Re: [Qemu-devel] segfault use VRingMemoryRegionCaches for avail and used ring vs num-queues

[Paolo Bonzini]

On 27/02/2017 16:37, Cornelia Huck wrote:
> With the following applied (probably whitespace damaged), my guest
> starts:
> 
> diff --git a/hw/virtio/virtio.c b/hw/virtio/virtio.c
> index e487e36..28906e5 100644
> --- a/hw/virtio/virtio.c
> +++ b/hw/virtio/virtio.c
> @@ -287,6 +287,9 @@ static inline void vring_set_avail_event(VirtQueue *vq, 
> uint16_t val)
>  void virtio_queue_set_notification(VirtQueue *vq, int enable)
>  {
>      vq->notification = enable;
> +    if (!vq->vring.desc) {
> +        return;
> +    }
>  
>      rcu_read_lock();
>      if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) {
> 
> Maybe introduction of caches just exposed bugs that were already there
> (trying to muck with vring state for queues that have not been setup?)

Yes, it did.  I had caught a few while writing the patches, but it does
feel like whack-a-mole...

Paolo

> Should we stick some asserts into the respective functions to help
> flush out the remaining bugs?