On Tue, 2006-11-14 at 20:28 -0800, David Miller wrote:From: Jeff Garzik <jeff@xxxxxxxxxx>
Date: Tue, 14 Nov 2006 23:24:04 -0500
I can't answer for the spec, but at least two independent device vendors recommended to write an MSI driver that way (disable intx, enable msi).Ok.
Completely independent of MSI though, a PCI 2.2 compliant driver should be nice and disable intx on exit, just to avoid any potential interrupt hassles after driver unload. And of course be aware that it might need to enable intx upon entry.This also sounds like it should occur in the generic PCI layer when a
PCI driver is unregistered.
Is this disable_intx() thingy something x86 specific ? I mean, you can't
just call disable_irq() for LSIs since you can be sharing it. If you
aren't sharing, free_irq() will mask for you.