[binutils] Special processing of R_X86_64_32 relocations

[binutils] Special processing of R_X86_64_32 relocations

R_X86_64_32 relocations in NaCl64 play the same role as R_X86_64_64 in native x86_64. They should be rewritten in case relocated symbol is hidden and will not exist in statically linked binary.

R_X86_64_64 are rewritten into R_X86_64_RELATIVE using the absolute value of the symbol. Unfortunately we don't have 32-bit alternative, thus we change relocation to use section instead of symbol. Usually this rewrite produces buggy relocations (why? there is corresponding comment in bfd, but it does not explain much), but in our special case we just create valid relocations.

BUG=http://code.google.com/p/nativeclient/issues/detail?id=1095
TEST=

[Mark Seaborn, 2010-10-26 11:13:30]

On 2010/10/26 10:39:23, Evgeny Eltsin wrote:
> BUG=
> TEST=

There's no point in putting BUG and TEST fields if they're empty.  "BUG=none"
and "TEST=none" would be preferred to empty fields.

Can you create a bug and explain on there how to reproduce this issue, e.g. how
to build an ELF executable/library that exhibits this and the expected/actual
output from readelf or objdump.

Mark

[eaeltsin, 2010-10-26 11:55:25]

BUG=http://code.google.com/p/nativeclient/issues/detail?id=1095

TEST= and BUG= are from git-cl. Next time I'll simply erase these lines.

I want to submit this CL as it blocks 64-bit glibc development. However, it is
likely to be rolled back somewhat soon, as from my current understanding it is a
hack.

[pasko-google - do not use, 2010-10-26 15:14:28]

LGTM