On Mon, May 16, 2016 at 01:48:01AM +0000, Tan, Jianfeng wrote:

quoted

On Fri, May 13, 2016 at 09:54:33AM +0800, Tan, Jianfeng wrote:

quoted

    So, I'd suggest something like following:

            if (is_vdev(..)) {


The blocker issue of your suggestion is that we have no such condition.

Previously, I use dev_type, but as David's comment said:

That's not the only option. There should be others, for example,
checking the existence of virtio_user_device. Or even, you could
add a new flag inside virtio hw while initiating your vdev.

quoted

    May I ask how many more such handling are needed, excluding the tx

queue

quoted

    header desc setup? And as stated, in generic, yes, we should try that.


Those which need special handling:
(1) vq->vq_ring_mem: it is set but never used, so it's out of question.
(2) vq->virtio_net_hdr_mem and vring_hdr_desc_init

vring_hdr_desc_init is common.

quoted

(3) vq->offset

Just (2) and (3) so far. And the question is quite clear: where to put these
two special handling.

Apparently, you can't put it into the queue_setup(). And I still think
my proposal works great here.

OK, since it's indeed inappropriate to put these special handlings inside queue_setup() from semantic perspective, I'll add them according to if hw->vdev_private == NULL in the driver.

I'm thinking maybe we could rename "vdev_private" to "virtio_user_dev"
(or something like that), to make it explicit that it's for virtio user
device. I mean, there should be no other vdevs after all. OTOH, using
"hw->vdev_private != NULL" to say it's a virtio-user device is a bit
weird; it doesn't even make too much sense.

	--yliu