[dpdk-dev] [PATCH 1/3] vhost: pre update used ring for Tx and Rx

[Michael S. Tsirkin]

On Wed, Jun 01, 2016 at 06:40:41AM +0000, Xie, Huawei wrote:
> On 5/3/2016 8:42 AM, Yuanhan Liu wrote:
> > Pre update and update used ring in batch for Tx and Rx at the stage
> > while fetching all avail desc idx. This would reduce some cache misses
> > and hence, increase the performance a bit.
> >
> > Pre update would be feasible as guest driver will not start processing
> > those entries as far as we don't update "used->idx". (I'm not 100%
> > certain I don't miss anything, though).
> >
> > Cc: Michael S. Tsirkin <mst at redhat.com>
> > Signed-off-by: Yuanhan Liu <yuanhan.liu at linux.intel.com>
> > ---
> >  lib/librte_vhost/vhost_rxtx.c | 58 +++++++++++++++++++++----------------------
> >  1 file changed, 28 insertions(+), 30 deletions(-)
> >
> > diff --git a/lib/librte_vhost/vhost_rxtx.c b/lib/librte_vhost/vhost_rxtx.c
> > index c9cd1c5..2c3b810 100644
> > --- a/lib/librte_vhost/vhost_rxtx.c
> > +++ b/lib/librte_vhost/vhost_rxtx.c
> > @@ -137,7 +137,7 @@ copy_virtio_net_hdr(struct virtio_net *dev, uint64_t desc_addr,
> >  
> >  static inline int __attribute__((always_inline))
> >  copy_mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
> > -		  struct rte_mbuf *m, uint16_t desc_idx, uint32_t *copied)
> > +		  struct rte_mbuf *m, uint16_t desc_idx)
> >  {
> >  	uint32_t desc_avail, desc_offset;
> >  	uint32_t mbuf_avail, mbuf_offset;
> > @@ -161,7 +161,6 @@ copy_mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
> >  	desc_offset = dev->vhost_hlen;
> >  	desc_avail  = desc->len - dev->vhost_hlen;
> >  
> > -	*copied = rte_pktmbuf_pkt_len(m);
> >  	mbuf_avail  = rte_pktmbuf_data_len(m);
> >  	mbuf_offset = 0;
> >  	while (mbuf_avail != 0 || m->next != NULL) {
> > @@ -262,6 +261,7 @@ virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
> >  	struct vhost_virtqueue *vq;
> >  	uint16_t res_start_idx, res_end_idx;
> >  	uint16_t desc_indexes[MAX_PKT_BURST];
> > +	uint16_t used_idx;
> >  	uint32_t i;
> >  
> >  	LOG_DEBUG(VHOST_DATA, "(%d) %s\n", dev->vid, __func__);
> > @@ -285,27 +285,29 @@ virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
> >  	/* Retrieve all of the desc indexes first to avoid caching issues. */
> >  	rte_prefetch0(&vq->avail->ring[res_start_idx & (vq->size - 1)]);
> >  	for (i = 0; i < count; i++) {
> > -		desc_indexes[i] = vq->avail->ring[(res_start_idx + i) &
> > -						  (vq->size - 1)];
> > +		used_idx = (res_start_idx + i) & (vq->size - 1);
> > +		desc_indexes[i] = vq->avail->ring[used_idx];
> > +		vq->used->ring[used_idx].id = desc_indexes[i];
> > +		vq->used->ring[used_idx].len = pkts[i]->pkt_len +
> > +					       dev->vhost_hlen;
> > +		vhost_log_used_vring(dev, vq,
> > +			offsetof(struct vring_used, ring[used_idx]),
> > +			sizeof(vq->used->ring[used_idx]));
> >  	}
> >  
> >  	rte_prefetch0(&vq->desc[desc_indexes[0]]);
> >  	for (i = 0; i < count; i++) {
> >  		uint16_t desc_idx = desc_indexes[i];
> > -		uint16_t used_idx = (res_start_idx + i) & (vq->size - 1);
> > -		uint32_t copied;
> >  		int err;
> >  
> > -		err = copy_mbuf_to_desc(dev, vq, pkts[i], desc_idx, &copied);
> > -
> > -		vq->used->ring[used_idx].id = desc_idx;
> > -		if (unlikely(err))
> > +		err = copy_mbuf_to_desc(dev, vq, pkts[i], desc_idx);
> > +		if (unlikely(err)) {
> > +			used_idx = (res_start_idx + i) & (vq->size - 1);
> >  			vq->used->ring[used_idx].len = dev->vhost_hlen;
> > -		else
> > -			vq->used->ring[used_idx].len = copied + dev->vhost_hlen;
> > -		vhost_log_used_vring(dev, vq,
> > -			offsetof(struct vring_used, ring[used_idx]),
> > -			sizeof(vq->used->ring[used_idx]));
> > +			vhost_log_used_vring(dev, vq,
> > +				offsetof(struct vring_used, ring[used_idx]),
> > +				sizeof(vq->used->ring[used_idx]));
> > +		}
> >  
> >  		if (i + 1 < count)
> >  			rte_prefetch0(&vq->desc[desc_indexes[i+1]]);
> > @@ -879,6 +881,7 @@ rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
> >  	/* Prefetch available ring to retrieve head indexes. */
> >  	used_idx = vq->last_used_idx & (vq->size - 1);
> >  	rte_prefetch0(&vq->avail->ring[used_idx]);
> > +	rte_prefetch0(&vq->used->ring[used_idx]);
> >  
> >  	count = RTE_MIN(count, MAX_PKT_BURST);
> >  	count = RTE_MIN(count, free_entries);
> > @@ -887,22 +890,23 @@ rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
> >  
> >  	/* Retrieve all of the head indexes first to avoid caching issues. */
> >  	for (i = 0; i < count; i++) {
> > -		desc_indexes[i] = vq->avail->ring[(vq->last_used_idx + i) &
> > -					(vq->size - 1)];
> > +		used_idx = (vq->last_used_idx + i) & (vq->size - 1);
> > +		desc_indexes[i] = vq->avail->ring[used_idx];
> > +
> > +		vq->used->ring[used_idx].id  = desc_indexes[i];
> > +		vq->used->ring[used_idx].len = 0;
> > +		vhost_log_used_vring(dev, vq,
> > +				offsetof(struct vring_used, ring[used_idx]),
> > +				sizeof(vq->used->ring[used_idx]));
> >  	}
> >  
> >  	/* Prefetch descriptor index. */
> >  	rte_prefetch0(&vq->desc[desc_indexes[0]]);
> > -	rte_prefetch0(&vq->used->ring[vq->last_used_idx & (vq->size - 1)]);
> > -
> >  	for (i = 0; i < count; i++) {
> >  		int err;
> >  
> > -		if (likely(i + 1 < count)) {
> > +		if (likely(i + 1 < count))
> >  			rte_prefetch0(&vq->desc[desc_indexes[i + 1]]);
> > -			rte_prefetch0(&vq->used->ring[(used_idx + 1) &
> > -						      (vq->size - 1)]);
> > -		}
> >  
> >  		pkts[i] = rte_pktmbuf_alloc(mbuf_pool);
> >  		if (unlikely(pkts[i] == NULL)) {
> > @@ -916,18 +920,12 @@ rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
> >  			rte_pktmbuf_free(pkts[i]);
> >  			break;
> >  		}
> > -
> > -		used_idx = vq->last_used_idx++ & (vq->size - 1);
> > -		vq->used->ring[used_idx].id  = desc_indexes[i];
> > -		vq->used->ring[used_idx].len = 0;
> > -		vhost_log_used_vring(dev, vq,
> > -				offsetof(struct vring_used, ring[used_idx]),
> > -				sizeof(vq->used->ring[used_idx]));
> >  	}
> 
> Had tried post-updating used ring in batch,  but forget the perf change.
> 
> One optimization would be on vhost_log_used_ring.
> I have two ideas,
> a) In QEMU side, we always assume use ring will be changed. so that we
> don't need to log used ring in VHOST.
> 
> Michael: feasible in QEMU? comments on this?

To avoid breaking old QEMU, we'll need a new protocol feature bit,
but generally this sounds reasonable.

> b) We could always mark the total used ring modified rather than entry
> by entry.
> 
> >  
> >  	rte_smp_wmb();
> >  	rte_smp_rmb();
> >  	vq->used->idx += i;
> > +	vq->last_used_idx += i;
> >  	vhost_log_used_vring(dev, vq, offsetof(struct vring_used, idx),
> >  			sizeof(vq->used->idx));
> >  
>