Re: [PATCH V2 1/2] arm64: dts: imx8mn-beacon: Add HDMI video with sound

From: Adam Ford
Date: Mon Jun 12 2023 - 13:08:13 EST

On Sat, Jun 10, 2023 at 1:11 PM Adam Ford <aford173@xxxxxxxxx> wrote:
>
> On Tue, Jun 6, 2023 at 7:57 AM Adam Ford <aford173@xxxxxxxxx> wrote:
> >
> > On Tue, Jun 6, 2023 at 2:13 AM Frieder Schrempf
> > <frieder.schrempf@xxxxxxxxxx> wrote:
> > >
> > > Hi Adam,
> > >
> > > On 06.06.23 00:33, Adam Ford wrote:
> > > > The Beacon Embedded imx8mn development kit has a DSI
> > > > to HDMI bridge chip. The bridge supports stereo audio
> > > > and hot-plug detection.
> > > >
> > > > Signed-off-by: Adam Ford <aford173@xxxxxxxxx>
> > > >
> > > > diff --git a/arch/arm64/boot/dts/freescale/imx8mn-beacon-kit.dts b/arch/arm64/boot/dts/freescale/imx8mn-beacon-kit.dts
> > > > index 1392ce02587b..2108ec8c019c 100644
> > > > --- a/arch/arm64/boot/dts/freescale/imx8mn-beacon-kit.dts
> > > > +++ b/arch/arm64/boot/dts/freescale/imx8mn-beacon-kit.dts
> > >
> > > I have to minor comments below, otherwise this looks good to me.
> > >
> > > As I'm trying to come up with similar changes for our boards I also have
> > > some questions below. Maybe you could share your knowledge on these.
> > >
> > > Thanks!
> > > Frieder
> > >
> > > > @@ -16,4 +16,138 @@ / {
> > > > chosen {
> > > > stdout-path = &uart2;
> > > > };
> > > > +
> > > > + connector {
> > > > + compatible = "hdmi-connector";
> > > > + type = "a";
> > > > +
> > > > + port {
> > > > + hdmi_connector_in: endpoint {
> > > > + remote-endpoint = <&adv7535_out>;
> > > > + };
> > > > + };
> > > > + };
> > > > +
> > > > + reg_hdmi: regulator-hdmi-dvdd {
> > > > + compatible = "regulator-fixed";
> > > > + pinctrl-names = "default";
> > > > + pinctrl-0 = <&pinctrl_reg_hdmi>;
> > > > + regulator-name = "hdmi_pwr_en";
> > > > + regulator-min-microvolt = <3300000>;
> > > > + regulator-max-microvolt = <3300000>;
> > > > + gpio = <&gpio2 11 GPIO_ACTIVE_HIGH>;
> > > > + enable-active-high;
> > > > + startup-delay-us = <70000>;
> > > > + regulator-always-on;
> > > > + };
> > > > +
> > > > + sound-hdmi {
> > > > + compatible = "simple-audio-card";
> > > > + simple-audio-card,name = "sound-hdmi";
> > > > + simple-audio-card,format = "i2s";
> > > > +
> > > > + simple-audio-card,cpu {
> > > > + sound-dai = <&sai5 0>;
> > > > + system-clock-direction-out;
> > > > + };
> > > > +
> > > > + simple-audio-card,codec {
> > > > + sound-dai = <&adv_bridge>;
> > > > + };
> > > > + };
> > > > +};
> > > > +
> > > > +&i2c2 {
> > > > + adv_bridge: hdmi@3d {
> > > > + compatible = "adi,adv7535";
> > > > + pinctrl-names = "default";
> > > > + pinctrl-0 = <&pinctrl_hdmi_bridge>;
> > > > + reg = <0x3d>, <0x3b>;
> > > > + reg-names = "main", "cec";
> > > > + adi,dsi-lanes = <4>;
> > >
> > > On our boards we have this working with 4 lanes. But we also have some
> > > boards that only have 2 DSI lanes connected. We don't get any image in
> > > on the display in this case. Did you ever try 2 lanes or do you have an
> > > idea what could be wrong?
> >

+ Lucas,

> > I didn't try 2-lane, but see below regarding clocks to see if any of
> > that makes any sense.
>
Frieder,

> I tried configuring my hardware for 2-lane operation, and it appears
> to not function, but I haven't been able to determine why. Have you
> made any progress on getting your 2-lane interface operational?

I looked at the NXP down-stream code to compare the HBP, HFP, and HSA
values [1] to what we're generating with this mainline code, but the
formula to achieve these values doesn't make sense to me. The
equations for the 4-lane calculation were pretty straight-forward.

I tried to do a look-up table similar to the method that NXP did, but
it was rejected in favor of a formula to calculate values. You might
try to see if using hard-coded values for HFP, HBP and HSA work for
your display which would at least narrow down where the problem might
be. NXP's downstream code also has a work-around for calculating the
PHY [2] which only appears to affect 720P@60 for some reason.

If someone has an idea of how to calculate the 2-lane timings for HFP,
HBP, and HSA, that would be useful.

720p@60
HFP HBP HSA
159 324 54 (calculated with current formula)
159 320 40 (downstream look-up table)

adam

[1] - https://github.com/nxp-imx/linux-imx/blob/lf-6.1.y/drivers/gpu/drm/bridge/sec-dsim.c#L369
[2] - https://github.com/nxp-imx/linux-imx/blob/lf-6.1.y/drivers/gpu/drm/bridge/sec-dsim.c#L1018
>
> 2-lane clocks probably require a pixel clock 2x the speed of the
> 4-lane modes. I'm really tight on time right now, but I can try to
> put some time in to investigating the 2-lane operation.
>
> adam
> >
> > >
> > > > + adi,fixed-lanes;
> > >
> > > I think this property comes from downstream and should be removed. I
> > > don't see it anywhere in the upstream driver or bindings.
> >
> > You're right. This came from some early debugging I was doing for a
> > patch I was going to propose when someone beat me to it. I'll submit
> > a V3 with this removed.
> > >
> > > > + dvdd-supply = <&reg_hdmi>;
> > > > + v3p3-supply = <&reg_hdmi>;
> > > > + v1p2-supply = <&reg_hdmi>;
> > > > + a2vdd-supply = <&reg_hdmi>;
> > > > + avdd-supply = <&reg_hdmi>;
> > > > + pvdd-supply = <&reg_hdmi>;
> > >
> > > Please sort the reg properties above alphabetically.
> >
> > OK.
> >
> > >
> > > > + interrupt-parent = <&gpio1>;
> > > > + interrupts = <9 IRQ_TYPE_LEVEL_LOW>;
> > > > + #sound-dai-cells = <0>;
> > > > +
> > > > + ports {
> > > > + #address-cells = <1>;
> > > > + #size-cells = <0>;
> > > > +
> > > > + port@0 {
> > > > + reg = <0>;
> > > > +
> > > > + adv7535_in: endpoint {
> > > > + remote-endpoint = <&dsi_out>;
> > > > + };
> > > > + };
> > > > +
> > > > + port@1 {
> > > > + reg = <1>;
> > > > +
> > > > + adv7535_out: endpoint {
> > > > + remote-endpoint = <&hdmi_connector_in>;
> > > > + };
> > > > + };
> > > > + };
> > > > + };
> > > > +};
> > > > +
> > > > +&lcdif {
> > > > + assigned-clocks = <&clk IMX8MN_VIDEO_PLL1>;
> > > > + assigned-clock-rates = <594000000>;
> > >
> > > Just out of interest: Why do you need to set the video PLL clock here
> > > and how did you determine the "correct" value? Why is this missing in
> > > the i.MX8MM dts?
> >
> > I am glad you asked this question. In an ideal world we would not
> > need to set this, because the display clock would propagate up to the
> > video_pll, but that's currently not happening yet, so one needs to
> > find a video-pll clock rate that nicely divides into the maximum
> > number of resoltions as possible.
> >
> > When you run modetest on a display, you'll get a list of resolutions
> > with their refresh rate and horizontal and vertical timings as well as
> > the pixel clock. The mxsfb needs its clock to match the pixel clock
> > in order for the displays to sync properly. With 594000000 set for
> > the video-pll, the disp-clock divdes down to achieve what it can to
> > hit the required pixel clock.
> >
> > Here is an example with some of the extra timing info removed. The
> > final column is the 594MHz / pix-clk
> > name refresh pix clk Divide by 594000000
> > 1920x1080 60 148500 4000
> > 1920x1080 59.94 148352 4003.990509
> > 1920x1080 50 148500 4000
> > 1920x1080 30 74250 8000
> > 1920x1080 29.97 74176 8007.981018
> > 1920x1080 24 74250 8000
> > 1920x1080 23.98 74176 8007.981018
> > 1600x900 60 108000 5500
> > 1280x1024 60.02 108000 5500
> > 1280x800 59.91 71000 8366.197183
> > 1152x864 59.97 81579 7281.285625
> > 1280x720 60 74250 8000
> > 1280x720 59.94 74176 8007.981018
> > 1024x768 60 65000 9138.461538
> > 800x600 60.32 40000 14850
> > 720x576 50 27000 22000
> > 720x480 60 27027 21978.02198
> > 720x480 59.94 27000 22000
> > 640x480 60 25200 23571.42857
> > 640x480 59.94 25175 23594.83615
> >
> >
> > As you can see, there are a number of resolutions which divide evenly,
> > but there are a bunch that do not. If you wanted to achieve
> > resolutions that do not divide evenly by 594MHz, you'd need to change
> > the video-pll clock to something else in order for this clock to
> > evenly divide but then it would break other resolutions.. I have
> > tested this, and I can get some of these additional resolutions to
> > work with a different video-pll clock rate. NXP's downstream kernel
> > masks this by blocking out clock rates that didn't divide evenly.
> >
> > The Mini has a different clock parent-child relationship, and the
> > clock appears to default to 594MHz already, but the Nano has a default
> > of 650MHz which is why I had to add the clock there. I didn't want to
> > put it into the imx8mn.dtsi file, because this may not be appropriate
> > for some people or people with a fixed display running at specific
> > resltion and/or pixel rate.
> >
> > I proposed a solution which would fix both Mini and Nano by allowing
> > the requested display clock to propagate back up to the video-pll.
> >
> > On the Mini and Nano, a patch I proposed can help achieve more accurate
> > lcdif clocks. For example, when trying to get a pixel clock of 31.500MHz
> > on an imx8m Nano, the clocks divided the 594MHz down, but
> > left the parent rate untouched which caused a calculation error.
> >
> > Before:
> > video_pll 594000000
> > video_pll_bypass 594000000
> > video_pll_out 594000000
> > disp_pixel 31263158
> > disp_pixel_clk 31263158
> >
> > Variance = -236842 Hz
> >
> > After this patch:
> > video_pll 31500000
> > video_pll_bypass 31500000
> > video_pll_out 31500000
> > disp_pixel 31500000
> > disp_pixel_clk 31500000
> >
> > Variance = 0 Hz
> >
> > If you want to check out the patch I proposed, look at [1]. I'd love
> > to get some tracking and/or feedback for this. This would allow the
> > Mini and Nano to sync more resolutions and allow people to not have to
> > specify the video-pll rate.
> > imx8mp is little more complicated, because there are two different
> > pixel clocks generated from the same video-pll (the DSI and the
> > LDB/LVDS)
> >
> > >
> > > > + status = "okay";
> > > > +};
> > > > +
> > > > +&mipi_dsi {
> > > > + samsung,esc-clock-frequency = <20000000>;
> > >
> > > Same here, I'm interested in how you determined the "correct" value for
> > > this property. Are there any rules to follow?
> >
> > 20MHz was the max clock rate supported by the DSI controller. It also
> > appears to be the clock speed set in the NXP downstream kernel, so I
> > used it. I wish I had a better explanation.
> >
> > adam
> >
> > [1] - https://lore.kernel.org/linux-arm-kernel/20230506195325.876871-1-aford173@xxxxxxxxx/
> >
> > >
> > > > + status = "okay";
> > > > +
> > > > + ports {
> > > > + port@1 {
> > > > + reg = <1>;
> > > > +
> > > > + dsi_out: endpoint {
> > > > + remote-endpoint = <&adv7535_in>;
> > > > + };
> > > > + };
> > > > + };
> > > > +};
> > > > +
> > > > +&sai5 {
> > > > + pinctrl-names = "default";
> > > > + pinctrl-0 = <&pinctrl_sai5>;
> > > > + assigned-clocks = <&clk IMX8MN_CLK_SAI5>;
> > > > + assigned-clock-parents = <&clk IMX8MN_AUDIO_PLL1_OUT>;
> > > > + assigned-clock-rates = <24576000>;
> > > > + #sound-dai-cells = <0>;
> > > > + status = "okay";
> > > > +};
> > > > +
> > > > +&iomuxc {
> > > > + pinctrl_hdmi_bridge: hdmibridgegrp {
> > > > + fsl,pins = <
> > > > + MX8MN_IOMUXC_GPIO1_IO09_GPIO1_IO9 0x19
> > > > + >;
> > > > + };
> > > > +
> > > > + pinctrl_reg_hdmi: reghdmigrp {
> > > > + fsl,pins = <
> > > > + MX8MN_IOMUXC_SD1_STROBE_GPIO2_IO11 0x16
> > > > + >;
> > > > + };
> > > > +
> > > > + pinctrl_sai5: sai5grp {
> > > > + fsl,pins = <
> > > > + MX8MN_IOMUXC_SAI5_RXD3_SAI5_TX_DATA0 0xd6
> > > > + MX8MN_IOMUXC_SAI5_RXD2_SAI5_TX_BCLK 0xd6
> > > > + MX8MN_IOMUXC_SAI5_RXD1_SAI5_TX_SYNC 0xd6
> > > > + >;
> > > > + };
> > > > };

Attachment: image.png
Description: PNG image