[PATCH v2] perf test: Fix test case Merge cpu map

* [PATCH v2] perf test: Fix test case Merge cpu map
@ 2020-01-20 13:20 Thomas Richter
  2020-01-20 13:20 ` [PATCH v2] perf probe: Add ustring support for perf probe command Thomas Richter
                   ` (2 more replies)
  0 siblings, 3 replies; 6+ messages in thread
From: Thomas Richter @ 2020-01-20 13:20 UTC (permalink / raw)
  To: linux-kernel, linux-perf-users, acme
  Cc: gor, sumanthk, heiko.carstens, Thomas Richter

V2: Added Reviewed-by statement

Commit a2408a70368a ("perf evlist: Maintain evlist->all_cpus")
introduces a test case for cpumap merge operation, see functions
perf_cpu_map__merge() and test__cpu_map_merge().

The test case fails on s390 with this error message:
 [root@m35lp76 perf]# ./perf test -Fvvvvv 52
 52: Merge cpu map                                         :
 --- start ---
 cpumask list: 1-2,4-5,7
 perf: /root/linux/tools/include/linux/refcount.h:131:\
          refcount_sub_and_test: Assertion `!(new > val)' failed.
 Aborted (core dumped)
 [root@m35lp76 perf]#

The root cause is in the function test__cpu_map_merge():
It creates two cpu_maps named 'a' and 'b':

  struct perf_cpu_map *a = perf_cpu_map__new("4,2,1");
  struct perf_cpu_map *b = perf_cpu_map__new("4,5,7");

and creates a third map named 'c' which is the result of
the merge of maps a and b:

  struct perf_cpu_map *c = perf_cpu_map__merge(a, b);

After some verifaction of the merged cpu_map all three
of them are have their reference count reduced and are
freed:

   perf_cpu_map__put(a); (1)
   perf_cpu_map__put(b);
   perf_cpu_map__put(c);

The release of perf_cpu_map__put(a) is wrong. The map
is already released and free'ed as part of the function

  perf_cpu_map__merge(struct perf_cpu_map *orig,
  |	              struct perf_cpu_map *other)
  +--> perf_cpu_map__put(orig);
       |
       +--> cpu_map__delete(orig)

At the end perf_cpu_map_put() is called for map 'orig'
alias 'a' and since the reference count is 1, the map
is deleted, as can be seen by the following gdb trace:

 (gdb) where
 #0  tcache_put (tc_idx=0, chunk=0x156cc30) at malloc.c:2940
 #1  _int_free (av=0x3fffd49ee80 <main_arena>, p=0x156cc30,
		     have_lock=<optimized out>) at malloc.c:4222
 #2  0x00000000012d5e78 in cpu_map__delete (map=0x156cc40) at cpumap.c:31
 #3  0x00000000012d5f7a in perf_cpu_map__put (map=0x156cc40) at cpumap.c:45
 #4  0x00000000012d723a in perf_cpu_map__merge (orig=0x156cc40,
     other=0x156cc60) at cpumap.c:343
 #5  0x000000000110cdd0 in test__cpu_map_merge (
     test=0x14ea6c8 <generic_tests+2856>, subtest=-1) at tests/cpumap.c:128

Thus the perf_cpu_map__put(a) (see (1) above) frees map 'a'
a second time and causes the failure. Fix this be removing that
function call.

Output after:
  [root@m35lp76 perf]# ./perf test -Fvvvvv 52
  52: Merge cpu map                                         :
  --- start ---
  cpumask list: 1-2,4-5,7
  ---- end ----
  Merge cpu map: Ok
  [root@m35lp76 perf]#

Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: Andi Kleen <ak@linux.intel.com>
---
 tools/perf/tests/cpumap.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/tools/perf/tests/cpumap.c b/tools/perf/tests/cpumap.c
index 4ac56741ac5f..29c793ac7d10 100644
--- a/tools/perf/tests/cpumap.c
+++ b/tools/perf/tests/cpumap.c
@@ -131,7 +131,6 @@ int test__cpu_map_merge(struct test *test __maybe_unused, int subtest __maybe_un
 	TEST_ASSERT_VAL("failed to merge map: bad nr", c->nr == 5);
 	cpu_map__snprint(c, buf, sizeof(buf));
 	TEST_ASSERT_VAL("failed to merge map: bad result", !strcmp(buf, "1-2,4-5,7"));
-	perf_cpu_map__put(a);
 	perf_cpu_map__put(b);
 	perf_cpu_map__put(c);
 	return 0;
-- 
2.21.0


^ permalink raw reply related	[flat|nested] 6+ messages in thread