+ linux-pci
- linux-wireless
On Fri, Jul 07, 2017 at 11:04:32AM +0800, Jeffy Chen wrote:
> On 07/07/2017 08:53 AM, Brian Norris wrote:
> >It kind of seems like we inadvertently conflicted with the PCI OF
> >interrupt spec [1]. There, the "interrupts" property for a device (if
> >present) is supposed to represent INT{A...D} with values of {1...4}.
> >IIUC, there should only be a single entry in this property.
> >
> >If we were to extend this properly, I guess that would mean we'd need a
> >second "interrupts" entry, with a different parent. I think we can use
> >"interrupts-extended" for that.
> >
> >So we'd need to document an optional "interrupt-names" for Marvell, and
> >have the driver try that first. The rough outline would be something
> >like this.
> >
> >For the device tree (e.g., rk3399-gru):
> >
> >- interrupt-parent = <&gpio0>;
> >- interrupts = <8 IRQ_TYPE_LEVEL_LOW>;
> >+ interrupts-extended = <&pcie0 1>, <&gpio0 8 IRQ_TYPE_LEVEL_LOW>;
> >+ interrupt-names = "int-A", "wake";
> This is a great idea.
>
> And how about also add a property to tell of_pci_irq to ignore of
> irq and force using PCI_INTERRUPT_PIN? Since there might be devices
> don't use pci irq, but using other irq(wowlan for example).
Even if they don't use the PCI IRQ, is there a way to provide an empty /
un-mapped entry?
I'm a little wary on trying to "extend" (which can sometimes be read
"break") the device tree spec here too much more. But perhaps DT or PCI
folks have a better recommendation (if they can hold their noses long
enough).
> Then we can specify this property and add a name("wake") to the wifi
> wake irq here. And interrupts-extended would still be an available
> option.
Brian
