Loading...
1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * Read-Copy Update mechanism for mutual exclusion
4 *
5 * Copyright IBM Corporation, 2001
6 *
7 * Authors: Dipankar Sarma <dipankar@in.ibm.com>
8 * Manfred Spraul <manfred@colorfullife.com>
9 *
10 * Based on the original work by Paul McKenney <paulmck@linux.ibm.com>
11 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
12 * Papers:
13 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
14 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
15 *
16 * For detailed explanation of Read-Copy Update mechanism see -
17 * http://lse.sourceforge.net/locking/rcupdate.html
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/atomic.h>
29#include <linux/bitops.h>
30#include <linux/percpu.h>
31#include <linux/notifier.h>
32#include <linux/cpu.h>
33#include <linux/mutex.h>
34#include <linux/export.h>
35#include <linux/hardirq.h>
36#include <linux/delay.h>
37#include <linux/moduleparam.h>
38#include <linux/kthread.h>
39#include <linux/tick.h>
40#include <linux/rcupdate_wait.h>
41#include <linux/sched/isolation.h>
42#include <linux/kprobes.h>
43#include <linux/slab.h>
44#include <linux/irq_work.h>
45#include <linux/rcupdate_trace.h>
46
47#define CREATE_TRACE_POINTS
48
49#include "rcu.h"
50
51#ifdef MODULE_PARAM_PREFIX
52#undef MODULE_PARAM_PREFIX
53#endif
54#define MODULE_PARAM_PREFIX "rcupdate."
55
56#ifndef CONFIG_TINY_RCU
57module_param(rcu_expedited, int, 0444);
58module_param(rcu_normal, int, 0444);
59static int rcu_normal_after_boot = IS_ENABLED(CONFIG_PREEMPT_RT);
60#if !defined(CONFIG_PREEMPT_RT) || defined(CONFIG_NO_HZ_FULL)
61module_param(rcu_normal_after_boot, int, 0444);
62#endif
63#endif /* #ifndef CONFIG_TINY_RCU */
64
65#ifdef CONFIG_DEBUG_LOCK_ALLOC
66/**
67 * rcu_read_lock_held_common() - might we be in RCU-sched read-side critical section?
68 * @ret: Best guess answer if lockdep cannot be relied on
69 *
70 * Returns true if lockdep must be ignored, in which case ``*ret`` contains
71 * the best guess described below. Otherwise returns false, in which
72 * case ``*ret`` tells the caller nothing and the caller should instead
73 * consult lockdep.
74 *
75 * If CONFIG_DEBUG_LOCK_ALLOC is selected, set ``*ret`` to nonzero iff in an
76 * RCU-sched read-side critical section. In absence of
77 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side
78 * critical section unless it can prove otherwise. Note that disabling
79 * of preemption (including disabling irqs) counts as an RCU-sched
80 * read-side critical section. This is useful for debug checks in functions
81 * that required that they be called within an RCU-sched read-side
82 * critical section.
83 *
84 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot
85 * and while lockdep is disabled.
86 *
87 * Note that if the CPU is in the idle loop from an RCU point of view (ie:
88 * that we are in the section between ct_idle_enter() and ct_idle_exit())
89 * then rcu_read_lock_held() sets ``*ret`` to false even if the CPU did an
90 * rcu_read_lock(). The reason for this is that RCU ignores CPUs that are
91 * in such a section, considering these as in extended quiescent state,
92 * so such a CPU is effectively never in an RCU read-side critical section
93 * regardless of what RCU primitives it invokes. This state of affairs is
94 * required --- we need to keep an RCU-free window in idle where the CPU may
95 * possibly enter into low power mode. This way we can notice an extended
96 * quiescent state to other CPUs that started a grace period. Otherwise
97 * we would delay any grace period as long as we run in the idle task.
98 *
99 * Similarly, we avoid claiming an RCU read lock held if the current
100 * CPU is offline.
101 */
102static bool rcu_read_lock_held_common(bool *ret)
103{
104 if (!debug_lockdep_rcu_enabled()) {
105 *ret = true;
106 return true;
107 }
108 if (!rcu_is_watching()) {
109 *ret = false;
110 return true;
111 }
112 if (!rcu_lockdep_current_cpu_online()) {
113 *ret = false;
114 return true;
115 }
116 return false;
117}
118
119int rcu_read_lock_sched_held(void)
120{
121 bool ret;
122
123 if (rcu_read_lock_held_common(&ret))
124 return ret;
125 return lock_is_held(&rcu_sched_lock_map) || !preemptible();
126}
127EXPORT_SYMBOL(rcu_read_lock_sched_held);
128#endif
129
130#ifndef CONFIG_TINY_RCU
131
132/*
133 * Should expedited grace-period primitives always fall back to their
134 * non-expedited counterparts? Intended for use within RCU. Note
135 * that if the user specifies both rcu_expedited and rcu_normal, then
136 * rcu_normal wins. (Except during the time period during boot from
137 * when the first task is spawned until the rcu_set_runtime_mode()
138 * core_initcall() is invoked, at which point everything is expedited.)
139 */
140bool rcu_gp_is_normal(void)
141{
142 return READ_ONCE(rcu_normal) &&
143 rcu_scheduler_active != RCU_SCHEDULER_INIT;
144}
145EXPORT_SYMBOL_GPL(rcu_gp_is_normal);
146
147static atomic_t rcu_expedited_nesting = ATOMIC_INIT(1);
148
149/*
150 * Should normal grace-period primitives be expedited? Intended for
151 * use within RCU. Note that this function takes the rcu_expedited
152 * sysfs/boot variable and rcu_scheduler_active into account as well
153 * as the rcu_expedite_gp() nesting. So looping on rcu_unexpedite_gp()
154 * until rcu_gp_is_expedited() returns false is a -really- bad idea.
155 */
156bool rcu_gp_is_expedited(void)
157{
158 return rcu_expedited || atomic_read(&rcu_expedited_nesting);
159}
160EXPORT_SYMBOL_GPL(rcu_gp_is_expedited);
161
162/**
163 * rcu_expedite_gp - Expedite future RCU grace periods
164 *
165 * After a call to this function, future calls to synchronize_rcu() and
166 * friends act as the corresponding synchronize_rcu_expedited() function
167 * had instead been called.
168 */
169void rcu_expedite_gp(void)
170{
171 atomic_inc(&rcu_expedited_nesting);
172}
173EXPORT_SYMBOL_GPL(rcu_expedite_gp);
174
175/**
176 * rcu_unexpedite_gp - Cancel prior rcu_expedite_gp() invocation
177 *
178 * Undo a prior call to rcu_expedite_gp(). If all prior calls to
179 * rcu_expedite_gp() are undone by a subsequent call to rcu_unexpedite_gp(),
180 * and if the rcu_expedited sysfs/boot parameter is not set, then all
181 * subsequent calls to synchronize_rcu() and friends will return to
182 * their normal non-expedited behavior.
183 */
184void rcu_unexpedite_gp(void)
185{
186 atomic_dec(&rcu_expedited_nesting);
187}
188EXPORT_SYMBOL_GPL(rcu_unexpedite_gp);
189
190static bool rcu_boot_ended __read_mostly;
191
192/*
193 * Inform RCU of the end of the in-kernel boot sequence.
194 */
195void rcu_end_inkernel_boot(void)
196{
197 rcu_unexpedite_gp();
198 if (rcu_normal_after_boot)
199 WRITE_ONCE(rcu_normal, 1);
200 rcu_boot_ended = true;
201}
202
203/*
204 * Let rcutorture know when it is OK to turn it up to eleven.
205 */
206bool rcu_inkernel_boot_has_ended(void)
207{
208 return rcu_boot_ended;
209}
210EXPORT_SYMBOL_GPL(rcu_inkernel_boot_has_ended);
211
212#endif /* #ifndef CONFIG_TINY_RCU */
213
214/*
215 * Test each non-SRCU synchronous grace-period wait API. This is
216 * useful just after a change in mode for these primitives, and
217 * during early boot.
218 */
219void rcu_test_sync_prims(void)
220{
221 if (!IS_ENABLED(CONFIG_PROVE_RCU))
222 return;
223 synchronize_rcu();
224 synchronize_rcu_expedited();
225}
226
227#if !defined(CONFIG_TINY_RCU)
228
229/*
230 * Switch to run-time mode once RCU has fully initialized.
231 */
232static int __init rcu_set_runtime_mode(void)
233{
234 rcu_test_sync_prims();
235 rcu_scheduler_active = RCU_SCHEDULER_RUNNING;
236 kfree_rcu_scheduler_running();
237 rcu_test_sync_prims();
238 return 0;
239}
240core_initcall(rcu_set_runtime_mode);
241
242#endif /* #if !defined(CONFIG_TINY_RCU) */
243
244#ifdef CONFIG_DEBUG_LOCK_ALLOC
245static struct lock_class_key rcu_lock_key;
246struct lockdep_map rcu_lock_map = {
247 .name = "rcu_read_lock",
248 .key = &rcu_lock_key,
249 .wait_type_outer = LD_WAIT_FREE,
250 .wait_type_inner = LD_WAIT_CONFIG, /* PREEMPT_RT implies PREEMPT_RCU */
251};
252EXPORT_SYMBOL_GPL(rcu_lock_map);
253
254static struct lock_class_key rcu_bh_lock_key;
255struct lockdep_map rcu_bh_lock_map = {
256 .name = "rcu_read_lock_bh",
257 .key = &rcu_bh_lock_key,
258 .wait_type_outer = LD_WAIT_FREE,
259 .wait_type_inner = LD_WAIT_CONFIG, /* PREEMPT_RT makes BH preemptible. */
260};
261EXPORT_SYMBOL_GPL(rcu_bh_lock_map);
262
263static struct lock_class_key rcu_sched_lock_key;
264struct lockdep_map rcu_sched_lock_map = {
265 .name = "rcu_read_lock_sched",
266 .key = &rcu_sched_lock_key,
267 .wait_type_outer = LD_WAIT_FREE,
268 .wait_type_inner = LD_WAIT_SPIN,
269};
270EXPORT_SYMBOL_GPL(rcu_sched_lock_map);
271
272// Tell lockdep when RCU callbacks are being invoked.
273static struct lock_class_key rcu_callback_key;
274struct lockdep_map rcu_callback_map =
275 STATIC_LOCKDEP_MAP_INIT("rcu_callback", &rcu_callback_key);
276EXPORT_SYMBOL_GPL(rcu_callback_map);
277
278noinstr int notrace debug_lockdep_rcu_enabled(void)
279{
280 return rcu_scheduler_active != RCU_SCHEDULER_INACTIVE && READ_ONCE(debug_locks) &&
281 current->lockdep_recursion == 0;
282}
283EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled);
284
285/**
286 * rcu_read_lock_held() - might we be in RCU read-side critical section?
287 *
288 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an RCU
289 * read-side critical section. In absence of CONFIG_DEBUG_LOCK_ALLOC,
290 * this assumes we are in an RCU read-side critical section unless it can
291 * prove otherwise. This is useful for debug checks in functions that
292 * require that they be called within an RCU read-side critical section.
293 *
294 * Checks debug_lockdep_rcu_enabled() to prevent false positives during boot
295 * and while lockdep is disabled.
296 *
297 * Note that rcu_read_lock() and the matching rcu_read_unlock() must
298 * occur in the same context, for example, it is illegal to invoke
299 * rcu_read_unlock() in process context if the matching rcu_read_lock()
300 * was invoked from within an irq handler.
301 *
302 * Note that rcu_read_lock() is disallowed if the CPU is either idle or
303 * offline from an RCU perspective, so check for those as well.
304 */
305int rcu_read_lock_held(void)
306{
307 bool ret;
308
309 if (rcu_read_lock_held_common(&ret))
310 return ret;
311 return lock_is_held(&rcu_lock_map);
312}
313EXPORT_SYMBOL_GPL(rcu_read_lock_held);
314
315/**
316 * rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section?
317 *
318 * Check for bottom half being disabled, which covers both the
319 * CONFIG_PROVE_RCU and not cases. Note that if someone uses
320 * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled)
321 * will show the situation. This is useful for debug checks in functions
322 * that require that they be called within an RCU read-side critical
323 * section.
324 *
325 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot.
326 *
327 * Note that rcu_read_lock_bh() is disallowed if the CPU is either idle or
328 * offline from an RCU perspective, so check for those as well.
329 */
330int rcu_read_lock_bh_held(void)
331{
332 bool ret;
333
334 if (rcu_read_lock_held_common(&ret))
335 return ret;
336 return in_softirq() || irqs_disabled();
337}
338EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held);
339
340int rcu_read_lock_any_held(void)
341{
342 bool ret;
343
344 if (rcu_read_lock_held_common(&ret))
345 return ret;
346 if (lock_is_held(&rcu_lock_map) ||
347 lock_is_held(&rcu_bh_lock_map) ||
348 lock_is_held(&rcu_sched_lock_map))
349 return 1;
350 return !preemptible();
351}
352EXPORT_SYMBOL_GPL(rcu_read_lock_any_held);
353
354#endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
355
356/**
357 * wakeme_after_rcu() - Callback function to awaken a task after grace period
358 * @head: Pointer to rcu_head member within rcu_synchronize structure
359 *
360 * Awaken the corresponding task now that a grace period has elapsed.
361 */
362void wakeme_after_rcu(struct rcu_head *head)
363{
364 struct rcu_synchronize *rcu;
365
366 rcu = container_of(head, struct rcu_synchronize, head);
367 complete(&rcu->completion);
368}
369EXPORT_SYMBOL_GPL(wakeme_after_rcu);
370
371void __wait_rcu_gp(bool checktiny, int n, call_rcu_func_t *crcu_array,
372 struct rcu_synchronize *rs_array)
373{
374 int i;
375 int j;
376
377 /* Initialize and register callbacks for each crcu_array element. */
378 for (i = 0; i < n; i++) {
379 if (checktiny &&
380 (crcu_array[i] == call_rcu)) {
381 might_sleep();
382 continue;
383 }
384 for (j = 0; j < i; j++)
385 if (crcu_array[j] == crcu_array[i])
386 break;
387 if (j == i) {
388 init_rcu_head_on_stack(&rs_array[i].head);
389 init_completion(&rs_array[i].completion);
390 (crcu_array[i])(&rs_array[i].head, wakeme_after_rcu);
391 }
392 }
393
394 /* Wait for all callbacks to be invoked. */
395 for (i = 0; i < n; i++) {
396 if (checktiny &&
397 (crcu_array[i] == call_rcu))
398 continue;
399 for (j = 0; j < i; j++)
400 if (crcu_array[j] == crcu_array[i])
401 break;
402 if (j == i) {
403 wait_for_completion(&rs_array[i].completion);
404 destroy_rcu_head_on_stack(&rs_array[i].head);
405 }
406 }
407}
408EXPORT_SYMBOL_GPL(__wait_rcu_gp);
409
410void finish_rcuwait(struct rcuwait *w)
411{
412 rcu_assign_pointer(w->task, NULL);
413 __set_current_state(TASK_RUNNING);
414}
415EXPORT_SYMBOL_GPL(finish_rcuwait);
416
417#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
418void init_rcu_head(struct rcu_head *head)
419{
420 debug_object_init(head, &rcuhead_debug_descr);
421}
422EXPORT_SYMBOL_GPL(init_rcu_head);
423
424void destroy_rcu_head(struct rcu_head *head)
425{
426 debug_object_free(head, &rcuhead_debug_descr);
427}
428EXPORT_SYMBOL_GPL(destroy_rcu_head);
429
430static bool rcuhead_is_static_object(void *addr)
431{
432 return true;
433}
434
435/**
436 * init_rcu_head_on_stack() - initialize on-stack rcu_head for debugobjects
437 * @head: pointer to rcu_head structure to be initialized
438 *
439 * This function informs debugobjects of a new rcu_head structure that
440 * has been allocated as an auto variable on the stack. This function
441 * is not required for rcu_head structures that are statically defined or
442 * that are dynamically allocated on the heap. This function has no
443 * effect for !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
444 */
445void init_rcu_head_on_stack(struct rcu_head *head)
446{
447 debug_object_init_on_stack(head, &rcuhead_debug_descr);
448}
449EXPORT_SYMBOL_GPL(init_rcu_head_on_stack);
450
451/**
452 * destroy_rcu_head_on_stack() - destroy on-stack rcu_head for debugobjects
453 * @head: pointer to rcu_head structure to be initialized
454 *
455 * This function informs debugobjects that an on-stack rcu_head structure
456 * is about to go out of scope. As with init_rcu_head_on_stack(), this
457 * function is not required for rcu_head structures that are statically
458 * defined or that are dynamically allocated on the heap. Also as with
459 * init_rcu_head_on_stack(), this function has no effect for
460 * !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
461 */
462void destroy_rcu_head_on_stack(struct rcu_head *head)
463{
464 debug_object_free(head, &rcuhead_debug_descr);
465}
466EXPORT_SYMBOL_GPL(destroy_rcu_head_on_stack);
467
468const struct debug_obj_descr rcuhead_debug_descr = {
469 .name = "rcu_head",
470 .is_static_object = rcuhead_is_static_object,
471};
472EXPORT_SYMBOL_GPL(rcuhead_debug_descr);
473#endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
474
475#if defined(CONFIG_TREE_RCU) || defined(CONFIG_RCU_TRACE)
476void do_trace_rcu_torture_read(const char *rcutorturename, struct rcu_head *rhp,
477 unsigned long secs,
478 unsigned long c_old, unsigned long c)
479{
480 trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c);
481}
482EXPORT_SYMBOL_GPL(do_trace_rcu_torture_read);
483#else
484#define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
485 do { } while (0)
486#endif
487
488#if IS_ENABLED(CONFIG_RCU_TORTURE_TEST) || IS_MODULE(CONFIG_RCU_TORTURE_TEST)
489/* Get rcutorture access to sched_setaffinity(). */
490long rcutorture_sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
491{
492 int ret;
493
494 ret = sched_setaffinity(pid, in_mask);
495 WARN_ONCE(ret, "%s: sched_setaffinity() returned %d\n", __func__, ret);
496 return ret;
497}
498EXPORT_SYMBOL_GPL(rcutorture_sched_setaffinity);
499#endif
500
501#ifdef CONFIG_RCU_STALL_COMMON
502int rcu_cpu_stall_ftrace_dump __read_mostly;
503module_param(rcu_cpu_stall_ftrace_dump, int, 0644);
504int rcu_cpu_stall_suppress __read_mostly; // !0 = suppress stall warnings.
505EXPORT_SYMBOL_GPL(rcu_cpu_stall_suppress);
506module_param(rcu_cpu_stall_suppress, int, 0644);
507int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT;
508module_param(rcu_cpu_stall_timeout, int, 0644);
509int rcu_exp_cpu_stall_timeout __read_mostly = CONFIG_RCU_EXP_CPU_STALL_TIMEOUT;
510module_param(rcu_exp_cpu_stall_timeout, int, 0644);
511#endif /* #ifdef CONFIG_RCU_STALL_COMMON */
512
513// Suppress boot-time RCU CPU stall warnings and rcutorture writer stall
514// warnings. Also used by rcutorture even if stall warnings are excluded.
515int rcu_cpu_stall_suppress_at_boot __read_mostly; // !0 = suppress boot stalls.
516EXPORT_SYMBOL_GPL(rcu_cpu_stall_suppress_at_boot);
517module_param(rcu_cpu_stall_suppress_at_boot, int, 0444);
518
519/**
520 * get_completed_synchronize_rcu - Return a pre-completed polled state cookie
521 *
522 * Returns a value that will always be treated by functions like
523 * poll_state_synchronize_rcu() as a cookie whose grace period has already
524 * completed.
525 */
526unsigned long get_completed_synchronize_rcu(void)
527{
528 return RCU_GET_STATE_COMPLETED;
529}
530EXPORT_SYMBOL_GPL(get_completed_synchronize_rcu);
531
532#ifdef CONFIG_PROVE_RCU
533
534/*
535 * Early boot self test parameters.
536 */
537static bool rcu_self_test;
538module_param(rcu_self_test, bool, 0444);
539
540static int rcu_self_test_counter;
541
542static void test_callback(struct rcu_head *r)
543{
544 rcu_self_test_counter++;
545 pr_info("RCU test callback executed %d\n", rcu_self_test_counter);
546}
547
548DEFINE_STATIC_SRCU(early_srcu);
549static unsigned long early_srcu_cookie;
550
551struct early_boot_kfree_rcu {
552 struct rcu_head rh;
553};
554
555static void early_boot_test_call_rcu(void)
556{
557 static struct rcu_head head;
558 static struct rcu_head shead;
559 struct early_boot_kfree_rcu *rhp;
560
561 call_rcu(&head, test_callback);
562 early_srcu_cookie = start_poll_synchronize_srcu(&early_srcu);
563 call_srcu(&early_srcu, &shead, test_callback);
564 rhp = kmalloc(sizeof(*rhp), GFP_KERNEL);
565 if (!WARN_ON_ONCE(!rhp))
566 kfree_rcu(rhp, rh);
567}
568
569void rcu_early_boot_tests(void)
570{
571 pr_info("Running RCU self tests\n");
572
573 if (rcu_self_test)
574 early_boot_test_call_rcu();
575 rcu_test_sync_prims();
576}
577
578static int rcu_verify_early_boot_tests(void)
579{
580 int ret = 0;
581 int early_boot_test_counter = 0;
582
583 if (rcu_self_test) {
584 early_boot_test_counter++;
585 rcu_barrier();
586 early_boot_test_counter++;
587 srcu_barrier(&early_srcu);
588 WARN_ON_ONCE(!poll_state_synchronize_srcu(&early_srcu, early_srcu_cookie));
589 }
590 if (rcu_self_test_counter != early_boot_test_counter) {
591 WARN_ON(1);
592 ret = -1;
593 }
594
595 return ret;
596}
597late_initcall(rcu_verify_early_boot_tests);
598#else
599void rcu_early_boot_tests(void) {}
600#endif /* CONFIG_PROVE_RCU */
601
602#include "tasks.h"
603
604#ifndef CONFIG_TINY_RCU
605
606/*
607 * Print any significant non-default boot-time settings.
608 */
609void __init rcupdate_announce_bootup_oddness(void)
610{
611 if (rcu_normal)
612 pr_info("\tNo expedited grace period (rcu_normal).\n");
613 else if (rcu_normal_after_boot)
614 pr_info("\tNo expedited grace period (rcu_normal_after_boot).\n");
615 else if (rcu_expedited)
616 pr_info("\tAll grace periods are expedited (rcu_expedited).\n");
617 if (rcu_cpu_stall_suppress)
618 pr_info("\tRCU CPU stall warnings suppressed (rcu_cpu_stall_suppress).\n");
619 if (rcu_cpu_stall_timeout != CONFIG_RCU_CPU_STALL_TIMEOUT)
620 pr_info("\tRCU CPU stall warnings timeout set to %d (rcu_cpu_stall_timeout).\n", rcu_cpu_stall_timeout);
621 rcu_tasks_bootup_oddness();
622}
623
624#endif /* #ifndef CONFIG_TINY_RCU */
1/*
2 * 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.
17 *
18 * Copyright IBM Corporation, 2001
19 *
20 * Authors: Dipankar Sarma <dipankar@in.ibm.com>
21 * Manfred Spraul <manfred@colorfullife.com>
22 *
23 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
24 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
25 * Papers:
26 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
27 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
28 *
29 * For detailed explanation of Read-Copy Update mechanism see -
30 * http://lse.sourceforge.net/locking/rcupdate.html
31 *
32 */
33#include <linux/types.h>
34#include <linux/kernel.h>
35#include <linux/init.h>
36#include <linux/spinlock.h>
37#include <linux/smp.h>
38#include <linux/interrupt.h>
39#include <linux/sched.h>
40#include <linux/atomic.h>
41#include <linux/bitops.h>
42#include <linux/percpu.h>
43#include <linux/notifier.h>
44#include <linux/cpu.h>
45#include <linux/mutex.h>
46#include <linux/export.h>
47#include <linux/hardirq.h>
48#include <linux/delay.h>
49#include <linux/module.h>
50
51#define CREATE_TRACE_POINTS
52
53#include "rcu.h"
54
55MODULE_ALIAS("rcupdate");
56#ifdef MODULE_PARAM_PREFIX
57#undef MODULE_PARAM_PREFIX
58#endif
59#define MODULE_PARAM_PREFIX "rcupdate."
60
61module_param(rcu_expedited, int, 0);
62
63#ifdef CONFIG_PREEMPT_RCU
64
65/*
66 * Preemptible RCU implementation for rcu_read_lock().
67 * Just increment ->rcu_read_lock_nesting, shared state will be updated
68 * if we block.
69 */
70void __rcu_read_lock(void)
71{
72 current->rcu_read_lock_nesting++;
73 barrier(); /* critical section after entry code. */
74}
75EXPORT_SYMBOL_GPL(__rcu_read_lock);
76
77/*
78 * Preemptible RCU implementation for rcu_read_unlock().
79 * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost
80 * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then
81 * invoke rcu_read_unlock_special() to clean up after a context switch
82 * in an RCU read-side critical section and other special cases.
83 */
84void __rcu_read_unlock(void)
85{
86 struct task_struct *t = current;
87
88 if (t->rcu_read_lock_nesting != 1) {
89 --t->rcu_read_lock_nesting;
90 } else {
91 barrier(); /* critical section before exit code. */
92 t->rcu_read_lock_nesting = INT_MIN;
93#ifdef CONFIG_PROVE_RCU_DELAY
94 udelay(10); /* Make preemption more probable. */
95#endif /* #ifdef CONFIG_PROVE_RCU_DELAY */
96 barrier(); /* assign before ->rcu_read_unlock_special load */
97 if (unlikely(ACCESS_ONCE(t->rcu_read_unlock_special)))
98 rcu_read_unlock_special(t);
99 barrier(); /* ->rcu_read_unlock_special load before assign */
100 t->rcu_read_lock_nesting = 0;
101 }
102#ifdef CONFIG_PROVE_LOCKING
103 {
104 int rrln = ACCESS_ONCE(t->rcu_read_lock_nesting);
105
106 WARN_ON_ONCE(rrln < 0 && rrln > INT_MIN / 2);
107 }
108#endif /* #ifdef CONFIG_PROVE_LOCKING */
109}
110EXPORT_SYMBOL_GPL(__rcu_read_unlock);
111
112#endif /* #ifdef CONFIG_PREEMPT_RCU */
113
114#ifdef CONFIG_DEBUG_LOCK_ALLOC
115static struct lock_class_key rcu_lock_key;
116struct lockdep_map rcu_lock_map =
117 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key);
118EXPORT_SYMBOL_GPL(rcu_lock_map);
119
120static struct lock_class_key rcu_bh_lock_key;
121struct lockdep_map rcu_bh_lock_map =
122 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_bh", &rcu_bh_lock_key);
123EXPORT_SYMBOL_GPL(rcu_bh_lock_map);
124
125static struct lock_class_key rcu_sched_lock_key;
126struct lockdep_map rcu_sched_lock_map =
127 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_sched", &rcu_sched_lock_key);
128EXPORT_SYMBOL_GPL(rcu_sched_lock_map);
129
130static struct lock_class_key rcu_callback_key;
131struct lockdep_map rcu_callback_map =
132 STATIC_LOCKDEP_MAP_INIT("rcu_callback", &rcu_callback_key);
133EXPORT_SYMBOL_GPL(rcu_callback_map);
134
135int notrace debug_lockdep_rcu_enabled(void)
136{
137 return rcu_scheduler_active && debug_locks &&
138 current->lockdep_recursion == 0;
139}
140EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled);
141
142/**
143 * rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section?
144 *
145 * Check for bottom half being disabled, which covers both the
146 * CONFIG_PROVE_RCU and not cases. Note that if someone uses
147 * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled)
148 * will show the situation. This is useful for debug checks in functions
149 * that require that they be called within an RCU read-side critical
150 * section.
151 *
152 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot.
153 *
154 * Note that rcu_read_lock() is disallowed if the CPU is either idle or
155 * offline from an RCU perspective, so check for those as well.
156 */
157int rcu_read_lock_bh_held(void)
158{
159 if (!debug_lockdep_rcu_enabled())
160 return 1;
161 if (!rcu_is_watching())
162 return 0;
163 if (!rcu_lockdep_current_cpu_online())
164 return 0;
165 return in_softirq() || irqs_disabled();
166}
167EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held);
168
169#endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
170
171struct rcu_synchronize {
172 struct rcu_head head;
173 struct completion completion;
174};
175
176/*
177 * Awaken the corresponding synchronize_rcu() instance now that a
178 * grace period has elapsed.
179 */
180static void wakeme_after_rcu(struct rcu_head *head)
181{
182 struct rcu_synchronize *rcu;
183
184 rcu = container_of(head, struct rcu_synchronize, head);
185 complete(&rcu->completion);
186}
187
188void wait_rcu_gp(call_rcu_func_t crf)
189{
190 struct rcu_synchronize rcu;
191
192 init_rcu_head_on_stack(&rcu.head);
193 init_completion(&rcu.completion);
194 /* Will wake me after RCU finished. */
195 crf(&rcu.head, wakeme_after_rcu);
196 /* Wait for it. */
197 wait_for_completion(&rcu.completion);
198 destroy_rcu_head_on_stack(&rcu.head);
199}
200EXPORT_SYMBOL_GPL(wait_rcu_gp);
201
202#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
203static inline void debug_init_rcu_head(struct rcu_head *head)
204{
205 debug_object_init(head, &rcuhead_debug_descr);
206}
207
208static inline void debug_rcu_head_free(struct rcu_head *head)
209{
210 debug_object_free(head, &rcuhead_debug_descr);
211}
212
213/*
214 * fixup_activate is called when:
215 * - an active object is activated
216 * - an unknown object is activated (might be a statically initialized object)
217 * Activation is performed internally by call_rcu().
218 */
219static int rcuhead_fixup_activate(void *addr, enum debug_obj_state state)
220{
221 struct rcu_head *head = addr;
222
223 switch (state) {
224
225 case ODEBUG_STATE_NOTAVAILABLE:
226 /*
227 * This is not really a fixup. We just make sure that it is
228 * tracked in the object tracker.
229 */
230 debug_object_init(head, &rcuhead_debug_descr);
231 debug_object_activate(head, &rcuhead_debug_descr);
232 return 0;
233 default:
234 return 1;
235 }
236}
237
238/**
239 * init_rcu_head_on_stack() - initialize on-stack rcu_head for debugobjects
240 * @head: pointer to rcu_head structure to be initialized
241 *
242 * This function informs debugobjects of a new rcu_head structure that
243 * has been allocated as an auto variable on the stack. This function
244 * is not required for rcu_head structures that are statically defined or
245 * that are dynamically allocated on the heap. This function has no
246 * effect for !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
247 */
248void init_rcu_head_on_stack(struct rcu_head *head)
249{
250 debug_object_init_on_stack(head, &rcuhead_debug_descr);
251}
252EXPORT_SYMBOL_GPL(init_rcu_head_on_stack);
253
254/**
255 * destroy_rcu_head_on_stack() - destroy on-stack rcu_head for debugobjects
256 * @head: pointer to rcu_head structure to be initialized
257 *
258 * This function informs debugobjects that an on-stack rcu_head structure
259 * is about to go out of scope. As with init_rcu_head_on_stack(), this
260 * function is not required for rcu_head structures that are statically
261 * defined or that are dynamically allocated on the heap. Also as with
262 * init_rcu_head_on_stack(), this function has no effect for
263 * !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
264 */
265void destroy_rcu_head_on_stack(struct rcu_head *head)
266{
267 debug_object_free(head, &rcuhead_debug_descr);
268}
269EXPORT_SYMBOL_GPL(destroy_rcu_head_on_stack);
270
271struct debug_obj_descr rcuhead_debug_descr = {
272 .name = "rcu_head",
273 .fixup_activate = rcuhead_fixup_activate,
274};
275EXPORT_SYMBOL_GPL(rcuhead_debug_descr);
276#endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
277
278#if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU) || defined(CONFIG_RCU_TRACE)
279void do_trace_rcu_torture_read(const char *rcutorturename, struct rcu_head *rhp,
280 unsigned long secs,
281 unsigned long c_old, unsigned long c)
282{
283 trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c);
284}
285EXPORT_SYMBOL_GPL(do_trace_rcu_torture_read);
286#else
287#define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
288 do { } while (0)
289#endif
290
291#ifdef CONFIG_RCU_STALL_COMMON
292
293#ifdef CONFIG_PROVE_RCU
294#define RCU_STALL_DELAY_DELTA (5 * HZ)
295#else
296#define RCU_STALL_DELAY_DELTA 0
297#endif
298
299int rcu_cpu_stall_suppress __read_mostly; /* 1 = suppress stall warnings. */
300static int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT;
301
302module_param(rcu_cpu_stall_suppress, int, 0644);
303module_param(rcu_cpu_stall_timeout, int, 0644);
304
305int rcu_jiffies_till_stall_check(void)
306{
307 int till_stall_check = ACCESS_ONCE(rcu_cpu_stall_timeout);
308
309 /*
310 * Limit check must be consistent with the Kconfig limits
311 * for CONFIG_RCU_CPU_STALL_TIMEOUT.
312 */
313 if (till_stall_check < 3) {
314 ACCESS_ONCE(rcu_cpu_stall_timeout) = 3;
315 till_stall_check = 3;
316 } else if (till_stall_check > 300) {
317 ACCESS_ONCE(rcu_cpu_stall_timeout) = 300;
318 till_stall_check = 300;
319 }
320 return till_stall_check * HZ + RCU_STALL_DELAY_DELTA;
321}
322
323static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr)
324{
325 rcu_cpu_stall_suppress = 1;
326 return NOTIFY_DONE;
327}
328
329static struct notifier_block rcu_panic_block = {
330 .notifier_call = rcu_panic,
331};
332
333static int __init check_cpu_stall_init(void)
334{
335 atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block);
336 return 0;
337}
338early_initcall(check_cpu_stall_init);
339
340#endif /* #ifdef CONFIG_RCU_STALL_COMMON */