diff --git a/contrib/ofed/libmlx5/mlx5.c b/contrib/ofed/libmlx5/mlx5.c
--- a/contrib/ofed/libmlx5/mlx5.c
+++ b/contrib/ofed/libmlx5/mlx5.c
@@ -363,8 +363,12 @@
 	mlx5_local_cpu_set(ibdev, &dev_local_cpus);
 
 	/* check if my cpu set is in dev cpu */
+#if __FreeBSD_version < 1400045 /* TO BE ADJUSTED */
 	CPU_OR(&result_set, &my_cpus);
 	CPU_OR(&result_set, &dev_local_cpus);
+#else
+	CPU_OR(&result_set, &my_cpus, &dev_local_cpus);
+#endif
 	stall_enable = CPU_EQUAL(&result_set, &dev_local_cpus) ? 0 : 1;
 
 out:
diff --git a/sys/amd64/amd64/pmap.c b/sys/amd64/amd64/pmap.c
--- a/sys/amd64/amd64/pmap.c
+++ b/sys/amd64/amd64/pmap.c
@@ -8290,7 +8290,7 @@
 		other_cpus = all_cpus;
 		critical_enter();
 		CPU_CLR(PCPU_GET(cpuid), &other_cpus);
-		CPU_AND(&other_cpus, &pmap->pm_active);
+		CPU_AND(&other_cpus, &other_cpus, &pmap->pm_active);
 		critical_exit();
 		KASSERT(CPU_EMPTY(&other_cpus), ("pmap active %p", pmap));
 	}
diff --git a/sys/amd64/vmm/vmm.c b/sys/amd64/vmm/vmm.c
--- a/sys/amd64/vmm/vmm.c
+++ b/sys/amd64/vmm/vmm.c
@@ -1304,7 +1304,7 @@
 	mtx_lock(&vm->rendezvous_mtx);
 	while (vm->rendezvous_func != NULL) {
 		/* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
-		CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus);
+		CPU_AND(&vm->rendezvous_req_cpus, &vm->rendezvous_req_cpus, &vm->active_cpus);
 
 		if (vcpuid != -1 &&
 		    CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
diff --git a/sys/dev/acpica/acpi.c b/sys/dev/acpica/acpi.c
--- a/sys/dev/acpica/acpi.c
+++ b/sys/dev/acpica/acpi.c
@@ -1173,7 +1173,7 @@
 			return (error);
 		if (setsize != sizeof(cpuset_t))
 			return (EINVAL);
-		CPU_AND(cpuset, &cpuset_domain[d]);
+		CPU_AND(cpuset, cpuset, &cpuset_domain[d]);
 		return (0);
 	default:
 		return (bus_generic_get_cpus(dev, child, op, setsize, cpuset));
diff --git a/sys/i386/i386/pmap.c b/sys/i386/i386/pmap.c
--- a/sys/i386/i386/pmap.c
+++ b/sys/i386/i386/pmap.c
@@ -1243,7 +1243,7 @@
 		cpuid = PCPU_GET(cpuid);
 		other_cpus = all_cpus;
 		CPU_CLR(cpuid, &other_cpus);
-		CPU_AND(&other_cpus, &pmap->pm_active);
+		CPU_AND(&other_cpus, &other_cpus, &pmap->pm_active);
 		mask = &other_cpus;
 	}
 	smp_masked_invlpg(*mask, va, pmap, pmap_curcpu_cb_dummy);
@@ -1276,7 +1276,7 @@
 		cpuid = PCPU_GET(cpuid);
 		other_cpus = all_cpus;
 		CPU_CLR(cpuid, &other_cpus);
-		CPU_AND(&other_cpus, &pmap->pm_active);
+		CPU_AND(&other_cpus, &other_cpus, &pmap->pm_active);
 		mask = &other_cpus;
 	}
 	smp_masked_invlpg_range(*mask, sva, eva, pmap, pmap_curcpu_cb_dummy);
@@ -1299,7 +1299,7 @@
 		cpuid = PCPU_GET(cpuid);
 		other_cpus = all_cpus;
 		CPU_CLR(cpuid, &other_cpus);
-		CPU_AND(&other_cpus, &pmap->pm_active);
+		CPU_AND(&other_cpus, &other_cpus, &pmap->pm_active);
 		mask = &other_cpus;
 	}
 	smp_masked_invltlb(*mask, pmap, pmap_curcpu_cb_dummy);
diff --git a/sys/i386/i386/vm_machdep.c b/sys/i386/i386/vm_machdep.c
--- a/sys/i386/i386/vm_machdep.c
+++ b/sys/i386/i386/vm_machdep.c
@@ -604,9 +604,9 @@
 	if ((flags & SFB_CPUPRIVATE) == 0) {
 		other_cpus = all_cpus;
 		CPU_CLR(cpuid, &other_cpus);
-		CPU_ANDNOT(&other_cpus, &sf->cpumask);
+		CPU_ANDNOT(&other_cpus, &other_cpus, &sf->cpumask);
 		if (!CPU_EMPTY(&other_cpus)) {
-			CPU_OR(&sf->cpumask, &other_cpus);
+			CPU_OR(&sf->cpumask, &sf->cpumask, &other_cpus);
 			smp_masked_invlpg(other_cpus, sf->kva, kernel_pmap,
 			    sf_buf_shootdown_curcpu_cb);
 		}
diff --git a/sys/kern/kern_cpuset.c b/sys/kern/kern_cpuset.c
--- a/sys/kern/kern_cpuset.c
+++ b/sys/kern/kern_cpuset.c
@@ -326,7 +326,7 @@
 	set->cs_flags = 0;
 	mtx_lock_spin(&cpuset_lock);
 	set->cs_domain = domain;
-	CPU_AND(&set->cs_mask, &parent->cs_mask);
+	CPU_AND(&set->cs_mask, &set->cs_mask, &parent->cs_mask);
 	set->cs_id = id;
 	set->cs_parent = cpuset_ref(parent);
 	LIST_INSERT_HEAD(&parent->cs_children, set, cs_siblings);
@@ -645,8 +645,7 @@
 	if (set->cs_flags & CPU_SET_RDONLY)
 		return (EPERM);
 	if (augment_mask) {
-		CPU_COPY(&set->cs_mask, &newmask);
-		CPU_AND(&newmask, mask);
+		CPU_AND(&newmask, &set->cs_mask, mask);
 	} else
 		CPU_COPY(mask, &newmask);
 
@@ -668,7 +667,7 @@
 	struct cpuset *nset;
 
 	mtx_assert(&cpuset_lock, MA_OWNED);
-	CPU_AND(&set->cs_mask, mask);
+	CPU_AND(&set->cs_mask, &set->cs_mask, mask);
 	LIST_FOREACH(nset, &set->cs_children, cs_siblings) 
 		cpuset_update(nset, &set->cs_mask);
 
@@ -1083,8 +1082,7 @@
 	 * restriction to the new set, otherwise take it wholesale.
 	 */
 	if (CPU_CMP(&tdset->cs_mask, &parent->cs_mask) != 0) {
-		CPU_COPY(&tdset->cs_mask, mask);
-		CPU_AND(mask, &set->cs_mask);
+		CPU_AND(mask, &tdset->cs_mask, &set->cs_mask);
 	} else
 		CPU_COPY(&set->cs_mask, mask);
 
@@ -1153,8 +1151,7 @@
 	pbase = cpuset_getbase(td->td_cpuset);
 
 	/* Copy process mask, then further apply the new root mask. */
-	CPU_COPY(&pbase->cs_mask, &nmask);
-	CPU_AND(&nmask, &nroot->cs_mask);
+	CPU_AND(&nmask, &pbase->cs_mask, &nroot->cs_mask);
 
 	domainset_copy(pbase->cs_domain, &ndomain);
 	DOMAINSET_AND(&ndomain.ds_mask, &set->cs_domain->ds_mask);
@@ -1946,7 +1943,7 @@
 		case CPU_WHICH_PID:
 			FOREACH_THREAD_IN_PROC(p, ttd) {
 				thread_lock(ttd);
-				CPU_OR(mask, &ttd->td_cpuset->cs_mask);
+				CPU_OR(mask, mask, &ttd->td_cpuset->cs_mask);
 				thread_unlock(ttd);
 			}
 			break;
diff --git a/sys/kern/kern_rmlock.c b/sys/kern/kern_rmlock.c
--- a/sys/kern/kern_rmlock.c
+++ b/sys/kern/kern_rmlock.c
@@ -548,8 +548,7 @@
 
 	if (CPU_CMP(&rm->rm_writecpus, &all_cpus)) {
 		/* Get all read tokens back */
-		readcpus = all_cpus;
-		CPU_ANDNOT(&readcpus, &rm->rm_writecpus);
+		CPU_ANDNOT(&readcpus, &all_cpus, &rm->rm_writecpus);
 		rm->rm_writecpus = all_cpus;
 
 		/*
diff --git a/sys/kern/sched_4bsd.c b/sys/kern/sched_4bsd.c
--- a/sys/kern/sched_4bsd.c
+++ b/sys/kern/sched_4bsd.c
@@ -1165,8 +1165,8 @@
 		return (0);
 
 	CPU_SETOF(me, &dontuse);
-	CPU_OR(&dontuse, &stopped_cpus);
-	CPU_OR(&dontuse, &hlt_cpus_mask);
+	CPU_OR(&dontuse, &dontuse, &stopped_cpus);
+	CPU_OR(&dontuse, &dontuse, &hlt_cpus_mask);
 	CPU_ZERO(&map2);
 	if (forward_wakeup_use_loop) {
 		STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
@@ -1179,8 +1179,7 @@
 	}
 
 	if (forward_wakeup_use_mask) {
-		map = idle_cpus_mask;
-		CPU_ANDNOT(&map, &dontuse);
+		CPU_ANDNOT(&map, &idle_cpus_mask, &dontuse);
 
 		/* If they are both on, compare and use loop if different. */
 		if (forward_wakeup_use_loop) {
@@ -1366,8 +1365,7 @@
 	        kick_other_cpu(td->td_priority, cpu);
 	} else {
 		if (!single_cpu) {
-			tidlemsk = idle_cpus_mask;
-			CPU_ANDNOT(&tidlemsk, &hlt_cpus_mask);
+			CPU_ANDNOT(&tidlemsk, &idle_cpus_mask, &hlt_cpus_mask);
 			CPU_CLR(cpuid, &tidlemsk);
 
 			if (!CPU_ISSET(cpuid, &idle_cpus_mask) &&
diff --git a/sys/kern/subr_kdb.c b/sys/kern/subr_kdb.c
--- a/sys/kern/subr_kdb.c
+++ b/sys/kern/subr_kdb.c
@@ -707,8 +707,7 @@
 
 	if (!SCHEDULER_STOPPED()) {
 #ifdef SMP
-		other_cpus = all_cpus;
-		CPU_ANDNOT(&other_cpus, &stopped_cpus);
+		CPU_ANDNOT(&other_cpus, &all_cpus, &stopped_cpus);
 		CPU_CLR(PCPU_GET(cpuid), &other_cpus);
 		stop_cpus_hard(other_cpus);
 #endif
@@ -746,7 +745,7 @@
 	if (did_stop_cpus) {
 		curthread->td_stopsched = 0;
 #ifdef SMP
-		CPU_AND(&other_cpus, &stopped_cpus);
+		CPU_AND(&other_cpus, &other_cpus, &stopped_cpus);
 		restart_cpus(other_cpus);
 #endif
 	}
diff --git a/sys/kern/subr_smp.c b/sys/kern/subr_smp.c
--- a/sys/kern/subr_smp.c
+++ b/sys/kern/subr_smp.c
@@ -761,7 +761,7 @@
 			    parent,
 			    cpusetobj_strprint(cpusetbuf, &parent->cg_mask),
 			    cpusetobj_strprint(cpusetbuf2, &child->cg_mask));
-		CPU_OR(&parent->cg_mask, &child->cg_mask);
+		CPU_OR(&parent->cg_mask, &parent->cg_mask, &child->cg_mask);
 		parent->cg_count += child->cg_count;
 	}
 
diff --git a/sys/net/iflib.c b/sys/net/iflib.c
--- a/sys/net/iflib.c
+++ b/sys/net/iflib.c
@@ -4969,7 +4969,7 @@
 	for (i = 0; i < scctx->isc_nrxqsets; i++)
 		CPU_SET(get_cpuid_for_queue(ctx, first_valid, i, false),
 		    &assigned_cpus);
-	CPU_AND(&assigned_cpus, &ctx->ifc_cpus);
+	CPU_AND(&assigned_cpus, &assigned_cpus, &ctx->ifc_cpus);
 	cores_consumed = CPU_COUNT(&assigned_cpus);
 
 	mtx_lock(&cpu_offset_mtx);
diff --git a/sys/powerpc/ofw/ofw_pcibus.c b/sys/powerpc/ofw/ofw_pcibus.c
--- a/sys/powerpc/ofw/ofw_pcibus.c
+++ b/sys/powerpc/ofw/ofw_pcibus.c
@@ -405,7 +405,7 @@
 			return (error);
 		if (setsize != sizeof(cpuset_t))
 			return (EINVAL);
-		CPU_AND(cpuset, &cpuset_domain[d]);
+		CPU_AND(cpuset, cpuset, &cpuset_domain[d]);
 		return (0);
 	default:
 		return (bus_generic_get_cpus(dev, child, op, setsize, cpuset));
diff --git a/sys/sys/bitset.h b/sys/sys/bitset.h
--- a/sys/sys/bitset.h
+++ b/sys/sys/bitset.h
@@ -312,7 +312,13 @@
 /*
  * Dynamically allocate a bitset.
  */
+#if defined(_KERNEL)
 #define __BITSET_ALLOC(_s, mt, mf) malloc(__BITSET_SIZE((_s)), mt, (mf))
+#define __BITSET_FREE(p, mt)	free(p, mt)
+#else /* _KERNEL */
+#define __BITSET_ALLOC(_s)	malloc(__BITSET_SIZE((_s)))
+#define __BITSET_FREE(p)	free(p)
+#endif /* _KERNEL */
 
 #define	BIT_AND(_s, d, s)			__BIT_AND(_s, d, s)
 #define	BIT_AND2(_s, d, s1, s2)			__BIT_AND2(_s, d, s1, s2)
@@ -352,6 +358,7 @@
 #define	BIT_ZERO(_s, p)				__BIT_ZERO(_s, p)
 
 #define BITSET_ALLOC(_s, mt, mf)		__BITSET_ALLOC(_s, mt, mf)
+#define	BITSET_FREE(p)				__BITSET_FREE(p)
 #define	BITSET_FSET(n)				__BITSET_FSET(n)
 #define	BITSET_SIZE(_s)				__BITSET_SIZE(_s)
 #define	BITSET_T_INITIALIZER(x)			__BITSET_T_INITIALIZER(x)
diff --git a/sys/sys/cpuset.h b/sys/sys/cpuset.h
--- a/sys/sys/cpuset.h
+++ b/sys/sys/cpuset.h
@@ -36,6 +36,7 @@
 
 #include <sys/_cpuset.h>
 
+#include <sys/_bitset.h>
 #include <sys/bitset.h>
 
 #define	_NCPUBITS	_BITSET_BITS
@@ -56,9 +57,10 @@
 #define	CPU_SUBSET(p, c)		__BIT_SUBSET(CPU_SETSIZE, p, c)
 #define	CPU_OVERLAP(p, c)		__BIT_OVERLAP(CPU_SETSIZE, p, c)
 #define	CPU_CMP(p, c)			__BIT_CMP(CPU_SETSIZE, p, c)
-#define	CPU_OR(d, s)			__BIT_OR(CPU_SETSIZE, d, s)
-#define	CPU_AND(d, s)			__BIT_AND(CPU_SETSIZE, d, s)
-#define	CPU_ANDNOT(d, s)		__BIT_ANDNOT(CPU_SETSIZE, d, s)
+#define	CPU_OR(d, s1, s2)		__BIT_OR2(CPU_SETSIZE, d, s1, s2)
+#define	CPU_AND(d, s1, s2)		__BIT_AND2(CPU_SETSIZE, d, s1, s2)
+#define	CPU_ANDNOT(d, s1, s2)		__BIT_ANDNOT2(CPU_SETSIZE, d, s1, s2)
+#define	CPU_XOR(d, s1, s2)		__BIT_XOR2(CPU_SETSIZE, d, s1, s2)
 #define	CPU_CLR_ATOMIC(n, p)		__BIT_CLR_ATOMIC(CPU_SETSIZE, n, p)
 #define	CPU_SET_ATOMIC(n, p)		__BIT_SET_ATOMIC(CPU_SETSIZE, n, p)
 #define	CPU_SET_ATOMIC_ACQ(n, p)	__BIT_SET_ATOMIC_ACQ(CPU_SETSIZE, n, p)
@@ -73,6 +75,20 @@
 #define	CPUSET_FSET			__BITSET_FSET(_NCPUWORDS)
 #define	CPUSET_T_INITIALIZER(x)		__BITSET_T_INITIALIZER(x)
 
+#if !defined(_KERNEL)
+#define CPU_ALLOC(_s)			__BITSET_ALLOC(_s)
+#define CPU_ALLOC_SIZE(_s)		__BITSET_SIZE(_s)
+#define CPU_FREE(p)			__BITSET_FREE(p)
+
+#define CPU_ISSET_S(n, _s, p)		__BIT_ISSET(_s, n, p)
+#define CPU_SET_S(n, _s, p)		__BIT_SET(_s, n, p)
+#define CPU_ZERO_S(_s, p)		__BIT_ZERO(_s, p)
+
+#define	CPU_OR_S(_s, d, s1, s2)		__BIT_OR2(_s, d, s1, s2)
+#define	CPU_AND_S(_s, d, s1, s2)	__BIT_AND2(_s, d, s1, s2)
+#define	CPU_XOR_S(_s, d, s1, s2)	__BIT_XOR2(_s, d, s1, s2)
+#endif
+
 /*
  * Valid cpulevel_t values.
  */
diff --git a/sys/x86/x86/cpu_machdep.c b/sys/x86/x86/cpu_machdep.c
--- a/sys/x86/x86/cpu_machdep.c
+++ b/sys/x86/x86/cpu_machdep.c
@@ -475,7 +475,7 @@
 	if (smp_started) {
 		map = all_cpus;
 		CPU_CLR(PCPU_GET(cpuid), &map);
-		CPU_ANDNOT(&map, &stopped_cpus);
+		CPU_ANDNOT(&map, &map, &stopped_cpus);
 		if (!CPU_EMPTY(&map)) {
 			printf("cpu_reset: Stopping other CPUs\n");
 			stop_cpus(map);