[PATCH v4] virtio_net: disable cb when NAPI is busy-polled

[PATCH v4] virtio_net: disable cb when NAPI is busy-polled

[Longjun Tang]

From: Longjun Tang <tanglongjun@kylinos.cn>

When busy-poll is active, napi_schedule_prep() returns false in
virtqueue_napi_schedule(), so virtqueue_disable_cb() is skipped.
The device may keep firing irqs until reaches virtqueue_napi_complete().
Under load (received == budget), it will lead to a large number
of spurious interrupts.

Fix it by disabling the callback at the virtnet_poll() entry.
This keeps the callback off while we poll and it is re-enabled by
virtqueue_napi_complete() when going idle.

Fixes: ceef438d613f ("virtio_net: remove custom busy_poll")
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Longjun Tang <tanglongjun@kylinos.cn>

---
V1 -> V2: Remain agnostic to busy polling
V2 -> V3: Add fixes tag
V3 -> V4: Update commit message and remove some comments
---
 drivers/net/virtio_net.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index f4adcfee7a80..569e4db187d1 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -3008,6 +3008,8 @@ static int virtnet_poll(struct napi_struct *napi, int budget)
 	unsigned int xdp_xmit = 0;
 	bool napi_complete;
 
+	virtqueue_disable_cb(rq->vq);
+
 	virtnet_poll_cleantx(rq, budget);
 
 	received = virtnet_receive(rq, budget, &xdp_xmit);
-- 
2.43.0

Re: [PATCH v4] virtio_net: disable cb when NAPI is busy-polled

[Simon Horman]

This is an AI-generated review of your patch. The human sending this
email has considered the AI review valid, or at least plausible.
Full review at: https://sashiko.dev
---
virtio_net: disable cb when NAPI is busy-polled

This commit disables the virtqueue callback at the entry of virtnet_poll().
It aims to prevent a large number of spurious interrupts when busy-poll is
active by keeping the callback off while polling.

> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> index 26afa6341d161..c1e252400c0fc 100644
> --- a/drivers/net/virtio_net.c
> +++ b/drivers/net/virtio_net.c
> @@ -3011,6 +3011,8 @@ static int virtnet_poll(struct napi_struct *napi, int budget)
>  	unsigned int xdp_xmit = 0;
>  	bool napi_complete;
>  
> +	virtqueue_disable_cb(rq->vq);
> +

[Severity: High]
Can this unconditionally disable the RX callback and cause a permanent network
stall when polled by netpoll?

If the netpoll subsystem (e.g., used by netconsole) calls the NAPI poll
function directly with a budget of 0, virtqueue_disable_cb() is called
immediately. virtnet_receive() will then process 0 packets and return 0.

Later in virtnet_poll(), the completion condition is checked:

	if (received < budget) {
		napi_complete = virtqueue_napi_complete(napi, rq->vq, received);
        ...
	}

When both received and budget are 0, this condition evaluates to false. This
skips virtqueue_napi_complete(), leaving the RX callback permanently disabled
and preventing the device from raising further RX interrupts.

>  	virtnet_poll_cleantx(rq, budget);
>  
>  	received = virtnet_receive(rq, budget, &xdp_xmit);