[PATCH 0/3] lockde,sched: Better pinning

mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed

* [PATCH 0/3] lockde,sched: Better pinning
@ 2016-04-28 15:40 Peter Zijlstra
  2016-04-28 15:41 ` [PATCH 1/3] sched: Out-of-line task_rq_lock() Peter Zijlstra
                   ` (2 more replies)
  0 siblings, 3 replies; 4+ messages in thread
From: Peter Zijlstra @ 2016-04-28 15:40 UTC (permalink / raw)
  To: mingo, tglx, mgalbraith; +Cc: linux-kernel

I've been sitting on these patches for a long while and finally got around to
finishing them.

They implement a better pinning, as requested by Ingo. I'm not entirely sure I
like what 2/3 does, but it does avoid extra code bloat, where all other
attempts resulted in changed code for !lockdep.

^ permalink raw reply	[flat|nested] 4+ messages in thread

* [PATCH 1/3] sched: Out-of-line task_rq_lock()
  2016-04-28 15:40 [PATCH 0/3] lockde,sched: Better pinning Peter Zijlstra
@ 2016-04-28 15:41 ` Peter Zijlstra
  2016-04-28 15:41 ` [PATCH 2/3] sched: Introduce struct rq_flags Peter Zijlstra
  2016-04-28 15:41 ` [PATCH 3/3] lockdep,sched: Implement a better lock pinning scheme Peter Zijlstra
  2 siblings, 0 replies; 4+ messages in thread
From: Peter Zijlstra @ 2016-04-28 15:41 UTC (permalink / raw)
  To: mingo, tglx, mgalbraith; +Cc: linux-kernel, Peter Zijlstra (Intel)

[-- Attachment #1: peterz-sched-out-of-line-task_rq_lock.patch --]
[-- Type: text/plain, Size: 4987 bytes --]

Its a rather large function, inline doesn't seems to make much sense.

$ size defconfig-build/kernel/sched/core.o{.orig,}
   text    data     bss     dec     hex filename
  56533   21037    2320   79890   13812 defconfig-build/kernel/sched/core.o.orig
  55733   21037    2320   79090   134f2 defconfig-build/kernel/sched/core.o

$ for i in /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor ; do echo performance > $i ; done
$ perf stat --null --repeat 25 -- perf bench sched messaging -g 40 -l 5000

pre:
       4.582798193 seconds time elapsed          ( +-  1.41% )
       4.733374877 seconds time elapsed          ( +-  2.10% )
       4.560955136 seconds time elapsed          ( +-  1.43% )
       4.631062303 seconds time elapsed          ( +-  1.40% )

post:
       4.364765213 seconds time elapsed          ( +-  0.91% )
       4.454442734 seconds time elapsed          ( +-  1.18% )
       4.448893817 seconds time elapsed          ( +-  1.41% )
       4.424346872 seconds time elapsed          ( +-  0.97% )

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -170,6 +170,71 @@ static struct rq *this_rq_lock(void)
 	return rq;
 }
 
+/*
+ * __task_rq_lock - lock the rq @p resides on.
+ */
+struct rq *__task_rq_lock(struct task_struct *p)
+	__acquires(rq->lock)
+{
+	struct rq *rq;
+
+	lockdep_assert_held(&p->pi_lock);
+
+	for (;;) {
+		rq = task_rq(p);
+		raw_spin_lock(&rq->lock);
+		if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
+			lockdep_pin_lock(&rq->lock);
+			return rq;
+		}
+		raw_spin_unlock(&rq->lock);
+
+		while (unlikely(task_on_rq_migrating(p)))
+			cpu_relax();
+	}
+}
+
+/*
+ * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
+ */
+struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
+	__acquires(p->pi_lock)
+	__acquires(rq->lock)
+{
+	struct rq *rq;
+
+	for (;;) {
+		raw_spin_lock_irqsave(&p->pi_lock, *flags);
+		rq = task_rq(p);
+		raw_spin_lock(&rq->lock);
+		/*
+		 *	move_queued_task()		task_rq_lock()
+		 *
+		 *	ACQUIRE (rq->lock)
+		 *	[S] ->on_rq = MIGRATING		[L] rq = task_rq()
+		 *	WMB (__set_task_cpu())		ACQUIRE (rq->lock);
+		 *	[S] ->cpu = new_cpu		[L] task_rq()
+		 *					[L] ->on_rq
+		 *	RELEASE (rq->lock)
+		 *
+		 * If we observe the old cpu in task_rq_lock, the acquire of
+		 * the old rq->lock will fully serialize against the stores.
+		 *
+		 * If we observe the new cpu in task_rq_lock, the acquire will
+		 * pair with the WMB to ensure we must then also see migrating.
+		 */
+		if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
+			lockdep_pin_lock(&rq->lock);
+			return rq;
+		}
+		raw_spin_unlock(&rq->lock);
+		raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
+
+		while (unlikely(task_on_rq_migrating(p)))
+			cpu_relax();
+	}
+}
+
 #ifdef CONFIG_SCHED_HRTICK
 /*
  * Use HR-timers to deliver accurate preemption points.
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -1451,70 +1451,11 @@ static inline void sched_rt_avg_update(s
 static inline void sched_avg_update(struct rq *rq) { }
 #endif
 
-/*
- * __task_rq_lock - lock the rq @p resides on.
- */
-static inline struct rq *__task_rq_lock(struct task_struct *p)
-	__acquires(rq->lock)
-{
-	struct rq *rq;
-
-	lockdep_assert_held(&p->pi_lock);
-
-	for (;;) {
-		rq = task_rq(p);
-		raw_spin_lock(&rq->lock);
-		if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
-			lockdep_pin_lock(&rq->lock);
-			return rq;
-		}
-		raw_spin_unlock(&rq->lock);
-
-		while (unlikely(task_on_rq_migrating(p)))
-			cpu_relax();
-	}
-}
-
-/*
- * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
- */
-static inline struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
+struct rq *__task_rq_lock(struct task_struct *p)
+	__acquires(rq->lock);
+struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
 	__acquires(p->pi_lock)
-	__acquires(rq->lock)
-{
-	struct rq *rq;
-
-	for (;;) {
-		raw_spin_lock_irqsave(&p->pi_lock, *flags);
-		rq = task_rq(p);
-		raw_spin_lock(&rq->lock);
-		/*
-		 *	move_queued_task()		task_rq_lock()
-		 *
-		 *	ACQUIRE (rq->lock)
-		 *	[S] ->on_rq = MIGRATING		[L] rq = task_rq()
-		 *	WMB (__set_task_cpu())		ACQUIRE (rq->lock);
-		 *	[S] ->cpu = new_cpu		[L] task_rq()
-		 *					[L] ->on_rq
-		 *	RELEASE (rq->lock)
-		 *
-		 * If we observe the old cpu in task_rq_lock, the acquire of
-		 * the old rq->lock will fully serialize against the stores.
-		 *
-		 * If we observe the new cpu in task_rq_lock, the acquire will
-		 * pair with the WMB to ensure we must then also see migrating.
-		 */
-		if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
-			lockdep_pin_lock(&rq->lock);
-			return rq;
-		}
-		raw_spin_unlock(&rq->lock);
-		raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
-
-		while (unlikely(task_on_rq_migrating(p)))
-			cpu_relax();
-	}
-}
+	__acquires(rq->lock);
 
 static inline void __task_rq_unlock(struct rq *rq)
 	__releases(rq->lock)

^ permalink raw reply	[flat|nested] 4+ messages in thread

* [PATCH 2/3] sched: Introduce struct rq_flags
  2016-04-28 15:40 [PATCH 0/3] lockde,sched: Better pinning Peter Zijlstra
  2016-04-28 15:41 ` [PATCH 1/3] sched: Out-of-line task_rq_lock() Peter Zijlstra
@ 2016-04-28 15:41 ` Peter Zijlstra
  2016-04-28 15:41 ` [PATCH 3/3] lockdep,sched: Implement a better lock pinning scheme Peter Zijlstra
  2 siblings, 0 replies; 4+ messages in thread
From: Peter Zijlstra @ 2016-04-28 15:41 UTC (permalink / raw)
  To: mingo, tglx, mgalbraith; +Cc: linux-kernel, Peter Zijlstra (Intel)

[-- Attachment #1: peterz-sched-task_rq_lock-flags.patch --]
[-- Type: text/plain, Size: 13375 bytes --]

In order to pass around more than just the IRQ flags, add a rq_flags
structure.

No difference in code generation for the x86_64-defconfig build I
tested.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 kernel/sched/core.c     |   98 ++++++++++++++++++++++++------------------------
 kernel/sched/deadline.c |    6 +-
 kernel/sched/sched.h    |   14 ++++--
 3 files changed, 62 insertions(+), 56 deletions(-)

--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -173,7 +173,7 @@ static struct rq *this_rq_lock(void)
 /*
  * __task_rq_lock - lock the rq @p resides on.
  */
-struct rq *__task_rq_lock(struct task_struct *p)
+struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
 	__acquires(rq->lock)
 {
 	struct rq *rq;
@@ -197,14 +197,14 @@ struct rq *__task_rq_lock(struct task_st
 /*
  * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
  */
-struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
+struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
 	__acquires(p->pi_lock)
 	__acquires(rq->lock)
 {
 	struct rq *rq;
 
 	for (;;) {
-		raw_spin_lock_irqsave(&p->pi_lock, *flags);
+		raw_spin_lock_irqsave(&p->pi_lock, rf->flags);
 		rq = task_rq(p);
 		raw_spin_lock(&rq->lock);
 		/*
@@ -228,7 +228,7 @@ struct rq *task_rq_lock(struct task_stru
 			return rq;
 		}
 		raw_spin_unlock(&rq->lock);
-		raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
+		raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
 
 		while (unlikely(task_on_rq_migrating(p)))
 			cpu_relax();
@@ -1147,12 +1147,12 @@ void do_set_cpus_allowed(struct task_str
 static int __set_cpus_allowed_ptr(struct task_struct *p,
 				  const struct cpumask *new_mask, bool check)
 {
-	unsigned long flags;
-	struct rq *rq;
 	unsigned int dest_cpu;
+	struct rq_flags rf;
+	struct rq *rq;
 	int ret = 0;
 
-	rq = task_rq_lock(p, &flags);
+	rq = task_rq_lock(p, &rf);
 
 	/*
 	 * Must re-check here, to close a race against __kthread_bind(),
@@ -1181,7 +1181,7 @@ static int __set_cpus_allowed_ptr(struct
 	if (task_running(rq, p) || p->state == TASK_WAKING) {
 		struct migration_arg arg = { p, dest_cpu };
 		/* Need help from migration thread: drop lock and wait. */
-		task_rq_unlock(rq, p, &flags);
+		task_rq_unlock(rq, p, &rf);
 		stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
 		tlb_migrate_finish(p->mm);
 		return 0;
@@ -1195,7 +1195,7 @@ static int __set_cpus_allowed_ptr(struct
 		lockdep_pin_lock(&rq->lock);
 	}
 out:
-	task_rq_unlock(rq, p, &flags);
+	task_rq_unlock(rq, p, &rf);
 
 	return ret;
 }
@@ -1379,8 +1379,8 @@ int migrate_swap(struct task_struct *cur
  */
 unsigned long wait_task_inactive(struct task_struct *p, long match_state)
 {
-	unsigned long flags;
 	int running, queued;
+	struct rq_flags rf;
 	unsigned long ncsw;
 	struct rq *rq;
 
@@ -1415,14 +1415,14 @@ unsigned long wait_task_inactive(struct
 		 * lock now, to be *sure*. If we're wrong, we'll
 		 * just go back and repeat.
 		 */
-		rq = task_rq_lock(p, &flags);
+		rq = task_rq_lock(p, &rf);
 		trace_sched_wait_task(p);
 		running = task_running(rq, p);
 		queued = task_on_rq_queued(p);
 		ncsw = 0;
 		if (!match_state || p->state == match_state)
 			ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
-		task_rq_unlock(rq, p, &flags);
+		task_rq_unlock(rq, p, &rf);
 
 		/*
 		 * If it changed from the expected state, bail out now.
@@ -1720,17 +1720,18 @@ ttwu_do_activate(struct rq *rq, struct t
  */
 static int ttwu_remote(struct task_struct *p, int wake_flags)
 {
+	struct rq_flags rf;
 	struct rq *rq;
 	int ret = 0;
 
-	rq = __task_rq_lock(p);
+	rq = __task_rq_lock(p, &rf);
 	if (task_on_rq_queued(p)) {
 		/* check_preempt_curr() may use rq clock */
 		update_rq_clock(rq);
 		ttwu_do_wakeup(rq, p, wake_flags);
 		ret = 1;
 	}
-	__task_rq_unlock(rq);
+	__task_rq_unlock(rq, &rf);
 
 	return ret;
 }
@@ -2483,12 +2484,12 @@ extern void init_dl_bw(struct dl_bw *dl_
  */
 void wake_up_new_task(struct task_struct *p)
 {
-	unsigned long flags;
+	struct rq_flags rf;
 	struct rq *rq;
 
-	raw_spin_lock_irqsave(&p->pi_lock, flags);
 	/* Initialize new task's runnable average */
 	init_entity_runnable_average(&p->se);
+	raw_spin_lock_irqsave(&p->pi_lock, rf.flags);
 #ifdef CONFIG_SMP
 	/*
 	 * Fork balancing, do it here and not earlier because:
@@ -2500,7 +2501,7 @@ void wake_up_new_task(struct task_struct
 	/* Post initialize new task's util average when its cfs_rq is set */
 	post_init_entity_util_avg(&p->se);
 
-	rq = __task_rq_lock(p);
+	rq = __task_rq_lock(p, &rf);
 	activate_task(rq, p, 0);
 	p->on_rq = TASK_ON_RQ_QUEUED;
 	trace_sched_wakeup_new(p);
@@ -2516,7 +2517,7 @@ void wake_up_new_task(struct task_struct
 		lockdep_pin_lock(&rq->lock);
 	}
 #endif
-	task_rq_unlock(rq, p, &flags);
+	task_rq_unlock(rq, p, &rf);
 }
 
 #ifdef CONFIG_PREEMPT_NOTIFIERS
@@ -2932,7 +2933,7 @@ EXPORT_PER_CPU_SYMBOL(kernel_cpustat);
  */
 unsigned long long task_sched_runtime(struct task_struct *p)
 {
-	unsigned long flags;
+	struct rq_flags rf;
 	struct rq *rq;
 	u64 ns;
 
@@ -2952,7 +2953,7 @@ unsigned long long task_sched_runtime(st
 		return p->se.sum_exec_runtime;
 #endif
 
-	rq = task_rq_lock(p, &flags);
+	rq = task_rq_lock(p, &rf);
 	/*
 	 * Must be ->curr _and_ ->on_rq.  If dequeued, we would
 	 * project cycles that may never be accounted to this
@@ -2963,7 +2964,7 @@ unsigned long long task_sched_runtime(st
 		p->sched_class->update_curr(rq);
 	}
 	ns = p->se.sum_exec_runtime;
-	task_rq_unlock(rq, p, &flags);
+	task_rq_unlock(rq, p, &rf);
 
 	return ns;
 }
@@ -3521,12 +3522,13 @@ EXPORT_SYMBOL(default_wake_function);
 void rt_mutex_setprio(struct task_struct *p, int prio)
 {
 	int oldprio, queued, running, queue_flag = DEQUEUE_SAVE | DEQUEUE_MOVE;
-	struct rq *rq;
 	const struct sched_class *prev_class;
+	struct rq_flags rf;
+	struct rq *rq;
 
 	BUG_ON(prio > MAX_PRIO);
 
-	rq = __task_rq_lock(p);
+	rq = __task_rq_lock(p, &rf);
 
 	/*
 	 * Idle task boosting is a nono in general. There is one
@@ -3602,7 +3604,7 @@ void rt_mutex_setprio(struct task_struct
 	check_class_changed(rq, p, prev_class, oldprio);
 out_unlock:
 	preempt_disable(); /* avoid rq from going away on us */
-	__task_rq_unlock(rq);
+	__task_rq_unlock(rq, &rf);
 
 	balance_callback(rq);
 	preempt_enable();
@@ -3612,7 +3614,7 @@ void rt_mutex_setprio(struct task_struct
 void set_user_nice(struct task_struct *p, long nice)
 {
 	int old_prio, delta, queued;
-	unsigned long flags;
+	struct rq_flags rf;
 	struct rq *rq;
 
 	if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE)
@@ -3621,7 +3623,7 @@ void set_user_nice(struct task_struct *p
 	 * We have to be careful, if called from sys_setpriority(),
 	 * the task might be in the middle of scheduling on another CPU.
 	 */
-	rq = task_rq_lock(p, &flags);
+	rq = task_rq_lock(p, &rf);
 	/*
 	 * The RT priorities are set via sched_setscheduler(), but we still
 	 * allow the 'normal' nice value to be set - but as expected
@@ -3652,7 +3654,7 @@ void set_user_nice(struct task_struct *p
 			resched_curr(rq);
 	}
 out_unlock:
-	task_rq_unlock(rq, p, &flags);
+	task_rq_unlock(rq, p, &rf);
 }
 EXPORT_SYMBOL(set_user_nice);
 
@@ -3949,11 +3951,11 @@ static int __sched_setscheduler(struct t
 		      MAX_RT_PRIO - 1 - attr->sched_priority;
 	int retval, oldprio, oldpolicy = -1, queued, running;
 	int new_effective_prio, policy = attr->sched_policy;
-	unsigned long flags;
 	const struct sched_class *prev_class;
-	struct rq *rq;
+	struct rq_flags rf;
 	int reset_on_fork;
 	int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE;
+	struct rq *rq;
 
 	/* may grab non-irq protected spin_locks */
 	BUG_ON(in_interrupt());
@@ -4048,13 +4050,13 @@ static int __sched_setscheduler(struct t
 	 * To be able to change p->policy safely, the appropriate
 	 * runqueue lock must be held.
 	 */
-	rq = task_rq_lock(p, &flags);
+	rq = task_rq_lock(p, &rf);
 
 	/*
 	 * Changing the policy of the stop threads its a very bad idea
 	 */
 	if (p == rq->stop) {
-		task_rq_unlock(rq, p, &flags);
+		task_rq_unlock(rq, p, &rf);
 		return -EINVAL;
 	}
 
@@ -4071,7 +4073,7 @@ static int __sched_setscheduler(struct t
 			goto change;
 
 		p->sched_reset_on_fork = reset_on_fork;
-		task_rq_unlock(rq, p, &flags);
+		task_rq_unlock(rq, p, &rf);
 		return 0;
 	}
 change:
@@ -4085,7 +4087,7 @@ static int __sched_setscheduler(struct t
 		if (rt_bandwidth_enabled() && rt_policy(policy) &&
 				task_group(p)->rt_bandwidth.rt_runtime == 0 &&
 				!task_group_is_autogroup(task_group(p))) {
-			task_rq_unlock(rq, p, &flags);
+			task_rq_unlock(rq, p, &rf);
 			return -EPERM;
 		}
 #endif
@@ -4100,7 +4102,7 @@ static int __sched_setscheduler(struct t
 			 */
 			if (!cpumask_subset(span, &p->cpus_allowed) ||
 			    rq->rd->dl_bw.bw == 0) {
-				task_rq_unlock(rq, p, &flags);
+				task_rq_unlock(rq, p, &rf);
 				return -EPERM;
 			}
 		}
@@ -4110,7 +4112,7 @@ static int __sched_setscheduler(struct t
 	/* recheck policy now with rq lock held */
 	if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
 		policy = oldpolicy = -1;
-		task_rq_unlock(rq, p, &flags);
+		task_rq_unlock(rq, p, &rf);
 		goto recheck;
 	}
 
@@ -4120,7 +4122,7 @@ static int __sched_setscheduler(struct t
 	 * is available.
 	 */
 	if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) {
-		task_rq_unlock(rq, p, &flags);
+		task_rq_unlock(rq, p, &rf);
 		return -EBUSY;
 	}
 
@@ -4165,7 +4167,7 @@ static int __sched_setscheduler(struct t
 
 	check_class_changed(rq, p, prev_class, oldprio);
 	preempt_disable(); /* avoid rq from going away on us */
-	task_rq_unlock(rq, p, &flags);
+	task_rq_unlock(rq, p, &rf);
 
 	if (pi)
 		rt_mutex_adjust_pi(p);
@@ -5018,10 +5020,10 @@ SYSCALL_DEFINE2(sched_rr_get_interval, p
 {
 	struct task_struct *p;
 	unsigned int time_slice;
-	unsigned long flags;
+	struct rq_flags rf;
+	struct timespec t;
 	struct rq *rq;
 	int retval;
-	struct timespec t;
 
 	if (pid < 0)
 		return -EINVAL;
@@ -5036,11 +5038,11 @@ SYSCALL_DEFINE2(sched_rr_get_interval, p
 	if (retval)
 		goto out_unlock;
 
-	rq = task_rq_lock(p, &flags);
+	rq = task_rq_lock(p, &rf);
 	time_slice = 0;
 	if (p->sched_class->get_rr_interval)
 		time_slice = p->sched_class->get_rr_interval(rq, p);
-	task_rq_unlock(rq, p, &flags);
+	task_rq_unlock(rq, p, &rf);
 
 	rcu_read_unlock();
 	jiffies_to_timespec(time_slice, &t);
@@ -5304,11 +5306,11 @@ int migrate_task_to(struct task_struct *
  */
 void sched_setnuma(struct task_struct *p, int nid)
 {
-	struct rq *rq;
-	unsigned long flags;
 	bool queued, running;
+	struct rq_flags rf;
+	struct rq *rq;
 
-	rq = task_rq_lock(p, &flags);
+	rq = task_rq_lock(p, &rf);
 	queued = task_on_rq_queued(p);
 	running = task_current(rq, p);
 
@@ -5323,7 +5325,7 @@ void sched_setnuma(struct task_struct *p
 		p->sched_class->set_curr_task(rq);
 	if (queued)
 		enqueue_task(rq, p, ENQUEUE_RESTORE);
-	task_rq_unlock(rq, p, &flags);
+	task_rq_unlock(rq, p, &rf);
 }
 #endif /* CONFIG_NUMA_BALANCING */
 
@@ -7754,10 +7756,10 @@ void sched_move_task(struct task_struct
 {
 	struct task_group *tg;
 	int queued, running;
-	unsigned long flags;
+	struct rq_flags rf;
 	struct rq *rq;
 
-	rq = task_rq_lock(tsk, &flags);
+	rq = task_rq_lock(tsk, &rf);
 
 	running = task_current(rq, tsk);
 	queued = task_on_rq_queued(tsk);
@@ -7789,7 +7791,7 @@ void sched_move_task(struct task_struct
 	if (queued)
 		enqueue_task(rq, tsk, ENQUEUE_RESTORE | ENQUEUE_MOVE);
 
-	task_rq_unlock(rq, tsk, &flags);
+	task_rq_unlock(rq, tsk, &rf);
 }
 #endif /* CONFIG_CGROUP_SCHED */
 
--- a/kernel/sched/deadline.c
+++ b/kernel/sched/deadline.c
@@ -591,10 +591,10 @@ static enum hrtimer_restart dl_task_time
 						     struct sched_dl_entity,
 						     dl_timer);
 	struct task_struct *p = dl_task_of(dl_se);
-	unsigned long flags;
+	struct rq_flags rf;
 	struct rq *rq;
 
-	rq = task_rq_lock(p, &flags);
+	rq = task_rq_lock(p, &rf);
 
 	/*
 	 * The task might have changed its scheduling policy to something
@@ -677,7 +677,7 @@ static enum hrtimer_restart dl_task_time
 #endif
 
 unlock:
-	task_rq_unlock(rq, p, &flags);
+	task_rq_unlock(rq, p, &rf);
 
 	/*
 	 * This can free the task_struct, including this hrtimer, do not touch
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -1451,13 +1451,17 @@ static inline void sched_rt_avg_update(s
 static inline void sched_avg_update(struct rq *rq) { }
 #endif
 
-struct rq *__task_rq_lock(struct task_struct *p)
+struct rq_flags {
+	unsigned long flags;
+};
+
+struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
 	__acquires(rq->lock);
-struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
+struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
 	__acquires(p->pi_lock)
 	__acquires(rq->lock);
 
-static inline void __task_rq_unlock(struct rq *rq)
+static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
 	__releases(rq->lock)
 {
 	lockdep_unpin_lock(&rq->lock);
@@ -1465,13 +1469,13 @@ static inline void __task_rq_unlock(stru
 }
 
 static inline void
-task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags)
+task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf)
 	__releases(rq->lock)
 	__releases(p->pi_lock)
 {
 	lockdep_unpin_lock(&rq->lock);
 	raw_spin_unlock(&rq->lock);
-	raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
+	raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
 }
 
 #ifdef CONFIG_SMP

^ permalink raw reply	[flat|nested] 4+ messages in thread

* [PATCH 3/3] lockdep,sched: Implement a better lock pinning scheme
  2016-04-28 15:40 [PATCH 0/3] lockde,sched: Better pinning Peter Zijlstra
  2016-04-28 15:41 ` [PATCH 1/3] sched: Out-of-line task_rq_lock() Peter Zijlstra
  2016-04-28 15:41 ` [PATCH 2/3] sched: Introduce struct rq_flags Peter Zijlstra
@ 2016-04-28 15:41 ` Peter Zijlstra
  2 siblings, 0 replies; 4+ messages in thread
From: Peter Zijlstra @ 2016-04-28 15:41 UTC (permalink / raw)
  To: mingo, tglx, mgalbraith; +Cc: linux-kernel, Peter Zijlstra (Intel)

[-- Attachment #1: peterz-lockdep-better-pin.patch --]
[-- Type: text/plain, Size: 20371 bytes --]

The problem with the existing lock pinning is that each pin is of
value 1; this mean you can simply unpin if you know its pinned,
without having any extra information.

This scheme generates a random (16 bit) cookie for each pin and
requires this same cookie to unpin. This means you have to keep the
cookie in context.

No objsize difference for !LOCKDEP kernels.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 include/linux/lockdep.h  |   23 ++++++++++---
 kernel/locking/lockdep.c |   71 ++++++++++++++++++++++++++++++++++++------
 kernel/sched/core.c      |   79 ++++++++++++++++++++++++-----------------------
 kernel/sched/deadline.c  |   11 +++---
 kernel/sched/fair.c      |    6 +--
 kernel/sched/idle_task.c |    2 -
 kernel/sched/rt.c        |    6 +--
 kernel/sched/sched.h     |    8 ++--
 kernel/sched/stop_task.c |    2 -
 9 files changed, 140 insertions(+), 68 deletions(-)

--- a/include/linux/lockdep.h
+++ b/include/linux/lockdep.h
@@ -356,8 +356,13 @@ extern void lockdep_set_current_reclaim_
 extern void lockdep_clear_current_reclaim_state(void);
 extern void lockdep_trace_alloc(gfp_t mask);
 
-extern void lock_pin_lock(struct lockdep_map *lock);
-extern void lock_unpin_lock(struct lockdep_map *lock);
+struct pin_cookie { unsigned int val; };
+
+#define NIL_COOKIE (struct pin_cookie){ .val = 0U, }
+
+extern struct pin_cookie lock_pin_lock(struct lockdep_map *lock);
+extern void lock_repin_lock(struct lockdep_map *lock, struct pin_cookie);
+extern void lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie);
 
 # define INIT_LOCKDEP				.lockdep_recursion = 0, .lockdep_reclaim_gfp = 0,
 
@@ -373,8 +378,9 @@ extern void lock_unpin_lock(struct lockd
 
 #define lockdep_recursing(tsk)	((tsk)->lockdep_recursion)
 
-#define lockdep_pin_lock(l)		lock_pin_lock(&(l)->dep_map)
-#define lockdep_unpin_lock(l)	lock_unpin_lock(&(l)->dep_map)
+#define lockdep_pin_lock(l)	lock_pin_lock(&(l)->dep_map)
+#define lockdep_repin_lock(l,c)	lock_repin_lock(&(l)->dep_map, (c))
+#define lockdep_unpin_lock(l,c)	lock_unpin_lock(&(l)->dep_map, (c))
 
 #else /* !CONFIG_LOCKDEP */
 
@@ -427,8 +433,13 @@ struct lock_class_key { };
 
 #define lockdep_recursing(tsk)			(0)
 
-#define lockdep_pin_lock(l)				do { (void)(l); } while (0)
-#define lockdep_unpin_lock(l)			do { (void)(l); } while (0)
+struct pin_cookie { };
+
+#define NIL_COOKIE (struct pin_cookie){ }
+
+#define lockdep_pin_lock(l)			({ struct pin_cookie cookie; cookie; })
+#define lockdep_repin_lock(l, c)		do { (void)(l); (void)(c); } while (0)
+#define lockdep_unpin_lock(l, c)		do { (void)(l); (void)(c); } while (0)
 
 #endif /* !LOCKDEP */
 
--- a/kernel/locking/lockdep.c
+++ b/kernel/locking/lockdep.c
@@ -45,6 +45,7 @@
 #include <linux/bitops.h>
 #include <linux/gfp.h>
 #include <linux/kmemcheck.h>
+#include <linux/random.h>
 
 #include <asm/sections.h>
 
@@ -3585,7 +3586,35 @@ static int __lock_is_held(struct lockdep
 	return 0;
 }
 
-static void __lock_pin_lock(struct lockdep_map *lock)
+static struct pin_cookie __lock_pin_lock(struct lockdep_map *lock)
+{
+	struct pin_cookie cookie = NIL_COOKIE;
+	struct task_struct *curr = current;
+	int i;
+
+	if (unlikely(!debug_locks))
+		return cookie;
+
+	for (i = 0; i < curr->lockdep_depth; i++) {
+		struct held_lock *hlock = curr->held_locks + i;
+
+		if (match_held_lock(hlock, lock)) {
+			/*
+			 * Grab 16bits of randomness; this is sufficient to not
+			 * be guessable and still allows some pin nesting in
+			 * our u32 pin_count.
+			 */
+			cookie.val = 1 + (prandom_u32() >> 16);
+			hlock->pin_count += cookie.val;
+			return cookie;
+		}
+	}
+
+	WARN(1, "pinning an unheld lock\n");
+	return cookie;
+}
+
+static void __lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
 {
 	struct task_struct *curr = current;
 	int i;
@@ -3597,7 +3626,7 @@ static void __lock_pin_lock(struct lockd
 		struct held_lock *hlock = curr->held_locks + i;
 
 		if (match_held_lock(hlock, lock)) {
-			hlock->pin_count++;
+			hlock->pin_count += cookie.val;
 			return;
 		}
 	}
@@ -3605,7 +3634,7 @@ static void __lock_pin_lock(struct lockd
 	WARN(1, "pinning an unheld lock\n");
 }
 
-static void __lock_unpin_lock(struct lockdep_map *lock)
+static void __lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
 {
 	struct task_struct *curr = current;
 	int i;
@@ -3620,7 +3649,11 @@ static void __lock_unpin_lock(struct loc
 			if (WARN(!hlock->pin_count, "unpinning an unpinned lock\n"))
 				return;
 
-			hlock->pin_count--;
+			hlock->pin_count -= cookie.val;
+
+			if (WARN((int)hlock->pin_count < 0, "pin count corrupted\n"))
+				hlock->pin_count = 0;
+
 			return;
 		}
 	}
@@ -3751,24 +3784,44 @@ int lock_is_held(struct lockdep_map *loc
 }
 EXPORT_SYMBOL_GPL(lock_is_held);
 
-void lock_pin_lock(struct lockdep_map *lock)
+struct pin_cookie lock_pin_lock(struct lockdep_map *lock)
 {
+	struct pin_cookie cookie = NIL_COOKIE;
 	unsigned long flags;
 
 	if (unlikely(current->lockdep_recursion))
-		return;
+		return cookie;
 
 	raw_local_irq_save(flags);
 	check_flags(flags);
 
 	current->lockdep_recursion = 1;
-	__lock_pin_lock(lock);
+	cookie = __lock_pin_lock(lock);
 	current->lockdep_recursion = 0;
 	raw_local_irq_restore(flags);
+
+	return cookie;
 }
 EXPORT_SYMBOL_GPL(lock_pin_lock);
 
-void lock_unpin_lock(struct lockdep_map *lock)
+void lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
+{
+	unsigned long flags;
+
+	if (unlikely(current->lockdep_recursion))
+		return;
+
+	raw_local_irq_save(flags);
+	check_flags(flags);
+
+	current->lockdep_recursion = 1;
+	__lock_repin_lock(lock, cookie);
+	current->lockdep_recursion = 0;
+	raw_local_irq_restore(flags);
+}
+EXPORT_SYMBOL_GPL(lock_repin_lock);
+
+void lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
 {
 	unsigned long flags;
 
@@ -3779,7 +3832,7 @@ void lock_unpin_lock(struct lockdep_map
 	check_flags(flags);
 
 	current->lockdep_recursion = 1;
-	__lock_unpin_lock(lock);
+	__lock_unpin_lock(lock, cookie);
 	current->lockdep_recursion = 0;
 	raw_local_irq_restore(flags);
 }
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -184,7 +184,7 @@ struct rq *__task_rq_lock(struct task_st
 		rq = task_rq(p);
 		raw_spin_lock(&rq->lock);
 		if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
-			lockdep_pin_lock(&rq->lock);
+			rf->cookie = lockdep_pin_lock(&rq->lock);
 			return rq;
 		}
 		raw_spin_unlock(&rq->lock);
@@ -224,7 +224,7 @@ struct rq *task_rq_lock(struct task_stru
 		 * pair with the WMB to ensure we must then also see migrating.
 		 */
 		if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
-			lockdep_pin_lock(&rq->lock);
+			rf->cookie = lockdep_pin_lock(&rq->lock);
 			return rq;
 		}
 		raw_spin_unlock(&rq->lock);
@@ -1190,9 +1190,9 @@ static int __set_cpus_allowed_ptr(struct
 		 * OK, since we're going to drop the lock immediately
 		 * afterwards anyway.
 		 */
-		lockdep_unpin_lock(&rq->lock);
+		lockdep_unpin_lock(&rq->lock, rf.cookie);
 		rq = move_queued_task(rq, p, dest_cpu);
-		lockdep_pin_lock(&rq->lock);
+		lockdep_repin_lock(&rq->lock, rf.cookie);
 	}
 out:
 	task_rq_unlock(rq, p, &rf);
@@ -1666,8 +1666,8 @@ static inline void ttwu_activate(struct
 /*
  * Mark the task runnable and perform wakeup-preemption.
  */
-static void
-ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
+static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags,
+			   struct pin_cookie cookie)
 {
 	check_preempt_curr(rq, p, wake_flags);
 	p->state = TASK_RUNNING;
@@ -1679,9 +1679,9 @@ ttwu_do_wakeup(struct rq *rq, struct tas
 		 * Our task @p is fully woken up and running; so its safe to
 		 * drop the rq->lock, hereafter rq is only used for statistics.
 		 */
-		lockdep_unpin_lock(&rq->lock);
+		lockdep_unpin_lock(&rq->lock, cookie);
 		p->sched_class->task_woken(rq, p);
-		lockdep_pin_lock(&rq->lock);
+		lockdep_repin_lock(&rq->lock, cookie);
 	}
 
 	if (rq->idle_stamp) {
@@ -1699,7 +1699,8 @@ ttwu_do_wakeup(struct rq *rq, struct tas
 }
 
 static void
-ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags)
+ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags,
+		 struct pin_cookie cookie)
 {
 	lockdep_assert_held(&rq->lock);
 
@@ -1709,7 +1710,7 @@ ttwu_do_activate(struct rq *rq, struct t
 #endif
 
 	ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING);
-	ttwu_do_wakeup(rq, p, wake_flags);
+	ttwu_do_wakeup(rq, p, wake_flags, cookie);
 }
 
 /*
@@ -1728,7 +1729,7 @@ static int ttwu_remote(struct task_struc
 	if (task_on_rq_queued(p)) {
 		/* check_preempt_curr() may use rq clock */
 		update_rq_clock(rq);
-		ttwu_do_wakeup(rq, p, wake_flags);
+		ttwu_do_wakeup(rq, p, wake_flags, rf.cookie);
 		ret = 1;
 	}
 	__task_rq_unlock(rq, &rf);
@@ -1741,6 +1742,7 @@ void sched_ttwu_pending(void)
 {
 	struct rq *rq = this_rq();
 	struct llist_node *llist = llist_del_all(&rq->wake_list);
+	struct pin_cookie cookie;
 	struct task_struct *p;
 	unsigned long flags;
 
@@ -1748,15 +1750,15 @@ void sched_ttwu_pending(void)
 		return;
 
 	raw_spin_lock_irqsave(&rq->lock, flags);
-	lockdep_pin_lock(&rq->lock);
+	cookie = lockdep_pin_lock(&rq->lock);
 
 	while (llist) {
 		p = llist_entry(llist, struct task_struct, wake_entry);
 		llist = llist_next(llist);
-		ttwu_do_activate(rq, p, 0);
+		ttwu_do_activate(rq, p, 0, cookie);
 	}
 
-	lockdep_unpin_lock(&rq->lock);
+	lockdep_unpin_lock(&rq->lock, cookie);
 	raw_spin_unlock_irqrestore(&rq->lock, flags);
 }
 
@@ -1843,6 +1845,7 @@ bool cpus_share_cache(int this_cpu, int
 static void ttwu_queue(struct task_struct *p, int cpu)
 {
 	struct rq *rq = cpu_rq(cpu);
+	struct pin_cookie cookie;
 
 #if defined(CONFIG_SMP)
 	if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) {
@@ -1853,9 +1856,9 @@ static void ttwu_queue(struct task_struc
 #endif
 
 	raw_spin_lock(&rq->lock);
-	lockdep_pin_lock(&rq->lock);
-	ttwu_do_activate(rq, p, 0);
-	lockdep_unpin_lock(&rq->lock);
+	cookie = lockdep_pin_lock(&rq->lock);
+	ttwu_do_activate(rq, p, 0, cookie);
+	lockdep_unpin_lock(&rq->lock, cookie);
 	raw_spin_unlock(&rq->lock);
 }
 
@@ -2052,7 +2055,7 @@ try_to_wake_up(struct task_struct *p, un
  * ensure that this_rq() is locked, @p is bound to this_rq() and not
  * the current task.
  */
-static void try_to_wake_up_local(struct task_struct *p)
+static void try_to_wake_up_local(struct task_struct *p, struct pin_cookie cookie)
 {
 	struct rq *rq = task_rq(p);
 
@@ -2069,11 +2072,11 @@ static void try_to_wake_up_local(struct
 		 * disabled avoiding further scheduler activity on it and we've
 		 * not yet picked a replacement task.
 		 */
-		lockdep_unpin_lock(&rq->lock);
+		lockdep_unpin_lock(&rq->lock, cookie);
 		raw_spin_unlock(&rq->lock);
 		raw_spin_lock(&p->pi_lock);
 		raw_spin_lock(&rq->lock);
-		lockdep_pin_lock(&rq->lock);
+		lockdep_repin_lock(&rq->lock, cookie);
 	}
 
 	if (!(p->state & TASK_NORMAL))
@@ -2084,7 +2087,7 @@ static void try_to_wake_up_local(struct
 	if (!task_on_rq_queued(p))
 		ttwu_activate(rq, p, ENQUEUE_WAKEUP);
 
-	ttwu_do_wakeup(rq, p, 0);
+	ttwu_do_wakeup(rq, p, 0, cookie);
 	if (schedstat_enabled())
 		ttwu_stat(p, smp_processor_id(), 0);
 out:
@@ -2512,9 +2515,9 @@ void wake_up_new_task(struct task_struct
 		 * Nothing relies on rq->lock after this, so its fine to
 		 * drop it.
 		 */
-		lockdep_unpin_lock(&rq->lock);
+		lockdep_unpin_lock(&rq->lock, rf.cookie);
 		p->sched_class->task_woken(rq, p);
-		lockdep_pin_lock(&rq->lock);
+		lockdep_repin_lock(&rq->lock, rf.cookie);
 	}
 #endif
 	task_rq_unlock(rq, p, &rf);
@@ -2779,7 +2782,7 @@ asmlinkage __visible void schedule_tail(
  */
 static __always_inline struct rq *
 context_switch(struct rq *rq, struct task_struct *prev,
-	       struct task_struct *next)
+	       struct task_struct *next, struct pin_cookie cookie)
 {
 	struct mm_struct *mm, *oldmm;
 
@@ -2811,7 +2814,7 @@ context_switch(struct rq *rq, struct tas
 	 * of the scheduler it's an obvious special-case), so we
 	 * do an early lockdep release here:
 	 */
-	lockdep_unpin_lock(&rq->lock);
+	lockdep_unpin_lock(&rq->lock, cookie);
 	spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
 
 	/* Here we just switch the register state and the stack. */
@@ -3151,7 +3154,7 @@ static inline void schedule_debug(struct
  * Pick up the highest-prio task:
  */
 static inline struct task_struct *
-pick_next_task(struct rq *rq, struct task_struct *prev)
+pick_next_task(struct rq *rq, struct task_struct *prev, struct pin_cookie cookie)
 {
 	const struct sched_class *class = &fair_sched_class;
 	struct task_struct *p;
@@ -3162,20 +3165,20 @@ pick_next_task(struct rq *rq, struct tas
 	 */
 	if (likely(prev->sched_class == class &&
 		   rq->nr_running == rq->cfs.h_nr_running)) {
-		p = fair_sched_class.pick_next_task(rq, prev);
+		p = fair_sched_class.pick_next_task(rq, prev, cookie);
 		if (unlikely(p == RETRY_TASK))
 			goto again;
 
 		/* assumes fair_sched_class->next == idle_sched_class */
 		if (unlikely(!p))
-			p = idle_sched_class.pick_next_task(rq, prev);
+			p = idle_sched_class.pick_next_task(rq, prev, cookie);
 
 		return p;
 	}
 
 again:
 	for_each_class(class) {
-		p = class->pick_next_task(rq, prev);
+		p = class->pick_next_task(rq, prev, cookie);
 		if (p) {
 			if (unlikely(p == RETRY_TASK))
 				goto again;
@@ -3229,6 +3232,7 @@ static void __sched notrace __schedule(b
 {
 	struct task_struct *prev, *next;
 	unsigned long *switch_count;
+	struct pin_cookie cookie;
 	struct rq *rq;
 	int cpu;
 
@@ -3262,7 +3266,7 @@ static void __sched notrace __schedule(b
 	 */
 	smp_mb__before_spinlock();
 	raw_spin_lock(&rq->lock);
-	lockdep_pin_lock(&rq->lock);
+	cookie = lockdep_pin_lock(&rq->lock);
 
 	rq->clock_skip_update <<= 1; /* promote REQ to ACT */
 
@@ -3284,7 +3288,7 @@ static void __sched notrace __schedule(b
 
 				to_wakeup = wq_worker_sleeping(prev);
 				if (to_wakeup)
-					try_to_wake_up_local(to_wakeup);
+					try_to_wake_up_local(to_wakeup, cookie);
 			}
 		}
 		switch_count = &prev->nvcsw;
@@ -3293,7 +3297,7 @@ static void __sched notrace __schedule(b
 	if (task_on_rq_queued(prev))
 		update_rq_clock(rq);
 
-	next = pick_next_task(rq, prev);
+	next = pick_next_task(rq, prev, cookie);
 	clear_tsk_need_resched(prev);
 	clear_preempt_need_resched();
 	rq->clock_skip_update = 0;
@@ -3304,9 +3308,9 @@ static void __sched notrace __schedule(b
 		++*switch_count;
 
 		trace_sched_switch(preempt, prev, next);
-		rq = context_switch(rq, prev, next); /* unlocks the rq */
+		rq = context_switch(rq, prev, next, cookie); /* unlocks the rq */
 	} else {
-		lockdep_unpin_lock(&rq->lock);
+		lockdep_unpin_lock(&rq->lock, cookie);
 		raw_spin_unlock_irq(&rq->lock);
 	}
 
@@ -5389,6 +5393,7 @@ static void migrate_tasks(struct rq *dea
 {
 	struct rq *rq = dead_rq;
 	struct task_struct *next, *stop = rq->stop;
+	struct pin_cookie cookie;
 	int dest_cpu;
 
 	/*
@@ -5420,8 +5425,8 @@ static void migrate_tasks(struct rq *dea
 		/*
 		 * pick_next_task assumes pinned rq->lock.
 		 */
-		lockdep_pin_lock(&rq->lock);
-		next = pick_next_task(rq, &fake_task);
+		cookie = lockdep_pin_lock(&rq->lock);
+		next = pick_next_task(rq, &fake_task, cookie);
 		BUG_ON(!next);
 		next->sched_class->put_prev_task(rq, next);
 
@@ -5434,7 +5439,7 @@ static void migrate_tasks(struct rq *dea
 		 * because !cpu_active at this point, which means load-balance
 		 * will not interfere. Also, stop-machine.
 		 */
-		lockdep_unpin_lock(&rq->lock);
+		lockdep_unpin_lock(&rq->lock, cookie);
 		raw_spin_unlock(&rq->lock);
 		raw_spin_lock(&next->pi_lock);
 		raw_spin_lock(&rq->lock);
--- a/kernel/sched/deadline.c
+++ b/kernel/sched/deadline.c
@@ -670,9 +670,9 @@ static enum hrtimer_restart dl_task_time
 		 * Nothing relies on rq->lock after this, so its safe to drop
 		 * rq->lock.
 		 */
-		lockdep_unpin_lock(&rq->lock);
+		lockdep_unpin_lock(&rq->lock, rf.cookie);
 		push_dl_task(rq);
-		lockdep_pin_lock(&rq->lock);
+		lockdep_repin_lock(&rq->lock, rf.cookie);
 	}
 #endif
 
@@ -1125,7 +1125,8 @@ static struct sched_dl_entity *pick_next
 	return rb_entry(left, struct sched_dl_entity, rb_node);
 }
 
-struct task_struct *pick_next_task_dl(struct rq *rq, struct task_struct *prev)
+struct task_struct *
+pick_next_task_dl(struct rq *rq, struct task_struct *prev, struct pin_cookie cookie)
 {
 	struct sched_dl_entity *dl_se;
 	struct task_struct *p;
@@ -1140,9 +1141,9 @@ struct task_struct *pick_next_task_dl(st
 		 * disabled avoiding further scheduler activity on it and we're
 		 * being very careful to re-start the picking loop.
 		 */
-		lockdep_unpin_lock(&rq->lock);
+		lockdep_unpin_lock(&rq->lock, cookie);
 		pull_dl_task(rq);
-		lockdep_pin_lock(&rq->lock);
+		lockdep_repin_lock(&rq->lock, cookie);
 		/*
 		 * pull_rt_task() can drop (and re-acquire) rq->lock; this
 		 * means a stop task can slip in, in which case we need to
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -5542,7 +5542,7 @@ static void check_preempt_wakeup(struct
 }
 
 static struct task_struct *
-pick_next_task_fair(struct rq *rq, struct task_struct *prev)
+pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct pin_cookie cookie)
 {
 	struct cfs_rq *cfs_rq = &rq->cfs;
 	struct sched_entity *se;
@@ -5655,9 +5655,9 @@ pick_next_task_fair(struct rq *rq, struc
 	 * further scheduler activity on it and we're being very careful to
 	 * re-start the picking loop.
 	 */
-	lockdep_unpin_lock(&rq->lock);
+	lockdep_unpin_lock(&rq->lock, cookie);
 	new_tasks = idle_balance(rq);
-	lockdep_pin_lock(&rq->lock);
+	lockdep_repin_lock(&rq->lock, cookie);
 	/*
 	 * Because idle_balance() releases (and re-acquires) rq->lock, it is
 	 * possible for any higher priority task to appear. In that case we
--- a/kernel/sched/idle_task.c
+++ b/kernel/sched/idle_task.c
@@ -24,7 +24,7 @@ static void check_preempt_curr_idle(stru
 }
 
 static struct task_struct *
-pick_next_task_idle(struct rq *rq, struct task_struct *prev)
+pick_next_task_idle(struct rq *rq, struct task_struct *prev, struct pin_cookie cookie)
 {
 	put_prev_task(rq, prev);
 
--- a/kernel/sched/rt.c
+++ b/kernel/sched/rt.c
@@ -1524,7 +1524,7 @@ static struct task_struct *_pick_next_ta
 }
 
 static struct task_struct *
-pick_next_task_rt(struct rq *rq, struct task_struct *prev)
+pick_next_task_rt(struct rq *rq, struct task_struct *prev, struct pin_cookie cookie)
 {
 	struct task_struct *p;
 	struct rt_rq *rt_rq = &rq->rt;
@@ -1536,9 +1536,9 @@ pick_next_task_rt(struct rq *rq, struct
 		 * disabled avoiding further scheduler activity on it and we're
 		 * being very careful to re-start the picking loop.
 		 */
-		lockdep_unpin_lock(&rq->lock);
+		lockdep_unpin_lock(&rq->lock, cookie);
 		pull_rt_task(rq);
-		lockdep_pin_lock(&rq->lock);
+		lockdep_repin_lock(&rq->lock, cookie);
 		/*
 		 * pull_rt_task() can drop (and re-acquire) rq->lock; this
 		 * means a dl or stop task can slip in, in which case we need
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -1202,7 +1202,8 @@ struct sched_class {
 	 * tasks.
 	 */
 	struct task_struct * (*pick_next_task) (struct rq *rq,
-						struct task_struct *prev);
+						struct task_struct *prev,
+						struct pin_cookie cookie);
 	void (*put_prev_task) (struct rq *rq, struct task_struct *p);
 
 #ifdef CONFIG_SMP
@@ -1453,6 +1454,7 @@ static inline void sched_avg_update(stru
 
 struct rq_flags {
 	unsigned long flags;
+	struct pin_cookie cookie;
 };
 
 struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
@@ -1464,7 +1466,7 @@ struct rq *task_rq_lock(struct task_stru
 static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf)
 	__releases(rq->lock)
 {
-	lockdep_unpin_lock(&rq->lock);
+	lockdep_unpin_lock(&rq->lock, rf->cookie);
 	raw_spin_unlock(&rq->lock);
 }
 
@@ -1473,7 +1475,7 @@ task_rq_unlock(struct rq *rq, struct tas
 	__releases(rq->lock)
 	__releases(p->pi_lock)
 {
-	lockdep_unpin_lock(&rq->lock);
+	lockdep_unpin_lock(&rq->lock, rf->cookie);
 	raw_spin_unlock(&rq->lock);
 	raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
 }
--- a/kernel/sched/stop_task.c
+++ b/kernel/sched/stop_task.c
@@ -24,7 +24,7 @@ check_preempt_curr_stop(struct rq *rq, s
 }
 
 static struct task_struct *
-pick_next_task_stop(struct rq *rq, struct task_struct *prev)
+pick_next_task_stop(struct rq *rq, struct task_struct *prev, struct pin_cookie cookie)
 {
 	struct task_struct *stop = rq->stop;
 

^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2016-04-28 15:45 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-04-28 15:40 [PATCH 0/3] lockde,sched: Better pinning Peter Zijlstra
2016-04-28 15:41 ` [PATCH 1/3] sched: Out-of-line task_rq_lock() Peter Zijlstra
2016-04-28 15:41 ` [PATCH 2/3] sched: Introduce struct rq_flags Peter Zijlstra
2016-04-28 15:41 ` [PATCH 3/3] lockdep,sched: Implement a better lock pinning scheme Peter Zijlstra

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®