2021-06-18 09:13:21

by Keiichi Watanabe

[permalink] [raw]
Subject: [PATCH] virtio_net: Enable MSI-X vector for ctrl queue

When we use vhost-user backend on the host, MSI-X vector should be set
so that the vmm can get an irq FD and send it to the backend device
process with vhost-user protocol.
Since whether the vector is set for a queue is determined depending on
the queue has a callback, this commit sets an empty callback for
virtio-net's control queue.

Signed-off-by: Keiichi Watanabe <[email protected]>
---
drivers/net/virtio_net.c | 7 ++++++-
1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 11f722460513..002e3695d4b3 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -2696,6 +2696,11 @@ static void virtnet_del_vqs(struct virtnet_info *vi)
virtnet_free_queues(vi);
}

+static void virtnet_ctrlq_done(struct virtqueue *rvq)
+{
+ /* Do nothing */
+}
+
/* How large should a single buffer be so a queue full of these can fit at
* least one full packet?
* Logic below assumes the mergeable buffer header is used.
@@ -2748,7 +2753,7 @@ static int virtnet_find_vqs(struct virtnet_info *vi)

/* Parameters for control virtqueue, if any */
if (vi->has_cvq) {
- callbacks[total_vqs - 1] = NULL;
+ callbacks[total_vqs - 1] = virtnet_ctrlq_done;
names[total_vqs - 1] = "control";
}

--
2.32.0.288.g62a8d224e6-goog


2021-06-18 16:06:58

by Michael S. Tsirkin

[permalink] [raw]
Subject: Re: [PATCH] virtio_net: Enable MSI-X vector for ctrl queue

On Fri, Jun 18, 2021 at 04:26:25PM +0900, Keiichi Watanabe wrote:
> When we use vhost-user backend on the host, MSI-X vector should be set
> so that the vmm can get an irq FD and send it to the backend device
> process with vhost-user protocol.
> Since whether the vector is set for a queue is determined depending on
> the queue has a callback, this commit sets an empty callback for
> virtio-net's control queue.
>
> Signed-off-by: Keiichi Watanabe <[email protected]>

I'm confused by this explanation. If the vmm wants to get
an interrupt it can do so - why change the guest driver?

> ---
> drivers/net/virtio_net.c | 7 ++++++-
> 1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> index 11f722460513..002e3695d4b3 100644
> --- a/drivers/net/virtio_net.c
> +++ b/drivers/net/virtio_net.c
> @@ -2696,6 +2696,11 @@ static void virtnet_del_vqs(struct virtnet_info *vi)
> virtnet_free_queues(vi);
> }
>
> +static void virtnet_ctrlq_done(struct virtqueue *rvq)
> +{
> + /* Do nothing */
> +}
> +
> /* How large should a single buffer be so a queue full of these can fit at
> * least one full packet?
> * Logic below assumes the mergeable buffer header is used.
> @@ -2748,7 +2753,7 @@ static int virtnet_find_vqs(struct virtnet_info *vi)
>
> /* Parameters for control virtqueue, if any */
> if (vi->has_cvq) {
> - callbacks[total_vqs - 1] = NULL;
> + callbacks[total_vqs - 1] = virtnet_ctrlq_done;
> names[total_vqs - 1] = "control";
> }
>
> --
> 2.32.0.288.g62a8d224e6-goog

2021-06-21 06:23:36

by Jason Wang

[permalink] [raw]
Subject: Re: [PATCH] virtio_net: Enable MSI-X vector for ctrl queue


?? 2021/6/18 ????8:38, Michael S. Tsirkin ะด??:
> On Fri, Jun 18, 2021 at 04:26:25PM +0900, Keiichi Watanabe wrote:
>> When we use vhost-user backend on the host, MSI-X vector should be set
>> so that the vmm can get an irq FD and send it to the backend device
>> process with vhost-user protocol.
>> Since whether the vector is set for a queue is determined depending on
>> the queue has a callback, this commit sets an empty callback for
>> virtio-net's control queue.
>>
>> Signed-off-by: Keiichi Watanabe <[email protected]>
> I'm confused by this explanation. If the vmm wants to get
> an interrupt it can do so - why change the guest driver?


+1, it sounds like a bug in the backend or we probably need more context
here.

Thanks


>
>> ---
>> drivers/net/virtio_net.c | 7 ++++++-
>> 1 file changed, 6 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
>> index 11f722460513..002e3695d4b3 100644
>> --- a/drivers/net/virtio_net.c
>> +++ b/drivers/net/virtio_net.c
>> @@ -2696,6 +2696,11 @@ static void virtnet_del_vqs(struct virtnet_info *vi)
>> virtnet_free_queues(vi);
>> }
>>
>> +static void virtnet_ctrlq_done(struct virtqueue *rvq)
>> +{
>> + /* Do nothing */
>> +}
>> +
>> /* How large should a single buffer be so a queue full of these can fit at
>> * least one full packet?
>> * Logic below assumes the mergeable buffer header is used.
>> @@ -2748,7 +2753,7 @@ static int virtnet_find_vqs(struct virtnet_info *vi)
>>
>> /* Parameters for control virtqueue, if any */
>> if (vi->has_cvq) {
>> - callbacks[total_vqs - 1] = NULL;
>> + callbacks[total_vqs - 1] = virtnet_ctrlq_done;
>> names[total_vqs - 1] = "control";
>> }
>>
>> --
>> 2.32.0.288.g62a8d224e6-goog