virtio: recover from vector assignment failure

This patch series fixes a regression reported by Amit Shah (see Subject: Re: qemu-kvm missing some msix capability check,
Message-Id: [email protected])

Since Amit's testing, I have refactored the change slightly to make the patch smaller.

Rusty, can this go in for 2.6.31, or do you think it's too big a change? As an alternative work-around, we possibly could do something in qemu in host to make it less likely that mapping vq to vector fails - but it seems that we still want the guest to handle the error as gracefully as possible, right?

MST


Michael S. Tsirkin (2):
  virtio: delete vq from list
  virtio: retry on vector assignment failure

 drivers/virtio/virtio_pci.c |  136 ++++++++++++++++++++++++-------------------
 1 files changed, 76 insertions(+), 60 deletions(-)