On Tue, Nov 24, 2015 at 06:00:02PM +0900, Tetsuya Mukawa wrote:
> +static uint16_t
> +eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
> +{
> + struct vhost_queue *r = q;
> + uint16_t i, nb_rx = 0;
> +
> + if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
> + return 0;
> +
> + rte_atomic32_set(&r->while_queuing, 1);
> +
> + if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
> + goto out;
I'm not quite famililar with PMD driver yet, but as far as I know,
rte_eth_rx/tx_burst() does not provide any garantee on concurrent
dequeuing/queuing. If that's true, vhost pmd could (and should)
not do that, to keep the consistency.
On the other hand, rte_vhost_dequeue/enqueue_burst() already has
such support since the beginning: above check is redundant then.
However, FYI, Huawei has just (internally) proposed to remove
it, as he thinks that's application's duty.
So, in neither way, we should not do that.
--yliu
