diff --git a/sys/amd64/amd64/mp_machdep.c b/sys/amd64/amd64/mp_machdep.c
--- a/sys/amd64/amd64/mp_machdep.c
+++ b/sys/amd64/amd64/mp_machdep.c
@@ -618,7 +618,7 @@
 smp_targeted_tlb_shootdown(cpuset_t mask, pmap_t pmap, vm_offset_t addr1,
     vm_offset_t addr2, smp_invl_cb_t curcpu_cb, enum invl_op_codes op)
 {
-	cpuset_t other_cpus, mask1;
+	cpuset_t other_cpus;
 	uint32_t generation, *p_cpudone;
 	int cpu;
 	bool is_all;
@@ -662,10 +662,7 @@
 	/* Fence between filling smp_tlb fields and clearing scoreboard. */
 	atomic_thread_fence_rel();
 
-	mask1 = mask;
-	while ((cpu = CPU_FFS(&mask1)) != 0) {
-		cpu--;
-		CPU_CLR(cpu, &mask1);
+	CPU_FOREACH_ISSET(cpu, &mask) {
 		KASSERT(*invl_scoreboard_slot(cpu) != 0,
 		    ("IPI scoreboard is zero, initiator %d target %d",
 		    PCPU_GET(cpuid), cpu));
@@ -686,9 +683,7 @@
 		ipi_selected(mask, IPI_INVLOP);
 	}
 	curcpu_cb(pmap, addr1, addr2);
-	while ((cpu = CPU_FFS(&other_cpus)) != 0) {
-		cpu--;
-		CPU_CLR(cpu, &other_cpus);
+	CPU_FOREACH_ISSET(cpu, &other_cpus) {
 		p_cpudone = invl_scoreboard_slot(cpu);
 		while (atomic_load_int(p_cpudone) != generation)
 			ia32_pause();
diff --git a/sys/amd64/vmm/io/vlapic.c b/sys/amd64/vmm/io/vlapic.c
--- a/sys/amd64/vmm/io/vlapic.c
+++ b/sys/amd64/vmm/io/vlapic.c
@@ -860,10 +860,7 @@
 		 */
 		CPU_ZERO(dmask);
 		amask = vm_active_cpus(vm);
-		while ((vcpuid = CPU_FFS(&amask)) != 0) {
-			vcpuid--;
-			CPU_CLR(vcpuid, &amask);
-
+		CPU_FOREACH_ISSET(vcpuid, &amask) {
 			vlapic = vm_lapic(vm, vcpuid);
 			dfr = vlapic->apic_page->dfr;
 			ldr = vlapic->apic_page->ldr;
@@ -1003,9 +1000,7 @@
 			break;
 		}
 
-		while ((i = CPU_FFS(&dmask)) != 0) {
-			i--;
-			CPU_CLR(i, &dmask);
+		CPU_FOREACH_ISSET(i, &dmask) {
 			if (mode == APIC_DELMODE_FIXED) {
 				lapic_intr_edge(vlapic->vm, i, vec);
 				vmm_stat_array_incr(vlapic->vm, vlapic->vcpuid,
@@ -1554,9 +1549,7 @@
 	 */
 	vlapic_calcdest(vm, &dmask, dest, phys, lowprio, false);
 
-	while ((vcpuid = CPU_FFS(&dmask)) != 0) {
-		vcpuid--;
-		CPU_CLR(vcpuid, &dmask);
+	CPU_FOREACH_ISSET(vcpuid, &dmask) {
 		if (delmode == IOART_DELEXINT) {
 			vm_inject_extint(vm, vcpuid);
 		} else {
diff --git a/sys/amd64/vmm/vmm_lapic.c b/sys/amd64/vmm/vmm_lapic.c
--- a/sys/amd64/vmm/vmm_lapic.c
+++ b/sys/amd64/vmm/vmm_lapic.c
@@ -87,9 +87,7 @@
 	else
 		CPU_SETOF(cpu, &dmask);
 	error = 0;
-	while ((cpu = CPU_FFS(&dmask)) != 0) {
-		cpu--;
-		CPU_CLR(cpu, &dmask);
+	CPU_FOREACH_ISSET(cpu, &dmask) {
 		vlapic = vm_lapic(vm, cpu);
 		error = vlapic_trigger_lvt(vlapic, vector);
 		if (error)
diff --git a/sys/i386/i386/mp_machdep.c b/sys/i386/i386/mp_machdep.c
--- a/sys/i386/i386/mp_machdep.c
+++ b/sys/i386/i386/mp_machdep.c
@@ -598,9 +598,7 @@
 		ipi_selected(mask, vector);
 	}
 	curcpu_cb(pmap, addr1, addr2);
-	while ((cpu = CPU_FFS(&other_cpus)) != 0) {
-		cpu--;
-		CPU_CLR(cpu, &other_cpus);
+	CPU_FOREACH_ISSET(cpu, &other_cpus) {
 		p_cpudone = &cpuid_to_pcpu[cpu]->pc_smp_tlb_done;
 		while (*p_cpudone != generation)
 			ia32_pause();
diff --git a/sys/sys/cpuset.h b/sys/sys/cpuset.h
--- a/sys/sys/cpuset.h
+++ b/sys/sys/cpuset.h
@@ -66,6 +66,8 @@
 #define	CPU_COPY_STORE_REL(f, t)	BIT_COPY_STORE_REL(CPU_SETSIZE, f, t)
 #define	CPU_FFS(p)			BIT_FFS(CPU_SETSIZE, p)
 #define	CPU_FLS(p)			BIT_FLS(CPU_SETSIZE, p)
+#define	CPU_FOREACH_ISSET(i, p)		BIT_FOREACH_ISSET(CPU_SETSIZE, i, p)
+#define	CPU_FOREACH_ISCLR(i, p)		BIT_FOREACH_ISCLR(CPU_SETSIZE, i, p)
 #define	CPU_COUNT(p)			((int)BIT_COUNT(CPU_SETSIZE, p))
 #define	CPUSET_FSET			BITSET_FSET(_NCPUWORDS)
 #define	CPUSET_T_INITIALIZER		BITSET_T_INITIALIZER
diff --git a/sys/x86/x86/mp_x86.c b/sys/x86/x86/mp_x86.c
--- a/sys/x86/x86/mp_x86.c
+++ b/sys/x86/x86/mp_x86.c
@@ -1290,9 +1290,7 @@
 	if (ipi == IPI_STOP_HARD)
 		CPU_OR_ATOMIC(&ipi_stop_nmi_pending, &cpus);
 
-	while ((cpu = CPU_FFS(&cpus)) != 0) {
-		cpu--;
-		CPU_CLR(cpu, &cpus);
+	CPU_FOREACH_ISSET(cpu, &cpus) {
 		CTR3(KTR_SMP, "%s: cpu: %d ipi: %x", __func__, cpu, ipi);
 		ipi_send_cpu(cpu, ipi);
 	}