irqwork: push most work into softirq context

Initially we defered all irqwork into softirq because we didn't want the
latency spikes if perf or another user was busy and delayed the RT task.
The NOHZ trigger (nohz_full_kick_work) was the first user that did not work
as expected if it did not run in the original irqwork context so we had to
bring it back somehow for it. push_irq_work_func is the second one that
requires this.

This patch adds the IRQ_WORK_HARD_IRQ which makes sure the callback runs
in raw-irq context. Everything else is defered into softirq context. Without
-RT we have the orignal behavior.

This patch incorporates tglx orignal work which revoked a little bringing back
the arch_irq_work_raise() if possible and a few fixes from Steven Rostedt and
Mike Galbraith,

[bigeasy: melt tglx's irq_work_tick_soft() which splits irq_work_tick() into a
          hard and soft variant]
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

include/linux/irq_work.h

@@ -64,4 +64,10 @@ static inline void irq_work_run(void) { }
 static inline void irq_work_single(void *arg) { }
 #endif
 
+#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_PREEMPT_RT)
+void irq_work_tick_soft(void);
+#else
+static inline void irq_work_tick_soft(void) { }
+#endif
+
 #endif /* _LINUX_IRQ_WORK_H */

kernel/irq_work.c

@@ -18,6 +18,7 @@
 #include <linux/cpu.h>
 #include <linux/notifier.h>
 #include <linux/smp.h>
+#include <linux/interrupt.h>
 #include <asm/processor.h>
 #include <linux/kasan.h>
 
@@ -52,13 +53,27 @@ void __weak arch_irq_work_raise(void)
 /* Enqueue on current CPU, work must already be claimed and preempt disabled */
 static void __irq_work_queue_local(struct irq_work *work)
 {
-	/* If the work is "lazy", handle it from next tick if any */
-	if (atomic_read(&work->node.a_flags) & IRQ_WORK_LAZY) {
-		if (llist_add(&work->node.llist, this_cpu_ptr(&lazy_list)) &&
-		    tick_nohz_tick_stopped())
-			arch_irq_work_raise();
-	} else {
-		if (llist_add(&work->node.llist, this_cpu_ptr(&raised_list)))
+	struct llist_head *list;
+	bool lazy_work;
+	int work_flags;
+
+	work_flags = atomic_read(&work->node.a_flags);
+	if (work_flags & IRQ_WORK_LAZY)
+		lazy_work = true;
+	else if (IS_ENABLED(CONFIG_PREEMPT_RT) &&
+		!(work_flags & IRQ_WORK_HARD_IRQ))
+			lazy_work = true;
+	else
+		lazy_work = false;
+
+	if (lazy_work)
+		list = this_cpu_ptr(&lazy_list);
+	else
+		list = this_cpu_ptr(&raised_list);
+
+	if (llist_add(&work->node.llist, list)) {
+		/* If the work is "lazy", handle it from next tick if any */
+		if (!lazy_work || tick_nohz_tick_stopped())
 			arch_irq_work_raise();
 	}
 }
@@ -104,7 +119,14 @@ bool irq_work_queue_on(struct irq_work *work, int cpu)
 	if (cpu != smp_processor_id()) {
 		/* Arch remote IPI send/receive backend aren't NMI safe */
 		WARN_ON_ONCE(in_nmi());
-		__smp_call_single_queue(cpu, &work->node.llist);
+
+		if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(atomic_read(&work->node.a_flags) & IRQ_WORK_HARD_IRQ)) {
+			if (llist_add(&work->node.llist, &per_cpu(lazy_list, cpu)))
+				/* && tick_nohz_tick_stopped_cpu(cpu) */
+				arch_send_call_function_single_ipi(cpu);
+		} else {
+			__smp_call_single_queue(cpu, &work->node.llist);
+		}
 	} else {
 		__irq_work_queue_local(work);
 	}
@@ -122,9 +144,8 @@ bool irq_work_needs_cpu(void)
 	raised = this_cpu_ptr(&raised_list);
 	lazy = this_cpu_ptr(&lazy_list);
 
-	if (llist_empty(raised) || arch_irq_work_has_interrupt())
-		if (llist_empty(lazy))
-			return false;
+	if (llist_empty(raised) && llist_empty(lazy))
+		return false;
 
 	/* All work should have been flushed before going offline */
 	WARN_ON_ONCE(cpu_is_offline(smp_processor_id()));
@@ -167,8 +188,12 @@ static void irq_work_run_list(struct llist_head *list)
 	struct irq_work *work, *tmp;
 	struct llist_node *llnode;
 
+#ifndef CONFIG_PREEMPT_RT
+	/*
+	 * nort: On RT IRQ-work may run in SOFTIRQ context.
+	 */
 	BUG_ON(!irqs_disabled());
-
+#endif
 	if (llist_empty(list))
 		return;
 
@@ -184,7 +209,16 @@ static void irq_work_run_list(struct llist_head *list)
 void irq_work_run(void)
 {
 	irq_work_run_list(this_cpu_ptr(&raised_list));
-	irq_work_run_list(this_cpu_ptr(&lazy_list));
+	if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
+		/*
+		 * NOTE: we raise softirq via IPI for safety,
+		 * and execute in irq_work_tick() to move the
+		 * overhead from hard to soft irq context.
+		 */
+		if (!llist_empty(this_cpu_ptr(&lazy_list)))
+			raise_softirq(TIMER_SOFTIRQ);
+	} else
+		irq_work_run_list(this_cpu_ptr(&lazy_list));
 }
 EXPORT_SYMBOL_GPL(irq_work_run);
 
@@ -194,8 +228,17 @@ void irq_work_tick(void)
 
 	if (!llist_empty(raised) && !arch_irq_work_has_interrupt())
 		irq_work_run_list(raised);
+
+	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
+		irq_work_run_list(this_cpu_ptr(&lazy_list));
+}
+
+#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_PREEMPT_RT)
+void irq_work_tick_soft(void)
+{
 	irq_work_run_list(this_cpu_ptr(&lazy_list));
 }
+#endif
 
 /*
  * Synchronize against the irq_work @entry, ensures the entry is not

kernel/sched/topology.c

@@ -526,7 +526,8 @@ static int init_rootdomain(struct root_domain *rd)
 #ifdef HAVE_RT_PUSH_IPI
 	rd->rto_cpu = -1;
 	raw_spin_lock_init(&rd->rto_lock);
-	init_irq_work(&rd->rto_push_work, rto_push_irq_work_func);
+//	init_irq_work(&rd->rto_push_work, rto_push_irq_work_func);
+	rd->rto_push_work = IRQ_WORK_INIT_HARD(rto_push_irq_work_func);
 #endif
 
 	rd->visit_gen = 0;

kernel/time/timer.c

@@ -1744,6 +1744,8 @@ static __latent_entropy void run_timer_softirq(struct softirq_action *h)
 {
 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
 
+	irq_work_tick_soft();
+
 	__run_timers(base);
 	if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
 		__run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));