Date: Mon, 22 Jan 2024 15:57:08 +0800
From: Jason Wang <jasowang@...hat.com>
To: Xuan Zhuo <xuanzhuo@...ux.alibaba.com>
Cc: Andrew Lunn <andrew@...n.ch>, Heng Qi <hengqi@...ux.alibaba.com>, 
	Paolo Abeni <pabeni@...hat.com>, Zhu Yanjun <yanjun.zhu@...el.com>, mst@...hat.com, 
	davem@...emloft.net, edumazet@...gle.com, kuba@...nel.org, 
	virtualization@...ts.linux.dev, netdev@...r.kernel.org, 
	Zhu Yanjun <yanjun.zhu@...ux.dev>
Subject: Re: [PATCH 1/1] virtio_net: Add timeout handler to avoid kernel hang

On Mon, Jan 22, 2024 at 3:36 PM Xuan Zhuo <xuanzhuo@...ux.alibaba.com> wrote:
>
> On Mon, 22 Jan 2024 15:19:12 +0800, Jason Wang <jasowang@...hat.com> wrote:
> > On Mon, Jan 22, 2024 at 3:07 PM Xuan Zhuo <xuanzhuo@...ux.alibaba.com> wrote:
> > >
> > > On Mon, 22 Jan 2024 14:58:09 +0800, Jason Wang <jasowang@...hat.com> wrote:
> > > > On Mon, Jan 22, 2024 at 2:55 PM Jason Wang <jasowang@...hat.com> wrote:
> > > > >
> > > > > On Mon, Jan 22, 2024 at 2:20 PM Xuan Zhuo <xuanzhuo@...ux.alibaba.com> wrote:
> > > > > >
> > > > > > On Mon, 22 Jan 2024 12:16:27 +0800, Jason Wang <jasowang@...hat.com> wrote:
> > > > > > > On Mon, Jan 22, 2024 at 12:00 PM Xuan Zhuo <xuanzhuo@...ux.alibaba.com> wrote:
> > > > > > > >
> > > > > > > > On Mon, 22 Jan 2024 11:14:30 +0800, Jason Wang <jasowang@...hat.com> wrote:
> > > > > > > > > On Mon, Jan 22, 2024 at 10:12 AM Zhu Yanjun <yanjun.zhu@...ux.dev> wrote:
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > 在 2024/1/20 1:29, Andrew Lunn 写道:
> > > > > > > > > > >>>>>        while (!virtqueue_get_buf(vi->cvq, &tmp) &&
> > > > > > > > > > >>>>> -           !virtqueue_is_broken(vi->cvq))
> > > > > > > > > > >>>>> +           !virtqueue_is_broken(vi->cvq)) {
> > > > > > > > > > >>>>> +        if (timeout)
> > > > > > > > > > >>>>> +            timeout--;
> > > > > > > > > > >>>> This is not really a timeout, just a loop counter. 200 iterations could
> > > > > > > > > > >>>> be a very short time on reasonable H/W. I guess this avoid the soft
> > > > > > > > > > >>>> lockup, but possibly (likely?) breaks the functionality when we need to
> > > > > > > > > > >>>> loop for some non negligible time.
> > > > > > > > > > >>>>
> > > > > > > > > > >>>> I fear we need a more complex solution, as mentioned by Micheal in the
> > > > > > > > > > >>>> thread you quoted.
> > > > > > > > > > >>> Got it. I also look forward to the more complex solution to this problem.
> > > > > > > > > > >> Can we add a device capability (new feature bit) such as ctrq_wait_timeout
> > > > > > > > > > >> to get a reasonable timeout？
> > > > > > > > > > > The usual solution to this is include/linux/iopoll.h. If you can sleep
> > > > > > > > > > > read_poll_timeout() otherwise read_poll_timeout_atomic().
> > > > > > > > > >
> > > > > > > > > > I read carefully the functions read_poll_timeout() and
> > > > > > > > > > read_poll_timeout_atomic(). The timeout is set by the caller of the 2
> > > > > > > > > > functions.
> > > > > > > > >
> > > > > > > > > FYI, in order to avoid a swtich of atomic or not, we need convert rx
> > > > > > > > > mode setting to workqueue first:
> > > > > > > > >
> > > > > > > > > https://www.mail-archive.com/virtualization@lists.linux-foundation.org/msg60298.html
> > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > As such, can we add a module parameter to customize this timeout value
> > > > > > > > > > by the user?
> > > > > > > > >
> > > > > > > > > Who is the "user" here, or how can the "user" know the value?
> > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > Or this timeout value is stored in device register, virtio_net driver
> > > > > > > > > > will read this timeout value at initialization?
> > > > > > > > >
> > > > > > > > > See another thread. The design needs to be general, or you can post a RFC.
> > > > > > > > >
> > > > > > > > > In another thought, we've already had a tx watchdog, maybe we can have
> > > > > > > > > something similar to cvq and use timeout + reset in that case.
> > > > > > > >
> > > > > > > > But we may block by the reset ^_^ if the device is broken?
> > > > > > >
> > > > > > > I mean vq reset here.
> > > > > >
> > > > > > I see.
> > > > > >
> > > > > > I mean when the deivce is broken, the vq reset also many be blocked.
> > > > > >
> > > > > >         void vp_modern_set_queue_reset(struct virtio_pci_modern_device *mdev, u16 index)
> > > > > >         {
> > > > > >                 struct virtio_pci_modern_common_cfg __iomem *cfg;
> > > > > >
> > > > > >                 cfg = (struct virtio_pci_modern_common_cfg __iomem *)mdev->common;
> > > > > >
> > > > > >                 vp_iowrite16(index, &cfg->cfg.queue_select);
> > > > > >                 vp_iowrite16(1, &cfg->queue_reset);
> > > > > >
> > > > > >                 while (vp_ioread16(&cfg->queue_reset))
> > > > > >                         msleep(1);
> > > > > >
> > > > > >                 while (vp_ioread16(&cfg->cfg.queue_enable))
> > > > > >                         msleep(1);
> > > > > >         }
> > > > > >         EXPORT_SYMBOL_GPL(vp_modern_set_queue_reset);
> > > > > >
> > > > > > In this function, for the broken device, we can not expect something.
> > > > >
> > > > > Yes, it's best effort, there's no guarantee then. But it doesn't harm to try.
> > > > >
> > > > > Thanks
> > > > >
> > > > > >
> > > > > >
> > > > > > >
> > > > > > > It looks like we have multiple goals here
> > > > > > >
> > > > > > > 1) avoid lockups, using workqueue + cond_resched() seems to be
> > > > > > > sufficient, it has issue but nothing new
> > > > > > > 2) recover from the unresponsive device, the issue for timeout is that
> > > > > > > it needs to deal with false positives
> > > > > >
> > > > > >
> > > > > > I agree.
> > > > > >
> > > > > > But I want to add a new goal, cvq async. In the netdim, we will
> > > > > > send many requests via the cvq, so the cvq async will be nice.
> > > >
> > > > Then you need an interrupt for cvq.
> > > >
> > > > FYI, I've posted a series that use interrupt for cvq in the past:
> > > >
> > > > https://lore.kernel.org/lkml/6026e801-6fda-fee9-a69b-d06a80368621@redhat.com/t/
> > >
> > > I know this. But the interrupt maybe not a good solution without new space.
> >
> > What do you mean by "new space"?
>
> Yes, I know, the cvq can work with interrupt by the virtio spec.
> But as I know, many hypervisors implement the cvq without supporting interrupt.

It's a bug of the hypervisor that needs to be fix. Interrupt is
provided by transport not the virtio itself.

Otherwise it can only support for Linux but not other OSes.

> If we let the cvq work with interrupt without negotiation then
> many hypervisors will hang on the new kernel.
>
> >
> > We can introduce something like enable_cb_delayed(), then you will
> > only get notified after several requests.
> >
> > >
> > > >
> > > > Haven't found time in working on this anymore, maybe we can start from
> > > > this or not.
> > >
> > >
> > > I said async, but my aim is to put many requests to the cvq before getting the
> > > response.
> >
> > It doesn't differ from TX/RX in this case.
> >
> > >
> > > Heng Qi posted this https://lore.kernel.org/all/1705410693-118895-4-git-send-email-hengqi@linux.alibaba.com/
> > >
> >
> > This seems like a hack, if interrupt is used, you can simply do that
> > in the callback.
>
> YES.
>
> I also want to change the code, I just want to say the async is a goal.
>
> For the rx mode, we have introduce a work queue, I want to move the
> sending command job to the work queue. The caller just wakeup
> the work queue.
>
> If the caller want to got the result sync, then the caller can wait for it.
> If not, the caller can register an function to the work queue.
>
> And I think it will be easy to implement the timeout inside the workqueue.

Looks much more complicated than a simple interrupt + timer/watchdog etc.

Thanks

>
> Thanks.
>
>
> >
> > Thanks
> >
> > > Thanks.
> > >
> > >
> > > >
> > > > Thanks
> > > >
> > > > > >
> > > > > > Thanks.
> > > > > >
> > > > > >
> > > > > > >
> > > > > > > Thanks
> > > > > > >
> > > > > > > >
> > > > > > > > Thanks.
> > > > > > > >
> > > > > > > >
> > > > > > > > >
> > > > > > > > > Thans
> > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > Zhu Yanjun
> > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > >       Andrew
> > > > > > > > > >
> > > > > > > > >
> > > > > > > >
> > > > > > >
> > > > > >
> > > >
> > >
> >
>

Powered by blists - more mailing lists