]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blob - kernel/rcu/update.c
Merge tag 'kvmarm-fixes-5.10-2' of git://git.kernel.org/pub/scm/linux/kernel/git...
[mirror_ubuntu-hirsute-kernel.git] / kernel / rcu / update.c
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
57 module_param(rcu_expedited, int, 0);
58 module_param(rcu_normal, int, 0);
59 static int rcu_normal_after_boot;
60 module_param(rcu_normal_after_boot, int, 0);
61 #endif /* #ifndef CONFIG_TINY_RCU */
62
63 #ifdef CONFIG_DEBUG_LOCK_ALLOC
64 /**
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
67 *
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
74 * RCU-sched read-side critical section. In absence of
75 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side
76 * critical section unless it can prove otherwise. Note that disabling
77 * of preemption (including disabling irqs) counts as an RCU-sched
78 * read-side critical section. This is useful for debug checks in functions
79 * that required that they be called within an RCU-sched read-side
80 * critical section.
81 *
82 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot
83 * and while lockdep is disabled.
84 *
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.
96 *
97 * Similarly, we avoid claiming an RCU read lock held if the current
98 * CPU is offline.
99 */
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
117 int rcu_read_lock_sched_held(void)
118 {
119 bool ret;
120
121 if (rcu_read_lock_held_common(&ret))
122 return ret;
123 return lock_is_held(&rcu_sched_lock_map) || !preemptible();
124 }
125 EXPORT_SYMBOL(rcu_read_lock_sched_held);
126 #endif
127
128 #ifndef CONFIG_TINY_RCU
129
130 /*
131 * Should expedited grace-period primitives always fall back to their
132 * non-expedited counterparts? Intended for use within RCU. Note
133 * that if the user specifies both rcu_expedited and rcu_normal, then
134 * rcu_normal wins. (Except during the time period during boot from
135 * when the first task is spawned until the rcu_set_runtime_mode()
136 * core_initcall() is invoked, at which point everything is expedited.)
137 */
138 bool rcu_gp_is_normal(void)
139 {
140 return READ_ONCE(rcu_normal) &&
141 rcu_scheduler_active != RCU_SCHEDULER_INIT;
142 }
143 EXPORT_SYMBOL_GPL(rcu_gp_is_normal);
144
145 static atomic_t rcu_expedited_nesting = ATOMIC_INIT(1);
146
147 /*
148 * Should normal grace-period primitives be expedited? Intended for
149 * use within RCU. Note that this function takes the rcu_expedited
150 * sysfs/boot variable and rcu_scheduler_active into account as well
151 * as the rcu_expedite_gp() nesting. So looping on rcu_unexpedite_gp()
152 * until rcu_gp_is_expedited() returns false is a -really- bad idea.
153 */
154 bool rcu_gp_is_expedited(void)
155 {
156 return rcu_expedited || atomic_read(&rcu_expedited_nesting);
157 }
158 EXPORT_SYMBOL_GPL(rcu_gp_is_expedited);
159
160 /**
161 * rcu_expedite_gp - Expedite future RCU grace periods
162 *
163 * After a call to this function, future calls to synchronize_rcu() and
164 * friends act as the corresponding synchronize_rcu_expedited() function
165 * had instead been called.
166 */
167 void rcu_expedite_gp(void)
168 {
169 atomic_inc(&rcu_expedited_nesting);
170 }
171 EXPORT_SYMBOL_GPL(rcu_expedite_gp);
172
173 /**
174 * rcu_unexpedite_gp - Cancel prior rcu_expedite_gp() invocation
175 *
176 * Undo a prior call to rcu_expedite_gp(). If all prior calls to
177 * rcu_expedite_gp() are undone by a subsequent call to rcu_unexpedite_gp(),
178 * and if the rcu_expedited sysfs/boot parameter is not set, then all
179 * subsequent calls to synchronize_rcu() and friends will return to
180 * their normal non-expedited behavior.
181 */
182 void rcu_unexpedite_gp(void)
183 {
184 atomic_dec(&rcu_expedited_nesting);
185 }
186 EXPORT_SYMBOL_GPL(rcu_unexpedite_gp);
187
188 static bool rcu_boot_ended __read_mostly;
189
190 /*
191 * Inform RCU of the end of the in-kernel boot sequence.
192 */
193 void rcu_end_inkernel_boot(void)
194 {
195 rcu_unexpedite_gp();
196 if (rcu_normal_after_boot)
197 WRITE_ONCE(rcu_normal, 1);
198 rcu_boot_ended = true;
199 }
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);
209
210 #endif /* #ifndef CONFIG_TINY_RCU */
211
212 /*
213 * Test each non-SRCU synchronous grace-period wait API. This is
214 * useful just after a change in mode for these primitives, and
215 * during early boot.
216 */
217 void rcu_test_sync_prims(void)
218 {
219 if (!IS_ENABLED(CONFIG_PROVE_RCU))
220 return;
221 synchronize_rcu();
222 synchronize_rcu_expedited();
223 }
224
225 #if !defined(CONFIG_TINY_RCU) || defined(CONFIG_SRCU)
226
227 /*
228 * Switch to run-time mode once RCU has fully initialized.
229 */
230 static int __init rcu_set_runtime_mode(void)
231 {
232 rcu_test_sync_prims();
233 rcu_scheduler_active = RCU_SCHEDULER_RUNNING;
234 kfree_rcu_scheduler_running();
235 rcu_test_sync_prims();
236 return 0;
237 }
238 core_initcall(rcu_set_runtime_mode);
239
240 #endif /* #if !defined(CONFIG_TINY_RCU) || defined(CONFIG_SRCU) */
241
242 #ifdef CONFIG_DEBUG_LOCK_ALLOC
243 static struct lock_class_key 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 };
250 EXPORT_SYMBOL_GPL(rcu_lock_map);
251
252 static struct lock_class_key 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 };
259 EXPORT_SYMBOL_GPL(rcu_bh_lock_map);
260
261 static struct lock_class_key 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 };
268 EXPORT_SYMBOL_GPL(rcu_sched_lock_map);
269
270 // Tell lockdep when RCU callbacks are being invoked.
271 static struct lock_class_key rcu_callback_key;
272 struct lockdep_map rcu_callback_map =
273 STATIC_LOCKDEP_MAP_INIT("rcu_callback", &rcu_callback_key);
274 EXPORT_SYMBOL_GPL(rcu_callback_map);
275
276 noinstr int notrace debug_lockdep_rcu_enabled(void)
277 {
278 return rcu_scheduler_active != RCU_SCHEDULER_INACTIVE && debug_locks &&
279 current->lockdep_recursion == 0;
280 }
281 EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled);
282
283 /**
284 * rcu_read_lock_held() - might we be in RCU read-side critical section?
285 *
286 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an RCU
287 * read-side critical section. In absence of CONFIG_DEBUG_LOCK_ALLOC,
288 * this assumes we are in an RCU read-side critical section unless it can
289 * prove otherwise. This is useful for debug checks in functions that
290 * require that they be called within an RCU read-side critical section.
291 *
292 * Checks debug_lockdep_rcu_enabled() to prevent false positives during boot
293 * and while lockdep is disabled.
294 *
295 * Note that rcu_read_lock() and the matching rcu_read_unlock() must
296 * occur in the same context, for example, it is illegal to invoke
297 * rcu_read_unlock() in process context if the matching rcu_read_lock()
298 * was invoked from within an irq handler.
299 *
300 * Note that rcu_read_lock() is disallowed if the CPU is either idle or
301 * offline from an RCU perspective, so check for those as well.
302 */
303 int rcu_read_lock_held(void)
304 {
305 bool ret;
306
307 if (rcu_read_lock_held_common(&ret))
308 return ret;
309 return lock_is_held(&rcu_lock_map);
310 }
311 EXPORT_SYMBOL_GPL(rcu_read_lock_held);
312
313 /**
314 * rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section?
315 *
316 * Check for bottom half being disabled, which covers both the
317 * CONFIG_PROVE_RCU and not cases. Note that if someone uses
318 * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled)
319 * will show the situation. This is useful for debug checks in functions
320 * that require that they be called within an RCU read-side critical
321 * section.
322 *
323 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot.
324 *
325 * Note that rcu_read_lock_bh() is disallowed if the CPU is either idle or
326 * offline from an RCU perspective, so check for those as well.
327 */
328 int rcu_read_lock_bh_held(void)
329 {
330 bool ret;
331
332 if (rcu_read_lock_held_common(&ret))
333 return ret;
334 return in_softirq() || irqs_disabled();
335 }
336 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);
351
352 #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
353
354 /**
355 * wakeme_after_rcu() - Callback function to awaken a task after grace period
356 * @head: Pointer to rcu_head member within rcu_synchronize structure
357 *
358 * Awaken the corresponding task now that a grace period has elapsed.
359 */
360 void wakeme_after_rcu(struct rcu_head *head)
361 {
362 struct rcu_synchronize *rcu;
363
364 rcu = container_of(head, struct rcu_synchronize, head);
365 complete(&rcu->completion);
366 }
367 EXPORT_SYMBOL_GPL(wakeme_after_rcu);
368
369 void __wait_rcu_gp(bool checktiny, int n, call_rcu_func_t *crcu_array,
370 struct rcu_synchronize *rs_array)
371 {
372 int i;
373 int j;
374
375 /* Initialize and register callbacks for each crcu_array element. */
376 for (i = 0; i < n; i++) {
377 if (checktiny &&
378 (crcu_array[i] == call_rcu)) {
379 might_sleep();
380 continue;
381 }
382 for (j = 0; j < i; j++)
383 if (crcu_array[j] == crcu_array[i])
384 break;
385 if (j == i) {
386 init_rcu_head_on_stack(&rs_array[i].head);
387 init_completion(&rs_array[i].completion);
388 (crcu_array[i])(&rs_array[i].head, wakeme_after_rcu);
389 }
390 }
391
392 /* Wait for all callbacks to be invoked. */
393 for (i = 0; i < n; i++) {
394 if (checktiny &&
395 (crcu_array[i] == call_rcu))
396 continue;
397 for (j = 0; j < i; j++)
398 if (crcu_array[j] == crcu_array[i])
399 break;
400 if (j == i) {
401 wait_for_completion(&rs_array[i].completion);
402 destroy_rcu_head_on_stack(&rs_array[i].head);
403 }
404 }
405 }
406 EXPORT_SYMBOL_GPL(__wait_rcu_gp);
407
408 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
409 void init_rcu_head(struct rcu_head *head)
410 {
411 debug_object_init(head, &rcuhead_debug_descr);
412 }
413 EXPORT_SYMBOL_GPL(init_rcu_head);
414
415 void destroy_rcu_head(struct rcu_head *head)
416 {
417 debug_object_free(head, &rcuhead_debug_descr);
418 }
419 EXPORT_SYMBOL_GPL(destroy_rcu_head);
420
421 static bool rcuhead_is_static_object(void *addr)
422 {
423 return true;
424 }
425
426 /**
427 * init_rcu_head_on_stack() - initialize on-stack rcu_head for debugobjects
428 * @head: pointer to rcu_head structure to be initialized
429 *
430 * This function informs debugobjects of a new rcu_head structure that
431 * has been allocated as an auto variable on the stack. This function
432 * is not required for rcu_head structures that are statically defined or
433 * that are dynamically allocated on the heap. This function has no
434 * effect for !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
435 */
436 void init_rcu_head_on_stack(struct rcu_head *head)
437 {
438 debug_object_init_on_stack(head, &rcuhead_debug_descr);
439 }
440 EXPORT_SYMBOL_GPL(init_rcu_head_on_stack);
441
442 /**
443 * destroy_rcu_head_on_stack() - destroy on-stack rcu_head for debugobjects
444 * @head: pointer to rcu_head structure to be initialized
445 *
446 * This function informs debugobjects that an on-stack rcu_head structure
447 * is about to go out of scope. As with init_rcu_head_on_stack(), this
448 * function is not required for rcu_head structures that are statically
449 * defined or that are dynamically allocated on the heap. Also as with
450 * init_rcu_head_on_stack(), this function has no effect for
451 * !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
452 */
453 void destroy_rcu_head_on_stack(struct rcu_head *head)
454 {
455 debug_object_free(head, &rcuhead_debug_descr);
456 }
457 EXPORT_SYMBOL_GPL(destroy_rcu_head_on_stack);
458
459 const struct debug_obj_descr rcuhead_debug_descr = {
460 .name = "rcu_head",
461 .is_static_object = rcuhead_is_static_object,
462 };
463 EXPORT_SYMBOL_GPL(rcuhead_debug_descr);
464 #endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
465
466 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_RCU_TRACE)
467 void do_trace_rcu_torture_read(const char *rcutorturename, struct rcu_head *rhp,
468 unsigned long secs,
469 unsigned long c_old, unsigned long c)
470 {
471 trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c);
472 }
473 EXPORT_SYMBOL_GPL(do_trace_rcu_torture_read);
474 #else
475 #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
476 do { } while (0)
477 #endif
478
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;
484
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);
490 #endif
491
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.
496 EXPORT_SYMBOL_GPL(rcu_cpu_stall_suppress);
497 module_param(rcu_cpu_stall_suppress, int, 0644);
498 int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT;
499 module_param(rcu_cpu_stall_timeout, int, 0644);
500 #endif /* #ifdef CONFIG_RCU_STALL_COMMON */
501
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);
507
508 #ifdef CONFIG_PROVE_RCU
509
510 /*
511 * Early boot self test parameters.
512 */
513 static bool rcu_self_test;
514 module_param(rcu_self_test, bool, 0444);
515
516 static int rcu_self_test_counter;
517
518 static void test_callback(struct rcu_head *r)
519 {
520 rcu_self_test_counter++;
521 pr_info("RCU test callback executed %d\n", rcu_self_test_counter);
522 }
523
524 DEFINE_STATIC_SRCU(early_srcu);
525
526 struct early_boot_kfree_rcu {
527 struct rcu_head rh;
528 };
529
530 static void early_boot_test_call_rcu(void)
531 {
532 static struct rcu_head head;
533 static struct rcu_head shead;
534 struct early_boot_kfree_rcu *rhp;
535
536 call_rcu(&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);
542 }
543
544 void rcu_early_boot_tests(void)
545 {
546 pr_info("Running RCU self tests\n");
547
548 if (rcu_self_test)
549 early_boot_test_call_rcu();
550 rcu_test_sync_prims();
551 }
552
553 static int rcu_verify_early_boot_tests(void)
554 {
555 int ret = 0;
556 int early_boot_test_counter = 0;
557
558 if (rcu_self_test) {
559 early_boot_test_counter++;
560 rcu_barrier();
561 if (IS_ENABLED(CONFIG_SRCU)) {
562 early_boot_test_counter++;
563 srcu_barrier(&early_srcu);
564 }
565 }
566 if (rcu_self_test_counter != early_boot_test_counter) {
567 WARN_ON(1);
568 ret = -1;
569 }
570
571 return ret;
572 }
573 late_initcall(rcu_verify_early_boot_tests);
574 #else
575 void rcu_early_boot_tests(void) {}
576 #endif /* CONFIG_PROVE_RCU */
577
578 #include "tasks.h"
579
580 #ifndef CONFIG_TINY_RCU
581
582 /*
583 * Print any significant non-default boot-time settings.
584 */
585 void __init rcupdate_announce_bootup_oddness(void)
586 {
587 if (rcu_normal)
588 pr_info("\tNo expedited grace period (rcu_normal).\n");
589 else if (rcu_normal_after_boot)
590 pr_info("\tNo expedited grace period (rcu_normal_after_boot).\n");
591 else if (rcu_expedited)
592 pr_info("\tAll grace periods are expedited (rcu_expedited).\n");
593 if (rcu_cpu_stall_suppress)
594 pr_info("\tRCU CPU stall warnings suppressed (rcu_cpu_stall_suppress).\n");
595 if (rcu_cpu_stall_timeout != CONFIG_RCU_CPU_STALL_TIMEOUT)
596 pr_info("\tRCU CPU stall warnings timeout set to %d (rcu_cpu_stall_timeout).\n", rcu_cpu_stall_timeout);
597 rcu_tasks_bootup_oddness();
598 }
599
600 #endif /* #ifndef CONFIG_TINY_RCU */