[PATCH vhost 17/17] virtio_net: sq support premapped mode

[Jason Wang]

On Thu, Feb 1, 2024 at 2:01 PM Xuan Zhuo <xuanzhuo at linux.alibaba.com> wrote:
>
> On Thu, 1 Feb 2024 13:36:46 +0800, Jason Wang <jasowang at redhat.com> wrote:
> > On Thu, Feb 1, 2024 at 11:28 AM Xuan Zhuo <xuanzhuo at linux.alibaba.com> wrote:
> > >
> > > On Wed, 31 Jan 2024 17:12:47 +0800, Jason Wang <jasowang at redhat.com> wrote:
> > > > On Tue, Jan 30, 2024 at 7:43 PM Xuan Zhuo <xuanzhuo at linux.alibaba.com> wrote:
> > > > >
> > > > > If the xsk is enabling, the xsk tx will share the send queue.
> > > > > But the xsk requires that the send queue use the premapped mode.
> > > > > So the send queue must support premapped mode.
> > > > >
> > > > > Signed-off-by: Xuan Zhuo <xuanzhuo at linux.alibaba.com>
> > > > > ---
> > > > >  drivers/net/virtio_net.c | 167 ++++++++++++++++++++++++++++++++++++++-
> > > > >  1 file changed, 163 insertions(+), 4 deletions(-)
> > > > >
> > > > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > > > > index 226ab830870e..cf0c67380b07 100644
> > > > > --- a/drivers/net/virtio_net.c
> > > > > +++ b/drivers/net/virtio_net.c
> > > > > @@ -46,6 +46,7 @@ module_param(napi_tx, bool, 0644);
> > > > >  #define VIRTIO_XDP_REDIR       BIT(1)
> > > > >
> > > > >  #define VIRTIO_XDP_FLAG        BIT(0)
> > > > > +#define VIRTIO_DMA_FLAG        BIT(1)
> > > > >
> > > > >  /* RX packet size EWMA. The average packet size is used to determine the packet
> > > > >   * buffer size when refilling RX rings. As the entire RX ring may be refilled
> > > > > @@ -140,6 +141,21 @@ struct virtnet_rq_dma {
> > > > >         u16 need_sync;
> > > > >  };
> > > > >
> > > > > +struct virtnet_sq_dma {
> > > > > +       union {
> > > > > +               struct virtnet_sq_dma *next;
> > > > > +               void *data;
> > > > > +       };
> > > > > +       dma_addr_t addr;
> > > > > +       u32 len;
> > > > > +       bool is_tail;
> > > > > +};
> > > > > +
> > > > > +struct virtnet_sq_dma_head {
> > > > > +       struct virtnet_sq_dma *free;
> > > > > +       struct virtnet_sq_dma *head;
> > > >
> > > > Any reason the head must be a pointer instead of a simple index?
> > >
> > >
> > > The head is used for kfree.
> > > Maybe I need to rename it.
> > >
> > > About the index(next) of the virtnet_sq_dma.
> > > If we use the index, the struct will be:
> > >
> > > struct virtnet_sq_dma {
> > >        dma_addr_t addr;
> > >        u32 len;
> > >
> > >        u32 next;
> > >        void *data
> > > };
> > >
> > > The size of virtnet_sq_dma is same.
> >
> > Ok.
> >
> > >
> > >
> > > >
> > > > > +};
> > > > > +
> > > > >  /* Internal representation of a send virtqueue */
> > > > >  struct send_queue {
> > > > >         /* Virtqueue associated with this send _queue */
> > > > > @@ -159,6 +175,8 @@ struct send_queue {
> > > > >
> > > > >         /* Record whether sq is in reset state. */
> > > > >         bool reset;
> > > > > +
> > > > > +       struct virtnet_sq_dma_head dmainfo;
> > > > >  };
> > > > >
> > >
> > > ....
> > >
> > > > > +
> > > > > +static int virtnet_sq_init_dma_mate(struct send_queue *sq)
> > > > > +{
> > > > > +       struct virtnet_sq_dma *d;
> > > > > +       int size, i;
> > > > > +
> > > > > +       size = virtqueue_get_vring_size(sq->vq);
> > > > > +
> > > > > +       size += MAX_SKB_FRAGS + 2;
> > > >
> > > > Is this enough for the case where an indirect descriptor is used?
> > >
> > >
> > > This is for the case, when the ring is full, the xmit_skb is called.
> > >
> > > I will add comment.
> >
> > Just to make sure we are at the same page.
> >
> > I meant, we could have more pending #sg than allocated here.
> >
> > For example, we can have up to (vring_size - 2 - MAX_SKB_FRAGS) *
> > MAX_SKB_FRAGS number of pending sgs?
> >
>
> Oh, my was wrong.
>
> But the max value a
> But shouldn't the maximum value be vring_size * (2 + MAX_SKB_FRAGS)?

This seems to be safer, yes.

>
> And for the reason above, we should allocate (vring_size + 1) * (2 + MAX_SKB_FRAGS);

Then we need to benchmark to see if it has an impact on the performance.

Thanks

>
> Thanks.
>
>
> > Thanks
> >
> > >
> > > Thanks.
> > >
> > >
> > > >
> > > > Thanks
> > > >
> > >
> >
>