Implement --delay on perf trace

PATCH In review Alexis Berlemont 3 participants 4 messages Oct 10, 2016 – Oct 24, 2016 perf events

lore

Share & tools

lore

Alexis Berlemont

author

2016-10-10 05:45 UTC #0 lore raw

PATCH 1/1 Implement --delay on perf trace

Hi,

In the perf wiki todo-list, there is an entry regarding initial-delay and perf trace; the following small patch tries to fulfill this point. It has been generated against the branch tip/perf/core.

It has only been implemented in the "trace__run" case.

Ex.:

$ sudo strace -- ./perf trace --delay 5 sleep 1 2>&1
...
fcntl(7, F_SETFL, O_RDONLY|O_NONBLOCK)  = 0
ioctl(7, PERF_EVENT_IOC_ID, 0x7ffc8fd35718) = 0
ioctl(11, PERF_EVENT_IOC_SET_OUTPUT, 0x7) = 0
fcntl(11, F_SETFL, O_RDONLY|O_NONBLOCK) = 0
ioctl(11, PERF_EVENT_IOC_ID, 0x7ffc8fd35718) = 0
write(6, "\0", 1)                       = 1
close(6)                                = 0
nanosleep({0, 5000000}, NULL)           = 0  # DELAY OF 5 MS BEFORE ENABLING THE EVNTS
ioctl(3, PERF_EVENT_IOC_ENABLE, 0)      = 0
ioctl(4, PERF_EVENT_IOC_ENABLE, 0)      = 0
ioctl(5, PERF_EVENT_IOC_ENABLE, 0)      = 0
ioctl(7, PERF_EVENT_IOC_ENABLE, 0)      = 0
...

Alexis.

Alexis Berlemont (1):
perf: implement --delay on perf trace

 tools/perf/builtin-trace.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

Signature · 1 line

2.10.0

Alexis Berlemont

author

2016-10-10 05:45 UTC #1 lore raw

PATCH 1/1 perf: implement --delay on perf trace

Signed-off-by Alexis Berlemont <[email protected]>

---
 tools/perf/builtin-trace.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

tools/perf/builtin-trace.c +9 -1

index c298bd3..0bae454 100644
@@ -2310,12 +2310,17 @@ static int trace__run(struct trace *trace, int argc, const char **argv)
 	if (err < 0)
 		goto out_error_mmap;
 
-	if (!target__none(&trace->opts.target))
+	if (!target__none(&trace->opts.target) && !trace->opts.initial_delay)
 		perf_evlist__enable(evlist);
 
 	if (forks)
 		perf_evlist__start_workload(evlist);
 
+	if (trace->opts.initial_delay) {
+		usleep(trace->opts.initial_delay * 1000);
+		perf_evlist__enable(evlist);
+	}
+
 	trace->multiple_threads = thread_map__pid(evlist->threads, 0) == -1 ||
 				  evlist->threads->nr > 1 ||
 				  perf_evlist__first(evlist)->attr.inherit;

@@ -2816,6 +2821,9 @@ int cmd_trace(int argc, const char **argv, const char *prefix __maybe_unused)
 		     "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
 	OPT_UINTEGER(0, "proc-map-timeout", &trace.opts.proc_map_timeout,
 			"per thread proc mmap processing timeout in ms"),
+	OPT_UINTEGER('D', "delay", &trace.opts.initial_delay,
+		     "ms to wait before starting measurement after program "
+		     "start"),
 	OPT_END()
 	};
 	bool __maybe_unused max_stack_user_set = true;

Signature · 1 line

2.10.0

In the perf wiki todo-list[1], there is an entry regarding initial-delay and 'perf trace'; the following small patch tries to fulfill this point. It has been generated against the branch tip/perf/core.

It has only been implemented in the "trace__run" case.

Ex.:

  $ sudo strace -- ./perf trace --delay 5 sleep 1 2>&1
  ...
  fcntl(7, F_SETFL, O_RDONLY|O_NONBLOCK)  = 0
  ioctl(7, PERF_EVENT_IOC_ID, 0x7ffc8fd35718) = 0
  ioctl(11, PERF_EVENT_IOC_SET_OUTPUT, 0x7) = 0
  fcntl(11, F_SETFL, O_RDONLY|O_NONBLOCK) = 0
  ioctl(11, PERF_EVENT_IOC_ID, 0x7ffc8fd35718) = 0
  write(6, "\0", 1)                       = 1
  close(6)                                = 0
  nanosleep({0, 5000000}, NULL)           = 0  # DELAY OF 5 MS BEFORE ENABLING THE EVENTS
  ioctl(3, PERF_EVENT_IOC_ENABLE, 0)      = 0
  ioctl(4, PERF_EVENT_IOC_ENABLE, 0)      = 0
  ioctl(5, PERF_EVENT_IOC_ENABLE, 0)      = 0
  ioctl(7, PERF_EVENT_IOC_ENABLE, 0)      = 0
  ...

[1]: https://perf.wiki.kernel.org/index.php/Todo

Signed-off-by Alexis Berlemont <[email protected]>

Suggested-and-Tested-by Arnaldo Carvalho de Melo <[email protected]>

Cc Alexander Shishkin <[email protected]>

Cc Peter Zijlstra <[email protected]>

Link http://lkml.kernel.org/r/[email protected] open here

[ Add entry to the manpage, cut'n'pasted from stat's and record's ]

Signed-off-by Arnaldo Carvalho de Melo <[email protected]>

---
 tools/perf/Documentation/perf-trace.txt |  5 +++++
 tools/perf/builtin-trace.c              | 10 +++++++++-
 2 files changed, 14 insertions(+), 1 deletion(-)

tools/perf/Documentation/perf-trace.txt +5 -0

index 1ab0782..781b019 100644
@@ -39,6 +39,11 @@ OPTIONS
 	Prefixing with ! shows all syscalls but the ones specified.  You may
 	need to escape it.
 
+-D msecs::
+--delay msecs::
+After starting the program, wait msecs before measuring. This is useful to
+filter out the startup phase of the program, which is often very different.
+
 -o::
 --output=::
 	Output file name.

tools/perf/builtin-trace.c +9 -1

index c298bd3..0bae454 100644
@@ -2310,12 +2310,17 @@ static int trace__run(struct trace *trace, int argc, const char **argv)
 	if (err < 0)
 		goto out_error_mmap;
 
-	if (!target__none(&trace->opts.target))
+	if (!target__none(&trace->opts.target) && !trace->opts.initial_delay)
 		perf_evlist__enable(evlist);
 
 	if (forks)
 		perf_evlist__start_workload(evlist);
 
+	if (trace->opts.initial_delay) {
+		usleep(trace->opts.initial_delay * 1000);
+		perf_evlist__enable(evlist);
+	}
+
 	trace->multiple_threads = thread_map__pid(evlist->threads, 0) == -1 ||
 				  evlist->threads->nr > 1 ||
 				  perf_evlist__first(evlist)->attr.inherit;

@@ -2816,6 +2821,9 @@ int cmd_trace(int argc, const char **argv, const char *prefix __maybe_unused)
 		     "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
 	OPT_UINTEGER(0, "proc-map-timeout", &trace.opts.proc_map_timeout,
 			"per thread proc mmap processing timeout in ms"),
+	OPT_UINTEGER('D', "delay", &trace.opts.initial_delay,
+		     "ms to wait before starting measurement after program "
+		     "start"),
 	OPT_END()
 	};
 	bool __maybe_unused max_stack_user_set = true;

Arnaldo Carvalho de Melo

maintainer

2016-10-14 16:05 UTC #3 lore raw

In reply to Alexis Berlemont

Em Mon, Oct 10, 2016 at 07:43:27AM +0200, Alexis Berlemont escreveu:

Quoted 7 lines

Hi,
In the perf wiki todo-list, there is an entry regarding initial-delay and perf trace; the following small patch tries to fulfill this point. It has been generated against the branch tip/perf/core.
It has only been implemented in the "trace__run" case.

Ok, moved the description to the cset comment and added the entry to tools/perf/Documentation/perf-trace.txt, i.e. to its man page, just copying from the exact same description in 'perf stat' and 'perf record's ones.

What is next? :-)

- Arnaldo

Quoted 28 lines

Ex.:

$ sudo strace -- ./perf trace --delay 5 sleep 1 2>&1
...
fcntl(7, F_SETFL, O_RDONLY|O_NONBLOCK)  = 0
ioctl(7, PERF_EVENT_IOC_ID, 0x7ffc8fd35718) = 0
ioctl(11, PERF_EVENT_IOC_SET_OUTPUT, 0x7) = 0
fcntl(11, F_SETFL, O_RDONLY|O_NONBLOCK) = 0
ioctl(11, PERF_EVENT_IOC_ID, 0x7ffc8fd35718) = 0
write(6, "\0", 1)                       = 1
close(6)                                = 0
nanosleep({0, 5000000}, NULL)           = 0  # DELAY OF 5 MS BEFORE ENABLING THE EVNTS
ioctl(3, PERF_EVENT_IOC_ENABLE, 0)      = 0
ioctl(4, PERF_EVENT_IOC_ENABLE, 0)      = 0
ioctl(5, PERF_EVENT_IOC_ENABLE, 0)      = 0
ioctl(7, PERF_EVENT_IOC_ENABLE, 0)      = 0
...

Alexis.

Alexis Berlemont (1):
perf: implement --delay on perf trace

 tools/perf/builtin-trace.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

Signature · 1 line

2.10.0