Linux Virtualization - Jul 2014 - [PATCH net-next V2 3/3] virtio-net: rx busy polling support

Hi all:

This series introduces the support for rx busy polling support. This
was useful for reduing the latency for a kvm guest. Patch 1-2
introduces helpers which is used for rx busy polling. Patch 3
implement the main function.

Test was done between a kvm guest and an external host. Two hosts were
connected through 40gb mlx4 cards. With both busy_poll and busy_read are
set to 50 in guest, 1 byte netperf tcp_rr shows 116% improvement:
transaction rate was increased from 9151.94 to 19787.37.

Changes from V1:
- split the patch info smaller ones
- add more details about test setup/configuration

Please review.

Jason Wang (3):
  virtio-net: introduce helpers to enable and disable all NAPIs
  virtio-net: introduce virtnet_receive()
  virtio-net: rx busy polling support

 drivers/net/virtio_net.c | 234 ++++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 221 insertions(+), 13 deletions(-)

-- 
1.8.3.1

This patch introduces helpers to disable and enable NAPI for all rx
queues. This will be used by rx busy polling support.

Cc: Rusty Russell <rusty at rustcorp.com.au>
Cc: Michael S. Tsirkin <mst at redhat.com>
Cc: Vlad Yasevich <vyasevic at redhat.com>
Cc: Eric Dumazet <eric.dumazet at gmail.com>
Signed-off-by: Jason Wang <jasowang at redhat.com>
---
 drivers/net/virtio_net.c | 34 ++++++++++++++++++++++++----------
 1 file changed, 24 insertions(+), 10 deletions(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 7d9f84a..e1e1691 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -760,6 +760,22 @@ again:
 	return received;
 }
 
+static void virtnet_napi_enable_all(struct virtnet_info *vi)
+{
+	int i;
+
+	for (i = 0; i < vi->max_queue_pairs; i++)
+		virtnet_napi_enable(&vi->rq[i]);
+}
+
+static void virtnet_napi_disable_all(struct virtnet_info *vi)
+{
+	int i;
+
+	for (i = 0; i < vi->max_queue_pairs; i++)
+		napi_disable(&vi->rq[i].napi);
+}
+
 static int virtnet_open(struct net_device *dev)
 {
 	struct virtnet_info *vi = netdev_priv(dev);
@@ -770,9 +786,10 @@ static int virtnet_open(struct net_device *dev)
 			/* Make sure we have some buffers: if oom use wq. */
 			if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
 				schedule_delayed_work(&vi->refill, 0);
-		virtnet_napi_enable(&vi->rq[i]);
 	}
 
+	virtnet_napi_enable_all(vi);
+
 	return 0;
 }
 
@@ -1076,13 +1093,11 @@ static int virtnet_set_queues(struct virtnet_info *vi,
u16 queue_pairs)
 static int virtnet_close(struct net_device *dev)
 {
 	struct virtnet_info *vi = netdev_priv(dev);
-	int i;
 
 	/* Make sure refill_work doesn't re-enable napi! */
 	cancel_delayed_work_sync(&vi->refill);
 
-	for (i = 0; i < vi->max_queue_pairs; i++)
-		napi_disable(&vi->rq[i].napi);
+	virtnet_napi_disable_all(vi);
 
 	return 0;
 }
@@ -1853,11 +1868,11 @@ static int virtnet_freeze(struct virtio_device *vdev)
 	netif_device_detach(vi->dev);
 	cancel_delayed_work_sync(&vi->refill);
 
-	if (netif_running(vi->dev))
-		for (i = 0; i < vi->max_queue_pairs; i++) {
-			napi_disable(&vi->rq[i].napi);
+	if (netif_running(vi->dev)) {
+		virtnet_napi_disable_all(vi);
+		for (i = 0; i < vi->max_queue_pairs; i++)
 			netif_napi_del(&vi->rq[i].napi);
-		}
+	}
 
 	remove_vq_common(vi);
 
@@ -1880,8 +1895,7 @@ static int virtnet_restore(struct virtio_device *vdev)
 			if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
 				schedule_delayed_work(&vi->refill, 0);
 
-		for (i = 0; i < vi->max_queue_pairs; i++)
-			virtnet_napi_enable(&vi->rq[i]);
+		virtnet_napi_enable_all(vi);
 	}
 
 	netif_device_attach(vi->dev);
-- 
1.8.3.1

Move common receive logic to a new helper virtnet_receive(). It will
also be used by rx busy polling method.

Cc: Rusty Russell <rusty at rustcorp.com.au>
Cc: Michael S. Tsirkin <mst at redhat.com>
Cc: Vlad Yasevich <vyasevic at redhat.com>
Cc: Eric Dumazet <eric.dumazet at gmail.com>
Signed-off-by: Jason Wang <jasowang at redhat.com>
---
 drivers/net/virtio_net.c | 20 +++++++++++++++-----
 1 file changed, 15 insertions(+), 5 deletions(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index e1e1691..e417d93 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -725,15 +725,12 @@ static void refill_work(struct work_struct *work)
 	}
 }
 
-static int virtnet_poll(struct napi_struct *napi, int budget)
+static int virtnet_receive(struct receive_queue *rq, int budget)
 {
-	struct receive_queue *rq -		container_of(napi, struct receive_queue, napi);
 	struct virtnet_info *vi = rq->vq->vdev->priv;
+	unsigned int len, received = 0;
 	void *buf;
-	unsigned int r, len, received = 0;
 
-again:
 	while (received < budget &&
 	       (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
 		receive_buf(rq, buf, len);
@@ -745,6 +742,18 @@ again:
 			schedule_delayed_work(&vi->refill, 0);
 	}
 
+	return received;
+}
+
+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 = 0;
+
+again:
+	received += virtnet_receive(rq, budget);
+
 	/* Out of packets? */
 	if (received < budget) {
 		r = virtqueue_enable_cb_prepare(rq->vq);
@@ -753,6 +762,7 @@ again:
 		    napi_schedule_prep(napi)) {
 			virtqueue_disable_cb(rq->vq);
 			__napi_schedule(napi);
+			budget -= received;
 			goto again;
 		}
 	}
-- 
1.8.3.1

Add basic support for rx busy polling.

Test was done between a kvm guest and an external host. Two hosts were
connected through 40gb mlx4 cards. With both busy_poll and busy_read
are set to 50 in guest, 1 byte netperf tcp_rr shows 116% improvement:
transaction rate was increased from 9151.94 to 19787.37.

Cc: Rusty Russell <rusty at rustcorp.com.au>
Cc: Michael S. Tsirkin <mst at redhat.com>
Cc: Vlad Yasevich <vyasevic at redhat.com>
Cc: Eric Dumazet <eric.dumazet at gmail.com>
Signed-off-by: Jason Wang <jasowang at redhat.com>
---
 drivers/net/virtio_net.c | 190 ++++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 187 insertions(+), 3 deletions(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index e417d93..4830713 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -27,6 +27,7 @@
 #include <linux/slab.h>
 #include <linux/cpu.h>
 #include <linux/average.h>
+#include <net/busy_poll.h>
 
 static int napi_weight = NAPI_POLL_WEIGHT;
 module_param(napi_weight, int, 0444);
@@ -94,8 +95,143 @@ struct receive_queue {
 
 	/* Name of this receive queue: input.$index */
 	char name[40];
+
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	unsigned int state;
+#define VIRTNET_RQ_STATE_IDLE        0
+#define VIRTNET_RQ_STATE_NAPI	     1    /* NAPI or refill owns this RQ */
+#define VIRTNET_RQ_STATE_POLL	     2    /* poll owns this RQ */
+#define VIRTNET_RQ_STATE_DISABLED    4    /* RQ is disabled */
+#define VIRTNET_RQ_OWNED (VIRTNET_RQ_STATE_NAPI | VIRTNET_RQ_STATE_POLL)
+#define VIRTNET_RQ_LOCKED (VIRTNET_RQ_OWNED | VIRTNET_RQ_STATE_DISABLED)
+#define VIRTNET_RQ_STATE_NAPI_YIELD  8    /* NAPI or refill yielded this RQ */
+#define VIRTNET_RQ_STATE_POLL_YIELD  16   /* poll yielded this RQ */
+	spinlock_t lock;
+#endif  /* CONFIG_NET_RX_BUSY_POLL */
 };
 
+#ifdef CONFIG_NET_RX_BUSY_POLL
+static inline void virtnet_rq_init_lock(struct receive_queue *rq)
+{
+
+	spin_lock_init(&rq->lock);
+	rq->state = VIRTNET_RQ_STATE_IDLE;
+}
+
+/* called from the device poll routine or refill routine to get ownership of a
+ * receive queue.
+ */
+static inline bool virtnet_rq_lock_napi_refill(struct receive_queue *rq)
+{
+	int rc = true;
+
+	spin_lock(&rq->lock);
+	if (rq->state & VIRTNET_RQ_LOCKED) {
+		WARN_ON(rq->state & VIRTNET_RQ_STATE_NAPI);
+		rq->state |= VIRTNET_RQ_STATE_NAPI_YIELD;
+		rc = false;
+	} else
+		/* we don't care if someone yielded */
+		rq->state = VIRTNET_RQ_STATE_NAPI;
+	spin_unlock(&rq->lock);
+	return rc;
+}
+
+/* returns true is someone tried to get the rq while napi or refill had it */
+static inline bool virtnet_rq_unlock_napi_refill(struct receive_queue *rq)
+{
+	int rc = false;
+
+	spin_lock(&rq->lock);
+	WARN_ON(rq->state & (VIRTNET_RQ_STATE_POLL |
+			     VIRTNET_RQ_STATE_NAPI_YIELD));
+
+	if (rq->state & VIRTNET_RQ_STATE_POLL_YIELD)
+		rc = true;
+	/* will reset state to idle, unless RQ is disabled */
+	rq->state &= VIRTNET_RQ_STATE_DISABLED;
+	spin_unlock(&rq->lock);
+	return rc;
+}
+
+/* called from virtnet_low_latency_recv() */
+static inline bool virtnet_rq_lock_poll(struct receive_queue *rq)
+{
+	int rc = true;
+
+	spin_lock_bh(&rq->lock);
+	if ((rq->state & VIRTNET_RQ_LOCKED)) {
+		rq->state |= VIRTNET_RQ_STATE_POLL_YIELD;
+		rc = false;
+	} else
+		/* preserve yield marks */
+		rq->state |= VIRTNET_RQ_STATE_POLL;
+	spin_unlock_bh(&rq->lock);
+	return rc;
+}
+
+/* returns true if someone tried to get the receive queue while it was locked
*/
+static inline bool virtnet_rq_unlock_poll(struct receive_queue *rq)
+{
+	int rc = false;
+
+	spin_lock_bh(&rq->lock);
+	WARN_ON(rq->state & (VIRTNET_RQ_STATE_NAPI));
+
+	if (rq->state & VIRTNET_RQ_STATE_POLL_YIELD)
+		rc = true;
+	/* will reset state to idle, unless RQ is disabled */
+	rq->state &= VIRTNET_RQ_STATE_DISABLED;
+	spin_unlock_bh(&rq->lock);
+	return rc;
+}
+
+/* return false if RQ is currently owned */
+static inline bool virtnet_rq_disable(struct receive_queue *rq)
+{
+	int rc = true;
+
+	spin_lock_bh(&rq->lock);
+	if (rq->state & VIRTNET_RQ_OWNED)
+		rc = false;
+	rq->state |= VIRTNET_RQ_STATE_DISABLED;
+	spin_unlock_bh(&rq->lock);
+
+	return rc;
+}
+
+#else /* CONFIG_NET_RX_BUSY_POLL */
+static inline void virtnet_rq_init_lock(struct receive_queue *rq)
+{
+}
+
+static inline bool virtnet_rq_lock_napi_refill(struct receive_queue *rq)
+{
+	return true;
+}
+
+static inline bool virtnet_rq_unlock_napi_refill(struct receive_queue *rq)
+{
+	return false;
+}
+
+static inline bool virtnet_rq_lock_poll(struct receive_queue *rq)
+{
+	return false;
+}
+
+static inline bool virtnet_rq_unlock_poll(struct receive_queue *rq)
+{
+	return false;
+}
+
+static inline bool virtnet_rq_disable(struct receive_queue *rq)
+{
+	return true;
+}
+
+#endif /* CONFIG_NET_RX_BUSY_POLL */
+
 struct virtnet_info {
 	struct virtio_device *vdev;
 	struct virtqueue *cvq;
@@ -521,6 +657,8 @@ static void receive_buf(struct receive_queue *rq, void *buf,
unsigned int len)
 		skb_shinfo(skb)->gso_segs = 0;
 	}
 
+	skb_mark_napi_id(skb, &rq->napi);
+
 	netif_receive_skb(skb);
 	return;
 
@@ -714,7 +852,12 @@ static void refill_work(struct work_struct *work)
 		struct receive_queue *rq = &vi->rq[i];
 
 		napi_disable(&rq->napi);
+		if (!virtnet_rq_lock_napi_refill(rq)) {
+			virtnet_napi_enable(rq);
+			continue;
+		}
 		still_empty = !try_fill_recv(rq, GFP_KERNEL);
+		virtnet_rq_unlock_napi_refill(rq);
 		virtnet_napi_enable(rq);
 
 		/* In theory, this can happen: if we don't get any buffers in
@@ -752,8 +895,13 @@ static int virtnet_poll(struct napi_struct *napi, int
budget)
 	unsigned int r, received = 0;
 
 again:
+	if (!virtnet_rq_lock_napi_refill(rq))
+		return budget;
+
 	received += virtnet_receive(rq, budget);
 
+	virtnet_rq_unlock_napi_refill(rq);
+
 	/* Out of packets? */
 	if (received < budget) {
 		r = virtqueue_enable_cb_prepare(rq->vq);
@@ -770,20 +918,50 @@ again:
 	return received;
 }
 
+#ifdef CONFIG_NET_RX_BUSY_POLL
+/* must be called with local_bh_disable()d */
+static int virtnet_low_latency_recv(struct napi_struct *napi)
+{
+	struct receive_queue *rq +		container_of(napi, struct receive_queue, napi);
+	struct virtnet_info *vi = rq->vq->vdev->priv;
+	int received;
+
+	if (!(vi->status & VIRTIO_NET_S_LINK_UP))
+		return LL_FLUSH_FAILED;
+
+	if (!virtnet_rq_lock_poll(rq))
+		return LL_FLUSH_BUSY;
+
+	received = virtnet_receive(rq, 4);
+
+	virtnet_rq_unlock_poll(rq);
+
+	return received;
+}
+#endif	/* CONFIG_NET_RX_BUSY_POLL */
+
 static void virtnet_napi_enable_all(struct virtnet_info *vi)
 {
 	int i;
 
-	for (i = 0; i < vi->max_queue_pairs; i++)
+	for (i = 0; i < vi->max_queue_pairs; i++) {
+		virtnet_rq_init_lock(&vi->rq[i]);
 		virtnet_napi_enable(&vi->rq[i]);
+	}
 }
 
 static void virtnet_napi_disable_all(struct virtnet_info *vi)
 {
 	int i;
 
-	for (i = 0; i < vi->max_queue_pairs; i++)
+	for (i = 0; i < vi->max_queue_pairs; i++) {
 		napi_disable(&vi->rq[i].napi);
+		while (!virtnet_rq_disable(&vi->rq[i])) {
+			pr_info("RQ %d locked\n", i);
+			usleep_range(1000, 20000);
+		}
+	}
 }
 
 static int virtnet_open(struct net_device *dev)
@@ -1372,6 +1550,9 @@ static const struct net_device_ops virtnet_netdev = {
 #ifdef CONFIG_NET_POLL_CONTROLLER
 	.ndo_poll_controller = virtnet_netpoll,
 #endif
+#ifdef CONFIG_NET_RX_BUSY_POLL
+	.ndo_busy_poll		= virtnet_low_latency_recv,
+#endif
 };
 
 static void virtnet_config_changed_work(struct work_struct *work)
@@ -1577,6 +1758,7 @@ 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);
+		napi_hash_add(&vi->rq[i].napi);
 
 		sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
 		ewma_init(&vi->rq[i].mrg_avg_pkt_len, 1, RECEIVE_AVG_WEIGHT);
@@ -1880,8 +2062,10 @@ static int virtnet_freeze(struct virtio_device *vdev)
 
 	if (netif_running(vi->dev)) {
 		virtnet_napi_disable_all(vi);
-		for (i = 0; i < vi->max_queue_pairs; i++)
+		for (i = 0; i < vi->max_queue_pairs; i++) {
+			napi_hash_del(&vi->rq[i].napi);
 			netif_napi_del(&vi->rq[i].napi);
+		}
 	}
 
 	remove_vq_common(vi);
-- 
1.8.3.1

On 07/16/2014 11:51 AM, Jason Wang wrote:
> Add basic support for rx busy polling.
>
> Test was done between a kvm guest and an external host. Two hosts were
> connected through 40gb mlx4 cards. With both busy_poll and busy_read
> are set to 50 in guest, 1 byte netperf tcp_rr shows 116% improvement:
> transaction rate was increased from 9151.94 to 19787.37.
>
> Cc: Rusty Russell <rusty at rustcorp.com.au>
> Cc: Michael S. Tsirkin <mst at redhat.com>
> Cc: Vlad Yasevich <vyasevic at redhat.com>
> Cc: Eric Dumazet <eric.dumazet at gmail.com>
> Signed-off-by: Jason Wang <jasowang at redhat.com>
> ---
>   drivers/net/virtio_net.c | 190
++++++++++++++++++++++++++++++++++++++++++++++-
>   1 file changed, 187 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> index e417d93..4830713 100644
> --- a/drivers/net/virtio_net.c
> +++ b/drivers/net/virtio_net.c
> @@ -27,6 +27,7 @@
>   #include <linux/slab.h>
>   #include <linux/cpu.h>
>   #include <linux/average.h>
> +#include <net/busy_poll.h>
>   
>   static int napi_weight = NAPI_POLL_WEIGHT;
>   module_param(napi_weight, int, 0444);
> @@ -94,8 +95,143 @@ struct receive_queue {
>   
>   	/* Name of this receive queue: input.$index */
>   	char name[40];
> +
> +#ifdef CONFIG_NET_RX_BUSY_POLL
> +	unsigned int state;
> +#define VIRTNET_RQ_STATE_IDLE        0
> +#define VIRTNET_RQ_STATE_NAPI	     1    /* NAPI or refill owns this RQ */
> +#define VIRTNET_RQ_STATE_POLL	     2    /* poll owns this RQ */
> +#define VIRTNET_RQ_STATE_DISABLED    4    /* RQ is disabled */
> +#define VIRTNET_RQ_OWNED (VIRTNET_RQ_STATE_NAPI | VIRTNET_RQ_STATE_POLL)
> +#define VIRTNET_RQ_LOCKED (VIRTNET_RQ_OWNED | VIRTNET_RQ_STATE_DISABLED)
> +#define VIRTNET_RQ_STATE_NAPI_YIELD  8    /* NAPI or refill yielded this
RQ */
> +#define VIRTNET_RQ_STATE_POLL_YIELD  16   /* poll yielded this RQ */
> +	spinlock_t lock;
> +#endif  /* CONFIG_NET_RX_BUSY_POLL */
>   };
>   
> +#ifdef CONFIG_NET_RX_BUSY_POLL
> +static inline void virtnet_rq_init_lock(struct receive_queue *rq)
> +{
> +
> +	spin_lock_init(&rq->lock);
> +	rq->state = VIRTNET_RQ_STATE_IDLE;
> +}
> +
> +/* called from the device poll routine or refill routine to get ownership
of a
> + * receive queue.
> + */
> +static inline bool virtnet_rq_lock_napi_refill(struct receive_queue *rq)
> +{
> +	int rc = true;
> +
bool instead of int...?
> +	spin_lock(&rq->lock);
> +	if (rq->state & VIRTNET_RQ_LOCKED) {
> +		WARN_ON(rq->state & VIRTNET_RQ_STATE_NAPI);
> +		rq->state |= VIRTNET_RQ_STATE_NAPI_YIELD;
> +		rc = false;
> +	} else
> +		/* we don't care if someone yielded */
> +		rq->state = VIRTNET_RQ_STATE_NAPI;
> +	spin_unlock(&rq->lock);
Lock for rq->state ...?

If yes:
spin_lock(&rq->lock);
if (rq->state & VIRTNET_RQ_LOCKED) {
	rq->state |= VIRTNET_RQ_STATE_NAPI_YIELD;
	spin_unlock(&rq->lock);
	WARN_ON(rq->state & VIRTNET_RQ_STATE_NAPI);
	rc = false;
} else {
	/* we don't care if someone yielded */
	rq->state = VIRTNET_RQ_STATE_NAPI;
	spin_unlock(&rq->lock);
}
> +	return rc;
> +}
> +
> +/* returns true is someone tried to get the rq while napi or refill had it
*/
> +static inline bool virtnet_rq_unlock_napi_refill(struct receive_queue *rq)
> +{
> +	int rc = false;
> +
> +	spin_lock(&rq->lock);
> +	WARN_ON(rq->state & (VIRTNET_RQ_STATE_POLL |
> +			     VIRTNET_RQ_STATE_NAPI_YIELD));
> +
> +	if (rq->state & VIRTNET_RQ_STATE_POLL_YIELD)
> +		rc = true;
> +	/* will reset state to idle, unless RQ is disabled */
> +	rq->state &= VIRTNET_RQ_STATE_DISABLED;
> +	spin_unlock(&rq->lock);
> +	return rc;
> +}
> +
> +/* called from virtnet_low_latency_recv() */
> +static inline bool virtnet_rq_lock_poll(struct receive_queue *rq)
> +{
> +	int rc = true;
> +
> +	spin_lock_bh(&rq->lock);
> +	if ((rq->state & VIRTNET_RQ_LOCKED)) {
> +		rq->state |= VIRTNET_RQ_STATE_POLL_YIELD;
> +		rc = false;
> +	} else
> +		/* preserve yield marks */
> +		rq->state |= VIRTNET_RQ_STATE_POLL;
> +	spin_unlock_bh(&rq->lock);
> +	return rc;
> +}
> +
> +/* returns true if someone tried to get the receive queue while it was
locked */
> +static inline bool virtnet_rq_unlock_poll(struct receive_queue *rq)
> +{
> +	int rc = false;
> +
> +	spin_lock_bh(&rq->lock);
> +	WARN_ON(rq->state & (VIRTNET_RQ_STATE_NAPI));
> +
> +	if (rq->state & VIRTNET_RQ_STATE_POLL_YIELD)
> +		rc = true;
> +	/* will reset state to idle, unless RQ is disabled */
> +	rq->state &= VIRTNET_RQ_STATE_DISABLED;
> +	spin_unlock_bh(&rq->lock);
> +	return rc;
> +}
> +
> +/* return false if RQ is currently owned */
> +static inline bool virtnet_rq_disable(struct receive_queue *rq)
> +{
> +	int rc = true;
> +
> +	spin_lock_bh(&rq->lock);
> +	if (rq->state & VIRTNET_RQ_OWNED)
> +		rc = false;
> +	rq->state |= VIRTNET_RQ_STATE_DISABLED;
> +	spin_unlock_bh(&rq->lock);
> +
> +	return rc;
> +}
> +
> +#else /* CONFIG_NET_RX_BUSY_POLL */
> +static inline void virtnet_rq_init_lock(struct receive_queue *rq)
> +{
> +}
> +
> +static inline bool virtnet_rq_lock_napi_refill(struct receive_queue *rq)
> +{
> +	return true;
> +}
> +
> +static inline bool virtnet_rq_unlock_napi_refill(struct receive_queue *rq)
> +{
> +	return false;
> +}
> +
> +static inline bool virtnet_rq_lock_poll(struct receive_queue *rq)
> +{
> +	return false;
> +}
> +
> +static inline bool virtnet_rq_unlock_poll(struct receive_queue *rq)
> +{
> +	return false;
> +}
> +
> +static inline bool virtnet_rq_disable(struct receive_queue *rq)
> +{
> +	return true;
> +}
> +
> +#endif /* CONFIG_NET_RX_BUSY_POLL */
> +
>   struct virtnet_info {
>   	struct virtio_device *vdev;
>   	struct virtqueue *cvq;
> @@ -521,6 +657,8 @@ static void receive_buf(struct receive_queue *rq, void
*buf, unsigned int len)
>   		skb_shinfo(skb)->gso_segs = 0;
>   	}
>   
> +	skb_mark_napi_id(skb, &rq->napi);
> +
>   	netif_receive_skb(skb);
>   	return;
>   
> @@ -714,7 +852,12 @@ static void refill_work(struct work_struct *work)
>   		struct receive_queue *rq = &vi->rq[i];
>   
>   		napi_disable(&rq->napi);
> +		if (!virtnet_rq_lock_napi_refill(rq)) {
> +			virtnet_napi_enable(rq);
> +			continue;
> +		}
>   		still_empty = !try_fill_recv(rq, GFP_KERNEL);
> +		virtnet_rq_unlock_napi_refill(rq);
>   		virtnet_napi_enable(rq);
>   
>   		/* In theory, this can happen: if we don't get any buffers in
> @@ -752,8 +895,13 @@ static int virtnet_poll(struct napi_struct *napi, int
budget)
>   	unsigned int r, received = 0;
>   
>   again:
> +	if (!virtnet_rq_lock_napi_refill(rq))
> +		return budget;
> +
>   	received += virtnet_receive(rq, budget);
>   
> +	virtnet_rq_unlock_napi_refill(rq);
> +
>   	/* Out of packets? */
>   	if (received < budget) {
>   		r = virtqueue_enable_cb_prepare(rq->vq);
> @@ -770,20 +918,50 @@ again:
>   	return received;
>   }
>   
> +#ifdef CONFIG_NET_RX_BUSY_POLL
> +/* must be called with local_bh_disable()d */
> +static int virtnet_low_latency_recv(struct napi_struct *napi)
> +{
> +	struct receive_queue *rq > +		container_of(napi, struct receive_queue,
napi);
> +	struct virtnet_info *vi = rq->vq->vdev->priv;
> +	int received;
> +
> +	if (!(vi->status & VIRTIO_NET_S_LINK_UP))
> +		return LL_FLUSH_FAILED;
> +
> +	if (!virtnet_rq_lock_poll(rq))
> +		return LL_FLUSH_BUSY;
> +
> +	received = virtnet_receive(rq, 4);
> +
> +	virtnet_rq_unlock_poll(rq);
> +
> +	return received;
> +}
> +#endif	/* CONFIG_NET_RX_BUSY_POLL */
> +
>   static void virtnet_napi_enable_all(struct virtnet_info *vi)
>   {
>   	int i;
>   
> -	for (i = 0; i < vi->max_queue_pairs; i++)
> +	for (i = 0; i < vi->max_queue_pairs; i++) {
> +		virtnet_rq_init_lock(&vi->rq[i]);
>   		virtnet_napi_enable(&vi->rq[i]);
> +	}
>   }
>   
>   static void virtnet_napi_disable_all(struct virtnet_info *vi)
>   {
>   	int i;
>   
> -	for (i = 0; i < vi->max_queue_pairs; i++)
> +	for (i = 0; i < vi->max_queue_pairs; i++) {
>   		napi_disable(&vi->rq[i].napi);
> +		while (!virtnet_rq_disable(&vi->rq[i])) {
> +			pr_info("RQ %d locked\n", i);
> +			usleep_range(1000, 20000);
> +		}
> +	}
>   }
>   
>   static int virtnet_open(struct net_device *dev)
> @@ -1372,6 +1550,9 @@ static const struct net_device_ops virtnet_netdev = {
>   #ifdef CONFIG_NET_POLL_CONTROLLER
>   	.ndo_poll_controller = virtnet_netpoll,
>   #endif
> +#ifdef CONFIG_NET_RX_BUSY_POLL
> +	.ndo_busy_poll		= virtnet_low_latency_recv,
> +#endif
>   };
>   
>   static void virtnet_config_changed_work(struct work_struct *work)
> @@ -1577,6 +1758,7 @@ 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);
> +		napi_hash_add(&vi->rq[i].napi);
>   
>   		sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
>   		ewma_init(&vi->rq[i].mrg_avg_pkt_len, 1, RECEIVE_AVG_WEIGHT);
> @@ -1880,8 +2062,10 @@ static int virtnet_freeze(struct virtio_device
*vdev)
>   
>   	if (netif_running(vi->dev)) {
>   		virtnet_napi_disable_all(vi);
> -		for (i = 0; i < vi->max_queue_pairs; i++)
> +		for (i = 0; i < vi->max_queue_pairs; i++) {
> +			napi_hash_del(&vi->rq[i].napi);
>   			netif_napi_del(&vi->rq[i].napi);
> +		}
>   	}
>   
>   	remove_vq_common(vi);

-- 
Regards,
Varka Bhadram.

From: Jason Wang <jasowang at redhat.com>
Date: Wed, 16 Jul 2014 14:21:44 +0800
> Hi all:
> 
> This series introduces the support for rx busy polling support. This
> was useful for reduing the latency for a kvm guest. Patch 1-2
> introduces helpers which is used for rx busy polling. Patch 3
> implement the main function.
> 
> Test was done between a kvm guest and an external host. Two hosts were
> connected through 40gb mlx4 cards. With both busy_poll and busy_read are
> set to 50 in guest, 1 byte netperf tcp_rr shows 116% improvement:
> transaction rate was increased from 9151.94 to 19787.37.
> 
> Changes from V1:
> - split the patch info smaller ones
> - add more details about test setup/configuration
> 
> Please review.
Looks like you can make some minor adjustments based upon review, and
anyways I'd like to wait for Michael to get back and review a change
of this nature.

Thanks.

On 07/17/2014 02:21 PM, David Miller wrote:
> From: Jason Wang <jasowang at redhat.com>
> Date: Wed, 16 Jul 2014 14:21:44 +0800
>
>> Hi all:
>>
>> This series introduces the support for rx busy polling support. This
>> was useful for reduing the latency for a kvm guest. Patch 1-2
>> introduces helpers which is used for rx busy polling. Patch 3
>> implement the main function.
>>
>> Test was done between a kvm guest and an external host. Two hosts were
>> connected through 40gb mlx4 cards. With both busy_poll and busy_read
are
>> set to 50 in guest, 1 byte netperf tcp_rr shows 116% improvement:
>> transaction rate was increased from 9151.94 to 19787.37.
>>
>> Changes from V1:
>> - split the patch info smaller ones
>> - add more details about test setup/configuration
>>
>> Please review.
> Looks like you can make some minor adjustments based upon review, and
> anyways I'd like to wait for Michael to get back and review a change
> of this nature.
>
> Thanks.
Ok, I will wait for Michael's comments and then post V3.

Thanks

On Wed, Jul 16, 2014 at 02:21:47PM +0800, Jason Wang
wrote:
> Add basic support for rx busy polling.
> 
> Test was done between a kvm guest and an external host. Two hosts were
> connected through 40gb mlx4 cards. With both busy_poll and busy_read
> are set to 50 in guest, 1 byte netperf tcp_rr shows 116% improvement:
> transaction rate was increased from 9151.94 to 19787.37.
Pls include data about non polling tests: any effect on
cpu utilization there?
There could be as we are adding locking.
> 
> Cc: Rusty Russell <rusty at rustcorp.com.au>
> Cc: Michael S. Tsirkin <mst at redhat.com>
> Cc: Vlad Yasevich <vyasevic at redhat.com>
> Cc: Eric Dumazet <eric.dumazet at gmail.com>
> Signed-off-by: Jason Wang <jasowang at redhat.com>
> ---
>  drivers/net/virtio_net.c | 190
++++++++++++++++++++++++++++++++++++++++++++++-
>  1 file changed, 187 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> index e417d93..4830713 100644
> --- a/drivers/net/virtio_net.c
> +++ b/drivers/net/virtio_net.c
> @@ -27,6 +27,7 @@
>  #include <linux/slab.h>
>  #include <linux/cpu.h>
>  #include <linux/average.h>
> +#include <net/busy_poll.h>
>  
>  static int napi_weight = NAPI_POLL_WEIGHT;
>  module_param(napi_weight, int, 0444);
> @@ -94,8 +95,143 @@ struct receive_queue {
>  
>  	/* Name of this receive queue: input.$index */
>  	char name[40];
> +
> +#ifdef CONFIG_NET_RX_BUSY_POLL
> +	unsigned int state;
> +#define VIRTNET_RQ_STATE_IDLE        0
> +#define VIRTNET_RQ_STATE_NAPI	     1    /* NAPI or refill owns this RQ */
> +#define VIRTNET_RQ_STATE_POLL	     2    /* poll owns this RQ */
> +#define VIRTNET_RQ_STATE_DISABLED    4    /* RQ is disabled */
> +#define VIRTNET_RQ_OWNED (VIRTNET_RQ_STATE_NAPI | VIRTNET_RQ_STATE_POLL)
> +#define VIRTNET_RQ_LOCKED (VIRTNET_RQ_OWNED | VIRTNET_RQ_STATE_DISABLED)
> +#define VIRTNET_RQ_STATE_NAPI_YIELD  8    /* NAPI or refill yielded this
RQ */
> +#define VIRTNET_RQ_STATE_POLL_YIELD  16   /* poll yielded this RQ */
> +	spinlock_t lock;
> +#endif  /* CONFIG_NET_RX_BUSY_POLL */
do we have to have a new state? no way to reuse the napi state
for this? two lock/unlock operations for a poll seems
excessive.
>  };
>  
> +#ifdef CONFIG_NET_RX_BUSY_POLL
> +static inline void virtnet_rq_init_lock(struct receive_queue *rq)
> +{
> +
> +	spin_lock_init(&rq->lock);
> +	rq->state = VIRTNET_RQ_STATE_IDLE;
> +}
> +
> +/* called from the device poll routine or refill routine to get ownership
of a
> + * receive queue.
> + */
> +static inline bool virtnet_rq_lock_napi_refill(struct receive_queue *rq)
> +{
> +	int rc = true;
> +
> +	spin_lock(&rq->lock);
> +	if (rq->state & VIRTNET_RQ_LOCKED) {
> +		WARN_ON(rq->state & VIRTNET_RQ_STATE_NAPI);
> +		rq->state |= VIRTNET_RQ_STATE_NAPI_YIELD;
> +		rc = false;
> +	} else
> +		/* we don't care if someone yielded */
> +		rq->state = VIRTNET_RQ_STATE_NAPI;
> +	spin_unlock(&rq->lock);
> +	return rc;
> +}
> +
> +/* returns true is someone tried to get the rq while napi or refill had it
*/
s/is/if/
> +static inline bool virtnet_rq_unlock_napi_refill(struct receive_queue *rq)
> +{
> +	int rc = false;
> +
> +	spin_lock(&rq->lock);
> +	WARN_ON(rq->state & (VIRTNET_RQ_STATE_POLL |
> +			     VIRTNET_RQ_STATE_NAPI_YIELD));
> +
> +	if (rq->state & VIRTNET_RQ_STATE_POLL_YIELD)
> +		rc = true;
> +	/* will reset state to idle, unless RQ is disabled */
> +	rq->state &= VIRTNET_RQ_STATE_DISABLED;
> +	spin_unlock(&rq->lock);
> +	return rc;
> +}
> +
> +/* called from virtnet_low_latency_recv() */
> +static inline bool virtnet_rq_lock_poll(struct receive_queue *rq)
> +{
> +	int rc = true;
> +
> +	spin_lock_bh(&rq->lock);
> +	if ((rq->state & VIRTNET_RQ_LOCKED)) {
> +		rq->state |= VIRTNET_RQ_STATE_POLL_YIELD;
> +		rc = false;
> +	} else
> +		/* preserve yield marks */
> +		rq->state |= VIRTNET_RQ_STATE_POLL;
> +	spin_unlock_bh(&rq->lock);
> +	return rc;
> +}
> +
> +/* returns true if someone tried to get the receive queue while it was
locked */
> +static inline bool virtnet_rq_unlock_poll(struct receive_queue *rq)
> +{
> +	int rc = false;
> +
> +	spin_lock_bh(&rq->lock);
> +	WARN_ON(rq->state & (VIRTNET_RQ_STATE_NAPI));
> +
> +	if (rq->state & VIRTNET_RQ_STATE_POLL_YIELD)
> +		rc = true;
> +	/* will reset state to idle, unless RQ is disabled */
> +	rq->state &= VIRTNET_RQ_STATE_DISABLED;
> +	spin_unlock_bh(&rq->lock);
> +	return rc;
> +}
> +
> +/* return false if RQ is currently owned */
> +static inline bool virtnet_rq_disable(struct receive_queue *rq)
> +{
> +	int rc = true;
> +
> +	spin_lock_bh(&rq->lock);
> +	if (rq->state & VIRTNET_RQ_OWNED)
> +		rc = false;
> +	rq->state |= VIRTNET_RQ_STATE_DISABLED;
> +	spin_unlock_bh(&rq->lock);
> +
> +	return rc;
> +}
> +
> +#else /* CONFIG_NET_RX_BUSY_POLL */
> +static inline void virtnet_rq_init_lock(struct receive_queue *rq)
> +{
> +}
> +
> +static inline bool virtnet_rq_lock_napi_refill(struct receive_queue *rq)
> +{
> +	return true;
> +}
> +
> +static inline bool virtnet_rq_unlock_napi_refill(struct receive_queue *rq)
> +{
> +	return false;
> +}
> +
> +static inline bool virtnet_rq_lock_poll(struct receive_queue *rq)
> +{
> +	return false;
> +}
> +
> +static inline bool virtnet_rq_unlock_poll(struct receive_queue *rq)
> +{
> +	return false;
> +}
> +
> +static inline bool virtnet_rq_disable(struct receive_queue *rq)
> +{
> +	return true;
> +}
> +
> +#endif /* CONFIG_NET_RX_BUSY_POLL */
> +
>  struct virtnet_info {
>  	struct virtio_device *vdev;
>  	struct virtqueue *cvq;
> @@ -521,6 +657,8 @@ static void receive_buf(struct receive_queue *rq, void
*buf, unsigned int len)
>  		skb_shinfo(skb)->gso_segs = 0;
>  	}
>  
> +	skb_mark_napi_id(skb, &rq->napi);
> +
>  	netif_receive_skb(skb);
>  	return;
>  
> @@ -714,7 +852,12 @@ static void refill_work(struct work_struct *work)
>  		struct receive_queue *rq = &vi->rq[i];
>  
>  		napi_disable(&rq->napi);
> +		if (!virtnet_rq_lock_napi_refill(rq)) {
> +			virtnet_napi_enable(rq);
> +			continue;
> +		}
>  		still_empty = !try_fill_recv(rq, GFP_KERNEL);
> +		virtnet_rq_unlock_napi_refill(rq);
>  		virtnet_napi_enable(rq);
>  
>  		/* In theory, this can happen: if we don't get any buffers in
> @@ -752,8 +895,13 @@ static int virtnet_poll(struct napi_struct *napi, int
budget)
>  	unsigned int r, received = 0;
>  
>  again:
> +	if (!virtnet_rq_lock_napi_refill(rq))
> +		return budget;
> +
>  	received += virtnet_receive(rq, budget);
>  
> +	virtnet_rq_unlock_napi_refill(rq);
> +
>  	/* Out of packets? */
>  	if (received < budget) {
>  		r = virtqueue_enable_cb_prepare(rq->vq);
> @@ -770,20 +918,50 @@ again:
>  	return received;
>  }
>  
> +#ifdef CONFIG_NET_RX_BUSY_POLL
> +/* must be called with local_bh_disable()d */
> +static int virtnet_low_latency_recv(struct napi_struct *napi)
let's call it busy poll :)
> +{
> +	struct receive_queue *rq > +		container_of(napi, struct receive_queue,
napi);
> +	struct virtnet_info *vi = rq->vq->vdev->priv;
> +	int received;
> +
> +	if (!(vi->status & VIRTIO_NET_S_LINK_UP))
> +		return LL_FLUSH_FAILED;
> +
> +	if (!virtnet_rq_lock_poll(rq))
> +		return LL_FLUSH_BUSY;
> +
> +	received = virtnet_receive(rq, 4);
Hmm why 4 exactly?
> +
> +	virtnet_rq_unlock_poll(rq);
> +
> +	return received;
> +}
> +#endif	/* CONFIG_NET_RX_BUSY_POLL */
> +
>  static void virtnet_napi_enable_all(struct virtnet_info *vi)
>  {
>  	int i;
>  
> -	for (i = 0; i < vi->max_queue_pairs; i++)
> +	for (i = 0; i < vi->max_queue_pairs; i++) {
> +		virtnet_rq_init_lock(&vi->rq[i]);
>  		virtnet_napi_enable(&vi->rq[i]);
> +	}
>  }
>  
>  static void virtnet_napi_disable_all(struct virtnet_info *vi)
>  {
>  	int i;
>  
> -	for (i = 0; i < vi->max_queue_pairs; i++)
> +	for (i = 0; i < vi->max_queue_pairs; i++) {
>  		napi_disable(&vi->rq[i].napi);
> +		while (!virtnet_rq_disable(&vi->rq[i])) {
> +			pr_info("RQ %d locked\n", i);
> +			usleep_range(1000, 20000);
What's going on here, exactly?
> +		}
> +	}
>  }
>  
>  static int virtnet_open(struct net_device *dev)
> @@ -1372,6 +1550,9 @@ static const struct net_device_ops virtnet_netdev = {
>  #ifdef CONFIG_NET_POLL_CONTROLLER
>  	.ndo_poll_controller = virtnet_netpoll,
>  #endif
> +#ifdef CONFIG_NET_RX_BUSY_POLL
> +	.ndo_busy_poll		= virtnet_low_latency_recv,
> +#endif
>  };
>  
>  static void virtnet_config_changed_work(struct work_struct *work)
> @@ -1577,6 +1758,7 @@ 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);
> +		napi_hash_add(&vi->rq[i].napi);
>  
>  		sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
>  		ewma_init(&vi->rq[i].mrg_avg_pkt_len, 1, RECEIVE_AVG_WEIGHT);
> @@ -1880,8 +2062,10 @@ static int virtnet_freeze(struct virtio_device
*vdev)
>  
>  	if (netif_running(vi->dev)) {
>  		virtnet_napi_disable_all(vi);
> -		for (i = 0; i < vi->max_queue_pairs; i++)
> +		for (i = 0; i < vi->max_queue_pairs; i++) {
> +			napi_hash_del(&vi->rq[i].napi);
>  			netif_napi_del(&vi->rq[i].napi);
> +		}
>  	}
>  
>  	remove_vq_common(vi);
> -- 
> 1.8.3.1

On Wed, Jul 16, 2014 at 02:21:44PM +0800, Jason Wang
wrote:
> Hi all:
> 
> This series introduces the support for rx busy polling support. This
> was useful for reduing the latency for a kvm guest. Patch 1-2
> introduces helpers which is used for rx busy polling. Patch 3
> implement the main function.
> 
> Test was done between a kvm guest and an external host. Two hosts were
> connected through 40gb mlx4 cards. With both busy_poll and busy_read are
> set to 50 in guest, 1 byte netperf tcp_rr shows 116% improvement:
> transaction rate was increased from 9151.94 to 19787.37.
> 
> Changes from V1:
> - split the patch info smaller ones
> - add more details about test setup/configuration
> 
> Please review.
Generally I think we should let host know we are polling.
For example, kick the rq or something. Or maybe add another
io address.
Something like this would need a new feature flag though, so I'm fine
with just polling in guest until that is available.
> Jason Wang (3):
>   virtio-net: introduce helpers to enable and disable all NAPIs
>   virtio-net: introduce virtnet_receive()
>   virtio-net: rx busy polling support
> 
>  drivers/net/virtio_net.c | 234
++++++++++++++++++++++++++++++++++++++++++++---
>  1 file changed, 221 insertions(+), 13 deletions(-)
> 
> -- 
> 1.8.3.1

On 07/21/2014 04:34 AM, Michael S. Tsirkin wrote:
> On Wed, Jul 16, 2014 at 02:21:44PM +0800, Jason Wang wrote:
>> Hi all:
>>
>> This series introduces the support for rx busy polling support. This
>> was useful for reduing the latency for a kvm guest. Patch 1-2
>> introduces helpers which is used for rx busy polling. Patch 3
>> implement the main function.
>>
>> Test was done between a kvm guest and an external host. Two hosts were
>> connected through 40gb mlx4 cards. With both busy_poll and busy_read
are
>> set to 50 in guest, 1 byte netperf tcp_rr shows 116% improvement:
>> transaction rate was increased from 9151.94 to 19787.37.
>>
>> Changes from V1:
>> - split the patch info smaller ones
>> - add more details about test setup/configuration
>>
>> Please review.
> Generally I think we should let host know we are polling.
> For example, kick the rq or something. Or maybe add another
> io address.
Yes, I'm also working on busy polling for tun and vhost which may also
help here.
> Something like this would need a new feature flag though, so I'm fine
> with just polling in guest until that is available.
Yes.
>
>> Jason Wang (3):
>>   virtio-net: introduce helpers to enable and disable all NAPIs
>>   virtio-net: introduce virtnet_receive()
>>   virtio-net: rx busy polling support
>>
>>  drivers/net/virtio_net.c | 234
++++++++++++++++++++++++++++++++++++++++++++---
>>  1 file changed, 221 insertions(+), 13 deletions(-)
>>
>> -- 
>> 1.8.3.1
> --
> To unsubscribe from this list: send the line "unsubscribe
linux-kernel" in
> the body of a message to majordomo at vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/