Re: git: 60f098f84119 - main - arch: Update to remove armv6

From: Warner Losh <imp_at_bsdimp.com>
Date: Mon, 15 Jul 2024 18:23:48 UTC
On Mon, Jul 15, 2024 at 7:03 AM Ed Maste <emaste@freebsd.org> wrote:

> On Sun, 14 Jul 2024 at 02:00, Warner Losh <imp@freebsd.org> wrote:
> >
> >  .Dv aarch64
> >  will support execution of
> > -.Dv armv6
> > -or
> >  .Dv armv7
> >  binaries if the CPU implements
> >  .Dv AArch32
> > -execution state, however older arm binaries aren't supported.
> > +execution state, however older arm binaries are not supported by
> > +.Fx .
>
> Do older 32-bit Arm userland binaries actually not work? Would
> "support execution of 32-bit Arm binaries if the CPU implements the
> AArch32 execution state" be accurate?
>

armv7 works.

armv6 likely works since they are quite similar to armv7. The only
difference is in userland APIs.

armv4 and v5 do not work if they used RAS (atomics). We don't implement the
kernel part of RAS anymore, so those cannot work today. So nothing threaded
will work. Some non-threaded things might work, but I don't think anybody
has tested it, or seen if there's other, currently unknown, issues.

So if you want to run an armv7 binary AND the CPU implements aarch32
execution states, then you will succeed. if armv6 and aarch64 it may work,
but isn't supported. Otherwise, it won't work (though some trivial examples
of armv4/armv5 might work).

So I omitted the edge case details: armv6 likely will work, but is
unsupported. armv4 and armv5 binaries generally won't work, though some
restricted cases might work. In both of these cases, if they don't work or
break in the future, we don't care (as a project).

Clear as mud, eh?

Warner