From: Willem de Bruijn <willemb at google.com> Add napi for virtio-net transmit completion processing. Changes: v2 -> v3: - convert __netif_tx_trylock to __netif_tx_lock on tx napi poll ensure that the handler always cleans, to avoid deadlock - unconditionally clean in start_xmit avoid adding an unnecessary "if (use_napi)" branch - remove virtqueue_disable_cb in patch 5/5 a noop in the common event_idx based loop - document affinity_hint_set constraint v1 -> v2: - disable by default - disable unless affinity_hint_set because cache misses add up to a third higher cycle cost, e.g., in TCP_RR tests. This is not limited to the patch that enables tx completion cleaning in rx napi. - use trylock to avoid contention between tx and rx napi - keep interrupts masked during xmit_more (new patch 5/5) this improves cycles especially for multi UDP_STREAM, which does not benefit from cleaning tx completions on rx napi. - move free_old_xmit_skbs (new patch 3/5) to avoid forward declaration not changed: - deduplicate virnet_poll_tx and virtnet_poll_txclean they look similar, but have differ too much to make it worthwhile. - delay netif_wake_subqueue for more than 2 + MAX_SKB_FRAGS evaluated, but made no difference - patch 1/5 RFC -> v1: - dropped vhost interrupt moderation patch: not needed and likely expensive at light load - remove tx napi weight - always clean all tx completions - use boolean to toggle tx-napi, instead - only clean tx in rx if tx-napi is enabled - then clean tx before rx - fix: add missing braces in virtnet_freeze_down - testing: add 4KB TCP_RR + UDP test results Based on previous patchsets by Jason Wang: [RFC V7 PATCH 0/7] enable tx interrupts for virtio-net http://lkml.iu.edu/hypermail/linux/kernel/1505.3/00245.html Before commit b0c39dbdc204 ("virtio_net: don't free buffers in xmit ring") the virtio-net driver would free transmitted packets on transmission of new packets in ndo_start_xmit and, to catch the edge case when no new packet is sent, also in a timer at 10HZ. A timer can cause long stalls. VIRTIO_F_NOTIFY_ON_EMPTY avoids stalls due to low free descriptor count. It does not address a stalls due to low socket SO_SNDBUF. Increasing timer frequency decreases that stall time, but increases interrupt rate and, thus, cycle count. Currently, with no timer, packets are freed only at ndo_start_xmit. Latency of consume_skb is now unbounded. To avoid a deadlock if a sock reaches SO_SNDBUF, packets are orphaned on tx. This breaks TCP small queues. Reenable TCP small queues by removing the orphan. Instead of using a timer, convert the driver to regular tx napi. This does not have the unresolved stall issue and does not have any frequency to tune. By keeping interrupts enabled by default, napi increases tx interrupt rate. VIRTIO_F_EVENT_IDX avoids sending an interrupt if one is already unacknowledged, so makes this more feasible today. Combine that with an optimization that brings interrupt rate back in line with the existing version for most workloads: Tx completion cleaning on rx interrupts elides most explicit tx interrupts by relying on the fact that many rx interrupts fire. Tested by running {1, 10, 100} {TCP, UDP} STREAM, RR, 4K_RR benchmarks from a guest to a server on the host, on an x86_64 Haswell. The guest runs 4 vCPUs pinned to 4 cores. vhost and the test server are pinned to a core each. All results are the median of 5 runs, with variance well < 10%. Used neper (github.com/google/neper) as test process. Napi increases single stream throughput, but increases cycle cost. The optimizations bring this down. The previous patchset saw a regression with UDP_STREAM, which does not benefit from cleaning tx interrupts in rx napi. This regression is now gone for 10x, 100x. Remaining difference is higher 1x TCP_STREAM, lower 1x UDP_STREAM. The latest results are with process, rx napi and tx napi affine to the same core. All numbers are lower than the previous patchset. upstream napi TCP_STREAM: 1x: Mbps 27816 39805 Gcycles 274 285 10x: Mbps 42947 42531 Gcycles 300 296 100x: Mbps 31830 28042 Gcycles 279 269 TCP_RR Latency (us): 1x: p50 21 21 p99 27 27 Gcycles 180 167 10x: p50 40 39 p99 52 52 Gcycles 214 211 100x: p50 281 241 p99 411 337 Gcycles 218 226 TCP_RR 4K: 1x: p50 28 29 p99 34 36 Gcycles 177 167 10x: p50 70 71 p99 85 134 Gcycles 213 214 100x: p50 442 611 p99 802 785 Gcycles 237 216 UDP_STREAM: 1x: Mbps 29468 26800 Gcycles 284 293 10x: Mbps 29891 29978 Gcycles 285 312 100x: Mbps 30269 30304 Gcycles 318 316 UDP_RR: 1x: p50 19 19 p99 23 23 Gcycles 180 173 10x: p50 35 40 p99 54 64 Gcycles 245 237 100x: p50 234 286 p99 484 473 Gcycles 224 214 Note that GSO is enabled, so 4K RR still translates to one packet per request. Lower throughput at 100x vs 10x can be (at least in part) explained by looking at bytes per packet sent (nstat). It likely also explains the lower throughput of 1x for some variants. upstream: N=1 bytes/pkt=16581 N=10 bytes/pkt=61513 N=100 bytes/pkt=51558 at_rx: N=1 bytes/pkt=65204 N=10 bytes/pkt=65148 N=100 bytes/pkt=56840 Willem de Bruijn (5): virtio-net: napi helper functions virtio-net: transmit napi virtio-net: move free_old_xmit_skbs virtio-net: clean tx descriptors from rx napi virtio-net: keep tx interrupts disabled unless kick drivers/net/virtio_net.c | 193 ++++++++++++++++++++++++++++++++--------------- 1 file changed, 132 insertions(+), 61 deletions(-) -- 2.12.2.816.g2cccc81164-goog
Willem de Bruijn
2017-Apr-24 17:49 UTC
[PATCH net-next v3 1/5] virtio-net: napi helper functions
From: Willem de Bruijn <willemb at google.com> Prepare virtio-net for tx napi by converting existing napi code to use helper functions. This also deduplicates some logic. Signed-off-by: Willem de Bruijn <willemb at google.com> Signed-off-by: Jason Wang <jasowang at redhat.com> --- drivers/net/virtio_net.c | 65 ++++++++++++++++++++++++++---------------------- 1 file changed, 35 insertions(+), 30 deletions(-) diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c index 666ada6130ab..b9c1df29892c 100644 --- a/drivers/net/virtio_net.c +++ b/drivers/net/virtio_net.c @@ -239,6 +239,26 @@ static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) return p; } +static void virtqueue_napi_schedule(struct napi_struct *napi, + struct virtqueue *vq) +{ + if (napi_schedule_prep(napi)) { + virtqueue_disable_cb(vq); + __napi_schedule(napi); + } +} + +static void virtqueue_napi_complete(struct napi_struct *napi, + struct virtqueue *vq, int processed) +{ + int opaque; + + opaque = virtqueue_enable_cb_prepare(vq); + if (napi_complete_done(napi, processed) && + unlikely(virtqueue_poll(vq, opaque))) + virtqueue_napi_schedule(napi, vq); +} + static void skb_xmit_done(struct virtqueue *vq) { struct virtnet_info *vi = vq->vdev->priv; @@ -936,27 +956,20 @@ static void skb_recv_done(struct virtqueue *rvq) struct virtnet_info *vi = rvq->vdev->priv; struct receive_queue *rq = &vi->rq[vq2rxq(rvq)]; - /* Schedule NAPI, Suppress further interrupts if successful. */ - if (napi_schedule_prep(&rq->napi)) { - virtqueue_disable_cb(rvq); - __napi_schedule(&rq->napi); - } + virtqueue_napi_schedule(&rq->napi, rvq); } -static void virtnet_napi_enable(struct receive_queue *rq) +static void virtnet_napi_enable(struct virtqueue *vq, struct napi_struct *napi) { - napi_enable(&rq->napi); + napi_enable(napi); /* If all buffers were filled by other side before we napi_enabled, we - * won't get another interrupt, so process any outstanding packets - * now. virtnet_poll wants re-enable the queue, so we disable here. - * We synchronize against interrupts via NAPI_STATE_SCHED */ - if (napi_schedule_prep(&rq->napi)) { - virtqueue_disable_cb(rq->vq); - local_bh_disable(); - __napi_schedule(&rq->napi); - local_bh_enable(); - } + * won't get another interrupt, so process any outstanding packets now. + * Call local_bh_enable after to trigger softIRQ processing. + */ + local_bh_disable(); + virtqueue_napi_schedule(napi, vq); + local_bh_enable(); } static void refill_work(struct work_struct *work) @@ -971,7 +984,7 @@ static void refill_work(struct work_struct *work) napi_disable(&rq->napi); still_empty = !try_fill_recv(vi, rq, GFP_KERNEL); - virtnet_napi_enable(rq); + virtnet_napi_enable(rq->vq, &rq->napi); /* In theory, this can happen: if we don't get any buffers in * we will *never* try to fill again. @@ -1011,21 +1024,13 @@ static int virtnet_poll(struct napi_struct *napi, int budget) { struct receive_queue *rq container_of(napi, struct receive_queue, napi); - unsigned int r, received; + unsigned int received; received = virtnet_receive(rq, budget); /* Out of packets? */ - if (received < budget) { - r = virtqueue_enable_cb_prepare(rq->vq); - if (napi_complete_done(napi, received)) { - if (unlikely(virtqueue_poll(rq->vq, r)) && - napi_schedule_prep(napi)) { - virtqueue_disable_cb(rq->vq); - __napi_schedule(napi); - } - } - } + if (received < budget) + virtqueue_napi_complete(napi, rq->vq, received); return received; } @@ -1040,7 +1045,7 @@ static int virtnet_open(struct net_device *dev) /* Make sure we have some buffers: if oom use wq. */ if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) schedule_delayed_work(&vi->refill, 0); - virtnet_napi_enable(&vi->rq[i]); + virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); } return 0; @@ -1747,7 +1752,7 @@ static int virtnet_restore_up(struct virtio_device *vdev) schedule_delayed_work(&vi->refill, 0); for (i = 0; i < vi->max_queue_pairs; i++) - virtnet_napi_enable(&vi->rq[i]); + virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); } netif_device_attach(vi->dev); -- 2.12.2.816.g2cccc81164-goog
From: Willem de Bruijn <willemb at google.com> Convert virtio-net to a standard napi tx completion path. This enables better TCP pacing using TCP small queues and increases single stream throughput. The virtio-net driver currently cleans tx descriptors on transmission of new packets in ndo_start_xmit. Latency depends on new traffic, so is unbounded. To avoid deadlock when a socket reaches its snd limit, packets are orphaned on tranmission. This breaks socket backpressure, including TSQ. Napi increases the number of interrupts generated compared to the current model, which keeps interrupts disabled as long as the ring has enough free descriptors. Keep tx napi optional and disabled for now. Follow-on patches will reduce the interrupt cost. Signed-off-by: Willem de Bruijn <willemb at google.com> Signed-off-by: Jason Wang <jasowang at redhat.com> --- drivers/net/virtio_net.c | 76 ++++++++++++++++++++++++++++++++++++++++++------ 1 file changed, 67 insertions(+), 9 deletions(-) diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c index b9c1df29892c..356d18481ee4 100644 --- a/drivers/net/virtio_net.c +++ b/drivers/net/virtio_net.c @@ -33,9 +33,10 @@ static int napi_weight = NAPI_POLL_WEIGHT; module_param(napi_weight, int, 0444); -static bool csum = true, gso = true; +static bool csum = true, gso = true, napi_tx; module_param(csum, bool, 0444); module_param(gso, bool, 0444); +module_param(napi_tx, bool, 0644); /* FIXME: MTU in config. */ #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN) @@ -86,6 +87,8 @@ struct send_queue { /* Name of the send queue: output.$index */ char name[40]; + + struct napi_struct napi; }; /* Internal representation of a receive virtqueue */ @@ -262,12 +265,16 @@ static void virtqueue_napi_complete(struct napi_struct *napi, static void skb_xmit_done(struct virtqueue *vq) { struct virtnet_info *vi = vq->vdev->priv; + struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi; /* Suppress further interrupts. */ virtqueue_disable_cb(vq); - /* We were probably waiting for more output buffers. */ - netif_wake_subqueue(vi->dev, vq2txq(vq)); + if (napi->weight) + virtqueue_napi_schedule(napi, vq); + else + /* We were probably waiting for more output buffers. */ + netif_wake_subqueue(vi->dev, vq2txq(vq)); } static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx) @@ -972,6 +979,24 @@ static void virtnet_napi_enable(struct virtqueue *vq, struct napi_struct *napi) local_bh_enable(); } +static void virtnet_napi_tx_enable(struct virtnet_info *vi, + struct virtqueue *vq, + struct napi_struct *napi) +{ + if (!napi->weight) + return; + + /* Tx napi touches cachelines on the cpu handling tx interrupts. Only + * enable the feature if this is likely affine with the transmit path. + */ + if (!vi->affinity_hint_set) { + napi->weight = 0; + return; + } + + return virtnet_napi_enable(vq, napi); +} + static void refill_work(struct work_struct *work) { struct virtnet_info *vi @@ -1046,6 +1071,7 @@ static int virtnet_open(struct net_device *dev) if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) schedule_delayed_work(&vi->refill, 0); virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); + virtnet_napi_tx_enable(vi, vi->sq[i].vq, &vi->sq[i].napi); } return 0; @@ -1081,6 +1107,24 @@ static void free_old_xmit_skbs(struct send_queue *sq) u64_stats_update_end(&stats->tx_syncp); } +static int virtnet_poll_tx(struct napi_struct *napi, int budget) +{ + struct send_queue *sq = container_of(napi, struct send_queue, napi); + struct virtnet_info *vi = sq->vq->vdev->priv; + struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, vq2txq(sq->vq)); + + __netif_tx_lock(txq, raw_smp_processor_id()); + free_old_xmit_skbs(sq); + __netif_tx_unlock(txq); + + virtqueue_napi_complete(napi, sq->vq, 0); + + if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) + netif_tx_wake_queue(txq); + + return 0; +} + static int xmit_skb(struct send_queue *sq, struct sk_buff *skb) { struct virtio_net_hdr_mrg_rxbuf *hdr; @@ -1130,6 +1174,7 @@ static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) int err; struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum); bool kick = !skb->xmit_more; + bool use_napi = sq->napi.weight; /* Free up any pending old buffers before queueing new ones. */ free_old_xmit_skbs(sq); @@ -1152,8 +1197,10 @@ static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) } /* Don't wait up for transmitted skbs to be freed. */ - skb_orphan(skb); - nf_reset(skb); + if (!use_napi) { + skb_orphan(skb); + nf_reset(skb); + } /* If running out of space, stop queue to avoid getting packets that we * are then unable to transmit. @@ -1167,7 +1214,8 @@ static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) */ if (sq->vq->num_free < 2+MAX_SKB_FRAGS) { netif_stop_subqueue(dev, qnum); - if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) { + if (!use_napi && + unlikely(!virtqueue_enable_cb_delayed(sq->vq))) { /* More just got used, free them then recheck. */ free_old_xmit_skbs(sq); if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) { @@ -1371,8 +1419,10 @@ static int virtnet_close(struct net_device *dev) /* Make sure refill_work doesn't re-enable napi! */ cancel_delayed_work_sync(&vi->refill); - for (i = 0; i < vi->max_queue_pairs; i++) + for (i = 0; i < vi->max_queue_pairs; i++) { napi_disable(&vi->rq[i].napi); + napi_disable(&vi->sq[i].napi); + } return 0; } @@ -1727,8 +1777,10 @@ static void virtnet_freeze_down(struct virtio_device *vdev) cancel_delayed_work_sync(&vi->refill); if (netif_running(vi->dev)) { - for (i = 0; i < vi->max_queue_pairs; i++) + for (i = 0; i < vi->max_queue_pairs; i++) { napi_disable(&vi->rq[i].napi); + napi_disable(&vi->sq[i].napi); + } } } @@ -1751,8 +1803,11 @@ static int virtnet_restore_up(struct virtio_device *vdev) if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) schedule_delayed_work(&vi->refill, 0); - for (i = 0; i < vi->max_queue_pairs; i++) + for (i = 0; i < vi->max_queue_pairs; i++) { virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); + virtnet_napi_tx_enable(vi, vi->sq[i].vq, + &vi->sq[i].napi); + } } netif_device_attach(vi->dev); @@ -1957,6 +2012,7 @@ static void virtnet_free_queues(struct virtnet_info *vi) for (i = 0; i < vi->max_queue_pairs; i++) { napi_hash_del(&vi->rq[i].napi); netif_napi_del(&vi->rq[i].napi); + netif_napi_del(&vi->sq[i].napi); } /* We called napi_hash_del() before netif_napi_del(), @@ -2142,6 +2198,8 @@ static int virtnet_alloc_queues(struct virtnet_info *vi) vi->rq[i].pages = NULL; netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll, napi_weight); + netif_napi_add(vi->dev, &vi->sq[i].napi, virtnet_poll_tx, + napi_tx ? napi_weight : 0); sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg)); ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len); -- 2.12.2.816.g2cccc81164-goog
Willem de Bruijn
2017-Apr-24 17:49 UTC
[PATCH net-next v3 3/5] virtio-net: move free_old_xmit_skbs
From: Willem de Bruijn <willemb at google.com> An upcoming patch will call free_old_xmit_skbs indirectly from virtnet_poll. Move the function above this to avoid having to introduce a forward declaration. This is a pure move: no code changes. Signed-off-by: Willem de Bruijn <willemb at google.com> --- drivers/net/virtio_net.c | 60 ++++++++++++++++++++++++------------------------ 1 file changed, 30 insertions(+), 30 deletions(-) diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c index 356d18481ee4..4ec79e5d7a86 100644 --- a/drivers/net/virtio_net.c +++ b/drivers/net/virtio_net.c @@ -1045,6 +1045,36 @@ static int virtnet_receive(struct receive_queue *rq, int budget) return received; } +static void free_old_xmit_skbs(struct send_queue *sq) +{ + struct sk_buff *skb; + unsigned int len; + struct virtnet_info *vi = sq->vq->vdev->priv; + struct virtnet_stats *stats = this_cpu_ptr(vi->stats); + unsigned int packets = 0; + unsigned int bytes = 0; + + while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) { + pr_debug("Sent skb %p\n", skb); + + bytes += skb->len; + packets++; + + dev_kfree_skb_any(skb); + } + + /* Avoid overhead when no packets have been processed + * happens when called speculatively from start_xmit. + */ + if (!packets) + return; + + u64_stats_update_begin(&stats->tx_syncp); + stats->tx_bytes += bytes; + stats->tx_packets += packets; + u64_stats_update_end(&stats->tx_syncp); +} + static int virtnet_poll(struct napi_struct *napi, int budget) { struct receive_queue *rq @@ -1077,36 +1107,6 @@ static int virtnet_open(struct net_device *dev) return 0; } -static void free_old_xmit_skbs(struct send_queue *sq) -{ - struct sk_buff *skb; - unsigned int len; - struct virtnet_info *vi = sq->vq->vdev->priv; - struct virtnet_stats *stats = this_cpu_ptr(vi->stats); - unsigned int packets = 0; - unsigned int bytes = 0; - - while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) { - pr_debug("Sent skb %p\n", skb); - - bytes += skb->len; - packets++; - - dev_kfree_skb_any(skb); - } - - /* Avoid overhead when no packets have been processed - * happens when called speculatively from start_xmit. - */ - if (!packets) - return; - - u64_stats_update_begin(&stats->tx_syncp); - stats->tx_bytes += bytes; - stats->tx_packets += packets; - u64_stats_update_end(&stats->tx_syncp); -} - static int virtnet_poll_tx(struct napi_struct *napi, int budget) { struct send_queue *sq = container_of(napi, struct send_queue, napi); -- 2.12.2.816.g2cccc81164-goog
Willem de Bruijn
2017-Apr-24 17:49 UTC
[PATCH net-next v3 4/5] virtio-net: clean tx descriptors from rx napi
From: Willem de Bruijn <willemb at google.com> Amortize the cost of virtual interrupts by doing both rx and tx work on reception of a receive interrupt if tx napi is enabled. With VIRTIO_F_EVENT_IDX, this suppresses most explicit tx completion interrupts for bidirectional workloads. Signed-off-by: Willem de Bruijn <willemb at google.com> --- drivers/net/virtio_net.c | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+) diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c index 4ec79e5d7a86..9dd978f34c1f 100644 --- a/drivers/net/virtio_net.c +++ b/drivers/net/virtio_net.c @@ -1075,12 +1075,33 @@ static void free_old_xmit_skbs(struct send_queue *sq) u64_stats_update_end(&stats->tx_syncp); } +static void virtnet_poll_cleantx(struct receive_queue *rq) +{ + struct virtnet_info *vi = rq->vq->vdev->priv; + unsigned int index = vq2rxq(rq->vq); + struct send_queue *sq = &vi->sq[index]; + struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, index); + + if (!sq->napi.weight) + return; + + if (__netif_tx_trylock(txq)) { + free_old_xmit_skbs(sq); + __netif_tx_unlock(txq); + } + + if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) + netif_tx_wake_queue(txq); +} + static int virtnet_poll(struct napi_struct *napi, int budget) { struct receive_queue *rq container_of(napi, struct receive_queue, napi); unsigned int received; + virtnet_poll_cleantx(rq); + received = virtnet_receive(rq, budget); /* Out of packets? */ -- 2.12.2.816.g2cccc81164-goog
Willem de Bruijn
2017-Apr-24 17:49 UTC
[PATCH net-next v3 5/5] virtio-net: keep tx interrupts disabled unless kick
From: Willem de Bruijn <willemb at google.com> Tx napi mode increases the rate of transmit interrupts. Suppress some by masking interrupts while more packets are expected. The interrupts will be reenabled before the last packet is sent. This optimization reduces the througput drop with tx napi for unidirectional flows such as UDP_STREAM that do not benefit from cleaning tx completions in the the receive napi handler. Signed-off-by: Willem de Bruijn <willemb at google.com> --- drivers/net/virtio_net.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c index 9dd978f34c1f..003143835766 100644 --- a/drivers/net/virtio_net.c +++ b/drivers/net/virtio_net.c @@ -1200,6 +1200,9 @@ static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) /* Free up any pending old buffers before queueing new ones. */ free_old_xmit_skbs(sq); + if (use_napi && kick) + virtqueue_enable_cb_delayed(sq->vq); + /* timestamp packet in software */ skb_tx_timestamp(skb); -- 2.12.2.816.g2cccc81164-goog
On Mon, Apr 24, 2017 at 01:49:25PM -0400, Willem de Bruijn wrote:> From: Willem de Bruijn <willemb at google.com> > > Add napi for virtio-net transmit completion processing.Acked-by: Michael S. Tsirkin <mst at redhat.com>> Changes: > v2 -> v3: > - convert __netif_tx_trylock to __netif_tx_lock on tx napi poll > ensure that the handler always cleans, to avoid deadlock > - unconditionally clean in start_xmit > avoid adding an unnecessary "if (use_napi)" branch > - remove virtqueue_disable_cb in patch 5/5 > a noop in the common event_idx based loop > - document affinity_hint_set constraint > > v1 -> v2: > - disable by default > - disable unless affinity_hint_set > because cache misses add up to a third higher cycle cost, > e.g., in TCP_RR tests. This is not limited to the patch > that enables tx completion cleaning in rx napi. > - use trylock to avoid contention between tx and rx napi > - keep interrupts masked during xmit_more (new patch 5/5) > this improves cycles especially for multi UDP_STREAM, which > does not benefit from cleaning tx completions on rx napi. > - move free_old_xmit_skbs (new patch 3/5) > to avoid forward declaration > > not changed: > - deduplicate virnet_poll_tx and virtnet_poll_txclean > they look similar, but have differ too much to make it > worthwhile. > - delay netif_wake_subqueue for more than 2 + MAX_SKB_FRAGS > evaluated, but made no difference > - patch 1/5 > > RFC -> v1: > - dropped vhost interrupt moderation patch: > not needed and likely expensive at light load > - remove tx napi weight > - always clean all tx completions > - use boolean to toggle tx-napi, instead > - only clean tx in rx if tx-napi is enabled > - then clean tx before rx > - fix: add missing braces in virtnet_freeze_down > - testing: add 4KB TCP_RR + UDP test results > > Based on previous patchsets by Jason Wang: > > [RFC V7 PATCH 0/7] enable tx interrupts for virtio-net > http://lkml.iu.edu/hypermail/linux/kernel/1505.3/00245.html > > > Before commit b0c39dbdc204 ("virtio_net: don't free buffers in xmit > ring") the virtio-net driver would free transmitted packets on > transmission of new packets in ndo_start_xmit and, to catch the edge > case when no new packet is sent, also in a timer at 10HZ. > > A timer can cause long stalls. VIRTIO_F_NOTIFY_ON_EMPTY avoids stalls > due to low free descriptor count. It does not address a stalls due to > low socket SO_SNDBUF. Increasing timer frequency decreases that stall > time, but increases interrupt rate and, thus, cycle count. > > Currently, with no timer, packets are freed only at ndo_start_xmit. > Latency of consume_skb is now unbounded. To avoid a deadlock if a sock > reaches SO_SNDBUF, packets are orphaned on tx. This breaks TCP small > queues. > > Reenable TCP small queues by removing the orphan. Instead of using a > timer, convert the driver to regular tx napi. This does not have the > unresolved stall issue and does not have any frequency to tune. > > By keeping interrupts enabled by default, napi increases tx > interrupt rate. VIRTIO_F_EVENT_IDX avoids sending an interrupt if > one is already unacknowledged, so makes this more feasible today. > Combine that with an optimization that brings interrupt rate > back in line with the existing version for most workloads: > > Tx completion cleaning on rx interrupts elides most explicit tx > interrupts by relying on the fact that many rx interrupts fire. > > Tested by running {1, 10, 100} {TCP, UDP} STREAM, RR, 4K_RR benchmarks > from a guest to a server on the host, on an x86_64 Haswell. The guest > runs 4 vCPUs pinned to 4 cores. vhost and the test server are > pinned to a core each. > > All results are the median of 5 runs, with variance well < 10%. > Used neper (github.com/google/neper) as test process. > > Napi increases single stream throughput, but increases cycle cost. > The optimizations bring this down. The previous patchset saw a > regression with UDP_STREAM, which does not benefit from cleaning tx > interrupts in rx napi. This regression is now gone for 10x, 100x. > Remaining difference is higher 1x TCP_STREAM, lower 1x UDP_STREAM. > > The latest results are with process, rx napi and tx napi affine to > the same core. All numbers are lower than the previous patchset. > > > upstream napi > TCP_STREAM: > 1x: > Mbps 27816 39805 > Gcycles 274 285 > > 10x: > Mbps 42947 42531 > Gcycles 300 296 > > 100x: > Mbps 31830 28042 > Gcycles 279 269 > > TCP_RR Latency (us): > 1x: > p50 21 21 > p99 27 27 > Gcycles 180 167 > > 10x: > p50 40 39 > p99 52 52 > Gcycles 214 211 > > 100x: > p50 281 241 > p99 411 337 > Gcycles 218 226 > > TCP_RR 4K: > 1x: > p50 28 29 > p99 34 36 > Gcycles 177 167 > > 10x: > p50 70 71 > p99 85 134 > Gcycles 213 214 > > 100x: > p50 442 611 > p99 802 785 > Gcycles 237 216 > > UDP_STREAM: > 1x: > Mbps 29468 26800 > Gcycles 284 293 > > 10x: > Mbps 29891 29978 > Gcycles 285 312 > > 100x: > Mbps 30269 30304 > Gcycles 318 316 > > UDP_RR: > 1x: > p50 19 19 > p99 23 23 > Gcycles 180 173 > > 10x: > p50 35 40 > p99 54 64 > Gcycles 245 237 > > 100x: > p50 234 286 > p99 484 473 > Gcycles 224 214 > > Note that GSO is enabled, so 4K RR still translates to one packet > per request. > > Lower throughput at 100x vs 10x can be (at least in part) > explained by looking at bytes per packet sent (nstat). It likely > also explains the lower throughput of 1x for some variants. > > upstream: > > N=1 bytes/pkt=16581 > N=10 bytes/pkt=61513 > N=100 bytes/pkt=51558 > > at_rx: > > N=1 bytes/pkt=65204 > N=10 bytes/pkt=65148 > N=100 bytes/pkt=56840 > > Willem de Bruijn (5): > virtio-net: napi helper functions > virtio-net: transmit napi > virtio-net: move free_old_xmit_skbs > virtio-net: clean tx descriptors from rx napi > virtio-net: keep tx interrupts disabled unless kick > > drivers/net/virtio_net.c | 193 ++++++++++++++++++++++++++++++++--------------- > 1 file changed, 132 insertions(+), 61 deletions(-) > > -- > 2.12.2.816.g2cccc81164-goog
On 2017?04?25? 01:49, Willem de Bruijn wrote:> @@ -1371,8 +1419,10 @@ static int virtnet_close(struct net_device *dev) > /* Make sure refill_work doesn't re-enable napi! */ > cancel_delayed_work_sync(&vi->refill); > > - for (i = 0; i < vi->max_queue_pairs; i++) > + for (i = 0; i < vi->max_queue_pairs; i++) { > napi_disable(&vi->rq[i].napi); > + napi_disable(&vi->sq[i].napi); > + }Looks like this will wait for ever if napi_tx is false because we never enable the NAPI so we will wait for NAPI_STATE_SCHED to be cleared. Thanks