[llvm-dev] [lld] R_MIPS_HI16 / R_MIPS_LO16 calculation

[Simon Atanasyan via llvm-dev]

On Sat, Nov 21, 2015 at 10:08 AM, Rui Ueyama <ruiu at google.com> wrote:
> On Fri, Nov 20, 2015 at 11:02 PM, Simon Atanasyan <simon at atanasyan.com>
> wrote:
>>
>> On Sat, Nov 21, 2015 at 9:28 AM, Rui Ueyama <ruiu at google.com> wrote:
>> > On Fri, Nov 20, 2015 at 10:13 PM, Simon Atanasyan <simon at atanasyan.com>
>> > wrote:
>> >>
>> >> In case of MIPS O32 ABI we have to find a matching R_MIPS_LO16
>> >> relocation to calculate R_MIPS_HI16 one because R_MIPS_HI16 uses
>> >> combined addend (AHI << 16) + (short)ALO where AHI is original
>> >> R_MIPS_HI16 addend and ALO is addend of the matching R_MIPS_LO16
>> >> relocation [1]. There are two methods to do matching and R_MIPS_HI16
>> >> calculation.
>> >>
>> >> Method A:
>> >> 1. Postpone R_MIPS_HI16 relocation calculation and record its
>> >> arguments.
>> >> 2. When R_MIPS_LO16 is found, iterate over recorded R_MIPS_HI16
>> >> relocations, calculate combined addend and apply relocations.
>> >> 3. At the end check orphaned (without R_MIPS_LO16 pair) R_MIPS_HI16
>> >> relocations, show warnings and apply them with zero addend.
>> >>
>> >> Method B:
>> >> 1. Each time we have found R_MIPS_HI16 relocation, iterate remaining
>> >> relocations list to find matching R_MIPS_LO16.
>> >> 2. Calculate combined adddend and apply relocation or show warning if
>> >> the R_MIPS_LO16 is not found.
>> >>
>> >> Method A requires some sort of container to keep postponed HI16
>> >> relocations. If we add the container to the `MipsTargetInfo` class we
>> >> will be able to hide all this unusual scheme inside MIPS specific code
>> >> and will not need to perform LO16 lookups. But the `MipsTargetInfo`
>> >> becomes stateful.
>> >>
>> >> Method B keeps the `MipsTargetInfo` stateless but requires forward
>> >> LO16 lookup for each HI16 relocation and requires to provide an
>> >> interface for such lookup to the `MipsTargetInfo`.
>> >>
>> >> Sure we can implement each of these methods somewhere in the
>> >> `InputSectionBase` class under `if (MIPS)` statements.
>> >>
>> >> Any opinions about the best method / approach?
>> >
>> >
>> > If I understand that spec correctly, an R_MIPS_HI16 should immediately
>> > be
>> > followed by an R_MIPS_LO16. Can't you use that property? It doesn't seem
>> > to
>> > me that you really have to search and pair up HI16 and LO16 relocations.
>>
>> It is a question what the ABI authors did mean by the "R_MIPS_HI16
>> must have an associated R_MIPS_LO16 entry immediately following it"
>> phrase. In fact you can get from a compiler this code:
>>
>> lui    $t0,%hi(sym1+4)      # R_MIPS_HI16
>> lui    $t0,%hi(sym1+8)      # R_MIPS_HI16
>> lui    $t0,%hi(sym1+12)     # R_MIPS_HI16
>> addi   $t0,$t0,%lo(sym1+16) # R_MIPS_LO16
>
>
> The first two relocations don't conform to the standard because there are no
> corresponding LO16 relocations, no?
>
>> and even such code:
>>
>> lui    $t0,%hi(sym1)     # R_MIPS_HI16 on sym1
>> lui    $t0,%hi(sym2)     # R_MIPS_HI16 on sym2
>> addi   $t0,$t0,%lo(sym1) # R_MIPS_LO16 on sym1
>> addi   $t0,$t0,%lo(sym2) # R_MIPS_LO16 on sym2
>
>
> Hmm, isn't this a violation of the ABI? My interpretation of "[e]ach
> relocation type of R_MIPS_HI16 must have an associated R_MIPS_LO16 entry
> immediately following it in the list of relocations" is not ambiguous to
> allow them. Is there any chance to fix the compiler? (I guess there isn't,
> though.)

Strictly speaking yes, it is a violation. But it is not a bug of the
single compiler. You can find such code everywhere from various
versions of libc compiled by different versions of gcc and to the code
produced by Clang.

Moreover, I scan through the libc code and found some places where
R_MIPS_HI16 / R_MIPS_LO16 pairs are interleaved with other types of
relocations.

-- 
Simon Atanasyan