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;
MIME-Version: 1.0
References: <20221222060427.21626-1-jasowang@redhat.com> <20221222060427.21626-5-jasowang@redhat.com>
 <CAJs=3_D6sug80Bb9tnAw5T0_NaL_b=u8ZMcwZtd-dy+AH_yqzQ@mail.gmail.com>
In-Reply-To: <CAJs=3_D6sug80Bb9tnAw5T0_NaL_b=u8ZMcwZtd-dy+AH_yqzQ@mail.gmail.com>
From:   Jason Wang <jasowang@redhat.com>
Date:   Thu, 22 Dec 2022 16:43:09 +0800
Message-ID: <CACGkMEv4YxuqrSx_HW2uWgXXSMOFCzTJCCD_EVhMwegsL8SoCg@mail.gmail.com>
Subject: Re: [RFC PATCH 4/4] virtio-net: sleep instead of busy waiting for cvq command
To:     Alvaro Karsz <alvaro.karsz@solid-run.com>
Cc:     mst@redhat.com, davem@davemloft.net, edumazet@google.com,
        kuba@kernel.org, pabeni@redhat.com,
        virtualization@lists.linux-foundation.org, netdev@vger.kernel.org,
        linux-kernel@vger.kernel.org, maxime.coquelin@redhat.com,
        eperezma@redhat.com
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi Alvaro:

On Thu, Dec 22, 2022 at 2:44 PM Alvaro Karsz <alvaro.karsz@solid-run.com> wrote:
>
> Hi Jason,
>
> Adding timeout to the cvq is a great idea IMO.
>
> > -       /* Spin for a response, the kick causes an ioport write, trapping
> > -        * into the hypervisor, so the request should be handled immediately.
> > -        */
> > -       while (!virtqueue_get_buf(vi->cvq, &tmp) &&
> > -              !virtqueue_is_broken(vi->cvq))
> > -               cpu_relax();
> > +       virtqueue_wait_for_used(vi->cvq, &tmp);
>
> Do you think that we should continue like nothing happened in case of a timeout?

We could, but we should not depend on a device to do this since it's
not reliable. More below.

> Shouldn't we reset the device?

We can't depend on device, there's probably another loop in reset():

E.g in vp_reset() we had:

        while (vp_modern_get_status(mdev))
                msleep(1);

> What happens if a device completes the control command after timeout?

Maybe we could have a BAD_RING() here in this case (and more check in
vq->broken in this case).

Thanks

>
> Thanks
>
> Alvaro
>