Re: [PATCH] arm64: dts: imx8mm: Add CSI nodes

From: Laurent Pinchart
Date: Fri Oct 15 2021 - 18:40:03 EST

Hi Adam,

On Fri, Oct 15, 2021 at 05:26:03PM -0500, Adam Ford wrote:
> On Thu, Oct 14, 2021 at 4:16 PM Laurent Pinchart wrote:
> > On Thu, Oct 14, 2021 at 03:32:33PM -0500, Adam Ford wrote:
> > > On Wed, Oct 13, 2021 at 9:52 PM Adam Ford wrote:
> > > >
> > > > There is a csi bridge and csis interface that tie together
> > > > to allow csi2 capture.
> > > >
> > > > Signed-off-by: Adam Ford <aford173@xxxxxxxxx>
> > >
> > > Laurent,
> > >
> > > Since you did some work to make the csis functional on the imx8mm, I
> > > was hoping you might have some insights. Please see below:
> > >
> > > > diff --git a/arch/arm64/boot/dts/freescale/imx8mm.dtsi b/arch/arm64/boot/dts/freescale/imx8mm.dtsi
> > > > index c2f3f118f82e..8a8a5d0a4a1e 100644
> > > > --- a/arch/arm64/boot/dts/freescale/imx8mm.dtsi
> > > > +++ b/arch/arm64/boot/dts/freescale/imx8mm.dtsi
> > > > @@ -1068,6 +1068,22 @@ aips4: bus@32c00000 {
> > > > #size-cells = <1>;
> > > > ranges = <0x32c00000 0x32c00000 0x400000>;
> > > >
> > > > + csi: csi@32e20000 {
> > > > + compatible = "fsl,imx7-csi";
> >
> > This should be
> >
> > compatible = "fsl,imx8mm-csi", "fsl,imx7-csi";;
> >
> > > > + reg = <0x32e20000 0x1000>;
> > > > + interrupts = <GIC_SPI 16 IRQ_TYPE_LEVEL_HIGH>;
> > > > + clocks = <&clk IMX8MM_CLK_CSI1_ROOT>;
> > > > + clock-names = "mclk";
> > > > + power-domains = <&disp_blk_ctrl IMX8MM_DISPBLK_PD_CSI_BRIDGE>;
> > > > + status = "disabled";
> > > > +
> > > > + port {
> > > > + csi_in: endpoint {
> > > > + remote-endpoint = <&imx8mm_mipi_csi_out>;
> > > > + };
> > > > + };
> > > > + };
> > > > +
> > > > disp_blk_ctrl: blk-ctrl@32e28000 {
> > > > compatible = "fsl,imx8mm-disp-blk-ctrl", "syscon";
> > > > reg = <0x32e28000 0x100>;
> > > > @@ -1095,6 +1111,41 @@ disp_blk_ctrl: blk-ctrl@32e28000 {
> > > > #power-domain-cells = <1>;
> > > > };
> > > >
> > > > + mipi_csi2: mipi-csi@32e30000 {
> > > > + compatible = "fsl,imx8mm-mipi-csi2";
> > > > + reg = <0x32e30000 0x1000>;
> > > > + interrupts = <GIC_SPI 17 IRQ_TYPE_LEVEL_HIGH>;
> > > > + clock-frequency = <333000000>;
> > > > + clocks = <&clk IMX8MM_CLK_DISP_APB_ROOT>,
> > > > + <&clk IMX8MM_CLK_CSI1_ROOT>,
> > > > + <&clk IMX8MM_CLK_CSI1_PHY_REF>,
> > > > + <&clk IMX8MM_CLK_DISP_AXI_ROOT>;
> > >
> > > When comparing clock parents and the clock rates to one of NXP's
> > > kernels, it appears we need assigned-clocks and assigned-clock-parents
> > > to allow the CSI1 clocks to run fast enough.
> > >
> > > assigned-clocks = <&clk IMX8MM_CLK_CSI1_PHY_REF>,
> > > <&clk IMX8MM_CLK_CSI1_CORE>,
> > > <&clk IMX8MM_CLK_CSI1_ROOT>,
> > > <&clk IMX8MM_CLK_CSI1_ESC>;
> > > assigned-clock-parents = <&clk IMX8MM_SYS_PLL2_1000M>,
> > > <&clk IMX8MM_SYS_PLL2_1000M>,
> > > <&clk IMX8MM_CLK_CSI1_CORE>,
> > > <&clk IMX8MM_SYS_PLL1_800M>;
> > >
> > > However, even with this, I am attempting to capture. While I can get
> > > the pipeline enabled, when I try to capture a file, I get no data. i
> > > can use control-c to exit gstreamer, so I know the kernel didn't hang.
> > > I don't get errors, but I don't get data either.
> > >
> > > Any ideas if what/if any other clock entries might be missing?
> >
> > Have you tried looking at the error interrupt counters to see if the
> > CSI-2 bus side operates as expected ? I'd start there before blaming the
> > clocks. Note that the issue could also be in the CSI bridge.
>
> I am not seeing interrupts on either the CSI, mipi-csi, or err:
>
> 54: 0 0 0 0 GICv3 48 Level csi
> 55: 0 0 0 0 GICv3 49 Level
> 32e30000.mipi-csi
> Err: 0
>
> > I've had issues capturing from one particular sensor on an i.MX8MM
> > board, where I would get images, but with heavy corruption. The problem
> > was solved by chance when trying a different SPL binary. I need to
>
> I wonder if the issue was related to different versions of ATF, and
> depending on the power domain controls used, it could have been using
> ATF to set the power domains.
> I've had issues in the past too, but I have tried a couple different
> binaries, and I am not seeing a change in behavior.

We've just finished the investigation, and it turned out to be related
to the clock frequencies. Here's the DT changes we had to make to get it
working:

assigned-clocks = <&clk IMX8MM_CLK_DISP_AXI>,
<&clk IMX8MM_CLK_DISP_AXI_ROOT>,
<&clk IMX8MM_CLK_DISP_APB>,
<&clk IMX8MM_CLK_DISP_APB_ROOT>;
assigned-clock-parents = <&clk IMX8MM_SYS_PLL2_1000M>,
<&clk IMX8MM_CLK_DISP_AXI>,
<&clk IMX8MM_SYS_PLL1_800M>,
<&clk IMX8MM_CLK_DISP_APB>;
assigned-clock-rates = <500000000>, <500000000>, <200000000>, <200000000>;

This results in the following clock configuration:

sys_pll2_1000m 1000000000
disp_axi 500000000
disp_axi_root_clk 500000000

sys_pll1_800m 800000000
disp_apb 200000000
disp_apb_root_clk 200000000

I'll check which clock is the culprit, and see if we could program the
clocks in the driver directly.

> > investigate the differences there, and possibly update the driver
> > accordingly. It may or may not be related to your issue.
> >
> > Apart from the compatible string issue mentioned above, the bindings
> > match what I am using, so, with the compatible string fixed,
>
> Out of curiosity, I noticed the default csi clock parents point to a
> 24MHz clock, but the csi-2 interface wants to set the clock to 333MHz.
> Any chance you could share your clk_summary?

Is the above information what you need ?

> > Reviewed-by: Laurent Pinchart <laurent.pinchart@xxxxxxxxxxxxxxxx>
>
> Thanks for the review
>
> > > > + clock-names = "pclk", "wrap", "phy", "axi";
> > > > + power-domains = <&disp_blk_ctrl IMX8MM_DISPBLK_PD_MIPI_CSI>;
> > > > + status = "disabled";
> > > > +
> > > > + ports {
> > > > + #address-cells = <1>;
> > > > + #size-cells = <0>;
> > > > +
> > > > + port@0 {
> > > > + reg = <0>;
> > > > +
> > > > + imx8mm_mipi_csi_in: endpoint {
> > > > + };
> > > > + };
> > > > +
> > > > + port@1 {
> > > > + reg = <1>;
> > > > +
> > > > + imx8mm_mipi_csi_out: endpoint {
> > > > + remote-endpoint = <&csi_in>;
> > > > + };
> > > > + };
> > > > + };
> > > > + };
> > > > +
> > > > +
> > > > usbotg1: usb@32e40000 {
> > > > compatible = "fsl,imx8mm-usb", "fsl,imx7d-usb";
> > > > reg = <0x32e40000 0x200>;

--
Regards,

Laurent Pinchart