Re: [PATCH V2 1/5] perf mem: Add mem_events into the supported perf_pmu

[Leo Yan]

On Thu, Dec 07, 2023 at 11:23:34AM -0800, kan.liang@xxxxxxxxxxxxxxx wrote:
> From: Kan Liang <kan.liang@xxxxxxxxxxxxxxx>
> 
> With the mem_events, perf doesn't need to read sysfs for each PMU to
> find the mem-events-supported PMU. The patch also makes it possible to
> clean up the related __weak functions later.
> 
> The patch is only to add the mem_events into the perf_pmu for all ARCHs.
> It will be used in the later cleanup patches.
> 
> Reviewed-by: Ian Rogers <irogers@xxxxxxxxxx>
> Tested-by: Ravi Bangoria <ravi.bangoria@xxxxxxx>
> Signed-off-by: Kan Liang <kan.liang@xxxxxxxxxxxxxxx>
> ---
>  tools/perf/arch/arm64/util/mem-events.c | 4 ++--
>  tools/perf/arch/arm64/util/mem-events.h | 7 +++++++
>  tools/perf/arch/arm64/util/pmu.c        | 6 ++++++
>  tools/perf/arch/s390/util/pmu.c         | 3 +++
>  tools/perf/arch/x86/util/mem-events.c   | 4 ++--
>  tools/perf/arch/x86/util/mem-events.h   | 9 +++++++++
>  tools/perf/arch/x86/util/pmu.c          | 7 +++++++
>  tools/perf/util/mem-events.c            | 2 +-
>  tools/perf/util/mem-events.h            | 1 +
>  tools/perf/util/pmu.c                   | 4 +++-
>  tools/perf/util/pmu.h                   | 7 +++++++
>  11 files changed, 48 insertions(+), 6 deletions(-)
>  create mode 100644 tools/perf/arch/arm64/util/mem-events.h
>  create mode 100644 tools/perf/arch/x86/util/mem-events.h
> 
> diff --git a/tools/perf/arch/arm64/util/mem-events.c b/tools/perf/arch/arm64/util/mem-events.c
> index 3bcc5c7035c2..aaa4804922b4 100644
> --- a/tools/perf/arch/arm64/util/mem-events.c
> +++ b/tools/perf/arch/arm64/util/mem-events.c
> @@ -4,7 +4,7 @@
>  
>  #define E(t, n, s) { .tag = t, .name = n, .sysfs_name = s }
>  
> -static struct perf_mem_event perf_mem_events[PERF_MEM_EVENTS__MAX] = {
> +struct perf_mem_event perf_mem_events_arm[PERF_MEM_EVENTS__MAX] = {
>  	E("spe-load",	"arm_spe_0/ts_enable=1,pa_enable=1,load_filter=1,store_filter=0,min_latency=%u/",	"arm_spe_0"),
>  	E("spe-store",	"arm_spe_0/ts_enable=1,pa_enable=1,load_filter=0,store_filter=1/",			"arm_spe_0"),
>  	E("spe-ldst",	"arm_spe_0/ts_enable=1,pa_enable=1,load_filter=1,store_filter=1,min_latency=%u/",	"arm_spe_0"),
> @@ -17,7 +17,7 @@ struct perf_mem_event *perf_mem_events__ptr(int i)
>  	if (i >= PERF_MEM_EVENTS__MAX)
>  		return NULL;
>  
> -	return &perf_mem_events[i];
> +	return &perf_mem_events_arm[i];

I recognized that it's hard code to "arm_spe_0", which might break if
system registers different Arm SPE groups.  But this is not the issue
introduced by this patch, we might need to consider to fix it later.

>  }
>  
>  const char *perf_mem_events__name(int i, const char *pmu_name __maybe_unused)
> diff --git a/tools/perf/arch/arm64/util/mem-events.h b/tools/perf/arch/arm64/util/mem-events.h
> new file mode 100644
> index 000000000000..5fc50be4be38
> --- /dev/null
> +++ b/tools/perf/arch/arm64/util/mem-events.h
> @@ -0,0 +1,7 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +#ifndef _ARM64_MEM_EVENTS_H
> +#define _ARM64_MEM_EVENTS_H
> +
> +extern struct perf_mem_event perf_mem_events_arm[PERF_MEM_EVENTS__MAX];
> +
> +#endif /* _ARM64_MEM_EVENTS_H */
> diff --git a/tools/perf/arch/arm64/util/pmu.c b/tools/perf/arch/arm64/util/pmu.c
> index 2a4eab2d160e..06ec9b838807 100644
> --- a/tools/perf/arch/arm64/util/pmu.c
> +++ b/tools/perf/arch/arm64/util/pmu.c
> @@ -8,6 +8,12 @@
>  #include <api/fs/fs.h>
>  #include <math.h>
>  
> +void perf_pmu__arch_init(struct perf_pmu *pmu)
> +{
> +	if (!strcmp(pmu->name, "arm_spe_0"))
> +		pmu->mem_events = perf_mem_events_arm;

This is not right and it should cause building failure on aarch64.

aarch64 reuses aarch32's file arch/arm/util/pmu.c, and this file has
already defined perf_pmu__arch_init(), you should add above change in
the file arch/arm/util/pmu.c.

Now I cannot access a machine for testing Arm SPE, but I will play
a bit for this patch set to ensure it can pass compilation.  After
that, I will seek Arm maintainers/reviewers help for the test.

Thanks,
Leo