Re: [PATCH] gpio-dwapb: reset mask register on probe

[Alexey Brodkin <Alexey.Brodkin@synopsys.com>]

Hi Linus,

On Tue, 2015-03-17 at 17:54 +0100, Linus Walleij wrote:
> >> +    /* Mask out and disable all interrupts */
> >> +    dwapb_write(gpio, GPIO_INTMASK, 0xffffffff);
> >> +    dwapb_write(gpio, GPIO_INTEN, 0);
> >
> > This looks good to me - it's always a good idea to make sure defaults
> > are set as we expect.
> 
> So should I cook a patch doing just these two lines?
> 
> But the initial patch unmasking all IRQs then? Is that even needed?

Oops, my bad here :)

I intentionally wrote 0 to MASK register to return it to default state
(if pre-bootloader messes it). As I explained if we really mask all
interrupts (together with disabling them all via INTEN) then interrupts
will never happen.

If we look at DW APB GPIO databook it says:
--->8---
Whenever a 1 is written to a bit of this register [GPIO_INTEN], it
configures the corresponding bit on Port A to become an interrupt;
otherwise, Port A operates as a normal GPIO signal.
--->8---

While for GPIO_INTMASK it says:
--->8---
Controls whether an interrupt on Port A can create an interrupt for the
interrupt controller by not masking it. By default, all interrupts bits
are unmasked.
--->8---

So IMHO we need to update "gpio-dwapb" driver in the following manner:

 [1] In dwapb_configure_irqs() in accordance to "snps,nr-gpios" in the
first bank set a value in GPIO_INTEN. This way we turn N pins in that
port/bank in "interrupt" mode from their default gpio mode.

 [2] Don't expose dwapb_irq_enable()/dwapb_irq_disable() through
ct->chip.irq_enable/ct->chip.irq_disable. Because we don't want to
toggle modes of pins, right?

 [3] Use ct->chip.irq_mask/ct->chip.irq_unmask for purpose of real
enabling/disabling interrupts.

If I'm not missing something that would be implementation that matches
real device specification.

If we do these changes then indeed we'll want to make sure GPIO_INTMASK
is initialized with all interrupts masked out:
--->8---
dwapb_write(gpio, GPIO_INTMASK, 0xffffffff);
--->8---

Let me know if my proposal makes sense and then I'll send patch that
implements it.

-Alexey