----- Original Message ----- > From: "jasowang" <jasowang@xxxxxxxxxx> > To: "Guo Zhi" <qtxuning1999@xxxxxxxxxxx>, "eperezma" <eperezma@xxxxxxxxxx>, "sgarzare" <sgarzare@xxxxxxxxxx>, "Michael > Tsirkin" <mst@xxxxxxxxxx> > Cc: "netdev" <netdev@xxxxxxxxxxxxxxx>, "linux-kernel" <linux-kernel@xxxxxxxxxxxxxxx>, "kvm list" <kvm@xxxxxxxxxxxxxxx>, > "virtualization" <virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx> > Sent: Thursday, August 25, 2022 3:08:58 PM > Subject: Re: [RFC v2 3/7] vsock: batch buffers in tx > 在 2022/8/17 21:57, Guo Zhi 写道: >> Vsock uses buffers in order, and for tx driver doesn't have to >> know the length of the buffer. So we can do a batch for vsock if >> in order negotiated, only write one used ring for a batch of buffers >> >> Signed-off-by: Guo Zhi <qtxuning1999@xxxxxxxxxxx> >> --- >> drivers/vhost/vsock.c | 9 ++++++++- >> 1 file changed, 8 insertions(+), 1 deletion(-) >> >> diff --git a/drivers/vhost/vsock.c b/drivers/vhost/vsock.c >> index 368330417bde..b0108009c39a 100644 >> --- a/drivers/vhost/vsock.c >> +++ b/drivers/vhost/vsock.c >> @@ -500,6 +500,7 @@ static void vhost_vsock_handle_tx_kick(struct vhost_work >> *work) >> int head, pkts = 0, total_len = 0; >> unsigned int out, in; >> bool added = false; >> + int last_head = -1; >> >> mutex_lock(&vq->mutex); >> >> @@ -551,10 +552,16 @@ static void vhost_vsock_handle_tx_kick(struct vhost_work >> *work) >> else >> virtio_transport_free_pkt(pkt); >> >> - vhost_add_used(vq, head, 0); >> + if (!vhost_has_feature(vq, VIRTIO_F_IN_ORDER)) >> + vhost_add_used(vq, head, 0); >> + else >> + last_head = head; >> added = true; >> } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len))); >> >> + /* If in order feature negotiaged, we can do a batch to increase performance >> */ >> + if (vhost_has_feature(vq, VIRTIO_F_IN_ORDER) && last_head != -1) >> + vhost_add_used(vq, last_head, 0); > > > I may miss something but spec said "The device then skips forward in the > ring according to the size of the batch. ". > > I don't see how it is done here. > > Thanks > It can skip them in __vhost_add_used_n if _F_IN_ORDER is negotiated. last_used_idx will be added by size of the batch. > >> no_more_replies: >> if (added) >> vhost_signal(&vsock->dev, vq);
