| 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: <b53c62a5-a752-436a-b039-cdc82b055fdd@huawei.com>
Date: Wed, 21 Jan 2026 11:15:07 +0800
From: Qinxin Xia <xiaqinxin@...wei.com>
To: Ian Rogers <irogers@...gle.com>
CC: <peterz@...radead.org>, <mingo@...hat.com>, <acme@...nel.org>,
<namhyung@...nel.org>, <wangyushan12@...wei.com>, <hejunhao3@...artners.com>,
<jonathan.cameron@...wei.com>, <linux-perf-users@...r.kernel.org>,
<linux-kernel@...r.kernel.org>, <mark.rutland@....com>,
<alexander.shishkin@...ux.intel.com>, <jolsa@...nel.org>,
<adrian.hunter@...el.com>, <james.clark@...aro.org>, <linuxarm@...wei.com>
Subject: Re: [RFC PATCH] perf: Add --uncorepmu option for filtering uncore
PMUs
On 2026/1/21 04:55:30, Ian Rogers <irogers@...gle.com> wrote:
> On Tue, Jan 20, 2026 at 1:51 AM Qinxin Xia <xiaqinxin@...wei.com> wrote:
>>
>> This patch adds a new --uncorepmu option to perf-stat command to allow
>> filtering events on specific uncore PMUs. This is useful when there are
>> multiple uncore PMUs with the same type (e.g. hisi_sicl2_cpa0 and
>> hisi_sicl0_cpa0).
>>
>> eg:
>> [root@...alhost tmp]# ./perf stat -M cpa_p0_avg_bw
>> Performance counter stats for 'system wide':
>>
>> 19,417,779,115 hisi_sicl0_cpa0/cpa_cycles/ # 0.00 cpa_p0_avg_bw
>> 0 hisi_sicl0_cpa0/cpa_p0_wr_dat/
>> 0 hisi_sicl0_cpa0/cpa_p0_rd_dat_64b/
>> 0 hisi_sicl0_cpa0/cpa_p0_rd_dat_32b/
>> 19,417,751,103 hisi_sicl10_cpa0/cpa_cycles/ # 0.00 cpa_p0_avg_bw
>> 0 hisi_sicl10_cpa0/cpa_p0_wr_dat/
>> 0 hisi_sicl10_cpa0/cpa_p0_rd_dat_64b/
>> 0 hisi_sicl10_cpa0/cpa_p0_rd_dat_32b/
>> 19,417,730,679 hisi_sicl2_cpa0/cpa_cycles/ # 0.31 cpa_p0_avg_bw
>> 75,635,749 hisi_sicl2_cpa0/cpa_p0_wr_dat/
>> 18,520,640 hisi_sicl2_cpa0/cpa_p0_rd_dat_64b/
>> 0 hisi_sicl2_cpa0/cpa_p0_rd_dat_32b/
>> 19,417,674,227 hisi_sicl8_cpa0/cpa_cycles/ # 0.00 cpa_p0_avg_bw
>> 0 hisi_sicl8_cpa0/cpa_p0_wr_dat/
>> 0 hisi_sicl8_cpa0/cpa_p0_rd_dat_64b/
>> 0 hisi_sicl8_cpa0/cpa_p0_rd_dat_32b/
>>
>> 19.417734480 seconds time elapsed
>>
>> [root@...alhost tmp]# ./perf stat --uncorepmu hisi_sicl2_cpa0 -M cpa_p0_avg_bw
>> Performance counter stats for 'system wide':
>>
>> 6,234,093,559 cpa_cycles # 0.60 cpa_p0_avg_bw
>> 50,548,465 cpa_p0_wr_dat
>> 7,552,182 cpa_p0_rd_dat_64b
>> 0 cpa_p0_rd_dat_32b
>>
>> 6.234139320 seconds time elapsed
>
> So normally for this case the -A/--no-merge will give you a per-PMU
> breakdown. It isn't restricted to a single PMU like the output here,
> but we similarly don't have options to restrict per-socket, per-dir,
> per-cluster and per-core. There is an option with --per-cache but that
> looks like a way to avoid having many --per-cache options rather than
> taking a value for a particular cache to only dump.
>
> I did deliberate make the --cputype code generic so it could be used
> in ways like this, but it'd be nice to glob match the names for
> consistency with event parsing:
> https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/util/pmu.c#n2647
> I needed to have --cputype as Intel's early hybrid patches had added
> it before I cleaned things up to be more PMU oriented.
>
Hello, Ian!
For core-related events and metric, we usually care about which core it
happens on, so we can filter it with -C.
Such as:
[root@...alhost ~]# perf stat -C 0 -e L1-dcache-loads sleep 3
Performance counter stats for 'CPU(s) 0':
136,578 L1-dcache-loads
3.099147550 seconds time elapsed
[root@...alhost tool]# ./perf stat -C 0 -M l1d_miss_rate sleep 3
Performance counter stats for 'CPU(s) 0':
18,708 L1-dcache-load-misses # 3.6 %
l1d_miss_rate
513,038 L1-dcache-loads
3.004649800 seconds time elapsed
For core-independent events and metrics such as ddrc, smmu, and cpa, we
care about which hardware module they occur. (The example provided in
commit info is sicl.) so we want to filter it by pmu, for events I can
filter it with'cputype' (This is a little strange because sicl does not
belong to any cputype.). However, metric does not support this mode.
[root@...alhost ~]# perf stat --cputype hisi_sicl0_cpa0 -e cpa_cycles
sleep 3
Performance counter stats for 'system wide':
3,003,492,731 cpa_cycles
3.003484270 seconds time elapsed
[root@...alhost ~]# perf stat --cputype hisi_sicl0_cpa0 -M
cpa_p0_wr_dat sleep 3
Cannot find metric or group `cpa_p0_wr_dat'
This patch supports the metric scenario through '--uncorepmu'.
> So I can see the purpose of this patch in being low overhead compared
> to -A, I'm just not sure whether this is a PMU or an aggregation
> option. I also think the term "uncore" is kind of confusing and so
> have tried to keep it away from command line options. Why isn't is it
> just a pmu filter?
>
> Thanks,
> Ian
>
Perhaps we could make cputype a little more generic so that it could
also include pmu-filter functionality.
>> Signed-off-by: Qinxin Xia <xiaqinxin@...wei.com>
>> ---
>> tools/perf/Documentation/perf-stat.txt | 4 ++++
>> tools/perf/builtin-stat.c | 26 ++++++++++++++++++++++++++
>> tools/perf/util/metricgroup.c | 18 +++++++++++++-----
>> 3 files changed, 43 insertions(+), 5 deletions(-)
>>
>> diff --git a/tools/perf/Documentation/perf-stat.txt b/tools/perf/Documentation/perf-stat.txt
>> index 1a766d4a2233..9c0952b43a57 100644
>> --- a/tools/perf/Documentation/perf-stat.txt
>> +++ b/tools/perf/Documentation/perf-stat.txt
>> @@ -573,6 +573,10 @@ $ perf config stat.no-csv-summary=true
>> Only enable events on applying cpu with this type for hybrid platform
>> (e.g. core or atom)"
>>
>> +--uncorepmu::
>> +Only enable events on applying uncore pmu with specified for multiple
>> +uncore pmus with same type (e.g. hisi_sicl2_cpa0 or hisi_sicl0_cpa0)
>> +
>> EXAMPLES
>> --------
>>
>> diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
>> index ab40d85fb125..3f34bfc79179 100644
>> --- a/tools/perf/builtin-stat.c
>> +++ b/tools/perf/builtin-stat.c
>> @@ -1235,6 +1235,28 @@ static int parse_cputype(const struct option *opt,
>> return 0;
>> }
>>
>> +static int parse_uncorepmu(const struct option *opt,
>> + const char *str,
>> + int unset __maybe_unused)
>> +{
>> + const struct perf_pmu *pmu;
>> + struct evlist *evlist = *(struct evlist **)opt->value;
>> +
>> + if (!list_empty(&evlist->core.entries)) {
>> + fprintf(stderr, "Must define uncorepmu before events/metrics\n");
>> + return -1;
>> + }
>> +
>> + pmu = perf_pmus__pmu_for_pmu_filter(str);
>> + if (!pmu || pmu->is_core) {
>> + fprintf(stderr, "--uncorepmu %s is not supported!\n", str);
>> + return -1;
>> + }
>> + parse_events_option_args.pmu_filter = pmu->name;
>> +
>> + return 0;
>> +}
>> +
>> static int parse_cache_level(const struct option *opt,
>> const char *str,
>> int unset __maybe_unused)
>> @@ -2579,6 +2601,10 @@ int cmd_stat(int argc, const char **argv)
>> "Only enable events on applying cpu with this type "
>> "for hybrid platform (e.g. core or atom)",
>> parse_cputype),
>> + OPT_CALLBACK(0, "uncorepmu", &evsel_list, "uncore pmu",
>> + "Only enable events on applying uncore pmu with specified "
>> + "for multiple uncore pmus with same type(e.g. hisi_sicl2_cpa0 or hisi_sicl0_cpa0)",
>> + parse_uncorepmu),
>> #ifdef HAVE_LIBPFM
>> OPT_CALLBACK(0, "pfm-events", &evsel_list, "event",
>> "libpfm4 event selector. use 'perf list' to list available events",
>> diff --git a/tools/perf/util/metricgroup.c b/tools/perf/util/metricgroup.c
>> index 25c75fdbfc52..59ef71df20bd 100644
>> --- a/tools/perf/util/metricgroup.c
>> +++ b/tools/perf/util/metricgroup.c
>> @@ -387,8 +387,13 @@ static bool match_pm_metric_or_groups(const struct pmu_metric *pm, const char *p
>> const char *metric_or_groups)
>> {
>> const char *pm_pmu = pm->pmu ?: "cpu";
>> + struct perf_pmu *perf_pmu = NULL;
>>
>> - if (strcmp(pmu, "all") && strcmp(pm_pmu, pmu))
>> + if (pm->pmu)
>> + perf_pmu = perf_pmus__find(pm->pmu);
>> +
>> + if (strcmp(pmu, "all") && strcmp(pm_pmu, pmu) &&
>> + (perf_pmu && !perf_pmu__name_wildcard_match(perf_pmu, pmu)))
>> return false;
>>
>> return match_metric_or_groups(pm->metric_group, metric_or_groups) ||
>> @@ -1260,7 +1265,8 @@ static int build_combined_expr_ctx(const struct list_head *metric_list,
>> static int parse_ids(bool metric_no_merge, bool fake_pmu,
>> struct expr_parse_ctx *ids, const char *modifier,
>> bool group_events, const bool tool_events[TOOL_PMU__EVENT_MAX],
>> - struct evlist **out_evlist)
>> + struct evlist **out_evlist,
>> + const char *filter_pmu)
>> {
>> struct parse_events_error parse_error;
>> struct evlist *parsed_evlist;
>> @@ -1314,7 +1320,7 @@ static int parse_ids(bool metric_no_merge, bool fake_pmu,
>> }
>> pr_debug("Parsing metric events '%s'\n", events.buf);
>> parse_events_error__init(&parse_error);
>> - ret = __parse_events(parsed_evlist, events.buf, /*pmu_filter=*/NULL,
>> + ret = __parse_events(parsed_evlist, events.buf, filter_pmu,
>> &parse_error, fake_pmu, /*warn_if_reordered=*/false,
>> /*fake_tp=*/false);
>> if (ret) {
>> @@ -1417,7 +1423,8 @@ static int parse_groups(struct evlist *perf_evlist,
>> /*modifier=*/NULL,
>> /*group_events=*/false,
>> tool_events,
>> - &combined_evlist);
>> + &combined_evlist,
>> + (pmu && strcmp(pmu, "all") == 0) ? NULL : pmu);
>> }
>> if (combined)
>> expr__ctx_free(combined);
>> @@ -1472,7 +1479,8 @@ static int parse_groups(struct evlist *perf_evlist,
>> }
>> if (!metric_evlist) {
>> ret = parse_ids(metric_no_merge, fake_pmu, m->pctx, m->modifier,
>> - m->group_events, tool_events, &m->evlist);
>> + m->group_events, tool_events, &m->evlist,
>> + (pmu && strcmp(pmu, "all") == 0) ? NULL : pmu);
>> if (ret)
>> goto out;
>>
>> --
>> 2.33.0
>>
--
Thanks,
Qinxin
Powered by blists - more mailing lists