Hi Doug, On Wed, Apr 15, 2020 at 04:49:00PM -0700, Doug Anderson wrote: > On Wed, Apr 15, 2020 at 1:33 PM Laurent Pinchart wrote: > > On Wed, Apr 15, 2020 at 12:53:02PM -0700, Stephen Boyd wrote: > > > Quoting Douglas Anderson (2020-04-15 08:48:40) > > > > Allow people to specify to use a GPIO for hot-plug-detect. Add an > > > > example. > > > > > > > > NOTE: The current patch adding support for hpd-gpios to the Linux > > > > driver for hpd-gpios only adds enough support to the driver so that > > > > the bridge can use one of its own GPIOs. The bindings, however, are > > > > written generically. > > > > > > > > Signed-off-by: Douglas Anderson <dianders@xxxxxxxxxxxx> > > > > --- > > > > > > > > .../bindings/display/bridge/ti,sn65dsi86.yaml | 10 +++++++++- > > > > 1 file changed, 9 insertions(+), 1 deletion(-) > > > > > > > > diff --git a/Documentation/devicetree/bindings/display/bridge/ti,sn65dsi86.yaml b/Documentation/devicetree/bindings/display/bridge/ti,sn65dsi86.yaml > > > > index 8cacc6db33a9..554bfd003000 100644 > > > > --- a/Documentation/devicetree/bindings/display/bridge/ti,sn65dsi86.yaml > > > > +++ b/Documentation/devicetree/bindings/display/bridge/ti,sn65dsi86.yaml > > > > @@ -60,6 +60,10 @@ properties: > > > > const: 1 > > > > description: See ../../pwm/pwm.yaml for description of the cell formats. > > > > > > > > + hpd-gpios: > > > > + maxItems: 1 > > > > + description: If present use the given GPIO for hot-plug-detect. > > > > > > Shouldn't this go in the panel node? And the panel driver should get the > > > gpio and poll it after powering up the panel? Presumably that's why we > > > have the no-hpd property in the simple panel binding vs. putting it here > > > in the bridge. > > > > Same question really, I think this belongs to the panel (or connector) > > node indeed. > > Hrm. > > To me "no-hpd" feels OK in the panel because the lack of a connection > is somewhat symmetric. Thus it's OK to say either "HPD isn't hooked > up to the panel in this system" or "HPD isn't hooked up to the bridge > in this system" and both express the same thing (AKA that there is no > HPD connection between the bridge and the panel). In the case of > "no-hpd" it's more convenient to express it on the panel side because > the panel driver is the one whose behavior has to change if HPD isn't > hooked up. The panel datasheet is the one that says how long of a > delay we need if HPD isn't hooked up. > > ...but when you're talking about where the bridge driver should look > to find the HPD signal that it needs, that really feels like it should > be described as part of the bridge. Specifically imagine we were > using our bridge for DP, not for eDP. In that case simple-panel > wouldn't be involved because we could get any type of display plugged > in. Thus it couldn't go in the panel node. Here it feels clearer > that hpd-gpio needs to be a property of the bridge driver. If you were using it for DP, you would need a DT node for the DP connector (with bindings to be added to Documentation/devicetree/bindings/display/connector/, similar to the ones we already have for other connectors). That DT node should reference the HPD pin GPIO. The bridge driver for the connector (drivers/gpu/drm/bridge/display-connector.c) would then handle HPD. The good news is that it already does :-) > Looking at other usages of "hpd-gpio" in the kernel, it seems like the > usage I'm proposing is also common. Grepping for "hpd-gpios" shows > numerous examples of "hpd-gpios" being defined at the display > controller level and (effectively) I believe the bridge is at the > equivalent level. Bridge drivers should only implement support for features available from the corresponding hardware. If an HPD signal is connected to a dedicated pin of the bridge, and the bridge can generate an interrupt and expose the HPD status through I2C, then it should implement HPD-related operations. If the HPD pin from the connector is hooked up to a GPIO of the SoC, it should be handled by the connector bridge driver. -- Regards, Laurent Pinchart
