Re: Bugfixes: PCI devices get assigned redundant IRQs

["Andreas Block" <job@andreasblock.de>]

Am Sun, 28 Jan 2007 09:04:41 +0100 schrieb H. Peter Anvin <hpa@zytor.com>:

> I think you're confusing the Interrupt Line register and the Interrupt  
> Pin register.  The Interrupt Line register is platform-dependent, but on  
> x86 platforms it generally contains the IRQ number (and IRQ 0 is valid,  
> although in practice it is never used since IRQ 0 is the system timer  
> and is never connected to the PCI bus), or 255 meaning "none" -- see the  
> footnote on page 223 of the PCI 3.0 spec.

No, I don't think so. I meant the PCI Interrupt Pin register and not the  
Interrupt line register. I do know, that the latter contains a platform  
dependent interrupt assignment. In the former a device states which  
interrupt "trace" the device is connected to (Int A-D).
Perhaps you take at look at the code, I think, it's dealing with the  
register I described (Interrupt pin).

Linus stated his opinion about the patch and thinks, it should be well  
tested in -mm kernel, because it might break devices, which do not comply  
to the PCI spec.
I second that. If there're indeed devices out there, which violate the  
spec in this point, the patch would hurt more, than it's doing any good.
As I wrote in my first message, the consequences of this rather small bug  
(if it is one and not a workaround for bad devices) are harmless.

Regards,
Andreas