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

[PATCH 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 move 'open_flags' from tools/perf to evsel::open_flags.

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

Third patch add support for overflow handling of sampling events.

Fourth patch adds a interface to check overflowed events.

Fifth patch adds a interface to perform IOC_REFRESH and IOC_PERIOD.

Sixth and seventh patch adds tests.


---
Previous version at:
https://lore.kernel.org/linux-perf-users/20220325043829.224045-1-nakamura.shun@fujitsu.com/

Changes in v1:
 - Move initialization/reference of evsel->open_flags from the first
   patch to the second patch
 - Move signal-related handling and related fields of the opts
   structure from the second patch to the third patch
 - Move _GNU_SOURCE from test-evlist.c to Makefile
 - Delete *_cpu() function
 - Refactor the fourth patch
 - Fix test to use real-time signals instead of standard signals

Changes in RFC 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: Move 'open_flags' from tools/perf to evsel::open_flags
  libperf: Introduce perf_{evsel, evlist}__open_opt with extensible
    struct opts
  libperf: Add support for overflow handling of sampling events
  libperf: Add perf_evsel__has_fd() functions
  libperf: Add perf_evsel__{refresh, period}() functions
  libperf test: Add test_stat_overflow()
  libperf test: Add test_stat_overflow_event()

 tools/lib/perf/Documentation/libperf.txt |  17 +++
 tools/lib/perf/Makefile                  |   1 +
 tools/lib/perf/evlist.c                  |  20 +++
 tools/lib/perf/evsel.c                   | 166 ++++++++++++++++++++++-
 tools/lib/perf/include/internal/evsel.h  |   2 +
 tools/lib/perf/include/perf/evlist.h     |   3 +
 tools/lib/perf/include/perf/evsel.h      |  29 ++++
 tools/lib/perf/internal.h                |  44 ++++++
 tools/lib/perf/libperf.map               |   5 +
 tools/lib/perf/tests/test-evlist.c       | 112 ++++++++++++++-
 tools/lib/perf/tests/test-evsel.c        | 108 +++++++++++++++
 tools/perf/util/evsel.c                  |  16 ++-
 tools/perf/util/evsel.h                  |   1 -
 13 files changed, 508 insertions(+), 16 deletions(-)

-- 
2.25.1

[PATCH 1/7] 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/include/internal/evsel.h |  2 ++
 tools/perf/util/evsel.c                 | 16 +++++++++-------
 tools/perf/util/evsel.h                 |  1 -
 3 files changed, 11 insertions(+), 8 deletions(-)

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 2a1729e7aee4..92f7a76ca9d5 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

[PATCH 2/7] 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. Currently, only open_flags is supported for the
opts structure.

Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
---
 tools/lib/perf/Documentation/libperf.txt | 10 ++++++
 tools/lib/perf/evlist.c                  | 20 +++++++++++
 tools/lib/perf/evsel.c                   | 25 +++++++++++++-
 tools/lib/perf/include/perf/evlist.h     |  3 ++
 tools/lib/perf/include/perf/evsel.h      | 22 ++++++++++++
 tools/lib/perf/internal.h                | 44 ++++++++++++++++++++++++
 tools/lib/perf/libperf.map               |  2 ++
 7 files changed, 125 insertions(+), 1 deletion(-)

diff --git a/tools/lib/perf/Documentation/libperf.txt b/tools/lib/perf/Documentation/libperf.txt
index a8f1a237931b..6ca91ca94e01 100644
--- a/tools/lib/perf/Documentation/libperf.txt
+++ b/tools/lib/perf/Documentation/libperf.txt
@@ -132,6 +132,16 @@ 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 */
+  };
+  #define perf_evsel_open_opts__last_field open_flags
+
+  #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 a09315538a30..bf9343acb799 100644
--- a/tools/lib/perf/evlist.c
+++ b/tools/lib/perf/evlist.c
@@ -665,3 +665,23 @@ 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 210ea7c06ce8..00c0cea43b52 100644
--- a/tools/lib/perf/evsel.c
+++ b/tools/lib/perf/evsel.c
@@ -16,8 +16,12 @@
 #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)
@@ -26,6 +30,7 @@ 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 +163,7 @@ 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;
@@ -454,3 +459,21 @@ void perf_counts_values__scale(struct perf_counts_values *count,
 	if (pscaled)
 		*pscaled = scaled;
 }
+
+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);
+
+	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 2a9516b42d15..1140df4d2578 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 +24,23 @@ 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 */
+};
+#define perf_evsel_open_opts__last_field open_flags
+
+#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,
@@ -43,5 +61,9 @@ 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..4b91a087ed62 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 190b56ae923a..eeeb3075e092 100644
--- a/tools/lib/perf/libperf.map
+++ b/tools/lib/perf/libperf.map
@@ -25,6 +25,7 @@ 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;
@@ -36,6 +37,7 @@ 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

[PATCH 3/7] libperf: Add support for overflow handling of sampling events

[Shunsuke Nakamura]

Extend the fields of the opts structure to set up overflow handling
for sampling events. Also, add processing to set signal handlers in
perf_evsel__open_opts.

Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
---
 tools/lib/perf/Documentation/libperf.txt |  6 +-
 tools/lib/perf/Makefile                  |  1 +
 tools/lib/perf/evsel.c                   | 79 ++++++++++++++++++++++++
 tools/lib/perf/include/perf/evsel.h      |  6 +-
 tools/lib/perf/tests/test-evlist.c       |  1 -
 5 files changed, 90 insertions(+), 3 deletions(-)

diff --git a/tools/lib/perf/Documentation/libperf.txt b/tools/lib/perf/Documentation/libperf.txt
index 6ca91ca94e01..ec93b1c75ebe 100644
--- a/tools/lib/perf/Documentation/libperf.txt
+++ b/tools/lib/perf/Documentation/libperf.txt
@@ -137,8 +137,12 @@ SYNOPSIS
           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 open_flags
+  #define perf_evsel_open_opts__last_field sig
 
   #define LIBPERF_OPTS(TYPE, NAME, ...)
 
diff --git a/tools/lib/perf/Makefile b/tools/lib/perf/Makefile
index 21df023a2103..4c8fae193cf5 100644
--- a/tools/lib/perf/Makefile
+++ b/tools/lib/perf/Makefile
@@ -75,6 +75,7 @@ override CFLAGS += -Werror -Wall
 override CFLAGS += -fPIC
 override CFLAGS += $(INCLUDES)
 override CFLAGS += -fvisibility=hidden
+override CFLAGS += -D_GNU_SOURCE
 
 all:
 
diff --git a/tools/lib/perf/evsel.c b/tools/lib/perf/evsel.c
index 00c0cea43b52..a289f6c44d7c 100644
--- a/tools/lib/perf/evsel.c
+++ b/tools/lib/perf/evsel.c
@@ -460,6 +460,79 @@ void perf_counts_values__scale(struct perf_counts_values *count,
 		*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)
@@ -474,6 +547,12 @@ int perf_evsel__open_opts(struct perf_evsel *evsel, struct perf_cpu_map *cpus,
 	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/evsel.h b/tools/lib/perf/include/perf/evsel.h
index 1140df4d2578..50662babfe97 100644
--- a/tools/lib/perf/include/perf/evsel.h
+++ b/tools/lib/perf/include/perf/evsel.h
@@ -29,8 +29,12 @@ struct perf_evsel_open_opts {
 	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 open_flags
+#define perf_evsel_open_opts__last_field sig
 
 #define LIBPERF_OPTS(TYPE, NAME, ...)			\
 	struct TYPE NAME = ({				\
diff --git a/tools/lib/perf/tests/test-evlist.c b/tools/lib/perf/tests/test-evlist.c
index ed616fc19b4f..d8f9493cd4d1 100644
--- a/tools/lib/perf/tests/test-evlist.c
+++ b/tools/lib/perf/tests/test-evlist.c
@@ -1,5 +1,4 @@
 // SPDX-License-Identifier: GPL-2.0
-#define _GNU_SOURCE // needed for sched.h to get sched_[gs]etaffinity and CPU_(ZERO,SET)
 #include <inttypes.h>
 #include <sched.h>
 #include <stdio.h>
-- 
2.25.1

[PATCH 4/7] libperf: Add perf_evsel__has_fd() functions

[Shunsuke Nakamura]

Add the following functions:

  perf_evsel__has_fd

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

Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
---
 tools/lib/perf/Documentation/libperf.txt |  1 +
 tools/lib/perf/evsel.c                   | 18 ++++++++++++++++++
 tools/lib/perf/include/perf/evsel.h      |  1 +
 tools/lib/perf/libperf.map               |  1 +
 4 files changed, 21 insertions(+)

diff --git a/tools/lib/perf/Documentation/libperf.txt b/tools/lib/perf/Documentation/libperf.txt
index ec93b1c75ebe..bc7881348c76 100644
--- a/tools/lib/perf/Documentation/libperf.txt
+++ b/tools/lib/perf/Documentation/libperf.txt
@@ -161,6 +161,7 @@ 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);
+  bool perf_evsel__has_fd(struct perf_evsel *evsel, int fd);
   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 a289f6c44d7c..446934c0e5e5 100644
--- a/tools/lib/perf/evsel.c
+++ b/tools/lib/perf/evsel.c
@@ -556,3 +556,21 @@ int perf_evsel__open_opts(struct perf_evsel *evsel, struct perf_cpu_map *cpus,
 
 	return err;
 }
+
+bool perf_evsel__has_fd(struct perf_evsel *evsel, int fd)
+{
+	int cpu_map_idx;
+	int thread;
+	int *evsel_fd;
+
+	for (cpu_map_idx = 0; cpu_map_idx < xyarray__max_x(evsel->fd); ++cpu_map_idx) {
+		for (thread = 0; thread < xyarray__max_y(evsel->fd); ++thread) {
+			evsel_fd = FD(evsel, cpu_map_idx, thread);
+
+			if (fd == *evsel_fd)
+				return true;
+		}
+	}
+
+	return false;
+}
diff --git a/tools/lib/perf/include/perf/evsel.h b/tools/lib/perf/include/perf/evsel.h
index 50662babfe97..19a7993d9021 100644
--- a/tools/lib/perf/include/perf/evsel.h
+++ b/tools/lib/perf/include/perf/evsel.h
@@ -69,5 +69,6 @@ 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 bool perf_evsel__has_fd(struct perf_evsel *evsel, int fd);
 
 #endif /* __LIBPERF_EVSEL_H */
diff --git a/tools/lib/perf/libperf.map b/tools/lib/perf/libperf.map
index eeeb3075e092..b2ace16bbc32 100644
--- a/tools/lib/perf/libperf.map
+++ b/tools/lib/perf/libperf.map
@@ -34,6 +34,7 @@ LIBPERF_0.0.1 {
 		perf_evsel__cpus;
 		perf_evsel__threads;
 		perf_evsel__attr;
+		perf_evsel__has_fd;
 		perf_evlist__new;
 		perf_evlist__delete;
 		perf_evlist__open;
-- 
2.25.1

[PATCH 5/7] libperf: Add perf_evsel__{refresh, period}() functions

[Shunsuke Nakamura]

Add the following functions:

  perf_evsel__refresh()
  perf_evsel__period()

to set the over flow limit and period.

Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
---
 tools/lib/perf/Documentation/libperf.txt |  2 ++
 tools/lib/perf/evsel.c                   | 44 ++++++++++++++++++++----
 tools/lib/perf/include/perf/evsel.h      |  2 ++
 tools/lib/perf/libperf.map               |  2 ++
 4 files changed, 44 insertions(+), 6 deletions(-)

diff --git a/tools/lib/perf/Documentation/libperf.txt b/tools/lib/perf/Documentation/libperf.txt
index bc7881348c76..773224a96bd3 100644
--- a/tools/lib/perf/Documentation/libperf.txt
+++ b/tools/lib/perf/Documentation/libperf.txt
@@ -162,6 +162,8 @@ SYNOPSIS
   int perf_evsel__disable(struct perf_evsel *evsel);
   int perf_evsel__disable_cpu(struct perf_evsel *evsel, int cpu_map_idx);
   bool perf_evsel__has_fd(struct perf_evsel *evsel, int fd);
+  int perf_evsel__refresh(struct perf_evsel *evsel, int refresh);
+  int perf_evsel__period(struct perf_evsel *evsel, int period);
   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 446934c0e5e5..578ae0050036 100644
--- a/tools/lib/perf/evsel.c
+++ b/tools/lib/perf/evsel.c
@@ -334,7 +334,7 @@ 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;
@@ -357,7 +357,7 @@ 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)
@@ -366,13 +366,13 @@ 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)
@@ -381,7 +381,39 @@ 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(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(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;
 }
 
@@ -392,7 +424,7 @@ 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 19a7993d9021..2c5e52c17d28 100644
--- a/tools/lib/perf/include/perf/evsel.h
+++ b/tools/lib/perf/include/perf/evsel.h
@@ -60,6 +60,8 @@ 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__period(struct perf_evsel *evsel, __u64 period);
 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 b2ace16bbc32..d97e208c5be8 100644
--- a/tools/lib/perf/libperf.map
+++ b/tools/lib/perf/libperf.map
@@ -31,6 +31,8 @@ LIBPERF_0.0.1 {
 		perf_evsel__munmap;
 		perf_evsel__mmap_base;
 		perf_evsel__read;
+		perf_evsel__refresh;
+		perf_evsel__period;
 		perf_evsel__cpus;
 		perf_evsel__threads;
 		perf_evsel__attr;
-- 
2.25.1

[PATCH 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 | 108 ++++++++++++++++++++++++++++++
 1 file changed, 108 insertions(+)

diff --git a/tools/lib/perf/tests/test-evsel.c b/tools/lib/perf/tests/test-evsel.c
index 89be89afb24d..4de5d5554836 100644
--- a/tools/lib/perf/tests/test-evsel.c
+++ b/tools/lib/perf/tests/test-evsel.c
@@ -1,6 +1,11 @@
 // SPDX-License-Identifier: GPL-2.0
+#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 +13,15 @@
 #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;
+static struct perf_evsel *s_evsel;
+
 static int libperf_print(enum libperf_print_level level,
 			 const char *fmt, va_list ap)
 {
@@ -189,6 +203,98 @@ 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	 = SIGRTMIN + 1,
+		     .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;
+
+		sig_count.in = 0;
+		sig_count.hup = 0;
+		sig_count.others = 0;
+		sig_count.overflow = 0;
+
+		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_HUP = %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);
+	}
+
+	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 +306,8 @@ 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

[PATCH 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 | 111 +++++++++++++++++++++++++++++
 1 file changed, 111 insertions(+)

diff --git a/tools/lib/perf/tests/test-evlist.c b/tools/lib/perf/tests/test-evlist.c
index d8f9493cd4d1..587364851432 100644
--- a/tools/lib/perf/tests/test-evlist.c
+++ b/tools/lib/perf/tests/test-evlist.c
@@ -5,6 +5,8 @@
 #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>
@@ -24,6 +26,13 @@
 #define EVENT_NUM 15
 #define WAIT_COUNT 100000000UL
 
+static unsigned int overflow_flag;
+static struct signal_counts {
+	int hup;
+	int others;
+} sig_count;
+static struct perf_evlist *s_evlist;
+
 static int libperf_print(enum libperf_print_level level,
 			 const char *fmt, va_list ap)
 {
@@ -570,6 +579,107 @@ static int test_stat_multiplexing(void)
 	return 0;
 }
 
+static void sig_handler(int signo, siginfo_t *info, void *uc)
+{
+	struct perf_evsel *evsel;
+	int i = 0;
+
+	perf_evlist__for_each_evsel(s_evlist, evsel) {
+		if (perf_evsel__has_fd(evsel, info->si_fd)) {
+			if (info->si_code == POLL_HUP)
+				sig_count.hup++;
+			else
+				sig_count.others++;
+
+			overflow_flag = (1U << i);
+			return;
+		}
+		i++;
+	}
+
+	__T_VERBOSE("Failed to get fd of overflowed event");
+}
+
+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	 = SIGRTMIN + 1,
+		     .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;
+
+		overflow_flag = 0;
+		sig_count.hup = 0;
+		sig_count.others = 0;
+
+		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_HUP = %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);
+
+		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;
@@ -582,6 +692,7 @@ 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: [PATCH 2/7] libperf: Introduce perf_{evsel, evlist}__open_opt with extensible struct opts

[Namhyung Kim]

Hello,

On Fri, Apr 22, 2022 at 2:44 AM Shunsuke Nakamura
<nakamura.shun@fujitsu.com> wrote:
>
> Introduce perf_{evsel, evlist}__open_opt with extensible structure opts.
> The mechanism of the extensible structure opts imitates
> tools/lib/bpf/libbpf.h. Currently, only open_flags is supported for the
> opts structure.
>
> Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
> ---
[SNIP]
> diff --git a/tools/lib/perf/evsel.c b/tools/lib/perf/evsel.c
> index 210ea7c06ce8..00c0cea43b52 100644
> --- a/tools/lib/perf/evsel.c
> +++ b/tools/lib/perf/evsel.c
> @@ -16,8 +16,12 @@
>  #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)
> @@ -26,6 +30,7 @@ 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;

In general, you don't need to reset it to zero as it's allocated with
zalloc().

>  }
>
[SNIP]
> diff --git a/tools/lib/perf/internal.h b/tools/lib/perf/internal.h
> index 2c27e158de6b..4b91a087ed62 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)) {

I don't think the cast is necessary since it'll be promoted again
to size_t due to opts_sz.  Instead, we can check if user_sz is
greater than opts_sz explicitly and call the function.

> +               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)

It'd be nice if you add a blank line between the macros.

Thanks,
Namhyung


> +#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 190b56ae923a..eeeb3075e092 100644
> --- a/tools/lib/perf/libperf.map
> +++ b/tools/lib/perf/libperf.map
> @@ -25,6 +25,7 @@ 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;
> @@ -36,6 +37,7 @@ 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: [PATCH 3/7] libperf: Add support for overflow handling of sampling events

[Namhyung Kim]

On Fri, Apr 22, 2022 at 2:44 AM Shunsuke Nakamura
<nakamura.shun@fujitsu.com> wrote:
>
> Extend the fields of the opts structure to set up overflow handling
> for sampling events. Also, add processing to set signal handlers in
> perf_evsel__open_opts.
>
> Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
> ---
>  tools/lib/perf/Documentation/libperf.txt |  6 +-
>  tools/lib/perf/Makefile                  |  1 +
>  tools/lib/perf/evsel.c                   | 79 ++++++++++++++++++++++++
>  tools/lib/perf/include/perf/evsel.h      |  6 +-
>  tools/lib/perf/tests/test-evlist.c       |  1 -
>  5 files changed, 90 insertions(+), 3 deletions(-)
>
> diff --git a/tools/lib/perf/Documentation/libperf.txt b/tools/lib/perf/Documentation/libperf.txt
> index 6ca91ca94e01..ec93b1c75ebe 100644
> --- a/tools/lib/perf/Documentation/libperf.txt
> +++ b/tools/lib/perf/Documentation/libperf.txt
> @@ -137,8 +137,12 @@ SYNOPSIS
>            size_t sz;
>
>            unsigned long open_flags;       /* perf_event_open flags */
> +          int flags;                      /* fcntl flags */

What about just rename it to fcntl_flags ?

> +          unsigned int signal;

int ?

> +          int owner_type;

Please add a comment that it's for F_SETOWN_EX.

> +          struct sigaction *sig;
>    };
> -  #define perf_evsel_open_opts__last_field open_flags
> +  #define perf_evsel_open_opts__last_field sig
>
>    #define LIBPERF_OPTS(TYPE, NAME, ...)
>
> diff --git a/tools/lib/perf/Makefile b/tools/lib/perf/Makefile
> index 21df023a2103..4c8fae193cf5 100644
> --- a/tools/lib/perf/Makefile
> +++ b/tools/lib/perf/Makefile
> @@ -75,6 +75,7 @@ override CFLAGS += -Werror -Wall
>  override CFLAGS += -fPIC
>  override CFLAGS += $(INCLUDES)
>  override CFLAGS += -fvisibility=hidden
> +override CFLAGS += -D_GNU_SOURCE
>
>  all:
>
> diff --git a/tools/lib/perf/evsel.c b/tools/lib/perf/evsel.c
> index 00c0cea43b52..a289f6c44d7c 100644
> --- a/tools/lib/perf/evsel.c
> +++ b/tools/lib/perf/evsel.c
> @@ -460,6 +460,79 @@ void perf_counts_values__scale(struct perf_counts_values *count,
>                 *pscaled = scaled;
>  }
>
> +static int perf_evsel__run_fcntl(struct perf_evsel *evsel,
> +                                unsigned int cmd, unsigned long arg,
> +                                int cpu_map_idx)

I think it'd be better to have _cpu suffix in the function name
as it handles on the specific cpu.

> +{
> +       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;

It can be confused with the signal number, how about sigact?


> +       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)

You man want to check owner.type without "!".

> +               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)
> @@ -474,6 +547,12 @@ int perf_evsel__open_opts(struct perf_evsel *evsel, struct perf_cpu_map *cpus,
>         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/evsel.h b/tools/lib/perf/include/perf/evsel.h
> index 1140df4d2578..50662babfe97 100644
> --- a/tools/lib/perf/include/perf/evsel.h
> +++ b/tools/lib/perf/include/perf/evsel.h
> @@ -29,8 +29,12 @@ struct perf_evsel_open_opts {
>         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 open_flags
> +#define perf_evsel_open_opts__last_field sig
>
>  #define LIBPERF_OPTS(TYPE, NAME, ...)                  \
>         struct TYPE NAME = ({                           \
> diff --git a/tools/lib/perf/tests/test-evlist.c b/tools/lib/perf/tests/test-evlist.c
> index ed616fc19b4f..d8f9493cd4d1 100644
> --- a/tools/lib/perf/tests/test-evlist.c
> +++ b/tools/lib/perf/tests/test-evlist.c
> @@ -1,5 +1,4 @@
>  // SPDX-License-Identifier: GPL-2.0
> -#define _GNU_SOURCE // needed for sched.h to get sched_[gs]etaffinity and CPU_(ZERO,SET)

Could be in a separate change.

Thanks,
Namhyung


>  #include <inttypes.h>
>  #include <sched.h>
>  #include <stdio.h>
> --
> 2.25.1
>

Re: [PATCH 5/7] libperf: Add perf_evsel__{refresh, period}() functions

[Namhyung Kim]

On Fri, Apr 22, 2022 at 2:45 AM Shunsuke Nakamura
<nakamura.shun@fujitsu.com> wrote:
>
> Add the following functions:
>
>   perf_evsel__refresh()
>   perf_evsel__period()
>
> to set the over flow limit and period.
>
> Signed-off-by: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
> ---
>  tools/lib/perf/Documentation/libperf.txt |  2 ++
>  tools/lib/perf/evsel.c                   | 44 ++++++++++++++++++++----
>  tools/lib/perf/include/perf/evsel.h      |  2 ++
>  tools/lib/perf/libperf.map               |  2 ++
>  4 files changed, 44 insertions(+), 6 deletions(-)
>
> diff --git a/tools/lib/perf/Documentation/libperf.txt b/tools/lib/perf/Documentation/libperf.txt
> index bc7881348c76..773224a96bd3 100644
> --- a/tools/lib/perf/Documentation/libperf.txt
> +++ b/tools/lib/perf/Documentation/libperf.txt
> @@ -162,6 +162,8 @@ SYNOPSIS
>    int perf_evsel__disable(struct perf_evsel *evsel);
>    int perf_evsel__disable_cpu(struct perf_evsel *evsel, int cpu_map_idx);
>    bool perf_evsel__has_fd(struct perf_evsel *evsel, int fd);
> +  int perf_evsel__refresh(struct perf_evsel *evsel, int refresh);
> +  int perf_evsel__period(struct perf_evsel *evsel, int period);
>    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 446934c0e5e5..578ae0050036 100644
> --- a/tools/lib/perf/evsel.c
> +++ b/tools/lib/perf/evsel.c
> @@ -334,7 +334,7 @@ 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;
> @@ -357,7 +357,7 @@ 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)
> @@ -366,13 +366,13 @@ 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)
> @@ -381,7 +381,39 @@ 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(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(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;

I don't think it'd return NULL..

Thanks,
Namhyung


> +
> +       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;
>  }
>
> @@ -392,7 +424,7 @@ 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 19a7993d9021..2c5e52c17d28 100644
> --- a/tools/lib/perf/include/perf/evsel.h
> +++ b/tools/lib/perf/include/perf/evsel.h
> @@ -60,6 +60,8 @@ 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__period(struct perf_evsel *evsel, __u64 period);
>  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 b2ace16bbc32..d97e208c5be8 100644
> --- a/tools/lib/perf/libperf.map
> +++ b/tools/lib/perf/libperf.map
> @@ -31,6 +31,8 @@ LIBPERF_0.0.1 {
>                 perf_evsel__munmap;
>                 perf_evsel__mmap_base;
>                 perf_evsel__read;
> +               perf_evsel__refresh;
> +               perf_evsel__period;
>                 perf_evsel__cpus;
>                 perf_evsel__threads;
>                 perf_evsel__attr;
> --
> 2.25.1
>