[RFC PATCH v2 0/7] libperf: Add interface for overflow check of sampling events

[RFC PATCH v2 0/7] libperf: Add interface for overflow check of sampling events

[Shunsuke Nakamura]

This patch series adds interface for overflow check of sampling events
to libperf.

First patch fixes a typo in the error message that I noticed.

Second patch move 'open_flags' from tools/perf to evsel::open_flags.

Third patch adds a interface to perform IOC_REFRESH and IOC_PERIOD.

Fourth patch introduce perf_{evsel, evlist}__open_opt() with extensible
structure opts.

Fifth patch adds a interface to check overflowed events.

Sixth and seventh patch adds tests.

---
Previous version at:
https://lore.kernel.org/linux-perf-users/20220225103114.144239-1-nakamura.shun@fujitsu.com/

Changes in v2:
 - Delete perf_evsel__set_close_on_exec() function
 - Introduce perf_{evsel, evlist}__open_opt() with extensible structure
   opts
 - Fix perf_evsel__set_signal() to a internal function
 - Add bool type argument to perf_evsel__check_{fd, fd_cpu}() to indicate
   overflow results


Shunsuke Nakamura (7):
  libperf tests: Fix typo in the error message
  libperf: Move 'open_flags' from tools/perf to evsel::open_flags
  libperf: Add perf_evsel__{refresh, period}() functions
  libperf: Introduce perf_{evsel, evlist}__open_opt with extensible
    struct opts
  libperf: Add perf_evsel__check_overflow() functions
  libperf test: Add test_stat_overflow()
  libperf test: Add test_stat_overflow_event()

 tools/lib/perf/Documentation/libperf.txt |  22 +++
 tools/lib/perf/evlist.c                  |  20 +++
 tools/lib/perf/evsel.c                   | 210 ++++++++++++++++++++++-
 tools/lib/perf/include/internal/evsel.h  |   2 +
 tools/lib/perf/include/perf/evlist.h     |   3 +
 tools/lib/perf/include/perf/evsel.h      |  35 ++++
 tools/lib/perf/internal.h                |  44 +++++
 tools/lib/perf/libperf.map               |   8 +
 tools/lib/perf/tests/test-evlist.c       | 135 ++++++++++++++-
 tools/lib/perf/tests/test-evsel.c        | 111 ++++++++++++
 tools/perf/util/evsel.c                  |  16 +-
 tools/perf/util/evsel.h                  |   1 -
 12 files changed, 588 insertions(+), 19 deletions(-)

-- 
2.25.1

[RFC PATCH v2 1/7] libperf tests: Fix typo in the error message

[Shunsuke Nakamura]

This patch corrects a typo in the error message.

Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
---
 tools/lib/perf/tests/test-evlist.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/tools/lib/perf/tests/test-evlist.c b/tools/lib/perf/tests/test-evlist.c
index fa854c83b7e7..ed616fc19b4f 100644
--- a/tools/lib/perf/tests/test-evlist.c
+++ b/tools/lib/perf/tests/test-evlist.c
@@ -69,7 +69,7 @@ static int test_stat_cpu(void)
 	perf_evlist__set_maps(evlist, cpus, NULL);
 
 	err = perf_evlist__open(evlist);
-	__T("failed to open evsel", err == 0);
+	__T("failed to open evlist", err == 0);
 
 	perf_evlist__for_each_evsel(evlist, evsel) {
 		cpus = perf_evsel__cpus(evsel);
@@ -130,7 +130,7 @@ static int test_stat_thread(void)
 	perf_evlist__set_maps(evlist, NULL, threads);
 
 	err = perf_evlist__open(evlist);
-	__T("failed to open evsel", err == 0);
+	__T("failed to open evlist", err == 0);
 
 	perf_evlist__for_each_evsel(evlist, evsel) {
 		perf_evsel__read(evsel, 0, 0, &counts);
@@ -187,7 +187,7 @@ static int test_stat_thread_enable(void)
 	perf_evlist__set_maps(evlist, NULL, threads);
 
 	err = perf_evlist__open(evlist);
-	__T("failed to open evsel", err == 0);
+	__T("failed to open evlist", err == 0);
 
 	perf_evlist__for_each_evsel(evlist, evsel) {
 		perf_evsel__read(evsel, 0, 0, &counts);
@@ -507,7 +507,7 @@ static int test_stat_multiplexing(void)
 	perf_evlist__set_maps(evlist, NULL, threads);
 
 	err = perf_evlist__open(evlist);
-	__T("failed to open evsel", err == 0);
+	__T("failed to open evlist", err == 0);
 
 	perf_evlist__enable(evlist);
 
-- 
2.25.1

Re: [RFC PATCH v2 1/7] libperf tests: Fix typo in the error message

[Arnaldo Carvalho de Melo]

Em Fri, Mar 25, 2022 at 01:38:23PM +0900, Shunsuke Nakamura escreveu:
> This patch corrects a typo in the error message.

Thanks, applied. Will wait for review for the other patches in this
series.

- Arnaldo

 
> Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
> ---
>  tools/lib/perf/tests/test-evlist.c | 8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/tools/lib/perf/tests/test-evlist.c b/tools/lib/perf/tests/test-evlist.c
> index fa854c83b7e7..ed616fc19b4f 100644
> --- a/tools/lib/perf/tests/test-evlist.c
> +++ b/tools/lib/perf/tests/test-evlist.c
> @@ -69,7 +69,7 @@ static int test_stat_cpu(void)
>  	perf_evlist__set_maps(evlist, cpus, NULL);
>  
>  	err = perf_evlist__open(evlist);
> -	__T("failed to open evsel", err == 0);
> +	__T("failed to open evlist", err == 0);
>  
>  	perf_evlist__for_each_evsel(evlist, evsel) {
>  		cpus = perf_evsel__cpus(evsel);
> @@ -130,7 +130,7 @@ static int test_stat_thread(void)
>  	perf_evlist__set_maps(evlist, NULL, threads);
>  
>  	err = perf_evlist__open(evlist);
> -	__T("failed to open evsel", err == 0);
> +	__T("failed to open evlist", err == 0);
>  
>  	perf_evlist__for_each_evsel(evlist, evsel) {
>  		perf_evsel__read(evsel, 0, 0, &counts);
> @@ -187,7 +187,7 @@ static int test_stat_thread_enable(void)
>  	perf_evlist__set_maps(evlist, NULL, threads);
>  
>  	err = perf_evlist__open(evlist);
> -	__T("failed to open evsel", err == 0);
> +	__T("failed to open evlist", err == 0);
>  
>  	perf_evlist__for_each_evsel(evlist, evsel) {
>  		perf_evsel__read(evsel, 0, 0, &counts);
> @@ -507,7 +507,7 @@ static int test_stat_multiplexing(void)
>  	perf_evlist__set_maps(evlist, NULL, threads);
>  
>  	err = perf_evlist__open(evlist);
> -	__T("failed to open evsel", err == 0);
> +	__T("failed to open evlist", err == 0);
>  
>  	perf_evlist__enable(evlist);
>  
> -- 
> 2.25.1

-- 

- Arnaldo

[RFC PATCH v2 2/7] libperf: Move 'open_flags' from tools/perf to evsel::open_flags

[Shunsuke Nakamura]

Move evsel::open_flags to perf_evsel::open_flags, so we can move
the open_flags interface to libperf.

Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
---
 tools/lib/perf/evsel.c                  |  3 ++-
 tools/lib/perf/include/internal/evsel.h |  2 ++
 tools/perf/util/evsel.c                 | 16 +++++++++-------
 tools/perf/util/evsel.h                 |  1 -
 4 files changed, 13 insertions(+), 9 deletions(-)

diff --git a/tools/lib/perf/evsel.c b/tools/lib/perf/evsel.c
index 210ea7c06ce8..6640a333e6d9 100644
--- a/tools/lib/perf/evsel.c
+++ b/tools/lib/perf/evsel.c
@@ -26,6 +26,7 @@ void perf_evsel__init(struct perf_evsel *evsel, struct perf_event_attr *attr,
 	evsel->attr = *attr;
 	evsel->idx  = idx;
 	evsel->leader = evsel;
+	evsel->open_flags = 0;
 }
 
 struct perf_evsel *perf_evsel__new(struct perf_event_attr *attr)
@@ -158,7 +159,7 @@ int perf_evsel__open(struct perf_evsel *evsel, struct perf_cpu_map *cpus,
 
 			fd = sys_perf_event_open(&evsel->attr,
 						 threads->map[thread].pid,
-						 cpu, group_fd, 0);
+						 cpu, group_fd, evsel->open_flags);
 
 			if (fd < 0)
 				return -errno;
diff --git a/tools/lib/perf/include/internal/evsel.h b/tools/lib/perf/include/internal/evsel.h
index cfc9ebd7968e..37a99cf261b3 100644
--- a/tools/lib/perf/include/internal/evsel.h
+++ b/tools/lib/perf/include/internal/evsel.h
@@ -51,6 +51,8 @@ struct perf_evsel {
 	int			 nr_members;
 	bool			 system_wide;
 	int			 idx;
+
+	unsigned long		 open_flags;
 };
 
 void perf_evsel__init(struct perf_evsel *evsel, struct perf_event_attr *attr,
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 22d3267ce294..52c34552ee76 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -1794,9 +1794,9 @@ static int __evsel__prepare_open(struct evsel *evsel, struct perf_cpu_map *cpus,
 	    perf_evsel__alloc_fd(&evsel->core, perf_cpu_map__nr(cpus), nthreads) < 0)
 		return -ENOMEM;
 
-	evsel->open_flags = PERF_FLAG_FD_CLOEXEC;
+	evsel->core.open_flags = PERF_FLAG_FD_CLOEXEC;
 	if (evsel->cgrp)
-		evsel->open_flags |= PERF_FLAG_PID_CGROUP;
+		evsel->core.open_flags |= PERF_FLAG_PID_CGROUP;
 
 	return 0;
 }
@@ -1814,7 +1814,7 @@ static void evsel__disable_missing_features(struct evsel *evsel)
 		evsel->core.attr.clockid = 0;
 	}
 	if (perf_missing_features.cloexec)
-		evsel->open_flags &= ~(unsigned long)PERF_FLAG_FD_CLOEXEC;
+		evsel->core.open_flags &= ~(unsigned long)PERF_FLAG_FD_CLOEXEC;
 	if (perf_missing_features.mmap2)
 		evsel->core.attr.mmap2 = 0;
 	if (evsel->pmu && evsel->pmu->missing_features.exclude_guest)
@@ -1902,7 +1902,8 @@ bool evsel__detect_missing_features(struct evsel *evsel)
 		perf_missing_features.clockid = true;
 		pr_debug2_peo("switching off use_clockid\n");
 		return true;
-	} else if (!perf_missing_features.cloexec && (evsel->open_flags & PERF_FLAG_FD_CLOEXEC)) {
+	} else if (!perf_missing_features.cloexec &&
+		   (evsel->core.open_flags & PERF_FLAG_FD_CLOEXEC)) {
 		perf_missing_features.cloexec = true;
 		pr_debug2_peo("switching off cloexec flag\n");
 		return true;
@@ -2029,11 +2030,12 @@ static int evsel__open_cpu(struct evsel *evsel, struct perf_cpu_map *cpus,
 			test_attr__ready();
 
 			pr_debug2_peo("sys_perf_event_open: pid %d  cpu %d  group_fd %d  flags %#lx",
-				pid, perf_cpu_map__cpu(cpus, idx).cpu, group_fd, evsel->open_flags);
+				pid, perf_cpu_map__cpu(cpus, idx).cpu, group_fd,
+				evsel->core.open_flags);
 
 			fd = sys_perf_event_open(&evsel->core.attr, pid,
 						perf_cpu_map__cpu(cpus, idx).cpu,
-						group_fd, evsel->open_flags);
+						group_fd, evsel->core.open_flags);
 
 			FD(evsel, idx, thread) = fd;
 
@@ -2050,7 +2052,7 @@ static int evsel__open_cpu(struct evsel *evsel, struct perf_cpu_map *cpus,
 			if (unlikely(test_attr__enabled)) {
 				test_attr__open(&evsel->core.attr, pid,
 						perf_cpu_map__cpu(cpus, idx),
-						fd, group_fd, evsel->open_flags);
+						fd, group_fd, evsel->core.open_flags);
 			}
 
 			pr_debug2_peo(" = %d\n", fd);
diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
index 041b42d33bf5..8a545954eec7 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -152,7 +152,6 @@ struct evsel {
 		struct bperf_leader_bpf *leader_skel;
 		struct bperf_follower_bpf *follower_skel;
 	};
-	unsigned long		open_flags;
 	int			precise_ip_original;
 
 	/* for missing_features */
-- 
2.25.1

Re: [RFC PATCH v2 2/7] libperf: Move 'open_flags' from tools/perf to evsel::open_flags

[Jiri Olsa]

On Fri, Mar 25, 2022 at 01:38:24PM +0900, Shunsuke Nakamura wrote:
> Move evsel::open_flags to perf_evsel::open_flags, so we can move
> the open_flags interface to libperf.
> 
> Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
> ---
>  tools/lib/perf/evsel.c                  |  3 ++-
>  tools/lib/perf/include/internal/evsel.h |  2 ++
>  tools/perf/util/evsel.c                 | 16 +++++++++-------
>  tools/perf/util/evsel.h                 |  1 -
>  4 files changed, 13 insertions(+), 9 deletions(-)
> 
> diff --git a/tools/lib/perf/evsel.c b/tools/lib/perf/evsel.c
> index 210ea7c06ce8..6640a333e6d9 100644
> --- a/tools/lib/perf/evsel.c
> +++ b/tools/lib/perf/evsel.c
> @@ -26,6 +26,7 @@ void perf_evsel__init(struct perf_evsel *evsel, struct perf_event_attr *attr,
>  	evsel->attr = *attr;
>  	evsel->idx  = idx;
>  	evsel->leader = evsel;
> +	evsel->open_flags = 0;
>  }
>  
>  struct perf_evsel *perf_evsel__new(struct perf_event_attr *attr)
> @@ -158,7 +159,7 @@ int perf_evsel__open(struct perf_evsel *evsel, struct perf_cpu_map *cpus,
>  
>  			fd = sys_perf_event_open(&evsel->attr,
>  						 threads->map[thread].pid,
> -						 cpu, group_fd, 0);
> +						 cpu, group_fd, evsel->open_flags);

please keep this change just open_flags field move,
this change should go to later patch

jirka

>  
>  			if (fd < 0)
>  				return -errno;
> diff --git a/tools/lib/perf/include/internal/evsel.h b/tools/lib/perf/include/internal/evsel.h
> index cfc9ebd7968e..37a99cf261b3 100644
> --- a/tools/lib/perf/include/internal/evsel.h
> +++ b/tools/lib/perf/include/internal/evsel.h
> @@ -51,6 +51,8 @@ struct perf_evsel {
>  	int			 nr_members;
>  	bool			 system_wide;
>  	int			 idx;
> +
> +	unsigned long		 open_flags;
>  };
>  
>  void perf_evsel__init(struct perf_evsel *evsel, struct perf_event_attr *attr,
> diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
> index 22d3267ce294..52c34552ee76 100644
> --- a/tools/perf/util/evsel.c
> +++ b/tools/perf/util/evsel.c
> @@ -1794,9 +1794,9 @@ static int __evsel__prepare_open(struct evsel *evsel, struct perf_cpu_map *cpus,
>  	    perf_evsel__alloc_fd(&evsel->core, perf_cpu_map__nr(cpus), nthreads) < 0)
>  		return -ENOMEM;
>  
> -	evsel->open_flags = PERF_FLAG_FD_CLOEXEC;
> +	evsel->core.open_flags = PERF_FLAG_FD_CLOEXEC;
>  	if (evsel->cgrp)
> -		evsel->open_flags |= PERF_FLAG_PID_CGROUP;
> +		evsel->core.open_flags |= PERF_FLAG_PID_CGROUP;
>  
>  	return 0;
>  }
> @@ -1814,7 +1814,7 @@ static void evsel__disable_missing_features(struct evsel *evsel)
>  		evsel->core.attr.clockid = 0;
>  	}
>  	if (perf_missing_features.cloexec)
> -		evsel->open_flags &= ~(unsigned long)PERF_FLAG_FD_CLOEXEC;
> +		evsel->core.open_flags &= ~(unsigned long)PERF_FLAG_FD_CLOEXEC;
>  	if (perf_missing_features.mmap2)
>  		evsel->core.attr.mmap2 = 0;
>  	if (evsel->pmu && evsel->pmu->missing_features.exclude_guest)
> @@ -1902,7 +1902,8 @@ bool evsel__detect_missing_features(struct evsel *evsel)
>  		perf_missing_features.clockid = true;
>  		pr_debug2_peo("switching off use_clockid\n");
>  		return true;
> -	} else if (!perf_missing_features.cloexec && (evsel->open_flags & PERF_FLAG_FD_CLOEXEC)) {
> +	} else if (!perf_missing_features.cloexec &&
> +		   (evsel->core.open_flags & PERF_FLAG_FD_CLOEXEC)) {
>  		perf_missing_features.cloexec = true;
>  		pr_debug2_peo("switching off cloexec flag\n");
>  		return true;
> @@ -2029,11 +2030,12 @@ static int evsel__open_cpu(struct evsel *evsel, struct perf_cpu_map *cpus,
>  			test_attr__ready();
>  
>  			pr_debug2_peo("sys_perf_event_open: pid %d  cpu %d  group_fd %d  flags %#lx",
> -				pid, perf_cpu_map__cpu(cpus, idx).cpu, group_fd, evsel->open_flags);
> +				pid, perf_cpu_map__cpu(cpus, idx).cpu, group_fd,
> +				evsel->core.open_flags);
>  
>  			fd = sys_perf_event_open(&evsel->core.attr, pid,
>  						perf_cpu_map__cpu(cpus, idx).cpu,
> -						group_fd, evsel->open_flags);
> +						group_fd, evsel->core.open_flags);
>  
>  			FD(evsel, idx, thread) = fd;
>  
> @@ -2050,7 +2052,7 @@ static int evsel__open_cpu(struct evsel *evsel, struct perf_cpu_map *cpus,
>  			if (unlikely(test_attr__enabled)) {
>  				test_attr__open(&evsel->core.attr, pid,
>  						perf_cpu_map__cpu(cpus, idx),
> -						fd, group_fd, evsel->open_flags);
> +						fd, group_fd, evsel->core.open_flags);
>  			}
>  
>  			pr_debug2_peo(" = %d\n", fd);
> diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
> index 041b42d33bf5..8a545954eec7 100644
> --- a/tools/perf/util/evsel.h
> +++ b/tools/perf/util/evsel.h
> @@ -152,7 +152,6 @@ struct evsel {
>  		struct bperf_leader_bpf *leader_skel;
>  		struct bperf_follower_bpf *follower_skel;
>  	};
> -	unsigned long		open_flags;
>  	int			precise_ip_original;
>  
>  	/* for missing_features */
> -- 
> 2.25.1
> 

RE: [RFC PATCH v2 2/7] libperf: Move 'open_flags' from tools/perf to evsel::open_flags

[nakamura.shun]

Hi jirka

> >  struct perf_evsel *perf_evsel__new(struct perf_event_attr *attr) @@
> > -158,7 +159,7 @@ int perf_evsel__open(struct perf_evsel *evsel, struct
> > perf_cpu_map *cpus,
> >
> >  			fd = sys_perf_event_open(&evsel->attr,
> >  						 threads->map[thread].pid,
> > -						 cpu, group_fd, 0);
> > +						 cpu, group_fd,
> evsel->open_flags);
> 
> please keep this change just open_flags field move, this change should go to
> later patch
> 
I'll fix it.


Best Regards
Shunsuke

[RFC PATCH v2 3/7] libperf: Add perf_evsel__{refresh, period}() functions

[Shunsuke Nakamura]

Add the following functions:

  perf_evsel__refresh()
  perf_evsel__refresh_cpu()
  perf_evsel__period()
  perf_evsel__period_cpu()

to set the over flow limit and period.

Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
---
 tools/lib/perf/Documentation/libperf.txt |  5 ++
 tools/lib/perf/evsel.c                   | 68 +++++++++++++++++++++---
 tools/lib/perf/include/perf/evsel.h      |  5 ++
 tools/lib/perf/libperf.map               |  4 ++
 4 files changed, 76 insertions(+), 6 deletions(-)

diff --git a/tools/lib/perf/Documentation/libperf.txt b/tools/lib/perf/Documentation/libperf.txt
index 32c5051c24eb..ae55e62fc4ce 100644
--- a/tools/lib/perf/Documentation/libperf.txt
+++ b/tools/lib/perf/Documentation/libperf.txt
@@ -146,6 +146,11 @@ SYNOPSIS
   int perf_evsel__enable_cpu(struct perf_evsel *evsel, int cpu_map_idx);
   int perf_evsel__disable(struct perf_evsel *evsel);
   int perf_evsel__disable_cpu(struct perf_evsel *evsel, int cpu_map_idx);
+  int perf_evsel__refresh(struct perf_evsel *evsel, int refresh);
+  int perf_evsel__refresh_cpu(struct perf_evsel *evsel, int refresh,
+                              int cpu_map_idx);
+  int perf_evsel__period(struct perf_evsel *evsel, int period);
+  int perf_evsel__period_cpu(struct perf_evsel *evsel, int period, int cpu_map_idx);
   struct perf_cpu_map *perf_evsel__cpus(struct perf_evsel *evsel);
   struct perf_thread_map *perf_evsel__threads(struct perf_evsel *evsel);
   struct perf_event_attr *perf_evsel__attr(struct perf_evsel *evsel);
diff --git a/tools/lib/perf/evsel.c b/tools/lib/perf/evsel.c
index 6640a333e6d9..b17842581dea 100644
--- a/tools/lib/perf/evsel.c
+++ b/tools/lib/perf/evsel.c
@@ -330,7 +330,7 @@ int perf_evsel__read(struct perf_evsel *evsel, int cpu_map_idx, int thread,
 }
 
 static int perf_evsel__run_ioctl(struct perf_evsel *evsel,
-				 int ioc,  void *arg,
+				 int ioc, unsigned long arg,
 				 int cpu_map_idx)
 {
 	int thread;
@@ -353,7 +353,7 @@ static int perf_evsel__run_ioctl(struct perf_evsel *evsel,
 
 int perf_evsel__enable_cpu(struct perf_evsel *evsel, int cpu_map_idx)
 {
-	return perf_evsel__run_ioctl(evsel, PERF_EVENT_IOC_ENABLE, NULL, cpu_map_idx);
+	return perf_evsel__run_ioctl(evsel, PERF_EVENT_IOC_ENABLE, 0, cpu_map_idx);
 }
 
 int perf_evsel__enable(struct perf_evsel *evsel)
@@ -362,13 +362,13 @@ int perf_evsel__enable(struct perf_evsel *evsel)
 	int err = 0;
 
 	for (i = 0; i < xyarray__max_x(evsel->fd) && !err; i++)
-		err = perf_evsel__run_ioctl(evsel, PERF_EVENT_IOC_ENABLE, NULL, i);
+		err = perf_evsel__run_ioctl(evsel, PERF_EVENT_IOC_ENABLE, 0, i);
 	return err;
 }
 
 int perf_evsel__disable_cpu(struct perf_evsel *evsel, int cpu_map_idx)
 {
-	return perf_evsel__run_ioctl(evsel, PERF_EVENT_IOC_DISABLE, NULL, cpu_map_idx);
+	return perf_evsel__run_ioctl(evsel, PERF_EVENT_IOC_DISABLE, 0, cpu_map_idx);
 }
 
 int perf_evsel__disable(struct perf_evsel *evsel)
@@ -377,7 +377,63 @@ int perf_evsel__disable(struct perf_evsel *evsel)
 	int err = 0;
 
 	for (i = 0; i < xyarray__max_x(evsel->fd) && !err; i++)
-		err = perf_evsel__run_ioctl(evsel, PERF_EVENT_IOC_DISABLE, NULL, i);
+		err = perf_evsel__run_ioctl(evsel, PERF_EVENT_IOC_DISABLE, 0, i);
+	return err;
+}
+
+int perf_evsel__refresh_cpu(struct perf_evsel *evsel, int refresh, int cpu_map_idx)
+{
+	return perf_evsel__run_ioctl(evsel, PERF_EVENT_IOC_REFRESH, refresh, cpu_map_idx);
+}
+
+int perf_evsel__refresh(struct perf_evsel *evsel, int refresh)
+{
+	int i;
+	int err = 0;
+
+	for (i = 0; i < xyarray__max_x(evsel->fd) && !err; i++)
+		err = perf_evsel__run_ioctl(evsel, PERF_EVENT_IOC_REFRESH, refresh, i);
+	return err;
+}
+
+int perf_evsel__period_cpu(struct perf_evsel *evsel, __u64 period, int cpu_map_idx)
+{
+	struct perf_event_attr *attr;
+	int err = 0;
+
+	attr = perf_evsel__attr(evsel);
+	if (!attr)
+		return -EINVAL;
+
+	err = perf_evsel__run_ioctl(evsel, PERF_EVENT_IOC_PERIOD,
+				    (unsigned long)&period, cpu_map_idx);
+	if (err)
+		return err;
+
+	attr->sample_period = period;
+
+	return err;
+}
+
+int perf_evsel__period(struct perf_evsel *evsel, __u64 period)
+{
+	struct perf_event_attr *attr;
+	int i;
+	int err = 0;
+
+	attr = perf_evsel__attr(evsel);
+	if (!attr)
+		return -EINVAL;
+
+	for (i = 0; i < xyarray__max_x(evsel->fd); i++) {
+		err = perf_evsel__run_ioctl(evsel, PERF_EVENT_IOC_PERIOD,
+					    (unsigned long)&period, i);
+		if (err)
+			return err;
+	}
+
+	attr->sample_period = period;
+
 	return err;
 }
 
@@ -388,7 +444,7 @@ int perf_evsel__apply_filter(struct perf_evsel *evsel, const char *filter)
 	for (i = 0; i < perf_cpu_map__nr(evsel->cpus) && !err; i++)
 		err = perf_evsel__run_ioctl(evsel,
 				     PERF_EVENT_IOC_SET_FILTER,
-				     (void *)filter, i);
+				     (unsigned long)filter, i);
 	return err;
 }
 
diff --git a/tools/lib/perf/include/perf/evsel.h b/tools/lib/perf/include/perf/evsel.h
index 2a9516b42d15..360ced734add 100644
--- a/tools/lib/perf/include/perf/evsel.h
+++ b/tools/lib/perf/include/perf/evsel.h
@@ -38,6 +38,11 @@ LIBPERF_API int perf_evsel__enable(struct perf_evsel *evsel);
 LIBPERF_API int perf_evsel__enable_cpu(struct perf_evsel *evsel, int cpu_map_idx);
 LIBPERF_API int perf_evsel__disable(struct perf_evsel *evsel);
 LIBPERF_API int perf_evsel__disable_cpu(struct perf_evsel *evsel, int cpu_map_idx);
+LIBPERF_API int perf_evsel__refresh(struct perf_evsel *evsel, int refresh);
+LIBPERF_API int perf_evsel__refresh_cpu(struct perf_evsel *evsel, int refresh,
+					int cpu_map_idx);
+LIBPERF_API int perf_evsel__period(struct perf_evsel *evsel, __u64 period);
+LIBPERF_API int perf_evsel__period_cpu(struct perf_evsel *evsel, __u64 period, int cpu_map_idx);
 LIBPERF_API struct perf_cpu_map *perf_evsel__cpus(struct perf_evsel *evsel);
 LIBPERF_API struct perf_thread_map *perf_evsel__threads(struct perf_evsel *evsel);
 LIBPERF_API struct perf_event_attr *perf_evsel__attr(struct perf_evsel *evsel);
diff --git a/tools/lib/perf/libperf.map b/tools/lib/perf/libperf.map
index 6fa0d651576b..ab0f44e9bb57 100644
--- a/tools/lib/perf/libperf.map
+++ b/tools/lib/perf/libperf.map
@@ -29,6 +29,10 @@ LIBPERF_0.0.1 {
 		perf_evsel__munmap;
 		perf_evsel__mmap_base;
 		perf_evsel__read;
+		perf_evsel__refresh;
+		perf_evsel__refresh_cpu;
+		perf_evsel__period;
+		perf_evsel__period_cpu;
 		perf_evsel__cpus;
 		perf_evsel__threads;
 		perf_evsel__attr;
-- 
2.25.1

Re: [RFC PATCH v2 3/7] libperf: Add perf_evsel__{refresh, period}() functions

[Jiri Olsa]

On Fri, Mar 25, 2022 at 01:38:25PM +0900, Shunsuke Nakamura wrote:

SNIP

> +int perf_evsel__refresh_cpu(struct perf_evsel *evsel, int refresh, int cpu_map_idx)
> +{
> +	return perf_evsel__run_ioctl(evsel, PERF_EVENT_IOC_REFRESH, refresh, cpu_map_idx);
> +}
> +
> +int perf_evsel__refresh(struct perf_evsel *evsel, int refresh)
> +{
> +	int i;
> +	int err = 0;
> +
> +	for (i = 0; i < xyarray__max_x(evsel->fd) && !err; i++)
> +		err = perf_evsel__run_ioctl(evsel, PERF_EVENT_IOC_REFRESH, refresh, i);
> +	return err;
> +}
> +
> +int perf_evsel__period_cpu(struct perf_evsel *evsel, __u64 period, int cpu_map_idx)
> +{
> +	struct perf_event_attr *attr;
> +	int err = 0;
> +
> +	attr = perf_evsel__attr(evsel);
> +	if (!attr)
> +		return -EINVAL;
> +
> +	err = perf_evsel__run_ioctl(evsel, PERF_EVENT_IOC_PERIOD,
> +				    (unsigned long)&period, cpu_map_idx);
> +	if (err)
> +		return err;
> +
> +	attr->sample_period = period;

what's the point in updating attr? this will be used after
event is created right?

> +
> +	return err;
> +}
> +
> +int perf_evsel__period(struct perf_evsel *evsel, __u64 period)
> +{
> +	struct perf_event_attr *attr;
> +	int i;
> +	int err = 0;
> +
> +	attr = perf_evsel__attr(evsel);
> +	if (!attr)
> +		return -EINVAL;
> +
> +	for (i = 0; i < xyarray__max_x(evsel->fd); i++) {
> +		err = perf_evsel__run_ioctl(evsel, PERF_EVENT_IOC_PERIOD,
> +					    (unsigned long)&period, i);
> +		if (err)
> +			return err;
> +	}
> +
> +	attr->sample_period = period;

same as above

thanks,
jirka

> +
>  	return err;
>  }
>  

SNIP

RE: [RFC PATCH v2 3/7] libperf: Add perf_evsel__{refresh, period}() functions

[nakamura.shun]

Hi jirka

> On Fri, Mar 25, 2022 at 01:38:25PM +0900, Shunsuke Nakamura wrote:
> 
> SNIP
> 
> > +int perf_evsel__refresh_cpu(struct perf_evsel *evsel, int refresh,
> > +int cpu_map_idx) {
> > +	return perf_evsel__run_ioctl(evsel, PERF_EVENT_IOC_REFRESH,
> refresh,
> > +cpu_map_idx); }
> > +
> > +int perf_evsel__refresh(struct perf_evsel *evsel, int refresh) {
> > +	int i;
> > +	int err = 0;
> > +
> > +	for (i = 0; i < xyarray__max_x(evsel->fd) && !err; i++)
> > +		err = perf_evsel__run_ioctl(evsel, PERF_EVENT_IOC_REFRESH,
> refresh, i);
> > +	return err;
> > +}
> > +
> > +int perf_evsel__period_cpu(struct perf_evsel *evsel, __u64 period,
> > +int cpu_map_idx) {
> > +	struct perf_event_attr *attr;
> > +	int err = 0;
> > +
> > +	attr = perf_evsel__attr(evsel);
> > +	if (!attr)
> > +		return -EINVAL;
> > +
> > +	err = perf_evsel__run_ioctl(evsel, PERF_EVENT_IOC_PERIOD,
> > +				    (unsigned long)&period, cpu_map_idx);
> > +	if (err)
> > +		return err;
> > +
> > +	attr->sample_period = period;
> 
> what's the point in updating attr? this will be used after event is created right?
> 
This is to maintain consistency between evsel->attr->sample_period and the sample_period of the created event.
Users can reference evsel->attr->sample_period using perf_evsel__attr().
So I thought it was necessary to update the sample_period of the event and the sample_period of evsel to avoid discrepancies.


Best Regards
Shunsuke

[RFC PATCH v2 4/7] libperf: Introduce perf_{evsel, evlist}__open_opt with extensible struct opts

[Shunsuke Nakamura]

Introduce perf_{evsel, evlist}__open_opt with extensible structure opts.
The mechanism of the extensible structure opts imitates
tools/lib/bpf/libbpf.h. This allows the user to set the open_flags
specified in perf_event_open and a signal handler to receive overflow
notifications for sampling events.

Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
---
 tools/lib/perf/Documentation/libperf.txt |  14 +++
 tools/lib/perf/evlist.c                  |  20 +++++
 tools/lib/perf/evsel.c                   | 105 +++++++++++++++++++++++
 tools/lib/perf/include/perf/evlist.h     |   3 +
 tools/lib/perf/include/perf/evsel.h      |  26 ++++++
 tools/lib/perf/internal.h                |  44 ++++++++++
 tools/lib/perf/libperf.map               |   2 +
 7 files changed, 214 insertions(+)

diff --git a/tools/lib/perf/Documentation/libperf.txt b/tools/lib/perf/Documentation/libperf.txt
index ae55e62fc4ce..700c1ce15159 100644
--- a/tools/lib/perf/Documentation/libperf.txt
+++ b/tools/lib/perf/Documentation/libperf.txt
@@ -131,6 +131,20 @@ SYNOPSIS
           };
   };
 
+  struct perf_evsel_open_opts {
+          /* size of this struct, for forward/backward compatibility */
+          size_t sz;
+
+          unsigned long open_flags;       /* perf_event_open flags */
+          int flags;                      /* fcntl flags */
+          unsigned int signal;
+          int owner_type;
+          struct sigaction *sig;
+  };
+  #define perf_evsel_open_opts__last_field sig
+
+  #define LIBPERF_OPTS(TYPE, NAME, ...)
+
   struct perf_evsel *perf_evsel__new(struct perf_event_attr *attr);
   void perf_evsel__delete(struct perf_evsel *evsel);
   int perf_evsel__open(struct perf_evsel *evsel, struct perf_cpu_map *cpus,
diff --git a/tools/lib/perf/evlist.c b/tools/lib/perf/evlist.c
index 9a770bfdc804..6f136905d249 100644
--- a/tools/lib/perf/evlist.c
+++ b/tools/lib/perf/evlist.c
@@ -666,3 +666,23 @@ void perf_evlist__set_leader(struct perf_evlist *evlist)
 		__perf_evlist__set_leader(&evlist->entries, first);
 	}
 }
+
+int perf_evlist__open_opts(struct perf_evlist *evlist,
+			   struct perf_evsel_open_opts *opts)
+{
+	struct perf_evsel *evsel;
+	int err;
+
+	perf_evlist__for_each_entry(evlist, evsel) {
+		err = perf_evsel__open_opts(evsel, evsel->cpus,
+					    evsel->threads, opts);
+		if (err < 0)
+			goto out_err;
+	}
+
+	return 0;
+
+out_err:
+	perf_evlist__close(evlist);
+	return err;
+}
diff --git a/tools/lib/perf/evsel.c b/tools/lib/perf/evsel.c
index b17842581dea..db9b7274feb5 100644
--- a/tools/lib/perf/evsel.c
+++ b/tools/lib/perf/evsel.c
@@ -1,4 +1,7 @@
 // SPDX-License-Identifier: GPL-2.0
+#ifndef _GNU_SOURCE
+#define _GNU_SOURCE
+#endif
 #include <errno.h>
 #include <unistd.h>
 #include <sys/syscall.h>
@@ -16,8 +19,12 @@
 #include <internal/lib.h>
 #include <linux/string.h>
 #include <sys/ioctl.h>
+#include <signal.h>
+#include <fcntl.h>
+#include <sys/types.h>
 #include <sys/mman.h>
 #include <asm/bug.h>
+#include "internal.h"
 
 void perf_evsel__init(struct perf_evsel *evsel, struct perf_event_attr *attr,
 		      int idx)
@@ -511,3 +518,101 @@ void perf_counts_values__scale(struct perf_counts_values *count,
 	if (pscaled)
 		*pscaled = scaled;
 }
+
+static int perf_evsel__run_fcntl(struct perf_evsel *evsel,
+				 unsigned int cmd, unsigned long arg,
+				 int cpu_map_idx)
+{
+	int thread;
+
+	for (thread = 0; thread < xyarray__max_y(evsel->fd); thread++) {
+		int err;
+		int *fd = FD(evsel, cpu_map_idx, thread);
+
+		if (!fd || *fd < 0)
+			return -1;
+
+		err = fcntl(*fd, cmd, arg);
+
+		if (err)
+			return err;
+	}
+
+	return 0;
+}
+
+static int perf_evsel__set_signal_handler(struct perf_evsel *evsel,
+					  struct perf_evsel_open_opts *opts)
+{
+	unsigned int flags;
+	unsigned int signal;
+	struct f_owner_ex owner;
+	struct sigaction *sig;
+	int cpu_map_idx;
+	int err = 0;
+
+	flags = OPTS_GET(opts, flags, (O_RDWR | O_NONBLOCK | O_ASYNC));
+	signal = OPTS_GET(opts, signal, SIGIO);
+	owner.type = OPTS_GET(opts, owner_type, F_OWNER_PID);
+	sig = OPTS_GET(opts, sig, NULL);
+
+	if (flags == 0 && signal == 0 && !owner.type == 0 && sig == 0)
+		return err;
+
+	err = sigaction(signal, sig, NULL);
+	if (err)
+		return err;
+
+	switch (owner.type) {
+	case F_OWNER_PID:
+		owner.pid = getpid();
+		break;
+	case F_OWNER_TID:
+		owner.pid = syscall(SYS_gettid);
+		break;
+	case F_OWNER_PGRP:
+	default:
+		return -1;
+	}
+
+	for (cpu_map_idx = 0; cpu_map_idx < xyarray__max_x(evsel->fd); cpu_map_idx++) {
+		err = perf_evsel__run_fcntl(evsel, F_SETFL, flags, cpu_map_idx);
+		if (err)
+			return err;
+
+		err = perf_evsel__run_fcntl(evsel, F_SETSIG, signal, cpu_map_idx);
+		if (err)
+			return err;
+
+		err = perf_evsel__run_fcntl(evsel, F_SETOWN_EX,
+					    (unsigned long)&owner, cpu_map_idx);
+		if (err)
+			return err;
+	}
+
+	return err;
+}
+
+int perf_evsel__open_opts(struct perf_evsel *evsel, struct perf_cpu_map *cpus,
+			  struct perf_thread_map *threads,
+			  struct perf_evsel_open_opts *opts)
+{
+	int err = 0;
+
+	if (!OPTS_VALID(opts, perf_evsel_open_opts)) {
+		err = -EINVAL;
+		return err;
+	}
+
+	evsel->open_flags = OPTS_GET(opts, open_flags, 0);
+
+	err = perf_evsel__open(evsel, cpus, threads);
+	if (err)
+		return err;
+
+	err = perf_evsel__set_signal_handler(evsel, opts);
+	if (err)
+		return err;
+
+	return err;
+}
diff --git a/tools/lib/perf/include/perf/evlist.h b/tools/lib/perf/include/perf/evlist.h
index 9ca399d49bb4..6eff1e9b3481 100644
--- a/tools/lib/perf/include/perf/evlist.h
+++ b/tools/lib/perf/include/perf/evlist.h
@@ -9,6 +9,7 @@ struct perf_evlist;
 struct perf_evsel;
 struct perf_cpu_map;
 struct perf_thread_map;
+struct perf_evsel_open_opts;
 
 LIBPERF_API void perf_evlist__add(struct perf_evlist *evlist,
 				  struct perf_evsel *evsel);
@@ -47,4 +48,6 @@ LIBPERF_API struct perf_mmap *perf_evlist__next_mmap(struct perf_evlist *evlist,
 	     (pos) = perf_evlist__next_mmap((evlist), (pos), overwrite))
 
 LIBPERF_API void perf_evlist__set_leader(struct perf_evlist *evlist);
+LIBPERF_API int perf_evlist__open_opts(struct perf_evlist *evlist,
+				       struct perf_evsel_open_opts *opts);
 #endif /* __LIBPERF_EVLIST_H */
diff --git a/tools/lib/perf/include/perf/evsel.h b/tools/lib/perf/include/perf/evsel.h
index 360ced734add..ecf30bc6303f 100644
--- a/tools/lib/perf/include/perf/evsel.h
+++ b/tools/lib/perf/include/perf/evsel.h
@@ -5,6 +5,7 @@
 #include <stdint.h>
 #include <perf/core.h>
 #include <stdbool.h>
+#include <signal.h>
 #include <linux/types.h>
 
 struct perf_evsel;
@@ -23,6 +24,27 @@ struct perf_counts_values {
 	};
 };
 
+struct perf_evsel_open_opts {
+	/* size of this struct, for forward/backward compatibility */
+	size_t sz;
+
+	unsigned long open_flags;	/* perf_event_open flags */
+	int flags;			/* fcntl flags */
+	unsigned int signal;
+	int owner_type;
+	struct sigaction *sig;
+};
+#define perf_evsel_open_opts__last_field sig
+
+#define LIBPERF_OPTS(TYPE, NAME, ...)			\
+	struct TYPE NAME = ({				\
+		memset(&NAME, 0, sizeof(struct TYPE));	\
+		(struct TYPE) {				\
+			.sz = sizeof(struct TYPE),	\
+			__VA_ARGS__			\
+		};					\
+	})
+
 LIBPERF_API struct perf_evsel *perf_evsel__new(struct perf_event_attr *attr);
 LIBPERF_API void perf_evsel__delete(struct perf_evsel *evsel);
 LIBPERF_API int perf_evsel__open(struct perf_evsel *evsel, struct perf_cpu_map *cpus,
@@ -48,5 +70,9 @@ LIBPERF_API struct perf_thread_map *perf_evsel__threads(struct perf_evsel *evsel
 LIBPERF_API struct perf_event_attr *perf_evsel__attr(struct perf_evsel *evsel);
 LIBPERF_API void perf_counts_values__scale(struct perf_counts_values *count,
 					   bool scale, __s8 *pscaled);
+LIBPERF_API int perf_evsel__open_opts(struct perf_evsel *evsel,
+				      struct perf_cpu_map *cpus,
+				      struct perf_thread_map *threads,
+				      struct perf_evsel_open_opts *opts);
 
 #endif /* __LIBPERF_EVSEL_H */
diff --git a/tools/lib/perf/internal.h b/tools/lib/perf/internal.h
index 2c27e158de6b..637bf6793c26 100644
--- a/tools/lib/perf/internal.h
+++ b/tools/lib/perf/internal.h
@@ -20,4 +20,48 @@ do {                            \
 #define pr_debug2(fmt, ...)     __pr(LIBPERF_DEBUG2, fmt, ##__VA_ARGS__)
 #define pr_debug3(fmt, ...)     __pr(LIBPERF_DEBUG3, fmt, ##__VA_ARGS__)
 
+static inline bool libperf_is_mem_zeroed(const char *p, ssize_t len)
+{
+	while (len > 0) {
+		if (*p)
+			return false;
+		p++;
+		len--;
+	}
+	return true;
+}
+
+static inline bool libperf_validate_opts(const char *opts,
+					 size_t opts_sz, size_t user_sz,
+					 const char *type_name)
+{
+	if (user_sz < sizeof(size_t)) {
+		pr_warning("%s size (%zu) is too small\n", type_name, user_sz);
+		return false;
+	}
+	if (!libperf_is_mem_zeroed(opts + opts_sz, (ssize_t)user_sz - opts_sz)) {
+		pr_warning("%s has non-zero extra bytes\n", type_name);
+		return false;
+	}
+	return true;
+}
+
+# define offsetofend(TYPE, FIELD) \
+	(offsetof(TYPE, FIELD) + sizeof(((TYPE *)0)->FIELD))
+
+#define OPTS_VALID(opts, type)							\
+	(!(opts) || libperf_validate_opts((const char *)opts,			\
+					  offsetofend(struct type,		\
+						      type##__last_field),	\
+						(opts)->sz, #type))
+#define OPTS_HAS(opts, field) \
+	((opts) && opts->sz >= offsetofend(typeof(*(opts)), field))
+#define OPTS_GET(opts, field, fallback_value) \
+	(OPTS_HAS(opts, field) ? (opts)->field : fallback_value)
+#define OPTS_SET(opts, field, value)		\
+	do {					\
+		if (OPTS_HAS(opts, field))	\
+			(opts)->field = value;	\
+	} while (0)
+
 #endif /* __LIBPERF_INTERNAL_H */
diff --git a/tools/lib/perf/libperf.map b/tools/lib/perf/libperf.map
index ab0f44e9bb57..534614fbbb26 100644
--- a/tools/lib/perf/libperf.map
+++ b/tools/lib/perf/libperf.map
@@ -24,6 +24,7 @@ LIBPERF_0.0.1 {
 		perf_evsel__enable;
 		perf_evsel__disable;
 		perf_evsel__open;
+		perf_evsel__open_opts;
 		perf_evsel__close;
 		perf_evsel__mmap;
 		perf_evsel__munmap;
@@ -39,6 +40,7 @@ LIBPERF_0.0.1 {
 		perf_evlist__new;
 		perf_evlist__delete;
 		perf_evlist__open;
+		perf_evlist__open_opts;
 		perf_evlist__close;
 		perf_evlist__enable;
 		perf_evlist__disable;
-- 
2.25.1

Re: [RFC PATCH v2 4/7] libperf: Introduce perf_{evsel, evlist}__open_opt with extensible struct opts

[Jiri Olsa]

On Fri, Mar 25, 2022 at 01:38:26PM +0900, Shunsuke Nakamura wrote:
> Introduce perf_{evsel, evlist}__open_opt with extensible structure opts.
> The mechanism of the extensible structure opts imitates
> tools/lib/bpf/libbpf.h. This allows the user to set the open_flags
> specified in perf_event_open and a signal handler to receive overflow
> notifications for sampling events.
> 
> Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
> ---
>  tools/lib/perf/Documentation/libperf.txt |  14 +++
>  tools/lib/perf/evlist.c                  |  20 +++++
>  tools/lib/perf/evsel.c                   | 105 +++++++++++++++++++++++
>  tools/lib/perf/include/perf/evlist.h     |   3 +
>  tools/lib/perf/include/perf/evsel.h      |  26 ++++++
>  tools/lib/perf/internal.h                |  44 ++++++++++
>  tools/lib/perf/libperf.map               |   2 +
>  7 files changed, 214 insertions(+)
> 
> diff --git a/tools/lib/perf/Documentation/libperf.txt b/tools/lib/perf/Documentation/libperf.txt
> index ae55e62fc4ce..700c1ce15159 100644
> --- a/tools/lib/perf/Documentation/libperf.txt
> +++ b/tools/lib/perf/Documentation/libperf.txt
> @@ -131,6 +131,20 @@ SYNOPSIS
>            };
>    };
>  
> +  struct perf_evsel_open_opts {
> +          /* size of this struct, for forward/backward compatibility */
> +          size_t sz;
> +
> +          unsigned long open_flags;       /* perf_event_open flags */
> +          int flags;                      /* fcntl flags */
> +          unsigned int signal;
> +          int owner_type;
> +          struct sigaction *sig;
> +  };
> +  #define perf_evsel_open_opts__last_field sig
> +
> +  #define LIBPERF_OPTS(TYPE, NAME, ...)
> +

SNIP

> +
> +int perf_evsel__open_opts(struct perf_evsel *evsel, struct perf_cpu_map *cpus,
> +			  struct perf_thread_map *threads,
> +			  struct perf_evsel_open_opts *opts)
> +{
> +	int err = 0;
> +
> +	if (!OPTS_VALID(opts, perf_evsel_open_opts)) {
> +		err = -EINVAL;
> +		return err;
> +	}
> +
> +	evsel->open_flags = OPTS_GET(opts, open_flags, 0);
> +
> +	err = perf_evsel__open(evsel, cpus, threads);
> +	if (err)
> +		return err;
> +
> +	err = perf_evsel__set_signal_handler(evsel, opts);
> +	if (err)
> +		return err;

please move the signal stuff handling into separate patch
together with the related fields in opts struct

thanks,
jirka

> +
> +	return err;
> +}
> diff --git a/tools/lib/perf/include/perf/evlist.h b/tools/lib/perf/include/perf/evlist.h
> index 9ca399d49bb4..6eff1e9b3481 100644
> --- a/tools/lib/perf/include/perf/evlist.h
> +++ b/tools/lib/perf/include/perf/evlist.h
> @@ -9,6 +9,7 @@ struct perf_evlist;
>  struct perf_evsel;
>  struct perf_cpu_map;
>  struct perf_thread_map;
> +struct perf_evsel_open_opts;
>  
>  LIBPERF_API void perf_evlist__add(struct perf_evlist *evlist,
>  				  struct perf_evsel *evsel);
> @@ -47,4 +48,6 @@ LIBPERF_API struct perf_mmap *perf_evlist__next_mmap(struct perf_evlist *evlist,
>  	     (pos) = perf_evlist__next_mmap((evlist), (pos), overwrite))
>  
>  LIBPERF_API void perf_evlist__set_leader(struct perf_evlist *evlist);
> +LIBPERF_API int perf_evlist__open_opts(struct perf_evlist *evlist,
> +				       struct perf_evsel_open_opts *opts);
>  #endif /* __LIBPERF_EVLIST_H */
> diff --git a/tools/lib/perf/include/perf/evsel.h b/tools/lib/perf/include/perf/evsel.h
> index 360ced734add..ecf30bc6303f 100644
> --- a/tools/lib/perf/include/perf/evsel.h
> +++ b/tools/lib/perf/include/perf/evsel.h
> @@ -5,6 +5,7 @@
>  #include <stdint.h>
>  #include <perf/core.h>
>  #include <stdbool.h>
> +#include <signal.h>
>  #include <linux/types.h>
>  
>  struct perf_evsel;
> @@ -23,6 +24,27 @@ struct perf_counts_values {
>  	};
>  };
>  
> +struct perf_evsel_open_opts {
> +	/* size of this struct, for forward/backward compatibility */
> +	size_t sz;
> +
> +	unsigned long open_flags;	/* perf_event_open flags */
> +	int flags;			/* fcntl flags */
> +	unsigned int signal;
> +	int owner_type;
> +	struct sigaction *sig;
> +};
> +#define perf_evsel_open_opts__last_field sig
> +
> +#define LIBPERF_OPTS(TYPE, NAME, ...)			\
> +	struct TYPE NAME = ({				\
> +		memset(&NAME, 0, sizeof(struct TYPE));	\
> +		(struct TYPE) {				\
> +			.sz = sizeof(struct TYPE),	\
> +			__VA_ARGS__			\
> +		};					\
> +	})
> +
>  LIBPERF_API struct perf_evsel *perf_evsel__new(struct perf_event_attr *attr);
>  LIBPERF_API void perf_evsel__delete(struct perf_evsel *evsel);
>  LIBPERF_API int perf_evsel__open(struct perf_evsel *evsel, struct perf_cpu_map *cpus,
> @@ -48,5 +70,9 @@ LIBPERF_API struct perf_thread_map *perf_evsel__threads(struct perf_evsel *evsel
>  LIBPERF_API struct perf_event_attr *perf_evsel__attr(struct perf_evsel *evsel);
>  LIBPERF_API void perf_counts_values__scale(struct perf_counts_values *count,
>  					   bool scale, __s8 *pscaled);
> +LIBPERF_API int perf_evsel__open_opts(struct perf_evsel *evsel,
> +				      struct perf_cpu_map *cpus,
> +				      struct perf_thread_map *threads,
> +				      struct perf_evsel_open_opts *opts);
>  
>  #endif /* __LIBPERF_EVSEL_H */
> diff --git a/tools/lib/perf/internal.h b/tools/lib/perf/internal.h
> index 2c27e158de6b..637bf6793c26 100644
> --- a/tools/lib/perf/internal.h
> +++ b/tools/lib/perf/internal.h
> @@ -20,4 +20,48 @@ do {                            \
>  #define pr_debug2(fmt, ...)     __pr(LIBPERF_DEBUG2, fmt, ##__VA_ARGS__)
>  #define pr_debug3(fmt, ...)     __pr(LIBPERF_DEBUG3, fmt, ##__VA_ARGS__)
>  
> +static inline bool libperf_is_mem_zeroed(const char *p, ssize_t len)
> +{
> +	while (len > 0) {
> +		if (*p)
> +			return false;
> +		p++;
> +		len--;
> +	}
> +	return true;
> +}
> +
> +static inline bool libperf_validate_opts(const char *opts,
> +					 size_t opts_sz, size_t user_sz,
> +					 const char *type_name)
> +{
> +	if (user_sz < sizeof(size_t)) {
> +		pr_warning("%s size (%zu) is too small\n", type_name, user_sz);
> +		return false;
> +	}
> +	if (!libperf_is_mem_zeroed(opts + opts_sz, (ssize_t)user_sz - opts_sz)) {
> +		pr_warning("%s has non-zero extra bytes\n", type_name);
> +		return false;
> +	}
> +	return true;
> +}
> +
> +# define offsetofend(TYPE, FIELD) \
> +	(offsetof(TYPE, FIELD) + sizeof(((TYPE *)0)->FIELD))
> +
> +#define OPTS_VALID(opts, type)							\
> +	(!(opts) || libperf_validate_opts((const char *)opts,			\
> +					  offsetofend(struct type,		\
> +						      type##__last_field),	\
> +						(opts)->sz, #type))
> +#define OPTS_HAS(opts, field) \
> +	((opts) && opts->sz >= offsetofend(typeof(*(opts)), field))
> +#define OPTS_GET(opts, field, fallback_value) \
> +	(OPTS_HAS(opts, field) ? (opts)->field : fallback_value)
> +#define OPTS_SET(opts, field, value)		\
> +	do {					\
> +		if (OPTS_HAS(opts, field))	\
> +			(opts)->field = value;	\
> +	} while (0)
> +
>  #endif /* __LIBPERF_INTERNAL_H */
> diff --git a/tools/lib/perf/libperf.map b/tools/lib/perf/libperf.map
> index ab0f44e9bb57..534614fbbb26 100644
> --- a/tools/lib/perf/libperf.map
> +++ b/tools/lib/perf/libperf.map
> @@ -24,6 +24,7 @@ LIBPERF_0.0.1 {
>  		perf_evsel__enable;
>  		perf_evsel__disable;
>  		perf_evsel__open;
> +		perf_evsel__open_opts;
>  		perf_evsel__close;
>  		perf_evsel__mmap;
>  		perf_evsel__munmap;
> @@ -39,6 +40,7 @@ LIBPERF_0.0.1 {
>  		perf_evlist__new;
>  		perf_evlist__delete;
>  		perf_evlist__open;
> +		perf_evlist__open_opts;
>  		perf_evlist__close;
>  		perf_evlist__enable;
>  		perf_evlist__disable;
> -- 
> 2.25.1
> 

RE: [RFC PATCH v2 4/7] libperf: Introduce perf_{evsel, evlist}__open_opt with extensible struct opts

[nakamura.shun]

Hi jirka

> On Fri, Mar 25, 2022 at 01:38:26PM +0900, Shunsuke Nakamura wrote:
> > Introduce perf_{evsel, evlist}__open_opt with extensible structure opts.
> > The mechanism of the extensible structure opts imitates
> > tools/lib/bpf/libbpf.h. This allows the user to set the open_flags
> > specified in perf_event_open and a signal handler to receive overflow
> > notifications for sampling events.
> >
> > Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
> > ---
> >  tools/lib/perf/Documentation/libperf.txt |  14 +++
> >  tools/lib/perf/evlist.c                  |  20 +++++
> >  tools/lib/perf/evsel.c                   | 105
> +++++++++++++++++++++++
> >  tools/lib/perf/include/perf/evlist.h     |   3 +
> >  tools/lib/perf/include/perf/evsel.h      |  26 ++++++
> >  tools/lib/perf/internal.h                |  44 ++++++++++
> >  tools/lib/perf/libperf.map               |   2 +
> >  7 files changed, 214 insertions(+)
> >
> > diff --git a/tools/lib/perf/Documentation/libperf.txt
> > b/tools/lib/perf/Documentation/libperf.txt
> > index ae55e62fc4ce..700c1ce15159 100644
> > --- a/tools/lib/perf/Documentation/libperf.txt
> > +++ b/tools/lib/perf/Documentation/libperf.txt
> > @@ -131,6 +131,20 @@ SYNOPSIS
> >            };
> >    };
> >
> > +  struct perf_evsel_open_opts {
> > +          /* size of this struct, for forward/backward compatibility */
> > +          size_t sz;
> > +
> > +          unsigned long open_flags;       /* perf_event_open flags */
> > +          int flags;                      /* fcntl flags */
> > +          unsigned int signal;
> > +          int owner_type;
> > +          struct sigaction *sig;
> > +  };
> > +  #define perf_evsel_open_opts__last_field sig
> > +
> > +  #define LIBPERF_OPTS(TYPE, NAME, ...)
> > +
> 
> SNIP
> 
> > +
> > +int perf_evsel__open_opts(struct perf_evsel *evsel, struct perf_cpu_map
> *cpus,
> > +			  struct perf_thread_map *threads,
> > +			  struct perf_evsel_open_opts *opts) {
> > +	int err = 0;
> > +
> > +	if (!OPTS_VALID(opts, perf_evsel_open_opts)) {
> > +		err = -EINVAL;
> > +		return err;
> > +	}
> > +
> > +	evsel->open_flags = OPTS_GET(opts, open_flags, 0);
> > +
> > +	err = perf_evsel__open(evsel, cpus, threads);
> > +	if (err)
> > +		return err;
> > +
> > +	err = perf_evsel__set_signal_handler(evsel, opts);
> > +	if (err)
> > +		return err;
> 
> please move the signal stuff handling into separate patch together with the
> related fields in opts struct
> 
I'll fix it.


Best Regards
Shunsuke

Re: [RFC PATCH v2 4/7] libperf: Introduce perf_{evsel, evlist}__open_opt with extensible struct opts

[Jiri Olsa]

On Fri, Mar 25, 2022 at 01:38:26PM +0900, Shunsuke Nakamura wrote:

SNIP

> diff --git a/tools/lib/perf/evsel.c b/tools/lib/perf/evsel.c
> index b17842581dea..db9b7274feb5 100644
> --- a/tools/lib/perf/evsel.c
> +++ b/tools/lib/perf/evsel.c
> @@ -1,4 +1,7 @@
>  // SPDX-License-Identifier: GPL-2.0
> +#ifndef _GNU_SOURCE
> +#define _GNU_SOURCE
> +#endif

I think we can make this global set in Makefile, like we do in perf

jirka

RE: [RFC PATCH v2 4/7] libperf: Introduce perf_{evsel, evlist}__open_opt with extensible struct opts

[nakamura.shun]

Hi jirka

> On Fri, Mar 25, 2022 at 01:38:26PM +0900, Shunsuke Nakamura wrote:
> 
> SNIP
> 
> > diff --git a/tools/lib/perf/evsel.c b/tools/lib/perf/evsel.c index
> > b17842581dea..db9b7274feb5 100644
> > --- a/tools/lib/perf/evsel.c
> > +++ b/tools/lib/perf/evsel.c
> > @@ -1,4 +1,7 @@
> >  // SPDX-License-Identifier: GPL-2.0
> > +#ifndef _GNU_SOURCE
> > +#define _GNU_SOURCE
> > +#endif
> 
> I think we can make this global set in Makefile, like we do in perf
> 
I'll fix it.


Best Regards
Shunsuke

[RFC PATCH v2 5/7] libperf: Add perf_evsel__check_overflow() functions

[Shunsuke Nakamura]

Add the following functions:

  perf_evsel__check_overflow
  perf_evsel__check_overflow_cpu

to check for perf events with the file descriptor specified in the
argument.
These functions can be used in signal handlers to check overflow.

Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
---
 tools/lib/perf/Documentation/libperf.txt |  3 +++
 tools/lib/perf/evsel.c                   | 34 ++++++++++++++++++++++++
 tools/lib/perf/include/perf/evsel.h      |  4 +++
 tools/lib/perf/libperf.map               |  2 ++
 4 files changed, 43 insertions(+)

diff --git a/tools/lib/perf/Documentation/libperf.txt b/tools/lib/perf/Documentation/libperf.txt
index 700c1ce15159..4ae8d738b948 100644
--- a/tools/lib/perf/Documentation/libperf.txt
+++ b/tools/lib/perf/Documentation/libperf.txt
@@ -165,6 +165,9 @@ SYNOPSIS
                               int cpu_map_idx);
   int perf_evsel__period(struct perf_evsel *evsel, int period);
   int perf_evsel__period_cpu(struct perf_evsel *evsel, int period, int cpu_map_idx);
+  int perf_evsel__check_overflow(struct perf_evsel *evsel, int sig_fd, bool *overflow);
+  int perf_evsel__check_overflow_cpu(struct perf_evsel *evsel, int cpu_map_idx,
+                                     int sig_fd, bool *overflow);
   struct perf_cpu_map *perf_evsel__cpus(struct perf_evsel *evsel);
   struct perf_thread_map *perf_evsel__threads(struct perf_evsel *evsel);
   struct perf_event_attr *perf_evsel__attr(struct perf_evsel *evsel);
diff --git a/tools/lib/perf/evsel.c b/tools/lib/perf/evsel.c
index db9b7274feb5..6ff3cf692df3 100644
--- a/tools/lib/perf/evsel.c
+++ b/tools/lib/perf/evsel.c
@@ -616,3 +616,37 @@ int perf_evsel__open_opts(struct perf_evsel *evsel, struct perf_cpu_map *cpus,
 
 	return err;
 }
+
+int perf_evsel__check_overflow_cpu(struct perf_evsel *evsel, int cpu_map_idx,
+				   int sig_fd, bool *overflow)
+{
+	int thread;
+	int *fd;
+	int err = 0;
+
+	if (!overflow)
+		return -EINVAL;
+
+	*overflow = false;
+
+	for (thread = 0; thread < xyarray__max_y(evsel->fd) && !err; ++thread) {
+		fd = FD(evsel, cpu_map_idx, thread);
+		if (sig_fd <= 0 || !fd || *fd < 0)
+			err = -EINVAL;
+
+		if (sig_fd == *fd)
+			*overflow = true;
+	}
+
+	return err;
+}
+
+int perf_evsel__check_overflow(struct perf_evsel *evsel, int sig_fd, bool *overflow)
+{
+	int cpu_map_idx;
+	int err = 0;
+
+	for (cpu_map_idx = 0; cpu_map_idx < xyarray__max_x(evsel->fd) && !err; cpu_map_idx++)
+		err = perf_evsel__check_overflow_cpu(evsel, cpu_map_idx, sig_fd, overflow);
+	return err;
+}
diff --git a/tools/lib/perf/include/perf/evsel.h b/tools/lib/perf/include/perf/evsel.h
index ecf30bc6303f..d686cfbd88a6 100644
--- a/tools/lib/perf/include/perf/evsel.h
+++ b/tools/lib/perf/include/perf/evsel.h
@@ -74,5 +74,9 @@ LIBPERF_API int perf_evsel__open_opts(struct perf_evsel *evsel,
 				      struct perf_cpu_map *cpus,
 				      struct perf_thread_map *threads,
 				      struct perf_evsel_open_opts *opts);
+LIBPERF_API int perf_evsel__check_overflow(struct perf_evsel *evsel, int sig_fd,
+					   bool *overflow);
+LIBPERF_API int perf_evsel__check_overflow_cpu(struct perf_evsel *evsel,
+					       int cpu_map_idx, int sig_fd, bool *overflow);
 
 #endif /* __LIBPERF_EVSEL_H */
diff --git a/tools/lib/perf/libperf.map b/tools/lib/perf/libperf.map
index 534614fbbb26..981eade34237 100644
--- a/tools/lib/perf/libperf.map
+++ b/tools/lib/perf/libperf.map
@@ -37,6 +37,8 @@ LIBPERF_0.0.1 {
 		perf_evsel__cpus;
 		perf_evsel__threads;
 		perf_evsel__attr;
+		perf_evsel__check_overflow;
+		perf_evsel__check_overflow_cpu;
 		perf_evlist__new;
 		perf_evlist__delete;
 		perf_evlist__open;
-- 
2.25.1

Re: [RFC PATCH v2 5/7] libperf: Add perf_evsel__check_overflow() functions

[Jiri Olsa]

On Fri, Mar 25, 2022 at 01:38:27PM +0900, Shunsuke Nakamura wrote:
> Add the following functions:
> 
>   perf_evsel__check_overflow
>   perf_evsel__check_overflow_cpu
> 
> to check for perf events with the file descriptor specified in the
> argument.
> These functions can be used in signal handlers to check overflow.
> 
> Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
> ---
>  tools/lib/perf/Documentation/libperf.txt |  3 +++
>  tools/lib/perf/evsel.c                   | 34 ++++++++++++++++++++++++
>  tools/lib/perf/include/perf/evsel.h      |  4 +++
>  tools/lib/perf/libperf.map               |  2 ++
>  4 files changed, 43 insertions(+)
> 
> diff --git a/tools/lib/perf/Documentation/libperf.txt b/tools/lib/perf/Documentation/libperf.txt
> index 700c1ce15159..4ae8d738b948 100644
> --- a/tools/lib/perf/Documentation/libperf.txt
> +++ b/tools/lib/perf/Documentation/libperf.txt
> @@ -165,6 +165,9 @@ SYNOPSIS
>                                int cpu_map_idx);
>    int perf_evsel__period(struct perf_evsel *evsel, int period);
>    int perf_evsel__period_cpu(struct perf_evsel *evsel, int period, int cpu_map_idx);
> +  int perf_evsel__check_overflow(struct perf_evsel *evsel, int sig_fd, bool *overflow);
> +  int perf_evsel__check_overflow_cpu(struct perf_evsel *evsel, int cpu_map_idx,
> +                                     int sig_fd, bool *overflow);

should this be more like:

  perf_evsel__has_fd(struct perf_evsel *evsel, int fd)

also why do we need to export *_cpu version for this?
I don't see it used in the test

>    struct perf_cpu_map *perf_evsel__cpus(struct perf_evsel *evsel);
>    struct perf_thread_map *perf_evsel__threads(struct perf_evsel *evsel);
>    struct perf_event_attr *perf_evsel__attr(struct perf_evsel *evsel);
> diff --git a/tools/lib/perf/evsel.c b/tools/lib/perf/evsel.c
> index db9b7274feb5..6ff3cf692df3 100644
> --- a/tools/lib/perf/evsel.c
> +++ b/tools/lib/perf/evsel.c
> @@ -616,3 +616,37 @@ int perf_evsel__open_opts(struct perf_evsel *evsel, struct perf_cpu_map *cpus,
>  
>  	return err;
>  }
> +
> +int perf_evsel__check_overflow_cpu(struct perf_evsel *evsel, int cpu_map_idx,
> +				   int sig_fd, bool *overflow)
> +{
> +	int thread;
> +	int *fd;
> +	int err = 0;
> +
> +	if (!overflow)
> +		return -EINVAL;
> +
> +	*overflow = false;
> +
> +	for (thread = 0; thread < xyarray__max_y(evsel->fd) && !err; ++thread) {
> +		fd = FD(evsel, cpu_map_idx, thread);
> +		if (sig_fd <= 0 || !fd || *fd < 0)
> +			err = -EINVAL;

sig_fd check should be before the loop

jirka

RE: [RFC PATCH v2 5/7] libperf: Add perf_evsel__check_overflow() functions

[nakamura.shun]

Hi jirka

> On Fri, Mar 25, 2022 at 01:38:27PM +0900, Shunsuke Nakamura wrote:
> > Add the following functions:
> >
> >   perf_evsel__check_overflow
> >   perf_evsel__check_overflow_cpu
> >
> > to check for perf events with the file descriptor specified in the
> > argument.
> > These functions can be used in signal handlers to check overflow.
> >
> > Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
> > ---
> >  tools/lib/perf/Documentation/libperf.txt |  3 +++
> >  tools/lib/perf/evsel.c                   | 34
> ++++++++++++++++++++++++
> >  tools/lib/perf/include/perf/evsel.h      |  4 +++
> >  tools/lib/perf/libperf.map               |  2 ++
> >  4 files changed, 43 insertions(+)
> >
> > diff --git a/tools/lib/perf/Documentation/libperf.txt
> > b/tools/lib/perf/Documentation/libperf.txt
> > index 700c1ce15159..4ae8d738b948 100644
> > --- a/tools/lib/perf/Documentation/libperf.txt
> > +++ b/tools/lib/perf/Documentation/libperf.txt
> > @@ -165,6 +165,9 @@ SYNOPSIS
> >                                int cpu_map_idx);
> >    int perf_evsel__period(struct perf_evsel *evsel, int period);
> >    int perf_evsel__period_cpu(struct perf_evsel *evsel, int period,
> > int cpu_map_idx);
> > +  int perf_evsel__check_overflow(struct perf_evsel *evsel, int
> > + sig_fd, bool *overflow);  int perf_evsel__check_overflow_cpu(struct
> perf_evsel *evsel, int cpu_map_idx,
> > +                                     int sig_fd, bool *overflow);
> 
> should this be more like:
> 
>   perf_evsel__has_fd(struct perf_evsel *evsel, int fd)
> 
I'll fix it.


> also why do we need to export *_cpu version for this?
> I don't see it used in the test
> 
Sorry, not necessary.
I'll fix it.

> >    struct perf_cpu_map *perf_evsel__cpus(struct perf_evsel *evsel);
> >    struct perf_thread_map *perf_evsel__threads(struct perf_evsel *evsel);
> >    struct perf_event_attr *perf_evsel__attr(struct perf_evsel *evsel);
> > diff --git a/tools/lib/perf/evsel.c b/tools/lib/perf/evsel.c index
> > db9b7274feb5..6ff3cf692df3 100644
> > --- a/tools/lib/perf/evsel.c
> > +++ b/tools/lib/perf/evsel.c
> > @@ -616,3 +616,37 @@ int perf_evsel__open_opts(struct perf_evsel
> > *evsel, struct perf_cpu_map *cpus,
> >
> >  	return err;
> >  }
> > +
> > +int perf_evsel__check_overflow_cpu(struct perf_evsel *evsel, int
> cpu_map_idx,
> > +				   int sig_fd, bool *overflow)
> > +{
> > +	int thread;
> > +	int *fd;
> > +	int err = 0;
> > +
> > +	if (!overflow)
> > +		return -EINVAL;
> > +
> > +	*overflow = false;
> > +
> > +	for (thread = 0; thread < xyarray__max_y(evsel->fd) && !err; ++thread)
> {
> > +		fd = FD(evsel, cpu_map_idx, thread);
> > +		if (sig_fd <= 0 || !fd || *fd < 0)
> > +			err = -EINVAL;
> 
> sig_fd check should be before the loop
> 
I'll fix it.


Best Regards
Shunsuke

[RFC PATCH v2 6/7] libperf test: Add test_stat_overflow()

[Shunsuke Nakamura]

Added overflow test using refresh and period.

Confirmation
 - That the overflow occurs the number of times specified by
   perf_evse__refresh()
 - That the period can be updated by perf_evsel__period()

Committer testing:

  $ sudo make tests -C ./tools/lib/perf V=1
  make: Entering directory '/home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/lib/perf'
  make -f /home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/build/Makefile.build dir=. obj=libperf
  make -C /home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/lib/api/ O= libapi.a
  make -f /home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/build/Makefile.build dir=./fd obj=libapi
  make -f /home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/build/Makefile.build dir=./fs obj=libapi
  make -f /home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/build/Makefile.build dir=. obj=tests
  make -f /home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/build/Makefile.build dir=./tests obj=tests
  running static:
  - running tests/test-cpumap.c...OK
  - running tests/test-threadmap.c...OK
  - running tests/test-evlist.c...

  <SNIP>

  OK
  - running tests/test-evsel.c...

  <SNIP>

          period = 1000000
          overflow limit = 3, overflow count = 3, POLL_IN = 2, POLL_UP = 1, other signal event = 0
          period = 2000000
          overflow limit = 3, overflow count = 3, POLL_IN = 2, POLL_UP = 1, other signal event = 0
          period = 1000000
          overflow limit = 3, overflow count = 3, POLL_IN = 2, POLL_UP = 1, other signal event = 0
          period = 2000000
          overflow limit = 3, overflow count = 3, POLL_IN = 2, POLL_UP = 1, other signal event = 0
  OK
  running dynamic:
  - running tests/test-cpumap.c...OK
  - running tests/test-threadmap.c...OK
  - running tests/test-evlist.c...

  <SNIP>

  OK
  - running tests/test-evsel.c...

  <SNIP>

          period = 1000000
          overflow limit = 3, overflow count = 3, POLL_IN = 2, POLL_UP = 1, other signal event = 0
          period = 2000000
          overflow limit = 3, overflow count = 3, POLL_IN = 2, POLL_UP = 1, other signal event = 0
          period = 1000000
          overflow limit = 3, overflow count = 3, POLL_IN = 2, POLL_UP = 1, other signal event = 0
          period = 2000000
          overflow limit = 3, overflow count = 3, POLL_IN = 2, POLL_UP = 1, other signal event = 0
  OK
  make: Leaving directory '/home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/lib/perf'

Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
---
 tools/lib/perf/tests/test-evsel.c | 111 ++++++++++++++++++++++++++++++
 1 file changed, 111 insertions(+)

diff --git a/tools/lib/perf/tests/test-evsel.c b/tools/lib/perf/tests/test-evsel.c
index 89be89afb24d..8e7cf336fcd7 100644
--- a/tools/lib/perf/tests/test-evsel.c
+++ b/tools/lib/perf/tests/test-evsel.c
@@ -1,6 +1,14 @@
 // SPDX-License-Identifier: GPL-2.0
+#ifndef _GNU_SOURCE
+#define _GNU_SOURCE
+#endif
+#include <inttypes.h>
 #include <stdarg.h>
 #include <stdio.h>
+#include <string.h>
+#include <signal.h>
+#include <unistd.h>
+#include <fcntl.h>
 #include <linux/perf_event.h>
 #include <perf/cpumap.h>
 #include <perf/threadmap.h>
@@ -8,6 +16,15 @@
 #include <internal/tests.h>
 #include "tests.h"
 
+#define WAIT_COUNT 10000000UL
+static struct signal_counts {
+	int in;
+	int hup;
+	int others;
+	int overflow;
+} sig_count = {0, 0, 0};
+static struct perf_evsel *s_evsel;
+
 static int libperf_print(enum libperf_print_level level,
 			 const char *fmt, va_list ap)
 {
@@ -189,6 +206,98 @@ static int test_stat_user_read(int event)
 	return 0;
 }
 
+static void sig_handler(int signo, siginfo_t *info, void *uc)
+{
+	switch (info->si_code) {
+	case POLL_IN:
+		sig_count.in++;
+		break;
+	case POLL_HUP:
+		sig_count.hup++;
+		break;
+	default:
+		sig_count.others++;
+	}
+
+	sig_count.overflow++;
+}
+
+static int test_stat_overflow(int owner)
+{
+	static struct sigaction sig;
+	u64 period = 1000000;
+	int overflow_limit = 3;
+
+	struct perf_thread_map *threads;
+	struct perf_event_attr attr = {
+		.type           = PERF_TYPE_SOFTWARE,
+		.config         = PERF_COUNT_SW_TASK_CLOCK,
+		.sample_type    = PERF_SAMPLE_PERIOD,
+		.sample_period  = period,
+		.disabled       = 1,
+	};
+	struct perf_event_attr *tmp_attr;
+	int err = 0, i;
+
+	LIBPERF_OPTS(perf_evsel_open_opts, opts,
+		     .open_flags = PERF_FLAG_FD_CLOEXEC,
+		     .flags	 = (O_RDWR | O_NONBLOCK | O_ASYNC),
+		     .signal	 = SIGIO,
+		     .owner_type = owner,
+		     .sig	 = &sig);
+
+	/* setup signal handler */
+	memset(&sig, 0, sizeof(struct sigaction));
+	sig.sa_sigaction = (void *)sig_handler;
+	sig.sa_flags = SA_SIGINFO;
+
+	threads = perf_thread_map__new_dummy();
+	__T("failed to create threads", threads);
+
+	perf_thread_map__set_pid(threads, 0, 0);
+
+	s_evsel = perf_evsel__new(&attr);
+	__T("failed to create evsel", s_evsel);
+
+	err = perf_evsel__open_opts(s_evsel, NULL, threads, &opts);
+	__T("failed to open evsel", err == 0);
+
+	for (i = 0; i < 2; i++) {
+		volatile unsigned int wait_count = WAIT_COUNT;
+
+		period = period << i;
+		err = perf_evsel__period(s_evsel, period);
+		__T("failed to period evsel", err == 0);
+
+		tmp_attr = perf_evsel__attr(s_evsel);
+		__T_VERBOSE("\tperiod = %llu\n", tmp_attr->sample_period);
+
+		err = perf_evsel__refresh(s_evsel, overflow_limit);
+		__T("failed to refresh evsel", err == 0);
+
+		while (wait_count--)
+			;
+
+		__T_VERBOSE("\toverflow limit = %d, overflow count = %d, ",
+			    overflow_limit, sig_count.overflow);
+		__T_VERBOSE("POLL_IN = %d, POLL_UP = %d, other signal event = %d\n",
+			    sig_count.in, sig_count.hup, sig_count.others);
+
+		__T("failed to overflow count", overflow_limit == sig_count.overflow);
+
+		sig_count.in = 0;
+		sig_count.hup = 0;
+		sig_count.others = 0;
+		sig_count.overflow = 0;
+	}
+
+	perf_evsel__close(s_evsel);
+	perf_evsel__delete(s_evsel);
+	perf_thread_map__put(threads);
+
+	return 0;
+}
+
 int test_evsel(int argc, char **argv)
 {
 	__T_START;
@@ -200,6 +309,8 @@ int test_evsel(int argc, char **argv)
 	test_stat_thread_enable();
 	test_stat_user_read(PERF_COUNT_HW_INSTRUCTIONS);
 	test_stat_user_read(PERF_COUNT_HW_CPU_CYCLES);
+	test_stat_overflow(F_OWNER_PID);
+	test_stat_overflow(F_OWNER_TID);
 
 	__T_END;
 	return tests_failed == 0 ? 0 : -1;
-- 
2.25.1

[RFC PATCH v2 7/7] libperf test: Add test_stat_overflow_event()

[Shunsuke Nakamura]

Add a test to check overflowed events.

Committer testing:

  $ sudo make tests -C ./tools/lib/perf V=1
  make: Entering directory '/home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/lib/perf'
  make -f /home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/build/Makefile.build dir=. obj=libperf
  make -C /home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/lib/api/ O= libapi.a
  make -f /home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/build/Makefile.build dir=./fd obj=libapi
  make -f /home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/build/Makefile.build dir=./fs obj=libapi
  make -f /home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/build/Makefile.build dir=. obj=tests
  make -f /home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/build/Makefile.build dir=./tests obj=tests
  running static:
  - running tests/test-cpumap.c...OK
  - running tests/test-threadmap.c...OK
  - running tests/test-evlist.c...

  <SNIP>

  Event  0 -- overflow flag = 0x1, POLL_UP = 1, other signal event = 0
  Event  1 -- overflow flag = 0x2, POLL_UP = 1, other signal event = 0
  Event  2 -- overflow flag = 0x4, POLL_UP = 1, other signal event = 0
  Event  3 -- overflow flag = 0x8, POLL_UP = 1, other signal event = 0
  OK
  - running tests/test-evsel.c...

  <SNIP>

  OK
  running dynamic:
  - running tests/test-cpumap.c...OK
  - running tests/test-threadmap.c...OK
  - running tests/test-evlist.c...

  <SNIP>

  Event  0 -- overflow flag = 0x1, POLL_UP = 1, other signal event = 0
  Event  1 -- overflow flag = 0x2, POLL_UP = 1, other signal event = 0
  Event  2 -- overflow flag = 0x4, POLL_UP = 1, other signal event = 0
  Event  3 -- overflow flag = 0x8, POLL_UP = 1, other signal event = 0
  OK
  - running tests/test-evsel.c...

  <SNIP>

  OK
  make: Leaving directory '/home/nakamura/build_work/build_kernel/linux-kernel/linux/tools/lib/perf'

Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
---
 tools/lib/perf/tests/test-evlist.c | 127 +++++++++++++++++++++++++++++
 1 file changed, 127 insertions(+)

diff --git a/tools/lib/perf/tests/test-evlist.c b/tools/lib/perf/tests/test-evlist.c
index ed616fc19b4f..ecfe35c64c40 100644
--- a/tools/lib/perf/tests/test-evlist.c
+++ b/tools/lib/perf/tests/test-evlist.c
@@ -6,6 +6,8 @@
 #include <stdarg.h>
 #include <unistd.h>
 #include <stdlib.h>
+#include <string.h>
+#include <fcntl.h>
 #include <linux/perf_event.h>
 #include <linux/limits.h>
 #include <sys/types.h>
@@ -25,6 +27,13 @@
 #define EVENT_NUM 15
 #define WAIT_COUNT 100000000UL
 
+static unsigned int overflow_flag;
+static struct signal_counts {
+	int hup;
+	int others;
+} sig_count = { 0, 0 };
+static struct perf_evlist *s_evlist;
+
 static int libperf_print(enum libperf_print_level level,
 			 const char *fmt, va_list ap)
 {
@@ -571,6 +580,123 @@ static int test_stat_multiplexing(void)
 	return 0;
 }
 
+static void sig_handler(int signo, siginfo_t *info, void *uc)
+{
+	struct perf_evsel *evsel;
+	int idx = 0;
+	bool overflow;
+	int err;
+
+	switch (info->si_code) {
+	case POLL_HUP:
+		perf_evlist__for_each_evsel(s_evlist, evsel) {
+			err = perf_evsel__check_overflow(evsel, info->si_fd, &overflow);
+			if (err)
+				fprintf(stderr, "failed to check evsel overflow %d\n", err);
+
+			if (overflow) {
+				overflow_flag = (1U << idx);
+				sig_count.hup++;
+				break;
+			}
+			idx++;
+		}
+		break;
+	default:
+		perf_evlist__for_each_evsel(s_evlist, evsel) {
+			err = perf_evsel__check_overflow(evsel, info->si_fd, &overflow);
+			if (err)
+				fprintf(stderr, "failed to check evsel overflow %d\n", err);
+
+			if (overflow) {
+				overflow_flag = (1U << idx);
+				sig_count.others++;
+				break;
+			}
+			idx++;
+		}
+	}
+}
+
+static int test_stat_overflow_event(void)
+{
+	static struct sigaction sig;
+
+	struct perf_thread_map *threads;
+	struct perf_evsel *evsel;
+	struct perf_event_attr attr = {
+		.type		= PERF_TYPE_SOFTWARE,
+		.config		= PERF_COUNT_SW_CPU_CLOCK,
+		.sample_type	= PERF_SAMPLE_PERIOD,
+		.sample_period	= 100000,
+		.disabled	= 1,
+	};
+	int err, i, event_num = 4;
+
+	LIBPERF_OPTS(perf_evsel_open_opts, opts,
+		     .open_flags = PERF_FLAG_FD_CLOEXEC,
+		     .flags	 = (O_RDWR | O_NONBLOCK | O_ASYNC),
+		     .signal	 = SIGIO,
+		     .owner_type = F_OWNER_PID,
+		     .sig	 = &sig);
+
+	/* setup signal handler */
+	memset(&sig, 0, sizeof(struct sigaction));
+	sig.sa_sigaction = (void *)sig_handler;
+	sig.sa_flags = SA_SIGINFO;
+
+	threads = perf_thread_map__new_dummy();
+	__T("failed to create threads", threads);
+
+	perf_thread_map__set_pid(threads, 0, 0);
+
+	s_evlist = perf_evlist__new();
+	__T("failed to create evlist", s_evlist);
+
+	for (i = 0; i < event_num; i++) {
+		evsel = perf_evsel__new(&attr);
+		__T("failed to create evsel", evsel);
+
+		perf_evlist__add(s_evlist, evsel);
+	}
+
+	perf_evlist__set_maps(s_evlist, NULL, threads);
+
+	err = perf_evlist__open_opts(s_evlist, &opts);
+	__T("failed to open evlist", err == 0);
+
+	i = 0;
+	perf_evlist__for_each_evsel(s_evlist, evsel) {
+		volatile unsigned int wait_count = WAIT_COUNT;
+
+		err = perf_evsel__refresh(evsel, 1);
+		__T("failed to refresh evsel", err == 0);
+
+		while (wait_count--)
+			;
+
+		__T_VERBOSE("Event %2d -- overflow flag = %#x, ",
+			    i, overflow_flag);
+		__T_VERBOSE("POLL_UP = %d, other signal event = %d\n",
+			    sig_count.hup, sig_count.others);
+
+		__T("unexpected event overflow detected", overflow_flag && (1U << i));
+		__T("unexpected signal event detected",
+		    sig_count.hup == 1 && sig_count.others == 0);
+
+		overflow_flag = 0;
+		sig_count.hup = 0;
+		sig_count.others = 0;
+		i++;
+	}
+
+	perf_evlist__close(s_evlist);
+	perf_evlist__delete(s_evlist);
+	perf_thread_map__put(threads);
+
+	return 0;
+}
+
 int test_evlist(int argc, char **argv)
 {
 	__T_START;
@@ -583,6 +709,7 @@ int test_evlist(int argc, char **argv)
 	test_mmap_thread();
 	test_mmap_cpus();
 	test_stat_multiplexing();
+	test_stat_overflow_event();
 
 	__T_END;
 	return tests_failed == 0 ? 0 : -1;
-- 
2.25.1

Re: [RFC PATCH v2 0/7] libperf: Add interface for overflow check of sampling events

[Jiri Olsa]

On Fri, Mar 25, 2022 at 01:38:22PM +0900, Shunsuke Nakamura wrote:
> This patch series adds interface for overflow check of sampling events
> to libperf.
> 
> First patch fixes a typo in the error message that I noticed.
> 
> Second patch move 'open_flags' from tools/perf to evsel::open_flags.
> 
> Third patch adds a interface to perform IOC_REFRESH and IOC_PERIOD.
> 
> Fourth patch introduce perf_{evsel, evlist}__open_opt() with extensible
> structure opts.
> 
> Fifth patch adds a interface to check overflowed events.
> 
> Sixth and seventh patch adds tests.
> 
> ---
> Previous version at:
> https://lore.kernel.org/linux-perf-users/20220225103114.144239-1-nakamura.shun@fujitsu.com/
> 
> Changes in v2:
>  - Delete perf_evsel__set_close_on_exec() function
>  - Introduce perf_{evsel, evlist}__open_opt() with extensible structure
>    opts
>  - Fix perf_evsel__set_signal() to a internal function
>  - Add bool type argument to perf_evsel__check_{fd, fd_cpu}() to indicate
>    overflow results

looks good.. it got more clear to me, sending some comments


> 
> 
> Shunsuke Nakamura (7):
>   libperf tests: Fix typo in the error message
>   libperf: Move 'open_flags' from tools/perf to evsel::open_flags
>   libperf: Add perf_evsel__{refresh, period}() functions
>   libperf: Introduce perf_{evsel, evlist}__open_opt with extensible
>     struct opts
>   libperf: Add perf_evsel__check_overflow() functions
>   libperf test: Add test_stat_overflow()
>   libperf test: Add test_stat_overflow_event()

I'm getting:

[root@krava perf]# make tests
running static:
- running tests/test-cpumap.c...OK
- running tests/test-threadmap.c...OK
- running tests/test-evlist.c...OK
- running tests/test-evsel.c...FAILED tests/test-evsel.c:286 failed to overflow count
FAILED tests/test-evsel.c:286 failed to overflow count
  FAILED (2)
FAILED tests/main.c:13 test evsel
make: *** [Makefile:162: tests] Error 255

thanks,
jirka

> 
>  tools/lib/perf/Documentation/libperf.txt |  22 +++
>  tools/lib/perf/evlist.c                  |  20 +++
>  tools/lib/perf/evsel.c                   | 210 ++++++++++++++++++++++-
>  tools/lib/perf/include/internal/evsel.h  |   2 +
>  tools/lib/perf/include/perf/evlist.h     |   3 +
>  tools/lib/perf/include/perf/evsel.h      |  35 ++++
>  tools/lib/perf/internal.h                |  44 +++++
>  tools/lib/perf/libperf.map               |   8 +
>  tools/lib/perf/tests/test-evlist.c       | 135 ++++++++++++++-
>  tools/lib/perf/tests/test-evsel.c        | 111 ++++++++++++
>  tools/perf/util/evsel.c                  |  16 +-
>  tools/perf/util/evsel.h                  |   1 -
>  12 files changed, 588 insertions(+), 19 deletions(-)
> 
> -- 
> 2.25.1
> 

RE: [RFC PATCH v2 0/7] libperf: Add interface for overflow check of sampling events

[nakamura.shun]

Hi jirka

Sorry for the late reply.

> >
> > Shunsuke Nakamura (7):
> >   libperf tests: Fix typo in the error message
> >   libperf: Move 'open_flags' from tools/perf to evsel::open_flags
> >   libperf: Add perf_evsel__{refresh, period}() functions
> >   libperf: Introduce perf_{evsel, evlist}__open_opt with extensible
> >     struct opts
> >   libperf: Add perf_evsel__check_overflow() functions
> >   libperf test: Add test_stat_overflow()
> >   libperf test: Add test_stat_overflow_event()
> 
> I'm getting:
> 
> [root@krava perf]# make tests
> running static:
> - running tests/test-cpumap.c...OK
> - running tests/test-threadmap.c...OK
> - running tests/test-evlist.c...OK
> - running tests/test-evsel.c...FAILED tests/test-evsel.c:286 failed to overflow
> count FAILED tests/test-evsel.c:286 failed to overflow count
>   FAILED (2)
> FAILED tests/main.c:13 test evsel
> make: *** [Makefile:162: tests] Error 255
> 
Thanks for telling me.
However, we could not reproduce the test failure in our environment.
Could you please tell me the results of your test with the addition of V=1?


Best Regards
Shunsuke

Re: [RFC PATCH v2 0/7] libperf: Add interface for overflow check of sampling events

[Jiri Olsa]

On Mon, Apr 11, 2022 at 08:23:54AM +0000, nakamura.shun@fujitsu.com wrote:
> Hi jirka
> 
> Sorry for the late reply.
> 
> > >
> > > Shunsuke Nakamura (7):
> > >   libperf tests: Fix typo in the error message
> > >   libperf: Move 'open_flags' from tools/perf to evsel::open_flags
> > >   libperf: Add perf_evsel__{refresh, period}() functions
> > >   libperf: Introduce perf_{evsel, evlist}__open_opt with extensible
> > >     struct opts
> > >   libperf: Add perf_evsel__check_overflow() functions
> > >   libperf test: Add test_stat_overflow()
> > >   libperf test: Add test_stat_overflow_event()
> > 
> > I'm getting:
> > 
> > [root@krava perf]# make tests
> > running static:
> > - running tests/test-cpumap.c...OK
> > - running tests/test-threadmap.c...OK
> > - running tests/test-evlist.c...OK
> > - running tests/test-evsel.c...FAILED tests/test-evsel.c:286 failed to overflow
> > count FAILED tests/test-evsel.c:286 failed to overflow count
> >   FAILED (2)
> > FAILED tests/main.c:13 test evsel
> > make: *** [Makefile:162: tests] Error 255
> > 
> Thanks for telling me.
> However, we could not reproduce the test failure in our environment.
> Could you please tell me the results of your test with the addition of V=1?

sorry, forgot to answer this one..

[root@krava perf]# LD_LIBRARY_PATH=. ./tests-shared -v
- running tests/test-cpumap.c...OK
- running tests/test-threadmap.c...OK
- running tests/test-evlist.c...
Event  0 -- Raw count = 231217013, run = 10196083, enable = 22183091
         Scaled count = 503046909 (45.96%, 10196083/22183091)
Event  1 -- Raw count = 247219790, run = 10946246, enable = 22182044
         Scaled count = 500979080 (49.35%, 10946246/22182044)
Event  2 -- Raw count = 268999685, run = 11943490, enable = 22180868
         Scaled count = 499573115 (53.85%, 11943490/22180868)
Event  3 -- Raw count = 291383974, run = 12941069, enable = 22179798
         Scaled count = 499405241 (58.35%, 12941069/22179798)
Event  4 -- Raw count = 314293846, run = 13938576, enable = 22178649
         Scaled count = 500095052 (62.85%, 13938576/22178649)
Event  5 -- Raw count = 337393782, run = 14935798, enable = 22177192
         Scaled count = 500974014 (67.35%, 14935798/22177192)
Event  6 -- Raw count = 360218294, run = 15933314, enable = 22175507
         Scaled count = 501340982 (71.85%, 15933314/22175507)
Event  7 -- Raw count = 365516630, run = 16179860, enable = 22173686
         Scaled count = 500922194 (72.97%, 16179860/22173686)
Event  8 -- Raw count = 365180177, run = 16178058, enable = 22171705
         Scaled count = 500472130 (72.97%, 16178058/22171705)
Event  9 -- Raw count = 341205844, run = 15233252, enable = 22169603
         Scaled count = 496571454 (68.71%, 15233252/22169603)
Event 10 -- Raw count = 318204949, run = 14234754, enable = 22167081
         Scaled count = 495524887 (64.22%, 14234754/22167081)
Event 11 -- Raw count = 296671131, run = 13236210, enable = 22164791
         Scaled count = 496792783 (59.72%, 13236210/22164791)
Event 12 -- Raw count = 276639189, run = 12237128, enable = 22162629
         Scaled count = 501020477 (55.22%, 12237128/22162629)
Event 13 -- Raw count = 255437518, run = 11238588, enable = 22160497
         Scaled count = 503677361 (50.71%, 11238588/22160497)
Event 14 -- Raw count = 232427791, run = 10239844, enable = 22158353
         Scaled count = 502958545 (46.21%, 10239844/22158353)
   Expected: 500221918
   High: 503677361   Low:  231217013   Average:  500223614
   Average Error = 0.00%
Event  0 -- overflow flag = 0x1, POLL_UP = 1, other signal event = 0
Event  1 -- overflow flag = 0x2, POLL_UP = 1, other signal event = 0
Event  2 -- overflow flag = 0x4, POLL_UP = 1, other signal event = 0
Event  3 -- overflow flag = 0x8, POLL_UP = 1, other signal event = 0
OK
- running tests/test-evsel.c...
        loop = 65536, count = 328183
        loop = 131072, count = 655863
        loop = 262144, count = 1320613
        loop = 524288, count = 2621943
        loop = 1048576, count = 5243383
        loop = 65536, count = 66349
        loop = 131072, count = 131789
        loop = 262144, count = 264697
        loop = 524288, count = 528484
        loop = 1048576, count = 1056485
        period = 1000000
        overflow limit = 3, overflow count = 2, POLL_IN = 2, POLL_UP = 0, other signal event = 0
FAILED tests/test-evsel.c:286 failed to overflow count
        period = 1000000
        overflow limit = 3, overflow count = 5, POLL_IN = 4, POLL_UP = 1, other signal event = 0
FAILED tests/test-evsel.c:286 failed to overflow count
  FAILED (2)
FAILED tests/main.c:13 test evsel


jirka

RE: [RFC PATCH v2 0/7] libperf: Add interface for overflow check of sampling events

[nakamura.shun]

Hi jirka

> On Mon, Apr 11, 2022 at 08:23:54AM +0000, nakamura.shun@fujitsu.com wrote:
> > Hi jirka
> >
> > Sorry for the late reply.
> >
> > > >
> > > > Shunsuke Nakamura (7):
> > > >   libperf tests: Fix typo in the error message
> > > >   libperf: Move 'open_flags' from tools/perf to evsel::open_flags
> > > >   libperf: Add perf_evsel__{refresh, period}() functions
> > > >   libperf: Introduce perf_{evsel, evlist}__open_opt with extensible
> > > >     struct opts
> > > >   libperf: Add perf_evsel__check_overflow() functions
> > > >   libperf test: Add test_stat_overflow()
> > > >   libperf test: Add test_stat_overflow_event()
> > >
> > > I'm getting:
> > >
> > > [root@krava perf]# make tests
> > > running static:
> > > - running tests/test-cpumap.c...OK
> > > - running tests/test-threadmap.c...OK
> > > - running tests/test-evlist.c...OK
> > > - running tests/test-evsel.c...FAILED tests/test-evsel.c:286 failed
> > > to overflow count FAILED tests/test-evsel.c:286 failed to overflow count
> > >   FAILED (2)
> > > FAILED tests/main.c:13 test evsel
> > > make: *** [Makefile:162: tests] Error 255
> > >
> > Thanks for telling me.
> > However, we could not reproduce the test failure in our environment.
> > Could you please tell me the results of your test with the addition of V=1?
> 
> sorry, forgot to answer this one..
> 
Thanks for the information.


> [root@krava perf]# LD_LIBRARY_PATH=. ./tests-shared -v
> - running tests/test-cpumap.c...OK
> - running tests/test-threadmap.c...OK
> - running tests/test-evlist.c...
> Event  0 -- Raw count = 231217013, run = 10196083, enable = 22183091
>          Scaled count = 503046909 (45.96%, 10196083/22183091) Event  1 --
> Raw count = 247219790, run = 10946246, enable = 22182044
>          Scaled count = 500979080 (49.35%, 10946246/22182044) Event  2 --
> Raw count = 268999685, run = 11943490, enable = 22180868
>          Scaled count = 499573115 (53.85%, 11943490/22180868) Event  3 --
> Raw count = 291383974, run = 12941069, enable = 22179798
>          Scaled count = 499405241 (58.35%, 12941069/22179798) Event  4 --
> Raw count = 314293846, run = 13938576, enable = 22178649
>          Scaled count = 500095052 (62.85%, 13938576/22178649) Event  5 --
> Raw count = 337393782, run = 14935798, enable = 22177192
>          Scaled count = 500974014 (67.35%, 14935798/22177192) Event  6 --
> Raw count = 360218294, run = 15933314, enable = 22175507
>          Scaled count = 501340982 (71.85%, 15933314/22175507) Event  7 --
> Raw count = 365516630, run = 16179860, enable = 22173686
>          Scaled count = 500922194 (72.97%, 16179860/22173686) Event  8 --
> Raw count = 365180177, run = 16178058, enable = 22171705
>          Scaled count = 500472130 (72.97%, 16178058/22171705) Event  9 --
> Raw count = 341205844, run = 15233252, enable = 22169603
>          Scaled count = 496571454 (68.71%, 15233252/22169603) Event 10 --
> Raw count = 318204949, run = 14234754, enable = 22167081
>          Scaled count = 495524887 (64.22%, 14234754/22167081) Event 11 --
> Raw count = 296671131, run = 13236210, enable = 22164791
>          Scaled count = 496792783 (59.72%, 13236210/22164791) Event 12 --
> Raw count = 276639189, run = 12237128, enable = 22162629
>          Scaled count = 501020477 (55.22%, 12237128/22162629) Event 13 --
> Raw count = 255437518, run = 11238588, enable = 22160497
>          Scaled count = 503677361 (50.71%, 11238588/22160497) Event 14 --
> Raw count = 232427791, run = 10239844, enable = 22158353
>          Scaled count = 502958545 (46.21%, 10239844/22158353)
>    Expected: 500221918
>    High: 503677361   Low:  231217013   Average:  500223614
>    Average Error = 0.00%
> Event  0 -- overflow flag = 0x1, POLL_UP = 1, other signal event = 0 Event  1 --
> overflow flag = 0x2, POLL_UP = 1, other signal event = 0 Event  2 -- overflow
> flag = 0x4, POLL_UP = 1, other signal event = 0 Event  3 -- overflow flag = 0x8,
> POLL_UP = 1, other signal event = 0 OK
> - running tests/test-evsel.c...
>         loop = 65536, count = 328183
>         loop = 131072, count = 655863
>         loop = 262144, count = 1320613
>         loop = 524288, count = 2621943
>         loop = 1048576, count = 5243383
>         loop = 65536, count = 66349
>         loop = 131072, count = 131789
>         loop = 262144, count = 264697
>         loop = 524288, count = 528484
>         loop = 1048576, count = 1056485
>         period = 1000000
>         overflow limit = 3, overflow count = 2, POLL_IN = 2, POLL_UP = 0, other
> signal event = 0 FAILED tests/test-evsel.c:286 failed to overflow count

This test was using standard signals (SIGIO), but after changing to real-time signals,
I confirmed that the test succeeded.
When using standard signals, it seems that if there is an overlapping timing of standard signals,
the signals generated later will be dropped.


>         period = 1000000
>         overflow limit = 3, overflow count = 5, POLL_IN = 4, POLL_UP = 1, other
> signal event = 0 FAILED tests/test-evsel.c:286 failed to overflow count

Fixed implementation of overflow counter not clearing on test failure.


>   FAILED (2)
> FAILED tests/main.c:13 test evsel
> 
I will post a v3 patch with the above fixes.

Best Regards
Shunsuke