Support the Sortix operating system

[Jan Beulich]

On 04.11.2024 12:22, Jonas 'Sortie' Termansen wrote:
> On 11/4/24 09:16, Jan Beulich wrote:
>> Looks largely okay to me, yet one important question: Can you point us
>> at where the ELFOSABI_SORTIX value can be seen to be an officially
>> assigned value? Otherwise imo it not being an official value at least
>> needs mentioning next to the #define-s.
> 
> Oh, nobody ever assigned it officially.
> 
> Who is the OSABI allocation authority?

Cary, from a while back I seem to be remembering that you were in the
process of giving the gABI site a new home. After that the process was
supposed to be better formalized again. Mind me asking what the state
of this is, or whether I'm simply misremembering?

> Serious question because it appears the System V ABI documents have been
> forked for new platforms and architectures time and time again. I could
> only find doubt about who's responsible when I looked around. I had been
> kinda assuming it was a de-facto consensus type deal where values
> becomes basically official if they're upstreamed to important projects
> like binutils.
> 
> Sortix has been using the 83 value since 2014. The program loader
> requires it to only load compatible executables. The value was picked to
> be unlikely to conflict with anyone else (and because it's the ascii
> value of 'S'). I just realized today that it's actually inside the
> machine reserved range 64-128, although that feature barely looks used
> and I don't see how machine extensions could combine with an OS value.

I do (albeit the project that we wanted such for has been dead for
over 20 years), and to me the existence of ELFOSABI_ARM also supports
that as something potentially wanted in practice.

Using a value from that range would, aiui, require that all psABI
owners for architectures you care about would agree to have that value
reserved there. Perhaps better to have an officially assigned value
from the legitimate, non-OS range.

>> I'm further somewhat puzzled by TE_Sortix and the comment next to its
>> definition: You don't use the symbol anywhere, so why is it important
>> to have it defined (distinguishing te-sortix.h from te-generic.h)? I
>> would have expected that you also need to alter the respective #if in
>> tc-i386.c then.
> 
> Oops I did not include that hunk to tc-i386.c. Well spotted.
> 
> I attached an updated patch with that hunk included.

One more (mechanical) thing: The copyright statement in the new
te-sortix.h wants its year range extended to 2024.

Jan