Commit | Line | Data |
---|---|---|
1da177e4 LT |
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, write to the Free Software | |
16 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
17 | * | |
01c1c660 | 18 | * Copyright IBM Corporation, 2001 |
1da177e4 LT |
19 | * |
20 | * Authors: Dipankar Sarma <dipankar@in.ibm.com> | |
21 | * Manfred Spraul <manfred@colorfullife.com> | |
a71fca58 | 22 | * |
1da177e4 LT |
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 - | |
a71fca58 | 30 | * http://lse.sourceforge.net/locking/rcupdate.html |
1da177e4 LT |
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> | |
60063497 | 40 | #include <linux/atomic.h> |
1da177e4 | 41 | #include <linux/bitops.h> |
1da177e4 LT |
42 | #include <linux/percpu.h> |
43 | #include <linux/notifier.h> | |
1da177e4 | 44 | #include <linux/cpu.h> |
9331b315 | 45 | #include <linux/mutex.h> |
9984de1a | 46 | #include <linux/export.h> |
e3818b8d | 47 | #include <linux/hardirq.h> |
e3ebfb96 | 48 | #include <linux/delay.h> |
3705b88d | 49 | #include <linux/module.h> |
1da177e4 | 50 | |
29c00b4a PM |
51 | #define CREATE_TRACE_POINTS |
52 | #include <trace/events/rcu.h> | |
53 | ||
54 | #include "rcu.h" | |
55 | ||
3705b88d AM |
56 | module_param(rcu_expedited, int, 0); |
57 | ||
9dd8fb16 PM |
58 | #ifdef CONFIG_PREEMPT_RCU |
59 | ||
2a3fa843 PM |
60 | /* |
61 | * Preemptible RCU implementation for rcu_read_lock(). | |
62 | * Just increment ->rcu_read_lock_nesting, shared state will be updated | |
63 | * if we block. | |
64 | */ | |
65 | void __rcu_read_lock(void) | |
66 | { | |
67 | current->rcu_read_lock_nesting++; | |
68 | barrier(); /* critical section after entry code. */ | |
69 | } | |
70 | EXPORT_SYMBOL_GPL(__rcu_read_lock); | |
71 | ||
72 | /* | |
73 | * Preemptible RCU implementation for rcu_read_unlock(). | |
74 | * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost | |
75 | * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then | |
76 | * invoke rcu_read_unlock_special() to clean up after a context switch | |
77 | * in an RCU read-side critical section and other special cases. | |
78 | */ | |
79 | void __rcu_read_unlock(void) | |
80 | { | |
81 | struct task_struct *t = current; | |
82 | ||
83 | if (t->rcu_read_lock_nesting != 1) { | |
84 | --t->rcu_read_lock_nesting; | |
85 | } else { | |
86 | barrier(); /* critical section before exit code. */ | |
87 | t->rcu_read_lock_nesting = INT_MIN; | |
e3ebfb96 PM |
88 | #ifdef CONFIG_PROVE_RCU_DELAY |
89 | udelay(10); /* Make preemption more probable. */ | |
90 | #endif /* #ifdef CONFIG_PROVE_RCU_DELAY */ | |
2a3fa843 PM |
91 | barrier(); /* assign before ->rcu_read_unlock_special load */ |
92 | if (unlikely(ACCESS_ONCE(t->rcu_read_unlock_special))) | |
93 | rcu_read_unlock_special(t); | |
94 | barrier(); /* ->rcu_read_unlock_special load before assign */ | |
95 | t->rcu_read_lock_nesting = 0; | |
96 | } | |
97 | #ifdef CONFIG_PROVE_LOCKING | |
98 | { | |
99 | int rrln = ACCESS_ONCE(t->rcu_read_lock_nesting); | |
100 | ||
101 | WARN_ON_ONCE(rrln < 0 && rrln > INT_MIN / 2); | |
102 | } | |
103 | #endif /* #ifdef CONFIG_PROVE_LOCKING */ | |
104 | } | |
105 | EXPORT_SYMBOL_GPL(__rcu_read_unlock); | |
106 | ||
2439b696 | 107 | #endif /* #ifdef CONFIG_PREEMPT_RCU */ |
9dd8fb16 | 108 | |
162cc279 PM |
109 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
110 | static struct lock_class_key rcu_lock_key; | |
111 | struct lockdep_map rcu_lock_map = | |
112 | STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key); | |
113 | EXPORT_SYMBOL_GPL(rcu_lock_map); | |
632ee200 PM |
114 | |
115 | static struct lock_class_key rcu_bh_lock_key; | |
116 | struct lockdep_map rcu_bh_lock_map = | |
117 | STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_bh", &rcu_bh_lock_key); | |
118 | EXPORT_SYMBOL_GPL(rcu_bh_lock_map); | |
119 | ||
120 | static struct lock_class_key rcu_sched_lock_key; | |
121 | struct lockdep_map rcu_sched_lock_map = | |
122 | STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_sched", &rcu_sched_lock_key); | |
123 | EXPORT_SYMBOL_GPL(rcu_sched_lock_map); | |
e3818b8d | 124 | |
bc293d62 PM |
125 | int debug_lockdep_rcu_enabled(void) |
126 | { | |
127 | return rcu_scheduler_active && debug_locks && | |
128 | current->lockdep_recursion == 0; | |
129 | } | |
130 | EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled); | |
131 | ||
e3818b8d | 132 | /** |
ca5ecddf | 133 | * rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section? |
e3818b8d PM |
134 | * |
135 | * Check for bottom half being disabled, which covers both the | |
136 | * CONFIG_PROVE_RCU and not cases. Note that if someone uses | |
137 | * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled) | |
ca5ecddf PM |
138 | * will show the situation. This is useful for debug checks in functions |
139 | * that require that they be called within an RCU read-side critical | |
140 | * section. | |
e3818b8d PM |
141 | * |
142 | * Check debug_lockdep_rcu_enabled() to prevent false positives during boot. | |
c0d6d01b PM |
143 | * |
144 | * Note that rcu_read_lock() is disallowed if the CPU is either idle or | |
145 | * offline from an RCU perspective, so check for those as well. | |
e3818b8d PM |
146 | */ |
147 | int rcu_read_lock_bh_held(void) | |
148 | { | |
149 | if (!debug_lockdep_rcu_enabled()) | |
150 | return 1; | |
e6b80a3b FW |
151 | if (rcu_is_cpu_idle()) |
152 | return 0; | |
c0d6d01b PM |
153 | if (!rcu_lockdep_current_cpu_online()) |
154 | return 0; | |
773e3f93 | 155 | return in_softirq() || irqs_disabled(); |
e3818b8d PM |
156 | } |
157 | EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held); | |
158 | ||
159 | #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ | |
160 | ||
2c42818e PM |
161 | struct rcu_synchronize { |
162 | struct rcu_head head; | |
163 | struct completion completion; | |
164 | }; | |
165 | ||
fbf6bfca PM |
166 | /* |
167 | * Awaken the corresponding synchronize_rcu() instance now that a | |
168 | * grace period has elapsed. | |
169 | */ | |
2c42818e | 170 | static void wakeme_after_rcu(struct rcu_head *head) |
21a1ea9e | 171 | { |
01c1c660 PM |
172 | struct rcu_synchronize *rcu; |
173 | ||
174 | rcu = container_of(head, struct rcu_synchronize, head); | |
175 | complete(&rcu->completion); | |
21a1ea9e | 176 | } |
ee84b824 | 177 | |
2c42818e PM |
178 | void wait_rcu_gp(call_rcu_func_t crf) |
179 | { | |
180 | struct rcu_synchronize rcu; | |
181 | ||
182 | init_rcu_head_on_stack(&rcu.head); | |
183 | init_completion(&rcu.completion); | |
184 | /* Will wake me after RCU finished. */ | |
185 | crf(&rcu.head, wakeme_after_rcu); | |
186 | /* Wait for it. */ | |
187 | wait_for_completion(&rcu.completion); | |
188 | destroy_rcu_head_on_stack(&rcu.head); | |
189 | } | |
190 | EXPORT_SYMBOL_GPL(wait_rcu_gp); | |
191 | ||
ee84b824 PM |
192 | #ifdef CONFIG_PROVE_RCU |
193 | /* | |
194 | * wrapper function to avoid #include problems. | |
195 | */ | |
196 | int rcu_my_thread_group_empty(void) | |
197 | { | |
198 | return thread_group_empty(current); | |
199 | } | |
200 | EXPORT_SYMBOL_GPL(rcu_my_thread_group_empty); | |
201 | #endif /* #ifdef CONFIG_PROVE_RCU */ | |
551d55a9 MD |
202 | |
203 | #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD | |
204 | static inline void debug_init_rcu_head(struct rcu_head *head) | |
205 | { | |
206 | debug_object_init(head, &rcuhead_debug_descr); | |
207 | } | |
208 | ||
209 | static inline void debug_rcu_head_free(struct rcu_head *head) | |
210 | { | |
211 | debug_object_free(head, &rcuhead_debug_descr); | |
212 | } | |
213 | ||
214 | /* | |
215 | * fixup_init is called when: | |
216 | * - an active object is initialized | |
217 | */ | |
218 | static int rcuhead_fixup_init(void *addr, enum debug_obj_state state) | |
219 | { | |
220 | struct rcu_head *head = addr; | |
221 | ||
222 | switch (state) { | |
223 | case ODEBUG_STATE_ACTIVE: | |
224 | /* | |
225 | * Ensure that queued callbacks are all executed. | |
226 | * If we detect that we are nested in a RCU read-side critical | |
227 | * section, we should simply fail, otherwise we would deadlock. | |
fc2ecf7e MD |
228 | * In !PREEMPT configurations, there is no way to tell if we are |
229 | * in a RCU read-side critical section or not, so we never | |
230 | * attempt any fixup and just print a warning. | |
551d55a9 | 231 | */ |
fc2ecf7e | 232 | #ifndef CONFIG_PREEMPT |
108aae22 | 233 | WARN_ON_ONCE(1); |
fc2ecf7e MD |
234 | return 0; |
235 | #endif | |
551d55a9 MD |
236 | if (rcu_preempt_depth() != 0 || preempt_count() != 0 || |
237 | irqs_disabled()) { | |
108aae22 | 238 | WARN_ON_ONCE(1); |
551d55a9 MD |
239 | return 0; |
240 | } | |
241 | rcu_barrier(); | |
242 | rcu_barrier_sched(); | |
243 | rcu_barrier_bh(); | |
244 | debug_object_init(head, &rcuhead_debug_descr); | |
245 | return 1; | |
246 | default: | |
247 | return 0; | |
248 | } | |
249 | } | |
250 | ||
251 | /* | |
252 | * fixup_activate is called when: | |
253 | * - an active object is activated | |
254 | * - an unknown object is activated (might be a statically initialized object) | |
255 | * Activation is performed internally by call_rcu(). | |
256 | */ | |
257 | static int rcuhead_fixup_activate(void *addr, enum debug_obj_state state) | |
258 | { | |
259 | struct rcu_head *head = addr; | |
260 | ||
261 | switch (state) { | |
262 | ||
263 | case ODEBUG_STATE_NOTAVAILABLE: | |
264 | /* | |
265 | * This is not really a fixup. We just make sure that it is | |
266 | * tracked in the object tracker. | |
267 | */ | |
268 | debug_object_init(head, &rcuhead_debug_descr); | |
269 | debug_object_activate(head, &rcuhead_debug_descr); | |
270 | return 0; | |
271 | ||
272 | case ODEBUG_STATE_ACTIVE: | |
273 | /* | |
274 | * Ensure that queued callbacks are all executed. | |
275 | * If we detect that we are nested in a RCU read-side critical | |
276 | * section, we should simply fail, otherwise we would deadlock. | |
fc2ecf7e MD |
277 | * In !PREEMPT configurations, there is no way to tell if we are |
278 | * in a RCU read-side critical section or not, so we never | |
279 | * attempt any fixup and just print a warning. | |
551d55a9 | 280 | */ |
fc2ecf7e | 281 | #ifndef CONFIG_PREEMPT |
108aae22 | 282 | WARN_ON_ONCE(1); |
fc2ecf7e MD |
283 | return 0; |
284 | #endif | |
551d55a9 MD |
285 | if (rcu_preempt_depth() != 0 || preempt_count() != 0 || |
286 | irqs_disabled()) { | |
108aae22 | 287 | WARN_ON_ONCE(1); |
551d55a9 MD |
288 | return 0; |
289 | } | |
290 | rcu_barrier(); | |
291 | rcu_barrier_sched(); | |
292 | rcu_barrier_bh(); | |
293 | debug_object_activate(head, &rcuhead_debug_descr); | |
294 | return 1; | |
295 | default: | |
296 | return 0; | |
297 | } | |
298 | } | |
299 | ||
300 | /* | |
301 | * fixup_free is called when: | |
302 | * - an active object is freed | |
303 | */ | |
304 | static int rcuhead_fixup_free(void *addr, enum debug_obj_state state) | |
305 | { | |
306 | struct rcu_head *head = addr; | |
307 | ||
308 | switch (state) { | |
309 | case ODEBUG_STATE_ACTIVE: | |
310 | /* | |
311 | * Ensure that queued callbacks are all executed. | |
312 | * If we detect that we are nested in a RCU read-side critical | |
313 | * section, we should simply fail, otherwise we would deadlock. | |
fc2ecf7e MD |
314 | * In !PREEMPT configurations, there is no way to tell if we are |
315 | * in a RCU read-side critical section or not, so we never | |
316 | * attempt any fixup and just print a warning. | |
551d55a9 | 317 | */ |
fc2ecf7e | 318 | #ifndef CONFIG_PREEMPT |
108aae22 | 319 | WARN_ON_ONCE(1); |
fc2ecf7e MD |
320 | return 0; |
321 | #endif | |
551d55a9 MD |
322 | if (rcu_preempt_depth() != 0 || preempt_count() != 0 || |
323 | irqs_disabled()) { | |
108aae22 | 324 | WARN_ON_ONCE(1); |
551d55a9 MD |
325 | return 0; |
326 | } | |
327 | rcu_barrier(); | |
328 | rcu_barrier_sched(); | |
329 | rcu_barrier_bh(); | |
330 | debug_object_free(head, &rcuhead_debug_descr); | |
331 | return 1; | |
551d55a9 MD |
332 | default: |
333 | return 0; | |
334 | } | |
335 | } | |
336 | ||
337 | /** | |
338 | * init_rcu_head_on_stack() - initialize on-stack rcu_head for debugobjects | |
339 | * @head: pointer to rcu_head structure to be initialized | |
340 | * | |
341 | * This function informs debugobjects of a new rcu_head structure that | |
342 | * has been allocated as an auto variable on the stack. This function | |
343 | * is not required for rcu_head structures that are statically defined or | |
344 | * that are dynamically allocated on the heap. This function has no | |
345 | * effect for !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds. | |
346 | */ | |
347 | void init_rcu_head_on_stack(struct rcu_head *head) | |
348 | { | |
349 | debug_object_init_on_stack(head, &rcuhead_debug_descr); | |
350 | } | |
351 | EXPORT_SYMBOL_GPL(init_rcu_head_on_stack); | |
352 | ||
353 | /** | |
354 | * destroy_rcu_head_on_stack() - destroy on-stack rcu_head for debugobjects | |
355 | * @head: pointer to rcu_head structure to be initialized | |
356 | * | |
357 | * This function informs debugobjects that an on-stack rcu_head structure | |
358 | * is about to go out of scope. As with init_rcu_head_on_stack(), this | |
359 | * function is not required for rcu_head structures that are statically | |
360 | * defined or that are dynamically allocated on the heap. Also as with | |
361 | * init_rcu_head_on_stack(), this function has no effect for | |
362 | * !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds. | |
363 | */ | |
364 | void destroy_rcu_head_on_stack(struct rcu_head *head) | |
365 | { | |
366 | debug_object_free(head, &rcuhead_debug_descr); | |
367 | } | |
368 | EXPORT_SYMBOL_GPL(destroy_rcu_head_on_stack); | |
369 | ||
370 | struct debug_obj_descr rcuhead_debug_descr = { | |
371 | .name = "rcu_head", | |
372 | .fixup_init = rcuhead_fixup_init, | |
373 | .fixup_activate = rcuhead_fixup_activate, | |
374 | .fixup_free = rcuhead_fixup_free, | |
375 | }; | |
376 | EXPORT_SYMBOL_GPL(rcuhead_debug_descr); | |
377 | #endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */ | |
91afaf30 PM |
378 | |
379 | #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU) || defined(CONFIG_RCU_TRACE) | |
52494535 PM |
380 | void do_trace_rcu_torture_read(char *rcutorturename, struct rcu_head *rhp, |
381 | unsigned long secs, | |
382 | unsigned long c_old, unsigned long c) | |
91afaf30 | 383 | { |
52494535 | 384 | trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c); |
91afaf30 PM |
385 | } |
386 | EXPORT_SYMBOL_GPL(do_trace_rcu_torture_read); | |
387 | #else | |
52494535 PM |
388 | #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \ |
389 | do { } while (0) | |
91afaf30 | 390 | #endif |
6bfc09e2 PM |
391 | |
392 | #ifdef CONFIG_RCU_STALL_COMMON | |
393 | ||
394 | #ifdef CONFIG_PROVE_RCU | |
395 | #define RCU_STALL_DELAY_DELTA (5 * HZ) | |
396 | #else | |
397 | #define RCU_STALL_DELAY_DELTA 0 | |
398 | #endif | |
399 | ||
400 | int rcu_cpu_stall_suppress __read_mostly; /* 1 = suppress stall warnings. */ | |
401 | int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT; | |
402 | ||
403 | module_param(rcu_cpu_stall_suppress, int, 0644); | |
404 | module_param(rcu_cpu_stall_timeout, int, 0644); | |
405 | ||
406 | int rcu_jiffies_till_stall_check(void) | |
407 | { | |
408 | int till_stall_check = ACCESS_ONCE(rcu_cpu_stall_timeout); | |
409 | ||
410 | /* | |
411 | * Limit check must be consistent with the Kconfig limits | |
412 | * for CONFIG_RCU_CPU_STALL_TIMEOUT. | |
413 | */ | |
414 | if (till_stall_check < 3) { | |
415 | ACCESS_ONCE(rcu_cpu_stall_timeout) = 3; | |
416 | till_stall_check = 3; | |
417 | } else if (till_stall_check > 300) { | |
418 | ACCESS_ONCE(rcu_cpu_stall_timeout) = 300; | |
419 | till_stall_check = 300; | |
420 | } | |
421 | return till_stall_check * HZ + RCU_STALL_DELAY_DELTA; | |
422 | } | |
423 | ||
424 | static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr) | |
425 | { | |
426 | rcu_cpu_stall_suppress = 1; | |
427 | return NOTIFY_DONE; | |
428 | } | |
429 | ||
430 | static struct notifier_block rcu_panic_block = { | |
431 | .notifier_call = rcu_panic, | |
432 | }; | |
433 | ||
434 | static int __init check_cpu_stall_init(void) | |
435 | { | |
436 | atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block); | |
437 | return 0; | |
438 | } | |
439 | early_initcall(check_cpu_stall_init); | |
440 | ||
441 | #endif /* #ifdef CONFIG_RCU_STALL_COMMON */ |