On 13/10/14 11:43, Joe.C wrote:

On Thu, 2014-10-09 at 17:59 +0100, Marc Zyngier wrote:

quoted

On 09/10/14 15:29, Joe.C wrote

quoted

@@ -952,7 +988,11 @@ void __init gic_init_bases(unsigned int gic_nr, int irq_start,
 
 	gic_irqs -= hwirq_base; /* calculate # of irqs to allocate */
 
-	if (of_property_read_u32(node, "arm,routable-irqs",
+	if (IS_ENABLED(CONFIG_IRQ_DOMAIN_HIERARCHY) &&
+		of_find_property(node, "arm,irq-domain-hierarchy", NULL))
+		gic->domain = irq_domain_add_linear(node, gic_irqs,
+					&gic_irq_domain_hierarchy_ops, gic);

I really think that looking for a property is the wrong thing to do. If
"node" is non-NULL, then we're pretty sure that we're initializing from
DT, and that a pure linear domain should be the right thing, leaving the
legacy stuff for the few non-DT platforms that are still around.

Thanks,

	M.

The only reason I introduce "arm,irq-domain-hierarchy" property is
trying to keep original behavior when hierarchy irq domain is not used.
Without this, when a board init GIC with DT, all driver will have to use
devicetree. I'm not sure we want to break things like this.

I don't think we want to support a "middle of the road" setup, where the
GIC is probed by DT, but some devices have hardcoded interrupts.

I will remove this and just use linear for all DT in my next version.

I came up with the attached patch, which allows me to boot my test
platform (together with the other fix I posted earlier).

Thanks,

	M.
-- 
Jazz is not dead. It just smells funny...