diff --git a/src/mutex.c b/src/mutex.c
index 1f8d41a..48b4589 100644
--- a/src/mutex.c
+++ b/src/mutex.c
@@ -14,7 +14,8 @@
 
 
 // Await the value of an `atomic_int`.
-static bool await_atomic_int(atomic_int *var, timestamp_us_t timeout, int expected, int new_value, memory_order order) {
+static bool
+    await_swap_atomic_int(atomic_int *var, timestamp_us_t timeout, int expected, int new_value, memory_order order) {
     do {
         int old_value = expected;
         if (atomic_compare_exchange_weak_explicit(var, &old_value, new_value, order, memory_order_relaxed)) {
@@ -27,10 +28,10 @@ static bool await_atomic_int(atomic_int *var, timestamp_us_t timeout, int expect
 }
 
 // Atomically check the value exceeds a threshold and subtract.
-static bool thresh_subtrach_atomic_int(atomic_int *var, int threshold, int subtract, memory_order order) {
+static bool thresh_sub_atomic_int(atomic_int *var, int threshold, int sub, memory_order order) {
     while (1) {
         int old_value = atomic_load(var);
-        int new_value = old_value - subtract;
+        int new_value = old_value - sub;
         if (old_value < threshold) {
             return false;
         } else if (atomic_compare_exchange_weak_explicit(var, &old_value, new_value, order, memory_order_relaxed)) {
@@ -41,6 +42,36 @@ static bool thresh_subtrach_atomic_int(atomic_int *var, int threshold, int subtr
     }
 }
 
+// Atomically check the value does not exceed a threshold and add.
+static bool thresh_add_atomic_int(atomic_int *var, int threshold, int add, memory_order order) {
+    while (1) {
+        int old_value = atomic_load(var);
+        int new_value = old_value + add;
+        if (old_value >= threshold || new_value >= threshold) {
+            return false;
+        } else if (atomic_compare_exchange_weak_explicit(var, &old_value, new_value, order, memory_order_relaxed)) {
+            return true;
+        } else {
+            sched_yield();
+        }
+    }
+}
+
+// Atomically check the value doesn't equal either illegal values and subtract.
+static bool unequal_sub_atomic_int(atomic_int *var, int unequal0, int unequal1, int sub, memory_order order) {
+    while (1) {
+        int old_value = atomic_load(var);
+        int new_value = old_value - sub;
+        if (old_value == unequal0 || old_value == unequal1) {
+            return false;
+        } else if (atomic_compare_exchange_weak_explicit(var, &old_value, new_value, order, memory_order_relaxed)) {
+            return true;
+        } else {
+            sched_yield();
+        }
+    }
+}
+
 
 
 // Initialise a mutex for unshared use.
@@ -88,7 +119,7 @@ bool mutex_acquire(badge_err_t *ec, mutex_t *mutex, timestamp_us_t timeout) {
     }
     timeout += time_us();
     // Await the shared portion to reach 0 and then lock.
-    if (await_atomic_int(&mutex->shares, timeout, 0, EXCLUSIVE_MAGIC, memory_order_acquire)) {
+    if (await_swap_atomic_int(&mutex->shares, timeout, 0, EXCLUSIVE_MAGIC, memory_order_acquire)) {
         // If that succeeds, the mutex was acquired.
         badge_err_set_ok(ec);
         return true;
@@ -106,10 +137,12 @@ bool mutex_release(badge_err_t *ec, mutex_t *mutex) {
         badge_err_set(ec, ELOC_UNKNOWN, ECAUSE_ILLEGAL);
         return false;
     }
-    if (thresh_subtrach_atomic_int(&mutex->shares, EXCLUSIVE_MAGIC, EXCLUSIVE_MAGIC, memory_order_release)) {
+    if (thresh_sub_atomic_int(&mutex->shares, EXCLUSIVE_MAGIC, EXCLUSIVE_MAGIC, memory_order_release)) {
+        // Successful release.
         badge_err_set_ok(ec);
         return true;
     } else {
+        // Mutex was not taken exclusively.
         badge_err_set(ec, ELOC_UNKNOWN, ECAUSE_ILLEGAL);
         return false;
     }
@@ -128,16 +161,13 @@ bool mutex_acquire_shared(badge_err_t *ec, mutex_t *mutex, timestamp_us_t timeou
     }
     timeout += time_us();
     // Take a share.
-    int val = atomic_fetch_add_explicit(&mutex->shares, 1, memory_order_acquire);
-    // Await the lock to be released.
-    if (val < EXCLUSIVE_MAGIC) {
+    if (thresh_add_atomic_int(&mutex->shares, EXCLUSIVE_MAGIC, 1, memory_order_acquire)) {
         // If that succeeds, the mutex was successfully acquired.
         badge_err_set_ok(ec);
         return true;
     } else {
         // If that fails, abort trying to lock.
         badge_err_set(ec, ELOC_UNKNOWN, ECAUSE_TIMEOUT);
-        atomic_fetch_sub_explicit(&mutex->shares, 1, memory_order_relaxed);
         return false;
     }
 }
@@ -150,13 +180,13 @@ bool mutex_release_shared(badge_err_t *ec, mutex_t *mutex) {
         return false;
     }
     int old = atomic_fetch_sub_explicit(&mutex->shares, 1, memory_order_release);
-    if (old == 0 || old == EXCLUSIVE_MAGIC) {
+    if (unequal_sub_atomic_int(&mutex->shares, 0, EXCLUSIVE_MAGIC, 1, memory_order_release)) {
         // Prevent the counter from underflowing.
         badge_err_set(ec, ELOC_UNKNOWN, ECAUSE_ILLEGAL);
-        atomic_fetch_add_explicit(&mutex->shares, 1, memory_order_relaxed);
         return false;
     } else {
         // Successful release.
+        badge_err_set_ok(ec);
         return true;
     }
 }