Resource Allocation for Option Rom

[Richard Smith]

Li-Ta Lo wrote:

>>Did this happen?  I'm trying to get a plug in PCI video card up with 
>>ADLO and I need the option rom to be enabled.  I can do it direcly in my 
>>mainboard code for a hack but If there was anything to test I though I 
>>would try.
>>
> How is testbios doing on your side ?

It's proving very useful.

I have tested it on another pc with my M1 eval PCI card and on that 
machine I can boot with the Vbios removed, run testbios and I get a 
signon message.  Yay.

So my other test pc must have some problem with not setting up access to 
  a0000-bffff properly when you boot a card with out a bios.  Just like 
you suspected.

So now that I have a golden path with the emulator I have been doing log 
comparisons between the eval card in the pc vs in our target board.

Comparing IO accesses I don't find any meaningfull differences. So I 
upped the logging to include memory reads/writes as well.  Looks like 
when the vbios trys to do a memory test on a0000 it fails.

I've also managed to make our COTS bios from the previous rev of this 
board boot as well.  It does option ROM scans and booting that bios with 
the PCI eval card stuck in does exactly the same thing as my linuxbios 
setup + emulator does.  The COTS bios also has a vid mem test and if I 
enable that it will fail as well.

So now I've got failure under both linux bios and a COTS bios using a 
PCI eval card that is known to work.

So its pretty much got to be some sort of hardware problem with access 
to the legacy VGA range.  Something fscked with the northbridge perhaps. 
  I'm hacking up the vbios source to do the a0000 test over and over and 
tomorow we are going to see where those accesses are really going if at all.

As far as testbios is concerned the only mod it would need to work with 
my stock ATI bios is the timer IO access trapping and reporting back a 
consistant value.

Its not really an issue for me since the ATI vbios works fine under 
testbios with the problematic ATI delay hacked to just return rather 
than delay.  The delay provided by the emulator appears to be enough. 
   Thats what I've been using for all my testing so far.