3 files changed, 15 insertions, 9 deletions
diff --git a/include/spinlock/dec.h b/include/spinlock/dec.h
index 11d36dd..3e36bf7 100644
--- a/include/spinlock/dec.h
+++ b/include/spinlock/dec.h
@@ -111,7 +111,8 @@ ck_spinlock_dec_lock_eb(struct ck_spinlock_dec *lock)
 		if (r == true)
 			break;
 
-		ck_backoff_eb(&backoff);
+		while (ck_pr_load_uint(&lock->value) != 1)
+			ck_backoff_eb(&backoff);
 	}
 
 	ck_pr_fence_lock();
diff --git a/include/spinlock/fas.h b/include/spinlock/fas.h
index 4e6c123..bfe91fe 100644
--- a/include/spinlock/fas.h
+++ b/include/spinlock/fas.h
@@ -77,10 +77,11 @@ CK_CC_INLINE static void
 ck_spinlock_fas_lock(struct ck_spinlock_fas *lock)
 {
 
-	while (ck_pr_fas_uint(&lock->value, true) == true) {
-		while (ck_pr_load_uint(&lock->value) == true)
-			ck_pr_stall();
-	}
+        while (CK_CC_UNLIKELY(ck_pr_fas_uint(&lock->value, true) == true)) {
+                do {
+                        ck_pr_stall();
+                } while (ck_pr_load_uint(&lock->value) == true);
+        }
 
 	ck_pr_fence_lock();
 	return;
diff --git a/include/spinlock/hclh.h b/include/spinlock/hclh.h
index 296448b..ece56c6 100644
--- a/include/spinlock/hclh.h
+++ b/include/spinlock/hclh.h
@@ -81,6 +81,8 @@ ck_spinlock_hclh_lock(struct ck_spinlock_hclh **glob_queue,
 	thread->wait = true;
 	thread->splice = false;
 	thread->cluster_id = (*local_queue)->cluster_id;
+	/* Make sure previous->previous doesn't appear to be NULL */
+	thread->previous = *local_queue;
 
 	/* Serialize with respect to update of local queue. */
 	ck_pr_fence_store_atomic();
@@ -91,13 +93,15 @@ ck_spinlock_hclh_lock(struct ck_spinlock_hclh **glob_queue,
 
 	/* Wait until previous thread from the local queue is done with lock. */
 	ck_pr_fence_load();
-	if (previous->previous != NULL &&
-	    previous->cluster_id == thread->cluster_id) {
-		while (ck_pr_load_uint(&previous->wait) == true)
+	if (previous->previous != NULL) {
+		while (ck_pr_load_uint(&previous->wait) == true &&
+			ck_pr_load_int(&previous->cluster_id) == thread->cluster_id &&
+			ck_pr_load_uint(&previous->splice) == false)
 			ck_pr_stall();
 
 		/* We're head of the global queue, we're done */
-		if (ck_pr_load_uint(&previous->splice) == false)
+		if (ck_pr_load_int(&previous->cluster_id) == thread->cluster_id &&
+				ck_pr_load_uint(&previous->splice) == false)
 			return;
 	}