[PATCH -tip v2 0/8] perf-probe: Updates for handling localfunctions correctly

[Masami Hiramatsu]

Hi,

Here is the 2nd version of the series for handling local
functions correctly with perf-probe.

In this version, I used "_stext" based probe point instead
of absolute address, because KASLR changes the address
offset randomly and the debuginfo doesn't know that offset.

Issue 1)
 Current perf-probe can't handle probe-points for kprobes,
 since it uses symbol-based probe definition. The symbol
 based definition is easy to read and robust for differnt
 kernel and modules. However, when user gives a local
 function name which has several different instances,
 it may put probes on wrong (or unexpected) address.
 On the other hand, since uprobe events are based on the
 actual address, it can avoid this issue.

 E.g.
In the case to probe t_show local functions (which has
4 different instances.
  ----
  # grep " t_show\$" /proc/kallsyms
  ffffffff810d9720 t t_show
  ffffffff810e2e40 t t_show
  ffffffff810ece30 t t_show
  ffffffff810f4ad0 t t_show
  # ./perf probe -fa "t_show \$vars"
  Added new events:
    probe:t_show         (on t_show with $vars)
    probe:t_show_1       (on t_show with $vars)
    probe:t_show_2       (on t_show with $vars)
    probe:t_show_3       (on t_show with $vars)

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

          perf record -e probe:t_show_3 -aR sleep 1
  ----
OK, we have 4 different t_show()s. All functions have
different arguments as below;
  ----
  # cat /sys/kernel/debug/tracing/kprobe_events
  p:probe/t_show t_show m=%di:u64 v=%si:u64
  p:probe/t_show_1 t_show m=%di:u64 v=%si:u64 t=%si:u64
  p:probe/t_show_2 t_show m=%di:u64 v=%si:u64 fmt=%si:u64
  p:probe/t_show_3 t_show m=%di:u64 v=%si:u64 file=%si:u64
  ----
However, all of them have been put on the *same* address.
  ----
  # cat /sys/kernel/debug/kprobes/list
  ffffffff810d9720  k  t_show+0x0    [DISABLED]
  ffffffff810d9720  k  t_show+0x0    [DISABLED]
  ffffffff810d9720  k  t_show+0x0    [DISABLED]
  ffffffff810d9720  k  t_show+0x0    [DISABLED]
  ----
 oops...

Issue 2)
 With the debuginfo, issue 1 can be solved by using
 _stext-based probe definition instead of local symbol-based.
 However, without debuginfo, perf-probe can only use
 symbol-map in the binary (or kallsyms). The map provides
 symbol find methods, but it returns only the first matched
 symbol. To put probes on all functions which have given
 symbol, we need a symbol-list iterator for the map.

 E.g. (built perf with NO_DWARF=1)
In the case to probe t_show and identity__map_ip in perf.
  ----
  # ./perf probe -a t_show
  Added new event:
    probe:t_show         (on t_show)

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

          perf record -e probe:t_show -aR sleep 1

  # ./perf probe -x perf -a identity__map_ip
  no symbols found in /kbuild/ksrc/linux-3/tools/perf/perf, maybe install a debug package?
  Failed to load map.
    Error: Failed to add events. (-22)
  ----
 oops.....


Solutions)
To solve the issue 1, this series changes perf probe to
use _stext-based probe definition. This means that we
also need to fix the --list options to analyze actual
probe address from _stext address. (and that has been
done in this series).

E.g. with this series;
  ----
  # ./perf probe -a "t_show \$vars"
  Added new events:
    probe:t_show         (on t_show with $vars)
    probe:t_show_1       (on t_show with $vars)
    probe:t_show_2       (on t_show with $vars)
    probe:t_show_3       (on t_show with $vars)

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

          perf record -e probe:t_show_3 -aR sleep 1

  # cat /sys/kernel/debug/tracing/kprobe_events
  p:probe/t_show _stext+889880 m=%di:u64 v=%si:u64
  p:probe/t_show_1 _stext+928568 m=%di:u64 v=%si:u64 t=%si:u64
  p:probe/t_show_2 _stext+969512 m=%di:u64 v=%si:u64 fmt=%si:u64
  p:probe/t_show_3 _stext+1001416 m=%di:u64 v=%si:u64 file=%si:u64

  # cat /sys/kernel/debug/kprobes/list
  ffffffffb50d95e0  k  t_show+0x0    [DISABLED]
  ffffffffb50e2d00  k  t_show+0x0    [DISABLED]
  ffffffffb50f4990  k  t_show+0x0    [DISABLED]
  ffffffffb50eccf0  k  t_show+0x0    [DISABLED]
  ----
This time we can see the events are set in different
addresses.

And for the issue 2, the last patch introduces symbol
iterators for map, dso and symbols (since the symbol
list is the symbols and it is included dso, and perf
probe accesses dso via map).

E.g. with this series (built perf with NO_DWARF=1);
  ----
  # ./perf probe -a t_show
  Added new events:
    probe:t_show         (on t_show)
    probe:t_show_1       (on t_show)
    probe:t_show_2       (on t_show)
    probe:t_show_3       (on t_show)

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

          perf record -e probe:t_show_3 -aR sleep 1

  # ./perf probe -x perf -a identity__map_ip
  Added new events:
    probe_perf:identity__map_ip (on identity__map_ip in /kbuild/ksrc/linux-3/tools/perf/perf)
    probe_perf:identity__map_ip_1 (on identity__map_ip in /kbuild/ksrc/linux-3/tools/perf/perf)
    probe_perf:identity__map_ip_2 (on identity__map_ip in /kbuild/ksrc/linux-3/tools/perf/perf)
    probe_perf:identity__map_ip_3 (on identity__map_ip in /kbuild/ksrc/linux-3/tools/perf/perf)

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

          perf record -e probe_perf:identity__map_ip_3 -aR sleep 1
  ----
Now, even without the debuginfo, both the kprobe and
uprobe are set 4 different places correctly.

BTW, while testing above, I've found some bugs and
another minor issue; perf-probe doesn't show the
modules and binaries in which probes are set.
I've also fixed it in this series as below.

Without the fix;

  # ./perf probe -m drm drm_av_sync_delay
  # ./perf probe -x perf dso__load_vmlinux

  # ./perf probe -l
    probe:drm_av_sync_delay (on drm_av_sync_delay)
    probe_perf:dso__load_vmlinux (on 0x000000000006d110)

With this fix;

  # ./perf probe -l
    probe:drm_av_sync_delay (on drm_av_sync_delay in drm)
    probe_perf:dso__load_vmlinux (on 0x000000000006d110 in /kbuild/ksrc/linux-3/tools/perf/perf)


TODO:
 - Support local functions in modules. This requires kernel
 side enhancement to allow setting probes by the relative
 addresses in modules too.
 - Uprobe-event MUST traces the change of given binary even
 when the event is disabled. I've found that user can replace
 the target binary after setting events and the events can be
 enabled on the different instructions...

---

Masami Hiramatsu (8):
      [BUGFIX] perf-probe: Fix to do exit call for symbol maps
      perf-probe: Remove incorrect symbol check for --list
      perf-probe: Show in what binaries/modules probes are set
      perf-probe: Use _stext based address instead of the symbol name
      perf-probe: Retry to find given address from offline dwarf
      perf-probe: Show appropriate symbol for _stext based kprobes
      perf-probe: Show source-level or symbol-level info for uprobes
      perf-probe: Allow to add events on the local functions


 tools/perf/util/dso.h         |   10 +
 tools/perf/util/map.h         |   10 +
 tools/perf/util/probe-event.c |  735 +++++++++++++++++++++++------------------
 tools/perf/util/symbol.h      |   11 +
 4 files changed, 440 insertions(+), 326 deletions(-)

-- 
Masami HIRAMATSU
IT Management Research Dept. Linux Technology Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: masami.hiramatsu.pt@xxxxxxxxxxx

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/