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 | |
6cc68793 | 4 | * or preemptible semantics. |
f41d911f PM |
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> |
62ab7072 | 28 | #include <linux/smpboot.h> |
f41d911f | 29 | |
5b61b0ba MG |
30 | #define RCU_KTHREAD_PRIO 1 |
31 | ||
32 | #ifdef CONFIG_RCU_BOOST | |
33 | #define RCU_BOOST_PRIO CONFIG_RCU_BOOST_PRIO | |
34 | #else | |
35 | #define RCU_BOOST_PRIO RCU_KTHREAD_PRIO | |
36 | #endif | |
37 | ||
26845c28 PM |
38 | /* |
39 | * Check the RCU kernel configuration parameters and print informative | |
40 | * messages about anything out of the ordinary. If you like #ifdef, you | |
41 | * will love this function. | |
42 | */ | |
43 | static void __init rcu_bootup_announce_oddness(void) | |
44 | { | |
45 | #ifdef CONFIG_RCU_TRACE | |
46 | printk(KERN_INFO "\tRCU debugfs-based tracing is enabled.\n"); | |
47 | #endif | |
48 | #if (defined(CONFIG_64BIT) && CONFIG_RCU_FANOUT != 64) || (!defined(CONFIG_64BIT) && CONFIG_RCU_FANOUT != 32) | |
49 | printk(KERN_INFO "\tCONFIG_RCU_FANOUT set to non-default value of %d\n", | |
50 | CONFIG_RCU_FANOUT); | |
51 | #endif | |
52 | #ifdef CONFIG_RCU_FANOUT_EXACT | |
53 | printk(KERN_INFO "\tHierarchical RCU autobalancing is disabled.\n"); | |
54 | #endif | |
55 | #ifdef CONFIG_RCU_FAST_NO_HZ | |
56 | printk(KERN_INFO | |
57 | "\tRCU dyntick-idle grace-period acceleration is enabled.\n"); | |
58 | #endif | |
59 | #ifdef CONFIG_PROVE_RCU | |
60 | printk(KERN_INFO "\tRCU lockdep checking is enabled.\n"); | |
61 | #endif | |
62 | #ifdef CONFIG_RCU_TORTURE_TEST_RUNNABLE | |
63 | printk(KERN_INFO "\tRCU torture testing starts during boot.\n"); | |
64 | #endif | |
81a294c4 | 65 | #if defined(CONFIG_TREE_PREEMPT_RCU) && !defined(CONFIG_RCU_CPU_STALL_VERBOSE) |
a858af28 PM |
66 | printk(KERN_INFO "\tDump stacks of tasks blocking RCU-preempt GP.\n"); |
67 | #endif | |
68 | #if defined(CONFIG_RCU_CPU_STALL_INFO) | |
69 | printk(KERN_INFO "\tAdditional per-CPU info printed with stalls.\n"); | |
26845c28 PM |
70 | #endif |
71 | #if NUM_RCU_LVL_4 != 0 | |
cc5df65b | 72 | printk(KERN_INFO "\tFour-level hierarchy is enabled.\n"); |
26845c28 | 73 | #endif |
f885b7f2 PM |
74 | if (rcu_fanout_leaf != CONFIG_RCU_FANOUT_LEAF) |
75 | printk(KERN_INFO "\tExperimental boot-time adjustment of leaf fanout to %d.\n", rcu_fanout_leaf); | |
cca6f393 PM |
76 | if (nr_cpu_ids != NR_CPUS) |
77 | printk(KERN_INFO "\tRCU restricting CPUs from NR_CPUS=%d to nr_cpu_ids=%d.\n", NR_CPUS, nr_cpu_ids); | |
26845c28 PM |
78 | } |
79 | ||
f41d911f PM |
80 | #ifdef CONFIG_TREE_PREEMPT_RCU |
81 | ||
037b64ed PM |
82 | struct rcu_state rcu_preempt_state = |
83 | RCU_STATE_INITIALIZER(rcu_preempt, call_rcu); | |
f41d911f | 84 | DEFINE_PER_CPU(struct rcu_data, rcu_preempt_data); |
27f4d280 | 85 | static struct rcu_state *rcu_state = &rcu_preempt_state; |
f41d911f | 86 | |
d9a3da06 PM |
87 | static int rcu_preempted_readers_exp(struct rcu_node *rnp); |
88 | ||
f41d911f PM |
89 | /* |
90 | * Tell them what RCU they are running. | |
91 | */ | |
0e0fc1c2 | 92 | static void __init rcu_bootup_announce(void) |
f41d911f | 93 | { |
6cc68793 | 94 | printk(KERN_INFO "Preemptible hierarchical RCU implementation.\n"); |
26845c28 | 95 | rcu_bootup_announce_oddness(); |
f41d911f PM |
96 | } |
97 | ||
98 | /* | |
99 | * Return the number of RCU-preempt batches processed thus far | |
100 | * for debug and statistics. | |
101 | */ | |
102 | long rcu_batches_completed_preempt(void) | |
103 | { | |
104 | return rcu_preempt_state.completed; | |
105 | } | |
106 | EXPORT_SYMBOL_GPL(rcu_batches_completed_preempt); | |
107 | ||
108 | /* | |
109 | * Return the number of RCU batches processed thus far for debug & stats. | |
110 | */ | |
111 | long rcu_batches_completed(void) | |
112 | { | |
113 | return rcu_batches_completed_preempt(); | |
114 | } | |
115 | EXPORT_SYMBOL_GPL(rcu_batches_completed); | |
116 | ||
bf66f18e PM |
117 | /* |
118 | * Force a quiescent state for preemptible RCU. | |
119 | */ | |
120 | void rcu_force_quiescent_state(void) | |
121 | { | |
122 | force_quiescent_state(&rcu_preempt_state, 0); | |
123 | } | |
124 | EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); | |
125 | ||
f41d911f | 126 | /* |
6cc68793 | 127 | * Record a preemptible-RCU quiescent state for the specified CPU. Note |
f41d911f PM |
128 | * that this just means that the task currently running on the CPU is |
129 | * not in a quiescent state. There might be any number of tasks blocked | |
130 | * while in an RCU read-side critical section. | |
25502a6c PM |
131 | * |
132 | * Unlike the other rcu_*_qs() functions, callers to this function | |
133 | * must disable irqs in order to protect the assignment to | |
134 | * ->rcu_read_unlock_special. | |
f41d911f | 135 | */ |
c3422bea | 136 | static void rcu_preempt_qs(int cpu) |
f41d911f PM |
137 | { |
138 | struct rcu_data *rdp = &per_cpu(rcu_preempt_data, cpu); | |
25502a6c | 139 | |
e4cc1f22 | 140 | rdp->passed_quiesce_gpnum = rdp->gpnum; |
c3422bea | 141 | barrier(); |
e4cc1f22 | 142 | if (rdp->passed_quiesce == 0) |
d4c08f2a | 143 | trace_rcu_grace_period("rcu_preempt", rdp->gpnum, "cpuqs"); |
e4cc1f22 | 144 | rdp->passed_quiesce = 1; |
25502a6c | 145 | current->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS; |
f41d911f PM |
146 | } |
147 | ||
148 | /* | |
c3422bea PM |
149 | * We have entered the scheduler, and the current task might soon be |
150 | * context-switched away from. If this task is in an RCU read-side | |
151 | * critical section, we will no longer be able to rely on the CPU to | |
12f5f524 PM |
152 | * record that fact, so we enqueue the task on the blkd_tasks list. |
153 | * The task will dequeue itself when it exits the outermost enclosing | |
154 | * RCU read-side critical section. Therefore, the current grace period | |
155 | * cannot be permitted to complete until the blkd_tasks list entries | |
156 | * predating the current grace period drain, in other words, until | |
157 | * rnp->gp_tasks becomes NULL. | |
c3422bea PM |
158 | * |
159 | * Caller must disable preemption. | |
f41d911f | 160 | */ |
cba6d0d6 | 161 | static void rcu_preempt_note_context_switch(int cpu) |
f41d911f PM |
162 | { |
163 | struct task_struct *t = current; | |
c3422bea | 164 | unsigned long flags; |
f41d911f PM |
165 | struct rcu_data *rdp; |
166 | struct rcu_node *rnp; | |
167 | ||
10f39bb1 | 168 | if (t->rcu_read_lock_nesting > 0 && |
f41d911f PM |
169 | (t->rcu_read_unlock_special & RCU_READ_UNLOCK_BLOCKED) == 0) { |
170 | ||
171 | /* Possibly blocking in an RCU read-side critical section. */ | |
cba6d0d6 | 172 | rdp = per_cpu_ptr(rcu_preempt_state.rda, cpu); |
f41d911f | 173 | rnp = rdp->mynode; |
1304afb2 | 174 | raw_spin_lock_irqsave(&rnp->lock, flags); |
f41d911f | 175 | t->rcu_read_unlock_special |= RCU_READ_UNLOCK_BLOCKED; |
86848966 | 176 | t->rcu_blocked_node = rnp; |
f41d911f PM |
177 | |
178 | /* | |
179 | * If this CPU has already checked in, then this task | |
180 | * will hold up the next grace period rather than the | |
181 | * current grace period. Queue the task accordingly. | |
182 | * If the task is queued for the current grace period | |
183 | * (i.e., this CPU has not yet passed through a quiescent | |
184 | * state for the current grace period), then as long | |
185 | * as that task remains queued, the current grace period | |
12f5f524 PM |
186 | * cannot end. Note that there is some uncertainty as |
187 | * to exactly when the current grace period started. | |
188 | * We take a conservative approach, which can result | |
189 | * in unnecessarily waiting on tasks that started very | |
190 | * slightly after the current grace period began. C'est | |
191 | * la vie!!! | |
b0e165c0 PM |
192 | * |
193 | * But first, note that the current CPU must still be | |
194 | * on line! | |
f41d911f | 195 | */ |
b0e165c0 | 196 | WARN_ON_ONCE((rdp->grpmask & rnp->qsmaskinit) == 0); |
e7d8842e | 197 | WARN_ON_ONCE(!list_empty(&t->rcu_node_entry)); |
12f5f524 PM |
198 | if ((rnp->qsmask & rdp->grpmask) && rnp->gp_tasks != NULL) { |
199 | list_add(&t->rcu_node_entry, rnp->gp_tasks->prev); | |
200 | rnp->gp_tasks = &t->rcu_node_entry; | |
27f4d280 PM |
201 | #ifdef CONFIG_RCU_BOOST |
202 | if (rnp->boost_tasks != NULL) | |
203 | rnp->boost_tasks = rnp->gp_tasks; | |
204 | #endif /* #ifdef CONFIG_RCU_BOOST */ | |
12f5f524 PM |
205 | } else { |
206 | list_add(&t->rcu_node_entry, &rnp->blkd_tasks); | |
207 | if (rnp->qsmask & rdp->grpmask) | |
208 | rnp->gp_tasks = &t->rcu_node_entry; | |
209 | } | |
d4c08f2a PM |
210 | trace_rcu_preempt_task(rdp->rsp->name, |
211 | t->pid, | |
212 | (rnp->qsmask & rdp->grpmask) | |
213 | ? rnp->gpnum | |
214 | : rnp->gpnum + 1); | |
1304afb2 | 215 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
10f39bb1 PM |
216 | } else if (t->rcu_read_lock_nesting < 0 && |
217 | t->rcu_read_unlock_special) { | |
218 | ||
219 | /* | |
220 | * Complete exit from RCU read-side critical section on | |
221 | * behalf of preempted instance of __rcu_read_unlock(). | |
222 | */ | |
223 | rcu_read_unlock_special(t); | |
f41d911f PM |
224 | } |
225 | ||
226 | /* | |
227 | * Either we were not in an RCU read-side critical section to | |
228 | * begin with, or we have now recorded that critical section | |
229 | * globally. Either way, we can now note a quiescent state | |
230 | * for this CPU. Again, if we were in an RCU read-side critical | |
231 | * section, and if that critical section was blocking the current | |
232 | * grace period, then the fact that the task has been enqueued | |
233 | * means that we continue to block the current grace period. | |
234 | */ | |
e7d8842e | 235 | local_irq_save(flags); |
cba6d0d6 | 236 | rcu_preempt_qs(cpu); |
e7d8842e | 237 | local_irq_restore(flags); |
f41d911f PM |
238 | } |
239 | ||
fc2219d4 PM |
240 | /* |
241 | * Check for preempted RCU readers blocking the current grace period | |
242 | * for the specified rcu_node structure. If the caller needs a reliable | |
243 | * answer, it must hold the rcu_node's ->lock. | |
244 | */ | |
27f4d280 | 245 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 | 246 | { |
12f5f524 | 247 | return rnp->gp_tasks != NULL; |
fc2219d4 PM |
248 | } |
249 | ||
b668c9cf PM |
250 | /* |
251 | * Record a quiescent state for all tasks that were previously queued | |
252 | * on the specified rcu_node structure and that were blocking the current | |
253 | * RCU grace period. The caller must hold the specified rnp->lock with | |
254 | * irqs disabled, and this lock is released upon return, but irqs remain | |
255 | * disabled. | |
256 | */ | |
d3f6bad3 | 257 | static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags) |
b668c9cf PM |
258 | __releases(rnp->lock) |
259 | { | |
260 | unsigned long mask; | |
261 | struct rcu_node *rnp_p; | |
262 | ||
27f4d280 | 263 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
1304afb2 | 264 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
b668c9cf PM |
265 | return; /* Still need more quiescent states! */ |
266 | } | |
267 | ||
268 | rnp_p = rnp->parent; | |
269 | if (rnp_p == NULL) { | |
270 | /* | |
271 | * Either there is only one rcu_node in the tree, | |
272 | * or tasks were kicked up to root rcu_node due to | |
273 | * CPUs going offline. | |
274 | */ | |
d3f6bad3 | 275 | rcu_report_qs_rsp(&rcu_preempt_state, flags); |
b668c9cf PM |
276 | return; |
277 | } | |
278 | ||
279 | /* Report up the rest of the hierarchy. */ | |
280 | mask = rnp->grpmask; | |
1304afb2 PM |
281 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
282 | raw_spin_lock(&rnp_p->lock); /* irqs already disabled. */ | |
d3f6bad3 | 283 | rcu_report_qs_rnp(mask, &rcu_preempt_state, rnp_p, flags); |
b668c9cf PM |
284 | } |
285 | ||
12f5f524 PM |
286 | /* |
287 | * Advance a ->blkd_tasks-list pointer to the next entry, instead | |
288 | * returning NULL if at the end of the list. | |
289 | */ | |
290 | static struct list_head *rcu_next_node_entry(struct task_struct *t, | |
291 | struct rcu_node *rnp) | |
292 | { | |
293 | struct list_head *np; | |
294 | ||
295 | np = t->rcu_node_entry.next; | |
296 | if (np == &rnp->blkd_tasks) | |
297 | np = NULL; | |
298 | return np; | |
299 | } | |
300 | ||
b668c9cf PM |
301 | /* |
302 | * Handle special cases during rcu_read_unlock(), such as needing to | |
303 | * notify RCU core processing or task having blocked during the RCU | |
304 | * read-side critical section. | |
305 | */ | |
2a3fa843 | 306 | void rcu_read_unlock_special(struct task_struct *t) |
f41d911f PM |
307 | { |
308 | int empty; | |
d9a3da06 | 309 | int empty_exp; |
389abd48 | 310 | int empty_exp_now; |
f41d911f | 311 | unsigned long flags; |
12f5f524 | 312 | struct list_head *np; |
82e78d80 PM |
313 | #ifdef CONFIG_RCU_BOOST |
314 | struct rt_mutex *rbmp = NULL; | |
315 | #endif /* #ifdef CONFIG_RCU_BOOST */ | |
f41d911f PM |
316 | struct rcu_node *rnp; |
317 | int special; | |
318 | ||
319 | /* NMI handlers cannot block and cannot safely manipulate state. */ | |
320 | if (in_nmi()) | |
321 | return; | |
322 | ||
323 | local_irq_save(flags); | |
324 | ||
325 | /* | |
326 | * If RCU core is waiting for this CPU to exit critical section, | |
327 | * let it know that we have done so. | |
328 | */ | |
329 | special = t->rcu_read_unlock_special; | |
330 | if (special & RCU_READ_UNLOCK_NEED_QS) { | |
c3422bea | 331 | rcu_preempt_qs(smp_processor_id()); |
f41d911f PM |
332 | } |
333 | ||
334 | /* Hardware IRQ handlers cannot block. */ | |
ec433f0c | 335 | if (in_irq() || in_serving_softirq()) { |
f41d911f PM |
336 | local_irq_restore(flags); |
337 | return; | |
338 | } | |
339 | ||
340 | /* Clean up if blocked during RCU read-side critical section. */ | |
341 | if (special & RCU_READ_UNLOCK_BLOCKED) { | |
342 | t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_BLOCKED; | |
343 | ||
dd5d19ba PM |
344 | /* |
345 | * Remove this task from the list it blocked on. The | |
346 | * task can migrate while we acquire the lock, but at | |
347 | * most one time. So at most two passes through loop. | |
348 | */ | |
349 | for (;;) { | |
86848966 | 350 | rnp = t->rcu_blocked_node; |
1304afb2 | 351 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
86848966 | 352 | if (rnp == t->rcu_blocked_node) |
dd5d19ba | 353 | break; |
1304afb2 | 354 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
dd5d19ba | 355 | } |
27f4d280 | 356 | empty = !rcu_preempt_blocked_readers_cgp(rnp); |
d9a3da06 PM |
357 | empty_exp = !rcu_preempted_readers_exp(rnp); |
358 | smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */ | |
12f5f524 | 359 | np = rcu_next_node_entry(t, rnp); |
f41d911f | 360 | list_del_init(&t->rcu_node_entry); |
82e78d80 | 361 | t->rcu_blocked_node = NULL; |
d4c08f2a PM |
362 | trace_rcu_unlock_preempted_task("rcu_preempt", |
363 | rnp->gpnum, t->pid); | |
12f5f524 PM |
364 | if (&t->rcu_node_entry == rnp->gp_tasks) |
365 | rnp->gp_tasks = np; | |
366 | if (&t->rcu_node_entry == rnp->exp_tasks) | |
367 | rnp->exp_tasks = np; | |
27f4d280 PM |
368 | #ifdef CONFIG_RCU_BOOST |
369 | if (&t->rcu_node_entry == rnp->boost_tasks) | |
370 | rnp->boost_tasks = np; | |
82e78d80 PM |
371 | /* Snapshot/clear ->rcu_boost_mutex with rcu_node lock held. */ |
372 | if (t->rcu_boost_mutex) { | |
373 | rbmp = t->rcu_boost_mutex; | |
374 | t->rcu_boost_mutex = NULL; | |
7765be2f | 375 | } |
27f4d280 | 376 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
f41d911f PM |
377 | |
378 | /* | |
379 | * If this was the last task on the current list, and if | |
380 | * we aren't waiting on any CPUs, report the quiescent state. | |
389abd48 PM |
381 | * Note that rcu_report_unblock_qs_rnp() releases rnp->lock, |
382 | * so we must take a snapshot of the expedited state. | |
f41d911f | 383 | */ |
389abd48 | 384 | empty_exp_now = !rcu_preempted_readers_exp(rnp); |
d4c08f2a PM |
385 | if (!empty && !rcu_preempt_blocked_readers_cgp(rnp)) { |
386 | trace_rcu_quiescent_state_report("preempt_rcu", | |
387 | rnp->gpnum, | |
388 | 0, rnp->qsmask, | |
389 | rnp->level, | |
390 | rnp->grplo, | |
391 | rnp->grphi, | |
392 | !!rnp->gp_tasks); | |
d3f6bad3 | 393 | rcu_report_unblock_qs_rnp(rnp, flags); |
c701d5d9 | 394 | } else { |
d4c08f2a | 395 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
c701d5d9 | 396 | } |
d9a3da06 | 397 | |
27f4d280 PM |
398 | #ifdef CONFIG_RCU_BOOST |
399 | /* Unboost if we were boosted. */ | |
82e78d80 PM |
400 | if (rbmp) |
401 | rt_mutex_unlock(rbmp); | |
27f4d280 PM |
402 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
403 | ||
d9a3da06 PM |
404 | /* |
405 | * If this was the last task on the expedited lists, | |
406 | * then we need to report up the rcu_node hierarchy. | |
407 | */ | |
389abd48 | 408 | if (!empty_exp && empty_exp_now) |
b40d293e | 409 | rcu_report_exp_rnp(&rcu_preempt_state, rnp, true); |
b668c9cf PM |
410 | } else { |
411 | local_irq_restore(flags); | |
f41d911f | 412 | } |
f41d911f PM |
413 | } |
414 | ||
1ed509a2 PM |
415 | #ifdef CONFIG_RCU_CPU_STALL_VERBOSE |
416 | ||
417 | /* | |
418 | * Dump detailed information for all tasks blocking the current RCU | |
419 | * grace period on the specified rcu_node structure. | |
420 | */ | |
421 | static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp) | |
422 | { | |
423 | unsigned long flags; | |
1ed509a2 PM |
424 | struct task_struct *t; |
425 | ||
27f4d280 | 426 | if (!rcu_preempt_blocked_readers_cgp(rnp)) |
12f5f524 PM |
427 | return; |
428 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
429 | t = list_entry(rnp->gp_tasks, | |
430 | struct task_struct, rcu_node_entry); | |
431 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) | |
432 | sched_show_task(t); | |
433 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
1ed509a2 PM |
434 | } |
435 | ||
436 | /* | |
437 | * Dump detailed information for all tasks blocking the current RCU | |
438 | * grace period. | |
439 | */ | |
440 | static void rcu_print_detail_task_stall(struct rcu_state *rsp) | |
441 | { | |
442 | struct rcu_node *rnp = rcu_get_root(rsp); | |
443 | ||
444 | rcu_print_detail_task_stall_rnp(rnp); | |
445 | rcu_for_each_leaf_node(rsp, rnp) | |
446 | rcu_print_detail_task_stall_rnp(rnp); | |
447 | } | |
448 | ||
449 | #else /* #ifdef CONFIG_RCU_CPU_STALL_VERBOSE */ | |
450 | ||
451 | static void rcu_print_detail_task_stall(struct rcu_state *rsp) | |
452 | { | |
453 | } | |
454 | ||
455 | #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_VERBOSE */ | |
456 | ||
a858af28 PM |
457 | #ifdef CONFIG_RCU_CPU_STALL_INFO |
458 | ||
459 | static void rcu_print_task_stall_begin(struct rcu_node *rnp) | |
460 | { | |
461 | printk(KERN_ERR "\tTasks blocked on level-%d rcu_node (CPUs %d-%d):", | |
462 | rnp->level, rnp->grplo, rnp->grphi); | |
463 | } | |
464 | ||
465 | static void rcu_print_task_stall_end(void) | |
466 | { | |
467 | printk(KERN_CONT "\n"); | |
468 | } | |
469 | ||
470 | #else /* #ifdef CONFIG_RCU_CPU_STALL_INFO */ | |
471 | ||
472 | static void rcu_print_task_stall_begin(struct rcu_node *rnp) | |
473 | { | |
474 | } | |
475 | ||
476 | static void rcu_print_task_stall_end(void) | |
477 | { | |
478 | } | |
479 | ||
480 | #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_INFO */ | |
481 | ||
f41d911f PM |
482 | /* |
483 | * Scan the current list of tasks blocked within RCU read-side critical | |
484 | * sections, printing out the tid of each. | |
485 | */ | |
9bc8b558 | 486 | static int rcu_print_task_stall(struct rcu_node *rnp) |
f41d911f | 487 | { |
f41d911f | 488 | struct task_struct *t; |
9bc8b558 | 489 | int ndetected = 0; |
f41d911f | 490 | |
27f4d280 | 491 | if (!rcu_preempt_blocked_readers_cgp(rnp)) |
9bc8b558 | 492 | return 0; |
a858af28 | 493 | rcu_print_task_stall_begin(rnp); |
12f5f524 PM |
494 | t = list_entry(rnp->gp_tasks, |
495 | struct task_struct, rcu_node_entry); | |
9bc8b558 | 496 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) { |
a858af28 | 497 | printk(KERN_CONT " P%d", t->pid); |
9bc8b558 PM |
498 | ndetected++; |
499 | } | |
a858af28 | 500 | rcu_print_task_stall_end(); |
9bc8b558 | 501 | return ndetected; |
f41d911f PM |
502 | } |
503 | ||
b0e165c0 PM |
504 | /* |
505 | * Check that the list of blocked tasks for the newly completed grace | |
506 | * period is in fact empty. It is a serious bug to complete a grace | |
507 | * period that still has RCU readers blocked! This function must be | |
508 | * invoked -before- updating this rnp's ->gpnum, and the rnp's ->lock | |
509 | * must be held by the caller. | |
12f5f524 PM |
510 | * |
511 | * Also, if there are blocked tasks on the list, they automatically | |
512 | * block the newly created grace period, so set up ->gp_tasks accordingly. | |
b0e165c0 PM |
513 | */ |
514 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) | |
515 | { | |
27f4d280 | 516 | WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)); |
12f5f524 PM |
517 | if (!list_empty(&rnp->blkd_tasks)) |
518 | rnp->gp_tasks = rnp->blkd_tasks.next; | |
28ecd580 | 519 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
520 | } |
521 | ||
33f76148 PM |
522 | #ifdef CONFIG_HOTPLUG_CPU |
523 | ||
dd5d19ba PM |
524 | /* |
525 | * Handle tasklist migration for case in which all CPUs covered by the | |
526 | * specified rcu_node have gone offline. Move them up to the root | |
527 | * rcu_node. The reason for not just moving them to the immediate | |
528 | * parent is to remove the need for rcu_read_unlock_special() to | |
529 | * make more than two attempts to acquire the target rcu_node's lock. | |
b668c9cf PM |
530 | * Returns true if there were tasks blocking the current RCU grace |
531 | * period. | |
dd5d19ba | 532 | * |
237c80c5 PM |
533 | * Returns 1 if there was previously a task blocking the current grace |
534 | * period on the specified rcu_node structure. | |
535 | * | |
dd5d19ba PM |
536 | * The caller must hold rnp->lock with irqs disabled. |
537 | */ | |
237c80c5 PM |
538 | static int rcu_preempt_offline_tasks(struct rcu_state *rsp, |
539 | struct rcu_node *rnp, | |
540 | struct rcu_data *rdp) | |
dd5d19ba | 541 | { |
dd5d19ba PM |
542 | struct list_head *lp; |
543 | struct list_head *lp_root; | |
d9a3da06 | 544 | int retval = 0; |
dd5d19ba | 545 | struct rcu_node *rnp_root = rcu_get_root(rsp); |
12f5f524 | 546 | struct task_struct *t; |
dd5d19ba | 547 | |
86848966 PM |
548 | if (rnp == rnp_root) { |
549 | WARN_ONCE(1, "Last CPU thought to be offlined?"); | |
237c80c5 | 550 | return 0; /* Shouldn't happen: at least one CPU online. */ |
86848966 | 551 | } |
12f5f524 PM |
552 | |
553 | /* If we are on an internal node, complain bitterly. */ | |
554 | WARN_ON_ONCE(rnp != rdp->mynode); | |
dd5d19ba PM |
555 | |
556 | /* | |
12f5f524 PM |
557 | * Move tasks up to root rcu_node. Don't try to get fancy for |
558 | * this corner-case operation -- just put this node's tasks | |
559 | * at the head of the root node's list, and update the root node's | |
560 | * ->gp_tasks and ->exp_tasks pointers to those of this node's, | |
561 | * if non-NULL. This might result in waiting for more tasks than | |
562 | * absolutely necessary, but this is a good performance/complexity | |
563 | * tradeoff. | |
dd5d19ba | 564 | */ |
2036d94a | 565 | if (rcu_preempt_blocked_readers_cgp(rnp) && rnp->qsmask == 0) |
d9a3da06 PM |
566 | retval |= RCU_OFL_TASKS_NORM_GP; |
567 | if (rcu_preempted_readers_exp(rnp)) | |
568 | retval |= RCU_OFL_TASKS_EXP_GP; | |
12f5f524 PM |
569 | lp = &rnp->blkd_tasks; |
570 | lp_root = &rnp_root->blkd_tasks; | |
571 | while (!list_empty(lp)) { | |
572 | t = list_entry(lp->next, typeof(*t), rcu_node_entry); | |
573 | raw_spin_lock(&rnp_root->lock); /* irqs already disabled */ | |
574 | list_del(&t->rcu_node_entry); | |
575 | t->rcu_blocked_node = rnp_root; | |
576 | list_add(&t->rcu_node_entry, lp_root); | |
577 | if (&t->rcu_node_entry == rnp->gp_tasks) | |
578 | rnp_root->gp_tasks = rnp->gp_tasks; | |
579 | if (&t->rcu_node_entry == rnp->exp_tasks) | |
580 | rnp_root->exp_tasks = rnp->exp_tasks; | |
27f4d280 PM |
581 | #ifdef CONFIG_RCU_BOOST |
582 | if (&t->rcu_node_entry == rnp->boost_tasks) | |
583 | rnp_root->boost_tasks = rnp->boost_tasks; | |
584 | #endif /* #ifdef CONFIG_RCU_BOOST */ | |
12f5f524 | 585 | raw_spin_unlock(&rnp_root->lock); /* irqs still disabled */ |
dd5d19ba | 586 | } |
27f4d280 PM |
587 | |
588 | #ifdef CONFIG_RCU_BOOST | |
589 | /* In case root is being boosted and leaf is not. */ | |
590 | raw_spin_lock(&rnp_root->lock); /* irqs already disabled */ | |
591 | if (rnp_root->boost_tasks != NULL && | |
592 | rnp_root->boost_tasks != rnp_root->gp_tasks) | |
593 | rnp_root->boost_tasks = rnp_root->gp_tasks; | |
594 | raw_spin_unlock(&rnp_root->lock); /* irqs still disabled */ | |
595 | #endif /* #ifdef CONFIG_RCU_BOOST */ | |
596 | ||
12f5f524 PM |
597 | rnp->gp_tasks = NULL; |
598 | rnp->exp_tasks = NULL; | |
237c80c5 | 599 | return retval; |
dd5d19ba PM |
600 | } |
601 | ||
e5601400 PM |
602 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ |
603 | ||
f41d911f PM |
604 | /* |
605 | * Check for a quiescent state from the current CPU. When a task blocks, | |
606 | * the task is recorded in the corresponding CPU's rcu_node structure, | |
607 | * which is checked elsewhere. | |
608 | * | |
609 | * Caller must disable hard irqs. | |
610 | */ | |
611 | static void rcu_preempt_check_callbacks(int cpu) | |
612 | { | |
613 | struct task_struct *t = current; | |
614 | ||
615 | if (t->rcu_read_lock_nesting == 0) { | |
c3422bea | 616 | rcu_preempt_qs(cpu); |
f41d911f PM |
617 | return; |
618 | } | |
10f39bb1 PM |
619 | if (t->rcu_read_lock_nesting > 0 && |
620 | per_cpu(rcu_preempt_data, cpu).qs_pending) | |
c3422bea | 621 | t->rcu_read_unlock_special |= RCU_READ_UNLOCK_NEED_QS; |
f41d911f PM |
622 | } |
623 | ||
a46e0899 PM |
624 | #ifdef CONFIG_RCU_BOOST |
625 | ||
09223371 SL |
626 | static void rcu_preempt_do_callbacks(void) |
627 | { | |
628 | rcu_do_batch(&rcu_preempt_state, &__get_cpu_var(rcu_preempt_data)); | |
629 | } | |
630 | ||
a46e0899 PM |
631 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
632 | ||
f41d911f | 633 | /* |
6cc68793 | 634 | * Queue a preemptible-RCU callback for invocation after a grace period. |
f41d911f PM |
635 | */ |
636 | void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
637 | { | |
486e2593 | 638 | __call_rcu(head, func, &rcu_preempt_state, 0); |
f41d911f PM |
639 | } |
640 | EXPORT_SYMBOL_GPL(call_rcu); | |
641 | ||
486e2593 PM |
642 | /* |
643 | * Queue an RCU callback for lazy invocation after a grace period. | |
644 | * This will likely be later named something like "call_rcu_lazy()", | |
645 | * but this change will require some way of tagging the lazy RCU | |
646 | * callbacks in the list of pending callbacks. Until then, this | |
647 | * function may only be called from __kfree_rcu(). | |
648 | */ | |
649 | void kfree_call_rcu(struct rcu_head *head, | |
650 | void (*func)(struct rcu_head *rcu)) | |
651 | { | |
652 | __call_rcu(head, func, &rcu_preempt_state, 1); | |
653 | } | |
654 | EXPORT_SYMBOL_GPL(kfree_call_rcu); | |
655 | ||
6ebb237b PM |
656 | /** |
657 | * synchronize_rcu - wait until a grace period has elapsed. | |
658 | * | |
659 | * Control will return to the caller some time after a full grace | |
660 | * period has elapsed, in other words after all currently executing RCU | |
77d8485a PM |
661 | * read-side critical sections have completed. Note, however, that |
662 | * upon return from synchronize_rcu(), the caller might well be executing | |
663 | * concurrently with new RCU read-side critical sections that began while | |
664 | * synchronize_rcu() was waiting. RCU read-side critical sections are | |
665 | * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested. | |
6ebb237b PM |
666 | */ |
667 | void synchronize_rcu(void) | |
668 | { | |
fe15d706 PM |
669 | rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map) && |
670 | !lock_is_held(&rcu_lock_map) && | |
671 | !lock_is_held(&rcu_sched_lock_map), | |
672 | "Illegal synchronize_rcu() in RCU read-side critical section"); | |
6ebb237b PM |
673 | if (!rcu_scheduler_active) |
674 | return; | |
2c42818e | 675 | wait_rcu_gp(call_rcu); |
6ebb237b PM |
676 | } |
677 | EXPORT_SYMBOL_GPL(synchronize_rcu); | |
678 | ||
d9a3da06 PM |
679 | static DECLARE_WAIT_QUEUE_HEAD(sync_rcu_preempt_exp_wq); |
680 | static long sync_rcu_preempt_exp_count; | |
681 | static DEFINE_MUTEX(sync_rcu_preempt_exp_mutex); | |
682 | ||
683 | /* | |
684 | * Return non-zero if there are any tasks in RCU read-side critical | |
685 | * sections blocking the current preemptible-RCU expedited grace period. | |
686 | * If there is no preemptible-RCU expedited grace period currently in | |
687 | * progress, returns zero unconditionally. | |
688 | */ | |
689 | static int rcu_preempted_readers_exp(struct rcu_node *rnp) | |
690 | { | |
12f5f524 | 691 | return rnp->exp_tasks != NULL; |
d9a3da06 PM |
692 | } |
693 | ||
694 | /* | |
695 | * return non-zero if there is no RCU expedited grace period in progress | |
696 | * for the specified rcu_node structure, in other words, if all CPUs and | |
697 | * tasks covered by the specified rcu_node structure have done their bit | |
698 | * for the current expedited grace period. Works only for preemptible | |
699 | * RCU -- other RCU implementation use other means. | |
700 | * | |
701 | * Caller must hold sync_rcu_preempt_exp_mutex. | |
702 | */ | |
703 | static int sync_rcu_preempt_exp_done(struct rcu_node *rnp) | |
704 | { | |
705 | return !rcu_preempted_readers_exp(rnp) && | |
706 | ACCESS_ONCE(rnp->expmask) == 0; | |
707 | } | |
708 | ||
709 | /* | |
710 | * Report the exit from RCU read-side critical section for the last task | |
711 | * that queued itself during or before the current expedited preemptible-RCU | |
712 | * grace period. This event is reported either to the rcu_node structure on | |
713 | * which the task was queued or to one of that rcu_node structure's ancestors, | |
714 | * recursively up the tree. (Calm down, calm down, we do the recursion | |
715 | * iteratively!) | |
716 | * | |
b40d293e TG |
717 | * Most callers will set the "wake" flag, but the task initiating the |
718 | * expedited grace period need not wake itself. | |
719 | * | |
d9a3da06 PM |
720 | * Caller must hold sync_rcu_preempt_exp_mutex. |
721 | */ | |
b40d293e TG |
722 | static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp, |
723 | bool wake) | |
d9a3da06 PM |
724 | { |
725 | unsigned long flags; | |
726 | unsigned long mask; | |
727 | ||
1304afb2 | 728 | raw_spin_lock_irqsave(&rnp->lock, flags); |
d9a3da06 | 729 | for (;;) { |
131906b0 PM |
730 | if (!sync_rcu_preempt_exp_done(rnp)) { |
731 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
d9a3da06 | 732 | break; |
131906b0 | 733 | } |
d9a3da06 | 734 | if (rnp->parent == NULL) { |
131906b0 | 735 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
b40d293e TG |
736 | if (wake) |
737 | wake_up(&sync_rcu_preempt_exp_wq); | |
d9a3da06 PM |
738 | break; |
739 | } | |
740 | mask = rnp->grpmask; | |
1304afb2 | 741 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
d9a3da06 | 742 | rnp = rnp->parent; |
1304afb2 | 743 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
d9a3da06 PM |
744 | rnp->expmask &= ~mask; |
745 | } | |
d9a3da06 PM |
746 | } |
747 | ||
748 | /* | |
749 | * Snapshot the tasks blocking the newly started preemptible-RCU expedited | |
750 | * grace period for the specified rcu_node structure. If there are no such | |
751 | * tasks, report it up the rcu_node hierarchy. | |
752 | * | |
753 | * Caller must hold sync_rcu_preempt_exp_mutex and rsp->onofflock. | |
754 | */ | |
755 | static void | |
756 | sync_rcu_preempt_exp_init(struct rcu_state *rsp, struct rcu_node *rnp) | |
757 | { | |
1217ed1b | 758 | unsigned long flags; |
12f5f524 | 759 | int must_wait = 0; |
d9a3da06 | 760 | |
1217ed1b | 761 | raw_spin_lock_irqsave(&rnp->lock, flags); |
c701d5d9 | 762 | if (list_empty(&rnp->blkd_tasks)) { |
1217ed1b | 763 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
c701d5d9 | 764 | } else { |
12f5f524 | 765 | rnp->exp_tasks = rnp->blkd_tasks.next; |
1217ed1b | 766 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock */ |
12f5f524 PM |
767 | must_wait = 1; |
768 | } | |
d9a3da06 | 769 | if (!must_wait) |
b40d293e | 770 | rcu_report_exp_rnp(rsp, rnp, false); /* Don't wake self. */ |
d9a3da06 PM |
771 | } |
772 | ||
236fefaf PM |
773 | /** |
774 | * synchronize_rcu_expedited - Brute-force RCU grace period | |
775 | * | |
776 | * Wait for an RCU-preempt grace period, but expedite it. The basic | |
777 | * idea is to invoke synchronize_sched_expedited() to push all the tasks to | |
778 | * the ->blkd_tasks lists and wait for this list to drain. This consumes | |
779 | * significant time on all CPUs and is unfriendly to real-time workloads, | |
780 | * so is thus not recommended for any sort of common-case code. | |
781 | * In fact, if you are using synchronize_rcu_expedited() in a loop, | |
782 | * please restructure your code to batch your updates, and then Use a | |
783 | * single synchronize_rcu() instead. | |
784 | * | |
785 | * Note that it is illegal to call this function while holding any lock | |
786 | * that is acquired by a CPU-hotplug notifier. And yes, it is also illegal | |
787 | * to call this function from a CPU-hotplug notifier. Failing to observe | |
788 | * these restriction will result in deadlock. | |
019129d5 PM |
789 | */ |
790 | void synchronize_rcu_expedited(void) | |
791 | { | |
d9a3da06 PM |
792 | unsigned long flags; |
793 | struct rcu_node *rnp; | |
794 | struct rcu_state *rsp = &rcu_preempt_state; | |
795 | long snap; | |
796 | int trycount = 0; | |
797 | ||
798 | smp_mb(); /* Caller's modifications seen first by other CPUs. */ | |
799 | snap = ACCESS_ONCE(sync_rcu_preempt_exp_count) + 1; | |
800 | smp_mb(); /* Above access cannot bleed into critical section. */ | |
801 | ||
802 | /* | |
803 | * Acquire lock, falling back to synchronize_rcu() if too many | |
804 | * lock-acquisition failures. Of course, if someone does the | |
805 | * expedited grace period for us, just leave. | |
806 | */ | |
807 | while (!mutex_trylock(&sync_rcu_preempt_exp_mutex)) { | |
c701d5d9 | 808 | if (trycount++ < 10) { |
d9a3da06 | 809 | udelay(trycount * num_online_cpus()); |
c701d5d9 | 810 | } else { |
d9a3da06 PM |
811 | synchronize_rcu(); |
812 | return; | |
813 | } | |
814 | if ((ACCESS_ONCE(sync_rcu_preempt_exp_count) - snap) > 0) | |
815 | goto mb_ret; /* Others did our work for us. */ | |
816 | } | |
817 | if ((ACCESS_ONCE(sync_rcu_preempt_exp_count) - snap) > 0) | |
818 | goto unlock_mb_ret; /* Others did our work for us. */ | |
819 | ||
12f5f524 | 820 | /* force all RCU readers onto ->blkd_tasks lists. */ |
d9a3da06 PM |
821 | synchronize_sched_expedited(); |
822 | ||
1304afb2 | 823 | raw_spin_lock_irqsave(&rsp->onofflock, flags); |
d9a3da06 PM |
824 | |
825 | /* Initialize ->expmask for all non-leaf rcu_node structures. */ | |
826 | rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) { | |
1304afb2 | 827 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
d9a3da06 | 828 | rnp->expmask = rnp->qsmaskinit; |
1304afb2 | 829 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
d9a3da06 PM |
830 | } |
831 | ||
12f5f524 | 832 | /* Snapshot current state of ->blkd_tasks lists. */ |
d9a3da06 PM |
833 | rcu_for_each_leaf_node(rsp, rnp) |
834 | sync_rcu_preempt_exp_init(rsp, rnp); | |
835 | if (NUM_RCU_NODES > 1) | |
836 | sync_rcu_preempt_exp_init(rsp, rcu_get_root(rsp)); | |
837 | ||
1304afb2 | 838 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); |
d9a3da06 | 839 | |
12f5f524 | 840 | /* Wait for snapshotted ->blkd_tasks lists to drain. */ |
d9a3da06 PM |
841 | rnp = rcu_get_root(rsp); |
842 | wait_event(sync_rcu_preempt_exp_wq, | |
843 | sync_rcu_preempt_exp_done(rnp)); | |
844 | ||
845 | /* Clean up and exit. */ | |
846 | smp_mb(); /* ensure expedited GP seen before counter increment. */ | |
847 | ACCESS_ONCE(sync_rcu_preempt_exp_count)++; | |
848 | unlock_mb_ret: | |
849 | mutex_unlock(&sync_rcu_preempt_exp_mutex); | |
850 | mb_ret: | |
851 | smp_mb(); /* ensure subsequent action seen after grace period. */ | |
019129d5 PM |
852 | } |
853 | EXPORT_SYMBOL_GPL(synchronize_rcu_expedited); | |
854 | ||
e74f4c45 PM |
855 | /** |
856 | * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete. | |
857 | */ | |
858 | void rcu_barrier(void) | |
859 | { | |
037b64ed | 860 | _rcu_barrier(&rcu_preempt_state); |
e74f4c45 PM |
861 | } |
862 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
863 | ||
1eba8f84 | 864 | /* |
6cc68793 | 865 | * Initialize preemptible RCU's state structures. |
1eba8f84 PM |
866 | */ |
867 | static void __init __rcu_init_preempt(void) | |
868 | { | |
394f99a9 | 869 | rcu_init_one(&rcu_preempt_state, &rcu_preempt_data); |
1eba8f84 PM |
870 | } |
871 | ||
f41d911f PM |
872 | #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */ |
873 | ||
27f4d280 PM |
874 | static struct rcu_state *rcu_state = &rcu_sched_state; |
875 | ||
f41d911f PM |
876 | /* |
877 | * Tell them what RCU they are running. | |
878 | */ | |
0e0fc1c2 | 879 | static void __init rcu_bootup_announce(void) |
f41d911f PM |
880 | { |
881 | printk(KERN_INFO "Hierarchical RCU implementation.\n"); | |
26845c28 | 882 | rcu_bootup_announce_oddness(); |
f41d911f PM |
883 | } |
884 | ||
885 | /* | |
886 | * Return the number of RCU batches processed thus far for debug & stats. | |
887 | */ | |
888 | long rcu_batches_completed(void) | |
889 | { | |
890 | return rcu_batches_completed_sched(); | |
891 | } | |
892 | EXPORT_SYMBOL_GPL(rcu_batches_completed); | |
893 | ||
bf66f18e PM |
894 | /* |
895 | * Force a quiescent state for RCU, which, because there is no preemptible | |
896 | * RCU, becomes the same as rcu-sched. | |
897 | */ | |
898 | void rcu_force_quiescent_state(void) | |
899 | { | |
900 | rcu_sched_force_quiescent_state(); | |
901 | } | |
902 | EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); | |
903 | ||
cba6d0d6 PM |
904 | /* |
905 | * Because preemptible RCU does not exist, we never have to check for | |
906 | * CPUs being in quiescent states. | |
907 | */ | |
908 | static void rcu_preempt_note_context_switch(int cpu) | |
909 | { | |
910 | } | |
911 | ||
fc2219d4 | 912 | /* |
6cc68793 | 913 | * Because preemptible RCU does not exist, there are never any preempted |
fc2219d4 PM |
914 | * RCU readers. |
915 | */ | |
27f4d280 | 916 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 PM |
917 | { |
918 | return 0; | |
919 | } | |
920 | ||
b668c9cf PM |
921 | #ifdef CONFIG_HOTPLUG_CPU |
922 | ||
923 | /* Because preemptible RCU does not exist, no quieting of tasks. */ | |
d3f6bad3 | 924 | static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags) |
b668c9cf | 925 | { |
1304afb2 | 926 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
b668c9cf PM |
927 | } |
928 | ||
929 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | |
930 | ||
1ed509a2 | 931 | /* |
6cc68793 | 932 | * Because preemptible RCU does not exist, we never have to check for |
1ed509a2 PM |
933 | * tasks blocked within RCU read-side critical sections. |
934 | */ | |
935 | static void rcu_print_detail_task_stall(struct rcu_state *rsp) | |
936 | { | |
937 | } | |
938 | ||
f41d911f | 939 | /* |
6cc68793 | 940 | * Because preemptible RCU does not exist, we never have to check for |
f41d911f PM |
941 | * tasks blocked within RCU read-side critical sections. |
942 | */ | |
9bc8b558 | 943 | static int rcu_print_task_stall(struct rcu_node *rnp) |
f41d911f | 944 | { |
9bc8b558 | 945 | return 0; |
f41d911f PM |
946 | } |
947 | ||
b0e165c0 | 948 | /* |
6cc68793 | 949 | * Because there is no preemptible RCU, there can be no readers blocked, |
49e29126 PM |
950 | * so there is no need to check for blocked tasks. So check only for |
951 | * bogus qsmask values. | |
b0e165c0 PM |
952 | */ |
953 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) | |
954 | { | |
49e29126 | 955 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
956 | } |
957 | ||
33f76148 PM |
958 | #ifdef CONFIG_HOTPLUG_CPU |
959 | ||
dd5d19ba | 960 | /* |
6cc68793 | 961 | * Because preemptible RCU does not exist, it never needs to migrate |
237c80c5 PM |
962 | * tasks that were blocked within RCU read-side critical sections, and |
963 | * such non-existent tasks cannot possibly have been blocking the current | |
964 | * grace period. | |
dd5d19ba | 965 | */ |
237c80c5 PM |
966 | static int rcu_preempt_offline_tasks(struct rcu_state *rsp, |
967 | struct rcu_node *rnp, | |
968 | struct rcu_data *rdp) | |
dd5d19ba | 969 | { |
237c80c5 | 970 | return 0; |
dd5d19ba PM |
971 | } |
972 | ||
e5601400 PM |
973 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ |
974 | ||
f41d911f | 975 | /* |
6cc68793 | 976 | * Because preemptible RCU does not exist, it never has any callbacks |
f41d911f PM |
977 | * to check. |
978 | */ | |
1eba8f84 | 979 | static void rcu_preempt_check_callbacks(int cpu) |
f41d911f PM |
980 | { |
981 | } | |
982 | ||
486e2593 PM |
983 | /* |
984 | * Queue an RCU callback for lazy invocation after a grace period. | |
985 | * This will likely be later named something like "call_rcu_lazy()", | |
986 | * but this change will require some way of tagging the lazy RCU | |
987 | * callbacks in the list of pending callbacks. Until then, this | |
988 | * function may only be called from __kfree_rcu(). | |
989 | * | |
990 | * Because there is no preemptible RCU, we use RCU-sched instead. | |
991 | */ | |
992 | void kfree_call_rcu(struct rcu_head *head, | |
993 | void (*func)(struct rcu_head *rcu)) | |
994 | { | |
995 | __call_rcu(head, func, &rcu_sched_state, 1); | |
996 | } | |
997 | EXPORT_SYMBOL_GPL(kfree_call_rcu); | |
998 | ||
019129d5 PM |
999 | /* |
1000 | * Wait for an rcu-preempt grace period, but make it happen quickly. | |
6cc68793 | 1001 | * But because preemptible RCU does not exist, map to rcu-sched. |
019129d5 PM |
1002 | */ |
1003 | void synchronize_rcu_expedited(void) | |
1004 | { | |
1005 | synchronize_sched_expedited(); | |
1006 | } | |
1007 | EXPORT_SYMBOL_GPL(synchronize_rcu_expedited); | |
1008 | ||
d9a3da06 PM |
1009 | #ifdef CONFIG_HOTPLUG_CPU |
1010 | ||
1011 | /* | |
6cc68793 | 1012 | * Because preemptible RCU does not exist, there is never any need to |
d9a3da06 PM |
1013 | * report on tasks preempted in RCU read-side critical sections during |
1014 | * expedited RCU grace periods. | |
1015 | */ | |
b40d293e TG |
1016 | static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp, |
1017 | bool wake) | |
d9a3da06 | 1018 | { |
d9a3da06 PM |
1019 | } |
1020 | ||
1021 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | |
1022 | ||
e74f4c45 | 1023 | /* |
6cc68793 | 1024 | * Because preemptible RCU does not exist, rcu_barrier() is just |
e74f4c45 PM |
1025 | * another name for rcu_barrier_sched(). |
1026 | */ | |
1027 | void rcu_barrier(void) | |
1028 | { | |
1029 | rcu_barrier_sched(); | |
1030 | } | |
1031 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
1032 | ||
1eba8f84 | 1033 | /* |
6cc68793 | 1034 | * Because preemptible RCU does not exist, it need not be initialized. |
1eba8f84 PM |
1035 | */ |
1036 | static void __init __rcu_init_preempt(void) | |
1037 | { | |
1038 | } | |
1039 | ||
f41d911f | 1040 | #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */ |
8bd93a2c | 1041 | |
27f4d280 PM |
1042 | #ifdef CONFIG_RCU_BOOST |
1043 | ||
1044 | #include "rtmutex_common.h" | |
1045 | ||
0ea1f2eb PM |
1046 | #ifdef CONFIG_RCU_TRACE |
1047 | ||
1048 | static void rcu_initiate_boost_trace(struct rcu_node *rnp) | |
1049 | { | |
1050 | if (list_empty(&rnp->blkd_tasks)) | |
1051 | rnp->n_balk_blkd_tasks++; | |
1052 | else if (rnp->exp_tasks == NULL && rnp->gp_tasks == NULL) | |
1053 | rnp->n_balk_exp_gp_tasks++; | |
1054 | else if (rnp->gp_tasks != NULL && rnp->boost_tasks != NULL) | |
1055 | rnp->n_balk_boost_tasks++; | |
1056 | else if (rnp->gp_tasks != NULL && rnp->qsmask != 0) | |
1057 | rnp->n_balk_notblocked++; | |
1058 | else if (rnp->gp_tasks != NULL && | |
a9f4793d | 1059 | ULONG_CMP_LT(jiffies, rnp->boost_time)) |
0ea1f2eb PM |
1060 | rnp->n_balk_notyet++; |
1061 | else | |
1062 | rnp->n_balk_nos++; | |
1063 | } | |
1064 | ||
1065 | #else /* #ifdef CONFIG_RCU_TRACE */ | |
1066 | ||
1067 | static void rcu_initiate_boost_trace(struct rcu_node *rnp) | |
1068 | { | |
1069 | } | |
1070 | ||
1071 | #endif /* #else #ifdef CONFIG_RCU_TRACE */ | |
1072 | ||
5d01bbd1 TG |
1073 | static void rcu_wake_cond(struct task_struct *t, int status) |
1074 | { | |
1075 | /* | |
1076 | * If the thread is yielding, only wake it when this | |
1077 | * is invoked from idle | |
1078 | */ | |
1079 | if (status != RCU_KTHREAD_YIELDING || is_idle_task(current)) | |
1080 | wake_up_process(t); | |
1081 | } | |
1082 | ||
27f4d280 PM |
1083 | /* |
1084 | * Carry out RCU priority boosting on the task indicated by ->exp_tasks | |
1085 | * or ->boost_tasks, advancing the pointer to the next task in the | |
1086 | * ->blkd_tasks list. | |
1087 | * | |
1088 | * Note that irqs must be enabled: boosting the task can block. | |
1089 | * Returns 1 if there are more tasks needing to be boosted. | |
1090 | */ | |
1091 | static int rcu_boost(struct rcu_node *rnp) | |
1092 | { | |
1093 | unsigned long flags; | |
1094 | struct rt_mutex mtx; | |
1095 | struct task_struct *t; | |
1096 | struct list_head *tb; | |
1097 | ||
1098 | if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) | |
1099 | return 0; /* Nothing left to boost. */ | |
1100 | ||
1101 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
1102 | ||
1103 | /* | |
1104 | * Recheck under the lock: all tasks in need of boosting | |
1105 | * might exit their RCU read-side critical sections on their own. | |
1106 | */ | |
1107 | if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) { | |
1108 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
1109 | return 0; | |
1110 | } | |
1111 | ||
1112 | /* | |
1113 | * Preferentially boost tasks blocking expedited grace periods. | |
1114 | * This cannot starve the normal grace periods because a second | |
1115 | * expedited grace period must boost all blocked tasks, including | |
1116 | * those blocking the pre-existing normal grace period. | |
1117 | */ | |
0ea1f2eb | 1118 | if (rnp->exp_tasks != NULL) { |
27f4d280 | 1119 | tb = rnp->exp_tasks; |
0ea1f2eb PM |
1120 | rnp->n_exp_boosts++; |
1121 | } else { | |
27f4d280 | 1122 | tb = rnp->boost_tasks; |
0ea1f2eb PM |
1123 | rnp->n_normal_boosts++; |
1124 | } | |
1125 | rnp->n_tasks_boosted++; | |
27f4d280 PM |
1126 | |
1127 | /* | |
1128 | * We boost task t by manufacturing an rt_mutex that appears to | |
1129 | * be held by task t. We leave a pointer to that rt_mutex where | |
1130 | * task t can find it, and task t will release the mutex when it | |
1131 | * exits its outermost RCU read-side critical section. Then | |
1132 | * simply acquiring this artificial rt_mutex will boost task | |
1133 | * t's priority. (Thanks to tglx for suggesting this approach!) | |
1134 | * | |
1135 | * Note that task t must acquire rnp->lock to remove itself from | |
1136 | * the ->blkd_tasks list, which it will do from exit() if from | |
1137 | * nowhere else. We therefore are guaranteed that task t will | |
1138 | * stay around at least until we drop rnp->lock. Note that | |
1139 | * rnp->lock also resolves races between our priority boosting | |
1140 | * and task t's exiting its outermost RCU read-side critical | |
1141 | * section. | |
1142 | */ | |
1143 | t = container_of(tb, struct task_struct, rcu_node_entry); | |
1144 | rt_mutex_init_proxy_locked(&mtx, t); | |
1145 | t->rcu_boost_mutex = &mtx; | |
27f4d280 PM |
1146 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1147 | rt_mutex_lock(&mtx); /* Side effect: boosts task t's priority. */ | |
1148 | rt_mutex_unlock(&mtx); /* Keep lockdep happy. */ | |
1149 | ||
4f89b336 PM |
1150 | return ACCESS_ONCE(rnp->exp_tasks) != NULL || |
1151 | ACCESS_ONCE(rnp->boost_tasks) != NULL; | |
27f4d280 PM |
1152 | } |
1153 | ||
27f4d280 PM |
1154 | /* |
1155 | * Priority-boosting kthread. One per leaf rcu_node and one for the | |
1156 | * root rcu_node. | |
1157 | */ | |
1158 | static int rcu_boost_kthread(void *arg) | |
1159 | { | |
1160 | struct rcu_node *rnp = (struct rcu_node *)arg; | |
1161 | int spincnt = 0; | |
1162 | int more2boost; | |
1163 | ||
385680a9 | 1164 | trace_rcu_utilization("Start boost kthread@init"); |
27f4d280 | 1165 | for (;;) { |
d71df90e | 1166 | rnp->boost_kthread_status = RCU_KTHREAD_WAITING; |
385680a9 | 1167 | trace_rcu_utilization("End boost kthread@rcu_wait"); |
08bca60a | 1168 | rcu_wait(rnp->boost_tasks || rnp->exp_tasks); |
385680a9 | 1169 | trace_rcu_utilization("Start boost kthread@rcu_wait"); |
d71df90e | 1170 | rnp->boost_kthread_status = RCU_KTHREAD_RUNNING; |
27f4d280 PM |
1171 | more2boost = rcu_boost(rnp); |
1172 | if (more2boost) | |
1173 | spincnt++; | |
1174 | else | |
1175 | spincnt = 0; | |
1176 | if (spincnt > 10) { | |
5d01bbd1 | 1177 | rnp->boost_kthread_status = RCU_KTHREAD_YIELDING; |
385680a9 | 1178 | trace_rcu_utilization("End boost kthread@rcu_yield"); |
5d01bbd1 | 1179 | schedule_timeout_interruptible(2); |
385680a9 | 1180 | trace_rcu_utilization("Start boost kthread@rcu_yield"); |
27f4d280 PM |
1181 | spincnt = 0; |
1182 | } | |
1183 | } | |
1217ed1b | 1184 | /* NOTREACHED */ |
385680a9 | 1185 | trace_rcu_utilization("End boost kthread@notreached"); |
27f4d280 PM |
1186 | return 0; |
1187 | } | |
1188 | ||
1189 | /* | |
1190 | * Check to see if it is time to start boosting RCU readers that are | |
1191 | * blocking the current grace period, and, if so, tell the per-rcu_node | |
1192 | * kthread to start boosting them. If there is an expedited grace | |
1193 | * period in progress, it is always time to boost. | |
1194 | * | |
1217ed1b PM |
1195 | * The caller must hold rnp->lock, which this function releases, |
1196 | * but irqs remain disabled. The ->boost_kthread_task is immortal, | |
1197 | * so we don't need to worry about it going away. | |
27f4d280 | 1198 | */ |
1217ed1b | 1199 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
27f4d280 PM |
1200 | { |
1201 | struct task_struct *t; | |
1202 | ||
0ea1f2eb PM |
1203 | if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) { |
1204 | rnp->n_balk_exp_gp_tasks++; | |
1217ed1b | 1205 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
27f4d280 | 1206 | return; |
0ea1f2eb | 1207 | } |
27f4d280 PM |
1208 | if (rnp->exp_tasks != NULL || |
1209 | (rnp->gp_tasks != NULL && | |
1210 | rnp->boost_tasks == NULL && | |
1211 | rnp->qsmask == 0 && | |
1212 | ULONG_CMP_GE(jiffies, rnp->boost_time))) { | |
1213 | if (rnp->exp_tasks == NULL) | |
1214 | rnp->boost_tasks = rnp->gp_tasks; | |
1217ed1b | 1215 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
27f4d280 | 1216 | t = rnp->boost_kthread_task; |
5d01bbd1 TG |
1217 | if (t) |
1218 | rcu_wake_cond(t, rnp->boost_kthread_status); | |
1217ed1b | 1219 | } else { |
0ea1f2eb | 1220 | rcu_initiate_boost_trace(rnp); |
1217ed1b PM |
1221 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1222 | } | |
27f4d280 PM |
1223 | } |
1224 | ||
a46e0899 PM |
1225 | /* |
1226 | * Wake up the per-CPU kthread to invoke RCU callbacks. | |
1227 | */ | |
1228 | static void invoke_rcu_callbacks_kthread(void) | |
1229 | { | |
1230 | unsigned long flags; | |
1231 | ||
1232 | local_irq_save(flags); | |
1233 | __this_cpu_write(rcu_cpu_has_work, 1); | |
1eb52121 | 1234 | if (__this_cpu_read(rcu_cpu_kthread_task) != NULL && |
5d01bbd1 TG |
1235 | current != __this_cpu_read(rcu_cpu_kthread_task)) { |
1236 | rcu_wake_cond(__this_cpu_read(rcu_cpu_kthread_task), | |
1237 | __this_cpu_read(rcu_cpu_kthread_status)); | |
1238 | } | |
a46e0899 PM |
1239 | local_irq_restore(flags); |
1240 | } | |
1241 | ||
dff1672d PM |
1242 | /* |
1243 | * Is the current CPU running the RCU-callbacks kthread? | |
1244 | * Caller must have preemption disabled. | |
1245 | */ | |
1246 | static bool rcu_is_callbacks_kthread(void) | |
1247 | { | |
1248 | return __get_cpu_var(rcu_cpu_kthread_task) == current; | |
1249 | } | |
1250 | ||
27f4d280 PM |
1251 | #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000) |
1252 | ||
1253 | /* | |
1254 | * Do priority-boost accounting for the start of a new grace period. | |
1255 | */ | |
1256 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) | |
1257 | { | |
1258 | rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES; | |
1259 | } | |
1260 | ||
27f4d280 PM |
1261 | /* |
1262 | * Create an RCU-boost kthread for the specified node if one does not | |
1263 | * already exist. We only create this kthread for preemptible RCU. | |
1264 | * Returns zero if all is well, a negated errno otherwise. | |
1265 | */ | |
1266 | static int __cpuinit rcu_spawn_one_boost_kthread(struct rcu_state *rsp, | |
5d01bbd1 | 1267 | struct rcu_node *rnp) |
27f4d280 | 1268 | { |
5d01bbd1 | 1269 | int rnp_index = rnp - &rsp->node[0]; |
27f4d280 PM |
1270 | unsigned long flags; |
1271 | struct sched_param sp; | |
1272 | struct task_struct *t; | |
1273 | ||
1274 | if (&rcu_preempt_state != rsp) | |
1275 | return 0; | |
5d01bbd1 TG |
1276 | |
1277 | if (!rcu_scheduler_fully_active || rnp->qsmaskinit == 0) | |
1278 | return 0; | |
1279 | ||
a46e0899 | 1280 | rsp->boost = 1; |
27f4d280 PM |
1281 | if (rnp->boost_kthread_task != NULL) |
1282 | return 0; | |
1283 | t = kthread_create(rcu_boost_kthread, (void *)rnp, | |
5b61b0ba | 1284 | "rcub/%d", rnp_index); |
27f4d280 PM |
1285 | if (IS_ERR(t)) |
1286 | return PTR_ERR(t); | |
1287 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
1288 | rnp->boost_kthread_task = t; | |
1289 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
5b61b0ba | 1290 | sp.sched_priority = RCU_BOOST_PRIO; |
27f4d280 | 1291 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); |
9a432736 | 1292 | wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */ |
27f4d280 PM |
1293 | return 0; |
1294 | } | |
1295 | ||
f8b7fc6b PM |
1296 | static void rcu_kthread_do_work(void) |
1297 | { | |
1298 | rcu_do_batch(&rcu_sched_state, &__get_cpu_var(rcu_sched_data)); | |
1299 | rcu_do_batch(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); | |
1300 | rcu_preempt_do_callbacks(); | |
1301 | } | |
1302 | ||
62ab7072 | 1303 | static void rcu_cpu_kthread_setup(unsigned int cpu) |
f8b7fc6b | 1304 | { |
f8b7fc6b | 1305 | struct sched_param sp; |
f8b7fc6b | 1306 | |
62ab7072 PM |
1307 | sp.sched_priority = RCU_KTHREAD_PRIO; |
1308 | sched_setscheduler_nocheck(current, SCHED_FIFO, &sp); | |
f8b7fc6b PM |
1309 | } |
1310 | ||
62ab7072 PM |
1311 | static void rcu_cpu_kthread_park(unsigned int cpu) |
1312 | { | |
1313 | per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU; | |
1314 | } | |
1315 | ||
1316 | static int rcu_cpu_kthread_should_run(unsigned int cpu) | |
1317 | { | |
1318 | return __get_cpu_var(rcu_cpu_has_work); | |
f8b7fc6b PM |
1319 | } |
1320 | ||
1321 | /* | |
1322 | * Per-CPU kernel thread that invokes RCU callbacks. This replaces the | |
e0f23060 PM |
1323 | * RCU softirq used in flavors and configurations of RCU that do not |
1324 | * support RCU priority boosting. | |
f8b7fc6b | 1325 | */ |
62ab7072 | 1326 | static void rcu_cpu_kthread(unsigned int cpu) |
f8b7fc6b | 1327 | { |
62ab7072 PM |
1328 | unsigned int *statusp = &__get_cpu_var(rcu_cpu_kthread_status); |
1329 | char work, *workp = &__get_cpu_var(rcu_cpu_has_work); | |
1330 | int spincnt; | |
f8b7fc6b | 1331 | |
62ab7072 | 1332 | for (spincnt = 0; spincnt < 10; spincnt++) { |
385680a9 | 1333 | trace_rcu_utilization("Start CPU kthread@rcu_wait"); |
f8b7fc6b | 1334 | local_bh_disable(); |
f8b7fc6b | 1335 | *statusp = RCU_KTHREAD_RUNNING; |
62ab7072 PM |
1336 | this_cpu_inc(rcu_cpu_kthread_loops); |
1337 | local_irq_disable(); | |
f8b7fc6b PM |
1338 | work = *workp; |
1339 | *workp = 0; | |
62ab7072 | 1340 | local_irq_enable(); |
f8b7fc6b PM |
1341 | if (work) |
1342 | rcu_kthread_do_work(); | |
1343 | local_bh_enable(); | |
62ab7072 PM |
1344 | if (*workp == 0) { |
1345 | trace_rcu_utilization("End CPU kthread@rcu_wait"); | |
1346 | *statusp = RCU_KTHREAD_WAITING; | |
1347 | return; | |
f8b7fc6b PM |
1348 | } |
1349 | } | |
62ab7072 PM |
1350 | *statusp = RCU_KTHREAD_YIELDING; |
1351 | trace_rcu_utilization("Start CPU kthread@rcu_yield"); | |
1352 | schedule_timeout_interruptible(2); | |
1353 | trace_rcu_utilization("End CPU kthread@rcu_yield"); | |
1354 | *statusp = RCU_KTHREAD_WAITING; | |
f8b7fc6b PM |
1355 | } |
1356 | ||
f8b7fc6b PM |
1357 | /* |
1358 | * Set the per-rcu_node kthread's affinity to cover all CPUs that are | |
1359 | * served by the rcu_node in question. The CPU hotplug lock is still | |
1360 | * held, so the value of rnp->qsmaskinit will be stable. | |
1361 | * | |
1362 | * We don't include outgoingcpu in the affinity set, use -1 if there is | |
1363 | * no outgoing CPU. If there are no CPUs left in the affinity set, | |
1364 | * this function allows the kthread to execute on any CPU. | |
1365 | */ | |
5d01bbd1 | 1366 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b | 1367 | { |
5d01bbd1 TG |
1368 | struct task_struct *t = rnp->boost_kthread_task; |
1369 | unsigned long mask = rnp->qsmaskinit; | |
f8b7fc6b PM |
1370 | cpumask_var_t cm; |
1371 | int cpu; | |
f8b7fc6b | 1372 | |
5d01bbd1 | 1373 | if (!t) |
f8b7fc6b | 1374 | return; |
5d01bbd1 | 1375 | if (!zalloc_cpumask_var(&cm, GFP_KERNEL)) |
f8b7fc6b | 1376 | return; |
f8b7fc6b PM |
1377 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1) |
1378 | if ((mask & 0x1) && cpu != outgoingcpu) | |
1379 | cpumask_set_cpu(cpu, cm); | |
1380 | if (cpumask_weight(cm) == 0) { | |
1381 | cpumask_setall(cm); | |
1382 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++) | |
1383 | cpumask_clear_cpu(cpu, cm); | |
1384 | WARN_ON_ONCE(cpumask_weight(cm) == 0); | |
1385 | } | |
5d01bbd1 | 1386 | set_cpus_allowed_ptr(t, cm); |
f8b7fc6b PM |
1387 | free_cpumask_var(cm); |
1388 | } | |
1389 | ||
62ab7072 PM |
1390 | static struct smp_hotplug_thread rcu_cpu_thread_spec = { |
1391 | .store = &rcu_cpu_kthread_task, | |
1392 | .thread_should_run = rcu_cpu_kthread_should_run, | |
1393 | .thread_fn = rcu_cpu_kthread, | |
1394 | .thread_comm = "rcuc/%u", | |
1395 | .setup = rcu_cpu_kthread_setup, | |
1396 | .park = rcu_cpu_kthread_park, | |
1397 | }; | |
1398 | ||
f8b7fc6b PM |
1399 | /* |
1400 | * Spawn all kthreads -- called as soon as the scheduler is running. | |
1401 | */ | |
1402 | static int __init rcu_spawn_kthreads(void) | |
1403 | { | |
f8b7fc6b | 1404 | struct rcu_node *rnp; |
5d01bbd1 | 1405 | int cpu; |
f8b7fc6b | 1406 | |
b0d30417 | 1407 | rcu_scheduler_fully_active = 1; |
62ab7072 | 1408 | for_each_possible_cpu(cpu) |
f8b7fc6b | 1409 | per_cpu(rcu_cpu_has_work, cpu) = 0; |
62ab7072 | 1410 | BUG_ON(smpboot_register_percpu_thread(&rcu_cpu_thread_spec)); |
f8b7fc6b | 1411 | rnp = rcu_get_root(rcu_state); |
5d01bbd1 | 1412 | (void)rcu_spawn_one_boost_kthread(rcu_state, rnp); |
f8b7fc6b PM |
1413 | if (NUM_RCU_NODES > 1) { |
1414 | rcu_for_each_leaf_node(rcu_state, rnp) | |
5d01bbd1 | 1415 | (void)rcu_spawn_one_boost_kthread(rcu_state, rnp); |
f8b7fc6b PM |
1416 | } |
1417 | return 0; | |
1418 | } | |
1419 | early_initcall(rcu_spawn_kthreads); | |
1420 | ||
1421 | static void __cpuinit rcu_prepare_kthreads(int cpu) | |
1422 | { | |
1423 | struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu); | |
1424 | struct rcu_node *rnp = rdp->mynode; | |
1425 | ||
1426 | /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */ | |
62ab7072 | 1427 | if (rcu_scheduler_fully_active) |
5d01bbd1 | 1428 | (void)rcu_spawn_one_boost_kthread(rcu_state, rnp); |
f8b7fc6b PM |
1429 | } |
1430 | ||
27f4d280 PM |
1431 | #else /* #ifdef CONFIG_RCU_BOOST */ |
1432 | ||
1217ed1b | 1433 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
27f4d280 | 1434 | { |
1217ed1b | 1435 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
27f4d280 PM |
1436 | } |
1437 | ||
a46e0899 | 1438 | static void invoke_rcu_callbacks_kthread(void) |
27f4d280 | 1439 | { |
a46e0899 | 1440 | WARN_ON_ONCE(1); |
27f4d280 PM |
1441 | } |
1442 | ||
dff1672d PM |
1443 | static bool rcu_is_callbacks_kthread(void) |
1444 | { | |
1445 | return false; | |
1446 | } | |
1447 | ||
27f4d280 PM |
1448 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) |
1449 | { | |
1450 | } | |
1451 | ||
5d01bbd1 | 1452 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b PM |
1453 | { |
1454 | } | |
1455 | ||
b0d30417 PM |
1456 | static int __init rcu_scheduler_really_started(void) |
1457 | { | |
1458 | rcu_scheduler_fully_active = 1; | |
1459 | return 0; | |
1460 | } | |
1461 | early_initcall(rcu_scheduler_really_started); | |
1462 | ||
f8b7fc6b PM |
1463 | static void __cpuinit rcu_prepare_kthreads(int cpu) |
1464 | { | |
1465 | } | |
1466 | ||
27f4d280 PM |
1467 | #endif /* #else #ifdef CONFIG_RCU_BOOST */ |
1468 | ||
8bd93a2c PM |
1469 | #if !defined(CONFIG_RCU_FAST_NO_HZ) |
1470 | ||
1471 | /* | |
1472 | * Check to see if any future RCU-related work will need to be done | |
1473 | * by the current CPU, even if none need be done immediately, returning | |
1474 | * 1 if so. This function is part of the RCU implementation; it is -not- | |
1475 | * an exported member of the RCU API. | |
1476 | * | |
7cb92499 PM |
1477 | * Because we not have RCU_FAST_NO_HZ, just check whether this CPU needs |
1478 | * any flavor of RCU. | |
8bd93a2c | 1479 | */ |
aa9b1630 | 1480 | int rcu_needs_cpu(int cpu, unsigned long *delta_jiffies) |
8bd93a2c | 1481 | { |
aa9b1630 | 1482 | *delta_jiffies = ULONG_MAX; |
aea1b35e PM |
1483 | return rcu_cpu_has_callbacks(cpu); |
1484 | } | |
1485 | ||
7cb92499 PM |
1486 | /* |
1487 | * Because we do not have RCU_FAST_NO_HZ, don't bother initializing for it. | |
1488 | */ | |
1489 | static void rcu_prepare_for_idle_init(int cpu) | |
1490 | { | |
1491 | } | |
1492 | ||
1493 | /* | |
1494 | * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up | |
1495 | * after it. | |
1496 | */ | |
1497 | static void rcu_cleanup_after_idle(int cpu) | |
1498 | { | |
1499 | } | |
1500 | ||
aea1b35e | 1501 | /* |
a858af28 | 1502 | * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n, |
aea1b35e PM |
1503 | * is nothing. |
1504 | */ | |
1505 | static void rcu_prepare_for_idle(int cpu) | |
1506 | { | |
1507 | } | |
1508 | ||
c57afe80 PM |
1509 | /* |
1510 | * Don't bother keeping a running count of the number of RCU callbacks | |
1511 | * posted because CONFIG_RCU_FAST_NO_HZ=n. | |
1512 | */ | |
1513 | static void rcu_idle_count_callbacks_posted(void) | |
1514 | { | |
1515 | } | |
1516 | ||
8bd93a2c PM |
1517 | #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
1518 | ||
f23f7fa1 PM |
1519 | /* |
1520 | * This code is invoked when a CPU goes idle, at which point we want | |
1521 | * to have the CPU do everything required for RCU so that it can enter | |
1522 | * the energy-efficient dyntick-idle mode. This is handled by a | |
1523 | * state machine implemented by rcu_prepare_for_idle() below. | |
1524 | * | |
1525 | * The following three proprocessor symbols control this state machine: | |
1526 | * | |
1527 | * RCU_IDLE_FLUSHES gives the maximum number of times that we will attempt | |
1528 | * to satisfy RCU. Beyond this point, it is better to incur a periodic | |
1529 | * scheduling-clock interrupt than to loop through the state machine | |
1530 | * at full power. | |
1531 | * RCU_IDLE_OPT_FLUSHES gives the number of RCU_IDLE_FLUSHES that are | |
1532 | * optional if RCU does not need anything immediately from this | |
1533 | * CPU, even if this CPU still has RCU callbacks queued. The first | |
1534 | * times through the state machine are mandatory: we need to give | |
1535 | * the state machine a chance to communicate a quiescent state | |
1536 | * to the RCU core. | |
1537 | * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted | |
1538 | * to sleep in dyntick-idle mode with RCU callbacks pending. This | |
1539 | * is sized to be roughly one RCU grace period. Those energy-efficiency | |
1540 | * benchmarkers who might otherwise be tempted to set this to a large | |
1541 | * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your | |
1542 | * system. And if you are -that- concerned about energy efficiency, | |
1543 | * just power the system down and be done with it! | |
778d250a PM |
1544 | * RCU_IDLE_LAZY_GP_DELAY gives the number of jiffies that a CPU is |
1545 | * permitted to sleep in dyntick-idle mode with only lazy RCU | |
1546 | * callbacks pending. Setting this too high can OOM your system. | |
f23f7fa1 PM |
1547 | * |
1548 | * The values below work well in practice. If future workloads require | |
1549 | * adjustment, they can be converted into kernel config parameters, though | |
1550 | * making the state machine smarter might be a better option. | |
1551 | */ | |
1552 | #define RCU_IDLE_FLUSHES 5 /* Number of dyntick-idle tries. */ | |
1553 | #define RCU_IDLE_OPT_FLUSHES 3 /* Optional dyntick-idle tries. */ | |
e84c48ae | 1554 | #define RCU_IDLE_GP_DELAY 4 /* Roughly one grace period. */ |
778d250a | 1555 | #define RCU_IDLE_LAZY_GP_DELAY (6 * HZ) /* Roughly six seconds. */ |
f23f7fa1 | 1556 | |
9d2ad243 PM |
1557 | extern int tick_nohz_enabled; |
1558 | ||
486e2593 PM |
1559 | /* |
1560 | * Does the specified flavor of RCU have non-lazy callbacks pending on | |
1561 | * the specified CPU? Both RCU flavor and CPU are specified by the | |
1562 | * rcu_data structure. | |
1563 | */ | |
1564 | static bool __rcu_cpu_has_nonlazy_callbacks(struct rcu_data *rdp) | |
1565 | { | |
1566 | return rdp->qlen != rdp->qlen_lazy; | |
1567 | } | |
1568 | ||
1569 | #ifdef CONFIG_TREE_PREEMPT_RCU | |
1570 | ||
1571 | /* | |
1572 | * Are there non-lazy RCU-preempt callbacks? (There cannot be if there | |
1573 | * is no RCU-preempt in the kernel.) | |
1574 | */ | |
1575 | static bool rcu_preempt_cpu_has_nonlazy_callbacks(int cpu) | |
1576 | { | |
1577 | struct rcu_data *rdp = &per_cpu(rcu_preempt_data, cpu); | |
1578 | ||
1579 | return __rcu_cpu_has_nonlazy_callbacks(rdp); | |
1580 | } | |
1581 | ||
1582 | #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */ | |
1583 | ||
1584 | static bool rcu_preempt_cpu_has_nonlazy_callbacks(int cpu) | |
1585 | { | |
1586 | return 0; | |
1587 | } | |
1588 | ||
1589 | #endif /* else #ifdef CONFIG_TREE_PREEMPT_RCU */ | |
1590 | ||
1591 | /* | |
1592 | * Does any flavor of RCU have non-lazy callbacks on the specified CPU? | |
1593 | */ | |
1594 | static bool rcu_cpu_has_nonlazy_callbacks(int cpu) | |
1595 | { | |
1596 | return __rcu_cpu_has_nonlazy_callbacks(&per_cpu(rcu_sched_data, cpu)) || | |
1597 | __rcu_cpu_has_nonlazy_callbacks(&per_cpu(rcu_bh_data, cpu)) || | |
1598 | rcu_preempt_cpu_has_nonlazy_callbacks(cpu); | |
1599 | } | |
1600 | ||
aa9b1630 PM |
1601 | /* |
1602 | * Allow the CPU to enter dyntick-idle mode if either: (1) There are no | |
1603 | * callbacks on this CPU, (2) this CPU has not yet attempted to enter | |
1604 | * dyntick-idle mode, or (3) this CPU is in the process of attempting to | |
1605 | * enter dyntick-idle mode. Otherwise, if we have recently tried and failed | |
1606 | * to enter dyntick-idle mode, we refuse to try to enter it. After all, | |
1607 | * it is better to incur scheduling-clock interrupts than to spin | |
1608 | * continuously for the same time duration! | |
1609 | * | |
1610 | * The delta_jiffies argument is used to store the time when RCU is | |
1611 | * going to need the CPU again if it still has callbacks. The reason | |
1612 | * for this is that rcu_prepare_for_idle() might need to post a timer, | |
1613 | * but if so, it will do so after tick_nohz_stop_sched_tick() has set | |
1614 | * the wakeup time for this CPU. This means that RCU's timer can be | |
1615 | * delayed until the wakeup time, which defeats the purpose of posting | |
1616 | * a timer. | |
1617 | */ | |
1618 | int rcu_needs_cpu(int cpu, unsigned long *delta_jiffies) | |
1619 | { | |
1620 | struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); | |
1621 | ||
1622 | /* Flag a new idle sojourn to the idle-entry state machine. */ | |
1623 | rdtp->idle_first_pass = 1; | |
1624 | /* If no callbacks, RCU doesn't need the CPU. */ | |
1625 | if (!rcu_cpu_has_callbacks(cpu)) { | |
1626 | *delta_jiffies = ULONG_MAX; | |
1627 | return 0; | |
1628 | } | |
1629 | if (rdtp->dyntick_holdoff == jiffies) { | |
1630 | /* RCU recently tried and failed, so don't try again. */ | |
1631 | *delta_jiffies = 1; | |
1632 | return 1; | |
1633 | } | |
1634 | /* Set up for the possibility that RCU will post a timer. */ | |
e84c48ae PM |
1635 | if (rcu_cpu_has_nonlazy_callbacks(cpu)) { |
1636 | *delta_jiffies = round_up(RCU_IDLE_GP_DELAY + jiffies, | |
1637 | RCU_IDLE_GP_DELAY) - jiffies; | |
1638 | } else { | |
1639 | *delta_jiffies = jiffies + RCU_IDLE_LAZY_GP_DELAY; | |
1640 | *delta_jiffies = round_jiffies(*delta_jiffies) - jiffies; | |
1641 | } | |
aa9b1630 PM |
1642 | return 0; |
1643 | } | |
1644 | ||
21e52e15 PM |
1645 | /* |
1646 | * Handler for smp_call_function_single(). The only point of this | |
1647 | * handler is to wake the CPU up, so the handler does only tracing. | |
1648 | */ | |
1649 | void rcu_idle_demigrate(void *unused) | |
1650 | { | |
1651 | trace_rcu_prep_idle("Demigrate"); | |
1652 | } | |
1653 | ||
7cb92499 PM |
1654 | /* |
1655 | * Timer handler used to force CPU to start pushing its remaining RCU | |
1656 | * callbacks in the case where it entered dyntick-idle mode with callbacks | |
1657 | * pending. The hander doesn't really need to do anything because the | |
1658 | * real work is done upon re-entry to idle, or by the next scheduling-clock | |
1659 | * interrupt should idle not be re-entered. | |
21e52e15 PM |
1660 | * |
1661 | * One special case: the timer gets migrated without awakening the CPU | |
1662 | * on which the timer was scheduled on. In this case, we must wake up | |
1663 | * that CPU. We do so with smp_call_function_single(). | |
7cb92499 | 1664 | */ |
21e52e15 | 1665 | static void rcu_idle_gp_timer_func(unsigned long cpu_in) |
7cb92499 | 1666 | { |
21e52e15 PM |
1667 | int cpu = (int)cpu_in; |
1668 | ||
7cb92499 | 1669 | trace_rcu_prep_idle("Timer"); |
21e52e15 PM |
1670 | if (cpu != smp_processor_id()) |
1671 | smp_call_function_single(cpu, rcu_idle_demigrate, NULL, 0); | |
1672 | else | |
1673 | WARN_ON_ONCE(1); /* Getting here can hang the system... */ | |
7cb92499 PM |
1674 | } |
1675 | ||
1676 | /* | |
1677 | * Initialize the timer used to pull CPUs out of dyntick-idle mode. | |
1678 | */ | |
1679 | static void rcu_prepare_for_idle_init(int cpu) | |
1680 | { | |
5955f7ee PM |
1681 | struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); |
1682 | ||
1683 | rdtp->dyntick_holdoff = jiffies - 1; | |
1684 | setup_timer(&rdtp->idle_gp_timer, rcu_idle_gp_timer_func, cpu); | |
1685 | rdtp->idle_gp_timer_expires = jiffies - 1; | |
1686 | rdtp->idle_first_pass = 1; | |
7cb92499 PM |
1687 | } |
1688 | ||
1689 | /* | |
1690 | * Clean up for exit from idle. Because we are exiting from idle, there | |
5955f7ee | 1691 | * is no longer any point to ->idle_gp_timer, so cancel it. This will |
7cb92499 PM |
1692 | * do nothing if this timer is not active, so just cancel it unconditionally. |
1693 | */ | |
1694 | static void rcu_cleanup_after_idle(int cpu) | |
1695 | { | |
5955f7ee PM |
1696 | struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); |
1697 | ||
1698 | del_timer(&rdtp->idle_gp_timer); | |
2fdbb31b | 1699 | trace_rcu_prep_idle("Cleanup after idle"); |
9d2ad243 | 1700 | rdtp->tick_nohz_enabled_snap = ACCESS_ONCE(tick_nohz_enabled); |
7cb92499 PM |
1701 | } |
1702 | ||
aea1b35e PM |
1703 | /* |
1704 | * Check to see if any RCU-related work can be done by the current CPU, | |
1705 | * and if so, schedule a softirq to get it done. This function is part | |
1706 | * of the RCU implementation; it is -not- an exported member of the RCU API. | |
8bd93a2c | 1707 | * |
aea1b35e PM |
1708 | * The idea is for the current CPU to clear out all work required by the |
1709 | * RCU core for the current grace period, so that this CPU can be permitted | |
1710 | * to enter dyntick-idle mode. In some cases, it will need to be awakened | |
1711 | * at the end of the grace period by whatever CPU ends the grace period. | |
1712 | * This allows CPUs to go dyntick-idle more quickly, and to reduce the | |
1713 | * number of wakeups by a modest integer factor. | |
a47cd880 PM |
1714 | * |
1715 | * Because it is not legal to invoke rcu_process_callbacks() with irqs | |
1716 | * disabled, we do one pass of force_quiescent_state(), then do a | |
a46e0899 | 1717 | * invoke_rcu_core() to cause rcu_process_callbacks() to be invoked |
5955f7ee | 1718 | * later. The ->dyntick_drain field controls the sequencing. |
aea1b35e PM |
1719 | * |
1720 | * The caller must have disabled interrupts. | |
8bd93a2c | 1721 | */ |
aea1b35e | 1722 | static void rcu_prepare_for_idle(int cpu) |
8bd93a2c | 1723 | { |
f511fc62 | 1724 | struct timer_list *tp; |
5955f7ee | 1725 | struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); |
9d2ad243 PM |
1726 | int tne; |
1727 | ||
1728 | /* Handle nohz enablement switches conservatively. */ | |
1729 | tne = ACCESS_ONCE(tick_nohz_enabled); | |
1730 | if (tne != rdtp->tick_nohz_enabled_snap) { | |
1731 | if (rcu_cpu_has_callbacks(cpu)) | |
1732 | invoke_rcu_core(); /* force nohz to see update. */ | |
1733 | rdtp->tick_nohz_enabled_snap = tne; | |
1734 | return; | |
1735 | } | |
1736 | if (!tne) | |
1737 | return; | |
f511fc62 | 1738 | |
c57afe80 PM |
1739 | /* |
1740 | * If this is an idle re-entry, for example, due to use of | |
1741 | * RCU_NONIDLE() or the new idle-loop tracing API within the idle | |
1742 | * loop, then don't take any state-machine actions, unless the | |
1743 | * momentary exit from idle queued additional non-lazy callbacks. | |
5955f7ee | 1744 | * Instead, repost the ->idle_gp_timer if this CPU has callbacks |
c57afe80 PM |
1745 | * pending. |
1746 | */ | |
5955f7ee PM |
1747 | if (!rdtp->idle_first_pass && |
1748 | (rdtp->nonlazy_posted == rdtp->nonlazy_posted_snap)) { | |
f511fc62 | 1749 | if (rcu_cpu_has_callbacks(cpu)) { |
5955f7ee PM |
1750 | tp = &rdtp->idle_gp_timer; |
1751 | mod_timer_pinned(tp, rdtp->idle_gp_timer_expires); | |
f511fc62 | 1752 | } |
c57afe80 PM |
1753 | return; |
1754 | } | |
5955f7ee PM |
1755 | rdtp->idle_first_pass = 0; |
1756 | rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted - 1; | |
c57afe80 | 1757 | |
3084f2f8 | 1758 | /* |
f535a607 PM |
1759 | * If there are no callbacks on this CPU, enter dyntick-idle mode. |
1760 | * Also reset state to avoid prejudicing later attempts. | |
3084f2f8 | 1761 | */ |
aea1b35e | 1762 | if (!rcu_cpu_has_callbacks(cpu)) { |
5955f7ee PM |
1763 | rdtp->dyntick_holdoff = jiffies - 1; |
1764 | rdtp->dyntick_drain = 0; | |
433cdddc | 1765 | trace_rcu_prep_idle("No callbacks"); |
aea1b35e | 1766 | return; |
77e38ed3 | 1767 | } |
3084f2f8 PM |
1768 | |
1769 | /* | |
1770 | * If in holdoff mode, just return. We will presumably have | |
1771 | * refrained from disabling the scheduling-clock tick. | |
1772 | */ | |
5955f7ee | 1773 | if (rdtp->dyntick_holdoff == jiffies) { |
433cdddc | 1774 | trace_rcu_prep_idle("In holdoff"); |
aea1b35e | 1775 | return; |
433cdddc | 1776 | } |
a47cd880 | 1777 | |
5955f7ee PM |
1778 | /* Check and update the ->dyntick_drain sequencing. */ |
1779 | if (rdtp->dyntick_drain <= 0) { | |
a47cd880 | 1780 | /* First time through, initialize the counter. */ |
5955f7ee PM |
1781 | rdtp->dyntick_drain = RCU_IDLE_FLUSHES; |
1782 | } else if (rdtp->dyntick_drain <= RCU_IDLE_OPT_FLUSHES && | |
c3ce910b PM |
1783 | !rcu_pending(cpu) && |
1784 | !local_softirq_pending()) { | |
7cb92499 | 1785 | /* Can we go dyntick-idle despite still having callbacks? */ |
5955f7ee PM |
1786 | rdtp->dyntick_drain = 0; |
1787 | rdtp->dyntick_holdoff = jiffies; | |
fd4b3526 PM |
1788 | if (rcu_cpu_has_nonlazy_callbacks(cpu)) { |
1789 | trace_rcu_prep_idle("Dyntick with callbacks"); | |
5955f7ee | 1790 | rdtp->idle_gp_timer_expires = |
e84c48ae PM |
1791 | round_up(jiffies + RCU_IDLE_GP_DELAY, |
1792 | RCU_IDLE_GP_DELAY); | |
fd4b3526 | 1793 | } else { |
5955f7ee | 1794 | rdtp->idle_gp_timer_expires = |
e84c48ae | 1795 | round_jiffies(jiffies + RCU_IDLE_LAZY_GP_DELAY); |
fd4b3526 PM |
1796 | trace_rcu_prep_idle("Dyntick with lazy callbacks"); |
1797 | } | |
5955f7ee PM |
1798 | tp = &rdtp->idle_gp_timer; |
1799 | mod_timer_pinned(tp, rdtp->idle_gp_timer_expires); | |
1800 | rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted; | |
f23f7fa1 | 1801 | return; /* Nothing more to do immediately. */ |
5955f7ee | 1802 | } else if (--(rdtp->dyntick_drain) <= 0) { |
a47cd880 | 1803 | /* We have hit the limit, so time to give up. */ |
5955f7ee | 1804 | rdtp->dyntick_holdoff = jiffies; |
433cdddc | 1805 | trace_rcu_prep_idle("Begin holdoff"); |
aea1b35e PM |
1806 | invoke_rcu_core(); /* Force the CPU out of dyntick-idle. */ |
1807 | return; | |
a47cd880 PM |
1808 | } |
1809 | ||
aea1b35e PM |
1810 | /* |
1811 | * Do one step of pushing the remaining RCU callbacks through | |
1812 | * the RCU core state machine. | |
1813 | */ | |
1814 | #ifdef CONFIG_TREE_PREEMPT_RCU | |
1815 | if (per_cpu(rcu_preempt_data, cpu).nxtlist) { | |
1816 | rcu_preempt_qs(cpu); | |
1817 | force_quiescent_state(&rcu_preempt_state, 0); | |
aea1b35e PM |
1818 | } |
1819 | #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */ | |
a47cd880 PM |
1820 | if (per_cpu(rcu_sched_data, cpu).nxtlist) { |
1821 | rcu_sched_qs(cpu); | |
1822 | force_quiescent_state(&rcu_sched_state, 0); | |
a47cd880 PM |
1823 | } |
1824 | if (per_cpu(rcu_bh_data, cpu).nxtlist) { | |
1825 | rcu_bh_qs(cpu); | |
1826 | force_quiescent_state(&rcu_bh_state, 0); | |
8bd93a2c PM |
1827 | } |
1828 | ||
433cdddc PM |
1829 | /* |
1830 | * If RCU callbacks are still pending, RCU still needs this CPU. | |
1831 | * So try forcing the callbacks through the grace period. | |
1832 | */ | |
3ad0decf | 1833 | if (rcu_cpu_has_callbacks(cpu)) { |
433cdddc | 1834 | trace_rcu_prep_idle("More callbacks"); |
a46e0899 | 1835 | invoke_rcu_core(); |
c701d5d9 | 1836 | } else { |
433cdddc | 1837 | trace_rcu_prep_idle("Callbacks drained"); |
c701d5d9 | 1838 | } |
8bd93a2c PM |
1839 | } |
1840 | ||
c57afe80 | 1841 | /* |
98248a0e PM |
1842 | * Keep a running count of the number of non-lazy callbacks posted |
1843 | * on this CPU. This running counter (which is never decremented) allows | |
1844 | * rcu_prepare_for_idle() to detect when something out of the idle loop | |
1845 | * posts a callback, even if an equal number of callbacks are invoked. | |
1846 | * Of course, callbacks should only be posted from within a trace event | |
1847 | * designed to be called from idle or from within RCU_NONIDLE(). | |
c57afe80 PM |
1848 | */ |
1849 | static void rcu_idle_count_callbacks_posted(void) | |
1850 | { | |
5955f7ee | 1851 | __this_cpu_add(rcu_dynticks.nonlazy_posted, 1); |
c57afe80 PM |
1852 | } |
1853 | ||
8bd93a2c | 1854 | #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
a858af28 PM |
1855 | |
1856 | #ifdef CONFIG_RCU_CPU_STALL_INFO | |
1857 | ||
1858 | #ifdef CONFIG_RCU_FAST_NO_HZ | |
1859 | ||
1860 | static void print_cpu_stall_fast_no_hz(char *cp, int cpu) | |
1861 | { | |
5955f7ee PM |
1862 | struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); |
1863 | struct timer_list *tltp = &rdtp->idle_gp_timer; | |
a858af28 | 1864 | |
2ee3dc80 | 1865 | sprintf(cp, "drain=%d %c timer=%lu", |
5955f7ee PM |
1866 | rdtp->dyntick_drain, |
1867 | rdtp->dyntick_holdoff == jiffies ? 'H' : '.', | |
2ee3dc80 | 1868 | timer_pending(tltp) ? tltp->expires - jiffies : -1); |
a858af28 PM |
1869 | } |
1870 | ||
1871 | #else /* #ifdef CONFIG_RCU_FAST_NO_HZ */ | |
1872 | ||
1873 | static void print_cpu_stall_fast_no_hz(char *cp, int cpu) | |
1874 | { | |
1c17e4d4 | 1875 | *cp = '\0'; |
a858af28 PM |
1876 | } |
1877 | ||
1878 | #endif /* #else #ifdef CONFIG_RCU_FAST_NO_HZ */ | |
1879 | ||
1880 | /* Initiate the stall-info list. */ | |
1881 | static void print_cpu_stall_info_begin(void) | |
1882 | { | |
1883 | printk(KERN_CONT "\n"); | |
1884 | } | |
1885 | ||
1886 | /* | |
1887 | * Print out diagnostic information for the specified stalled CPU. | |
1888 | * | |
1889 | * If the specified CPU is aware of the current RCU grace period | |
1890 | * (flavor specified by rsp), then print the number of scheduling | |
1891 | * clock interrupts the CPU has taken during the time that it has | |
1892 | * been aware. Otherwise, print the number of RCU grace periods | |
1893 | * that this CPU is ignorant of, for example, "1" if the CPU was | |
1894 | * aware of the previous grace period. | |
1895 | * | |
1896 | * Also print out idle and (if CONFIG_RCU_FAST_NO_HZ) idle-entry info. | |
1897 | */ | |
1898 | static void print_cpu_stall_info(struct rcu_state *rsp, int cpu) | |
1899 | { | |
1900 | char fast_no_hz[72]; | |
1901 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); | |
1902 | struct rcu_dynticks *rdtp = rdp->dynticks; | |
1903 | char *ticks_title; | |
1904 | unsigned long ticks_value; | |
1905 | ||
1906 | if (rsp->gpnum == rdp->gpnum) { | |
1907 | ticks_title = "ticks this GP"; | |
1908 | ticks_value = rdp->ticks_this_gp; | |
1909 | } else { | |
1910 | ticks_title = "GPs behind"; | |
1911 | ticks_value = rsp->gpnum - rdp->gpnum; | |
1912 | } | |
1913 | print_cpu_stall_fast_no_hz(fast_no_hz, cpu); | |
1914 | printk(KERN_ERR "\t%d: (%lu %s) idle=%03x/%llx/%d %s\n", | |
1915 | cpu, ticks_value, ticks_title, | |
1916 | atomic_read(&rdtp->dynticks) & 0xfff, | |
1917 | rdtp->dynticks_nesting, rdtp->dynticks_nmi_nesting, | |
1918 | fast_no_hz); | |
1919 | } | |
1920 | ||
1921 | /* Terminate the stall-info list. */ | |
1922 | static void print_cpu_stall_info_end(void) | |
1923 | { | |
1924 | printk(KERN_ERR "\t"); | |
1925 | } | |
1926 | ||
1927 | /* Zero ->ticks_this_gp for all flavors of RCU. */ | |
1928 | static void zero_cpu_stall_ticks(struct rcu_data *rdp) | |
1929 | { | |
1930 | rdp->ticks_this_gp = 0; | |
1931 | } | |
1932 | ||
1933 | /* Increment ->ticks_this_gp for all flavors of RCU. */ | |
1934 | static void increment_cpu_stall_ticks(void) | |
1935 | { | |
1936 | __get_cpu_var(rcu_sched_data).ticks_this_gp++; | |
1937 | __get_cpu_var(rcu_bh_data).ticks_this_gp++; | |
1938 | #ifdef CONFIG_TREE_PREEMPT_RCU | |
1939 | __get_cpu_var(rcu_preempt_data).ticks_this_gp++; | |
1940 | #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */ | |
1941 | } | |
1942 | ||
1943 | #else /* #ifdef CONFIG_RCU_CPU_STALL_INFO */ | |
1944 | ||
1945 | static void print_cpu_stall_info_begin(void) | |
1946 | { | |
1947 | printk(KERN_CONT " {"); | |
1948 | } | |
1949 | ||
1950 | static void print_cpu_stall_info(struct rcu_state *rsp, int cpu) | |
1951 | { | |
1952 | printk(KERN_CONT " %d", cpu); | |
1953 | } | |
1954 | ||
1955 | static void print_cpu_stall_info_end(void) | |
1956 | { | |
1957 | printk(KERN_CONT "} "); | |
1958 | } | |
1959 | ||
1960 | static void zero_cpu_stall_ticks(struct rcu_data *rdp) | |
1961 | { | |
1962 | } | |
1963 | ||
1964 | static void increment_cpu_stall_ticks(void) | |
1965 | { | |
1966 | } | |
1967 | ||
1968 | #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_INFO */ |