]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Read-Copy Update mechanism for mutual exclusion (tree-based version) | |
3 | * Internal non-public definitions that provide either classic | |
4 | * or preemptable semantics. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License as published by | |
8 | * the Free Software Foundation; either version 2 of the License, or | |
9 | * (at your option) any later version. | |
10 | * | |
11 | * This program is distributed in the hope that it will be useful, | |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
14 | * GNU General Public License for more details. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * along with this program; if not, write to the Free Software | |
18 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
19 | * | |
20 | * Copyright Red Hat, 2009 | |
21 | * Copyright IBM Corporation, 2009 | |
22 | * | |
23 | * Author: Ingo Molnar <mingo@elte.hu> | |
24 | * Paul E. McKenney <paulmck@linux.vnet.ibm.com> | |
25 | */ | |
26 | ||
27 | #include <linux/delay.h> | |
28 | ||
29 | #ifdef CONFIG_TREE_PREEMPT_RCU | |
30 | ||
31 | struct rcu_state rcu_preempt_state = RCU_STATE_INITIALIZER(rcu_preempt_state); | |
32 | DEFINE_PER_CPU(struct rcu_data, rcu_preempt_data); | |
33 | ||
34 | static int rcu_preempted_readers_exp(struct rcu_node *rnp); | |
35 | ||
36 | /* | |
37 | * Tell them what RCU they are running. | |
38 | */ | |
39 | static void __init rcu_bootup_announce(void) | |
40 | { | |
41 | printk(KERN_INFO | |
42 | "Experimental preemptable hierarchical RCU implementation.\n"); | |
43 | } | |
44 | ||
45 | /* | |
46 | * Return the number of RCU-preempt batches processed thus far | |
47 | * for debug and statistics. | |
48 | */ | |
49 | long rcu_batches_completed_preempt(void) | |
50 | { | |
51 | return rcu_preempt_state.completed; | |
52 | } | |
53 | EXPORT_SYMBOL_GPL(rcu_batches_completed_preempt); | |
54 | ||
55 | /* | |
56 | * Return the number of RCU batches processed thus far for debug & stats. | |
57 | */ | |
58 | long rcu_batches_completed(void) | |
59 | { | |
60 | return rcu_batches_completed_preempt(); | |
61 | } | |
62 | EXPORT_SYMBOL_GPL(rcu_batches_completed); | |
63 | ||
64 | /* | |
65 | * Force a quiescent state for preemptible RCU. | |
66 | */ | |
67 | void rcu_force_quiescent_state(void) | |
68 | { | |
69 | force_quiescent_state(&rcu_preempt_state, 0); | |
70 | } | |
71 | EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); | |
72 | ||
73 | /* | |
74 | * Record a preemptable-RCU quiescent state for the specified CPU. Note | |
75 | * that this just means that the task currently running on the CPU is | |
76 | * not in a quiescent state. There might be any number of tasks blocked | |
77 | * while in an RCU read-side critical section. | |
78 | */ | |
79 | static void rcu_preempt_qs(int cpu) | |
80 | { | |
81 | struct rcu_data *rdp = &per_cpu(rcu_preempt_data, cpu); | |
82 | rdp->passed_quiesc_completed = rdp->gpnum - 1; | |
83 | barrier(); | |
84 | rdp->passed_quiesc = 1; | |
85 | } | |
86 | ||
87 | /* | |
88 | * We have entered the scheduler, and the current task might soon be | |
89 | * context-switched away from. If this task is in an RCU read-side | |
90 | * critical section, we will no longer be able to rely on the CPU to | |
91 | * record that fact, so we enqueue the task on the appropriate entry | |
92 | * of the blocked_tasks[] array. The task will dequeue itself when | |
93 | * it exits the outermost enclosing RCU read-side critical section. | |
94 | * Therefore, the current grace period cannot be permitted to complete | |
95 | * until the blocked_tasks[] entry indexed by the low-order bit of | |
96 | * rnp->gpnum empties. | |
97 | * | |
98 | * Caller must disable preemption. | |
99 | */ | |
100 | static void rcu_preempt_note_context_switch(int cpu) | |
101 | { | |
102 | struct task_struct *t = current; | |
103 | unsigned long flags; | |
104 | int phase; | |
105 | struct rcu_data *rdp; | |
106 | struct rcu_node *rnp; | |
107 | ||
108 | if (t->rcu_read_lock_nesting && | |
109 | (t->rcu_read_unlock_special & RCU_READ_UNLOCK_BLOCKED) == 0) { | |
110 | ||
111 | /* Possibly blocking in an RCU read-side critical section. */ | |
112 | rdp = rcu_preempt_state.rda[cpu]; | |
113 | rnp = rdp->mynode; | |
114 | spin_lock_irqsave(&rnp->lock, flags); | |
115 | t->rcu_read_unlock_special |= RCU_READ_UNLOCK_BLOCKED; | |
116 | t->rcu_blocked_node = rnp; | |
117 | ||
118 | /* | |
119 | * If this CPU has already checked in, then this task | |
120 | * will hold up the next grace period rather than the | |
121 | * current grace period. Queue the task accordingly. | |
122 | * If the task is queued for the current grace period | |
123 | * (i.e., this CPU has not yet passed through a quiescent | |
124 | * state for the current grace period), then as long | |
125 | * as that task remains queued, the current grace period | |
126 | * cannot end. | |
127 | * | |
128 | * But first, note that the current CPU must still be | |
129 | * on line! | |
130 | */ | |
131 | WARN_ON_ONCE((rdp->grpmask & rnp->qsmaskinit) == 0); | |
132 | WARN_ON_ONCE(!list_empty(&t->rcu_node_entry)); | |
133 | phase = (rnp->gpnum + !(rnp->qsmask & rdp->grpmask)) & 0x1; | |
134 | list_add(&t->rcu_node_entry, &rnp->blocked_tasks[phase]); | |
135 | spin_unlock_irqrestore(&rnp->lock, flags); | |
136 | } | |
137 | ||
138 | /* | |
139 | * Either we were not in an RCU read-side critical section to | |
140 | * begin with, or we have now recorded that critical section | |
141 | * globally. Either way, we can now note a quiescent state | |
142 | * for this CPU. Again, if we were in an RCU read-side critical | |
143 | * section, and if that critical section was blocking the current | |
144 | * grace period, then the fact that the task has been enqueued | |
145 | * means that we continue to block the current grace period. | |
146 | */ | |
147 | rcu_preempt_qs(cpu); | |
148 | local_irq_save(flags); | |
149 | t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS; | |
150 | local_irq_restore(flags); | |
151 | } | |
152 | ||
153 | /* | |
154 | * Tree-preemptable RCU implementation for rcu_read_lock(). | |
155 | * Just increment ->rcu_read_lock_nesting, shared state will be updated | |
156 | * if we block. | |
157 | */ | |
158 | void __rcu_read_lock(void) | |
159 | { | |
160 | ACCESS_ONCE(current->rcu_read_lock_nesting)++; | |
161 | barrier(); /* needed if we ever invoke rcu_read_lock in rcutree.c */ | |
162 | } | |
163 | EXPORT_SYMBOL_GPL(__rcu_read_lock); | |
164 | ||
165 | /* | |
166 | * Check for preempted RCU readers blocking the current grace period | |
167 | * for the specified rcu_node structure. If the caller needs a reliable | |
168 | * answer, it must hold the rcu_node's ->lock. | |
169 | */ | |
170 | static int rcu_preempted_readers(struct rcu_node *rnp) | |
171 | { | |
172 | int phase = rnp->gpnum & 0x1; | |
173 | ||
174 | return !list_empty(&rnp->blocked_tasks[phase]) || | |
175 | !list_empty(&rnp->blocked_tasks[phase + 2]); | |
176 | } | |
177 | ||
178 | /* | |
179 | * Record a quiescent state for all tasks that were previously queued | |
180 | * on the specified rcu_node structure and that were blocking the current | |
181 | * RCU grace period. The caller must hold the specified rnp->lock with | |
182 | * irqs disabled, and this lock is released upon return, but irqs remain | |
183 | * disabled. | |
184 | */ | |
185 | static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags) | |
186 | __releases(rnp->lock) | |
187 | { | |
188 | unsigned long mask; | |
189 | struct rcu_node *rnp_p; | |
190 | ||
191 | if (rnp->qsmask != 0 || rcu_preempted_readers(rnp)) { | |
192 | spin_unlock_irqrestore(&rnp->lock, flags); | |
193 | return; /* Still need more quiescent states! */ | |
194 | } | |
195 | ||
196 | rnp_p = rnp->parent; | |
197 | if (rnp_p == NULL) { | |
198 | /* | |
199 | * Either there is only one rcu_node in the tree, | |
200 | * or tasks were kicked up to root rcu_node due to | |
201 | * CPUs going offline. | |
202 | */ | |
203 | rcu_report_qs_rsp(&rcu_preempt_state, flags); | |
204 | return; | |
205 | } | |
206 | ||
207 | /* Report up the rest of the hierarchy. */ | |
208 | mask = rnp->grpmask; | |
209 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ | |
210 | spin_lock(&rnp_p->lock); /* irqs already disabled. */ | |
211 | rcu_report_qs_rnp(mask, &rcu_preempt_state, rnp_p, flags); | |
212 | } | |
213 | ||
214 | /* | |
215 | * Handle special cases during rcu_read_unlock(), such as needing to | |
216 | * notify RCU core processing or task having blocked during the RCU | |
217 | * read-side critical section. | |
218 | */ | |
219 | static void rcu_read_unlock_special(struct task_struct *t) | |
220 | { | |
221 | int empty; | |
222 | int empty_exp; | |
223 | unsigned long flags; | |
224 | struct rcu_node *rnp; | |
225 | int special; | |
226 | ||
227 | /* NMI handlers cannot block and cannot safely manipulate state. */ | |
228 | if (in_nmi()) | |
229 | return; | |
230 | ||
231 | local_irq_save(flags); | |
232 | ||
233 | /* | |
234 | * If RCU core is waiting for this CPU to exit critical section, | |
235 | * let it know that we have done so. | |
236 | */ | |
237 | special = t->rcu_read_unlock_special; | |
238 | if (special & RCU_READ_UNLOCK_NEED_QS) { | |
239 | t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS; | |
240 | rcu_preempt_qs(smp_processor_id()); | |
241 | } | |
242 | ||
243 | /* Hardware IRQ handlers cannot block. */ | |
244 | if (in_irq()) { | |
245 | local_irq_restore(flags); | |
246 | return; | |
247 | } | |
248 | ||
249 | /* Clean up if blocked during RCU read-side critical section. */ | |
250 | if (special & RCU_READ_UNLOCK_BLOCKED) { | |
251 | t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_BLOCKED; | |
252 | ||
253 | /* | |
254 | * Remove this task from the list it blocked on. The | |
255 | * task can migrate while we acquire the lock, but at | |
256 | * most one time. So at most two passes through loop. | |
257 | */ | |
258 | for (;;) { | |
259 | rnp = t->rcu_blocked_node; | |
260 | spin_lock(&rnp->lock); /* irqs already disabled. */ | |
261 | if (rnp == t->rcu_blocked_node) | |
262 | break; | |
263 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ | |
264 | } | |
265 | empty = !rcu_preempted_readers(rnp); | |
266 | empty_exp = !rcu_preempted_readers_exp(rnp); | |
267 | smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */ | |
268 | list_del_init(&t->rcu_node_entry); | |
269 | t->rcu_blocked_node = NULL; | |
270 | ||
271 | /* | |
272 | * If this was the last task on the current list, and if | |
273 | * we aren't waiting on any CPUs, report the quiescent state. | |
274 | * Note that rcu_report_unblock_qs_rnp() releases rnp->lock. | |
275 | */ | |
276 | if (empty) | |
277 | spin_unlock_irqrestore(&rnp->lock, flags); | |
278 | else | |
279 | rcu_report_unblock_qs_rnp(rnp, flags); | |
280 | ||
281 | /* | |
282 | * If this was the last task on the expedited lists, | |
283 | * then we need to report up the rcu_node hierarchy. | |
284 | */ | |
285 | if (!empty_exp && !rcu_preempted_readers_exp(rnp)) | |
286 | rcu_report_exp_rnp(&rcu_preempt_state, rnp); | |
287 | } else { | |
288 | local_irq_restore(flags); | |
289 | } | |
290 | } | |
291 | ||
292 | /* | |
293 | * Tree-preemptable RCU implementation for rcu_read_unlock(). | |
294 | * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost | |
295 | * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then | |
296 | * invoke rcu_read_unlock_special() to clean up after a context switch | |
297 | * in an RCU read-side critical section and other special cases. | |
298 | */ | |
299 | void __rcu_read_unlock(void) | |
300 | { | |
301 | struct task_struct *t = current; | |
302 | ||
303 | barrier(); /* needed if we ever invoke rcu_read_unlock in rcutree.c */ | |
304 | if (--ACCESS_ONCE(t->rcu_read_lock_nesting) == 0 && | |
305 | unlikely(ACCESS_ONCE(t->rcu_read_unlock_special))) | |
306 | rcu_read_unlock_special(t); | |
307 | } | |
308 | EXPORT_SYMBOL_GPL(__rcu_read_unlock); | |
309 | ||
310 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR | |
311 | ||
312 | /* | |
313 | * Scan the current list of tasks blocked within RCU read-side critical | |
314 | * sections, printing out the tid of each. | |
315 | */ | |
316 | static void rcu_print_task_stall(struct rcu_node *rnp) | |
317 | { | |
318 | unsigned long flags; | |
319 | struct list_head *lp; | |
320 | int phase; | |
321 | struct task_struct *t; | |
322 | ||
323 | if (rcu_preempted_readers(rnp)) { | |
324 | spin_lock_irqsave(&rnp->lock, flags); | |
325 | phase = rnp->gpnum & 0x1; | |
326 | lp = &rnp->blocked_tasks[phase]; | |
327 | list_for_each_entry(t, lp, rcu_node_entry) | |
328 | printk(" P%d", t->pid); | |
329 | spin_unlock_irqrestore(&rnp->lock, flags); | |
330 | } | |
331 | } | |
332 | ||
333 | #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | |
334 | ||
335 | /* | |
336 | * Check that the list of blocked tasks for the newly completed grace | |
337 | * period is in fact empty. It is a serious bug to complete a grace | |
338 | * period that still has RCU readers blocked! This function must be | |
339 | * invoked -before- updating this rnp's ->gpnum, and the rnp's ->lock | |
340 | * must be held by the caller. | |
341 | */ | |
342 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) | |
343 | { | |
344 | WARN_ON_ONCE(rcu_preempted_readers(rnp)); | |
345 | WARN_ON_ONCE(rnp->qsmask); | |
346 | } | |
347 | ||
348 | #ifdef CONFIG_HOTPLUG_CPU | |
349 | ||
350 | /* | |
351 | * Handle tasklist migration for case in which all CPUs covered by the | |
352 | * specified rcu_node have gone offline. Move them up to the root | |
353 | * rcu_node. The reason for not just moving them to the immediate | |
354 | * parent is to remove the need for rcu_read_unlock_special() to | |
355 | * make more than two attempts to acquire the target rcu_node's lock. | |
356 | * Returns true if there were tasks blocking the current RCU grace | |
357 | * period. | |
358 | * | |
359 | * Returns 1 if there was previously a task blocking the current grace | |
360 | * period on the specified rcu_node structure. | |
361 | * | |
362 | * The caller must hold rnp->lock with irqs disabled. | |
363 | */ | |
364 | static int rcu_preempt_offline_tasks(struct rcu_state *rsp, | |
365 | struct rcu_node *rnp, | |
366 | struct rcu_data *rdp) | |
367 | { | |
368 | int i; | |
369 | struct list_head *lp; | |
370 | struct list_head *lp_root; | |
371 | int retval = 0; | |
372 | struct rcu_node *rnp_root = rcu_get_root(rsp); | |
373 | struct task_struct *tp; | |
374 | ||
375 | if (rnp == rnp_root) { | |
376 | WARN_ONCE(1, "Last CPU thought to be offlined?"); | |
377 | return 0; /* Shouldn't happen: at least one CPU online. */ | |
378 | } | |
379 | WARN_ON_ONCE(rnp != rdp->mynode && | |
380 | (!list_empty(&rnp->blocked_tasks[0]) || | |
381 | !list_empty(&rnp->blocked_tasks[1]) || | |
382 | !list_empty(&rnp->blocked_tasks[2]) || | |
383 | !list_empty(&rnp->blocked_tasks[3]))); | |
384 | ||
385 | /* | |
386 | * Move tasks up to root rcu_node. Rely on the fact that the | |
387 | * root rcu_node can be at most one ahead of the rest of the | |
388 | * rcu_nodes in terms of gp_num value. This fact allows us to | |
389 | * move the blocked_tasks[] array directly, element by element. | |
390 | */ | |
391 | if (rcu_preempted_readers(rnp)) | |
392 | retval |= RCU_OFL_TASKS_NORM_GP; | |
393 | if (rcu_preempted_readers_exp(rnp)) | |
394 | retval |= RCU_OFL_TASKS_EXP_GP; | |
395 | for (i = 0; i < 4; i++) { | |
396 | lp = &rnp->blocked_tasks[i]; | |
397 | lp_root = &rnp_root->blocked_tasks[i]; | |
398 | while (!list_empty(lp)) { | |
399 | tp = list_entry(lp->next, typeof(*tp), rcu_node_entry); | |
400 | spin_lock(&rnp_root->lock); /* irqs already disabled */ | |
401 | list_del(&tp->rcu_node_entry); | |
402 | tp->rcu_blocked_node = rnp_root; | |
403 | list_add(&tp->rcu_node_entry, lp_root); | |
404 | spin_unlock(&rnp_root->lock); /* irqs remain disabled */ | |
405 | } | |
406 | } | |
407 | return retval; | |
408 | } | |
409 | ||
410 | /* | |
411 | * Do CPU-offline processing for preemptable RCU. | |
412 | */ | |
413 | static void rcu_preempt_offline_cpu(int cpu) | |
414 | { | |
415 | __rcu_offline_cpu(cpu, &rcu_preempt_state); | |
416 | } | |
417 | ||
418 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | |
419 | ||
420 | /* | |
421 | * Check for a quiescent state from the current CPU. When a task blocks, | |
422 | * the task is recorded in the corresponding CPU's rcu_node structure, | |
423 | * which is checked elsewhere. | |
424 | * | |
425 | * Caller must disable hard irqs. | |
426 | */ | |
427 | static void rcu_preempt_check_callbacks(int cpu) | |
428 | { | |
429 | struct task_struct *t = current; | |
430 | ||
431 | if (t->rcu_read_lock_nesting == 0) { | |
432 | t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS; | |
433 | rcu_preempt_qs(cpu); | |
434 | return; | |
435 | } | |
436 | if (per_cpu(rcu_preempt_data, cpu).qs_pending) | |
437 | t->rcu_read_unlock_special |= RCU_READ_UNLOCK_NEED_QS; | |
438 | } | |
439 | ||
440 | /* | |
441 | * Process callbacks for preemptable RCU. | |
442 | */ | |
443 | static void rcu_preempt_process_callbacks(void) | |
444 | { | |
445 | __rcu_process_callbacks(&rcu_preempt_state, | |
446 | &__get_cpu_var(rcu_preempt_data)); | |
447 | } | |
448 | ||
449 | /* | |
450 | * Queue a preemptable-RCU callback for invocation after a grace period. | |
451 | */ | |
452 | void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
453 | { | |
454 | __call_rcu(head, func, &rcu_preempt_state); | |
455 | } | |
456 | EXPORT_SYMBOL_GPL(call_rcu); | |
457 | ||
458 | /** | |
459 | * synchronize_rcu - wait until a grace period has elapsed. | |
460 | * | |
461 | * Control will return to the caller some time after a full grace | |
462 | * period has elapsed, in other words after all currently executing RCU | |
463 | * read-side critical sections have completed. RCU read-side critical | |
464 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), | |
465 | * and may be nested. | |
466 | */ | |
467 | void synchronize_rcu(void) | |
468 | { | |
469 | struct rcu_synchronize rcu; | |
470 | ||
471 | if (!rcu_scheduler_active) | |
472 | return; | |
473 | ||
474 | init_completion(&rcu.completion); | |
475 | /* Will wake me after RCU finished. */ | |
476 | call_rcu(&rcu.head, wakeme_after_rcu); | |
477 | /* Wait for it. */ | |
478 | wait_for_completion(&rcu.completion); | |
479 | } | |
480 | EXPORT_SYMBOL_GPL(synchronize_rcu); | |
481 | ||
482 | static DECLARE_WAIT_QUEUE_HEAD(sync_rcu_preempt_exp_wq); | |
483 | static long sync_rcu_preempt_exp_count; | |
484 | static DEFINE_MUTEX(sync_rcu_preempt_exp_mutex); | |
485 | ||
486 | /* | |
487 | * Return non-zero if there are any tasks in RCU read-side critical | |
488 | * sections blocking the current preemptible-RCU expedited grace period. | |
489 | * If there is no preemptible-RCU expedited grace period currently in | |
490 | * progress, returns zero unconditionally. | |
491 | */ | |
492 | static int rcu_preempted_readers_exp(struct rcu_node *rnp) | |
493 | { | |
494 | return !list_empty(&rnp->blocked_tasks[2]) || | |
495 | !list_empty(&rnp->blocked_tasks[3]); | |
496 | } | |
497 | ||
498 | /* | |
499 | * return non-zero if there is no RCU expedited grace period in progress | |
500 | * for the specified rcu_node structure, in other words, if all CPUs and | |
501 | * tasks covered by the specified rcu_node structure have done their bit | |
502 | * for the current expedited grace period. Works only for preemptible | |
503 | * RCU -- other RCU implementation use other means. | |
504 | * | |
505 | * Caller must hold sync_rcu_preempt_exp_mutex. | |
506 | */ | |
507 | static int sync_rcu_preempt_exp_done(struct rcu_node *rnp) | |
508 | { | |
509 | return !rcu_preempted_readers_exp(rnp) && | |
510 | ACCESS_ONCE(rnp->expmask) == 0; | |
511 | } | |
512 | ||
513 | /* | |
514 | * Report the exit from RCU read-side critical section for the last task | |
515 | * that queued itself during or before the current expedited preemptible-RCU | |
516 | * grace period. This event is reported either to the rcu_node structure on | |
517 | * which the task was queued or to one of that rcu_node structure's ancestors, | |
518 | * recursively up the tree. (Calm down, calm down, we do the recursion | |
519 | * iteratively!) | |
520 | * | |
521 | * Caller must hold sync_rcu_preempt_exp_mutex. | |
522 | */ | |
523 | static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp) | |
524 | { | |
525 | unsigned long flags; | |
526 | unsigned long mask; | |
527 | ||
528 | spin_lock_irqsave(&rnp->lock, flags); | |
529 | for (;;) { | |
530 | if (!sync_rcu_preempt_exp_done(rnp)) | |
531 | break; | |
532 | if (rnp->parent == NULL) { | |
533 | wake_up(&sync_rcu_preempt_exp_wq); | |
534 | break; | |
535 | } | |
536 | mask = rnp->grpmask; | |
537 | spin_unlock(&rnp->lock); /* irqs remain disabled */ | |
538 | rnp = rnp->parent; | |
539 | spin_lock(&rnp->lock); /* irqs already disabled */ | |
540 | rnp->expmask &= ~mask; | |
541 | } | |
542 | spin_unlock_irqrestore(&rnp->lock, flags); | |
543 | } | |
544 | ||
545 | /* | |
546 | * Snapshot the tasks blocking the newly started preemptible-RCU expedited | |
547 | * grace period for the specified rcu_node structure. If there are no such | |
548 | * tasks, report it up the rcu_node hierarchy. | |
549 | * | |
550 | * Caller must hold sync_rcu_preempt_exp_mutex and rsp->onofflock. | |
551 | */ | |
552 | static void | |
553 | sync_rcu_preempt_exp_init(struct rcu_state *rsp, struct rcu_node *rnp) | |
554 | { | |
555 | int must_wait; | |
556 | ||
557 | spin_lock(&rnp->lock); /* irqs already disabled */ | |
558 | list_splice_init(&rnp->blocked_tasks[0], &rnp->blocked_tasks[2]); | |
559 | list_splice_init(&rnp->blocked_tasks[1], &rnp->blocked_tasks[3]); | |
560 | must_wait = rcu_preempted_readers_exp(rnp); | |
561 | spin_unlock(&rnp->lock); /* irqs remain disabled */ | |
562 | if (!must_wait) | |
563 | rcu_report_exp_rnp(rsp, rnp); | |
564 | } | |
565 | ||
566 | /* | |
567 | * Wait for an rcu-preempt grace period, but expedite it. The basic idea | |
568 | * is to invoke synchronize_sched_expedited() to push all the tasks to | |
569 | * the ->blocked_tasks[] lists, move all entries from the first set of | |
570 | * ->blocked_tasks[] lists to the second set, and finally wait for this | |
571 | * second set to drain. | |
572 | */ | |
573 | void synchronize_rcu_expedited(void) | |
574 | { | |
575 | unsigned long flags; | |
576 | struct rcu_node *rnp; | |
577 | struct rcu_state *rsp = &rcu_preempt_state; | |
578 | long snap; | |
579 | int trycount = 0; | |
580 | ||
581 | smp_mb(); /* Caller's modifications seen first by other CPUs. */ | |
582 | snap = ACCESS_ONCE(sync_rcu_preempt_exp_count) + 1; | |
583 | smp_mb(); /* Above access cannot bleed into critical section. */ | |
584 | ||
585 | /* | |
586 | * Acquire lock, falling back to synchronize_rcu() if too many | |
587 | * lock-acquisition failures. Of course, if someone does the | |
588 | * expedited grace period for us, just leave. | |
589 | */ | |
590 | while (!mutex_trylock(&sync_rcu_preempt_exp_mutex)) { | |
591 | if (trycount++ < 10) | |
592 | udelay(trycount * num_online_cpus()); | |
593 | else { | |
594 | synchronize_rcu(); | |
595 | return; | |
596 | } | |
597 | if ((ACCESS_ONCE(sync_rcu_preempt_exp_count) - snap) > 0) | |
598 | goto mb_ret; /* Others did our work for us. */ | |
599 | } | |
600 | if ((ACCESS_ONCE(sync_rcu_preempt_exp_count) - snap) > 0) | |
601 | goto unlock_mb_ret; /* Others did our work for us. */ | |
602 | ||
603 | /* force all RCU readers onto blocked_tasks[]. */ | |
604 | synchronize_sched_expedited(); | |
605 | ||
606 | spin_lock_irqsave(&rsp->onofflock, flags); | |
607 | ||
608 | /* Initialize ->expmask for all non-leaf rcu_node structures. */ | |
609 | rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) { | |
610 | spin_lock(&rnp->lock); /* irqs already disabled. */ | |
611 | rnp->expmask = rnp->qsmaskinit; | |
612 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ | |
613 | } | |
614 | ||
615 | /* Snapshot current state of ->blocked_tasks[] lists. */ | |
616 | rcu_for_each_leaf_node(rsp, rnp) | |
617 | sync_rcu_preempt_exp_init(rsp, rnp); | |
618 | if (NUM_RCU_NODES > 1) | |
619 | sync_rcu_preempt_exp_init(rsp, rcu_get_root(rsp)); | |
620 | ||
621 | spin_unlock_irqrestore(&rsp->onofflock, flags); | |
622 | ||
623 | /* Wait for snapshotted ->blocked_tasks[] lists to drain. */ | |
624 | rnp = rcu_get_root(rsp); | |
625 | wait_event(sync_rcu_preempt_exp_wq, | |
626 | sync_rcu_preempt_exp_done(rnp)); | |
627 | ||
628 | /* Clean up and exit. */ | |
629 | smp_mb(); /* ensure expedited GP seen before counter increment. */ | |
630 | ACCESS_ONCE(sync_rcu_preempt_exp_count)++; | |
631 | unlock_mb_ret: | |
632 | mutex_unlock(&sync_rcu_preempt_exp_mutex); | |
633 | mb_ret: | |
634 | smp_mb(); /* ensure subsequent action seen after grace period. */ | |
635 | } | |
636 | EXPORT_SYMBOL_GPL(synchronize_rcu_expedited); | |
637 | ||
638 | /* | |
639 | * Check to see if there is any immediate preemptable-RCU-related work | |
640 | * to be done. | |
641 | */ | |
642 | static int rcu_preempt_pending(int cpu) | |
643 | { | |
644 | return __rcu_pending(&rcu_preempt_state, | |
645 | &per_cpu(rcu_preempt_data, cpu)); | |
646 | } | |
647 | ||
648 | /* | |
649 | * Does preemptable RCU need the CPU to stay out of dynticks mode? | |
650 | */ | |
651 | static int rcu_preempt_needs_cpu(int cpu) | |
652 | { | |
653 | return !!per_cpu(rcu_preempt_data, cpu).nxtlist; | |
654 | } | |
655 | ||
656 | /** | |
657 | * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete. | |
658 | */ | |
659 | void rcu_barrier(void) | |
660 | { | |
661 | _rcu_barrier(&rcu_preempt_state, call_rcu); | |
662 | } | |
663 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
664 | ||
665 | /* | |
666 | * Initialize preemptable RCU's per-CPU data. | |
667 | */ | |
668 | static void __cpuinit rcu_preempt_init_percpu_data(int cpu) | |
669 | { | |
670 | rcu_init_percpu_data(cpu, &rcu_preempt_state, 1); | |
671 | } | |
672 | ||
673 | /* | |
674 | * Move preemptable RCU's callbacks to ->orphan_cbs_list. | |
675 | */ | |
676 | static void rcu_preempt_send_cbs_to_orphanage(void) | |
677 | { | |
678 | rcu_send_cbs_to_orphanage(&rcu_preempt_state); | |
679 | } | |
680 | ||
681 | /* | |
682 | * Initialize preemptable RCU's state structures. | |
683 | */ | |
684 | static void __init __rcu_init_preempt(void) | |
685 | { | |
686 | RCU_INIT_FLAVOR(&rcu_preempt_state, rcu_preempt_data); | |
687 | } | |
688 | ||
689 | /* | |
690 | * Check for a task exiting while in a preemptable-RCU read-side | |
691 | * critical section, clean up if so. No need to issue warnings, | |
692 | * as debug_check_no_locks_held() already does this if lockdep | |
693 | * is enabled. | |
694 | */ | |
695 | void exit_rcu(void) | |
696 | { | |
697 | struct task_struct *t = current; | |
698 | ||
699 | if (t->rcu_read_lock_nesting == 0) | |
700 | return; | |
701 | t->rcu_read_lock_nesting = 1; | |
702 | rcu_read_unlock(); | |
703 | } | |
704 | ||
705 | #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */ | |
706 | ||
707 | /* | |
708 | * Tell them what RCU they are running. | |
709 | */ | |
710 | static void __init rcu_bootup_announce(void) | |
711 | { | |
712 | printk(KERN_INFO "Hierarchical RCU implementation.\n"); | |
713 | } | |
714 | ||
715 | /* | |
716 | * Return the number of RCU batches processed thus far for debug & stats. | |
717 | */ | |
718 | long rcu_batches_completed(void) | |
719 | { | |
720 | return rcu_batches_completed_sched(); | |
721 | } | |
722 | EXPORT_SYMBOL_GPL(rcu_batches_completed); | |
723 | ||
724 | /* | |
725 | * Force a quiescent state for RCU, which, because there is no preemptible | |
726 | * RCU, becomes the same as rcu-sched. | |
727 | */ | |
728 | void rcu_force_quiescent_state(void) | |
729 | { | |
730 | rcu_sched_force_quiescent_state(); | |
731 | } | |
732 | EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); | |
733 | ||
734 | /* | |
735 | * Because preemptable RCU does not exist, we never have to check for | |
736 | * CPUs being in quiescent states. | |
737 | */ | |
738 | static void rcu_preempt_note_context_switch(int cpu) | |
739 | { | |
740 | } | |
741 | ||
742 | /* | |
743 | * Because preemptable RCU does not exist, there are never any preempted | |
744 | * RCU readers. | |
745 | */ | |
746 | static int rcu_preempted_readers(struct rcu_node *rnp) | |
747 | { | |
748 | return 0; | |
749 | } | |
750 | ||
751 | #ifdef CONFIG_HOTPLUG_CPU | |
752 | ||
753 | /* Because preemptible RCU does not exist, no quieting of tasks. */ | |
754 | static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags) | |
755 | { | |
756 | spin_unlock_irqrestore(&rnp->lock, flags); | |
757 | } | |
758 | ||
759 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | |
760 | ||
761 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR | |
762 | ||
763 | /* | |
764 | * Because preemptable RCU does not exist, we never have to check for | |
765 | * tasks blocked within RCU read-side critical sections. | |
766 | */ | |
767 | static void rcu_print_task_stall(struct rcu_node *rnp) | |
768 | { | |
769 | } | |
770 | ||
771 | #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | |
772 | ||
773 | /* | |
774 | * Because there is no preemptable RCU, there can be no readers blocked, | |
775 | * so there is no need to check for blocked tasks. So check only for | |
776 | * bogus qsmask values. | |
777 | */ | |
778 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) | |
779 | { | |
780 | WARN_ON_ONCE(rnp->qsmask); | |
781 | } | |
782 | ||
783 | #ifdef CONFIG_HOTPLUG_CPU | |
784 | ||
785 | /* | |
786 | * Because preemptable RCU does not exist, it never needs to migrate | |
787 | * tasks that were blocked within RCU read-side critical sections, and | |
788 | * such non-existent tasks cannot possibly have been blocking the current | |
789 | * grace period. | |
790 | */ | |
791 | static int rcu_preempt_offline_tasks(struct rcu_state *rsp, | |
792 | struct rcu_node *rnp, | |
793 | struct rcu_data *rdp) | |
794 | { | |
795 | return 0; | |
796 | } | |
797 | ||
798 | /* | |
799 | * Because preemptable RCU does not exist, it never needs CPU-offline | |
800 | * processing. | |
801 | */ | |
802 | static void rcu_preempt_offline_cpu(int cpu) | |
803 | { | |
804 | } | |
805 | ||
806 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | |
807 | ||
808 | /* | |
809 | * Because preemptable RCU does not exist, it never has any callbacks | |
810 | * to check. | |
811 | */ | |
812 | static void rcu_preempt_check_callbacks(int cpu) | |
813 | { | |
814 | } | |
815 | ||
816 | /* | |
817 | * Because preemptable RCU does not exist, it never has any callbacks | |
818 | * to process. | |
819 | */ | |
820 | static void rcu_preempt_process_callbacks(void) | |
821 | { | |
822 | } | |
823 | ||
824 | /* | |
825 | * In classic RCU, call_rcu() is just call_rcu_sched(). | |
826 | */ | |
827 | void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
828 | { | |
829 | call_rcu_sched(head, func); | |
830 | } | |
831 | EXPORT_SYMBOL_GPL(call_rcu); | |
832 | ||
833 | /* | |
834 | * Wait for an rcu-preempt grace period, but make it happen quickly. | |
835 | * But because preemptable RCU does not exist, map to rcu-sched. | |
836 | */ | |
837 | void synchronize_rcu_expedited(void) | |
838 | { | |
839 | synchronize_sched_expedited(); | |
840 | } | |
841 | EXPORT_SYMBOL_GPL(synchronize_rcu_expedited); | |
842 | ||
843 | #ifdef CONFIG_HOTPLUG_CPU | |
844 | ||
845 | /* | |
846 | * Because preemptable RCU does not exist, there is never any need to | |
847 | * report on tasks preempted in RCU read-side critical sections during | |
848 | * expedited RCU grace periods. | |
849 | */ | |
850 | static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp) | |
851 | { | |
852 | return; | |
853 | } | |
854 | ||
855 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | |
856 | ||
857 | /* | |
858 | * Because preemptable RCU does not exist, it never has any work to do. | |
859 | */ | |
860 | static int rcu_preempt_pending(int cpu) | |
861 | { | |
862 | return 0; | |
863 | } | |
864 | ||
865 | /* | |
866 | * Because preemptable RCU does not exist, it never needs any CPU. | |
867 | */ | |
868 | static int rcu_preempt_needs_cpu(int cpu) | |
869 | { | |
870 | return 0; | |
871 | } | |
872 | ||
873 | /* | |
874 | * Because preemptable RCU does not exist, rcu_barrier() is just | |
875 | * another name for rcu_barrier_sched(). | |
876 | */ | |
877 | void rcu_barrier(void) | |
878 | { | |
879 | rcu_barrier_sched(); | |
880 | } | |
881 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
882 | ||
883 | /* | |
884 | * Because preemptable RCU does not exist, there is no per-CPU | |
885 | * data to initialize. | |
886 | */ | |
887 | static void __cpuinit rcu_preempt_init_percpu_data(int cpu) | |
888 | { | |
889 | } | |
890 | ||
891 | /* | |
892 | * Because there is no preemptable RCU, there are no callbacks to move. | |
893 | */ | |
894 | static void rcu_preempt_send_cbs_to_orphanage(void) | |
895 | { | |
896 | } | |
897 | ||
898 | /* | |
899 | * Because preemptable RCU does not exist, it need not be initialized. | |
900 | */ | |
901 | static void __init __rcu_init_preempt(void) | |
902 | { | |
903 | } | |
904 | ||
905 | #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */ |