diff mbox

[dpdk-dev,RFC] vhost: Add indirect descriptors support to the TX path

Message ID 1468333932-26509-1-git-send-email-maxime.coquelin@redhat.com (mailing list archive)
State Superseded, archived
Delegated to: Yuanhan Liu
Headers show

Commit Message

Maxime Coquelin July 12, 2016, 2:32 p.m. UTC
Indirect descriptors are usually supported by virtio-net devices,
allowing to dispatch a large number of large requests.

When the virtio device sends a packet using indirect descriptors,
only one slot is used in the ring, even for large packets.

Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
I have a two questions regarding the implementation of this feature:

1. Should I add a check to ensure the indirect feature is supported
(i.e. the negociation succeeded) when having an indirect desc?

2. Should I check in copy_desc_to_mbuf() that we don't have a nested
indirect descriptor?

Both these sanity checks are recommended from the virtio spec, but
since it is to be done in the hot path, it may introduce some
performance penalties.

Note that the first check is not done in the Kernel vhost driver, whereas
the second one is.

Cheers,
Maxime
---
 lib/librte_vhost/vhost_rxtx.c | 31 ++++++++++++++++++++++---------
 lib/librte_vhost/virtio-net.c |  3 ++-
 2 files changed, 24 insertions(+), 10 deletions(-)

Comments

Yuanhan Liu Aug. 3, 2016, 2:03 p.m. UTC | #1
On Tue, Jul 12, 2016 at 04:32:12PM +0200, Maxime Coquelin wrote:
> Indirect descriptors are usually supported by virtio-net devices,
> allowing to dispatch a large number of large requests.
> 
> When the virtio device sends a packet using indirect descriptors,
> only one slot is used in the ring, even for large packets.
> 
> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
> ---
> I have a two questions regarding the implementation of this feature:
> 
> 1. Should I add a check to ensure the indirect feature is supported
> (i.e. the negociation succeeded) when having an indirect desc?
> 
> 2. Should I check in copy_desc_to_mbuf() that we don't have a nested
> indirect descriptor?
> 
> Both these sanity checks are recommended from the virtio spec, but
> since it is to be done in the hot path, it may introduce some
> performance penalties.
> 
> Note that the first check is not done in the Kernel vhost driver, whereas
> the second one is.

I think we could firstly follow the Linux kernel implementation.

> +		if (vq->desc[desc_indexes[i]].flags & VRING_DESC_F_INDIRECT) {
> +			desc = (struct vring_desc *)gpa_to_vva(dev,
> +					vq->desc[desc_indexes[i]].addr);
> +			rte_prefetch0(desc);
> +			size = vq->desc[desc_indexes[i]].len / sizeof(*desc);
> +			idx = 0;
> +		} else {
> +			desc = vq->desc;
> +			size = vq->size;
> +			idx = desc_indexes[i];

Why not fetching the desc here and feeding it to copy_desc_to_mbuf(),
instead of using "desc" and "idx" combination and fetching it inside
that function?

Overall, this patch looks good to me. As stated in IRC, It'd be great
if you could offer some performance data.

Thanks for the work, and apologize for the late review!

	--yliu
Maxime Coquelin Aug. 5, 2016, 12:18 p.m. UTC | #2
On 08/03/2016 04:03 PM, Yuanhan Liu wrote:
> On Tue, Jul 12, 2016 at 04:32:12PM +0200, Maxime Coquelin wrote:
>> Indirect descriptors are usually supported by virtio-net devices,
>> allowing to dispatch a large number of large requests.
>>
>> When the virtio device sends a packet using indirect descriptors,
>> only one slot is used in the ring, even for large packets.
>>
>> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
>> ---
>> I have a two questions regarding the implementation of this feature:
>>
>> 1. Should I add a check to ensure the indirect feature is supported
>> (i.e. the negociation succeeded) when having an indirect desc?
>>
>> 2. Should I check in copy_desc_to_mbuf() that we don't have a nested
>> indirect descriptor?
>>
>> Both these sanity checks are recommended from the virtio spec, but
>> since it is to be done in the hot path, it may introduce some
>> performance penalties.
>>
>> Note that the first check is not done in the Kernel vhost driver, whereas
>> the second one is.
>
> I think we could firstly follow the Linux kernel implementation.
OK, I can do that in the v2.
>
>> +		if (vq->desc[desc_indexes[i]].flags & VRING_DESC_F_INDIRECT) {
>> +			desc = (struct vring_desc *)gpa_to_vva(dev,
>> +					vq->desc[desc_indexes[i]].addr);
>> +			rte_prefetch0(desc);
>> +			size = vq->desc[desc_indexes[i]].len / sizeof(*desc);
>> +			idx = 0;
>> +		} else {
>> +			desc = vq->desc;
>> +			size = vq->size;
>> +			idx = desc_indexes[i];
>
> Why not fetching the desc here and feeding it to copy_desc_to_mbuf(),
> instead of using "desc" and "idx" combination and fetching it inside
> that function?
Because the desc we pass to copy_desc_to_mbuf() is the descriptor that 
will be used as the base later to get the next descriptors.

In case of indirect descriptors, the first descriptor of the chain is 
always the first element of the descriptors array, which is not the case 
for direct descriptors.

>
> Overall, this patch looks good to me. As stated in IRC, It'd be great
> if you could offer some performance data.
Sure.
I lacked time this week, and next week will be likely the same.
But I agree we need the perf data before it gets merged.

> Thanks for the work, and apologize for the late review!
Not a problem, you review is appreciated.

Thanks,
Maxime
Yuanhan Liu Sept. 21, 2016, 2:41 a.m. UTC | #3
On Fri, Aug 05, 2016 at 02:18:42PM +0200, Maxime Coquelin wrote:
> 
> 
> On 08/03/2016 04:03 PM, Yuanhan Liu wrote:
> >On Tue, Jul 12, 2016 at 04:32:12PM +0200, Maxime Coquelin wrote:
> >>Indirect descriptors are usually supported by virtio-net devices,
> >>allowing to dispatch a large number of large requests.
> >>
> >>When the virtio device sends a packet using indirect descriptors,
> >>only one slot is used in the ring, even for large packets.
> >>
> >>Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
> >>---
> >>I have a two questions regarding the implementation of this feature:
> >>
> >>1. Should I add a check to ensure the indirect feature is supported
> >>(i.e. the negociation succeeded) when having an indirect desc?
> >>
> >>2. Should I check in copy_desc_to_mbuf() that we don't have a nested
> >>indirect descriptor?
> >>
> >>Both these sanity checks are recommended from the virtio spec, but
> >>since it is to be done in the hot path, it may introduce some
> >>performance penalties.
> >>
> >>Note that the first check is not done in the Kernel vhost driver, whereas
> >>the second one is.
> >
> >I think we could firstly follow the Linux kernel implementation.
> OK, I can do that in the v2.

FYI, I'm waiting for it.

	--yliu
diff mbox

Patch

diff --git a/lib/librte_vhost/vhost_rxtx.c b/lib/librte_vhost/vhost_rxtx.c
index 15ca956..74cbc75 100644
--- a/lib/librte_vhost/vhost_rxtx.c
+++ b/lib/librte_vhost/vhost_rxtx.c
@@ -669,8 +669,8 @@  make_rarp_packet(struct rte_mbuf *rarp_mbuf, const struct ether_addr *mac)
 }
 
 static inline int __attribute__((always_inline))
-copy_desc_to_mbuf(struct virtio_net *dev, struct vhost_virtqueue *vq,
-		  struct rte_mbuf *m, uint16_t desc_idx,
+copy_desc_to_mbuf(struct virtio_net *dev, struct vring_desc *descs,
+		  uint16_t max_desc, struct rte_mbuf *m, uint16_t desc_idx,
 		  struct rte_mempool *mbuf_pool)
 {
 	struct vring_desc *desc;
@@ -683,7 +683,7 @@  copy_desc_to_mbuf(struct virtio_net *dev, struct vhost_virtqueue *vq,
 	/* A counter to avoid desc dead loop chain */
 	uint32_t nr_desc = 1;
 
-	desc = &vq->desc[desc_idx];
+	desc = &descs[desc_idx];
 	if (unlikely(desc->len < dev->vhost_hlen))
 		return -1;
 
@@ -698,7 +698,7 @@  copy_desc_to_mbuf(struct virtio_net *dev, struct vhost_virtqueue *vq,
 	 */
 	if (likely((desc->len == dev->vhost_hlen) &&
 		   (desc->flags & VRING_DESC_F_NEXT) != 0)) {
-		desc = &vq->desc[desc->next];
+		desc = &descs[desc->next];
 
 		desc_addr = gpa_to_vva(dev, desc->addr);
 		rte_prefetch0((void *)(uintptr_t)desc_addr);
@@ -731,10 +731,10 @@  copy_desc_to_mbuf(struct virtio_net *dev, struct vhost_virtqueue *vq,
 			if ((desc->flags & VRING_DESC_F_NEXT) == 0)
 				break;
 
-			if (unlikely(desc->next >= vq->size ||
-				     ++nr_desc >= vq->size))
+			if (unlikely(desc->next >= max_desc ||
+				     ++nr_desc >= max_desc))
 				return -1;
-			desc = &vq->desc[desc->next];
+			desc = &descs[desc->next];
 
 			desc_addr = gpa_to_vva(dev, desc->addr);
 			rte_prefetch0((void *)(uintptr_t)desc_addr);
@@ -859,19 +859,32 @@  rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
 	/* Prefetch descriptor index. */
 	rte_prefetch0(&vq->desc[desc_indexes[0]]);
 	for (i = 0; i < count; i++) {
+		struct vring_desc *desc;
+		uint16_t size, idx;
 		int err;
 
 		if (likely(i + 1 < count))
 			rte_prefetch0(&vq->desc[desc_indexes[i + 1]]);
 
+		if (vq->desc[desc_indexes[i]].flags & VRING_DESC_F_INDIRECT) {
+			desc = (struct vring_desc *)gpa_to_vva(dev,
+					vq->desc[desc_indexes[i]].addr);
+			rte_prefetch0(desc);
+			size = vq->desc[desc_indexes[i]].len / sizeof(*desc);
+			idx = 0;
+		} else {
+			desc = vq->desc;
+			size = vq->size;
+			idx = desc_indexes[i];
+		}
+
 		pkts[i] = rte_pktmbuf_alloc(mbuf_pool);
 		if (unlikely(pkts[i] == NULL)) {
 			RTE_LOG(ERR, VHOST_DATA,
 				"Failed to allocate memory for mbuf.\n");
 			break;
 		}
-		err = copy_desc_to_mbuf(dev, vq, pkts[i], desc_indexes[i],
-					mbuf_pool);
+		err = copy_desc_to_mbuf(dev, desc, size, pkts[i], idx, mbuf_pool);
 		if (unlikely(err)) {
 			rte_pktmbuf_free(pkts[i]);
 			break;
diff --git a/lib/librte_vhost/virtio-net.c b/lib/librte_vhost/virtio-net.c
index 1785695..0304812 100644
--- a/lib/librte_vhost/virtio-net.c
+++ b/lib/librte_vhost/virtio-net.c
@@ -76,7 +76,8 @@  struct virtio_net_device_ops const *notify_ops;
 				(1ULL << VIRTIO_NET_F_CSUM)    | \
 				(1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
 				(1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
-				(1ULL << VIRTIO_NET_F_GUEST_TSO6))
+				(1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
+				(1ULL << VIRTIO_RING_F_INDIRECT_DESC))
 
 static uint64_t VHOST_FEATURES = VHOST_SUPPORTED_FEATURES;