[PATCH] PR30437 aarch64: make RELA relocs idempotent

[Richard Earnshaw (lists)]

On 25/05/2023 13:29, Michael Matz via Binutils wrote:
> Hello,
> 
> On Thu, 25 May 2023, Richard Earnshaw (lists) wrote:
> 
>> Thanks for the detailed explanation.  That makes sense, overall.  However...
>>
>> I didn't think we had separate relocation rules for REL/RELA format
>> relocs.  I know that we recommend RELA for aarch64, but how would an
>> object file with REL format relocs be handled if src_mask is always
>> zero?
> 
> Right now the aarch64 bfd backend only support RELA
> (elf_backend_may_use_rel_p is off).  elf64-mips is an example that
> supports both, and it has separate howto tables for rel and rela
> descriptors (e.g. mips_elf64_howto_table_rel and
> mips_elf64_howto_table_rela), with the appropriate
> src_mask/partial_inplace settings depending on rel/rela:
> 
>    HOWTO (R_MIPS_32,             /* type */
>           ...
>           true,                  /* partial_inplace */
>           0xffffffff,            /* src_mask */
>           0xffffffff,            /* dst_mask */
> 
> vs.
> 
>    HOWTO (R_MIPS_32,             /* type */
>           ...
>           false,                 /* partial_inplace */
>           0,                     /* src_mask */
>           0xffffffff,            /* dst_mask */
> 
> the rtype_to_howto backend routine appropriately switches between those
> depending on side-info passed in parameters.  (Whereas the name_lookup and
> type_lookup routines always use the rela tables and contain a FIXME
> comment to that effect :) ).  Various non-mips embedded targets try to
> also support this, but often only for read-in, not for generating both
> types of relocs, and so get away with only one howto table.  (I believe (!)
> mips is the only one that actively tries to get REL+RELA support correct).
> 
> 
> Ciao,
> Michael.

I'd be surprised if even that is totally correct.  ELF permits 
relocations of a single 'place' to be chained provided that they are of 
the same type (all REL or all RELA); not that I've ever seen that done 
in practice.  When this is done the addend for the second and subsequent 
relocations use the result of the previous relocation - they must be 
processed in order.  Only the last relocation in a sequence writes a 
value back to the location being relocated.

So really, we shouldn't need two tables, we just need to know where to 
get the addend value from at the start of a chain.

R.