On 2024/11/27 17:29, Sasha Finkelstein wrote: > On Wed, 27 Nov 2024 at 02:51, Dmitry Torokhov <dmitry.torokhov@xxxxxxxxx> wrote: >> I believe this needs to be done at the SPI controller level. See >> "cs-gpiods" property in >> Documentation/devicetree/bindings/spi/spi-controller.yaml that, as far >> as I understand, allows overriding controller's native CS handling with >> a GPIO when needed. > > I have already tried doing that (adding the relevant gpio as cs-gpios > on the controller) > and for some reason none of my attempts worked. Since there is no hardware > documentation, I can't really tell why, could be possible that we need both > native CS and that gpio, could be memory barrier issues somewhere in > the driver core, > but the method above is the only one i could get to work. Are you sure this isn't just a pinmux problem, i.e. the bootloader initializes the pinmux for hardware CS only on one device and not the other? See spi3_pins in the DTS in our downstream tree (and the reference from the SPI controller). If the rest of the SPI pins are already working properly you can just try with the CS pin (same index as on the gpio spec). Ideally we'd list the 4 pins, but someone needs to reverse engineer the mapping with m1n1 gpiola since we don't know what it is. If it really doesn't work with native CS and proper pinmux then cs-gpios on the controller should work. If it doesn't something weird is going on elsewhere. There's only one CS line, needing both makes no sense. - Hector
