Index: sys/net/if_lagg.h
===================================================================
--- sys/net/if_lagg.h
+++ sys/net/if_lagg.h
@@ -206,6 +206,7 @@
 	struct ifnet			*sc_ifp;	/* virtual interface */
 	struct rmlock			sc_mtx;
 	struct sx			sc_sx;
+	_Bool				sc_slowpath;
 	int				sc_proto;	/* lagg protocol */
 	u_int				sc_count;	/* number of ports */
 	u_int				sc_active;	/* active port count */
@@ -257,20 +258,59 @@
 
 #define	LAGG_LOCK_INIT(_sc)	rm_init(&(_sc)->sc_mtx, "if_lagg rmlock")
 #define	LAGG_LOCK_DESTROY(_sc)	rm_destroy(&(_sc)->sc_mtx)
-#define	LAGG_RLOCK(_sc, _p)	rm_rlock(&(_sc)->sc_mtx, (_p))
-#define	LAGG_WLOCK(_sc)		rm_wlock(&(_sc)->sc_mtx)
+#define	LAGG_RLOCK(_sc, _p)	do {		\
+	rm_rlock(&(_sc)->sc_mtx, (_p));		\
+	if ((_sc)->sc_slowpath == 0)		\
+		break;				\
+	rm_runlock(&(_sc)->sc_mtx, (_p));	\
+	DELAY(1);				\
+} while(1)
 #define	LAGG_RUNLOCK(_sc, _p)	rm_runlock(&(_sc)->sc_mtx, (_p))
-#define	LAGG_WUNLOCK(_sc)	rm_wunlock(&(_sc)->sc_mtx)
-#define	LAGG_RLOCK_ASSERT(_sc)	rm_assert(&(_sc)->sc_mtx, RA_RLOCKED)
-#define	LAGG_WLOCK_ASSERT(_sc)	rm_assert(&(_sc)->sc_mtx, RA_WLOCKED)
-#define	LAGG_UNLOCK_ASSERT(_sc)	rm_assert(&(_sc)->sc_mtx, RA_UNLOCKED)
 
 #define	LAGG_SX_INIT(_sc)	sx_init(&(_sc)->sc_sx, "if_lagg sx")
 #define	LAGG_SX_DESTROY(_sc)	sx_destroy(&(_sc)->sc_sx)
-#define	LAGG_SLOCK(_sc)		sx_slock(&(_sc)->sc_sx)
-#define	LAGG_XLOCK(_sc)		sx_xlock(&(_sc)->sc_sx)
-#define	LAGG_SUNLOCK(_sc)	sx_sunlock(&(_sc)->sc_sx)
-#define	LAGG_XUNLOCK(_sc)	sx_xunlock(&(_sc)->sc_sx)
+#define	LAGG_SLOCK(_sc)		do {			\
+	do {						\
+		rm_wlock(&(_sc)->sc_mtx);		\
+		if ((_sc)->sc_slowpath == 0) {		\
+			(_sc)->sc_slowpath = 1;		\
+			rm_wunlock(&(_sc)->sc_mtx);	\
+			break;				\
+		}					\
+		DELAY(1);				\
+		rm_wunlock(&(_sc)->sc_mtx);		\
+	} while(1);					\
+	sx_slock(&(_sc)->sc_sx);			\
+} while(0)
+
+#define	LAGG_XLOCK(_sc)	do {				\
+	do {						\
+		rm_wlock(&(_sc)->sc_mtx);		\
+		if ((_sc)->sc_slowpath == 0) {		\
+			(_sc)->sc_slowpath = 1;		\
+			rm_wunlock(&(_sc)->sc_mtx);	\
+			break;				\
+		}					\
+		DELAY(1);				\
+		rm_wunlock(&(_sc)->sc_mtx);		\
+	} while(1);					\
+	sx_xlock(&(_sc)->sc_sx);			\
+} while(0)
+
+#define	LAGG_SUNLOCK(_sc)	do {	\
+	sx_sunlock(&(_sc)->sc_sx);	\
+	rm_wlock(&(_sc)->sc_mtx);	\
+	(_sc)->sc_slowpath = 0;		\
+	rm_wunlock(&(_sc)->sc_mtx);	\
+} while(0)
+
+#define	LAGG_XUNLOCK(_sc)	do {	\
+	sx_xunlock(&(_sc)->sc_sx);	\
+	rm_wlock(&(_sc)->sc_mtx);	\
+	(_sc)->sc_slowpath = 0;		\
+	rm_wunlock(&(_sc)->sc_mtx);	\
+} while(0)
+
 #define	LAGG_SXLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_LOCKED)
 #define	LAGG_SLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_SLOCKED)
 #define	LAGG_XLOCK_ASSERT(_sc)	sx_assert(&(_sc)->sc_sx, SA_XLOCKED)
Index: sys/net/if_lagg.c
===================================================================
--- sys/net/if_lagg.c
+++ sys/net/if_lagg.c
@@ -334,14 +334,11 @@
 	lagg_proto pr;
 
 	LAGG_XLOCK_ASSERT(sc);
-	LAGG_WLOCK_ASSERT(sc);
 	pr = sc->sc_proto;
 	sc->sc_proto = LAGG_PROTO_NONE;
 
 	if (lagg_protos[pr].pr_detach != NULL)
 		lagg_protos[pr].pr_detach(sc);
-	else
-		LAGG_WUNLOCK(sc);
 }
 
 static int
@@ -550,9 +547,7 @@
 		lagg_port_destroy(lp, 1);
 
 	/* Unhook the aggregation protocol */
-	LAGG_WLOCK(sc);
 	lagg_proto_detach(sc);
-	LAGG_UNLOCK_ASSERT(sc);
 	LAGG_XUNLOCK(sc);
 
 	ifmedia_removeall(&sc->sc_media);
@@ -564,7 +559,6 @@
 	LAGG_LIST_UNLOCK();
 
 	LAGG_SX_DESTROY(sc);
-	LAGG_LOCK_DESTROY(sc);
 	free(sc, M_DEVBUF);
 }
 
@@ -689,17 +683,14 @@
 	bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN);
 	lp->lp_ifcapenable = ifp->if_capenable;
 	if (SLIST_EMPTY(&sc->sc_ports)) {
-		LAGG_WLOCK(sc);
 		bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
 		lagg_proto_lladdr(sc);
-		LAGG_WUNLOCK(sc);
 		EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
 	} else {
 		if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
 	}
 	lagg_setflags(lp, 1);
 
-	LAGG_WLOCK(sc);
 	if (SLIST_EMPTY(&sc->sc_ports))
 		sc->sc_primary = lp;
 
@@ -738,13 +729,9 @@
 
 	lagg_setmulti(lp);
 
-	LAGG_WUNLOCK(sc);
-
 	if ((error = lagg_proto_addport(sc, lp)) != 0) {
 		/* Remove the port, without calling pr_delport. */
-		LAGG_WLOCK(sc);
 		lagg_port_destroy(lp, 0);
-		LAGG_UNLOCK_ASSERT(sc);
 		return (error);
 	}
 
@@ -786,11 +773,8 @@
 
 	LAGG_XLOCK_ASSERT(sc);
 
-	if (rundelport) {
-		LAGG_WLOCK(sc);
+	if (rundelport)
 		lagg_proto_delport(sc, lp);
-	} else
-		LAGG_WLOCK_ASSERT(sc);
 
 	if (lp->lp_detaching == 0)
 		lagg_clrmulti(lp);
@@ -824,10 +808,8 @@
 		if (sc->sc_destroying == 0) {
 			bcopy(lladdr, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN);
 			lagg_proto_lladdr(sc);
-			LAGG_WUNLOCK(sc);
 			EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp);
-		} else
-			LAGG_WUNLOCK(sc);
+		}
 
 		/*
 		 * Update lladdr for each port (new primary needs update
@@ -835,8 +817,7 @@
 		 */
 		SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries)
 			if_setlladdr(lp_ptr->lp_ifp, lladdr, ETHER_ADDR_LEN);
-	} else
-		LAGG_WUNLOCK(sc);
+	}
 
 	if (lp->lp_ifflags)
 		if_printf(ifp, "%s: lp_ifflags unclean\n", __func__);
@@ -942,8 +923,8 @@
 	struct lagg_softc *sc;
 	struct lagg_port *lp;
 	struct ifnet *lpifp;
-	struct rm_priotracker tracker;
 	uint64_t newval, oldval, vsum;
+	struct rm_priotracker tracker;
 
 	/* Revise this when we've got non-generic counters. */
 	KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
@@ -1158,9 +1139,7 @@
 		}
 
 		LAGG_XLOCK(sc);
-		LAGG_WLOCK(sc);
 		lagg_proto_detach(sc);
-		LAGG_UNLOCK_ASSERT(sc);
 		lagg_proto_attach(sc, ra->ra_proto);
 		LAGG_XUNLOCK(sc);
 		break;
@@ -1430,12 +1409,12 @@
 		break;
 	case SIOCADDMULTI:
 	case SIOCDELMULTI:
-		LAGG_WLOCK(sc);
+		LAGG_XLOCK(sc);
 		SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) {
 			lagg_clrmulti(lp);
 			lagg_setmulti(lp);
 		}
-		LAGG_WUNLOCK(sc);
+		LAGG_XUNLOCK(sc);
 		error = 0;
 		break;
 	case SIOCSIFMEDIA:
@@ -1523,7 +1502,7 @@
 	struct ifmultiaddr *ifma;
 	int error;
 
-	LAGG_WLOCK_ASSERT(sc);
+	LAGG_XLOCK_ASSERT(sc);
 	IF_ADDR_WLOCK(scifp);
 	TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) {
 		if (ifma->ifma_addr->sa_family != AF_LINK)
@@ -1554,7 +1533,7 @@
 {
 	struct lagg_mc *mc;
 
-	LAGG_WLOCK_ASSERT(lp->lp_softc);
+	LAGG_XLOCK_ASSERT(lp->lp_softc);
 	while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) {
 		SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries);
 		if (mc->mc_ifma && lp->lp_detaching == 0)
@@ -2014,8 +1993,8 @@
 {
 	struct lagg_lb *lb;
 
+	LAGG_XLOCK_ASSERT(sc);
 	lb = (struct lagg_lb *)sc->sc_psc;
-	LAGG_WUNLOCK(sc);
 	if (lb != NULL)
 		free(lb, M_DEVBUF);
 }
@@ -2114,12 +2093,12 @@
 	struct lagg_port *lp;
 	void *psc;
 
+	LAGG_XLOCK_ASSERT(sc);
 	SLIST_FOREACH(lp, &sc->sc_ports, lp_entries)
 		lacp_port_destroy(lp);
 
 	psc = sc->sc_psc;
 	sc->sc_psc = NULL;
-	LAGG_WUNLOCK(sc);
 
 	lacp_detach(psc);
 }