Make sure that DRD does not complain about mutexes being held too long on systems where the clock can go backward. Apparently this happens frequently when Linux is running inside a virtual machine.

git-svn-id: svn://svn.valgrind.org/valgrind/trunk@9521 a5019735-40e9-0310-863c-91ae7b9d1cf9
diff --git a/drd/drd_rwlock.c b/drd/drd_rwlock.c
index a2cfdaf..82f28ac 100644
--- a/drd/drd_rwlock.c
+++ b/drd/drd_rwlock.c
@@ -521,7 +521,7 @@
       q->reader_nesting_count--;
       if (q->reader_nesting_count == 0 && DRD_(s_shared_threshold_ms) > 0)
       {
-         ULong held = VG_(read_millisecond_timer)() - p->acquiry_time_ms;
+         Long held = VG_(read_millisecond_timer)() - p->acquiry_time_ms;
          if (held > DRD_(s_shared_threshold_ms))
          {
             HoldtimeErrInfo HEI
@@ -539,7 +539,7 @@
       q->writer_nesting_count--;
       if (q->writer_nesting_count == 0 && DRD_(s_exclusive_threshold_ms) > 0)
       {
-         ULong held = VG_(read_millisecond_timer)() - p->acquiry_time_ms;
+         Long held = VG_(read_millisecond_timer)() - p->acquiry_time_ms;
          if (held > DRD_(s_exclusive_threshold_ms))
          {
             HoldtimeErrInfo HEI