genirq: Free irq_desc with rcu

The new VMD device driver needs to iterate over a list of
"demultiplexing" interrupts. Protecting that list with a lock is not
possible because the list is also required in code pathes which hold
irq descriptor lock. Therefor the demultiplexing interrupt handler
would create a lock inversion scenario if it calls a demux handler
with the list protection lock held.

A solution for this is to free the irq descriptor via RCU, so the
list can be walked with rcu read lock held.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Keith Busch <keith.busch@intel.com>

2 files changed, 22 insertions, 3 deletions

diff --git a/include/linux/irqdesc.h b/include/linux/irqdesc.h
index a587a33363c7..dcca77c4b9d2 100644
--- a/include/linux/irqdesc.h
+++ b/include/linux/irqdesc.h
@@ -1,6 +1,8 @@
 #ifndef _LINUX_IRQDESC_H
 #define _LINUX_IRQDESC_H
 
+#include <linux/rcupdate.h>
+
 /*
  * Core internal functions to deal with irq descriptors
  */
@@ -40,6 +42,7 @@ struct pt_regs;
  *			IRQF_NO_SUSPEND set
  * @force_resume_depth:	number of irqactions on a irq descriptor with
  *			IRQF_FORCE_RESUME set
+ * @rcu:		rcu head for delayed free
  * @dir:		/proc/irq/ procfs entry
  * @name:		flow handler name for /proc/interrupts output
  */
@@ -82,6 +85,9 @@ struct irq_desc {
 #ifdef CONFIG_PROC_FS
 	struct proc_dir_entry	*dir;
 #endif
+#ifdef CONFIG_SPARSE_IRQ
+	struct rcu_head		rcu;
+#endif
 	int			parent_irq;
 	struct module		*owner;
 	const char		*name;
diff --git a/kernel/irq/irqdesc.c b/kernel/irq/irqdesc.c
index 239e2ae2c947..0409da0bcc33 100644
--- a/kernel/irq/irqdesc.c
+++ b/kernel/irq/irqdesc.c
@@ -159,6 +159,7 @@ static struct irq_desc *alloc_desc(int irq, int node, struct module *owner)
 
 	raw_spin_lock_init(&desc->lock);
 	lockdep_set_class(&desc->lock, &irq_desc_lock_class);
+	init_rcu_head(&desc->rcu);
 
 	desc_set_defaults(irq, desc, node, owner);
 
@@ -171,6 +172,15 @@ err_desc:
 	return NULL;
 }
 
+static void delayed_free_desc(struct rcu_head *rhp)
+{
+	struct irq_desc *desc = container_of(rhp, struct irq_desc, rcu);
+
+	free_masks(desc);
+	free_percpu(desc->kstat_irqs);
+	kfree(desc);
+}
+
 static void free_desc(unsigned int irq)
 {
 	struct irq_desc *desc = irq_to_desc(irq);
@@ -187,9 +197,12 @@ static void free_desc(unsigned int irq)
 	delete_irq_desc(irq);
 	mutex_unlock(&sparse_irq_lock);
 
-	free_masks(desc);
-	free_percpu(desc->kstat_irqs);
-	kfree(desc);
+	/*
+	 * We free the descriptor, masks and stat fields via RCU. That
+	 * allows demultiplex interrupts to do rcu based management of
+	 * the child interrupts.
+	 */
+	call_rcu(&desc->rcu, delayed_free_desc);
 }
 
 static int alloc_descs(unsigned int start, unsigned int cnt, int node,