Re: Debian ARM achitectures in Sid

[Jim Pick <jim@jimpick.com>]

Martin Schulze <joey@finlandia.Infodrom.North.DE> writes:

> Jim Pick wrote:
> > 
> > Hi,
> > 
> > After discussions with the Linux/ARM people (Russell King, and
> > friends), we've come to the decision that it would be best to have two
> > separate ARM architectures:
> > 
> >   armv3  - primarily for use on Acorn RiscPC machines - including
> >            ones which use the ARM6, ARM7, and StrongARM (without halfword
> >            support) cpus.  This could be a very popular port, as Acorn
> >            has decided to discontinue manufacturing PCs - so the existing
> >            customers will be looking for a new OS to install.
> > 
> >   armv4  - for StrongARM machines, such as the Corel NetWinder.  With Intel's
> >            backing, there are probably going to be a lot of StrongARM machines
> >            coming onto the market in the next year or so.

 
> Are they really binary-incompatible?

The armv3 instruction set is compatible with the armv4 instruction
set, however, on the linux-arm list, Scott Bambrough
<scottb@corelcomputer.com> posted:

  The NetWinder requires -march=armv4 to ensure some cpu cache issues are
  handled correctly, however, the cpu could be left as default.

So there might actually be some incompatibility there.  I'm not
exactly sure what the issues are.

As I understand it, there is going to be a significant performance
difference between binaries built for armv3 and armv4, because of the
addition of several new instructions.  As the StrongARM is probably
going to be a very popular chip going forward (Intel is now pushing
it), I don't think it would be wise to not optimize for it.

There was a pretty long discussion about which arm architectures
should be supported on the linux-arm mailing list over in the past two
weeks.  There are even people with old 26-bit armv2 machines
(vs. 32-bit for the armv3 and beyond), but I don't think anybody is
serious about having a distribution for those.

Cheers,

 - Jim