[LinuxBIOS] MB1030 / 3036 VGA: disabling 00:12:4

[Goodrich,Steven]

I've looked at the original BIOS source and the internal versions of the
documentation, and here's what I've determined:  As near as I can tell,
you cannot disable the PCI header for the device at 00:12:4.  The
original BIOS (written by others on the BIOS team here at AMD in
Colorado) virtualized the PCI headers via SMM code, and that header was
masked when internal graphics were disabled.  Without SMM code, you
cannot do the same thing.

Interesting info/notes:
If you disable memory space via the PCI header (PCI header register 04,
bit 1 -- clear to zero), you should disable it somewhat.

If the F4 Video Configuration trap bit (F0, Index 42h, bit 1) is set,
the system will generate SMIs when trying to access the PCI header
registers.  This should be clear for a non-SMM system.

I suppose you could special-case the device in your PCI scan/enumeration
code, but that is probably more kludgy than you'd like.  Could you
reduce the device's resource requirements to zero?

-- Steve G.

Steve Goodrich ([e-mail] 
)
Advanced Micro Devices, Inc.
1351 S. Sunset Street
Mail Stop 83-36
Longmont, CO  80501
303-774-5129    (fax: [phone])
http://www.amd.com/embeddedprocessors

[Richard Smith]

On 4/28/06, Goodrich,Steven  wrote:
> I suppose you could special-case the device in your PCI scan/enumeration
> code, but that is probably more kludgy than you'd like.  Could you reduce
> the device's resource requirements to zero?

Well we don't really know if having the device enabled is a problem. 
We were worried that perhaps there was some sort of conflict with the
2 video devices and thats why the factory bios had it disabled.

So If its left alone and not allocated any memory or IO resources it
should effectively not be there.  It it really behaves as proper PCI
device then this should be true.

I think the re-cap on this is if the Config.lb is correct then 0:9.0
will get the VGA resources.  And we just need to disable 12.4s SMI on
config access.  Although it seems to enumerate fine so that may not be
an issue.

That just leaves this pesky 90 second video init issue.

[Christian Sühs]

> That just leaves this pesky 90 second video init issue.

Yeeeeaaaaah,
tomorrow I will have a few more tests.

If you have any ideas, let me know.
I do all what you want ;)

First I will try to read out the factory settings.

Next, a question.
Is there something different after a linux boot.
Means, does linux anything what LB not do?
Sets Linux any special registers, I/O ranges or whatever?

Needs/use the emu special resources?

What about the RAM init. It seems, the settings for the memory 
controller are not well done.

and more and more ...

chris
>

[Christian Sühs]

> What about the RAM init. It seems, the settings for the memory 
> controller are not well done.

The settings for the memory config registers are ok.

chris

[Christian Sühs]

Many thanks for that,

Goodrich,Steven schrieb:
> I've looked at the original BIOS source and the internal versions of the 
> documentation, and here's what I've determined:  As near as I can tell, 
> you cannot disable the PCI header for the device at 00:12:4.  The 
> original BIOS (written by others on the BIOS team here at AMD in 
> Colorado) virtualized the PCI headers via SMM code, and that header was 
> masked when internal graphics were disabled.  Without SMM code, you 
> cannot do the same thing.

Okay, First we have thought, that the enabled but not used device causes 
the trouble we have had at the beginning. At the moment it seems there 
is somewhere an other reason for our problems.
However, I will find it ;)
> Interesting info/notes:
> If you disable memory space via the PCI header (PCI header register 04, 
> bit 1 -- clear to zero), you should disable it somewhat.
>
> If the F4 Video Configuration trap bit (F0, Index 42h, bit 1) is set, 
> the system will generate SMIs when trying to access the PCI header 
> registers.  This should be clear for a non-SMM system.

Yeah, I have read that and it could be useful.
> I suppose you could special-case the device in your PCI scan/enumeration 
> code, but that is probably more kludgy than you'd like.  Could you 
> reduce the device's resource requirements to zero?

I think we could. It is shown, but not allocated at the moment, 
therefore it uses no resources.

regards chris
> -- Steve G.
>
> *****Steve Goodrich* ([e-mail]_ 
> )
> Advanced Micro Devices, Inc.
> 1351 S. Sunset Street
> Mail Stop 83-36
> Longmont, CO  80501
> 303-774-5129    (fax: [phone])
> http://www.amd.com/embeddedprocessors