1 // SPDX-License-Identifier: GPL-2.0+ 1
2 /*
3 * Read-Copy Update mechanism for mutual exclu
4 *
5 * Copyright IBM Corporation, 2001
6 *
7 * Authors: Dipankar Sarma <dipankar@in.ibm.co
8 * Manfred Spraul <manfred@colorfulli
9 *
10 * Based on the original work by Paul McKenney
11 * and inputs from Rusty Russell, Andrea Arcan
12 * Papers:
13 * http://www.rdrop.com/users/paulmck/paper/rc
14 * http://lse.sourceforge.net/locking/rclock_O
15 *
16 * For detailed explanation of Read-Copy Updat
17 * http://lse.sourceforge.net/loc
18 *
19 */
20 #include <linux/types.h>
21 #include <linux/kernel.h>
22 #include <linux/init.h>
23 #include <linux/spinlock.h>
24 #include <linux/smp.h>
25 #include <linux/interrupt.h>
26 #include <linux/sched/signal.h>
27 #include <linux/sched/debug.h>
28 #include <linux/torture.h>
29 #include <linux/atomic.h>
30 #include <linux/bitops.h>
31 #include <linux/percpu.h>
32 #include <linux/notifier.h>
33 #include <linux/cpu.h>
34 #include <linux/mutex.h>
35 #include <linux/export.h>
36 #include <linux/hardirq.h>
37 #include <linux/delay.h>
38 #include <linux/moduleparam.h>
39 #include <linux/kthread.h>
40 #include <linux/tick.h>
41 #include <linux/rcupdate_wait.h>
42 #include <linux/sched/isolation.h>
43 #include <linux/kprobes.h>
44 #include <linux/slab.h>
45 #include <linux/irq_work.h>
46 #include <linux/rcupdate_trace.h>
47
48 #define CREATE_TRACE_POINTS
49
50 #include "rcu.h"
51
52 #ifdef MODULE_PARAM_PREFIX
53 #undef MODULE_PARAM_PREFIX
54 #endif
55 #define MODULE_PARAM_PREFIX "rcupdate."
56
57 #ifndef CONFIG_TINY_RCU
58 module_param(rcu_expedited, int, 0444);
59 module_param(rcu_normal, int, 0444);
60 static int rcu_normal_after_boot = IS_ENABLED(
61 #if !defined(CONFIG_PREEMPT_RT) || defined(CON
62 module_param(rcu_normal_after_boot, int, 0444)
63 #endif
64 #endif /* #ifndef CONFIG_TINY_RCU */
65
66 #ifdef CONFIG_DEBUG_LOCK_ALLOC
67 /**
68 * rcu_read_lock_held_common() - might we be i
69 * @ret: Best guess answer if lockdep c
70 *
71 * Returns true if lockdep must be ignored, in
72 * the best guess described below. Otherwise
73 * case ``*ret`` tells the caller nothing and
74 * consult lockdep.
75 *
76 * If CONFIG_DEBUG_LOCK_ALLOC is selected, set
77 * RCU-sched read-side critical section. In a
78 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we ar
79 * critical section unless it can prove otherw
80 * of preemption (including disabling irqs) co
81 * read-side critical section. This is useful
82 * that required that they be called within an
83 * critical section.
84 *
85 * Check debug_lockdep_rcu_enabled() to preven
86 * and while lockdep is disabled.
87 *
88 * Note that if the CPU is in the idle loop fr
89 * that we are in the section between ct_idle_
90 * then rcu_read_lock_held() sets ``*ret`` to
91 * rcu_read_lock(). The reason for this is th
92 * in such a section, considering these as in
93 * so such a CPU is effectively never in an RC
94 * regardless of what RCU primitives it invoke
95 * required --- we need to keep an RCU-free wi
96 * possibly enter into low power mode. This wa
97 * quiescent state to other CPUs that started
98 * we would delay any grace period as long as
99 *
100 * Similarly, we avoid claiming an RCU read lo
101 * CPU is offline.
102 */
103 static bool rcu_read_lock_held_common(bool *re
104 {
105 if (!debug_lockdep_rcu_enabled()) {
106 *ret = true;
107 return true;
108 }
109 if (!rcu_is_watching()) {
110 *ret = false;
111 return true;
112 }
113 if (!rcu_lockdep_current_cpu_online())
114 *ret = false;
115 return true;
116 }
117 return false;
118 }
119
120 int rcu_read_lock_sched_held(void)
121 {
122 bool ret;
123
124 if (rcu_read_lock_held_common(&ret))
125 return ret;
126 return lock_is_held(&rcu_sched_lock_ma
127 }
128 EXPORT_SYMBOL(rcu_read_lock_sched_held);
129 #endif
130
131 #ifndef CONFIG_TINY_RCU
132
133 /*
134 * Should expedited grace-period primitives al
135 * non-expedited counterparts? Intended for u
136 * that if the user specifies both rcu_expedit
137 * rcu_normal wins. (Except during the time p
138 * when the first task is spawned until the rc
139 * core_initcall() is invoked, at which point
140 */
141 bool rcu_gp_is_normal(void)
142 {
143 return READ_ONCE(rcu_normal) &&
144 rcu_scheduler_active != RCU_SCH
145 }
146 EXPORT_SYMBOL_GPL(rcu_gp_is_normal);
147
148 static atomic_t rcu_async_hurry_nesting = ATOM
149 /*
150 * Should call_rcu() callbacks be processed wi
151 * they OK being executed with arbitrary delay
152 */
153 bool rcu_async_should_hurry(void)
154 {
155 return !IS_ENABLED(CONFIG_RCU_LAZY) ||
156 atomic_read(&rcu_async_hurry_ne
157 }
158 EXPORT_SYMBOL_GPL(rcu_async_should_hurry);
159
160 /**
161 * rcu_async_hurry - Make future async RCU cal
162 *
163 * After a call to this function, future calls
164 * will be processed in a timely fashion.
165 */
166 void rcu_async_hurry(void)
167 {
168 if (IS_ENABLED(CONFIG_RCU_LAZY))
169 atomic_inc(&rcu_async_hurry_ne
170 }
171 EXPORT_SYMBOL_GPL(rcu_async_hurry);
172
173 /**
174 * rcu_async_relax - Make future async RCU cal
175 *
176 * After a call to this function, future calls
177 * will be processed in a lazy fashion.
178 */
179 void rcu_async_relax(void)
180 {
181 if (IS_ENABLED(CONFIG_RCU_LAZY))
182 atomic_dec(&rcu_async_hurry_ne
183 }
184 EXPORT_SYMBOL_GPL(rcu_async_relax);
185
186 static atomic_t rcu_expedited_nesting = ATOMIC
187 /*
188 * Should normal grace-period primitives be ex
189 * use within RCU. Note that this function ta
190 * sysfs/boot variable and rcu_scheduler_activ
191 * as the rcu_expedite_gp() nesting. So loopi
192 * until rcu_gp_is_expedited() returns false i
193 */
194 bool rcu_gp_is_expedited(void)
195 {
196 return rcu_expedited || atomic_read(&r
197 }
198 EXPORT_SYMBOL_GPL(rcu_gp_is_expedited);
199
200 /**
201 * rcu_expedite_gp - Expedite future RCU grace
202 *
203 * After a call to this function, future calls
204 * friends act as the corresponding synchroniz
205 * had instead been called.
206 */
207 void rcu_expedite_gp(void)
208 {
209 atomic_inc(&rcu_expedited_nesting);
210 }
211 EXPORT_SYMBOL_GPL(rcu_expedite_gp);
212
213 /**
214 * rcu_unexpedite_gp - Cancel prior rcu_expedi
215 *
216 * Undo a prior call to rcu_expedite_gp(). If
217 * rcu_expedite_gp() are undone by a subsequen
218 * and if the rcu_expedited sysfs/boot paramet
219 * subsequent calls to synchronize_rcu() and f
220 * their normal non-expedited behavior.
221 */
222 void rcu_unexpedite_gp(void)
223 {
224 atomic_dec(&rcu_expedited_nesting);
225 }
226 EXPORT_SYMBOL_GPL(rcu_unexpedite_gp);
227
228 static bool rcu_boot_ended __read_mostly;
229
230 /*
231 * Inform RCU of the end of the in-kernel boot
232 */
233 void rcu_end_inkernel_boot(void)
234 {
235 rcu_unexpedite_gp();
236 rcu_async_relax();
237 if (rcu_normal_after_boot)
238 WRITE_ONCE(rcu_normal, 1);
239 rcu_boot_ended = true;
240 }
241
242 /*
243 * Let rcutorture know when it is OK to turn i
244 */
245 bool rcu_inkernel_boot_has_ended(void)
246 {
247 return rcu_boot_ended;
248 }
249 EXPORT_SYMBOL_GPL(rcu_inkernel_boot_has_ended)
250
251 #endif /* #ifndef CONFIG_TINY_RCU */
252
253 /*
254 * Test each non-SRCU synchronous grace-period
255 * useful just after a change in mode for thes
256 * during early boot.
257 */
258 void rcu_test_sync_prims(void)
259 {
260 if (!IS_ENABLED(CONFIG_PROVE_RCU))
261 return;
262 pr_info("Running RCU synchronous self
263 synchronize_rcu();
264 synchronize_rcu_expedited();
265 }
266
267 #if !defined(CONFIG_TINY_RCU)
268
269 /*
270 * Switch to run-time mode once RCU has fully
271 */
272 static int __init rcu_set_runtime_mode(void)
273 {
274 rcu_test_sync_prims();
275 rcu_scheduler_active = RCU_SCHEDULER_R
276 kfree_rcu_scheduler_running();
277 rcu_test_sync_prims();
278 return 0;
279 }
280 core_initcall(rcu_set_runtime_mode);
281
282 #endif /* #if !defined(CONFIG_TINY_RCU) */
283
284 #ifdef CONFIG_DEBUG_LOCK_ALLOC
285 static struct lock_class_key rcu_lock_key;
286 struct lockdep_map rcu_lock_map = {
287 .name = "rcu_read_lock",
288 .key = &rcu_lock_key,
289 .wait_type_outer = LD_WAIT_FREE,
290 .wait_type_inner = LD_WAIT_CONFIG, /*
291 };
292 EXPORT_SYMBOL_GPL(rcu_lock_map);
293
294 static struct lock_class_key rcu_bh_lock_key;
295 struct lockdep_map rcu_bh_lock_map = {
296 .name = "rcu_read_lock_bh",
297 .key = &rcu_bh_lock_key,
298 .wait_type_outer = LD_WAIT_FREE,
299 .wait_type_inner = LD_WAIT_CONFIG, /*
300 };
301 EXPORT_SYMBOL_GPL(rcu_bh_lock_map);
302
303 static struct lock_class_key rcu_sched_lock_ke
304 struct lockdep_map rcu_sched_lock_map = {
305 .name = "rcu_read_lock_sched",
306 .key = &rcu_sched_lock_key,
307 .wait_type_outer = LD_WAIT_FREE,
308 .wait_type_inner = LD_WAIT_SPIN,
309 };
310 EXPORT_SYMBOL_GPL(rcu_sched_lock_map);
311
312 // Tell lockdep when RCU callbacks are being i
313 static struct lock_class_key rcu_callback_key;
314 struct lockdep_map rcu_callback_map =
315 STATIC_LOCKDEP_MAP_INIT("rcu_callback"
316 EXPORT_SYMBOL_GPL(rcu_callback_map);
317
318 noinstr int notrace debug_lockdep_rcu_enabled(
319 {
320 return rcu_scheduler_active != RCU_SCH
321 current->lockdep_recursion == 0
322 }
323 EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled);
324
325 /**
326 * rcu_read_lock_held() - might we be in RCU r
327 *
328 * If CONFIG_DEBUG_LOCK_ALLOC is selected, ret
329 * read-side critical section. In absence of
330 * this assumes we are in an RCU read-side cri
331 * prove otherwise. This is useful for debug
332 * require that they be called within an RCU r
333 *
334 * Checks debug_lockdep_rcu_enabled() to preve
335 * and while lockdep is disabled.
336 *
337 * Note that rcu_read_lock() and the matching
338 * occur in the same context, for example, it
339 * rcu_read_unlock() in process context if the
340 * was invoked from within an irq handler.
341 *
342 * Note that rcu_read_lock() is disallowed if
343 * offline from an RCU perspective, so check f
344 */
345 int rcu_read_lock_held(void)
346 {
347 bool ret;
348
349 if (rcu_read_lock_held_common(&ret))
350 return ret;
351 return lock_is_held(&rcu_lock_map);
352 }
353 EXPORT_SYMBOL_GPL(rcu_read_lock_held);
354
355 /**
356 * rcu_read_lock_bh_held() - might we be in RC
357 *
358 * Check for bottom half being disabled, which
359 * CONFIG_PROVE_RCU and not cases. Note that
360 * rcu_read_lock_bh(), but then later enables
361 * will show the situation. This is useful fo
362 * that require that they be called within an
363 * section.
364 *
365 * Check debug_lockdep_rcu_enabled() to preven
366 *
367 * Note that rcu_read_lock_bh() is disallowed
368 * offline from an RCU perspective, so check f
369 */
370 int rcu_read_lock_bh_held(void)
371 {
372 bool ret;
373
374 if (rcu_read_lock_held_common(&ret))
375 return ret;
376 return in_softirq() || irqs_disabled()
377 }
378 EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held);
379
380 int rcu_read_lock_any_held(void)
381 {
382 bool ret;
383
384 if (rcu_read_lock_held_common(&ret))
385 return ret;
386 if (lock_is_held(&rcu_lock_map) ||
387 lock_is_held(&rcu_bh_lock_map) ||
388 lock_is_held(&rcu_sched_lock_map))
389 return 1;
390 return !preemptible();
391 }
392 EXPORT_SYMBOL_GPL(rcu_read_lock_any_held);
393
394 #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
395
396 /**
397 * wakeme_after_rcu() - Callback function to a
398 * @head: Pointer to rcu_head member within rc
399 *
400 * Awaken the corresponding task now that a gr
401 */
402 void wakeme_after_rcu(struct rcu_head *head)
403 {
404 struct rcu_synchronize *rcu;
405
406 rcu = container_of(head, struct rcu_sy
407 complete(&rcu->completion);
408 }
409 EXPORT_SYMBOL_GPL(wakeme_after_rcu);
410
411 void __wait_rcu_gp(bool checktiny, unsigned in
412 struct rcu_synchronize *rs_
413 {
414 int i;
415 int j;
416
417 /* Initialize and register callbacks f
418 for (i = 0; i < n; i++) {
419 if (checktiny &&
420 (crcu_array[i] == call_rcu
421 might_sleep();
422 continue;
423 }
424 for (j = 0; j < i; j++)
425 if (crcu_array[j] == c
426 break;
427 if (j == i) {
428 init_rcu_head_on_stack
429 init_completion(&rs_ar
430 (crcu_array[i])(&rs_ar
431 }
432 }
433
434 /* Wait for all callbacks to be invoke
435 for (i = 0; i < n; i++) {
436 if (checktiny &&
437 (crcu_array[i] == call_rcu
438 continue;
439 for (j = 0; j < i; j++)
440 if (crcu_array[j] == c
441 break;
442 if (j == i) {
443 wait_for_completion_st
444 destroy_rcu_head_on_st
445 }
446 }
447 }
448 EXPORT_SYMBOL_GPL(__wait_rcu_gp);
449
450 void finish_rcuwait(struct rcuwait *w)
451 {
452 rcu_assign_pointer(w->task, NULL);
453 __set_current_state(TASK_RUNNING);
454 }
455 EXPORT_SYMBOL_GPL(finish_rcuwait);
456
457 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
458 void init_rcu_head(struct rcu_head *head)
459 {
460 debug_object_init(head, &rcuhead_debug
461 }
462 EXPORT_SYMBOL_GPL(init_rcu_head);
463
464 void destroy_rcu_head(struct rcu_head *head)
465 {
466 debug_object_free(head, &rcuhead_debug
467 }
468 EXPORT_SYMBOL_GPL(destroy_rcu_head);
469
470 static bool rcuhead_is_static_object(void *add
471 {
472 return true;
473 }
474
475 /**
476 * init_rcu_head_on_stack() - initialize on-st
477 * @head: pointer to rcu_head structure to be
478 *
479 * This function informs debugobjects of a new
480 * has been allocated as an auto variable on t
481 * is not required for rcu_head structures tha
482 * that are dynamically allocated on the heap.
483 * effect for !CONFIG_DEBUG_OBJECTS_RCU_HEAD k
484 */
485 void init_rcu_head_on_stack(struct rcu_head *h
486 {
487 debug_object_init_on_stack(head, &rcuh
488 }
489 EXPORT_SYMBOL_GPL(init_rcu_head_on_stack);
490
491 /**
492 * destroy_rcu_head_on_stack() - destroy on-st
493 * @head: pointer to rcu_head structure to be
494 *
495 * This function informs debugobjects that an
496 * is about to go out of scope. As with init_
497 * function is not required for rcu_head struc
498 * defined or that are dynamically allocated o
499 * init_rcu_head_on_stack(), this function has
500 * !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel build
501 */
502 void destroy_rcu_head_on_stack(struct rcu_head
503 {
504 debug_object_free(head, &rcuhead_debug
505 }
506 EXPORT_SYMBOL_GPL(destroy_rcu_head_on_stack);
507
508 const struct debug_obj_descr rcuhead_debug_des
509 .name = "rcu_head",
510 .is_static_object = rcuhead_is_static_
511 };
512 EXPORT_SYMBOL_GPL(rcuhead_debug_descr);
513 #endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
514
515 #if defined(CONFIG_TREE_RCU) || defined(CONFIG
516 void do_trace_rcu_torture_read(const char *rcu
517 unsigned long s
518 unsigned long c
519 {
520 trace_rcu_torture_read(rcutorturename,
521 }
522 EXPORT_SYMBOL_GPL(do_trace_rcu_torture_read);
523 #else
524 #define do_trace_rcu_torture_read(rcutorturena
525 do { } while (0)
526 #endif
527
528 #if IS_ENABLED(CONFIG_RCU_TORTURE_TEST) || IS_
529 /* Get rcutorture access to sched_setaffinity(
530 long torture_sched_setaffinity(pid_t pid, cons
531 {
532 int ret;
533
534 ret = sched_setaffinity(pid, in_mask);
535 WARN_ONCE(ret, "%s: sched_setaffinity(
536 return ret;
537 }
538 EXPORT_SYMBOL_GPL(torture_sched_setaffinity);
539 #endif
540
541 int rcu_cpu_stall_notifiers __read_mostly; //
542 EXPORT_SYMBOL_GPL(rcu_cpu_stall_notifiers);
543
544 #ifdef CONFIG_RCU_STALL_COMMON
545 int rcu_cpu_stall_ftrace_dump __read_mostly;
546 module_param(rcu_cpu_stall_ftrace_dump, int, 0
547 #ifdef CONFIG_RCU_CPU_STALL_NOTIFIER
548 module_param(rcu_cpu_stall_notifiers, int, 044
549 #endif // #ifdef CONFIG_RCU_CPU_STALL_NOTIFIER
550 int rcu_cpu_stall_suppress __read_mostly; // !
551 EXPORT_SYMBOL_GPL(rcu_cpu_stall_suppress);
552 module_param(rcu_cpu_stall_suppress, int, 0644
553 int rcu_cpu_stall_timeout __read_mostly = CONF
554 module_param(rcu_cpu_stall_timeout, int, 0644)
555 int rcu_exp_cpu_stall_timeout __read_mostly =
556 module_param(rcu_exp_cpu_stall_timeout, int, 0
557 int rcu_cpu_stall_cputime __read_mostly = IS_E
558 module_param(rcu_cpu_stall_cputime, int, 0644)
559 bool rcu_exp_stall_task_details __read_mostly;
560 module_param(rcu_exp_stall_task_details, bool,
561 #endif /* #ifdef CONFIG_RCU_STALL_COMMON */
562
563 // Suppress boot-time RCU CPU stall warnings a
564 // warnings. Also used by rcutorture even if
565 int rcu_cpu_stall_suppress_at_boot __read_most
566 EXPORT_SYMBOL_GPL(rcu_cpu_stall_suppress_at_bo
567 module_param(rcu_cpu_stall_suppress_at_boot, i
568
569 /**
570 * get_completed_synchronize_rcu - Return a pr
571 *
572 * Returns a value that will always be treated
573 * poll_state_synchronize_rcu() as a cookie wh
574 * completed.
575 */
576 unsigned long get_completed_synchronize_rcu(vo
577 {
578 return RCU_GET_STATE_COMPLETED;
579 }
580 EXPORT_SYMBOL_GPL(get_completed_synchronize_rc
581
582 #ifdef CONFIG_PROVE_RCU
583
584 /*
585 * Early boot self test parameters.
586 */
587 static bool rcu_self_test;
588 module_param(rcu_self_test, bool, 0444);
589
590 static int rcu_self_test_counter;
591
592 static void test_callback(struct rcu_head *r)
593 {
594 rcu_self_test_counter++;
595 pr_info("RCU test callback executed %d
596 }
597
598 DEFINE_STATIC_SRCU(early_srcu);
599 static unsigned long early_srcu_cookie;
600
601 struct early_boot_kfree_rcu {
602 struct rcu_head rh;
603 };
604
605 static void early_boot_test_call_rcu(void)
606 {
607 static struct rcu_head head;
608 int idx;
609 static struct rcu_head shead;
610 struct early_boot_kfree_rcu *rhp;
611
612 idx = srcu_down_read(&early_srcu);
613 srcu_up_read(&early_srcu, idx);
614 call_rcu(&head, test_callback);
615 early_srcu_cookie = start_poll_synchro
616 call_srcu(&early_srcu, &shead, test_ca
617 rhp = kmalloc(sizeof(*rhp), GFP_KERNEL
618 if (!WARN_ON_ONCE(!rhp))
619 kfree_rcu(rhp, rh);
620 }
621
622 void rcu_early_boot_tests(void)
623 {
624 pr_info("Running RCU self tests\n");
625
626 if (rcu_self_test)
627 early_boot_test_call_rcu();
628 rcu_test_sync_prims();
629 }
630
631 static int rcu_verify_early_boot_tests(void)
632 {
633 int ret = 0;
634 int early_boot_test_counter = 0;
635
636 if (rcu_self_test) {
637 early_boot_test_counter++;
638 rcu_barrier();
639 early_boot_test_counter++;
640 srcu_barrier(&early_srcu);
641 WARN_ON_ONCE(!poll_state_synch
642 cleanup_srcu_struct(&early_src
643 }
644 if (rcu_self_test_counter != early_boo
645 WARN_ON(1);
646 ret = -1;
647 }
648
649 return ret;
650 }
651 late_initcall(rcu_verify_early_boot_tests);
652 #else
653 void rcu_early_boot_tests(void) {}
654 #endif /* CONFIG_PROVE_RCU */
655
656 #include "tasks.h"
657
658 #ifndef CONFIG_TINY_RCU
659
660 /*
661 * Print any significant non-default boot-time
662 */
663 void __init rcupdate_announce_bootup_oddness(v
664 {
665 if (rcu_normal)
666 pr_info("\tNo expedited grace
667 else if (rcu_normal_after_boot)
668 pr_info("\tNo expedited grace
669 else if (rcu_expedited)
670 pr_info("\tAll grace periods a
671 if (rcu_cpu_stall_suppress)
672 pr_info("\tRCU CPU stall warni
673 if (rcu_cpu_stall_timeout != CONFIG_RC
674 pr_info("\tRCU CPU stall warni
675 rcu_tasks_bootup_oddness();
676 }
677
678 #endif /* #ifndef CONFIG_TINY_RCU */
679
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.