1. 26 Dec, 2013 2 commits
    • Masami Hiramatsu's avatar
      perf probe: Support basic dwarf-based operations on uprobe events · fb7345bb
      Masami Hiramatsu authored
      Support basic dwarf(debuginfo) based operations for uprobe events.  With
      this change, perf probe can analyze debuginfo of user application binary
      to set up new uprobe event.
      
      This allows perf-probe --add(with local variables, line numbers) and
      --line works with -x option.  (Actually, --vars has already accepted -x
      option)
      
      For example, the following command shows the probe-able lines of a given
      user space function. Something that so far was only available in the
      'perf probe' tool for kernel space functions:
      
        # ./perf probe -x perf --line map__load
        <map__load@/home/fedora/ksrc/linux-2.6/tools/perf/util/map.c:0>
              0  int map__load(struct map *map, symbol_filter_t filter)
              1  {
              2         const char *name = map->dso->long_name;
                        int nr;
      
              5         if (dso__loaded(map->dso, map->type))
              6                 return 0;
      
              8         nr = dso__load(map->dso, map, filter);
              9         if (nr < 0) {
             10                 if (map->dso->has_build_id) {
      
      And this shows the available variables at the given line of the
      function.
      
        # ./perf probe -x perf --vars map__load:8
        Available variables at map__load:8
                @<map__load+96>
                        char*   name
                        struct map*     map
                        symbol_filter_t filter
                @<map__find_symbol+112>
                        char*   name
                        symbol_filter_t filter
                @<map__find_symbol_by_name+136>
                        char*   name
                        symbol_filter_t filter
                @<map_groups__find_symbol_by_name+176>
                        char*   name
                        struct map*     map
                        symbol_filter_t filter
      
      And lastly, we can now define probe(s) with all available
      variables on the given line:
      
        # ./perf probe -x perf --add 'map__load:8 $vars'
      
        Added new events:
          probe_perf:map__load (on map__load:8 with $vars)
          probe_perf:map__load_1 (on map__load:8 with $vars)
          probe_perf:map__load_2 (on map__load:8 with $vars)
          probe_perf:map__load_3 (on map__load:8 with $vars)
      
        You can now use it in all perf tools, such as:
      
                perf record -e probe_perf:map__load_3 -aR sleep 1
      
        Changes from previous version:
         - Add examples in the patch description.
         - Use .text section start address and dwarf symbol address
           for calculating the offset of given symbol, instead of
           searching the symbol in symtab again.
           With this change, we can safely handle multiple local
           function instances (e.g. scnprintf in perf).
      Signed-off-by: default avatarMasami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
      Cc: David Ahern <dsahern@gmail.com>
      Cc: David A. Long <dave.long@linaro.org>
      Cc: Ingo Molnar <mingo@kernel.org>
      Cc: Namhyung Kim <namhyung@kernel.org>
      Cc: Oleg Nesterov <oleg@redhat.com>
      Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
      Cc: Steven Rostedt <rostedt@goodmis.org>
      Cc: systemtap@sourceware.org
      Cc: yrl.pp-manager.tt@hitachi.com
      Link: http://lkml.kernel.org/r/20131226054152.22364.47021.stgit@kbuild-fedora.novalocalSigned-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      fb7345bb
    • Masami Hiramatsu's avatar
      perf probe: Expand given path to absolute path · 8a613d40
      Masami Hiramatsu authored
      Expand given path to absolute path in the option parser, except for a
      module name.
      
      Since realpath at later stage in processing several probe point, can be
      called several times (even if currently doesn't, it can happen when we
      expands the feature), it is waste of the performance.
      
      Processing it once at the early stage can avoid that.
      
      Changes from previous one:
       - Fix not to print null string.
       - Allocate memory for given path/module name everytime.
      Signed-off-by: default avatarMasami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
      Cc: "David A. Long" <dave.long@linaro.org>
      Cc: "Steven Rostedt (Red Hat)" <rostedt@goodmis.org>
      Cc: David Ahern <dsahern@gmail.com>
      Cc: Ingo Molnar <mingo@kernel.org>
      Cc: Namhyung Kim <namhyung@kernel.org>
      Cc: Oleg Nesterov <oleg@redhat.com>
      Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
      Cc: systemtap@sourceware.org
      Cc: yrl.pp-manager.tt@hitachi.com
      Link: http://lkml.kernel.org/r/20131226054150.22364.12187.stgit@kbuild-fedora.novalocal
      [ Clarified the pr_warning message as per David Ahern's suggestion ]
      Signed-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      8a613d40
  2. 23 Dec, 2013 7 commits
  3. 20 Dec, 2013 2 commits
  4. 19 Dec, 2013 28 commits
  5. 18 Dec, 2013 1 commit