Hi Gary, On Tue, 16 Apr 2019, Gary Zibrat wrote:
> In virtio_alloc_vq, virtio_setup_queue is called before the queue is filled > in completely by virtio_init_vq. Yes, you are right. That bug has been there from the very start. I have committed a fix. > The reasons I suspect it is an OS bug are due to the ways I was able to fix > the issue: > In the virtual hardware I simply put a sleep before the first access and it > worked fine. > The other way I was able to fix it was by having the firmware zero the > pages the OS ends up using for the virtio queue. I guess this bug is not visible with other virtio devices (in qemu, vmm, ...) because those do not look at the virtqueues before the driver sends the first notify to the device. Maybe that would be a way you could be compatible with older openbsd versions. Cheers, Stefan
