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> |
7b27d547 | 28 | #include <linux/stop_machine.h> |
f41d911f | 29 | |
26845c28 PM |
30 | /* |
31 | * Check the RCU kernel configuration parameters and print informative | |
32 | * messages about anything out of the ordinary. If you like #ifdef, you | |
33 | * will love this function. | |
34 | */ | |
35 | static void __init rcu_bootup_announce_oddness(void) | |
36 | { | |
37 | #ifdef CONFIG_RCU_TRACE | |
38 | printk(KERN_INFO "\tRCU debugfs-based tracing is enabled.\n"); | |
39 | #endif | |
40 | #if (defined(CONFIG_64BIT) && CONFIG_RCU_FANOUT != 64) || (!defined(CONFIG_64BIT) && CONFIG_RCU_FANOUT != 32) | |
41 | printk(KERN_INFO "\tCONFIG_RCU_FANOUT set to non-default value of %d\n", | |
42 | CONFIG_RCU_FANOUT); | |
43 | #endif | |
44 | #ifdef CONFIG_RCU_FANOUT_EXACT | |
45 | printk(KERN_INFO "\tHierarchical RCU autobalancing is disabled.\n"); | |
46 | #endif | |
47 | #ifdef CONFIG_RCU_FAST_NO_HZ | |
48 | printk(KERN_INFO | |
49 | "\tRCU dyntick-idle grace-period acceleration is enabled.\n"); | |
50 | #endif | |
51 | #ifdef CONFIG_PROVE_RCU | |
52 | printk(KERN_INFO "\tRCU lockdep checking is enabled.\n"); | |
53 | #endif | |
54 | #ifdef CONFIG_RCU_TORTURE_TEST_RUNNABLE | |
55 | printk(KERN_INFO "\tRCU torture testing starts during boot.\n"); | |
56 | #endif | |
81a294c4 | 57 | #if defined(CONFIG_TREE_PREEMPT_RCU) && !defined(CONFIG_RCU_CPU_STALL_VERBOSE) |
26845c28 PM |
58 | printk(KERN_INFO "\tVerbose stalled-CPUs detection is disabled.\n"); |
59 | #endif | |
60 | #if NUM_RCU_LVL_4 != 0 | |
61 | printk(KERN_INFO "\tExperimental four-level hierarchy is enabled.\n"); | |
62 | #endif | |
63 | } | |
64 | ||
f41d911f PM |
65 | #ifdef CONFIG_TREE_PREEMPT_RCU |
66 | ||
67 | struct rcu_state rcu_preempt_state = RCU_STATE_INITIALIZER(rcu_preempt_state); | |
68 | DEFINE_PER_CPU(struct rcu_data, rcu_preempt_data); | |
69 | ||
d9a3da06 PM |
70 | static int rcu_preempted_readers_exp(struct rcu_node *rnp); |
71 | ||
f41d911f PM |
72 | /* |
73 | * Tell them what RCU they are running. | |
74 | */ | |
0e0fc1c2 | 75 | static void __init rcu_bootup_announce(void) |
f41d911f | 76 | { |
26845c28 PM |
77 | printk(KERN_INFO "Preemptable hierarchical RCU implementation.\n"); |
78 | rcu_bootup_announce_oddness(); | |
f41d911f PM |
79 | } |
80 | ||
81 | /* | |
82 | * Return the number of RCU-preempt batches processed thus far | |
83 | * for debug and statistics. | |
84 | */ | |
85 | long rcu_batches_completed_preempt(void) | |
86 | { | |
87 | return rcu_preempt_state.completed; | |
88 | } | |
89 | EXPORT_SYMBOL_GPL(rcu_batches_completed_preempt); | |
90 | ||
91 | /* | |
92 | * Return the number of RCU batches processed thus far for debug & stats. | |
93 | */ | |
94 | long rcu_batches_completed(void) | |
95 | { | |
96 | return rcu_batches_completed_preempt(); | |
97 | } | |
98 | EXPORT_SYMBOL_GPL(rcu_batches_completed); | |
99 | ||
bf66f18e PM |
100 | /* |
101 | * Force a quiescent state for preemptible RCU. | |
102 | */ | |
103 | void rcu_force_quiescent_state(void) | |
104 | { | |
105 | force_quiescent_state(&rcu_preempt_state, 0); | |
106 | } | |
107 | EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); | |
108 | ||
f41d911f PM |
109 | /* |
110 | * Record a preemptable-RCU quiescent state for the specified CPU. Note | |
111 | * that this just means that the task currently running on the CPU is | |
112 | * not in a quiescent state. There might be any number of tasks blocked | |
113 | * while in an RCU read-side critical section. | |
25502a6c PM |
114 | * |
115 | * Unlike the other rcu_*_qs() functions, callers to this function | |
116 | * must disable irqs in order to protect the assignment to | |
117 | * ->rcu_read_unlock_special. | |
f41d911f | 118 | */ |
c3422bea | 119 | static void rcu_preempt_qs(int cpu) |
f41d911f PM |
120 | { |
121 | struct rcu_data *rdp = &per_cpu(rcu_preempt_data, cpu); | |
25502a6c | 122 | |
c64ac3ce | 123 | rdp->passed_quiesc_completed = rdp->gpnum - 1; |
c3422bea PM |
124 | barrier(); |
125 | rdp->passed_quiesc = 1; | |
25502a6c | 126 | current->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS; |
f41d911f PM |
127 | } |
128 | ||
129 | /* | |
c3422bea PM |
130 | * We have entered the scheduler, and the current task might soon be |
131 | * context-switched away from. If this task is in an RCU read-side | |
132 | * critical section, we will no longer be able to rely on the CPU to | |
133 | * record that fact, so we enqueue the task on the appropriate entry | |
134 | * of the blocked_tasks[] array. The task will dequeue itself when | |
135 | * it exits the outermost enclosing RCU read-side critical section. | |
136 | * Therefore, the current grace period cannot be permitted to complete | |
137 | * until the blocked_tasks[] entry indexed by the low-order bit of | |
138 | * rnp->gpnum empties. | |
139 | * | |
140 | * Caller must disable preemption. | |
f41d911f | 141 | */ |
c3422bea | 142 | static void rcu_preempt_note_context_switch(int cpu) |
f41d911f PM |
143 | { |
144 | struct task_struct *t = current; | |
c3422bea | 145 | unsigned long flags; |
f41d911f PM |
146 | int phase; |
147 | struct rcu_data *rdp; | |
148 | struct rcu_node *rnp; | |
149 | ||
150 | if (t->rcu_read_lock_nesting && | |
151 | (t->rcu_read_unlock_special & RCU_READ_UNLOCK_BLOCKED) == 0) { | |
152 | ||
153 | /* Possibly blocking in an RCU read-side critical section. */ | |
394f99a9 | 154 | rdp = per_cpu_ptr(rcu_preempt_state.rda, cpu); |
f41d911f | 155 | rnp = rdp->mynode; |
1304afb2 | 156 | raw_spin_lock_irqsave(&rnp->lock, flags); |
f41d911f | 157 | t->rcu_read_unlock_special |= RCU_READ_UNLOCK_BLOCKED; |
86848966 | 158 | t->rcu_blocked_node = rnp; |
f41d911f PM |
159 | |
160 | /* | |
161 | * If this CPU has already checked in, then this task | |
162 | * will hold up the next grace period rather than the | |
163 | * current grace period. Queue the task accordingly. | |
164 | * If the task is queued for the current grace period | |
165 | * (i.e., this CPU has not yet passed through a quiescent | |
166 | * state for the current grace period), then as long | |
167 | * as that task remains queued, the current grace period | |
168 | * cannot end. | |
b0e165c0 PM |
169 | * |
170 | * But first, note that the current CPU must still be | |
171 | * on line! | |
f41d911f | 172 | */ |
b0e165c0 | 173 | WARN_ON_ONCE((rdp->grpmask & rnp->qsmaskinit) == 0); |
e7d8842e PM |
174 | WARN_ON_ONCE(!list_empty(&t->rcu_node_entry)); |
175 | phase = (rnp->gpnum + !(rnp->qsmask & rdp->grpmask)) & 0x1; | |
f41d911f | 176 | list_add(&t->rcu_node_entry, &rnp->blocked_tasks[phase]); |
1304afb2 | 177 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
f41d911f PM |
178 | } |
179 | ||
180 | /* | |
181 | * Either we were not in an RCU read-side critical section to | |
182 | * begin with, or we have now recorded that critical section | |
183 | * globally. Either way, we can now note a quiescent state | |
184 | * for this CPU. Again, if we were in an RCU read-side critical | |
185 | * section, and if that critical section was blocking the current | |
186 | * grace period, then the fact that the task has been enqueued | |
187 | * means that we continue to block the current grace period. | |
188 | */ | |
e7d8842e | 189 | local_irq_save(flags); |
25502a6c | 190 | rcu_preempt_qs(cpu); |
e7d8842e | 191 | local_irq_restore(flags); |
f41d911f PM |
192 | } |
193 | ||
194 | /* | |
195 | * Tree-preemptable RCU implementation for rcu_read_lock(). | |
196 | * Just increment ->rcu_read_lock_nesting, shared state will be updated | |
197 | * if we block. | |
198 | */ | |
199 | void __rcu_read_lock(void) | |
200 | { | |
80dcf60e | 201 | current->rcu_read_lock_nesting++; |
f41d911f PM |
202 | barrier(); /* needed if we ever invoke rcu_read_lock in rcutree.c */ |
203 | } | |
204 | EXPORT_SYMBOL_GPL(__rcu_read_lock); | |
205 | ||
fc2219d4 PM |
206 | /* |
207 | * Check for preempted RCU readers blocking the current grace period | |
208 | * for the specified rcu_node structure. If the caller needs a reliable | |
209 | * answer, it must hold the rcu_node's ->lock. | |
210 | */ | |
211 | static int rcu_preempted_readers(struct rcu_node *rnp) | |
212 | { | |
d9a3da06 PM |
213 | int phase = rnp->gpnum & 0x1; |
214 | ||
215 | return !list_empty(&rnp->blocked_tasks[phase]) || | |
216 | !list_empty(&rnp->blocked_tasks[phase + 2]); | |
fc2219d4 PM |
217 | } |
218 | ||
b668c9cf PM |
219 | /* |
220 | * Record a quiescent state for all tasks that were previously queued | |
221 | * on the specified rcu_node structure and that were blocking the current | |
222 | * RCU grace period. The caller must hold the specified rnp->lock with | |
223 | * irqs disabled, and this lock is released upon return, but irqs remain | |
224 | * disabled. | |
225 | */ | |
d3f6bad3 | 226 | static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags) |
b668c9cf PM |
227 | __releases(rnp->lock) |
228 | { | |
229 | unsigned long mask; | |
230 | struct rcu_node *rnp_p; | |
231 | ||
232 | if (rnp->qsmask != 0 || rcu_preempted_readers(rnp)) { | |
1304afb2 | 233 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
b668c9cf PM |
234 | return; /* Still need more quiescent states! */ |
235 | } | |
236 | ||
237 | rnp_p = rnp->parent; | |
238 | if (rnp_p == NULL) { | |
239 | /* | |
240 | * Either there is only one rcu_node in the tree, | |
241 | * or tasks were kicked up to root rcu_node due to | |
242 | * CPUs going offline. | |
243 | */ | |
d3f6bad3 | 244 | rcu_report_qs_rsp(&rcu_preempt_state, flags); |
b668c9cf PM |
245 | return; |
246 | } | |
247 | ||
248 | /* Report up the rest of the hierarchy. */ | |
249 | mask = rnp->grpmask; | |
1304afb2 PM |
250 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
251 | raw_spin_lock(&rnp_p->lock); /* irqs already disabled. */ | |
d3f6bad3 | 252 | rcu_report_qs_rnp(mask, &rcu_preempt_state, rnp_p, flags); |
b668c9cf PM |
253 | } |
254 | ||
255 | /* | |
256 | * Handle special cases during rcu_read_unlock(), such as needing to | |
257 | * notify RCU core processing or task having blocked during the RCU | |
258 | * read-side critical section. | |
259 | */ | |
f41d911f PM |
260 | static void rcu_read_unlock_special(struct task_struct *t) |
261 | { | |
262 | int empty; | |
d9a3da06 | 263 | int empty_exp; |
f41d911f | 264 | unsigned long flags; |
f41d911f PM |
265 | struct rcu_node *rnp; |
266 | int special; | |
267 | ||
268 | /* NMI handlers cannot block and cannot safely manipulate state. */ | |
269 | if (in_nmi()) | |
270 | return; | |
271 | ||
272 | local_irq_save(flags); | |
273 | ||
274 | /* | |
275 | * If RCU core is waiting for this CPU to exit critical section, | |
276 | * let it know that we have done so. | |
277 | */ | |
278 | special = t->rcu_read_unlock_special; | |
279 | if (special & RCU_READ_UNLOCK_NEED_QS) { | |
c3422bea | 280 | rcu_preempt_qs(smp_processor_id()); |
f41d911f PM |
281 | } |
282 | ||
283 | /* Hardware IRQ handlers cannot block. */ | |
284 | if (in_irq()) { | |
285 | local_irq_restore(flags); | |
286 | return; | |
287 | } | |
288 | ||
289 | /* Clean up if blocked during RCU read-side critical section. */ | |
290 | if (special & RCU_READ_UNLOCK_BLOCKED) { | |
291 | t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_BLOCKED; | |
292 | ||
dd5d19ba PM |
293 | /* |
294 | * Remove this task from the list it blocked on. The | |
295 | * task can migrate while we acquire the lock, but at | |
296 | * most one time. So at most two passes through loop. | |
297 | */ | |
298 | for (;;) { | |
86848966 | 299 | rnp = t->rcu_blocked_node; |
1304afb2 | 300 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
86848966 | 301 | if (rnp == t->rcu_blocked_node) |
dd5d19ba | 302 | break; |
1304afb2 | 303 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
dd5d19ba | 304 | } |
fc2219d4 | 305 | empty = !rcu_preempted_readers(rnp); |
d9a3da06 PM |
306 | empty_exp = !rcu_preempted_readers_exp(rnp); |
307 | smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */ | |
f41d911f | 308 | list_del_init(&t->rcu_node_entry); |
dd5d19ba | 309 | t->rcu_blocked_node = NULL; |
f41d911f PM |
310 | |
311 | /* | |
312 | * If this was the last task on the current list, and if | |
313 | * we aren't waiting on any CPUs, report the quiescent state. | |
d3f6bad3 | 314 | * Note that rcu_report_unblock_qs_rnp() releases rnp->lock. |
f41d911f | 315 | */ |
b668c9cf | 316 | if (empty) |
1304afb2 | 317 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
b668c9cf | 318 | else |
d3f6bad3 | 319 | rcu_report_unblock_qs_rnp(rnp, flags); |
d9a3da06 PM |
320 | |
321 | /* | |
322 | * If this was the last task on the expedited lists, | |
323 | * then we need to report up the rcu_node hierarchy. | |
324 | */ | |
325 | if (!empty_exp && !rcu_preempted_readers_exp(rnp)) | |
326 | rcu_report_exp_rnp(&rcu_preempt_state, rnp); | |
b668c9cf PM |
327 | } else { |
328 | local_irq_restore(flags); | |
f41d911f | 329 | } |
f41d911f PM |
330 | } |
331 | ||
332 | /* | |
333 | * Tree-preemptable RCU implementation for rcu_read_unlock(). | |
334 | * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost | |
335 | * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then | |
336 | * invoke rcu_read_unlock_special() to clean up after a context switch | |
337 | * in an RCU read-side critical section and other special cases. | |
338 | */ | |
339 | void __rcu_read_unlock(void) | |
340 | { | |
341 | struct task_struct *t = current; | |
342 | ||
343 | barrier(); /* needed if we ever invoke rcu_read_unlock in rcutree.c */ | |
80dcf60e PM |
344 | --t->rcu_read_lock_nesting; |
345 | barrier(); /* decrement before load of ->rcu_read_unlock_special */ | |
346 | if (t->rcu_read_lock_nesting == 0 && | |
f41d911f PM |
347 | unlikely(ACCESS_ONCE(t->rcu_read_unlock_special))) |
348 | rcu_read_unlock_special(t); | |
cba8244a PM |
349 | #ifdef CONFIG_PROVE_LOCKING |
350 | WARN_ON_ONCE(ACCESS_ONCE(t->rcu_read_lock_nesting) < 0); | |
351 | #endif /* #ifdef CONFIG_PROVE_LOCKING */ | |
f41d911f PM |
352 | } |
353 | EXPORT_SYMBOL_GPL(__rcu_read_unlock); | |
354 | ||
1ed509a2 PM |
355 | #ifdef CONFIG_RCU_CPU_STALL_VERBOSE |
356 | ||
357 | /* | |
358 | * Dump detailed information for all tasks blocking the current RCU | |
359 | * grace period on the specified rcu_node structure. | |
360 | */ | |
361 | static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp) | |
362 | { | |
363 | unsigned long flags; | |
364 | struct list_head *lp; | |
365 | int phase; | |
366 | struct task_struct *t; | |
367 | ||
368 | if (rcu_preempted_readers(rnp)) { | |
369 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
370 | phase = rnp->gpnum & 0x1; | |
371 | lp = &rnp->blocked_tasks[phase]; | |
372 | list_for_each_entry(t, lp, rcu_node_entry) | |
373 | sched_show_task(t); | |
374 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
375 | } | |
376 | } | |
377 | ||
378 | /* | |
379 | * Dump detailed information for all tasks blocking the current RCU | |
380 | * grace period. | |
381 | */ | |
382 | static void rcu_print_detail_task_stall(struct rcu_state *rsp) | |
383 | { | |
384 | struct rcu_node *rnp = rcu_get_root(rsp); | |
385 | ||
386 | rcu_print_detail_task_stall_rnp(rnp); | |
387 | rcu_for_each_leaf_node(rsp, rnp) | |
388 | rcu_print_detail_task_stall_rnp(rnp); | |
389 | } | |
390 | ||
391 | #else /* #ifdef CONFIG_RCU_CPU_STALL_VERBOSE */ | |
392 | ||
393 | static void rcu_print_detail_task_stall(struct rcu_state *rsp) | |
394 | { | |
395 | } | |
396 | ||
397 | #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_VERBOSE */ | |
398 | ||
f41d911f PM |
399 | /* |
400 | * Scan the current list of tasks blocked within RCU read-side critical | |
401 | * sections, printing out the tid of each. | |
402 | */ | |
403 | static void rcu_print_task_stall(struct rcu_node *rnp) | |
404 | { | |
f41d911f | 405 | struct list_head *lp; |
fc2219d4 | 406 | int phase; |
f41d911f PM |
407 | struct task_struct *t; |
408 | ||
fc2219d4 | 409 | if (rcu_preempted_readers(rnp)) { |
fc2219d4 | 410 | phase = rnp->gpnum & 0x1; |
f41d911f PM |
411 | lp = &rnp->blocked_tasks[phase]; |
412 | list_for_each_entry(t, lp, rcu_node_entry) | |
413 | printk(" P%d", t->pid); | |
f41d911f PM |
414 | } |
415 | } | |
416 | ||
53d84e00 PM |
417 | /* |
418 | * Suppress preemptible RCU's CPU stall warnings by pushing the | |
419 | * time of the next stall-warning message comfortably far into the | |
420 | * future. | |
421 | */ | |
422 | static void rcu_preempt_stall_reset(void) | |
423 | { | |
424 | rcu_preempt_state.jiffies_stall = jiffies + ULONG_MAX / 2; | |
425 | } | |
426 | ||
b0e165c0 PM |
427 | /* |
428 | * Check that the list of blocked tasks for the newly completed grace | |
429 | * period is in fact empty. It is a serious bug to complete a grace | |
430 | * period that still has RCU readers blocked! This function must be | |
431 | * invoked -before- updating this rnp's ->gpnum, and the rnp's ->lock | |
432 | * must be held by the caller. | |
433 | */ | |
434 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) | |
435 | { | |
fc2219d4 | 436 | WARN_ON_ONCE(rcu_preempted_readers(rnp)); |
28ecd580 | 437 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
438 | } |
439 | ||
33f76148 PM |
440 | #ifdef CONFIG_HOTPLUG_CPU |
441 | ||
dd5d19ba PM |
442 | /* |
443 | * Handle tasklist migration for case in which all CPUs covered by the | |
444 | * specified rcu_node have gone offline. Move them up to the root | |
445 | * rcu_node. The reason for not just moving them to the immediate | |
446 | * parent is to remove the need for rcu_read_unlock_special() to | |
447 | * make more than two attempts to acquire the target rcu_node's lock. | |
b668c9cf PM |
448 | * Returns true if there were tasks blocking the current RCU grace |
449 | * period. | |
dd5d19ba | 450 | * |
237c80c5 PM |
451 | * Returns 1 if there was previously a task blocking the current grace |
452 | * period on the specified rcu_node structure. | |
453 | * | |
dd5d19ba PM |
454 | * The caller must hold rnp->lock with irqs disabled. |
455 | */ | |
237c80c5 PM |
456 | static int rcu_preempt_offline_tasks(struct rcu_state *rsp, |
457 | struct rcu_node *rnp, | |
458 | struct rcu_data *rdp) | |
dd5d19ba PM |
459 | { |
460 | int i; | |
461 | struct list_head *lp; | |
462 | struct list_head *lp_root; | |
d9a3da06 | 463 | int retval = 0; |
dd5d19ba PM |
464 | struct rcu_node *rnp_root = rcu_get_root(rsp); |
465 | struct task_struct *tp; | |
466 | ||
86848966 PM |
467 | if (rnp == rnp_root) { |
468 | WARN_ONCE(1, "Last CPU thought to be offlined?"); | |
237c80c5 | 469 | return 0; /* Shouldn't happen: at least one CPU online. */ |
86848966 | 470 | } |
28ecd580 PM |
471 | WARN_ON_ONCE(rnp != rdp->mynode && |
472 | (!list_empty(&rnp->blocked_tasks[0]) || | |
d9a3da06 PM |
473 | !list_empty(&rnp->blocked_tasks[1]) || |
474 | !list_empty(&rnp->blocked_tasks[2]) || | |
475 | !list_empty(&rnp->blocked_tasks[3]))); | |
dd5d19ba PM |
476 | |
477 | /* | |
478 | * Move tasks up to root rcu_node. Rely on the fact that the | |
479 | * root rcu_node can be at most one ahead of the rest of the | |
480 | * rcu_nodes in terms of gp_num value. This fact allows us to | |
481 | * move the blocked_tasks[] array directly, element by element. | |
482 | */ | |
d9a3da06 PM |
483 | if (rcu_preempted_readers(rnp)) |
484 | retval |= RCU_OFL_TASKS_NORM_GP; | |
485 | if (rcu_preempted_readers_exp(rnp)) | |
486 | retval |= RCU_OFL_TASKS_EXP_GP; | |
487 | for (i = 0; i < 4; i++) { | |
dd5d19ba PM |
488 | lp = &rnp->blocked_tasks[i]; |
489 | lp_root = &rnp_root->blocked_tasks[i]; | |
490 | while (!list_empty(lp)) { | |
491 | tp = list_entry(lp->next, typeof(*tp), rcu_node_entry); | |
1304afb2 | 492 | raw_spin_lock(&rnp_root->lock); /* irqs already disabled */ |
dd5d19ba PM |
493 | list_del(&tp->rcu_node_entry); |
494 | tp->rcu_blocked_node = rnp_root; | |
495 | list_add(&tp->rcu_node_entry, lp_root); | |
1304afb2 | 496 | raw_spin_unlock(&rnp_root->lock); /* irqs remain disabled */ |
dd5d19ba PM |
497 | } |
498 | } | |
237c80c5 | 499 | return retval; |
dd5d19ba PM |
500 | } |
501 | ||
33f76148 PM |
502 | /* |
503 | * Do CPU-offline processing for preemptable RCU. | |
504 | */ | |
505 | static void rcu_preempt_offline_cpu(int cpu) | |
506 | { | |
507 | __rcu_offline_cpu(cpu, &rcu_preempt_state); | |
508 | } | |
509 | ||
510 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | |
511 | ||
f41d911f PM |
512 | /* |
513 | * Check for a quiescent state from the current CPU. When a task blocks, | |
514 | * the task is recorded in the corresponding CPU's rcu_node structure, | |
515 | * which is checked elsewhere. | |
516 | * | |
517 | * Caller must disable hard irqs. | |
518 | */ | |
519 | static void rcu_preempt_check_callbacks(int cpu) | |
520 | { | |
521 | struct task_struct *t = current; | |
522 | ||
523 | if (t->rcu_read_lock_nesting == 0) { | |
c3422bea | 524 | rcu_preempt_qs(cpu); |
f41d911f PM |
525 | return; |
526 | } | |
a71fca58 | 527 | if (per_cpu(rcu_preempt_data, cpu).qs_pending) |
c3422bea | 528 | t->rcu_read_unlock_special |= RCU_READ_UNLOCK_NEED_QS; |
f41d911f PM |
529 | } |
530 | ||
531 | /* | |
532 | * Process callbacks for preemptable RCU. | |
533 | */ | |
534 | static void rcu_preempt_process_callbacks(void) | |
535 | { | |
536 | __rcu_process_callbacks(&rcu_preempt_state, | |
537 | &__get_cpu_var(rcu_preempt_data)); | |
538 | } | |
539 | ||
540 | /* | |
541 | * Queue a preemptable-RCU callback for invocation after a grace period. | |
542 | */ | |
543 | void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
544 | { | |
545 | __call_rcu(head, func, &rcu_preempt_state); | |
546 | } | |
547 | EXPORT_SYMBOL_GPL(call_rcu); | |
548 | ||
6ebb237b PM |
549 | /** |
550 | * synchronize_rcu - wait until a grace period has elapsed. | |
551 | * | |
552 | * Control will return to the caller some time after a full grace | |
553 | * period has elapsed, in other words after all currently executing RCU | |
77d8485a PM |
554 | * read-side critical sections have completed. Note, however, that |
555 | * upon return from synchronize_rcu(), the caller might well be executing | |
556 | * concurrently with new RCU read-side critical sections that began while | |
557 | * synchronize_rcu() was waiting. RCU read-side critical sections are | |
558 | * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested. | |
6ebb237b PM |
559 | */ |
560 | void synchronize_rcu(void) | |
561 | { | |
562 | struct rcu_synchronize rcu; | |
563 | ||
564 | if (!rcu_scheduler_active) | |
565 | return; | |
566 | ||
72d5a9f7 | 567 | init_rcu_head_on_stack(&rcu.head); |
6ebb237b PM |
568 | init_completion(&rcu.completion); |
569 | /* Will wake me after RCU finished. */ | |
570 | call_rcu(&rcu.head, wakeme_after_rcu); | |
571 | /* Wait for it. */ | |
572 | wait_for_completion(&rcu.completion); | |
72d5a9f7 | 573 | destroy_rcu_head_on_stack(&rcu.head); |
6ebb237b PM |
574 | } |
575 | EXPORT_SYMBOL_GPL(synchronize_rcu); | |
576 | ||
d9a3da06 PM |
577 | static DECLARE_WAIT_QUEUE_HEAD(sync_rcu_preempt_exp_wq); |
578 | static long sync_rcu_preempt_exp_count; | |
579 | static DEFINE_MUTEX(sync_rcu_preempt_exp_mutex); | |
580 | ||
581 | /* | |
582 | * Return non-zero if there are any tasks in RCU read-side critical | |
583 | * sections blocking the current preemptible-RCU expedited grace period. | |
584 | * If there is no preemptible-RCU expedited grace period currently in | |
585 | * progress, returns zero unconditionally. | |
586 | */ | |
587 | static int rcu_preempted_readers_exp(struct rcu_node *rnp) | |
588 | { | |
589 | return !list_empty(&rnp->blocked_tasks[2]) || | |
590 | !list_empty(&rnp->blocked_tasks[3]); | |
591 | } | |
592 | ||
593 | /* | |
594 | * return non-zero if there is no RCU expedited grace period in progress | |
595 | * for the specified rcu_node structure, in other words, if all CPUs and | |
596 | * tasks covered by the specified rcu_node structure have done their bit | |
597 | * for the current expedited grace period. Works only for preemptible | |
598 | * RCU -- other RCU implementation use other means. | |
599 | * | |
600 | * Caller must hold sync_rcu_preempt_exp_mutex. | |
601 | */ | |
602 | static int sync_rcu_preempt_exp_done(struct rcu_node *rnp) | |
603 | { | |
604 | return !rcu_preempted_readers_exp(rnp) && | |
605 | ACCESS_ONCE(rnp->expmask) == 0; | |
606 | } | |
607 | ||
608 | /* | |
609 | * Report the exit from RCU read-side critical section for the last task | |
610 | * that queued itself during or before the current expedited preemptible-RCU | |
611 | * grace period. This event is reported either to the rcu_node structure on | |
612 | * which the task was queued or to one of that rcu_node structure's ancestors, | |
613 | * recursively up the tree. (Calm down, calm down, we do the recursion | |
614 | * iteratively!) | |
615 | * | |
616 | * Caller must hold sync_rcu_preempt_exp_mutex. | |
617 | */ | |
618 | static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp) | |
619 | { | |
620 | unsigned long flags; | |
621 | unsigned long mask; | |
622 | ||
1304afb2 | 623 | raw_spin_lock_irqsave(&rnp->lock, flags); |
d9a3da06 PM |
624 | for (;;) { |
625 | if (!sync_rcu_preempt_exp_done(rnp)) | |
626 | break; | |
627 | if (rnp->parent == NULL) { | |
628 | wake_up(&sync_rcu_preempt_exp_wq); | |
629 | break; | |
630 | } | |
631 | mask = rnp->grpmask; | |
1304afb2 | 632 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
d9a3da06 | 633 | rnp = rnp->parent; |
1304afb2 | 634 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
d9a3da06 PM |
635 | rnp->expmask &= ~mask; |
636 | } | |
1304afb2 | 637 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
d9a3da06 PM |
638 | } |
639 | ||
640 | /* | |
641 | * Snapshot the tasks blocking the newly started preemptible-RCU expedited | |
642 | * grace period for the specified rcu_node structure. If there are no such | |
643 | * tasks, report it up the rcu_node hierarchy. | |
644 | * | |
645 | * Caller must hold sync_rcu_preempt_exp_mutex and rsp->onofflock. | |
646 | */ | |
647 | static void | |
648 | sync_rcu_preempt_exp_init(struct rcu_state *rsp, struct rcu_node *rnp) | |
649 | { | |
650 | int must_wait; | |
651 | ||
1304afb2 | 652 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
d9a3da06 PM |
653 | list_splice_init(&rnp->blocked_tasks[0], &rnp->blocked_tasks[2]); |
654 | list_splice_init(&rnp->blocked_tasks[1], &rnp->blocked_tasks[3]); | |
655 | must_wait = rcu_preempted_readers_exp(rnp); | |
1304afb2 | 656 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
d9a3da06 PM |
657 | if (!must_wait) |
658 | rcu_report_exp_rnp(rsp, rnp); | |
659 | } | |
660 | ||
019129d5 | 661 | /* |
d9a3da06 PM |
662 | * Wait for an rcu-preempt grace period, but expedite it. The basic idea |
663 | * is to invoke synchronize_sched_expedited() to push all the tasks to | |
664 | * the ->blocked_tasks[] lists, move all entries from the first set of | |
665 | * ->blocked_tasks[] lists to the second set, and finally wait for this | |
666 | * second set to drain. | |
019129d5 PM |
667 | */ |
668 | void synchronize_rcu_expedited(void) | |
669 | { | |
d9a3da06 PM |
670 | unsigned long flags; |
671 | struct rcu_node *rnp; | |
672 | struct rcu_state *rsp = &rcu_preempt_state; | |
673 | long snap; | |
674 | int trycount = 0; | |
675 | ||
676 | smp_mb(); /* Caller's modifications seen first by other CPUs. */ | |
677 | snap = ACCESS_ONCE(sync_rcu_preempt_exp_count) + 1; | |
678 | smp_mb(); /* Above access cannot bleed into critical section. */ | |
679 | ||
680 | /* | |
681 | * Acquire lock, falling back to synchronize_rcu() if too many | |
682 | * lock-acquisition failures. Of course, if someone does the | |
683 | * expedited grace period for us, just leave. | |
684 | */ | |
685 | while (!mutex_trylock(&sync_rcu_preempt_exp_mutex)) { | |
686 | if (trycount++ < 10) | |
687 | udelay(trycount * num_online_cpus()); | |
688 | else { | |
689 | synchronize_rcu(); | |
690 | return; | |
691 | } | |
692 | if ((ACCESS_ONCE(sync_rcu_preempt_exp_count) - snap) > 0) | |
693 | goto mb_ret; /* Others did our work for us. */ | |
694 | } | |
695 | if ((ACCESS_ONCE(sync_rcu_preempt_exp_count) - snap) > 0) | |
696 | goto unlock_mb_ret; /* Others did our work for us. */ | |
697 | ||
698 | /* force all RCU readers onto blocked_tasks[]. */ | |
699 | synchronize_sched_expedited(); | |
700 | ||
1304afb2 | 701 | raw_spin_lock_irqsave(&rsp->onofflock, flags); |
d9a3da06 PM |
702 | |
703 | /* Initialize ->expmask for all non-leaf rcu_node structures. */ | |
704 | rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) { | |
1304afb2 | 705 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
d9a3da06 | 706 | rnp->expmask = rnp->qsmaskinit; |
1304afb2 | 707 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
d9a3da06 PM |
708 | } |
709 | ||
710 | /* Snapshot current state of ->blocked_tasks[] lists. */ | |
711 | rcu_for_each_leaf_node(rsp, rnp) | |
712 | sync_rcu_preempt_exp_init(rsp, rnp); | |
713 | if (NUM_RCU_NODES > 1) | |
714 | sync_rcu_preempt_exp_init(rsp, rcu_get_root(rsp)); | |
715 | ||
1304afb2 | 716 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); |
d9a3da06 PM |
717 | |
718 | /* Wait for snapshotted ->blocked_tasks[] lists to drain. */ | |
719 | rnp = rcu_get_root(rsp); | |
720 | wait_event(sync_rcu_preempt_exp_wq, | |
721 | sync_rcu_preempt_exp_done(rnp)); | |
722 | ||
723 | /* Clean up and exit. */ | |
724 | smp_mb(); /* ensure expedited GP seen before counter increment. */ | |
725 | ACCESS_ONCE(sync_rcu_preempt_exp_count)++; | |
726 | unlock_mb_ret: | |
727 | mutex_unlock(&sync_rcu_preempt_exp_mutex); | |
728 | mb_ret: | |
729 | smp_mb(); /* ensure subsequent action seen after grace period. */ | |
019129d5 PM |
730 | } |
731 | EXPORT_SYMBOL_GPL(synchronize_rcu_expedited); | |
732 | ||
f41d911f PM |
733 | /* |
734 | * Check to see if there is any immediate preemptable-RCU-related work | |
735 | * to be done. | |
736 | */ | |
737 | static int rcu_preempt_pending(int cpu) | |
738 | { | |
739 | return __rcu_pending(&rcu_preempt_state, | |
740 | &per_cpu(rcu_preempt_data, cpu)); | |
741 | } | |
742 | ||
743 | /* | |
744 | * Does preemptable RCU need the CPU to stay out of dynticks mode? | |
745 | */ | |
746 | static int rcu_preempt_needs_cpu(int cpu) | |
747 | { | |
748 | return !!per_cpu(rcu_preempt_data, cpu).nxtlist; | |
749 | } | |
750 | ||
e74f4c45 PM |
751 | /** |
752 | * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete. | |
753 | */ | |
754 | void rcu_barrier(void) | |
755 | { | |
756 | _rcu_barrier(&rcu_preempt_state, call_rcu); | |
757 | } | |
758 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
759 | ||
f41d911f PM |
760 | /* |
761 | * Initialize preemptable RCU's per-CPU data. | |
762 | */ | |
763 | static void __cpuinit rcu_preempt_init_percpu_data(int cpu) | |
764 | { | |
765 | rcu_init_percpu_data(cpu, &rcu_preempt_state, 1); | |
766 | } | |
767 | ||
e74f4c45 | 768 | /* |
2d999e03 | 769 | * Move preemptable RCU's callbacks from dying CPU to other online CPU. |
e74f4c45 | 770 | */ |
29494be7 | 771 | static void rcu_preempt_send_cbs_to_online(void) |
e74f4c45 | 772 | { |
29494be7 | 773 | rcu_send_cbs_to_online(&rcu_preempt_state); |
e74f4c45 PM |
774 | } |
775 | ||
1eba8f84 PM |
776 | /* |
777 | * Initialize preemptable RCU's state structures. | |
778 | */ | |
779 | static void __init __rcu_init_preempt(void) | |
780 | { | |
394f99a9 | 781 | rcu_init_one(&rcu_preempt_state, &rcu_preempt_data); |
1eba8f84 PM |
782 | } |
783 | ||
f41d911f PM |
784 | /* |
785 | * Check for a task exiting while in a preemptable-RCU read-side | |
786 | * critical section, clean up if so. No need to issue warnings, | |
787 | * as debug_check_no_locks_held() already does this if lockdep | |
788 | * is enabled. | |
789 | */ | |
790 | void exit_rcu(void) | |
791 | { | |
792 | struct task_struct *t = current; | |
793 | ||
794 | if (t->rcu_read_lock_nesting == 0) | |
795 | return; | |
796 | t->rcu_read_lock_nesting = 1; | |
797 | rcu_read_unlock(); | |
798 | } | |
799 | ||
800 | #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */ | |
801 | ||
802 | /* | |
803 | * Tell them what RCU they are running. | |
804 | */ | |
0e0fc1c2 | 805 | static void __init rcu_bootup_announce(void) |
f41d911f PM |
806 | { |
807 | printk(KERN_INFO "Hierarchical RCU implementation.\n"); | |
26845c28 | 808 | rcu_bootup_announce_oddness(); |
f41d911f PM |
809 | } |
810 | ||
811 | /* | |
812 | * Return the number of RCU batches processed thus far for debug & stats. | |
813 | */ | |
814 | long rcu_batches_completed(void) | |
815 | { | |
816 | return rcu_batches_completed_sched(); | |
817 | } | |
818 | EXPORT_SYMBOL_GPL(rcu_batches_completed); | |
819 | ||
bf66f18e PM |
820 | /* |
821 | * Force a quiescent state for RCU, which, because there is no preemptible | |
822 | * RCU, becomes the same as rcu-sched. | |
823 | */ | |
824 | void rcu_force_quiescent_state(void) | |
825 | { | |
826 | rcu_sched_force_quiescent_state(); | |
827 | } | |
828 | EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); | |
829 | ||
f41d911f PM |
830 | /* |
831 | * Because preemptable RCU does not exist, we never have to check for | |
832 | * CPUs being in quiescent states. | |
833 | */ | |
c3422bea | 834 | static void rcu_preempt_note_context_switch(int cpu) |
f41d911f PM |
835 | { |
836 | } | |
837 | ||
fc2219d4 PM |
838 | /* |
839 | * Because preemptable RCU does not exist, there are never any preempted | |
840 | * RCU readers. | |
841 | */ | |
842 | static int rcu_preempted_readers(struct rcu_node *rnp) | |
843 | { | |
844 | return 0; | |
845 | } | |
846 | ||
b668c9cf PM |
847 | #ifdef CONFIG_HOTPLUG_CPU |
848 | ||
849 | /* Because preemptible RCU does not exist, no quieting of tasks. */ | |
d3f6bad3 | 850 | static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags) |
b668c9cf | 851 | { |
1304afb2 | 852 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
b668c9cf PM |
853 | } |
854 | ||
855 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | |
856 | ||
1ed509a2 PM |
857 | /* |
858 | * Because preemptable RCU does not exist, we never have to check for | |
859 | * tasks blocked within RCU read-side critical sections. | |
860 | */ | |
861 | static void rcu_print_detail_task_stall(struct rcu_state *rsp) | |
862 | { | |
863 | } | |
864 | ||
f41d911f PM |
865 | /* |
866 | * Because preemptable RCU does not exist, we never have to check for | |
867 | * tasks blocked within RCU read-side critical sections. | |
868 | */ | |
869 | static void rcu_print_task_stall(struct rcu_node *rnp) | |
870 | { | |
871 | } | |
872 | ||
53d84e00 PM |
873 | /* |
874 | * Because preemptible RCU does not exist, there is no need to suppress | |
875 | * its CPU stall warnings. | |
876 | */ | |
877 | static void rcu_preempt_stall_reset(void) | |
878 | { | |
879 | } | |
880 | ||
b0e165c0 PM |
881 | /* |
882 | * Because there is no preemptable RCU, there can be no readers blocked, | |
49e29126 PM |
883 | * so there is no need to check for blocked tasks. So check only for |
884 | * bogus qsmask values. | |
b0e165c0 PM |
885 | */ |
886 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) | |
887 | { | |
49e29126 | 888 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
889 | } |
890 | ||
33f76148 PM |
891 | #ifdef CONFIG_HOTPLUG_CPU |
892 | ||
dd5d19ba PM |
893 | /* |
894 | * Because preemptable RCU does not exist, it never needs to migrate | |
237c80c5 PM |
895 | * tasks that were blocked within RCU read-side critical sections, and |
896 | * such non-existent tasks cannot possibly have been blocking the current | |
897 | * grace period. | |
dd5d19ba | 898 | */ |
237c80c5 PM |
899 | static int rcu_preempt_offline_tasks(struct rcu_state *rsp, |
900 | struct rcu_node *rnp, | |
901 | struct rcu_data *rdp) | |
dd5d19ba | 902 | { |
237c80c5 | 903 | return 0; |
dd5d19ba PM |
904 | } |
905 | ||
33f76148 PM |
906 | /* |
907 | * Because preemptable RCU does not exist, it never needs CPU-offline | |
908 | * processing. | |
909 | */ | |
910 | static void rcu_preempt_offline_cpu(int cpu) | |
911 | { | |
912 | } | |
913 | ||
914 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | |
915 | ||
f41d911f PM |
916 | /* |
917 | * Because preemptable RCU does not exist, it never has any callbacks | |
918 | * to check. | |
919 | */ | |
1eba8f84 | 920 | static void rcu_preempt_check_callbacks(int cpu) |
f41d911f PM |
921 | { |
922 | } | |
923 | ||
924 | /* | |
925 | * Because preemptable RCU does not exist, it never has any callbacks | |
926 | * to process. | |
927 | */ | |
1eba8f84 | 928 | static void rcu_preempt_process_callbacks(void) |
f41d911f PM |
929 | { |
930 | } | |
931 | ||
019129d5 PM |
932 | /* |
933 | * Wait for an rcu-preempt grace period, but make it happen quickly. | |
934 | * But because preemptable RCU does not exist, map to rcu-sched. | |
935 | */ | |
936 | void synchronize_rcu_expedited(void) | |
937 | { | |
938 | synchronize_sched_expedited(); | |
939 | } | |
940 | EXPORT_SYMBOL_GPL(synchronize_rcu_expedited); | |
941 | ||
d9a3da06 PM |
942 | #ifdef CONFIG_HOTPLUG_CPU |
943 | ||
944 | /* | |
945 | * Because preemptable RCU does not exist, there is never any need to | |
946 | * report on tasks preempted in RCU read-side critical sections during | |
947 | * expedited RCU grace periods. | |
948 | */ | |
949 | static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp) | |
950 | { | |
951 | return; | |
952 | } | |
953 | ||
954 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | |
955 | ||
f41d911f PM |
956 | /* |
957 | * Because preemptable RCU does not exist, it never has any work to do. | |
958 | */ | |
959 | static int rcu_preempt_pending(int cpu) | |
960 | { | |
961 | return 0; | |
962 | } | |
963 | ||
964 | /* | |
965 | * Because preemptable RCU does not exist, it never needs any CPU. | |
966 | */ | |
967 | static int rcu_preempt_needs_cpu(int cpu) | |
968 | { | |
969 | return 0; | |
970 | } | |
971 | ||
e74f4c45 PM |
972 | /* |
973 | * Because preemptable RCU does not exist, rcu_barrier() is just | |
974 | * another name for rcu_barrier_sched(). | |
975 | */ | |
976 | void rcu_barrier(void) | |
977 | { | |
978 | rcu_barrier_sched(); | |
979 | } | |
980 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
981 | ||
f41d911f PM |
982 | /* |
983 | * Because preemptable RCU does not exist, there is no per-CPU | |
984 | * data to initialize. | |
985 | */ | |
986 | static void __cpuinit rcu_preempt_init_percpu_data(int cpu) | |
987 | { | |
988 | } | |
989 | ||
e74f4c45 PM |
990 | /* |
991 | * Because there is no preemptable RCU, there are no callbacks to move. | |
992 | */ | |
29494be7 | 993 | static void rcu_preempt_send_cbs_to_online(void) |
e74f4c45 PM |
994 | { |
995 | } | |
996 | ||
1eba8f84 PM |
997 | /* |
998 | * Because preemptable RCU does not exist, it need not be initialized. | |
999 | */ | |
1000 | static void __init __rcu_init_preempt(void) | |
1001 | { | |
1002 | } | |
1003 | ||
f41d911f | 1004 | #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */ |
8bd93a2c | 1005 | |
7b27d547 LJ |
1006 | #ifndef CONFIG_SMP |
1007 | ||
1008 | void synchronize_sched_expedited(void) | |
1009 | { | |
1010 | cond_resched(); | |
1011 | } | |
1012 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
1013 | ||
1014 | #else /* #ifndef CONFIG_SMP */ | |
1015 | ||
e27fc964 TH |
1016 | static atomic_t sync_sched_expedited_started = ATOMIC_INIT(0); |
1017 | static atomic_t sync_sched_expedited_done = ATOMIC_INIT(0); | |
7b27d547 LJ |
1018 | |
1019 | static int synchronize_sched_expedited_cpu_stop(void *data) | |
1020 | { | |
1021 | /* | |
1022 | * There must be a full memory barrier on each affected CPU | |
1023 | * between the time that try_stop_cpus() is called and the | |
1024 | * time that it returns. | |
1025 | * | |
1026 | * In the current initial implementation of cpu_stop, the | |
1027 | * above condition is already met when the control reaches | |
1028 | * this point and the following smp_mb() is not strictly | |
1029 | * necessary. Do smp_mb() anyway for documentation and | |
1030 | * robustness against future implementation changes. | |
1031 | */ | |
1032 | smp_mb(); /* See above comment block. */ | |
1033 | return 0; | |
1034 | } | |
1035 | ||
1036 | /* | |
1037 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
1038 | * approach to force grace period to end quickly. This consumes | |
1039 | * significant time on all CPUs, and is thus not recommended for | |
1040 | * any sort of common-case code. | |
1041 | * | |
1042 | * Note that it is illegal to call this function while holding any | |
1043 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
1044 | * observe this restriction will result in deadlock. | |
db3a8920 | 1045 | * |
e27fc964 TH |
1046 | * This implementation can be thought of as an application of ticket |
1047 | * locking to RCU, with sync_sched_expedited_started and | |
1048 | * sync_sched_expedited_done taking on the roles of the halves | |
1049 | * of the ticket-lock word. Each task atomically increments | |
1050 | * sync_sched_expedited_started upon entry, snapshotting the old value, | |
1051 | * then attempts to stop all the CPUs. If this succeeds, then each | |
1052 | * CPU will have executed a context switch, resulting in an RCU-sched | |
1053 | * grace period. We are then done, so we use atomic_cmpxchg() to | |
1054 | * update sync_sched_expedited_done to match our snapshot -- but | |
1055 | * only if someone else has not already advanced past our snapshot. | |
1056 | * | |
1057 | * On the other hand, if try_stop_cpus() fails, we check the value | |
1058 | * of sync_sched_expedited_done. If it has advanced past our | |
1059 | * initial snapshot, then someone else must have forced a grace period | |
1060 | * some time after we took our snapshot. In this case, our work is | |
1061 | * done for us, and we can simply return. Otherwise, we try again, | |
1062 | * but keep our initial snapshot for purposes of checking for someone | |
1063 | * doing our work for us. | |
1064 | * | |
1065 | * If we fail too many times in a row, we fall back to synchronize_sched(). | |
7b27d547 LJ |
1066 | */ |
1067 | void synchronize_sched_expedited(void) | |
1068 | { | |
e27fc964 | 1069 | int firstsnap, s, snap, trycount = 0; |
7b27d547 | 1070 | |
e27fc964 TH |
1071 | /* Note that atomic_inc_return() implies full memory barrier. */ |
1072 | firstsnap = snap = atomic_inc_return(&sync_sched_expedited_started); | |
7b27d547 | 1073 | get_online_cpus(); |
e27fc964 TH |
1074 | |
1075 | /* | |
1076 | * Each pass through the following loop attempts to force a | |
1077 | * context switch on each CPU. | |
1078 | */ | |
7b27d547 LJ |
1079 | while (try_stop_cpus(cpu_online_mask, |
1080 | synchronize_sched_expedited_cpu_stop, | |
1081 | NULL) == -EAGAIN) { | |
1082 | put_online_cpus(); | |
e27fc964 TH |
1083 | |
1084 | /* No joy, try again later. Or just synchronize_sched(). */ | |
7b27d547 LJ |
1085 | if (trycount++ < 10) |
1086 | udelay(trycount * num_online_cpus()); | |
1087 | else { | |
1088 | synchronize_sched(); | |
1089 | return; | |
1090 | } | |
e27fc964 TH |
1091 | |
1092 | /* Check to see if someone else did our work for us. */ | |
1093 | s = atomic_read(&sync_sched_expedited_done); | |
1094 | if (UINT_CMP_GE((unsigned)s, (unsigned)firstsnap)) { | |
7b27d547 LJ |
1095 | smp_mb(); /* ensure test happens before caller kfree */ |
1096 | return; | |
1097 | } | |
e27fc964 TH |
1098 | |
1099 | /* | |
1100 | * Refetching sync_sched_expedited_started allows later | |
1101 | * callers to piggyback on our grace period. We subtract | |
1102 | * 1 to get the same token that the last incrementer got. | |
1103 | * We retry after they started, so our grace period works | |
1104 | * for them, and they started after our first try, so their | |
1105 | * grace period works for us. | |
1106 | */ | |
7b27d547 | 1107 | get_online_cpus(); |
e27fc964 TH |
1108 | snap = atomic_read(&sync_sched_expedited_started) - 1; |
1109 | smp_mb(); /* ensure read is before try_stop_cpus(). */ | |
7b27d547 | 1110 | } |
e27fc964 TH |
1111 | |
1112 | /* | |
1113 | * Everyone up to our most recent fetch is covered by our grace | |
1114 | * period. Update the counter, but only if our work is still | |
1115 | * relevant -- which it won't be if someone who started later | |
1116 | * than we did beat us to the punch. | |
1117 | */ | |
1118 | do { | |
1119 | s = atomic_read(&sync_sched_expedited_done); | |
1120 | if (UINT_CMP_GE((unsigned)s, (unsigned)snap)) { | |
1121 | smp_mb(); /* ensure test happens before caller kfree */ | |
1122 | break; | |
1123 | } | |
1124 | } while (atomic_cmpxchg(&sync_sched_expedited_done, s, snap) != s); | |
1125 | ||
7b27d547 LJ |
1126 | put_online_cpus(); |
1127 | } | |
1128 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
1129 | ||
1130 | #endif /* #else #ifndef CONFIG_SMP */ | |
1131 | ||
8bd93a2c PM |
1132 | #if !defined(CONFIG_RCU_FAST_NO_HZ) |
1133 | ||
1134 | /* | |
1135 | * Check to see if any future RCU-related work will need to be done | |
1136 | * by the current CPU, even if none need be done immediately, returning | |
1137 | * 1 if so. This function is part of the RCU implementation; it is -not- | |
1138 | * an exported member of the RCU API. | |
1139 | * | |
1140 | * Because we have preemptible RCU, just check whether this CPU needs | |
1141 | * any flavor of RCU. Do not chew up lots of CPU cycles with preemption | |
1142 | * disabled in a most-likely vain attempt to cause RCU not to need this CPU. | |
1143 | */ | |
1144 | int rcu_needs_cpu(int cpu) | |
1145 | { | |
1146 | return rcu_needs_cpu_quick_check(cpu); | |
1147 | } | |
1148 | ||
a47cd880 PM |
1149 | /* |
1150 | * Check to see if we need to continue a callback-flush operations to | |
1151 | * allow the last CPU to enter dyntick-idle mode. But fast dyntick-idle | |
1152 | * entry is not configured, so we never do need to. | |
1153 | */ | |
1154 | static void rcu_needs_cpu_flush(void) | |
1155 | { | |
1156 | } | |
1157 | ||
8bd93a2c PM |
1158 | #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
1159 | ||
1160 | #define RCU_NEEDS_CPU_FLUSHES 5 | |
a47cd880 | 1161 | static DEFINE_PER_CPU(int, rcu_dyntick_drain); |
71da8132 | 1162 | static DEFINE_PER_CPU(unsigned long, rcu_dyntick_holdoff); |
8bd93a2c PM |
1163 | |
1164 | /* | |
1165 | * Check to see if any future RCU-related work will need to be done | |
1166 | * by the current CPU, even if none need be done immediately, returning | |
1167 | * 1 if so. This function is part of the RCU implementation; it is -not- | |
1168 | * an exported member of the RCU API. | |
1169 | * | |
1170 | * Because we are not supporting preemptible RCU, attempt to accelerate | |
1171 | * any current grace periods so that RCU no longer needs this CPU, but | |
1172 | * only if all other CPUs are already in dynticks-idle mode. This will | |
1173 | * allow the CPU cores to be powered down immediately, as opposed to after | |
1174 | * waiting many milliseconds for grace periods to elapse. | |
a47cd880 PM |
1175 | * |
1176 | * Because it is not legal to invoke rcu_process_callbacks() with irqs | |
1177 | * disabled, we do one pass of force_quiescent_state(), then do a | |
1178 | * raise_softirq() to cause rcu_process_callbacks() to be invoked later. | |
1179 | * The per-cpu rcu_dyntick_drain variable controls the sequencing. | |
8bd93a2c PM |
1180 | */ |
1181 | int rcu_needs_cpu(int cpu) | |
1182 | { | |
a47cd880 | 1183 | int c = 0; |
77e38ed3 | 1184 | int snap; |
8bd93a2c PM |
1185 | int thatcpu; |
1186 | ||
622ea685 PM |
1187 | /* Check for being in the holdoff period. */ |
1188 | if (per_cpu(rcu_dyntick_holdoff, cpu) == jiffies) | |
1189 | return rcu_needs_cpu_quick_check(cpu); | |
1190 | ||
8bd93a2c | 1191 | /* Don't bother unless we are the last non-dyntick-idle CPU. */ |
77e38ed3 PM |
1192 | for_each_online_cpu(thatcpu) { |
1193 | if (thatcpu == cpu) | |
1194 | continue; | |
e59fb312 PM |
1195 | snap = atomic_add_return(0, &per_cpu(rcu_dynticks, |
1196 | thatcpu).dynticks); | |
77e38ed3 | 1197 | smp_mb(); /* Order sampling of snap with end of grace period. */ |
e59fb312 | 1198 | if ((snap & 0x1) != 0) { |
a47cd880 | 1199 | per_cpu(rcu_dyntick_drain, cpu) = 0; |
71da8132 | 1200 | per_cpu(rcu_dyntick_holdoff, cpu) = jiffies - 1; |
8bd93a2c | 1201 | return rcu_needs_cpu_quick_check(cpu); |
8bd93a2c | 1202 | } |
77e38ed3 | 1203 | } |
a47cd880 PM |
1204 | |
1205 | /* Check and update the rcu_dyntick_drain sequencing. */ | |
1206 | if (per_cpu(rcu_dyntick_drain, cpu) <= 0) { | |
1207 | /* First time through, initialize the counter. */ | |
1208 | per_cpu(rcu_dyntick_drain, cpu) = RCU_NEEDS_CPU_FLUSHES; | |
1209 | } else if (--per_cpu(rcu_dyntick_drain, cpu) <= 0) { | |
1210 | /* We have hit the limit, so time to give up. */ | |
71da8132 | 1211 | per_cpu(rcu_dyntick_holdoff, cpu) = jiffies; |
a47cd880 PM |
1212 | return rcu_needs_cpu_quick_check(cpu); |
1213 | } | |
1214 | ||
1215 | /* Do one step pushing remaining RCU callbacks through. */ | |
1216 | if (per_cpu(rcu_sched_data, cpu).nxtlist) { | |
1217 | rcu_sched_qs(cpu); | |
1218 | force_quiescent_state(&rcu_sched_state, 0); | |
1219 | c = c || per_cpu(rcu_sched_data, cpu).nxtlist; | |
1220 | } | |
1221 | if (per_cpu(rcu_bh_data, cpu).nxtlist) { | |
1222 | rcu_bh_qs(cpu); | |
1223 | force_quiescent_state(&rcu_bh_state, 0); | |
1224 | c = c || per_cpu(rcu_bh_data, cpu).nxtlist; | |
8bd93a2c PM |
1225 | } |
1226 | ||
1227 | /* If RCU callbacks are still pending, RCU still needs this CPU. */ | |
622ea685 | 1228 | if (c) |
a47cd880 | 1229 | raise_softirq(RCU_SOFTIRQ); |
8bd93a2c PM |
1230 | return c; |
1231 | } | |
1232 | ||
a47cd880 PM |
1233 | /* |
1234 | * Check to see if we need to continue a callback-flush operations to | |
1235 | * allow the last CPU to enter dyntick-idle mode. | |
1236 | */ | |
1237 | static void rcu_needs_cpu_flush(void) | |
1238 | { | |
1239 | int cpu = smp_processor_id(); | |
71da8132 | 1240 | unsigned long flags; |
a47cd880 PM |
1241 | |
1242 | if (per_cpu(rcu_dyntick_drain, cpu) <= 0) | |
1243 | return; | |
71da8132 | 1244 | local_irq_save(flags); |
a47cd880 | 1245 | (void)rcu_needs_cpu(cpu); |
71da8132 | 1246 | local_irq_restore(flags); |
a47cd880 PM |
1247 | } |
1248 | ||
8bd93a2c | 1249 | #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |