On 2012-08-14 at 19:52:08 +0200, Catalin Marinas [off-list ref] wrote:

+void cpu_idle(void)
+{
+	local_fiq_enable();
+
+	/* endless idle loop with no priority at all */
+	while (1) {
+		tick_nohz_idle_enter();
+		rcu_idle_enter();
+		while (!need_resched()) {
+			/*
+			 * We need to disable interrupts here to ensure
+			 * we don't miss a wakeup call.
+			 */
+			local_irq_disable();
+			if (!need_resched()) {
+				stop_critical_timings();
+				pm_idle();
+				start_critical_timings();
+				/*
+				 * pm_idle functions should always return
+				 * with IRQs enabled.
+				 */
+				WARN_ON(irqs_disabled());
+			} else {
+				local_irq_enable();
+			}
+		}
+		rcu_idle_exit();
+		tick_nohz_idle_exit();
+		preempt_enable_no_resched();
+		schedule();
+		preempt_disable();

You could use schedule_preempt_disabled() instead of the above 3 calls.
See http://lkml.kernel.org/n/tip-24swm5zut3h9c4a6s46x8rws@git.kernel.org

Cheers
Tobias