irq/work: Use llist_for_each_entry_safe

The llist_for_each_entry() loop in irq_work_run_list() is unsafe because
once the works PENDING bit is cleared it can be requeued on another CPU.

Use llist_for_each_entry_safe() instead.

Fixes: 16c0890 ("irq/work: Don't reinvent the wheel but use existing llist API")
Reported-by:Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Frederic Weisbecker <frederic@kernel.org>
Cc: Byungchul Park <byungchul.park@lge.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Petri Latvala <petri.latvala@intel.com>
Link: http://lkml.kernel.org/r/151027307351.14762.4611888896020658384@mail.alporthouse.com

kernel/irq_work.c

@@ -128,17 +128,17 @@ bool irq_work_needs_cpu(void)
 
 static void irq_work_run_list(struct llist_head *list)
 {
-	unsigned long flags;
-	struct irq_work *work;
+	struct irq_work *work, *tmp;
 	struct llist_node *llnode;
+	unsigned long flags;
 
 	BUG_ON(!irqs_disabled());
 
 	if (llist_empty(list))
 		return;
 
 	llnode = llist_del_all(list);
-	llist_for_each_entry(work, llnode, llnode) {
+	llist_for_each_entry_safe(work, tmp, llnode, llnode) {
 		/*
 		 * Clear the PENDING bit, after this point the @work
 		 * can be re-used.