Geert,
I forgot to mention one alternative: add a handle_polled_irq() to kernel/chip.c that does explicitly fudge the handler return value. Or, in the alternative, set IRQS_POLL_INPROGRESS on the IRQ state for all interrupts multiplexed from the timer D IRQ chip, which will make note_interrupt shut up about spurious interrupts, but may have other unintended side effects. I'll try both, but rate the chance to get such a hack to pass review as minimal (you would not support that yourself, now would you?).Neither do I - but the IRQ being disabled is the device specific one, not the master chip interrupt dispatching the device specific one. The only way around that is to share this one with a dummy handler (and Paul did veto that).In the mean time, have you found a better solution that the noirqdebug kernel option? I don't like that one.
In order to get the driver into Debian kernels via Linus' tree, the noirqdebug option might be the least atrocious of all.
Cheers, Michael