Subject: softirq: Adapt NOHZ softirq pending check to new RT scheme
From: Thomas Gleixner <tglx@linutronix.de>
Date: Sun, 28 Oct 2012 13:46:16 +0000

We can't rely on ksoftirqd anymore and we need to check the tasks
which run a particular softirq and if such a task is pi blocked ignore
the other pending bits of that task as well.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 kernel/softirq.c |   83 ++++++++++++++++++++++++++++++++++++++-----------------
 1 file changed, 58 insertions(+), 25 deletions(-)

--- a/kernel/softirq.c
+++ b/kernel/softirq.c
@@ -66,46 +66,71 @@ char *softirq_to_name[NR_SOFTIRQS] = {
 
 #ifdef CONFIG_NO_HZ_COMMON
 # ifdef CONFIG_PREEMPT_RT_FULL
+
+struct softirq_runner {
+	struct task_struct *runner[NR_SOFTIRQS];
+};
+
+static DEFINE_PER_CPU(struct softirq_runner, softirq_runners);
+
+static inline void softirq_set_runner(unsigned int sirq)
+{
+	struct softirq_runner *sr = &__get_cpu_var(softirq_runners);
+
+	sr->runner[sirq] = current;
+}
+
+static inline void softirq_clr_runner(unsigned int sirq)
+{
+	struct softirq_runner *sr = &__get_cpu_var(softirq_runners);
+
+	sr->runner[sirq] = NULL;
+}
+
 /*
- * On preempt-rt a softirq might be blocked on a lock. There might be
- * no other runnable task on this CPU because the lock owner runs on
- * some other CPU. So we have to go into idle with the pending bit
- * set. Therefor we need to check this otherwise we warn about false
- * positives which confuses users and defeats the whole purpose of
- * this test.
+ * On preempt-rt a softirq running context might be blocked on a
+ * lock. There might be no other runnable task on this CPU because the
+ * lock owner runs on some other CPU. So we have to go into idle with
+ * the pending bit set. Therefor we need to check this otherwise we
+ * warn about false positives which confuses users and defeats the
+ * whole purpose of this test.
  *
  * This code is called with interrupts disabled.
  */
 void softirq_check_pending_idle(void)
 {
 	static int rate_limit;
-	u32 warnpending = 0, pending;
+	struct softirq_runner *sr = &__get_cpu_var(softirq_runners);
+	u32 warnpending;
+	int i;
 
 	if (rate_limit >= 10)
 		return;
 
-	pending = local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK;
-	if (pending) {
-		struct task_struct *tsk;
+	warnpending = local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK;
+	for (i = 0; i < NR_SOFTIRQS; i++) {
+		struct task_struct *tsk = sr->runner[i];
 
-		tsk = __get_cpu_var(ksoftirqd);
 		/*
 		 * The wakeup code in rtmutex.c wakes up the task
 		 * _before_ it sets pi_blocked_on to NULL under
 		 * tsk->pi_lock. So we need to check for both: state
 		 * and pi_blocked_on.
 		 */
-		raw_spin_lock(&tsk->pi_lock);
-
-		if (!tsk->pi_blocked_on && !(tsk->state == TASK_RUNNING))
-			warnpending = 1;
-
-		raw_spin_unlock(&tsk->pi_lock);
+		if (tsk) {
+			raw_spin_lock(&tsk->pi_lock);
+			if (tsk->pi_blocked_on || tsk->state == TASK_RUNNING) {
+				/* Clear all bits pending in that task */
+				warnpending &= ~(tsk->softirqs_raised);
+				warnpending &= ~(1 << i);
+			}
+			raw_spin_unlock(&tsk->pi_lock);
+		}
 	}
 
 	if (warnpending) {
 		printk(KERN_ERR "NOHZ: local_softirq_pending %02x\n",
-		       pending);
+		       warnpending);
 		rate_limit++;
 	}
 }
@@ -125,6 +150,10 @@ void softirq_check_pending_idle(void)
 	}
 }
 # endif
+
+#else /* !CONFIG_NO_HZ_COMMON */
+static inline void softirq_set_runner(unsigned int sirq) { }
+static inline void softirq_clr_runner(unsigned int sirq) { }
 #endif
 
 /*
@@ -486,6 +515,7 @@ static void do_current_softirqs(int need
 		 */
 		lock_softirq(i);
 		local_irq_disable();
+		softirq_set_runner(i);
 		/*
 		 * Check with the local_softirq_pending() bits,
 		 * whether we need to process this still or if someone
@@ -496,6 +526,7 @@ static void do_current_softirqs(int need
 			set_softirq_pending(pending & ~mask);
 			do_single_softirq(i, need_rcu_bh_qs);
 		}
+		softirq_clr_runner(i);
 		unlock_softirq(i);
 		WARN_ON(current->softirq_nestcnt != 1);
 	}
@@ -566,7 +597,7 @@ void thread_do_softirq(void)
 	}
 }
 
-void __raise_softirq_irqoff(unsigned int nr)
+static void do_raise_softirq_irqoff(unsigned int nr)
 {
 	trace_softirq_raise(nr);
 	or_softirq_pending(1UL << nr);
@@ -583,12 +614,19 @@ void __raise_softirq_irqoff(unsigned int
 		__this_cpu_read(ksoftirqd)->softirqs_raised |= (1U << nr);
 }
 
+void __raise_softirq_irqoff(unsigned int nr)
+{
+	do_raise_softirq_irqoff(nr);
+	if (!in_irq() && !current->softirq_nestcnt)
+		wakeup_softirqd();
+}
+
 /*
  * This function must run with irqs disabled!
  */
 void raise_softirq_irqoff(unsigned int nr)
 {
-	__raise_softirq_irqoff(nr);
+	do_raise_softirq_irqoff(nr);
 
 	/*
 	 * If we're in an hard interrupt we let irq return code deal
@@ -610,11 +648,6 @@ void raise_softirq_irqoff(unsigned int n
 		wakeup_softirqd();
 }
 
-void do_raise_softirq_irqoff(unsigned int nr)
-{
-	raise_softirq_irqoff(nr);
-}
-
 static inline int ksoftirqd_softirq_pending(void)
 {
 	return current->softirqs_raised;