Re: [PATCH 2/3] vdpa/mlx5: Only return vq ready if vq is initialized

From: Eli Cohen
Date: Wed Jun 02 2021 - 03:44:12 EST


On Wed, Jun 02, 2021 at 03:29:52PM +0800, Jason Wang wrote:
>
> 在 2021/6/1 下午6:21, Eli Cohen 写道:
> > On Tue, Jun 01, 2021 at 01:03:36PM +0800, Jason Wang wrote:
> > > 在 2021/6/1 下午12:13, Eli Cohen 写道:
> > > > On Tue, Jun 01, 2021 at 10:18:04AM +0800, Jason Wang wrote:
> > > > > 在 2021/6/1 上午12:04, Eli Cohen 写道:
> > > > > > Only return the value of the ready field if the VQ is initialized in
> > > > > > which case the value of the field is valid.
> > > > > >
> > > > > > Failing to do so can result in virtio_vdpa failing to load if the device
> > > > > > was previously used by vhost_vdpa and the old values are ready.
> > > > > > virtio_vdpa expects to find VQs in "not ready" state.
> > > > > >
> > > > > > Fixes: 1a86b377aa21 ("vdpa/mlx5: Add VDPA driver for supported mlx5 devices")
> > > > > > Signed-off-by: Eli Cohen <elic@xxxxxxxxxx>
> > > > > > ---
> > > > > > drivers/vdpa/mlx5/net/mlx5_vnet.c | 2 +-
> > > > > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > > > > >
> > > > > > diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > > > > > index 02a05492204c..f6b680d2ab1c 100644
> > > > > > --- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > > > > > +++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > > > > > @@ -1407,7 +1407,7 @@ static bool mlx5_vdpa_get_vq_ready(struct vdpa_device *vdev, u16 idx)
> > > > > > struct mlx5_vdpa_net *ndev = to_mlx5_vdpa_ndev(mvdev);
> > > > > > struct mlx5_vdpa_virtqueue *mvq = &ndev->vqs[idx];
> > > > > > - return mvq->ready;
> > > > > > + return mvq->initialized && mvq->ready;
> > > > > I think the more suitable fix is to reset mvq->ready during reset. The
> > > > > vq_ready should follow the queue_enable semantic in virtio-pci:
> > > > >
> > > > > "
> > > > > The device MUST present a 0 in queue_enable on reset.
> > > > > "
> > > > Thinking again, I think we should call set_vq_ready() from
> > > > qemu/virtio_vdpa etc. after reset to explicitly set ready to false.
> > >
> > > This is not what I read from the spec and how the current driver behave.
> > >
> > > And I don't see why we need to stick to 1 after the reset.
> > Like you said below, this must be clarified by the spec.
>
>
> I will add this to my todo list.

Thanks.

>
>
> > Because my
> > understanding of the text is that the ready flag is a means for the
> > driver to tell the device whether it may or may not execute requests. So
> > following this reasoning, the driver should tell the device when ready
> > should be zero. As you suggest, after reset.
>
>
> Another thought, reset should prevent the device from executing requests on
> each queue. In this case, presenting vq_ready to zero make sense. Otherwise
> there would be a conflict.
>

Right.

> Thanks
>
>
> >
> > Meanwhile I am going to change the code to reset ready after device
> > reset.
> >
> > >
> > > > The ready indication is not necessairily a reflection of the hardware
> > > > queue:
> > > >
> > > > "Virtual queue ready bit
> > > > Writing one (0x1) to this register notifies the device that it can
> > > > execute requests from this virtual queue. Reading from this register
> > > > returns the last value written to it. Both read and write accesses apply
> > > > to the queue selected by writing to QueueSel."
> > >
> > > My understanding that this applies if not reset in the middle. We can
> > > clarify this in the spec if needed.
> > >
> > > Thanks
> > >
> > >
> > > >
> > > > > Thanks
> > > > >
> > > > >
> > > > > > }
> > > > > > static int mlx5_vdpa_set_vq_state(struct vdpa_device *vdev, u16 idx,
>