From: Cheng Jiang <cheng1.jiang@intel.com>
To: maxime.coquelin@redhat.com, chenbo.xia@intel.com
Cc: dev@dpdk.org, jiayu.hu@intel.com, xuan.ding@intel.com,
wenwux.ma@intel.com, yuanx.wang@intel.com,
yvonnex.yang@intel.com, xingguang.he@intel.com,
Cheng Jiang <cheng1.jiang@intel.com>
Subject: [RFC 2/3] vhost: add batch enqueue in async vhost packed ring
Date: Fri, 16 Dec 2022 07:38:43 +0000 [thread overview]
Message-ID: <20221216073844.24448-3-cheng1.jiang@intel.com> (raw)
In-Reply-To: <20221216073844.24448-1-cheng1.jiang@intel.com>
Add batch enqueue function in asynchronous vhost packed ring to
improve the performance. Chained mbufs are not supported, it will
be handled in single enqueue function.
Signed-off-by: Cheng Jiang <cheng1.jiang@intel.com>
---
lib/vhost/virtio_net.c | 159 ++++++++++++++++++++++++++++++++++++++++-
1 file changed, 157 insertions(+), 2 deletions(-)
diff --git a/lib/vhost/virtio_net.c b/lib/vhost/virtio_net.c
index 22f97d4f77..b87405ba54 100644
--- a/lib/vhost/virtio_net.c
+++ b/lib/vhost/virtio_net.c
@@ -432,6 +432,24 @@ vhost_flush_enqueue_batch_packed(struct virtio_net *dev,
vq_inc_last_used_packed(vq, PACKED_BATCH_SIZE);
}
+static __rte_always_inline void
+vhost_async_shadow_enqueue_packed_batch(struct vhost_virtqueue *vq,
+ uint64_t *lens,
+ uint16_t *ids)
+{
+ uint16_t i;
+ struct vhost_async *async = vq->async;
+
+ vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
+ async->buffers_packed[async->buffer_idx_packed].id = ids[i];
+ async->buffers_packed[async->buffer_idx_packed].len = lens[i];
+ async->buffers_packed[async->buffer_idx_packed].count = 1;
+ async->buffer_idx_packed++;
+ if (async->buffer_idx_packed >= vq->size)
+ async->buffer_idx_packed -= vq->size;
+ }
+}
+
static __rte_always_inline void
vhost_shadow_dequeue_batch_packed_inorder(struct vhost_virtqueue *vq,
uint16_t id)
@@ -1451,6 +1469,58 @@ virtio_dev_rx_sync_batch_check(struct virtio_net *dev,
return 0;
}
+static __rte_always_inline int
+virtio_dev_rx_async_batch_check(struct vhost_virtqueue *vq,
+ struct rte_mbuf **pkts,
+ uint64_t *desc_addrs,
+ uint64_t *lens,
+ int16_t dma_id,
+ uint16_t vchan_id)
+{
+ bool wrap_counter = vq->avail_wrap_counter;
+ struct vring_packed_desc *descs = vq->desc_packed;
+ uint16_t avail_idx = vq->last_avail_idx;
+ uint32_t buf_offset = sizeof(struct virtio_net_hdr_mrg_rxbuf);
+ uint16_t i;
+
+ if (unlikely(avail_idx & PACKED_BATCH_MASK))
+ return -1;
+
+ if (unlikely((avail_idx + PACKED_BATCH_SIZE) > vq->size))
+ return -1;
+
+ vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
+ if (unlikely(pkts[i]->next != NULL))
+ return -1;
+ if (unlikely(!desc_is_avail(&descs[avail_idx + i],
+ wrap_counter)))
+ return -1;
+ }
+
+ vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
+ lens[i] = descs[avail_idx + i].len;
+
+ vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
+ if (unlikely(pkts[i]->pkt_len > (lens[i] - buf_offset)))
+ return -1;
+ }
+
+ vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
+ desc_addrs[i] = descs[avail_idx + i].addr;
+
+ vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
+ if (unlikely(!desc_addrs[i]))
+ return -1;
+ if (unlikely(lens[i] != descs[avail_idx + i].len))
+ return -1;
+ }
+
+ if (rte_dma_burst_capacity(dma_id, vchan_id) < PACKED_BATCH_SIZE)
+ return -1;
+
+ return 0;
+}
+
static __rte_always_inline void
virtio_dev_rx_batch_packed_copy(struct virtio_net *dev,
struct vhost_virtqueue *vq,
@@ -1850,6 +1920,78 @@ virtio_dev_rx_async_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
return 0;
}
+static __rte_always_inline void
+virtio_dev_rx_async_packed_batch_enqueue(struct virtio_net *dev,
+ struct vhost_virtqueue *vq,
+ struct rte_mbuf **pkts,
+ uint64_t *desc_addrs,
+ uint64_t *lens)
+{
+ uint32_t buf_offset = sizeof(struct virtio_net_hdr_mrg_rxbuf);
+ struct virtio_net_hdr_mrg_rxbuf *hdrs[PACKED_BATCH_SIZE];
+ struct vring_packed_desc *descs = vq->desc_packed;
+ struct vhost_async *async = vq->async;
+ uint16_t avail_idx = vq->last_avail_idx;
+ uint32_t mbuf_offset = 0;
+ uint16_t ids[PACKED_BATCH_SIZE];
+ uint64_t mapped_len[PACKED_BATCH_SIZE];
+ void *host_iova[PACKED_BATCH_SIZE];
+ uintptr_t desc;
+ uint16_t i;
+
+ vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
+ rte_prefetch0((void *)(uintptr_t)desc_addrs[i]);
+ desc = vhost_iova_to_vva(dev, vq, desc_addrs[i], &lens[i], VHOST_ACCESS_RW);
+ hdrs[i] = (struct virtio_net_hdr_mrg_rxbuf *)(uintptr_t)desc;
+ lens[i] = pkts[i]->pkt_len +
+ sizeof(struct virtio_net_hdr_mrg_rxbuf);
+ }
+
+ vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
+ virtio_enqueue_offload(pkts[i], &hdrs[i]->hdr);
+
+ vq_inc_last_avail_packed(vq, PACKED_BATCH_SIZE);
+
+ vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
+ host_iova[i] = (void *)(uintptr_t)gpa_to_first_hpa(dev,
+ desc_addrs[i] + buf_offset, lens[i], &mapped_len[i]);
+ }
+
+ vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
+ async_iter_initialize(dev, async);
+ async_iter_add_iovec(dev, async,
+ (void *)(uintptr_t)rte_pktmbuf_iova_offset(pkts[i], mbuf_offset),
+ host_iova[i],
+ mapped_len[i]);
+ async->iter_idx++;
+ }
+
+ vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
+ vhost_log_cache_write_iova(dev, vq, descs[avail_idx + i].addr, lens[i]);
+
+ vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
+ ids[i] = descs[avail_idx + i].id;
+
+ vhost_async_shadow_enqueue_packed_batch(vq, lens, ids);
+}
+
+static __rte_always_inline int
+virtio_dev_rx_async_packed_batch(struct virtio_net *dev,
+ struct vhost_virtqueue *vq,
+ struct rte_mbuf **pkts,
+ int16_t dma_id, uint16_t vchan_id)
+{
+ uint64_t desc_addrs[PACKED_BATCH_SIZE];
+ uint64_t lens[PACKED_BATCH_SIZE];
+
+ if (virtio_dev_rx_async_batch_check(vq, pkts, desc_addrs, lens, dma_id, vchan_id) == -1)
+ return -1;
+
+ virtio_dev_rx_async_packed_batch_enqueue(dev, vq, pkts, desc_addrs, lens);
+
+ return 0;
+}
+
static __rte_always_inline void
dma_error_handler_packed(struct vhost_virtqueue *vq, uint16_t slot_idx,
uint32_t nr_err, uint32_t *pkt_idx)
@@ -1885,7 +2027,6 @@ virtio_dev_rx_async_submit_packed(struct virtio_net *dev, struct vhost_virtqueue
struct rte_mbuf **pkts, uint32_t count, int16_t dma_id, uint16_t vchan_id)
{
uint32_t pkt_idx = 0;
- uint32_t remained = count;
uint16_t n_xfer;
uint16_t num_buffers;
uint16_t num_descs;
@@ -1894,10 +2035,25 @@ virtio_dev_rx_async_submit_packed(struct virtio_net *dev, struct vhost_virtqueue
struct async_inflight_info *pkts_info = async->pkts_info;
uint32_t pkt_err = 0;
uint16_t slot_idx = 0;
+ uint16_t i;
do {
rte_prefetch0(&vq->desc_packed[vq->last_avail_idx]);
+ if (count - pkt_idx >= PACKED_BATCH_SIZE) {
+ if (!virtio_dev_rx_async_packed_batch(dev, vq, &pkts[pkt_idx],
+ dma_id, vchan_id)) {
+ for (i = 0; i < PACKED_BATCH_SIZE; i++) {
+ slot_idx = (async->pkts_idx + pkt_idx) % vq->size;
+ pkts_info[slot_idx].descs = 1;
+ pkts_info[slot_idx].nr_buffers = 1;
+ pkts_info[slot_idx].mbuf = pkts[pkt_idx];
+ pkt_idx++;
+ }
+ continue;
+ }
+ }
+
num_buffers = 0;
num_descs = 0;
if (unlikely(virtio_dev_rx_async_packed(dev, vq, pkts[pkt_idx],
@@ -1911,7 +2067,6 @@ virtio_dev_rx_async_submit_packed(struct virtio_net *dev, struct vhost_virtqueue
pkts_info[slot_idx].mbuf = pkts[pkt_idx];
pkt_idx++;
- remained--;
vq_inc_last_avail_packed(vq, num_descs);
} while (pkt_idx < count);
--
2.35.1
next prev parent reply other threads:[~2022-12-16 8:25 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-12-16 7:38 [RFC 0/3] Async vhost packed ring optimization Cheng Jiang
2022-12-16 7:38 ` [RFC 1/3] vhost: remove redundant copy for packed shadow used ring Cheng Jiang
2022-12-16 7:38 ` Cheng Jiang [this message]
2022-12-16 7:38 ` [RFC 3/3] vhost: add batch dequeue in async vhost packed ring Cheng Jiang
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20221216073844.24448-3-cheng1.jiang@intel.com \
--to=cheng1.jiang@intel.com \
--cc=chenbo.xia@intel.com \
--cc=dev@dpdk.org \
--cc=jiayu.hu@intel.com \
--cc=maxime.coquelin@redhat.com \
--cc=wenwux.ma@intel.com \
--cc=xingguang.he@intel.com \
--cc=xuan.ding@intel.com \
--cc=yuanx.wang@intel.com \
--cc=yvonnex.yang@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).