Hello,

> > > Using LKMs?  It looks like an interrupt handler jumped into the mapped
> > > kernel space.
> >Yes, I'm working on a driver module.
> >Ok, what did I do wrong? The Interrupt handler doesn't do anything besides 
> >printf(some_register_content) and the card it's supposed to handle has 
> >interrupts disabled ( although the IRQ is shared because the card sits 
> >behind a PCI-PCI bridge )
> 
> The shared IRQ will cause your handler to be invoked. 
Ah, but apparently not always - I have other devices behind this bridge that should produce quite a lot of interrupts ( like an USB controller ) and stray interrupts in the lkm are very rare.

> Can you see if the attached patch helps?
Applied it, didn't see the panic yet but that doesn't mean a lot since the panic occured after longer idle time which isn't really going to happen in the next few hours ( bug squashing time... )

> The current LKM code doesn't know how to synchronize the instruction cache.
Ouch, shouldn't that bite lots of other lkms too?

have fun
Michael