[plug] agp/pci interrupt problems

Trevor Phillips phillips at central.murdoch.edu.au
Thu Mar 15 16:56:17 WST 2001


"Simon Scott (SSC)" wrote:
> 
> Just for interest:
> 
> I found a site that explains a little more than the motherboard manual.
> Basically PCI devices *should* be able to share interrupts, altho obviously
> some cheaper or less well designed cards have a little problem. The spec
> doesnt recommend it, it insists upon it.
> 
> AGP cards are notoriously bad at sharing interrupts, which has nothing to do
> with the AGP spec but rather the manufacturers.
> 
> In particular you should never share an IRQ between a Soundblaster Live and
> anything else, and also alot of NICs dont like it much.
> 
> Basically I have to disable any devices Im not using if possible (eg USB,
> COM ports) to free up interrupts. Each PCI (and AGP) port is connected to
> one of four 'Int Pins' which can have an IRQ assigned to them. Obviously you
> have to shift devices around so the more sensitive ones are on their own
> pin. For example, it appears that on my board it works as follows:
> 
> Int Pin #1 - AGP Slot and PCI Slot 1 (closest to AGP)
> Int Pin #2 - PCI Slot 2, PCI Slot 3 and ACPI (what the hell is ACPI?)
> Int Pin #3 - PCI Slot 5 (and if you have the version of the K7T with onboard
> RAID it uses this)
> Int Pin #4 - PCI Slot 4, PCI Slot 6 and USB.
> 
> So if a disable USB and ACPI (and maybe COM ports) in the bios, and shuffle
> a few cards around, I should get each device to have its own IRQ, which
> negates the need to track down my sensitive card that causes the crash.
> 
> They also recommend disabling 'PCI Delay Transitions' in the BIOS.
> 
> This doesnt really solve my 'AGP card doesnt have an IRQ' problem, but I was
> interested to see that on a motherboard review site the guy had taken a
> screen dump of the IRQs in Windows and no graphics card appeared either.
> 
> Can someone on a AGP machine do a quick cat /proc/interrupts or have a look
> in the control panel and let me know if this is a common thing?

Mine does have an entry:
  16:      11112      11236   IO-APIC-level  mga at PCI:1:0:0  

This is a G400, on an SMP (BP6, Dual Celeron 500) system, with agpgart, DRI,
and all the works enabled. ^_^

-- 
. Trevor Phillips             -           http://jurai.murdoch.edu.au/ . 
: CWIS Systems Administrator     -           T.Phillips at murdoch.edu.au : 
| IT Services                       -               Murdoch University | 
 >------------------- Member of the #SAS# & #CFC# --------------------<
| On nights such as this, evil deeds are done. And good deeds, of     /
| course. But mostly evil, on the whole.                             /
 \      -- (Terry Pratchett, Wyrd Sisters)                          /



More information about the plug mailing list