[PATCH v2 3/9] spl: Convert fat to spl_load

[Sean Anderson]

On 5/5/22 2:58 PM, Tom Rini wrote:
> On Thu, May 05, 2022 at 02:53:55PM -0400, Sean Anderson wrote:
>> 
>> 
>> On 5/5/22 11:38 AM, Tom Rini wrote:
>> > On Thu, May 05, 2022 at 10:51:39AM -0400, Sean Anderson wrote:
>> >> Hi Tom,
>> >> 
>> >> On 5/4/22 8:06 PM, Tom Rini wrote:
>> >> > On Fri, Apr 22, 2022 at 02:27:41PM -0400, Sean Anderson wrote:
>> >> > 
>> >> >> This converts the fat loader to use spl_load.
>> >> >> 
>> >> >> Signed-off-by: Sean Anderson <sean.anderson at seco.com>
>> >> >> Reviewed-by: Stefan Roese <sr at denx.de>
>> >> > 
>> >> > On am335x_evm_defconfig, and booting from a FAT SD card I see:
>> >> > Trying to boot from MMC1
>> >> > spl_load_image_fat: error reading image u-boot.img, err - 0
>> >> > SPL: failed to boot from all boot devices
>> >> > ### ERROR ### Please RESET the board ###
>> >> 
>> >> Looks like spl_load_image_fat was expecting bytes read and
>> >> not an errno (and so turned your success into a failure). Will fix.
>> > 
>> > OK.
>> > 
>> >> > Note that while not _this_ patch (but likely the one for raw) breaks
>> >> > booting from my mx6cuboxi via SD card with:
>> >> > Trying to boot from MMC1
>> >> > [hang]
>> >> > 
>> >> 
>> >> Hm, if there's a problem you should at least see
>> >> "mmc_load_image_raw_sector: mmc block read error\n"
>> >> 
>> >> Could you enable DEBUG in common/spl/spl.c and common/spl/spl_mmc.c?
>> > 
>> > U-Boot SPL 2022.07-rc1-00267-g064334e21dde-dirty (May 05 2022 - 11:33:38 -0400)
>> >>>SPL: board_init_r()
>> > spl_init
>> > Trying to boot from MMC1
>> > spl: mmc boot mode: raw
>> > hdr read sector 8a, count=1
>> > 
>> 
>> Looks like there's a bug in spl_simple_read
>> 
>> 	size_t bl_len = info->filename ? ARCH_DMA_MINALIGN : bl_len;
>> 
>> should be
>> 
>> 	size_t bl_len = info->filename ? ARCH_DMA_MINALIGN : info->bl_len;
>> 
>> The funny thing is that this kind of bug results in a -Wuninitialized warning
>> on godbolt [1] but *not* from my installed gcc 9.4... Not sure what's going on
>> there.
>> 
>> In other bugs, there was also a problem calculating bl_shift (ffs instead of fls).
>> I probably should have tested this the first two times around, but I now have a "raw"
>> MMC working. I expect that there are probably some similar bugs lurking in spi/net/nor.
>> I will try and test net, but I don't think I have any spi/nor boards I can test with.
> 
> OK.  When you're reasonably confident SPI should work I'll go fight
> around with the am335x_evm and check my notes on how to make it SPI
> boot.  Looking at am335x_evm_spiboot_defconfig it would cover what
> you're wanting to check, yes?
> 

I'll send v3 soon and that should be good to test with. Falcon mode SPI boot also needs
to be tested, which seems to be rather unpopular:

git grep -l SPI_LOAD $(git grep -l OS_BOOT configs/)
configs/am57xx_evm_defconfig
configs/display5_defconfig
configs/display5_factory_defconfig
configs/dra7xx_evm_defconfig
configs/imx28_xea_defconfig
configs/xilinx_zynq_virt_defconfig
configs/xilinx_zynqmp_virt_defconfig

I'm not sure whether the zynq stuff is virtual as in qemu or something else.

--Sean