[PATCH 00/16] perf tools: Add libdw DWARF unwind support

[Jiri Olsa <jolsa@redhat.com>]

hi,
adding libdw DWARF unwind support, which is part of just released
elfutils package 0.158.

We can now compile perf to have either libunwind DWARF unwind,
(which is still default) or the new one libdw unwind.

Examples:
  - compile in libdw unwinder if present:
    $ make NO_LIBUNWIND=1

  - compile in libdw (with libdw installation directory) unwinder if present:
    $ make LIBDW_DIR=/opt/elfutils/ NO_LIBUNWIND=1

  - disable post dwarf unwind completely:
    $ make NO_LIBUNWIND=1 NO_LIBDW_DWARF_UNWIND=1

Performance wise the libdw unwinder seems to be faster.
Following results are from 2GB perf.data file with 249156
samples:

  The libdw unwind perf:
  $ time perf report -i ~/perf.data > perf.data.ldw

  real    0m35.746s
  user    0m24.211s
  sys     0m11.214s

  The libunwind unwind perf:
  $ time perf report -i ~/perf.data > perf.data.unw

  real    1m41.149s
  user    1m9.694s
  sys     0m30.595s

In example above I've got 2206 differences (perf.data.ldw and
perf.data.unw files) in the callchains output out of the 249156
samples which is below 1%. Together with some other tips for speed
increase it's on my TODO list for future.

Git tree:
  https://git.kernel.org/cgit/linux/kernel/git/jolsa/perf.git/
  perf/core_libdw_unwind

Kudos to Jan Kratochvil for helping me with the new interface.

Jean,
I haven't tested on arm.. so not sure I broke anything there.
Any chance you could test for me, and maybe make the automated
test I added work there? ;-)

thanks,
jirka


Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jan Kratochvil <jan.kratochvil@redhat.com>
Cc: Jean Pihet <jean.pihet@linaro.org>
---
Jiri Olsa (16):
      perf tools: Automate setup of FEATURE_CHECK_(C|LD)FLAGS-all variables
      perf tools: Fix machine initialization
      perf tools: Make perf_event__synthesize_mmap_events global
      perf tests x86: Introduce perf_regs_load function
      perf tests x86: Add dwarf unwind test
      perf tools: Fix dwarf unwind max_stack processing
      perf tools: Do not report zero address in unwind
      perf tools: Add mask into struct regs_dump
      perf tools: Separate libunwind code to special object
      perf tools: Rename unwind__arch_reg_id into libunwind__arch_reg_id
      perf tools: Introduce HAVE_DWARF_UNWIND_SUPPORT macro
      perf tools: Separate perf_reg_value function in perf_regs object
      perf tools: Add feature check for libdw dwarf unwind
      perf tools: Add libdw DWARF post unwind support
      perf tools: Setup default dwarf post unwinder
      perf tests: Add NO_LIBDW_DWARF_UNWIND make test

 tools/perf/Makefile.perf                                   |  19 ++++++-
 tools/perf/arch/arm/Makefile                               |   2 +-
 tools/perf/arch/arm/util/{unwind.c => unwind-libunwind.c}  |   2 +-
 tools/perf/arch/x86/Makefile                               |   9 +++-
 tools/perf/arch/x86/include/perf_regs.h                    |   6 +++
 tools/perf/arch/x86/tests/dwarf-unwind.c                   |  59 ++++++++++++++++++++
 tools/perf/arch/x86/tests/regs_load.S                      |  92 ++++++++++++++++++++++++++++++++
 tools/perf/arch/x86/util/unwind-libdw.c                    |  51 ++++++++++++++++++
 tools/perf/arch/x86/util/{unwind.c => unwind-libunwind.c}  |   4 +-
 tools/perf/builtin-inject.c                                |   1 -
 tools/perf/builtin-record.c                                |  10 ++--
 tools/perf/config/Makefile                                 |  94 +++++++++++++++++++++++---------
 tools/perf/config/feature-checks/Makefile                  |   6 ++-
 tools/perf/config/feature-checks/test-all.c                |   5 ++
 tools/perf/config/feature-checks/test-libdw-dwarf-unwind.c |  13 +++++
 tools/perf/tests/builtin-test.c                            |   8 +++
 tools/perf/tests/dwarf-unwind.c                            | 144 +++++++++++++++++++++++++++++++++++++++++++++++++
 tools/perf/tests/make                                      |  63 +++++++++++-----------
 tools/perf/tests/sample-parsing.c                          |  17 +++---
 tools/perf/tests/tests.h                                   |   9 ++++
 tools/perf/util/event.c                                    |  12 ++---
 tools/perf/util/event.h                                    |  12 ++++-
 tools/perf/util/evsel.c                                    |  13 ++---
 tools/perf/util/machine.c                                  |   4 +-
 tools/perf/util/perf_regs.c                                |  19 +++++++
 tools/perf/util/perf_regs.h                                |  13 +++++
 tools/perf/util/session.c                                  |   5 +-
 tools/perf/util/unwind-libdw.c                             | 210 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 tools/perf/util/unwind-libdw.h                             |  21 ++++++++
 tools/perf/util/{unwind.c => unwind-libunwind.c}           |  50 ++++++-----------
 tools/perf/util/unwind.h                                   |  11 ++--
 31 files changed, 849 insertions(+), 135 deletions(-)
 rename tools/perf/arch/arm/util/{unwind.c => unwind-libunwind.c} (95%)
 create mode 100644 tools/perf/arch/x86/tests/dwarf-unwind.c
 create mode 100644 tools/perf/arch/x86/tests/regs_load.S
 create mode 100644 tools/perf/arch/x86/util/unwind-libdw.c
 rename tools/perf/arch/x86/util/{unwind.c => unwind-libunwind.c} (95%)
 create mode 100644 tools/perf/config/feature-checks/test-libdw-dwarf-unwind.c
 create mode 100644 tools/perf/tests/dwarf-unwind.c
 create mode 100644 tools/perf/util/perf_regs.c
 create mode 100644 tools/perf/util/unwind-libdw.c
 create mode 100644 tools/perf/util/unwind-libdw.h
 rename tools/perf/util/{unwind.c => unwind-libunwind.c} (92%)