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 | * | |
18 | * Copyright (C) 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> | |
e56d0903 | 38 | #include <linux/rcupdate.h> |
1da177e4 LT |
39 | #include <linux/interrupt.h> |
40 | #include <linux/sched.h> | |
41 | #include <asm/atomic.h> | |
42 | #include <linux/bitops.h> | |
43 | #include <linux/module.h> | |
44 | #include <linux/completion.h> | |
45 | #include <linux/moduleparam.h> | |
46 | #include <linux/percpu.h> | |
47 | #include <linux/notifier.h> | |
48 | #include <linux/rcupdate.h> | |
49 | #include <linux/cpu.h> | |
9331b315 | 50 | #include <linux/mutex.h> |
1da177e4 | 51 | |
851a67b8 PZ |
52 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
53 | static struct lock_class_key rcu_lock_key; | |
54 | struct lockdep_map rcu_lock_map = | |
55 | STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key); | |
56 | ||
57 | EXPORT_SYMBOL_GPL(rcu_lock_map); | |
58 | #endif | |
59 | ||
1da177e4 | 60 | /* Definition for rcupdate control block. */ |
2178426d | 61 | static struct rcu_ctrlblk rcu_ctrlblk = { |
69a0b315 ON |
62 | .cur = -300, |
63 | .completed = -300, | |
e4d91918 | 64 | .lock = __SPIN_LOCK_UNLOCKED(&rcu_ctrlblk.lock), |
69a0b315 ON |
65 | .cpumask = CPU_MASK_NONE, |
66 | }; | |
2178426d | 67 | static struct rcu_ctrlblk rcu_bh_ctrlblk = { |
69a0b315 ON |
68 | .cur = -300, |
69 | .completed = -300, | |
e4d91918 | 70 | .lock = __SPIN_LOCK_UNLOCKED(&rcu_bh_ctrlblk.lock), |
69a0b315 | 71 | .cpumask = CPU_MASK_NONE, |
1da177e4 | 72 | }; |
1da177e4 LT |
73 | |
74 | DEFINE_PER_CPU(struct rcu_data, rcu_data) = { 0L }; | |
75 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data) = { 0L }; | |
76 | ||
77 | /* Fake initialization required by compiler */ | |
78 | static DEFINE_PER_CPU(struct tasklet_struct, rcu_tasklet) = {NULL}; | |
21a1ea9e DS |
79 | static int blimit = 10; |
80 | static int qhimark = 10000; | |
81 | static int qlowmark = 100; | |
21a1ea9e DS |
82 | |
83 | static atomic_t rcu_barrier_cpu_count; | |
9331b315 | 84 | static DEFINE_MUTEX(rcu_barrier_mutex); |
21a1ea9e DS |
85 | static struct completion rcu_barrier_completion; |
86 | ||
87 | #ifdef CONFIG_SMP | |
88 | static void force_quiescent_state(struct rcu_data *rdp, | |
89 | struct rcu_ctrlblk *rcp) | |
90 | { | |
91 | int cpu; | |
92 | cpumask_t cpumask; | |
93 | set_need_resched(); | |
20e9751b ON |
94 | if (unlikely(!rcp->signaled)) { |
95 | rcp->signaled = 1; | |
21a1ea9e DS |
96 | /* |
97 | * Don't send IPI to itself. With irqs disabled, | |
98 | * rdp->cpu is the current cpu. | |
99 | */ | |
100 | cpumask = rcp->cpumask; | |
101 | cpu_clear(rdp->cpu, cpumask); | |
102 | for_each_cpu_mask(cpu, cpumask) | |
103 | smp_send_reschedule(cpu); | |
104 | } | |
105 | } | |
106 | #else | |
107 | static inline void force_quiescent_state(struct rcu_data *rdp, | |
108 | struct rcu_ctrlblk *rcp) | |
109 | { | |
110 | set_need_resched(); | |
111 | } | |
112 | #endif | |
1da177e4 LT |
113 | |
114 | /** | |
115 | * call_rcu - Queue an RCU callback for invocation after a grace period. | |
116 | * @head: structure to be used for queueing the RCU updates. | |
117 | * @func: actual update function to be invoked after the grace period | |
118 | * | |
119 | * The update function will be invoked some time after a full grace | |
120 | * period elapses, in other words after all currently executing RCU | |
121 | * read-side critical sections have completed. RCU read-side critical | |
122 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), | |
123 | * and may be nested. | |
124 | */ | |
125 | void fastcall call_rcu(struct rcu_head *head, | |
126 | void (*func)(struct rcu_head *rcu)) | |
127 | { | |
128 | unsigned long flags; | |
129 | struct rcu_data *rdp; | |
130 | ||
131 | head->func = func; | |
132 | head->next = NULL; | |
133 | local_irq_save(flags); | |
134 | rdp = &__get_cpu_var(rcu_data); | |
135 | *rdp->nxttail = head; | |
136 | rdp->nxttail = &head->next; | |
21a1ea9e DS |
137 | if (unlikely(++rdp->qlen > qhimark)) { |
138 | rdp->blimit = INT_MAX; | |
139 | force_quiescent_state(rdp, &rcu_ctrlblk); | |
140 | } | |
1da177e4 LT |
141 | local_irq_restore(flags); |
142 | } | |
143 | ||
144 | /** | |
145 | * call_rcu_bh - Queue an RCU for invocation after a quicker grace period. | |
146 | * @head: structure to be used for queueing the RCU updates. | |
147 | * @func: actual update function to be invoked after the grace period | |
148 | * | |
149 | * The update function will be invoked some time after a full grace | |
150 | * period elapses, in other words after all currently executing RCU | |
151 | * read-side critical sections have completed. call_rcu_bh() assumes | |
152 | * that the read-side critical sections end on completion of a softirq | |
153 | * handler. This means that read-side critical sections in process | |
154 | * context must not be interrupted by softirqs. This interface is to be | |
155 | * used when most of the read-side critical sections are in softirq context. | |
156 | * RCU read-side critical sections are delimited by rcu_read_lock() and | |
157 | * rcu_read_unlock(), * if in interrupt context or rcu_read_lock_bh() | |
158 | * and rcu_read_unlock_bh(), if in process context. These may be nested. | |
159 | */ | |
160 | void fastcall call_rcu_bh(struct rcu_head *head, | |
161 | void (*func)(struct rcu_head *rcu)) | |
162 | { | |
163 | unsigned long flags; | |
164 | struct rcu_data *rdp; | |
165 | ||
166 | head->func = func; | |
167 | head->next = NULL; | |
168 | local_irq_save(flags); | |
169 | rdp = &__get_cpu_var(rcu_bh_data); | |
170 | *rdp->nxttail = head; | |
171 | rdp->nxttail = &head->next; | |
21a1ea9e DS |
172 | |
173 | if (unlikely(++rdp->qlen > qhimark)) { | |
174 | rdp->blimit = INT_MAX; | |
175 | force_quiescent_state(rdp, &rcu_bh_ctrlblk); | |
176 | } | |
177 | ||
1da177e4 LT |
178 | local_irq_restore(flags); |
179 | } | |
180 | ||
a241ec65 PM |
181 | /* |
182 | * Return the number of RCU batches processed thus far. Useful | |
183 | * for debug and statistics. | |
184 | */ | |
185 | long rcu_batches_completed(void) | |
186 | { | |
187 | return rcu_ctrlblk.completed; | |
188 | } | |
189 | ||
c32e0660 PM |
190 | /* |
191 | * Return the number of RCU batches processed thus far. Useful | |
192 | * for debug and statistics. | |
193 | */ | |
194 | long rcu_batches_completed_bh(void) | |
195 | { | |
196 | return rcu_bh_ctrlblk.completed; | |
197 | } | |
198 | ||
ab4720ec DS |
199 | static void rcu_barrier_callback(struct rcu_head *notused) |
200 | { | |
201 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
202 | complete(&rcu_barrier_completion); | |
203 | } | |
204 | ||
205 | /* | |
206 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
207 | */ | |
208 | static void rcu_barrier_func(void *notused) | |
209 | { | |
210 | int cpu = smp_processor_id(); | |
211 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); | |
212 | struct rcu_head *head; | |
213 | ||
214 | head = &rdp->barrier; | |
215 | atomic_inc(&rcu_barrier_cpu_count); | |
216 | call_rcu(head, rcu_barrier_callback); | |
217 | } | |
218 | ||
219 | /** | |
220 | * rcu_barrier - Wait until all the in-flight RCUs are complete. | |
221 | */ | |
222 | void rcu_barrier(void) | |
223 | { | |
224 | BUG_ON(in_interrupt()); | |
9331b315 IM |
225 | /* Take cpucontrol mutex to protect against CPU hotplug */ |
226 | mutex_lock(&rcu_barrier_mutex); | |
ab4720ec DS |
227 | init_completion(&rcu_barrier_completion); |
228 | atomic_set(&rcu_barrier_cpu_count, 0); | |
229 | on_each_cpu(rcu_barrier_func, NULL, 0, 1); | |
230 | wait_for_completion(&rcu_barrier_completion); | |
9331b315 | 231 | mutex_unlock(&rcu_barrier_mutex); |
ab4720ec DS |
232 | } |
233 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
234 | ||
1da177e4 LT |
235 | /* |
236 | * Invoke the completed RCU callbacks. They are expected to be in | |
237 | * a per-cpu list. | |
238 | */ | |
239 | static void rcu_do_batch(struct rcu_data *rdp) | |
240 | { | |
241 | struct rcu_head *next, *list; | |
242 | int count = 0; | |
243 | ||
244 | list = rdp->donelist; | |
245 | while (list) { | |
1c69d921 ED |
246 | next = list->next; |
247 | prefetch(next); | |
1da177e4 LT |
248 | list->func(list); |
249 | list = next; | |
21a1ea9e | 250 | if (++count >= rdp->blimit) |
1da177e4 LT |
251 | break; |
252 | } | |
1c69d921 | 253 | rdp->donelist = list; |
dd9daa22 ON |
254 | |
255 | local_irq_disable(); | |
256 | rdp->qlen -= count; | |
257 | local_irq_enable(); | |
21a1ea9e DS |
258 | if (rdp->blimit == INT_MAX && rdp->qlen <= qlowmark) |
259 | rdp->blimit = blimit; | |
dd9daa22 | 260 | |
1da177e4 LT |
261 | if (!rdp->donelist) |
262 | rdp->donetail = &rdp->donelist; | |
263 | else | |
264 | tasklet_schedule(&per_cpu(rcu_tasklet, rdp->cpu)); | |
265 | } | |
266 | ||
267 | /* | |
268 | * Grace period handling: | |
269 | * The grace period handling consists out of two steps: | |
270 | * - A new grace period is started. | |
271 | * This is done by rcu_start_batch. The start is not broadcasted to | |
272 | * all cpus, they must pick this up by comparing rcp->cur with | |
273 | * rdp->quiescbatch. All cpus are recorded in the | |
69a0b315 | 274 | * rcu_ctrlblk.cpumask bitmap. |
1da177e4 LT |
275 | * - All cpus must go through a quiescent state. |
276 | * Since the start of the grace period is not broadcasted, at least two | |
277 | * calls to rcu_check_quiescent_state are required: | |
278 | * The first call just notices that a new grace period is running. The | |
279 | * following calls check if there was a quiescent state since the beginning | |
69a0b315 | 280 | * of the grace period. If so, it updates rcu_ctrlblk.cpumask. If |
1da177e4 LT |
281 | * the bitmap is empty, then the grace period is completed. |
282 | * rcu_check_quiescent_state calls rcu_start_batch(0) to start the next grace | |
283 | * period (if necessary). | |
284 | */ | |
285 | /* | |
286 | * Register a new batch of callbacks, and start it up if there is currently no | |
287 | * active batch and the batch to be registered has not already occurred. | |
69a0b315 | 288 | * Caller must hold rcu_ctrlblk.lock. |
1da177e4 | 289 | */ |
69a0b315 | 290 | static void rcu_start_batch(struct rcu_ctrlblk *rcp) |
1da177e4 | 291 | { |
1da177e4 LT |
292 | if (rcp->next_pending && |
293 | rcp->completed == rcp->cur) { | |
1da177e4 | 294 | rcp->next_pending = 0; |
c3f59023 SV |
295 | /* |
296 | * next_pending == 0 must be visible in | |
297 | * __rcu_process_callbacks() before it can see new value of cur. | |
1da177e4 LT |
298 | */ |
299 | smp_wmb(); | |
300 | rcp->cur++; | |
c3f59023 SV |
301 | |
302 | /* | |
303 | * Accessing nohz_cpu_mask before incrementing rcp->cur needs a | |
304 | * Barrier Otherwise it can cause tickless idle CPUs to be | |
69a0b315 | 305 | * included in rcp->cpumask, which will extend graceperiods |
c3f59023 SV |
306 | * unnecessarily. |
307 | */ | |
308 | smp_mb(); | |
69a0b315 | 309 | cpus_andnot(rcp->cpumask, cpu_online_map, nohz_cpu_mask); |
c3f59023 | 310 | |
20e9751b | 311 | rcp->signaled = 0; |
1da177e4 LT |
312 | } |
313 | } | |
314 | ||
315 | /* | |
316 | * cpu went through a quiescent state since the beginning of the grace period. | |
317 | * Clear it from the cpu mask and complete the grace period if it was the last | |
318 | * cpu. Start another grace period if someone has further entries pending | |
319 | */ | |
69a0b315 | 320 | static void cpu_quiet(int cpu, struct rcu_ctrlblk *rcp) |
1da177e4 | 321 | { |
69a0b315 ON |
322 | cpu_clear(cpu, rcp->cpumask); |
323 | if (cpus_empty(rcp->cpumask)) { | |
1da177e4 LT |
324 | /* batch completed ! */ |
325 | rcp->completed = rcp->cur; | |
69a0b315 | 326 | rcu_start_batch(rcp); |
1da177e4 LT |
327 | } |
328 | } | |
329 | ||
330 | /* | |
331 | * Check if the cpu has gone through a quiescent state (say context | |
332 | * switch). If so and if it already hasn't done so in this RCU | |
333 | * quiescent cycle, then indicate that it has done so. | |
334 | */ | |
335 | static void rcu_check_quiescent_state(struct rcu_ctrlblk *rcp, | |
69a0b315 | 336 | struct rcu_data *rdp) |
1da177e4 LT |
337 | { |
338 | if (rdp->quiescbatch != rcp->cur) { | |
339 | /* start new grace period: */ | |
340 | rdp->qs_pending = 1; | |
341 | rdp->passed_quiesc = 0; | |
342 | rdp->quiescbatch = rcp->cur; | |
343 | return; | |
344 | } | |
345 | ||
346 | /* Grace period already completed for this cpu? | |
347 | * qs_pending is checked instead of the actual bitmap to avoid | |
348 | * cacheline trashing. | |
349 | */ | |
350 | if (!rdp->qs_pending) | |
351 | return; | |
352 | ||
353 | /* | |
354 | * Was there a quiescent state since the beginning of the grace | |
355 | * period? If no, then exit and wait for the next call. | |
356 | */ | |
357 | if (!rdp->passed_quiesc) | |
358 | return; | |
359 | rdp->qs_pending = 0; | |
360 | ||
69a0b315 | 361 | spin_lock(&rcp->lock); |
1da177e4 LT |
362 | /* |
363 | * rdp->quiescbatch/rcp->cur and the cpu bitmap can come out of sync | |
364 | * during cpu startup. Ignore the quiescent state. | |
365 | */ | |
366 | if (likely(rdp->quiescbatch == rcp->cur)) | |
69a0b315 | 367 | cpu_quiet(rdp->cpu, rcp); |
1da177e4 | 368 | |
69a0b315 | 369 | spin_unlock(&rcp->lock); |
1da177e4 LT |
370 | } |
371 | ||
372 | ||
373 | #ifdef CONFIG_HOTPLUG_CPU | |
374 | ||
375 | /* warning! helper for rcu_offline_cpu. do not use elsewhere without reviewing | |
376 | * locking requirements, the list it's pulling from has to belong to a cpu | |
377 | * which is dead and hence not processing interrupts. | |
378 | */ | |
379 | static void rcu_move_batch(struct rcu_data *this_rdp, struct rcu_head *list, | |
380 | struct rcu_head **tail) | |
381 | { | |
382 | local_irq_disable(); | |
383 | *this_rdp->nxttail = list; | |
384 | if (list) | |
385 | this_rdp->nxttail = tail; | |
386 | local_irq_enable(); | |
387 | } | |
388 | ||
389 | static void __rcu_offline_cpu(struct rcu_data *this_rdp, | |
69a0b315 | 390 | struct rcu_ctrlblk *rcp, struct rcu_data *rdp) |
1da177e4 LT |
391 | { |
392 | /* if the cpu going offline owns the grace period | |
393 | * we can block indefinitely waiting for it, so flush | |
394 | * it here | |
395 | */ | |
69a0b315 | 396 | spin_lock_bh(&rcp->lock); |
1da177e4 | 397 | if (rcp->cur != rcp->completed) |
69a0b315 ON |
398 | cpu_quiet(rdp->cpu, rcp); |
399 | spin_unlock_bh(&rcp->lock); | |
1da177e4 LT |
400 | rcu_move_batch(this_rdp, rdp->curlist, rdp->curtail); |
401 | rcu_move_batch(this_rdp, rdp->nxtlist, rdp->nxttail); | |
a9c82815 | 402 | rcu_move_batch(this_rdp, rdp->donelist, rdp->donetail); |
1da177e4 | 403 | } |
a9c82815 | 404 | |
1da177e4 LT |
405 | static void rcu_offline_cpu(int cpu) |
406 | { | |
407 | struct rcu_data *this_rdp = &get_cpu_var(rcu_data); | |
408 | struct rcu_data *this_bh_rdp = &get_cpu_var(rcu_bh_data); | |
409 | ||
69a0b315 | 410 | __rcu_offline_cpu(this_rdp, &rcu_ctrlblk, |
1da177e4 | 411 | &per_cpu(rcu_data, cpu)); |
69a0b315 | 412 | __rcu_offline_cpu(this_bh_rdp, &rcu_bh_ctrlblk, |
1da177e4 LT |
413 | &per_cpu(rcu_bh_data, cpu)); |
414 | put_cpu_var(rcu_data); | |
415 | put_cpu_var(rcu_bh_data); | |
416 | tasklet_kill_immediate(&per_cpu(rcu_tasklet, cpu), cpu); | |
417 | } | |
418 | ||
419 | #else | |
420 | ||
421 | static void rcu_offline_cpu(int cpu) | |
422 | { | |
423 | } | |
424 | ||
425 | #endif | |
426 | ||
427 | /* | |
428 | * This does the RCU processing work from tasklet context. | |
429 | */ | |
430 | static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp, | |
69a0b315 | 431 | struct rcu_data *rdp) |
1da177e4 LT |
432 | { |
433 | if (rdp->curlist && !rcu_batch_before(rcp->completed, rdp->batch)) { | |
434 | *rdp->donetail = rdp->curlist; | |
435 | rdp->donetail = rdp->curtail; | |
436 | rdp->curlist = NULL; | |
437 | rdp->curtail = &rdp->curlist; | |
438 | } | |
439 | ||
1da177e4 | 440 | if (rdp->nxtlist && !rdp->curlist) { |
caa9ee77 | 441 | local_irq_disable(); |
1da177e4 LT |
442 | rdp->curlist = rdp->nxtlist; |
443 | rdp->curtail = rdp->nxttail; | |
444 | rdp->nxtlist = NULL; | |
445 | rdp->nxttail = &rdp->nxtlist; | |
446 | local_irq_enable(); | |
447 | ||
448 | /* | |
449 | * start the next batch of callbacks | |
450 | */ | |
451 | ||
452 | /* determine batch number */ | |
453 | rdp->batch = rcp->cur + 1; | |
454 | /* see the comment and corresponding wmb() in | |
455 | * the rcu_start_batch() | |
456 | */ | |
457 | smp_rmb(); | |
458 | ||
459 | if (!rcp->next_pending) { | |
460 | /* and start it/schedule start if it's a new batch */ | |
69a0b315 | 461 | spin_lock(&rcp->lock); |
dbc1651f | 462 | rcp->next_pending = 1; |
69a0b315 ON |
463 | rcu_start_batch(rcp); |
464 | spin_unlock(&rcp->lock); | |
1da177e4 | 465 | } |
1da177e4 | 466 | } |
caa9ee77 | 467 | |
69a0b315 | 468 | rcu_check_quiescent_state(rcp, rdp); |
1da177e4 LT |
469 | if (rdp->donelist) |
470 | rcu_do_batch(rdp); | |
471 | } | |
472 | ||
473 | static void rcu_process_callbacks(unsigned long unused) | |
474 | { | |
69a0b315 ON |
475 | __rcu_process_callbacks(&rcu_ctrlblk, &__get_cpu_var(rcu_data)); |
476 | __rcu_process_callbacks(&rcu_bh_ctrlblk, &__get_cpu_var(rcu_bh_data)); | |
1da177e4 LT |
477 | } |
478 | ||
67751777 ON |
479 | static int __rcu_pending(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) |
480 | { | |
481 | /* This cpu has pending rcu entries and the grace period | |
482 | * for them has completed. | |
483 | */ | |
484 | if (rdp->curlist && !rcu_batch_before(rcp->completed, rdp->batch)) | |
485 | return 1; | |
486 | ||
487 | /* This cpu has no pending entries, but there are new entries */ | |
488 | if (!rdp->curlist && rdp->nxtlist) | |
489 | return 1; | |
490 | ||
491 | /* This cpu has finished callbacks to invoke */ | |
492 | if (rdp->donelist) | |
493 | return 1; | |
494 | ||
495 | /* The rcu core waits for a quiescent state from the cpu */ | |
496 | if (rdp->quiescbatch != rcp->cur || rdp->qs_pending) | |
497 | return 1; | |
498 | ||
499 | /* nothing to do */ | |
500 | return 0; | |
501 | } | |
502 | ||
986733e0 HC |
503 | /* |
504 | * Check to see if there is any immediate RCU-related work to be done | |
505 | * by the current CPU, returning 1 if so. This function is part of the | |
506 | * RCU implementation; it is -not- an exported member of the RCU API. | |
507 | */ | |
67751777 ON |
508 | int rcu_pending(int cpu) |
509 | { | |
510 | return __rcu_pending(&rcu_ctrlblk, &per_cpu(rcu_data, cpu)) || | |
511 | __rcu_pending(&rcu_bh_ctrlblk, &per_cpu(rcu_bh_data, cpu)); | |
512 | } | |
513 | ||
986733e0 HC |
514 | /* |
515 | * Check to see if any future RCU-related work will need to be done | |
516 | * by the current CPU, even if none need be done immediately, returning | |
517 | * 1 if so. This function is part of the RCU implementation; it is -not- | |
518 | * an exported member of the RCU API. | |
519 | */ | |
520 | int rcu_needs_cpu(int cpu) | |
521 | { | |
522 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); | |
523 | struct rcu_data *rdp_bh = &per_cpu(rcu_bh_data, cpu); | |
524 | ||
525 | return (!!rdp->curlist || !!rdp_bh->curlist || rcu_pending(cpu)); | |
526 | } | |
527 | ||
1da177e4 LT |
528 | void rcu_check_callbacks(int cpu, int user) |
529 | { | |
530 | if (user || | |
531 | (idle_cpu(cpu) && !in_softirq() && | |
532 | hardirq_count() <= (1 << HARDIRQ_SHIFT))) { | |
533 | rcu_qsctr_inc(cpu); | |
534 | rcu_bh_qsctr_inc(cpu); | |
535 | } else if (!in_softirq()) | |
536 | rcu_bh_qsctr_inc(cpu); | |
537 | tasklet_schedule(&per_cpu(rcu_tasklet, cpu)); | |
538 | } | |
539 | ||
540 | static void rcu_init_percpu_data(int cpu, struct rcu_ctrlblk *rcp, | |
541 | struct rcu_data *rdp) | |
542 | { | |
543 | memset(rdp, 0, sizeof(*rdp)); | |
544 | rdp->curtail = &rdp->curlist; | |
545 | rdp->nxttail = &rdp->nxtlist; | |
546 | rdp->donetail = &rdp->donelist; | |
547 | rdp->quiescbatch = rcp->completed; | |
548 | rdp->qs_pending = 0; | |
549 | rdp->cpu = cpu; | |
21a1ea9e | 550 | rdp->blimit = blimit; |
1da177e4 LT |
551 | } |
552 | ||
553 | static void __devinit rcu_online_cpu(int cpu) | |
554 | { | |
555 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); | |
556 | struct rcu_data *bh_rdp = &per_cpu(rcu_bh_data, cpu); | |
557 | ||
558 | rcu_init_percpu_data(cpu, &rcu_ctrlblk, rdp); | |
559 | rcu_init_percpu_data(cpu, &rcu_bh_ctrlblk, bh_rdp); | |
560 | tasklet_init(&per_cpu(rcu_tasklet, cpu), rcu_process_callbacks, 0UL); | |
561 | } | |
562 | ||
8c78f307 | 563 | static int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
1da177e4 LT |
564 | unsigned long action, void *hcpu) |
565 | { | |
566 | long cpu = (long)hcpu; | |
567 | switch (action) { | |
568 | case CPU_UP_PREPARE: | |
8bb78442 | 569 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 LT |
570 | rcu_online_cpu(cpu); |
571 | break; | |
572 | case CPU_DEAD: | |
8bb78442 | 573 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
574 | rcu_offline_cpu(cpu); |
575 | break; | |
576 | default: | |
577 | break; | |
578 | } | |
579 | return NOTIFY_OK; | |
580 | } | |
581 | ||
8c78f307 | 582 | static struct notifier_block __cpuinitdata rcu_nb = { |
1da177e4 LT |
583 | .notifier_call = rcu_cpu_notify, |
584 | }; | |
585 | ||
586 | /* | |
587 | * Initializes rcu mechanism. Assumed to be called early. | |
588 | * That is before local timer(SMP) or jiffie timer (uniproc) is setup. | |
589 | * Note that rcu_qsctr and friends are implicitly | |
590 | * initialized due to the choice of ``0'' for RCU_CTR_INVALID. | |
591 | */ | |
592 | void __init rcu_init(void) | |
593 | { | |
594 | rcu_cpu_notify(&rcu_nb, CPU_UP_PREPARE, | |
595 | (void *)(long)smp_processor_id()); | |
596 | /* Register notifier for non-boot CPUs */ | |
597 | register_cpu_notifier(&rcu_nb); | |
598 | } | |
599 | ||
600 | struct rcu_synchronize { | |
601 | struct rcu_head head; | |
602 | struct completion completion; | |
603 | }; | |
604 | ||
605 | /* Because of FASTCALL declaration of complete, we use this wrapper */ | |
606 | static void wakeme_after_rcu(struct rcu_head *head) | |
607 | { | |
608 | struct rcu_synchronize *rcu; | |
609 | ||
610 | rcu = container_of(head, struct rcu_synchronize, head); | |
611 | complete(&rcu->completion); | |
612 | } | |
613 | ||
614 | /** | |
9b06e818 | 615 | * synchronize_rcu - wait until a grace period has elapsed. |
1da177e4 LT |
616 | * |
617 | * Control will return to the caller some time after a full grace | |
618 | * period has elapsed, in other words after all currently executing RCU | |
619 | * read-side critical sections have completed. RCU read-side critical | |
620 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), | |
621 | * and may be nested. | |
9b06e818 PM |
622 | * |
623 | * If your read-side code is not protected by rcu_read_lock(), do -not- | |
624 | * use synchronize_rcu(). | |
1da177e4 | 625 | */ |
9b06e818 | 626 | void synchronize_rcu(void) |
1da177e4 LT |
627 | { |
628 | struct rcu_synchronize rcu; | |
629 | ||
630 | init_completion(&rcu.completion); | |
631 | /* Will wake me after RCU finished */ | |
632 | call_rcu(&rcu.head, wakeme_after_rcu); | |
633 | ||
634 | /* Wait for it */ | |
635 | wait_for_completion(&rcu.completion); | |
636 | } | |
637 | ||
21a1ea9e DS |
638 | module_param(blimit, int, 0); |
639 | module_param(qhimark, int, 0); | |
640 | module_param(qlowmark, int, 0); | |
a241ec65 | 641 | EXPORT_SYMBOL_GPL(rcu_batches_completed); |
c32e0660 | 642 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); |
d83015b8 PM |
643 | EXPORT_SYMBOL_GPL(call_rcu); |
644 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
9b06e818 | 645 | EXPORT_SYMBOL_GPL(synchronize_rcu); |