--- a/usr/src/uts/sun4v/io/vsw.c	Thu Jul 03 12:07:41 2008 -0700
+++ b/usr/src/uts/sun4v/io/vsw.c	Thu Jul 03 13:18:44 2008 -0700
@@ -211,6 +211,13 @@
  */
 uint32_t vsw_pri_tx_nmblks = 64;
 
+/*
+ * Number of RARP packets sent to announce macaddr to the physical switch,
+ * after vsw's physical device is changed dynamically or after a guest (client
+ * vnet) is live migrated in.
+ */
+uint32_t vsw_publish_macaddr_count = 3;
+
 boolean_t vsw_hio_enabled = B_TRUE;	/* Enable/disable HybridIO */
 int vsw_hio_max_cleanup_retries = 10;	/* Max retries for HybridIO cleanp */
 int vsw_hio_cleanup_delay = 10000;	/* 10ms */

--- a/usr/src/uts/sun4v/io/vsw_ldc.c	Thu Jul 03 12:07:41 2008 -0700
+++ b/usr/src/uts/sun4v/io/vsw_ldc.c	Thu Jul 03 13:18:44 2008 -0700
@@ -205,6 +205,7 @@
 extern void vsw_hio_stop(vsw_t *vswp, vsw_ldc_t *ldcp);
 extern void vsw_process_dds_msg(vsw_t *vswp, vsw_ldc_t *ldcp, void *msg);
 extern void vsw_hio_stop_port(vsw_port_t *portp);
+extern void vsw_publish_macaddr(vsw_t *vswp, uint8_t *addr);
 
 #define	VSW_NUM_VMPOOLS		3	/* number of vio mblk pools */
 
@@ -229,6 +230,7 @@
 extern uint32_t vsw_num_mblks2;
 extern uint32_t vsw_num_mblks3;
 extern boolean_t vsw_obp_ver_proto_workaround;
+extern uint32_t vsw_publish_macaddr_count;
 
 #define	LDC_ENTER_LOCK(ldcp)	\
 				mutex_enter(&((ldcp)->ldc_cblock));\
@@ -379,6 +381,11 @@
 	 */
 	(void) vsw_init_ldcs(port);
 
+	/* announce macaddr of vnet to the physical switch */
+	if (vsw_publish_macaddr_count != 0) {	/* enabled */
+		vsw_publish_macaddr(vswp, (uint8_t *)&(port->p_macaddr));
+	}
+
 	D1(vswp, "%s: exit", __func__);
 	return (0);
 }

--- a/usr/src/uts/sun4v/io/vsw_phys.c	Thu Jul 03 12:07:41 2008 -0700
+++ b/usr/src/uts/sun4v/io/vsw_phys.c	Thu Jul 03 13:18:44 2008 -0700
@@ -51,6 +51,8 @@
 #include <sys/dlpi.h>
 #include <sys/ethernet.h>
 #include <net/if.h>
+#include <netinet/arp.h>
+#include <inet/arp.h>
 #include <sys/varargs.h>
 #include <sys/machsystm.h>
 #include <sys/modctl.h>
@@ -96,6 +98,7 @@
 void vsw_set_addrs(vsw_t *vswp);
 int vsw_get_hw_maddr(vsw_t *);
 mblk_t *vsw_tx_msg(vsw_t *, mblk_t *);
+void vsw_publish_macaddr(vsw_t *vswp, uint8_t *addr);
 
 /*
  * Tunables used in this file.
@@ -103,6 +106,7 @@
 extern int vsw_mac_open_retries;
 extern boolean_t vsw_multi_ring_enable;
 extern int vsw_mac_rx_rings;
+extern uint32_t vsw_publish_macaddr_count;
 
 /*
  * Check to see if the card supports the setting of multiple unicst
@@ -226,6 +230,13 @@
 		mutex_exit(&port->mca_lock);
 	}
 
+	/* announce macaddr of vnets to the physical switch */
+	if (vsw_publish_macaddr_count != 0) {	/* enabled */
+		for (port = plist->head; port != NULL; port = port->p_next) {
+			vsw_publish_macaddr(vswp, (uint8_t *)&port->p_macaddr);
+		}
+	}
+
 	RW_EXIT(&plist->lockrw);
 }
 
@@ -1352,3 +1363,83 @@
 
 	return (mp);
 }
+
+#define	ARH_FIXED_LEN	8    /* Length of fixed part of ARP header(see arp.h) */
+
+/*
+ * Send a gratuitous RARP packet to notify the physical switch to update its
+ * Layer2 forwarding table for the given mac address. This is done to allow the
+ * switch to quickly learn the macaddr-port association when a guest is live
+ * migrated or when vsw's physical device is changed dynamically. Any protocol
+ * packet would serve this purpose, but we choose RARP, as it allows us to
+ * accomplish this within L2 (ie, no need to specify IP addr etc in the packet)
+ * The macaddr of vnet is retained across migration. Hence, we don't need to
+ * update the arp cache of other hosts within the broadcast domain. Note that
+ * it is harmless to send these RARP packets during normal port attach of a
+ * client vnet. This can can be turned off if needed, by setting
+ * vsw_publish_macaddr_count to zero in /etc/system.
+ */
+void
+vsw_publish_macaddr(vsw_t *vswp, uint8_t *addr)
+{
+	mblk_t			*mp;
+	mblk_t			*bp;
+	struct arphdr		*arh;
+	struct	ether_header 	*ehp;
+	int			count = 0;
+	int			plen = 4;
+	uint8_t			*cp;
+
+	mp = allocb(ETHERMIN, BPRI_MED);
+	if (mp == NULL) {
+		return;
+	}
+
+	/* Initialize eth header */
+	ehp = (struct  ether_header *)mp->b_rptr;
+	bcopy(&etherbroadcastaddr, &ehp->ether_dhost, ETHERADDRL);
+	bcopy(addr, &ehp->ether_shost, ETHERADDRL);
+	ehp->ether_type = htons(ETHERTYPE_REVARP);
+
+	/* Initialize arp packet */
+	arh = (struct arphdr *)(mp->b_rptr + sizeof (struct ether_header));
+	cp = (uint8_t *)arh;
+
+	arh->ar_hrd = htons(ARPHRD_ETHER);	/* Hardware type:  ethernet */
+	arh->ar_pro = htons(ETHERTYPE_IP);	/* Protocol type:  IP */
+	arh->ar_hln = ETHERADDRL;	/* Length of hardware address:  6 */
+	arh->ar_pln = plen;		/* Length of protocol address:  4 */
+	arh->ar_op = htons(REVARP_REQUEST);	/* Opcode: REVARP Request */
+
+	cp += ARH_FIXED_LEN;
+
+	/* Sender's hardware address and protocol address */
+	bcopy(addr, cp, ETHERADDRL);
+	cp += ETHERADDRL;
+	bzero(cp, plen);	/* INADDR_ANY */
+	cp += plen;
+
+	/* Target hardware address and protocol address */
+	bcopy(addr, cp, ETHERADDRL);
+	cp += ETHERADDRL;
+	bzero(cp, plen);	/* INADDR_ANY */
+	cp += plen;
+
+	mp->b_wptr += ETHERMIN;	/* total size is 42; round up to ETHERMIN */
+
+	for (count = 0; count < vsw_publish_macaddr_count; count++) {
+
+		bp = dupmsg(mp);
+		if (bp == NULL) {
+			continue;
+		}
+
+		/* transmit the packet */
+		bp = vsw_tx_msg(vswp, bp);
+		if (bp != NULL) {
+			freemsg(bp);
+		}
+	}
+
+	freemsg(mp);
+}