does virtio-pci support more queues?

[Zhenwei Pi]

Hi,

I did a very simple test :
1, modify qemu/hw/block/virtio-blk.c, let virtio-blk use 256 queue:
*s->vq = virtio_add_queue(vdev, 128, virtio_blk_handle_output);*
--> *s->vq = virtio_add_queue(vdev, 256 , virtio_blk_handle_output);*

**
2, compile qemu and run a VM which boots from virtio-blk disk.

The VM can NOT boot because seabios use macro MAX_QUEUE_NUM as 128.
So, does virtio-pci support more queues?
Looking forward your reply!

[Stefan Hajnoczi]

On Wed, Apr 19, 2017 at 2:08 PM, Zhenwei Pi  wrote:
> I did a very simple test :
> 1, modify qemu/hw/block/virtio-blk.c, let virtio-blk use 256 queue:
> s->vq = virtio_add_queue(vdev, 128, virtio_blk_handle_output);
> --> s->vq = virtio_add_queue(vdev, 256 , virtio_blk_handle_output);

This change does *not* make virtio-blk use 256 queues.  It increases
the descriptor ring size for the one and only virtqueue.

You are using an old version of QEMU.  Recent QEMU releases support
virtio-blk multiqueue using the -device virtio-blk-pci,num-queues=N
parameter.

Keep in mind that QEMU's block layer does not support true multiqueue
yet.  I/O requests from all virtqueues in a device are processed by
just one thread inside QEMU.  There is work in progress to remove this
limitation and support true multiqueue block layer in QEMU.

Regarding SeaBIOS, recent SeaBIOS versions that support VIRTIO 1.0
clamp descriptor ring size to 128.  The BIOS typically does not
perform a lot of parallel I/O requests so this doesn't matter.  When
the guest OS boots up its drivers will reset the virtio-blk device and
can use the full descriptor ring size offered by the hypervisor.
Therefore you don't need to worry about SeaBIOS.  The default machine
type in recent QEMU releases automatically enables VIRTIO 1.0.