Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Laurent Pinchart <laurent.pinchart@ideasonboard.com>
To:     Daniel Vetter <daniel@ffwll.ch>
Cc:     Peter Rosin <peda@axentia.se>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Mark Rutland <mark.rutland@arm.com>,
        Boris Brezillon <boris.brezillon@free-electrons.com>,
        Alexandre Belloni <alexandre.belloni@bootlin.com>,
        devicetree@vger.kernel.org, David Airlie <airlied@linux.ie>,
        Nicolas Ferre <nicolas.ferre@microchip.com>,
        dri-devel <dri-devel@lists.freedesktop.org>,
        Rob Herring <robh+dt@kernel.org>,
        Jacopo Mondi <jacopo@jmondi.org>,
        Daniel Vetter <daniel.vetter@intel.com>,
        Linux ARM <linux-arm-kernel@lists.infradead.org>
Subject: Re: [PATCH v2 0/5] allow override of bus format in bridges
Date:   Wed, 04 Apr 2018 12:07:15 +0300
Message-ID: <1563250.XEJxFDuW0h@avalon>
Organization: Ideas on Board Oy
In-Reply-To: <CAKMK7uH-=BqU+vuePTq4joY3XS2ypvtSKYvpv1LLyovPpSXv6A@mail.gmail.com>
References: <20180326212447.7380-1-peda@axentia.se> <2233231.my6BbD3pcT@avalon> <CAKMK7uH-=BqU+vuePTq4joY3XS2ypvtSKYvpv1LLyovPpSXv6A@mail.gmail.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 7Bit
Content-Type: text/plain; charset="us-ascii"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi Daniel,

On Wednesday, 4 April 2018 09:34:41 EEST Daniel Vetter wrote:
> On Wed, Apr 4, 2018 at 12:28 AM, Laurent Pinchart wrote:
> > On Wednesday, 28 March 2018 10:08:26 EEST Daniel Vetter wrote:
> >> On Mon, Mar 26, 2018 at 11:24:42PM +0200, Peter Rosin wrote:
> >>> Hi!
> >>> 
> >>> [I got to v2 sooner than expected]
> >>> 
> >>> I have an Atmel sama5d31 hooked up to an lvds encoder and then
> >>> on to an lvds panel. Which seems like something that has been
> >>> done one or two times before...
> >>> 
> >>> The problem is that the bus_format of the SoC and the panel do
> >>> not agree. The SoC driver (atmel-hlcdc) can handle the
> >>> rgb444, rgb565, rgb666 and rgb888 bus formats. The hardware is
> >>> wired for the rgb565 case. The lvds encoder supports rgb888 on
> >>> its input side with the LSB wires for each color simply pulled
> >>> down internally in the encoder in my case which means that the
> >>> rgb565 bus_format is the format that works best. And the panel
> >>> is expecting lvds (vesa-24), which is what the encoder outputs.
> >>> 
> >>> The reason I "blame" the bus_format of the drm_connector is that
> >>> with the below DT snippet, things do not work *exactly* due to
> >>> that. At least, it starts to work if I hack the panel-lvds driver
> >>> to report the rgb565 bus_format instead of vesa-24.
> >>> 
> >>>     panel: panel {
> >>>             compatible = "panel-lvds";
> >>>             
> >>>             width-mm = <304>;
> >>>             height-mm = <228;
> >>>             
> >>>             data-mapping = "vesa-24";
> >>>             
> >>>             panel-timing {
> >>>                     // 1024x768 @ 60Hz (typical)
> >>>                     clock-frequency = <52140000 65000000 71100000>;
> >>>                     hactive = <1024>;
> >>>                     vactive = <768>;
> >>>                     hfront-porch = <48 88 88>;
> >>>                     hback-porch = <96 168 168>;
> >>>                     hsync-len = <32 64 64>;
> >>>                     vfront-porch = <8 13 14>;
> >>>                     vback-porch = <8 13 14>;
> >>>                     vsync-len = <8 12 14>;
> >>>             };
> >>>             
> >>>             port {
> >>>                     panel_input: endpoint {
> >>>                             remote-endpoint = <&lvds_encoder_output>;
> >>>                     };
> >>>             };
> >>>     };
> >>>     
> >>>     lvds-encoder {
> >>>             compatible = "ti,ds90c185", "lvds-encoder";
> >>>             
> >>>             ports {
> >>>                     #address-cells = <1>;
> >>>                     #size-cells = <0>;
> >>>                     
> >>>                     port@0 {
> >>>                             reg = <0>;
> >>>                             
> >>>                             lvds_encoder_input: endpoint {
> >>>                                     remote-endpoint = <&hlcdc_output>;
> >>>                             };
> >>>                     };
> >>>                     
> >>>                     port@1 {
> >>>                             reg = <1>;
> >>>                             
> >>>                             lvds_encoder_output: endpoint {
> >>>                                     remote-endpoint = <&panel_input>;
> >>>                             };
> >>>                     };
> >>>             };
> >>>     };
> >>> 
> >>> But, instead of perverting the panel-lvds driver with support
> >>> for a totally fake non-lvds bus_format, I intruduce an API that allows
> >>> display controller drivers to query the required bus_format of any
> >>> intermediate bridges, and match up with that instead of the formats
> >>> given by the drm_connector. I trigger this with this addition to the
> >>> 
> >>> lvds-encoder DT node:
> >>>             interface-pix-fmt = "rgb565";
> >>> 
> >>> Naming is hard though, so I'm not sure if that's good?
> >>> 
> >>> I threw in the first patch, since that is the actual lvds encoder
> >>> I have in this case.
> >>> 
> >>> Suggestions welcome.
> >> 
> >> Took a quick look, feels rather un-atomic. And there's beend discussing
> >> for other bridge related state that we might want to track (like the full
> >> adjusted_mode that might need to be adjusted at each stage in the chain).
> >> So here's my suggestions:
> >> 
> >> - Add an optional per-bridge internal state struct using the support in
> >>   https://dri.freedesktop.org/docs/drm/gpu/drm-kms.html#handling-driver-> >>   private-state
> >> 
> >>   Yes it says "driver private", but since bridge is just helper stuff
> >>   that's all included. "driver private" == "not exposed as uapi" here.
> >>   Include all the usual convenience wrappers to get at the state for a
> >>   bridge.
> >> 
> >> - Then stuff your bus_format into that new drm_bridge_state struct.
> >> 
> >> - Add a new bridge callback atomic_check, which gets that bridge state as
> >>   parameter (similar to all the other atomic_check functions).
> >> 
> >> This way we can even handle the bus_format dynamically, through the
> >> atomic framework your bridge's atomic_check callback can look at the
> >> entire atomic state (both up and down the chain if needed), it all neatly
> >> fits into atomic overall and it's much easier to extend.
> > 
> > While I think we'll eventually need bridge states, I don't think that's
> > need yet. The bus formats reported by this patch series are static. We're
> > not talking about the currently configured format for a bridge, but about
> > the list of supported formats. This is similar to the bus_formats field
> > present in the drm_display_info structure. There is thus in my opinion no
> > need to interface this with atomic until we need to track the current
> > format (and I think that will indeed happen at some point, but I don't
> > think Peter needs this feature for now). That's why I've told Peter that
> > I would like a bridge API to report the information and haven't requested
> > a state-based implementation.
> 
> If it's static, why a dynamic callback? Just add it to struct
> drm_bridge, require it's filled out before registering the bridge,
> done.

If I remember correctly I mentioned both options in my initial proposal, 
without a personal preference. A new field in struct drm_bridge would 
certainly work for me.

> >> Please also cc Laurent Pinchart on this.
> >> 
> >>> Changes since v1 https://lkml.org/lkml/2018/3/17/221
> >>> - Add a proper bridge API to query the bus_format instead of abusing
> >>>   the ->get_modes part of the code. This is cleaner but requires
> >>>   changes to all display controller drivers wishing to participate.
> >>> - Add patch to adjust the atmel-hlcdc driver according to the above.
> >>> - Hook the new info into the bridge local to the lvds-encoder instead
> >>>   of messing about with new interfaces for the panel-bridge driver.
> >>> - Add patch with a DT parsing function of bus_formats in a central
> >>>   place.
> >>> - Rephrase the addition of ti,ds90c185 to the lvds-transmitter binding.
> >>> 
> >>> Peter Rosin (5):
> >>>   dt-bindings: display: bridge: lvds-transmitter: add ti,ds90c185
> >>>   drm: bridge: add API to query the expected input formats of bridges
> >>>   drm: of: add display bus-format parser
> >>>   drm: bridge: lvds-encoder: allow specifying the input bus format
> >>>   drm/atmel-hlcdc: take bridges into account when selecting output
> >>>     format
> >>>  
> >>>  .../bindings/display/bridge/lvds-transmitter.txt   | 14 ++++-
> >>>  .../devicetree/bindings/display/bus-format.txt     | 35 +++++++++++++
> >>>  drivers/gpu/drm/atmel-hlcdc/atmel_hlcdc_crtc.c     | 49 +++++++++++++--
> >>>  drivers/gpu/drm/bridge/lvds-encoder.c              | 25 +++++++++
> >>>  drivers/gpu/drm/drm_bridge.c                       | 32 ++++++++++++
> >>>  drivers/gpu/drm/drm_of.c                           | 59 +++++++++++++++
> >>>  include/drm/drm_bridge.h                           | 18 +++++++
> >>>  include/drm/drm_of.h                               |  9 ++++
> >>>  8 files changed, 237 insertions(+), 4 deletions(-)
> >>>  create mode 100644
> >>>  Documentation/devicetree/bindings/display/bus-format.txt

-- 
Regards,

Laurent Pinchart