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 | |
15 | * along with this program; if not, write to the Free Software | |
16 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
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 <trace/events/rcu.h> |
40 | ||
41 | #include "rcu.h" | |
42 | ||
931ea9d1 LJ |
43 | /* |
44 | * Initialize an rcu_batch structure to empty. | |
45 | */ | |
46 | static inline void rcu_batch_init(struct rcu_batch *b) | |
47 | { | |
48 | b->head = NULL; | |
49 | b->tail = &b->head; | |
50 | } | |
51 | ||
52 | /* | |
53 | * Enqueue a callback onto the tail of the specified rcu_batch structure. | |
54 | */ | |
55 | static inline void rcu_batch_queue(struct rcu_batch *b, struct rcu_head *head) | |
56 | { | |
57 | *b->tail = head; | |
58 | b->tail = &head->next; | |
59 | } | |
60 | ||
61 | /* | |
62 | * Is the specified rcu_batch structure empty? | |
63 | */ | |
64 | static inline bool rcu_batch_empty(struct rcu_batch *b) | |
65 | { | |
66 | return b->tail == &b->head; | |
67 | } | |
68 | ||
69 | /* | |
70 | * Remove the callback at the head of the specified rcu_batch structure | |
71 | * and return a pointer to it, or return NULL if the structure is empty. | |
72 | */ | |
73 | static inline struct rcu_head *rcu_batch_dequeue(struct rcu_batch *b) | |
74 | { | |
75 | struct rcu_head *head; | |
76 | ||
77 | if (rcu_batch_empty(b)) | |
78 | return NULL; | |
79 | ||
80 | head = b->head; | |
81 | b->head = head->next; | |
82 | if (b->tail == &head->next) | |
83 | rcu_batch_init(b); | |
84 | ||
85 | return head; | |
86 | } | |
87 | ||
88 | /* | |
89 | * Move all callbacks from the rcu_batch structure specified by "from" to | |
90 | * the structure specified by "to". | |
91 | */ | |
92 | static inline void rcu_batch_move(struct rcu_batch *to, struct rcu_batch *from) | |
93 | { | |
94 | if (!rcu_batch_empty(from)) { | |
95 | *to->tail = from->head; | |
96 | to->tail = from->tail; | |
97 | rcu_batch_init(from); | |
98 | } | |
99 | } | |
100 | ||
632ee200 PM |
101 | static int init_srcu_struct_fields(struct srcu_struct *sp) |
102 | { | |
103 | sp->completed = 0; | |
931ea9d1 LJ |
104 | spin_lock_init(&sp->queue_lock); |
105 | sp->running = false; | |
106 | rcu_batch_init(&sp->batch_queue); | |
107 | rcu_batch_init(&sp->batch_check0); | |
108 | rcu_batch_init(&sp->batch_check1); | |
109 | rcu_batch_init(&sp->batch_done); | |
110 | INIT_DELAYED_WORK(&sp->work, process_srcu); | |
632ee200 PM |
111 | sp->per_cpu_ref = alloc_percpu(struct srcu_struct_array); |
112 | return sp->per_cpu_ref ? 0 : -ENOMEM; | |
113 | } | |
114 | ||
115 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
116 | ||
117 | int __init_srcu_struct(struct srcu_struct *sp, const char *name, | |
118 | struct lock_class_key *key) | |
119 | { | |
632ee200 PM |
120 | /* Don't re-initialize a lock while it is held. */ |
121 | debug_check_no_locks_freed((void *)sp, sizeof(*sp)); | |
122 | lockdep_init_map(&sp->dep_map, name, key, 0); | |
632ee200 PM |
123 | return init_srcu_struct_fields(sp); |
124 | } | |
125 | EXPORT_SYMBOL_GPL(__init_srcu_struct); | |
126 | ||
127 | #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ | |
128 | ||
621934ee PM |
129 | /** |
130 | * init_srcu_struct - initialize a sleep-RCU structure | |
131 | * @sp: structure to initialize. | |
132 | * | |
133 | * Must invoke this on a given srcu_struct before passing that srcu_struct | |
134 | * to any other function. Each srcu_struct represents a separate domain | |
135 | * of SRCU protection. | |
136 | */ | |
e6a92013 | 137 | int init_srcu_struct(struct srcu_struct *sp) |
621934ee | 138 | { |
632ee200 | 139 | return init_srcu_struct_fields(sp); |
621934ee | 140 | } |
0cd397d3 | 141 | EXPORT_SYMBOL_GPL(init_srcu_struct); |
621934ee | 142 | |
632ee200 PM |
143 | #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ |
144 | ||
b52ce066 LJ |
145 | /* |
146 | * Returns approximate total of the readers' ->seq[] values for the | |
147 | * rank of per-CPU counters specified by idx. | |
148 | */ | |
149 | static unsigned long srcu_readers_seq_idx(struct srcu_struct *sp, int idx) | |
150 | { | |
151 | int cpu; | |
152 | unsigned long sum = 0; | |
153 | unsigned long t; | |
154 | ||
155 | for_each_possible_cpu(cpu) { | |
156 | t = ACCESS_ONCE(per_cpu_ptr(sp->per_cpu_ref, cpu)->seq[idx]); | |
157 | sum += t; | |
158 | } | |
159 | return sum; | |
160 | } | |
161 | ||
621934ee | 162 | /* |
cef50120 | 163 | * Returns approximate number of readers active on the specified rank |
b52ce066 | 164 | * of the per-CPU ->c[] counters. |
621934ee | 165 | */ |
cef50120 PM |
166 | static unsigned long srcu_readers_active_idx(struct srcu_struct *sp, int idx) |
167 | { | |
168 | int cpu; | |
169 | unsigned long sum = 0; | |
170 | unsigned long t; | |
621934ee | 171 | |
cef50120 PM |
172 | for_each_possible_cpu(cpu) { |
173 | t = ACCESS_ONCE(per_cpu_ptr(sp->per_cpu_ref, cpu)->c[idx]); | |
174 | sum += t; | |
cef50120 | 175 | } |
b52ce066 | 176 | return sum; |
cef50120 PM |
177 | } |
178 | ||
179 | /* | |
b52ce066 LJ |
180 | * Return true if the number of pre-existing readers is determined to |
181 | * be stably zero. An example unstable zero can occur if the call | |
182 | * to srcu_readers_active_idx() misses an __srcu_read_lock() increment, | |
183 | * but due to task migration, sees the corresponding __srcu_read_unlock() | |
184 | * decrement. This can happen because srcu_readers_active_idx() takes | |
185 | * time to sum the array, and might in fact be interrupted or preempted | |
186 | * partway through the summation. | |
cef50120 PM |
187 | */ |
188 | static bool srcu_readers_active_idx_check(struct srcu_struct *sp, int idx) | |
621934ee | 189 | { |
b52ce066 LJ |
190 | unsigned long seq; |
191 | ||
192 | seq = srcu_readers_seq_idx(sp, idx); | |
193 | ||
194 | /* | |
195 | * The following smp_mb() A pairs with the smp_mb() B located in | |
196 | * __srcu_read_lock(). This pairing ensures that if an | |
197 | * __srcu_read_lock() increments its counter after the summation | |
198 | * in srcu_readers_active_idx(), then the corresponding SRCU read-side | |
199 | * critical section will see any changes made prior to the start | |
200 | * of the current SRCU grace period. | |
201 | * | |
202 | * Also, if the above call to srcu_readers_seq_idx() saw the | |
203 | * increment of ->seq[], then the call to srcu_readers_active_idx() | |
204 | * must see the increment of ->c[]. | |
205 | */ | |
206 | smp_mb(); /* A */ | |
621934ee | 207 | |
cef50120 PM |
208 | /* |
209 | * Note that srcu_readers_active_idx() can incorrectly return | |
210 | * zero even though there is a pre-existing reader throughout. | |
211 | * To see this, suppose that task A is in a very long SRCU | |
212 | * read-side critical section that started on CPU 0, and that | |
b52ce066 | 213 | * no other reader exists, so that the sum of the counters |
cef50120 PM |
214 | * is equal to one. Then suppose that task B starts executing |
215 | * srcu_readers_active_idx(), summing up to CPU 1, and then that | |
216 | * task C starts reading on CPU 0, so that its increment is not | |
217 | * summed, but finishes reading on CPU 2, so that its decrement | |
218 | * -is- summed. Then when task B completes its sum, it will | |
219 | * incorrectly get zero, despite the fact that task A has been | |
220 | * in its SRCU read-side critical section the whole time. | |
221 | * | |
222 | * We therefore do a validation step should srcu_readers_active_idx() | |
223 | * return zero. | |
224 | */ | |
225 | if (srcu_readers_active_idx(sp, idx) != 0) | |
226 | return false; | |
227 | ||
228 | /* | |
b52ce066 LJ |
229 | * The remainder of this function is the validation step. |
230 | * The following smp_mb() D pairs with the smp_mb() C in | |
231 | * __srcu_read_unlock(). If the __srcu_read_unlock() was seen | |
232 | * by srcu_readers_active_idx() above, then any destructive | |
233 | * operation performed after the grace period will happen after | |
234 | * the corresponding SRCU read-side critical section. | |
cef50120 | 235 | * |
b52ce066 LJ |
236 | * Note that there can be at most NR_CPUS worth of readers using |
237 | * the old index, which is not enough to overflow even a 32-bit | |
238 | * integer. (Yes, this does mean that systems having more than | |
239 | * a billion or so CPUs need to be 64-bit systems.) Therefore, | |
240 | * the sum of the ->seq[] counters cannot possibly overflow. | |
241 | * Therefore, the only way that the return values of the two | |
242 | * calls to srcu_readers_seq_idx() can be equal is if there were | |
243 | * no increments of the corresponding rank of ->seq[] counts | |
244 | * in the interim. But the missed-increment scenario laid out | |
245 | * above includes an increment of the ->seq[] counter by | |
246 | * the corresponding __srcu_read_lock(). Therefore, if this | |
247 | * scenario occurs, the return values from the two calls to | |
248 | * srcu_readers_seq_idx() will differ, and thus the validation | |
249 | * step below suffices. | |
cef50120 | 250 | */ |
b52ce066 LJ |
251 | smp_mb(); /* D */ |
252 | ||
253 | return srcu_readers_seq_idx(sp, idx) == seq; | |
621934ee PM |
254 | } |
255 | ||
256 | /** | |
257 | * srcu_readers_active - returns approximate number of readers. | |
258 | * @sp: which srcu_struct to count active readers (holding srcu_read_lock). | |
259 | * | |
260 | * Note that this is not an atomic primitive, and can therefore suffer | |
261 | * severe errors when invoked on an active srcu_struct. That said, it | |
262 | * can be useful as an error check at cleanup time. | |
263 | */ | |
bb695170 | 264 | static int srcu_readers_active(struct srcu_struct *sp) |
621934ee | 265 | { |
dc879175 LJ |
266 | int cpu; |
267 | unsigned long sum = 0; | |
268 | ||
269 | for_each_possible_cpu(cpu) { | |
270 | sum += ACCESS_ONCE(per_cpu_ptr(sp->per_cpu_ref, cpu)->c[0]); | |
271 | sum += ACCESS_ONCE(per_cpu_ptr(sp->per_cpu_ref, cpu)->c[1]); | |
272 | } | |
273 | return sum; | |
621934ee PM |
274 | } |
275 | ||
276 | /** | |
277 | * cleanup_srcu_struct - deconstruct a sleep-RCU structure | |
278 | * @sp: structure to clean up. | |
279 | * | |
280 | * Must invoke this after you are finished using a given srcu_struct that | |
281 | * was initialized via init_srcu_struct(), else you leak memory. | |
282 | */ | |
283 | void cleanup_srcu_struct(struct srcu_struct *sp) | |
284 | { | |
285 | int sum; | |
286 | ||
287 | sum = srcu_readers_active(sp); | |
288 | WARN_ON(sum); /* Leakage unless caller handles error. */ | |
289 | if (sum != 0) | |
290 | return; | |
291 | free_percpu(sp->per_cpu_ref); | |
292 | sp->per_cpu_ref = NULL; | |
293 | } | |
0cd397d3 | 294 | EXPORT_SYMBOL_GPL(cleanup_srcu_struct); |
621934ee | 295 | |
632ee200 | 296 | /* |
621934ee PM |
297 | * Counts the new reader in the appropriate per-CPU element of the |
298 | * srcu_struct. Must be called from process context. | |
299 | * Returns an index that must be passed to the matching srcu_read_unlock(). | |
300 | */ | |
632ee200 | 301 | int __srcu_read_lock(struct srcu_struct *sp) |
621934ee PM |
302 | { |
303 | int idx; | |
304 | ||
305 | preempt_disable(); | |
cef50120 PM |
306 | idx = rcu_dereference_index_check(sp->completed, |
307 | rcu_read_lock_sched_held()) & 0x1; | |
b52ce066 | 308 | ACCESS_ONCE(this_cpu_ptr(sp->per_cpu_ref)->c[idx]) += 1; |
cef50120 | 309 | smp_mb(); /* B */ /* Avoid leaking the critical section. */ |
b52ce066 | 310 | ACCESS_ONCE(this_cpu_ptr(sp->per_cpu_ref)->seq[idx]) += 1; |
621934ee PM |
311 | preempt_enable(); |
312 | return idx; | |
313 | } | |
632ee200 | 314 | EXPORT_SYMBOL_GPL(__srcu_read_lock); |
621934ee | 315 | |
632ee200 | 316 | /* |
621934ee PM |
317 | * Removes the count for the old reader from the appropriate per-CPU |
318 | * element of the srcu_struct. Note that this may well be a different | |
319 | * CPU than that which was incremented by the corresponding srcu_read_lock(). | |
320 | * Must be called from process context. | |
321 | */ | |
632ee200 | 322 | void __srcu_read_unlock(struct srcu_struct *sp, int idx) |
621934ee | 323 | { |
cef50120 | 324 | smp_mb(); /* C */ /* Avoid leaking the critical section. */ |
5a41344a | 325 | this_cpu_dec(sp->per_cpu_ref->c[idx]); |
621934ee | 326 | } |
632ee200 | 327 | EXPORT_SYMBOL_GPL(__srcu_read_unlock); |
621934ee | 328 | |
c072a388 PM |
329 | /* |
330 | * We use an adaptive strategy for synchronize_srcu() and especially for | |
331 | * synchronize_srcu_expedited(). We spin for a fixed time period | |
332 | * (defined below) to allow SRCU readers to exit their read-side critical | |
333 | * sections. If there are still some readers after 10 microseconds, | |
334 | * we repeatedly block for 1-millisecond time periods. This approach | |
335 | * has done well in testing, so there is no need for a config parameter. | |
336 | */ | |
931ea9d1 | 337 | #define SRCU_RETRY_CHECK_DELAY 5 |
d9792edd LJ |
338 | #define SYNCHRONIZE_SRCU_TRYCOUNT 2 |
339 | #define SYNCHRONIZE_SRCU_EXP_TRYCOUNT 12 | |
cef50120 | 340 | |
18108ebf | 341 | /* |
931ea9d1 | 342 | * @@@ Wait until all pre-existing readers complete. Such readers |
18108ebf | 343 | * will have used the index specified by "idx". |
931ea9d1 LJ |
344 | * the caller should ensures the ->completed is not changed while checking |
345 | * and idx = (->completed & 1) ^ 1 | |
18108ebf | 346 | */ |
931ea9d1 | 347 | static bool try_check_zero(struct srcu_struct *sp, int idx, int trycount) |
cef50120 | 348 | { |
931ea9d1 LJ |
349 | for (;;) { |
350 | if (srcu_readers_active_idx_check(sp, idx)) | |
351 | return true; | |
352 | if (--trycount <= 0) | |
353 | return false; | |
354 | udelay(SRCU_RETRY_CHECK_DELAY); | |
cef50120 | 355 | } |
cef50120 | 356 | } |
c072a388 | 357 | |
931ea9d1 LJ |
358 | /* |
359 | * Increment the ->completed counter so that future SRCU readers will | |
360 | * use the other rank of the ->c[] and ->seq[] arrays. This allows | |
361 | * us to wait for pre-existing readers in a starvation-free manner. | |
362 | */ | |
18108ebf | 363 | static void srcu_flip(struct srcu_struct *sp) |
944ce9af | 364 | { |
18108ebf | 365 | sp->completed++; |
944ce9af LJ |
366 | } |
367 | ||
931ea9d1 LJ |
368 | /* |
369 | * Enqueue an SRCU callback on the specified srcu_struct structure, | |
370 | * initiating grace-period processing if it is not already running. | |
371 | */ | |
372 | void call_srcu(struct srcu_struct *sp, struct rcu_head *head, | |
373 | void (*func)(struct rcu_head *head)) | |
374 | { | |
375 | unsigned long flags; | |
376 | ||
377 | head->next = NULL; | |
378 | head->func = func; | |
379 | spin_lock_irqsave(&sp->queue_lock, flags); | |
380 | rcu_batch_queue(&sp->batch_queue, head); | |
381 | if (!sp->running) { | |
382 | sp->running = true; | |
3b07e9ca | 383 | schedule_delayed_work(&sp->work, 0); |
931ea9d1 LJ |
384 | } |
385 | spin_unlock_irqrestore(&sp->queue_lock, flags); | |
386 | } | |
387 | EXPORT_SYMBOL_GPL(call_srcu); | |
388 | ||
389 | struct rcu_synchronize { | |
390 | struct rcu_head head; | |
391 | struct completion completion; | |
392 | }; | |
393 | ||
394 | /* | |
395 | * Awaken the corresponding synchronize_srcu() instance now that a | |
396 | * grace period has elapsed. | |
397 | */ | |
398 | static void wakeme_after_rcu(struct rcu_head *head) | |
399 | { | |
400 | struct rcu_synchronize *rcu; | |
401 | ||
402 | rcu = container_of(head, struct rcu_synchronize, head); | |
403 | complete(&rcu->completion); | |
404 | } | |
405 | ||
406 | static void srcu_advance_batches(struct srcu_struct *sp, int trycount); | |
407 | static void srcu_reschedule(struct srcu_struct *sp); | |
408 | ||
0cd397d3 PM |
409 | /* |
410 | * Helper function for synchronize_srcu() and synchronize_srcu_expedited(). | |
621934ee | 411 | */ |
d9792edd | 412 | static void __synchronize_srcu(struct srcu_struct *sp, int trycount) |
621934ee | 413 | { |
931ea9d1 LJ |
414 | struct rcu_synchronize rcu; |
415 | struct rcu_head *head = &rcu.head; | |
416 | bool done = false; | |
18108ebf | 417 | |
fe15d706 PM |
418 | rcu_lockdep_assert(!lock_is_held(&sp->dep_map) && |
419 | !lock_is_held(&rcu_bh_lock_map) && | |
420 | !lock_is_held(&rcu_lock_map) && | |
421 | !lock_is_held(&rcu_sched_lock_map), | |
422 | "Illegal synchronize_srcu() in same-type SRCU (or RCU) read-side critical section"); | |
423 | ||
931ea9d1 LJ |
424 | init_completion(&rcu.completion); |
425 | ||
426 | head->next = NULL; | |
427 | head->func = wakeme_after_rcu; | |
428 | spin_lock_irq(&sp->queue_lock); | |
429 | if (!sp->running) { | |
430 | /* steal the processing owner */ | |
431 | sp->running = true; | |
432 | rcu_batch_queue(&sp->batch_check0, head); | |
433 | spin_unlock_irq(&sp->queue_lock); | |
434 | ||
435 | srcu_advance_batches(sp, trycount); | |
436 | if (!rcu_batch_empty(&sp->batch_done)) { | |
437 | BUG_ON(sp->batch_done.head != head); | |
438 | rcu_batch_dequeue(&sp->batch_done); | |
439 | done = true; | |
440 | } | |
441 | /* give the processing owner to work_struct */ | |
442 | srcu_reschedule(sp); | |
443 | } else { | |
444 | rcu_batch_queue(&sp->batch_queue, head); | |
445 | spin_unlock_irq(&sp->queue_lock); | |
446 | } | |
944ce9af | 447 | |
931ea9d1 LJ |
448 | if (!done) |
449 | wait_for_completion(&rcu.completion); | |
621934ee PM |
450 | } |
451 | ||
0cd397d3 PM |
452 | /** |
453 | * synchronize_srcu - wait for prior SRCU read-side critical-section completion | |
454 | * @sp: srcu_struct with which to synchronize. | |
455 | * | |
456 | * Flip the completed counter, and wait for the old count to drain to zero. | |
457 | * As with classic RCU, the updater must use some separate means of | |
458 | * synchronizing concurrent updates. Can block; must be called from | |
459 | * process context. | |
460 | * | |
461 | * Note that it is illegal to call synchronize_srcu() from the corresponding | |
462 | * SRCU read-side critical section; doing so will result in deadlock. | |
463 | * However, it is perfectly legal to call synchronize_srcu() on one | |
464 | * srcu_struct from some other srcu_struct's read-side critical section. | |
465 | */ | |
466 | void synchronize_srcu(struct srcu_struct *sp) | |
467 | { | |
3705b88d AM |
468 | __synchronize_srcu(sp, rcu_expedited |
469 | ? SYNCHRONIZE_SRCU_EXP_TRYCOUNT | |
470 | : SYNCHRONIZE_SRCU_TRYCOUNT); | |
0cd397d3 PM |
471 | } |
472 | EXPORT_SYMBOL_GPL(synchronize_srcu); | |
473 | ||
474 | /** | |
236fefaf | 475 | * synchronize_srcu_expedited - Brute-force SRCU grace period |
0cd397d3 PM |
476 | * @sp: srcu_struct with which to synchronize. |
477 | * | |
cef50120 PM |
478 | * Wait for an SRCU grace period to elapse, but be more aggressive about |
479 | * spinning rather than blocking when waiting. | |
0cd397d3 | 480 | * |
236fefaf | 481 | * Note that it is illegal to call this function while holding any lock |
cef50120 | 482 | * that is acquired by a CPU-hotplug notifier. It is also illegal to call |
236fefaf PM |
483 | * synchronize_srcu_expedited() from the corresponding SRCU read-side |
484 | * critical section; doing so will result in deadlock. However, it is | |
485 | * perfectly legal to call synchronize_srcu_expedited() on one srcu_struct | |
486 | * from some other srcu_struct's read-side critical section, as long as | |
487 | * the resulting graph of srcu_structs is acyclic. | |
0cd397d3 PM |
488 | */ |
489 | void synchronize_srcu_expedited(struct srcu_struct *sp) | |
490 | { | |
d9792edd | 491 | __synchronize_srcu(sp, SYNCHRONIZE_SRCU_EXP_TRYCOUNT); |
0cd397d3 PM |
492 | } |
493 | EXPORT_SYMBOL_GPL(synchronize_srcu_expedited); | |
494 | ||
931ea9d1 LJ |
495 | /** |
496 | * srcu_barrier - Wait until all in-flight call_srcu() callbacks complete. | |
497 | */ | |
498 | void srcu_barrier(struct srcu_struct *sp) | |
499 | { | |
500 | synchronize_srcu(sp); | |
501 | } | |
502 | EXPORT_SYMBOL_GPL(srcu_barrier); | |
503 | ||
621934ee PM |
504 | /** |
505 | * srcu_batches_completed - return batches completed. | |
506 | * @sp: srcu_struct on which to report batch completion. | |
507 | * | |
508 | * Report the number of batches, correlated with, but not necessarily | |
509 | * precisely the same as, the number of grace periods that have elapsed. | |
510 | */ | |
621934ee PM |
511 | long srcu_batches_completed(struct srcu_struct *sp) |
512 | { | |
513 | return sp->completed; | |
514 | } | |
621934ee | 515 | EXPORT_SYMBOL_GPL(srcu_batches_completed); |
931ea9d1 LJ |
516 | |
517 | #define SRCU_CALLBACK_BATCH 10 | |
518 | #define SRCU_INTERVAL 1 | |
519 | ||
520 | /* | |
521 | * Move any new SRCU callbacks to the first stage of the SRCU grace | |
522 | * period pipeline. | |
523 | */ | |
524 | static void srcu_collect_new(struct srcu_struct *sp) | |
525 | { | |
526 | if (!rcu_batch_empty(&sp->batch_queue)) { | |
527 | spin_lock_irq(&sp->queue_lock); | |
528 | rcu_batch_move(&sp->batch_check0, &sp->batch_queue); | |
529 | spin_unlock_irq(&sp->queue_lock); | |
530 | } | |
531 | } | |
532 | ||
533 | /* | |
534 | * Core SRCU state machine. Advance callbacks from ->batch_check0 to | |
535 | * ->batch_check1 and then to ->batch_done as readers drain. | |
536 | */ | |
537 | static void srcu_advance_batches(struct srcu_struct *sp, int trycount) | |
538 | { | |
539 | int idx = 1 ^ (sp->completed & 1); | |
540 | ||
541 | /* | |
542 | * Because readers might be delayed for an extended period after | |
543 | * fetching ->completed for their index, at any point in time there | |
544 | * might well be readers using both idx=0 and idx=1. We therefore | |
545 | * need to wait for readers to clear from both index values before | |
546 | * invoking a callback. | |
547 | */ | |
548 | ||
549 | if (rcu_batch_empty(&sp->batch_check0) && | |
550 | rcu_batch_empty(&sp->batch_check1)) | |
551 | return; /* no callbacks need to be advanced */ | |
552 | ||
553 | if (!try_check_zero(sp, idx, trycount)) | |
554 | return; /* failed to advance, will try after SRCU_INTERVAL */ | |
555 | ||
556 | /* | |
557 | * The callbacks in ->batch_check1 have already done with their | |
558 | * first zero check and flip back when they were enqueued on | |
559 | * ->batch_check0 in a previous invocation of srcu_advance_batches(). | |
560 | * (Presumably try_check_zero() returned false during that | |
561 | * invocation, leaving the callbacks stranded on ->batch_check1.) | |
562 | * They are therefore ready to invoke, so move them to ->batch_done. | |
563 | */ | |
564 | rcu_batch_move(&sp->batch_done, &sp->batch_check1); | |
565 | ||
566 | if (rcu_batch_empty(&sp->batch_check0)) | |
567 | return; /* no callbacks need to be advanced */ | |
568 | srcu_flip(sp); | |
569 | ||
570 | /* | |
571 | * The callbacks in ->batch_check0 just finished their | |
572 | * first check zero and flip, so move them to ->batch_check1 | |
573 | * for future checking on the other idx. | |
574 | */ | |
575 | rcu_batch_move(&sp->batch_check1, &sp->batch_check0); | |
576 | ||
577 | /* | |
578 | * SRCU read-side critical sections are normally short, so check | |
579 | * at least twice in quick succession after a flip. | |
580 | */ | |
581 | trycount = trycount < 2 ? 2 : trycount; | |
582 | if (!try_check_zero(sp, idx^1, trycount)) | |
583 | return; /* failed to advance, will try after SRCU_INTERVAL */ | |
584 | ||
585 | /* | |
586 | * The callbacks in ->batch_check1 have now waited for all | |
587 | * pre-existing readers using both idx values. They are therefore | |
588 | * ready to invoke, so move them to ->batch_done. | |
589 | */ | |
590 | rcu_batch_move(&sp->batch_done, &sp->batch_check1); | |
591 | } | |
592 | ||
593 | /* | |
594 | * Invoke a limited number of SRCU callbacks that have passed through | |
595 | * their grace period. If there are more to do, SRCU will reschedule | |
596 | * the workqueue. | |
597 | */ | |
598 | static void srcu_invoke_callbacks(struct srcu_struct *sp) | |
599 | { | |
600 | int i; | |
601 | struct rcu_head *head; | |
602 | ||
603 | for (i = 0; i < SRCU_CALLBACK_BATCH; i++) { | |
604 | head = rcu_batch_dequeue(&sp->batch_done); | |
605 | if (!head) | |
606 | break; | |
607 | local_bh_disable(); | |
608 | head->func(head); | |
609 | local_bh_enable(); | |
610 | } | |
611 | } | |
612 | ||
613 | /* | |
614 | * Finished one round of SRCU grace period. Start another if there are | |
615 | * more SRCU callbacks queued, otherwise put SRCU into not-running state. | |
616 | */ | |
617 | static void srcu_reschedule(struct srcu_struct *sp) | |
618 | { | |
619 | bool pending = true; | |
620 | ||
621 | if (rcu_batch_empty(&sp->batch_done) && | |
622 | rcu_batch_empty(&sp->batch_check1) && | |
623 | rcu_batch_empty(&sp->batch_check0) && | |
624 | rcu_batch_empty(&sp->batch_queue)) { | |
625 | spin_lock_irq(&sp->queue_lock); | |
626 | if (rcu_batch_empty(&sp->batch_done) && | |
627 | rcu_batch_empty(&sp->batch_check1) && | |
628 | rcu_batch_empty(&sp->batch_check0) && | |
629 | rcu_batch_empty(&sp->batch_queue)) { | |
630 | sp->running = false; | |
631 | pending = false; | |
632 | } | |
633 | spin_unlock_irq(&sp->queue_lock); | |
634 | } | |
635 | ||
636 | if (pending) | |
3b07e9ca | 637 | schedule_delayed_work(&sp->work, SRCU_INTERVAL); |
931ea9d1 LJ |
638 | } |
639 | ||
640 | /* | |
641 | * This is the work-queue function that handles SRCU grace periods. | |
642 | */ | |
f2ebfbc9 | 643 | void process_srcu(struct work_struct *work) |
931ea9d1 LJ |
644 | { |
645 | struct srcu_struct *sp; | |
646 | ||
647 | sp = container_of(work, struct srcu_struct, work.work); | |
648 | ||
649 | srcu_collect_new(sp); | |
650 | srcu_advance_batches(sp, 1); | |
651 | srcu_invoke_callbacks(sp); | |
652 | srcu_reschedule(sp); | |
653 | } | |
f2ebfbc9 | 654 | EXPORT_SYMBOL_GPL(process_srcu); |