[virtio-dev] Re: [PATCH v4 2/3] virtio: pci support virtqueue reset

[Cornelia Huck <cohuck@redhat.com>]

On Thu, Sep 30 2021, Jason Wang <jasowang@redhat.com> wrote:

> On Thu, Sep 30, 2021 at 12:33 AM Cornelia Huck <cohuck@redhat.com> wrote:
>>
>> On Wed, Sep 29 2021, Xuan Zhuo <xuanzhuo@linux.alibaba.com> wrote:
>>
>> > On Tue, 28 Sep 2021 12:20:46 +0200, Cornelia Huck <cohuck@redhat.com> wrote:
>> >> On Tue, Sep 28 2021, Xuan Zhuo <xuanzhuo@linux.alibaba.com> wrote:
>> >> > +The device MUST reset the queue when 1 is written to \field{queue_reset}, and
>> >> > +present a 1 in \field{queue_reset} after the queue has been reset, until the
>> >> > +driver re-enables the queue via \field{queue_enable}. (see \ref{sec:Virtqueues / Virtqueue Reset}).
>> >>
>> >> "...or the device is reset." ?
>> >>
>> >
>> >     The device MUST reset the queue when 1 is written to \field{queue_reset}, and
>> >     present a 1 in \field{queue_reset} after the queue has been reset, until the
>> >     driver re-enables the queue via \field{queue_enable} or the device is reset. (see \ref{sec:Virtqueues / Virtqueue Reset}).
>> >
>> > Is that so? Doesn't it feel necessary?
>>
>> Is queue_reset supposed to persist across device reset? That feels a bit
>> odd to me.
>
> I think it's better to follow e.g the pci "status".
>
> Driver writes 1 to queue_reset. And the device notified the completion
> by presenting 0. This looks easier to be dealt with during device
> reset.

Indeed, that might be more clear, let's go with that (same with MMIO.)

>
>
>>
>> >
>> >
>> >> Maybe also
>> >>
>> >> "The device MAY change the value of \field{queue_size} if the queue has
>> >> been reset." ?
>> >>
>> >> Should it always set that field to the currently maximum supported queue
>> >> size (assuming that can change dynamically)? Do we need some kind of
>> >> synchronization for those changes?
>> >
>> > When the queue is reset, all states of this queue MUST be modified to the
>> > initial value. For example, queue_size MUST be reset to the maximum value
>> > supported by the device. Because in the last reset queue or the entire device
>> > reset process, the driver will modify the queue_size of the device so that its
>> > value may be less than the maximum value.
>>
>> I think the question is whether the device may choose a different
>> initial maximum value once the queue has been reset. If it does change
>> the value, there may be a race where the driver has reset the queue,
>> read queue_reset back, read the max queue size, and the device only then
>> changing the max queue size. Maybe I'm overthinking this.
>
> I think the driver should do the exact same steps as the device initialization.
>
> Not sure it's worth mentioning here.

Hm, we might have similar problems there... but they are probably not
really problems in practice. So just a note that the device may fill in
the fields with different values might be enough. Like

"The device MAY present different default values after queue reset."

?


---------------------------------------------------------------------
To unsubscribe, e-mail: virtio-dev-unsubscribe@lists.oasis-open.org
For additional commands, e-mail: virtio-dev-help@lists.oasis-open.org