https://sourceware.org/bugzilla/show_bug.cgi?id=33177
--- Comment #21 from Ali Bahrami <ali_swbugzilla at emvision dot com> ---
(In reply to Alan Modra from comment #20)
> OK, getting back to this. I have some reservations about making the global
> ABI symbols SHN_ABS. (I made the patch to demonstrate how it is done,
> that's all.) The problem with SHN_ABS symbols holding addresses as is the
> case with all of these symbols, is that in a shared libary or PIE those
> symbols must be relocated by the base address of the library if they are
> used in dynamic relocations. Which is wrong according to the ELF gABI:
>
> SHN_ABS
> The symbol has an absolute value that will not change because of
> relocation.
>
> For that reason GNU binutils has moved away from making
> _GLOBAL_OFFSET_TABLE_ and similar symbols SHN_ABS on a number of targets.
I just noticed this, from a couple of days ago, sorry for
the delay. I may have missed something, but in case there's
a question, the Solaris ld doesn't use SHN_ABS for these symbols:
% elfdump -sN.dynsym /lib/64/libc.so.1 | egrep
'GLOBAL|PROCED|DYNAM|_end|_fini|_init|_start'
[86] 0x1352c0 0x2b FUNC LOCL H 0 .fini _fini
[89] 0x135290 0x26 FUNC LOCL H 0 .init _init
[12813] 0x3c28f4 0 OBJT GLOB D 0 .bss _end
[13395] 0x7590 0 OBJT GLOB D 0 .dynamic _DYNAMIC
[13625] 0x134f08 0 OBJT GLOB D 0 .plt
_PROCEDURE_LINKAGE_TABLE_
I don't know how much using ABS would really break for our
linkers, but it does seem that these symbols should be
marked as data or functions as appropriate instead.
--
You are receiving this mail because:
You are on the CC list for the bug.
