openrisc: Fix IPIs on simple multicore systems

Commit c056718 ("openrisc: sleep instead of spin on secondary
wait") fixed OpenRISC SMP Linux for QEMU. However, stability was never
achieved on FPGA development boards.  This is because the above patch
has a step to unmask IPIs on non-boot cpu's but on hardware without
power management, IPIs remain masked.

This meant that IPI's were never actually working on the simple SMP
systems we run on development boards.  The systems booted but stability
was very suspect.

Add the ability to unmask IPI's on the non-boot cores.  This is done by
making the OMPIC IRQs proper percpu IRQs.  We can then use the
enabled_percpu_irq() to unmask IRQ on the non-boot cpus.

Update the or1k PIC driver to use a flow handler that can switch between
percpu and the configured level or edge flow handlers at runtime.
This mechanism is inspired by that done in the J-Core AIC driver.

Signed-off-by: Stafford Horne <shorne@gmail.com>
Acked-by: Thomas Gleixner <tglx@kernel.org>

Author: stffrdhrn

arch/openrisc/include/asm/smp.h

@@ -20,7 +20,8 @@ extern void smp_init_cpus(void);
 extern void arch_send_call_function_single_ipi(int cpu);
 extern void arch_send_call_function_ipi_mask(const struct cpumask *mask);
 
-extern void set_smp_cross_call(void (*)(const struct cpumask *, unsigned int));
+extern void set_smp_cross_call(void (*)(const struct cpumask *, unsigned int),
+			       unsigned int irq);
 extern void handle_IPI(unsigned int ipi_msg);
 
 #endif /* __ASM_OPENRISC_SMP_H */

arch/openrisc/kernel/smp.c

@@ -13,6 +13,7 @@
 
 #include <linux/smp.h>
 #include <linux/cpu.h>
+#include <linux/interrupt.h>
 #include <linux/sched.h>
 #include <linux/sched/mm.h>
 #include <linux/irq.h>
@@ -25,6 +26,7 @@
 
 asmlinkage __init void secondary_start_kernel(void);
 
+static unsigned int ipi_irq __ro_after_init;
 static void (*smp_cross_call)(const struct cpumask *, unsigned int);
 
 unsigned long secondary_release = -1;
@@ -39,6 +41,14 @@ enum ipi_msg_type {
 
 static DEFINE_SPINLOCK(boot_lock);
 
+static void or1k_ipi_enable(void)
+{
+	if (WARN_ON_ONCE(!ipi_irq))
+		return;
+
+	enable_percpu_irq(ipi_irq, 0);
+}
+
 static void boot_secondary(unsigned int cpu, struct task_struct *idle)
 {
 	/*
@@ -136,6 +146,7 @@ asmlinkage __init void secondary_start_kernel(void)
 	complete(&cpu_running);
 
 	synchronise_count_slave(cpu);
+	or1k_ipi_enable();
 	set_cpu_online(cpu, true);
 
 	local_irq_enable();
@@ -195,9 +206,18 @@ void smp_send_stop(void)
 	smp_call_function(stop_this_cpu, NULL, 0);
 }
 
-void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int))
+void __init set_smp_cross_call(void (*fn)(const struct cpumask *, unsigned int),
+			       unsigned int irq)
 {
+	if (WARN_ON(ipi_irq))
+		return;
+
 	smp_cross_call = fn;
+
+	ipi_irq = irq;
+
+	/* Enabled IPIs for boot CPU immediately */
+	or1k_ipi_enable();
 }
 
 void arch_send_call_function_single_ipi(int cpu)

drivers/irqchip/irq-ompic.c

@@ -84,6 +84,8 @@ DEFINE_PER_CPU(unsigned long, ops);
 
 static void __iomem *ompic_base;
 
+static DEFINE_PER_CPU_READ_MOSTLY(int, ipi_dummy_dev);
+
 static inline u32 ompic_readreg(void __iomem *base, loff_t offset)
 {
 	return ioread32be(base + offset);
@@ -183,12 +185,17 @@ static int __init ompic_of_init(struct device_node *node,
 		goto out_unmap;
 	}
 
-	ret = request_irq(irq, ompic_ipi_handler, IRQF_PERCPU,
-				"ompic_ipi", NULL);
-	if (ret)
+	irq_set_percpu_devid(irq);
+	ret = request_percpu_irq(irq, ompic_ipi_handler, "ompic_ipi",
+				 &ipi_dummy_dev);
+
+	if (ret) {
+		pr_err("ompic: failed to request irq %d, error: %d",
+		       irq, ret);
 		goto out_irq_disp;
+	}
 
-	set_smp_cross_call(ompic_raise_softirq);
+	set_smp_cross_call(ompic_raise_softirq, irq);
 
 	return 0;
 

drivers/irqchip/irq-or1k-pic.c

@@ -118,11 +118,36 @@ static void or1k_pic_handle_irq(struct pt_regs *regs)
 		generic_handle_domain_irq(root_domain, irq);
 }
 
+/*
+ * The OR1K PIC is a cpu-local interrupt controller and does not distinguish or
+ * use distinct irq number ranges for per-cpu event interrupts (IPI). Since
+ * information to determine whether a particular irq number should be treated as
+ * per-cpu is not available at mapping time, we use a wrapper handler function
+ * which chooses the right handler at runtime based on whether IRQF_PERCPU was
+ * used when requesting the irq.  Borrowed from J-Core AIC.
+ */
+static void or1k_irq_flow_handler(struct irq_desc *desc)
+{
+#ifdef CONFIG_SMP
+	struct irq_data *data = irq_desc_get_irq_data(desc);
+	struct or1k_pic_dev *pic = data->domain->host_data;
+
+	if (irqd_is_per_cpu(data))
+		handle_percpu_devid_irq(desc);
+	else
+		pic->handle(desc);
+#endif
+}
+
 static int or1k_map(struct irq_domain *d, unsigned int irq, irq_hw_number_t hw)
 {
 	struct or1k_pic_dev *pic = d->host_data;
 
-	irq_set_chip_and_handler(irq, &pic->chip, pic->handle);
+	if (IS_ENABLED(CONFIG_SMP))
+		irq_set_chip_and_handler(irq, &pic->chip, or1k_irq_flow_handler);
+	else
+		irq_set_chip_and_handler(irq, &pic->chip, pic->handle);
+
 	irq_set_status_flags(irq, pic->flags);
 
 	return 0;