Commit | Line | Data |
---|---|---|
621934ee PM |
1 | /* |
2 | * Sleepable 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 | |
87de1cfd PM |
15 | * along with this program; if not, you can access it online at |
16 | * http://www.gnu.org/licenses/gpl-2.0.html. | |
621934ee PM |
17 | * |
18 | * Copyright (C) IBM Corporation, 2006 | |
4e87b2d7 | 19 | * Copyright (C) Fujitsu, 2012 |
621934ee PM |
20 | * |
21 | * Author: Paul McKenney <paulmck@us.ibm.com> | |
4e87b2d7 | 22 | * Lai Jiangshan <laijs@cn.fujitsu.com> |
621934ee PM |
23 | * |
24 | * For detailed explanation of Read-Copy Update mechanism see - | |
25 | * Documentation/RCU/ *.txt | |
26 | * | |
27 | */ | |
28 | ||
9984de1a | 29 | #include <linux/export.h> |
621934ee PM |
30 | #include <linux/mutex.h> |
31 | #include <linux/percpu.h> | |
32 | #include <linux/preempt.h> | |
33 | #include <linux/rcupdate.h> | |
34 | #include <linux/sched.h> | |
621934ee | 35 | #include <linux/smp.h> |
46fdb093 | 36 | #include <linux/delay.h> |
621934ee PM |
37 | #include <linux/srcu.h> |
38 | ||
3705b88d AM |
39 | #include "rcu.h" |
40 | ||
931ea9d1 LJ |
41 | /* |
42 | * Initialize an rcu_batch structure to empty. | |
43 | */ | |
44 | static inline void rcu_batch_init(struct rcu_batch *b) | |
45 | { | |
46 | b->head = NULL; | |
47 | b->tail = &b->head; | |
48 | } | |
49 | ||
50 | /* | |
51 | * Enqueue a callback onto the tail of the specified rcu_batch structure. | |
52 | */ | |
53 | static inline void rcu_batch_queue(struct rcu_batch *b, struct rcu_head *head) | |
54 | { | |
55 | *b->tail = head; | |
56 | b->tail = &head->next; | |
57 | } | |
58 | ||
59 | /* | |
60 | * Is the specified rcu_batch structure empty? | |
61 | */ | |
62 | static inline bool rcu_batch_empty(struct rcu_batch *b) | |
63 | { | |
64 | return b->tail == &b->head; | |
65 | } | |
66 | ||
67 | /* | |
68 | * Remove the callback at the head of the specified rcu_batch structure | |
69 | * and return a pointer to it, or return NULL if the structure is empty. | |
70 | */ | |
71 | static inline struct rcu_head *rcu_batch_dequeue(struct rcu_batch *b) | |
72 | { | |
73 | struct rcu_head *head; | |
74 | ||
75 | if (rcu_batch_empty(b)) | |
76 | return NULL; | |
77 | ||
78 | head = b->head; | |
79 | b->head = head->next; | |
80 | if (b->tail == &head->next) | |
81 | rcu_batch_init(b); | |
82 | ||
83 | return head; | |
84 | } | |
85 | ||
86 | /* | |
87 | * Move all callbacks from the rcu_batch structure specified by "from" to | |
88 | * the structure specified by "to". | |
89 | */ | |
90 | static inline void rcu_batch_move(struct rcu_batch *to, struct rcu_batch *from) | |
91 | { | |
92 | if (!rcu_batch_empty(from)) { | |
93 | *to->tail = from->head; | |
94 | to->tail = from->tail; | |
95 | rcu_batch_init(from); | |
96 | } | |
97 | } | |
98 | ||
632ee200 PM |
99 | static int init_srcu_struct_fields(struct srcu_struct *sp) |
100 | { | |
101 | sp->completed = 0; | |
931ea9d1 LJ |
102 | spin_lock_init(&sp->queue_lock); |
103 | sp->running = false; | |
104 | rcu_batch_init(&sp->batch_queue); | |
105 | rcu_batch_init(&sp->batch_check0); | |
106 | rcu_batch_init(&sp->batch_check1); | |
107 | rcu_batch_init(&sp->batch_done); | |
108 | INIT_DELAYED_WORK(&sp->work, process_srcu); | |
632ee200 PM |
109 | sp->per_cpu_ref = alloc_percpu(struct srcu_struct_array); |
110 | return sp->per_cpu_ref ? 0 : -ENOMEM; | |
111 | } | |
112 | ||
113 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
114 | ||
115 | int __init_srcu_struct(struct srcu_struct *sp, const char *name, | |
116 | struct lock_class_key *key) | |
117 | { | |
632ee200 PM |
118 | /* Don't re-initialize a lock while it is held. */ |
119 | debug_check_no_locks_freed((void *)sp, sizeof(*sp)); | |
120 | lockdep_init_map(&sp->dep_map, name, key, 0); | |
632ee200 PM |
121 | return init_srcu_struct_fields(sp); |
122 | } | |
123 | EXPORT_SYMBOL_GPL(__init_srcu_struct); | |
124 | ||
125 | #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ | |
126 | ||
621934ee PM |
127 | /** |
128 | * init_srcu_struct - initialize a sleep-RCU structure | |
129 | * @sp: structure to initialize. | |
130 | * | |
131 | * Must invoke this on a given srcu_struct before passing that srcu_struct | |
132 | * to any other function. Each srcu_struct represents a separate domain | |
133 | * of SRCU protection. | |
134 | */ | |
e6a92013 | 135 | int init_srcu_struct(struct srcu_struct *sp) |
621934ee | 136 | { |
632ee200 | 137 | return init_srcu_struct_fields(sp); |
621934ee | 138 | } |
0cd397d3 | 139 | EXPORT_SYMBOL_GPL(init_srcu_struct); |
621934ee | 140 | |
632ee200 PM |
141 | #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ |
142 | ||
b52ce066 LJ |
143 | /* |
144 | * Returns approximate total of the readers' ->seq[] values for the | |
145 | * rank of per-CPU counters specified by idx. | |
146 | */ | |
147 | static unsigned long srcu_readers_seq_idx(struct srcu_struct *sp, int idx) | |
148 | { | |
149 | int cpu; | |
150 | unsigned long sum = 0; | |
151 | unsigned long t; | |
152 | ||
153 | for_each_possible_cpu(cpu) { | |
7d0ae808 | 154 | t = READ_ONCE(per_cpu_ptr(sp->per_cpu_ref, cpu)->seq[idx]); |
b52ce066 LJ |
155 | sum += t; |
156 | } | |
157 | return sum; | |
158 | } | |
159 | ||
621934ee | 160 | /* |
cef50120 | 161 | * Returns approximate number of readers active on the specified rank |
b52ce066 | 162 | * of the per-CPU ->c[] counters. |
621934ee | 163 | */ |
cef50120 PM |
164 | static unsigned long srcu_readers_active_idx(struct srcu_struct *sp, int idx) |
165 | { | |
166 | int cpu; | |
167 | unsigned long sum = 0; | |
168 | unsigned long t; | |
621934ee | 169 | |
cef50120 | 170 | for_each_possible_cpu(cpu) { |
7d0ae808 | 171 | t = READ_ONCE(per_cpu_ptr(sp->per_cpu_ref, cpu)->c[idx]); |
cef50120 | 172 | sum += t; |
cef50120 | 173 | } |
b52ce066 | 174 | return sum; |
cef50120 PM |
175 | } |
176 | ||
177 | /* | |
b52ce066 LJ |
178 | * Return true if the number of pre-existing readers is determined to |
179 | * be stably zero. An example unstable zero can occur if the call | |
180 | * to srcu_readers_active_idx() misses an __srcu_read_lock() increment, | |
181 | * but due to task migration, sees the corresponding __srcu_read_unlock() | |
182 | * decrement. This can happen because srcu_readers_active_idx() takes | |
183 | * time to sum the array, and might in fact be interrupted or preempted | |
184 | * partway through the summation. | |
cef50120 PM |
185 | */ |
186 | static bool srcu_readers_active_idx_check(struct srcu_struct *sp, int idx) | |
621934ee | 187 | { |
b52ce066 LJ |
188 | unsigned long seq; |
189 | ||
190 | seq = srcu_readers_seq_idx(sp, idx); | |
191 | ||
192 | /* | |
193 | * The following smp_mb() A pairs with the smp_mb() B located in | |
194 | * __srcu_read_lock(). This pairing ensures that if an | |
195 | * __srcu_read_lock() increments its counter after the summation | |
196 | * in srcu_readers_active_idx(), then the corresponding SRCU read-side | |
197 | * critical section will see any changes made prior to the start | |
198 | * of the current SRCU grace period. | |
199 | * | |
200 | * Also, if the above call to srcu_readers_seq_idx() saw the | |
201 | * increment of ->seq[], then the call to srcu_readers_active_idx() | |
202 | * must see the increment of ->c[]. | |
203 | */ | |
204 | smp_mb(); /* A */ | |
621934ee | 205 | |
cef50120 PM |
206 | /* |
207 | * Note that srcu_readers_active_idx() can incorrectly return | |
208 | * zero even though there is a pre-existing reader throughout. | |
209 | * To see this, suppose that task A is in a very long SRCU | |
210 | * read-side critical section that started on CPU 0, and that | |
b52ce066 | 211 | * no other reader exists, so that the sum of the counters |
cef50120 PM |
212 | * is equal to one. Then suppose that task B starts executing |
213 | * srcu_readers_active_idx(), summing up to CPU 1, and then that | |
214 | * task C starts reading on CPU 0, so that its increment is not | |
215 | * summed, but finishes reading on CPU 2, so that its decrement | |
216 | * -is- summed. Then when task B completes its sum, it will | |
217 | * incorrectly get zero, despite the fact that task A has been | |
218 | * in its SRCU read-side critical section the whole time. | |
219 | * | |
220 | * We therefore do a validation step should srcu_readers_active_idx() | |
221 | * return zero. | |
222 | */ | |
223 | if (srcu_readers_active_idx(sp, idx) != 0) | |
224 | return false; | |
225 | ||
226 | /* | |
b52ce066 LJ |
227 | * The remainder of this function is the validation step. |
228 | * The following smp_mb() D pairs with the smp_mb() C in | |
229 | * __srcu_read_unlock(). If the __srcu_read_unlock() was seen | |
230 | * by srcu_readers_active_idx() above, then any destructive | |
231 | * operation performed after the grace period will happen after | |
232 | * the corresponding SRCU read-side critical section. | |
cef50120 | 233 | * |
b52ce066 LJ |
234 | * Note that there can be at most NR_CPUS worth of readers using |
235 | * the old index, which is not enough to overflow even a 32-bit | |
236 | * integer. (Yes, this does mean that systems having more than | |
237 | * a billion or so CPUs need to be 64-bit systems.) Therefore, | |
238 | * the sum of the ->seq[] counters cannot possibly overflow. | |
239 | * Therefore, the only way that the return values of the two | |
240 | * calls to srcu_readers_seq_idx() can be equal is if there were | |
241 | * no increments of the corresponding rank of ->seq[] counts | |
242 | * in the interim. But the missed-increment scenario laid out | |
243 | * above includes an increment of the ->seq[] counter by | |
244 | * the corresponding __srcu_read_lock(). Therefore, if this | |
245 | * scenario occurs, the return values from the two calls to | |
246 | * srcu_readers_seq_idx() will differ, and thus the validation | |
247 | * step below suffices. | |
cef50120 | 248 | */ |
b52ce066 LJ |
249 | smp_mb(); /* D */ |
250 | ||
251 | return srcu_readers_seq_idx(sp, idx) == seq; | |
621934ee PM |
252 | } |
253 | ||
254 | /** | |
f765d113 NMG |
255 | * srcu_readers_active - returns true if there are readers. and false |
256 | * otherwise | |
621934ee PM |
257 | * @sp: which srcu_struct to count active readers (holding srcu_read_lock). |
258 | * | |
259 | * Note that this is not an atomic primitive, and can therefore suffer | |
260 | * severe errors when invoked on an active srcu_struct. That said, it | |
261 | * can be useful as an error check at cleanup time. | |
262 | */ | |
f765d113 | 263 | static bool srcu_readers_active(struct srcu_struct *sp) |
621934ee | 264 | { |
dc879175 LJ |
265 | int cpu; |
266 | unsigned long sum = 0; | |
267 | ||
268 | for_each_possible_cpu(cpu) { | |
7d0ae808 PM |
269 | sum += READ_ONCE(per_cpu_ptr(sp->per_cpu_ref, cpu)->c[0]); |
270 | sum += READ_ONCE(per_cpu_ptr(sp->per_cpu_ref, cpu)->c[1]); | |
dc879175 LJ |
271 | } |
272 | return sum; | |
621934ee PM |
273 | } |
274 | ||
275 | /** | |
276 | * cleanup_srcu_struct - deconstruct a sleep-RCU structure | |
277 | * @sp: structure to clean up. | |
278 | * | |
279 | * Must invoke this after you are finished using a given srcu_struct that | |
280 | * was initialized via init_srcu_struct(), else you leak memory. | |
281 | */ | |
282 | void cleanup_srcu_struct(struct srcu_struct *sp) | |
283 | { | |
ab4d2986 LJ |
284 | if (WARN_ON(srcu_readers_active(sp))) |
285 | return; /* Leakage unless caller handles error. */ | |
621934ee PM |
286 | free_percpu(sp->per_cpu_ref); |
287 | sp->per_cpu_ref = NULL; | |
288 | } | |
0cd397d3 | 289 | EXPORT_SYMBOL_GPL(cleanup_srcu_struct); |
621934ee | 290 | |
632ee200 | 291 | /* |
621934ee PM |
292 | * Counts the new reader in the appropriate per-CPU element of the |
293 | * srcu_struct. Must be called from process context. | |
294 | * Returns an index that must be passed to the matching srcu_read_unlock(). | |
295 | */ | |
632ee200 | 296 | int __srcu_read_lock(struct srcu_struct *sp) |
621934ee PM |
297 | { |
298 | int idx; | |
299 | ||
7d0ae808 | 300 | idx = READ_ONCE(sp->completed) & 0x1; |
a792563b | 301 | __this_cpu_inc(sp->per_cpu_ref->c[idx]); |
cef50120 | 302 | smp_mb(); /* B */ /* Avoid leaking the critical section. */ |
a792563b | 303 | __this_cpu_inc(sp->per_cpu_ref->seq[idx]); |
621934ee PM |
304 | return idx; |
305 | } | |
632ee200 | 306 | EXPORT_SYMBOL_GPL(__srcu_read_lock); |
621934ee | 307 | |
632ee200 | 308 | /* |
621934ee PM |
309 | * Removes the count for the old reader from the appropriate per-CPU |
310 | * element of the srcu_struct. Note that this may well be a different | |
311 | * CPU than that which was incremented by the corresponding srcu_read_lock(). | |
312 | * Must be called from process context. | |
313 | */ | |
632ee200 | 314 | void __srcu_read_unlock(struct srcu_struct *sp, int idx) |
621934ee | 315 | { |
cef50120 | 316 | smp_mb(); /* C */ /* Avoid leaking the critical section. */ |
5a41344a | 317 | this_cpu_dec(sp->per_cpu_ref->c[idx]); |
621934ee | 318 | } |
632ee200 | 319 | EXPORT_SYMBOL_GPL(__srcu_read_unlock); |
621934ee | 320 | |
c072a388 PM |
321 | /* |
322 | * We use an adaptive strategy for synchronize_srcu() and especially for | |
323 | * synchronize_srcu_expedited(). We spin for a fixed time period | |
324 | * (defined below) to allow SRCU readers to exit their read-side critical | |
325 | * sections. If there are still some readers after 10 microseconds, | |
326 | * we repeatedly block for 1-millisecond time periods. This approach | |
327 | * has done well in testing, so there is no need for a config parameter. | |
328 | */ | |
931ea9d1 | 329 | #define SRCU_RETRY_CHECK_DELAY 5 |
d9792edd LJ |
330 | #define SYNCHRONIZE_SRCU_TRYCOUNT 2 |
331 | #define SYNCHRONIZE_SRCU_EXP_TRYCOUNT 12 | |
cef50120 | 332 | |
18108ebf | 333 | /* |
931ea9d1 | 334 | * @@@ Wait until all pre-existing readers complete. Such readers |
18108ebf | 335 | * will have used the index specified by "idx". |
931ea9d1 LJ |
336 | * the caller should ensures the ->completed is not changed while checking |
337 | * and idx = (->completed & 1) ^ 1 | |
18108ebf | 338 | */ |
931ea9d1 | 339 | static bool try_check_zero(struct srcu_struct *sp, int idx, int trycount) |
cef50120 | 340 | { |
931ea9d1 LJ |
341 | for (;;) { |
342 | if (srcu_readers_active_idx_check(sp, idx)) | |
343 | return true; | |
344 | if (--trycount <= 0) | |
345 | return false; | |
346 | udelay(SRCU_RETRY_CHECK_DELAY); | |
cef50120 | 347 | } |
cef50120 | 348 | } |
c072a388 | 349 | |
931ea9d1 LJ |
350 | /* |
351 | * Increment the ->completed counter so that future SRCU readers will | |
352 | * use the other rank of the ->c[] and ->seq[] arrays. This allows | |
353 | * us to wait for pre-existing readers in a starvation-free manner. | |
354 | */ | |
18108ebf | 355 | static void srcu_flip(struct srcu_struct *sp) |
944ce9af | 356 | { |
18108ebf | 357 | sp->completed++; |
944ce9af LJ |
358 | } |
359 | ||
931ea9d1 LJ |
360 | /* |
361 | * Enqueue an SRCU callback on the specified srcu_struct structure, | |
362 | * initiating grace-period processing if it is not already running. | |
bc72d962 PM |
363 | * |
364 | * Note that all CPUs must agree that the grace period extended beyond | |
365 | * all pre-existing SRCU read-side critical section. On systems with | |
366 | * more than one CPU, this means that when "func()" is invoked, each CPU | |
367 | * is guaranteed to have executed a full memory barrier since the end of | |
368 | * its last corresponding SRCU read-side critical section whose beginning | |
369 | * preceded the call to call_rcu(). It also means that each CPU executing | |
370 | * an SRCU read-side critical section that continues beyond the start of | |
371 | * "func()" must have executed a memory barrier after the call_rcu() | |
372 | * but before the beginning of that SRCU read-side critical section. | |
373 | * Note that these guarantees include CPUs that are offline, idle, or | |
374 | * executing in user mode, as well as CPUs that are executing in the kernel. | |
375 | * | |
376 | * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the | |
377 | * resulting SRCU callback function "func()", then both CPU A and CPU | |
378 | * B are guaranteed to execute a full memory barrier during the time | |
379 | * interval between the call to call_rcu() and the invocation of "func()". | |
380 | * This guarantee applies even if CPU A and CPU B are the same CPU (but | |
381 | * again only if the system has more than one CPU). | |
382 | * | |
383 | * Of course, these guarantees apply only for invocations of call_srcu(), | |
384 | * srcu_read_lock(), and srcu_read_unlock() that are all passed the same | |
385 | * srcu_struct structure. | |
931ea9d1 LJ |
386 | */ |
387 | void call_srcu(struct srcu_struct *sp, struct rcu_head *head, | |
b6a4ae76 | 388 | rcu_callback_t func) |
931ea9d1 LJ |
389 | { |
390 | unsigned long flags; | |
391 | ||
392 | head->next = NULL; | |
393 | head->func = func; | |
394 | spin_lock_irqsave(&sp->queue_lock, flags); | |
395 | rcu_batch_queue(&sp->batch_queue, head); | |
396 | if (!sp->running) { | |
397 | sp->running = true; | |
ae167033 | 398 | queue_delayed_work(system_power_efficient_wq, &sp->work, 0); |
931ea9d1 LJ |
399 | } |
400 | spin_unlock_irqrestore(&sp->queue_lock, flags); | |
401 | } | |
402 | EXPORT_SYMBOL_GPL(call_srcu); | |
403 | ||
931ea9d1 LJ |
404 | static void srcu_advance_batches(struct srcu_struct *sp, int trycount); |
405 | static void srcu_reschedule(struct srcu_struct *sp); | |
406 | ||
0cd397d3 PM |
407 | /* |
408 | * Helper function for synchronize_srcu() and synchronize_srcu_expedited(). | |
621934ee | 409 | */ |
d9792edd | 410 | static void __synchronize_srcu(struct srcu_struct *sp, int trycount) |
621934ee | 411 | { |
931ea9d1 LJ |
412 | struct rcu_synchronize rcu; |
413 | struct rcu_head *head = &rcu.head; | |
414 | bool done = false; | |
18108ebf | 415 | |
f78f5b90 PM |
416 | RCU_LOCKDEP_WARN(lock_is_held(&sp->dep_map) || |
417 | lock_is_held(&rcu_bh_lock_map) || | |
418 | lock_is_held(&rcu_lock_map) || | |
419 | lock_is_held(&rcu_sched_lock_map), | |
420 | "Illegal synchronize_srcu() in same-type SRCU (or in RCU) read-side critical section"); | |
fe15d706 | 421 | |
6e6f1b30 | 422 | might_sleep(); |
931ea9d1 LJ |
423 | init_completion(&rcu.completion); |
424 | ||
425 | head->next = NULL; | |
426 | head->func = wakeme_after_rcu; | |
427 | spin_lock_irq(&sp->queue_lock); | |
428 | if (!sp->running) { | |
429 | /* steal the processing owner */ | |
430 | sp->running = true; | |
431 | rcu_batch_queue(&sp->batch_check0, head); | |
432 | spin_unlock_irq(&sp->queue_lock); | |
433 | ||
434 | srcu_advance_batches(sp, trycount); | |
435 | if (!rcu_batch_empty(&sp->batch_done)) { | |
436 | BUG_ON(sp->batch_done.head != head); | |
437 | rcu_batch_dequeue(&sp->batch_done); | |
438 | done = true; | |
439 | } | |
440 | /* give the processing owner to work_struct */ | |
441 | srcu_reschedule(sp); | |
442 | } else { | |
443 | rcu_batch_queue(&sp->batch_queue, head); | |
444 | spin_unlock_irq(&sp->queue_lock); | |
445 | } | |
944ce9af | 446 | |
931ea9d1 LJ |
447 | if (!done) |
448 | wait_for_completion(&rcu.completion); | |
621934ee PM |
449 | } |
450 | ||
0cd397d3 PM |
451 | /** |
452 | * synchronize_srcu - wait for prior SRCU read-side critical-section completion | |
453 | * @sp: srcu_struct with which to synchronize. | |
454 | * | |
34a64b6b LJ |
455 | * Wait for the count to drain to zero of both indexes. To avoid the |
456 | * possible starvation of synchronize_srcu(), it waits for the count of | |
457 | * the index=((->completed & 1) ^ 1) to drain to zero at first, | |
458 | * and then flip the completed and wait for the count of the other index. | |
459 | * | |
460 | * Can block; must be called from process context. | |
0cd397d3 PM |
461 | * |
462 | * Note that it is illegal to call synchronize_srcu() from the corresponding | |
463 | * SRCU read-side critical section; doing so will result in deadlock. | |
464 | * However, it is perfectly legal to call synchronize_srcu() on one | |
bc72d962 PM |
465 | * srcu_struct from some other srcu_struct's read-side critical section, |
466 | * as long as the resulting graph of srcu_structs is acyclic. | |
467 | * | |
468 | * There are memory-ordering constraints implied by synchronize_srcu(). | |
469 | * On systems with more than one CPU, when synchronize_srcu() returns, | |
470 | * each CPU is guaranteed to have executed a full memory barrier since | |
471 | * the end of its last corresponding SRCU-sched read-side critical section | |
472 | * whose beginning preceded the call to synchronize_srcu(). In addition, | |
473 | * each CPU having an SRCU read-side critical section that extends beyond | |
474 | * the return from synchronize_srcu() is guaranteed to have executed a | |
475 | * full memory barrier after the beginning of synchronize_srcu() and before | |
476 | * the beginning of that SRCU read-side critical section. Note that these | |
477 | * guarantees include CPUs that are offline, idle, or executing in user mode, | |
478 | * as well as CPUs that are executing in the kernel. | |
479 | * | |
480 | * Furthermore, if CPU A invoked synchronize_srcu(), which returned | |
481 | * to its caller on CPU B, then both CPU A and CPU B are guaranteed | |
482 | * to have executed a full memory barrier during the execution of | |
483 | * synchronize_srcu(). This guarantee applies even if CPU A and CPU B | |
484 | * are the same CPU, but again only if the system has more than one CPU. | |
485 | * | |
486 | * Of course, these memory-ordering guarantees apply only when | |
487 | * synchronize_srcu(), srcu_read_lock(), and srcu_read_unlock() are | |
488 | * passed the same srcu_struct structure. | |
0cd397d3 PM |
489 | */ |
490 | void synchronize_srcu(struct srcu_struct *sp) | |
491 | { | |
5a9be7c6 | 492 | __synchronize_srcu(sp, (rcu_gp_is_expedited() && !rcu_gp_is_normal()) |
3705b88d AM |
493 | ? SYNCHRONIZE_SRCU_EXP_TRYCOUNT |
494 | : SYNCHRONIZE_SRCU_TRYCOUNT); | |
0cd397d3 PM |
495 | } |
496 | EXPORT_SYMBOL_GPL(synchronize_srcu); | |
497 | ||
498 | /** | |
236fefaf | 499 | * synchronize_srcu_expedited - Brute-force SRCU grace period |
0cd397d3 PM |
500 | * @sp: srcu_struct with which to synchronize. |
501 | * | |
cef50120 PM |
502 | * Wait for an SRCU grace period to elapse, but be more aggressive about |
503 | * spinning rather than blocking when waiting. | |
0cd397d3 | 504 | * |
bc72d962 PM |
505 | * Note that synchronize_srcu_expedited() has the same deadlock and |
506 | * memory-ordering properties as does synchronize_srcu(). | |
0cd397d3 PM |
507 | */ |
508 | void synchronize_srcu_expedited(struct srcu_struct *sp) | |
509 | { | |
d9792edd | 510 | __synchronize_srcu(sp, SYNCHRONIZE_SRCU_EXP_TRYCOUNT); |
0cd397d3 PM |
511 | } |
512 | EXPORT_SYMBOL_GPL(synchronize_srcu_expedited); | |
513 | ||
931ea9d1 LJ |
514 | /** |
515 | * srcu_barrier - Wait until all in-flight call_srcu() callbacks complete. | |
4461212a | 516 | * @sp: srcu_struct on which to wait for in-flight callbacks. |
931ea9d1 LJ |
517 | */ |
518 | void srcu_barrier(struct srcu_struct *sp) | |
519 | { | |
520 | synchronize_srcu(sp); | |
521 | } | |
522 | EXPORT_SYMBOL_GPL(srcu_barrier); | |
523 | ||
621934ee PM |
524 | /** |
525 | * srcu_batches_completed - return batches completed. | |
526 | * @sp: srcu_struct on which to report batch completion. | |
527 | * | |
528 | * Report the number of batches, correlated with, but not necessarily | |
529 | * precisely the same as, the number of grace periods that have elapsed. | |
530 | */ | |
a5c198f4 | 531 | unsigned long srcu_batches_completed(struct srcu_struct *sp) |
621934ee PM |
532 | { |
533 | return sp->completed; | |
534 | } | |
621934ee | 535 | EXPORT_SYMBOL_GPL(srcu_batches_completed); |
931ea9d1 LJ |
536 | |
537 | #define SRCU_CALLBACK_BATCH 10 | |
538 | #define SRCU_INTERVAL 1 | |
539 | ||
540 | /* | |
541 | * Move any new SRCU callbacks to the first stage of the SRCU grace | |
542 | * period pipeline. | |
543 | */ | |
544 | static void srcu_collect_new(struct srcu_struct *sp) | |
545 | { | |
546 | if (!rcu_batch_empty(&sp->batch_queue)) { | |
547 | spin_lock_irq(&sp->queue_lock); | |
548 | rcu_batch_move(&sp->batch_check0, &sp->batch_queue); | |
549 | spin_unlock_irq(&sp->queue_lock); | |
550 | } | |
551 | } | |
552 | ||
553 | /* | |
554 | * Core SRCU state machine. Advance callbacks from ->batch_check0 to | |
555 | * ->batch_check1 and then to ->batch_done as readers drain. | |
556 | */ | |
557 | static void srcu_advance_batches(struct srcu_struct *sp, int trycount) | |
558 | { | |
559 | int idx = 1 ^ (sp->completed & 1); | |
560 | ||
561 | /* | |
562 | * Because readers might be delayed for an extended period after | |
563 | * fetching ->completed for their index, at any point in time there | |
564 | * might well be readers using both idx=0 and idx=1. We therefore | |
565 | * need to wait for readers to clear from both index values before | |
566 | * invoking a callback. | |
567 | */ | |
568 | ||
569 | if (rcu_batch_empty(&sp->batch_check0) && | |
570 | rcu_batch_empty(&sp->batch_check1)) | |
571 | return; /* no callbacks need to be advanced */ | |
572 | ||
573 | if (!try_check_zero(sp, idx, trycount)) | |
574 | return; /* failed to advance, will try after SRCU_INTERVAL */ | |
575 | ||
576 | /* | |
577 | * The callbacks in ->batch_check1 have already done with their | |
578 | * first zero check and flip back when they were enqueued on | |
579 | * ->batch_check0 in a previous invocation of srcu_advance_batches(). | |
580 | * (Presumably try_check_zero() returned false during that | |
581 | * invocation, leaving the callbacks stranded on ->batch_check1.) | |
582 | * They are therefore ready to invoke, so move them to ->batch_done. | |
583 | */ | |
584 | rcu_batch_move(&sp->batch_done, &sp->batch_check1); | |
585 | ||
586 | if (rcu_batch_empty(&sp->batch_check0)) | |
587 | return; /* no callbacks need to be advanced */ | |
588 | srcu_flip(sp); | |
589 | ||
590 | /* | |
591 | * The callbacks in ->batch_check0 just finished their | |
592 | * first check zero and flip, so move them to ->batch_check1 | |
593 | * for future checking on the other idx. | |
594 | */ | |
595 | rcu_batch_move(&sp->batch_check1, &sp->batch_check0); | |
596 | ||
597 | /* | |
598 | * SRCU read-side critical sections are normally short, so check | |
599 | * at least twice in quick succession after a flip. | |
600 | */ | |
601 | trycount = trycount < 2 ? 2 : trycount; | |
602 | if (!try_check_zero(sp, idx^1, trycount)) | |
603 | return; /* failed to advance, will try after SRCU_INTERVAL */ | |
604 | ||
605 | /* | |
606 | * The callbacks in ->batch_check1 have now waited for all | |
607 | * pre-existing readers using both idx values. They are therefore | |
608 | * ready to invoke, so move them to ->batch_done. | |
609 | */ | |
610 | rcu_batch_move(&sp->batch_done, &sp->batch_check1); | |
611 | } | |
612 | ||
613 | /* | |
614 | * Invoke a limited number of SRCU callbacks that have passed through | |
615 | * their grace period. If there are more to do, SRCU will reschedule | |
616 | * the workqueue. | |
617 | */ | |
618 | static void srcu_invoke_callbacks(struct srcu_struct *sp) | |
619 | { | |
620 | int i; | |
621 | struct rcu_head *head; | |
622 | ||
623 | for (i = 0; i < SRCU_CALLBACK_BATCH; i++) { | |
624 | head = rcu_batch_dequeue(&sp->batch_done); | |
625 | if (!head) | |
626 | break; | |
627 | local_bh_disable(); | |
628 | head->func(head); | |
629 | local_bh_enable(); | |
630 | } | |
631 | } | |
632 | ||
633 | /* | |
634 | * Finished one round of SRCU grace period. Start another if there are | |
635 | * more SRCU callbacks queued, otherwise put SRCU into not-running state. | |
636 | */ | |
637 | static void srcu_reschedule(struct srcu_struct *sp) | |
638 | { | |
639 | bool pending = true; | |
640 | ||
641 | if (rcu_batch_empty(&sp->batch_done) && | |
642 | rcu_batch_empty(&sp->batch_check1) && | |
643 | rcu_batch_empty(&sp->batch_check0) && | |
644 | rcu_batch_empty(&sp->batch_queue)) { | |
645 | spin_lock_irq(&sp->queue_lock); | |
646 | if (rcu_batch_empty(&sp->batch_done) && | |
647 | rcu_batch_empty(&sp->batch_check1) && | |
648 | rcu_batch_empty(&sp->batch_check0) && | |
649 | rcu_batch_empty(&sp->batch_queue)) { | |
650 | sp->running = false; | |
651 | pending = false; | |
652 | } | |
653 | spin_unlock_irq(&sp->queue_lock); | |
654 | } | |
655 | ||
656 | if (pending) | |
ae167033 SD |
657 | queue_delayed_work(system_power_efficient_wq, |
658 | &sp->work, SRCU_INTERVAL); | |
931ea9d1 LJ |
659 | } |
660 | ||
661 | /* | |
662 | * This is the work-queue function that handles SRCU grace periods. | |
663 | */ | |
f2ebfbc9 | 664 | void process_srcu(struct work_struct *work) |
931ea9d1 LJ |
665 | { |
666 | struct srcu_struct *sp; | |
667 | ||
668 | sp = container_of(work, struct srcu_struct, work.work); | |
669 | ||
670 | srcu_collect_new(sp); | |
671 | srcu_advance_batches(sp, 1); | |
672 | srcu_invoke_callbacks(sp); | |
673 | srcu_reschedule(sp); | |
674 | } | |
f2ebfbc9 | 675 | EXPORT_SYMBOL_GPL(process_srcu); |