Re: [PATCH v2 0/8] qcom: support wakeup capable GPIOs

[Linus Walleij]

On Thu, Jan 24, 2019 at 9:22 PM Lina Iyer <ilina@xxxxxxxxxxxxxx> wrote:

> This is a bug fix submission of the v1 posted here [1]. The discussion on how
> to represent the wakeup-parent interrupt controller is on-going [2] here. The
> reiew comments in [1], from Doug and Stephen are addressed in this patch.
>
> The series attempts to add GPIO chip in hierarchy with PDC interrupt controller
> that can detect and wakeup the GPIOs routed to it, when the system is
> suspend/deep sleep mode.

I kind of start to get the hang of this now. This is starting to
look finished. Some words on the hierarchical GPIO IRQs:

I have started to look into hierarchical GPIO+irqchip since
the usage of such is spreading.

I have been on to Thierry patches trying to make him implement
more generic helpers in the gpiolib irqchip library functions.

In short I see the following:

- Hierarchical gpiochips all have .alloc() and .translate() functions
  that look more or less the same.

- They all fall down to remapping either tuples or entire ranges
  of IRQs from parent to child with a linear look-up table on
  allocation.

So my idea would be to add support for this into the gpiolib
hierarchical irqchip helper: by supplying a parent irqdomain
and a list of translation tuples, we should be able to handle
pretty much any hierarchical GPIO irqdomain (famous last
words).

Right now I am converting the IXP4xx platform to hierarchical
IRQ from the ground up (it is not even using device tree
so it is a bit of a challenge) but it seems to be working.

So I will try to post this in some hopefully working form, and
on top of those changes or before them introduce some
helpers in the core for hierarchical irqs. Or I fail.

Anyways, I do not think my ambitions for refactoring the
helpers can stand in the way of support for these use cases
and new hardware, so maybe we need to merge a few
more hierarchical drivers just bypassing the gpiolib
helpers. I don't want to block development, and I suspect
Thierry might be getting annoyed at me at this point, so
we should maybe just pile up a few more hierarchical
chips so I can refactor them later.

What do you think?

Yours,
Linus Walleij