ARM: IMX: fix ability to use pcie interrupt-map

When using interrupt-maps, the size of a map entry is #address-cells +
#interrupt-cells for the parent interrupt controller. For the ARM GIC
address-cells should be 0 as this is not used.

This patch fixes the example by correctly specifying #address-cells = 0.

Quoted 5 lines · Tim Harvey

When using interrupt-maps, the size of a map entry is #address-cells +
#interrupt-cells for the parent interrupt controller. For the ARM GIC
address-cells should be 0 as this is not used.

This patch fixes the example by correctly specifying #address-cells = 0.

If the #address-cells property is required (well, or even optional...) in the node, shouldn't it be included in the list of required/optional properties above, and not solely included in the example?

Quoted 10 lines · Stephen Warren

Quoted 5 lines · Tim Harvey

When using interrupt-maps, the size of a map entry is #address-cells +
#interrupt-cells for the parent interrupt controller. For the ARM GIC
address-cells should be 0 as this is not used.

This patch fixes the example by correctly specifying #address-cells = 0.

If the #address-cells property is required (well, or even optional...) in the node, shouldn't it be included in the list of required/optional properties above, and not solely included in the example?

AFAIK, #address-cells = 0 is the same as not including a #address-cells at all.

Omitting entirely is what other interrupt-controller bindings are doing, so I'd just drop mention of #address-cells completely.

Jason

Quoted 12 lines · Jason Gunthorpe

Quoted 10 lines · Stephen Warren

Quoted 5 lines · Tim Harvey

When using interrupt-maps, the size of a map entry is #address-cells +
#interrupt-cells for the parent interrupt controller. For the ARM GIC
address-cells should be 0 as this is not used.

This patch fixes the example by correctly specifying #address-cells = 0.

If the #address-cells property is required (well, or even optional...) in the node, shouldn't it be included in the list of required/optional properties above, and not solely included in the example?

Quoted 7 lines

AFAIK, #address-cells = 0 is the same as not including a #address-cells at all.

Omitting entirely is what other interrupt-controller bindings are doing, so I'd just drop mention of #address-cells completely.

Jason

Jason / Stephen,

Yes, it does indeed default to 0 if not specified. For arm-gic it would appear #address-cells must be 0, so should we remove all places its set explicitly to 0 in dts as wel as the example in Documentation/devicetree/bindings?

Looking through arch/arm/boot/dts I'm seeing other dts configs using 'arm,cortex-a9-gic' which I'm guessing are equally as wrong as the imx6 situation:
- arch/arm/boot/dts/highbank.dts
- arch/arm/boot/dts/sh73a0.dtsi

I'm also seeing a number of places where its set explicitly to 0:
- Documentation/devicetree/bindings/arm/gic.txt
- Documentation/devicetree/bindings/arm/vexpress.txt
- arch/arm/boot/dts/bcm11351.dtsi
- arch/arm/boot/dts/hi3620.dtsi
- arch/arm/boot/dts/imx6qdl.dtsi
- arch/arm/boot/dts/vexpress-v2p-ca15_a7.dts

If I extend this search to the other compatible gic interrupt controllers implemented in drivers/irqchip/irq-gic.c (arm,cortex-a15-gic, qcom,msm-8660-qgic, qcom,msm-qgic2) I find even more examples where #address-cells is set explicitly to 0 or to 1 (and I would believe a value of 1 is invalid, just as it is in my case).

I'm happy to broaden my patch, but I certainly want to make sure I'm doing the right thing with it, as I don't have the ability to test all the other devicetree configs and am by no means a devicetree expert.

Recommendations?

Quoted 10 lines · Stephen Warren

Quoted 5 lines · Tim Harvey

When using interrupt-maps, the size of a map entry is #address-cells +
#interrupt-cells for the parent interrupt controller. For the ARM GIC
address-cells should be 0 as this is not used.

This patch fixes the example by correctly specifying #address-cells = 0.

If the #address-cells property is required (well, or even optional...) in the node, shouldn't it be included in the list of required/optional properties above, and not solely included in the example?

AFAIK, #address-cells = 0 is the same as not including a #address-cells at all.

Omitting entirely is what other interrupt-controller bindings are doing, so I'd just drop mention of #address-cells completely.

The invalid value of #address-cells in the imx6 pcie host controller node causes of_irq_parse_raw() to incorrectly advance through an interrupt-map table of more than one interrupt.

This patch resolves this issue and allows proper interrupt mapping for an imx6 pcie host connected to a P2P bridge when using legacy interrupts.

Signed-off-by Tim Harvey <[email protected]>

Cc Jason Gunthorpe <[email protected]>

Cc Jingoo Han <[email protected]>

Cc Lucas Stach <[email protected]>

Cc Mark Rutland <[email protected]>

Cc linux-samsung-soc <[email protected]>

Cc Richard Zhu <[email protected]>

Cc Sascha Hauer <[email protected]>

Cc Arnd Bergmann <[email protected]>

Cc Stephen Warren <[email protected]>

Cc Bjorn Helgaas <[email protected]>

Cc Simon Horman <[email protected]>

Cc Thierry Reding <[email protected]>

Cc Ben Dooks <[email protected]>

Cc linux-tegra <[email protected]>

Cc Kukjin Kim <[email protected]>

Cc Shawn Guo <[email protected]>

Cc Grant Likely <[email protected]>

---
 arch/arm/boot/dts/imx6qdl.dtsi | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

arch/arm/boot/dts/imx6qdl.dtsi +0 -0

index af8d28d..2285299 100644

arch/arm/boot/dts/imx6qdl.dtsi +0 -0

 		interrupt-controller;
 		reg = <0x00a01000 0x1000>,

Signature · 7 lines

1.8.3.2


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel