[PATCH V2 1/2] virtio-net: fix the set affinity bug when CPU IDs are not consecutive

As M.S.T mentioned, set affinity will not work very well when
CPU IDs are not consecutive, this can happen with hot unplug.
Fix this bug by traversal the online CPUs.

Cc: Rusty Russell <redacted>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <redacted>
Cc: Eric Dumazet <redacted>
Cc: virtualization@lists.linux-foundation.org
Cc: netdev@vger.kernel.org
Signed-off-by: Wanlong Gao <redacted>
---
 drivers/net/virtio_net.c | 24 +++++++++++++++++-------
 1 file changed, 17 insertions(+), 7 deletions(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index a6fcf15..b483fb5 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c

@@ -1016,6 +1016,7 @@ static int virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
 static void virtnet_set_affinity(struct virtnet_info *vi, bool set)
 {
 	int i;
+	int cpu;
 
 	/* In multiqueue mode, when the number of cpu is equal to the number of
 	 * queue pairs, we let the queue pairs to be private to one cpu by

@@ -1029,16 +1030,25 @@ static void virtnet_set_affinity(struct virtnet_info *vi, bool set)
 			return;
 	}
 
-	for (i = 0; i < vi->max_queue_pairs; i++) {
-		int cpu = set ? i : -1;
-		virtqueue_set_affinity(vi->rq[i].vq, cpu);
-		virtqueue_set_affinity(vi->sq[i].vq, cpu);
-	}
+	if (set) {
+		i = 0;
+		for_each_online_cpu(cpu) {
+			virtqueue_set_affinity(vi->rq[i].vq, cpu);
+			virtqueue_set_affinity(vi->sq[i].vq, cpu);
+			i++;
+			if (i >= vi->max_queue_pairs)
+				break;
+		}
 
-	if (set)
 		vi->affinity_hint_set = true;
-	else
+	} else {
+		for(i = 0; i < vi->max_queue_pairs; i++) {
+			virtqueue_set_affinity(vi->rq[i].vq, -1);
+			virtqueue_set_affinity(vi->sq[i].vq, -1);
+		}
+
 		vi->affinity_hint_set = false;
+	}
 }
 
 static void virtnet_get_ringparam(struct net_device *dev,

-- 
1.8.1

Add a cpu notifier to virtio-net, so that we can reset the
virtqueue affinity if the cpu hotplug happens. It improve
the performance through enabling or disabling the virtqueue
affinity after doing cpu hotplug.
Adding the notifier block to virtnet_info is suggested by
Jason, thank you.

Cc: Rusty Russell <redacted>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <redacted>
Cc: Eric Dumazet <redacted>
Cc: virtualization@lists.linux-foundation.org
Cc: netdev@vger.kernel.org
Signed-off-by: Wanlong Gao <redacted>
---
 drivers/net/virtio_net.c | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index b483fb5..9547b4c 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c

@@ -26,6 +26,7 @@
 #include <linux/scatterlist.h>
 #include <linux/if_vlan.h>
 #include <linux/slab.h>
+#include <linux/cpu.h>
 
 static int napi_weight = 128;
 module_param(napi_weight, int, 0444);

@@ -123,6 +124,9 @@ struct virtnet_info {
 
 	/* Does the affinity hint is set for virtqueues? */
 	bool affinity_hint_set;
+
+	/* CPU hot plug notifier */
+	struct notifier_block nb;
 };
 
 struct skb_vnet_hdr {

@@ -1051,6 +1055,23 @@ static void virtnet_set_affinity(struct virtnet_info *vi, bool set)
 	}
 }
 
+static int virtnet_cpu_callback(struct notifier_block *nfb,
+			        unsigned long action, void *hcpu)
+{
+	struct virtnet_info *vi = container_of(nfb, struct virtnet_info, nb);
+	switch(action) {
+	case CPU_ONLINE:
+	case CPU_ONLINE_FROZEN:
+	case CPU_DEAD:
+	case CPU_DEAD_FROZEN:
+		virtnet_set_affinity(vi, true);
+		break;
+	default:
+		break;
+	}
+	return NOTIFY_OK;
+}
+
 static void virtnet_get_ringparam(struct net_device *dev,
 				struct ethtool_ringparam *ring)
 {

@@ -1509,6 +1530,13 @@ static int virtnet_probe(struct virtio_device *vdev)
 		}
 	}
 
+	vi->nb.notifier_call = &virtnet_cpu_callback;
+	err = register_hotcpu_notifier(&vi->nb);
+	if (err) {
+		pr_debug("virtio_net: registering cpu notifier failed\n");
+		goto free_recv_bufs;
+	}
+
 	/* Assume link up if device can't report link status,
 	   otherwise get link status from config. */
 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {

@@ -1553,6 +1581,8 @@ static void virtnet_remove(struct virtio_device *vdev)
 {
 	struct virtnet_info *vi = vdev->priv;
 
+	unregister_hotcpu_notifier(&vi->nb);
+
 	/* Prevent config work handler from accessing the device. */
 	mutex_lock(&vi->config_lock);
 	vi->config_enable = false;

-- 
1.8.1

On 01/07/2013 03:15 PM, Wanlong Gao wrote:

I think you'd better fix the .ndo_select_queue() as well (as Michael
said in your V1) since it currently uses smp processor id which may not
work very well in this case also.

Thanks

On 01/07/2013 03:28 PM, Jason Wang wrote:

On 01/07/2013 03:15 PM, Wanlong Gao wrote:

quoted

Add a cpu notifier to virtio-net, so that we can reset the
virtqueue affinity if the cpu hotplug happens. It improve
the performance through enabling or disabling the virtqueue
affinity after doing cpu hotplug.
Adding the notifier block to virtnet_info is suggested by
Jason, thank you.

Cc: Rusty Russell <redacted>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <redacted>
Cc: Eric Dumazet <redacted>
Cc: virtualization@lists.linux-foundation.org
Cc: netdev@vger.kernel.org
Signed-off-by: Wanlong Gao <redacted>
---
 drivers/net/virtio_net.c | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index b483fb5..9547b4c 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c

@@ -26,6 +26,7 @@
 #include <linux/scatterlist.h>
 #include <linux/if_vlan.h>
 #include <linux/slab.h>
+#include <linux/cpu.h>
 
 static int napi_weight = 128;
 module_param(napi_weight, int, 0444);

@@ -123,6 +124,9 @@ struct virtnet_info {
 
 	/* Does the affinity hint is set for virtqueues? */
 	bool affinity_hint_set;
+
+	/* CPU hot plug notifier */
+	struct notifier_block nb;
 };
 
 struct skb_vnet_hdr {

@@ -1051,6 +1055,23 @@ static void virtnet_set_affinity(struct virtnet_info *vi, bool set)
 	}
 }
 
+static int virtnet_cpu_callback(struct notifier_block *nfb,
+			        unsigned long action, void *hcpu)
+{
+	struct virtnet_info *vi = container_of(nfb, struct virtnet_info, nb);
+	switch(action) {
+	case CPU_ONLINE:
+	case CPU_ONLINE_FROZEN:
+	case CPU_DEAD:
+	case CPU_DEAD_FROZEN:
+		virtnet_set_affinity(vi, true);
+		break;
+	default:
+		break;
+	}
+	return NOTIFY_OK;
+}
+

I think you'd better fix the .ndo_select_queue() as well (as Michael
said in your V1) since it currently uses smp processor id which may not
work very well in this case also.

The bug is we can't get the right txq if the CPU IDs are not consecutive,
right? Do you have any good idea about fixing this? 

Thanks,
Wanlong Gao

Thanks

quoted

 static void virtnet_get_ringparam(struct net_device *dev,
 				struct ethtool_ringparam *ring)
 {

@@ -1509,6 +1530,13 @@ static int virtnet_probe(struct virtio_device *vdev)
 		}
 	}
 
+	vi->nb.notifier_call = &virtnet_cpu_callback;
+	err = register_hotcpu_notifier(&vi->nb);
+	if (err) {
+		pr_debug("virtio_net: registering cpu notifier failed\n");
+		goto free_recv_bufs;
+	}
+
 	/* Assume link up if device can't report link status,
 	   otherwise get link status from config. */
 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {

@@ -1553,6 +1581,8 @@ static void virtnet_remove(struct virtio_device *vdev)
 {
 	struct virtnet_info *vi = vdev->priv;
 
+	unregister_hotcpu_notifier(&vi->nb);
+
 	/* Prevent config work handler from accessing the device. */
 	mutex_lock(&vi->config_lock);
 	vi->config_enable = false;

On 01/07/2013 03:48 PM, Wanlong Gao wrote:

On 01/07/2013 03:28 PM, Jason Wang wrote:

quoted

On 01/07/2013 03:15 PM, Wanlong Gao wrote:

quoted

Add a cpu notifier to virtio-net, so that we can reset the
virtqueue affinity if the cpu hotplug happens. It improve
the performance through enabling or disabling the virtqueue
affinity after doing cpu hotplug.
Adding the notifier block to virtnet_info is suggested by
Jason, thank you.

Cc: Rusty Russell <redacted>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <redacted>
Cc: Eric Dumazet <redacted>
Cc: virtualization@lists.linux-foundation.org
Cc: netdev@vger.kernel.org
Signed-off-by: Wanlong Gao <redacted>
---
 drivers/net/virtio_net.c | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index b483fb5..9547b4c 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c

@@ -26,6 +26,7 @@
 #include <linux/scatterlist.h>
 #include <linux/if_vlan.h>
 #include <linux/slab.h>
+#include <linux/cpu.h>
 
 static int napi_weight = 128;
 module_param(napi_weight, int, 0444);

@@ -123,6 +124,9 @@ struct virtnet_info {
 
 	/* Does the affinity hint is set for virtqueues? */
 	bool affinity_hint_set;
+
+	/* CPU hot plug notifier */
+	struct notifier_block nb;
 };
 
 struct skb_vnet_hdr {

@@ -1051,6 +1055,23 @@ static void virtnet_set_affinity(struct virtnet_info *vi, bool set)
 	}
 }
 
+static int virtnet_cpu_callback(struct notifier_block *nfb,
+			        unsigned long action, void *hcpu)
+{
+	struct virtnet_info *vi = container_of(nfb, struct virtnet_info, nb);
+	switch(action) {
+	case CPU_ONLINE:
+	case CPU_ONLINE_FROZEN:
+	case CPU_DEAD:
+	case CPU_DEAD_FROZEN:
+		virtnet_set_affinity(vi, true);
+		break;
+	default:
+		break;
+	}
+	return NOTIFY_OK;
+}
+

I think you'd better fix the .ndo_select_queue() as well (as Michael
said in your V1) since it currently uses smp processor id which may not
work very well in this case also.

The bug is we can't get the right txq if the CPU IDs are not consecutive,
right? Do you have any good idea about fixing this? 

Thanks,
Wanlong Gao

The point is make the virtqueue private to a specific cpu when the
number of queue pairs is equal to the number of cpus. So after you bind
the vq affinity to a specific cpu, you'd better use the reverse mapping
of this affinity to do .ndo_select_queue(). One possible idea, as
Michael suggested, is a per-cpu structure to record the preferable
virtqueue and do both .ndo_select_queue() and affinity hint setting
based on this.

quoted

Thanks

quoted

 static void virtnet_get_ringparam(struct net_device *dev,
 				struct ethtool_ringparam *ring)
 {

@@ -1509,6 +1530,13 @@ static int virtnet_probe(struct virtio_device *vdev)
 		}
 	}
 
+	vi->nb.notifier_call = &virtnet_cpu_callback;
+	err = register_hotcpu_notifier(&vi->nb);
+	if (err) {
+		pr_debug("virtio_net: registering cpu notifier failed\n");
+		goto free_recv_bufs;
+	}
+
 	/* Assume link up if device can't report link status,
 	   otherwise get link status from config. */
 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {

@@ -1553,6 +1581,8 @@ static void virtnet_remove(struct virtio_device *vdev)
 {
 	struct virtnet_info *vi = vdev->priv;
 
+	unregister_hotcpu_notifier(&vi->nb);
+
 	/* Prevent config work handler from accessing the device. */
 	mutex_lock(&vi->config_lock);
 	vi->config_enable = false;

On 01/07/2013 03:55 PM, Jason Wang wrote:

On 01/07/2013 03:48 PM, Wanlong Gao wrote:

quoted

On 01/07/2013 03:28 PM, Jason Wang wrote:

quoted

On 01/07/2013 03:15 PM, Wanlong Gao wrote:

quoted

Add a cpu notifier to virtio-net, so that we can reset the
virtqueue affinity if the cpu hotplug happens. It improve
the performance through enabling or disabling the virtqueue
affinity after doing cpu hotplug.
Adding the notifier block to virtnet_info is suggested by
Jason, thank you.

Cc: Rusty Russell <redacted>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <redacted>
Cc: Eric Dumazet <redacted>
Cc: virtualization@lists.linux-foundation.org
Cc: netdev@vger.kernel.org
Signed-off-by: Wanlong Gao <redacted>
---
 drivers/net/virtio_net.c | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index b483fb5..9547b4c 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c

@@ -26,6 +26,7 @@
 #include <linux/scatterlist.h>
 #include <linux/if_vlan.h>
 #include <linux/slab.h>
+#include <linux/cpu.h>
 
 static int napi_weight = 128;
 module_param(napi_weight, int, 0444);

@@ -123,6 +124,9 @@ struct virtnet_info {
 
 	/* Does the affinity hint is set for virtqueues? */
 	bool affinity_hint_set;
+
+	/* CPU hot plug notifier */
+	struct notifier_block nb;
 };
 
 struct skb_vnet_hdr {

@@ -1051,6 +1055,23 @@ static void virtnet_set_affinity(struct virtnet_info *vi, bool set)
 	}
 }
 
+static int virtnet_cpu_callback(struct notifier_block *nfb,
+			        unsigned long action, void *hcpu)
+{
+	struct virtnet_info *vi = container_of(nfb, struct virtnet_info, nb);
+	switch(action) {
+	case CPU_ONLINE:
+	case CPU_ONLINE_FROZEN:
+	case CPU_DEAD:
+	case CPU_DEAD_FROZEN:
+		virtnet_set_affinity(vi, true);
+		break;
+	default:
+		break;
+	}
+	return NOTIFY_OK;
+}
+

I think you'd better fix the .ndo_select_queue() as well (as Michael
said in your V1) since it currently uses smp processor id which may not
work very well in this case also.

The bug is we can't get the right txq if the CPU IDs are not consecutive,
right? Do you have any good idea about fixing this? 

Thanks,
Wanlong Gao

The point is make the virtqueue private to a specific cpu when the
number of queue pairs is equal to the number of cpus. So after you bind
the vq affinity to a specific cpu, you'd better use the reverse mapping
of this affinity to do .ndo_select_queue(). One possible idea, as
Michael suggested, is a per-cpu structure to record the preferable
virtqueue and do both .ndo_select_queue() and affinity hint setting
based on this.

Yeah, I think I got it now, will address it in V3. thank you. ;)

Regards,
Wanlong Gao

quoted

quoted

Thanks

quoted

 static void virtnet_get_ringparam(struct net_device *dev,
 				struct ethtool_ringparam *ring)
 {

@@ -1509,6 +1530,13 @@ static int virtnet_probe(struct virtio_device *vdev)
 		}
 	}
 
+	vi->nb.notifier_call = &virtnet_cpu_callback;
+	err = register_hotcpu_notifier(&vi->nb);
+	if (err) {
+		pr_debug("virtio_net: registering cpu notifier failed\n");
+		goto free_recv_bufs;
+	}
+
 	/* Assume link up if device can't report link status,
 	   otherwise get link status from config. */
 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {

@@ -1553,6 +1581,8 @@ static void virtnet_remove(struct virtio_device *vdev)
 {
 	struct virtnet_info *vi = vdev->priv;
 
+	unregister_hotcpu_notifier(&vi->nb);
+
 	/* Prevent config work handler from accessing the device. */
 	mutex_lock(&vi->config_lock);
 	vi->config_enable = false;