Re: Attempt to fix #189031
Greg Lindahl wrote:
> On Thu, Apr 17, 2003 at 08:28:09PM +0200, Thiemo Seufer wrote:
>
> > I guess you are talking about the work already contributed by
> > Alexandre Oliva. The "simple solution" mentioned before was a
> > part of it. The relocation handling is still horrible.
>
> I suppose I should say that "not as bad as before" means "can mostly
> generate working n32/n64 programs", not that it is any less horrible.
With 'horrible', I meant the binutils implementation, not the
resulting code. :-)
> > > By the way, the 32-bit kernel could have an N32 mode added; it's
> > > fairly straightforward to merge the bits currently in the 64-bit
> > > kernel that you'd need. You'd also need a very modern glibc.
> >
> > It definitely can't. You are missing the fact that N32 has 64bit
> > register widths.
>
> Why does that make a difference? I suppose that I don't know the
> details of what mode things like the context switch routines run in.
> >From looking at the code, it would seem that it's the 32-bit mode in
> the 32-bit kernel.
What anwers your question. N32 will need e.g. to save&restore 64bit
wide registers, does syscalls with 64bit registers and so on.
A 32bit kernel doesn't know how to handle this.
Thiemo
Reply to: