diff mbox

[V2] virtio-net: calculate proper msix vectors on init

Message ID 1392973519-5371-1-git-send-email-jasowang@redhat.com
State New
Headers show

Commit Message

Jason Wang Feb. 21, 2014, 9:05 a.m. UTC 
Currently, the default msix vectors for virtio-net-pci is 3 which is
obvious not suitable for multiqueue guest, so we depends on the user
or management tools to pass a correct vectors parameter. In fact, we
can simplifying this by calculate the number of vectors on init.

Consider we have N queues, the number of vectors needed is 2*N + 2
(plus one config interrupt and control vq). We didn't check whether or
not host support control vq because it was added unconditionally by
qemu to avoid breaking legacy guests such as Minix.

Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
---
 hw/virtio/virtio-pci.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

Comments

Gonglei (Arei) March 4, 2014, 3:26 a.m. UTC | #1 
> -----Original Message-----
> From: qemu-devel-bounces+arei.gonglei=huawei.com@nongnu.org
> [mailto:qemu-devel-bounces+arei.gonglei=huawei.com@nongnu.org] On
> Behalf Of Jason Wang
> Sent: Friday, February 21, 2014 5:05 PM
> To: aliguori@amazon.com; mst@redhat.com; qemu-devel@nongnu.org
> Cc: Paolo Bonzini; Jason Wang
> Subject: [Qemu-devel] [PATCH V2] virtio-net: calculate proper msix vectors on
> init
> 
> Currently, the default msix vectors for virtio-net-pci is 3 which is
> obvious not suitable for multiqueue guest, so we depends on the user
> or management tools to pass a correct vectors parameter. In fact, we
> can simplifying this by calculate the number of vectors on init.
> 
> Consider we have N queues, the number of vectors needed is 2*N + 2
> (plus one config interrupt and control vq). We didn't check whether or
> not host support control vq because it was added unconditionally by
> qemu to avoid breaking legacy guests such as Minix.
> 
> Cc: Paolo Bonzini <pbonzini@redhat.com>
> Cc: Michael S. Tsirkin <mst@redhat.com>
> Signed-off-by: Jason Wang <jasowang@redhat.com>
> ---
>  hw/virtio/virtio-pci.c | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
> 
> diff --git a/hw/virtio/virtio-pci.c b/hw/virtio/virtio-pci.c
> index 7b91841..3b3b0e2 100644
> --- a/hw/virtio/virtio-pci.c
> +++ b/hw/virtio/virtio-pci.c
> @@ -1416,7 +1416,8 @@ static const TypeInfo virtio_serial_pci_info = {
>  static Property virtio_net_properties[] = {
>      DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags,
>                      VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, false),
> -    DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors, 3),
> +    DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors,
> +                       DEV_NVECTORS_UNSPECIFIED),
>      DEFINE_VIRTIO_NET_FEATURES(VirtIOPCIProxy, host_features),
>      DEFINE_NIC_PROPERTIES(VirtIONetPCI, vdev.nic_conf),
>      DEFINE_VIRTIO_NET_PROPERTIES(VirtIONetPCI, vdev.net_conf),
> @@ -1428,6 +1429,11 @@ static int virtio_net_pci_init(VirtIOPCIProxy
> *vpci_dev)
>      DeviceState *qdev = DEVICE(vpci_dev);
>      VirtIONetPCI *dev = VIRTIO_NET_PCI(vpci_dev);
>      DeviceState *vdev = DEVICE(&dev->vdev);
> +    VirtIONet *net = VIRTIO_NET(&dev->vdev);
> +
> +    if (vpci_dev->nvectors == DEV_NVECTORS_UNSPECIFIED) {
> +        vpci_dev->nvectors = 2 * MAX(net->nic_conf.queues, 1) + 2;
> +    }
> 
>      virtio_net_set_config_size(&dev->vdev, vpci_dev->host_features);
>      virtio_net_set_netclient_name(&dev->vdev, qdev->id,
> --
> 1.8.3.2
> 

Good catch.

Reviewed-by: Gonglei <arei.gonglei@huawei.com>


Best regards,
-Gonglei
Jason Wang March 4, 2014, 7:34 a.m. UTC | #2 
On 03/04/2014 11:26 AM, Gonglei (Arei) wrote:
>> -----Original Message-----
>> From: qemu-devel-bounces+arei.gonglei=huawei.com@nongnu.org
>> [mailto:qemu-devel-bounces+arei.gonglei=huawei.com@nongnu.org] On
>> Behalf Of Jason Wang
>> Sent: Friday, February 21, 2014 5:05 PM
>> To: aliguori@amazon.com; mst@redhat.com; qemu-devel@nongnu.org
>> Cc: Paolo Bonzini; Jason Wang
>> Subject: [Qemu-devel] [PATCH V2] virtio-net: calculate proper msix vectors on
>> init
>>
>> Currently, the default msix vectors for virtio-net-pci is 3 which is
>> obvious not suitable for multiqueue guest, so we depends on the user
>> or management tools to pass a correct vectors parameter. In fact, we
>> can simplifying this by calculate the number of vectors on init.
>>
>> Consider we have N queues, the number of vectors needed is 2*N + 2
>> (plus one config interrupt and control vq). We didn't check whether or
>> not host support control vq because it was added unconditionally by
>> qemu to avoid breaking legacy guests such as Minix.
>>
>> Cc: Paolo Bonzini <pbonzini@redhat.com>
>> Cc: Michael S. Tsirkin <mst@redhat.com>
>> Signed-off-by: Jason Wang <jasowang@redhat.com>
>> ---
>>  hw/virtio/virtio-pci.c | 8 +++++++-
>>  1 file changed, 7 insertions(+), 1 deletion(-)
>>
>> diff --git a/hw/virtio/virtio-pci.c b/hw/virtio/virtio-pci.c
>> index 7b91841..3b3b0e2 100644
>> --- a/hw/virtio/virtio-pci.c
>> +++ b/hw/virtio/virtio-pci.c
>> @@ -1416,7 +1416,8 @@ static const TypeInfo virtio_serial_pci_info = {
>>  static Property virtio_net_properties[] = {
>>      DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags,
>>                      VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, false),
>> -    DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors, 3),
>> +    DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors,
>> +                       DEV_NVECTORS_UNSPECIFIED),
>>      DEFINE_VIRTIO_NET_FEATURES(VirtIOPCIProxy, host_features),
>>      DEFINE_NIC_PROPERTIES(VirtIONetPCI, vdev.nic_conf),
>>      DEFINE_VIRTIO_NET_PROPERTIES(VirtIONetPCI, vdev.net_conf),
>> @@ -1428,6 +1429,11 @@ static int virtio_net_pci_init(VirtIOPCIProxy
>> *vpci_dev)
>>      DeviceState *qdev = DEVICE(vpci_dev);
>>      VirtIONetPCI *dev = VIRTIO_NET_PCI(vpci_dev);
>>      DeviceState *vdev = DEVICE(&dev->vdev);
>> +    VirtIONet *net = VIRTIO_NET(&dev->vdev);
>> +
>> +    if (vpci_dev->nvectors == DEV_NVECTORS_UNSPECIFIED) {
>> +        vpci_dev->nvectors = 2 * MAX(net->nic_conf.queues, 1) + 2;
>> +    }
>>
>>      virtio_net_set_config_size(&dev->vdev, vpci_dev->host_features);
>>      virtio_net_set_netclient_name(&dev->vdev, qdev->id,
>> --
>> 1.8.3.2
>>
> Good catch.
>
> Reviewed-by: Gonglei <arei.gonglei@huawei.com>
>
>
> Best regards,
> -Gonglei
>

Thanks for the reviewing.

The patch has one issue that it may breaks the migration if "vectors="
was not specified.

I will post a new version.

Thanks
Jason Wang March 4, 2014, 7:41 a.m. UTC | #3 
On 03/04/2014 11:26 AM, Gonglei (Arei) wrote:
>> -----Original Message-----
>> From: qemu-devel-bounces+arei.gonglei=huawei.com@nongnu.org
>> [mailto:qemu-devel-bounces+arei.gonglei=huawei.com@nongnu.org] On
>> Behalf Of Jason Wang
>> Sent: Friday, February 21, 2014 5:05 PM
>> To: aliguori@amazon.com; mst@redhat.com; qemu-devel@nongnu.org
>> Cc: Paolo Bonzini; Jason Wang
>> Subject: [Qemu-devel] [PATCH V2] virtio-net: calculate proper msix vectors on
>> init
>>
>> Currently, the default msix vectors for virtio-net-pci is 3 which is
>> obvious not suitable for multiqueue guest, so we depends on the user
>> or management tools to pass a correct vectors parameter. In fact, we
>> can simplifying this by calculate the number of vectors on init.
>>
>> Consider we have N queues, the number of vectors needed is 2*N + 2
>> (plus one config interrupt and control vq). We didn't check whether or
>> not host support control vq because it was added unconditionally by
>> qemu to avoid breaking legacy guests such as Minix.
>>
>> Cc: Paolo Bonzini <pbonzini@redhat.com>
>> Cc: Michael S. Tsirkin <mst@redhat.com>
>> Signed-off-by: Jason Wang <jasowang@redhat.com>
>> ---
>>  hw/virtio/virtio-pci.c | 8 +++++++-
>>  1 file changed, 7 insertions(+), 1 deletion(-)
>>
>> diff --git a/hw/virtio/virtio-pci.c b/hw/virtio/virtio-pci.c
>> index 7b91841..3b3b0e2 100644
>> --- a/hw/virtio/virtio-pci.c
>> +++ b/hw/virtio/virtio-pci.c
>> @@ -1416,7 +1416,8 @@ static const TypeInfo virtio_serial_pci_info = {
>>  static Property virtio_net_properties[] = {
>>      DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags,
>>                      VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, false),
>> -    DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors, 3),
>> +    DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors,
>> +                       DEV_NVECTORS_UNSPECIFIED),
>>      DEFINE_VIRTIO_NET_FEATURES(VirtIOPCIProxy, host_features),
>>      DEFINE_NIC_PROPERTIES(VirtIONetPCI, vdev.nic_conf),
>>      DEFINE_VIRTIO_NET_PROPERTIES(VirtIONetPCI, vdev.net_conf),
>> @@ -1428,6 +1429,11 @@ static int virtio_net_pci_init(VirtIOPCIProxy
>> *vpci_dev)
>>      DeviceState *qdev = DEVICE(vpci_dev);
>>      VirtIONetPCI *dev = VIRTIO_NET_PCI(vpci_dev);
>>      DeviceState *vdev = DEVICE(&dev->vdev);
>> +    VirtIONet *net = VIRTIO_NET(&dev->vdev);
>> +
>> +    if (vpci_dev->nvectors == DEV_NVECTORS_UNSPECIFIED) {
>> +        vpci_dev->nvectors = 2 * MAX(net->nic_conf.queues, 1) + 2;
>> +    }
>>
>>      virtio_net_set_config_size(&dev->vdev, vpci_dev->host_features);
>>      virtio_net_set_netclient_name(&dev->vdev, qdev->id,
>> --
>> 1.8.3.2
>>
> Good catch.
>
> Reviewed-by: Gonglei <arei.gonglei@huawei.com>
>
>
> Best regards,
> -Gonglei
>

Hi Gonglei:

Thanks for the reviewing. Looks like this patch has a bug which will
break the migration if "vectors=" is not specified.

Will post a new version.
Gonglei (Arei) March 4, 2014, 11:18 a.m. UTC | #4 
On 2014/3/4 15:41, Jason Wang wrote:

> On 03/04/2014 11:26 AM, Gonglei (Arei) wrote:
>>> -----Original Message-----
>>> From: qemu-devel-bounces+arei.gonglei=huawei.com@nongnu.org
>>> [mailto:qemu-devel-bounces+arei.gonglei=huawei.com@nongnu.org] On
>>> Behalf Of Jason Wang
>>> Sent: Friday, February 21, 2014 5:05 PM
>>> To: aliguori@amazon.com; mst@redhat.com; qemu-devel@nongnu.org
>>> Cc: Paolo Bonzini; Jason Wang
>>> Subject: [Qemu-devel] [PATCH V2] virtio-net: calculate proper msix vectors on
>>> init
>>>
>>> Currently, the default msix vectors for virtio-net-pci is 3 which is
>>> obvious not suitable for multiqueue guest, so we depends on the user
>>> or management tools to pass a correct vectors parameter. In fact, we
>>> can simplifying this by calculate the number of vectors on init.
>>>
>>> Consider we have N queues, the number of vectors needed is 2*N + 2
>>> (plus one config interrupt and control vq). We didn't check whether or
>>> not host support control vq because it was added unconditionally by
>>> qemu to avoid breaking legacy guests such as Minix.
>>>
>>> Cc: Paolo Bonzini <pbonzini@redhat.com>
>>> Cc: Michael S. Tsirkin <mst@redhat.com>
>>> Signed-off-by: Jason Wang <jasowang@redhat.com>
>>> ---
>>>  hw/virtio/virtio-pci.c | 8 +++++++-
>>>  1 file changed, 7 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/hw/virtio/virtio-pci.c b/hw/virtio/virtio-pci.c
>>> index 7b91841..3b3b0e2 100644
>>> --- a/hw/virtio/virtio-pci.c
>>> +++ b/hw/virtio/virtio-pci.c
>>> @@ -1416,7 +1416,8 @@ static const TypeInfo virtio_serial_pci_info = {
>>>  static Property virtio_net_properties[] = {
>>>      DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags,
>>>                      VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, false),
>>> -    DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors, 3),
>>> +    DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors,
>>> +                       DEV_NVECTORS_UNSPECIFIED),
>>>      DEFINE_VIRTIO_NET_FEATURES(VirtIOPCIProxy, host_features),
>>>      DEFINE_NIC_PROPERTIES(VirtIONetPCI, vdev.nic_conf),
>>>      DEFINE_VIRTIO_NET_PROPERTIES(VirtIONetPCI, vdev.net_conf),
>>> @@ -1428,6 +1429,11 @@ static int virtio_net_pci_init(VirtIOPCIProxy
>>> *vpci_dev)
>>>      DeviceState *qdev = DEVICE(vpci_dev);
>>>      VirtIONetPCI *dev = VIRTIO_NET_PCI(vpci_dev);
>>>      DeviceState *vdev = DEVICE(&dev->vdev);
>>> +    VirtIONet *net = VIRTIO_NET(&dev->vdev);
>>> +
>>> +    if (vpci_dev->nvectors == DEV_NVECTORS_UNSPECIFIED) {
>>> +        vpci_dev->nvectors = 2 * MAX(net->nic_conf.queues, 1) + 2;
>>> +    }
>>>
>>>      virtio_net_set_config_size(&dev->vdev, vpci_dev->host_features);
>>>      virtio_net_set_netclient_name(&dev->vdev, qdev->id,
>>> --
>>> 1.8.3.2
>>>
>> Good catch.
>>
>> Reviewed-by: Gonglei <arei.gonglei@huawei.com>
>>
>>
>> Best regards,
>> -Gonglei
>>
> 
> Hi Gonglei:
> 
> Thanks for the reviewing. Looks like this patch has a bug which will
> break the migration if "vectors=" is not specified.
> 
> Will post a new version.
> 


Yes, you are right, Jason.
You need to add backwards-compatibility properties in hw/pc.h

Best regards,
-Gonglei
diff mbox

Patch

diff --git a/hw/virtio/virtio-pci.c b/hw/virtio/virtio-pci.c
index 7b91841..3b3b0e2 100644
--- a/hw/virtio/virtio-pci.c
+++ b/hw/virtio/virtio-pci.c
@@ -1416,7 +1416,8 @@  static const TypeInfo virtio_serial_pci_info = {
 static Property virtio_net_properties[] = {
     DEFINE_PROP_BIT("ioeventfd", VirtIOPCIProxy, flags,
                     VIRTIO_PCI_FLAG_USE_IOEVENTFD_BIT, false),
-    DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors, 3),
+    DEFINE_PROP_UINT32("vectors", VirtIOPCIProxy, nvectors,
+                       DEV_NVECTORS_UNSPECIFIED),
     DEFINE_VIRTIO_NET_FEATURES(VirtIOPCIProxy, host_features),
     DEFINE_NIC_PROPERTIES(VirtIONetPCI, vdev.nic_conf),
     DEFINE_VIRTIO_NET_PROPERTIES(VirtIONetPCI, vdev.net_conf),
@@ -1428,6 +1429,11 @@  static int virtio_net_pci_init(VirtIOPCIProxy *vpci_dev)
     DeviceState *qdev = DEVICE(vpci_dev);
     VirtIONetPCI *dev = VIRTIO_NET_PCI(vpci_dev);
     DeviceState *vdev = DEVICE(&dev->vdev);
+    VirtIONet *net = VIRTIO_NET(&dev->vdev);
+
+    if (vpci_dev->nvectors == DEV_NVECTORS_UNSPECIFIED) {
+        vpci_dev->nvectors = 2 * MAX(net->nic_conf.queues, 1) + 2;
+    }
 
     virtio_net_set_config_size(&dev->vdev, vpci_dev->host_features);
     virtio_net_set_netclient_name(&dev->vdev, qdev->id,