Use less TSan ignores for base/atomic*...

tools/valgrind/tsan/ignores.txt

 # This file lists the functions, object files and source files
 # which should be ignored (i.e. not instrumented) by ThreadSanitizer.
 # See http://code.google.com/p/data-race-test/wiki/ThreadSanitizerIgnores.
 
 # ignore these libraries
 obj:*/ld-2*
 obj:*/libpthread-*
 obj:*/libfreetype*
 
 # we ignore the whole NSS library for now since
@@ skipping 35 matching lines @@
 fun:*_M_is_shared*
 fun:*_M_set_leaked*
 fun:*_M_set_sharable*
 
 # Don't instrument intercepts
 src:*ts_valgrind_intercepts.c
 
 ##################################################################
 # Don't instrument synchronization code
 src:*base/waitable_event*
-src:*base/atomic*
 src:*base/condition_variable*
 src:*base/lock*
 src:*base/stats_counters*
 src:*base/thread_local_storage*
 
+# Don't instrument code dealing with atomics (base::subtle)
+fun:*base*subtle*Release_Store*
+fun:*base*subtle*NoBarrier_CompareAndSwap*
+
 # Don't instrument tcmalloc
 src:*/tcmalloc/*
 
 # This function is heavy in net_unittests
 fun_r:*disk_cache*BackendImpl*CheckAllEntries*
 
 # V8 is a hot-spot under ThreadSanitizer.
 # Lots of tiny functions there...
 # TODO(timurrrr):
 # Can we miss data races on V8 objects due to non thread-safe API calls
@@ skipping 25 matching lines @@
 # There's some weird failure test going on in this tiny test function in sqlite
 fun_r:threadLockingTest
 
 # Ignore accesses below GetCurrentThreadIdentifier.
 # There is a benign race which is hard to suppress properly,
 # see http://crbug.com/44580
 fun_r:*BrowserThread*GetCurrentThreadIdentifier*
 
 # zlib is smarter than we are, see http://www.zlib.net/zlib_faq.html#faq36
 fun_r:inflate