]>
Commit | Line | Data |
---|---|---|
f41d911f PM |
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 | ||
d9a3da06 | 27 | #include <linux/delay.h> |
f41d911f PM |
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 | ||
d9a3da06 PM |
34 | static int rcu_preempted_readers_exp(struct rcu_node *rnp); |
35 | ||
f41d911f PM |
36 | /* |
37 | * Tell them what RCU they are running. | |
38 | */ | |
0e0fc1c2 | 39 | static void __init rcu_bootup_announce(void) |
f41d911f PM |
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 | ||
bf66f18e PM |
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 | ||
f41d911f PM |
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 | */ | |
c3422bea | 79 | static void rcu_preempt_qs(int cpu) |
f41d911f PM |
80 | { |
81 | struct rcu_data *rdp = &per_cpu(rcu_preempt_data, cpu); | |
c64ac3ce | 82 | rdp->passed_quiesc_completed = rdp->gpnum - 1; |
c3422bea PM |
83 | barrier(); |
84 | rdp->passed_quiesc = 1; | |
f41d911f PM |
85 | } |
86 | ||
87 | /* | |
c3422bea PM |
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. | |
f41d911f | 99 | */ |
c3422bea | 100 | static void rcu_preempt_note_context_switch(int cpu) |
f41d911f PM |
101 | { |
102 | struct task_struct *t = current; | |
c3422bea | 103 | unsigned long flags; |
f41d911f PM |
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; | |
c3422bea | 114 | spin_lock_irqsave(&rnp->lock, flags); |
f41d911f | 115 | t->rcu_read_unlock_special |= RCU_READ_UNLOCK_BLOCKED; |
86848966 | 116 | t->rcu_blocked_node = rnp; |
f41d911f PM |
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. | |
b0e165c0 PM |
127 | * |
128 | * But first, note that the current CPU must still be | |
129 | * on line! | |
f41d911f | 130 | */ |
b0e165c0 | 131 | WARN_ON_ONCE((rdp->grpmask & rnp->qsmaskinit) == 0); |
e7d8842e PM |
132 | WARN_ON_ONCE(!list_empty(&t->rcu_node_entry)); |
133 | phase = (rnp->gpnum + !(rnp->qsmask & rdp->grpmask)) & 0x1; | |
f41d911f | 134 | list_add(&t->rcu_node_entry, &rnp->blocked_tasks[phase]); |
c3422bea | 135 | spin_unlock_irqrestore(&rnp->lock, flags); |
f41d911f PM |
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 | */ | |
c3422bea | 147 | rcu_preempt_qs(cpu); |
e7d8842e | 148 | local_irq_save(flags); |
c3422bea | 149 | t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS; |
e7d8842e | 150 | local_irq_restore(flags); |
f41d911f PM |
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 | ||
fc2219d4 PM |
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 | { | |
d9a3da06 PM |
172 | int phase = rnp->gpnum & 0x1; |
173 | ||
174 | return !list_empty(&rnp->blocked_tasks[phase]) || | |
175 | !list_empty(&rnp->blocked_tasks[phase + 2]); | |
fc2219d4 PM |
176 | } |
177 | ||
b668c9cf PM |
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 | */ | |
d3f6bad3 | 185 | static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags) |
b668c9cf PM |
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 | */ | |
d3f6bad3 | 203 | rcu_report_qs_rsp(&rcu_preempt_state, flags); |
b668c9cf PM |
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. */ | |
d3f6bad3 | 211 | rcu_report_qs_rnp(mask, &rcu_preempt_state, rnp_p, flags); |
b668c9cf PM |
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 | */ | |
f41d911f PM |
219 | static void rcu_read_unlock_special(struct task_struct *t) |
220 | { | |
221 | int empty; | |
d9a3da06 | 222 | int empty_exp; |
f41d911f | 223 | unsigned long flags; |
f41d911f PM |
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; | |
c3422bea | 240 | rcu_preempt_qs(smp_processor_id()); |
f41d911f PM |
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 | ||
dd5d19ba PM |
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 (;;) { | |
86848966 | 259 | rnp = t->rcu_blocked_node; |
e7d8842e | 260 | spin_lock(&rnp->lock); /* irqs already disabled. */ |
86848966 | 261 | if (rnp == t->rcu_blocked_node) |
dd5d19ba | 262 | break; |
e7d8842e | 263 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
dd5d19ba | 264 | } |
fc2219d4 | 265 | empty = !rcu_preempted_readers(rnp); |
d9a3da06 PM |
266 | empty_exp = !rcu_preempted_readers_exp(rnp); |
267 | smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */ | |
f41d911f | 268 | list_del_init(&t->rcu_node_entry); |
dd5d19ba | 269 | t->rcu_blocked_node = NULL; |
f41d911f PM |
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. | |
d3f6bad3 | 274 | * Note that rcu_report_unblock_qs_rnp() releases rnp->lock. |
f41d911f | 275 | */ |
b668c9cf | 276 | if (empty) |
f41d911f | 277 | spin_unlock_irqrestore(&rnp->lock, flags); |
b668c9cf | 278 | else |
d3f6bad3 | 279 | rcu_report_unblock_qs_rnp(rnp, flags); |
d9a3da06 PM |
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); | |
b668c9cf PM |
287 | } else { |
288 | local_irq_restore(flags); | |
f41d911f | 289 | } |
f41d911f PM |
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); | |
cba8244a PM |
307 | #ifdef CONFIG_PROVE_LOCKING |
308 | WARN_ON_ONCE(ACCESS_ONCE(t->rcu_read_lock_nesting) < 0); | |
309 | #endif /* #ifdef CONFIG_PROVE_LOCKING */ | |
f41d911f PM |
310 | } |
311 | EXPORT_SYMBOL_GPL(__rcu_read_unlock); | |
312 | ||
313 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR | |
314 | ||
315 | /* | |
316 | * Scan the current list of tasks blocked within RCU read-side critical | |
317 | * sections, printing out the tid of each. | |
318 | */ | |
319 | static void rcu_print_task_stall(struct rcu_node *rnp) | |
320 | { | |
321 | unsigned long flags; | |
322 | struct list_head *lp; | |
fc2219d4 | 323 | int phase; |
f41d911f PM |
324 | struct task_struct *t; |
325 | ||
fc2219d4 | 326 | if (rcu_preempted_readers(rnp)) { |
f41d911f | 327 | spin_lock_irqsave(&rnp->lock, flags); |
fc2219d4 | 328 | phase = rnp->gpnum & 0x1; |
f41d911f PM |
329 | lp = &rnp->blocked_tasks[phase]; |
330 | list_for_each_entry(t, lp, rcu_node_entry) | |
331 | printk(" P%d", t->pid); | |
332 | spin_unlock_irqrestore(&rnp->lock, flags); | |
333 | } | |
334 | } | |
335 | ||
336 | #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | |
337 | ||
b0e165c0 PM |
338 | /* |
339 | * Check that the list of blocked tasks for the newly completed grace | |
340 | * period is in fact empty. It is a serious bug to complete a grace | |
341 | * period that still has RCU readers blocked! This function must be | |
342 | * invoked -before- updating this rnp's ->gpnum, and the rnp's ->lock | |
343 | * must be held by the caller. | |
344 | */ | |
345 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) | |
346 | { | |
fc2219d4 | 347 | WARN_ON_ONCE(rcu_preempted_readers(rnp)); |
28ecd580 | 348 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
349 | } |
350 | ||
33f76148 PM |
351 | #ifdef CONFIG_HOTPLUG_CPU |
352 | ||
dd5d19ba PM |
353 | /* |
354 | * Handle tasklist migration for case in which all CPUs covered by the | |
355 | * specified rcu_node have gone offline. Move them up to the root | |
356 | * rcu_node. The reason for not just moving them to the immediate | |
357 | * parent is to remove the need for rcu_read_unlock_special() to | |
358 | * make more than two attempts to acquire the target rcu_node's lock. | |
b668c9cf PM |
359 | * Returns true if there were tasks blocking the current RCU grace |
360 | * period. | |
dd5d19ba | 361 | * |
237c80c5 PM |
362 | * Returns 1 if there was previously a task blocking the current grace |
363 | * period on the specified rcu_node structure. | |
364 | * | |
dd5d19ba PM |
365 | * The caller must hold rnp->lock with irqs disabled. |
366 | */ | |
237c80c5 PM |
367 | static int rcu_preempt_offline_tasks(struct rcu_state *rsp, |
368 | struct rcu_node *rnp, | |
369 | struct rcu_data *rdp) | |
dd5d19ba PM |
370 | { |
371 | int i; | |
372 | struct list_head *lp; | |
373 | struct list_head *lp_root; | |
d9a3da06 | 374 | int retval = 0; |
dd5d19ba PM |
375 | struct rcu_node *rnp_root = rcu_get_root(rsp); |
376 | struct task_struct *tp; | |
377 | ||
86848966 PM |
378 | if (rnp == rnp_root) { |
379 | WARN_ONCE(1, "Last CPU thought to be offlined?"); | |
237c80c5 | 380 | return 0; /* Shouldn't happen: at least one CPU online. */ |
86848966 | 381 | } |
28ecd580 PM |
382 | WARN_ON_ONCE(rnp != rdp->mynode && |
383 | (!list_empty(&rnp->blocked_tasks[0]) || | |
d9a3da06 PM |
384 | !list_empty(&rnp->blocked_tasks[1]) || |
385 | !list_empty(&rnp->blocked_tasks[2]) || | |
386 | !list_empty(&rnp->blocked_tasks[3]))); | |
dd5d19ba PM |
387 | |
388 | /* | |
389 | * Move tasks up to root rcu_node. Rely on the fact that the | |
390 | * root rcu_node can be at most one ahead of the rest of the | |
391 | * rcu_nodes in terms of gp_num value. This fact allows us to | |
392 | * move the blocked_tasks[] array directly, element by element. | |
393 | */ | |
d9a3da06 PM |
394 | if (rcu_preempted_readers(rnp)) |
395 | retval |= RCU_OFL_TASKS_NORM_GP; | |
396 | if (rcu_preempted_readers_exp(rnp)) | |
397 | retval |= RCU_OFL_TASKS_EXP_GP; | |
398 | for (i = 0; i < 4; i++) { | |
dd5d19ba PM |
399 | lp = &rnp->blocked_tasks[i]; |
400 | lp_root = &rnp_root->blocked_tasks[i]; | |
401 | while (!list_empty(lp)) { | |
402 | tp = list_entry(lp->next, typeof(*tp), rcu_node_entry); | |
403 | spin_lock(&rnp_root->lock); /* irqs already disabled */ | |
404 | list_del(&tp->rcu_node_entry); | |
405 | tp->rcu_blocked_node = rnp_root; | |
406 | list_add(&tp->rcu_node_entry, lp_root); | |
407 | spin_unlock(&rnp_root->lock); /* irqs remain disabled */ | |
408 | } | |
409 | } | |
237c80c5 | 410 | return retval; |
dd5d19ba PM |
411 | } |
412 | ||
33f76148 PM |
413 | /* |
414 | * Do CPU-offline processing for preemptable RCU. | |
415 | */ | |
416 | static void rcu_preempt_offline_cpu(int cpu) | |
417 | { | |
418 | __rcu_offline_cpu(cpu, &rcu_preempt_state); | |
419 | } | |
420 | ||
421 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | |
422 | ||
f41d911f PM |
423 | /* |
424 | * Check for a quiescent state from the current CPU. When a task blocks, | |
425 | * the task is recorded in the corresponding CPU's rcu_node structure, | |
426 | * which is checked elsewhere. | |
427 | * | |
428 | * Caller must disable hard irqs. | |
429 | */ | |
430 | static void rcu_preempt_check_callbacks(int cpu) | |
431 | { | |
432 | struct task_struct *t = current; | |
433 | ||
434 | if (t->rcu_read_lock_nesting == 0) { | |
c3422bea PM |
435 | t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS; |
436 | rcu_preempt_qs(cpu); | |
f41d911f PM |
437 | return; |
438 | } | |
a71fca58 | 439 | if (per_cpu(rcu_preempt_data, cpu).qs_pending) |
c3422bea | 440 | t->rcu_read_unlock_special |= RCU_READ_UNLOCK_NEED_QS; |
f41d911f PM |
441 | } |
442 | ||
443 | /* | |
444 | * Process callbacks for preemptable RCU. | |
445 | */ | |
446 | static void rcu_preempt_process_callbacks(void) | |
447 | { | |
448 | __rcu_process_callbacks(&rcu_preempt_state, | |
449 | &__get_cpu_var(rcu_preempt_data)); | |
450 | } | |
451 | ||
452 | /* | |
453 | * Queue a preemptable-RCU callback for invocation after a grace period. | |
454 | */ | |
455 | void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
456 | { | |
457 | __call_rcu(head, func, &rcu_preempt_state); | |
458 | } | |
459 | EXPORT_SYMBOL_GPL(call_rcu); | |
460 | ||
6ebb237b PM |
461 | /** |
462 | * synchronize_rcu - wait until a grace period has elapsed. | |
463 | * | |
464 | * Control will return to the caller some time after a full grace | |
465 | * period has elapsed, in other words after all currently executing RCU | |
466 | * read-side critical sections have completed. RCU read-side critical | |
467 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), | |
468 | * and may be nested. | |
469 | */ | |
470 | void synchronize_rcu(void) | |
471 | { | |
472 | struct rcu_synchronize rcu; | |
473 | ||
474 | if (!rcu_scheduler_active) | |
475 | return; | |
476 | ||
477 | init_completion(&rcu.completion); | |
478 | /* Will wake me after RCU finished. */ | |
479 | call_rcu(&rcu.head, wakeme_after_rcu); | |
480 | /* Wait for it. */ | |
481 | wait_for_completion(&rcu.completion); | |
482 | } | |
483 | EXPORT_SYMBOL_GPL(synchronize_rcu); | |
484 | ||
d9a3da06 PM |
485 | static DECLARE_WAIT_QUEUE_HEAD(sync_rcu_preempt_exp_wq); |
486 | static long sync_rcu_preempt_exp_count; | |
487 | static DEFINE_MUTEX(sync_rcu_preempt_exp_mutex); | |
488 | ||
489 | /* | |
490 | * Return non-zero if there are any tasks in RCU read-side critical | |
491 | * sections blocking the current preemptible-RCU expedited grace period. | |
492 | * If there is no preemptible-RCU expedited grace period currently in | |
493 | * progress, returns zero unconditionally. | |
494 | */ | |
495 | static int rcu_preempted_readers_exp(struct rcu_node *rnp) | |
496 | { | |
497 | return !list_empty(&rnp->blocked_tasks[2]) || | |
498 | !list_empty(&rnp->blocked_tasks[3]); | |
499 | } | |
500 | ||
501 | /* | |
502 | * return non-zero if there is no RCU expedited grace period in progress | |
503 | * for the specified rcu_node structure, in other words, if all CPUs and | |
504 | * tasks covered by the specified rcu_node structure have done their bit | |
505 | * for the current expedited grace period. Works only for preemptible | |
506 | * RCU -- other RCU implementation use other means. | |
507 | * | |
508 | * Caller must hold sync_rcu_preempt_exp_mutex. | |
509 | */ | |
510 | static int sync_rcu_preempt_exp_done(struct rcu_node *rnp) | |
511 | { | |
512 | return !rcu_preempted_readers_exp(rnp) && | |
513 | ACCESS_ONCE(rnp->expmask) == 0; | |
514 | } | |
515 | ||
516 | /* | |
517 | * Report the exit from RCU read-side critical section for the last task | |
518 | * that queued itself during or before the current expedited preemptible-RCU | |
519 | * grace period. This event is reported either to the rcu_node structure on | |
520 | * which the task was queued or to one of that rcu_node structure's ancestors, | |
521 | * recursively up the tree. (Calm down, calm down, we do the recursion | |
522 | * iteratively!) | |
523 | * | |
524 | * Caller must hold sync_rcu_preempt_exp_mutex. | |
525 | */ | |
526 | static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp) | |
527 | { | |
528 | unsigned long flags; | |
529 | unsigned long mask; | |
530 | ||
531 | spin_lock_irqsave(&rnp->lock, flags); | |
532 | for (;;) { | |
533 | if (!sync_rcu_preempt_exp_done(rnp)) | |
534 | break; | |
535 | if (rnp->parent == NULL) { | |
536 | wake_up(&sync_rcu_preempt_exp_wq); | |
537 | break; | |
538 | } | |
539 | mask = rnp->grpmask; | |
540 | spin_unlock(&rnp->lock); /* irqs remain disabled */ | |
541 | rnp = rnp->parent; | |
542 | spin_lock(&rnp->lock); /* irqs already disabled */ | |
543 | rnp->expmask &= ~mask; | |
544 | } | |
545 | spin_unlock_irqrestore(&rnp->lock, flags); | |
546 | } | |
547 | ||
548 | /* | |
549 | * Snapshot the tasks blocking the newly started preemptible-RCU expedited | |
550 | * grace period for the specified rcu_node structure. If there are no such | |
551 | * tasks, report it up the rcu_node hierarchy. | |
552 | * | |
553 | * Caller must hold sync_rcu_preempt_exp_mutex and rsp->onofflock. | |
554 | */ | |
555 | static void | |
556 | sync_rcu_preempt_exp_init(struct rcu_state *rsp, struct rcu_node *rnp) | |
557 | { | |
558 | int must_wait; | |
559 | ||
560 | spin_lock(&rnp->lock); /* irqs already disabled */ | |
561 | list_splice_init(&rnp->blocked_tasks[0], &rnp->blocked_tasks[2]); | |
562 | list_splice_init(&rnp->blocked_tasks[1], &rnp->blocked_tasks[3]); | |
563 | must_wait = rcu_preempted_readers_exp(rnp); | |
564 | spin_unlock(&rnp->lock); /* irqs remain disabled */ | |
565 | if (!must_wait) | |
566 | rcu_report_exp_rnp(rsp, rnp); | |
567 | } | |
568 | ||
019129d5 | 569 | /* |
d9a3da06 PM |
570 | * Wait for an rcu-preempt grace period, but expedite it. The basic idea |
571 | * is to invoke synchronize_sched_expedited() to push all the tasks to | |
572 | * the ->blocked_tasks[] lists, move all entries from the first set of | |
573 | * ->blocked_tasks[] lists to the second set, and finally wait for this | |
574 | * second set to drain. | |
019129d5 PM |
575 | */ |
576 | void synchronize_rcu_expedited(void) | |
577 | { | |
d9a3da06 PM |
578 | unsigned long flags; |
579 | struct rcu_node *rnp; | |
580 | struct rcu_state *rsp = &rcu_preempt_state; | |
581 | long snap; | |
582 | int trycount = 0; | |
583 | ||
584 | smp_mb(); /* Caller's modifications seen first by other CPUs. */ | |
585 | snap = ACCESS_ONCE(sync_rcu_preempt_exp_count) + 1; | |
586 | smp_mb(); /* Above access cannot bleed into critical section. */ | |
587 | ||
588 | /* | |
589 | * Acquire lock, falling back to synchronize_rcu() if too many | |
590 | * lock-acquisition failures. Of course, if someone does the | |
591 | * expedited grace period for us, just leave. | |
592 | */ | |
593 | while (!mutex_trylock(&sync_rcu_preempt_exp_mutex)) { | |
594 | if (trycount++ < 10) | |
595 | udelay(trycount * num_online_cpus()); | |
596 | else { | |
597 | synchronize_rcu(); | |
598 | return; | |
599 | } | |
600 | if ((ACCESS_ONCE(sync_rcu_preempt_exp_count) - snap) > 0) | |
601 | goto mb_ret; /* Others did our work for us. */ | |
602 | } | |
603 | if ((ACCESS_ONCE(sync_rcu_preempt_exp_count) - snap) > 0) | |
604 | goto unlock_mb_ret; /* Others did our work for us. */ | |
605 | ||
606 | /* force all RCU readers onto blocked_tasks[]. */ | |
607 | synchronize_sched_expedited(); | |
608 | ||
609 | spin_lock_irqsave(&rsp->onofflock, flags); | |
610 | ||
611 | /* Initialize ->expmask for all non-leaf rcu_node structures. */ | |
612 | rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) { | |
613 | spin_lock(&rnp->lock); /* irqs already disabled. */ | |
614 | rnp->expmask = rnp->qsmaskinit; | |
615 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ | |
616 | } | |
617 | ||
618 | /* Snapshot current state of ->blocked_tasks[] lists. */ | |
619 | rcu_for_each_leaf_node(rsp, rnp) | |
620 | sync_rcu_preempt_exp_init(rsp, rnp); | |
621 | if (NUM_RCU_NODES > 1) | |
622 | sync_rcu_preempt_exp_init(rsp, rcu_get_root(rsp)); | |
623 | ||
624 | spin_unlock_irqrestore(&rsp->onofflock, flags); | |
625 | ||
626 | /* Wait for snapshotted ->blocked_tasks[] lists to drain. */ | |
627 | rnp = rcu_get_root(rsp); | |
628 | wait_event(sync_rcu_preempt_exp_wq, | |
629 | sync_rcu_preempt_exp_done(rnp)); | |
630 | ||
631 | /* Clean up and exit. */ | |
632 | smp_mb(); /* ensure expedited GP seen before counter increment. */ | |
633 | ACCESS_ONCE(sync_rcu_preempt_exp_count)++; | |
634 | unlock_mb_ret: | |
635 | mutex_unlock(&sync_rcu_preempt_exp_mutex); | |
636 | mb_ret: | |
637 | smp_mb(); /* ensure subsequent action seen after grace period. */ | |
019129d5 PM |
638 | } |
639 | EXPORT_SYMBOL_GPL(synchronize_rcu_expedited); | |
640 | ||
f41d911f PM |
641 | /* |
642 | * Check to see if there is any immediate preemptable-RCU-related work | |
643 | * to be done. | |
644 | */ | |
645 | static int rcu_preempt_pending(int cpu) | |
646 | { | |
647 | return __rcu_pending(&rcu_preempt_state, | |
648 | &per_cpu(rcu_preempt_data, cpu)); | |
649 | } | |
650 | ||
651 | /* | |
652 | * Does preemptable RCU need the CPU to stay out of dynticks mode? | |
653 | */ | |
654 | static int rcu_preempt_needs_cpu(int cpu) | |
655 | { | |
656 | return !!per_cpu(rcu_preempt_data, cpu).nxtlist; | |
657 | } | |
658 | ||
e74f4c45 PM |
659 | /** |
660 | * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete. | |
661 | */ | |
662 | void rcu_barrier(void) | |
663 | { | |
664 | _rcu_barrier(&rcu_preempt_state, call_rcu); | |
665 | } | |
666 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
667 | ||
f41d911f PM |
668 | /* |
669 | * Initialize preemptable RCU's per-CPU data. | |
670 | */ | |
671 | static void __cpuinit rcu_preempt_init_percpu_data(int cpu) | |
672 | { | |
673 | rcu_init_percpu_data(cpu, &rcu_preempt_state, 1); | |
674 | } | |
675 | ||
e74f4c45 PM |
676 | /* |
677 | * Move preemptable RCU's callbacks to ->orphan_cbs_list. | |
678 | */ | |
679 | static void rcu_preempt_send_cbs_to_orphanage(void) | |
680 | { | |
681 | rcu_send_cbs_to_orphanage(&rcu_preempt_state); | |
682 | } | |
683 | ||
1eba8f84 PM |
684 | /* |
685 | * Initialize preemptable RCU's state structures. | |
686 | */ | |
687 | static void __init __rcu_init_preempt(void) | |
688 | { | |
1eba8f84 PM |
689 | RCU_INIT_FLAVOR(&rcu_preempt_state, rcu_preempt_data); |
690 | } | |
691 | ||
f41d911f PM |
692 | /* |
693 | * Check for a task exiting while in a preemptable-RCU read-side | |
694 | * critical section, clean up if so. No need to issue warnings, | |
695 | * as debug_check_no_locks_held() already does this if lockdep | |
696 | * is enabled. | |
697 | */ | |
698 | void exit_rcu(void) | |
699 | { | |
700 | struct task_struct *t = current; | |
701 | ||
702 | if (t->rcu_read_lock_nesting == 0) | |
703 | return; | |
704 | t->rcu_read_lock_nesting = 1; | |
705 | rcu_read_unlock(); | |
706 | } | |
707 | ||
708 | #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */ | |
709 | ||
710 | /* | |
711 | * Tell them what RCU they are running. | |
712 | */ | |
0e0fc1c2 | 713 | static void __init rcu_bootup_announce(void) |
f41d911f PM |
714 | { |
715 | printk(KERN_INFO "Hierarchical RCU implementation.\n"); | |
716 | } | |
717 | ||
718 | /* | |
719 | * Return the number of RCU batches processed thus far for debug & stats. | |
720 | */ | |
721 | long rcu_batches_completed(void) | |
722 | { | |
723 | return rcu_batches_completed_sched(); | |
724 | } | |
725 | EXPORT_SYMBOL_GPL(rcu_batches_completed); | |
726 | ||
bf66f18e PM |
727 | /* |
728 | * Force a quiescent state for RCU, which, because there is no preemptible | |
729 | * RCU, becomes the same as rcu-sched. | |
730 | */ | |
731 | void rcu_force_quiescent_state(void) | |
732 | { | |
733 | rcu_sched_force_quiescent_state(); | |
734 | } | |
735 | EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); | |
736 | ||
f41d911f PM |
737 | /* |
738 | * Because preemptable RCU does not exist, we never have to check for | |
739 | * CPUs being in quiescent states. | |
740 | */ | |
c3422bea | 741 | static void rcu_preempt_note_context_switch(int cpu) |
f41d911f PM |
742 | { |
743 | } | |
744 | ||
fc2219d4 PM |
745 | /* |
746 | * Because preemptable RCU does not exist, there are never any preempted | |
747 | * RCU readers. | |
748 | */ | |
749 | static int rcu_preempted_readers(struct rcu_node *rnp) | |
750 | { | |
751 | return 0; | |
752 | } | |
753 | ||
b668c9cf PM |
754 | #ifdef CONFIG_HOTPLUG_CPU |
755 | ||
756 | /* Because preemptible RCU does not exist, no quieting of tasks. */ | |
d3f6bad3 | 757 | static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags) |
b668c9cf PM |
758 | { |
759 | spin_unlock_irqrestore(&rnp->lock, flags); | |
760 | } | |
761 | ||
762 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | |
763 | ||
f41d911f PM |
764 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR |
765 | ||
766 | /* | |
767 | * Because preemptable RCU does not exist, we never have to check for | |
768 | * tasks blocked within RCU read-side critical sections. | |
769 | */ | |
770 | static void rcu_print_task_stall(struct rcu_node *rnp) | |
771 | { | |
772 | } | |
773 | ||
774 | #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ | |
775 | ||
b0e165c0 PM |
776 | /* |
777 | * Because there is no preemptable RCU, there can be no readers blocked, | |
49e29126 PM |
778 | * so there is no need to check for blocked tasks. So check only for |
779 | * bogus qsmask values. | |
b0e165c0 PM |
780 | */ |
781 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) | |
782 | { | |
49e29126 | 783 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
784 | } |
785 | ||
33f76148 PM |
786 | #ifdef CONFIG_HOTPLUG_CPU |
787 | ||
dd5d19ba PM |
788 | /* |
789 | * Because preemptable RCU does not exist, it never needs to migrate | |
237c80c5 PM |
790 | * tasks that were blocked within RCU read-side critical sections, and |
791 | * such non-existent tasks cannot possibly have been blocking the current | |
792 | * grace period. | |
dd5d19ba | 793 | */ |
237c80c5 PM |
794 | static int rcu_preempt_offline_tasks(struct rcu_state *rsp, |
795 | struct rcu_node *rnp, | |
796 | struct rcu_data *rdp) | |
dd5d19ba | 797 | { |
237c80c5 | 798 | return 0; |
dd5d19ba PM |
799 | } |
800 | ||
33f76148 PM |
801 | /* |
802 | * Because preemptable RCU does not exist, it never needs CPU-offline | |
803 | * processing. | |
804 | */ | |
805 | static void rcu_preempt_offline_cpu(int cpu) | |
806 | { | |
807 | } | |
808 | ||
809 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | |
810 | ||
f41d911f PM |
811 | /* |
812 | * Because preemptable RCU does not exist, it never has any callbacks | |
813 | * to check. | |
814 | */ | |
1eba8f84 | 815 | static void rcu_preempt_check_callbacks(int cpu) |
f41d911f PM |
816 | { |
817 | } | |
818 | ||
819 | /* | |
820 | * Because preemptable RCU does not exist, it never has any callbacks | |
821 | * to process. | |
822 | */ | |
1eba8f84 | 823 | static void rcu_preempt_process_callbacks(void) |
f41d911f PM |
824 | { |
825 | } | |
826 | ||
827 | /* | |
828 | * In classic RCU, call_rcu() is just call_rcu_sched(). | |
829 | */ | |
830 | void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
831 | { | |
832 | call_rcu_sched(head, func); | |
833 | } | |
834 | EXPORT_SYMBOL_GPL(call_rcu); | |
835 | ||
019129d5 PM |
836 | /* |
837 | * Wait for an rcu-preempt grace period, but make it happen quickly. | |
838 | * But because preemptable RCU does not exist, map to rcu-sched. | |
839 | */ | |
840 | void synchronize_rcu_expedited(void) | |
841 | { | |
842 | synchronize_sched_expedited(); | |
843 | } | |
844 | EXPORT_SYMBOL_GPL(synchronize_rcu_expedited); | |
845 | ||
d9a3da06 PM |
846 | #ifdef CONFIG_HOTPLUG_CPU |
847 | ||
848 | /* | |
849 | * Because preemptable RCU does not exist, there is never any need to | |
850 | * report on tasks preempted in RCU read-side critical sections during | |
851 | * expedited RCU grace periods. | |
852 | */ | |
853 | static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp) | |
854 | { | |
855 | return; | |
856 | } | |
857 | ||
858 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | |
859 | ||
f41d911f PM |
860 | /* |
861 | * Because preemptable RCU does not exist, it never has any work to do. | |
862 | */ | |
863 | static int rcu_preempt_pending(int cpu) | |
864 | { | |
865 | return 0; | |
866 | } | |
867 | ||
868 | /* | |
869 | * Because preemptable RCU does not exist, it never needs any CPU. | |
870 | */ | |
871 | static int rcu_preempt_needs_cpu(int cpu) | |
872 | { | |
873 | return 0; | |
874 | } | |
875 | ||
e74f4c45 PM |
876 | /* |
877 | * Because preemptable RCU does not exist, rcu_barrier() is just | |
878 | * another name for rcu_barrier_sched(). | |
879 | */ | |
880 | void rcu_barrier(void) | |
881 | { | |
882 | rcu_barrier_sched(); | |
883 | } | |
884 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
885 | ||
f41d911f PM |
886 | /* |
887 | * Because preemptable RCU does not exist, there is no per-CPU | |
888 | * data to initialize. | |
889 | */ | |
890 | static void __cpuinit rcu_preempt_init_percpu_data(int cpu) | |
891 | { | |
892 | } | |
893 | ||
e74f4c45 PM |
894 | /* |
895 | * Because there is no preemptable RCU, there are no callbacks to move. | |
896 | */ | |
897 | static void rcu_preempt_send_cbs_to_orphanage(void) | |
898 | { | |
899 | } | |
900 | ||
1eba8f84 PM |
901 | /* |
902 | * Because preemptable RCU does not exist, it need not be initialized. | |
903 | */ | |
904 | static void __init __rcu_init_preempt(void) | |
905 | { | |
906 | } | |
907 | ||
f41d911f | 908 | #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */ |
8bd93a2c PM |
909 | |
910 | #if !defined(CONFIG_RCU_FAST_NO_HZ) | |
911 | ||
912 | /* | |
913 | * Check to see if any future RCU-related work will need to be done | |
914 | * by the current CPU, even if none need be done immediately, returning | |
915 | * 1 if so. This function is part of the RCU implementation; it is -not- | |
916 | * an exported member of the RCU API. | |
917 | * | |
918 | * Because we have preemptible RCU, just check whether this CPU needs | |
919 | * any flavor of RCU. Do not chew up lots of CPU cycles with preemption | |
920 | * disabled in a most-likely vain attempt to cause RCU not to need this CPU. | |
921 | */ | |
922 | int rcu_needs_cpu(int cpu) | |
923 | { | |
924 | return rcu_needs_cpu_quick_check(cpu); | |
925 | } | |
926 | ||
927 | #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */ | |
928 | ||
929 | #define RCU_NEEDS_CPU_FLUSHES 5 | |
930 | ||
931 | /* | |
932 | * Check to see if any future RCU-related work will need to be done | |
933 | * by the current CPU, even if none need be done immediately, returning | |
934 | * 1 if so. This function is part of the RCU implementation; it is -not- | |
935 | * an exported member of the RCU API. | |
936 | * | |
937 | * Because we are not supporting preemptible RCU, attempt to accelerate | |
938 | * any current grace periods so that RCU no longer needs this CPU, but | |
939 | * only if all other CPUs are already in dynticks-idle mode. This will | |
940 | * allow the CPU cores to be powered down immediately, as opposed to after | |
941 | * waiting many milliseconds for grace periods to elapse. | |
942 | */ | |
943 | int rcu_needs_cpu(int cpu) | |
944 | { | |
945 | int c = 1; | |
946 | int i; | |
947 | int thatcpu; | |
948 | ||
949 | /* Don't bother unless we are the last non-dyntick-idle CPU. */ | |
950 | for_each_cpu_not(thatcpu, nohz_cpu_mask) | |
951 | if (thatcpu != cpu) | |
952 | return rcu_needs_cpu_quick_check(cpu); | |
953 | ||
954 | /* Try to push remaining RCU-sched and RCU-bh callbacks through. */ | |
955 | for (i = 0; i < RCU_NEEDS_CPU_FLUSHES && c; i++) { | |
956 | c = 0; | |
957 | if (per_cpu(rcu_sched_data, cpu).nxtlist) { | |
958 | rcu_sched_qs(cpu); | |
959 | force_quiescent_state(&rcu_sched_state, 0); | |
960 | __rcu_process_callbacks(&rcu_sched_state, | |
961 | &per_cpu(rcu_sched_data, cpu)); | |
962 | c = !!per_cpu(rcu_sched_data, cpu).nxtlist; | |
963 | } | |
964 | if (per_cpu(rcu_bh_data, cpu).nxtlist) { | |
965 | rcu_bh_qs(cpu); | |
966 | force_quiescent_state(&rcu_bh_state, 0); | |
967 | __rcu_process_callbacks(&rcu_bh_state, | |
968 | &per_cpu(rcu_bh_data, cpu)); | |
969 | c = !!per_cpu(rcu_bh_data, cpu).nxtlist; | |
970 | } | |
971 | } | |
972 | ||
973 | /* If RCU callbacks are still pending, RCU still needs this CPU. */ | |
974 | return c; | |
975 | } | |
976 | ||
977 | #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |