forked from ~ljy/RK356X_SDK_RELEASE

blame | history | patch | commit | commitdiff
hc
2024-12-19 9370bb92b2d16684ee45cf24e879c93c509162da 
 kernel/kernel/time/timer.c
....@@ -1,6 +1,5 @@
1
+// SPDX-License-Identifier: GPL-2.0
12 /*
2
- *  linux/kernel/timer.c
3
- *
43  *  Kernel internal timers
54  *
65  *  Copyright (C) 1991, 1992  Linus Torvalds
....@@ -56,6 +55,11 @@
5655 
5756 #define CREATE_TRACE_POINTS
5857 #include <trace/events/timer.h>
58
+#undef CREATE_TRACE_POINTS
59
+#include <trace/hooks/timer.h>
60
+
61
+EXPORT_TRACEPOINT_SYMBOL_GPL(hrtimer_expire_entry);
62
+EXPORT_TRACEPOINT_SYMBOL_GPL(hrtimer_expire_exit);
5963 
6064 __visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
6165 
....@@ -158,7 +162,8 @@
158162 
159163 /*
160164  * The time start value for each level to select the bucket at enqueue
161
- * time.
165
+ * time. We start from the last possible delta of the previous level
166
+ * so that we can later add an extra LVL_GRAN(n) to n (see calc_index()).
162167  */
163168 #define LVL_START(n)	((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT))
164169 
....@@ -198,12 +203,16 @@
198203 struct timer_base {
199204 	raw_spinlock_t		lock;
200205 	struct timer_list	*running_timer;
206
+#ifdef CONFIG_PREEMPT_RT
201207 	spinlock_t		expiry_lock;
208
+	atomic_t		timer_waiters;
209
+#endif
202210 	unsigned long		clk;
203211 	unsigned long		next_expiry;
204212 	unsigned int		cpu;
213
+	bool			next_expiry_recalc;
205214 	bool			is_idle;
206
-	bool			must_forward_clk;
215
+	bool			timers_pending;
207216 	DECLARE_BITMAP(pending_map, WHEEL_SIZE);
208217 	struct hlist_head	vectors[WHEEL_SIZE];
209218 } ____cacheline_aligned;
....@@ -215,7 +224,8 @@
215224 static DEFINE_STATIC_KEY_FALSE(timers_nohz_active);
216225 static DEFINE_MUTEX(timer_keys_mutex);
217226 
218
-static struct swork_event timer_update_swork;
227
+static void timer_update_keys(struct work_struct *work);
228
+static DECLARE_WORK(timer_update_work, timer_update_keys);
219229 
220230 #ifdef CONFIG_SMP
221231 unsigned int sysctl_timer_migration = 1;
....@@ -233,7 +243,7 @@
233243 static inline void timers_update_migration(void) { }
234244 #endif /* !CONFIG_SMP */
235245 
236
-static void timer_update_keys(struct swork_event *event)
246
+static void timer_update_keys(struct work_struct *work)
237247 {
238248 	mutex_lock(&timer_keys_mutex);
239249 	timers_update_migration();
....@@ -243,20 +253,11 @@
243253 
244254 void timers_update_nohz(void)
245255 {
246
-	swork_queue(&timer_update_swork);
256
+	schedule_work(&timer_update_work);
247257 }
248
-
249
-static __init int hrtimer_init_thread(void)
250
-{
251
-	WARN_ON(swork_get());
252
-	INIT_SWORK(&timer_update_swork, timer_update_keys);
253
-	return 0;
254
-}
255
-early_initcall(hrtimer_init_thread);
256258 
257259 int timer_migration_handler(struct ctl_table *table, int write,
258
-			    void __user *buffer, size_t *lenp,
259
-			    loff_t *ppos)
260
+			    void *buffer, size_t *lenp, loff_t *ppos)
260261 {
261262 	int ret;
262263 
....@@ -494,35 +495,49 @@
494495  * Helper function to calculate the array index for a given expiry
495496  * time.
496497  */
497
-static inline unsigned calc_index(unsigned expires, unsigned lvl)
498
+static inline unsigned calc_index(unsigned long expires, unsigned lvl,
499
+				  unsigned long *bucket_expiry)
498500 {
501
+
502
+	/*
503
+	 * The timer wheel has to guarantee that a timer does not fire
504
+	 * early. Early expiry can happen due to:
505
+	 * - Timer is armed at the edge of a tick
506
+	 * - Truncation of the expiry time in the outer wheel levels
507
+	 *
508
+	 * Round up with level granularity to prevent this.
509
+	 */
510
+	trace_android_vh_timer_calc_index(lvl, &expires);
499511 	expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl);
512
+	*bucket_expiry = expires << LVL_SHIFT(lvl);
500513 	return LVL_OFFS(lvl) + (expires & LVL_MASK);
501514 }
502515 
503
-static int calc_wheel_index(unsigned long expires, unsigned long clk)
516
+static int calc_wheel_index(unsigned long expires, unsigned long clk,
517
+			    unsigned long *bucket_expiry)
504518 {
505519 	unsigned long delta = expires - clk;
506520 	unsigned int idx;
507521 
508522 	if (delta < LVL_START(1)) {
509
-		idx = calc_index(expires, 0);
523
+		idx = calc_index(expires, 0, bucket_expiry);
510524 	} else if (delta < LVL_START(2)) {
511
-		idx = calc_index(expires, 1);
525
+		idx = calc_index(expires, 1, bucket_expiry);
512526 	} else if (delta < LVL_START(3)) {
513
-		idx = calc_index(expires, 2);
527
+		idx = calc_index(expires, 2, bucket_expiry);
514528 	} else if (delta < LVL_START(4)) {
515
-		idx = calc_index(expires, 3);
529
+		idx = calc_index(expires, 3, bucket_expiry);
516530 	} else if (delta < LVL_START(5)) {
517
-		idx = calc_index(expires, 4);
531
+		idx = calc_index(expires, 4, bucket_expiry);
518532 	} else if (delta < LVL_START(6)) {
519
-		idx = calc_index(expires, 5);
533
+		idx = calc_index(expires, 5, bucket_expiry);
520534 	} else if (delta < LVL_START(7)) {
521
-		idx = calc_index(expires, 6);
535
+		idx = calc_index(expires, 6, bucket_expiry);
522536 	} else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
523
-		idx = calc_index(expires, 7);
537
+		idx = calc_index(expires, 7, bucket_expiry);
524538 	} else if ((long) delta < 0) {
525539 		idx = clk & LVL_MASK;
540
+		*bucket_expiry = clk;
526541 	} else {
527542 		/*
528543 		 * Force expire obscene large timeouts to expire at the
....@@ -531,30 +546,9 @@
531546 		if (delta >= WHEEL_TIMEOUT_CUTOFF)
532547 			expires = clk + WHEEL_TIMEOUT_MAX;
533548 
534
-		idx = calc_index(expires, LVL_DEPTH - 1);
549
+		idx = calc_index(expires, LVL_DEPTH - 1, bucket_expiry);
535550 	}
536551 	return idx;
537
-}
538
-
539
-/*
540
- * Enqueue the timer into the hash bucket, mark it pending in
541
- * the bitmap and store the index in the timer flags.
542
- */
543
-static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
544
-			  unsigned int idx)
545
-{
546
-	hlist_add_head(&timer->entry, base->vectors + idx);
547
-	__set_bit(idx, base->pending_map);
548
-	timer_set_idx(timer, idx);
549
-}
550
-
551
-static void
552
-__internal_add_timer(struct timer_base *base, struct timer_list *timer)
553
-{
554
-	unsigned int idx;
555
-
556
-	idx = calc_wheel_index(timer->expires, base->clk);
557
-	enqueue_timer(base, timer, idx);
558552 }
559553 
560554 static void
....@@ -578,39 +572,54 @@
578572 	 * timer is not deferrable. If the other CPU is on the way to idle
579573 	 * then it can't set base->is_idle as we hold the base lock:
580574 	 */
581
-	if (!base->is_idle)
582
-		return;
583
-
584
-	/* Check whether this is the new first expiring timer: */
585
-	if (time_after_eq(timer->expires, base->next_expiry))
586
-		return;
587
-
588
-	/*
589
-	 * Set the next expiry time and kick the CPU so it can reevaluate the
590
-	 * wheel:
591
-	 */
592
-	if (time_before(timer->expires, base->clk)) {
593
-		/*
594
-		 * Prevent from forward_timer_base() moving the base->clk
595
-		 * backward
596
-		 */
597
-		base->next_expiry = base->clk;
598
-	} else {
599
-		base->next_expiry = timer->expires;
600
-	}
601
-	wake_up_nohz_cpu(base->cpu);
575
+	if (base->is_idle)
576
+		wake_up_nohz_cpu(base->cpu);
602577 }
603578 
604
-static void
605
-internal_add_timer(struct timer_base *base, struct timer_list *timer)
579
+/*
580
+ * Enqueue the timer into the hash bucket, mark it pending in
581
+ * the bitmap, store the index in the timer flags then wake up
582
+ * the target CPU if needed.
583
+ */
584
+static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
585
+			  unsigned int idx, unsigned long bucket_expiry)
606586 {
607
-	__internal_add_timer(base, timer);
608
-	trigger_dyntick_cpu(base, timer);
587
+
588
+	hlist_add_head(&timer->entry, base->vectors + idx);
589
+	__set_bit(idx, base->pending_map);
590
+	timer_set_idx(timer, idx);
591
+
592
+	trace_timer_start(timer, timer->expires, timer->flags);
593
+
594
+	/*
595
+	 * Check whether this is the new first expiring timer. The
596
+	 * effective expiry time of the timer is required here
597
+	 * (bucket_expiry) instead of timer->expires.
598
+	 */
599
+	if (time_before(bucket_expiry, base->next_expiry)) {
600
+		/*
601
+		 * Set the next expiry time and kick the CPU so it
602
+		 * can reevaluate the wheel:
603
+		 */
604
+		base->next_expiry = bucket_expiry;
605
+		base->timers_pending = true;
606
+		base->next_expiry_recalc = false;
607
+		trigger_dyntick_cpu(base, timer);
608
+	}
609
+}
610
+
611
+static void internal_add_timer(struct timer_base *base, struct timer_list *timer)
612
+{
613
+	unsigned long bucket_expiry;
614
+	unsigned int idx;
615
+
616
+	idx = calc_wheel_index(timer->expires, base->clk, &bucket_expiry);
617
+	enqueue_timer(base, timer, idx, bucket_expiry);
609618 }
610619 
611620 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
612621 
613
-static struct debug_obj_descr timer_debug_descr;
622
+static const struct debug_obj_descr timer_debug_descr;
614623 
615624 static void *timer_debug_hint(void *addr)
616625 {
....@@ -665,7 +674,7 @@
665674 
666675 	case ODEBUG_STATE_ACTIVE:
667676 		WARN_ON(1);
668
-
677
+		fallthrough;
669678 	default:
670679 		return false;
671680 	}
....@@ -706,7 +715,7 @@
706715 	}
707716 }
708717 
709
-static struct debug_obj_descr timer_debug_descr = {
718
+static const struct debug_obj_descr timer_debug_descr = {
710719 	.name			= "timer_list",
711720 	.debug_hint		= timer_debug_hint,
712721 	.is_static_object	= timer_is_static_object,
....@@ -729,11 +738,6 @@
729738 static inline void debug_timer_deactivate(struct timer_list *timer)
730739 {
731740 	debug_object_deactivate(timer, &timer_debug_descr);
732
-}
733
-
734
-static inline void debug_timer_free(struct timer_list *timer)
735
-{
736
-	debug_object_free(timer, &timer_debug_descr);
737741 }
738742 
739743 static inline void debug_timer_assert_init(struct timer_list *timer)
....@@ -775,13 +779,6 @@
775779 	trace_timer_init(timer);
776780 }
777781 
778
-static inline void
779
-debug_activate(struct timer_list *timer, unsigned long expires)
780
-{
781
-	debug_timer_activate(timer);
782
-	trace_timer_start(timer, expires, timer->flags);
783
-}
784
-
785782 static inline void debug_deactivate(struct timer_list *timer)
786783 {
787784 	debug_timer_deactivate(timer);
....@@ -800,6 +797,8 @@
800797 {
801798 	timer->entry.pprev = NULL;
802799 	timer->function = func;
800
+	if (WARN_ON_ONCE(flags & ~TIMER_INIT_FLAGS))
801
+		flags &= TIMER_INIT_FLAGS;
803802 	timer->flags = flags | raw_smp_processor_id();
804803 	lockdep_init_map(&timer->lockdep_map, name, key, 0);
805804 }
....@@ -845,8 +844,10 @@
845844 	if (!timer_pending(timer))
846845 		return 0;
847846 
848
-	if (hlist_is_singular_node(&timer->entry, base->vectors + idx))
847
+	if (hlist_is_singular_node(&timer->entry, base->vectors + idx)) {
849848 		__clear_bit(idx, base->pending_map);
849
+		base->next_expiry_recalc = true;
850
+	}
850851 
851852 	detach_timer(timer, clear_pending);
852853 	return 1;
....@@ -896,20 +897,14 @@
896897 
897898 static inline void forward_timer_base(struct timer_base *base)
898899 {
899
-#ifdef CONFIG_NO_HZ_COMMON
900
-	unsigned long jnow;
900
+	unsigned long jnow = READ_ONCE(jiffies);
901901 
902902 	/*
903
-	 * We only forward the base when we are idle or have just come out of
904
-	 * idle (must_forward_clk logic), and have a delta between base clock
905
-	 * and jiffies. In the common case, run_timers will take care of it.
903
+	 * No need to forward if we are close enough below jiffies.
904
+	 * Also while executing timers, base->clk is 1 offset ahead
905
+	 * of jiffies to avoid endless requeuing to current jffies.
906906 	 */
907
-	if (likely(!base->must_forward_clk))
908
-		return;
909
-
910
-	jnow = READ_ONCE(jiffies);
911
-	base->must_forward_clk = base->is_idle;
912
-	if ((long)(jnow - base->clk) < 2)
907
+	if ((long)(jnow - base->clk) < 1)
913908 		return;
914909 
915910 	/*
....@@ -923,7 +918,6 @@
923918 			return;
924919 		base->clk = base->next_expiry;
925920 	}
926
-#endif
927921 }
928922 
929923 
....@@ -966,13 +960,14 @@
966960 
967961 #define MOD_TIMER_PENDING_ONLY		0x01
968962 #define MOD_TIMER_REDUCE		0x02
963
+#define MOD_TIMER_NOTPENDING		0x04
969964 
970965 static inline int
971966 __mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)
972967 {
968
+	unsigned long clk = 0, flags, bucket_expiry;
973969 	struct timer_base *base, *new_base;
974970 	unsigned int idx = UINT_MAX;
975
-	unsigned long clk = 0, flags;
976971 	int ret = 0;
977972 
978973 	BUG_ON(!timer->function);
....@@ -982,7 +977,7 @@
982977 	 * the timer is re-modified to have the same timeout or ends up in the
983978 	 * same array bucket then just return:
984979 	 */
985
-	if (timer_pending(timer)) {
980
+	if (!(options & MOD_TIMER_NOTPENDING) && timer_pending(timer)) {
986981 		/*
987982 		 * The downside of this optimization is that it can result in
988983 		 * larger granularity than you would get from adding a new
....@@ -1011,7 +1006,7 @@
10111006 		}
10121007 
10131008 		clk = base->clk;
1014
-		idx = calc_wheel_index(expires, clk);
1009
+		idx = calc_wheel_index(expires, clk, &bucket_expiry);
10151010 
10161011 		/*
10171012 		 * Retrieve and compare the array index of the pending
....@@ -1058,22 +1053,19 @@
10581053 		}
10591054 	}
10601055 
1061
-	debug_activate(timer, expires);
1056
+	debug_timer_activate(timer);
10621057 
10631058 	timer->expires = expires;
10641059 	/*
10651060 	 * If 'idx' was calculated above and the base time did not advance
10661061 	 * between calculating 'idx' and possibly switching the base, only
1067
-	 * enqueue_timer() and trigger_dyntick_cpu() is required. Otherwise
1068
-	 * we need to (re)calculate the wheel index via
1069
-	 * internal_add_timer().
1062
+	 * enqueue_timer() is required. Otherwise we need to (re)calculate
1063
+	 * the wheel index via internal_add_timer().
10701064 	 */
1071
-	if (idx != UINT_MAX && clk == base->clk) {
1072
-		enqueue_timer(base, timer, idx);
1073
-		trigger_dyntick_cpu(base, timer);
1074
-	} else {
1065
+	if (idx != UINT_MAX && clk == base->clk)
1066
+		enqueue_timer(base, timer, idx, bucket_expiry);
1067
+	else
10751068 		internal_add_timer(base, timer);
1076
-	}
10771069 
10781070 out_unlock:
10791071 	raw_spin_unlock_irqrestore(&base->lock, flags);
....@@ -1155,7 +1147,7 @@
11551147 void add_timer(struct timer_list *timer)
11561148 {
11571149 	BUG_ON(timer_pending(timer));
1158
-	mod_timer(timer, timer->expires);
1150
+	__mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING);
11591151 }
11601152 EXPORT_SYMBOL(add_timer);
11611153 
....@@ -1192,7 +1184,7 @@
11921184 	}
11931185 	forward_timer_base(base);
11941186 
1195
-	debug_activate(timer, timer->expires);
1187
+	debug_timer_activate(timer);
11961188 	internal_add_timer(base, timer);
11971189 	raw_spin_unlock_irqrestore(&base->lock, flags);
11981190 }
....@@ -1227,25 +1219,6 @@
12271219 }
12281220 EXPORT_SYMBOL(del_timer);
12291221 
1230
-static int __try_to_del_timer_sync(struct timer_list *timer,
1231
-				   struct timer_base **basep)
1232
-{
1233
-	struct timer_base *base;
1234
-	unsigned long flags;
1235
-	int ret = -1;
1236
-
1237
-	debug_assert_init(timer);
1238
-
1239
-	*basep = base = lock_timer_base(timer, &flags);
1240
-
1241
-	if (base->running_timer != timer)
1242
-		ret = detach_if_pending(timer, base, true);
1243
-
1244
-	raw_spin_unlock_irqrestore(&base->lock, flags);
1245
-
1246
-	return ret;
1247
-}
1248
-
12491222 /**
12501223  * try_to_del_timer_sync - Try to deactivate a timer
12511224  * @timer: timer to delete
....@@ -1256,31 +1229,96 @@
12561229 int try_to_del_timer_sync(struct timer_list *timer)
12571230 {
12581231 	struct timer_base *base;
1232
+	unsigned long flags;
1233
+	int ret = -1;
12591234 
1260
-	return __try_to_del_timer_sync(timer, &base);
1235
+	debug_assert_init(timer);
1236
+
1237
+	base = lock_timer_base(timer, &flags);
1238
+
1239
+	if (base->running_timer != timer)
1240
+		ret = detach_if_pending(timer, base, true);
1241
+
1242
+	raw_spin_unlock_irqrestore(&base->lock, flags);
1243
+
1244
+	return ret;
12611245 }
12621246 EXPORT_SYMBOL(try_to_del_timer_sync);
12631247 
1264
-#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT_FULL)
1265
-static int __del_timer_sync(struct timer_list *timer)
1248
+#ifdef CONFIG_PREEMPT_RT
1249
+static __init void timer_base_init_expiry_lock(struct timer_base *base)
12661250 {
1267
-	struct timer_base *base;
1268
-	int ret;
1251
+	spin_lock_init(&base->expiry_lock);
1252
+}
12691253 
1270
-	for (;;) {
1271
-		ret = __try_to_del_timer_sync(timer, &base);
1272
-		if (ret >= 0)
1273
-			return ret;
1254
+static inline void timer_base_lock_expiry(struct timer_base *base)
1255
+{
1256
+	spin_lock(&base->expiry_lock);
1257
+}
12741258 
1275
-		/*
1276
-		 * When accessing the lock, timers of base are no longer expired
1277
-		 * and so timer is no longer running.
1278
-		 */
1279
-		spin_lock(&base->expiry_lock);
1259
+static inline void timer_base_unlock_expiry(struct timer_base *base)
1260
+{
1261
+	spin_unlock(&base->expiry_lock);
1262
+}
1263
+
1264
+/*
1265
+ * The counterpart to del_timer_wait_running().
1266
+ *
1267
+ * If there is a waiter for base->expiry_lock, then it was waiting for the
1268
+ * timer callback to finish. Drop expiry_lock and reaquire it. That allows
1269
+ * the waiter to acquire the lock and make progress.
1270
+ */
1271
+static void timer_sync_wait_running(struct timer_base *base)
1272
+{
1273
+	if (atomic_read(&base->timer_waiters)) {
1274
+		raw_spin_unlock_irq(&base->lock);
12801275 		spin_unlock(&base->expiry_lock);
1276
+		spin_lock(&base->expiry_lock);
1277
+		raw_spin_lock_irq(&base->lock);
12811278 	}
12821279 }
12831280 
1281
+/*
1282
+ * This function is called on PREEMPT_RT kernels when the fast path
1283
+ * deletion of a timer failed because the timer callback function was
1284
+ * running.
1285
+ *
1286
+ * This prevents priority inversion, if the softirq thread on a remote CPU
1287
+ * got preempted, and it prevents a life lock when the task which tries to
1288
+ * delete a timer preempted the softirq thread running the timer callback
1289
+ * function.
1290
+ */
1291
+static void del_timer_wait_running(struct timer_list *timer)
1292
+{
1293
+	u32 tf;
1294
+
1295
+	tf = READ_ONCE(timer->flags);
1296
+	if (!(tf & TIMER_MIGRATING)) {
1297
+		struct timer_base *base = get_timer_base(tf);
1298
+
1299
+		/*
1300
+		 * Mark the base as contended and grab the expiry lock,
1301
+		 * which is held by the softirq across the timer
1302
+		 * callback. Drop the lock immediately so the softirq can
1303
+		 * expire the next timer. In theory the timer could already
1304
+		 * be running again, but that's more than unlikely and just
1305
+		 * causes another wait loop.
1306
+		 */
1307
+		atomic_inc(&base->timer_waiters);
1308
+		spin_lock_bh(&base->expiry_lock);
1309
+		atomic_dec(&base->timer_waiters);
1310
+		spin_unlock_bh(&base->expiry_lock);
1311
+	}
1312
+}
1313
+#else
1314
+static inline void timer_base_init_expiry_lock(struct timer_base *base) { }
1315
+static inline void timer_base_lock_expiry(struct timer_base *base) { }
1316
+static inline void timer_base_unlock_expiry(struct timer_base *base) { }
1317
+static inline void timer_sync_wait_running(struct timer_base *base) { }
1318
+static inline void del_timer_wait_running(struct timer_list *timer) { }
1319
+#endif
1320
+
1321
+#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT)
12841322 /**
12851323  * del_timer_sync - deactivate a timer and wait for the handler to finish.
12861324  * @timer: the timer to be deactivated
....@@ -1319,6 +1357,8 @@
13191357  */
13201358 int del_timer_sync(struct timer_list *timer)
13211359 {
1360
+	int ret;
1361
+
13221362 #ifdef CONFIG_LOCKDEP
13231363 	unsigned long flags;
13241364 
....@@ -1337,12 +1377,23 @@
13371377 	 */
13381378 	WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
13391379 
1340
-	return __del_timer_sync(timer);
1380
+	do {
1381
+		ret = try_to_del_timer_sync(timer);
1382
+
1383
+		if (unlikely(ret < 0)) {
1384
+			del_timer_wait_running(timer);
1385
+			cpu_relax();
1386
+		}
1387
+	} while (ret < 0);
1388
+
1389
+	return ret;
13411390 }
13421391 EXPORT_SYMBOL(del_timer_sync);
13431392 #endif
13441393 
1345
-static void call_timer_fn(struct timer_list *timer, void (*fn)(struct timer_list *))
1394
+static void call_timer_fn(struct timer_list *timer,
1395
+			  void (*fn)(struct timer_list *),
1396
+			  unsigned long baseclk)
13461397 {
13471398 	int count = preempt_count();
13481399 
....@@ -1365,14 +1416,14 @@
13651416 	 */
13661417 	lock_map_acquire(&lockdep_map);
13671418 
1368
-	trace_timer_expire_entry(timer);
1419
+	trace_timer_expire_entry(timer, baseclk);
13691420 	fn(timer);
13701421 	trace_timer_expire_exit(timer);
13711422 
13721423 	lock_map_release(&lockdep_map);
13731424 
13741425 	if (count != preempt_count()) {
1375
-		WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1426
+		WARN_ONCE(1, "timer: %pS preempt leak: %08x -> %08x\n",
13761427 			  fn, count, preempt_count());
13771428 		/*
13781429 		 * Restore the preempt count. That gives us a decent
....@@ -1386,6 +1437,13 @@
13861437 
13871438 static void expire_timers(struct timer_base *base, struct hlist_head *head)
13881439 {
1440
+	/*
1441
+	 * This value is required only for tracing. base->clk was
1442
+	 * incremented directly before expire_timers was called. But expiry
1443
+	 * is related to the old base->clk value.
1444
+	 */
1445
+	unsigned long baseclk = base->clk - 1;
1446
+
13891447 	while (!hlist_empty(head)) {
13901448 		struct timer_list *timer;
13911449 		void (*fn)(struct timer_list *);
....@@ -1399,26 +1457,23 @@
13991457 
14001458 		if (timer->flags & TIMER_IRQSAFE) {
14011459 			raw_spin_unlock(&base->lock);
1402
-			call_timer_fn(timer, fn);
1403
-			base->running_timer = NULL;
1404
-			spin_unlock(&base->expiry_lock);
1405
-			spin_lock(&base->expiry_lock);
1460
+			call_timer_fn(timer, fn, baseclk);
14061461 			raw_spin_lock(&base->lock);
1462
+			base->running_timer = NULL;
14071463 		} else {
14081464 			raw_spin_unlock_irq(&base->lock);
1409
-			call_timer_fn(timer, fn);
1410
-			base->running_timer = NULL;
1411
-			spin_unlock(&base->expiry_lock);
1412
-			spin_lock(&base->expiry_lock);
1465
+			call_timer_fn(timer, fn, baseclk);
14131466 			raw_spin_lock_irq(&base->lock);
1467
+			base->running_timer = NULL;
1468
+			timer_sync_wait_running(base);
14141469 		}
14151470 	}
14161471 }
14171472 
1418
-static int __collect_expired_timers(struct timer_base *base,
1419
-				    struct hlist_head *heads)
1473
+static int collect_expired_timers(struct timer_base *base,
1474
+				  struct hlist_head *heads)
14201475 {
1421
-	unsigned long clk = base->clk;
1476
+	unsigned long clk = base->clk = base->next_expiry;
14221477 	struct hlist_head *vec;
14231478 	int i, levels = 0;
14241479 	unsigned int idx;
....@@ -1440,7 +1495,6 @@
14401495 	return levels;
14411496 }
14421497 
1443
-#ifdef CONFIG_NO_HZ_COMMON
14441498 /*
14451499  * Find the next pending bucket of a level. Search from level start (@offset)
14461500  * + @clk upwards and if nothing there, search from start of the level
....@@ -1473,6 +1527,7 @@
14731527 	clk = base->clk;
14741528 	for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
14751529 		int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
1530
+		unsigned long lvl_clk = clk & LVL_CLK_MASK;
14761531 
14771532 		if (pos >= 0) {
14781533 			unsigned long tmp = clk + (unsigned long) pos;
....@@ -1480,6 +1535,13 @@
14801535 			tmp <<= LVL_SHIFT(lvl);
14811536 			if (time_before(tmp, next))
14821537 				next = tmp;
1538
+
1539
+			/*
1540
+			 * If the next expiration happens before we reach
1541
+			 * the next level, no need to check further.
1542
+			 */
1543
+			if (pos <= ((LVL_CLK_DIV - lvl_clk) & LVL_CLK_MASK))
1544
+				break;
14831545 		}
14841546 		/*
14851547 		 * Clock for the next level. If the current level clock lower
....@@ -1517,13 +1579,18 @@
15171579 		 * So the simple check whether the lower bits of the current
15181580 		 * level are 0 or not is sufficient for all cases.
15191581 		 */
1520
-		adj = clk & LVL_CLK_MASK ? 1 : 0;
1582
+		adj = lvl_clk ? 1 : 0;
15211583 		clk >>= LVL_CLK_SHIFT;
15221584 		clk += adj;
15231585 	}
1586
+
1587
+	base->next_expiry_recalc = false;
1588
+	base->timers_pending = !(next == base->clk + NEXT_TIMER_MAX_DELTA);
1589
+
15241590 	return next;
15251591 }
15261592 
1593
+#ifdef CONFIG_NO_HZ_COMMON
15271594 /*
15281595  * Check, if the next hrtimer event is before the next timer wheel
15291596  * event:
....@@ -1570,7 +1637,6 @@
15701637 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
15711638 	u64 expires = KTIME_MAX;
15721639 	unsigned long nextevt;
1573
-	bool is_max_delta;
15741640 
15751641 	/*
15761642 	 * Pretend that there is no timer pending if the cpu is offline.
....@@ -1580,9 +1646,10 @@
15801646 		return expires;
15811647 
15821648 	raw_spin_lock(&base->lock);
1583
-	nextevt = __next_timer_interrupt(base);
1584
-	is_max_delta = (nextevt == base->clk + NEXT_TIMER_MAX_DELTA);
1585
-	base->next_expiry = nextevt;
1649
+	if (base->next_expiry_recalc)
1650
+		base->next_expiry = __next_timer_interrupt(base);
1651
+	nextevt = base->next_expiry;
1652
+
15861653 	/*
15871654 	 * We have a fresh next event. Check whether we can forward the
15881655 	 * base. We can only do that when @basej is past base->clk
....@@ -1599,7 +1666,7 @@
15991666 		expires = basem;
16001667 		base->is_idle = false;
16011668 	} else {
1602
-		if (!is_max_delta)
1669
+		if (base->timers_pending)
16031670 			expires = basem + (u64)(nextevt - basej) * TICK_NSEC;
16041671 		/*
16051672 		 * If we expect to sleep more than a tick, mark the base idle.
....@@ -1608,10 +1675,8 @@
16081675 		 * logic is only maintained for the BASE_STD base, deferrable
16091676 		 * timers may still see large granularity skew (by design).
16101677 		 */
1611
-		if ((expires - basem) > TICK_NSEC) {
1612
-			base->must_forward_clk = true;
1678
+		if ((expires - basem) > TICK_NSEC)
16131679 			base->is_idle = true;
1614
-		}
16151680 	}
16161681 	raw_spin_unlock(&base->lock);
16171682 
....@@ -1635,42 +1700,6 @@
16351700 	 */
16361701 	base->is_idle = false;
16371702 }
1638
-
1639
-static int collect_expired_timers(struct timer_base *base,
1640
-				  struct hlist_head *heads)
1641
-{
1642
-	unsigned long now = READ_ONCE(jiffies);
1643
-
1644
-	/*
1645
-	 * NOHZ optimization. After a long idle sleep we need to forward the
1646
-	 * base to current jiffies. Avoid a loop by searching the bitfield for
1647
-	 * the next expiring timer.
1648
-	 */
1649
-	if ((long)(now - base->clk) > 2) {
1650
-		unsigned long next = __next_timer_interrupt(base);
1651
-
1652
-		/*
1653
-		 * If the next timer is ahead of time forward to current
1654
-		 * jiffies, otherwise forward to the next expiry time:
1655
-		 */
1656
-		if (time_after(next, now)) {
1657
-			/*
1658
-			 * The call site will increment base->clk and then
1659
-			 * terminate the expiry loop immediately.
1660
-			 */
1661
-			base->clk = now;
1662
-			return 0;
1663
-		}
1664
-		base->clk = next;
1665
-	}
1666
-	return __collect_expired_timers(base, heads);
1667
-}
1668
-#else
1669
-static inline int collect_expired_timers(struct timer_base *base,
1670
-					 struct hlist_head *heads)
1671
-{
1672
-	return __collect_expired_timers(base, heads);
1673
-}
16741703 #endif
16751704 
16761705 /*
....@@ -1681,17 +1710,19 @@
16811710 {
16821711 	struct task_struct *p = current;
16831712 
1713
+	PRANDOM_ADD_NOISE(jiffies, user_tick, p, 0);
1714
+
16841715 	/* Note: this timer irq context must be accounted for as well. */
16851716 	account_process_tick(p, user_tick);
16861717 	run_local_timers();
1687
-	rcu_check_callbacks(user_tick);
1718
+	rcu_sched_clock_irq(user_tick);
16881719 #ifdef CONFIG_IRQ_WORK
16891720 	if (in_irq())
16901721 		irq_work_tick();
16911722 #endif
16921723 	scheduler_tick();
16931724 	if (IS_ENABLED(CONFIG_POSIX_TIMERS))
1694
-		run_posix_cpu_timers(p);
1725
+		run_posix_cpu_timers();
16951726 }
16961727 
16971728 /**
....@@ -1703,38 +1734,32 @@
17031734 	struct hlist_head heads[LVL_DEPTH];
17041735 	int levels;
17051736 
1706
-	if (!time_after_eq(jiffies, base->clk))
1737
+	if (time_before(jiffies, base->next_expiry))
17071738 		return;
17081739 
1709
-	spin_lock(&base->expiry_lock);
1740
+	timer_base_lock_expiry(base);
17101741 	raw_spin_lock_irq(&base->lock);
17111742 
1712
-	/*
1713
-	 * timer_base::must_forward_clk must be cleared before running
1714
-	 * timers so that any timer functions that call mod_timer() will
1715
-	 * not try to forward the base. Idle tracking / clock forwarding
1716
-	 * logic is only used with BASE_STD timers.
1717
-	 *
1718
-	 * The must_forward_clk flag is cleared unconditionally also for
1719
-	 * the deferrable base. The deferrable base is not affected by idle
1720
-	 * tracking and never forwarded, so clearing the flag is a NOOP.
1721
-	 *
1722
-	 * The fact that the deferrable base is never forwarded can cause
1723
-	 * large variations in granularity for deferrable timers, but they
1724
-	 * can be deferred for long periods due to idle anyway.
1725
-	 */
1726
-	base->must_forward_clk = false;
1727
-
1728
-	while (time_after_eq(jiffies, base->clk)) {
1729
-
1743
+	while (time_after_eq(jiffies, base->clk) &&
1744
+	       time_after_eq(jiffies, base->next_expiry)) {
17301745 		levels = collect_expired_timers(base, heads);
1746
+		/*
1747
+		 * The two possible reasons for not finding any expired
1748
+		 * timer at this clk are that all matching timers have been
1749
+		 * dequeued or no timer has been queued since
1750
+		 * base::next_expiry was set to base::clk +
1751
+		 * NEXT_TIMER_MAX_DELTA.
1752
+		 */
1753
+		WARN_ON_ONCE(!levels && !base->next_expiry_recalc
1754
+			     && base->timers_pending);
17311755 		base->clk++;
1756
+		base->next_expiry = __next_timer_interrupt(base);
17321757 
17331758 		while (levels--)
17341759 			expire_timers(base, heads + levels);
17351760 	}
17361761 	raw_spin_unlock_irq(&base->lock);
1737
-	spin_unlock(&base->expiry_lock);
1762
+	timer_base_unlock_expiry(base);
17381763 }
17391764 
17401765 /*
....@@ -1743,8 +1768,6 @@
17431768 static __latent_entropy void run_timer_softirq(struct softirq_action *h)
17441769 {
17451770 	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1746
-
1747
-	irq_work_tick_soft();
17481771 
17491772 	__run_timers(base);
17501773 	if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
....@@ -1760,12 +1783,12 @@
17601783 
17611784 	hrtimer_run_queues();
17621785 	/* Raise the softirq only if required. */
1763
-	if (time_before(jiffies, base->clk)) {
1786
+	if (time_before(jiffies, base->next_expiry)) {
17641787 		if (!IS_ENABLED(CONFIG_NO_HZ_COMMON))
17651788 			return;
17661789 		/* CPU is awake, so check the deferrable base. */
17671790 		base++;
1768
-		if (time_before(jiffies, base->clk))
1791
+		if (time_before(jiffies, base->next_expiry))
17691792 			return;
17701793 	}
17711794 	raise_softirq(TIMER_SOFTIRQ);
....@@ -1791,21 +1814,23 @@
17911814  * schedule_timeout - sleep until timeout
17921815  * @timeout: timeout value in jiffies
17931816  *
1794
- * Make the current task sleep until @timeout jiffies have
1795
- * elapsed. The routine will return immediately unless
1796
- * the current task state has been set (see set_current_state()).
1817
+ * Make the current task sleep until @timeout jiffies have elapsed.
1818
+ * The function behavior depends on the current task state
1819
+ * (see also set_current_state() description):
17971820  *
1798
- * You can set the task state as follows -
1821
+ * %TASK_RUNNING - the scheduler is called, but the task does not sleep
1822
+ * at all. That happens because sched_submit_work() does nothing for
1823
+ * tasks in %TASK_RUNNING state.
17991824  *
18001825  * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
18011826  * pass before the routine returns unless the current task is explicitly
1802
- * woken up, (e.g. by wake_up_process())".
1827
+ * woken up, (e.g. by wake_up_process()).
18031828  *
18041829  * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
18051830  * delivered to the current task or the current task is explicitly woken
18061831  * up.
18071832  *
1808
- * The current task state is guaranteed to be TASK_RUNNING when this
1833
+ * The current task state is guaranteed to be %TASK_RUNNING when this
18091834  * routine returns.
18101835  *
18111836  * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
....@@ -1813,7 +1838,7 @@
18131838  * value will be %MAX_SCHEDULE_TIMEOUT.
18141839  *
18151840  * Returns 0 when the timer has expired otherwise the remaining time in
1816
- * jiffies will be returned.  In all cases the return value is guaranteed
1841
+ * jiffies will be returned. In all cases the return value is guaranteed
18171842  * to be non-negative.
18181843  */
18191844 signed long __sched schedule_timeout(signed long timeout)
....@@ -1854,7 +1879,7 @@
18541879 
18551880 	timer.task = current;
18561881 	timer_setup_on_stack(&timer.timer, process_timeout, 0);
1857
-	__mod_timer(&timer.timer, expire, 0);
1882
+	__mod_timer(&timer.timer, expire, MOD_TIMER_NOTPENDING);
18581883 	schedule();
18591884 	del_singleshot_timer_sync(&timer.timer);
18601885 
....@@ -1927,8 +1952,8 @@
19271952 		base = per_cpu_ptr(&timer_bases[b], cpu);
19281953 		base->clk = jiffies;
19291954 		base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
1955
+		base->timers_pending = false;
19301956 		base->is_idle = false;
1931
-		base->must_forward_clk = true;
19321957 	}
19331958 	return 0;
19341959 }
....@@ -1981,7 +2006,8 @@
19812006 		base->cpu = cpu;
19822007 		raw_spin_lock_init(&base->lock);
19832008 		base->clk = jiffies;
1984
-		spin_lock_init(&base->expiry_lock);
2009
+		base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
2010
+		timer_base_init_expiry_lock(base);
19852011 	}
19862012 }
19872013 
....@@ -1996,6 +2022,7 @@
19962022 void __init init_timers(void)
19972023 {
19982024 	init_timer_cpus();
2025
+	posix_cputimers_init_work();
19992026 	open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
20002027 }
20012028 
....@@ -2029,6 +2056,32 @@
20292056 EXPORT_SYMBOL(msleep_interruptible);
20302057 
20312058 /**
2059
+ * usleep_range_state - Sleep for an approximate time in a given state
2060
+ * @min:	Minimum time in usecs to sleep
2061
+ * @max:	Maximum time in usecs to sleep
2062
+ * @state:	State of the current task that will be while sleeping
2063
+ *
2064
+ * In non-atomic context where the exact wakeup time is flexible, use
2065
+ * usleep_range_state() instead of udelay().  The sleep improves responsiveness
2066
+ * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
2067
+ * power usage by allowing hrtimers to take advantage of an already-
2068
+ * scheduled interrupt instead of scheduling a new one just for this sleep.
2069
+ */
2070
+void __sched usleep_range_state(unsigned long min, unsigned long max,
2071
+				unsigned int state)
2072
+{
2073
+	ktime_t exp = ktime_add_us(ktime_get(), min);
2074
+	u64 delta = (u64)(max - min) * NSEC_PER_USEC;
2075
+
2076
+	for (;;) {
2077
+		__set_current_state(state);
2078
+		/* Do not return before the requested sleep time has elapsed */
2079
+		if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
2080
+			break;
2081
+	}
2082
+}
2083
+
2084
+/**
20322085  * usleep_range - Sleep for an approximate time
20332086  * @min: Minimum time in usecs to sleep
20342087  * @max: Maximum time in usecs to sleep
....@@ -2041,14 +2094,6 @@
20412094  */
20422095 void __sched usleep_range(unsigned long min, unsigned long max)
20432096 {
2044
-	ktime_t exp = ktime_add_us(ktime_get(), min);
2045
-	u64 delta = (u64)(max - min) * NSEC_PER_USEC;
2046
-
2047
-	for (;;) {
2048
-		__set_current_state(TASK_UNINTERRUPTIBLE);
2049
-		/* Do not return before the requested sleep time has elapsed */
2050
-		if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
2051
-			break;
2052
-	}
2097
+	usleep_range_state(min, max, TASK_UNINTERRUPTIBLE);
20532098 }
20542099 EXPORT_SYMBOL(usleep_range);