On Mon, Aug 29, 2022 at 3:36 PM Neil Armstrong <narmstrong@xxxxxxxxxxxx> wrote:
> On 25/08/2022 14:48, Linus Walleij wrote:
> > On Wed, Aug 17, 2022 at 3:31 PM Neil Armstrong <narmstrong@xxxxxxxxxxxx> wrote:
> >> On 15/08/2022 02:15, Dmitry Osipenko wrote:
> >>> 08.08.2022 12:51, Neil Armstrong пишет:
> >>>> On 08/08/2022 11:15, Neil Armstrong wrote:
> >
> >>>>>> This patch broke ARM/QEMU vexpress display because of_drm_find_bridge()
> >>>>>> always fail with -EPROBE_DEFER. Reverting this patch returns display
> >>>>>> back. Please fix or revert, thanks in advance.
> >>>>>
> >>>>> Can you share a QEMU cmdline to reproduce ?
> >>>>
> >>>> Actually the vexpress DT has multiple input ports instead of a single
> >>>> input port at @0
> >>>> and an output port at @1 like documented in the bindings:
> >>>>
> >>>> vexpress-v2m.dtsi#L303-L307:
> >>>> ports {
> >>>> #address-cells = <1>;
> >>>> #size-cells = <0>;
> >>>>
> >>>> /*
> >>>> * Both the core tile and the motherboard routes their output
> >>>> * pads to this transmitter. The motherboard system controller
> >>>> * can select one of them as input using a mux register in
> >>>> * "arm,vexpress-muxfpga". The Vexpress with the CA9 core tile is
> >>>> * the only platform with this specific set-up.
> >>>> */
> >>>> port@0 {
> >>>> reg = <0>;
> >>>> dvi_bridge_in_ct: endpoint {
> >>>> remote-endpoint = <&clcd_pads_ct>;
> >>>> };
> >>>> };
> >>>> port@1 {
> >>>> reg = <1>;
> >>>> dvi_bridge_in_mb: endpoint {
> >>>> remote-endpoint = <&clcd_pads_mb>;
> >>>> };
> >>>> };
> >>>> };
> >>>>
> >>>> bindings:
> >>>> ports:
> >>>> $ref: /schemas/graph.yaml#/properties/ports
> >>>>
> >>>> properties:
> >>>> port@0:
> >>>> $ref: /schemas/graph.yaml#/properties/port
> >>>> description: Parallel RGB input port
> >>>>
> >>>> port@1:
> >>>> $ref: /schemas/graph.yaml#/properties/port
> >>>> description: HDMI output port
> >>>>
> >>>> port@3:
> >>>> $ref: /schemas/graph.yaml#/properties/port
> >>>> description: Sound input port
> >>>>
> >>>> The patch is conform to the bindings, the DT was working but is actually
> >>>> not valid.
> >>>
> >>> I haven't looked closely at how to fix this properly, but if we can fix
> >>> it using of_machine_is_compatible("arm,vexpress") workaround in the
> >>> driver, then it will be good enough at least as a temporal fix, IMO.
> >>
> >> If other maintainers are ok with that, it can be temporary fix until the DT gets fixed.
> >
> > That's fine with me, will you send a patch?
>
> Who, me ?
Whoever you were referring to with the "temporary fix" :D
> > I don't know how you expect the DT to get "fixed" though.
> >
> > The hardware looks like this, it's maybe not the most elegant
> > electronics design but it exists, so... I wrote this DT with two
> > inputs, see commit f1fe12c8bf332, the code handling this
> > awkward mux is part of the DRM driver, see
> > drivers/gpu/drm/pl111/pl111_versatile.c function
> > pl111_vexpress_clcd_init() for an idea of how it works.
>
> The proper fix would be the other way around, adding a mux bridge before the sii902x
> returning the next bridge or nothing to the right controller.
Hm you mean the vexpress PLD should be a bridge and not just
some magic registers poked by the DRM driver?
OK fair point. But it will need proper representing in the DT then,
I guess that is what you mean by "DT gets fixed". It's not just
the DT that needs fixing then but also the driver(s).
Yours,
Linus Walleij
