On Tue, Apr 9, 2019 at 5:26 PM Brian Norris <briannorris@xxxxxxxxxxxx> wrote:
>
> So, I think the problem is still potentially present no matter when we
> request the IRQ. The "uninitialized" state of the hardware (or,
> firmware) just exposes the issue extremely clearly.
Well, I think that as long as you don't request the irq, and it's not
shared with anything else, the bogus state of the irq line simply
doesn't matter. So the NOAUTOEN shouldn't matter if the irq is
requested properly late.
Either it's edge-triggered and you'll get one possibly spurious
interrupt for an old issue, or it's level-triggered and setting up the
hw should bring the irq line inactive and you'll be ok.
But I've applied your patch for now simply because it seems to be a
smaller change.
But I think you should look into whether it can be fixed by just
requesting the irq once the hardware is really up (which may indeed be
as late as open time).
Linus
