ARM64: Round-Robin dispatch IRQs between CPUs.

raspberrypi inclusion
category: feature
bugzilla: 50432

--------------------------------

IRQ-CPU mapping is round robined on ARM64 to increase
concurrency and allow multiple interrupts to be serviced
at a time.  This reduces the need for FIQ.
Signed-off-by: NMichael Zoran <mzoran@crowfest.net>
Signed-off-by: NFang Yafen <yafen@iscas.ac.cn>
Signed-off-by: NZheng Zengkai <zhengzengkai@huawei.com>

drivers/irqchip/irq-bcm2835.c

@@ -154,10 +154,23 @@ static void armctrl_unmask_irq(struct irq_data *d)
 	}
 }
 
+#ifdef CONFIG_ARM64
+void bcm2836_arm_irqchip_spin_gpu_irq(void);
+
+static void armctrl_ack_irq(struct irq_data *d)
+{
+	bcm2836_arm_irqchip_spin_gpu_irq();
+}
+
+#endif
+
 static struct irq_chip armctrl_chip = {
 	.name = "ARMCTRL-level",
 	.irq_mask = armctrl_mask_irq,
-	.irq_unmask = armctrl_unmask_irq
+	.irq_unmask = armctrl_unmask_irq,
+#ifdef CONFIG_ARM64
+	.irq_ack    = armctrl_ack_irq
+#endif
 };
 
 static int armctrl_xlate(struct irq_domain *d, struct device_node *ctrlr,

drivers/irqchip/irq-bcm2836.c

@@ -87,6 +87,27 @@ static void bcm2836_arm_irqchip_unmask_gpu_irq(struct irq_data *d)
 {
 }
 
+#ifdef CONFIG_ARM64
+
+void bcm2836_arm_irqchip_spin_gpu_irq(void)
+{
+	u32 i;
+	void __iomem *gpurouting = (intc.base + LOCAL_GPU_ROUTING);
+	u32 routing_val = readl(gpurouting);
+
+	for (i = 1; i <= 3; i++) {
+		u32 new_routing_val = (routing_val + i) & 3;
+
+		if (cpu_active(new_routing_val)) {
+			writel(new_routing_val, gpurouting);
+			return;
+		}
+	}
+}
+EXPORT_SYMBOL(bcm2836_arm_irqchip_spin_gpu_irq);
+
+#endif
+
 static struct irq_chip bcm2836_arm_irqchip_gpu = {
 	.name		= "bcm2836-gpu",
 	.irq_mask	= bcm2836_arm_irqchip_mask_gpu_irq,