Re: [patch/rfc,rfa:ppc64] Add osabi wildcard support

[Daniel Jacobowitz <drow at mvista dot com>]

On Thu, Oct 23, 2003 at 07:25:37PM -0400, Andrew Cagney wrote:
> Hello,
> 
> The attached patch adds the ability to specify a wildcard machine when 
> registering an OSABI / arch / machine.  It then updates PPC64 GNU/Linux 
> to specify that wild card (-1) instead of zero as the default machine.
> 
> Looking at the PPC64 GNU/Linux code:
> 
>   gdbarch_register_osabi (bfd_arch_powerpc, 0, GDB_OSABI_LINUX,
>                           ppc_linux_init_abi);
> 
> I believe that the call is attempting to register ppc_linux_init_abi as 
> the OSABI handler for all arch/machine conbinations.  The problem is 
> that machine "0" gets turned into bfd_mach_ppc or bfd_mach_ppc64 
> dependant on how GDB is built, and they are both incompatible with each 
> other and incompatible .  And that in turn restricts the support to just 
> one half of the ISA family making it impossible for GDB to debug both 32 
> and 64 bit :-(
> 
> I know of two ways to fix this.  First is the attached patch which 
> modifies osabi.[hc] so that a wildcard machine (-1) can be specified vis:
> 
>   gdbarch_register_osabi (bfd_arch_powerpc, -1, GDB_OSABI_LINUX,
>                           ppc_linux_init_abi);
> 
> and the second is to explicitly register both of these architecture 
> variants vis:
> 
>   gdbarch_register_osabi (bfd_arch_powerpc, bfd_mach_ppc, ...
>   gdbarch_register_osabi (bfd_arch_powerpc, bfd_mach_ppc64, ...
> 
> (possibly also splitting ppc_linux_init_abi into ppc_linux_init_abi_32 
> and ppc_linux_init_abi_32).
> 
> There are pros and cons to both.
> 
> The former will always match, making the code somewhat future proof, the 
> later is far simplier.
> 
> preferences?

My preference is for registering both.  This is really a special case;
normally 0 is correct.  It wouldn't have worked before but will with
Jim's recent change to can_run_code_for.

> Andrew
> 
> PS: I'm also wondering if the existing ppc arch/machine table is 
> complete.  I think it will indicate that e500 is "compatible" with "ppc" 
> when it is not.

The table is almost certainly not complete, but that's not the problem
here.  The e500 is compatible with ppc - the e500 can run ppc code. 
Vice versa is not necessarily true.  There's a FIXME in osabi.c saying
that we should look for the most compatible architecture:

>  	 type that is compatible with the desired machine type.  Right
>  	 now we simply return the first match, which is fine for now.
>  	 However, we might want to do something smarter in the future.  */
> -      compatible = arch_info->compatible (arch_info, handler->arch_info);
> +      compatible = info.bfd_arch_info->compatible (info.bfd_arch_info,
> +						   handler->arch_info);
>        if (compatible == handler->arch_info)
>  	{
>  	  (*handler->init_osabi) (info, gdbarch);



-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer