diff --git a/configure b/configure
index 2b46ab8367..3d1519f075 100755
--- a/configure
+++ b/configure
@@ -343,6 +343,7 @@ CYGWIN*)
   # Flags below are still necessary mostly for MinGW.
   socklen_t="yes"
   sfaa="yes"
+  svcas="yes"
   rusage_thread="yes"
   fdatasync="yes"
   clock_gettime="yes" # clock_monotonic probe has dependency on this
@@ -706,6 +707,30 @@ if compile_prog "" "" "__sync_fetch_and_add()" ; then
 fi
 print_config "__sync_fetch_and_add" "$sfaa"
 
+##########################################
+# __sync_val_compare_and_swap test
+if test "$svcas" != "yes" ; then
+  svcas="no"
+fi
+cat > $TMPC << EOF
+#include <inttypes.h>
+static int svcas(uint32_t *ptr)
+{
+  return __sync_val_compare_and_swap(ptr, 0, 0);
+}
+
+int main(int argc, char **argv)
+{
+  uint32_t val = 42;
+  svcas(&val);
+  return val;
+}
+EOF
+if compile_prog "" "" "__sync_val_compare_and_swap()" ; then
+    svcas="yes"
+fi
+print_config "__sync_val_compare_and_swap" "$svcas"
+
 ##########################################
 # libverbs probe
 if test "$libverbs" != "yes" ; then
@@ -2108,6 +2133,9 @@ fi
 if test "$sfaa" = "yes" ; then
   output_sym "CONFIG_SFAA"
 fi
+if test "$svcas" = "yes" ; then
+  output_sym "CONFIG_SVCAS"
+fi
 if test "$libverbs" = "yes" -a "$rdmacm" = "yes" ; then
   output_sym "CONFIG_RDMA"
 fi
diff --git a/gettime.c b/gettime.c
index 390d2e70b7..49d55718a7 100644
--- a/gettime.c
+++ b/gettime.c
@@ -547,8 +547,8 @@ uint64_t time_since_now(const struct timespec *s)
 	return mtime_since_now(s) / 1000;
 }
 
-#if defined(FIO_HAVE_CPU_AFFINITY) && defined(ARCH_HAVE_CPU_CLOCK)  && \
-    defined(CONFIG_SFAA)
+#if defined(FIO_HAVE_CPU_AFFINITY) && defined(ARCH_HAVE_CPU_CLOCK) && \
+    defined(CONFIG_SFAA) && defined(CONFIG_SVCAS)
 
 #define CLOCK_ENTRIES_DEBUG	100000
 #define CLOCK_ENTRIES_TEST	1000
@@ -570,11 +570,19 @@ struct clock_thread {
 	struct clock_entry *entries;
 };
 
+/* Note: the following acts as a full barrier */
 static inline uint32_t atomic32_inc_return(uint32_t *seq)
 {
 	return 1 + __sync_fetch_and_add(seq, 1);
 }
 
+/* Note: the following acts as a full barrier */
+static inline uint32_t atomic32_cas_return(uint32_t *ptr, uint32_t old,
+		uint32_t new)
+{
+	return __sync_val_compare_and_swap(ptr, old, new);
+}
+
 static void *clock_thread_fn(void *data)
 {
 	struct clock_thread *t = data;
@@ -607,16 +615,18 @@ static void *clock_thread_fn(void *data)
 	last_seq = 0;
 	c = &t->entries[0];
 	for (i = 0; i < t->nr_entries; i++, c++) {
-		uint32_t seq;
+		uint32_t seq, end_seq;
 		uint64_t tsc;
 
 		c->cpu = t->cpu;
 		do {
 			seq = atomic32_inc_return(t->seq);
+			tsc = get_cpu_clock();
+			end_seq = atomic32_cas_return(t->seq, seq, seq);
 			if (seq < last_seq)
 				break;
-			tsc = get_cpu_clock();
-		} while (seq != *t->seq);
+			last_seq = end_seq;
+		} while (seq != end_seq);
 
 		c->seq = seq;
 		c->tsc = tsc;
@@ -778,7 +788,8 @@ int fio_monotonic_clocktest(int debug)
 	return !!failed;
 }
 
-#else /* defined(FIO_HAVE_CPU_AFFINITY) && defined(ARCH_HAVE_CPU_CLOCK) */
+#else /* defined(FIO_HAVE_CPU_AFFINITY) && defined(ARCH_HAVE_CPU_CLOCK) &&
+       * defined(CONFIG_SFAA) && defined(CONFIG_SVCAS)*/
 
 int fio_monotonic_clocktest(int debug)
 {
@@ -787,4 +798,5 @@ int fio_monotonic_clocktest(int debug)
 	return 1;
 }
 
-#endif
+#endif /* defined(FIO_HAVE_CPU_AFFINITY) && defined(ARCH_HAVE_CPU_CLOCK) &&
+        * defined(CONFIG_SFAA) && defined(CONFIG_SVCAS)*/