Skip to content

RFC function profiler

Steven Rostedt lkml 1 version In review

Get series with b4
b4 am [email protected]

Versions newest first

  1. v1 RFC Latest5 patches 19 repliesIn review

Patches in v1 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/5tracing: add function profilerinclude/linux/ftrace.h kernel/trace/Kconfig kernel/trace/ftrace.c+334 -25 replies
2/5tracing: move function profiler data out of function structinclude/linux/ftrace.h kernel/trace/Kconfig kernel/trace/ftrace.c+263 -1919 replies
3/5tracing: adding function timings to function profilerkernel/trace/ftrace.c kernel/trace/trace.c kernel/trace/trace.h +3 more+117 -191 reply
4/5tracing: make the function profiler per cpukernel/trace/ftrace.c+130 -690 replies
5/5function-graph: add option to calculate graph time or notinclude/linux/ftrace.h kernel/trace/ftrace.c kernel/trace/trace.c +2 more+29 -82 replies

Reviews trailers per patch

No Reviewed-by, Acked-by or Tested-by trailers for v1 yet.

Read the v1 thread

Cover letter [PATCH 0/5][RFC] [RFC] function profiler

The following patches are RFC, and to discuss is this work would be useful.

This patch series adds a function profiler hook to the function tracer and function graph tracer. That is, it uses the mcount calls to track the functions being called. Now with the function graph tracer it also keeps track of the times the functions have been executing.

Note, this is similar to oprofile, but is not a sampling approach, but a true recorder. Of course, by recording every function the overhead compared to oprofile is higher. With the use of dynamic ftrace, you could just trace certain functions, or just a certain module. The flexibility is a bit greater with this profiling.

The last patch adds another feature to the function graph. It gives the ability to only trace the amount of time inside a function and not the amount spent in sub functions that are also recorded. That is, if A calls B, you can decide if the time spent in A should include the time spent in B. This is an option added to trace_options.

Again, this is all RFC, but if there is any interest in having this feature, prehaps we should include it.

-- Steve

The following patches are in:

git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-2.6-trace.git

branch: rfc/tip/tracing/function-profiler

Steven Rostedt (5):
tracing: add function profiler
tracing: move function profiler data out of function struct
tracing: adding function timings to function profiler
tracing: make the function profiler per cpu
function-graph: add option to calculate graph time or not

----

 include/linux/ftrace.h               |    3 +-
 kernel/trace/Kconfig                 |   15 +
 kernel/trace/ftrace.c                |  558 +++++++++++++++++++++++++++++++++-
 kernel/trace/trace.c                 |   15 +-
 kernel/trace/trace.h                 |    4 +-
 kernel/trace/trace_functions_graph.c |   25 ++-
 kernel/trace/trace_output.c          |   10 +
 kernel/trace/trace_output.h          |    2 +
 8 files changed, 608 insertions(+), 24 deletions(-)