A similar issue has been resolved by Herbert Xu in the kernel source (net/ipv4/udp.c) by substituting spin_lock_irq/spin_unlock_irq with spin_lock_bh/spin_unlock_bh.
1) the patch from Herbert Xu,<http://www.ussg.iu.edu/hypermail/linux/kernel/0505.3/1654.html>, claims to be fixing a problem with highmem interaction, not xen interaction. xen was involved in that mailing list thread, but only because it increased the likelihood that some udp packets were fragmented.(?!)
2) Xen appears to support spin_lock_irq{,save} and local_irq_{enable,disable}, so I don't know what the user is actually complaining about. I suppose I could boot my xen kernel and try to compile/load afs on it...
p7s0kYan7iND1.p7s
Description: S/MIME cryptographic signature
