[coreboot] coreboot.v2+seabios on 440bx: option roms not found

[Marc Jones]

Andriy Gapon wrote:
> on 30/10/2008 16:52 Myles Watson said the following:
>   
>>> -----Original Message-----
>>> From: Andriy Gapon [mailto:avg at icyb.net.ua]
>>> Sent: Thursday, October 30, 2008 8:43 AM
>>> To: Myles Watson
>>> Cc: 'Coreboot'
>>> Subject: Re: [coreboot] coreboot.v2+seabios on 440bx: option roms not
>>> found
>>>
>>> on 30/10/2008 16:34 Myles Watson said the following:
>>>       
>>>>> -----Original Message-----
>>>>> From: coreboot-bounces at coreboot.org [mailto:coreboot-
>>>>>           
>>> bounces at coreboot.org]
>>>       
>>>>> On Behalf Of Andriy Gapon
>>>>> Sent: Thursday, October 30, 2008 8:25 AM
>>>>> To: Coreboot
>>>>> Subject: [coreboot] coreboot.v2+seabios on 440bx: option roms not found
>>>>>
>>>>>
>>>>> In northbridge/intel/i440bx/raminit.c:sdram_set_spd_registers all PAM
>>>>> registers are programmed for RAM R/W access (0x33).
>>>>> When SeaBIOS searches for option ROMs (including VGA ROM) it doesn't do
>>>>> anything about PAM, so it sees empty memory instead of the ROMs.
>>>>>
>>>>> I am not sure what is the best solution here. It is debatable how
>>>>> coreboot should set PAM register, and it is not right to make SeaBIOS
>>>>> too hardware dependent.
>>>>>
>>>>> Maybe coreboot could somehow export functions for setting access to
>>>>> option ROM space (aka legacy memory segments) and SeaBIOS could call
>>>>>           
>>> them.
>>>       
>>>> I think SeaBIOS expects Coreboot to copy the option ROM to 0xc0000 in
>>>>         
>>> RAM.
>>>       
>>>> That way SeaBIOS gets to stay hardware agnostic.
>>>>
>>>>         
>>> Makes sense!
>>> But it doesn't look like this does actually happen (by default).
>>> I am playing with msi/ms6147 target (under heavily tweaked qemu).
>>>       
>> You're right.  It doesn't happen by default.  Kevin sent a simple patch to
>> the list, which I can't find right now :(
>>
>> Basically you tell coreboot to run the ROM, but comment out the execution
>> right after the copy.  Let me know if you can't find it, and I'll try to
>> help you dig.
>>     
>
> I found a post where Kevin talks about this but no patch.
>
> I am thinking about this approach for 440BX: set PAM registers to "read
> from ROM, write to RAM" mode, copy the whole range onto itself, set PAM
> register to R/W RAM mode.
> I am not sure if this better be done for all legacy segments (0xC0000 -
> 0xFFFFF range) or only for option ROM segments (0xC0000 - 0xDFFFF range)
> excluding BIOS Area and BIOS Extension segments.
>
> P.S.
> This approach won't work in stock qemu because of incorrect
> implementation (absence, actually) of emulation of this "mixed" PAM mode.
>
>   
I think PAM should be set for the entire space. That would be consistent 
with K8 and Geode. We don't scan the legacy space for ISA ROMs (and if 
we did and they were there we should copy them to memory and set the space).

Marc


-- 
Marc Jones
Senior Firmware Engineer
(970) 226-9684 Office
mailto:Marc.Jones at amd.com
http://www.amd.com/embeddedprocessors