SHT_UNWIND instead of SHT_X86_64_UNWIND? (was: RFC: Usefulness of SHT_X86_64_UNWIND)

[Fāng-ruì Sòng]

On 2020-03-16, Michael Matz wrote:
>Hello,
>
>On Fri, 13 Mar 2020, Fāng-ruì Sòng via Gnu-gabi wrote:
>
>> OK, so it is unfortunate that x86-64 psABI says "The call frame
>> information needed for unwinding the stack is output into one or more
>> ELF sections of type SHT_X86_64_UNWIND." while there is no corresponding
>> change made to the most widely assembler (GNU as). This sentence
>> triggered https://reviews.llvm.org/rL252300 which made clang integrated
>> assembler diverge.
>>
>> At this point, I agree that the world is not going to be simplified.
>> Toolchain has to continue to support SHT_X86_64_UNWIND. However, I think
>> clarifying the canonical section type can guide future assembly files
>> and toolchain support.
>
>I think realistically this is the only thing we can do for the x86-64
>psABI: clarify and add acceptable section types, nothing of that will
>simplify anything.  So, I'd add SHT_PROGBITS as an additional acceptable
>type for .eh_frame, but continue to recommend SHT_X86_64_UNWIND (because
>that's in spirit), linkers will have to continue accepting both types for
>the next umpteen years.  So, that would document the de-facto state of the
>psABI with a little nudging towards a better future (the recommendation).

+1 for clarifying that SHT_PROBITS .eh_frame is acceptable.
SHT_X86_64_UNWIND is still the recommended type.

I'll update my https://reviews.llvm.org/D76151 accordingly to allow

.section .eh_frame,"a",@progbits (canonical one is @unwind)

To GNU as maintainers, should @unwind be accepted for non-x86?

>Adding a whole new general section type (SHT_UNWIND) seems to accomplish
>nothing than additional code for all existing psABIs.  For _future_ psABIs
>it might make sense to allocate and document an SHT_UNWIND now, but for
>existing ones it doesn't seem to make much sense.  (And for the general
>type: would we then require this section type to be forever associated
>with dwarf unwind info?  What about ARM unwind info, that couldn't use
>SHT_UNWIND then.  Or would we leave the specific format of SHT_UNWIND to
>the psABI, but still allow them to use that common section type despite
>principal difference to other ABIs?  All of those questions can be
>answered in multiple ways with pros and cons for each, but they need to be
>answered before a generic SHT_UNWIND could be introduced, at which point
>it's even less obvious if we should even bother)
>(FWIW, my personal opinion would be to document SHT_UNWIND in the gABI,
>with psABI to clarify content; but to _not_ make use of it in existing
>psABIs)

+1
Clarifying in gABI that 0x70000001 (processor specific) could not be
used for unrelated purposes would be nice... Sadly Xinuos (formerly SCO)
stopped maintaing the website a few years ago..

>Ciao,
>Michael.
>P.S: I wish there would have been more implementations of the x86-64 psABI
>right from the beginning ;-)