From 47958f7cbf74dc59a22a7aec2b65cd721679ca52 Mon Sep 17 00:00:00 2001 From: Patrick Fu Date: Tue, 21 Jul 2020 13:47:20 +0800 Subject: [PATCH] vhost: fix async completion of multi-seg packets In async enqueue copy, a packet could be split into multiple copy segments. When polling the copy completion status, current async data path assumes the async device callbacks are aware of the packet boundary and return completed segments only if all segments belonging to the same packet are done. Such assumption are not generic to common async devices and may degrade the copy performance if async callbacks have to implement it in software manner. This patch adds tracking of the completed copy segments at vhost side. If async copy device reports partial completion of a packets, only vhost internal record is updated and vring status keeps unchanged until remaining segments of the packet are also finished. The async copy device is no longer necessary to care about the packet boundary. Fixes: cd6760da1076 ("vhost: introduce async enqueue for split ring") Signed-off-by: Patrick Fu Reviewed-by: Maxime Coquelin --- lib/librte_vhost/vhost.h | 3 +++ lib/librte_vhost/virtio_net.c | 27 +++++++++++++++++---------- 2 files changed, 20 insertions(+), 10 deletions(-) diff --git a/lib/librte_vhost/vhost.h b/lib/librte_vhost/vhost.h index 8c01cee42b..0f7212f888 100644 --- a/lib/librte_vhost/vhost.h +++ b/lib/librte_vhost/vhost.h @@ -46,6 +46,8 @@ #define MAX_PKT_BURST 32 +#define ASYNC_MAX_POLL_SEG 255 + #define VHOST_MAX_ASYNC_IT (MAX_PKT_BURST * 2) #define VHOST_MAX_ASYNC_VEC (BUF_VECTOR_MAX * 2) @@ -225,6 +227,7 @@ struct vhost_virtqueue { uint64_t *async_pending_info; uint16_t async_pkts_idx; uint16_t async_pkts_inflight_n; + uint16_t async_last_seg_n; /* vq async features */ bool async_inorder; diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c index 62a5965fc4..95a0bc19f8 100644 --- a/lib/librte_vhost/virtio_net.c +++ b/lib/librte_vhost/virtio_net.c @@ -1631,8 +1631,9 @@ uint16_t rte_vhost_poll_enqueue_completed(int vid, uint16_t queue_id, { struct virtio_net *dev = get_device(vid); struct vhost_virtqueue *vq; - uint16_t n_pkts_cpl, n_pkts_put = 0, n_descs = 0; + uint16_t n_segs_cpl, n_pkts_put = 0, n_descs = 0; uint16_t start_idx, pkts_idx, vq_size; + uint16_t n_inflight; uint64_t *async_pending_info; if (!dev) @@ -1649,46 +1650,52 @@ uint16_t rte_vhost_poll_enqueue_completed(int vid, uint16_t queue_id, rte_spinlock_lock(&vq->access_lock); + n_inflight = vq->async_pkts_inflight_n; pkts_idx = vq->async_pkts_idx; async_pending_info = vq->async_pending_info; vq_size = vq->size; start_idx = virtio_dev_rx_async_get_info_idx(pkts_idx, vq_size, vq->async_pkts_inflight_n); - n_pkts_cpl = - vq->async_ops.check_completed_copies(vid, queue_id, 0, count); + n_segs_cpl = vq->async_ops.check_completed_copies(vid, queue_id, + 0, ASYNC_MAX_POLL_SEG - vq->async_last_seg_n) + + vq->async_last_seg_n; rte_smp_wmb(); - while (likely(((start_idx + n_pkts_put) & (vq_size - 1)) != pkts_idx)) { + while (likely((n_pkts_put < count) && n_inflight)) { uint64_t info = async_pending_info[ (start_idx + n_pkts_put) & (vq_size - 1)]; uint64_t n_segs; n_pkts_put++; + n_inflight--; n_descs += info & ASYNC_PENDING_INFO_N_MSK; n_segs = info >> ASYNC_PENDING_INFO_N_SFT; if (n_segs) { - if (!n_pkts_cpl || n_pkts_cpl < n_segs) { + if (unlikely(n_segs_cpl < n_segs)) { n_pkts_put--; + n_inflight++; n_descs -= info & ASYNC_PENDING_INFO_N_MSK; - if (n_pkts_cpl) { + if (n_segs_cpl) { async_pending_info[ (start_idx + n_pkts_put) & (vq_size - 1)] = - ((n_segs - n_pkts_cpl) << + ((n_segs - n_segs_cpl) << ASYNC_PENDING_INFO_N_SFT) | (info & ASYNC_PENDING_INFO_N_MSK); - n_pkts_cpl = 0; + n_segs_cpl = 0; } break; } - n_pkts_cpl -= n_segs; + n_segs_cpl -= n_segs; } } + vq->async_last_seg_n = n_segs_cpl; + if (n_pkts_put) { - vq->async_pkts_inflight_n -= n_pkts_put; + vq->async_pkts_inflight_n = n_inflight; if (likely(vq->enabled && vq->access_ok)) { __atomic_add_fetch(&vq->used->idx, n_descs, __ATOMIC_RELEASE); -- 2.20.1