Re: Booting on RiscPC

[Stephen Borrill <netbsd%precedence.co.uk@localhost>]

On Tue, 26 Mar 2019, Mike Pumford wrote:
[snip]
> The podule problem is a lot more awkward to solve as it relates to the fact 
> that NetBSD-current now uses GCC 7. GCC7 has dropped support for the armv3 
> instruction set which means that the underlying compiler architecture has 
> changed for acorn32. Whilst the strongARM is a v4 instruction set CPU there 
> is a hardware issue on the RiscPC that stops the ldrh instruction from 
> working. This meant that the podule identification code was losing the top 
> byte of the podule vendor/device IDs.
>
> build.sh can be persuaded to revert to GCC 6. To do this you need to set the 
> environment:
> HAVE_GCC=6
>
> You also need to add '-a arm' to the build.sh options when building things.
>
> These force build.sh to use gcc.old rather than gcc and to select the armv3 
> backend. This may make it very difficult for the acorn32 port to continue to 
> exist as its going to require local patches to gcc to keep things going.
>
> If we can persuade gcc7 not not generate ldrh but otherwise be armv4 it might 
> be workable but thats way outside my skillset. :(

OK, so -current drops support for ARM6/7 unless GCC6 is used. This
kills A7000, NC and RiscPC 600/700.

It may be possible to forward port the GCC5/6 code to future GCCs for the
armv3 target, but it would require someone to maintain it. I don't think 
llvm will help us here. Anyway, this part is probably better aimed at 
port-arm%netbsd.org@localhost (Cc: added).

--
Stephen