Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <d05564fd4fc47ed212be2c02be4d7c05167fa2b3.camel@redhat.com>
Subject: Re: [PATCH v3] virtio_net: suppress cpu stall when free_unused_bufs
From:   Paolo Abeni <pabeni@redhat.com>
To:     Wenliang Wang <wangwenliang.1995@bytedance.com>, mst@redhat.com,
        jasowang@redhat.com, davem@davemloft.net, edumazet@google.com,
        kuba@kernel.org, zhengqi.arch@bytedance.com,
        willemdebruijn.kernel@gmail.com
Cc:     virtualization@lists.linux-foundation.org, netdev@vger.kernel.org,
        linux-kernel@vger.kernel.org
Date:   Tue, 02 May 2023 11:05:40 +0200
In-Reply-To: <1682783278-12819-1-git-send-email-wangwenliang.1995@bytedance.com>
References: <1682783278-12819-1-git-send-email-wangwenliang.1995@bytedance.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.46.4 (3.46.4-1.fc37) 
MIME-Version: 1.0
Precedence: bulk

On Sat, 2023-04-29 at 23:47 +0800, Wenliang Wang wrote:
> For multi-queue and large ring-size use case, the following error
> occurred when free_unused_bufs:
> rcu: INFO: rcu_sched self-detected stall on CPU.
>=20
> Signed-off-by: Wenliang Wang <wangwenliang.1995@bytedance.com>

Net next is currently closed, but this patch could arguably land on the
net tree - assuming Micheal agrees.

In that case you should include a suitable Fixes tag - reposting a new
version.

> ---
> v2:
> -add need_resched check.
> -apply same logic to sq.
> v3:
> -use cond_resched instead.
> ---
>  drivers/net/virtio_net.c | 2 ++
>  1 file changed, 2 insertions(+)
>=20
> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> index ea1bd4bb326d..744bdc8a1abd 100644
> --- a/drivers/net/virtio_net.c
> +++ b/drivers/net/virtio_net.c
> @@ -3559,12 +3559,14 @@ static void free_unused_bufs(struct virtnet_info =
*vi)
>  		struct virtqueue *vq =3D vi->sq[i].vq;
>  		while ((buf =3D virtqueue_detach_unused_buf(vq)) !=3D NULL)
>  			virtnet_sq_free_unused_buf(vq, buf);
> +		cond_resched();

Note that on v1 Xuan Zhuo suggested to do the above check only every a
low fixed number of buffer. I think the current code is a better
option, as the need_resched() check is very cheap and the overall code
simpler.

In any case, when submitting a new revision, please explicitly CC
whoever gave feedback on previous ones.

Thanks,

Paolo