Skip to contentSkip to conversation

perf tools: Update pmu scan using openat() (v1)

PATCH Applied Namhyung Kim 5 participants 19 messages Mar 31, 2023 – Apr 4, 2023 perf events

lore
Share & tools
lore

Namhyung Kim

author maintainer
#0 lore raw

cover perf tools: Update pmu scan using openat() (v1)

Hello,

This patchset changes PMU info scanning on sysfs using openat() basically. I got reports of occasional contention on the opening files in sysfs. While the root cause was a separate issue, I discovered some inefficiencies in the perf code.

To scan PMUs, it roughly does something like below:

  dir = opendir("/sys/bus/event_source/devices");
  while (dentry = readdir(dir)) {
    char buf[PATH_MAX];
    snprintf(buf, sizeof(buf), "%s/%s",
             "/sys/bus/event_source/devices", dentry->d_name);
    fd = open(buf, O_RDONLY);
    ...

}

But this is not good since it needs to copy the string to build the absolute pathname, and it makes redundant pathname walk (from the /sys) in the kernel unnecessarily. We can use openat(2) to open the file in the given directory.

Add a couple of new helper to return the file descriptor of PMU directory so that it can use it with relative paths.

* perf_pmu__event_source_devices_fd()
- returns a fd for the PMU root ("/sys/bus/event_source/devices")

* perf_pmu__pathname_fd()
- returns a fd for "<pmu>/<file>" under the PMU root

Now the above code can be converted something like below:

  dirfd = perf_pmu__event_source_devices_fd();
  dir = fdopendir(dirfd);
  while (dentry = readdir(dir)) {
    fd = openat(dirfd, dentry->d_name, O_RDONLY);
    ...
  }

I added a benchmark for pmu-scan and it showed a slight speedup in the normal case too.

  $ ./perf.old bench internals pmu-scan
  # Running 'internals/pmu-scan' benchmark:
  Computing performance of sysfs PMU event scan for 100 times
    Average PMU scanning took: 6670.970 usec (+- 13.022 usec)
  $ ./perf.new bench internals pmu-scan
  # Running 'internals/pmu-scan' benchmark:
  Computing performance of sysfs PMU event scan for 100 times
    Average PMU scanning took: 6296.980 usec (+- 14.891 usec)

The 5~6% of improvement might be small but it may have bigger impact when the system is contended.

You can get the code from 'perf/pmu-scan-v1' branch in

git://git.kernel.org/pub/scm/linux/kernel/git/namhyung/linux-perf.git

Thanks,
Namhyung

Namhyung Kim (9):
perf list: Use relative path for tracepoint scan perf tools: Fix a asan issue in parse_events_multi_pmu_add() perf pmu: Add perf_pmu__destroy() function
perf bench: Add pmu-scan benchmark
perf pmu: Use relative path for sysfs scan
perf pmu: Use relative path in perf_pmu__caps_parse() perf pmu: Use relative path in setup_pmu_alias_list() perf pmu: Add perf_pmu__{open,scan}_file_at()
perf intel-pt: Use perf_pmu__scan_file_at() if possible

 tools/perf/arch/x86/util/intel-pt.c |  52 ++++--
 tools/perf/arch/x86/util/pmu.c      |  13 +-
 tools/perf/bench/Build              |   1 +
 tools/perf/bench/bench.h            |   1 +
 tools/perf/bench/pmu-scan.c         | 184 ++++++++++++++++++
 tools/perf/builtin-bench.c          |   1 +
 tools/perf/tests/pmu.c              |   9 +-
 tools/perf/util/parse-events.c      |   2 +-
 tools/perf/util/pmu.c               | 278 ++++++++++++++++++++--------
 tools/perf/util/pmu.h               |  12 +-
 tools/perf/util/print-events.c      |  26 ++-
 11 files changed, 466 insertions(+), 113 deletions(-)
 create mode 100644 tools/perf/bench/pmu-scan.c

base-commit: 417c6adfb155f906f0441cc1034827f6e2b3c372

Signature · 1 line
2.40.0.348.gf938b09366-goog
author maintainer
#1 lore raw

PATCH 1/9 perf list: Use relative path for tracepoint scan

---
 tools/perf/util/print-events.c | 26 ++++++++++++++++++++------
 1 file changed, 20 insertions(+), 6 deletions(-)
tools/perf/util/print-events.c +20 -6
index 62e9ea7dcf40..26a7e017c928 100644
@@ -4,6 +4,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+#include <fcntl.h>
 #include <sys/param.h>
 
 #include <api/fs/tracing_path.h>
@@ -59,12 +60,20 @@ static const struct event_symbol event_symbols_tool[PERF_TOOL_MAX] = {
 void print_tracepoint_events(const struct print_callbacks *print_cb, void *print_state)
 {
 	struct dirent **sys_namelist = NULL;
+	char *events_path = get_tracing_file("events");
 	int sys_items = tracing_events__scandir_alphasort(&sys_namelist);
+	int events_fd = open(events_path, O_PATH);
+
+	put_tracing_file(events_path);
+	if (events_fd < 0) {
+		printf("Error: failed to open tracing events directory\n");
+		return;
+	}
 
 	for (int i = 0; i < sys_items; i++) {
 		struct dirent *sys_dirent = sys_namelist[i];
 		struct dirent **evt_namelist = NULL;
-		char *dir_path;
+		int dir_fd;
 		int evt_items;
 
 		if (sys_dirent->d_type != DT_DIR ||
@@ -72,22 +81,26 @@ void print_tracepoint_events(const struct print_callbacks *print_cb, void *print
 		    !strcmp(sys_dirent->d_name, ".."))
 			continue;
 
-		dir_path = get_events_file(sys_dirent->d_name);
-		if (!dir_path)
+		dir_fd = openat(events_fd, sys_dirent->d_name, O_PATH);
+		if (dir_fd < 0)
 			continue;
 
-		evt_items = scandir(dir_path, &evt_namelist, NULL, alphasort);
+		evt_items = scandirat(events_fd, sys_dirent->d_name, &evt_namelist, NULL, alphasort);
 		for (int j = 0; j < evt_items; j++) {
 			struct dirent *evt_dirent = evt_namelist[j];
 			char evt_path[MAXPATHLEN];
+			int evt_fd;
 
 			if (evt_dirent->d_type != DT_DIR ||
 			    !strcmp(evt_dirent->d_name, ".") ||
 			    !strcmp(evt_dirent->d_name, ".."))
 				continue;
 
-			if (tp_event_has_id(dir_path, evt_dirent) != 0)
+			snprintf(evt_path, sizeof(evt_path), "%s/id", evt_dirent->d_name);
+			evt_fd = openat(dir_fd, evt_path, O_RDONLY);
+			if (evt_fd < 0)
 				continue;
+			close(evt_fd);
 
 			snprintf(evt_path, MAXPATHLEN, "%s:%s",
 				 sys_dirent->d_name, evt_dirent->d_name);
@@ -103,10 +116,11 @@ void print_tracepoint_events(const struct print_callbacks *print_cb, void *print
 					/*long_desc=*/NULL,
 					/*encoding_desc=*/NULL);
 		}
-		free(dir_path);
+		close(dir_fd);
 		free(evt_namelist);
 	}
 	free(sys_namelist);
+	close(events_fd);
 }
 
 void print_sdt_events(const struct print_callbacks *print_cb, void *print_state)
Signature · 1 line
2.40.0.348.gf938b09366-goog
#2 lore raw

Re: [PATCH 1/9] perf list: Use relative path for tracepoint scan

Em Fri, Mar 31, 2023 at 01:29:41PM -0700, Namhyung Kim escreveu:

Quoted 4 lines
---
 tools/perf/util/print-events.c | 26 ++++++++++++++++++++------
 1 file changed, 20 insertions(+), 6 deletions(-)

Add to add this to fix on Alma Linux 8, and scandirat isn't being found on musl libc (Alpine Linux), probably we'll need some scaffolding...

- Arnaldo

tools/perf/util/print-events.c +1 -0
index 26a7e017c9284c01..28aa0b9300253d0a 100644
@@ -6,6 +6,7 @@
 #include <string.h>
 #include <fcntl.h>
 #include <sys/param.h>
+#include <unistd.h>
 
 #include <api/fs/tracing_path.h>
 #include <linux/stddef.h>
Quoted 81 lines
tools/perf/util/print-events.c +20 -6
index 62e9ea7dcf40..26a7e017c928 100644
@@ -4,6 +4,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+#include <fcntl.h>
 #include <sys/param.h>
 
 #include <api/fs/tracing_path.h>
@@ -59,12 +60,20 @@ static const struct event_symbol event_symbols_tool[PERF_TOOL_MAX] = {
 void print_tracepoint_events(const struct print_callbacks *print_cb, void *print_state)
 {
 	struct dirent **sys_namelist = NULL;
+	char *events_path = get_tracing_file("events");
 	int sys_items = tracing_events__scandir_alphasort(&sys_namelist);
+	int events_fd = open(events_path, O_PATH);
+
+	put_tracing_file(events_path);
+	if (events_fd < 0) {
+		printf("Error: failed to open tracing events directory\n");
+		return;
+	}
 
 	for (int i = 0; i < sys_items; i++) {
 		struct dirent *sys_dirent = sys_namelist[i];
 		struct dirent **evt_namelist = NULL;
-		char *dir_path;
+		int dir_fd;
 		int evt_items;
 
 		if (sys_dirent->d_type != DT_DIR ||
@@ -72,22 +81,26 @@ void print_tracepoint_events(const struct print_callbacks *print_cb, void *print
 		    !strcmp(sys_dirent->d_name, ".."))
 			continue;
 
-		dir_path = get_events_file(sys_dirent->d_name);
-		if (!dir_path)
+		dir_fd = openat(events_fd, sys_dirent->d_name, O_PATH);
+		if (dir_fd < 0)
 			continue;
 
-		evt_items = scandir(dir_path, &evt_namelist, NULL, alphasort);
+		evt_items = scandirat(events_fd, sys_dirent->d_name, &evt_namelist, NULL, alphasort);
 		for (int j = 0; j < evt_items; j++) {
 			struct dirent *evt_dirent = evt_namelist[j];
 			char evt_path[MAXPATHLEN];
+			int evt_fd;
 
 			if (evt_dirent->d_type != DT_DIR ||
 			    !strcmp(evt_dirent->d_name, ".") ||
 			    !strcmp(evt_dirent->d_name, ".."))
 				continue;
 
-			if (tp_event_has_id(dir_path, evt_dirent) != 0)
+			snprintf(evt_path, sizeof(evt_path), "%s/id", evt_dirent->d_name);
+			evt_fd = openat(dir_fd, evt_path, O_RDONLY);
+			if (evt_fd < 0)
 				continue;
+			close(evt_fd);
 
 			snprintf(evt_path, MAXPATHLEN, "%s:%s",
 				 sys_dirent->d_name, evt_dirent->d_name);
@@ -103,10 +116,11 @@ void print_tracepoint_events(const struct print_callbacks *print_cb, void *print
 					/*long_desc=*/NULL,
 					/*encoding_desc=*/NULL);
 		}
-		free(dir_path);
+		close(dir_fd);
 		free(evt_namelist);
 	}
 	free(sys_namelist);
+	close(events_fd);
 }
 
 void print_sdt_events(const struct print_callbacks *print_cb, void *print_state)
Signature · 1 line
2.40.0.348.gf938b09366-goog
Signature · 1 line
- Arnaldo
#3 lore raw

Re: [PATCH 1/9] perf list: Use relative path for tracepoint scan

Em Mon, Apr 03, 2023 at 06:59:10PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 8 lines

Em Fri, Mar 31, 2023 at 01:29:41PM -0700, Namhyung Kim escreveu:

Quoted 4 lines
---
 tools/perf/util/print-events.c | 26 ++++++++++++++++++++------
 1 file changed, 20 insertions(+), 6 deletions(-)

Add to add this to fix on Alma Linux 8, and scandirat isn't being found on musl libc (Alpine Linux), probably we'll need some scaffolding...

Some discussion about this for some other project:

https://gitlab.com/apparmor/apparmor/-/merge_requests/107

Quoted 101 lines

- Arnaldo

tools/perf/util/print-events.c +1 -0
index 26a7e017c9284c01..28aa0b9300253d0a 100644
@@ -6,6 +6,7 @@
 #include <string.h>
 #include <fcntl.h>
 #include <sys/param.h>
+#include <unistd.h>
 
 #include <api/fs/tracing_path.h>
 #include <linux/stddef.h>
Quoted 82 lines
tools/perf/util/print-events.c +20 -6
index 62e9ea7dcf40..26a7e017c928 100644
@@ -4,6 +4,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+#include <fcntl.h>
 #include <sys/param.h>
 
 #include <api/fs/tracing_path.h>
@@ -59,12 +60,20 @@ static const struct event_symbol event_symbols_tool[PERF_TOOL_MAX] = {
 void print_tracepoint_events(const struct print_callbacks *print_cb, void *print_state)
 {
 	struct dirent **sys_namelist = NULL;
+	char *events_path = get_tracing_file("events");
 	int sys_items = tracing_events__scandir_alphasort(&sys_namelist);
+	int events_fd = open(events_path, O_PATH);
+
+	put_tracing_file(events_path);
+	if (events_fd < 0) {
+		printf("Error: failed to open tracing events directory\n");
+		return;
+	}
 
 	for (int i = 0; i < sys_items; i++) {
 		struct dirent *sys_dirent = sys_namelist[i];
 		struct dirent **evt_namelist = NULL;
-		char *dir_path;
+		int dir_fd;
 		int evt_items;
 
 		if (sys_dirent->d_type != DT_DIR ||
@@ -72,22 +81,26 @@ void print_tracepoint_events(const struct print_callbacks *print_cb, void *print
 		    !strcmp(sys_dirent->d_name, ".."))
 			continue;
 
-		dir_path = get_events_file(sys_dirent->d_name);
-		if (!dir_path)
+		dir_fd = openat(events_fd, sys_dirent->d_name, O_PATH);
+		if (dir_fd < 0)
 			continue;
 
-		evt_items = scandir(dir_path, &evt_namelist, NULL, alphasort);
+		evt_items = scandirat(events_fd, sys_dirent->d_name, &evt_namelist, NULL, alphasort);
 		for (int j = 0; j < evt_items; j++) {
 			struct dirent *evt_dirent = evt_namelist[j];
 			char evt_path[MAXPATHLEN];
+			int evt_fd;
 
 			if (evt_dirent->d_type != DT_DIR ||
 			    !strcmp(evt_dirent->d_name, ".") ||
 			    !strcmp(evt_dirent->d_name, ".."))
 				continue;
 
-			if (tp_event_has_id(dir_path, evt_dirent) != 0)
+			snprintf(evt_path, sizeof(evt_path), "%s/id", evt_dirent->d_name);
+			evt_fd = openat(dir_fd, evt_path, O_RDONLY);
+			if (evt_fd < 0)
 				continue;
+			close(evt_fd);
 
 			snprintf(evt_path, MAXPATHLEN, "%s:%s",
 				 sys_dirent->d_name, evt_dirent->d_name);
@@ -103,10 +116,11 @@ void print_tracepoint_events(const struct print_callbacks *print_cb, void *print
 					/*long_desc=*/NULL,
 					/*encoding_desc=*/NULL);
 		}
-		free(dir_path);
+		close(dir_fd);
 		free(evt_namelist);
 	}
 	free(sys_namelist);
+	close(events_fd);
 }
 
 void print_sdt_events(const struct print_callbacks *print_cb, void *print_state)
Signature · 1 line
2.40.0.348.gf938b09366-goog
Signature · 1 line
- Arnaldo
Signature · 1 line
- Arnaldo
#4 lore raw

Re: [PATCH 1/9] perf list: Use relative path for tracepoint scan

In reply to Arnaldo Carvalho de Melo

Em Mon, Apr 03, 2023 at 06:59:10PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 8 lines

Em Fri, Mar 31, 2023 at 01:29:41PM -0700, Namhyung Kim escreveu:

Quoted 4 lines
---
 tools/perf/util/print-events.c | 26 ++++++++++++++++++++------
 1 file changed, 20 insertions(+), 6 deletions(-)

Add to add this to fix on Alma Linux 8, and scandirat isn't being found on musl libc (Alpine Linux), probably we'll need some scaffolding...

About scandirat I'm adding the feature test below and this patch to the patch that starts using scandirat():

tools/perf/util/print-events.c +12 -3
index 28aa0b9300253d0a..386b1ab0b60e1d93 100644
@@ -58,11 +58,9 @@ static const struct event_symbol event_symbols_tool[PERF_TOOL_MAX] = {
 /*
  * Print the events from <debugfs_mount_point>/tracing/events
  */
-void print_tracepoint_events(const struct print_callbacks *print_cb, void *print_state)
+void print_tracepoint_events(const struct print_callbacks *print_cb __maybe_unused, void *print_state __maybe_unused)
 {
-	struct dirent **sys_namelist = NULL;
 	char *events_path = get_tracing_file("events");
-	int sys_items = tracing_events__scandir_alphasort(&sys_namelist);
 	int events_fd = open(events_path, O_PATH);
 
 	put_tracing_file(events_path);
@@ -71,6 +69,11 @@ void print_tracepoint_events(const struct print_callbacks *print_cb, void *print
 		return;
 	}
 
+#ifdef HAVE_SCANDIRAT_SUPPORT
+{
+	struct dirent **sys_namelist = NULL;
+	int sys_items = tracing_events__scandir_alphasort(&sys_namelist);
+
 	for (int i = 0; i < sys_items; i++) {
 		struct dirent *sys_dirent = sys_namelist[i];
 		struct dirent **evt_namelist = NULL;
@@ -120,7 +123,13 @@ void print_tracepoint_events(const struct print_callbacks *print_cb, void *print
 		close(dir_fd);
 		free(evt_namelist);
 	}
+
 	free(sys_namelist);
+}
+#else
+	printf("\nWARNING: Your libc doesn't have the scandir function, please ask its maintainers to implement it.\n"
+	       "         As a rough fallback, please do 'ls %s' to see the available tracepoint events.\n", events_path);
+#endif
 	close(events_fd);
 }
 

From b25f9de2f944a550b063322b7210f4392e622a5e Mon Sep 17 00:00:00 2001
From: Arnaldo Carvalho de Melo <[email protected]>
Date: Tue, 4 Apr 2023 11:05:57 -0300
Subject: [PATCH 1/1] tools build: Add a feature test for scandirat(), that is
not implemented so far in musl and uclibc

We use it just when listing tracepoint events, and for root, so just emit a warning about it to get users to ask the library maintainers to implement it, as suggested in this systemd ticket:

https://github.com/systemd/casync/issues/129

Cc Adrian Hunter <[email protected]>
Cc Ian Rogers <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
---
 tools/build/Makefile.feature   | 1 +
 tools/build/feature/Makefile   | 4 ++++
 tools/build/feature/test-all.c | 5 +++++
 tools/perf/Makefile.config     | 4 ++++
 4 files changed, 14 insertions(+)
tools/build/Makefile.feature +1 -0
index 214622d7537cc56b..934e2777a2dbcd90 100644
@@ -64,6 +64,7 @@ FEATURE_TESTS_BASIC :=                  \
         lzma                            \
         get_cpuid                       \
         bpf                             \
+        scandirat			\
         sched_getcpu			\
         sdt				\
         setns				\
tools/build/feature/Makefile +4 -0
index 0a3b9281f8b08000..0f0aa9b7d7b5e43c 100644
@@ -65,6 +65,7 @@ FILES=                                          \
          test-gettid.bin			\
          test-jvmti.bin				\
          test-jvmti-cmlr.bin			\
+         test-scandirat.bin			\
          test-sched_getcpu.bin			\
          test-setns.bin				\
          test-libopencsd.bin			\
@@ -129,6 +130,9 @@ $(OUTPUT)test-get_current_dir_name.bin:
 $(OUTPUT)test-glibc.bin:
 	$(BUILD)
 
+$(OUTPUT)test-scandirat.bin:
+	$(BUILD)
+
 $(OUTPUT)test-sched_getcpu.bin:
 	$(BUILD)
 
tools/build/feature/test-all.c +5 -0
index 957c02c7b163b579..6f4bf386a3b5c4b0 100644
@@ -114,6 +114,10 @@
 # include "test-pthread-barrier.c"
 #undef main
 
+#define main main_test_scandirat
+# include "test-scandirat.c"
+#undef main
+
 #define main main_test_sched_getcpu
 # include "test-sched_getcpu.c"
 #undef main
@@ -206,6 +210,7 @@ int main(int argc, char *argv[])
 	main_test_get_cpuid();
 	main_test_bpf();
 	main_test_libcrypto();
+	main_test_scandirat();
 	main_test_sched_getcpu();
 	main_test_sdt();
 	main_test_setns();
tools/perf/Makefile.config +4 -0
index dd203f0a2b7e64e1..16bea51f0bcd95ea 100644
@@ -478,6 +478,10 @@ ifdef NO_DWARF
   NO_LIBDW_DWARF_UNWIND := 1
 endif
 
+ifeq ($(feature-scandirat), 1)
+  CFLAGS += -DHAVE_SCANDIRAT_SUPPORT
+endif
+
 ifeq ($(feature-sched_getcpu), 1)
   CFLAGS += -DHAVE_SCHED_GETCPU_SUPPORT
 endif
Signature · 1 line
2.39.2
author maintainer
#5 lore raw

Re: [PATCH 1/9] perf list: Use relative path for tracepoint scan

Hi Arnaldo,

Quoted 8 lines · Arnaldo Carvalho de Melo

Em Mon, Apr 03, 2023 at 06:59:10PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 6 lines

Em Fri, Mar 31, 2023 at 01:29:41PM -0700, Namhyung Kim escreveu:

Quoted 4 lines
---
 tools/perf/util/print-events.c | 26 ++++++++++++++++++++------
 1 file changed, 20 insertions(+), 6 deletions(-)
Quoted 5 lines
Quoted 2 lines

Add to add this to fix on Alma Linux 8, and scandirat isn't being found on musl libc (Alpine Linux), probably we'll need some scaffolding...

About scandirat I'm adding the feature test below and this patch to the patch that starts using scandirat():

Thanks for doing this!

Quoted 62 lines
tools/perf/util/print-events.c +12 -3
index 28aa0b9300253d0a..386b1ab0b60e1d93 100644
@@ -58,11 +58,9 @@ static const struct event_symbol event_symbols_tool[PERF_TOOL_MAX] = {
 /*
  * Print the events from <debugfs_mount_point>/tracing/events
  */
-void print_tracepoint_events(const struct print_callbacks *print_cb, void *print_state)
+void print_tracepoint_events(const struct print_callbacks *print_cb __maybe_unused, void *print_state __maybe_unused)
 {
-       struct dirent **sys_namelist = NULL;
        char *events_path = get_tracing_file("events");
-       int sys_items = tracing_events__scandir_alphasort(&sys_namelist);
        int events_fd = open(events_path, O_PATH);
 
        put_tracing_file(events_path);
@@ -71,6 +69,11 @@ void print_tracepoint_events(const struct print_callbacks *print_cb, void *print
                return;
        }
 
+#ifdef HAVE_SCANDIRAT_SUPPORT
+{
+       struct dirent **sys_namelist = NULL;
+       int sys_items = tracing_events__scandir_alphasort(&sys_namelist);
+
        for (int i = 0; i < sys_items; i++) {
                struct dirent *sys_dirent = sys_namelist[i];
                struct dirent **evt_namelist = NULL;
@@ -120,7 +123,13 @@ void print_tracepoint_events(const struct print_callbacks *print_cb, void *print
                close(dir_fd);
                free(evt_namelist);
        }
+
        free(sys_namelist);
+}
+#else
+       printf("\nWARNING: Your libc doesn't have the scandir function, please ask its maintainers to implement it.\n"
+              "         As a rough fallback, please do 'ls %s' to see the available tracepoint events.\n", events_path);
+#endif
        close(events_fd);
 }
 

From b25f9de2f944a550b063322b7210f4392e622a5e Mon Sep 17 00:00:00 2001
From: Arnaldo Carvalho de Melo <[email protected]>
Date: Tue, 4 Apr 2023 11:05:57 -0300
Subject: [PATCH 1/1] tools build: Add a feature test for scandirat(), that is
not implemented so far in musl and uclibc

We use it just when listing tracepoint events, and for root, so just emit a warning about it to get users to ask the library maintainers to implement it, as suggested in this systemd ticket:

https://github.com/systemd/casync/issues/129

Cc Adrian Hunter <[email protected]>
Cc Ian Rogers <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
Quoted 5 lines
---
 tools/build/Makefile.feature   | 1 +
 tools/build/feature/Makefile   | 4 ++++
 tools/build/feature/test-all.c | 5 +++++
 tools/perf/Makefile.config     | 4 ++++

Maybe you forgot to add the tools/build/feature/test-scandirat.c :)

Thanks,
Namhyung

Quoted 77 lines
 4 files changed, 14 insertions(+)
tools/build/Makefile.feature +1 -0
index 214622d7537cc56b..934e2777a2dbcd90 100644
@@ -64,6 +64,7 @@ FEATURE_TESTS_BASIC :=                  \
         lzma                            \
         get_cpuid                       \
         bpf                             \
+        scandirat                      \
         sched_getcpu                   \
         sdt                            \
         setns                          \
tools/build/feature/Makefile +4 -0
index 0a3b9281f8b08000..0f0aa9b7d7b5e43c 100644
@@ -65,6 +65,7 @@ FILES=                                          \
          test-gettid.bin                       \
          test-jvmti.bin                                \
          test-jvmti-cmlr.bin                   \
+         test-scandirat.bin                    \
          test-sched_getcpu.bin                 \
          test-setns.bin                                \
          test-libopencsd.bin                   \
@@ -129,6 +130,9 @@ $(OUTPUT)test-get_current_dir_name.bin:
 $(OUTPUT)test-glibc.bin:
        $(BUILD)
 
+$(OUTPUT)test-scandirat.bin:
+       $(BUILD)
+
 $(OUTPUT)test-sched_getcpu.bin:
        $(BUILD)
 
tools/build/feature/test-all.c +5 -0
index 957c02c7b163b579..6f4bf386a3b5c4b0 100644
@@ -114,6 +114,10 @@
 # include "test-pthread-barrier.c"
 #undef main
 
+#define main main_test_scandirat
+# include "test-scandirat.c"
+#undef main
+
 #define main main_test_sched_getcpu
 # include "test-sched_getcpu.c"
 #undef main
@@ -206,6 +210,7 @@ int main(int argc, char *argv[])
        main_test_get_cpuid();
        main_test_bpf();
        main_test_libcrypto();
+       main_test_scandirat();
        main_test_sched_getcpu();
        main_test_sdt();
        main_test_setns();
tools/perf/Makefile.config +4 -1
index dd203f0a2b7e64e1..16bea51f0bcd95ea 100644
@@ -478,6 +478,10 @@ ifdef NO_DWARF
   NO_LIBDW_DWARF_UNWIND := 1
 endif
 
+ifeq ($(feature-scandirat), 1)
+  CFLAGS += -DHAVE_SCANDIRAT_SUPPORT
+endif
+
 ifeq ($(feature-sched_getcpu), 1)
   CFLAGS += -DHAVE_SCHED_GETCPU_SUPPORT
 endif
--

2.39.2

#6 lore raw

Re: [PATCH 1/9] perf list: Use relative path for tracepoint scan

Quoted 7 lines · Namhyung Kim
Quoted 5 lines
---
 tools/build/Makefile.feature   | 1 +
 tools/build/feature/Makefile   | 4 ++++
 tools/build/feature/test-all.c | 5 +++++
 tools/perf/Makefile.config     | 4 ++++

Maybe you forgot to add the tools/build/feature/test-scandirat.c :)

Yeah , I noticed that and fixed it:

https://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git/tree/tools/build/feature/test-scandirat.c?h=tmp.perf-tools-next

- Arnaldo

author maintainer
#7 lore raw

PATCH 2/9 perf tools: Fix a asan issue in parse_events_multi_pmu_add()

In reply to Namhyung Kim

In the parse_events_multi_pmu_add() it passes the 'config' variable twice to parse_events_term__num() - one for config and another for loc_term. I'm not sure about the second one as it's converted to YYLTYPE variable. Asan reports it like below:

In function ‘parse_events_term__num’,
inlined from ‘parse_events_multi_pmu_add’ at util/parse-events.c:1602:6:
util/parse-events.c:2653:64: error: array subscript ‘YYLTYPE[0]’ is partly outside
array bounds of ‘char[8]’ [-Werror=array-bounds]
2653 | .err_term = loc_term ? loc_term->first_column : 0,

        |                              ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~
  util/parse-events.c: In function ‘parse_events_multi_pmu_add’:
  util/parse-events.c:1587:15: note: object ‘config’ of size 8
   1587 |         char *config;
        |               ^~~~~~

cc1: all warnings being treated as errors

---
 tools/perf/util/parse-events.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/parse-events.c +1 -1
index cc8e8766ca30..0010e5e0ee68 100644
@@ -1601,7 +1601,7 @@ int parse_events_multi_pmu_add(struct parse_events_state *parse_state,
 
 	if (parse_events_term__num(&term,
 				   PARSE_EVENTS__TERM_TYPE_USER,
-				   config, 1, false, &config,
+				   config, 1, false, NULL,
 					NULL) < 0) {
 		free(config);
 		goto out_err;
Signature · 1 line
2.40.0.348.gf938b09366-goog
author maintainer
#8 lore raw

PATCH 3/9 perf pmu: Add perf_pmu__destroy() function

In reply to Namhyung Kim

It seems there's no function to delete the perf pmu struct. Add the perf_pmu__destroy() to do the job. While at it, add some more helper functions to delete pmu aliases and caps.

---
 tools/perf/util/pmu.c | 50 +++++++++++++++++++++++++++++++++++++++++++
 tools/perf/util/pmu.h |  2 ++
 2 files changed, 52 insertions(+)
tools/perf/util/pmu.c +50 -0
index e3aae731bd6f..b112606f36ec 100644
@@ -300,6 +300,16 @@ void perf_pmu_free_alias(struct perf_pmu_alias *newalias)
 	free(newalias);
 }
 
+static void perf_pmu__del_aliases(struct perf_pmu *pmu)
+{
+	struct perf_pmu_alias *alias, *tmp;
+
+	list_for_each_entry_safe(alias, tmp, &pmu->aliases, list) {
+		list_del(&alias->list);
+		perf_pmu_free_alias(alias);
+	}
+}
+
 /* Merge an alias, search in alias list. If this name is already
  * present merge both of them to combine all information.
  */
@@ -921,6 +931,8 @@ static struct perf_pmu *pmu_lookup(const char *lookup_name)
 
 	if (is_hybrid)
 		list_add_tail(&pmu->hybrid_list, &perf_pmu__hybrid_pmus);
+	else
+		INIT_LIST_HEAD(&pmu->hybrid_list);
 
 	pmu->default_config = perf_pmu__get_default_config(pmu);
 
@@ -1750,6 +1762,18 @@ static int perf_pmu__new_caps(struct list_head *list, char *name, char *value)
 	return -ENOMEM;
 }
 
+static void perf_pmu__del_caps(struct perf_pmu *pmu)
+{
+	struct perf_pmu_caps *caps, *tmp;
+
+	list_for_each_entry_safe(caps, tmp, &pmu->caps, list) {
+		list_del(&caps->list);
+		free(caps->name);
+		free(caps->value);
+		free(caps);
+	}
+}
+
 /*
  * Reading/parsing the given pmu capabilities, which should be located at:
  * /sys/bus/event_source/devices/<dev>/caps as sysfs group attributes.
@@ -1932,3 +1956,29 @@ int perf_pmu__pathname_scnprintf(char *buf, size_t size,
 		return 0;
 	return scnprintf(buf, size, "%s%s/%s", base_path, pmu_name, filename);
 }
+
+static void perf_pmu__delete(struct perf_pmu *pmu)
+{
+	perf_pmu__del_formats(&pmu->format);
+	perf_pmu__del_aliases(pmu);
+	perf_pmu__del_caps(pmu);
+
+	perf_cpu_map__put(pmu->cpus);
+
+	free(pmu->default_config);
+	free(pmu->name);
+	free(pmu->alias_name);
+	free(pmu);
+}
+
+void perf_pmu__destroy(void)
+{
+	struct perf_pmu *pmu, *tmp;
+
+	list_for_each_entry_safe(pmu, tmp, &pmus, list) {
+		list_del(&pmu->list);
+		list_del(&pmu->hybrid_list);
+
+		perf_pmu__delete(pmu);
+	}
+}
tools/perf/util/pmu.h +2 -0
index 24cf69ab32cd..72fd5de334c0 100644
@@ -259,4 +259,6 @@ int perf_pmu__pathname_scnprintf(char *buf, size_t size,
 				 const char *pmu_name, const char *filename);
 FILE *perf_pmu__open_file(struct perf_pmu *pmu, const char *name);
 
+void perf_pmu__destroy(void);
+
 #endif /* __PMU_H */
Signature · 1 line
2.40.0.348.gf938b09366-goog
author maintainer
#9 lore raw

PATCH 4/9 perf bench: Add pmu-scan benchmark

In reply to Namhyung Kim

The pmu-scan benchmark will repeatedly scan the sysfs to get the available PMU information.

  $ ./perf bench internals pmu-scan
  # Running 'internals/pmu-scan' benchmark:
  Computing performance of sysfs PMU event scan for 100 times
    Average PMU scanning took: 6850.990 usec (+- 48.445 usec)
---
 tools/perf/bench/Build      |   1 +
 tools/perf/bench/bench.h    |   1 +
 tools/perf/bench/pmu-scan.c | 184 ++++++++++++++++++++++++++++++++++++
 tools/perf/builtin-bench.c  |   1 +
 4 files changed, 187 insertions(+)
 create mode 100644 tools/perf/bench/pmu-scan.c
tools/perf/bench/Build +1 -0
index 6b6155a8ad09..0f158dc8139b 100644
@@ -15,6 +15,7 @@ perf-y += find-bit-bench.o
 perf-y += inject-buildid.o
 perf-y += evlist-open-close.o
 perf-y += breakpoint.o
+perf-y += pmu-scan.o
 
 perf-$(CONFIG_X86_64) += mem-memcpy-x86-64-asm.o
 perf-$(CONFIG_X86_64) += mem-memset-x86-64-asm.o
tools/perf/bench/bench.h +1 -0
index edfc25793cca..0d2b65976212 100644
@@ -42,6 +42,7 @@ int bench_inject_build_id(int argc, const char **argv);
 int bench_evlist_open_close(int argc, const char **argv);
 int bench_breakpoint_thread(int argc, const char **argv);
 int bench_breakpoint_enable(int argc, const char **argv);
+int bench_pmu_scan(int argc, const char **argv);
 
 #define BENCH_FORMAT_DEFAULT_STR	"default"
 #define BENCH_FORMAT_DEFAULT		0
tools/perf/bench/pmu-scan.c new file +184 -0

184 lines not on this page · raw message

tools/perf/builtin-bench.c +1 -0
index d0fcc3cdc185..58f1cfe1eb34 100644
@@ -92,6 +92,7 @@ static struct bench internals_benchmarks[] = {
 	{ "kallsyms-parse", "Benchmark kallsyms parsing",	bench_kallsyms_parse	},
 	{ "inject-build-id", "Benchmark build-id injection",	bench_inject_build_id	},
 	{ "evlist-open-close", "Benchmark evlist open and close",	bench_evlist_open_close	},
+	{ "pmu-scan", "Benchmark sysfs PMU info scanning",	bench_pmu_scan		},
 	{ NULL,		NULL,					NULL			}
 };
 
Signature · 1 line
2.40.0.348.gf938b09366-goog
author maintainer
#10 lore raw

PATCH 5/9 perf pmu: Use relative path for sysfs scan

In reply to Namhyung Kim

The PMU information is in the kernel sysfs so it needs to scan the directory to get the whole information like event aliases, formats and so on. During the traversal, it opens a lot of files and directories like below:

  dir = opendir("/sys/bus/event_source/devices");
  while (dentry = readdir(dir)) {
    char buf[PATH_MAX];
    snprintf(buf, sizeof(buf), "%s/%s",
             "/sys/bus/event_source/devices", dentry->d_name);
    fd = open(buf, O_RDONLY);
    ...

}

But this is not good since it needs to copy the string to build the absolute pathname, and it makes redundant pathname walk (from the /sys) unnecessarily. We can use openat(2) to open the file in the given directory. While it's not a problem ususally, it can be a problem when the kernel has contentions on the sysfs.

Add a couple of new helper to return the file descriptor of PMU directory so that it can use it with relative paths.

* perf_pmu__event_source_devices_fd()
- returns a fd for the PMU root ("/sys/bus/event_source/devices")

* perf_pmu__pathname_fd()
- returns a fd for "<pmu>/<file>" under the PMU root

Now the above code can be converted something like below:

  dirfd = perf_pmu__event_source_devices_fd();
  dir = fdopendir(dirfd);
  while (dentry = readdir(dir)) {
    fd = openat(dirfd, dentry->d_name, O_RDONLY);
    ...
  }
---
 tools/perf/tests/pmu.c |   9 ++-
 tools/perf/util/pmu.c  | 161 +++++++++++++++++++++++++----------------
 tools/perf/util/pmu.h  |   4 +-
 3 files changed, 111 insertions(+), 63 deletions(-)
tools/perf/tests/pmu.c +8 -1
index 8507bd615e97..3cf25f883df7 100644
@@ -3,6 +3,7 @@
 #include "pmu.h"
 #include "tests.h"
 #include <errno.h>
+#include <fcntl.h>
 #include <stdio.h>
 #include <linux/kernel.h>
 #include <linux/limits.h>
@@ -149,10 +150,16 @@ static int test__pmu(struct test_suite *test __maybe_unused, int subtest __maybe
 
 	do {
 		struct perf_event_attr attr;
+		int fd;
 
 		memset(&attr, 0, sizeof(attr));
 
-		ret = perf_pmu__format_parse(format, &formats);
+		fd = open(format, O_DIRECTORY);
+		if (fd < 0) {
+			ret = fd;
+			break;
+		}
+		ret = perf_pmu__format_parse(fd, &formats);
 		if (ret)
 			break;
 
tools/perf/util/pmu.c +100 -61

378 lines not on this page · raw message

tools/perf/util/pmu.h +3 -1
index 72fd5de334c0..751c7016e7b6 100644
@@ -211,7 +211,7 @@ void perf_pmu_error(struct list_head *list, char *name, char const *msg);
 int perf_pmu__new_format(struct list_head *list, char *name,
 			 int config, unsigned long *bits);
 void perf_pmu__set_format(unsigned long *bits, long from, long to);
-int perf_pmu__format_parse(char *dir, struct list_head *head);
+int perf_pmu__format_parse(int dirfd, struct list_head *head);
 void perf_pmu__del_formats(struct list_head *formats);
 
 struct perf_pmu *perf_pmu__scan(struct perf_pmu *pmu);
@@ -257,6 +257,8 @@ double perf_pmu__cpu_slots_per_cycle(void);
 int perf_pmu__event_source_devices_scnprintf(char *pathname, size_t size);
 int perf_pmu__pathname_scnprintf(char *buf, size_t size,
 				 const char *pmu_name, const char *filename);
+int perf_pmu__event_source_devices_fd(void);
+int perf_pmu__pathname_fd(int dirfd, const char *pmu_name, const char *filename, int flags);
 FILE *perf_pmu__open_file(struct perf_pmu *pmu, const char *name);
 
 void perf_pmu__destroy(void);
Signature · 1 line
2.40.0.348.gf938b09366-goog
reviewer
#11 lore raw

Re: [PATCH 5/9] perf pmu: Use relative path for sysfs scan

Quoted 116 lines · Namhyung Kim

The PMU information is in the kernel sysfs so it needs to scan the directory to get the whole information like event aliases, formats and so on. During the traversal, it opens a lot of files and directories like below:

  dir = opendir("/sys/bus/event_source/devices");
  while (dentry = readdir(dir)) {
    char buf[PATH_MAX];
    snprintf(buf, sizeof(buf), "%s/%s",
             "/sys/bus/event_source/devices", dentry->d_name);
    fd = open(buf, O_RDONLY);
    ...

}

But this is not good since it needs to copy the string to build the absolute pathname, and it makes redundant pathname walk (from the /sys) unnecessarily. We can use openat(2) to open the file in the given directory. While it's not a problem ususally, it can be a problem when the kernel has contentions on the sysfs.

Add a couple of new helper to return the file descriptor of PMU directory so that it can use it with relative paths.

* perf_pmu__event_source_devices_fd()
- returns a fd for the PMU root ("/sys/bus/event_source/devices")

* perf_pmu__pathname_fd()
- returns a fd for "<pmu>/<file>" under the PMU root

Now the above code can be converted something like below:

  dirfd = perf_pmu__event_source_devices_fd();
  dir = fdopendir(dirfd);
  while (dentry = readdir(dir)) {
    fd = openat(dirfd, dentry->d_name, O_RDONLY);
    ...
  }
---
 tools/perf/tests/pmu.c |   9 ++-
 tools/perf/util/pmu.c  | 161 +++++++++++++++++++++++++----------------
 tools/perf/util/pmu.h  |   4 +-
 3 files changed, 111 insertions(+), 63 deletions(-)
tools/perf/tests/pmu.c +8 -1
index 8507bd615e97..3cf25f883df7 100644
@@ -3,6 +3,7 @@
 #include "pmu.h"
 #include "tests.h"
 #include <errno.h>
+#include <fcntl.h>
 #include <stdio.h>
 #include <linux/kernel.h>
 #include <linux/limits.h>
@@ -149,10 +150,16 @@ static int test__pmu(struct test_suite *test __maybe_unused, int subtest __maybe
 
        do {
                struct perf_event_attr attr;
+               int fd;
 
                memset(&attr, 0, sizeof(attr));
 
-               ret = perf_pmu__format_parse(format, &formats);
+               fd = open(format, O_DIRECTORY);
+               if (fd < 0) {
+                       ret = fd;
+                       break;
+               }
+               ret = perf_pmu__format_parse(fd, &formats);
                if (ret)
                        break;
 
tools/perf/util/pmu.c +5 -6
index b112606f36ec..9fc6b8b5732b 100644
@@ -62,38 +62,38 @@ extern FILE *perf_pmu_in;
 
 static bool hybrid_scanned;
 
+static struct perf_pmu *perf_pmu__find2(int dirfd, const char *name);
+
 /*
  * Parse & process all the sysfs attributes located under
  * the directory specified in 'dir' parameter.
  */
-int perf_pmu__format_parse(char *dir, struct list_head *head)
+int perf_pmu__format_parse(int dirfd, struct list_head *head)
 {
        struct dirent *evt_ent;
        DIR *format_dir;
        int ret = 0;
 
-       format_dir = opendir(dir);
+       format_dir = fdopendir(dirfd);
        if (!format_dir)
                return -EINVAL;
 
        while (!ret && (evt_ent = readdir(format_dir))) {
-               char path[PATH_MAX];
                char *name = evt_ent->d_name;
-               FILE *file;
+               int fd;
 
                if (!strcmp(name, ".") || !strcmp(name, ".."))
                        continue;
 
-               snprintf(path, PATH_MAX, "%s/%s", dir, name);
 
                ret = -EINVAL;
-               file = fopen(path, "r");

tools/perf/util/pmu.c extern FILE *perf_pmu_in; · line 96

Quoted 10 lines
tools/perf/util/pmu.c +4 -3
-               if (!file)
+               fd = openat(dirfd, name, O_RDONLY);
+               if (fd < 0)
                        break;
 
-               perf_pmu_in = file;
+               perf_pmu_in = fdopen(fd, "r");
                ret = perf_pmu_parse(head, name);
-               fclose(file);
+               fclose(perf_pmu_in);

Eek, I hadn't realized that the pmu parser wasn't reentrant. I sent out:
https://lore.kernel.org/lkml/[email protected]/
to fix this.

Thanks,
Ian

Quoted 382 lines
tools/perf/util/pmu.c +91 -52
        }
 
        closedir(format_dir);
@@ -105,17 +105,16 @@ int perf_pmu__format_parse(char *dir, struct list_head *head)
  * located at:
  * /sys/bus/event_source/devices/<dev>/format as sysfs group attributes.
  */
-static int pmu_format(const char *name, struct list_head *format)
+static int pmu_format(int dirfd, const char *name, struct list_head *format)
 {
-       char path[PATH_MAX];
-
-       if (!perf_pmu__pathname_scnprintf(path, sizeof(path), name, "format"))
-               return -1;
+       int fd;
 
-       if (!file_available(path))
+       fd = perf_pmu__pathname_fd(dirfd, name, "format", O_DIRECTORY);
+       if (fd < 0)
                return 0;
 
-       if (perf_pmu__format_parse(path, format))
+       /* it'll close the fd */
+       if (perf_pmu__format_parse(fd, format))
                return -1;
 
        return 0;
@@ -158,7 +157,7 @@ int perf_pmu__convert_scale(const char *scale, char **end, double *sval)
        return ret;
 }
 
-static int perf_pmu__parse_scale(struct perf_pmu_alias *alias, char *dir, char *name)
+static int perf_pmu__parse_scale(struct perf_pmu_alias *alias, int dirfd, char *name)
 {
        struct stat st;
        ssize_t sret;
@@ -166,9 +165,9 @@ static int perf_pmu__parse_scale(struct perf_pmu_alias *alias, char *dir, char *
        int fd, ret = -1;
        char path[PATH_MAX];
 
-       scnprintf(path, PATH_MAX, "%s/%s.scale", dir, name);
+       scnprintf(path, PATH_MAX, "%s.scale", name);
 
-       fd = open(path, O_RDONLY);
+       fd = openat(dirfd, path, O_RDONLY);
        if (fd == -1)
                return -1;
 
@@ -190,15 +189,15 @@ static int perf_pmu__parse_scale(struct perf_pmu_alias *alias, char *dir, char *
        return ret;
 }
 
-static int perf_pmu__parse_unit(struct perf_pmu_alias *alias, char *dir, char *name)
+static int perf_pmu__parse_unit(struct perf_pmu_alias *alias, int dirfd, char *name)
 {
        char path[PATH_MAX];
        ssize_t sret;
        int fd;
 
-       scnprintf(path, PATH_MAX, "%s/%s.unit", dir, name);
+       scnprintf(path, PATH_MAX, "%s.unit", name);
 
-       fd = open(path, O_RDONLY);
+       fd = openat(dirfd, path, O_RDONLY);
        if (fd == -1)
                return -1;
 
@@ -221,14 +220,14 @@ static int perf_pmu__parse_unit(struct perf_pmu_alias *alias, char *dir, char *n
 }
 
 static int
-perf_pmu__parse_per_pkg(struct perf_pmu_alias *alias, char *dir, char *name)
+perf_pmu__parse_per_pkg(struct perf_pmu_alias *alias, int dirfd, char *name)
 {
        char path[PATH_MAX];
        int fd;
 
-       scnprintf(path, PATH_MAX, "%s/%s.per-pkg", dir, name);
+       scnprintf(path, PATH_MAX, "%s.per-pkg", name);
 
-       fd = open(path, O_RDONLY);
+       fd = openat(dirfd, path, O_RDONLY);
        if (fd == -1)
                return -1;
 
@@ -239,14 +238,14 @@ perf_pmu__parse_per_pkg(struct perf_pmu_alias *alias, char *dir, char *name)
 }
 
 static int perf_pmu__parse_snapshot(struct perf_pmu_alias *alias,
-                                   char *dir, char *name)
+                                   int dirfd, char *name)
 {
        char path[PATH_MAX];
        int fd;
 
-       scnprintf(path, PATH_MAX, "%s/%s.snapshot", dir, name);
+       scnprintf(path, PATH_MAX, "%s.snapshot", name);
 
-       fd = open(path, O_RDONLY);
+       fd = openat(dirfd, path, O_RDONLY);
        if (fd == -1)
                return -1;
 
@@ -332,7 +331,7 @@ static bool perf_pmu_merge_alias(struct perf_pmu_alias *newalias,
        return false;
 }
 
-static int __perf_pmu__new_alias(struct list_head *list, char *dir, char *name,
+static int __perf_pmu__new_alias(struct list_head *list, int dirfd, char *name,
                                 char *desc, char *val, const struct pmu_event *pe)
 {
        struct parse_events_term *term;
@@ -391,14 +390,14 @@ static int __perf_pmu__new_alias(struct list_head *list, char *dir, char *name,
        }
 
        alias->name = strdup(name);
-       if (dir) {
+       if (dirfd >= 0) {
                /*
                 * load unit name and scale if available
                 */
-               perf_pmu__parse_unit(alias, dir, name);
-               perf_pmu__parse_scale(alias, dir, name);
-               perf_pmu__parse_per_pkg(alias, dir, name);
-               perf_pmu__parse_snapshot(alias, dir, name);
+               perf_pmu__parse_unit(alias, dirfd, name);
+               perf_pmu__parse_scale(alias, dirfd, name);
+               perf_pmu__parse_per_pkg(alias, dirfd, name);
+               perf_pmu__parse_snapshot(alias, dirfd, name);
        }
 
        alias->desc = desc ? strdup(desc) : NULL;
@@ -419,7 +418,7 @@ static int __perf_pmu__new_alias(struct list_head *list, char *dir, char *name,
        return 0;
 }
 
-static int perf_pmu__new_alias(struct list_head *list, char *dir, char *name, FILE *file)
+static int perf_pmu__new_alias(struct list_head *list, int dirfd, char *name, FILE *file)
 {
        char buf[256];
        int ret;
@@ -433,7 +432,7 @@ static int perf_pmu__new_alias(struct list_head *list, char *dir, char *name, FI
        /* Remove trailing newline from sysfs file */
        strim(buf);
 
-       return __perf_pmu__new_alias(list, dir, name, NULL, buf, NULL);
+       return __perf_pmu__new_alias(list, dirfd, name, NULL, buf, NULL);
 }
 
 static inline bool pmu_alias_info_file(char *name)
@@ -457,17 +456,17 @@ static inline bool pmu_alias_info_file(char *name)
  * Process all the sysfs attributes located under the directory
  * specified in 'dir' parameter.
  */
-static int pmu_aliases_parse(char *dir, struct list_head *head)
+static int pmu_aliases_parse(int dirfd, struct list_head *head)
 {
        struct dirent *evt_ent;
        DIR *event_dir;
+       int fd;
 
-       event_dir = opendir(dir);
+       event_dir = fdopendir(dirfd);
        if (!event_dir)
                return -EINVAL;
 
        while ((evt_ent = readdir(event_dir))) {
-               char path[PATH_MAX];
                char *name = evt_ent->d_name;
                FILE *file;
 
@@ -480,15 +479,14 @@ static int pmu_aliases_parse(char *dir, struct list_head *head)
                if (pmu_alias_info_file(name))
                        continue;
 
-               scnprintf(path, PATH_MAX, "%s/%s", dir, name);
-
-               file = fopen(path, "r");
+               fd = openat(dirfd, name, O_RDONLY);
+               file = fdopen(fd, "r");
                if (!file) {
-                       pr_debug("Cannot open %s\n", path);
+                       pr_debug("Cannot open %s\n", name);
                        continue;
                }
 
-               if (perf_pmu__new_alias(head, dir, name, file) < 0)
+               if (perf_pmu__new_alias(head, dirfd, name, file) < 0)
                        pr_debug("Cannot set up %s\n", name);
                fclose(file);
        }
@@ -501,17 +499,16 @@ static int pmu_aliases_parse(char *dir, struct list_head *head)
  * Reading the pmu event aliases definition, which should be located at:
  * /sys/bus/event_source/devices/<dev>/events as sysfs group attributes.
  */
-static int pmu_aliases(const char *name, struct list_head *head)
+static int pmu_aliases(int dirfd, const char *name, struct list_head *head)
 {
-       char path[PATH_MAX];
-
-       if (!perf_pmu__pathname_scnprintf(path, sizeof(path), name, "events"))
-               return -1;
+       int fd;
 
-       if (!file_available(path))
+       fd = perf_pmu__pathname_fd(dirfd, name, "events", O_DIRECTORY);
+       if (fd < 0)
                return 0;
 
-       if (pmu_aliases_parse(path, head))
+       /* it'll close the fd */
+       if (pmu_aliases_parse(fd, head))
                return -1;
 
        return 0;
@@ -544,14 +541,15 @@ static int pmu_alias_terms(struct perf_pmu_alias *alias,
 /* Add all pmus in sysfs to pmu list: */
 static void pmu_read_sysfs(void)
 {
-       char path[PATH_MAX];
+       int fd;
        DIR *dir;
        struct dirent *dent;
 
-       if (!perf_pmu__event_source_devices_scnprintf(path, sizeof(path)))
+       fd = perf_pmu__event_source_devices_fd();
+       if (fd < 0)
                return;
 
-       dir = opendir(path);
+       dir = fdopendir(fd);
        if (!dir)
                return;
 
@@ -559,7 +557,7 @@ static void pmu_read_sysfs(void)
                if (!strcmp(dent->d_name, ".") || !strcmp(dent->d_name, ".."))
                        continue;
                /* add to static LIST_HEAD(pmus): */
-               perf_pmu__find(dent->d_name);
+               perf_pmu__find2(fd, dent->d_name);
        }
 
        closedir(dir);
@@ -763,7 +761,7 @@ static int pmu_add_cpu_aliases_map_callback(const struct pmu_event *pe,
 
 new_alias:
        /* need type casts to override 'const' */
-       __perf_pmu__new_alias(data->head, NULL, (char *)pe->name, (char *)pe->desc,
+       __perf_pmu__new_alias(data->head, -1, (char *)pe->name, (char *)pe->desc,
                              (char *)pe->event, pe);
        return 0;
 }
@@ -814,7 +812,7 @@ static int pmu_add_sys_aliases_iter_fn(const struct pmu_event *pe,
 
        if (!strcmp(pmu->id, pe->compat) &&
            pmu_uncore_alias_match(pe->pmu, pmu->name)) {
-               __perf_pmu__new_alias(idata->head, NULL,
+               __perf_pmu__new_alias(idata->head, -1,
                                      (char *)pe->name,
                                      (char *)pe->desc,
                                      (char *)pe->event,
@@ -863,7 +861,7 @@ static int pmu_max_precise(struct perf_pmu *pmu)
        return max_precise;
 }
 
-static struct perf_pmu *pmu_lookup(const char *lookup_name)
+static struct perf_pmu *pmu_lookup(int dirfd, const char *lookup_name)
 {
        struct perf_pmu *pmu;
        LIST_HEAD(format);
@@ -884,13 +882,13 @@ static struct perf_pmu *pmu_lookup(const char *lookup_name)
         * type value and format definitions. Load both right
         * now.
         */
-       if (pmu_format(name, &format))
+       if (pmu_format(dirfd, name, &format))
                return NULL;
 
        /*
         * Check the aliases first to avoid unnecessary work.
         */
-       if (pmu_aliases(name, &aliases))
+       if (pmu_aliases(dirfd, name, &aliases))
                return NULL;
 
        pmu = zalloc(sizeof(*pmu));
@@ -1024,6 +1022,27 @@ bool evsel__is_aux_event(const struct evsel *evsel)
 }
 
 struct perf_pmu *perf_pmu__find(const char *name)
+{
+       struct perf_pmu *pmu;
+       int dirfd;
+
+       /*
+        * Once PMU is loaded it stays in the list,
+        * so we keep us from multiple reading/parsing
+        * the pmu format definitions.
+        */
+       pmu = pmu_find(name);
+       if (pmu)
+               return pmu;
+
+       dirfd = perf_pmu__event_source_devices_fd();
+       pmu = pmu_lookup(dirfd, name);
+       close(dirfd);
+
+       return pmu;
+}
+
+static struct perf_pmu *perf_pmu__find2(int dirfd, const char *name)
 {
        struct perf_pmu *pmu;
 
@@ -1036,7 +1055,7 @@ struct perf_pmu *perf_pmu__find(const char *name)
        if (pmu)
                return pmu;
 
-       return pmu_lookup(name);
+       return pmu_lookup(dirfd, name);
 }
 
 static struct perf_pmu_format *
@@ -1938,6 +1957,18 @@ int perf_pmu__event_source_devices_scnprintf(char *pathname, size_t size)
        return scnprintf(pathname, size, "%s/bus/event_source/devices/", sysfs);
 }
 
+int perf_pmu__event_source_devices_fd(void)
+{
+       char path[PATH_MAX];
+       const char *sysfs = sysfs__mountpoint();
+
+       if (!sysfs)
+               return -1;
+
+       scnprintf(path, sizeof(path), "%s/bus/event_source/devices/", sysfs);
+       return open(path, O_DIRECTORY);
+}
+
 /*
  * Fill 'buf' with the path to a file or folder in 'pmu_name' in
  * sysfs. For example if pmu_name = "cs_etm" and 'filename' = "format"
@@ -1957,6 +1988,14 @@ int perf_pmu__pathname_scnprintf(char *buf, size_t size,
        return scnprintf(buf, size, "%s%s/%s", base_path, pmu_name, filename);
 }
 
+int perf_pmu__pathname_fd(int dirfd, const char *pmu_name, const char *filename, int flags)
+{
+       char path[PATH_MAX];
+
+       scnprintf(path, sizeof(path), "%s/%s", pmu_name, filename);
+       return openat(dirfd, path, flags);
+}
+
 static void perf_pmu__delete(struct perf_pmu *pmu)
 {
        perf_pmu__del_formats(&pmu->format);
tools/perf/util/pmu.h +3 -2
index 72fd5de334c0..751c7016e7b6 100644
@@ -211,7 +211,7 @@ void perf_pmu_error(struct list_head *list, char *name, char const *msg);
 int perf_pmu__new_format(struct list_head *list, char *name,
                         int config, unsigned long *bits);
 void perf_pmu__set_format(unsigned long *bits, long from, long to);
-int perf_pmu__format_parse(char *dir, struct list_head *head);
+int perf_pmu__format_parse(int dirfd, struct list_head *head);
 void perf_pmu__del_formats(struct list_head *formats);
 
 struct perf_pmu *perf_pmu__scan(struct perf_pmu *pmu);
@@ -257,6 +257,8 @@ double perf_pmu__cpu_slots_per_cycle(void);
 int perf_pmu__event_source_devices_scnprintf(char *pathname, size_t size);
 int perf_pmu__pathname_scnprintf(char *buf, size_t size,
                                 const char *pmu_name, const char *filename);
+int perf_pmu__event_source_devices_fd(void);
+int perf_pmu__pathname_fd(int dirfd, const char *pmu_name, const char *filename, int flags);
 FILE *perf_pmu__open_file(struct perf_pmu *pmu, const char *name);
 
 void perf_pmu__destroy(void);
--

2.40.0.348.gf938b09366-goog

author maintainer
#12 lore raw

PATCH 6/9 perf pmu: Use relative path in perf_pmu__caps_parse()

In reply to Namhyung Kim

Likewise, it needs to traverse the pmu/caps directory, let's use openat() with the dirfd instead of open() using the absolute path.

---
 tools/perf/util/pmu.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)
tools/perf/util/pmu.c +6 -4
index 9fc6b8b5732b..0c1d87f10b23 100644
@@ -1804,6 +1804,7 @@ int perf_pmu__caps_parse(struct perf_pmu *pmu)
 	char caps_path[PATH_MAX];
 	DIR *caps_dir;
 	struct dirent *evt_ent;
+	int caps_fd;
 
 	if (pmu->caps_initialized)
 		return pmu->nr_caps;
@@ -1822,18 +1823,19 @@ int perf_pmu__caps_parse(struct perf_pmu *pmu)
 	if (!caps_dir)
 		return -EINVAL;
 
+	caps_fd = dirfd(caps_dir);
+
 	while ((evt_ent = readdir(caps_dir)) != NULL) {
-		char path[PATH_MAX + NAME_MAX + 1];
 		char *name = evt_ent->d_name;
 		char value[128];
 		FILE *file;
+		int fd;
 
 		if (!strcmp(name, ".") || !strcmp(name, ".."))
 			continue;
 
-		snprintf(path, sizeof(path), "%s/%s", caps_path, name);
-
-		file = fopen(path, "r");
+		fd = openat(caps_fd, name, O_RDONLY);
+		file = fdopen(fd, "r");
 		if (!file)
 			continue;
 
Signature · 1 line
2.40.0.348.gf938b09366-goog
author maintainer
#13 lore raw

PATCH 7/9 perf pmu: Use relative path in setup_pmu_alias_list()

In reply to Namhyung Kim

Likewise, x86 needs to traverse the PMU list to build alias. Let's use the new helpers to use relative paths.

---
 tools/perf/arch/x86/util/pmu.c | 13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)
tools/perf/arch/x86/util/pmu.c +7 -6
index f73b80dcd8bd..3c0de3370d7e 100644
@@ -71,7 +71,7 @@ static struct pmu_alias *pmu_alias__new(char *name, char *alias)
 
 static int setup_pmu_alias_list(void)
 {
-	char path[PATH_MAX];
+	int fd, dirfd;
 	DIR *dir;
 	struct dirent *dent;
 	struct pmu_alias *pmu_alias;
@@ -79,10 +79,11 @@ static int setup_pmu_alias_list(void)
 	FILE *file;
 	int ret = -ENOMEM;
 
-	if (!perf_pmu__event_source_devices_scnprintf(path, sizeof(path)))
+	dirfd = perf_pmu__event_source_devices_fd();
+	if (dirfd < 0)
 		return -1;
 
-	dir = opendir(path);
+	dir = fdopendir(dirfd);
 	if (!dir)
 		return -errno;
 
@@ -91,11 +92,11 @@ static int setup_pmu_alias_list(void)
 		    !strcmp(dent->d_name, ".."))
 			continue;
 
-		perf_pmu__pathname_scnprintf(path, sizeof(path), dent->d_name, "alias");
-		if (!file_available(path))
+		fd = perf_pmu__pathname_fd(dirfd, dent->d_name, "alias", O_RDONLY);
+		if (fd < 0)
 			continue;
 
-		file = fopen(path, "r");
+		file = fdopen(fd, "r");
 		if (!file)
 			continue;
 
Signature · 1 line
2.40.0.348.gf938b09366-goog
author maintainer
#14 lore raw

PATCH 8/9 perf pmu: Add perf_pmu__{open,scan}_file_at()

In reply to Namhyung Kim

These two helpers will also use openat() to reduce the overhead with relative pathnames. Convert other functions in pmu_lookup() to use the new helpers.

---
 tools/perf/util/pmu.c | 57 ++++++++++++++++++++++++++++++++++---------
 tools/perf/util/pmu.h |  6 ++++-
 2 files changed, 50 insertions(+), 13 deletions(-)
tools/perf/util/pmu.c +45 -12
index 0c1d87f10b23..78a407b42ad1 100644
@@ -567,7 +567,7 @@ static void pmu_read_sysfs(void)
  * Uncore PMUs have a "cpumask" file under sysfs. CPU PMUs (e.g. on arm/arm64)
  * may have a "cpus" file.
  */
-static struct perf_cpu_map *pmu_cpumask(const char *name)
+static struct perf_cpu_map *pmu_cpumask(int dirfd, const char *name)
 {
 	struct perf_cpu_map *cpus;
 	const char *templates[] = {
@@ -582,10 +582,11 @@ static struct perf_cpu_map *pmu_cpumask(const char *name)
 
 	strlcpy(pmu_name, name, sizeof(pmu_name));
 	for (template = templates; *template; template++) {
-		file = perf_pmu__open_file(&pmu, *template);
+		file = perf_pmu__open_file_at(&pmu, dirfd, *template);
 		if (!file)
 			continue;
 		cpus = perf_cpu_map__read(file);
+		fclose(file);
 		if (cpus)
 			return cpus;
 	}
@@ -593,15 +594,19 @@ static struct perf_cpu_map *pmu_cpumask(const char *name)
 	return NULL;
 }
 
-static bool pmu_is_uncore(const char *name)
+static bool pmu_is_uncore(int dirfd, const char *name)
 {
-	char path[PATH_MAX];
+	int fd;
 
 	if (perf_pmu__hybrid_mounted(name))
 		return false;
 
-	perf_pmu__pathname_scnprintf(path, sizeof(path), name, "cpumask");
-	return file_available(path);
+	fd = perf_pmu__pathname_fd(dirfd, name, "cpumask", O_PATH);
+	if (fd < 0)
+		return false;
+
+	close(fd);
+	return true;
 }
 
 static char *pmu_id(const char *name)
@@ -853,11 +858,11 @@ pmu_find_alias_name(const char *name __maybe_unused)
 	return NULL;
 }
 
-static int pmu_max_precise(struct perf_pmu *pmu)
+static int pmu_max_precise(int dirfd, struct perf_pmu *pmu)
 {
 	int max_precise = -1;
 
-	perf_pmu__scan_file(pmu, "caps/max_precise", "%d", &max_precise);
+	perf_pmu__scan_file_at(pmu, dirfd, "caps/max_precise", "%d", &max_precise);
 	return max_precise;
 }
 
@@ -895,14 +900,14 @@ static struct perf_pmu *pmu_lookup(int dirfd, const char *lookup_name)
 	if (!pmu)
 		return NULL;
 
-	pmu->cpus = pmu_cpumask(name);
+	pmu->cpus = pmu_cpumask(dirfd, name);
 	pmu->name = strdup(name);
 
 	if (!pmu->name)
 		goto err;
 
 	/* Read type, and ensure that type value is successfully assigned (return 1) */
-	if (perf_pmu__scan_file(pmu, "type", "%u", &type) != 1)
+	if (perf_pmu__scan_file_at(pmu, dirfd, "type", "%u", &type) != 1)
 		goto err;
 
 	alias_name = pmu_find_alias_name(name);
@@ -913,10 +918,10 @@ static struct perf_pmu *pmu_lookup(int dirfd, const char *lookup_name)
 	}
 
 	pmu->type = type;
-	pmu->is_uncore = pmu_is_uncore(name);
+	pmu->is_uncore = pmu_is_uncore(dirfd, name);
 	if (pmu->is_uncore)
 		pmu->id = pmu_id(name);
-	pmu->max_precise = pmu_max_precise(pmu);
+	pmu->max_precise = pmu_max_precise(dirfd, pmu);
 	pmu_add_cpu_aliases(&aliases, pmu);
 	pmu_add_sys_aliases(&aliases, pmu);
 
@@ -1730,6 +1735,17 @@ FILE *perf_pmu__open_file(struct perf_pmu *pmu, const char *name)
 	return fopen(path, "r");
 }
 
+FILE *perf_pmu__open_file_at(struct perf_pmu *pmu, int dirfd, const char *name)
+{
+	int fd;
+
+	fd = perf_pmu__pathname_fd(dirfd, pmu->name, name, O_RDONLY);
+	if (fd < 0)
+		return NULL;
+
+	return fdopen(fd, "r");
+}
+
 int perf_pmu__scan_file(struct perf_pmu *pmu, const char *name, const char *fmt,
 			...)
 {
@@ -1747,6 +1763,23 @@ int perf_pmu__scan_file(struct perf_pmu *pmu, const char *name, const char *fmt,
 	return ret;
 }
 
+int perf_pmu__scan_file_at(struct perf_pmu *pmu, int dirfd, const char *name,
+			   const char *fmt, ...)
+{
+	va_list args;
+	FILE *file;
+	int ret = EOF;
+
+	va_start(args, fmt);
+	file = perf_pmu__open_file_at(pmu, dirfd, name);
+	if (file) {
+		ret = vfscanf(file, fmt, args);
+		fclose(file);
+	}
+	va_end(args);
+	return ret;
+}
+
 bool perf_pmu__file_exists(struct perf_pmu *pmu, const char *name)
 {
 	char path[PATH_MAX];
tools/perf/util/pmu.h +5 -1
index 751c7016e7b6..32c3a75bca0e 100644
@@ -220,7 +220,12 @@ bool is_pmu_core(const char *name);
 void print_pmu_events(const struct print_callbacks *print_cb, void *print_state);
 bool pmu_have_event(const char *pname, const char *name);
 
+FILE *perf_pmu__open_file(struct perf_pmu *pmu, const char *name);
+FILE *perf_pmu__open_file_at(struct perf_pmu *pmu, int dirfd, const char *name);
+
 int perf_pmu__scan_file(struct perf_pmu *pmu, const char *name, const char *fmt, ...) __scanf(3, 4);
+int perf_pmu__scan_file_at(struct perf_pmu *pmu, int dirfd, const char *name,
+			   const char *fmt, ...) __scanf(4, 5);
 
 bool perf_pmu__file_exists(struct perf_pmu *pmu, const char *name);
 
@@ -259,7 +264,6 @@ int perf_pmu__pathname_scnprintf(char *buf, size_t size,
 				 const char *pmu_name, const char *filename);
 int perf_pmu__event_source_devices_fd(void);
 int perf_pmu__pathname_fd(int dirfd, const char *pmu_name, const char *filename, int flags);
-FILE *perf_pmu__open_file(struct perf_pmu *pmu, const char *name);
 
 void perf_pmu__destroy(void);
 
Signature · 1 line
2.40.0.348.gf938b09366-goog
author maintainer
#15 lore raw

PATCH 9/9 perf intel-pt: Use perf_pmu__scan_file_at() if possible

In reply to Namhyung Kim

Intel-PT calls perf_pmu__scan_file() a lot, let's use relative address when it accesses multiple files at one place.

---
 tools/perf/arch/x86/util/intel-pt.c | 52 ++++++++++++++++++-----------
 1 file changed, 32 insertions(+), 20 deletions(-)
tools/perf/arch/x86/util/intel-pt.c +32 -20
index 1e39a034cee9..2cff11de9d8a 100644
@@ -194,16 +194,19 @@ static u64 intel_pt_default_config(struct perf_pmu *intel_pt_pmu)
 	int pos = 0;
 	u64 config;
 	char c;
+	int dirfd;
+
+	dirfd = perf_pmu__event_source_devices_fd();
 
 	pos += scnprintf(buf + pos, sizeof(buf) - pos, "tsc");
 
-	if (perf_pmu__scan_file(intel_pt_pmu, "caps/mtc", "%d",
-				&mtc) != 1)
+	if (perf_pmu__scan_file_at(intel_pt_pmu, dirfd, "caps/mtc", "%d",
+				   &mtc) != 1)
 		mtc = 1;
 
 	if (mtc) {
-		if (perf_pmu__scan_file(intel_pt_pmu, "caps/mtc_periods", "%x",
-					&mtc_periods) != 1)
+		if (perf_pmu__scan_file_at(intel_pt_pmu, dirfd, "caps/mtc_periods", "%x",
+					   &mtc_periods) != 1)
 			mtc_periods = 0;
 		if (mtc_periods) {
 			mtc_period = intel_pt_pick_bit(mtc_periods, 3);
@@ -212,13 +215,13 @@ static u64 intel_pt_default_config(struct perf_pmu *intel_pt_pmu)
 		}
 	}
 
-	if (perf_pmu__scan_file(intel_pt_pmu, "caps/psb_cyc", "%d",
-				&psb_cyc) != 1)
+	if (perf_pmu__scan_file_at(intel_pt_pmu, dirfd, "caps/psb_cyc", "%d",
+				   &psb_cyc) != 1)
 		psb_cyc = 1;
 
 	if (psb_cyc && mtc_periods) {
-		if (perf_pmu__scan_file(intel_pt_pmu, "caps/psb_periods", "%x",
-					&psb_periods) != 1)
+		if (perf_pmu__scan_file_at(intel_pt_pmu, dirfd, "caps/psb_periods", "%x",
+					   &psb_periods) != 1)
 			psb_periods = 0;
 		if (psb_periods) {
 			psb_period = intel_pt_pick_bit(psb_periods, 3);
@@ -227,8 +230,8 @@ static u64 intel_pt_default_config(struct perf_pmu *intel_pt_pmu)
 		}
 	}
 
-	if (perf_pmu__scan_file(intel_pt_pmu, "format/pt", "%c", &c) == 1 &&
-	    perf_pmu__scan_file(intel_pt_pmu, "format/branch", "%c", &c) == 1)
+	if (perf_pmu__scan_file_at(intel_pt_pmu, dirfd, "format/pt", "%c", &c) == 1 &&
+	    perf_pmu__scan_file_at(intel_pt_pmu, dirfd, "format/branch", "%c", &c) == 1)
 		pos += scnprintf(buf + pos, sizeof(buf) - pos, ",pt,branch");
 
 	pr_debug2("%s default config: %s\n", intel_pt_pmu->name, buf);
@@ -236,6 +239,7 @@ static u64 intel_pt_default_config(struct perf_pmu *intel_pt_pmu)
 	intel_pt_parse_terms(intel_pt_pmu->name, &intel_pt_pmu->format, buf,
 			     &config);
 
+	close(dirfd);
 	return config;
 }
 
@@ -488,7 +492,7 @@ static void intel_pt_valid_str(char *str, size_t len, u64 valid)
 	}
 }
 
-static int intel_pt_val_config_term(struct perf_pmu *intel_pt_pmu,
+static int intel_pt_val_config_term(struct perf_pmu *intel_pt_pmu, int dirfd,
 				    const char *caps, const char *name,
 				    const char *supported, u64 config)
 {
@@ -498,11 +502,11 @@ static int intel_pt_val_config_term(struct perf_pmu *intel_pt_pmu,
 	u64 bits;
 	int ok;
 
-	if (perf_pmu__scan_file(intel_pt_pmu, caps, "%llx", &valid) != 1)
+	if (perf_pmu__scan_file_at(intel_pt_pmu, dirfd, caps, "%llx", &valid) != 1)
 		valid = 0;
 
 	if (supported &&
-	    perf_pmu__scan_file(intel_pt_pmu, supported, "%d", &ok) == 1 && !ok)
+	    perf_pmu__scan_file_at(intel_pt_pmu, dirfd, supported, "%d", &ok) == 1 && !ok)
 		valid = 0;
 
 	valid |= 1;
@@ -531,37 +535,45 @@ static int intel_pt_val_config_term(struct perf_pmu *intel_pt_pmu,
 static int intel_pt_validate_config(struct perf_pmu *intel_pt_pmu,
 				    struct evsel *evsel)
 {
-	int err;
+	int err, dirfd;
 	char c;
 
 	if (!evsel)
 		return 0;
 
+	dirfd = perf_pmu__event_source_devices_fd();
+	if (dirfd < 0)
+		return dirfd;
+
 	/*
 	 * If supported, force pass-through config term (pt=1) even if user
 	 * sets pt=0, which avoids senseless kernel errors.
 	 */
-	if (perf_pmu__scan_file(intel_pt_pmu, "format/pt", "%c", &c) == 1 &&
+	if (perf_pmu__scan_file_at(intel_pt_pmu, dirfd, "format/pt", "%c", &c) == 1 &&
 	    !(evsel->core.attr.config & 1)) {
 		pr_warning("pt=0 doesn't make sense, forcing pt=1\n");
 		evsel->core.attr.config |= 1;
 	}
 
-	err = intel_pt_val_config_term(intel_pt_pmu, "caps/cycle_thresholds",
+	err = intel_pt_val_config_term(intel_pt_pmu, dirfd, "caps/cycle_thresholds",
 				       "cyc_thresh", "caps/psb_cyc",
 				       evsel->core.attr.config);
 	if (err)
-		return err;
+		goto out;
 
-	err = intel_pt_val_config_term(intel_pt_pmu, "caps/mtc_periods",
+	err = intel_pt_val_config_term(intel_pt_pmu, dirfd, "caps/mtc_periods",
 				       "mtc_period", "caps/mtc",
 				       evsel->core.attr.config);
 	if (err)
-		return err;
+		goto out;
 
-	return intel_pt_val_config_term(intel_pt_pmu, "caps/psb_periods",
+	err = intel_pt_val_config_term(intel_pt_pmu, dirfd, "caps/psb_periods",
 					"psb_period", "caps/psb_cyc",
 					evsel->core.attr.config);
+
+out:
+	close(dirfd);
+	return err;
 }
 
 static void intel_pt_config_sample_mode(struct perf_pmu *intel_pt_pmu,
Signature · 1 line
2.40.0.348.gf938b09366-goog
reviewer Acked-by on v1 9/9
#16 lore raw

Re: [PATCH 9/9] perf intel-pt: Use perf_pmu__scan_file_at() if possible

Quoted 4 lines · Namhyung Kim

Intel-PT calls perf_pmu__scan_file() a lot, let's use relative address when it accesses multiple files at one place.

Quoted 145 lines
---
 tools/perf/arch/x86/util/intel-pt.c | 52 ++++++++++++++++++-----------
 1 file changed, 32 insertions(+), 20 deletions(-)
tools/perf/arch/x86/util/intel-pt.c +32 -20
index 1e39a034cee9..2cff11de9d8a 100644
@@ -194,16 +194,19 @@ static u64 intel_pt_default_config(struct perf_pmu *intel_pt_pmu)
 	int pos = 0;
 	u64 config;
 	char c;
+	int dirfd;
+
+	dirfd = perf_pmu__event_source_devices_fd();
 
 	pos += scnprintf(buf + pos, sizeof(buf) - pos, "tsc");
 
-	if (perf_pmu__scan_file(intel_pt_pmu, "caps/mtc", "%d",
-				&mtc) != 1)
+	if (perf_pmu__scan_file_at(intel_pt_pmu, dirfd, "caps/mtc", "%d",
+				   &mtc) != 1)
 		mtc = 1;
 
 	if (mtc) {
-		if (perf_pmu__scan_file(intel_pt_pmu, "caps/mtc_periods", "%x",
-					&mtc_periods) != 1)
+		if (perf_pmu__scan_file_at(intel_pt_pmu, dirfd, "caps/mtc_periods", "%x",
+					   &mtc_periods) != 1)
 			mtc_periods = 0;
 		if (mtc_periods) {
 			mtc_period = intel_pt_pick_bit(mtc_periods, 3);
@@ -212,13 +215,13 @@ static u64 intel_pt_default_config(struct perf_pmu *intel_pt_pmu)
 		}
 	}
 
-	if (perf_pmu__scan_file(intel_pt_pmu, "caps/psb_cyc", "%d",
-				&psb_cyc) != 1)
+	if (perf_pmu__scan_file_at(intel_pt_pmu, dirfd, "caps/psb_cyc", "%d",
+				   &psb_cyc) != 1)
 		psb_cyc = 1;
 
 	if (psb_cyc && mtc_periods) {
-		if (perf_pmu__scan_file(intel_pt_pmu, "caps/psb_periods", "%x",
-					&psb_periods) != 1)
+		if (perf_pmu__scan_file_at(intel_pt_pmu, dirfd, "caps/psb_periods", "%x",
+					   &psb_periods) != 1)
 			psb_periods = 0;
 		if (psb_periods) {
 			psb_period = intel_pt_pick_bit(psb_periods, 3);
@@ -227,8 +230,8 @@ static u64 intel_pt_default_config(struct perf_pmu *intel_pt_pmu)
 		}
 	}
 
-	if (perf_pmu__scan_file(intel_pt_pmu, "format/pt", "%c", &c) == 1 &&
-	    perf_pmu__scan_file(intel_pt_pmu, "format/branch", "%c", &c) == 1)
+	if (perf_pmu__scan_file_at(intel_pt_pmu, dirfd, "format/pt", "%c", &c) == 1 &&
+	    perf_pmu__scan_file_at(intel_pt_pmu, dirfd, "format/branch", "%c", &c) == 1)
 		pos += scnprintf(buf + pos, sizeof(buf) - pos, ",pt,branch");
 
 	pr_debug2("%s default config: %s\n", intel_pt_pmu->name, buf);
@@ -236,6 +239,7 @@ static u64 intel_pt_default_config(struct perf_pmu *intel_pt_pmu)
 	intel_pt_parse_terms(intel_pt_pmu->name, &intel_pt_pmu->format, buf,
 			     &config);
 
+	close(dirfd);
 	return config;
 }
 
@@ -488,7 +492,7 @@ static void intel_pt_valid_str(char *str, size_t len, u64 valid)
 	}
 }
 
-static int intel_pt_val_config_term(struct perf_pmu *intel_pt_pmu,
+static int intel_pt_val_config_term(struct perf_pmu *intel_pt_pmu, int dirfd,
 				    const char *caps, const char *name,
 				    const char *supported, u64 config)
 {
@@ -498,11 +502,11 @@ static int intel_pt_val_config_term(struct perf_pmu *intel_pt_pmu,
 	u64 bits;
 	int ok;
 
-	if (perf_pmu__scan_file(intel_pt_pmu, caps, "%llx", &valid) != 1)
+	if (perf_pmu__scan_file_at(intel_pt_pmu, dirfd, caps, "%llx", &valid) != 1)
 		valid = 0;
 
 	if (supported &&
-	    perf_pmu__scan_file(intel_pt_pmu, supported, "%d", &ok) == 1 && !ok)
+	    perf_pmu__scan_file_at(intel_pt_pmu, dirfd, supported, "%d", &ok) == 1 && !ok)
 		valid = 0;
 
 	valid |= 1;
@@ -531,37 +535,45 @@ static int intel_pt_val_config_term(struct perf_pmu *intel_pt_pmu,
 static int intel_pt_validate_config(struct perf_pmu *intel_pt_pmu,
 				    struct evsel *evsel)
 {
-	int err;
+	int err, dirfd;
 	char c;
 
 	if (!evsel)
 		return 0;
 
+	dirfd = perf_pmu__event_source_devices_fd();
+	if (dirfd < 0)
+		return dirfd;
+
 	/*
 	 * If supported, force pass-through config term (pt=1) even if user
 	 * sets pt=0, which avoids senseless kernel errors.
 	 */
-	if (perf_pmu__scan_file(intel_pt_pmu, "format/pt", "%c", &c) == 1 &&
+	if (perf_pmu__scan_file_at(intel_pt_pmu, dirfd, "format/pt", "%c", &c) == 1 &&
 	    !(evsel->core.attr.config & 1)) {
 		pr_warning("pt=0 doesn't make sense, forcing pt=1\n");
 		evsel->core.attr.config |= 1;
 	}
 
-	err = intel_pt_val_config_term(intel_pt_pmu, "caps/cycle_thresholds",
+	err = intel_pt_val_config_term(intel_pt_pmu, dirfd, "caps/cycle_thresholds",
 				       "cyc_thresh", "caps/psb_cyc",
 				       evsel->core.attr.config);
 	if (err)
-		return err;
+		goto out;
 
-	err = intel_pt_val_config_term(intel_pt_pmu, "caps/mtc_periods",
+	err = intel_pt_val_config_term(intel_pt_pmu, dirfd, "caps/mtc_periods",
 				       "mtc_period", "caps/mtc",
 				       evsel->core.attr.config);
 	if (err)
-		return err;
+		goto out;
 
-	return intel_pt_val_config_term(intel_pt_pmu, "caps/psb_periods",
+	err = intel_pt_val_config_term(intel_pt_pmu, dirfd, "caps/psb_periods",
 					"psb_period", "caps/psb_cyc",
 					evsel->core.attr.config);
+
+out:
+	close(dirfd);
+	return err;
 }
 
 static void intel_pt_config_sample_mode(struct perf_pmu *intel_pt_pmu,
reviewer Acked-by on the v1 series
#17 lore raw
In reply to Namhyung Kim
Quoted 59 lines · Namhyung Kim

Hello,

This patchset changes PMU info scanning on sysfs using openat() basically. I got reports of occasional contention on the opening files in sysfs. While the root cause was a separate issue, I discovered some inefficiencies in the perf code.

To scan PMUs, it roughly does something like below:

  dir = opendir("/sys/bus/event_source/devices");
  while (dentry = readdir(dir)) {
    char buf[PATH_MAX];
    snprintf(buf, sizeof(buf), "%s/%s",
             "/sys/bus/event_source/devices", dentry->d_name);
    fd = open(buf, O_RDONLY);
    ...

}

But this is not good since it needs to copy the string to build the absolute pathname, and it makes redundant pathname walk (from the /sys) in the kernel unnecessarily. We can use openat(2) to open the file in the given directory.

Add a couple of new helper to return the file descriptor of PMU directory so that it can use it with relative paths.

* perf_pmu__event_source_devices_fd()
- returns a fd for the PMU root ("/sys/bus/event_source/devices")

* perf_pmu__pathname_fd()
- returns a fd for "<pmu>/<file>" under the PMU root

Now the above code can be converted something like below:

  dirfd = perf_pmu__event_source_devices_fd();
  dir = fdopendir(dirfd);
  while (dentry = readdir(dir)) {
    fd = openat(dirfd, dentry->d_name, O_RDONLY);
    ...
  }

I added a benchmark for pmu-scan and it showed a slight speedup in the normal case too.

  $ ./perf.old bench internals pmu-scan
  # Running 'internals/pmu-scan' benchmark:
  Computing performance of sysfs PMU event scan for 100 times
    Average PMU scanning took: 6670.970 usec (+- 13.022 usec)
  $ ./perf.new bench internals pmu-scan
  # Running 'internals/pmu-scan' benchmark:
  Computing performance of sysfs PMU event scan for 100 times
    Average PMU scanning took: 6296.980 usec (+- 14.891 usec)

The 5~6% of improvement might be small but it may have bigger impact when the system is contended.

Quoted 6 lines

You can get the code from 'perf/pmu-scan-v1' branch in

git://git.kernel.org/pub/scm/linux/kernel/git/namhyung/linux-perf.git

Thanks,
Namhyung

Some nice stack size savings too. The reentrant fix was pre-existing. Series:

Thanks,
Ian

Quoted 30 lines

Namhyung Kim (9):
perf list: Use relative path for tracepoint scan perf tools: Fix a asan issue in parse_events_multi_pmu_add() perf pmu: Add perf_pmu__destroy() function
perf bench: Add pmu-scan benchmark
perf pmu: Use relative path for sysfs scan
perf pmu: Use relative path in perf_pmu__caps_parse() perf pmu: Use relative path in setup_pmu_alias_list() perf pmu: Add perf_pmu__{open,scan}_file_at()
perf intel-pt: Use perf_pmu__scan_file_at() if possible

 tools/perf/arch/x86/util/intel-pt.c |  52 ++++--
 tools/perf/arch/x86/util/pmu.c      |  13 +-
 tools/perf/bench/Build              |   1 +
 tools/perf/bench/bench.h            |   1 +
 tools/perf/bench/pmu-scan.c         | 184 ++++++++++++++++++
 tools/perf/builtin-bench.c          |   1 +
 tools/perf/tests/pmu.c              |   9 +-
 tools/perf/util/parse-events.c      |   2 +-
 tools/perf/util/pmu.c               | 278 ++++++++++++++++++++--------
 tools/perf/util/pmu.h               |  12 +-
 tools/perf/util/print-events.c      |  26 ++-
 11 files changed, 466 insertions(+), 113 deletions(-)
 create mode 100644 tools/perf/bench/pmu-scan.c

base-commit: 417c6adfb155f906f0441cc1034827f6e2b3c372 --
2.40.0.348.gf938b09366-goog