On Saturday 04 August 2001  6:41 pm, Mike Pumford wrote:
> > Hi,
> >
> > I was just looking over the code and noticed that we currently use a per
> > process undefined instruction stack.  I was wondering if anyone knew why?
>
> Could it be remains from the time when the port used FPE rather than
> soft-float for floating point support. When doing FP by instruction
> emulation wouldn't every process need its own state whilst processing an
> undefined instruction to maintains FP state for each process?

I'd didn't think that instruction emulation could be interrupted, but yes it 
could be for that, but I'd have hoped that the fpe would allocate up memory 
it needs (although I've not idea where from)

> > We appear to actually setup a global one but then create one for every
> > process created.
> >
> > I'd rather like to switch to using a global one as it means that we don't
> > have to keep saving and restoring the undef instr stack pointer, and so
> > simplifies the code.
>
> Unfortunately the only thing I can suggest is to try it and see if anything
> breaks.

I'm tempted to, it's one of the diffs between the arm26 and arm32 pcbs.

Cheers,
Chris