On Tue, 2016-10-25 at 20:10 +0200, Linus Walleij wrote:

On Tue, Oct 25, 2016 at 4:47 PM, Marc Zyngier [off-list ref]
wrote:

quoted

On 25/10/16 15:22, Jerome Brunet wrote:

quoted

quoted

There is a few problems to guarantee that gpio == hwirq.
1. We have 2 instances of pinctrl, to guarantee that the linux
gpio
number == hwirq, we would have to guarantee the order in which
they are
probed. At least this my understanding

Maybe I wasn't clear enough, and my use of gpio is probably wrong.
So
Linux has a gpio number, which is obviously an abstract number
(just
like the Linux irq number). But the pad number, in the context of
given
SoC, is constant. So we have:

????????pad->gpio
????????hwirq->irq

Why can't you have pad == hwirq, always? This is already what you
have
in the irqchip driver. This would simplify a lot of things.

My thought as well.

We usually refer to the local numberspace on the GPIO controller
as "offsets", so line offsets 0...31 on a gpiochip with 31 lines.

The ngpio in struct gpio_chip is the number of lines on that
controller,
and should nominally map 1:1 to hwirq sources.

Indeed it should be the the case, and for meson, it is pretty close.
The irqchip controller provide a number of hwirq. Each hwirq maps to
one, and only one, pin. But since not every pins are connected to the
irqchip controller, the opposite is not true.

Taking an example with 16 gpios, here is what it could look like with
the exception we have on meson :

gpio offset [ 0 ?1 ?2 ?3 ?4 ?5 ?6 ?7 ?8 ?9 ?10 11 12 13 14 15 ]
hwirq num ? [ 0 ?1 ?2 ?3] NC NC[4 ?5 ?6 ?7 ?8 ?9 ?10]NC NC NC

Like gpio offset are used (internally) in the driver to find
appropriate gpio registers and bit, the hwirq has a meaning too.
It is the setting you put in the channel multiplexer of the controller
to select the proper pin to spy on.

In the end, these gpio offset and hwirq number are different. I would
prefer to have hwirq == gpio and go your way, it would make my life
easier, but I don't see how it would work.

The irqchip controller cares only about the hwirq number. You can
actually request an interrupt directly to the controller by asking the
proper hwirq number (in DT for example), without involving the gpio
driver (tested).

The relation between the pins and the interrupt number is provided by
the manufacturer in the Datasheet [1], in the section GPIO Interrupt.

Looking at other gpio drivers, it is not uncommon to have some simple
calculation to get from gpio offset to the hwirq number. I don't get
what is the specific problem here ?

If I missed something, feel free to point it out.

[1]?http://dn.odroid.com/S805/Datasheet/S805_Datasheet%20V0.8%202015012
6.pdf

Yours,
Linus Walleij