Re: [PATCH] perf record: Support "--cputype" option for hybrid events

From: Xing Zhengjun
Date: Tue Jun 21 2022 - 01:14:02 EST

Next message: Neeraj Upadhyay: "Re: [PATCH rcu 3/3] doc: Document rcutree.nocb_nobypass_lim_per_jiffy kernel parameter"
Previous message: Luck, Tony: "Re: [RFC PATCH 4/5] x86/mce: Move storm handling to core."
In reply to: Liang, Kan: "Re: [PATCH] perf record: Support "--cputype" option for hybrid events"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Hi Ian,

On 6/16/2022 10:31 PM, Liang, Kan wrote:

On 6/16/2022 4:31 AM, Xing Zhengjun wrote:

Hi Ian,

On 6/15/2022 11:16 PM, Ian Rogers wrote:

On Wed, Jun 15, 2022 at 8:07 AM <zhengjun.xing@xxxxxxxxxxxxxxx> wrote:

From: Zhengjun Xing <zhengjun.xing@xxxxxxxxxxxxxxx>

perf stat already has the "--cputype" option to enable events only on the
specified PMU for the hybrid platform, this commit extends the "--cputype"
support to perf record.

Without "--cputype", it reports events for both cpu_core and cpu_atom.

# ./perf record -e cycles -a sleep 1 | ./perf report

# To display the perf.data header info, please use --header/--header-only options.
#
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.000 MB (null) ]
#
# Total Lost Samples: 0
#
# Samples: 335 of event 'cpu_core/cycles/'
# Event count (approx.): 35855267
#
# Overhead Command          Shared Object      Symbol
# ........ ............... ................. .........................................
#
      10.31% swapper          [kernel.kallsyms] [k] poll_idle
       9.42% swapper          [kernel.kallsyms] [k] menu_select
       ...    ...               ...               ... ...

# Samples: 61 of event 'cpu_atom/cycles/'
# Event count (approx.): 16453825
#
# Overhead Command        Shared Object      Symbol
# ........ ............. ................. ......................................
#
      26.36% snapd          [unknown]          [.] 0x0000563cc6d03841
       7.43% migration/13   [kernel.kallsyms] [k] update_sd_lb_stats.constprop.0
       ...    ...            ...                ... ...

With "--cputype", it reports events only for the specified PMU.

# ./perf record --cputype core -e cycles -a sleep 1 | ./perf report

# To display the perf.data header info, please use --header/--header-only options.
#
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.000 MB (null) ]
#
# Total Lost Samples: 0
#
# Samples: 221 of event 'cpu_core/cycles/'
# Event count (approx.): 27121818
#
# Overhead Command          Shared Object      Symbol
# ........ ............... ................. .........................................
#
      11.24% swapper          [kernel.kallsyms] [k] e1000_irq_enable
       7.77% swapper          [kernel.kallsyms] [k] mwait_idle_with_hints.constprop.0
       ...    ...              ...                ... ...

This is already possible by having the PMU name as part of the event,
cpu_atom/cycles/ or cpu_core/cycles/. I don't know why we're adding
"hybrid" all over the code base, I wish it would stop. You are asking
for an option here for an implied PMU for events that don't specify a
PMU. The option should be called --pmutype and consider all PMUs.

The --pmutype should be redundant because other PMUs have to specify the PMU name with the event. Only the CPU type of PMUs have events which doesn't have a PMU name prefix, e.g., cycles, instructions.
So the "--cputype" was introduced for perf stat to avoid the annoying pmu prefix for the CPU type of PMU.

This patch just extends the "--cputype" to perf record to address the request from the community. It reuses the existing function.

Yes, "--cputype" is better. Ian, what do you think? Thanks.

We
should remove the "hybrid" PMU list and make all of the "hybrid" code
generic.

We already start to rework on the "hybrid" code.

We recently rework the perf stat default
https://lore.kernel.org/lkml/20220610025449.2089232-1-zhengjun.xing@xxxxxxxxxxxxxxx/

With the help of Ravi, we clean up the hybrid CPU in the perf header.
https://lore.kernel.org/all/20220604044519.594-6-ravi.bangoria@xxxxxxx/

I think Zhengjun will work on the perf record default cleanup shortly.

Then I guess we can further clean up the "--cputype", e.g., removing hybrid_pmu_name from evlist... as next step.

Thanks,
Kan

I can change the option name from "cputype" to "pmutype". We have planned to clean up the hybrid code, and try to reduce the code with "if perf_pmu__has_hybrid()", this will be done step by step, from this patch, we have begun to do some clean-up, move the hybrid API from the common file to the hybrid-specific files. For the clean-up, Do you have any ideas? Thanks.

Thanks,
Ian

Signed-off-by: Zhengjun Xing <zhengjun.xing@xxxxxxxxxxxxxxx>
Reviewed-by: Kan Liang <kan.liang@xxxxxxxxxxxxxxx>
---
tools/perf/Documentation/perf-record.txt | 4 ++++
tools/perf/builtin-record.c              | 3 +++
tools/perf/builtin-stat.c                | 20 --------------------
tools/perf/util/pmu-hybrid.c             | 19 +++++++++++++++++++
tools/perf/util/pmu-hybrid.h             | 2 ++
5 files changed, 28 insertions(+), 20 deletions(-)

diff --git a/tools/perf/Documentation/perf-record.txt b/tools/perf/Documentation/perf-record.txt
index cf8ad50f3de1..ba8d680da1ac 100644
--- a/tools/perf/Documentation/perf-record.txt
+++ b/tools/perf/Documentation/perf-record.txt
@@ -402,6 +402,10 @@ Enable weightened sampling. An additional weight is recorded per sample and can
displayed with the weight and local_weight sort keys. This currently works for TSX
abort events and some memory events in precise mode on modern Intel CPUs.

+--cputype::
+Only enable events on applying cpu with this type for hybrid platform(e.g. core or atom).
+For non-hybrid events, it should be no effect.
+
--namespaces::
Record events of type PERF_RECORD_NAMESPACES. This enables 'cgroup_id' sort key.

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 9a71f0330137..e1edd4e98358 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -3183,6 +3183,9 @@ static struct option __record_options[] = {
         OPT_INCR('v', "verbose", &verbose,
                     "be more verbose (show counter open errors, etc)"),
         OPT_BOOLEAN('q', "quiet", &quiet, "don't print any message"),
+       OPT_CALLBACK(0, "cputype", &record.evlist, "hybrid cpu type",
+                    "Only enable events on applying cpu with this type for hybrid platform (e.g. core or atom)",
+                    parse_hybrid_type),
         OPT_BOOLEAN('s', "stat", &record.opts.inherit_stat,
                     "per thread counts"),
         OPT_BOOLEAN('d', "data", &record.opts.sample_address, "Record the sample addresses"),
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 4ce87a8eb7d7..0d95b29273f4 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -1184,26 +1184,6 @@ static int parse_stat_cgroups(const struct option *opt,
         return parse_cgroups(opt, str, unset);
}

-static int parse_hybrid_type(const struct option *opt,
-                            const char *str,
-                            int unset __maybe_unused)
-{
-       struct evlist *evlist = *(struct evlist **)opt->value;
-
-       if (!list_empty(&evlist->core.entries)) {
-               fprintf(stderr, "Must define cputype before events/metrics\n");
-               return -1;
-       }
-
-       evlist->hybrid_pmu_name = perf_pmu__hybrid_type_to_pmu(str);
-       if (!evlist->hybrid_pmu_name) {
-               fprintf(stderr, "--cputype %s is not supported!\n", str);
-               return -1;
-       }
-
-       return 0;
-}
-
static struct option stat_options[] = {
         OPT_BOOLEAN('T', "transaction", &transaction_run,
                     "hardware transaction statistics"),
diff --git a/tools/perf/util/pmu-hybrid.c b/tools/perf/util/pmu-hybrid.c
index f51ccaac60ee..5c490b5201b7 100644
--- a/tools/perf/util/pmu-hybrid.c
+++ b/tools/perf/util/pmu-hybrid.c
@@ -13,6 +13,7 @@
#include <stdarg.h>
#include <locale.h>
#include <api/fs/fs.h>
+#include "util/evlist.h"
#include "fncache.h"
#include "pmu-hybrid.h"

@@ -87,3 +88,21 @@ char *perf_pmu__hybrid_type_to_pmu(const char *type)
         free(pmu_name);
         return NULL;
}
+
+int parse_hybrid_type(const struct option *opt, const char *str, int unset __maybe_unused)
+{
+       struct evlist *evlist = *(struct evlist **)opt->value;
+
+       if (!list_empty(&evlist->core.entries)) {
+               fprintf(stderr, "Must define cputype before events/metrics\n");
+               return -1;
+       }
+
+       evlist->hybrid_pmu_name = perf_pmu__hybrid_type_to_pmu(str);
+       if (!evlist->hybrid_pmu_name) {
+               fprintf(stderr, "--cputype %s is not supported!\n", str);
+               return -1;
+       }
+
+       return 0;
+}
diff --git a/tools/perf/util/pmu-hybrid.h b/tools/perf/util/pmu-hybrid.h
index 2b186c26a43e..26101f134a3a 100644
--- a/tools/perf/util/pmu-hybrid.h
+++ b/tools/perf/util/pmu-hybrid.h
@@ -5,6 +5,7 @@
#include <linux/perf_event.h>
#include <linux/compiler.h>
#include <linux/list.h>
+#include <subcmd/parse-options.h>
#include <stdbool.h>
#include "pmu.h"

@@ -18,6 +19,7 @@ bool perf_pmu__hybrid_mounted(const char *name);
struct perf_pmu *perf_pmu__find_hybrid_pmu(const char *name);
bool perf_pmu__is_hybrid(const char *name);
char *perf_pmu__hybrid_type_to_pmu(const char *type);
+int parse_hybrid_type(const struct option *opt, const char *str, int unset __maybe_unused);

static inline int perf_pmu__hybrid_pmu_num(void)
{
--
2.25.1

--
Zhengjun Xing

Next message: Neeraj Upadhyay: "Re: [PATCH rcu 3/3] doc: Document rcutree.nocb_nobypass_lim_per_jiffy kernel parameter"
Previous message: Luck, Tony: "Re: [RFC PATCH 4/5] x86/mce: Move storm handling to core."
In reply to: Liang, Kan: "Re: [PATCH] perf record: Support "--cputype" option for hybrid events"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]