Kick out of the idle loop if there are softirqs pending, avoiding trying to call nohz idle and service them on exit from schedule.

4 files changed, 39 insertions, 6 deletions

diff --git a/kernel/sched/MuQSS.c b/kernel/sched/MuQSS.c
index 57d1db9c7213..6d1d5a420bc4 100644
--- a/kernel/sched/MuQSS.c
+++ b/kernel/sched/MuQSS.c
@@ -752,6 +752,13 @@ static inline bool task_queued(struct task_struct *p)
 static void enqueue_task(struct rq *rq, struct task_struct *p, int flags);
 static inline void resched_if_idle(struct rq *rq);
 
+/* Dodgy workaround till we figure out where the softirqs are going */
+static inline void do_pending_softirq(struct rq *rq, struct task_struct *next)
+{
+	if (unlikely(next == rq->idle && local_softirq_pending() && !in_interrupt()))
+		do_softirq_own_stack();
+}
+
 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
 {
 #ifdef CONFIG_SMP
@@ -808,9 +815,7 @@ static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
 #endif
 	rq_unlock(rq);
 
-	/* Dodgy workaround till we figure out where the softirqs are going */
-	if (unlikely(current == rq->idle && local_softirq_pending() && !in_interrupt()))
-		do_softirq_own_stack();
+	do_pending_softirq(rq, current);
 
 	local_irq_enable();
 }
@@ -3852,7 +3857,9 @@ static void __sched notrace __schedule(bool preempt)
 		context_switch(rq, prev, next); /* unlocks the rq */
 	} else {
 		check_siblings(rq);
-		rq_unlock_irq(rq);
+		rq_unlock(rq);
+		do_pending_softirq(rq, next);
+		local_irq_enable();
 	}
 }
 
diff --git a/kernel/sched/MuQSS.h b/kernel/sched/MuQSS.h
index f9510d739777..3565a7d8a5ee 100644
--- a/kernel/sched/MuQSS.h
+++ b/kernel/sched/MuQSS.h
@@ -1,5 +1,6 @@
 #include <linux/sched.h>
 #include <linux/cpuidle.h>
+#include <linux/interrupt.h>
 #include <linux/skip_list.h>
 #include <linux/stop_machine.h>
 #include "cpuacct.h"
@@ -325,4 +326,18 @@ static inline void cpufreq_trigger(u64 time, unsigned long util)
 #define arch_scale_freq_invariant()	(false)
 #endif
 
+/*
+ * This should only be called when current == rq->idle. Dodgy workaround for
+ * when softirqs are pending and we are in the idle loop. Setting current to
+ * resched will kick us out of the idle loop and the softirqs will be serviced
+ * on our next pass through schedule().
+ */
+static inline bool softirq_pending(int cpu)
+{
+	if (likely(!local_softirq_pending()))
+		return false;
+	set_tsk_need_resched(current);
+	return true;
+}
+
 #endif /* MUQSS_SCHED_H */
diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c
index 060b76d854ec..51264e6b120e 100644
--- a/kernel/sched/idle.c
+++ b/kernel/sched/idle.c
@@ -208,6 +208,8 @@ static void cpu_idle_loop(void)
 	int cpu = smp_processor_id();
 
 	while (1) {
+		bool pending = false;
+
 		/*
 		 * If the arch has a polling bit, we maintain an invariant:
 		 *
@@ -219,7 +221,10 @@ static void cpu_idle_loop(void)
 
 		__current_set_polling();
 		quiet_vmstat();
-		tick_nohz_idle_enter();
+		if (unlikely(softirq_pending(cpu)))
+			pending = true;
+		else
+			tick_nohz_idle_enter();
 
 		while (!need_resched()) {
 			check_pgt_cache();
@@ -259,7 +264,8 @@ static void cpu_idle_loop(void)
 		 * not have had an IPI to fold the state for us.
 		 */
 		preempt_set_need_resched();
-		tick_nohz_idle_exit();
+		if (!pending)
+			tick_nohz_idle_exit();
 		__current_clr_polling();
 
 		/*
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index c64fc5114004..cdefab6df124 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -1813,3 +1813,8 @@ static inline void cpufreq_trigger_update(u64 time) {}
 #else /* arch_scale_freq_capacity */
 #define arch_scale_freq_invariant()	(false)
 #endif
+
+static inline bool softirq_pending(int cpu)
+{
+	return false;
+}