diff --git a/arch/x86_64/kernel/syscall.c b/arch/x86_64/kernel/syscall.c
index 0d60b120..012f88a3 100644
--- a/arch/x86_64/kernel/syscall.c
+++ b/arch/x86_64/kernel/syscall.c
@@ -93,21 +93,23 @@ extern int num_processors;
 int obtain_clone_cpuid(cpu_set_t *cpu_set) {
 	int min_queue_len = -1;
 	int cpu, min_cpu = -1;
+	unsigned long irqstate;
 
+	irqstate = ihk_mc_spinlock_lock(&runq_reservation_lock);
 	/* Find the first allowed core with the shortest run queue */
 	for (cpu = 0; cpu < num_processors; ++cpu) {
 		struct cpu_local_var *v;
-		unsigned long irqstate;
 
 		if (!CPU_ISSET(cpu, cpu_set)) continue;
 
 		v = get_cpu_local_var(cpu);
-		irqstate = ihk_mc_spinlock_lock(&v->runq_lock);
-		if (min_queue_len == -1 || v->runq_len < min_queue_len) {
-			min_queue_len = v->runq_len;
+		ihk_mc_spinlock_lock_noirq(&v->runq_lock);
+		dkprintf("%s: cpu=%d,runq_len=%d,runq_reserved=%d\n", __FUNCTION__, cpu, v->runq_len, v->runq_reserved);
+		if (min_queue_len == -1 || v->runq_len + v->runq_reserved < min_queue_len) {
+			min_queue_len = v->runq_len + v->runq_reserved;
 			min_cpu = cpu;
 		}
-		ihk_mc_spinlock_unlock(&v->runq_lock, irqstate);
+		ihk_mc_spinlock_unlock_noirq(&v->runq_lock);
 
 		if (min_queue_len == 0)
 			break;
@@ -116,7 +118,9 @@ int obtain_clone_cpuid(cpu_set_t *cpu_set) {
 	if (min_cpu != -1) {
 		if (get_cpu_local_var(min_cpu)->status != CPU_STATUS_RESERVED)
 			get_cpu_local_var(min_cpu)->status = CPU_STATUS_RESERVED;
+		__sync_fetch_and_add(&get_cpu_local_var(min_cpu)->runq_reserved, 1);
 	}
+	ihk_mc_spinlock_unlock(&runq_reservation_lock, irqstate);
 
     return min_cpu;
 }
diff --git a/kernel/include/cls.h b/kernel/include/cls.h
index 909c4db6..920914cd 100644
--- a/kernel/include/cls.h
+++ b/kernel/include/cls.h
@@ -74,6 +74,7 @@ struct cpu_local_var {
 	struct thread *current;
 	struct list_head runq;
 	size_t runq_len;
+	size_t runq_reserved; /* Number of threads which are about to be added to runq */
 
 	struct ihk_ikc_channel_desc *ikc2linux;
 
diff --git a/kernel/include/process.h b/kernel/include/process.h
index 6e4a0912..a2db49c8 100644
--- a/kernel/include/process.h
+++ b/kernel/include/process.h
@@ -275,6 +275,7 @@ extern struct list_head	resource_set_list;
 extern mcs_rwlock_lock_t	resource_set_lock;
 extern int idle_halt;
 extern int allow_oversubscribe;
+extern ihk_spinlock_t runq_reservation_lock; /* To serialize runq reservations for competeing fork()s */
 
 struct process_hash {
 	struct list_head	list[HASH_SIZE];
diff --git a/kernel/process.c b/kernel/process.c
index a8a023d1..d765c8f3 100644
--- a/kernel/process.c
+++ b/kernel/process.c
@@ -88,6 +88,7 @@ extern void procfs_delete_thread(struct thread *);
 
 struct list_head resource_set_list;
 mcs_rwlock_lock_t    resource_set_lock;
+ihk_spinlock_t runq_reservation_lock;
 
 int idle_halt = 0;
 int allow_oversubscribe = 0;
@@ -3268,8 +3269,15 @@ void schedule(void)
 void
 release_cpuid(int cpuid)
 {
-	if (!get_cpu_local_var(cpuid)->runq_len)
-		get_cpu_local_var(cpuid)->status = CPU_STATUS_IDLE;
+	unsigned long irqstate;
+    struct cpu_local_var *v = get_cpu_local_var(cpuid);
+    irqstate = ihk_mc_spinlock_lock(&runq_reservation_lock);
+    ihk_mc_spinlock_lock_noirq(&(v->runq_lock));
+	if (!v->runq_len)
+		v->status = CPU_STATUS_IDLE;
+	__sync_fetch_and_sub(&v->runq_reserved, 1);
+    ihk_mc_spinlock_unlock_noirq(&(v->runq_lock));
+    ihk_mc_spinlock_unlock(&runq_reservation_lock, irqstate);
 }
 
 void check_need_resched(void)
@@ -3438,10 +3446,12 @@ void runq_add_thread(struct thread *thread, int cpu_id)
 {
 	struct cpu_local_var *v = get_cpu_local_var(cpu_id);
 	unsigned long irqstate;
-	
-	irqstate = ihk_mc_spinlock_lock(&(v->runq_lock));
+	irqstate = ihk_mc_spinlock_lock(&runq_reservation_lock);
+	ihk_mc_spinlock_lock_noirq(&(v->runq_lock));
 	__runq_add_thread(thread, cpu_id);
-	ihk_mc_spinlock_unlock(&(v->runq_lock), irqstate);
+	__sync_fetch_and_sub(&v->runq_reserved, 1);
+	ihk_mc_spinlock_unlock_noirq(&(v->runq_lock));
+	ihk_mc_spinlock_unlock(&runq_reservation_lock, irqstate);
 
 	procfs_create_thread(thread);