diff options
Diffstat (limited to 'kernel/rcupdate.c')
-rw-r--r-- | kernel/rcupdate.c | 575 |
1 files changed, 25 insertions, 550 deletions
diff --git a/kernel/rcupdate.c b/kernel/rcupdate.c index 4dfa0b792efa..0ccd0095ebdc 100644 --- a/kernel/rcupdate.c +++ b/kernel/rcupdate.c | |||
@@ -15,7 +15,7 @@ | |||
15 | * along with this program; if not, write to the Free Software | 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. | 16 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. |
17 | * | 17 | * |
18 | * Copyright (C) IBM Corporation, 2001 | 18 | * Copyright IBM Corporation, 2001 |
19 | * | 19 | * |
20 | * Authors: Dipankar Sarma <dipankar@in.ibm.com> | 20 | * Authors: Dipankar Sarma <dipankar@in.ibm.com> |
21 | * Manfred Spraul <manfred@colorfullife.com> | 21 | * Manfred Spraul <manfred@colorfullife.com> |
@@ -35,163 +35,57 @@ | |||
35 | #include <linux/init.h> | 35 | #include <linux/init.h> |
36 | #include <linux/spinlock.h> | 36 | #include <linux/spinlock.h> |
37 | #include <linux/smp.h> | 37 | #include <linux/smp.h> |
38 | #include <linux/rcupdate.h> | ||
39 | #include <linux/interrupt.h> | 38 | #include <linux/interrupt.h> |
40 | #include <linux/sched.h> | 39 | #include <linux/sched.h> |
41 | #include <asm/atomic.h> | 40 | #include <asm/atomic.h> |
42 | #include <linux/bitops.h> | 41 | #include <linux/bitops.h> |
43 | #include <linux/module.h> | ||
44 | #include <linux/completion.h> | 42 | #include <linux/completion.h> |
45 | #include <linux/moduleparam.h> | ||
46 | #include <linux/percpu.h> | 43 | #include <linux/percpu.h> |
47 | #include <linux/notifier.h> | 44 | #include <linux/notifier.h> |
48 | #include <linux/cpu.h> | 45 | #include <linux/cpu.h> |
49 | #include <linux/mutex.h> | 46 | #include <linux/mutex.h> |
47 | #include <linux/module.h> | ||
50 | 48 | ||
51 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | 49 | struct rcu_synchronize { |
52 | static struct lock_class_key rcu_lock_key; | 50 | struct rcu_head head; |
53 | struct lockdep_map rcu_lock_map = | 51 | struct completion completion; |
54 | STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key); | ||
55 | |||
56 | EXPORT_SYMBOL_GPL(rcu_lock_map); | ||
57 | #endif | ||
58 | |||
59 | /* Definition for rcupdate control block. */ | ||
60 | static struct rcu_ctrlblk rcu_ctrlblk = { | ||
61 | .cur = -300, | ||
62 | .completed = -300, | ||
63 | .lock = __SPIN_LOCK_UNLOCKED(&rcu_ctrlblk.lock), | ||
64 | .cpumask = CPU_MASK_NONE, | ||
65 | }; | ||
66 | static struct rcu_ctrlblk rcu_bh_ctrlblk = { | ||
67 | .cur = -300, | ||
68 | .completed = -300, | ||
69 | .lock = __SPIN_LOCK_UNLOCKED(&rcu_bh_ctrlblk.lock), | ||
70 | .cpumask = CPU_MASK_NONE, | ||
71 | }; | 52 | }; |
72 | 53 | ||
73 | DEFINE_PER_CPU(struct rcu_data, rcu_data) = { 0L }; | 54 | static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL}; |
74 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data) = { 0L }; | ||
75 | |||
76 | static int blimit = 10; | ||
77 | static int qhimark = 10000; | ||
78 | static int qlowmark = 100; | ||
79 | |||
80 | static atomic_t rcu_barrier_cpu_count; | 55 | static atomic_t rcu_barrier_cpu_count; |
81 | static DEFINE_MUTEX(rcu_barrier_mutex); | 56 | static DEFINE_MUTEX(rcu_barrier_mutex); |
82 | static struct completion rcu_barrier_completion; | 57 | static struct completion rcu_barrier_completion; |
83 | 58 | ||
84 | #ifdef CONFIG_SMP | 59 | /* Because of FASTCALL declaration of complete, we use this wrapper */ |
85 | static void force_quiescent_state(struct rcu_data *rdp, | 60 | static void wakeme_after_rcu(struct rcu_head *head) |
86 | struct rcu_ctrlblk *rcp) | ||
87 | { | ||
88 | int cpu; | ||
89 | cpumask_t cpumask; | ||
90 | set_need_resched(); | ||
91 | if (unlikely(!rcp->signaled)) { | ||
92 | rcp->signaled = 1; | ||
93 | /* | ||
94 | * Don't send IPI to itself. With irqs disabled, | ||
95 | * rdp->cpu is the current cpu. | ||
96 | */ | ||
97 | cpumask = rcp->cpumask; | ||
98 | cpu_clear(rdp->cpu, cpumask); | ||
99 | for_each_cpu_mask(cpu, cpumask) | ||
100 | smp_send_reschedule(cpu); | ||
101 | } | ||
102 | } | ||
103 | #else | ||
104 | static inline void force_quiescent_state(struct rcu_data *rdp, | ||
105 | struct rcu_ctrlblk *rcp) | ||
106 | { | 61 | { |
107 | set_need_resched(); | 62 | struct rcu_synchronize *rcu; |
63 | |||
64 | rcu = container_of(head, struct rcu_synchronize, head); | ||
65 | complete(&rcu->completion); | ||
108 | } | 66 | } |
109 | #endif | ||
110 | 67 | ||
111 | /** | 68 | /** |
112 | * call_rcu - Queue an RCU callback for invocation after a grace period. | 69 | * synchronize_rcu - wait until a grace period has elapsed. |
113 | * @head: structure to be used for queueing the RCU updates. | ||
114 | * @func: actual update function to be invoked after the grace period | ||
115 | * | 70 | * |
116 | * The update function will be invoked some time after a full grace | 71 | * Control will return to the caller some time after a full grace |
117 | * period elapses, in other words after all currently executing RCU | 72 | * period has elapsed, in other words after all currently executing RCU |
118 | * read-side critical sections have completed. RCU read-side critical | 73 | * read-side critical sections have completed. RCU read-side critical |
119 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), | 74 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), |
120 | * and may be nested. | 75 | * and may be nested. |
121 | */ | 76 | */ |
122 | void fastcall call_rcu(struct rcu_head *head, | 77 | void synchronize_rcu(void) |
123 | void (*func)(struct rcu_head *rcu)) | ||
124 | { | ||
125 | unsigned long flags; | ||
126 | struct rcu_data *rdp; | ||
127 | |||
128 | head->func = func; | ||
129 | head->next = NULL; | ||
130 | local_irq_save(flags); | ||
131 | rdp = &__get_cpu_var(rcu_data); | ||
132 | *rdp->nxttail = head; | ||
133 | rdp->nxttail = &head->next; | ||
134 | if (unlikely(++rdp->qlen > qhimark)) { | ||
135 | rdp->blimit = INT_MAX; | ||
136 | force_quiescent_state(rdp, &rcu_ctrlblk); | ||
137 | } | ||
138 | local_irq_restore(flags); | ||
139 | } | ||
140 | |||
141 | /** | ||
142 | * call_rcu_bh - Queue an RCU for invocation after a quicker grace period. | ||
143 | * @head: structure to be used for queueing the RCU updates. | ||
144 | * @func: actual update function to be invoked after the grace period | ||
145 | * | ||
146 | * The update function will be invoked some time after a full grace | ||
147 | * period elapses, in other words after all currently executing RCU | ||
148 | * read-side critical sections have completed. call_rcu_bh() assumes | ||
149 | * that the read-side critical sections end on completion of a softirq | ||
150 | * handler. This means that read-side critical sections in process | ||
151 | * context must not be interrupted by softirqs. This interface is to be | ||
152 | * used when most of the read-side critical sections are in softirq context. | ||
153 | * RCU read-side critical sections are delimited by rcu_read_lock() and | ||
154 | * rcu_read_unlock(), * if in interrupt context or rcu_read_lock_bh() | ||
155 | * and rcu_read_unlock_bh(), if in process context. These may be nested. | ||
156 | */ | ||
157 | void fastcall call_rcu_bh(struct rcu_head *head, | ||
158 | void (*func)(struct rcu_head *rcu)) | ||
159 | { | 78 | { |
160 | unsigned long flags; | 79 | struct rcu_synchronize rcu; |
161 | struct rcu_data *rdp; | ||
162 | |||
163 | head->func = func; | ||
164 | head->next = NULL; | ||
165 | local_irq_save(flags); | ||
166 | rdp = &__get_cpu_var(rcu_bh_data); | ||
167 | *rdp->nxttail = head; | ||
168 | rdp->nxttail = &head->next; | ||
169 | |||
170 | if (unlikely(++rdp->qlen > qhimark)) { | ||
171 | rdp->blimit = INT_MAX; | ||
172 | force_quiescent_state(rdp, &rcu_bh_ctrlblk); | ||
173 | } | ||
174 | |||
175 | local_irq_restore(flags); | ||
176 | } | ||
177 | 80 | ||
178 | /* | 81 | init_completion(&rcu.completion); |
179 | * Return the number of RCU batches processed thus far. Useful | 82 | /* Will wake me after RCU finished */ |
180 | * for debug and statistics. | 83 | call_rcu(&rcu.head, wakeme_after_rcu); |
181 | */ | ||
182 | long rcu_batches_completed(void) | ||
183 | { | ||
184 | return rcu_ctrlblk.completed; | ||
185 | } | ||
186 | 84 | ||
187 | /* | 85 | /* Wait for it */ |
188 | * Return the number of RCU batches processed thus far. Useful | 86 | wait_for_completion(&rcu.completion); |
189 | * for debug and statistics. | ||
190 | */ | ||
191 | long rcu_batches_completed_bh(void) | ||
192 | { | ||
193 | return rcu_bh_ctrlblk.completed; | ||
194 | } | 87 | } |
88 | EXPORT_SYMBOL_GPL(synchronize_rcu); | ||
195 | 89 | ||
196 | static void rcu_barrier_callback(struct rcu_head *notused) | 90 | static void rcu_barrier_callback(struct rcu_head *notused) |
197 | { | 91 | { |
@@ -205,10 +99,8 @@ static void rcu_barrier_callback(struct rcu_head *notused) | |||
205 | static void rcu_barrier_func(void *notused) | 99 | static void rcu_barrier_func(void *notused) |
206 | { | 100 | { |
207 | int cpu = smp_processor_id(); | 101 | int cpu = smp_processor_id(); |
208 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); | 102 | struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu); |
209 | struct rcu_head *head; | ||
210 | 103 | ||
211 | head = &rdp->barrier; | ||
212 | atomic_inc(&rcu_barrier_cpu_count); | 104 | atomic_inc(&rcu_barrier_cpu_count); |
213 | call_rcu(head, rcu_barrier_callback); | 105 | call_rcu(head, rcu_barrier_callback); |
214 | } | 106 | } |
@@ -229,425 +121,8 @@ void rcu_barrier(void) | |||
229 | } | 121 | } |
230 | EXPORT_SYMBOL_GPL(rcu_barrier); | 122 | EXPORT_SYMBOL_GPL(rcu_barrier); |
231 | 123 | ||
232 | /* Raises the softirq for processing rcu_callbacks. */ | ||
233 | static inline void raise_rcu_softirq(void) | ||
234 | { | ||
235 | raise_softirq(RCU_SOFTIRQ); | ||
236 | /* | ||
237 | * The smp_mb() here is required to ensure that this cpu's | ||
238 | * __rcu_process_callbacks() reads the most recently updated | ||
239 | * value of rcu->cur. | ||
240 | */ | ||
241 | smp_mb(); | ||
242 | } | ||
243 | |||
244 | /* | ||
245 | * Invoke the completed RCU callbacks. They are expected to be in | ||
246 | * a per-cpu list. | ||
247 | */ | ||
248 | static void rcu_do_batch(struct rcu_data *rdp) | ||
249 | { | ||
250 | struct rcu_head *next, *list; | ||
251 | int count = 0; | ||
252 | |||
253 | list = rdp->donelist; | ||
254 | while (list) { | ||
255 | next = list->next; | ||
256 | prefetch(next); | ||
257 | list->func(list); | ||
258 | list = next; | ||
259 | if (++count >= rdp->blimit) | ||
260 | break; | ||
261 | } | ||
262 | rdp->donelist = list; | ||
263 | |||
264 | local_irq_disable(); | ||
265 | rdp->qlen -= count; | ||
266 | local_irq_enable(); | ||
267 | if (rdp->blimit == INT_MAX && rdp->qlen <= qlowmark) | ||
268 | rdp->blimit = blimit; | ||
269 | |||
270 | if (!rdp->donelist) | ||
271 | rdp->donetail = &rdp->donelist; | ||
272 | else | ||
273 | raise_rcu_softirq(); | ||
274 | } | ||
275 | |||
276 | /* | ||
277 | * Grace period handling: | ||
278 | * The grace period handling consists out of two steps: | ||
279 | * - A new grace period is started. | ||
280 | * This is done by rcu_start_batch. The start is not broadcasted to | ||
281 | * all cpus, they must pick this up by comparing rcp->cur with | ||
282 | * rdp->quiescbatch. All cpus are recorded in the | ||
283 | * rcu_ctrlblk.cpumask bitmap. | ||
284 | * - All cpus must go through a quiescent state. | ||
285 | * Since the start of the grace period is not broadcasted, at least two | ||
286 | * calls to rcu_check_quiescent_state are required: | ||
287 | * The first call just notices that a new grace period is running. The | ||
288 | * following calls check if there was a quiescent state since the beginning | ||
289 | * of the grace period. If so, it updates rcu_ctrlblk.cpumask. If | ||
290 | * the bitmap is empty, then the grace period is completed. | ||
291 | * rcu_check_quiescent_state calls rcu_start_batch(0) to start the next grace | ||
292 | * period (if necessary). | ||
293 | */ | ||
294 | /* | ||
295 | * Register a new batch of callbacks, and start it up if there is currently no | ||
296 | * active batch and the batch to be registered has not already occurred. | ||
297 | * Caller must hold rcu_ctrlblk.lock. | ||
298 | */ | ||
299 | static void rcu_start_batch(struct rcu_ctrlblk *rcp) | ||
300 | { | ||
301 | if (rcp->next_pending && | ||
302 | rcp->completed == rcp->cur) { | ||
303 | rcp->next_pending = 0; | ||
304 | /* | ||
305 | * next_pending == 0 must be visible in | ||
306 | * __rcu_process_callbacks() before it can see new value of cur. | ||
307 | */ | ||
308 | smp_wmb(); | ||
309 | rcp->cur++; | ||
310 | |||
311 | /* | ||
312 | * Accessing nohz_cpu_mask before incrementing rcp->cur needs a | ||
313 | * Barrier Otherwise it can cause tickless idle CPUs to be | ||
314 | * included in rcp->cpumask, which will extend graceperiods | ||
315 | * unnecessarily. | ||
316 | */ | ||
317 | smp_mb(); | ||
318 | cpus_andnot(rcp->cpumask, cpu_online_map, nohz_cpu_mask); | ||
319 | |||
320 | rcp->signaled = 0; | ||
321 | } | ||
322 | } | ||
323 | |||
324 | /* | ||
325 | * cpu went through a quiescent state since the beginning of the grace period. | ||
326 | * Clear it from the cpu mask and complete the grace period if it was the last | ||
327 | * cpu. Start another grace period if someone has further entries pending | ||
328 | */ | ||
329 | static void cpu_quiet(int cpu, struct rcu_ctrlblk *rcp) | ||
330 | { | ||
331 | cpu_clear(cpu, rcp->cpumask); | ||
332 | if (cpus_empty(rcp->cpumask)) { | ||
333 | /* batch completed ! */ | ||
334 | rcp->completed = rcp->cur; | ||
335 | rcu_start_batch(rcp); | ||
336 | } | ||
337 | } | ||
338 | |||
339 | /* | ||
340 | * Check if the cpu has gone through a quiescent state (say context | ||
341 | * switch). If so and if it already hasn't done so in this RCU | ||
342 | * quiescent cycle, then indicate that it has done so. | ||
343 | */ | ||
344 | static void rcu_check_quiescent_state(struct rcu_ctrlblk *rcp, | ||
345 | struct rcu_data *rdp) | ||
346 | { | ||
347 | if (rdp->quiescbatch != rcp->cur) { | ||
348 | /* start new grace period: */ | ||
349 | rdp->qs_pending = 1; | ||
350 | rdp->passed_quiesc = 0; | ||
351 | rdp->quiescbatch = rcp->cur; | ||
352 | return; | ||
353 | } | ||
354 | |||
355 | /* Grace period already completed for this cpu? | ||
356 | * qs_pending is checked instead of the actual bitmap to avoid | ||
357 | * cacheline trashing. | ||
358 | */ | ||
359 | if (!rdp->qs_pending) | ||
360 | return; | ||
361 | |||
362 | /* | ||
363 | * Was there a quiescent state since the beginning of the grace | ||
364 | * period? If no, then exit and wait for the next call. | ||
365 | */ | ||
366 | if (!rdp->passed_quiesc) | ||
367 | return; | ||
368 | rdp->qs_pending = 0; | ||
369 | |||
370 | spin_lock(&rcp->lock); | ||
371 | /* | ||
372 | * rdp->quiescbatch/rcp->cur and the cpu bitmap can come out of sync | ||
373 | * during cpu startup. Ignore the quiescent state. | ||
374 | */ | ||
375 | if (likely(rdp->quiescbatch == rcp->cur)) | ||
376 | cpu_quiet(rdp->cpu, rcp); | ||
377 | |||
378 | spin_unlock(&rcp->lock); | ||
379 | } | ||
380 | |||
381 | |||
382 | #ifdef CONFIG_HOTPLUG_CPU | ||
383 | |||
384 | /* warning! helper for rcu_offline_cpu. do not use elsewhere without reviewing | ||
385 | * locking requirements, the list it's pulling from has to belong to a cpu | ||
386 | * which is dead and hence not processing interrupts. | ||
387 | */ | ||
388 | static void rcu_move_batch(struct rcu_data *this_rdp, struct rcu_head *list, | ||
389 | struct rcu_head **tail) | ||
390 | { | ||
391 | local_irq_disable(); | ||
392 | *this_rdp->nxttail = list; | ||
393 | if (list) | ||
394 | this_rdp->nxttail = tail; | ||
395 | local_irq_enable(); | ||
396 | } | ||
397 | |||
398 | static void __rcu_offline_cpu(struct rcu_data *this_rdp, | ||
399 | struct rcu_ctrlblk *rcp, struct rcu_data *rdp) | ||
400 | { | ||
401 | /* if the cpu going offline owns the grace period | ||
402 | * we can block indefinitely waiting for it, so flush | ||
403 | * it here | ||
404 | */ | ||
405 | spin_lock_bh(&rcp->lock); | ||
406 | if (rcp->cur != rcp->completed) | ||
407 | cpu_quiet(rdp->cpu, rcp); | ||
408 | spin_unlock_bh(&rcp->lock); | ||
409 | rcu_move_batch(this_rdp, rdp->curlist, rdp->curtail); | ||
410 | rcu_move_batch(this_rdp, rdp->nxtlist, rdp->nxttail); | ||
411 | rcu_move_batch(this_rdp, rdp->donelist, rdp->donetail); | ||
412 | } | ||
413 | |||
414 | static void rcu_offline_cpu(int cpu) | ||
415 | { | ||
416 | struct rcu_data *this_rdp = &get_cpu_var(rcu_data); | ||
417 | struct rcu_data *this_bh_rdp = &get_cpu_var(rcu_bh_data); | ||
418 | |||
419 | __rcu_offline_cpu(this_rdp, &rcu_ctrlblk, | ||
420 | &per_cpu(rcu_data, cpu)); | ||
421 | __rcu_offline_cpu(this_bh_rdp, &rcu_bh_ctrlblk, | ||
422 | &per_cpu(rcu_bh_data, cpu)); | ||
423 | put_cpu_var(rcu_data); | ||
424 | put_cpu_var(rcu_bh_data); | ||
425 | } | ||
426 | |||
427 | #else | ||
428 | |||
429 | static void rcu_offline_cpu(int cpu) | ||
430 | { | ||
431 | } | ||
432 | |||
433 | #endif | ||
434 | |||
435 | /* | ||
436 | * This does the RCU processing work from softirq context. | ||
437 | */ | ||
438 | static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp, | ||
439 | struct rcu_data *rdp) | ||
440 | { | ||
441 | if (rdp->curlist && !rcu_batch_before(rcp->completed, rdp->batch)) { | ||
442 | *rdp->donetail = rdp->curlist; | ||
443 | rdp->donetail = rdp->curtail; | ||
444 | rdp->curlist = NULL; | ||
445 | rdp->curtail = &rdp->curlist; | ||
446 | } | ||
447 | |||
448 | if (rdp->nxtlist && !rdp->curlist) { | ||
449 | local_irq_disable(); | ||
450 | rdp->curlist = rdp->nxtlist; | ||
451 | rdp->curtail = rdp->nxttail; | ||
452 | rdp->nxtlist = NULL; | ||
453 | rdp->nxttail = &rdp->nxtlist; | ||
454 | local_irq_enable(); | ||
455 | |||
456 | /* | ||
457 | * start the next batch of callbacks | ||
458 | */ | ||
459 | |||
460 | /* determine batch number */ | ||
461 | rdp->batch = rcp->cur + 1; | ||
462 | /* see the comment and corresponding wmb() in | ||
463 | * the rcu_start_batch() | ||
464 | */ | ||
465 | smp_rmb(); | ||
466 | |||
467 | if (!rcp->next_pending) { | ||
468 | /* and start it/schedule start if it's a new batch */ | ||
469 | spin_lock(&rcp->lock); | ||
470 | rcp->next_pending = 1; | ||
471 | rcu_start_batch(rcp); | ||
472 | spin_unlock(&rcp->lock); | ||
473 | } | ||
474 | } | ||
475 | |||
476 | rcu_check_quiescent_state(rcp, rdp); | ||
477 | if (rdp->donelist) | ||
478 | rcu_do_batch(rdp); | ||
479 | } | ||
480 | |||
481 | static void rcu_process_callbacks(struct softirq_action *unused) | ||
482 | { | ||
483 | __rcu_process_callbacks(&rcu_ctrlblk, &__get_cpu_var(rcu_data)); | ||
484 | __rcu_process_callbacks(&rcu_bh_ctrlblk, &__get_cpu_var(rcu_bh_data)); | ||
485 | } | ||
486 | |||
487 | static int __rcu_pending(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) | ||
488 | { | ||
489 | /* This cpu has pending rcu entries and the grace period | ||
490 | * for them has completed. | ||
491 | */ | ||
492 | if (rdp->curlist && !rcu_batch_before(rcp->completed, rdp->batch)) | ||
493 | return 1; | ||
494 | |||
495 | /* This cpu has no pending entries, but there are new entries */ | ||
496 | if (!rdp->curlist && rdp->nxtlist) | ||
497 | return 1; | ||
498 | |||
499 | /* This cpu has finished callbacks to invoke */ | ||
500 | if (rdp->donelist) | ||
501 | return 1; | ||
502 | |||
503 | /* The rcu core waits for a quiescent state from the cpu */ | ||
504 | if (rdp->quiescbatch != rcp->cur || rdp->qs_pending) | ||
505 | return 1; | ||
506 | |||
507 | /* nothing to do */ | ||
508 | return 0; | ||
509 | } | ||
510 | |||
511 | /* | ||
512 | * Check to see if there is any immediate RCU-related work to be done | ||
513 | * by the current CPU, returning 1 if so. This function is part of the | ||
514 | * RCU implementation; it is -not- an exported member of the RCU API. | ||
515 | */ | ||
516 | int rcu_pending(int cpu) | ||
517 | { | ||
518 | return __rcu_pending(&rcu_ctrlblk, &per_cpu(rcu_data, cpu)) || | ||
519 | __rcu_pending(&rcu_bh_ctrlblk, &per_cpu(rcu_bh_data, cpu)); | ||
520 | } | ||
521 | |||
522 | /* | ||
523 | * Check to see if any future RCU-related work will need to be done | ||
524 | * by the current CPU, even if none need be done immediately, returning | ||
525 | * 1 if so. This function is part of the RCU implementation; it is -not- | ||
526 | * an exported member of the RCU API. | ||
527 | */ | ||
528 | int rcu_needs_cpu(int cpu) | ||
529 | { | ||
530 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); | ||
531 | struct rcu_data *rdp_bh = &per_cpu(rcu_bh_data, cpu); | ||
532 | |||
533 | return (!!rdp->curlist || !!rdp_bh->curlist || rcu_pending(cpu)); | ||
534 | } | ||
535 | |||
536 | void rcu_check_callbacks(int cpu, int user) | ||
537 | { | ||
538 | if (user || | ||
539 | (idle_cpu(cpu) && !in_softirq() && | ||
540 | hardirq_count() <= (1 << HARDIRQ_SHIFT))) { | ||
541 | rcu_qsctr_inc(cpu); | ||
542 | rcu_bh_qsctr_inc(cpu); | ||
543 | } else if (!in_softirq()) | ||
544 | rcu_bh_qsctr_inc(cpu); | ||
545 | raise_rcu_softirq(); | ||
546 | } | ||
547 | |||
548 | static void rcu_init_percpu_data(int cpu, struct rcu_ctrlblk *rcp, | ||
549 | struct rcu_data *rdp) | ||
550 | { | ||
551 | memset(rdp, 0, sizeof(*rdp)); | ||
552 | rdp->curtail = &rdp->curlist; | ||
553 | rdp->nxttail = &rdp->nxtlist; | ||
554 | rdp->donetail = &rdp->donelist; | ||
555 | rdp->quiescbatch = rcp->completed; | ||
556 | rdp->qs_pending = 0; | ||
557 | rdp->cpu = cpu; | ||
558 | rdp->blimit = blimit; | ||
559 | } | ||
560 | |||
561 | static void __cpuinit rcu_online_cpu(int cpu) | ||
562 | { | ||
563 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); | ||
564 | struct rcu_data *bh_rdp = &per_cpu(rcu_bh_data, cpu); | ||
565 | |||
566 | rcu_init_percpu_data(cpu, &rcu_ctrlblk, rdp); | ||
567 | rcu_init_percpu_data(cpu, &rcu_bh_ctrlblk, bh_rdp); | ||
568 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks, NULL); | ||
569 | } | ||
570 | |||
571 | static int __cpuinit rcu_cpu_notify(struct notifier_block *self, | ||
572 | unsigned long action, void *hcpu) | ||
573 | { | ||
574 | long cpu = (long)hcpu; | ||
575 | switch (action) { | ||
576 | case CPU_UP_PREPARE: | ||
577 | case CPU_UP_PREPARE_FROZEN: | ||
578 | rcu_online_cpu(cpu); | ||
579 | break; | ||
580 | case CPU_DEAD: | ||
581 | case CPU_DEAD_FROZEN: | ||
582 | rcu_offline_cpu(cpu); | ||
583 | break; | ||
584 | default: | ||
585 | break; | ||
586 | } | ||
587 | return NOTIFY_OK; | ||
588 | } | ||
589 | |||
590 | static struct notifier_block __cpuinitdata rcu_nb = { | ||
591 | .notifier_call = rcu_cpu_notify, | ||
592 | }; | ||
593 | |||
594 | /* | ||
595 | * Initializes rcu mechanism. Assumed to be called early. | ||
596 | * That is before local timer(SMP) or jiffie timer (uniproc) is setup. | ||
597 | * Note that rcu_qsctr and friends are implicitly | ||
598 | * initialized due to the choice of ``0'' for RCU_CTR_INVALID. | ||
599 | */ | ||
600 | void __init rcu_init(void) | 124 | void __init rcu_init(void) |
601 | { | 125 | { |
602 | rcu_cpu_notify(&rcu_nb, CPU_UP_PREPARE, | 126 | __rcu_init(); |
603 | (void *)(long)smp_processor_id()); | ||
604 | /* Register notifier for non-boot CPUs */ | ||
605 | register_cpu_notifier(&rcu_nb); | ||
606 | } | ||
607 | |||
608 | struct rcu_synchronize { | ||
609 | struct rcu_head head; | ||
610 | struct completion completion; | ||
611 | }; | ||
612 | |||
613 | /* Because of FASTCALL declaration of complete, we use this wrapper */ | ||
614 | static void wakeme_after_rcu(struct rcu_head *head) | ||
615 | { | ||
616 | struct rcu_synchronize *rcu; | ||
617 | |||
618 | rcu = container_of(head, struct rcu_synchronize, head); | ||
619 | complete(&rcu->completion); | ||
620 | } | ||
621 | |||
622 | /** | ||
623 | * synchronize_rcu - wait until a grace period has elapsed. | ||
624 | * | ||
625 | * Control will return to the caller some time after a full grace | ||
626 | * period has elapsed, in other words after all currently executing RCU | ||
627 | * read-side critical sections have completed. RCU read-side critical | ||
628 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), | ||
629 | * and may be nested. | ||
630 | * | ||
631 | * If your read-side code is not protected by rcu_read_lock(), do -not- | ||
632 | * use synchronize_rcu(). | ||
633 | */ | ||
634 | void synchronize_rcu(void) | ||
635 | { | ||
636 | struct rcu_synchronize rcu; | ||
637 | |||
638 | init_completion(&rcu.completion); | ||
639 | /* Will wake me after RCU finished */ | ||
640 | call_rcu(&rcu.head, wakeme_after_rcu); | ||
641 | |||
642 | /* Wait for it */ | ||
643 | wait_for_completion(&rcu.completion); | ||
644 | } | 127 | } |
645 | 128 | ||
646 | module_param(blimit, int, 0); | ||
647 | module_param(qhimark, int, 0); | ||
648 | module_param(qlowmark, int, 0); | ||
649 | EXPORT_SYMBOL_GPL(rcu_batches_completed); | ||
650 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); | ||
651 | EXPORT_SYMBOL_GPL(call_rcu); | ||
652 | EXPORT_SYMBOL_GPL(call_rcu_bh); | ||
653 | EXPORT_SYMBOL_GPL(synchronize_rcu); | ||