Added a suppression for another benign race in ICU, widen one old suppression.

(see third_party/icu/source/common/umutex.c) 
When two concurrent threads try to call umtx_lock(mutex) for an uninitialized mutex simultaneously, *mutex is read without taking any lock.
However, if *mutex is NULL, a DCLP inside umtx_init(mutex) guarantees that *mutex is initialized exactly once. Otherwise, if *mutex is already a valid pthread_mutex_t, the behavior of the program is not affected.

This patch was prepared by Alexander Potapenko (cc'ed)
TBR=dank
Review URL: http://codereview.chromium.org/303019

git-svn-id: svn://svn.chromium.org/chrome/trunk/src@29633 0039d316-1c4b-4281-b951-d872f2087c98

1 files changed, 8 insertions, 0 deletions

diff --git a/tools/valgrind/tsan/suppressions.txt b/tools/valgrind/tsan/suppressions.txt
index 14b46ae..4557452 100644
--- a/tools/valgrind/tsan/suppressions.txt
+++ b/tools/valgrind/tsan/suppressions.txt
@@ -167,6 +167,7 @@
 {
   Two writes, same value (ICU gGlobalMutex, gMutexesInUse)
   ThreadSanitizer:Race
+  ...
   fun:umtx_init_4_2
 }
 
@@ -176,6 +177,13 @@
   fun:uprv_malloc_4_2
 }
 
+# Reading a pointer to a mutex being initialized in a concurrent thread.
+{
+  A benign race in umtx_lock_4_2
+  ThreadSanitizer:Race
+  fun:umtx_lock_4_2
+}
+
 ############################
 # Benign race in SQLLite
 {