[U-Boot] [PATCH v2] [x86] Don't relocate symbols which point to things that aren't relocated

[Graeme Russ]

Hi Mike,

On Wed, Nov 9, 2011 at 12:46 AM, Mike Frysinger <vapier at gentoo.org> wrote:
> On Tuesday 08 November 2011 02:08:42 Gabe Black wrote:
>> This change adds an upper bound for symbols which are fixed up after u-boot
>> is relocated into RAM. This way portions that are left at their original
>> location can be referred to without having to manually fix up any pointers.
>
> i'm not familiar with the u-boot relocation handling on x86, but if a symbol
> shouldn't be relocated, then why does it have an elf reloc ?  shouldn't it be
> an absolute symbol ?

As Gabe pointed out, all x86 source files are compiled with the same
options (i.e. with relocation enabled). This includes the 16-bit realmode
sections which are actually designed to run at a fixed location in memory.
Now this introduces a few issues:

 a) There are entries in the relocation table that point to relocations in
    lower memory (i.e. below TEXT_BASE) - That's the first test
 b) Where the 32-bit bit code transitions to 16-bit, there is a relocation
    entry which points to a location above TEXT_BASE, but because the jump
    is to a fixed location in memory (below TEXT_BASE) it must not be
    adjusted (this is the second test)

There is also a couple of relocations that do not need adjusting in the
reset vector code and real-to-protected mode switch, but these are a don't
care case because by that stage, the code has already been run and will not
be run again so it really does not matter what we do to them

Now I think we might be able to compile the 16-bit code without relocations
(would need to tweak how the compiler flags are handled and split them up
a bit) and also change the 32-to-16 bit jump so it does not get a relocation
entry (I think I have a patch somewhere already as part of my massive
32-to-16 bit switch change) so we might be able to get rid of both test at
some stage...

Regards,

Graeme