diff options
Diffstat (limited to 'kernel/rcu/update.c')
-rw-r--r-- | kernel/rcu/update.c | 347 |
1 files changed, 347 insertions, 0 deletions
diff --git a/kernel/rcu/update.c b/kernel/rcu/update.c new file mode 100644 index 000000000000..6cb3dff89e2b --- /dev/null +++ b/kernel/rcu/update.c | |||
@@ -0,0 +1,347 @@ | |||
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 | * | ||
18 | * Copyright IBM Corporation, 2001 | ||
19 | * | ||
20 | * Authors: Dipankar Sarma <dipankar@in.ibm.com> | ||
21 | * Manfred Spraul <manfred@colorfullife.com> | ||
22 | * | ||
23 | * Based on the original work by Paul McKenney <paulmck@us.ibm.com> | ||
24 | * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen. | ||
25 | * Papers: | ||
26 | * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf | ||
27 | * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001) | ||
28 | * | ||
29 | * For detailed explanation of Read-Copy Update mechanism see - | ||
30 | * http://lse.sourceforge.net/locking/rcupdate.html | ||
31 | * | ||
32 | */ | ||
33 | #include <linux/types.h> | ||
34 | #include <linux/kernel.h> | ||
35 | #include <linux/init.h> | ||
36 | #include <linux/spinlock.h> | ||
37 | #include <linux/smp.h> | ||
38 | #include <linux/interrupt.h> | ||
39 | #include <linux/sched.h> | ||
40 | #include <linux/atomic.h> | ||
41 | #include <linux/bitops.h> | ||
42 | #include <linux/percpu.h> | ||
43 | #include <linux/notifier.h> | ||
44 | #include <linux/cpu.h> | ||
45 | #include <linux/mutex.h> | ||
46 | #include <linux/export.h> | ||
47 | #include <linux/hardirq.h> | ||
48 | #include <linux/delay.h> | ||
49 | #include <linux/module.h> | ||
50 | |||
51 | #define CREATE_TRACE_POINTS | ||
52 | #include <trace/events/rcu.h> | ||
53 | |||
54 | #include "rcu.h" | ||
55 | |||
56 | MODULE_ALIAS("rcupdate"); | ||
57 | #ifdef MODULE_PARAM_PREFIX | ||
58 | #undef MODULE_PARAM_PREFIX | ||
59 | #endif | ||
60 | #define MODULE_PARAM_PREFIX "rcupdate." | ||
61 | |||
62 | module_param(rcu_expedited, int, 0); | ||
63 | |||
64 | #ifdef CONFIG_PREEMPT_RCU | ||
65 | |||
66 | /* | ||
67 | * Preemptible RCU implementation for rcu_read_lock(). | ||
68 | * Just increment ->rcu_read_lock_nesting, shared state will be updated | ||
69 | * if we block. | ||
70 | */ | ||
71 | void __rcu_read_lock(void) | ||
72 | { | ||
73 | current->rcu_read_lock_nesting++; | ||
74 | barrier(); /* critical section after entry code. */ | ||
75 | } | ||
76 | EXPORT_SYMBOL_GPL(__rcu_read_lock); | ||
77 | |||
78 | /* | ||
79 | * Preemptible RCU implementation for rcu_read_unlock(). | ||
80 | * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost | ||
81 | * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then | ||
82 | * invoke rcu_read_unlock_special() to clean up after a context switch | ||
83 | * in an RCU read-side critical section and other special cases. | ||
84 | */ | ||
85 | void __rcu_read_unlock(void) | ||
86 | { | ||
87 | struct task_struct *t = current; | ||
88 | |||
89 | if (t->rcu_read_lock_nesting != 1) { | ||
90 | --t->rcu_read_lock_nesting; | ||
91 | } else { | ||
92 | barrier(); /* critical section before exit code. */ | ||
93 | t->rcu_read_lock_nesting = INT_MIN; | ||
94 | #ifdef CONFIG_PROVE_RCU_DELAY | ||
95 | udelay(10); /* Make preemption more probable. */ | ||
96 | #endif /* #ifdef CONFIG_PROVE_RCU_DELAY */ | ||
97 | barrier(); /* assign before ->rcu_read_unlock_special load */ | ||
98 | if (unlikely(ACCESS_ONCE(t->rcu_read_unlock_special))) | ||
99 | rcu_read_unlock_special(t); | ||
100 | barrier(); /* ->rcu_read_unlock_special load before assign */ | ||
101 | t->rcu_read_lock_nesting = 0; | ||
102 | } | ||
103 | #ifdef CONFIG_PROVE_LOCKING | ||
104 | { | ||
105 | int rrln = ACCESS_ONCE(t->rcu_read_lock_nesting); | ||
106 | |||
107 | WARN_ON_ONCE(rrln < 0 && rrln > INT_MIN / 2); | ||
108 | } | ||
109 | #endif /* #ifdef CONFIG_PROVE_LOCKING */ | ||
110 | } | ||
111 | EXPORT_SYMBOL_GPL(__rcu_read_unlock); | ||
112 | |||
113 | #endif /* #ifdef CONFIG_PREEMPT_RCU */ | ||
114 | |||
115 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | ||
116 | static struct lock_class_key rcu_lock_key; | ||
117 | struct lockdep_map rcu_lock_map = | ||
118 | STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key); | ||
119 | EXPORT_SYMBOL_GPL(rcu_lock_map); | ||
120 | |||
121 | static struct lock_class_key rcu_bh_lock_key; | ||
122 | struct lockdep_map rcu_bh_lock_map = | ||
123 | STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_bh", &rcu_bh_lock_key); | ||
124 | EXPORT_SYMBOL_GPL(rcu_bh_lock_map); | ||
125 | |||
126 | static struct lock_class_key rcu_sched_lock_key; | ||
127 | struct lockdep_map rcu_sched_lock_map = | ||
128 | STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_sched", &rcu_sched_lock_key); | ||
129 | EXPORT_SYMBOL_GPL(rcu_sched_lock_map); | ||
130 | |||
131 | int notrace debug_lockdep_rcu_enabled(void) | ||
132 | { | ||
133 | return rcu_scheduler_active && debug_locks && | ||
134 | current->lockdep_recursion == 0; | ||
135 | } | ||
136 | EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled); | ||
137 | |||
138 | /** | ||
139 | * rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section? | ||
140 | * | ||
141 | * Check for bottom half being disabled, which covers both the | ||
142 | * CONFIG_PROVE_RCU and not cases. Note that if someone uses | ||
143 | * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled) | ||
144 | * will show the situation. This is useful for debug checks in functions | ||
145 | * that require that they be called within an RCU read-side critical | ||
146 | * section. | ||
147 | * | ||
148 | * Check debug_lockdep_rcu_enabled() to prevent false positives during boot. | ||
149 | * | ||
150 | * Note that rcu_read_lock() is disallowed if the CPU is either idle or | ||
151 | * offline from an RCU perspective, so check for those as well. | ||
152 | */ | ||
153 | int rcu_read_lock_bh_held(void) | ||
154 | { | ||
155 | if (!debug_lockdep_rcu_enabled()) | ||
156 | return 1; | ||
157 | if (!rcu_is_watching()) | ||
158 | return 0; | ||
159 | if (!rcu_lockdep_current_cpu_online()) | ||
160 | return 0; | ||
161 | return in_softirq() || irqs_disabled(); | ||
162 | } | ||
163 | EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held); | ||
164 | |||
165 | #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ | ||
166 | |||
167 | struct rcu_synchronize { | ||
168 | struct rcu_head head; | ||
169 | struct completion completion; | ||
170 | }; | ||
171 | |||
172 | /* | ||
173 | * Awaken the corresponding synchronize_rcu() instance now that a | ||
174 | * grace period has elapsed. | ||
175 | */ | ||
176 | static void wakeme_after_rcu(struct rcu_head *head) | ||
177 | { | ||
178 | struct rcu_synchronize *rcu; | ||
179 | |||
180 | rcu = container_of(head, struct rcu_synchronize, head); | ||
181 | complete(&rcu->completion); | ||
182 | } | ||
183 | |||
184 | void wait_rcu_gp(call_rcu_func_t crf) | ||
185 | { | ||
186 | struct rcu_synchronize rcu; | ||
187 | |||
188 | init_rcu_head_on_stack(&rcu.head); | ||
189 | init_completion(&rcu.completion); | ||
190 | /* Will wake me after RCU finished. */ | ||
191 | crf(&rcu.head, wakeme_after_rcu); | ||
192 | /* Wait for it. */ | ||
193 | wait_for_completion(&rcu.completion); | ||
194 | destroy_rcu_head_on_stack(&rcu.head); | ||
195 | } | ||
196 | EXPORT_SYMBOL_GPL(wait_rcu_gp); | ||
197 | |||
198 | #ifdef CONFIG_PROVE_RCU | ||
199 | /* | ||
200 | * wrapper function to avoid #include problems. | ||
201 | */ | ||
202 | int rcu_my_thread_group_empty(void) | ||
203 | { | ||
204 | return thread_group_empty(current); | ||
205 | } | ||
206 | EXPORT_SYMBOL_GPL(rcu_my_thread_group_empty); | ||
207 | #endif /* #ifdef CONFIG_PROVE_RCU */ | ||
208 | |||
209 | #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD | ||
210 | static inline void debug_init_rcu_head(struct rcu_head *head) | ||
211 | { | ||
212 | debug_object_init(head, &rcuhead_debug_descr); | ||
213 | } | ||
214 | |||
215 | static inline void debug_rcu_head_free(struct rcu_head *head) | ||
216 | { | ||
217 | debug_object_free(head, &rcuhead_debug_descr); | ||
218 | } | ||
219 | |||
220 | /* | ||
221 | * fixup_activate is called when: | ||
222 | * - an active object is activated | ||
223 | * - an unknown object is activated (might be a statically initialized object) | ||
224 | * Activation is performed internally by call_rcu(). | ||
225 | */ | ||
226 | static int rcuhead_fixup_activate(void *addr, enum debug_obj_state state) | ||
227 | { | ||
228 | struct rcu_head *head = addr; | ||
229 | |||
230 | switch (state) { | ||
231 | |||
232 | case ODEBUG_STATE_NOTAVAILABLE: | ||
233 | /* | ||
234 | * This is not really a fixup. We just make sure that it is | ||
235 | * tracked in the object tracker. | ||
236 | */ | ||
237 | debug_object_init(head, &rcuhead_debug_descr); | ||
238 | debug_object_activate(head, &rcuhead_debug_descr); | ||
239 | return 0; | ||
240 | default: | ||
241 | return 1; | ||
242 | } | ||
243 | } | ||
244 | |||
245 | /** | ||
246 | * init_rcu_head_on_stack() - initialize on-stack rcu_head for debugobjects | ||
247 | * @head: pointer to rcu_head structure to be initialized | ||
248 | * | ||
249 | * This function informs debugobjects of a new rcu_head structure that | ||
250 | * has been allocated as an auto variable on the stack. This function | ||
251 | * is not required for rcu_head structures that are statically defined or | ||
252 | * that are dynamically allocated on the heap. This function has no | ||
253 | * effect for !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds. | ||
254 | */ | ||
255 | void init_rcu_head_on_stack(struct rcu_head *head) | ||
256 | { | ||
257 | debug_object_init_on_stack(head, &rcuhead_debug_descr); | ||
258 | } | ||
259 | EXPORT_SYMBOL_GPL(init_rcu_head_on_stack); | ||
260 | |||
261 | /** | ||
262 | * destroy_rcu_head_on_stack() - destroy on-stack rcu_head for debugobjects | ||
263 | * @head: pointer to rcu_head structure to be initialized | ||
264 | * | ||
265 | * This function informs debugobjects that an on-stack rcu_head structure | ||
266 | * is about to go out of scope. As with init_rcu_head_on_stack(), this | ||
267 | * function is not required for rcu_head structures that are statically | ||
268 | * defined or that are dynamically allocated on the heap. Also as with | ||
269 | * init_rcu_head_on_stack(), this function has no effect for | ||
270 | * !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds. | ||
271 | */ | ||
272 | void destroy_rcu_head_on_stack(struct rcu_head *head) | ||
273 | { | ||
274 | debug_object_free(head, &rcuhead_debug_descr); | ||
275 | } | ||
276 | EXPORT_SYMBOL_GPL(destroy_rcu_head_on_stack); | ||
277 | |||
278 | struct debug_obj_descr rcuhead_debug_descr = { | ||
279 | .name = "rcu_head", | ||
280 | .fixup_activate = rcuhead_fixup_activate, | ||
281 | }; | ||
282 | EXPORT_SYMBOL_GPL(rcuhead_debug_descr); | ||
283 | #endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */ | ||
284 | |||
285 | #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU) || defined(CONFIG_RCU_TRACE) | ||
286 | void do_trace_rcu_torture_read(const char *rcutorturename, struct rcu_head *rhp, | ||
287 | unsigned long secs, | ||
288 | unsigned long c_old, unsigned long c) | ||
289 | { | ||
290 | trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c); | ||
291 | } | ||
292 | EXPORT_SYMBOL_GPL(do_trace_rcu_torture_read); | ||
293 | #else | ||
294 | #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \ | ||
295 | do { } while (0) | ||
296 | #endif | ||
297 | |||
298 | #ifdef CONFIG_RCU_STALL_COMMON | ||
299 | |||
300 | #ifdef CONFIG_PROVE_RCU | ||
301 | #define RCU_STALL_DELAY_DELTA (5 * HZ) | ||
302 | #else | ||
303 | #define RCU_STALL_DELAY_DELTA 0 | ||
304 | #endif | ||
305 | |||
306 | int rcu_cpu_stall_suppress __read_mostly; /* 1 = suppress stall warnings. */ | ||
307 | static int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT; | ||
308 | |||
309 | module_param(rcu_cpu_stall_suppress, int, 0644); | ||
310 | module_param(rcu_cpu_stall_timeout, int, 0644); | ||
311 | |||
312 | int rcu_jiffies_till_stall_check(void) | ||
313 | { | ||
314 | int till_stall_check = ACCESS_ONCE(rcu_cpu_stall_timeout); | ||
315 | |||
316 | /* | ||
317 | * Limit check must be consistent with the Kconfig limits | ||
318 | * for CONFIG_RCU_CPU_STALL_TIMEOUT. | ||
319 | */ | ||
320 | if (till_stall_check < 3) { | ||
321 | ACCESS_ONCE(rcu_cpu_stall_timeout) = 3; | ||
322 | till_stall_check = 3; | ||
323 | } else if (till_stall_check > 300) { | ||
324 | ACCESS_ONCE(rcu_cpu_stall_timeout) = 300; | ||
325 | till_stall_check = 300; | ||
326 | } | ||
327 | return till_stall_check * HZ + RCU_STALL_DELAY_DELTA; | ||
328 | } | ||
329 | |||
330 | static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr) | ||
331 | { | ||
332 | rcu_cpu_stall_suppress = 1; | ||
333 | return NOTIFY_DONE; | ||
334 | } | ||
335 | |||
336 | static struct notifier_block rcu_panic_block = { | ||
337 | .notifier_call = rcu_panic, | ||
338 | }; | ||
339 | |||
340 | static int __init check_cpu_stall_init(void) | ||
341 | { | ||
342 | atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block); | ||
343 | return 0; | ||
344 | } | ||
345 | early_initcall(check_cpu_stall_init); | ||
346 | |||
347 | #endif /* #ifdef CONFIG_RCU_STALL_COMMON */ | ||