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 LT |
51 | |
52 | /* Definition for rcupdate control block. */ | |
2178426d | 53 | static struct rcu_ctrlblk rcu_ctrlblk = { |
69a0b315 ON |
54 | .cur = -300, |
55 | .completed = -300, | |
e4d91918 | 56 | .lock = __SPIN_LOCK_UNLOCKED(&rcu_ctrlblk.lock), |
69a0b315 ON |
57 | .cpumask = CPU_MASK_NONE, |
58 | }; | |
2178426d | 59 | static struct rcu_ctrlblk rcu_bh_ctrlblk = { |
69a0b315 ON |
60 | .cur = -300, |
61 | .completed = -300, | |
e4d91918 | 62 | .lock = __SPIN_LOCK_UNLOCKED(&rcu_bh_ctrlblk.lock), |
69a0b315 | 63 | .cpumask = CPU_MASK_NONE, |
1da177e4 | 64 | }; |
1da177e4 LT |
65 | |
66 | DEFINE_PER_CPU(struct rcu_data, rcu_data) = { 0L }; | |
67 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data) = { 0L }; | |
68 | ||
69 | /* Fake initialization required by compiler */ | |
70 | static DEFINE_PER_CPU(struct tasklet_struct, rcu_tasklet) = {NULL}; | |
21a1ea9e DS |
71 | static int blimit = 10; |
72 | static int qhimark = 10000; | |
73 | static int qlowmark = 100; | |
74 | #ifdef CONFIG_SMP | |
75 | static int rsinterval = 1000; | |
76 | #endif | |
77 | ||
78 | static atomic_t rcu_barrier_cpu_count; | |
9331b315 | 79 | static DEFINE_MUTEX(rcu_barrier_mutex); |
21a1ea9e DS |
80 | static struct completion rcu_barrier_completion; |
81 | ||
82 | #ifdef CONFIG_SMP | |
83 | static void force_quiescent_state(struct rcu_data *rdp, | |
84 | struct rcu_ctrlblk *rcp) | |
85 | { | |
86 | int cpu; | |
87 | cpumask_t cpumask; | |
88 | set_need_resched(); | |
89 | if (unlikely(rdp->qlen - rdp->last_rs_qlen > rsinterval)) { | |
90 | rdp->last_rs_qlen = rdp->qlen; | |
91 | /* | |
92 | * Don't send IPI to itself. With irqs disabled, | |
93 | * rdp->cpu is the current cpu. | |
94 | */ | |
95 | cpumask = rcp->cpumask; | |
96 | cpu_clear(rdp->cpu, cpumask); | |
97 | for_each_cpu_mask(cpu, cpumask) | |
98 | smp_send_reschedule(cpu); | |
99 | } | |
100 | } | |
101 | #else | |
102 | static inline void force_quiescent_state(struct rcu_data *rdp, | |
103 | struct rcu_ctrlblk *rcp) | |
104 | { | |
105 | set_need_resched(); | |
106 | } | |
107 | #endif | |
1da177e4 LT |
108 | |
109 | /** | |
110 | * call_rcu - Queue an RCU callback for invocation after a grace period. | |
111 | * @head: structure to be used for queueing the RCU updates. | |
112 | * @func: actual update function to be invoked after the grace period | |
113 | * | |
114 | * The update function will be invoked some time after a full grace | |
115 | * period elapses, in other words after all currently executing RCU | |
116 | * read-side critical sections have completed. RCU read-side critical | |
117 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), | |
118 | * and may be nested. | |
119 | */ | |
120 | void fastcall call_rcu(struct rcu_head *head, | |
121 | void (*func)(struct rcu_head *rcu)) | |
122 | { | |
123 | unsigned long flags; | |
124 | struct rcu_data *rdp; | |
125 | ||
126 | head->func = func; | |
127 | head->next = NULL; | |
128 | local_irq_save(flags); | |
129 | rdp = &__get_cpu_var(rcu_data); | |
130 | *rdp->nxttail = head; | |
131 | rdp->nxttail = &head->next; | |
21a1ea9e DS |
132 | if (unlikely(++rdp->qlen > qhimark)) { |
133 | rdp->blimit = INT_MAX; | |
134 | force_quiescent_state(rdp, &rcu_ctrlblk); | |
135 | } | |
1da177e4 LT |
136 | local_irq_restore(flags); |
137 | } | |
138 | ||
139 | /** | |
140 | * call_rcu_bh - Queue an RCU for invocation after a quicker grace period. | |
141 | * @head: structure to be used for queueing the RCU updates. | |
142 | * @func: actual update function to be invoked after the grace period | |
143 | * | |
144 | * The update function will be invoked some time after a full grace | |
145 | * period elapses, in other words after all currently executing RCU | |
146 | * read-side critical sections have completed. call_rcu_bh() assumes | |
147 | * that the read-side critical sections end on completion of a softirq | |
148 | * handler. This means that read-side critical sections in process | |
149 | * context must not be interrupted by softirqs. This interface is to be | |
150 | * used when most of the read-side critical sections are in softirq context. | |
151 | * RCU read-side critical sections are delimited by rcu_read_lock() and | |
152 | * rcu_read_unlock(), * if in interrupt context or rcu_read_lock_bh() | |
153 | * and rcu_read_unlock_bh(), if in process context. These may be nested. | |
154 | */ | |
155 | void fastcall call_rcu_bh(struct rcu_head *head, | |
156 | void (*func)(struct rcu_head *rcu)) | |
157 | { | |
158 | unsigned long flags; | |
159 | struct rcu_data *rdp; | |
160 | ||
161 | head->func = func; | |
162 | head->next = NULL; | |
163 | local_irq_save(flags); | |
164 | rdp = &__get_cpu_var(rcu_bh_data); | |
165 | *rdp->nxttail = head; | |
166 | rdp->nxttail = &head->next; | |
21a1ea9e DS |
167 | |
168 | if (unlikely(++rdp->qlen > qhimark)) { | |
169 | rdp->blimit = INT_MAX; | |
170 | force_quiescent_state(rdp, &rcu_bh_ctrlblk); | |
171 | } | |
172 | ||
1da177e4 LT |
173 | local_irq_restore(flags); |
174 | } | |
175 | ||
a241ec65 PM |
176 | /* |
177 | * Return the number of RCU batches processed thus far. Useful | |
178 | * for debug and statistics. | |
179 | */ | |
180 | long rcu_batches_completed(void) | |
181 | { | |
182 | return rcu_ctrlblk.completed; | |
183 | } | |
184 | ||
c32e0660 PM |
185 | /* |
186 | * Return the number of RCU batches processed thus far. Useful | |
187 | * for debug and statistics. | |
188 | */ | |
189 | long rcu_batches_completed_bh(void) | |
190 | { | |
191 | return rcu_bh_ctrlblk.completed; | |
192 | } | |
193 | ||
ab4720ec DS |
194 | static void rcu_barrier_callback(struct rcu_head *notused) |
195 | { | |
196 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
197 | complete(&rcu_barrier_completion); | |
198 | } | |
199 | ||
200 | /* | |
201 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
202 | */ | |
203 | static void rcu_barrier_func(void *notused) | |
204 | { | |
205 | int cpu = smp_processor_id(); | |
206 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); | |
207 | struct rcu_head *head; | |
208 | ||
209 | head = &rdp->barrier; | |
210 | atomic_inc(&rcu_barrier_cpu_count); | |
211 | call_rcu(head, rcu_barrier_callback); | |
212 | } | |
213 | ||
214 | /** | |
215 | * rcu_barrier - Wait until all the in-flight RCUs are complete. | |
216 | */ | |
217 | void rcu_barrier(void) | |
218 | { | |
219 | BUG_ON(in_interrupt()); | |
9331b315 IM |
220 | /* Take cpucontrol mutex to protect against CPU hotplug */ |
221 | mutex_lock(&rcu_barrier_mutex); | |
ab4720ec DS |
222 | init_completion(&rcu_barrier_completion); |
223 | atomic_set(&rcu_barrier_cpu_count, 0); | |
224 | on_each_cpu(rcu_barrier_func, NULL, 0, 1); | |
225 | wait_for_completion(&rcu_barrier_completion); | |
9331b315 | 226 | mutex_unlock(&rcu_barrier_mutex); |
ab4720ec DS |
227 | } |
228 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
229 | ||
1da177e4 LT |
230 | /* |
231 | * Invoke the completed RCU callbacks. They are expected to be in | |
232 | * a per-cpu list. | |
233 | */ | |
234 | static void rcu_do_batch(struct rcu_data *rdp) | |
235 | { | |
236 | struct rcu_head *next, *list; | |
237 | int count = 0; | |
238 | ||
239 | list = rdp->donelist; | |
240 | while (list) { | |
241 | next = rdp->donelist = list->next; | |
242 | list->func(list); | |
243 | list = next; | |
21a1ea9e | 244 | if (++count >= rdp->blimit) |
1da177e4 LT |
245 | break; |
246 | } | |
dd9daa22 ON |
247 | |
248 | local_irq_disable(); | |
249 | rdp->qlen -= count; | |
250 | local_irq_enable(); | |
21a1ea9e DS |
251 | if (rdp->blimit == INT_MAX && rdp->qlen <= qlowmark) |
252 | rdp->blimit = blimit; | |
dd9daa22 | 253 | |
1da177e4 LT |
254 | if (!rdp->donelist) |
255 | rdp->donetail = &rdp->donelist; | |
256 | else | |
257 | tasklet_schedule(&per_cpu(rcu_tasklet, rdp->cpu)); | |
258 | } | |
259 | ||
260 | /* | |
261 | * Grace period handling: | |
262 | * The grace period handling consists out of two steps: | |
263 | * - A new grace period is started. | |
264 | * This is done by rcu_start_batch. The start is not broadcasted to | |
265 | * all cpus, they must pick this up by comparing rcp->cur with | |
266 | * rdp->quiescbatch. All cpus are recorded in the | |
69a0b315 | 267 | * rcu_ctrlblk.cpumask bitmap. |
1da177e4 LT |
268 | * - All cpus must go through a quiescent state. |
269 | * Since the start of the grace period is not broadcasted, at least two | |
270 | * calls to rcu_check_quiescent_state are required: | |
271 | * The first call just notices that a new grace period is running. The | |
272 | * following calls check if there was a quiescent state since the beginning | |
69a0b315 | 273 | * of the grace period. If so, it updates rcu_ctrlblk.cpumask. If |
1da177e4 LT |
274 | * the bitmap is empty, then the grace period is completed. |
275 | * rcu_check_quiescent_state calls rcu_start_batch(0) to start the next grace | |
276 | * period (if necessary). | |
277 | */ | |
278 | /* | |
279 | * Register a new batch of callbacks, and start it up if there is currently no | |
280 | * active batch and the batch to be registered has not already occurred. | |
69a0b315 | 281 | * Caller must hold rcu_ctrlblk.lock. |
1da177e4 | 282 | */ |
69a0b315 | 283 | static void rcu_start_batch(struct rcu_ctrlblk *rcp) |
1da177e4 | 284 | { |
1da177e4 LT |
285 | if (rcp->next_pending && |
286 | rcp->completed == rcp->cur) { | |
1da177e4 | 287 | rcp->next_pending = 0; |
c3f59023 SV |
288 | /* |
289 | * next_pending == 0 must be visible in | |
290 | * __rcu_process_callbacks() before it can see new value of cur. | |
1da177e4 LT |
291 | */ |
292 | smp_wmb(); | |
293 | rcp->cur++; | |
c3f59023 SV |
294 | |
295 | /* | |
296 | * Accessing nohz_cpu_mask before incrementing rcp->cur needs a | |
297 | * Barrier Otherwise it can cause tickless idle CPUs to be | |
69a0b315 | 298 | * included in rcp->cpumask, which will extend graceperiods |
c3f59023 SV |
299 | * unnecessarily. |
300 | */ | |
301 | smp_mb(); | |
69a0b315 | 302 | cpus_andnot(rcp->cpumask, cpu_online_map, nohz_cpu_mask); |
c3f59023 | 303 | |
1da177e4 LT |
304 | } |
305 | } | |
306 | ||
307 | /* | |
308 | * cpu went through a quiescent state since the beginning of the grace period. | |
309 | * Clear it from the cpu mask and complete the grace period if it was the last | |
310 | * cpu. Start another grace period if someone has further entries pending | |
311 | */ | |
69a0b315 | 312 | static void cpu_quiet(int cpu, struct rcu_ctrlblk *rcp) |
1da177e4 | 313 | { |
69a0b315 ON |
314 | cpu_clear(cpu, rcp->cpumask); |
315 | if (cpus_empty(rcp->cpumask)) { | |
1da177e4 LT |
316 | /* batch completed ! */ |
317 | rcp->completed = rcp->cur; | |
69a0b315 | 318 | rcu_start_batch(rcp); |
1da177e4 LT |
319 | } |
320 | } | |
321 | ||
322 | /* | |
323 | * Check if the cpu has gone through a quiescent state (say context | |
324 | * switch). If so and if it already hasn't done so in this RCU | |
325 | * quiescent cycle, then indicate that it has done so. | |
326 | */ | |
327 | static void rcu_check_quiescent_state(struct rcu_ctrlblk *rcp, | |
69a0b315 | 328 | struct rcu_data *rdp) |
1da177e4 LT |
329 | { |
330 | if (rdp->quiescbatch != rcp->cur) { | |
331 | /* start new grace period: */ | |
332 | rdp->qs_pending = 1; | |
333 | rdp->passed_quiesc = 0; | |
334 | rdp->quiescbatch = rcp->cur; | |
335 | return; | |
336 | } | |
337 | ||
338 | /* Grace period already completed for this cpu? | |
339 | * qs_pending is checked instead of the actual bitmap to avoid | |
340 | * cacheline trashing. | |
341 | */ | |
342 | if (!rdp->qs_pending) | |
343 | return; | |
344 | ||
345 | /* | |
346 | * Was there a quiescent state since the beginning of the grace | |
347 | * period? If no, then exit and wait for the next call. | |
348 | */ | |
349 | if (!rdp->passed_quiesc) | |
350 | return; | |
351 | rdp->qs_pending = 0; | |
352 | ||
69a0b315 | 353 | spin_lock(&rcp->lock); |
1da177e4 LT |
354 | /* |
355 | * rdp->quiescbatch/rcp->cur and the cpu bitmap can come out of sync | |
356 | * during cpu startup. Ignore the quiescent state. | |
357 | */ | |
358 | if (likely(rdp->quiescbatch == rcp->cur)) | |
69a0b315 | 359 | cpu_quiet(rdp->cpu, rcp); |
1da177e4 | 360 | |
69a0b315 | 361 | spin_unlock(&rcp->lock); |
1da177e4 LT |
362 | } |
363 | ||
364 | ||
365 | #ifdef CONFIG_HOTPLUG_CPU | |
366 | ||
367 | /* warning! helper for rcu_offline_cpu. do not use elsewhere without reviewing | |
368 | * locking requirements, the list it's pulling from has to belong to a cpu | |
369 | * which is dead and hence not processing interrupts. | |
370 | */ | |
371 | static void rcu_move_batch(struct rcu_data *this_rdp, struct rcu_head *list, | |
372 | struct rcu_head **tail) | |
373 | { | |
374 | local_irq_disable(); | |
375 | *this_rdp->nxttail = list; | |
376 | if (list) | |
377 | this_rdp->nxttail = tail; | |
378 | local_irq_enable(); | |
379 | } | |
380 | ||
381 | static void __rcu_offline_cpu(struct rcu_data *this_rdp, | |
69a0b315 | 382 | struct rcu_ctrlblk *rcp, struct rcu_data *rdp) |
1da177e4 LT |
383 | { |
384 | /* if the cpu going offline owns the grace period | |
385 | * we can block indefinitely waiting for it, so flush | |
386 | * it here | |
387 | */ | |
69a0b315 | 388 | spin_lock_bh(&rcp->lock); |
1da177e4 | 389 | if (rcp->cur != rcp->completed) |
69a0b315 ON |
390 | cpu_quiet(rdp->cpu, rcp); |
391 | spin_unlock_bh(&rcp->lock); | |
1da177e4 LT |
392 | rcu_move_batch(this_rdp, rdp->curlist, rdp->curtail); |
393 | rcu_move_batch(this_rdp, rdp->nxtlist, rdp->nxttail); | |
a9c82815 | 394 | rcu_move_batch(this_rdp, rdp->donelist, rdp->donetail); |
1da177e4 | 395 | } |
a9c82815 | 396 | |
1da177e4 LT |
397 | static void rcu_offline_cpu(int cpu) |
398 | { | |
399 | struct rcu_data *this_rdp = &get_cpu_var(rcu_data); | |
400 | struct rcu_data *this_bh_rdp = &get_cpu_var(rcu_bh_data); | |
401 | ||
69a0b315 | 402 | __rcu_offline_cpu(this_rdp, &rcu_ctrlblk, |
1da177e4 | 403 | &per_cpu(rcu_data, cpu)); |
69a0b315 | 404 | __rcu_offline_cpu(this_bh_rdp, &rcu_bh_ctrlblk, |
1da177e4 LT |
405 | &per_cpu(rcu_bh_data, cpu)); |
406 | put_cpu_var(rcu_data); | |
407 | put_cpu_var(rcu_bh_data); | |
408 | tasklet_kill_immediate(&per_cpu(rcu_tasklet, cpu), cpu); | |
409 | } | |
410 | ||
411 | #else | |
412 | ||
413 | static void rcu_offline_cpu(int cpu) | |
414 | { | |
415 | } | |
416 | ||
417 | #endif | |
418 | ||
419 | /* | |
420 | * This does the RCU processing work from tasklet context. | |
421 | */ | |
422 | static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp, | |
69a0b315 | 423 | struct rcu_data *rdp) |
1da177e4 LT |
424 | { |
425 | if (rdp->curlist && !rcu_batch_before(rcp->completed, rdp->batch)) { | |
426 | *rdp->donetail = rdp->curlist; | |
427 | rdp->donetail = rdp->curtail; | |
428 | rdp->curlist = NULL; | |
429 | rdp->curtail = &rdp->curlist; | |
430 | } | |
431 | ||
1da177e4 | 432 | if (rdp->nxtlist && !rdp->curlist) { |
caa9ee77 | 433 | local_irq_disable(); |
1da177e4 LT |
434 | rdp->curlist = rdp->nxtlist; |
435 | rdp->curtail = rdp->nxttail; | |
436 | rdp->nxtlist = NULL; | |
437 | rdp->nxttail = &rdp->nxtlist; | |
438 | local_irq_enable(); | |
439 | ||
440 | /* | |
441 | * start the next batch of callbacks | |
442 | */ | |
443 | ||
444 | /* determine batch number */ | |
445 | rdp->batch = rcp->cur + 1; | |
446 | /* see the comment and corresponding wmb() in | |
447 | * the rcu_start_batch() | |
448 | */ | |
449 | smp_rmb(); | |
450 | ||
451 | if (!rcp->next_pending) { | |
452 | /* and start it/schedule start if it's a new batch */ | |
69a0b315 | 453 | spin_lock(&rcp->lock); |
dbc1651f | 454 | rcp->next_pending = 1; |
69a0b315 ON |
455 | rcu_start_batch(rcp); |
456 | spin_unlock(&rcp->lock); | |
1da177e4 | 457 | } |
1da177e4 | 458 | } |
caa9ee77 | 459 | |
69a0b315 | 460 | rcu_check_quiescent_state(rcp, rdp); |
1da177e4 LT |
461 | if (rdp->donelist) |
462 | rcu_do_batch(rdp); | |
463 | } | |
464 | ||
465 | static void rcu_process_callbacks(unsigned long unused) | |
466 | { | |
69a0b315 ON |
467 | __rcu_process_callbacks(&rcu_ctrlblk, &__get_cpu_var(rcu_data)); |
468 | __rcu_process_callbacks(&rcu_bh_ctrlblk, &__get_cpu_var(rcu_bh_data)); | |
1da177e4 LT |
469 | } |
470 | ||
67751777 ON |
471 | static int __rcu_pending(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) |
472 | { | |
473 | /* This cpu has pending rcu entries and the grace period | |
474 | * for them has completed. | |
475 | */ | |
476 | if (rdp->curlist && !rcu_batch_before(rcp->completed, rdp->batch)) | |
477 | return 1; | |
478 | ||
479 | /* This cpu has no pending entries, but there are new entries */ | |
480 | if (!rdp->curlist && rdp->nxtlist) | |
481 | return 1; | |
482 | ||
483 | /* This cpu has finished callbacks to invoke */ | |
484 | if (rdp->donelist) | |
485 | return 1; | |
486 | ||
487 | /* The rcu core waits for a quiescent state from the cpu */ | |
488 | if (rdp->quiescbatch != rcp->cur || rdp->qs_pending) | |
489 | return 1; | |
490 | ||
491 | /* nothing to do */ | |
492 | return 0; | |
493 | } | |
494 | ||
986733e0 HC |
495 | /* |
496 | * Check to see if there is any immediate RCU-related work to be done | |
497 | * by the current CPU, returning 1 if so. This function is part of the | |
498 | * RCU implementation; it is -not- an exported member of the RCU API. | |
499 | */ | |
67751777 ON |
500 | int rcu_pending(int cpu) |
501 | { | |
502 | return __rcu_pending(&rcu_ctrlblk, &per_cpu(rcu_data, cpu)) || | |
503 | __rcu_pending(&rcu_bh_ctrlblk, &per_cpu(rcu_bh_data, cpu)); | |
504 | } | |
505 | ||
986733e0 HC |
506 | /* |
507 | * Check to see if any future RCU-related work will need to be done | |
508 | * by the current CPU, even if none need be done immediately, returning | |
509 | * 1 if so. This function is part of the RCU implementation; it is -not- | |
510 | * an exported member of the RCU API. | |
511 | */ | |
512 | int rcu_needs_cpu(int cpu) | |
513 | { | |
514 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); | |
515 | struct rcu_data *rdp_bh = &per_cpu(rcu_bh_data, cpu); | |
516 | ||
517 | return (!!rdp->curlist || !!rdp_bh->curlist || rcu_pending(cpu)); | |
518 | } | |
519 | ||
1da177e4 LT |
520 | void rcu_check_callbacks(int cpu, int user) |
521 | { | |
522 | if (user || | |
523 | (idle_cpu(cpu) && !in_softirq() && | |
524 | hardirq_count() <= (1 << HARDIRQ_SHIFT))) { | |
525 | rcu_qsctr_inc(cpu); | |
526 | rcu_bh_qsctr_inc(cpu); | |
527 | } else if (!in_softirq()) | |
528 | rcu_bh_qsctr_inc(cpu); | |
529 | tasklet_schedule(&per_cpu(rcu_tasklet, cpu)); | |
530 | } | |
531 | ||
532 | static void rcu_init_percpu_data(int cpu, struct rcu_ctrlblk *rcp, | |
533 | struct rcu_data *rdp) | |
534 | { | |
535 | memset(rdp, 0, sizeof(*rdp)); | |
536 | rdp->curtail = &rdp->curlist; | |
537 | rdp->nxttail = &rdp->nxtlist; | |
538 | rdp->donetail = &rdp->donelist; | |
539 | rdp->quiescbatch = rcp->completed; | |
540 | rdp->qs_pending = 0; | |
541 | rdp->cpu = cpu; | |
21a1ea9e | 542 | rdp->blimit = blimit; |
1da177e4 LT |
543 | } |
544 | ||
545 | static void __devinit rcu_online_cpu(int cpu) | |
546 | { | |
547 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); | |
548 | struct rcu_data *bh_rdp = &per_cpu(rcu_bh_data, cpu); | |
549 | ||
550 | rcu_init_percpu_data(cpu, &rcu_ctrlblk, rdp); | |
551 | rcu_init_percpu_data(cpu, &rcu_bh_ctrlblk, bh_rdp); | |
552 | tasklet_init(&per_cpu(rcu_tasklet, cpu), rcu_process_callbacks, 0UL); | |
553 | } | |
554 | ||
8c78f307 | 555 | static int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
1da177e4 LT |
556 | unsigned long action, void *hcpu) |
557 | { | |
558 | long cpu = (long)hcpu; | |
559 | switch (action) { | |
560 | case CPU_UP_PREPARE: | |
561 | rcu_online_cpu(cpu); | |
562 | break; | |
563 | case CPU_DEAD: | |
564 | rcu_offline_cpu(cpu); | |
565 | break; | |
566 | default: | |
567 | break; | |
568 | } | |
569 | return NOTIFY_OK; | |
570 | } | |
571 | ||
8c78f307 | 572 | static struct notifier_block __cpuinitdata rcu_nb = { |
1da177e4 LT |
573 | .notifier_call = rcu_cpu_notify, |
574 | }; | |
575 | ||
576 | /* | |
577 | * Initializes rcu mechanism. Assumed to be called early. | |
578 | * That is before local timer(SMP) or jiffie timer (uniproc) is setup. | |
579 | * Note that rcu_qsctr and friends are implicitly | |
580 | * initialized due to the choice of ``0'' for RCU_CTR_INVALID. | |
581 | */ | |
582 | void __init rcu_init(void) | |
583 | { | |
584 | rcu_cpu_notify(&rcu_nb, CPU_UP_PREPARE, | |
585 | (void *)(long)smp_processor_id()); | |
586 | /* Register notifier for non-boot CPUs */ | |
587 | register_cpu_notifier(&rcu_nb); | |
588 | } | |
589 | ||
590 | struct rcu_synchronize { | |
591 | struct rcu_head head; | |
592 | struct completion completion; | |
593 | }; | |
594 | ||
595 | /* Because of FASTCALL declaration of complete, we use this wrapper */ | |
596 | static void wakeme_after_rcu(struct rcu_head *head) | |
597 | { | |
598 | struct rcu_synchronize *rcu; | |
599 | ||
600 | rcu = container_of(head, struct rcu_synchronize, head); | |
601 | complete(&rcu->completion); | |
602 | } | |
603 | ||
604 | /** | |
9b06e818 | 605 | * synchronize_rcu - wait until a grace period has elapsed. |
1da177e4 LT |
606 | * |
607 | * Control will return to the caller some time after a full grace | |
608 | * period has elapsed, in other words after all currently executing RCU | |
609 | * read-side critical sections have completed. RCU read-side critical | |
610 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), | |
611 | * and may be nested. | |
9b06e818 PM |
612 | * |
613 | * If your read-side code is not protected by rcu_read_lock(), do -not- | |
614 | * use synchronize_rcu(). | |
1da177e4 | 615 | */ |
9b06e818 | 616 | void synchronize_rcu(void) |
1da177e4 LT |
617 | { |
618 | struct rcu_synchronize rcu; | |
619 | ||
620 | init_completion(&rcu.completion); | |
621 | /* Will wake me after RCU finished */ | |
622 | call_rcu(&rcu.head, wakeme_after_rcu); | |
623 | ||
624 | /* Wait for it */ | |
625 | wait_for_completion(&rcu.completion); | |
626 | } | |
627 | ||
21a1ea9e DS |
628 | module_param(blimit, int, 0); |
629 | module_param(qhimark, int, 0); | |
630 | module_param(qlowmark, int, 0); | |
631 | #ifdef CONFIG_SMP | |
632 | module_param(rsinterval, int, 0); | |
633 | #endif | |
a241ec65 | 634 | EXPORT_SYMBOL_GPL(rcu_batches_completed); |
c32e0660 | 635 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); |
d83015b8 PM |
636 | EXPORT_SYMBOL_GPL(call_rcu); |
637 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
9b06e818 | 638 | EXPORT_SYMBOL_GPL(synchronize_rcu); |