Re: [PATCH net-next v2 3/4] virtio_net: Map NAPIs to queues

[Gerhard Engleder <gerhard@engleder-embedded.com>]

On 16.01.25 17:09, Joe Damato wrote:
> On Thu, Jan 16, 2025 at 03:53:14PM +0800, Xuan Zhuo wrote:
>> On Thu, 16 Jan 2025 05:52:58 +0000, Joe Damato <jdamato@fastly.com> wrote:
>>> Use netif_queue_set_napi to map NAPIs to queue IDs so that the mapping
>>> can be accessed by user apps.
>>>
>>> $ ethtool -i ens4 | grep driver
>>> driver: virtio_net
>>>
>>> $ sudo ethtool -L ens4 combined 4
>>>
>>> $ ./tools/net/ynl/pyynl/cli.py \
>>>         --spec Documentation/netlink/specs/netdev.yaml \
>>>         --dump queue-get --json='{"ifindex": 2}'
>>> [{'id': 0, 'ifindex': 2, 'napi-id': 8289, 'type': 'rx'},
>>>   {'id': 1, 'ifindex': 2, 'napi-id': 8290, 'type': 'rx'},
>>>   {'id': 2, 'ifindex': 2, 'napi-id': 8291, 'type': 'rx'},
>>>   {'id': 3, 'ifindex': 2, 'napi-id': 8292, 'type': 'rx'},
>>>   {'id': 0, 'ifindex': 2, 'type': 'tx'},
>>>   {'id': 1, 'ifindex': 2, 'type': 'tx'},
>>>   {'id': 2, 'ifindex': 2, 'type': 'tx'},
>>>   {'id': 3, 'ifindex': 2, 'type': 'tx'}]
>>>
>>> Note that virtio_net has TX-only NAPIs which do not have NAPI IDs, so
>>> the lack of 'napi-id' in the above output is expected.
>>>
>>> Signed-off-by: Joe Damato <jdamato@fastly.com>
>>> ---
>>>   v2:
>>>     - Eliminate RTNL code paths using the API Jakub introduced in patch 1
>>>       of this v2.
>>>     - Added virtnet_napi_disable to reduce code duplication as
>>>       suggested by Jason Wang.
>>>
>>>   drivers/net/virtio_net.c | 34 +++++++++++++++++++++++++++++-----
>>>   1 file changed, 29 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
>>> index cff18c66b54a..c6fda756dd07 100644
>>> --- a/drivers/net/virtio_net.c
>>> +++ b/drivers/net/virtio_net.c
>>> @@ -2803,9 +2803,18 @@ static void virtnet_napi_do_enable(struct virtqueue *vq,
>>>   	local_bh_enable();
>>>   }
>>>
>>> -static void virtnet_napi_enable(struct virtqueue *vq, struct napi_struct *napi)
>>> +static void virtnet_napi_enable(struct virtqueue *vq,
>>> +				struct napi_struct *napi)
>>>   {
>>> +	struct virtnet_info *vi = vq->vdev->priv;
>>> +	int q = vq2rxq(vq);
>>> +	u16 curr_qs;
>>> +
>>>   	virtnet_napi_do_enable(vq, napi);
>>> +
>>> +	curr_qs = vi->curr_queue_pairs - vi->xdp_queue_pairs;
>>> +	if (!vi->xdp_enabled || q < curr_qs)
>>> +		netif_queue_set_napi(vi->dev, q, NETDEV_QUEUE_TYPE_RX, napi);
>>
>> So what case the check of xdp_enabled is for?
> 
> Based on a previous discussion [1], the NAPIs should not be linked
> for in-kernel XDP, but they _should_ be linked for XSK.
> 
> I could certainly have misread the virtio_net code (please let me
> know if I've gotten it wrong, I'm not an expert), but the three
> cases I have in mind are:
> 
>    - vi->xdp_enabled = false, which happens when no XDP is being
>      used, so the queue number will be < vi->curr_queue_pairs.
> 
>    - vi->xdp_enabled = false, which I believe is what happens in the
>      XSK case. In this case, the NAPI is linked.
> 
>    - vi->xdp_enabled = true, which I believe only happens for
>      in-kernel XDP - but not XSK - and in this case, the NAPI should
>      NOT be linked.

My interpretation based on [1] is that an in-kernel XDP Tx queue is a
queue that is only used if XDP is attached and is not visible to
userspace. The in-kernel XDP Tx queue is used to not load stack Tx
queues with XDP packets. IIRC fbnic has additional queues only for
XDP Tx. So for stack RX queues I would always link napi, no matter if
XDP is attached or not. I think most driver do not have in-kernel XDP
Tx queues. But I'm also not an expert.

Gerhard