| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <6e3eae74-c89f-53a7-f954-bf36c7763316@linux.intel.com>
Date: Thu, 24 Aug 2023 15:43:59 -0400
From: "Liang, Kan" <kan.liang@...ux.intel.com>
To: Ian Rogers <irogers@...gle.com>
Cc: Peter Zijlstra <peterz@...radead.org>,
Ingo Molnar <mingo@...hat.com>,
Arnaldo Carvalho de Melo <acme@...nel.org>,
Mark Rutland <mark.rutland@....com>,
Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
Jiri Olsa <jolsa@...nel.org>,
Namhyung Kim <namhyung@...nel.org>,
Adrian Hunter <adrian.hunter@...el.com>,
Ravi Bangoria <ravi.bangoria@....com>,
Kajol Jain <kjain@...ux.ibm.com>,
John Garry <john.g.garry@...cle.com>,
linux-perf-users@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v3 2/3] perf pmus: Add scan that ignores duplicates, use
for perf list
On 2023-08-24 1:30 p.m., Ian Rogers wrote:
> On Thu, Aug 24, 2023 at 7:01 AM Liang, Kan <kan.liang@...ux.intel.com> wrote:
>>
>>
>>
>> On 2023-08-14 12:37 p.m., Ian Rogers wrote:
>>> When there are multiple PMUs that differ only by suffix, by default
>>> just list the first one and skip all others. As the PMUs are sorted,
>>> the scan routine checks that the PMU names match and the numbers are
>>> consecutive.
>>
>> The suffix number may not be consecutive, especially for SPR and later
>> platforms. Because the IDs are from the discovery table now, which is
>> assigned by the HW. The physic IDs are not guaranteed to be consecutive.
>>
>> I don't think there is a plan to change it to logical IDs. Because
>> sometimes people want to know the physic IDs. So they can locate the
>> specific unit quickly.
>
> Thanks Kan,
>
> I think this could lead to perf list merging some PMUs into one name
> and not doing this for others. We could keep the existing behavior by
> checking the numbers are consecutive but it'd come with some
> complexity and runtime cost. We could just ignore the consecutive
> property. We could just not try to solve the problem. What do you
> think is the right strategy?
I like the idea of merging the duplicate PMUs. My only concern is that
the assumption of the consecutive may not work for all the uncore cases.
If the IDs are 0,2,4,6, they cannot be merged successfully, right?
Can we just ignore the consecutive check?
Is there a problem if we just simply remove the "(last_pmu_num + 1 ==
pmu_num) &&", and only compare the no_suffix name?
Thanks,
Kan
>
> On other architectures they encode these numbers in different places
> but generally with no underscore, so this change has no impact for
> them. I'm keen to solve this problem as we're seeing large numbers of
> PMUs that cause perf list to be spammy and the all PMU events test to
> run for too long.>
> Thanks,
> Ian
>
>> Thanks,
>> Kan
>>
>>> If "-v" is passed to "perf list" then list all PMUs.
>>>
>>> Signed-off-by: Ian Rogers <irogers@...gle.com>
>>> ---
>>> tools/perf/builtin-list.c | 8 +++++
>>> tools/perf/util/pmus.c | 54 ++++++++++++++++++++++++++++++++--
>>> tools/perf/util/print-events.h | 1 +
>>> 3 files changed, 61 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/tools/perf/builtin-list.c b/tools/perf/builtin-list.c
>>> index 7fec2cca759f..8fe4ddf02c14 100644
>>> --- a/tools/perf/builtin-list.c
>>> +++ b/tools/perf/builtin-list.c
>>> @@ -423,6 +423,13 @@ static void json_print_metric(void *ps __maybe_unused, const char *group,
>>> strbuf_release(&buf);
>>> }
>>>
>>> +static bool default_skip_duplicate_pmus(void *ps)
>>> +{
>>> + struct print_state *print_state = ps;
>>> +
>>> + return !print_state->long_desc;
>>> +}
>>> +
>>> int cmd_list(int argc, const char **argv)
>>> {
>>> int i, ret = 0;
>>> @@ -434,6 +441,7 @@ int cmd_list(int argc, const char **argv)
>>> .print_end = default_print_end,
>>> .print_event = default_print_event,
>>> .print_metric = default_print_metric,
>>> + .skip_duplicate_pmus = default_skip_duplicate_pmus,
>>> };
>>> const char *cputype = NULL;
>>> const char *unit_name = NULL;
>>> diff --git a/tools/perf/util/pmus.c b/tools/perf/util/pmus.c
>>> index 3581710667b0..5073843aca19 100644
>>> --- a/tools/perf/util/pmus.c
>>> +++ b/tools/perf/util/pmus.c
>>> @@ -275,6 +275,50 @@ struct perf_pmu *perf_pmus__scan_core(struct perf_pmu *pmu)
>>> return NULL;
>>> }
>>>
>>> +static struct perf_pmu *perf_pmus__scan_skip_duplicates(struct perf_pmu *pmu)
>>> +{
>>> + bool use_core_pmus = !pmu || pmu->is_core;
>>> + int last_pmu_name_len = 0;
>>> + unsigned long last_pmu_num = 0;
>>> + const char *last_pmu_name = (pmu && pmu->name) ? pmu->name : "";
>>> +
>>> + if (!pmu) {
>>> + pmu_read_sysfs(/*core_only=*/false);
>>> + pmu = list_prepare_entry(pmu, &core_pmus, list);
>>> + } else
>>> + last_pmu_name_len = pmu_name_len_no_suffix(pmu->name ?: "", &last_pmu_num);
>>> +
>>> + if (use_core_pmus) {
>>> + list_for_each_entry_continue(pmu, &core_pmus, list) {
>>> + unsigned long pmu_num = 0;
>>> + int pmu_name_len = pmu_name_len_no_suffix(pmu->name ?: "", &pmu_num);
>>> +
>>> + if (last_pmu_name_len == pmu_name_len &&
>>> + (last_pmu_num + 1 == pmu_num) &&
>>> + !strncmp(last_pmu_name, pmu->name ?: "", pmu_name_len)) {
>>> + last_pmu_num++;
>>> + continue;
>>> + }
>>> + return pmu;
>>> + }
>>> + pmu = NULL;
>>> + pmu = list_prepare_entry(pmu, &other_pmus, list);
>>> + }
>>> + list_for_each_entry_continue(pmu, &other_pmus, list) {
>>> + unsigned long pmu_num = 0;
>>> + int pmu_name_len = pmu_name_len_no_suffix(pmu->name ?: "", &pmu_num);
>>> +
>>> + if (last_pmu_name_len == pmu_name_len &&
>>> + (last_pmu_num + 1 == pmu_num) &&
>>> + !strncmp(last_pmu_name, pmu->name ?: "", pmu_name_len)) {
>>> + last_pmu_num++;
>>> + continue;
>>> + }
>>> + return pmu;
>>> + }
>>> + return NULL;
>>> +}
>>> +
>>> const struct perf_pmu *perf_pmus__pmu_for_pmu_filter(const char *str)
>>> {
>>> struct perf_pmu *pmu = NULL;
>>> @@ -429,10 +473,16 @@ void perf_pmus__print_pmu_events(const struct print_callbacks *print_cb, void *p
>>> int printed = 0;
>>> int len, j;
>>> struct sevent *aliases;
>>> + struct perf_pmu *(*scan_fn)(struct perf_pmu *);
>>> +
>>> + if (print_cb->skip_duplicate_pmus(print_state))
>>> + scan_fn = perf_pmus__scan_skip_duplicates;
>>> + else
>>> + scan_fn = perf_pmus__scan;
>>>
>>> pmu = NULL;
>>> len = 0;
>>> - while ((pmu = perf_pmus__scan(pmu)) != NULL) {
>>> + while ((pmu = scan_fn(pmu)) != NULL) {
>>> list_for_each_entry(event, &pmu->aliases, list)
>>> len++;
>>> if (pmu->selectable)
>>> @@ -445,7 +495,7 @@ void perf_pmus__print_pmu_events(const struct print_callbacks *print_cb, void *p
>>> }
>>> pmu = NULL;
>>> j = 0;
>>> - while ((pmu = perf_pmus__scan(pmu)) != NULL) {
>>> + while ((pmu = scan_fn(pmu)) != NULL) {
>>> bool is_cpu = pmu->is_core;
>>>
>>> list_for_each_entry(event, &pmu->aliases, list) {
>>> diff --git a/tools/perf/util/print-events.h b/tools/perf/util/print-events.h
>>> index d7fab411e75c..bf4290bef0cd 100644
>>> --- a/tools/perf/util/print-events.h
>>> +++ b/tools/perf/util/print-events.h
>>> @@ -26,6 +26,7 @@ struct print_callbacks {
>>> const char *expr,
>>> const char *threshold,
>>> const char *unit);
>>> + bool (*skip_duplicate_pmus)(void *print_state);
>>> };
>>>
>>> /** Print all events, the default when no options are specified. */
Powered by blists - more mailing lists