On Tue, 2016-10-25 at 20:10 +0200, Linus Walleij wrote: > On Tue, Oct 25, 2016 at 4:47 PM, Marc Zyngier <marc.zyngier@xxxxxxx> > wrote: > > > > On 25/10/16 15:22, Jerome Brunet wrote: > > > > > > > > > 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 -- To unsubscribe from this list: send the line "unsubscribe devicetree" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
