On Sat, Dec 7, 2019 at 1:13 PM Sam Ravnborg sam@ravnborg.org wrote:

On Sat, Dec 07, 2019 at 12:35:50PM -0800, Rob Clark wrote:

...

From: Rob Clark robdclark@chromium.org

For devices that have one of several possible panels installed, the panel-id property gives firmware a generic way to locate and enable the panel node corresponding to the installed panel.

For display timings there is something similar. Here the property is named native-mode and is a phandle to the preferred timing. And it is documented that if no native-mode is specified the first timing in the tree is chosen. So a different concept than this.

I could not from your otherwise well-documented changelog see why you wanted to go for an opauge integer and status rather than a phandle to the active display.

I think a lot of cases, panel-id could simply be an integer 0..N, but for the snapdragon windows devices, they seem to assign each panel a unique id. For example, the two possible panels that we've seen on the c630 are 0xc4 and 0xc5. I think all the values we've seen so far on other aarch64 laptops fit in an u8, but the actual value is defined as u32. The meaning behind those values is not really terribly important (and might well be arbitrary.. I'm not sure why they didn't go with a GUID). All that matters is they match what DtbLoader pulls out of the u32 PanelId field in the UEFIDisplayInfo variable. The intention behind describing the value as "opaque" was simply "don't assume it has to be 0..N".

As far as using phandles, I had toyed around with the idea.. the ideal thing would be if I could compile the dtb with an unresolved phandle link, and then fixup that link in DtbLoader based on panel-id. But this seems not to be possible, afaict I'd have to create a dummy node for the phandle to point to. Maybe I'm missing something, if there were a way to do this then I could make this work without any drm patches. Ofc I'm open to suggestions.

The panel-id, if I get it right, is optional and the important part is that the first panel with staus = "okay" is selected.

yup, this was to ensure that the other panels don't probe, which was a problem pointed out by robher with my previous approach

This would cover my usecase fine. I have a target with four different displays and the bootloader knows what display is used (based on gpio etc). The bootloader (barebox in my case) uses a simple variant of the DT, but reads in the DT used by the kernel and can modify the DT before it is passed to the kernel.

I'd be pretty happy if this (or whatever the eventual solution is) covers all the possible multi-sourced panel cases.. this comes up in nearly all consumer devices (laptops, phones, etc) and we pretty badly need an upstream solution for this.

BR, -R

    Sam

...

Example of how to use this property:

ivo_panel {
    compatible = "ivo,m133nwf4-r0";
    panel-id = <0xc5>;
    status = "disabled";

    ports {
        port {
            ivo_panel_in_edp: endpoint {
                remote-endpoint = <&sn65dsi86_out_ivo>;
            };
        };
    };
};

boe_panel {
    compatible = "boe,nv133fhm-n61";
    panel-id = <0xc4>;
    status = "disabled";

    ports {
        port {
            boe_panel_in_edp: endpoint {
                remote-endpoint = <&sn65dsi86_out_boe>;
            };
        };
    };
};

sn65dsi86: bridge@2c {
    compatible = "ti,sn65dsi86";

    ports {
        #address-cells = <1>;
        #size-cells = <0>;

        port@0 {
            reg = <0>;
            sn65dsi86_in_a: endpoint {
                remote-endpoint = <&dsi0_out>;
            };
        };

        port@1 {
            reg = <1>;

            sn65dsi86_out_boe: endpoint@c4 {
                remote-endpoint = <&boe_panel_in_edp>;
            };

            sn65dsi86_out_ivo: endpoint@c5 {
                remote-endpoint = <&ivo_panel_in_edp>;
            };
        };
    };
};

Signed-off-by: Rob Clark robdclark@chromium.org

.../bindings/display/panel/panel-common.yaml | 26 +++++++++++++++++++ 1 file changed, 26 insertions(+)

diff --git a/Documentation/devicetree/bindings/display/panel/panel-common.yaml b/Documentation/devicetree/bindings/display/panel/panel-common.yaml index ef8d8cdfcede..6113319b91dd 100644 --- a/Documentation/devicetree/bindings/display/panel/panel-common.yaml +++ b/Documentation/devicetree/bindings/display/panel/panel-common.yaml @@ -75,6 +75,32 @@ properties: in the device graph bindings defined in Documentation/devicetree/bindings/graph.txt.

• panel-id:
• description:

•  To support the case where one of several different panels can be installed
  

•  on a device, the panel-id property can be used by the firmware to identify
  

•  which panel should have it's status changed to "ok".  This property is not
  

•  used by the HLOS itself.
  

•  For a device with multiple potential panels, a node for each potential
  

•  should be defined with status = "disabled", and an appropriate panel-id
  

•  property.  The video data producer should be setup with endpoints going to
  

•  each possible panel.  The firmware will find the dt node with a panel-id
  

•  matching the actual panel installed, and change it's status to "ok".
  

•  The exact method the firmware uses to determine the panel-id of the installed
  

•  panel is outside the scope of this binding, but a few examples are
  

•  1) u-boot module reading a value from a u-boot env var
  

•  2) EFI driver module reading a value from an EFI variable
  

•  3) device specific firmware reading some device specific GPIOs or
  

•     e-fuse
  

•  The panel-id values are an opaque integer.  They can be sparse.  The only
  

•  important thing is that each possible panel in the system has a unique
  

•  panel-id, and that the values configured in the device's DTB match the
  

•  values that the firmware is looking for.
  

• ddc-i2c-bus: $ref: /schemas/types.yaml#/definitions/phandle description:

-- 2.23.0

---

dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel

Hi Rob,

Thank you for the patch.

On Sat, Dec 07, 2019 at 12:35:50PM -0800, Rob Clark wrote:

From: Rob Clark robdclark@chromium.org

For devices that have one of several possible panels installed, the panel-id property gives firmware a generic way to locate and enable the panel node corresponding to the installed panel. Example of how to use this property:

ivo_panel {
    compatible = "ivo,m133nwf4-r0";
    panel-id = <0xc5>;
    status = "disabled";

    ports {
        port {
            ivo_panel_in_edp: endpoint {
                remote-endpoint = <&sn65dsi86_out_ivo>;
            };
        };
    };
};

boe_panel {
    compatible = "boe,nv133fhm-n61";
    panel-id = <0xc4>;
    status = "disabled";

    ports {
        port {
            boe_panel_in_edp: endpoint {
                remote-endpoint = <&sn65dsi86_out_boe>;
            };
        };
    };
};

sn65dsi86: bridge@2c {
    compatible = "ti,sn65dsi86";

    ports {
        #address-cells = <1>;
        #size-cells = <0>;

        port@0 {
            reg = <0>;
            sn65dsi86_in_a: endpoint {
                remote-endpoint = <&dsi0_out>;
            };
        };

        port@1 {
            reg = <1>;

            sn65dsi86_out_boe: endpoint@c4 {
                remote-endpoint = <&boe_panel_in_edp>;
            };

            sn65dsi86_out_ivo: endpoint@c5 {
                remote-endpoint = <&ivo_panel_in_edp>;
            };
        };
    };
};

Signed-off-by: Rob Clark robdclark@chromium.org

.../bindings/display/panel/panel-common.yaml | 26 +++++++++++++++++++ 1 file changed, 26 insertions(+)

diff --git a/Documentation/devicetree/bindings/display/panel/panel-common.yaml b/Documentation/devicetree/bindings/display/panel/panel-common.yaml index ef8d8cdfcede..6113319b91dd 100644 --- a/Documentation/devicetree/bindings/display/panel/panel-common.yaml +++ b/Documentation/devicetree/bindings/display/panel/panel-common.yaml @@ -75,6 +75,32 @@ properties: in the device graph bindings defined in Documentation/devicetree/bindings/graph.txt.

• panel-id:
• description:

•  To support the case where one of several different panels can be installed
  

•  on a device, the panel-id property can be used by the firmware to identify
  

•  which panel should have it's status changed to "ok".  This property is not
  

•  used by the HLOS itself.
  

If your firmware can modify the status property of a panel, it can also add DT nodes. As discussed before, I don't think this belongs to DT. Even if panel-id isn't used by the operating system, you have Linux kernel patches in this series that show that this isn't transparent.

This needs to be handled in the firmware (or, if not possible, in a kernel board driver). The above DT fragment, visible to the kernel, doesn't describe the actual hardware. Furthermore, you would require all bridge drivers to be patched to support this method, which shows again that the issue isn't handled in the right place.

Finally, unless I'm mistaken, this series is meant to support display for an ACPI-based ARM machine. Using DT as a stop-gap measure because ACPI support isn't there yet is fine out-of-tree, and fine by me in-tree provided that the DT bindings are clean, but not when DT is abused like this.

I'm sorry, but this is a NACK from me. Please handle this transparently in the firmware if you want DT-based boot, or with ACPI.

•  For a device with multiple potential panels, a node for each potential
  

•  should be defined with status = "disabled", and an appropriate panel-id
  

•  property.  The video data producer should be setup with endpoints going to
  

•  each possible panel.  The firmware will find the dt node with a panel-id
  

•  matching the actual panel installed, and change it's status to "ok".
  

•  The exact method the firmware uses to determine the panel-id of the installed
  

•  panel is outside the scope of this binding, but a few examples are
  

•  1) u-boot module reading a value from a u-boot env var
  

•  2) EFI driver module reading a value from an EFI variable
  

•  3) device specific firmware reading some device specific GPIOs or
  

•     e-fuse
  

•  The panel-id values are an opaque integer.  They can be sparse.  The only
  

•  important thing is that each possible panel in the system has a unique
  

•  panel-id, and that the values configured in the device's DTB match the
  

•  values that the firmware is looking for.
  

• ddc-i2c-bus: $ref: /schemas/types.yaml#/definitions/phandle description:

Hi Rob,

On Sun, Dec 08, 2019 at 08:50:32AM -0800, Rob Clark wrote:

On Sun, Dec 8, 2019 at 6:45 AM Laurent Pinchart wrote:

...

On Sat, Dec 07, 2019 at 12:35:50PM -0800, Rob Clark wrote:

...

From: Rob Clark robdclark@chromium.org

For devices that have one of several possible panels installed, the panel-id property gives firmware a generic way to locate and enable the panel node corresponding to the installed panel. Example of how to use this property:

ivo_panel {
    compatible = "ivo,m133nwf4-r0";
    panel-id = <0xc5>;
    status = "disabled";

    ports {
        port {
            ivo_panel_in_edp: endpoint {
                remote-endpoint = <&sn65dsi86_out_ivo>;
            };
        };
    };
};

boe_panel {
    compatible = "boe,nv133fhm-n61";
    panel-id = <0xc4>;
    status = "disabled";

    ports {
        port {
            boe_panel_in_edp: endpoint {
                remote-endpoint = <&sn65dsi86_out_boe>;
            };
        };
    };
};

sn65dsi86: bridge@2c {
    compatible = "ti,sn65dsi86";

    ports {
        #address-cells = <1>;
        #size-cells = <0>;

        port@0 {
            reg = <0>;
            sn65dsi86_in_a: endpoint {
                remote-endpoint = <&dsi0_out>;
            };
        };

        port@1 {
            reg = <1>;

            sn65dsi86_out_boe: endpoint@c4 {
                remote-endpoint = <&boe_panel_in_edp>;
            };

            sn65dsi86_out_ivo: endpoint@c5 {
                remote-endpoint = <&ivo_panel_in_edp>;
            };
        };
    };
};

Signed-off-by: Rob Clark robdclark@chromium.org

.../bindings/display/panel/panel-common.yaml | 26 +++++++++++++++++++ 1 file changed, 26 insertions(+)

diff --git a/Documentation/devicetree/bindings/display/panel/panel-common.yaml b/Documentation/devicetree/bindings/display/panel/panel-common.yaml index ef8d8cdfcede..6113319b91dd 100644 --- a/Documentation/devicetree/bindings/display/panel/panel-common.yaml +++ b/Documentation/devicetree/bindings/display/panel/panel-common.yaml @@ -75,6 +75,32 @@ properties: in the device graph bindings defined in Documentation/devicetree/bindings/graph.txt.

• panel-id:
• description:

•  To support the case where one of several different panels can be installed
  

•  on a device, the panel-id property can be used by the firmware to identify
  

•  which panel should have it's status changed to "ok".  This property is not
  

•  used by the HLOS itself.
  

If your firmware can modify the status property of a panel, it can also add DT nodes. As discussed before, I don't think this belongs to DT. Even if panel-id isn't used by the operating system, you have Linux kernel patches in this series that show that this isn't transparent.

I've already explained several times why this is not feasible. It would require DtbLoader to be familiar with each individual device, and be rev'd every time a new device appears. That is not practical at all.

(And fwiw, the ACPI tables describe each panel.. with an ACPI method that is passed the the panel-id and returns the appropriate table.. since DT doesn't have methods, this is the solution.)

I stand by this patch, we can't keep running away from this problem and wave the magic firmware wand.

I believe in firmware solutions more than firmware magic wands :-)

While device-specific knowledge in DtbLoader is indeed not practical, you still need device-specific knowledge at the firmware level in the sense that you pass a device-specific DT binary to DtbLoader to be patched based on the panel ID. The device-specific information required at the firmware level can thus be expressed as data instead of code.

I understand it wouldn't be practical for DtbLoader to receive two independent pieces of device-specific data (the DT binary and another custom data blob). Why couldn't however DtbLoader get a DT binary as described in the commit message of this patch, and strip off all the panel nodes that are not applicable to the platform, as well as the panel-id property ? This would be completely transparent on the OS side, and would not require patching DtbLoader for every device, as all the information required would be present in a single DT binary, encoded using DT syntax.

Ths would create a dichotomy in the DT bindings, in the sense that we would have bindings applicable to the boot loader only, and bindings for the OS, but this is already the case in what you're proposing here as the panel-id property is documented as not used by the OS itself.

...

This needs to be handled in the firmware (or, if not possible, in a kernel board driver). The above DT fragment, visible to the kernel, doesn't describe the actual hardware. Furthermore, you would require all bridge drivers to be patched to support this method, which shows again that the issue isn't handled in the right place.

Finally, unless I'm mistaken, this series is meant to support display for an ACPI-based ARM machine. Using DT as a stop-gap measure because ACPI support isn't there yet is fine out-of-tree, and fine by me in-tree provided that the DT bindings are clean, but not when DT is abused like this.

I'm sorry, but this is a NACK from me. Please handle this transparently in the firmware if you want DT-based boot, or with ACPI.

...

•  For a device with multiple potential panels, a node for each potential
  

•  should be defined with status = "disabled", and an appropriate panel-id
  

•  property.  The video data producer should be setup with endpoints going to
  

•  each possible panel.  The firmware will find the dt node with a panel-id
  

•  matching the actual panel installed, and change it's status to "ok".
  

•  The exact method the firmware uses to determine the panel-id of the installed
  

•  panel is outside the scope of this binding, but a few examples are
  

•  1) u-boot module reading a value from a u-boot env var
  

•  2) EFI driver module reading a value from an EFI variable
  

•  3) device specific firmware reading some device specific GPIOs or
  

•     e-fuse
  

•  The panel-id values are an opaque integer.  They can be sparse.  The only
  

•  important thing is that each possible panel in the system has a unique
  

•  panel-id, and that the values configured in the device's DTB match the
  

•  values that the firmware is looking for.
  

• ddc-i2c-bus: $ref: /schemas/types.yaml#/definitions/phandle description:

Hi Rob,

On Sun, Dec 08, 2019 at 01:23:59PM -0800, Rob Clark wrote:

On Sun, Dec 8, 2019 at 10:28 AM Laurent Pinchart wrote:

...

On Sun, Dec 08, 2019 at 08:50:32AM -0800, Rob Clark wrote:

...

On Sun, Dec 8, 2019 at 6:45 AM Laurent Pinchart wrote:

...

On Sat, Dec 07, 2019 at 12:35:50PM -0800, Rob Clark wrote:

...

From: Rob Clark robdclark@chromium.org

For devices that have one of several possible panels installed, the panel-id property gives firmware a generic way to locate and enable the panel node corresponding to the installed panel. Example of how to use this property:

ivo_panel {
    compatible = "ivo,m133nwf4-r0";
    panel-id = <0xc5>;
    status = "disabled";

    ports {
        port {
            ivo_panel_in_edp: endpoint {
                remote-endpoint = <&sn65dsi86_out_ivo>;
            };
        };
    };
};

boe_panel {
    compatible = "boe,nv133fhm-n61";
    panel-id = <0xc4>;
    status = "disabled";

    ports {
        port {
            boe_panel_in_edp: endpoint {
                remote-endpoint = <&sn65dsi86_out_boe>;
            };
        };
    };
};

sn65dsi86: bridge@2c {
    compatible = "ti,sn65dsi86";

    ports {
        #address-cells = <1>;
        #size-cells = <0>;

        port@0 {
            reg = <0>;
            sn65dsi86_in_a: endpoint {
                remote-endpoint = <&dsi0_out>;
            };
        };

        port@1 {
            reg = <1>;

            sn65dsi86_out_boe: endpoint@c4 {
                remote-endpoint = <&boe_panel_in_edp>;
            };

            sn65dsi86_out_ivo: endpoint@c5 {
                remote-endpoint = <&ivo_panel_in_edp>;
            };
        };
    };
};

Signed-off-by: Rob Clark robdclark@chromium.org

.../bindings/display/panel/panel-common.yaml | 26 +++++++++++++++++++ 1 file changed, 26 insertions(+)

diff --git a/Documentation/devicetree/bindings/display/panel/panel-common.yaml b/Documentation/devicetree/bindings/display/panel/panel-common.yaml index ef8d8cdfcede..6113319b91dd 100644 --- a/Documentation/devicetree/bindings/display/panel/panel-common.yaml +++ b/Documentation/devicetree/bindings/display/panel/panel-common.yaml @@ -75,6 +75,32 @@ properties: in the device graph bindings defined in Documentation/devicetree/bindings/graph.txt.

• panel-id:
• description:

•  To support the case where one of several different panels can be installed
  

•  on a device, the panel-id property can be used by the firmware to identify
  

•  which panel should have it's status changed to "ok".  This property is not
  

•  used by the HLOS itself.
  

If your firmware can modify the status property of a panel, it can also add DT nodes. As discussed before, I don't think this belongs to DT. Even if panel-id isn't used by the operating system, you have Linux kernel patches in this series that show that this isn't transparent.

I've already explained several times why this is not feasible. It would require DtbLoader to be familiar with each individual device, and be rev'd every time a new device appears. That is not practical at all.

(And fwiw, the ACPI tables describe each panel.. with an ACPI method that is passed the the panel-id and returns the appropriate table.. since DT doesn't have methods, this is the solution.)

I stand by this patch, we can't keep running away from this problem and wave the magic firmware wand.

I believe in firmware solutions more than firmware magic wands :-)

and with that in mind, I think I've come up with a firmware solution, in the form of dtb overlays :-)

I've managed to get DtbLoader to find and load a panel overlay based on the panel-id it reads, which drops all patches in the patchset except the last one, which now has this delta:

Thank you for looking into this, I really like the outcome :-)

---

diff --git a/arch/arm64/boot/dts/qcom/Makefile b/arch/arm64/boot/dts/qcom/Makefile index 6498a1ec893f..1a61e8da2521 100644 --- a/arch/arm64/boot/dts/qcom/Makefile +++ b/arch/arm64/boot/dts/qcom/Makefile @@ -1,4 +1,5 @@ # SPDX-License-Identifier: GPL-2.0 +subdir-y += panels dtb-$(CONFIG_ARCH_QCOM) += apq8016-sbc.dtb dtb-$(CONFIG_ARCH_QCOM) += apq8096-db820c.dtb dtb-$(CONFIG_ARCH_QCOM) += ipq8074-hk01.dtb diff --git a/arch/arm64/boot/dts/qcom/panels/Makefile b/arch/arm64/boot/dts/qcom/panels/Makefile new file mode 100644 index 000000000000..dbf55f423555 --- /dev/null +++ b/arch/arm64/boot/dts/qcom/panels/Makefile @@ -0,0 +1,3 @@ +# SPDX-License-Identifier: GPL-2.0 +dtb-$(CONFIG_ARCH_QCOM) += panel-c4.dtb +dtb-$(CONFIG_ARCH_QCOM) += panel-c5.dtb diff --git a/arch/arm64/boot/dts/qcom/panels/panel-c4.dts b/arch/arm64/boot/dts/qcom/panels/panel-c4.dts new file mode 100644 index 000000000000..ebcf65419dad --- /dev/null +++ b/arch/arm64/boot/dts/qcom/panels/panel-c4.dts @@ -0,0 +1,17 @@ +// SPDX-License-Identifier: BSD-3-Clause +/*

• • Panel overlay for panel-id 0xc4
• • Copyright (c) 2019, Linaro Ltd.
• */

+/dts-v1/; +/plugin/; +/ {

• fragment@0 {

•    target-path = "/panel";
  

•    __overlay__ {
  

•        compatible = "boe,nv133fhm-n61";
  

•    };
  

• };

+}; diff --git a/arch/arm64/boot/dts/qcom/panels/panel-c5.dts b/arch/arm64/boot/dts/qcom/panels/panel-c5.dts new file mode 100644 index 000000000000..0ad5bb6003e3 --- /dev/null +++ b/arch/arm64/boot/dts/qcom/panels/panel-c5.dts @@ -0,0 +1,17 @@ +// SPDX-License-Identifier: BSD-3-Clause +/*

• • Panel overlay for panel-id 0xc5
• • Copyright (c) 2019, Linaro Ltd.
• */

+/dts-v1/; +/plugin/; +/ {

• fragment@0 {

•    target-path = "/panel";
  

•    __overlay__ {
  

•        compatible = "ivo,m133nwf4-r0";
  

•    };
  

• };

+}; diff --git a/arch/arm64/boot/dts/qcom/sdm850-lenovo-yoga-c630.dts b/arch/arm64/boot/dts/qcom/sdm850-lenovo-yoga-c630.dts index c35d8099d8eb..92c76afb721c 100644 --- a/arch/arm64/boot/dts/qcom/sdm850-lenovo-yoga-c630.dts +++ b/arch/arm64/boot/dts/qcom/sdm850-lenovo-yoga-c630.dts @@ -22,11 +22,13 @@ hsuart0 = &uart6; };

• /*

• * stub node which defines how panel is connected to bridge, which
  

• * will be updated by panel specific overlay
  

• */
  

  panel {

•    compatible = "ivo,m133nwf4-r0";
     power-supply = <&vlcm_3v3>;
     no-hpd;
  

•    ports {
         port {
             panel_in_edp: endpoint {
  

---

Side note, try as I might, I couldn't get the 'target = <&phandle>' approach to work in the overlays, so I ended up going with target-path instead. From digging thru the fdt_overlay code, I *think* it is because I end up w/ an overlay dtb without symbols. In the end, I guess target-path works just as well.

On Mon, Dec 9, 2019 at 7:31 AM Rob Herring robh+dt@kernel.org wrote:

On Sun, Dec 8, 2019 at 3:24 PM Rob Clark robdclark@gmail.com wrote:

...

On Sun, Dec 8, 2019 at 10:28 AM Laurent Pinchart laurent.pinchart@ideasonboard.com wrote:

...

Hi Rob,

On Sun, Dec 08, 2019 at 08:50:32AM -0800, Rob Clark wrote:

...

On Sun, Dec 8, 2019 at 6:45 AM Laurent Pinchart wrote:

...

On Sat, Dec 07, 2019 at 12:35:50PM -0800, Rob Clark wrote:

...

From: Rob Clark robdclark@chromium.org

For devices that have one of several possible panels installed, the panel-id property gives firmware a generic way to locate and enable the panel node corresponding to the installed panel. Example of how to use this property:

ivo_panel {
    compatible = "ivo,m133nwf4-r0";
    panel-id = <0xc5>;
    status = "disabled";

    ports {
        port {
            ivo_panel_in_edp: endpoint {
                remote-endpoint = <&sn65dsi86_out_ivo>;
            };
        };
    };
};

boe_panel {
    compatible = "boe,nv133fhm-n61";
    panel-id = <0xc4>;
    status = "disabled";

    ports {
        port {
            boe_panel_in_edp: endpoint {
                remote-endpoint = <&sn65dsi86_out_boe>;
            };
        };
    };
};

sn65dsi86: bridge@2c {
    compatible = "ti,sn65dsi86";

    ports {
        #address-cells = <1>;
        #size-cells = <0>;

        port@0 {
            reg = <0>;
            sn65dsi86_in_a: endpoint {
                remote-endpoint = <&dsi0_out>;
            };
        };

        port@1 {
            reg = <1>;

            sn65dsi86_out_boe: endpoint@c4 {
                remote-endpoint = <&boe_panel_in_edp>;
            };

            sn65dsi86_out_ivo: endpoint@c5 {
                remote-endpoint = <&ivo_panel_in_edp>;
            };
        };
    };
};

Signed-off-by: Rob Clark robdclark@chromium.org

.../bindings/display/panel/panel-common.yaml | 26 +++++++++++++++++++ 1 file changed, 26 insertions(+)

diff --git a/Documentation/devicetree/bindings/display/panel/panel-common.yaml b/Documentation/devicetree/bindings/display/panel/panel-common.yaml index ef8d8cdfcede..6113319b91dd 100644 --- a/Documentation/devicetree/bindings/display/panel/panel-common.yaml +++ b/Documentation/devicetree/bindings/display/panel/panel-common.yaml @@ -75,6 +75,32 @@ properties: in the device graph bindings defined in Documentation/devicetree/bindings/graph.txt.

• panel-id:
• description:

•  To support the case where one of several different panels can be installed
  

•  on a device, the panel-id property can be used by the firmware to identify
  

•  which panel should have it's status changed to "ok".  This property is not
  

•  used by the HLOS itself.
  

If your firmware can modify the status property of a panel, it can also add DT nodes. As discussed before, I don't think this belongs to DT. Even if panel-id isn't used by the operating system, you have Linux kernel patches in this series that show that this isn't transparent.

I've already explained several times why this is not feasible. It would require DtbLoader to be familiar with each individual device, and be rev'd every time a new device appears. That is not practical at all.

(And fwiw, the ACPI tables describe each panel.. with an ACPI method that is passed the the panel-id and returns the appropriate table.. since DT doesn't have methods, this is the solution.)

I stand by this patch, we can't keep running away from this problem and wave the magic firmware wand.

I believe in firmware solutions more than firmware magic wands :-)

and with that in mind, I think I've come up with a firmware solution, in the form of dtb overlays :-)

I've managed to get DtbLoader to find and load a panel overlay based on the panel-id it reads, which drops all patches in the patchset except the last one, which now has this delta:

This looks good to me. The only slight concern I have with it is making the overlay filename an ABI. I don't have a better suggestion though. How would this work for other vendors or the same panel ID (for different panels) used on different platforms? For different vendors at least, I guess dtbloader gets the base dtb path somehow and the overlay's are relative to that?

Not sure if "different vendors" in this context means different OEMs/ODMs, or different SoC's? This solution is snapdragon specific.. and in this case the panel id seems to be a flat namespace (I don't see re-use for different panels). But in DtbLoader I attempt loading a device specific path first, just in case. In particular the paths DtbLoader uses (for dtb and panel overlay) are (where $SysVendor/$ProductName/$BoardName come from SMBIOS tables and $PanelId comes from a qcom specific UEFIDisplayInfo variable) described below:

It tries to load dtb from (in order, paths relative to EFI system partition where DtbLoader is installed):

1) \dtb$SysVendor$ProductName-$BoardName.dtb 2) \dtb$SysVendor$ProductName.dtb

and panel overlay dtb from:

1) \dtb$SysVendor$ProductName-$BoardName-panel-$PanelId.dtb 2) \dtb$SysVendor$ProductName-panel-$PanelId.dtb 3) \dtb\panels\panel-$PanelId.dtb

We are already using different dtb names for the main dtb compared to what the kernel uses. Which isn't great. At some point we might want to add SysVendor/ProductName/BoardName fields in the dtb so we could automate renaming and stuffing the dtb's into the correct layout.

...

---

diff --git a/arch/arm64/boot/dts/qcom/Makefile b/arch/arm64/boot/dts/qcom/Makefile index 6498a1ec893f..1a61e8da2521 100644 --- a/arch/arm64/boot/dts/qcom/Makefile +++ b/arch/arm64/boot/dts/qcom/Makefile @@ -1,4 +1,5 @@ # SPDX-License-Identifier: GPL-2.0 +subdir-y += panels dtb-$(CONFIG_ARCH_QCOM) += apq8016-sbc.dtb dtb-$(CONFIG_ARCH_QCOM) += apq8096-db820c.dtb dtb-$(CONFIG_ARCH_QCOM) += ipq8074-hk01.dtb diff --git a/arch/arm64/boot/dts/qcom/panels/Makefile b/arch/arm64/boot/dts/qcom/panels/Makefile new file mode 100644 index 000000000000..dbf55f423555 --- /dev/null +++ b/arch/arm64/boot/dts/qcom/panels/Makefile @@ -0,0 +1,3 @@ +# SPDX-License-Identifier: GPL-2.0 +dtb-$(CONFIG_ARCH_QCOM) += panel-c4.dtb +dtb-$(CONFIG_ARCH_QCOM) += panel-c5.dtb diff --git a/arch/arm64/boot/dts/qcom/panels/panel-c4.dts b/arch/arm64/boot/dts/qcom/panels/panel-c4.dts new file mode 100644 index 000000000000..ebcf65419dad --- /dev/null +++ b/arch/arm64/boot/dts/qcom/panels/panel-c4.dts @@ -0,0 +1,17 @@ +// SPDX-License-Identifier: BSD-3-Clause +/*

• • Panel overlay for panel-id 0xc4
• • Copyright (c) 2019, Linaro Ltd.
• */

+/dts-v1/; +/plugin/; +/ {

• fragment@0 {

•    target-path = "/panel";
  

•    __overlay__ {
  

•        compatible = "boe,nv133fhm-n61";
  

•    };
  

• };

+}; diff --git a/arch/arm64/boot/dts/qcom/panels/panel-c5.dts b/arch/arm64/boot/dts/qcom/panels/panel-c5.dts new file mode 100644 index 000000000000..0ad5bb6003e3 --- /dev/null +++ b/arch/arm64/boot/dts/qcom/panels/panel-c5.dts @@ -0,0 +1,17 @@ +// SPDX-License-Identifier: BSD-3-Clause +/*

• • Panel overlay for panel-id 0xc5
• • Copyright (c) 2019, Linaro Ltd.
• */

+/dts-v1/; +/plugin/; +/ {

• fragment@0 {

•    target-path = "/panel";
  

•    __overlay__ {
  

•        compatible = "ivo,m133nwf4-r0";
  

•    };
  

• };

+}; diff --git a/arch/arm64/boot/dts/qcom/sdm850-lenovo-yoga-c630.dts b/arch/arm64/boot/dts/qcom/sdm850-lenovo-yoga-c630.dts index c35d8099d8eb..92c76afb721c 100644 --- a/arch/arm64/boot/dts/qcom/sdm850-lenovo-yoga-c630.dts +++ b/arch/arm64/boot/dts/qcom/sdm850-lenovo-yoga-c630.dts @@ -22,11 +22,13 @@ hsuart0 = &uart6; };

• /*

• * stub node which defines how panel is connected to bridge, which
  

• * will be updated by panel specific overlay
  

• */
  

  panel {

•    compatible = "ivo,m133nwf4-r0";
     power-supply = <&vlcm_3v3>;
     no-hpd;
  

•    ports {
         port {
             panel_in_edp: endpoint {
  

---

Side note, try as I might, I couldn't get the 'target = <&phandle>' approach to work in the overlays, so I ended up going with target-path instead. From digging thru the fdt_overlay code, I *think* it is because I end up w/ an overlay dtb without symbols. In the end, I guess target-path works just as well.

It's the base dtb that needs the symbols I think.

BTW, to answer the question on #dri-devel, if you wanted to put the full panel into an overlay, the way to solve the problem of having bridge specific knowledge is defining a connector node. That should provide enough abstraction. Presumably the connector is actually the same across panels in this situation, so that should match up with the actual h/w. It could be possible to have a different physical connector populated for each possible panel, but hopefully that's not the common case.

ok, I'm not too familiar with this connector node thing. I think in the end, it is really just the compatible string that differs (ie. power-supply, etc would all be the same for each possible panel). But it might be worth trying this connector node thing

BR, -R

Hi Rob.

Patch looks good, one small improvement proposal.

On Sat, Dec 07, 2019 at 12:35:51PM -0800, Rob Clark wrote:

From: Rob Clark robdclark@chromium.org

To handle the case where there are multiple panel endpoints, only one of which is enabled/installed, add support for a wildcard endpoint value to request finding whichever endpoint is enabled.

Signed-off-by: Rob Clark robdclark@chromium.org

drivers/gpu/drm/drm_of.c | 41 +++++++++++++++++++++++++++++++++++++++- 1 file changed, 40 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/drm_of.c b/drivers/gpu/drm/drm_of.c index 0ca58803ba46..2baf44e401b8 100644 --- a/drivers/gpu/drm/drm_of.c +++ b/drivers/gpu/drm/drm_of.c @@ -219,11 +219,44 @@ int drm_of_encoder_active_endpoint(struct device_node *node, } EXPORT_SYMBOL_GPL(drm_of_encoder_active_endpoint); +static int find_enabled_endpoint(const struct device_node *node, u32 port) +{

• struct device_node *endpoint_node, *remote;
• u32 endpoint = 0;
• for (endpoint = 0; ; endpoint++) {

• endpoint_node = of_graph_get_endpoint_by_regs(node, port, endpoint);
  

• if (!endpoint_node) {
  

• 	pr_debug("No more endpoints!\n");
  

• 	return -ENODEV;
  

• }
  

• remote = of_graph_get_remote_port_parent(endpoint_node);
  

• of_node_put(endpoint_node);
  

• if (!remote) {
  

• 	pr_debug("no valid remote node\n");
  

• 	continue;
  

• }
  

• if (!of_device_is_available(remote)) {
  

• 	pr_debug("not available for remote node\n");
  

• 	of_node_put(remote);
  

• 	continue;
  

• }
  

• pr_debug("found enabled endpoint %d for %s\n", endpoint, remote->name);
  

• of_node_put(remote);
  

• return endpoint;
  

• }
• return -ENODEV;

+}

This function seems pretty generic. Should this be part of drivers/of/property.c - as others may have the same need?

/**

• drm_of_find_panel_or_bridge - return connected panel or bridge device
• @np: device tree node containing encoder output ports
• @port: port in the device tree node

• • @endpoint: endpoint in the device tree node

• • @endpoint: endpoint in the device tree node, or -1 to find an enabled endpoint
  • @panel: pointer to hold returned drm_panel
  • @bridge: pointer to hold returned drm_bridge

Introducing a define would make it easier to use this function in the right way. For example: #define OF_ENDPOINT_FIRST -1

Then we would see this code in the user side: + ret = drm_of_find_panel_or_bridge(pdata->dev->of_node, 1, OF_ENDPOINT_FIRST, &pdata->panel, NULL);

Or something like this.

Sam

@@ -246,6 +279,12 @@ int drm_of_find_panel_or_bridge(const struct device_node *np, if (panel) *panel = NULL;

• if (endpoint == -1) {

• endpoint = find_enabled_endpoint(np, port);
  

• if (endpoint < 0)
  

• 	return endpoint;
  

• }
• remote = of_graph_get_remote_node(np, port, endpoint); if (!remote) return -ENODEV;

-- 2.23.0

---

dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel