Subject: Re: unaligned access in in_delayed_cksum()
To: Manuel Bouyer <email@example.com>
From: Jason R Thorpe <firstname.lastname@example.org>
Date: 09/29/2002 17:32:32
On Sun, Sep 29, 2002 at 06:16:51PM +0200, Manuel Bouyer wrote:
> First I have a question: are hadware-assised checksums supposed to work
> with VLAN ? I found nothing in the code that could pass ec_capenable
That really depends on the interface. I suppose we need another cap flag
For example, interfaces which do VLAN encap/decap in hardware can generally
also do the checksum offload. Other interfaces, like the Intel i8254x, can
do the checksum offload even if it's not doing the VLAN encap/decap, because
the driver tells it explicitly where in the frame to start the checksum
(actually, it looks like I need to add VLAN support to the wm driver's
checksum offload :-)
However, interfaces like the 3c905, which have a very simple (a few flag
bits only) checksum offload api won't be able to handle checksum offload
> from the parent interface to the ip stack though if_vlan. Maybe I should
> just set *4csum on all vlan interfaces too.
> I tried enabling ip4csum|tcp4csum|udp4csum on ti0 only (not the vlan,
> didn't have time).
> Almost immediatly the machine paniced with:
> CPU 0: fatal kernel trap:
> CPU 0 trap entry = 0x4 (unaligned access fault)
> CPU 0 a0 = 0xfffffc000599757b
> CPU 0 a1 = 0xd
> CPU 0 a2 = 0x2
> CPU 0 pc = 0xfffffc0000493770
> CPU 0 ra = 0xfffffc00004936f0
> CPU 0 pv = 0xffffffffffff0001
> CPU 0 curproc = 0x0
> gdb shows that pc is in in_delayed_cksum() line 814:
> 0xfffffc0000493770 is in in_delayed_cksum (/home/src/sys/arch/alpha/compile/DISCO/../../../../netinet/ip_output.c:814).
> 809 printf("in_delayed_cksum: pullup len %d off %d proto %d\n",
> 810 m->m_len, offset, ip->ip_p);
> 811 */
> 812 m_copyback(m, offset, sizeof(csum), (caddr_t) &csum);
> 813 } else
> 814 *(u_int16_t *)(mtod(m, caddr_t) + offset) = csum;
> 815 }
> 817 /*
> 818 * Determine the maximum length of the options to be inserted;
> Any idea ?
This is a little surprising -- if the interface doesn't have any addresses,
what is it doing sending out packets?
> If I enable ip4csum|tcp4csum|udp4csum on both ti0 and the associated vlans
> before assigning IP addresses, all seem to go well.
Well, we need some method of communicating flag changes like this between
the vlan interface and the parent interface.
-- Jason R. Thorpe <email@example.com>