Re: [PATCH v2 06/18] perf s390 s390_cpumcfdg_dump: Don't scan all PMUs

[Ian Rogers]

On Fri, Aug 25, 2023 at 1:56 PM Arnaldo Carvalho de Melo
<acme@xxxxxxxxxx> wrote:
>
> Em Fri, Aug 25, 2023 at 04:39:22PM +0200, Thomas Richter escreveu:
> > On 8/25/23 15:14, Ian Rogers wrote:
> > > On Fri, Aug 25, 2023 at 1:20 AM Thomas Richter <tmricht@xxxxxxxxxxxxx> wrote:
>
> > >> On 8/24/23 15:59, Arnaldo Carvalho de Melo wrote:
> > >>> Em Wed, Aug 23, 2023 at 09:13:18PM -0700, Ian Rogers escreveu:
> > >>>> Rather than scanning all PMUs for a counter name, scan the PMU
> > >>>> associated with the evsel of the sample. This is done to remove a
> > >>>> dependence on pmu-events.h.
>
> > >>> I'm applying this one, and CCing the S/390 developers so that they can
> > >>> try this and maybe provide an Acked-by/Tested-by,
>
> > >> I have downloaded this patch set of 18 patches (using b4), but they do not
> > >> apply on my git tree.
>
> > >> Which git branch do I have to use to test this. Thanks a lot.
>
> > > the changes are in the perf-tools-next tree:
> > > https://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/
>
> > Unfurtunately this patch set fails again on s390.
> > Here is the test output from the current 6.5.0rc7 kernel:
> >
> > # ./perf test 6 10 'perf all metricgroups test' 'perf all metrics test'
> >   6: Parse event definition strings                                  :
> >   6.1: Test event parsing                                            : Ok
> >   6.2: Parsing of all PMU events from sysfs                          : Ok
> >   6.3: Parsing of given PMU events from sysfs                        : Ok
> >   6.4: Parsing of aliased events from sysfs                          : Skip (no aliases in sysfs)
> >   6.5: Parsing of aliased events                                     : Ok
> >   6.6: Parsing of terms (event modifiers)                            : Ok
> >  10: PMU events                                                      :
> >  10.1: PMU event table sanity                                        : Ok
> >  10.2: PMU event map aliases                                         : Ok
> >  10.3: Parsing of PMU event table metrics                            : Ok
> >  10.4: Parsing of PMU event table metrics with fake PMUs             : Ok
> >  10.5: Parsing of metric thresholds with fake PMUs                   : Ok
> >  95: perf all metricgroups test                                      : Ok
> >  96: perf all metrics test                                           : Ok
> > #
> >
> > This looks good.
>
> Reproduced:
>
> # grep -E vendor_id\|^processor -m2 /proc/cpuinfo
> vendor_id       : IBM/S390
> processor 0: version = 00,  identification = 1A33E8,  machine = 2964
> #
> # perf test 6 10 'perf all metricgroups test' 'perf all metrics test'
>   6: Parse event definition strings                                  :
>   6.1: Test event parsing                                            : Ok
>   6.2: Parsing of all PMU events from sysfs                          : Ok
>   6.3: Parsing of given PMU events from sysfs                        : Ok
>   6.4: Parsing of aliased events from sysfs                          : Skip (no aliases in sysfs)
>   6.5: Parsing of aliased events                                     : Ok
>   6.6: Parsing of terms (event modifiers)                            : Ok
>  10: PMU events                                                      :
>  10.1: PMU event table sanity                                        : Ok
>  10.2: PMU event map aliases                                         : Ok
>  10.3: Parsing of PMU event table metrics                            : Ok
>  10.4: Parsing of PMU event table metrics with fake PMUs             : Ok
>  10.5: Parsing of metric thresholds with fake PMUs                   : Ok
>  95: perf all metricgroups test                                      : Ok
>  96: perf all metrics test                                           : Ok
> # perf -v
> perf version 6.5.rc7.g6f0edbb833ec
> #
>
> > However when I use the check-out from perf-tools-next, I get this output:
> > # ./perf test 6 10 'perf all metricgroups test' 'perf all metrics test'
> >   6: Parse event definition strings                                  :
> >   6.1: Test event parsing                                            : Ok
> >   6.2: Parsing of all PMU events from sysfs                          : FAILED!
> >   6.3: Parsing of given PMU events from sysfs                        : Ok
> >   6.4: Parsing of aliased events from sysfs                          : Skip (no aliases in sysfs)
> >   6.5: Parsing of aliased events                                     : FAILED!
> >   6.6: Parsing of terms (event modifiers)                            : Ok
> >  10: PMU events                                                      :
> >  10.1: PMU event table sanity                                        : Ok
> >  10.2: PMU event map aliases                                         : FAILED!
> >  10.3: Parsing of PMU event table metrics                            : Ok
> >  10.4: Parsing of PMU event table metrics with fake PMUs             : FAILED!
> >  10.5: Parsing of metric thresholds with fake PMUs                   : Ok
> >  93: perf all metricgroups test                                      : FAILED!
> >  94: perf all metrics test                                           : FAILED!
> > #
> >
> > So some tests are failing again.
> >
> > I am out for the next two weeks, Sumanth Korikkar (on to list) might be able to help.
> > Thanks a lot.
>
> [root@kernelqe3 linux]# git checkout perf-tools-next
> git Switched to branch 'perf-tools-next'
> Your branch is up to date with 'perf-tools-next/perf-tools-next'.
> [root@kernelqe3 linux]# git log --oneline -5
> eeb6b12992c4 (HEAD -> perf-tools-next, perf-tools-next/perf-tools-next) perf jevents: Don't append Unit to desc
> f208b2c6f984 (perf-tools-next/tmp.perf-tools-next) perf scripts python gecko: Launch the profiler UI on the default browser with the appropriate URL
> 43803cb16f99 perf scripts python: Add support for input args in gecko script
> f85d120c46e7 perf jevents: Sort strings in the big C string to reduce faults
> 8d4b6d37ea78 perf pmu: Lazily load sysfs aliases
> [root@kernelqe3 linux]# make BUILD_BPF_SKEL=1 -C tools/perf O=/tmp/build/perf install-bin
>
> [root@kernelqe3 linux]# perf -v
> perf version 6.5.rc5.geeb6b12992c4
> [root@kernelqe3 linux]# git log --oneline -1
> eeb6b12992c4 (HEAD -> perf-tools-next, perf-tools-next/perf-tools-next) perf jevents: Don't append Unit to desc
> [root@kernelqe3 linux]# perf test 6 10 'perf all metricgroups test' 'perf all metrics test'
>   6: Parse event definition strings                                  :
>   6.1: Test event parsing                                            : Ok
>   6.2: Parsing of all PMU events from sysfs                          : FAILED!
>   6.3: Parsing of given PMU events from sysfs                        : Ok
>   6.4: Parsing of aliased events from sysfs                          : Skip (no aliases in sysfs)
>   6.5: Parsing of aliased events                                     : FAILED!
>   6.6: Parsing of terms (event modifiers)                            : Ok
>  10: PMU events                                                      :
>  10.1: PMU event table sanity                                        : Ok
>  10.2: PMU event map aliases                                         : FAILED!
>  10.3: Parsing of PMU event table metrics                            : Ok
>  10.4: Parsing of PMU event table metrics with fake PMUs             : FAILED!
>  10.5: Parsing of metric thresholds with fake PMUs                   : Ok
>  93: perf all metricgroups test                                      : FAILED!
>  94: perf all metrics test                                           : FAILED!
> [root@kernelqe3 linux]#
>
> Bisecting the first problem:
>
>  10.2: PMU event map aliases                                         : FAILED!
>
> make: Leaving directory '/root/git/linux/tools/perf'
>   6: Parse event definition strings                                  :
>   6.1: Test event parsing                                            : Ok
>   6.2: Parsing of all PMU events from sysfs                          : Ok
>   6.3: Parsing of given PMU events from sysfs                        : Ok
>   6.4: Parsing of aliased events from sysfs                          : Skip (no aliases in sysfs)
>   6.5: Parsing of aliased events                                     : Ok
>   6.6: Parsing of terms (event modifiers)                            : Ok
>  10: PMU events                                                      :
>  10.1: PMU event table sanity                                        : Ok
>  10.2: PMU event map aliases                                         : FAILED!
>  10.3: Parsing of PMU event table metrics                            : Ok
>  10.4: Parsing of PMU event table metrics with fake PMUs             : Ok
>  10.5: Parsing of metric thresholds with fake PMUs                   : Ok
>  93: perf all metricgroups test                                      : Ok
>  94: perf all metrics test                                           : Ok
> [root@kernelqe3 linux]# git bisect bad
> 2e255b4f9f41f137d9e3dc4fae3603a9c2c3dd28 is the first bad commit
> commit 2e255b4f9f41f137d9e3dc4fae3603a9c2c3dd28
> Author: Ian Rogers <irogers@xxxxxxxxxx>
> Date:   Wed Aug 23 21:13:16 2023 -0700
>
>     perf jevents: Group events by PMU
>
>     Prior to this change a cpuid would map to a list of events where the PMU
>     would be encoded alongside the event information. This change breaks
>     apart each group of events so that there is a group per PMU. A new table
>     is added with the PMU's name and the list of events, the original table
>     now holding an array of these per PMU tables.
>
>     These changes are to make it easier to get per PMU information about
>     events, rather than the current approach of scanning all events. The
>     perf binary size with BPF skeletons on x86 is reduced by about 1%. The
>     unidentified PMU is now always expanded to "cpu".
>
>     Signed-off-by: Ian Rogers <irogers@xxxxxxxxxx>
>     Cc: Adrian Hunter <adrian.hunter@xxxxxxxxx>
>     Cc: Alexander Shishkin <alexander.shishkin@xxxxxxxxxxxxxxx>
>     Cc: Gaosheng Cui <cuigaosheng1@xxxxxxxxxx>
>     Cc: Ingo Molnar <mingo@xxxxxxxxxx>
>     Cc: James Clark <james.clark@xxxxxxx>
>     Cc: Jing Zhang <renyu.zj@xxxxxxxxxxxxxxxxx>
>     Cc: Jiri Olsa <jolsa@xxxxxxxxxx>
>     Cc: John Garry <john.g.garry@xxxxxxxxxx>
>     Cc: Kajol Jain <kjain@xxxxxxxxxxxxx>
>     Cc: Kan Liang <kan.liang@xxxxxxxxxxxxxxx>
>     Cc: Mark Rutland <mark.rutland@xxxxxxx>
>     Cc: Namhyung Kim <namhyung@xxxxxxxxxx>
>     Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
>     Cc: Ravi Bangoria <ravi.bangoria@xxxxxxx>
>     Cc: Rob Herring <robh@xxxxxxxxxx>
>     Link: https://lore.kernel.org/r/20230824041330.266337-5-irogers@xxxxxxxxxx
>     Signed-off-by: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>
>
>  tools/perf/pmu-events/jevents.py | 181 +++++++++++++++++++++++++++++----------
>  tools/perf/tests/pmu-events.c    |  30 ++++---
>  2 files changed, 154 insertions(+), 57 deletions(-)
> [root@kernelqe3 linux]#
>

This change defaulted events without a specified PMU to being for the
PMU 'cpu', so that events in pmu-events.c were associated with a PMU
and we could find per-PMU information easily. The test events have no
PMU and so this has broken s390 where the the PMU should be "cpum_cf".
It has probably also broken x86 hybrid and arm where their default PMU
isn't cpu. I'll work on a fix, but the problem will be limited to the
test.

Thanks,
Ian