perf script: Allow +- operator for type specific fields option
authorJiri Olsa <jolsa@kernel.org>
Wed, 20 Feb 2019 12:27:57 +0000 (13:27 +0100)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Wed, 20 Feb 2019 19:15:35 +0000 (16:15 -0300)
Add support to add/remove fields for specific event types in -F option.
It's now possible to use '+-' after event type, like:

  # cat > test.c
  #include <stdio.h>

  int main(void)
  {
     printf("Hello world\n");
     while(1) {}
  }
  ^D
  # gcc -g -o test test.c
  # perf probe -x test 'test.c:5'
  # perf record -e '{cpu/cpu-cycles,period=10000/,probe_test:main}:S' ./test
  ...

  # perf script -Ftrace:+period,-cpu
            test  3859 396291.117343:      10275 cpu/cpu-cycles,period=10000/:      7f..
            test  3859 396291.118234:      11041 cpu/cpu-cycles,period=10000/:  ffffff..
            test  3859 396291.118234:          1              probe_test:main:
            test  3859 396291.118248:       8668 cpu/cpu-cycles,period=10000/:  ffffff..
            test  3859 396291.118263:      10139 cpu/cpu-cycles,period=10000/:  ffffff..

Committer testing:

Couldn't make the test above work, but tested it with:

  # perf probe -x hello main
  Added new event:
    probe_hello:main     (on main in /home/acme/c/hello)

  You can now use it in all perf tools, such as:

  perf record -e probe_hello:main -aR sleep 1

  # perf record -e probe_hello:main ./hello
  hello, world
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.025 MB perf.data (1 samples) ]
  # perf script
           hello 21454 [002] 254116.874005: probe_hello:main: (401126)
  #
  # perf script -Ftrace:+period,-cpu
           hello 21454 254116.874005:          1 probe_hello:main: (401126)

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20190220122800.864-4-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/Documentation/perf-script.txt
tools/perf/builtin-script.c

index 9e4def08d5693683b6646431378d9d31de3b4a8b..2e19fd7ffe35135a94fb55f66ea367171cc3fb2d 100644 (file)
@@ -159,6 +159,12 @@ OPTIONS
        the override, and the result of the above is that only S/W and H/W
        events are displayed with the given fields.
 
+       It's possible tp add/remove fields only for specific event type:
+
+               -Fsw:-cpu,-period
+
+       removes cpu and period from software events.
+
        For the 'wildcard' option if a user selected field is invalid for an
        event type, a message is displayed to the user that the option is
        ignored for that type. For example:
index 8d5fe092525c66fb0cf303a1990e2c52083e940d..373ea151dc602147f58a9ed21b56f9267e0d7a9d 100644 (file)
@@ -2560,6 +2560,10 @@ static int parse_output_fields(const struct option *opt __maybe_unused,
                        pr_warning("Overriding previous field request for %s events.\n",
                                   event_type(type));
 
+               /* Don't override defaults for +- */
+               if (strchr(tok, '+') || strchr(tok, '-'))
+                       goto parse;
+
                output[type].fields = 0;
                output[type].user_set = true;
                output[type].wildcard_set = false;
@@ -2644,6 +2648,10 @@ parse:
                                rc = -EINVAL;
                                goto out;
                        }
+                       if (change == REMOVE)
+                               output[type].fields &= ~all_output_options[i].field;
+                       else
+                               output[type].fields |= all_output_options[i].field;
                        output[type].user_set = true;
                        output[type].wildcard_set = true;
                }