[syslinux] Question about syslinux EFI alpha version

[Matt Fleming]

On Thu, 09 May, at 08:02:24AM, H. Peter Anvin wrote:
> On 05/09/2013 07:48 AM, Matt Fleming wrote:
> > On Thu, 09 May, at 07:03:21AM, H. Peter Anvin wrote:
> >> On 05/09/2013 05:51 AM, Matt Fleming wrote:
> >>>
> >>> Yeah, that could be a good idea. It would save us from having to invent
> >>> a scheme to ensure we load the correct ldlinux.c32 depending on whether
> >>> we're booting 32-bit or 64-bit.
> >>>
> >>
> >> The question is if we don't need that anyway for modules... after all,
> >> every module, not just ldlinux.c32, is affected here.
> >>
> >> Additionally, I do believe BIOS and EFI have different .c32 modules, no?
> >  
> > Yes, the .c32 modules for BIOS, efi32 and efi64 are all different.
> > Notably the BIOS and efi32 modules are compiled with different CFLAGS
> > (the immediate thing that comes to mind is -mno-red-zone).
> > 
> 
> The redzone is an x86-64 ABI feature... -mno-red-zone should have no
> effect.  I presume you meant -mregparm=3?
> 
> Either way, it is likely that low-level accesses will seep into modules,
> so even if the ABIs are compatible it doesn't make much sense.
 
Right.

> >> Unless they are in separate trees, I think we'll need separate names.
> >  
> > Once ldlinux.c32 is loaded it can parse the PATH directive in a config
> > file, which can be used to point to separate trees. Of course, you need
> > to be able to load *different* config files for BIOS, efi32 and efi64,
> > so that moves the problem rather than solving it.
> 
> Although we may end up doing something like macros for that (people are
> asking for that) ... so it is worth thinking about it.

Yeah, I'll take a look at that.

-- 
Matt Fleming, Intel Open Source Technology Center