Compile error with SPL_FIT_FULL_CHECK and SPL_LOAD_FIT_FULL enabled

[Simon Glass]

Hi Philippe,

On Tue, 8 Feb 2022 at 10:12, Philippe REYNES
<philippe.reynes at softathome.com> wrote:
>
> Hi Johann,
>
> Le 08/02/2022 à 16:43, Johann Neuhauser a écrit :
> > Dear developers and Simon,
> >
> > we wanna run secure boot with U-Boot's SPL_FIT_SIGNATURE and FIT_SIGNATURE on our STM32MP1 boards and discovered the CVE-2021-27097.
> > To mitigate this vulnerability we wanna enable SPL_LOAD_FIT_FULL and SPL_FIT_FULL_CHECK.
> > If I compile any U-Boot SPL with the mentioned config symbols after commit 6f3c2d8a, it fails always with the following error message:
> >
> > Used defconfig: stm32mp15_dhcom_basic_defconfig (+ mentioned configs enabled)
> > ```
> > ...
> >    LD      spl/lib/built-in.o
> >    LD      spl/u-boot-spl
> > /usr/bin/arm-linux-gnueabihf-ld.bfd: common/built-in.o: in function `fit_check_format':
> > /mnt/work/dev/u-boot/common/image-fit.c:1591: undefined reference to `fdt_check_full'
> > make[1]: *** [scripts/Makefile.spl:432: spl/u-boot-spl] Error 1
> > make: *** [Makefile:1941: spl/u-boot-spl] Error 2
> > ```
> > After diging around to find the cause, we're out of ideas.
> > Does anyone have a clue why the needed function is not compiled in libfdt for the spl build?
> I have reproduced this issue with this config
> (stm32mp15_dhcom_basic_defconfig (+ mentioned configs enabled)).
> The function fdt_check_full is build conditionnaly:
>
> #if !defined(FDT_ASSUME_MASK) || FDT_ASSUME_MASK != 0xff
> int fdt_check_full(const void *fdt, size_t bufsize)
>
> {
>
> ...
>
> }
>
> In this config FDT_ASSUME_MASK is 0xff, so the function fdt_check_full
> is not compiled.
> I succeed to build (not tested) with this patch:
>
> --- a/scripts/dtc/libfdt/fdt_ro.c
> +++ b/scripts/dtc/libfdt/fdt_ro.c
> @@ -937,4 +937,9 @@ int fdt_check_full(const void *fdt, size_t bufsize)
>                  }
>          }
>   }
> +#else
> +int fdt_check_full(const void *fdt, size_t bufsize)
> +{
> +       return 0;
> +}
>   #endif
>
>
> If simon agrees with this fix, I may sent a patch.

I suppose that is OK. It really should go upstream though, where this
code is slightly different at present.

Regards,
Simon