RFC: mips target support for Juniper Apollo

[Eric Christopher]

> The Juniper Apollo is a bit of an oddball target.  It is based on the
> r4000, but is a 32-bit part, so the only thing that really looks like
> the r4000 is the coprocessor 0 support.  It adds a few instructions
> which are handled by this patch.  It removed a few instructions which I
> haven't tried to handle here.  I didn't see a mechanism for this.
> 

It's not hard, but unfortunately is tedious.  It should be done before
you add the port into binutils though. The tx39 is a good example - it's
a mips2 processor minus a few instructions. As you'll see we had to add
most of them by hand in there. It makes for annoyingly long diffs but...

> It is also missing branch delay slots.  This is a serious
> incompatibility which requires having separate magic numbers for Apollo,
> as it won't be compatible with anything else.

Eek.

> 
> I'm concerned about the choice of the magic numbers.  I don't want to
> conflict with number that Red Hat is using, but only a Red Hat employee
> can check that.  This primarily means E_MIPS_MACH_APOLLO in the
> include/elf/mips.h file.
> 

Thanks for checking, but no, it doesn't conflict with anything we've
got.

> I am looking for comments, particularly from MIPS maintainers, as to
> whether the patch is reasonable, whether anything should be changed,
> etc.

It's ok with the addition of the mechanical job I mentioned above. I see
you turned off the branch delay slot optimizations and so that would
have been my only other worry.

-eric

-- 
Eric Christopher <echristo@redhat.com>