[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <eec7cdd2-e74b-4845-a898-320c911832fa@linaro.org>
Date: Fri, 4 Oct 2024 16:40:18 +0100
From: James Clark <james.clark@...aro.org>
To: Namhyung Kim <namhyung@...nel.org>
Cc: Jiri Olsa <jolsa@...nel.org>, Adrian Hunter <adrian.hunter@...el.com>,
Peter Zijlstra <peterz@...radead.org>, Ingo Molnar <mingo@...nel.org>,
LKML <linux-kernel@...r.kernel.org>, linux-perf-users@...r.kernel.org,
Ravi Bangoria <ravi.bangoria@....com>, Mark Rutland <mark.rutland@....com>,
James Clark <james.clark@....com>, Kajol Jain <kjain@...ux.ibm.com>,
Thomas Richter <tmricht@...ux.ibm.com>, Atish Patra <atishp@...shpatra.org>,
Palmer Dabbelt <palmer@...osinc.com>, Mingwei Zhang <mizhang@...gle.com>,
Arnaldo Carvalho de Melo <acme@...nel.org>, Ian Rogers <irogers@...gle.com>,
Kan Liang <kan.liang@...ux.intel.com>
Subject: Re: [PATCHSET 0/8] perf tools: Do not set attr.exclude_guest by
default (v4)
On 02/10/2024 19:29, Namhyung Kim wrote:
> On Wed, Oct 02, 2024 at 10:49:36AM +0100, James Clark wrote:
>>
>>
>> On 01/10/2024 1:20 am, Namhyung Kim wrote:
>>> Hello,
>>>
>>> I found perf tools set exclude_guest bit inconsistently. It used to
>>> set the bit but now the default event for perf record doesn't. So I'm
>>> wondering why we want the bit in the first place.
>>>
>>> Actually it's not good for PMUs don't support any exclusion like AMD
>>> IBS because it disables new features after the exclude_guest due to
>>> the missing feature detection logic.
>>>
>>> v4 changes)
>>>
>>> * handle EOPNOTSUPP error in compatible way (Kan)
>>> * drop --exclude-guest option in perf stat
>>> * not to separate exclude_hv fallback
>>> * rename to exclude_GH_default (Kan)
>>> * drop the RFC from the subject
>>>
>>> v3) https://lore.kernel.org/lkml/20240905202426.2690105-1-namhyung@kernel.org/
>>>
>>> * move exclude_guest fallback to the front
>>> * fix precise_max handling on AMD
>>> * simplify the default event for perf record
>>>
>>> v2) https://lore.kernel.org/lkml/20240904064131.2377873-1-namhyung@kernel.org/
>>>
>>> * update the missing feature detection logic
>>> * separate exclude_hv fallback
>>> * add new fallback for exclude_guest
>>>
>>> v1) https://lore.kernel.org/lkml/20240902014621.2002343-1-namhyung@kernel.org/
>>>
>>> AFAIK it doesn't matter for the most cases but perf kvm. If users
>>> need to set the bit, they can still use :H modifier. For vPMU pass-
>>> through or Apple M1, it'd add the exclude_guest during the fallback
>>> logic.
>>>
>>> Also the kernel feature detection logic should be separated from the
>>> exclude bit tests since it depends on the PMU implementation rather
>>> than the core kernel features. So I changed it to use a software
>>> event for the detection and factor out some hw-specific checks.
>>>
>>> The code is available at 'perf/exclude-v4' branch in
>>> git://git.kernel.org/pub/scm/linux/kernel/git/namhyung/linux-perf.git
>>>
>>> Thanks,
>>> Namhyung
>>>
>>>
>>
>> Looks like you need to allow for :H in the perf stat test on M1 now:
>>
>> diff --git a/tools/perf/tests/shell/stat.sh b/tools/perf/tests/shell/stat.sh
>> index 5a2ca2bcf94d..77cb95859649 100755
>> --- a/tools/perf/tests/shell/stat.sh
>> +++ b/tools/perf/tests/shell/stat.sh
>> @@ -161,7 +161,7 @@ test_hybrid() {
>> fi
>>
>> # Run default Perf stat
>> - cycles_events=$(perf stat -- true 2>&1 | grep -E "/cycles/| cycles " |
>> wc -l)
>> + cycles_events=$(perf stat -- true 2>&1 | grep -E "/cycles/| cycles | cycles:H " | wc -l)
>
> Ok, what about "u"? Probably it can be
>
> cycles_events=$(perf stat -- true 2>&1 | grep -E "/cycles/(uH)*| cycles(:[uH])* " | wc -l)
>
Yep that probably works too and is a bit more robust
>>
>> if [ "$pmus" -ne "$cycles_events" ]
>> then
>>
>> Other than that:
>>
>> Reviewed-by: James Clark <james.clark@...aro.org>
>
> Thanks for the review!
> Namhyung
>
>>
>>> Namhyung Kim (8):
>>> perf tools: Add fallback for exclude_guest
>>> perf tools: Don't set attr.exclude_guest by default
>>> perf tools: Simplify evsel__add_modifier()
>>> perf tools: Do not set exclude_guest for precise_ip
>>> perf tools: Detect missing kernel features properly
>>> perf tools: Move x86__is_amd_cpu() to util/env.c
>>> perf tools: Check fallback error and order
>>> perf record: Just use "cycles:P" as the default event
>>>
>>> tools/perf/arch/x86/util/Build | 1 -
>>> tools/perf/arch/x86/util/env.c | 19 -
>>> tools/perf/arch/x86/util/env.h | 7 -
>>> tools/perf/arch/x86/util/pmu.c | 2 +-
>>> tools/perf/builtin-kvm.c | 1 +
>>> tools/perf/builtin-record.c | 4 +-
>>> tools/perf/builtin-stat.c | 18 +-
>>> tools/perf/dlfilters/dlfilter-test-api-v0.c | 2 +-
>>> tools/perf/dlfilters/dlfilter-test-api-v2.c | 2 +-
>>> tools/perf/tests/attr/test-record-dummy-C0 | 2 +-
>>> tools/perf/tests/parse-events.c | 30 +-
>>> tools/perf/util/env.c | 24 ++
>>> tools/perf/util/env.h | 4 +
>>> tools/perf/util/evsel.c | 394 ++++++++++++++------
>>> tools/perf/util/evsel.h | 1 -
>>> tools/perf/util/parse-events.c | 6 +-
>>> tools/perf/util/util.c | 10 +-
>>> tools/perf/util/util.h | 3 +
>>> 18 files changed, 364 insertions(+), 166 deletions(-)
>>> delete mode 100644 tools/perf/arch/x86/util/env.c
>>> delete mode 100644 tools/perf/arch/x86/util/env.h
>>>
>>
Powered by blists - more mailing lists