anolis: genirq/cpuhotplug: Handle managed IRQs when the last CPU hotplug out in the affinity

ANBZ: #10929

Once we limit the number of managed interrupts, if the last online CPU in
the affinity goes offline, it will result in the interrupt becoming unavailable
util one of the assigned CPUs comes online again. So prevent the last online
CPU in the affinity from going offline, and return -EBUSY in this situation.

Signed-off-by: Guanjun <guanjun@linux.alibaba.com>
Reviewed-by: Zelin Deng <zelin.deng@linux.alibaba.com>
Reviewed-by: Cruz Zhao <CruzZhao@linux.alibaba.com>
Link: https://gitee.com/anolis/cloud-kernel/pulls/3856

Documentation/admin-guide/kernel-parameters.txt

@@ -2723,6 +2723,9 @@
 			interrupts cannot be properly allocated where a large
 			number of devices are present. The default number is 0,
 			that means no limit to the number of managed irqs.
+			Once we limit the number of managed interrupts, the last
+			online CPU in the affinity goes offline will fail with
+			the error code -EBUSY.
 			Format: integer between 0 and num_possible_cpus() / num_possible_nodes()
 			Default: 0
 

include/linux/irq.h

@@ -628,8 +628,10 @@ extern int irq_set_vcpu_affinity(unsigned int irq, void *vcpu_info);
 #if defined(CONFIG_SMP) && defined(CONFIG_GENERIC_IRQ_MIGRATION)
 extern void irq_migrate_all_off_this_cpu(void);
 extern int irq_affinity_online_cpu(unsigned int cpu);
+extern int irq_affinity_offline_cpu(unsigned int cpu);
 #else
 # define irq_affinity_online_cpu	NULL
+# define irq_affinity_offline_cpu	NULL
 #endif
 
 #if defined(CONFIG_SMP) && defined(CONFIG_GENERIC_PENDING_IRQ)

kernel/cpu.c

@@ -1699,7 +1699,7 @@ static struct cpuhp_step cpuhp_hp_states[] = {
 	[CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
 		.name			= "irq/affinity:online",
 		.startup.single		= irq_affinity_online_cpu,
-		.teardown.single	= NULL,
+		.teardown.single	= irq_affinity_offline_cpu,
 	},
 	[CPUHP_AP_PERF_ONLINE] = {
 		.name			= "perf:online",

kernel/irq/cpuhotplug.c

@@ -211,6 +211,31 @@ static void irq_restore_affinity_of_irq(struct irq_desc *desc, unsigned int cpu)
 		irq_set_affinity_locked(data, affinity, false);
 }
 
+static int irq_check_affinity_of_irq(struct irq_desc *desc, unsigned int cpu)
+{
+	struct irq_data *data = irq_desc_get_irq_data(desc);
+	const struct cpumask *affinity = irq_data_get_affinity_mask(data);
+	unsigned int cur;
+
+	if (!irqd_affinity_is_managed(data) || !desc->action ||
+	    !irq_data_get_irq_chip(data) || !cpumask_test_cpu(cpu, affinity))
+		return 0;
+
+	for_each_cpu(cur, affinity)
+		if (cur != cpu && cpumask_test_cpu(cur, cpu_online_mask))
+			return 0;
+
+	/*
+	 * If the onging offline CPU is the last one in the affinity,
+	 * the managed interrupts will be unavailable until one of
+	 * the assigned CPUs comes online. To prevent this unavailability,
+	 * return -EBUSY directly in this case.
+	 */
+	pr_warn("Affinity %*pbl of managed IRQ%u contains only one CPU%u that online\n",
+		cpumask_pr_args(affinity), data->irq, cpu);
+	return -EBUSY;
+}
+
 /**
  * irq_affinity_online_cpu - Restore affinity for managed interrupts
  * @cpu:	Upcoming CPU for which interrupts should be restored
@@ -231,3 +256,32 @@ int irq_affinity_online_cpu(unsigned int cpu)
 
 	return 0;
 }
+
+/**
+ * irq_affinity_offline_cpu - Check affinity for managed interrupts
+ * to prevent the unavailability caused by taking the last CPU in the
+ * affinity offline.
+ * @cpu:	Upcoming CPU for which interrupts should be checked
+ */
+int irq_affinity_offline_cpu(unsigned int cpu)
+{
+	struct irq_desc *desc;
+	unsigned int irq;
+	int ret = 0;
+
+	if (!managed_irqs_per_node)
+		return 0;
+
+	irq_lock_sparse();
+	for_each_active_irq(irq) {
+		desc = irq_to_desc(irq);
+		raw_spin_lock_irq(&desc->lock);
+		ret = irq_check_affinity_of_irq(desc, cpu);
+		raw_spin_unlock_irq(&desc->lock);
+		if (ret < 0)
+			break;
+	}
+	irq_unlock_sparse();
+
+	return ret;
+}