vhost: fix async completion of multi-seg packets
authorPatrick Fu <patrick.fu@intel.com>
Tue, 21 Jul 2020 05:47:20 +0000 (13:47 +0800)
committerFerruh Yigit <ferruh.yigit@intel.com>
Tue, 21 Jul 2020 14:54:58 +0000 (16:54 +0200)
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 <patrick.fu@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
lib/librte_vhost/vhost.h
lib/librte_vhost/virtio_net.c

index 8c01cee..0f7212f 100644 (file)
@@ -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;
index 62a5965..95a0bc1 100644 (file)
@@ -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);