diff --git a/include/jemalloc/internal/mutex_inlines.h b/include/jemalloc/internal/mutex_inlines.h
index d4703d23..7adcff4e 100644
--- a/include/jemalloc/internal/mutex_inlines.h
+++ b/include/jemalloc/internal/mutex_inlines.h
@@ -52,11 +52,13 @@ malloc_mutex_lock(tsdn_t *tsdn, malloc_mutex_t *mutex) {
 			malloc_mutex_lock_slow(mutex);
 		}
 		/* We own the lock now.  Update a few counters. */
-		mutex_prof_data_t *data = &mutex->prof_data;
-		data->n_lock_ops++;
-		if (data->prev_owner != tsdn) {
-			data->prev_owner = tsdn;
-			data->n_owner_switches++;
+		if (config_stats) {
+			mutex_prof_data_t *data = &mutex->prof_data;
+			data->n_lock_ops++;
+			if (data->prev_owner != tsdn) {
+				data->prev_owner = tsdn;
+				data->n_owner_switches++;
+			}
 		}
 	}
 	witness_lock(tsdn, &mutex->witness);
diff --git a/include/jemalloc/internal/mutex_types.h b/include/jemalloc/internal/mutex_types.h
index 257f69ca..3cc7bc2b 100644
--- a/include/jemalloc/internal/mutex_types.h
+++ b/include/jemalloc/internal/mutex_types.h
@@ -4,6 +4,12 @@
 typedef struct mutex_prof_data_s mutex_prof_data_t;
 typedef struct malloc_mutex_s malloc_mutex_t;
 
+/*
+ * Based on benchmark results, a fixed spin with this amount of retries works
+ * well for our critical sections.
+ */
+#define MALLOC_MUTEX_MAX_SPIN 250
+
 #ifdef _WIN32
 #  if _WIN32_WINNT >= 0x0600
 #    define MALLOC_MUTEX_LOCK(m)    AcquireSRWLockExclusive(&(m)->lock)
@@ -45,20 +51,10 @@ typedef struct malloc_mutex_s malloc_mutex_t;
      {{{LOCK_PROF_DATA_INITIALIZER, PTHREAD_MUTEX_INITIALIZER, NULL}},	\
       WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)}
 #else
-/* TODO: get rid of adaptive mutex once we do our own spin. */
-#  if (defined(JEMALLOC_HAVE_PTHREAD_MUTEX_ADAPTIVE_NP) &&		\
-       defined(PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP))
-#    define MALLOC_MUTEX_TYPE PTHREAD_MUTEX_ADAPTIVE_NP
-#    define MALLOC_MUTEX_INITIALIZER					\
-       {{{LOCK_PROF_DATA_INITIALIZER,					\
-          PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP}},			\
-        WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)}
-#  else
 #    define MALLOC_MUTEX_TYPE PTHREAD_MUTEX_DEFAULT
 #    define MALLOC_MUTEX_INITIALIZER					\
        {{{LOCK_PROF_DATA_INITIALIZER, PTHREAD_MUTEX_INITIALIZER}},	\
         WITNESS_INITIALIZER("mutex", WITNESS_RANK_OMIT)}
-#  endif
 #endif
 
 #endif /* JEMALLOC_INTERNAL_MUTEX_TYPES_H */
diff --git a/src/mutex.c b/src/mutex.c
index 82a5fa3e..06ccd42c 100644
--- a/src/mutex.c
+++ b/src/mutex.c
@@ -69,14 +69,26 @@ void
 malloc_mutex_lock_slow(malloc_mutex_t *mutex) {
 	mutex_prof_data_t *data = &mutex->prof_data;
 
-	{//TODO: a smart spin policy
+	if (ncpus == 1) {
+		goto label_spin_done;
+	}
+
+	int cnt = 0, max_cnt = MALLOC_MUTEX_MAX_SPIN;
+	do {
+		CPU_SPINWAIT;
 		if (!malloc_mutex_trylock(mutex)) {
 			data->n_spin_acquired++;
 			return;
 		}
-	}
+	} while (cnt++ < max_cnt);
 
+	if (!config_stats) {
+		/* Only spin is useful when stats is off. */
+		malloc_mutex_lock_final(mutex);
+		return;
+	}
 	nstime_t now, before;
+label_spin_done:
 	nstime_init(&now, 0);
 	nstime_update(&now);
 	nstime_copy(&before, &now);