Hi Fabio,
On 2/8/2026 10:26 PM, Fabio Estevam wrote:
> From: Fabio Estevam <[email protected]>
>
> Make spl-boot-order to accept boot from SPI NAND as well.
>
> With this change, it is possible to specify spi_nand
> as the boot medium like this:
>
> u-boot,spl-boot-order = &spi_nand;
You should probably include something like following (after DT has been
updated as per DT review) in rv1103b.c to also allow "same-as-spl" to
work.
const char * const boot_devices[BROM_LAST_BOOTSOURCE + 1] = {
[BROM_BOOTSOURCE_EMMC] = "/soc/mmc@20d30000",
[BROM_BOOTSOURCE_SPINOR] = "/soc/spi@20d40000/flash@0",
[BROM_BOOTSOURCE_SD] = "/soc/mmc@20d20000",
};
> Signed-off-by: Fabio Estevam <[email protected]>
> ---
> Changes since v1:
> - None.
>
> arch/arm/mach-rockchip/spl-boot-order.c | 11 +++++++++--
> 1 file changed, 9 insertions(+), 2 deletions(-)
>
> diff --git a/arch/arm/mach-rockchip/spl-boot-order.c
> b/arch/arm/mach-rockchip/spl-boot-order.c
> index 6572dde29f65..e5897536934a 100644
> --- a/arch/arm/mach-rockchip/spl-boot-order.c
> +++ b/arch/arm/mach-rockchip/spl-boot-order.c
> @@ -4,6 +4,7 @@
> */
>
> #include <dm.h>
> +#include <dm/device_compat.h>
> #include <fdt_support.h>
> #include <log.h>
> #include <mmc.h>
> @@ -75,6 +76,8 @@ static int spl_node_to_boot_device(int node)
> */
> if (!uclass_find_device_by_of_offset(UCLASS_SPI_FLASH, node, &parent))
> return BOOT_DEVICE_SPI;
> + if (fdt_node_check_compatible(gd->fdt_blob, node, "spi-nand") == 0)
> + return BOOT_DEVICE_SPI;
I think something like following will be better:
if (!uclass_find_device_by_of_offset(UCLASS_MTD, node, &parent))
return BOOT_DEVICE_MTD;
And possible also with a protect check (should probably be fixed for
MMC/SPI_FLASH in a separate future patch/series).
if (CONFIG_IS_ENABLED(MTD_LOAD_SUPPORT) &&
!uclass_find_device_by_of_offset(UCLASS_MTD, node, &parent))
return BOOT_DEVICE_MTD;
Or whatever Kconfig option is used to ensure both UCLASS and SPL loader
is supported in the build. To avoid error messages about missing UCLASS
like following (from a different rk soc/board):
board_spl_was_booted_from: brom_bootdevice_id 3 maps to
'/soc/spi@ffac0000/flash@0'
Cannot find uclass for id 119: please add the UCLASS_DRIVER() declaration for
this UCLASS_... id
board_boot_order: could not map node /soc/spi@ffac0000/flash@0 to a
boot-device
Cannot find uclass for id 119: please add the UCLASS_DRIVER() declaration for
this UCLASS_... id
>
> return -1;
> }
> @@ -221,8 +224,12 @@ int spl_decode_boot_device(u32 boot_device, char *buf,
> size_t buflen)
>
> ret = uclass_find_device_by_of_offset(UCLASS_SPI_FLASH,
> node, &dev);
> if (ret) {
> - debug("%s: could not find udevice for %s\n",
> __func__, conf);
> - continue;
> + ret =
> uclass_find_device_by_of_offset(UCLASS_MTD, node, &dev);
> + if (ret || !device_is_compatible(dev,
> "spi-nand")) {
I do not think we need to check for spi-nand compatible here, make it
generic and only check for UCLASS_MTD. If the node binds to a UCLASS_MTD
driver it is likely we could use a generic SPL MTD loader that used mtd
read ops to load next stage.
With that we could possible also support SPI_FLASH_MTD and consolidate
and/or simplify SPL code, something that can be looked at in future once
we have a more generic SPL MTD loader in place.
Regards,
Jonas
> + debug("%s: could not find udevice for
> %s\n",
> + __func__, conf);
> + continue;
> + }
> }
>
> return ofnode_get_path(dev_ofnode(dev), buf, buflen);
