perf thread-stack: Hide x86 retpolines
authorAdrian Hunter <adrian.hunter@intel.com>
Wed, 9 Jan 2019 09:18:35 +0000 (11:18 +0200)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Fri, 22 Feb 2019 19:49:49 +0000 (16:49 -0300)
commit3c0cd952cf051903929cd57b89034cc5f71f451d
treee8ec94ca7393b2a1ba35b54e6795218317e5033a
parent1f35cd65386e02430546727bd3cc2508d052e3ec
perf thread-stack: Hide x86 retpolines

x86 retpoline functions pollute the call graph by showing up everywhere
there is an indirect branch, but they do not really mean anything. Make
changes so that the default retpoline functions will no longer appear in
the call graph. Note this only affects the call graph, since all the
original branches are left unchanged.

This does not handle function return thunks, nor is there any
improvement for the handling of inline thunks or extern thunks.

Example:

  $ cat simple-retpoline.c
  __attribute__((noinline)) int bar(void)
  {
          return -1;
  }

  int foo(void)
  {
          return bar() + 1;
  }

  __attribute__((indirect_branch("thunk"))) int main()
  {
          int (*volatile fn)(void) = foo;

          fn();
          return fn();
  }
  $ gcc -ggdb3 -Wall -Wextra -O2 -o simple-retpoline simple-retpoline.c
  $ objdump -d simple-retpoline
  <SNIP>
  0000000000001040 <main>:
      1040:       48 83 ec 18             sub    $0x18,%rsp
      1044:       48 8d 05 25 01 00 00    lea    0x125(%rip),%rax        # 1170 <foo>
      104b:       48 89 44 24 08          mov    %rax,0x8(%rsp)
      1050:       48 8b 44 24 08          mov    0x8(%rsp),%rax
      1055:       e8 1f 01 00 00          callq  1179 <__x86_indirect_thunk_rax>
      105a:       48 8b 44 24 08          mov    0x8(%rsp),%rax
      105f:       48 83 c4 18             add    $0x18,%rsp
      1063:       e9 11 01 00 00          jmpq   1179 <__x86_indirect_thunk_rax>
  <SNIP>
  0000000000001160 <bar>:
      1160:       b8 ff ff ff ff          mov    $0xffffffff,%eax
      1165:       c3                      retq
  <SNIP>
  0000000000001170 <foo>:
      1170:       e8 eb ff ff ff          callq  1160 <bar>
      1175:       83 c0 01                add    $0x1,%eax
      1178:       c3                      retq
  0000000000001179 <__x86_indirect_thunk_rax>:
      1179:       e8 07 00 00 00          callq  1185 <__x86_indirect_thunk_rax+0xc>
      117e:       f3 90                   pause
      1180:       0f ae e8                lfence
      1183:       eb f9                   jmp    117e <__x86_indirect_thunk_rax+0x5>
      1185:       48 89 04 24             mov    %rax,(%rsp)
      1189:       c3                      retq
  <SNIP>
  $ perf record -o simple-retpoline.perf.data -e intel_pt/cyc/u ./simple-retpoline
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0,017 MB simple-retpoline.perf.data ]
  $ perf script -i simple-retpoline.perf.data --itrace=be -s ~/libexec/perf-core/scripts/python/export-to-sqlite.py simple-retpoline.db branches calls
  2019-01-08 14:03:37.851655 Creating database...
  2019-01-08 14:03:37.863256 Writing records...
  2019-01-08 14:03:38.069750 Adding indexes
  2019-01-08 14:03:38.078799 Done
  $ ~/libexec/perf-core/scripts/python/exported-sql-viewer.py simple-retpoline.db

Before:

    main
        -> __x86_indirect_thunk_rax
            -> __x86_indirect_thunk_rax
                -> foo
                    -> bar

After:

    main
        -> foo
            -> bar

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/20190109091835.5570-7-adrian.hunter@intel.com
[ Remove (sym->name != NULL) test, this is not a pointer and breaks the build with clang version 7.0.1 (Fedora 7.0.1-2.fc30) ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/util/thread-stack.c