Re: [PATCH v2] spl: Add generic SPL MTD loader support

[Sean Anderson <seanga2@gmail.com>]

On 2/19/26 22:52, Sean Anderson wrote:
> On 2/19/26 22:12, Fabio Estevam wrote:
>> From: Fabio Estevam <festevam@nabladev.com>
>>
>> Add support for loading the next stage from an MTD device in SPL.
>>
>> Introduce CONFIG_SPL_MTD_LOAD and a generic SPL MTD loader
>> implementation that uses the MTD subsystem to read the U-Boot payload.
>>
>> The loader works with any MTD-backed storage, including raw NAND and
>> SPI NAND, without being tied to a specific NAND type.
>>
>> The payload offset defaults to CONFIG_SYS_MTD_U_BOOT_OFFS and can be
>> overridden via the device tree property:
>>
>>      u-boot,spl-payload-offset
>>
>> To support both raw NAND and SPI NAND boot flows, the loader is
>> registered for BOOT_DEVICE_NAND and BOOT_DEVICE_SPI. This allows it
>> to operate correctly on platforms where the ROM reports either NAND
>> or SPI as the boot source while using the same MTD-based loading
>> infrastructure.
>>
>> The required NAND core and SPI NAND drivers are built for SPL when
>> CONFIG_SPL_MTD_LOAD is enabled.
>>
>> This provides reusable infrastructure for boards that boot from MTD
>> devices without relying on SPI-specific or NAND-specific SPL loaders.
>>
>> Signed-off-by: Fabio Estevam <festevam@nabladev.com>
>> ---
>> Changes since v1:
>> - Use uclass_get_device_by_seq().
>> - Use puts() instead of debug() for error.
>> - Include the new loader to include/spl_load.h.
>> - Introduce SPL_MTD_SPI_NAND.
>>
>>   common/spl/Kconfig        | 30 ++++++++++++++
>>   common/spl/Makefile       |  1 +
>>   common/spl/spl_mtd.c      | 83 +++++++++++++++++++++++++++++++++++++++
>>   drivers/mtd/Makefile      |  1 +
>>   drivers/mtd/nand/Makefile |  4 +-
>>   include/spl_load.h        |  1 +
>>   6 files changed, 119 insertions(+), 1 deletion(-)
>>   create mode 100644 common/spl/spl_mtd.c
>>
>> diff --git a/common/spl/Kconfig b/common/spl/Kconfig
>> index 2998b7acb75f..16dbc5b54324 100644
>> --- a/common/spl/Kconfig
>> +++ b/common/spl/Kconfig
>> @@ -933,6 +933,12 @@ config SYS_MMCSD_FS_BOOT_PARTITION
>>         used in fs mode.
>>         Use -1 as a special value to use the first bootable partition.
>> +config SYS_SPL_MTD_SEQ
>> +    int "MTD device number for the SPL load"
>> +    default 0
>> +    help
>> +      MTD device number used for the SPL load.
>> +
>>   config SPL_MMC_TINY
>>       bool "Tiny MMC framework in SPL"
>>       depends on SPL_MMC
>> @@ -1578,6 +1584,30 @@ config SPL_SPI_LOAD
>>   endif # SPL_SPI_FLASH_SUPPORT
>> +config SPL_MTD_LOAD
>> +    bool "Support loading from a generic MTD device"
>> +    depends on SPL
>> +    depends on MTD && DM_MTD
>> +    help
>> +      Enable support for loading the next stage from an MTD device
>> +      using the MTD subsystem in SPL.
>> +
>> +      This supports raw NAND and SPI NAND devices.
>> +
>> +config SPL_MTD_SPI_NAND
>> +    bool "Enable SPI NAND support in SPL"
>> +    depends on SPL_MTD_LOAD
>> +    select MTD_SPI_NAND
>> +    help
>> +     Build SPI NAND support for SPL.
>> +
>> +config SYS_MTD_U_BOOT_OFFS
>> +    hex "address of U-boot payload in the MTD device"
>> +    default 0x0
>> +    help
>> +     Address within the MTD device where the U-boot payload is fetched
>> +     from.
>> +
> 
> Maybe default to SYS_SPI_U_BOOT_OFFS? or CONFIG_SYS_NAND_U_BOOT_OFFS?
> 
>>   config SYS_SPI_U_BOOT_OFFS
>>       hex "address of u-boot payload in SPI flash"
>>       default 0x8000 if ARCH_SUNXI
>> diff --git a/common/spl/Makefile b/common/spl/Makefile
>> index 4c9482bd3096..67fc1cd1b396 100644
>> --- a/common/spl/Makefile
>> +++ b/common/spl/Makefile
>> @@ -35,6 +35,7 @@ obj-$(CONFIG_$(PHASE_)NVME) += spl_nvme.o
>>   obj-$(CONFIG_$(PHASE_)SEMIHOSTING) += spl_semihosting.o
>>   obj-$(CONFIG_$(PHASE_)DFU) += spl_dfu.o
>>   obj-$(CONFIG_$(PHASE_)SPI_LOAD) += spl_spi.o
>> +obj-$(CONFIG_SPL_MTD_LOAD) += spl_mtd.o
> 
> Does this need a $(PHASE_)? (I don't know, but the others have it)
> 
>>   obj-$(CONFIG_$(PHASE_)RAM_SUPPORT) += spl_ram.o
>>   obj-$(CONFIG_$(PHASE_)USB_SDP_SUPPORT) += spl_sdp.o
>>   endif
>> diff --git a/common/spl/spl_mtd.c b/common/spl/spl_mtd.c
>> new file mode 100644
>> index 000000000000..904f1fbad834
>> --- /dev/null
>> +++ b/common/spl/spl_mtd.c
>> @@ -0,0 +1,83 @@
>> +// SPDX-License-Identifier: GPL-2.0+
>> +
>> +/*
>> + * Generic SPL loader for MTD devices.
>> + *
>> + * Based on spl_spi.c, which is:
>> + *
>> + * Copyright (C) 2011 OMICRON electronics GmbH
>> + *
>> + * based on drivers/mtd/nand/raw/nand_spl_load.c
>> + *
>> + * Copyright (C) 2011
>> + * Heiko Schocher, DENX Software Engineering, hs@denx.de.
>> + */
>> +
>> +#include <config.h>
>> +#include <errno.h>
>> +#include <image.h>
>> +#include <log.h>
>> +#include <spl.h>
>> +#include <spl_load.h>
>> +
>> +#include <dm.h>
>> +#include <dm/ofnode.h>
>> +#include <dm/uclass.h>
>> +#include <mtd.h>
>> +
>> +static struct mtd_info *spl_mtd_get_device(void)
>> +{
>> +    struct udevice *dev;
>> +    int ret;
>> +
>> +    ret = uclass_get_device_by_seq(UCLASS_MTD, CONFIG_SYS_SPL_MTD_SEQ, &dev);

Actually, I don't think this is correct. NAND devices have have multiple MTDs for each
CS, and SPI-NOR devices are UCLASS_SPI_FLASH and not UCLASS_MTD. So I think it's better
to do something like

	mtd_probe_devices();
	switch (bootdev->boot_device) {
	case BOOT_DEVICE_NAND:
		mtd = get_mtd_device_nm(CONFIG_SPL_MTD_NAND_NAME);
		break;
	case BOOT_DEVICE_SPI:
		mtd = get_mtd_device_nm(CONFIG_SPL_MTD_SPI_FLASH_NAME);
		break;
	default:
		mtd = ERR_PTR(-ENODEV);
		break;
	}
	if (IS_ERR(mtd)) {
		ret = PTR_ERR(mtd);
		goto out;
	}

>> +    if (ret)
>> +        return NULL;
>> +
>> +    return dev_get_uclass_priv(dev);
>> +}
>> +
>> +static ulong spl_mtd_read(struct spl_load_info *load,
>> +              ulong offs, ulong size, void *buf)
>> +{
>> +    struct mtd_info *mtd = load->priv;
>> +    size_t retlen;
>> +    int ret;
>> +
>> +    ret = mtd_read(mtd, offs, size, &retlen, buf);
>> +    if (ret && !mtd_is_bitflip(ret))
>> +        return 0;
>> +
>> +    if (retlen != size)
>> +        return 0;
> 
> You can just return retlen here. We check it against size in spl_load.
> 
>> +
>> +    return retlen;
>> +}
>> +
>> +static int spl_mtd_load_image(struct spl_image_info *spl_image,
>> +                  struct spl_boot_device *bootdev)
>> +{
>> +    struct spl_load_info load;
>> +    struct mtd_info *mtd;
>> +    ulong offset;
>> +
>> +    mtd = spl_mtd_get_device();
>> +    if (!mtd) {
>> +        puts("SPL: No MTD device found\n");
>> +        return -ENODEV;
>> +    }
>> +
>> +    spl_load_init(&load, spl_mtd_read, mtd, mtd->writesize);
>> +
>> +    offset = CONFIG_SYS_MTD_U_BOOT_OFFS;
>> +
>> +    if (CONFIG_IS_ENABLED(OF_REAL))
>> +        offset = ofnode_conf_read_int("u-boot,spl-payload-offset",
>> +                          offset);
>> +
>> +    return spl_load(spl_image, bootdev, &load, 0, offset);
>> +}
>> +
>> +/* Priority 1 so boards may override */
>> +SPL_LOAD_IMAGE_METHOD("MTD-NAND", 1, BOOT_DEVICE_NAND, spl_mtd_load_image);
>> +SPL_LOAD_IMAGE_METHOD("MTD-SPI-NAND", 1, BOOT_DEVICE_SPI, spl_mtd_load_image);
> 
> Should these be a different priority than the "native" loaders? Does it make sense
> to have both enabled or should we disable native loaders in Kconfig so we don't have
> to consider this?
> 
> Actually, if CONFIG_SPL_NAND_SUPPORT is enabled then the native nand loader is compiled.
> Maybe we should add a separate Kconfig for it (CONFIG_SPL_NAND_LOAD) defaulting to
> SPL_NAND_SUPPORT. Then users can disable the "native" load method if they want to use MTD.

I have a patch for this that I will try to send this week.

--Sean

>> diff --git a/drivers/mtd/Makefile b/drivers/mtd/Makefile
>> index ce05e206073d..0856a8f68732 100644
>> --- a/drivers/mtd/Makefile
>> +++ b/drivers/mtd/Makefile
>> @@ -34,6 +34,7 @@ else
>>   ifneq ($(mtd-y),)
>>   obj-$(CONFIG_SPL_MTD) += mtd.o
>>   endif
>> +obj-$(CONFIG_SPL_MTD_LOAD) += nand/
> 
> Is this necessary? Shouldn't the user just enable CONFIG_SPL_NAND_SUPPORT?
> 
>>   obj-$(CONFIG_$(PHASE_)NAND_SUPPORT) += nand/
>>   obj-$(CONFIG_SPL_ONENAND_SUPPORT) += onenand/
>>   obj-$(CONFIG_$(PHASE_)SPI_FLASH_SUPPORT) += spi/
>> diff --git a/drivers/mtd/nand/Makefile b/drivers/mtd/nand/Makefile
>> index c8169cf73902..4bc11054ad74 100644
>> --- a/drivers/mtd/nand/Makefile
>> +++ b/drivers/mtd/nand/Makefile
>> @@ -1,10 +1,12 @@
>>   # SPDX-License-Identifier: GPL-2.0+
>> -ifeq ($(CONFIG_XPL_BUILD)$(CONFIG_TPL_BUILD),)
>>   nandcore-objs := core.o bbt.o
>> +ifeq ($(CONFIG_XPL_BUILD)$(CONFIG_TPL_BUILD),)
>>   obj-$(CONFIG_MTD_NAND_CORE) += nandcore.o
>>   obj-$(CONFIG_MTD_RAW_NAND) += raw/
>>   obj-$(CONFIG_MTD_SPI_NAND) += spi/
>>   else
>> +obj-$(CONFIG_SPL_MTD_LOAD) += nandcore.o
> 
> Shouldn't this be CONFIG_SPL_MTD_SPI_NAND too?
> 
>> +obj-$(CONFIG_SPL_MTD_SPI_NAND) += spi/
>>   obj-$(CONFIG_$(PHASE_)NAND_SUPPORT) += raw/
>>   endif
>> diff --git a/include/spl_load.h b/include/spl_load.h
>> index 525e0c9e86c6..5b06fa419dbf 100644
>> --- a/include/spl_load.h
>> +++ b/include/spl_load.h
>> @@ -113,6 +113,7 @@ static inline int _spl_load(struct spl_image_info *spl_image,
>>       IS_ENABLED(CONFIG_SPL_NOR_SUPPORT) + \
>>       IS_ENABLED(CONFIG_SPL_SEMIHOSTING) + \
>>       IS_ENABLED(CONFIG_SPL_SPI_LOAD) + \
>> +    IS_ENABLED(CONFIG_SPL_MTD_LOAD) + \
> 
> Please keep this alphabetical.
> 
> And still needs tests, but hopefully those should be simple to implement based
> on my reply to v1.
> 
> --Sean
> 
>>       0
>>   #if SPL_LOAD_USERS > 1
>