Re: [PATCH] drm/virtio: implement virtio_gpu_shutdown

[Maxime Ripard <mripard@kernel.org>]

[-- Attachment #1: Type: text/plain, Size: 2052 bytes --]

On Sun, May 18, 2025 at 05:53:59PM -0400, Michael S. Tsirkin wrote:
> On Tue, May 13, 2025 at 12:18:44PM +0200, Gerd Hoffmann wrote:
> > > > diff --git a/drivers/gpu/drm/virtio/virtgpu_drv.c b/drivers/gpu/drm/virtio/virtgpu_drv.c
> > > > index e32e680c7197..71c6ccad4b99 100644
> > > > --- a/drivers/gpu/drm/virtio/virtgpu_drv.c
> > > > +++ b/drivers/gpu/drm/virtio/virtgpu_drv.c
> > > > @@ -130,10 +130,10 @@ static void virtio_gpu_remove(struct virtio_device *vdev)
> > > >  
> > > >  static void virtio_gpu_shutdown(struct virtio_device *vdev)
> > > >  {
> > > > -	/*
> > > > -	 * drm does its own synchronization on shutdown.
> > > > -	 * Do nothing here, opt out of device reset.
> > > > -	 */
> > > > +	struct drm_device *dev = vdev->priv;
> > > > +
> > > > +	/* stop talking to the device */
> > > > +	drm_dev_unplug(dev);
> > > 
> > > I'm not necessarily opposed to using drm_dev_unplug() here, but it's
> > > still pretty surprising to me. It's typically used in remove, not
> > > shutdown. The typical helper to use at shutdown is
> > > drm_atomic_helper_shutdown.
> > > 
> > > So if the latter isn't enough or wrong, we should at least document why.
> > 
> > The intention of this is to make sure the driver stops talking to the
> > device (as the comment already says).
> > 
> > There are checks in place in the virt queue functions which will make
> > sure the driver will not try place new requests in the queues after
> > drm_dev_unplug() has been called.  Which why I decided to implement it
> > that way.
> > 
> > drm_atomic_helper_shutdown() tears down all outputs according to the
> > documentation.  Which is something different.  I don't think calling
> > drm_atomic_helper_shutdown() will do what I need here.  Calling it in
> > addition to drm_dev_unplug() might make sense, not sure.
> > 
> > Suggestions are welcome.
> > 
> > take care,
> >   Gerd
> 
> 
> I suggest adding comments in code explaining why it's approriate here.
> Want to try?

Yes, that would be great

Maxime

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 273 bytes --]