diff --git a/headers/private/kernel/smp.h b/headers/private/kernel/smp.h
index 43126eb697..0da58a0cd3 100644
--- a/headers/private/kernel/smp.h
+++ b/headers/private/kernel/smp.h
@@ -50,6 +50,9 @@ public:
 	inline	void		SetBit(int32 cpu);
 	inline	void		ClearBit(int32 cpu);
 
+	inline	void		SetBitAtomic(int32 cpu);
+	inline	void		ClearBitAtomic(int32 cpu);
+
 	inline	bool		GetBit(int32 cpu) const;
 
 	inline	bool		IsEmpty() const;
@@ -118,12 +121,28 @@ inline void
 CPUSet::SetBit(int32 cpu)
 {
 	int32* element = (int32*)&fBitmap[cpu % kArraySize];
-	atomic_or(element, 1u << (cpu / kArraySize));
+	*element |= 1u << (cpu / kArraySize);
 }
 
 
 inline void
 CPUSet::ClearBit(int32 cpu)
+{
+	int32* element = (int32*)&fBitmap[cpu % kArraySize];
+	*element &= ~uint32(1u << (cpu / kArraySize));
+}
+
+
+inline void
+CPUSet::SetBitAtomic(int32 cpu)
+{
+	int32* element = (int32*)&fBitmap[cpu % kArraySize];
+	atomic_or(element, 1u << (cpu / kArraySize));
+}
+
+
+inline void
+CPUSet::ClearBitAtomic(int32 cpu)
 {
 	int32* element = (int32*)&fBitmap[cpu % kArraySize];
 	atomic_and(element, ~uint32(1u << (cpu / kArraySize)));
diff --git a/src/system/kernel/arch/x86/arch_thread.cpp b/src/system/kernel/arch/x86/arch_thread.cpp
index 9fbbaa3ad3..11031b5264 100644
--- a/src/system/kernel/arch/x86/arch_thread.cpp
+++ b/src/system/kernel/arch/x86/arch_thread.cpp
@@ -229,8 +229,8 @@ arch_thread_context_switch(Thread* from, Thread* to)
 					!= activePagingStructures) {
 		// update on which CPUs the address space is used
 		int cpu = cpuData->cpu_num;
-		activePagingStructures->active_on_cpus.ClearBit(cpu);
-		toPagingStructures->active_on_cpus.SetBit(cpu);
+		activePagingStructures->active_on_cpus.ClearBitAtomic(cpu);
+		toPagingStructures->active_on_cpus.SetBitAtomic(cpu);
 
 		// assign the new paging structures to the CPU
 		toPagingStructures->AddReference();
diff --git a/src/system/kernel/smp.cpp b/src/system/kernel/smp.cpp
index 325fa7206d..4c4b357f68 100644
--- a/src/system/kernel/smp.cpp
+++ b/src/system/kernel/smp.cpp
@@ -822,7 +822,7 @@ check_for_message(int currentCPU, mailbox_source& sourceMailbox)
 			}
 
 			// mark it so we wont try to process this one again
-			msg->proc_bitmap.ClearBit(currentCPU);
+			msg->proc_bitmap.ClearBitAtomic(currentCPU);
 			atomic_add(&gCPU[currentCPU].ici_counter, 1);
 
 			sourceMailbox = MAILBOX_BCAST;
@@ -1000,7 +1000,7 @@ static void
 process_early_cpu_call(int32 cpu)
 {
 	sEarlyCPUCallFunction(sEarlyCPUCallCookie, cpu);
-	sEarlyCPUCallSet.ClearBit(cpu);
+	sEarlyCPUCallSet.ClearBitAtomic(cpu);
 	atomic_add(&sEarlyCPUCallCount, 1);
 }
 
@@ -1112,21 +1112,25 @@ smp_send_multicast_ici(CPUSet& cpuMask, int32 message, addr_t data,
 		return;
 
 	int currentCPU = smp_get_current_cpu();
-	bool broadcast = true;
-
-	// count target CPUs
-	int32 targetCPUs = 0;
-	for (int32 i = 0; i < sNumCPUs; i++) {
-		if (cpuMask.GetBit(i))
-			targetCPUs++;
-		else if (i != currentCPU)
-			broadcast = false;
-	}
 
 	// find_free_message leaves interrupts disabled
 	struct smp_msg *msg;
 	int state = find_free_message(&msg);
 
+	msg->proc_bitmap = cpuMask;
+	msg->proc_bitmap.ClearBit(currentCPU);
+
+	int32 targetCPUs = 0;
+	for (int32 i = 0; i < sNumCPUs; i++) {
+		if (msg->proc_bitmap.GetBit(i))
+			targetCPUs++;
+	}
+
+	if (targetCPUs == 0) {
+		panic("smp_send_multicast_ici(): 0 CPU mask");
+		return;
+	}
+
 	msg->message = message;
 	msg->data = data;
 	msg->data2 = data2;
@@ -1136,13 +1140,7 @@ smp_send_multicast_ici(CPUSet& cpuMask, int32 message, addr_t data,
 	msg->flags = flags;
 	msg->done = 0;
 
-	msg->proc_bitmap = cpuMask;
-	msg->proc_bitmap.ClearBit(currentCPU);
-	if (msg->proc_bitmap.IsEmpty()) {
-		panic("smp_send_multicast_ici(): 0 CPU mask");
-		return;
-	}
-
+	bool broadcast = targetCPUs == sNumCPUs - 1;
 
 	// stick it in the broadcast mailbox
 	acquire_spinlock_nocheck(&sBroadcastMessageSpinlock);