Subject: irqwork: push most work into softirq context
From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Date: Tue, 23 Jun 2015 15:32:51 +0200
Git-repo: git://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-rt-devel.git
Git-commit: da77ceac3d20f27310a07a7c346a4ee6b40d6c28
Patch-mainline: Queued in subsystem maintainer repository
References: SLE Realtime Extension

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: Daniel Wagner <dwagner@suse.de>
---
 include/linux/irq_work.h |    8 ++++++
 kernel/irq_work.c        |   59 +++++++++++++++++++++++++++++++++++++----------
 kernel/rcu/tree.c        |    1 
 kernel/sched/topology.c  |    1 
 kernel/time/tick-sched.c |    1 
 kernel/time/timer.c      |    2 +
 6 files changed, 60 insertions(+), 12 deletions(-)

--- a/include/linux/irq_work.h
+++ b/include/linux/irq_work.h
@@ -18,6 +18,8 @@
 
 /* Doesn't want IPI, wait for tick: */
 #define IRQ_WORK_LAZY		BIT(2)
+/* Run hard IRQ context, even on RT */
+#define IRQ_WORK_HARD_IRQ	BIT(3)
 
 #define IRQ_WORK_CLAIMED	(IRQ_WORK_PENDING | IRQ_WORK_BUSY)
 
@@ -56,4 +58,10 @@ static inline bool irq_work_needs_cpu(vo
 static inline void irq_work_run(void) { }
 #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 */
--- a/kernel/irq_work.c
+++ b/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>
 
 
@@ -52,13 +53,19 @@ 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)
 {
+	struct llist_head *list;
+	bool lazy_work, realtime = IS_ENABLED(CONFIG_PREEMPT_RT);
+
+	lazy_work = atomic_read(&work->flags) & IRQ_WORK_LAZY;
+
 	/* If the work is "lazy", handle it from next tick if any */
-	if (atomic_read(&work->flags) & IRQ_WORK_LAZY) {
-		if (llist_add(&work->llnode, this_cpu_ptr(&lazy_list)) &&
-		    tick_nohz_tick_stopped())
-			arch_irq_work_raise();
-	} else {
-		if (llist_add(&work->llnode, this_cpu_ptr(&raised_list)))
+	if (lazy_work || (realtime && !(atomic_read(&work->flags) & IRQ_WORK_HARD_IRQ)))
+		list = this_cpu_ptr(&lazy_list);
+	else
+		list = this_cpu_ptr(&raised_list);
+
+	if (llist_add(&work->llnode, list)) {
+		if (!lazy_work || tick_nohz_tick_stopped())
 			arch_irq_work_raise();
 	}
 }
@@ -100,9 +107,16 @@ bool irq_work_queue_on(struct irq_work *
 
 	preempt_disable();
 	if (cpu != smp_processor_id()) {
+		struct llist_head *list;
+
 		/* Arch remote IPI send/receive backend aren't NMI safe */
 		WARN_ON_ONCE(in_nmi());
-		if (llist_add(&work->llnode, &per_cpu(raised_list, cpu)))
+		if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(atomic_read(&work->flags) & IRQ_WORK_HARD_IRQ))
+			list = &per_cpu(lazy_list, cpu);
+		else
+			list = &per_cpu(raised_list, cpu);
+
+		if (llist_add(&work->llnode, list))
 			arch_send_call_function_single_ipi(cpu);
 	} else {
 		__irq_work_queue_local(work);
@@ -121,9 +135,8 @@ 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()));
@@ -136,8 +149,12 @@ static void irq_work_run_list(struct lli
 	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;
 
@@ -170,7 +187,16 @@ static void irq_work_run_list(struct lli
 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);
 
@@ -180,8 +206,17 @@ 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
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -1090,6 +1090,7 @@ static int rcu_implicit_dynticks_qs(stru
 		    !rdp->rcu_iw_pending && rdp->rcu_iw_gp_seq != rnp->gp_seq &&
 		    (rnp->ffmask & rdp->grpmask)) {
 			init_irq_work(&rdp->rcu_iw, rcu_iw_handler);
+			atomic_set(&rdp->rcu_iw.flags, IRQ_WORK_HARD_IRQ);
 			rdp->rcu_iw_pending = true;
 			rdp->rcu_iw_gp_seq = rnp->gp_seq;
 			irq_work_queue_on(&rdp->rcu_iw, rdp->cpu);
--- a/kernel/sched/topology.c
+++ b/kernel/sched/topology.c
@@ -510,6 +510,7 @@ static int init_rootdomain(struct root_d
 	rd->rto_cpu = -1;
 	raw_spin_lock_init(&rd->rto_lock);
 	init_irq_work(&rd->rto_push_work, rto_push_irq_work_func);
+	atomic_or(IRQ_WORK_HARD_IRQ, &rd->rto_push_work.flags);
 #endif
 
 	init_dl_bw(&rd->dl_bw);
--- a/kernel/time/tick-sched.c
+++ b/kernel/time/tick-sched.c
@@ -239,6 +239,7 @@ static void nohz_full_kick_func(struct i
 
 static DEFINE_PER_CPU(struct irq_work, nohz_full_kick_work) = {
 	.func = nohz_full_kick_func,
+	.flags = IRQ_WORK_HARD_IRQ,
 };
 
 /*
--- a/kernel/time/timer.c
+++ b/kernel/time/timer.c
@@ -1792,6 +1792,8 @@ static __latent_entropy void run_timer_s
 {
 	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]));