Re: [ping] Re: [Patch] arch recognition fix for osabi.c

[Daniel Jacobowitz <drow@mvista.com>]

On Fri, Jul 11, 2003 at 06:26:44PM +0200, Mark Kettenis wrote:
>    From: "Kris Warkentin" <kewarken@qnx.com>
>    Date: Fri, 11 Jul 2003 11:04:29 -0400
> 
>    Mark, Daniel had suggested that you were the end of the line as far as
>    approving this change.
> 
>    Thread starts here:
>    http://sources.redhat.com/ml/gdb-patches/2003-06/msg00562.html
> 
> Hmm, I folloed that thread, and I have some things to say about it.  I
> just never did :-(.
> 
> The code is deliberately written the way it is.  The idea is that for
> each CPU architecture we have a number of processors.  Some of these
> are compatible with each other.  Others aren't.  The trick is to find
> the ISA that is most compatible.  Most processor families out there
> are backwards compatible, which basically means that a new processor
> is based on ISA of its predecessor and adds some functionality to
> that.  For processor families that work like that,
> bfd_default_compatible does the right thing, and the code in
> gdbarch_init_osabi selects an OSABI/ISA that's a subset of the desired
> ISA.
> 
> Apparently this doesn't work for MIPS, since BFD declares different
> processors (which it calls "machines") to be incompatible.  I'm not
> quite familiar with MIPS, but I suppose this is not quite true, but
> that the various MIPS processors cannot be mapped on a one-dimensional
> quantity that expresses the features of the various CPU's.  That could
> be a valid reason why the MIPS "compatibility function" is written the
> way it is.  Perhaps it can be improved?  If so, I think that's the way
> to go.  Otherwise, I think you should register for all CPU types that
> you support.

It doesn't declare different machines to be incompatible.  Take another
look at mips_compatible; it is actually more lenient than the default. 
But it _always_ returns a, and osabi.c expects it to return b.  I
suppose that we could change mips_compatible to return the subset type
if one has mach == 0.  But even the default doesn't do that right now.

> One could argue that if one registers a handler for the "default
> machine" that this should apply for all machines within an
> architecture, but that breaks down for architectures that support both
> 32-bit and 64-bit machines.

Which is why the word size is checked separately in
bfd_default_compatible.  mips_compatible is a whole other kettle of
worms.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer