> Shouldn't there be that "lost interrupt" notify at console if that's the
> case?

Only if the driver is designed so to detect the condition.  tsleep/wakeup
framework can not handle "lost notification" symdrome in principle.  It's
the reason why "timeout" feature was added for the original sleep()
primitive.

As you already wonder I suspect GPIO/IRQ "line fiddling" ends up with
failing to catch or forgetting interrupt notification with your case.

Toru Nishimura/ALKYL Technology