Re: [RFC] Crash on modpost, addend_386_rel()

["Mika Penttilä" <mika.penttila@kolumbus.fi>]

Atsushi Nemoto wrote:
> On Tue, 22 May 2007 17:48:09 +0300, Mika_Penttilä <mika.penttila@kolumbus.fi> wrote:
>   
>> I can't see how this use of r_attend is going to work. find_elf_symbol
>> compares relsym->st_value to Elf_Rela->r_attend. I think it doesn't work
>> for RELA archs and even with this patch for REL.
>>     
>
> It seems works fine with RELA archs, at least mips64.
>
> For example, set_up_list3s is correctly reported.
>
> WARNING: mm/built-in.o - Section mismatch: reference to .init.text:set_up_list3s from .text between 'kmem_cache_create' (at offset 0x26358) and 'cache_reap'
>
> Here is excerpt from readelf output.  Addend value 0x21d8 matches
> st_value of its target symbol.
>
> $ mips64-linux-readelf -sr ../build-sb1250/mm/built-in.o
> Relocation section '.rela.text' at offset 0x33fe0 contains 5100 entries:
>   Offset          Info           Type           Sym. Value    Sym. Name + Addend
> ...
> 000000026358  000400000004 R_MIPS_26         0000000000000000 .init.text + 21d8
>                     Type2: R_MIPS_NONE
>                     Type3: R_MIPS_NONE
> ...
> Symbol table '.symtab' contains 1652 entries:
>    Num:    Value          Size Type    Bind   Vis      Ndx Name
> ...
>    746: 00000000000021d8   148 FUNC    LOCAL  DEFAULT    4 set_up_list3s
>
> ---
> Atsushi Nemoto
>
>   
So with mips64 you are lucky because the relocation symbol is .init.text 
and hence addend matches  (has to match) symbol's offset. I can't find 
any spec where it is stated that addend == address, maybe it's in mips64 
abi or something. It is quite normal to have addend of 0.

--Mika