Subject: Re: NetBSD/xen network problems (need help)
To: Mike M. Volokhov <mishka@intostroy.com>
From: Manuel Bouyer <bouyer@antioche.eu.org>
List: port-xen
Date: 01/24/2006 20:51:19
--wRRV7LY7NUeQGEoC
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline

On Tue, Jan 24, 2006 at 04:38:48PM +0200, Mike M. Volokhov wrote:
> 
> Well, behaviour has been changed. I've rebuilt all kernels and
> restarted the system. After that all worked fine for some time, and
> after a hour or so the whole system (including dom0) has been frozen
> again. Network stats:
> 
> Name     Address              Ipkts Ierrs    Opkts Oerrs Colls
> bge0     00:30:48:84:cf:98    31867  3714    24943     0     0
> bge1     00:30:48:84:cf:99    23117   989    28967     0     0
> lo0                             206     0      206     0     0
> bridge0                       56841     0    58398  1312     0
> bridge1                       52077     0    52074     0     0
> xvif1.0  aa:00:00:17:e9:7f      532    28     1444     0     0
> xvif2.0  aa:00:00:51:08:e4    24353     0    29746     0     0
> xvif2.1  aa:00:00:51:08:e5    28960     0    22879     0     0
> xvif3.0  aa:00:00:08:e9:d2       90     0     1055     0     0
> xvif4.0  aa:00:00:49:06:01        4     0      954     0     0

Hum, nothing looks wrong here. Still no messages in the consoles ?

> 
> 
> Heh, and with this patch I've another panic for dom2 (that one with two
> xennet interfaces):
> 
> panic: kernel diagnostic assertion "((pa ^ (pa + m->m_pkthdr.len)) & PG_FRAME) == 0" failed: file "../../../../arch/xen/xen/if_xennet.c", line 1036

I can't see how my patch could be related to this, nor how this can
happen. This looks like memory corruption, or m_pkthdr.len being way
too large. If this happens again we'll have to put more printfs to see what
really happens.


On Tue, Jan 24, 2006 at 05:00:30PM +0200, Mike M. Volokhov wrote:
> 
> And yet another obscure with this patch. Right now I've lost connection
> with all my single-attached domains (i.e. with only one xennet, dom[134]),
> and dom0 showing the output errors gradually increased on bridge0.

OK, this I found why. when dropping packets with the wrong ether address
in if_xennet.c, the receive buffer was not recycled. After some time,
there were no RX buffer at xennet, and the transmit on the xvif stalls.

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--

--wRRV7LY7NUeQGEoC
Content-Type: text/plain; charset=us-ascii
Content-Disposition: attachment; filename=diff

Index: if_xennet.c
===================================================================
RCS file: /cvsroot/src/sys/arch/xen/xen/if_xennet.c,v
retrieving revision 1.13.2.19
diff -u -r1.13.2.19 if_xennet.c
--- if_xennet.c	22 Jan 2006 10:41:59 -0000	1.13.2.19
+++ if_xennet.c	24 Jan 2006 19:29:10 -0000
@@ -638,6 +638,7 @@
 	mmu_update_t *mmu = rx_mmu;
 	multicall_entry_t *mcl = rx_mcl;
 	struct mbuf *m;
+	void *pktp;
 
 	network_tx_buf_gc(sc);
 	ifp->if_flags &= ~IFF_OACTIVE;
@@ -660,7 +661,7 @@
 		/* XXXcl check rx->status for error */
 
                 MGETHDR(m, M_DONTWAIT, MT_DATA);
-                if (m == NULL) {
+                if (__predict_false(m == NULL)) {
 			printf("xennet: rx no mbuf\n");
 			break;
 		}
@@ -687,14 +688,12 @@
 
 		/* Do all the remapping work, and M->P updates, in one
 		 * big hypercall. */
-		if ((mcl - rx_mcl) != 0) {
-			mcl->op = __HYPERVISOR_mmu_update;
-			mcl->args[0] = (unsigned long)rx_mmu;
-			mcl->args[1] = mmu - rx_mmu;
-			mcl->args[2] = 0;
-			mcl++;
-			(void)HYPERVISOR_multicall(rx_mcl, mcl - rx_mcl);
-		}
+		mcl->op = __HYPERVISOR_mmu_update;
+		mcl->args[0] = (unsigned long)rx_mmu;
+		mcl->args[1] = mmu - rx_mmu;
+		mcl->args[2] = 0;
+		mcl++;
+		(void)HYPERVISOR_multicall(rx_mcl, mcl - rx_mcl);
 		if (0)
 		printf("page mapped at va %08lx -> %08x/%08lx\n",
 		    sc->sc_rx_bufa[rx->id].xb_rx.xbrx_va,
@@ -711,11 +710,23 @@
 		    (void *)(PTE_BASE[x86_btop
 			(sc->sc_rx_bufa[rx->id].xb_rx.xbrx_va)] & PG_FRAME)));
 
+		pktp = (void *)(sc->sc_rx_bufa[rx->id].xb_rx.xbrx_va +
+		    (rx->addr & PAGE_MASK));
+		if ((ifp->if_flags & IFF_PROMISC) == 0) {
+			struct ether_header *eh = pktp;
+			if (ETHER_IS_MULTICAST(eh->ether_dhost) == 0 &&
+			    memcmp(LLADDR(ifp->if_sadl), eh->ether_dhost,
+			    ETHER_ADDR_LEN) != 0) {
+				m_freem(m);
+				xennet_rx_push_buffer(sc, rx->id);
+				continue; /* packet not for us */
+			}
+		}
 		m->m_pkthdr.rcvif = ifp;
-		if (sc->sc_rx->req_prod != sc->sc_rx->resp_prod) {
+		if (__predict_true(
+		    sc->sc_rx->req_prod != sc->sc_rx->resp_prod)) {
 			m->m_len = m->m_pkthdr.len = rx->status;
-			MEXTADD(m, (void *)(sc->sc_rx_bufa[rx->id].xb_rx.
-			    xbrx_va + (rx->addr & PAGE_MASK)), rx->status,
+			MEXTADD(m, pktp, rx->status,
 			    M_DEVBUF, xennet_rx_mbuf_free,
 			    &sc->sc_rx_bufa[rx->id]);
 		} else {
@@ -726,14 +737,13 @@
 			 */
 			m->m_len = MHLEN;
 			m->m_pkthdr.len = 0;
-			m_copyback(m, 0, rx->status, 
-			    (caddr_t)(sc->sc_rx_bufa[rx->id].xb_rx.xbrx_va +
-			    (rx->addr & PAGE_MASK)));
+			m_copyback(m, 0, rx->status, pktp);
 			xennet_rx_push_buffer(sc, rx->id);
 			if (m->m_pkthdr.len < rx->status) {
+				/* out of memory, just drop packets */
 				ifp->if_ierrors++;
 				m_freem(m);
-				break;
+				continue;
 			}
 		}
 
Index: xennetback.c
===================================================================
RCS file: /cvsroot/src/sys/arch/xen/xen/xennetback.c,v
retrieving revision 1.4.2.7
diff -u -r1.4.2.7 xennetback.c
--- xennetback.c	20 Jan 2006 21:14:47 -0000	1.4.2.7
+++ xennetback.c	24 Jan 2006 19:29:10 -0000
@@ -496,8 +496,8 @@
 			do_event = 1;
 
 		XENPRINTF(("%s pkg size %d\n", xneti->xni_if.if_xname, txreq->size));
-		if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) !=
-		    (IFF_UP | IFF_RUNNING)) {
+		if (__predict_false((ifp->if_flags & (IFF_UP | IFF_RUNNING)) !=
+		    (IFF_UP | IFF_RUNNING))) {
 			/* interface not up, drop */
 			txresp->id = txreq->id;
 			txresp->status = NETIF_RSP_DROPPED;
@@ -506,8 +506,8 @@
 		/*
 		 * Do some sanity checks, and map the packet's page.
 		 */
-		if (txreq->size < ETHER_HDR_LEN ||
-		   txreq->size > (ETHER_MAX_LEN - ETHER_CRC_LEN)) {
+		if (__predict_false(txreq->size < ETHER_HDR_LEN ||
+		   txreq->size > (ETHER_MAX_LEN - ETHER_CRC_LEN))) {
 			printf("%s: packet size %d too big\n",
 			    ifp->if_xname, txreq->size);
 			txresp->id = txreq->id;
@@ -516,7 +516,8 @@
 			continue;
 		}
 		/* don't cross page boundaries */
-		if ((txreq->addr & PAGE_MASK) + txreq->size > PAGE_SIZE) {
+		if (__predict_false(
+		    (txreq->addr & PAGE_MASK) + txreq->size > PAGE_SIZE)) {
 			printf("%s: packet cross page boundary\n",
 			    ifp->if_xname);
 			txresp->id = txreq->id;
@@ -526,7 +527,8 @@
 		}
 
 		ma = txreq->addr  & ~PAGE_MASK;
-		if (xen_shm_map(&ma, 1, xneti->domid, &pkt, 0) != 0) {
+		if (__predict_false(
+		    xen_shm_map(&ma, 1, xneti->domid, &pkt, 0) != 0)) {
 			printf("%s: can't map packet page\n", ifp->if_xname);
 			txresp->id = txreq->id;
 			txresp->status = NETIF_RSP_ERROR;
@@ -534,6 +536,18 @@
 			continue;
 		}
 		pkt |= (txreq->addr & PAGE_MASK);
+		if ((ifp->if_flags & IFF_PROMISC) == 0) {
+			struct ether_header *eh = (void *)pkt;
+			if (ETHER_IS_MULTICAST(eh->ether_dhost) == 0 &&
+			    memcmp(LLADDR(ifp->if_sadl), eh->ether_dhost,
+			    ETHER_ADDR_LEN) != 0) {
+				/* packet not for us */
+				xen_shm_unmap(pkt, &ma, 1, xneti->domid);
+				txresp->id = txreq->id;
+				txresp->status = NETIF_RSP_OKAY;
+				continue;
+			}
+		}
 		/* get a mbuf for this packet */
 		MGETHDR(m, M_DONTWAIT, MT_DATA);
 		if (m != NULL) {

--wRRV7LY7NUeQGEoC--