linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH] perf tools: Add sample types for bpf-output event
@ 2016-04-01 13:26 Wang Nan
  2016-04-01 15:16 ` Arnaldo Carvalho de Melo
  2016-04-06  7:10 ` [tip:perf/core] perf bpf: Add sample types for 'bpf-output' event tip-bot for Wang Nan
  0 siblings, 2 replies; 4+ messages in thread
From: Wang Nan @ 2016-04-01 13:26 UTC (permalink / raw)
  To: acme
  Cc: linux-kernel, pi3orama, Wang Nan, Arnaldo Carvalho de Melo,
	Adrian Hunter, David Ahern, Jiri Olsa, Milian Wolff,
	Namhyung Kim

Before this patch we can see very large time in the events before
bpf-output event. For example:

 # ./perf trace -vv -T --ev sched:sched_switch \
                       --ev bpf-output/no-inherit,name=evt/ \
                       --ev ./test_bpf_trace.c/map:channel.event=evt/ \
                       usleep 10
 ...
 18446744073709.551 (18446564645918.480 ms): usleep/4157 nanosleep(rqtp: 0x7ffd3f0dc4e0                                        ) ...
 18446744073709.551 (         ): evt:Raise a BPF event!..)
 179427791.076 (         ): perf_bpf_probe:func_begin:(ffffffff810eb9a0))
 179427791.081 (         ): sched:sched_switch:usleep:4157 [120] S ==> swapper/2:0 [120])
 ...

We can also see the differences between bpf-output events and
breakpoint events:

For bpf output event:
   sample_type                    IP|TID|RAW|IDENTIFIER

For tracepoint events:
   sample_type                    IP|TID|TIME|CPU|PERIOD|RAW|IDENTIFIER

This patch fix this differences by adding more sample type for
bpf-output events.

After this patch:

 # ./perf trace -vv -T --ev sched:sched_switch \
                       --ev bpf-output/no-inherit,name=evt/ \
                       --ev ./test_bpf_trace.c/map:channel.event=evt/ \
                       usleep 10
 ...
 179877370.878 ( 0.003 ms): usleep/5336 nanosleep(rqtp: 0x7ffff866c450                                        ) ...
 179877370.878 (         ): evt:Raise a BPF event!..)
 179877370.878 (         ): perf_bpf_probe:func_begin:(ffffffff810eb9a0))
 179877370.882 (         ): sched:sched_switch:usleep:5336 [120] S ==> swapper/4:0 [120])
 179877370.945 (         ): evt:Raise a BPF event!..)
 ...

 # ./perf trace -vv -T --ev sched:sched_switch \
                       --ev bpf-output/no-inherit,name=evt/ \
                       --ev ./test_bpf_trace.c/map:channel.event=evt/ \
                       usleep 10 2>&1 | grep sample_type
  sample_type                      IP|TID|TIME|ID|CPU|PERIOD|RAW
  sample_type                      IP|TID|TIME|ID|CPU|PERIOD|RAW
  sample_type                      IP|TID|TIME|ID|CPU|PERIOD|RAW
  sample_type                      IP|TID|TIME|ID|CPU|PERIOD|RAW
  sample_type                      IP|TID|TIME|ID|CPU|PERIOD|RAW
  sample_type                      IP|TID|TIME|ID|CPU|PERIOD|RAW

The 'IDENTIFIER' is not required because all events have the
same sample_type.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Milian Wolff <milian.wolff@kdab.com>
Cc: Namhyung Kim <namhyung@kernel.org>
---
 tools/perf/util/evsel.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 738ce22..3fd7c2c 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -226,7 +226,8 @@ struct perf_evsel *perf_evsel__new_idx(struct perf_event_attr *attr, int idx)
 		perf_evsel__init(evsel, attr, idx);
 
 	if (perf_evsel__is_bpf_output(evsel)) {
-		evsel->attr.sample_type |= PERF_SAMPLE_RAW;
+		evsel->attr.sample_type |= (PERF_SAMPLE_RAW | PERF_SAMPLE_TIME |
+					    PERF_SAMPLE_CPU | PERF_SAMPLE_PERIOD),
 		evsel->attr.sample_period = 1;
 	}
 
-- 
1.8.3.4

^ permalink raw reply related	[flat|nested] 4+ messages in thread

* Re: [PATCH] perf tools: Add sample types for bpf-output event
  2016-04-01 13:26 [PATCH] perf tools: Add sample types for bpf-output event Wang Nan
@ 2016-04-01 15:16 ` Arnaldo Carvalho de Melo
  2016-04-01 15:19   ` Arnaldo Carvalho de Melo
  2016-04-06  7:10 ` [tip:perf/core] perf bpf: Add sample types for 'bpf-output' event tip-bot for Wang Nan
  1 sibling, 1 reply; 4+ messages in thread
From: Arnaldo Carvalho de Melo @ 2016-04-01 15:16 UTC (permalink / raw)
  To: Wang Nan
  Cc: linux-kernel, pi3orama, Arnaldo Carvalho de Melo, Adrian Hunter,
	Ingo Molnar, David Ahern, Jiri Olsa, Milian Wolff, Namhyung Kim,
	Brendan Gregg

Em Fri, Apr 01, 2016 at 01:26:42PM +0000, Wang Nan escreveu:
> Before this patch we can see very large time in the events before
> bpf-output event. For example:
> 
>  # perf trace --ev bpf-output/no-inherit,name=evt/ \
>               --ev ./test_bpf_trace.c/map:channel.event=evt/ \
>               usleep 10

Thanks, applied and tested, now one idea that occurred to me to shorten
the above command line: automagically create a
"__perf_trace_bpf_stdout__" bpf-output event when a .c bpf event is
specified and no bpf-output is present, i.e. the following command line
would produce the same result as the one above:

  # trace --ev test_bpf_trace.c usleep 10

Well, it would have to build test_bpf_trace.c and see if it references
the equivalent of a "stdout", i.e. it expects a bpf-output event to be
present to send output to.

I.e. in this example we have a:

 struct bpf_map_def SEC("maps") channel = {

That later on you use to do "puts(msg)" like operations, i.e. to a
"stdout" of sorts:

 func(void *ctx, int type)
 {
	char output_str[] = "Raise a BPF event!";
	char err_str[] = "BAD %d\n";
	int err;

        err = perf_event_output(ctx, &channel, get_smp_processor_id(),
			        &output_str, sizeof(output_str));
	if (err)
		trace_printk(err_str, sizeof(err_str), err);
        return 1;
 }

Perhaps, to make all more familiar we could even define equivalents to
stdio.h functions like puts, printf, fputs, etc, that would send to this
bpf-output based "stdout" "channel", then the above would end up being:


 func(void *ctx, int type)
 {
        char err_str[] = "BAD %d\n";
        int err;

        err = puts("Raise a BPF event!");
        if (err)
                trace_printk(err_str, sizeof(err_str), err);
        return 1;
 }

This trace_printk() in turn could become error() (glibc's error.h header), i.e.
the error mechanism would use the equivalent to userland's "syslog", i.e.
trace_printk :-)

In general trying to make BPF C scriptlets fed via perf to be as compact as
possible, hiding all these details while allowing them to be used, if desired.

- Arnaldo

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [PATCH] perf tools: Add sample types for bpf-output event
  2016-04-01 15:16 ` Arnaldo Carvalho de Melo
@ 2016-04-01 15:19   ` Arnaldo Carvalho de Melo
  0 siblings, 0 replies; 4+ messages in thread
From: Arnaldo Carvalho de Melo @ 2016-04-01 15:19 UTC (permalink / raw)
  To: Arnaldo Carvalho de Melo
  Cc: Wang Nan, linux-kernel, pi3orama, Adrian Hunter, Ingo Molnar,
	David Ahern, Jiri Olsa, Milian Wolff, Namhyung Kim,
	Brendan Gregg

Em Fri, Apr 01, 2016 at 12:16:10PM -0300, Arnaldo Carvalho de Melo escreveu:
> Perhaps, to make all more familiar we could even define equivalents to
> stdio.h functions like puts, printf, fputs, etc, that would send to this
> bpf-output based "stdout" "channel", then the above would end up being:
> 
>  func(void *ctx, int type)
>  {
>         char err_str[] = "BAD %d\n";
>         int err;
> 
>         err = puts("Raise a BPF event!");
>         if (err)
>                 trace_printk(err_str, sizeof(err_str), err);
>         return 1;
>  }
> 
> This trace_printk() in turn could become error() (glibc's error.h header), i.e.
> the error mechanism would use the equivalent to userland's "syslog", i.e.
> trace_printk :-)
> 
> In general trying to make BPF C scriptlets fed via perf to be as compact as
> possible, hiding all these details while allowing them to be used, if desired.

One extra possibility would be that it would look so much like a user
space C program that we could test it without loading it to the kernel,
with just building it with a different header and feeding it data as it
would get inside the kernel :-)

- Arnaldo

^ permalink raw reply	[flat|nested] 4+ messages in thread

* [tip:perf/core] perf bpf: Add sample types for 'bpf-output' event
  2016-04-01 13:26 [PATCH] perf tools: Add sample types for bpf-output event Wang Nan
  2016-04-01 15:16 ` Arnaldo Carvalho de Melo
@ 2016-04-06  7:10 ` tip-bot for Wang Nan
  1 sibling, 0 replies; 4+ messages in thread
From: tip-bot for Wang Nan @ 2016-04-06  7:10 UTC (permalink / raw)
  To: linux-tip-commits
  Cc: adrian.hunter, acme, milian.wolff, dsahern, hpa, linux-kernel,
	jolsa, namhyung, tglx, mingo, wangnan0

Commit-ID:  d37ba880598654fda10b312331377cdca3edd574
Gitweb:     http://git.kernel.org/tip/d37ba880598654fda10b312331377cdca3edd574
Author:     Wang Nan <wangnan0@huawei.com>
AuthorDate: Fri, 1 Apr 2016 13:26:42 +0000
Committer:  Arnaldo Carvalho de Melo <acme@redhat.com>
CommitDate: Fri, 1 Apr 2016 18:46:25 -0300

perf bpf: Add sample types for 'bpf-output' event

Before this patch we can see very large time in the events before the
'bpf-output' event. For example:

  # perf trace -vv -T --ev sched:sched_switch \
                      --ev bpf-output/no-inherit,name=evt/ \
                      --ev ./test_bpf_trace.c/map:channel.event=evt/ \
                      usleep 10
  ...
  18446744073709.551 (18446564645918.480 ms): usleep/4157 nanosleep(rqtp: 0x7ffd3f0dc4e0) ...
  18446744073709.551 (         ): evt:Raise a BPF event!..)
  179427791.076 (         ): perf_bpf_probe:func_begin:(ffffffff810eb9a0))
  179427791.081 (         ): sched:sched_switch:usleep:4157 [120] S ==> swapper/2:0 [120])
  ...

We can also see the differences between bpf-output events and
breakpoint events:

For bpf output event:
   sample_type                    IP|TID|RAW|IDENTIFIER

For tracepoint events:
   sample_type                    IP|TID|TIME|CPU|PERIOD|RAW|IDENTIFIER

This patch fix this differences by adding more sample type for
bpf-output events.

After this patch:

  # perf trace -vv -T --ev sched:sched_switch \
                      --ev bpf-output/no-inherit,name=evt/ \
                      --ev ./test_bpf_trace.c/map:channel.event=evt/ \
                      usleep 10
  ...
  179877370.878 ( 0.003 ms): usleep/5336 nanosleep(rqtp: 0x7ffff866c450) ...
  179877370.878 (         ): evt:Raise a BPF event!..)
  179877370.878 (         ): perf_bpf_probe:func_begin:(ffffffff810eb9a0))
  179877370.882 (         ): sched:sched_switch:usleep:5336 [120] S ==> swapper/4:0 [120])
  179877370.945 (         ): evt:Raise a BPF event!..)
  ...

  # ./perf trace -vv -T --ev sched:sched_switch \
                        --ev bpf-output/no-inherit,name=evt/ \
                        --ev ./test_bpf_trace.c/map:channel.event=evt/ \
                        usleep 10 2>&1 | grep sample_type
  sample_type                      IP|TID|TIME|ID|CPU|PERIOD|RAW
  sample_type                      IP|TID|TIME|ID|CPU|PERIOD|RAW
  sample_type                      IP|TID|TIME|ID|CPU|PERIOD|RAW
  sample_type                      IP|TID|TIME|ID|CPU|PERIOD|RAW
  sample_type                      IP|TID|TIME|ID|CPU|PERIOD|RAW
  sample_type                      IP|TID|TIME|ID|CPU|PERIOD|RAW

The 'IDENTIFIER' info is not required because all events have the same
sample_type.

Committer notes:

Further testing, on top of the changes making 'perf trace' avoid samples
from events without PERF_SAMPLE_TIME:

Before:

  # trace --ev bpf-output/no-inherit,name=evt/ --ev /home/acme/bpf/test_bpf_trace.c/map:channel.event=evt/ usleep 10
  <SNIP>
    0.560 ( 0.001 ms): brk(                                                   ) = 0x55e5a1df8000
    18446640227439.430 (18446640227438.859 ms): nanosleep(rqtp: 0x7ffc96643370) ...
    18446640227439.430 (         ): evt:Raise a BPF event!..)
    0.576 (         ): perf_bpf_probe:func_begin:(ffffffff81112460))
    18446640227439.430 (         ): evt:Raise a BPF event!..)
    0.645 (         ): perf_bpf_probe:func_end:(ffffffff81112460 <- ffffffff81003d92))
    0.646 ( 0.076 ms):  ... [continued]: nanosleep()) = 0
  #

After:

  # trace --ev bpf-output/no-inherit,name=evt/ --ev /home/acme/bpf/test_bpf_trace.c/map:channel.event=evt/ usleep 10
  <SNIP>
     0.292 ( 0.001 ms): brk(                          ) = 0x55c7cd6e1000
     0.302 ( 0.004 ms): nanosleep(rqtp: 0x7ffedd8bc0f0) ...
     0.302 (         ): evt:Raise a BPF event!..)
     0.303 (         ): perf_bpf_probe:func_begin:(ffffffff81112460))
     0.397 (         ): evt:Raise a BPF event!..)
     0.397 (         ): perf_bpf_probe:func_end:(ffffffff81112460 <- ffffffff81003d92))
     0.398 ( 0.100 ms):  ... [continued]: nanosleep()) = 0

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Reported-and-Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Milian Wolff <milian.wolff@kdab.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1459517202-42320-1-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/util/evsel.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 738ce22..3fd7c2c 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -226,7 +226,8 @@ struct perf_evsel *perf_evsel__new_idx(struct perf_event_attr *attr, int idx)
 		perf_evsel__init(evsel, attr, idx);
 
 	if (perf_evsel__is_bpf_output(evsel)) {
-		evsel->attr.sample_type |= PERF_SAMPLE_RAW;
+		evsel->attr.sample_type |= (PERF_SAMPLE_RAW | PERF_SAMPLE_TIME |
+					    PERF_SAMPLE_CPU | PERF_SAMPLE_PERIOD),
 		evsel->attr.sample_period = 1;
 	}
 

^ permalink raw reply related	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2016-04-06  7:10 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-04-01 13:26 [PATCH] perf tools: Add sample types for bpf-output event Wang Nan
2016-04-01 15:16 ` Arnaldo Carvalho de Melo
2016-04-01 15:19   ` Arnaldo Carvalho de Melo
2016-04-06  7:10 ` [tip:perf/core] perf bpf: Add sample types for 'bpf-output' event tip-bot for Wang Nan

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).