On Fri, May 29, 2020 at 10:02 AM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote: > > Hi Rob, > > On Fri, May 29, 2020 at 5:54 PM Rob Herring <robh+dt@xxxxxxxxxx> wrote: > > On Thu, May 28, 2020 at 7:23 AM Geert Uytterhoeven > > <geert+renesas@xxxxxxxxx> wrote: > > > When an interrupt controller has an "interrupt-map" property, an "is > > > valid under each of" error is triggered. > > > > > > Fix this by allowing "interrupt-controller" and "interrupt-map" to > > > coexist, in both the interrrupts meta-schema and the > > > interrupt-controller schema. > > > > But both should not be present. If 'interrupt-controller' is present, > > Why not? Well, maybe I'm wrong. If you have more than just transparent remapping (i.e. mask/unmask/clear), then perhaps both are appropriate because you want get the irq domain for the first irq parent. > > the Linux irq parsing code will ignore 'interrupt-map'. Seems like > > that's backwards, but this parsing code is older than dirt and we'd > > probably break some 1990s machine changing it. > > That's fine. rza1_irqc_parse_map() parses the interrupt-map itself, > to map from downstream to upstream interrupts. You shouldn't really be parsing interrupt-map yourself. The code there doesn't account for #address-cells which can be a factor for interrupt-map. dtc is gaining some checks for 'interrupt-map', so let's hope you have it right. Rob
