[U-Boot] [PATCH] spl: fix entry_point equal to load_addr

[Tom Rini]

On Fri, Dec 06, 2019 at 03:05:55PM +0100, Simon Goldschmidt wrote:
> On Fri, Dec 6, 2019 at 2:49 PM Giulio Benetti
> <giulio.benetti at benettiengineering.com> wrote:
> >
> > Hello Tom, all,
> >
> > On 12/6/19 2:34 PM, Tom Rini wrote:
> > > On Fri, Dec 06, 2019 at 01:58:46PM +0100, Simon Goldschmidt wrote:
> > >> On Fri, Dec 6, 2019 at 1:46 PM Patrice CHOTARD <patrice.chotard at st.com> wrote:
> > >>>
> > >>> Hi
> > >>>
> > >>> This patch is breaking the STM32MP15 basic boot (spl => u-boot).
> > >>
> > >> Looking at socfpga gen5 u-boot.img, this is probably broken as well.
> > >>
> > >> And I don't even see any RB or TB tags here :-(
> > >
> > > Ugh, what the heck?  I applied this because I looked over the changes
> > > and they seemed correct.  I'm quite willing to just revert this but I
> > > would like to know how it's breaking.  Sorry all!
> > >
> >
> > IMHO this is due to wrong images creation with mkinage, especially when
> > passing parameters with -a and -e flags.
> >
> > In my case I need them to be:
> > -a 0x80002000 (load address) => CONFIG_SYS_TEXT_BASE
> > -e 0x800023FD (entry point where SPL jumps to) => CONFIG_SYS_UBOOT_START
> 
> Well, CONFIG_SYS_UBOOT_START is only set in 15 files in include/configs, so I
> guess a lot more boards might be broken...
> 
> >
> > So *maybe* on STM32MP1 and other broken boards -e
> > (CONFIG_SYS_UBOOT_START) is not equal to -a (CONFIG_SYS_TEXT_BASE) as
> > was assumed before(but wrong).
> >
> > Indeed CONFIG_SYS_UBOOT_START is set to 0 if not specified in
> > u-boot/Makefile:
> > `
> > # U-Boot entry point, needed for booting of full-blown U-Boot
> > # from the SPL U-Boot version.
> > #
> > ifndef CONFIG_SYS_UBOOT_START
> > CONFIG_SYS_UBOOT_START := 0
> > endif
> > `
> >
> > So probably broken boards try to jump to absolute 0.
> > A solving patch would be:
> > ifndef CONFIG_SYS_UBOOT_START
> > CONFIG_SYS_UBOOT_START := CONFIG_SYS_TEXT_BASE
> > endif
> 
> That might work, but I wonder if this is the correct time in the release to do
> so.

Yes, at this point in the cycle the best option is to revert the
original commit and for the next release bring it back after applying
Patrice's series to fix the bogus default to CONFIG_SYS_UBOOT_START and
cleaning up defconfigs.  Sorry again for all the troubles!

-- 
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL: <https://lists.denx.de/pipermail/u-boot/attachments/20191206/d6beab66/attachment.sig>