Skip to contentSkip to conversation

fix unaligned access of netfron/back

PATCH In review Akio Takebe 2 participants 2 messages Oct 17, 2006 – Nov 9, 2006

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 fix unaligned access of netfron/back

Hi,

We make a patch to avoid the following warnings.
The noisy warnings are happend
only when we enabele network of domU on xen/ia64.

kernel unaligned access to 0xe000000019247a1e, ip=0xa00000010093d681 kernel unaligned access to 0xe000000019247a1e, ip=0xa00000010093d751 kernel unaligned access to 0xe000000019247a1e, ip=0xa00000010093d800 kernel unaligned access to 0xe000000019247a2e, ip=0xa00000010093da10 kernel unaligned access to 0xe000000019247a1e, ip=0xa00000010093da90

Best Regards,

Akio Takebe

# HG changeset patch
# User root@tiger154
# Date 1161050883 -32400
# Node ID 77c6472e23c54dd44b832b29bfbcb622ca57cdb2
# Parent fcd746cf4647e06b8e88e620c29610ba43e3ad7c
This patche fix unaligned access when we enable network by using netfront. This problem is only happened on Xen/ia64.

linux-2.6-xen-sparse/drivers/xen/netback/netback.c +3 -3
diff -r fcd746cf4647 -r 77c6472e23c5 linux-2.6-xen-sparse/drivers/xen/netback/netback.c
@@ -187,7 +187,7 @@ static struct sk_buff *netbk_copy_skb(st
 	if (unlikely(!nskb))
 		goto err;
 
-	skb_reserve(nskb, 16);
+	skb_reserve(nskb, 16 + NET_IP_ALIGN);
 	headlen = nskb->end - nskb->data;
 	if (headlen > skb_headlen(skb))
 		headlen = skb_headlen(skb);
@@ -1210,7 +1210,7 @@ static void net_tx_action(unsigned long 
 			    ret < MAX_SKB_FRAGS) ?
 			PKT_PROT_LEN : txreq.size;
 
-		skb = alloc_skb(data_len+16, GFP_ATOMIC);
+		skb = alloc_skb(data_len + 16 + NET_IP_ALIGN, GFP_ATOMIC);
 		if (unlikely(skb == NULL)) {
 			DPRINTK("Can't allocate a skb in start_xmit.\n");
 			netbk_tx_err(netif, &txreq, i);
@@ -1218,7 +1218,7 @@ static void net_tx_action(unsigned long 
 		}
 
 		/* Packets passed to netif_rx() must have some headroom. */
-		skb_reserve(skb, 16);
+		skb_reserve(skb, 16 + NET_IP_ALIGN);
 
 		if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
 			struct netif_extra_info *gso;
linux-2.6-xen-sparse/drivers/xen/netfront/netfront.c +2 -2
diff -r fcd746cf4647 -r 77c6472e23c5 linux-2.6-xen-sparse/drivers/xen/netfront/netfront.c
@@ -684,7 +684,7 @@ static void network_alloc_rx_buffers(str
 		 * necessary here.
 		 * 16 bytes added as necessary headroom for netif_receive_skb.
 		 */
-		skb = alloc_skb(RX_COPY_THRESHOLD + 16,
+		skb = alloc_skb(RX_COPY_THRESHOLD + 16 + NET_IP_ALIGN,
 				GFP_ATOMIC | __GFP_NOWARN);
 		if (unlikely(!skb))
 			goto no_skb;
@@ -702,7 +702,7 @@ no_skb:
 			break;
 		}
 
-		skb_reserve(skb, 16); /* mimic dev_alloc_skb() */
+		skb_reserve(skb, 16 + NET_IP_ALIGN); /* mimic dev_alloc_skb() */
 		skb_shinfo(skb)->frags[0].page = page;
 		skb_shinfo(skb)->nr_frags = 1;
 		__skb_queue_tail(&np->rx_batch, skb);

_______________________________________________
Xen-ia64-devel mailing list
[email protected]
http://lists.xensource.com/xen-ia64-devel

#1 lore raw

LSM'07

Hi.

I'm working on Libre Software Meeting '07 (RMLL), which may occur in july, and will be in the town of Amiens, in France as usual.

I am the leader of the conferences cycle about clustering and virtualization.

For this topic, I would have one or two persons able to make a keynote about XEN, and would be able to organize a meeting between some of the open source actors of clustering & virtualization.

Would you be interested in ?
Will you be able to travel to Amiens in France for July ?

Thanks in advance.

PS: i'm a regular reader of the mailing list.

Signature · 3 lines
Il vaut mieux pomper même s'il ne se passe rien, que de risquer quelque chose de bien pire en ne pompant pas. -- Devise Shadok.

Gaétan RYCKEBOER