Re: PCI ROM resource allocation issue with 2.6.17-rc2

[Arjan van de Ven <arjan@infradead.org>]

On Mon, 2006-04-24 at 10:07 -0700, Linus Torvalds wrote:
> 
> On Mon, 24 Apr 2006, Dave Airlie wrote:
> > 
> > however not doing pci_enable_device causes interrupts to not work on the cards
> > in a lot of circumstances..
> 
> Well, you could use "pci_enable_device_bars(0)" instead.
> 
> That will set up interrupt routing _without_ enabling any BAR's, however, 
> that's probably crazy too, since that means that if an interrupt happens, 
> you're really really screwed and can't do anything about it. So that's 
> probably even more broken than what you do now.
> 
> How about delaying the "pci_enable_device()" until when you actually need 
> it, ie do it in drm_irq_install() instead?
> 
> Of course, I wonder how you could POST the device without having the BAR's 
> enabled, 

you haven't spent enough time reading the X pci code then ;)
(or rather, you've done the same thing but hey who's counting)

X does all that *itself* based on what X thinks is best.

Yes that's silly and X should be taken out and shot for that.
What's worse, this is the kind of thing that is really hard to work
around in a away that isn't going to make having a fixed X work as
well... you can't not enable it for old X and enable it for not-insane X
at the same time ;)