The perf-record and perf-stat have supported the option '-C/--cpus'
to count or collect only on the list of CPUs provided. This option
needs to be supported for hybrid as well.
v4:
---
Rename evlist__use_cpu_list to evlist__fix_hybrid_cpus
Add comments to evlist__fix_hybrid_cpus.
v3:
---
Rebase to latest perf/core branch.
v2:
---
Automatically map to hybrid pmu.
For example,
If cpu0-7 are 'cpu_core' and cpu9-11 are 'cpu_atom',
# perf stat -e cycles -C0,11 -- sleep 1
WARNING: use 0 in 'cpu_core' for 'cycles', skip other cpus in list.
WARNING: use 11 in 'cpu_atom' for 'cycles', skip other cpus in list.
Performance counter stats for 'CPU(s) 0,11':
1,914,704 cpu_core/cycles/
2,036,983 cpu_atom/cycles/
1.005815641 seconds time elapsed
It automatically selects cpu0 for cpu_core/cycles/, selects cpu11 for
cpu_atom/cycles/, also with some warnings output.
Jin Yao (3):
libperf: Add perf_cpu_map__default_new()
perf tools: Create hybrid flag in target
perf tools: Enable on a list of CPUs for hybrid
tools/lib/perf/cpumap.c | 5 ++
tools/lib/perf/include/perf/cpumap.h | 1 +
tools/perf/builtin-record.c | 7 +++
tools/perf/builtin-stat.c | 6 +++
tools/perf/util/evlist-hybrid.c | 73 ++++++++++++++++++++++++++++
tools/perf/util/evlist-hybrid.h | 1 +
tools/perf/util/evlist.c | 3 +-
tools/perf/util/pmu.c | 35 +++++++++++++
tools/perf/util/pmu.h | 4 ++
tools/perf/util/target.h | 1 +
10 files changed, 135 insertions(+), 1 deletion(-)
--
2.17.1
The libperf already has a static function 'cpu_map__default_new()'.
Add a new API perf_cpu_map__default_new() to export the function.
Signed-off-by: Jin Yao <[email protected]>
---
v3/v4:
- No change.
v2:
- New in v2.
tools/lib/perf/cpumap.c | 5 +++++
tools/lib/perf/include/perf/cpumap.h | 1 +
2 files changed, 6 insertions(+)
diff --git a/tools/lib/perf/cpumap.c b/tools/lib/perf/cpumap.c
index ca0215047c32..51b6553912e0 100644
--- a/tools/lib/perf/cpumap.c
+++ b/tools/lib/perf/cpumap.c
@@ -68,6 +68,11 @@ static struct perf_cpu_map *cpu_map__default_new(void)
return cpus;
}
+struct perf_cpu_map *perf_cpu_map__default_new(void)
+{
+ return cpu_map__default_new();
+}
+
static int cmp_int(const void *a, const void *b)
{
return *(const int *)a - *(const int*)b;
diff --git a/tools/lib/perf/include/perf/cpumap.h b/tools/lib/perf/include/perf/cpumap.h
index 6a17ad730cbc..7c27766ea0bf 100644
--- a/tools/lib/perf/include/perf/cpumap.h
+++ b/tools/lib/perf/include/perf/cpumap.h
@@ -9,6 +9,7 @@
struct perf_cpu_map;
LIBPERF_API struct perf_cpu_map *perf_cpu_map__dummy_new(void);
+LIBPERF_API struct perf_cpu_map *perf_cpu_map__default_new(void);
LIBPERF_API struct perf_cpu_map *perf_cpu_map__new(const char *cpu_list);
LIBPERF_API struct perf_cpu_map *perf_cpu_map__read(FILE *file);
LIBPERF_API struct perf_cpu_map *perf_cpu_map__get(struct perf_cpu_map *map);
--
2.17.1
User may count or collect only on a cpu list via '-C/--cpus' option.
Previously cpus for evsel was retrieved from pmu sysfs. But if the
target cpu list is defined, the retrieved cpus are not kept and the
target cpu list is used instead.
But for hybrid system, we can't directly use target cpu list. The
cpu list may not available on hybrid pmu (e.g. cpu_core or cpu_atom).
So we should not set 'has_user_cpus' flag for hybrid system.
The difficulity is that we can't call perf_pmu__has_hybrid() in evlist.c
to check hybrid system otherwise 'perf test python' would be failed
(undefined symbol for perf_pmu__has_hybrid). If we add pmu.c to
python-ext-sources, too many symbol dependencies are hard to resolve.
We use an alternative method by using a new 'hybrid' flag in target
for hybrid system checking.
Signed-off-by: Jin Yao <[email protected]>
---
v3/v4:
- No change.
v2:
- New in v2.
tools/perf/util/evlist.c | 2 +-
tools/perf/util/target.h | 1 +
2 files changed, 2 insertions(+), 1 deletion(-)
diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index 47581a237c7a..06f8890816c3 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -1002,7 +1002,7 @@ int evlist__create_maps(struct evlist *evlist, struct target *target)
if (!cpus)
goto out_delete_threads;
- evlist->core.has_user_cpus = !!target->cpu_list;
+ evlist->core.has_user_cpus = !!target->cpu_list && !target->hybrid;
perf_evlist__set_maps(&evlist->core, cpus, threads);
diff --git a/tools/perf/util/target.h b/tools/perf/util/target.h
index 4ff56217f2a6..daec6cba500d 100644
--- a/tools/perf/util/target.h
+++ b/tools/perf/util/target.h
@@ -17,6 +17,7 @@ struct target {
bool default_per_cpu;
bool per_thread;
bool use_bpf;
+ bool hybrid;
const char *attr_map;
};
--
2.17.1
The perf-record and perf-stat have supported the option '-C/--cpus'
to count or collect only on the list of CPUs provided. This option
needs to be supported for hybrid as well.
For hybrid support, it needs to check that the cpu list are available
on hybrid PMU. One example for AlderLake, cpu0-7 is 'cpu_core', cpu8-11
is 'cpu_atom'.
Before:
# perf stat -e cpu_core/cycles/ -C11 -- sleep 1
Performance counter stats for 'CPU(s) 11':
<not supported> cpu_core/cycles/
1.006179431 seconds time elapsed
The perf-stat silently returned "<not supported>" without any helpful
information. It should error out that cpu11 was not 'cpu_core'.
After:
# perf stat -e cpu_core/cycles/ -C11 -- sleep 1
WARNING: 11 isn't a 'cpu_core', please use a CPU list in the 'cpu_core' range (0-7)
failed to use cpu list 11
We also need to support the events without pmu prefix specified.
# perf stat -e cycles -C11 -- sleep 1
WARNING: 11 isn't a 'cpu_core', please use a CPU list in the 'cpu_core' range (0-7)
Performance counter stats for 'CPU(s) 11':
1,067,373 cpu_atom/cycles/
1.005544738 seconds time elapsed
The perf tool creates two cycles events automatically, cpu_core/cycles/ and
cpu_atom/cycles/. It checks that cpu11 is not 'cpu_core', then shows a warning
for cpu_core/cycles/ and only count the cpu_atom/cycles/.
If part of cpus are 'cpu_core' and part of cpus are 'cpu_atom', for example,
# perf stat -e cycles -C0,11 -- sleep 1
WARNING: use 0 in 'cpu_core' for 'cycles', skip other cpus in list.
WARNING: use 11 in 'cpu_atom' for 'cycles', skip other cpus in list.
Performance counter stats for 'CPU(s) 0,11':
1,914,704 cpu_core/cycles/
2,036,983 cpu_atom/cycles/
1.005815641 seconds time elapsed
It now automatically selects cpu0 for cpu_core/cycles/, selects cpu11 for
cpu_atom/cycles/, and output with some warnings.
Some more complex examples,
# perf stat -e cycles,instructions -C0,11 -- sleep 1
WARNING: use 0 in 'cpu_core' for 'cycles', skip other cpus in list.
WARNING: use 11 in 'cpu_atom' for 'cycles', skip other cpus in list.
WARNING: use 0 in 'cpu_core' for 'instructions', skip other cpus in list.
WARNING: use 11 in 'cpu_atom' for 'instructions', skip other cpus in list.
Performance counter stats for 'CPU(s) 0,11':
2,780,387 cpu_core/cycles/
1,583,432 cpu_atom/cycles/
3,957,277 cpu_core/instructions/
1,167,089 cpu_atom/instructions/
1.006005124 seconds time elapsed
# perf stat -e cycles,cpu_atom/instructions/ -C0,11 -- sleep 1
WARNING: use 0 in 'cpu_core' for 'cycles', skip other cpus in list.
WARNING: use 11 in 'cpu_atom' for 'cycles', skip other cpus in list.
WARNING: use 11 in 'cpu_atom' for 'cpu_atom/instructions/', skip other cpus in list.
Performance counter stats for 'CPU(s) 0,11':
3,290,301 cpu_core/cycles/
1,953,073 cpu_atom/cycles/
1,407,869 cpu_atom/instructions/
1.006260912 seconds time elapsed
Signed-off-by: Jin Yao <[email protected]>
---
v4:
- Rename evlist__use_cpu_list to evlist__fix_hybrid_cpus
- Add comments to evlist__fix_hybrid_cpus.
v3:
- Rebase to perf/core.
- No functional change.
v2:
- Automatically map to hybrid pmu.
tools/perf/builtin-record.c | 7 ++++
tools/perf/builtin-stat.c | 6 +++
tools/perf/util/evlist-hybrid.c | 73 +++++++++++++++++++++++++++++++++
tools/perf/util/evlist-hybrid.h | 1 +
tools/perf/util/evlist.c | 1 +
tools/perf/util/pmu.c | 35 ++++++++++++++++
tools/perf/util/pmu.h | 4 ++
7 files changed, 127 insertions(+)
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 671a21c9ee4d..4065bc8eba3c 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -2884,6 +2884,13 @@ int cmd_record(int argc, const char **argv)
/* Enable ignoring missing threads when -u/-p option is defined. */
rec->opts.ignore_missing_thread = rec->opts.target.uid != UINT_MAX || rec->opts.target.pid;
+ if (evlist__fix_hybrid_cpus(rec->evlist, rec->opts.target.cpu_list)) {
+ pr_err("failed to use cpu list %s\n",
+ rec->opts.target.cpu_list);
+ goto out;
+ }
+
+ rec->opts.target.hybrid = perf_pmu__has_hybrid();
err = -ENOMEM;
if (evlist__create_maps(rec->evlist, &rec->opts.target) < 0)
usage_with_options(record_usage, record_options);
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index d25cb8088e8c..8153468065ac 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -2430,6 +2430,12 @@ int cmd_stat(int argc, const char **argv)
if ((stat_config.aggr_mode == AGGR_THREAD) && (target.system_wide))
target.per_thread = true;
+ if (evlist__fix_hybrid_cpus(evsel_list, target.cpu_list)) {
+ pr_err("failed to use cpu list %s\n", target.cpu_list);
+ goto out;
+ }
+
+ target.hybrid = perf_pmu__has_hybrid();
if (evlist__create_maps(evsel_list, &target) < 0) {
if (target__has_task(&target)) {
pr_err("Problems finding threads of monitor\n");
diff --git a/tools/perf/util/evlist-hybrid.c b/tools/perf/util/evlist-hybrid.c
index db3f5fbdebe1..7c554234b43d 100644
--- a/tools/perf/util/evlist-hybrid.c
+++ b/tools/perf/util/evlist-hybrid.c
@@ -86,3 +86,76 @@ bool evlist__has_hybrid(struct evlist *evlist)
return false;
}
+
+int evlist__fix_hybrid_cpus(struct evlist *evlist, const char *cpu_list)
+{
+ struct perf_cpu_map *cpus;
+ struct evsel *evsel, *tmp;
+ struct perf_pmu *pmu;
+ int ret, unmatched_count = 0, events_nr = 0;
+
+ if (!perf_pmu__has_hybrid() || !cpu_list)
+ return 0;
+
+ cpus = perf_cpu_map__new(cpu_list);
+ if (!cpus)
+ return -1;
+
+ /*
+ * The evsels are created with hybrid pmu's cpus. But now we
+ * need to check and adjust the cpus of evsel by cpu_list because
+ * cpu_list may cause conflicts with cpus of evsel. For example,
+ * cpus of evsel is cpu0-7, but the cpu_list is cpu6-8, we need
+ * to adjust the cpus of evsel to cpu6-7. And then propatate maps
+ * in evlist__create_maps().
+ */
+ evlist__for_each_entry_safe(evlist, tmp, evsel) {
+ struct perf_cpu_map *matched_cpus, *unmatched_cpus;
+ char buf1[128], buf2[128];
+
+ pmu = perf_pmu__find_hybrid_pmu(evsel->pmu_name);
+ if (!pmu)
+ continue;
+
+ ret = perf_pmu__cpus_match(pmu, cpus, &matched_cpus,
+ &unmatched_cpus);
+ if (ret)
+ goto out;
+
+ events_nr++;
+
+ if (matched_cpus->nr > 0 && (unmatched_cpus->nr > 0 ||
+ matched_cpus->nr < cpus->nr ||
+ matched_cpus->nr < pmu->cpus->nr)) {
+ perf_cpu_map__put(evsel->core.cpus);
+ perf_cpu_map__put(evsel->core.own_cpus);
+ evsel->core.cpus = perf_cpu_map__get(matched_cpus);
+ evsel->core.own_cpus = perf_cpu_map__get(matched_cpus);
+
+ if (unmatched_cpus->nr > 0) {
+ cpu_map__snprint(matched_cpus, buf1, sizeof(buf1));
+ pr_warning("WARNING: use %s in '%s' for '%s', skip other cpus in list.\n",
+ buf1, pmu->name, evsel->name);
+ }
+ }
+
+ if (matched_cpus->nr == 0) {
+ evlist__remove(evlist, evsel);
+ evsel__delete(evsel);
+
+ cpu_map__snprint(cpus, buf1, sizeof(buf1));
+ cpu_map__snprint(pmu->cpus, buf2, sizeof(buf2));
+ pr_warning("WARNING: %s isn't a '%s', please use a CPU list in the '%s' range (%s)\n",
+ buf1, pmu->name, pmu->name, buf2);
+ unmatched_count++;
+ }
+
+ perf_cpu_map__put(matched_cpus);
+ perf_cpu_map__put(unmatched_cpus);
+ }
+
+ ret = (unmatched_count == events_nr) ? -1 : 0;
+out:
+ perf_cpu_map__put(cpus);
+ return ret;
+}
diff --git a/tools/perf/util/evlist-hybrid.h b/tools/perf/util/evlist-hybrid.h
index 19f74b4c340a..aacdb1b0f948 100644
--- a/tools/perf/util/evlist-hybrid.h
+++ b/tools/perf/util/evlist-hybrid.h
@@ -10,5 +10,6 @@
int evlist__add_default_hybrid(struct evlist *evlist, bool precise);
void evlist__warn_hybrid_group(struct evlist *evlist);
bool evlist__has_hybrid(struct evlist *evlist);
+int evlist__fix_hybrid_cpus(struct evlist *evlist, const char *cpu_list);
#endif /* __PERF_EVLIST_HYBRID_H */
diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index 06f8890816c3..5f92319ce258 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -27,6 +27,7 @@
#include "util/perf_api_probe.h"
#include "util/evsel_fprintf.h"
#include "util/evlist-hybrid.h"
+#include "util/pmu.h"
#include <signal.h>
#include <unistd.h>
#include <sched.h>
diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c
index 44b90d638ad5..229cb975c5d7 100644
--- a/tools/perf/util/pmu.c
+++ b/tools/perf/util/pmu.c
@@ -1906,3 +1906,38 @@ int perf_pmu__match(char *pattern, char *name, char *tok)
return 0;
}
+
+int perf_pmu__cpus_match(struct perf_pmu *pmu, struct perf_cpu_map *cpus,
+ struct perf_cpu_map **mcpus_ptr,
+ struct perf_cpu_map **ucpus_ptr)
+{
+ struct perf_cpu_map *pmu_cpus = pmu->cpus;
+ struct perf_cpu_map *matched_cpus, *unmatched_cpus;
+ int matched_nr = 0, unmatched_nr = 0;
+
+ matched_cpus = perf_cpu_map__default_new();
+ if (!matched_cpus)
+ return -1;
+
+ unmatched_cpus = perf_cpu_map__default_new();
+ if (!unmatched_cpus) {
+ perf_cpu_map__put(matched_cpus);
+ return -1;
+ }
+
+ for (int i = 0; i < cpus->nr; i++) {
+ int cpu;
+
+ cpu = perf_cpu_map__idx(pmu_cpus, cpus->map[i]);
+ if (cpu == -1)
+ unmatched_cpus->map[unmatched_nr++] = cpus->map[i];
+ else
+ matched_cpus->map[matched_nr++] = cpus->map[i];
+ }
+
+ unmatched_cpus->nr = unmatched_nr;
+ matched_cpus->nr = matched_nr;
+ *mcpus_ptr = matched_cpus;
+ *ucpus_ptr = unmatched_cpus;
+ return 0;
+}
diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h
index 926da483a141..e05ee1906f65 100644
--- a/tools/perf/util/pmu.h
+++ b/tools/perf/util/pmu.h
@@ -11,6 +11,7 @@
#include "pmu-events/pmu-events.h"
struct evsel_config_term;
+struct perf_cpu_map;
enum {
PERF_PMU_FORMAT_VALUE_CONFIG,
@@ -135,4 +136,7 @@ void perf_pmu__warn_invalid_config(struct perf_pmu *pmu, __u64 config,
bool perf_pmu__has_hybrid(void);
int perf_pmu__match(char *pattern, char *name, char *tok);
+int perf_pmu__cpus_match(struct perf_pmu *pmu, struct perf_cpu_map *cpus,
+ struct perf_cpu_map **mcpus_ptr,
+ struct perf_cpu_map **ucpus_ptr);
#endif /* __PMU_H */
--
2.17.1
Hi Jiri,
Any comments for v4 series?
Thanks
Jin Yao
On 7/23/2021 2:34 PM, Jin Yao wrote:
> The perf-record and perf-stat have supported the option '-C/--cpus'
> to count or collect only on the list of CPUs provided. This option
> needs to be supported for hybrid as well.
>
> v4:
> ---
> Rename evlist__use_cpu_list to evlist__fix_hybrid_cpus
> Add comments to evlist__fix_hybrid_cpus.
>
> v3:
> ---
> Rebase to latest perf/core branch.
>
> v2:
> ---
> Automatically map to hybrid pmu.
>
> For example,
>
> If cpu0-7 are 'cpu_core' and cpu9-11 are 'cpu_atom',
>
> # perf stat -e cycles -C0,11 -- sleep 1
> WARNING: use 0 in 'cpu_core' for 'cycles', skip other cpus in list.
> WARNING: use 11 in 'cpu_atom' for 'cycles', skip other cpus in list.
>
> Performance counter stats for 'CPU(s) 0,11':
>
> 1,914,704 cpu_core/cycles/
> 2,036,983 cpu_atom/cycles/
>
> 1.005815641 seconds time elapsed
>
> It automatically selects cpu0 for cpu_core/cycles/, selects cpu11 for
> cpu_atom/cycles/, also with some warnings output.
>
> Jin Yao (3):
> libperf: Add perf_cpu_map__default_new()
> perf tools: Create hybrid flag in target
> perf tools: Enable on a list of CPUs for hybrid
>
> tools/lib/perf/cpumap.c | 5 ++
> tools/lib/perf/include/perf/cpumap.h | 1 +
> tools/perf/builtin-record.c | 7 +++
> tools/perf/builtin-stat.c | 6 +++
> tools/perf/util/evlist-hybrid.c | 73 ++++++++++++++++++++++++++++
> tools/perf/util/evlist-hybrid.h | 1 +
> tools/perf/util/evlist.c | 3 +-
> tools/perf/util/pmu.c | 35 +++++++++++++
> tools/perf/util/pmu.h | 4 ++
> tools/perf/util/target.h | 1 +
> 10 files changed, 135 insertions(+), 1 deletion(-)
>
On Fri, Jul 23, 2021 at 02:34:30PM +0800, Jin Yao wrote:
> The perf-record and perf-stat have supported the option '-C/--cpus'
> to count or collect only on the list of CPUs provided. This option
> needs to be supported for hybrid as well.
>
> v4:
> ---
> Rename evlist__use_cpu_list to evlist__fix_hybrid_cpus
> Add comments to evlist__fix_hybrid_cpus.
Acked-by: Jiri Olsa <[email protected]>
thanks,
jirka
>
> v3:
> ---
> Rebase to latest perf/core branch.
>
> v2:
> ---
> Automatically map to hybrid pmu.
>
> For example,
>
> If cpu0-7 are 'cpu_core' and cpu9-11 are 'cpu_atom',
>
> # perf stat -e cycles -C0,11 -- sleep 1
> WARNING: use 0 in 'cpu_core' for 'cycles', skip other cpus in list.
> WARNING: use 11 in 'cpu_atom' for 'cycles', skip other cpus in list.
>
> Performance counter stats for 'CPU(s) 0,11':
>
> 1,914,704 cpu_core/cycles/
> 2,036,983 cpu_atom/cycles/
>
> 1.005815641 seconds time elapsed
>
> It automatically selects cpu0 for cpu_core/cycles/, selects cpu11 for
> cpu_atom/cycles/, also with some warnings output.
>
> Jin Yao (3):
> libperf: Add perf_cpu_map__default_new()
> perf tools: Create hybrid flag in target
> perf tools: Enable on a list of CPUs for hybrid
>
> tools/lib/perf/cpumap.c | 5 ++
> tools/lib/perf/include/perf/cpumap.h | 1 +
> tools/perf/builtin-record.c | 7 +++
> tools/perf/builtin-stat.c | 6 +++
> tools/perf/util/evlist-hybrid.c | 73 ++++++++++++++++++++++++++++
> tools/perf/util/evlist-hybrid.h | 1 +
> tools/perf/util/evlist.c | 3 +-
> tools/perf/util/pmu.c | 35 +++++++++++++
> tools/perf/util/pmu.h | 4 ++
> tools/perf/util/target.h | 1 +
> 10 files changed, 135 insertions(+), 1 deletion(-)
>
> --
> 2.17.1
>
Em Wed, Aug 11, 2021 at 09:01:00PM +0200, Jiri Olsa escreveu:
> On Fri, Jul 23, 2021 at 02:34:30PM +0800, Jin Yao wrote:
> > The perf-record and perf-stat have supported the option '-C/--cpus'
> > to count or collect only on the list of CPUs provided. This option
> > needs to be supported for hybrid as well.
> >
> > v4:
> > ---
> > Rename evlist__use_cpu_list to evlist__fix_hybrid_cpus
> > Add comments to evlist__fix_hybrid_cpus.
>
> Acked-by: Jiri Olsa <[email protected]>
Thanks, applied.
- Arnaldo