Re: [rfc] Use target descriptions for PowerPC

[Thiago Jung Bauermann <bauerman@br.ibm.com>]

On Fri, 2007-10-05 at 19:59 -0400, Daniel Jacobowitz wrote:
> On Fri, Oct 05, 2007 at 06:40:46PM -0300, Thiago Jung Bauermann wrote:
> > Hi Daniel,
> > 
> > On Fri, 2007-10-05 at 12:37 -0400, Daniel Jacobowitz wrote:
> > > +  /* If we have a 64-bit binary on a 32-bit target, complain.  Also
> > > +     complain for a 32-bit binary on a 64-bit target; we do not yet
> > > +     support that.  */
> > > +  if (tdesc_wordsize != -1 && tdesc_wordsize != wordsize)
> > > +    {
> > > +      tdesc_data_cleanup (tdesc_data);
> > > +      return NULL;
> > > +    }
> > 
> > Sorry if I am confusing things here. This sounds like it affects 64 bit
> > GDB debugging 32 bit binary. Is this the case? Or only for remote
> > debugging?
> 
> Good noticing, it could be a problem, but it works out fine - we don't
> have a supplied target description in this case, so we pick a default
> based on the 32-bit binary, and get a 32-bit binary; everything
> matches.  Then the GNU/Linux native code handles the details.  I've
> just tested that.
> 
> The problem I'm trying to prevent is that we will always use registers
> of the size described by the target description.  And we don't support
> implementing the 32-bit ABI on top of 64-bit GPRs; we'll copy
> arguments into the high half of registers when calling functions, for
> instance.  Try connecting a GDB to a 64-bit gdbserver and it will go
> wrong in a couple of ways.  The native targets fake this by knowing
> how to supply 32-bit registers when expected, even if you'd think
> there ought to be 64-bit registers.

Nice, thanks for the explanation. Would you mind expanding the comment
in the code above to include it?
-- 
[]'s
Thiago Jung Bauermann
Software Engineer
IBM Linux Technology Center