Am 27.03.24 um 13:17 schrieb Christian Gmeiner: > Hi Josua > > My bisect showed me that after a device-tree sync the ethernet broke. > >> please take a look at this patch, I suspect it will (hack-)fix your >> ethernet issue. >> > Yes.. it fixes the problem I am seeing. > >> Unfortunately I had no time to revisit this yet and implement a correct >> solution. >> > Would it be okay for you if I look into a proper solution?
Sure. I am swamped by other products at the moment. However I will provide a rough overview what needs to be done: Background: i.MX6 SoMs originally had a an atheros phy at unstable address, either 0 or 4 depending on electrical noise on floating configuration signals. Linux had solved this by placing 2 phy nodes in device-tree. During boot the kernel would attempt in order to probe the phys, then link the successful one to the ethernet netdev. As a side-effect there is always an error in the kernel log for one of the addresses. U-Boot had something similar in that with a special address (I think 0xff) in device-tree, the code will probe mdio bus for all addresses, but only for a single phy node in dts. With release of SoM 2.0 we changed to an analog devices phy at address 1, which most importantly uses a different driver, and requires a different description in device-tree. When adding this new phy, as a third node in device-tree, kernel maintainers requested a better solution, and we got u-boot to runtime patch dtb to update status properties of the dtb for linux, after probing mdio bus for phys. Now - what u-boot needs to do is probe the mdio bus, and then runtime-patch its own DTB. Either with status properties, or for adding the phy-handle property (not sure which method will work). This somehow has to happen after probing mdio driver, but before probing ethernet driver. > I have a > handful of such devices here > that are already or will be used in a CI farm so I am interested in > using the latest U-Boot for them. > >> sincerely >> Josua Mayer >> >> Am 28.07.22 um 09:08 schrieb Josua Mayer: >>> Please hold off merging this patch until someone tested it, I can not do so >>> this week. >>> @Tom Can you confirm if this fixes the networking on your Cubox? >>> Also note that the phy-handle property may or may not be required, I am not >>> sure. >>> >>> sincerely >>> Josua Mayer >>> >>> On Thu, Jul 28, 2022 at 7:05 AM Josua Mayer <jo...@solid-run.com> wrote: >>> >>> The i.MX6 Cubox-i and HummingBoards can have different PHYs at varying >>> addresses. U-Boot needs to auto-detect which phy is actually present, >>> and at which address it is responding. >>> >>> Auto-detection from multiple phy nodes specified in device-tree does not >>> currently work correct. As a work-around merge all three possible phys >>> into one node with the special address 0xffffffff which indicates to the >>> generic phy driver to probe all addresses. >>> Also fixup this fake address before booting Linux, *if* booting with >>> U-Boot's internal dtb. >>> >>> Signed-off-by: Josua Mayer <jo...@solid-run.com> >>> Fixes: d0399a46e7cd >>> --- >>> arch/arm/dts/imx6qdl-sr-som.dtsi | 30 +++++++++------------------- >>> board/solidrun/mx6cuboxi/mx6cuboxi.c | 6 +++++- >>> 2 files changed, 14 insertions(+), 22 deletions(-) >>> >>> diff --git a/arch/arm/dts/imx6qdl-sr-som.dtsi >>> b/arch/arm/dts/imx6qdl-sr-som.dtsi >>> index ce543e325c..2d7cbc26b3 100644 >>> ---_a/arch/arm/dts/imx6qdl-sr-som.dtsi >>> +++ b/arch/arm/dts/imx6qdl-sr-som.dtsi >>> @@ -53,6 +53,7 @@ >>> &fec { >>> pinctrl-names = "default"; >>> pinctrl-0 = <&pinctrl_microsom_enet_ar8035>; >>> + phy-handle = <&phy>; >>> phy-mode = "rgmii-id"; >>> >>> /* >>> @@ -68,30 +69,17 @@ >>> #address-cells = <1>; >>> #size-cells = <0>; >>> >>> - /* >>> - * The PHY can appear at either address 0 or 4 due to >>> the >>> - * configuration (LED) pin not being pulled >>> sufficiently. >>> - */ >>> - ethernet-phy@0 { >>> - reg = <0>; >>> + phy: ethernet-phy@0 { >>> + /* >>> + * The PHY can appear either: >>> + * - AR8035: at address 0 or 4 >>> + * - ADIN1300: at address 1 >>> + * Actual address being detected at runtime. >>> + */ >>> + reg = <0xffffffff>; >>> qca,clk-out-frequency = <125000000>; >>> qca,smarteee-tw-us-1g = <24>; >>> - }; >>> - >>> - ethernet-phy@4 { >>> - reg = <4>; >>> - qca,clk-out-frequency = <125000000>; >>> - qca,smarteee-tw-us-1g = <24>; >>> - }; >>> - >>> - /* >>> - * ADIN1300 (som rev 1.9 or later) is always at address >>> 1. It >>> - * will be enabled automatically by U-Boot if detected. >>> - */ >>> - ethernet-phy@1 { >>> - reg = <1>; >>> adi,phy-output-clock = "125mhz-free-running"; >>> - status = "disabled"; >>> }; >>> }; >>> }; >>> diff --git a/board/solidrun/mx6cuboxi/mx6cuboxi.c >>> b/board/solidrun/mx6cuboxi/mx6cuboxi.c >>> index debf4f6a3b..52172a03b1 100644 >>> ---_a/board/solidrun/mx6cuboxi/mx6cuboxi.c >>> +++ b/board/solidrun/mx6cuboxi/mx6cuboxi.c >>> @@ -446,7 +446,7 @@ static int find_ethernet_phy(void) >>> */ >>> int ft_board_setup(void *fdt, struct bd_info *bd) >>> { >>> - int node_phy0, node_phy1, node_phy4; >>> + int node_phy, node_phy0, node_phy1, node_phy4; >>> int ret, phy; >>> bool enable_phy0 = false, enable_phy1 = false, enable_phy4 = >>> false; >>> enum board_type board; >>> @@ -478,6 +478,10 @@ int ft_board_setup(void *fdt, struct bd_info *bd) >>> return 0; >>> } >>> >>> + // update U-Boot's own unified phy node phy address, if present >>> + node_phy = fdt_path_offset(fdt, >>> "/soc/bus@2100000/ethernet@2188000/mdio/phy"); >>> + ret = fdt_setprop_u32(fdt, node_phy, "reg", phy); >>> + >>> // update all phy nodes status >>> node_phy0 = fdt_path_offset(fdt, >>> "/soc/bus@2100000/ethernet@2188000/mdio/ethernet-phy@0"); >>> ret = fdt_setprop_string(fdt, node_phy0, "status", enable_phy0 >>> ? "okay" : "disabled"); >>> --_ >>> 2.37.1 >>>