diff --git a/arch/lkl/include/uapi/asm/host_ops.h b/arch/lkl/include/uapi/asm/host_ops.h
index d5de01bc7120dd..88ef9c9288f6eb 100644
--- a/arch/lkl/include/uapi/asm/host_ops.h
+++ b/arch/lkl/include/uapi/asm/host_ops.h
@@ -151,4 +151,13 @@ int lkl_is_running(void);
 int lkl_printf(const char *, ...);
 void lkl_bug(const char *, ...);
 
+/* atomic ops */
+int lkl__sync_fetch_and_sub(int *ptr, int value);
+int lkl__sync_fetch_and_add(int *ptr, int value);
+long lkl__sync_fetch_and_or(long *ptr, long value);
+long lkl__sync_fetch_and_and(long *ptr, long value);
+void lkl__sync_synchronize(void);
+void atomic_ops_init(void);
+void atomic_ops_cleanup(void);
+
 #endif
diff --git a/arch/lkl/kernel/Makefile b/arch/lkl/kernel/Makefile
index ef489f2f717618..0021141dba280c 100644
--- a/arch/lkl/kernel/Makefile
+++ b/arch/lkl/kernel/Makefile
@@ -1,4 +1,4 @@
 extra-y := vmlinux.lds
 
 obj-y = setup.o threads.o irq.o time.o syscalls.o misc.o console.o \
-	syscalls_32.o cpu.o
+	syscalls_32.o cpu.o atomic.o
diff --git a/arch/lkl/kernel/atomic.c b/arch/lkl/kernel/atomic.c
new file mode 100644
index 00000000000000..a1129e40ee3cdb
--- /dev/null
+++ b/arch/lkl/kernel/atomic.c
@@ -0,0 +1,97 @@
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <asm/host_ops.h>
+
+#if defined(__ARMEL__)
+static void *atomic_lock;
+
+long lkl__sync_fetch_and_or(long *ptr, long value)
+{
+	lkl_ops->sem_down(atomic_lock);
+	*ptr = value;
+	lkl_ops->sem_up(atomic_lock);
+	return 0;
+}
+
+long lkl__sync_fetch_and_and(long *ptr, long value)
+{
+	int tmp;
+
+	lkl_ops->sem_down(atomic_lock);
+	tmp = *ptr;
+	*ptr *= value;
+	lkl_ops->sem_up(atomic_lock);
+	return tmp;
+}
+
+int lkl__sync_fetch_and_add(int *ptr, int value)
+{
+	int tmp;
+
+	lkl_ops->sem_down(atomic_lock);
+	tmp = *ptr;
+	*ptr += value;
+	lkl_ops->sem_up(atomic_lock);
+	return tmp;
+}
+
+int lkl__sync_fetch_and_sub(int *ptr, int value)
+{
+	int tmp;
+
+	lkl_ops->sem_down(atomic_lock);
+	tmp = *ptr;
+	*ptr -= value;
+	lkl_ops->sem_up(atomic_lock);
+	return tmp;
+}
+
+void lkl__sync_synchronize(void)
+{
+}
+
+void atomic_ops_init(void)
+{
+	atomic_lock = lkl_ops->sem_alloc(1);
+}
+
+void atomic_ops_cleanup(void)
+{
+	lkl_ops->sem_free(atomic_lock);
+}
+
+#else
+long lkl__sync_fetch_and_or(long *ptr, long value)
+{
+	return __sync_fetch_and_or(ptr, value);
+}
+
+long lkl__sync_fetch_and_and(long *ptr, long value)
+{
+	return __sync_fetch_and_and(ptr, value);
+}
+
+int lkl__sync_fetch_and_add(int *ptr, int value)
+{
+	return __sync_fetch_and_add(ptr, value);
+}
+
+int lkl__sync_fetch_and_sub(int *ptr, int value)
+{
+	return __sync_fetch_and_sub(ptr, value);
+}
+
+void lkl__sync_synchronize(void)
+{
+	return __sync_synchronize();
+}
+
+void atomic_ops_init(void)
+{
+}
+
+void atomic_ops_cleanup(void)
+{
+}
+#endif
+
diff --git a/arch/lkl/kernel/cpu.c b/arch/lkl/kernel/cpu.c
index 2c315262a935e3..dbc47f8b6f76a9 100644
--- a/arch/lkl/kernel/cpu.c
+++ b/arch/lkl/kernel/cpu.c
@@ -66,7 +66,7 @@ static int __cpu_try_get_lock(int n)
 {
 	lkl_thread_t self;
 
-	if (__sync_fetch_and_add(&cpu.shutdown_gate, n) >= MAX_THREADS)
+	if (lkl__sync_fetch_and_add(&cpu.shutdown_gate, n) >= MAX_THREADS)
 		return -2;
 
 	lkl_ops->mutex_lock(cpu.lock);
@@ -89,7 +89,7 @@ static void __cpu_try_get_unlock(int lock_ret, int n)
 {
 	if (lock_ret >= -1)
 		lkl_ops->mutex_unlock(cpu.lock);
-	__sync_fetch_and_sub(&cpu.shutdown_gate, n);
+	lkl__sync_fetch_and_sub(&cpu.shutdown_gate, n);
 }
 
 void lkl_cpu_change_owner(lkl_thread_t owner)
@@ -173,7 +173,7 @@ int lkl_cpu_try_run_irq(int irq)
 
 void lkl_cpu_shutdown(void)
 {
-	__sync_fetch_and_add(&cpu.shutdown_gate, MAX_THREADS);
+	lkl__sync_fetch_and_add(&cpu.shutdown_gate, MAX_THREADS);
 }
 
 void lkl_cpu_wait_shutdown(void)
@@ -184,7 +184,7 @@ void lkl_cpu_wait_shutdown(void)
 
 static void lkl_cpu_cleanup(bool shutdown)
 {
-	while (__sync_fetch_and_add(&cpu.shutdown_gate, 0) > MAX_THREADS)
+	while (lkl__sync_fetch_and_add(&cpu.shutdown_gate, 0) > MAX_THREADS)
 		;
 
 	if (shutdown)
diff --git a/arch/lkl/kernel/irq.c b/arch/lkl/kernel/irq.c
index b5dbbaa7ebba39..0635afbaaeda70 100644
--- a/arch/lkl/kernel/irq.c
+++ b/arch/lkl/kernel/irq.c
@@ -27,14 +27,14 @@ static inline unsigned long test_and_clear_irq_index_status(void)
 {
 	if (!irq_index_status)
 		return 0;
-	return __sync_fetch_and_and(&irq_index_status, 0);
+	return lkl__sync_fetch_and_and(&irq_index_status, 0);
 }
 
 static inline unsigned long test_and_clear_irq_status(int index)
 {
 	if (!&irq_status[index])
 		return 0;
-	return __sync_fetch_and_and(&irq_status[index], 0);
+	return lkl__sync_fetch_and_and(&irq_status[index], 0);
 }
 
 void set_irq_pending(int irq)
@@ -42,8 +42,8 @@ void set_irq_pending(int irq)
 	int index = irq / IRQ_STATUS_BITS;
 	int bit = irq % IRQ_STATUS_BITS;
 
-	__sync_fetch_and_or(&irq_status[index], BIT(bit));
-	__sync_fetch_and_or(&irq_index_status, BIT(index));
+	lkl__sync_fetch_and_or(&irq_status[index], BIT(bit));
+	lkl__sync_fetch_and_or(&irq_index_status, BIT(index));
 }
 
 static struct irq_info {
diff --git a/arch/lkl/kernel/setup.c b/arch/lkl/kernel/setup.c
index 501d32175dc426..37edb46e2a75a2 100644
--- a/arch/lkl/kernel/setup.c
+++ b/arch/lkl/kernel/setup.c
@@ -35,6 +35,7 @@ void __init setup_arch(char **cl)
 
 static void __init lkl_run_kernel(void *arg)
 {
+	atomic_ops_init();
 	threads_init();
 	lkl_cpu_get();
 	start_kernel();
@@ -126,6 +127,7 @@ long lkl_sys_halt(void)
 
 	syscalls_cleanup();
 	threads_cleanup();
+	atomic_ops_cleanup();
 	/* Shutdown the clockevents source. */
 	tick_suspend_local();
 	free_mem();
diff --git a/arch/lkl/kernel/threads.c b/arch/lkl/kernel/threads.c
index db8a6b7e34f015..82e0e6825d9c2a 100644
--- a/arch/lkl/kernel/threads.c
+++ b/arch/lkl/kernel/threads.c
@@ -5,7 +5,7 @@
 #include <asm/cpu.h>
 #include <asm/sched.h>
 
-static volatile int threads_counter;
+static int threads_counter;
 
 static int init_ti(struct thread_info *ti)
 {
@@ -123,7 +123,7 @@ struct task_struct *__switch_to(struct task_struct *prev,
 	}
 
 	if (_prev->dead) {
-		__sync_fetch_and_sub(&threads_counter, 1);
+		lkl__sync_fetch_and_sub(&threads_counter, 1);
 		lkl_ops->thread_exit();
 	}
 
@@ -193,7 +193,7 @@ int copy_thread(unsigned long clone_flags, unsigned long esp,
 		return -ENOMEM;
 	}
 
-	__sync_fetch_and_add(&threads_counter, 1);
+	lkl__sync_fetch_and_add(&threads_counter, 1);
 
 	return 0;
 }
@@ -220,7 +220,7 @@ void threads_init(void)
 
 void threads_cnt_dec(void)
 {
-	__sync_fetch_and_sub(&threads_counter, 1);
+	lkl__sync_fetch_and_sub(&threads_counter, 1);
 }
 
 void threads_cleanup(void)