[RFC 0/4] trace,perf: Enabling ftrace:function tracepoint for perf

[Jiri Olsa <jolsa@redhat.com>]

hi,

attached patches add perf support for function trace.

- Patch 1 adds support to register ftrace:function tracepoint.

- Patch 2 adds a way to used filtering for ip and parent_ip fields.

- Patches 3 and 4 are adding new ioctl to specify the filter on the
  "struct ftrace" level.

  I know this is probably not a good idea, but I dont see a good way
  to combine ftrace_ops filtering with tracepoint filtering other than
  keep them separated.

  My other guess is instead of ioctl use some keyword to specify functions
  followed by standard filter.. something like:

    "FUNCTION_TRACE(sys_open sys_close sys_read) (ip == sys_open)"

  Please take this more as a question about interface than forcing
  new ioctl.. ;)


attached patches:
 - 1/4 perf, ftrace: Add perf support to use function tracepoint
 - 2/4 perf, ftrace: Add filter support
 - 3/4 perf, ftrace: Add new perf ioctl for function trace filter
 - 4/4 perf, ftrace: Add ftrace option for perf stat command


thanks for comments,
jirka
---
 include/linux/ftrace.h             |    7 +-
 include/linux/ftrace_event.h       |    9 +-
 include/linux/perf_event.h         |    2 +
 kernel/events/core.c               |   14 +++
 kernel/trace/ftrace.c              |   34 +++++-
 kernel/trace/trace.h               |    5 +
 kernel/trace/trace_event_perf.c    |  225 +++++++++++++++++++++++++++++-------
 kernel/trace/trace_events.c        |   10 +-
 kernel/trace/trace_events_filter.c |   32 ++++-
 kernel/trace/trace_export.c        |   21 ++++
 kernel/trace/trace_kprobe.c        |    6 +-
 kernel/trace/trace_syscalls.c      |   14 ++-
 tools/perf/builtin-stat.c          |    2 +
 tools/perf/util/evlist.c           |   26 ++++-
 tools/perf/util/evlist.h           |    2 +
 tools/perf/util/evsel.h            |    1 +
 tools/perf/util/parse-events.c     |   37 ++++++-
 tools/perf/util/parse-events.h     |    1 +
 18 files changed, 376 insertions(+), 72 deletions(-)