Re: EFI and 32bit vs. 64bit

[James Jarvis <James.Jarvis@ed.ac.uk>]

Vladimir 'phcoder' Serbinenko wrote:
> On Wed, Jun 3, 2009 at 11:32 PM, Felix Zielcke <fzielcke@z-51.de> wrote:
>   
>> Hello,
>>
>> just talked with a guy from the debian-boot people about using grub-efi
>> for the Intel Mac Users.
>> Is there any way avaible in Linux to detect somehow if the EFI is 64bit
>> or 32bit?
>>     
> If you EFI-boot linux then its architecture is also the one of efi
> (afaik linux doesn't support efi64 with i386 kernel or otherway round)
> If it's not the only reasonable way is to use dmidecode's output
> (especially ProductName)
>   
I have a 2.6.29.2 i386 linux kernel boot on a 64bit computer (Apple iMac 
8,1). With fakebios it all works fine (although one cannot access 
efivars). When I boot the i386 kernel (and initrd and root in memory) in 
EFI mode on 64 bit it boots but not all modules insert. It loads and can 
start the init script. That suggests that Felix is right the assumption 
may not be absolute.

Also, will the results of dmidecode be reliable if using fakebios???
>> Maybe some list of CPUs which could be grep'ed from /proc/cpuinfo or so?
>> I read on refit page that Apple introducted fat binaries, which would
>> work on both, but can they actually be created outside of MacOSX?
>>     
> fat binary is just a concateneting of 2 binaries with an extra header.
> So it's quite easy to create one
>   
I have already done this using the fatglue.py python script from refit. 
It works for grub.efi but not for modules. Another solution would be at 
module insertion that the process tries <name>.<arch>.mod first before 
<name>.mod. This means i386 and x86_64 modules could coexist in the same 
directory and be inserted accordingly.

James

-- 
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.