[PATCH v2] platform: generic: Keep some empty space in FDT passed to next stage

[Anup Patel]

On Tue, Dec 9, 2025 at 12:03 PM Samuel Holland
<samuel.holland at sifive.com> wrote:
>
> Hi Anup,
>
> On 2025-12-07 11:53 PM, Anup Patel wrote:
> > On Sun, Dec 7, 2025 at 4:29 PM Samuel Holland <samuel.holland at sifive.com> wrote:
> >> On 2025-11-30 11:00 PM, Anup Patel wrote:
> >>> Leaving no empty space in the FDT passed to the next booting stage
> >>> causes the following U-Boot crash on Ventana internal platforms:
> >>>
> >>> Unhandled exception: Load access fault
> >>> EPC: 00000000fffa6372 RA: 00000000fffa7418 TVAL: 0001746174730068
> >>> EPC: 0000000080245372 RA: 0000000080246418 reloc adjusted
> >>>
> >>> SP:  00000000fef38440 GP:  00000000fef40e60 TP:  0000000000000000
> >>> T0:  00000000fef40a70 T1:  000000000000ff00 T2:  0000000000000000
> >>> S0:  00000000fffc17a8 S1:  00000000fef38d40 A0:  7375746174730068
> >>> A1:  00000000fffc17a8 A2:  0000000000000010 A3:  0000000000000010
> >>> A4:  0000000000000000 A5:  00000000fffc17b8 A6:  0000000000ff0000
> >>> A7:  000000000000b100 S2:  0000000000000000 S3:  0000000000000001
> >>> S4:  00000000fef38d40 S5:  7375746174730068 S6:  0000000000000000
> >>> S7:  00000000fef4eef0 S8:  00000000fef4ef90 S9:  0000000000000000
> >>> S10: 0000000000000000 S11: 00000000fef4efc0 T3:  00000000fef40ea8
> >>> T4:  0000000000ff0000 T5:  00000000fef40a60 T6:  00000000fef40a6c
> >>>
> >>> To address the above issue, keep some minimal empty space in the
> >>> FDT instead of no empty space.
> >>>
> >>> Fixes: bbe9a23060e9 ("platform: generic: Pack the FDT after applying fixups")
> >>> Signed-off-by: Anup Patel <apatel at ventanamicro.com>
> >>> ---
> >>> Changes since v1:
> >>> - Add Kconfig option for amount of empty space in FDT with default
> >>>   value as 4KB
> >>> ---
> >>>  platform/generic/Kconfig    | 5 +++++
> >>>  platform/generic/platform.c | 3 +++
> >>>  2 files changed, 8 insertions(+)
> >>>
> >>> diff --git a/platform/generic/Kconfig b/platform/generic/Kconfig
> >>> index 25b8886b..b1808012 100644
> >>> --- a/platform/generic/Kconfig
> >>> +++ b/platform/generic/Kconfig
> >>> @@ -23,6 +23,11 @@ config PLATFORM_GENERIC_MINOR_VER
> >>>       range 0 65535
> >>>       default 1
> >>>
> >>> +config PLATFORM_GENERIC_FDT_EMPTY_SPACE
> >>> +     int "Platform FDT empty space (KB)"
> >>> +     range 0 1024
> >>> +     default 4
> >>
> >> Since this is a workaround for a specific platform, and the workaround should
> >> eventually be obsoleted by a U-Boot fix, please default this to zero.
> >
> > Although the issue manifests on Ventana internal platforms,
> > the issue is not Ventana specific and can happen on any platform
> > where U-Boot tries to modify FDT passed by OpenSBI without
> > adding extra space.
>
> It's still a U-Boot bug, no? Is your plan to keep the workaround indefinitely?

It is certainly a U-Boot bug. We will keep the workaround for atleast
6 months (one OpenSBI release) after U-Boot has fixed it.

>
> >>> +
> >>>  config PLATFORM_ALLWINNER_D1
> >>>       bool "Allwinner D1 support"
> >>>       depends on FDT_IRQCHIP_PLIC
> >>> diff --git a/platform/generic/platform.c b/platform/generic/platform.c
> >>> index e66f99fa..ddfc9e80 100644
> >>> --- a/platform/generic/platform.c
> >>> +++ b/platform/generic/platform.c
> >>> @@ -245,7 +245,10 @@ int generic_final_init(bool cold_boot)
> >>>       fdt_fixups(fdt);
> >>>       fdt_domain_fixup(fdt);
> >>>
> >>> +     /* Minimize the empty space in FDT to 4KB */
> >>
> >> This comment is no longer accurate, since the padding is not a fixed size.
> >
> > Sure, I will update.
> >
> >>
> >>>       fdt_pack(fdt);
> >>> +     fdt_open_into(fdt, fdt, fdt_totalsize(fdt) +
> >>> +                   CONFIG_PLATFORM_GENERIC_FDT_EMPTY_SPACE * 1024);
> >>
> >> This change reintroduces the problem that motivated adding fdt_pack().
> >> fdt_open_into() does not initialize the unused space, so once again the range
> >> [fdt, fdt + fdt_totalsize(fdt)) contains unitialized data.
> >>
> >> If having unitialized data in the FDT blob is not a problem on the platform
> >> requiring this workaround, then I suppose the code is fine. Otherwise, this
> >> would need a memset().
> >
> > I think it is better to do memset() on additional space in fdt_open_into()
> > itself but that would be a separate change in libfdt.
>
> I don't have an opinion on whether it is better to change libfdt, but we need to
> add the memset() in at least one place to avoid regressing here.
>

Understood. My suggestion is to do it as a separate patch irrespective
of where we do it.

Regards,
Anup