ARM: irq migration: ensure migration is handled safely
Ensure appropriate locks are taken to ensure that IRQ migration off the current CPU is race-free. We may have a concurrent set_affinity via procfs running on another CPU in parallel with the IRQ migration, resulting in unpredictable results. Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
This commit is contained in:
parent
1dbfa187da
commit
617912440b
|
@ -179,14 +179,21 @@ int __init arch_probe_nr_irqs(void)
|
|||
|
||||
#ifdef CONFIG_HOTPLUG_CPU
|
||||
|
||||
static void route_irq(struct irq_desc *desc, unsigned int irq, unsigned int cpu)
|
||||
static bool migrate_one_irq(struct irq_data *d)
|
||||
{
|
||||
pr_debug("IRQ%u: moving from cpu%u to cpu%u\n", irq, desc->irq_data.node, cpu);
|
||||
unsigned int cpu = cpumask_any_and(d->affinity, cpu_online_mask);
|
||||
bool ret = false;
|
||||
|
||||
raw_spin_lock_irq(&desc->lock);
|
||||
desc->irq_data.chip->irq_set_affinity(&desc->irq_data,
|
||||
cpumask_of(cpu), true);
|
||||
raw_spin_unlock_irq(&desc->lock);
|
||||
if (cpu >= nr_cpu_ids) {
|
||||
cpu = cpumask_any(cpu_online_mask);
|
||||
ret = true;
|
||||
}
|
||||
|
||||
pr_debug("IRQ%u: moving from cpu%u to cpu%u\n", d->irq, d->node, cpu);
|
||||
|
||||
d->chip->irq_set_affinity(d, cpumask_of(cpu), true);
|
||||
|
||||
return ret;
|
||||
}
|
||||
|
||||
/*
|
||||
|
@ -198,25 +205,30 @@ void migrate_irqs(void)
|
|||
{
|
||||
unsigned int i, cpu = smp_processor_id();
|
||||
struct irq_desc *desc;
|
||||
unsigned long flags;
|
||||
|
||||
local_irq_save(flags);
|
||||
|
||||
for_each_irq_desc(i, desc) {
|
||||
struct irq_data *d = &desc->irq_data;
|
||||
bool affinity_broken = false;
|
||||
|
||||
if (d->node == cpu) {
|
||||
unsigned int newcpu = cpumask_any_and(d->affinity,
|
||||
cpu_online_mask);
|
||||
if (newcpu >= nr_cpu_ids) {
|
||||
if (printk_ratelimit())
|
||||
printk(KERN_INFO "IRQ%u no longer affine to CPU%u\n",
|
||||
i, cpu);
|
||||
raw_spin_lock(&desc->lock);
|
||||
do {
|
||||
if (desc->action == NULL)
|
||||
break;
|
||||
|
||||
cpumask_setall(d->affinity);
|
||||
newcpu = cpumask_any_and(d->affinity,
|
||||
cpu_online_mask);
|
||||
}
|
||||
if (d->node != cpu)
|
||||
break;
|
||||
|
||||
route_irq(desc, i, newcpu);
|
||||
}
|
||||
affinity_broken = migrate_one_irq(d);
|
||||
} while (0);
|
||||
raw_spin_unlock(&desc->lock);
|
||||
|
||||
if (affinity_broken && printk_ratelimit())
|
||||
pr_warning("IRQ%u no longer affine to CPU%u\n", i, cpu);
|
||||
}
|
||||
|
||||
local_irq_restore(flags);
|
||||
}
|
||||
#endif /* CONFIG_HOTPLUG_CPU */
|
||||
|
|
Loading…
Reference in New Issue