....com>
Cc: linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] perf: arm_pmuv3: Use BR_RETIRED for HW branch event if
enabled
On 9/6/24 02:17, Ilkka Koskinen wrote:
> The PMU driver attempts to use PC_WRITE_RETIRED for the HW branch event,
> if enabled. However, PC_WRITE_RETIRED counts only taken branches,
> whereas BR_RETIRED counts also non-taken ones.
>
> Furthermore, perf uses HW branch event to calculate branch misses ratio,
> implying BR_RETIRED is the correct event to count.
But is the event BR_RETIRED always guaranteed to be available. Should not
armpmu->pmceid_bitmap be checked first ?
>
> Signed-off-by: Ilkka Koskinen <ilkka@...amperecomputing.com>
> ---
> drivers/perf/arm_pmuv3.c | 27 ++++-----------------------
> 1 file changed, 4 insertions(+), 23 deletions(-)
>
> diff --git a/drivers/perf/arm_pmuv3.c b/drivers/perf/arm_pmuv3.c
> index d246840797b6..a8ed08df1411 100644
> --- a/drivers/perf/arm_pmuv3.c
> +++ b/drivers/perf/arm_pmuv3.c
> @@ -46,6 +46,7 @@ static const unsigned armv8_pmuv3_perf_map[PERF_COUNT_HW_MAX] = {
> [PERF_COUNT_HW_INSTRUCTIONS] = ARMV8_PMUV3_PERFCTR_INST_RETIRED,
> [PERF_COUNT_HW_CACHE_REFERENCES] = ARMV8_PMUV3_PERFCTR_L1D_CACHE,
> [PERF_COUNT_HW_CACHE_MISSES] = ARMV8_PMUV3_PERFCTR_L1D_CACHE_REFILL,
> + [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = ARMV8_PMUV3_PERFCTR_BR_RETIRED,
> [PERF_COUNT_HW_BRANCH_MISSES] = ARMV8_PMUV3_PERFCTR_BR_MIS_PRED,
> [PERF_COUNT_HW_BUS_CYCLES] = ARMV8_PMUV3_PERFCTR_BUS_CYCLES,
> [PERF_COUNT_HW_STALLED_CYCLES_FRONTEND] = ARMV8_PMUV3_PERFCTR_STALL_FRONTEND,
> @@ -1083,28 +1084,6 @@ static void armv8pmu_reset(void *info)
> armv8pmu_pmcr_write(pmcr);
> }
>
> -static int __armv8_pmuv3_map_event_id(struct arm_pmu *armpmu,
> - struct perf_event *event)
> -{
> - if (event->attr.type == PERF_TYPE_HARDWARE &&
> - event->attr.config == PERF_COUNT_HW_BRANCH_INSTRUCTIONS) {
> -
> - if (test_bit(ARMV8_PMUV3_PERFCTR_PC_WRITE_RETIRED,
> - armpmu->pmceid_bitmap))
> - return ARMV8_PMUV3_PERFCTR_PC_WRITE_RETIRED;
> -
> - if (test_bit(ARMV8_PMUV3_PERFCTR_BR_RETIRED,
> - armpmu->pmceid_bitmap))
> - return ARMV8_PMUV3_PERFCTR_BR_RETIRED;
If BR_RETIRED event is absent on the platform, PC_WRITE_RETIRED still remains
a good alternative to fallback on. Hence wondering if the above order could
just be changed to use BR_RETIRED first when available.
> -
> - return HW_OP_UNSUPPORTED;
> - }
> -
> - return armpmu_map_event(event, &armv8_pmuv3_perf_map,
> - &armv8_pmuv3_perf_cache_map,
> - ARMV8_PMU_EVTYPE_EVENT);
> -}
> -
> static int __armv8_pmuv3_map_event(struct perf_event *event,
> const unsigned (*extra_event_map)
> [PERF_COUNT_HW_MAX],
> @@ -1116,7 +1095,9 @@ static int __armv8_pmuv3_map_event(struct perf_event *event,
> int hw_event_id;
> struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
>
> - hw_event_id = __armv8_pmuv3_map_event_id(armpmu, event);
> + hw_event_id = armpmu_map_event(event, &armv8_pmuv3_perf_map,
> + &armv8_pmuv3_perf_cache_map,
> + ARMV8_PMU_EVTYPE_EVENT);
>
> /*
> * CHAIN events only work when paired with an adjacent counter, and it
Powered by blists - more mailing lists