[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <dd55c313-632e-cf18-161a-50ac6632feb4@arm.com>
Date: Thu, 19 Oct 2023 11:38:57 +0100
From: James Clark <james.clark@....com>
To: Leo Yan <leo.yan@...aro.org>
Cc: John Garry <john.g.garry@...cle.com>,
Will Deacon <will@...nel.org>,
Peter Zijlstra <peterz@...radead.org>,
Ingo Molnar <mingo@...hat.com>,
Mark Rutland <mark.rutland@....com>,
Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
Jiri Olsa <jolsa@...nel.org>,
Namhyung Kim <namhyung@...nel.org>,
Ian Rogers <irogers@...gle.com>,
Adrian Hunter <adrian.hunter@...el.com>,
coresight@...ts.linaro.org, linux-arm-kernel@...ts.infradead.org,
linux-perf-users@...r.kernel.org, linux-kernel@...r.kernel.org,
Arnaldo Carvalho de Melo <acme@...nel.org>,
Suzuki K Poulose <suzuki.poulose@....com>,
Mike Leach <mike.leach@...aro.org>
Subject: Re: [PATCH v1 2/2] perf cs-etm: Enable itrace option 'T'
On 14/10/2023 08:45, Leo Yan wrote:
> Prior to Armv8.4, the feature FEAT_TRF is not supported by Arm CPUs.
> Consequently, the sysfs node 'ts_source' will not be set as 1 by the
> CoreSight ETM driver. On the other hand, the perf tool relies on the
> 'ts_source' node to determine whether the kernel timestamp is traced.
> Since the 'ts_source' is not set for Arm CPUs prior to Armv8.4,
> platforms in this case cannot utilize the traced timestamp as the kernel
> time.
>
> This patch enables the 'T' itrace option, which forcibly utilizes the
> traced timestamp as the kernel time. If users are aware that their
> working platform's Arm CoreSight shares the same counter with the kernel
> time, they can specify 'T' option to decode the traced timestamp as the
> kernel time.
>
> An usage example is:
>
> # perf record -e cs_etm// -- test_program
> # perf script --itrace=i10ibT
> # perf report --itrace=i10ibT
>
> Signed-off-by: Leo Yan <leo.yan@...aro.org>
> ---
> tools/perf/util/cs-etm.c | 21 ++++++++++++++++++---
> 1 file changed, 18 insertions(+), 3 deletions(-)
>
> diff --git a/tools/perf/util/cs-etm.c b/tools/perf/util/cs-etm.c
> index 9729d006550d..4a37fdeb1795 100644
> --- a/tools/perf/util/cs-etm.c
> +++ b/tools/perf/util/cs-etm.c
> @@ -3322,12 +3322,27 @@ int cs_etm__process_auxtrace_info_full(union perf_event *event,
> etm->metadata = metadata;
> etm->auxtrace_type = auxtrace_info->type;
>
> - /* Use virtual timestamps if all ETMs report ts_source = 1 */
> - etm->has_virtual_ts = cs_etm__has_virtual_ts(metadata, num_cpu);
> + if (etm->synth_opts.use_timestamp)
> + /*
> + * Prior to Armv8.4, Arm CPUs don't support FEAT_TRF feature,
> + * therefore the decoder cannot know if the timestamp trace is
> + * same with the kernel time.
> + *
> + * If a user has knowledge for the working platform and can
> + * specify itrace option 'T' to tell decoder to forcely use the
> + * traced timestamp as the kernel time.
> + */
> + etm->has_virtual_ts = true;
> + else
> + /* Use virtual timestamps if all ETMs report ts_source = 1 */
> + etm->has_virtual_ts = cs_etm__has_virtual_ts(metadata, num_cpu);
>
> if (!etm->has_virtual_ts)
> ui__warning("Virtual timestamps are not enabled, or not supported by the traced system.\n"
> - "The time field of the samples will not be set accurately.\n\n");
> + "The time field of the samples will not be set accurately.\n"
> + "For Arm CPUs prior to Armv8.4 or without support FEAT_TRF,\n"
> + "you can specify the itrace option 'T' for timestamp decoding\n"
> + "if the Coresight timestamp on the platform is same with the kernel time.\n\n");
>
> etm->auxtrace.process_event = cs_etm__process_event;
> etm->auxtrace.process_auxtrace_event = cs_etm__process_auxtrace_event;
Reviewed-by: James Clark <james.clark@....com>
Powered by blists - more mailing lists