e1000: Problem with "disable CRC stripping workaround" patch
Hi Jesse,
I just came across this:
http://www.mail-archive.com/[email protected]/msg14547.html
I'm seeing a problem with this currently under Xen's bridging
configuration.
Basically, with the patch, packets are being dropped at this point innet/bridge/br_forward.c:
---
int br_dev_queue_push_xmit(struct sk_buff *skb)
{
/* drop mtu oversized packets except gso */
if (packet_length(skb) > skb->dev->mtu && !skb_is_gso(skb)) {
kfree_skb(skb);--
What's happening that a 1500 byte packet comes in from e1000, onto the
bridge and is to be forwarded to a device whose mtu is 1500. Because the CRC hasn't been stripped, skb->len is 1504 and the packet is dropped.
One option is to fix this specific problem is to subtract the CRC
length from skb->len in e1000, another is to raise the MTU on the receive side of Xen's loopback interface. I've attached a patch for the latter, but I've no real opinion on which is more correct.
Cheers,
Mark.
xen/netback/loopback.c renamed from xen/netback/loopback.c.jumbo-mtu-on-vif +10 -10
@@ -161,16 +161,6 @@ NETIF_F_IP_CSUM);SET_ETHTOOL_OPS (dev, &network_ethtool_ops); - -/* -* We do not set a jumbo MTU on the interface. Otherwise the network -* stack will try to send large packets that will get dropped by the -* Ethernet bridge (unless the physical Ethernet interface is -* configured to transfer jumbo packets). If a larger MTU is desired -* then the system administrator can specify it using the 'ifconfig' -* command. -*/ -/*dev->mtu = 16*1024;*/ }static int __init make_loopback (int i)
@@ -193,6 +183,16 @@loopback_construct (dev2, dev1);/* +* We only set a jumbo MTU on vif0.0. Otherwise the network +* stack will try to send large packets that will get dropped by the +* Ethernet bridge (unless the physical Ethernet interface is +* configured to transfer jumbo packets). If a larger MTU is desired +* then the system administrator can specify it using the 'ifconfig' +* command. +*/ + dev1->mtu =16 *1024 ; + +/* * Initialise a dummy MAC address for the 'dummy backend' interface. We * choose the numerically largest non-broadcast address to prevent the * address getting stolen by an Ethernet bridge for STP purposes.
Quoted 5 lines · Mark McLoughlin
Hi Jesse,
I just came across this:
Quoted 8 lines
I'm seeing a problem with this currently under Xen's bridging
configuration.One option is to fix this specific problem is to subtract the CRC
length fromskb->lenin e1000, another is to raise the MTU on the receive side of Xen's loopback interface. I've attached a patch for the latter, but I've no real opinion on which is more correct.
We were sort of expecting this and sent the following patch upstream already - it's already queued for 2.6.18-rc3:
Please give it a try and let us know if it fixes the issue for you, it should be much better than patching xen's code.
Cheers,
Auke
Quoted 3 lines · Auke Kok
Please give it a try and let us know if it fixes the issue for you, it should be much better than patching xen's code.
I don't really care on way or another whether we get the FCS. However, the important thing to note here for Xen is that MTU != MRU. Just because you've set the MTU to 1500 does not imply that you won't receive packets > 1500.
Cheers,
Signature · 4 lines
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txtHi,
Quoted 14 lines · Auke Kok
Quoted 5 lines · Mark McLoughlin
Hi Jesse,
I just came across this:Quoted 7 lines
I'm seeing a problem with this currently under Xen's bridging
configuration.One option is to fix this specific problem is to subtract the CRC
length fromskb->lenin e1000, another is to raise the MTU on the receive side of Xen's loopback interface. I've attached a patch for the latter, but I've no real opinion on which is more correct.
Quoted 7 lines
We were sort of expecting this and sent the following patch upstream already - it's already queued for 2.6.18-rc3:
Please give it a try and let us know if it fixes the issue for you, it should be much better than patching xen's code.
Yep, this fixes the problem for me.
Thanks,
Mark.