Re: [PATCH v6 21/23] drm: rockchip: Add VOP2 driver

[Sascha Hauer <s.hauer@pengutronix.de>]

On Fri, Feb 18, 2022 at 11:50:32AM +0800, Andy Yan wrote:
> Hi Sascha:
> 
> On 2/17/22 22:06, Heiko Stübner wrote:
> > Am Donnerstag, 17. Februar 2022, 14:58:23 CET schrieb Sascha Hauer:
> > > Hi Andy,
> > > 
> > > Please trim the context in your answers to the relevant parts, it makes
> > > it easier to find the things you said.
> > > 
> > > On Thu, Feb 17, 2022 at 08:00:11PM +0800, Andy Yan wrote:
> > > > Hi Sascha:
> > > > 
> > > > > +
> > > > > +	drm_for_each_encoder_mask(encoder, crtc->dev, crtc_state->encoder_mask) {
> > > > > +		struct rockchip_encoder *rkencoder = to_rockchip_encoder(encoder);
> > > > > +		struct device_node *node, *parent;
> > > > > +
> > > > > +		parent = of_get_parent(rkencoder->port);
> > > > > +
> > > > > +		for_each_endpoint_of_node(parent, node) {
> > > > Is there any hurt directly use our downstream vendor kernel method here: use
> > > > vcstate->output_if set by encoder driver to get which interface we should
> > > > enable here?
> > > There is no vcstate->output_if in mainline currently. Ok, we could add
> > > that. The other thing is that there are multiple HDMI interfaces and
> > > the id of the HDMI encoder is encoded into output_if. Downstream kernel
> > > adds OF aliases to the HDMI ports. I didn't want to go that route
> > > because it doesn't seem to be very elegant to me.
> aliases is a very comm strategy in device tree world.

Yes, but not for retrieving bit offsets into registers. Normally aliases
can be changed at board level without confusing drivers.

> And your method also
> add need additional dt binds to define RK3568_VOP2_EP_xxx
> > > > You method is ok with device tree,  but it tied up this driver to device
> > > > tree, we are now tring to extend vop2 driver work with ACPI, so we hope this
> > > > driver can be much more flexible.
> > > The current rockchip drm driver seems to be pretty much tied to device
> > > tree. There are probably many other places that need parallel paths for
> > > ACPI support, I think we can delay this particular part until we see the
> > > whole picture. In the end we can still retrieve the output_if
> > > information differently with ACPI while still retrieving the information
> > > from the device tree the way we are doing currently.
> 
> The current driver only reference device thee at driver initial, we not wrap
> 
> device tree related things in other parts, so if we extend it to support
> ACPI,
> 
> we just need modify the initial code, this make things easier.

The device tree parsing could be moved out of vop2_crtc_atomic_enable()
into some initialisation path. In the end it's static information,
there's no need to do it repeatedly in atomic_enable.

Sascha

-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |