On Thu, Sep 30, 2021 at 4:11 PM Marc Zyngier <maz@xxxxxxxxxx> wrote: > On Thu, 30 Sep 2021 14:59:24 +0100, > Daniel Palmer <daniel@xxxxxxxx> wrote: > > gpioirqchip->handler = handle_bad_irq; > > gpioirqchip->default_type = IRQ_TYPE_NONE; > > Right. I have no idea why this is a requirement, and I would normally > set it to whatever is the normal flow handler on this HW, but this > looks like the GPIO subsystem has some expectations here. > > I'll let Linus comment on it. The handle_bad_irq() as default handler is because many GPIO IRQ controllers in difference from on-chip IRQ controllers support two levels and two edges of triggers, and there is nothing "normal" (it is "general purpose" after all) so we need to defer the selection of a suitable flow handler to .set_type(). With device tree the trigger is often specified in the second cell in the DTS, so .set_type() will be called for any consumer as part of the request_irq() and the appropriate handler is set from .set_type(). In my mind this is following the DT (ACPI) usage model of allocating and initializing resources dynamically as they are requested. I don't know if this reasoning is wrong in some way, what should we do otherwise? (Confused...) Yours, Linus Walleij
