hc
2024-05-10 cde9070d9970eef1f7ec2360586c802a16230ad8
kernel/kernel/rcu/update.c
....@@ -1,26 +1,13 @@
1
+// SPDX-License-Identifier: GPL-2.0+
12 /*
23 * Read-Copy Update mechanism for mutual exclusion
3
- *
4
- * This program is free software; you can redistribute it and/or modify
5
- * it under the terms of the GNU General Public License as published by
6
- * the Free Software Foundation; either version 2 of the License, or
7
- * (at your option) any later version.
8
- *
9
- * This program is distributed in the hope that it will be useful,
10
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
11
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12
- * GNU General Public License for more details.
13
- *
14
- * You should have received a copy of the GNU General Public License
15
- * along with this program; if not, you can access it online at
16
- * http://www.gnu.org/licenses/gpl-2.0.html.
174 *
185 * Copyright IBM Corporation, 2001
196 *
207 * Authors: Dipankar Sarma <dipankar@in.ibm.com>
218 * Manfred Spraul <manfred@colorfullife.com>
229 *
23
- * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
10
+ * Based on the original work by Paul McKenney <paulmck@linux.ibm.com>
2411 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
2512 * Papers:
2613 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
....@@ -53,6 +40,9 @@
5340 #include <linux/rcupdate_wait.h>
5441 #include <linux/sched/isolation.h>
5542 #include <linux/kprobes.h>
43
+#include <linux/slab.h>
44
+#include <linux/irq_work.h>
45
+#include <linux/rcupdate_trace.h>
5646
5747 #define CREATE_TRACE_POINTS
5848
....@@ -64,9 +54,7 @@
6454 #define MODULE_PARAM_PREFIX "rcupdate."
6555
6656 #ifndef CONFIG_TINY_RCU
67
-extern int rcu_expedited; /* from sysctl */
6857 module_param(rcu_expedited, int, 0);
69
-extern int rcu_normal; /* from sysctl */
7058 module_param(rcu_normal, int, 0);
7159 static int rcu_normal_after_boot;
7260 module_param(rcu_normal_after_boot, int, 0);
....@@ -74,9 +62,15 @@
7462
7563 #ifdef CONFIG_DEBUG_LOCK_ALLOC
7664 /**
77
- * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section?
65
+ * rcu_read_lock_held_common() - might we be in RCU-sched read-side critical section?
66
+ * @ret: Best guess answer if lockdep cannot be relied on
7867 *
79
- * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an
68
+ * Returns true if lockdep must be ignored, in which case ``*ret`` contains
69
+ * the best guess described below. Otherwise returns false, in which
70
+ * case ``*ret`` tells the caller nothing and the caller should instead
71
+ * consult lockdep.
72
+ *
73
+ * If CONFIG_DEBUG_LOCK_ALLOC is selected, set ``*ret`` to nonzero iff in an
8074 * RCU-sched read-side critical section. In absence of
8175 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side
8276 * critical section unless it can prove otherwise. Note that disabling
....@@ -88,35 +82,45 @@
8882 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot
8983 * and while lockdep is disabled.
9084 *
91
- * Note that if the CPU is in the idle loop from an RCU point of
92
- * view (ie: that we are in the section between rcu_idle_enter() and
93
- * rcu_idle_exit()) then rcu_read_lock_held() returns false even if the CPU
94
- * did an rcu_read_lock(). The reason for this is that RCU ignores CPUs
95
- * that are in such a section, considering these as in extended quiescent
96
- * state, so such a CPU is effectively never in an RCU read-side critical
97
- * section regardless of what RCU primitives it invokes. This state of
98
- * affairs is required --- we need to keep an RCU-free window in idle
99
- * where the CPU may possibly enter into low power mode. This way we can
100
- * notice an extended quiescent state to other CPUs that started a grace
101
- * period. Otherwise we would delay any grace period as long as we run in
102
- * the idle task.
85
+ * Note that if the CPU is in the idle loop from an RCU point of view (ie:
86
+ * that we are in the section between rcu_idle_enter() and rcu_idle_exit())
87
+ * then rcu_read_lock_held() sets ``*ret`` to false even if the CPU did an
88
+ * rcu_read_lock(). The reason for this is that RCU ignores CPUs that are
89
+ * in such a section, considering these as in extended quiescent state,
90
+ * so such a CPU is effectively never in an RCU read-side critical section
91
+ * regardless of what RCU primitives it invokes. This state of affairs is
92
+ * required --- we need to keep an RCU-free window in idle where the CPU may
93
+ * possibly enter into low power mode. This way we can notice an extended
94
+ * quiescent state to other CPUs that started a grace period. Otherwise
95
+ * we would delay any grace period as long as we run in the idle task.
10396 *
104
- * Similarly, we avoid claiming an SRCU read lock held if the current
97
+ * Similarly, we avoid claiming an RCU read lock held if the current
10598 * CPU is offline.
10699 */
100
+static bool rcu_read_lock_held_common(bool *ret)
101
+{
102
+ if (!debug_lockdep_rcu_enabled()) {
103
+ *ret = true;
104
+ return true;
105
+ }
106
+ if (!rcu_is_watching()) {
107
+ *ret = false;
108
+ return true;
109
+ }
110
+ if (!rcu_lockdep_current_cpu_online()) {
111
+ *ret = false;
112
+ return true;
113
+ }
114
+ return false;
115
+}
116
+
107117 int rcu_read_lock_sched_held(void)
108118 {
109
- int lockdep_opinion = 0;
119
+ bool ret;
110120
111
- if (!debug_lockdep_rcu_enabled())
112
- return 1;
113
- if (!rcu_is_watching())
114
- return 0;
115
- if (!rcu_lockdep_current_cpu_online())
116
- return 0;
117
- if (debug_locks)
118
- lockdep_opinion = lock_is_held(&rcu_sched_lock_map);
119
- return lockdep_opinion || !preemptible();
121
+ if (rcu_read_lock_held_common(&ret))
122
+ return ret;
123
+ return lock_is_held(&rcu_sched_lock_map) || !preemptible();
120124 }
121125 EXPORT_SYMBOL(rcu_read_lock_sched_held);
122126 #endif
....@@ -149,8 +153,7 @@
149153 */
150154 bool rcu_gp_is_expedited(void)
151155 {
152
- return rcu_expedited || atomic_read(&rcu_expedited_nesting) ||
153
- rcu_scheduler_active == RCU_SCHEDULER_INIT;
156
+ return rcu_expedited || atomic_read(&rcu_expedited_nesting);
154157 }
155158 EXPORT_SYMBOL_GPL(rcu_gp_is_expedited);
156159
....@@ -182,6 +185,8 @@
182185 }
183186 EXPORT_SYMBOL_GPL(rcu_unexpedite_gp);
184187
188
+static bool rcu_boot_ended __read_mostly;
189
+
185190 /*
186191 * Inform RCU of the end of the in-kernel boot sequence.
187192 */
....@@ -190,7 +195,17 @@
190195 rcu_unexpedite_gp();
191196 if (rcu_normal_after_boot)
192197 WRITE_ONCE(rcu_normal, 1);
198
+ rcu_boot_ended = true;
193199 }
200
+
201
+/*
202
+ * Let rcutorture know when it is OK to turn it up to eleven.
203
+ */
204
+bool rcu_inkernel_boot_has_ended(void)
205
+{
206
+ return rcu_boot_ended;
207
+}
208
+EXPORT_SYMBOL_GPL(rcu_inkernel_boot_has_ended);
194209
195210 #endif /* #ifndef CONFIG_TINY_RCU */
196211
....@@ -204,11 +219,7 @@
204219 if (!IS_ENABLED(CONFIG_PROVE_RCU))
205220 return;
206221 synchronize_rcu();
207
- synchronize_rcu_bh();
208
- synchronize_sched();
209222 synchronize_rcu_expedited();
210
- synchronize_rcu_bh_expedited();
211
- synchronize_sched_expedited();
212223 }
213224
214225 #if !defined(CONFIG_TINY_RCU) || defined(CONFIG_SRCU)
....@@ -220,6 +231,7 @@
220231 {
221232 rcu_test_sync_prims();
222233 rcu_scheduler_active = RCU_SCHEDULER_RUNNING;
234
+ kfree_rcu_scheduler_running();
223235 rcu_test_sync_prims();
224236 return 0;
225237 }
....@@ -229,32 +241,44 @@
229241
230242 #ifdef CONFIG_DEBUG_LOCK_ALLOC
231243 static struct lock_class_key rcu_lock_key;
232
-struct lockdep_map rcu_lock_map =
233
- STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key);
244
+struct lockdep_map rcu_lock_map = {
245
+ .name = "rcu_read_lock",
246
+ .key = &rcu_lock_key,
247
+ .wait_type_outer = LD_WAIT_FREE,
248
+ .wait_type_inner = LD_WAIT_CONFIG, /* XXX PREEMPT_RCU ? */
249
+};
234250 EXPORT_SYMBOL_GPL(rcu_lock_map);
235251
236252 static struct lock_class_key rcu_bh_lock_key;
237
-struct lockdep_map rcu_bh_lock_map =
238
- STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_bh", &rcu_bh_lock_key);
253
+struct lockdep_map rcu_bh_lock_map = {
254
+ .name = "rcu_read_lock_bh",
255
+ .key = &rcu_bh_lock_key,
256
+ .wait_type_outer = LD_WAIT_FREE,
257
+ .wait_type_inner = LD_WAIT_CONFIG, /* PREEMPT_LOCK also makes BH preemptible */
258
+};
239259 EXPORT_SYMBOL_GPL(rcu_bh_lock_map);
240260
241261 static struct lock_class_key rcu_sched_lock_key;
242
-struct lockdep_map rcu_sched_lock_map =
243
- STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_sched", &rcu_sched_lock_key);
262
+struct lockdep_map rcu_sched_lock_map = {
263
+ .name = "rcu_read_lock_sched",
264
+ .key = &rcu_sched_lock_key,
265
+ .wait_type_outer = LD_WAIT_FREE,
266
+ .wait_type_inner = LD_WAIT_SPIN,
267
+};
244268 EXPORT_SYMBOL_GPL(rcu_sched_lock_map);
245269
270
+// Tell lockdep when RCU callbacks are being invoked.
246271 static struct lock_class_key rcu_callback_key;
247272 struct lockdep_map rcu_callback_map =
248273 STATIC_LOCKDEP_MAP_INIT("rcu_callback", &rcu_callback_key);
249274 EXPORT_SYMBOL_GPL(rcu_callback_map);
250275
251
-int notrace debug_lockdep_rcu_enabled(void)
276
+noinstr int notrace debug_lockdep_rcu_enabled(void)
252277 {
253
- return rcu_scheduler_active != RCU_SCHEDULER_INACTIVE && debug_locks &&
278
+ return rcu_scheduler_active != RCU_SCHEDULER_INACTIVE && READ_ONCE(debug_locks) &&
254279 current->lockdep_recursion == 0;
255280 }
256281 EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled);
257
-NOKPROBE_SYMBOL(debug_lockdep_rcu_enabled);
258282
259283 /**
260284 * rcu_read_lock_held() - might we be in RCU read-side critical section?
....@@ -278,12 +302,10 @@
278302 */
279303 int rcu_read_lock_held(void)
280304 {
281
- if (!debug_lockdep_rcu_enabled())
282
- return 1;
283
- if (!rcu_is_watching())
284
- return 0;
285
- if (!rcu_lockdep_current_cpu_online())
286
- return 0;
305
+ bool ret;
306
+
307
+ if (rcu_read_lock_held_common(&ret))
308
+ return ret;
287309 return lock_is_held(&rcu_lock_map);
288310 }
289311 EXPORT_SYMBOL_GPL(rcu_read_lock_held);
....@@ -300,20 +322,32 @@
300322 *
301323 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot.
302324 *
303
- * Note that rcu_read_lock() is disallowed if the CPU is either idle or
325
+ * Note that rcu_read_lock_bh() is disallowed if the CPU is either idle or
304326 * offline from an RCU perspective, so check for those as well.
305327 */
306328 int rcu_read_lock_bh_held(void)
307329 {
308
- if (!debug_lockdep_rcu_enabled())
309
- return 1;
310
- if (!rcu_is_watching())
311
- return 0;
312
- if (!rcu_lockdep_current_cpu_online())
313
- return 0;
330
+ bool ret;
331
+
332
+ if (rcu_read_lock_held_common(&ret))
333
+ return ret;
314334 return in_softirq() || irqs_disabled();
315335 }
316336 EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held);
337
+
338
+int rcu_read_lock_any_held(void)
339
+{
340
+ bool ret;
341
+
342
+ if (rcu_read_lock_held_common(&ret))
343
+ return ret;
344
+ if (lock_is_held(&rcu_lock_map) ||
345
+ lock_is_held(&rcu_bh_lock_map) ||
346
+ lock_is_held(&rcu_sched_lock_map))
347
+ return 1;
348
+ return !preemptible();
349
+}
350
+EXPORT_SYMBOL_GPL(rcu_read_lock_any_held);
317351
318352 #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
319353
....@@ -338,35 +372,35 @@
338372 int i;
339373 int j;
340374
341
- /* Initialize and register callbacks for each flavor specified. */
375
+ /* Initialize and register callbacks for each crcu_array element. */
342376 for (i = 0; i < n; i++) {
343377 if (checktiny &&
344
- (crcu_array[i] == call_rcu ||
345
- crcu_array[i] == call_rcu_bh)) {
378
+ (crcu_array[i] == call_rcu)) {
346379 might_sleep();
347380 continue;
348381 }
349
- init_rcu_head_on_stack(&rs_array[i].head);
350
- init_completion(&rs_array[i].completion);
351382 for (j = 0; j < i; j++)
352383 if (crcu_array[j] == crcu_array[i])
353384 break;
354
- if (j == i)
385
+ if (j == i) {
386
+ init_rcu_head_on_stack(&rs_array[i].head);
387
+ init_completion(&rs_array[i].completion);
355388 (crcu_array[i])(&rs_array[i].head, wakeme_after_rcu);
389
+ }
356390 }
357391
358392 /* Wait for all callbacks to be invoked. */
359393 for (i = 0; i < n; i++) {
360394 if (checktiny &&
361
- (crcu_array[i] == call_rcu ||
362
- crcu_array[i] == call_rcu_bh))
395
+ (crcu_array[i] == call_rcu))
363396 continue;
364397 for (j = 0; j < i; j++)
365398 if (crcu_array[j] == crcu_array[i])
366399 break;
367
- if (j == i)
400
+ if (j == i) {
368401 wait_for_completion(&rs_array[i].completion);
369
- destroy_rcu_head_on_stack(&rs_array[i].head);
402
+ destroy_rcu_head_on_stack(&rs_array[i].head);
403
+ }
370404 }
371405 }
372406 EXPORT_SYMBOL_GPL(__wait_rcu_gp);
....@@ -422,14 +456,14 @@
422456 }
423457 EXPORT_SYMBOL_GPL(destroy_rcu_head_on_stack);
424458
425
-struct debug_obj_descr rcuhead_debug_descr = {
459
+const struct debug_obj_descr rcuhead_debug_descr = {
426460 .name = "rcu_head",
427461 .is_static_object = rcuhead_is_static_object,
428462 };
429463 EXPORT_SYMBOL_GPL(rcuhead_debug_descr);
430464 #endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
431465
432
-#if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU) || defined(CONFIG_RCU_TRACE)
466
+#if defined(CONFIG_TREE_RCU) || defined(CONFIG_RCU_TRACE)
433467 void do_trace_rcu_torture_read(const char *rcutorturename, struct rcu_head *rhp,
434468 unsigned long secs,
435469 unsigned long c_old, unsigned long c)
....@@ -442,445 +476,42 @@
442476 do { } while (0)
443477 #endif
444478
445
-#ifdef CONFIG_RCU_STALL_COMMON
479
+#if IS_ENABLED(CONFIG_RCU_TORTURE_TEST) || IS_MODULE(CONFIG_RCU_TORTURE_TEST)
480
+/* Get rcutorture access to sched_setaffinity(). */
481
+long rcutorture_sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
482
+{
483
+ int ret;
446484
447
-#ifdef CONFIG_PROVE_RCU
448
-#define RCU_STALL_DELAY_DELTA (5 * HZ)
449
-#else
450
-#define RCU_STALL_DELAY_DELTA 0
485
+ ret = sched_setaffinity(pid, in_mask);
486
+ WARN_ONCE(ret, "%s: sched_setaffinity() returned %d\n", __func__, ret);
487
+ return ret;
488
+}
489
+EXPORT_SYMBOL_GPL(rcutorture_sched_setaffinity);
451490 #endif
452491
453
-int rcu_cpu_stall_suppress __read_mostly; /* 1 = suppress stall warnings. */
492
+#ifdef CONFIG_RCU_STALL_COMMON
493
+int rcu_cpu_stall_ftrace_dump __read_mostly;
494
+module_param(rcu_cpu_stall_ftrace_dump, int, 0644);
495
+int rcu_cpu_stall_suppress __read_mostly; // !0 = suppress stall warnings.
454496 EXPORT_SYMBOL_GPL(rcu_cpu_stall_suppress);
455
-static int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT;
456
-
457497 module_param(rcu_cpu_stall_suppress, int, 0644);
498
+int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT;
458499 module_param(rcu_cpu_stall_timeout, int, 0644);
459
-
460
-int rcu_jiffies_till_stall_check(void)
461
-{
462
- int till_stall_check = READ_ONCE(rcu_cpu_stall_timeout);
463
-
464
- /*
465
- * Limit check must be consistent with the Kconfig limits
466
- * for CONFIG_RCU_CPU_STALL_TIMEOUT.
467
- */
468
- if (till_stall_check < 3) {
469
- WRITE_ONCE(rcu_cpu_stall_timeout, 3);
470
- till_stall_check = 3;
471
- } else if (till_stall_check > 300) {
472
- WRITE_ONCE(rcu_cpu_stall_timeout, 300);
473
- till_stall_check = 300;
474
- }
475
- return till_stall_check * HZ + RCU_STALL_DELAY_DELTA;
476
-}
477
-
478
-void rcu_sysrq_start(void)
479
-{
480
- if (!rcu_cpu_stall_suppress)
481
- rcu_cpu_stall_suppress = 2;
482
-}
483
-
484
-void rcu_sysrq_end(void)
485
-{
486
- if (rcu_cpu_stall_suppress == 2)
487
- rcu_cpu_stall_suppress = 0;
488
-}
489
-
490
-static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr)
491
-{
492
- rcu_cpu_stall_suppress = 1;
493
- return NOTIFY_DONE;
494
-}
495
-
496
-static struct notifier_block rcu_panic_block = {
497
- .notifier_call = rcu_panic,
498
-};
499
-
500
-static int __init check_cpu_stall_init(void)
501
-{
502
- atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block);
503
- return 0;
504
-}
505
-early_initcall(check_cpu_stall_init);
506
-
507500 #endif /* #ifdef CONFIG_RCU_STALL_COMMON */
508501
509
-#ifdef CONFIG_TASKS_RCU
510
-
511
-/*
512
- * Simple variant of RCU whose quiescent states are voluntary context
513
- * switch, cond_resched_rcu_qs(), user-space execution, and idle.
514
- * As such, grace periods can take one good long time. There are no
515
- * read-side primitives similar to rcu_read_lock() and rcu_read_unlock()
516
- * because this implementation is intended to get the system into a safe
517
- * state for some of the manipulations involved in tracing and the like.
518
- * Finally, this implementation does not support high call_rcu_tasks()
519
- * rates from multiple CPUs. If this is required, per-CPU callback lists
520
- * will be needed.
521
- */
522
-
523
-/* Global list of callbacks and associated lock. */
524
-static struct rcu_head *rcu_tasks_cbs_head;
525
-static struct rcu_head **rcu_tasks_cbs_tail = &rcu_tasks_cbs_head;
526
-static DECLARE_WAIT_QUEUE_HEAD(rcu_tasks_cbs_wq);
527
-static DEFINE_RAW_SPINLOCK(rcu_tasks_cbs_lock);
528
-
529
-/* Track exiting tasks in order to allow them to be waited for. */
530
-DEFINE_STATIC_SRCU(tasks_rcu_exit_srcu);
531
-
532
-/* Control stall timeouts. Disable with <= 0, otherwise jiffies till stall. */
533
-#define RCU_TASK_STALL_TIMEOUT (HZ * 60 * 10)
534
-static int rcu_task_stall_timeout __read_mostly = RCU_TASK_STALL_TIMEOUT;
535
-module_param(rcu_task_stall_timeout, int, 0644);
536
-
537
-static struct task_struct *rcu_tasks_kthread_ptr;
538
-
539
-/**
540
- * call_rcu_tasks() - Queue an RCU for invocation task-based grace period
541
- * @rhp: structure to be used for queueing the RCU updates.
542
- * @func: actual callback function to be invoked after the grace period
543
- *
544
- * The callback function will be invoked some time after a full grace
545
- * period elapses, in other words after all currently executing RCU
546
- * read-side critical sections have completed. call_rcu_tasks() assumes
547
- * that the read-side critical sections end at a voluntary context
548
- * switch (not a preemption!), cond_resched_rcu_qs(), entry into idle,
549
- * or transition to usermode execution. As such, there are no read-side
550
- * primitives analogous to rcu_read_lock() and rcu_read_unlock() because
551
- * this primitive is intended to determine that all tasks have passed
552
- * through a safe state, not so much for data-strcuture synchronization.
553
- *
554
- * See the description of call_rcu() for more detailed information on
555
- * memory ordering guarantees.
556
- */
557
-void call_rcu_tasks(struct rcu_head *rhp, rcu_callback_t func)
558
-{
559
- unsigned long flags;
560
- bool needwake;
561
-
562
- rhp->next = NULL;
563
- rhp->func = func;
564
- raw_spin_lock_irqsave(&rcu_tasks_cbs_lock, flags);
565
- needwake = !rcu_tasks_cbs_head;
566
- *rcu_tasks_cbs_tail = rhp;
567
- rcu_tasks_cbs_tail = &rhp->next;
568
- raw_spin_unlock_irqrestore(&rcu_tasks_cbs_lock, flags);
569
- /* We can't create the thread unless interrupts are enabled. */
570
- if (needwake && READ_ONCE(rcu_tasks_kthread_ptr))
571
- wake_up(&rcu_tasks_cbs_wq);
572
-}
573
-EXPORT_SYMBOL_GPL(call_rcu_tasks);
574
-
575
-/**
576
- * synchronize_rcu_tasks - wait until an rcu-tasks grace period has elapsed.
577
- *
578
- * Control will return to the caller some time after a full rcu-tasks
579
- * grace period has elapsed, in other words after all currently
580
- * executing rcu-tasks read-side critical sections have elapsed. These
581
- * read-side critical sections are delimited by calls to schedule(),
582
- * cond_resched_tasks_rcu_qs(), idle execution, userspace execution, calls
583
- * to synchronize_rcu_tasks(), and (in theory, anyway) cond_resched().
584
- *
585
- * This is a very specialized primitive, intended only for a few uses in
586
- * tracing and other situations requiring manipulation of function
587
- * preambles and profiling hooks. The synchronize_rcu_tasks() function
588
- * is not (yet) intended for heavy use from multiple CPUs.
589
- *
590
- * Note that this guarantee implies further memory-ordering guarantees.
591
- * On systems with more than one CPU, when synchronize_rcu_tasks() returns,
592
- * each CPU is guaranteed to have executed a full memory barrier since the
593
- * end of its last RCU-tasks read-side critical section whose beginning
594
- * preceded the call to synchronize_rcu_tasks(). In addition, each CPU
595
- * having an RCU-tasks read-side critical section that extends beyond
596
- * the return from synchronize_rcu_tasks() is guaranteed to have executed
597
- * a full memory barrier after the beginning of synchronize_rcu_tasks()
598
- * and before the beginning of that RCU-tasks read-side critical section.
599
- * Note that these guarantees include CPUs that are offline, idle, or
600
- * executing in user mode, as well as CPUs that are executing in the kernel.
601
- *
602
- * Furthermore, if CPU A invoked synchronize_rcu_tasks(), which returned
603
- * to its caller on CPU B, then both CPU A and CPU B are guaranteed
604
- * to have executed a full memory barrier during the execution of
605
- * synchronize_rcu_tasks() -- even if CPU A and CPU B are the same CPU
606
- * (but again only if the system has more than one CPU).
607
- */
608
-void synchronize_rcu_tasks(void)
609
-{
610
- /* Complain if the scheduler has not started. */
611
- RCU_LOCKDEP_WARN(rcu_scheduler_active == RCU_SCHEDULER_INACTIVE,
612
- "synchronize_rcu_tasks called too soon");
613
-
614
- /* Wait for the grace period. */
615
- wait_rcu_gp(call_rcu_tasks);
616
-}
617
-EXPORT_SYMBOL_GPL(synchronize_rcu_tasks);
618
-
619
-/**
620
- * rcu_barrier_tasks - Wait for in-flight call_rcu_tasks() callbacks.
621
- *
622
- * Although the current implementation is guaranteed to wait, it is not
623
- * obligated to, for example, if there are no pending callbacks.
624
- */
625
-void rcu_barrier_tasks(void)
626
-{
627
- /* There is only one callback queue, so this is easy. ;-) */
628
- synchronize_rcu_tasks();
629
-}
630
-EXPORT_SYMBOL_GPL(rcu_barrier_tasks);
631
-
632
-/* See if tasks are still holding out, complain if so. */
633
-static void check_holdout_task(struct task_struct *t,
634
- bool needreport, bool *firstreport)
635
-{
636
- int cpu;
637
-
638
- if (!READ_ONCE(t->rcu_tasks_holdout) ||
639
- t->rcu_tasks_nvcsw != READ_ONCE(t->nvcsw) ||
640
- !READ_ONCE(t->on_rq) ||
641
- (IS_ENABLED(CONFIG_NO_HZ_FULL) &&
642
- !is_idle_task(t) && t->rcu_tasks_idle_cpu >= 0)) {
643
- WRITE_ONCE(t->rcu_tasks_holdout, false);
644
- list_del_init(&t->rcu_tasks_holdout_list);
645
- put_task_struct(t);
646
- return;
647
- }
648
- rcu_request_urgent_qs_task(t);
649
- if (!needreport)
650
- return;
651
- if (*firstreport) {
652
- pr_err("INFO: rcu_tasks detected stalls on tasks:\n");
653
- *firstreport = false;
654
- }
655
- cpu = task_cpu(t);
656
- pr_alert("%p: %c%c nvcsw: %lu/%lu holdout: %d idle_cpu: %d/%d\n",
657
- t, ".I"[is_idle_task(t)],
658
- "N."[cpu < 0 || !tick_nohz_full_cpu(cpu)],
659
- t->rcu_tasks_nvcsw, t->nvcsw, t->rcu_tasks_holdout,
660
- t->rcu_tasks_idle_cpu, cpu);
661
- sched_show_task(t);
662
-}
663
-
664
-/* RCU-tasks kthread that detects grace periods and invokes callbacks. */
665
-static int __noreturn rcu_tasks_kthread(void *arg)
666
-{
667
- unsigned long flags;
668
- struct task_struct *g, *t;
669
- unsigned long lastreport;
670
- struct rcu_head *list;
671
- struct rcu_head *next;
672
- LIST_HEAD(rcu_tasks_holdouts);
673
- int fract;
674
-
675
- /* Run on housekeeping CPUs by default. Sysadm can move if desired. */
676
- housekeeping_affine(current, HK_FLAG_RCU);
677
-
678
- /*
679
- * Each pass through the following loop makes one check for
680
- * newly arrived callbacks, and, if there are some, waits for
681
- * one RCU-tasks grace period and then invokes the callbacks.
682
- * This loop is terminated by the system going down. ;-)
683
- */
684
- for (;;) {
685
-
686
- /* Pick up any new callbacks. */
687
- raw_spin_lock_irqsave(&rcu_tasks_cbs_lock, flags);
688
- list = rcu_tasks_cbs_head;
689
- rcu_tasks_cbs_head = NULL;
690
- rcu_tasks_cbs_tail = &rcu_tasks_cbs_head;
691
- raw_spin_unlock_irqrestore(&rcu_tasks_cbs_lock, flags);
692
-
693
- /* If there were none, wait a bit and start over. */
694
- if (!list) {
695
- wait_event_interruptible(rcu_tasks_cbs_wq,
696
- rcu_tasks_cbs_head);
697
- if (!rcu_tasks_cbs_head) {
698
- WARN_ON(signal_pending(current));
699
- schedule_timeout_interruptible(HZ/10);
700
- }
701
- continue;
702
- }
703
-
704
- /*
705
- * Wait for all pre-existing t->on_rq and t->nvcsw
706
- * transitions to complete. Invoking synchronize_sched()
707
- * suffices because all these transitions occur with
708
- * interrupts disabled. Without this synchronize_sched(),
709
- * a read-side critical section that started before the
710
- * grace period might be incorrectly seen as having started
711
- * after the grace period.
712
- *
713
- * This synchronize_sched() also dispenses with the
714
- * need for a memory barrier on the first store to
715
- * ->rcu_tasks_holdout, as it forces the store to happen
716
- * after the beginning of the grace period.
717
- */
718
- synchronize_sched();
719
-
720
- /*
721
- * There were callbacks, so we need to wait for an
722
- * RCU-tasks grace period. Start off by scanning
723
- * the task list for tasks that are not already
724
- * voluntarily blocked. Mark these tasks and make
725
- * a list of them in rcu_tasks_holdouts.
726
- */
727
- rcu_read_lock();
728
- for_each_process_thread(g, t) {
729
- if (t != current && READ_ONCE(t->on_rq) &&
730
- !is_idle_task(t)) {
731
- get_task_struct(t);
732
- t->rcu_tasks_nvcsw = READ_ONCE(t->nvcsw);
733
- WRITE_ONCE(t->rcu_tasks_holdout, true);
734
- list_add(&t->rcu_tasks_holdout_list,
735
- &rcu_tasks_holdouts);
736
- }
737
- }
738
- rcu_read_unlock();
739
-
740
- /*
741
- * Wait for tasks that are in the process of exiting.
742
- * This does only part of the job, ensuring that all
743
- * tasks that were previously exiting reach the point
744
- * where they have disabled preemption, allowing the
745
- * later synchronize_sched() to finish the job.
746
- */
747
- synchronize_srcu(&tasks_rcu_exit_srcu);
748
-
749
- /*
750
- * Each pass through the following loop scans the list
751
- * of holdout tasks, removing any that are no longer
752
- * holdouts. When the list is empty, we are done.
753
- */
754
- lastreport = jiffies;
755
-
756
- /* Start off with HZ/10 wait and slowly back off to 1 HZ wait*/
757
- fract = 10;
758
-
759
- for (;;) {
760
- bool firstreport;
761
- bool needreport;
762
- int rtst;
763
- struct task_struct *t1;
764
-
765
- if (list_empty(&rcu_tasks_holdouts))
766
- break;
767
-
768
- /* Slowly back off waiting for holdouts */
769
- schedule_timeout_interruptible(HZ/fract);
770
-
771
- if (fract > 1)
772
- fract--;
773
-
774
- rtst = READ_ONCE(rcu_task_stall_timeout);
775
- needreport = rtst > 0 &&
776
- time_after(jiffies, lastreport + rtst);
777
- if (needreport)
778
- lastreport = jiffies;
779
- firstreport = true;
780
- WARN_ON(signal_pending(current));
781
- list_for_each_entry_safe(t, t1, &rcu_tasks_holdouts,
782
- rcu_tasks_holdout_list) {
783
- check_holdout_task(t, needreport, &firstreport);
784
- cond_resched();
785
- }
786
- }
787
-
788
- /*
789
- * Because ->on_rq and ->nvcsw are not guaranteed
790
- * to have a full memory barriers prior to them in the
791
- * schedule() path, memory reordering on other CPUs could
792
- * cause their RCU-tasks read-side critical sections to
793
- * extend past the end of the grace period. However,
794
- * because these ->nvcsw updates are carried out with
795
- * interrupts disabled, we can use synchronize_sched()
796
- * to force the needed ordering on all such CPUs.
797
- *
798
- * This synchronize_sched() also confines all
799
- * ->rcu_tasks_holdout accesses to be within the grace
800
- * period, avoiding the need for memory barriers for
801
- * ->rcu_tasks_holdout accesses.
802
- *
803
- * In addition, this synchronize_sched() waits for exiting
804
- * tasks to complete their final preempt_disable() region
805
- * of execution, cleaning up after the synchronize_srcu()
806
- * above.
807
- */
808
- synchronize_sched();
809
-
810
- /* Invoke the callbacks. */
811
- while (list) {
812
- next = list->next;
813
- local_bh_disable();
814
- list->func(list);
815
- local_bh_enable();
816
- list = next;
817
- cond_resched();
818
- }
819
- /* Paranoid sleep to keep this from entering a tight loop */
820
- schedule_timeout_uninterruptible(HZ/10);
821
- }
822
-}
823
-
824
-/* Spawn rcu_tasks_kthread() at core_initcall() time. */
825
-static int __init rcu_spawn_tasks_kthread(void)
826
-{
827
- struct task_struct *t;
828
-
829
- t = kthread_run(rcu_tasks_kthread, NULL, "rcu_tasks_kthread");
830
- BUG_ON(IS_ERR(t));
831
- smp_mb(); /* Ensure others see full kthread. */
832
- WRITE_ONCE(rcu_tasks_kthread_ptr, t);
833
- return 0;
834
-}
835
-core_initcall(rcu_spawn_tasks_kthread);
836
-
837
-/* Do the srcu_read_lock() for the above synchronize_srcu(). */
838
-void exit_tasks_rcu_start(void)
839
-{
840
- preempt_disable();
841
- current->rcu_tasks_idx = __srcu_read_lock(&tasks_rcu_exit_srcu);
842
- preempt_enable();
843
-}
844
-
845
-/* Do the srcu_read_unlock() for the above synchronize_srcu(). */
846
-void exit_tasks_rcu_finish(void)
847
-{
848
- preempt_disable();
849
- __srcu_read_unlock(&tasks_rcu_exit_srcu, current->rcu_tasks_idx);
850
- preempt_enable();
851
-}
852
-
853
-#endif /* #ifdef CONFIG_TASKS_RCU */
854
-
855
-#ifndef CONFIG_TINY_RCU
856
-
857
-/*
858
- * Print any non-default Tasks RCU settings.
859
- */
860
-static void __init rcu_tasks_bootup_oddness(void)
861
-{
862
-#ifdef CONFIG_TASKS_RCU
863
- if (rcu_task_stall_timeout != RCU_TASK_STALL_TIMEOUT)
864
- pr_info("\tTasks-RCU CPU stall warnings timeout set to %d (rcu_task_stall_timeout).\n", rcu_task_stall_timeout);
865
- else
866
- pr_info("\tTasks RCU enabled.\n");
867
-#endif /* #ifdef CONFIG_TASKS_RCU */
868
-}
869
-
870
-#endif /* #ifndef CONFIG_TINY_RCU */
502
+// Suppress boot-time RCU CPU stall warnings and rcutorture writer stall
503
+// warnings. Also used by rcutorture even if stall warnings are excluded.
504
+int rcu_cpu_stall_suppress_at_boot __read_mostly; // !0 = suppress boot stalls.
505
+EXPORT_SYMBOL_GPL(rcu_cpu_stall_suppress_at_boot);
506
+module_param(rcu_cpu_stall_suppress_at_boot, int, 0444);
871507
872508 #ifdef CONFIG_PROVE_RCU
873509
874510 /*
875
- * Early boot self test parameters, one for each flavor
511
+ * Early boot self test parameters.
876512 */
877513 static bool rcu_self_test;
878
-static bool rcu_self_test_bh;
879
-static bool rcu_self_test_sched;
880
-
881514 module_param(rcu_self_test, bool, 0444);
882
-module_param(rcu_self_test_bh, bool, 0444);
883
-module_param(rcu_self_test_sched, bool, 0444);
884515
885516 static int rcu_self_test_counter;
886517
....@@ -890,25 +521,24 @@
890521 pr_info("RCU test callback executed %d\n", rcu_self_test_counter);
891522 }
892523
524
+DEFINE_STATIC_SRCU(early_srcu);
525
+
526
+struct early_boot_kfree_rcu {
527
+ struct rcu_head rh;
528
+};
529
+
893530 static void early_boot_test_call_rcu(void)
894531 {
895532 static struct rcu_head head;
533
+ static struct rcu_head shead;
534
+ struct early_boot_kfree_rcu *rhp;
896535
897536 call_rcu(&head, test_callback);
898
-}
899
-
900
-static void early_boot_test_call_rcu_bh(void)
901
-{
902
- static struct rcu_head head;
903
-
904
- call_rcu_bh(&head, test_callback);
905
-}
906
-
907
-static void early_boot_test_call_rcu_sched(void)
908
-{
909
- static struct rcu_head head;
910
-
911
- call_rcu_sched(&head, test_callback);
537
+ if (IS_ENABLED(CONFIG_SRCU))
538
+ call_srcu(&early_srcu, &shead, test_callback);
539
+ rhp = kmalloc(sizeof(*rhp), GFP_KERNEL);
540
+ if (!WARN_ON_ONCE(!rhp))
541
+ kfree_rcu(rhp, rh);
912542 }
913543
914544 void rcu_early_boot_tests(void)
....@@ -917,10 +547,6 @@
917547
918548 if (rcu_self_test)
919549 early_boot_test_call_rcu();
920
- if (rcu_self_test_bh)
921
- early_boot_test_call_rcu_bh();
922
- if (rcu_self_test_sched)
923
- early_boot_test_call_rcu_sched();
924550 rcu_test_sync_prims();
925551 }
926552
....@@ -932,16 +558,11 @@
932558 if (rcu_self_test) {
933559 early_boot_test_counter++;
934560 rcu_barrier();
561
+ if (IS_ENABLED(CONFIG_SRCU)) {
562
+ early_boot_test_counter++;
563
+ srcu_barrier(&early_srcu);
564
+ }
935565 }
936
- if (rcu_self_test_bh) {
937
- early_boot_test_counter++;
938
- rcu_barrier_bh();
939
- }
940
- if (rcu_self_test_sched) {
941
- early_boot_test_counter++;
942
- rcu_barrier_sched();
943
- }
944
-
945566 if (rcu_self_test_counter != early_boot_test_counter) {
946567 WARN_ON(1);
947568 ret = -1;
....@@ -954,6 +575,8 @@
954575 void rcu_early_boot_tests(void) {}
955576 #endif /* CONFIG_PROVE_RCU */
956577
578
+#include "tasks.h"
579
+
957580 #ifndef CONFIG_TINY_RCU
958581
959582 /*