Subject: Re: Hard reset error on dual PCI buses
To: Nathan J. Williams <>
From: David Hopper <>
List: port-alpha
Date: 03/07/2003 11:27:52
Nathan J. Williams wrote:
> David Hopper <> writes:
>>Besides, don't all these different chipsets need to map video memory,
>>texture memory, and address spaces differently?  Making a unified "ask
>>the console" device would mean creating kernel module drivers for
>>individual chipsets, and is the main reason why fb-dev in Linux is in
>>such a quagmire...
> I don't think Ross was suggesting going that far, but rather just "ask
> the console what PCI device to use" instead of scanning. We've already
> got some hooks in XFree86 to use our PCI interface code, so this would
> be a moderately localized change.

Ah, okay...  I didn't think the console was bright enough to cough up what the 
driver interface needs.  I really am operating under the assumption that the 
present codebase "should just work" with multiple PCI buses, and go from 
there.  Then we can call it a bug, and track 4.3.0 without forking for our 
tree.  (By the way, how does Open/FreeBSD do this stuff?  Just what is all 
that hae_* frobbing stuff that we've thrown out?).

> For that matter, XF86Config already has a way to directly specify a
> device's location on a bus with the BusID flag; can you try using that
> instead of letting the server scan? (It's not immediately clear if the
> existing syntax for BusID is really general enough for a system with
> more than one PCI hierarchy, like your system)

BusID is general enough right now (BusID PCI:01:xx:xx and BusID PCI:00:xx:xx), 
but it just doesn't function correctly in the new scanpci code (and it needs 
to, to tell me what the BusID parameters should be).  I _have_ seen XFree4 
work on this AS1200 once before; it was with Simon Burge's initial patches to 
the XF code on alpha (January 2001).  That was a different scanpci codebase-- 
it's all changed quite a bit since then.

>         - Nathan

Thanks v.v. much,