[U-Boot] [PATCH] Fix Freescale link scripts for newer GCCs

[Trent Piepho]

On Sat, 31 Jan 2009, Wolfgang Denk wrote:
> In message <Pine.LNX.4.64.0901310259430.8685 at t2.domain.actdsltmp> you wrote:
> > > so, what's the status of this patch?  I've seen this fail on 83xx.
> > > Most of these types of patches that fix the newer gcc's behaviour have
> > > been dropped, one of which even did sorting on name and alignment:
> > >
> > > http://www.mail-archive.com/u-boot@lists.denx.de/msg03193.html
> > >
> > > Do I need to make one for 83xx and apply it myself?
> >
> > Wolfgang rejected mine because he wanted a patch that fixed every single
> > board in one shot.  I didn't feel like patching a hundred other linker
> > scripts I knew nothing about.
>
> Why not? If you know what you are doing (and I have strong resons to
> believe that you do) then I think your fix should be applied to all
> systems that are affected by the problem.

But what systems are affected by the problem?  For instance ms7750se only
has ".rodata", not even ".rodata.str1.4".  So should it get changed too?

There are CPUs that have a special segment for byte aligned data.  Does
u-boot support any of these CPUs?  I don't know.  But changing ".rodata" to
".rodata*" could break a system if ".rodata1" was supposed to be picked up
by a ".*data*1" somewhere else in the linker script.  It seems like people
who know the platform would be less likely to get caught by this.

Or take something like mpc8220, which has this for a text segment:
    *(.text)
    *(.fixup)
    *(.got1)
    . = ALIGN(16);
    *(.rodata)
    *(.rodata1)
    *(.rodata.str1.4)
    *(.eh_frame)

Why is the start of rodata aligned to 16 bytes?  I'd guess some sort of I/D
cache issue.  Though aren't fixup and got1 data, not code?  I can't
find any docs on what the differences between fixup, got, got1, and got2
are.  It seems like current gcc only uses got2.  The mpc8220 ld script also
lists fixup a second time, in the reloc section, a mistake that probably
doesn't break anything since I think gcc hasn't used fixup in some time.

Anyway, not knowing if it's ok to delete fixup and got1, I would change
this to:
    *(.text)
    *(.fixup)
    *(.got1)
    . = ALIGN(16);
    *(.eh_frame)
    *(SORT_BY_ALIGNMENT(SORT_BY_NAME(.rodata*)))

eh_frame should go before rodata* since it has has a larger alignment, but
now rodata isn't 16 byte aligned like it was before.  Does that matter?  I
doubt it, but I don't know for sure.