This is the mail archive of the
libc-ports@sources.redhat.com
mailing list for the libc-ports project.
Re: [patch] handle unaligned arm abs relocs
On 12/14/2011 5:29 PM, Mike Frysinger wrote:
> On Wednesday 14 December 2011 17:05:25 Carlos O'Donell wrote:
>> On 12/12/2011 7:20 PM, Mike Frysinger wrote:
>>> background can be found here:
>>> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51456
>>>
>>> 2011-12-12 Mike Frysinger <vapier@gentoo.org>
>>>
>>> * dl-machine.h (elf_machine_rel, R_ARM_ABS32): Declare "reloc_value".
>>> Replace reloc_addr addition with memcpy's.
>>>
>>> diff --git a/sysdeps/arm/dl-machine.h b/sysdeps/arm/dl-machine.h
>>> index 5ceeaa9..fe64800 100644
>>> --- a/sysdeps/arm/dl-machine.h
>>> +++ b/sysdeps/arm/dl-machine.h
>>> @@ -413,6 +413,7 @@ elf_machine_rel (struct link_map *map, const
>>> break;
>>> case R_ARM_ABS32:
>>> {
>>> + Elf32_Addr reloc_value;
>>> # ifndef RTLD_BOOTSTRAP
>>> /* This is defined in rtld.c, but nowhere in the static
>>> libc.a; make the reference weak so static programs can
>>> @@ -431,7 +432,10 @@ elf_machine_rel (struct link_map *map, const
>>> used while loading those libraries. */
>>> value -= map->l_addr + refsym->st_value;
>>> # endif
>>> - *reloc_addr += value;
>>> + /* Support relocations on mis-aligned offsets. */
>>> + memcpy (&reloc_value, reloc_addr_arg, sizeof (reloc_value));
>>> + reloc_value += value;
>>> + memcpy (reloc_addr_arg, &reloc_value, sizeof (reloc_value));
>>
>> I don't believe that memcpy is safe this early in the loader.
>>
>> You might be lucky and get an inlined memcpy or builtin, but you
>> might not and if you go through the PLT you'll fault since
>> it's not yet setup.
>
> i thought it should be OK because R_ARM_COPY already calls memcpy() a few
> lines up, but i don't know the ARM loader that well to say what is necessary
COPY relocs are only for non-PIC executables accessing data from other
PIC objects. Almost all executables have a COPY reloc against stderr/stdout.
See this for some hand waving:
http://docs.oracle.com/cd/E19082-01/819-0690/chapter4-84604/index.html
> looking at the relocs that exist on my local arm ldso and i only see:
> R_ARM_RELATIVE
> R_ARM_GLOB_DAT
> R_ARM_JUMP_SLOT
>
> hopefully someone who understands ARM better can comment ...
If it works, and you test that it works, then you're fine.
If you use mempcy and it's inlined then that's great, but it's
a possible point of failure when compiling glibc (or eglibc)
with options that don't inline memcpy e.g. -Os.
My warning to you is this: be very very careful calling any
library functions in dl-machine.h.
Cheers
Carlos.
--
Carlos O'Donell
Mentor Graphics / CodeSourcery
carlos@codesourcery.com
+1 (613) 963 1026