]>
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> |
3fbfbf7a | 28 | #include <linux/gfp.h> |
b626c1b6 | 29 | #include <linux/oom.h> |
62ab7072 | 30 | #include <linux/smpboot.h> |
0edd1b17 | 31 | #include "time/tick-internal.h" |
f41d911f | 32 | |
5b61b0ba MG |
33 | #define RCU_KTHREAD_PRIO 1 |
34 | ||
35 | #ifdef CONFIG_RCU_BOOST | |
36 | #define RCU_BOOST_PRIO CONFIG_RCU_BOOST_PRIO | |
37 | #else | |
38 | #define RCU_BOOST_PRIO RCU_KTHREAD_PRIO | |
39 | #endif | |
40 | ||
3fbfbf7a PM |
41 | #ifdef CONFIG_RCU_NOCB_CPU |
42 | static cpumask_var_t rcu_nocb_mask; /* CPUs to have callbacks offloaded. */ | |
43 | static bool have_rcu_nocb_mask; /* Was rcu_nocb_mask allocated? */ | |
1b0048a4 | 44 | static bool __read_mostly rcu_nocb_poll; /* Offload kthread are to poll. */ |
3fbfbf7a PM |
45 | static char __initdata nocb_buf[NR_CPUS * 5]; |
46 | #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ | |
47 | ||
26845c28 PM |
48 | /* |
49 | * Check the RCU kernel configuration parameters and print informative | |
50 | * messages about anything out of the ordinary. If you like #ifdef, you | |
51 | * will love this function. | |
52 | */ | |
53 | static void __init rcu_bootup_announce_oddness(void) | |
54 | { | |
55 | #ifdef CONFIG_RCU_TRACE | |
efc151c3 | 56 | pr_info("\tRCU debugfs-based tracing is enabled.\n"); |
26845c28 PM |
57 | #endif |
58 | #if (defined(CONFIG_64BIT) && CONFIG_RCU_FANOUT != 64) || (!defined(CONFIG_64BIT) && CONFIG_RCU_FANOUT != 32) | |
efc151c3 | 59 | pr_info("\tCONFIG_RCU_FANOUT set to non-default value of %d\n", |
26845c28 PM |
60 | CONFIG_RCU_FANOUT); |
61 | #endif | |
62 | #ifdef CONFIG_RCU_FANOUT_EXACT | |
efc151c3 | 63 | pr_info("\tHierarchical RCU autobalancing is disabled.\n"); |
26845c28 PM |
64 | #endif |
65 | #ifdef CONFIG_RCU_FAST_NO_HZ | |
efc151c3 | 66 | pr_info("\tRCU dyntick-idle grace-period acceleration is enabled.\n"); |
26845c28 PM |
67 | #endif |
68 | #ifdef CONFIG_PROVE_RCU | |
efc151c3 | 69 | pr_info("\tRCU lockdep checking is enabled.\n"); |
26845c28 PM |
70 | #endif |
71 | #ifdef CONFIG_RCU_TORTURE_TEST_RUNNABLE | |
efc151c3 | 72 | pr_info("\tRCU torture testing starts during boot.\n"); |
26845c28 | 73 | #endif |
81a294c4 | 74 | #if defined(CONFIG_TREE_PREEMPT_RCU) && !defined(CONFIG_RCU_CPU_STALL_VERBOSE) |
efc151c3 | 75 | pr_info("\tDump stacks of tasks blocking RCU-preempt GP.\n"); |
a858af28 PM |
76 | #endif |
77 | #if defined(CONFIG_RCU_CPU_STALL_INFO) | |
efc151c3 | 78 | pr_info("\tAdditional per-CPU info printed with stalls.\n"); |
26845c28 PM |
79 | #endif |
80 | #if NUM_RCU_LVL_4 != 0 | |
efc151c3 | 81 | pr_info("\tFour-level hierarchy is enabled.\n"); |
26845c28 | 82 | #endif |
f885b7f2 | 83 | if (rcu_fanout_leaf != CONFIG_RCU_FANOUT_LEAF) |
9a5739d7 | 84 | pr_info("\tBoot-time adjustment of leaf fanout to %d.\n", rcu_fanout_leaf); |
cca6f393 | 85 | if (nr_cpu_ids != NR_CPUS) |
efc151c3 | 86 | pr_info("\tRCU restricting CPUs from NR_CPUS=%d to nr_cpu_ids=%d.\n", NR_CPUS, nr_cpu_ids); |
3fbfbf7a | 87 | #ifdef CONFIG_RCU_NOCB_CPU |
911af505 PM |
88 | #ifndef CONFIG_RCU_NOCB_CPU_NONE |
89 | if (!have_rcu_nocb_mask) { | |
615ee544 | 90 | zalloc_cpumask_var(&rcu_nocb_mask, GFP_KERNEL); |
911af505 PM |
91 | have_rcu_nocb_mask = true; |
92 | } | |
93 | #ifdef CONFIG_RCU_NOCB_CPU_ZERO | |
9a5739d7 | 94 | pr_info("\tOffload RCU callbacks from CPU 0\n"); |
911af505 PM |
95 | cpumask_set_cpu(0, rcu_nocb_mask); |
96 | #endif /* #ifdef CONFIG_RCU_NOCB_CPU_ZERO */ | |
97 | #ifdef CONFIG_RCU_NOCB_CPU_ALL | |
9a5739d7 | 98 | pr_info("\tOffload RCU callbacks from all CPUs\n"); |
5d5a0800 | 99 | cpumask_copy(rcu_nocb_mask, cpu_possible_mask); |
911af505 PM |
100 | #endif /* #ifdef CONFIG_RCU_NOCB_CPU_ALL */ |
101 | #endif /* #ifndef CONFIG_RCU_NOCB_CPU_NONE */ | |
3fbfbf7a | 102 | if (have_rcu_nocb_mask) { |
5d5a0800 KT |
103 | if (!cpumask_subset(rcu_nocb_mask, cpu_possible_mask)) { |
104 | pr_info("\tNote: kernel parameter 'rcu_nocbs=' contains nonexistent CPUs.\n"); | |
105 | cpumask_and(rcu_nocb_mask, cpu_possible_mask, | |
106 | rcu_nocb_mask); | |
107 | } | |
3fbfbf7a | 108 | cpulist_scnprintf(nocb_buf, sizeof(nocb_buf), rcu_nocb_mask); |
9a5739d7 | 109 | pr_info("\tOffload RCU callbacks from CPUs: %s.\n", nocb_buf); |
3fbfbf7a | 110 | if (rcu_nocb_poll) |
9a5739d7 | 111 | pr_info("\tPoll for callbacks from no-CBs CPUs.\n"); |
3fbfbf7a PM |
112 | } |
113 | #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ | |
26845c28 PM |
114 | } |
115 | ||
f41d911f PM |
116 | #ifdef CONFIG_TREE_PREEMPT_RCU |
117 | ||
a41bfeb2 | 118 | RCU_STATE_INITIALIZER(rcu_preempt, 'p', call_rcu); |
27f4d280 | 119 | static struct rcu_state *rcu_state = &rcu_preempt_state; |
f41d911f | 120 | |
d9a3da06 PM |
121 | static int rcu_preempted_readers_exp(struct rcu_node *rnp); |
122 | ||
f41d911f PM |
123 | /* |
124 | * Tell them what RCU they are running. | |
125 | */ | |
0e0fc1c2 | 126 | static void __init rcu_bootup_announce(void) |
f41d911f | 127 | { |
efc151c3 | 128 | pr_info("Preemptible hierarchical RCU implementation.\n"); |
26845c28 | 129 | rcu_bootup_announce_oddness(); |
f41d911f PM |
130 | } |
131 | ||
132 | /* | |
133 | * Return the number of RCU-preempt batches processed thus far | |
134 | * for debug and statistics. | |
135 | */ | |
136 | long rcu_batches_completed_preempt(void) | |
137 | { | |
138 | return rcu_preempt_state.completed; | |
139 | } | |
140 | EXPORT_SYMBOL_GPL(rcu_batches_completed_preempt); | |
141 | ||
142 | /* | |
143 | * Return the number of RCU batches processed thus far for debug & stats. | |
144 | */ | |
145 | long rcu_batches_completed(void) | |
146 | { | |
147 | return rcu_batches_completed_preempt(); | |
148 | } | |
149 | EXPORT_SYMBOL_GPL(rcu_batches_completed); | |
150 | ||
bf66f18e PM |
151 | /* |
152 | * Force a quiescent state for preemptible RCU. | |
153 | */ | |
154 | void rcu_force_quiescent_state(void) | |
155 | { | |
4cdfc175 | 156 | force_quiescent_state(&rcu_preempt_state); |
bf66f18e PM |
157 | } |
158 | EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); | |
159 | ||
f41d911f | 160 | /* |
6cc68793 | 161 | * Record a preemptible-RCU quiescent state for the specified CPU. Note |
f41d911f PM |
162 | * that this just means that the task currently running on the CPU is |
163 | * not in a quiescent state. There might be any number of tasks blocked | |
164 | * while in an RCU read-side critical section. | |
25502a6c PM |
165 | * |
166 | * Unlike the other rcu_*_qs() functions, callers to this function | |
167 | * must disable irqs in order to protect the assignment to | |
168 | * ->rcu_read_unlock_special. | |
f41d911f | 169 | */ |
c3422bea | 170 | static void rcu_preempt_qs(int cpu) |
f41d911f PM |
171 | { |
172 | struct rcu_data *rdp = &per_cpu(rcu_preempt_data, cpu); | |
25502a6c | 173 | |
e4cc1f22 | 174 | if (rdp->passed_quiesce == 0) |
f7f7bac9 | 175 | trace_rcu_grace_period(TPS("rcu_preempt"), rdp->gpnum, TPS("cpuqs")); |
e4cc1f22 | 176 | rdp->passed_quiesce = 1; |
25502a6c | 177 | current->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS; |
f41d911f PM |
178 | } |
179 | ||
180 | /* | |
c3422bea PM |
181 | * We have entered the scheduler, and the current task might soon be |
182 | * context-switched away from. If this task is in an RCU read-side | |
183 | * critical section, we will no longer be able to rely on the CPU to | |
12f5f524 PM |
184 | * record that fact, so we enqueue the task on the blkd_tasks list. |
185 | * The task will dequeue itself when it exits the outermost enclosing | |
186 | * RCU read-side critical section. Therefore, the current grace period | |
187 | * cannot be permitted to complete until the blkd_tasks list entries | |
188 | * predating the current grace period drain, in other words, until | |
189 | * rnp->gp_tasks becomes NULL. | |
c3422bea PM |
190 | * |
191 | * Caller must disable preemption. | |
f41d911f | 192 | */ |
cba6d0d6 | 193 | static void rcu_preempt_note_context_switch(int cpu) |
f41d911f PM |
194 | { |
195 | struct task_struct *t = current; | |
c3422bea | 196 | unsigned long flags; |
f41d911f PM |
197 | struct rcu_data *rdp; |
198 | struct rcu_node *rnp; | |
199 | ||
10f39bb1 | 200 | if (t->rcu_read_lock_nesting > 0 && |
f41d911f PM |
201 | (t->rcu_read_unlock_special & RCU_READ_UNLOCK_BLOCKED) == 0) { |
202 | ||
203 | /* Possibly blocking in an RCU read-side critical section. */ | |
cba6d0d6 | 204 | rdp = per_cpu_ptr(rcu_preempt_state.rda, cpu); |
f41d911f | 205 | rnp = rdp->mynode; |
1304afb2 | 206 | raw_spin_lock_irqsave(&rnp->lock, flags); |
f41d911f | 207 | t->rcu_read_unlock_special |= RCU_READ_UNLOCK_BLOCKED; |
86848966 | 208 | t->rcu_blocked_node = rnp; |
f41d911f PM |
209 | |
210 | /* | |
211 | * If this CPU has already checked in, then this task | |
212 | * will hold up the next grace period rather than the | |
213 | * current grace period. Queue the task accordingly. | |
214 | * If the task is queued for the current grace period | |
215 | * (i.e., this CPU has not yet passed through a quiescent | |
216 | * state for the current grace period), then as long | |
217 | * as that task remains queued, the current grace period | |
12f5f524 PM |
218 | * cannot end. Note that there is some uncertainty as |
219 | * to exactly when the current grace period started. | |
220 | * We take a conservative approach, which can result | |
221 | * in unnecessarily waiting on tasks that started very | |
222 | * slightly after the current grace period began. C'est | |
223 | * la vie!!! | |
b0e165c0 PM |
224 | * |
225 | * But first, note that the current CPU must still be | |
226 | * on line! | |
f41d911f | 227 | */ |
b0e165c0 | 228 | WARN_ON_ONCE((rdp->grpmask & rnp->qsmaskinit) == 0); |
e7d8842e | 229 | WARN_ON_ONCE(!list_empty(&t->rcu_node_entry)); |
12f5f524 PM |
230 | if ((rnp->qsmask & rdp->grpmask) && rnp->gp_tasks != NULL) { |
231 | list_add(&t->rcu_node_entry, rnp->gp_tasks->prev); | |
232 | rnp->gp_tasks = &t->rcu_node_entry; | |
27f4d280 PM |
233 | #ifdef CONFIG_RCU_BOOST |
234 | if (rnp->boost_tasks != NULL) | |
235 | rnp->boost_tasks = rnp->gp_tasks; | |
236 | #endif /* #ifdef CONFIG_RCU_BOOST */ | |
12f5f524 PM |
237 | } else { |
238 | list_add(&t->rcu_node_entry, &rnp->blkd_tasks); | |
239 | if (rnp->qsmask & rdp->grpmask) | |
240 | rnp->gp_tasks = &t->rcu_node_entry; | |
241 | } | |
d4c08f2a PM |
242 | trace_rcu_preempt_task(rdp->rsp->name, |
243 | t->pid, | |
244 | (rnp->qsmask & rdp->grpmask) | |
245 | ? rnp->gpnum | |
246 | : rnp->gpnum + 1); | |
1304afb2 | 247 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
10f39bb1 PM |
248 | } else if (t->rcu_read_lock_nesting < 0 && |
249 | t->rcu_read_unlock_special) { | |
250 | ||
251 | /* | |
252 | * Complete exit from RCU read-side critical section on | |
253 | * behalf of preempted instance of __rcu_read_unlock(). | |
254 | */ | |
255 | rcu_read_unlock_special(t); | |
f41d911f PM |
256 | } |
257 | ||
258 | /* | |
259 | * Either we were not in an RCU read-side critical section to | |
260 | * begin with, or we have now recorded that critical section | |
261 | * globally. Either way, we can now note a quiescent state | |
262 | * for this CPU. Again, if we were in an RCU read-side critical | |
263 | * section, and if that critical section was blocking the current | |
264 | * grace period, then the fact that the task has been enqueued | |
265 | * means that we continue to block the current grace period. | |
266 | */ | |
e7d8842e | 267 | local_irq_save(flags); |
cba6d0d6 | 268 | rcu_preempt_qs(cpu); |
e7d8842e | 269 | local_irq_restore(flags); |
f41d911f PM |
270 | } |
271 | ||
fc2219d4 PM |
272 | /* |
273 | * Check for preempted RCU readers blocking the current grace period | |
274 | * for the specified rcu_node structure. If the caller needs a reliable | |
275 | * answer, it must hold the rcu_node's ->lock. | |
276 | */ | |
27f4d280 | 277 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 | 278 | { |
12f5f524 | 279 | return rnp->gp_tasks != NULL; |
fc2219d4 PM |
280 | } |
281 | ||
b668c9cf PM |
282 | /* |
283 | * Record a quiescent state for all tasks that were previously queued | |
284 | * on the specified rcu_node structure and that were blocking the current | |
285 | * RCU grace period. The caller must hold the specified rnp->lock with | |
286 | * irqs disabled, and this lock is released upon return, but irqs remain | |
287 | * disabled. | |
288 | */ | |
d3f6bad3 | 289 | static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags) |
b668c9cf PM |
290 | __releases(rnp->lock) |
291 | { | |
292 | unsigned long mask; | |
293 | struct rcu_node *rnp_p; | |
294 | ||
27f4d280 | 295 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
1304afb2 | 296 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
b668c9cf PM |
297 | return; /* Still need more quiescent states! */ |
298 | } | |
299 | ||
300 | rnp_p = rnp->parent; | |
301 | if (rnp_p == NULL) { | |
302 | /* | |
303 | * Either there is only one rcu_node in the tree, | |
304 | * or tasks were kicked up to root rcu_node due to | |
305 | * CPUs going offline. | |
306 | */ | |
d3f6bad3 | 307 | rcu_report_qs_rsp(&rcu_preempt_state, flags); |
b668c9cf PM |
308 | return; |
309 | } | |
310 | ||
311 | /* Report up the rest of the hierarchy. */ | |
312 | mask = rnp->grpmask; | |
1304afb2 PM |
313 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
314 | raw_spin_lock(&rnp_p->lock); /* irqs already disabled. */ | |
d3f6bad3 | 315 | rcu_report_qs_rnp(mask, &rcu_preempt_state, rnp_p, flags); |
b668c9cf PM |
316 | } |
317 | ||
12f5f524 PM |
318 | /* |
319 | * Advance a ->blkd_tasks-list pointer to the next entry, instead | |
320 | * returning NULL if at the end of the list. | |
321 | */ | |
322 | static struct list_head *rcu_next_node_entry(struct task_struct *t, | |
323 | struct rcu_node *rnp) | |
324 | { | |
325 | struct list_head *np; | |
326 | ||
327 | np = t->rcu_node_entry.next; | |
328 | if (np == &rnp->blkd_tasks) | |
329 | np = NULL; | |
330 | return np; | |
331 | } | |
332 | ||
b668c9cf PM |
333 | /* |
334 | * Handle special cases during rcu_read_unlock(), such as needing to | |
335 | * notify RCU core processing or task having blocked during the RCU | |
336 | * read-side critical section. | |
337 | */ | |
2a3fa843 | 338 | void rcu_read_unlock_special(struct task_struct *t) |
f41d911f PM |
339 | { |
340 | int empty; | |
d9a3da06 | 341 | int empty_exp; |
389abd48 | 342 | int empty_exp_now; |
f41d911f | 343 | unsigned long flags; |
12f5f524 | 344 | struct list_head *np; |
82e78d80 PM |
345 | #ifdef CONFIG_RCU_BOOST |
346 | struct rt_mutex *rbmp = NULL; | |
347 | #endif /* #ifdef CONFIG_RCU_BOOST */ | |
f41d911f PM |
348 | struct rcu_node *rnp; |
349 | int special; | |
350 | ||
351 | /* NMI handlers cannot block and cannot safely manipulate state. */ | |
352 | if (in_nmi()) | |
353 | return; | |
354 | ||
355 | local_irq_save(flags); | |
356 | ||
357 | /* | |
358 | * If RCU core is waiting for this CPU to exit critical section, | |
359 | * let it know that we have done so. | |
360 | */ | |
361 | special = t->rcu_read_unlock_special; | |
362 | if (special & RCU_READ_UNLOCK_NEED_QS) { | |
c3422bea | 363 | rcu_preempt_qs(smp_processor_id()); |
f41d911f PM |
364 | } |
365 | ||
366 | /* Hardware IRQ handlers cannot block. */ | |
ec433f0c | 367 | if (in_irq() || in_serving_softirq()) { |
f41d911f PM |
368 | local_irq_restore(flags); |
369 | return; | |
370 | } | |
371 | ||
372 | /* Clean up if blocked during RCU read-side critical section. */ | |
373 | if (special & RCU_READ_UNLOCK_BLOCKED) { | |
374 | t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_BLOCKED; | |
375 | ||
dd5d19ba PM |
376 | /* |
377 | * Remove this task from the list it blocked on. The | |
378 | * task can migrate while we acquire the lock, but at | |
379 | * most one time. So at most two passes through loop. | |
380 | */ | |
381 | for (;;) { | |
86848966 | 382 | rnp = t->rcu_blocked_node; |
1304afb2 | 383 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
86848966 | 384 | if (rnp == t->rcu_blocked_node) |
dd5d19ba | 385 | break; |
1304afb2 | 386 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
dd5d19ba | 387 | } |
27f4d280 | 388 | empty = !rcu_preempt_blocked_readers_cgp(rnp); |
d9a3da06 PM |
389 | empty_exp = !rcu_preempted_readers_exp(rnp); |
390 | smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */ | |
12f5f524 | 391 | np = rcu_next_node_entry(t, rnp); |
f41d911f | 392 | list_del_init(&t->rcu_node_entry); |
82e78d80 | 393 | t->rcu_blocked_node = NULL; |
f7f7bac9 | 394 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt"), |
d4c08f2a | 395 | rnp->gpnum, t->pid); |
12f5f524 PM |
396 | if (&t->rcu_node_entry == rnp->gp_tasks) |
397 | rnp->gp_tasks = np; | |
398 | if (&t->rcu_node_entry == rnp->exp_tasks) | |
399 | rnp->exp_tasks = np; | |
27f4d280 PM |
400 | #ifdef CONFIG_RCU_BOOST |
401 | if (&t->rcu_node_entry == rnp->boost_tasks) | |
402 | rnp->boost_tasks = np; | |
82e78d80 PM |
403 | /* Snapshot/clear ->rcu_boost_mutex with rcu_node lock held. */ |
404 | if (t->rcu_boost_mutex) { | |
405 | rbmp = t->rcu_boost_mutex; | |
406 | t->rcu_boost_mutex = NULL; | |
7765be2f | 407 | } |
27f4d280 | 408 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
f41d911f PM |
409 | |
410 | /* | |
411 | * If this was the last task on the current list, and if | |
412 | * we aren't waiting on any CPUs, report the quiescent state. | |
389abd48 PM |
413 | * Note that rcu_report_unblock_qs_rnp() releases rnp->lock, |
414 | * so we must take a snapshot of the expedited state. | |
f41d911f | 415 | */ |
389abd48 | 416 | empty_exp_now = !rcu_preempted_readers_exp(rnp); |
d4c08f2a | 417 | if (!empty && !rcu_preempt_blocked_readers_cgp(rnp)) { |
f7f7bac9 | 418 | trace_rcu_quiescent_state_report(TPS("preempt_rcu"), |
d4c08f2a PM |
419 | rnp->gpnum, |
420 | 0, rnp->qsmask, | |
421 | rnp->level, | |
422 | rnp->grplo, | |
423 | rnp->grphi, | |
424 | !!rnp->gp_tasks); | |
d3f6bad3 | 425 | rcu_report_unblock_qs_rnp(rnp, flags); |
c701d5d9 | 426 | } else { |
d4c08f2a | 427 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
c701d5d9 | 428 | } |
d9a3da06 | 429 | |
27f4d280 PM |
430 | #ifdef CONFIG_RCU_BOOST |
431 | /* Unboost if we were boosted. */ | |
82e78d80 PM |
432 | if (rbmp) |
433 | rt_mutex_unlock(rbmp); | |
27f4d280 PM |
434 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
435 | ||
d9a3da06 PM |
436 | /* |
437 | * If this was the last task on the expedited lists, | |
438 | * then we need to report up the rcu_node hierarchy. | |
439 | */ | |
389abd48 | 440 | if (!empty_exp && empty_exp_now) |
b40d293e | 441 | rcu_report_exp_rnp(&rcu_preempt_state, rnp, true); |
b668c9cf PM |
442 | } else { |
443 | local_irq_restore(flags); | |
f41d911f | 444 | } |
f41d911f PM |
445 | } |
446 | ||
1ed509a2 PM |
447 | #ifdef CONFIG_RCU_CPU_STALL_VERBOSE |
448 | ||
449 | /* | |
450 | * Dump detailed information for all tasks blocking the current RCU | |
451 | * grace period on the specified rcu_node structure. | |
452 | */ | |
453 | static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp) | |
454 | { | |
455 | unsigned long flags; | |
1ed509a2 PM |
456 | struct task_struct *t; |
457 | ||
12f5f524 | 458 | raw_spin_lock_irqsave(&rnp->lock, flags); |
5fd4dc06 PM |
459 | if (!rcu_preempt_blocked_readers_cgp(rnp)) { |
460 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
461 | return; | |
462 | } | |
12f5f524 PM |
463 | t = list_entry(rnp->gp_tasks, |
464 | struct task_struct, rcu_node_entry); | |
465 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) | |
466 | sched_show_task(t); | |
467 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
1ed509a2 PM |
468 | } |
469 | ||
470 | /* | |
471 | * Dump detailed information for all tasks blocking the current RCU | |
472 | * grace period. | |
473 | */ | |
474 | static void rcu_print_detail_task_stall(struct rcu_state *rsp) | |
475 | { | |
476 | struct rcu_node *rnp = rcu_get_root(rsp); | |
477 | ||
478 | rcu_print_detail_task_stall_rnp(rnp); | |
479 | rcu_for_each_leaf_node(rsp, rnp) | |
480 | rcu_print_detail_task_stall_rnp(rnp); | |
481 | } | |
482 | ||
483 | #else /* #ifdef CONFIG_RCU_CPU_STALL_VERBOSE */ | |
484 | ||
485 | static void rcu_print_detail_task_stall(struct rcu_state *rsp) | |
486 | { | |
487 | } | |
488 | ||
489 | #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_VERBOSE */ | |
490 | ||
a858af28 PM |
491 | #ifdef CONFIG_RCU_CPU_STALL_INFO |
492 | ||
493 | static void rcu_print_task_stall_begin(struct rcu_node *rnp) | |
494 | { | |
efc151c3 | 495 | pr_err("\tTasks blocked on level-%d rcu_node (CPUs %d-%d):", |
a858af28 PM |
496 | rnp->level, rnp->grplo, rnp->grphi); |
497 | } | |
498 | ||
499 | static void rcu_print_task_stall_end(void) | |
500 | { | |
efc151c3 | 501 | pr_cont("\n"); |
a858af28 PM |
502 | } |
503 | ||
504 | #else /* #ifdef CONFIG_RCU_CPU_STALL_INFO */ | |
505 | ||
506 | static void rcu_print_task_stall_begin(struct rcu_node *rnp) | |
507 | { | |
508 | } | |
509 | ||
510 | static void rcu_print_task_stall_end(void) | |
511 | { | |
512 | } | |
513 | ||
514 | #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_INFO */ | |
515 | ||
f41d911f PM |
516 | /* |
517 | * Scan the current list of tasks blocked within RCU read-side critical | |
518 | * sections, printing out the tid of each. | |
519 | */ | |
9bc8b558 | 520 | static int rcu_print_task_stall(struct rcu_node *rnp) |
f41d911f | 521 | { |
f41d911f | 522 | struct task_struct *t; |
9bc8b558 | 523 | int ndetected = 0; |
f41d911f | 524 | |
27f4d280 | 525 | if (!rcu_preempt_blocked_readers_cgp(rnp)) |
9bc8b558 | 526 | return 0; |
a858af28 | 527 | rcu_print_task_stall_begin(rnp); |
12f5f524 PM |
528 | t = list_entry(rnp->gp_tasks, |
529 | struct task_struct, rcu_node_entry); | |
9bc8b558 | 530 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) { |
efc151c3 | 531 | pr_cont(" P%d", t->pid); |
9bc8b558 PM |
532 | ndetected++; |
533 | } | |
a858af28 | 534 | rcu_print_task_stall_end(); |
9bc8b558 | 535 | return ndetected; |
f41d911f PM |
536 | } |
537 | ||
b0e165c0 PM |
538 | /* |
539 | * Check that the list of blocked tasks for the newly completed grace | |
540 | * period is in fact empty. It is a serious bug to complete a grace | |
541 | * period that still has RCU readers blocked! This function must be | |
542 | * invoked -before- updating this rnp's ->gpnum, and the rnp's ->lock | |
543 | * must be held by the caller. | |
12f5f524 PM |
544 | * |
545 | * Also, if there are blocked tasks on the list, they automatically | |
546 | * block the newly created grace period, so set up ->gp_tasks accordingly. | |
b0e165c0 PM |
547 | */ |
548 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) | |
549 | { | |
27f4d280 | 550 | WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)); |
12f5f524 PM |
551 | if (!list_empty(&rnp->blkd_tasks)) |
552 | rnp->gp_tasks = rnp->blkd_tasks.next; | |
28ecd580 | 553 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
554 | } |
555 | ||
33f76148 PM |
556 | #ifdef CONFIG_HOTPLUG_CPU |
557 | ||
dd5d19ba PM |
558 | /* |
559 | * Handle tasklist migration for case in which all CPUs covered by the | |
560 | * specified rcu_node have gone offline. Move them up to the root | |
561 | * rcu_node. The reason for not just moving them to the immediate | |
562 | * parent is to remove the need for rcu_read_unlock_special() to | |
563 | * make more than two attempts to acquire the target rcu_node's lock. | |
b668c9cf PM |
564 | * Returns true if there were tasks blocking the current RCU grace |
565 | * period. | |
dd5d19ba | 566 | * |
237c80c5 PM |
567 | * Returns 1 if there was previously a task blocking the current grace |
568 | * period on the specified rcu_node structure. | |
569 | * | |
dd5d19ba PM |
570 | * The caller must hold rnp->lock with irqs disabled. |
571 | */ | |
237c80c5 PM |
572 | static int rcu_preempt_offline_tasks(struct rcu_state *rsp, |
573 | struct rcu_node *rnp, | |
574 | struct rcu_data *rdp) | |
dd5d19ba | 575 | { |
dd5d19ba PM |
576 | struct list_head *lp; |
577 | struct list_head *lp_root; | |
d9a3da06 | 578 | int retval = 0; |
dd5d19ba | 579 | struct rcu_node *rnp_root = rcu_get_root(rsp); |
12f5f524 | 580 | struct task_struct *t; |
dd5d19ba | 581 | |
86848966 PM |
582 | if (rnp == rnp_root) { |
583 | WARN_ONCE(1, "Last CPU thought to be offlined?"); | |
237c80c5 | 584 | return 0; /* Shouldn't happen: at least one CPU online. */ |
86848966 | 585 | } |
12f5f524 PM |
586 | |
587 | /* If we are on an internal node, complain bitterly. */ | |
588 | WARN_ON_ONCE(rnp != rdp->mynode); | |
dd5d19ba PM |
589 | |
590 | /* | |
12f5f524 PM |
591 | * Move tasks up to root rcu_node. Don't try to get fancy for |
592 | * this corner-case operation -- just put this node's tasks | |
593 | * at the head of the root node's list, and update the root node's | |
594 | * ->gp_tasks and ->exp_tasks pointers to those of this node's, | |
595 | * if non-NULL. This might result in waiting for more tasks than | |
596 | * absolutely necessary, but this is a good performance/complexity | |
597 | * tradeoff. | |
dd5d19ba | 598 | */ |
2036d94a | 599 | if (rcu_preempt_blocked_readers_cgp(rnp) && rnp->qsmask == 0) |
d9a3da06 PM |
600 | retval |= RCU_OFL_TASKS_NORM_GP; |
601 | if (rcu_preempted_readers_exp(rnp)) | |
602 | retval |= RCU_OFL_TASKS_EXP_GP; | |
12f5f524 PM |
603 | lp = &rnp->blkd_tasks; |
604 | lp_root = &rnp_root->blkd_tasks; | |
605 | while (!list_empty(lp)) { | |
606 | t = list_entry(lp->next, typeof(*t), rcu_node_entry); | |
607 | raw_spin_lock(&rnp_root->lock); /* irqs already disabled */ | |
608 | list_del(&t->rcu_node_entry); | |
609 | t->rcu_blocked_node = rnp_root; | |
610 | list_add(&t->rcu_node_entry, lp_root); | |
611 | if (&t->rcu_node_entry == rnp->gp_tasks) | |
612 | rnp_root->gp_tasks = rnp->gp_tasks; | |
613 | if (&t->rcu_node_entry == rnp->exp_tasks) | |
614 | rnp_root->exp_tasks = rnp->exp_tasks; | |
27f4d280 PM |
615 | #ifdef CONFIG_RCU_BOOST |
616 | if (&t->rcu_node_entry == rnp->boost_tasks) | |
617 | rnp_root->boost_tasks = rnp->boost_tasks; | |
618 | #endif /* #ifdef CONFIG_RCU_BOOST */ | |
12f5f524 | 619 | raw_spin_unlock(&rnp_root->lock); /* irqs still disabled */ |
dd5d19ba | 620 | } |
27f4d280 | 621 | |
1e3fd2b3 PM |
622 | rnp->gp_tasks = NULL; |
623 | rnp->exp_tasks = NULL; | |
27f4d280 | 624 | #ifdef CONFIG_RCU_BOOST |
1e3fd2b3 | 625 | rnp->boost_tasks = NULL; |
5cc900cf PM |
626 | /* |
627 | * In case root is being boosted and leaf was not. Make sure | |
628 | * that we boost the tasks blocking the current grace period | |
629 | * in this case. | |
630 | */ | |
27f4d280 PM |
631 | raw_spin_lock(&rnp_root->lock); /* irqs already disabled */ |
632 | if (rnp_root->boost_tasks != NULL && | |
5cc900cf PM |
633 | rnp_root->boost_tasks != rnp_root->gp_tasks && |
634 | rnp_root->boost_tasks != rnp_root->exp_tasks) | |
27f4d280 PM |
635 | rnp_root->boost_tasks = rnp_root->gp_tasks; |
636 | raw_spin_unlock(&rnp_root->lock); /* irqs still disabled */ | |
637 | #endif /* #ifdef CONFIG_RCU_BOOST */ | |
638 | ||
237c80c5 | 639 | return retval; |
dd5d19ba PM |
640 | } |
641 | ||
e5601400 PM |
642 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ |
643 | ||
f41d911f PM |
644 | /* |
645 | * Check for a quiescent state from the current CPU. When a task blocks, | |
646 | * the task is recorded in the corresponding CPU's rcu_node structure, | |
647 | * which is checked elsewhere. | |
648 | * | |
649 | * Caller must disable hard irqs. | |
650 | */ | |
651 | static void rcu_preempt_check_callbacks(int cpu) | |
652 | { | |
653 | struct task_struct *t = current; | |
654 | ||
655 | if (t->rcu_read_lock_nesting == 0) { | |
c3422bea | 656 | rcu_preempt_qs(cpu); |
f41d911f PM |
657 | return; |
658 | } | |
10f39bb1 PM |
659 | if (t->rcu_read_lock_nesting > 0 && |
660 | per_cpu(rcu_preempt_data, cpu).qs_pending) | |
c3422bea | 661 | t->rcu_read_unlock_special |= RCU_READ_UNLOCK_NEED_QS; |
f41d911f PM |
662 | } |
663 | ||
a46e0899 PM |
664 | #ifdef CONFIG_RCU_BOOST |
665 | ||
09223371 SL |
666 | static void rcu_preempt_do_callbacks(void) |
667 | { | |
c9d4b0af | 668 | rcu_do_batch(&rcu_preempt_state, this_cpu_ptr(&rcu_preempt_data)); |
09223371 SL |
669 | } |
670 | ||
a46e0899 PM |
671 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
672 | ||
f41d911f | 673 | /* |
6cc68793 | 674 | * Queue a preemptible-RCU callback for invocation after a grace period. |
f41d911f PM |
675 | */ |
676 | void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
677 | { | |
3fbfbf7a | 678 | __call_rcu(head, func, &rcu_preempt_state, -1, 0); |
f41d911f PM |
679 | } |
680 | EXPORT_SYMBOL_GPL(call_rcu); | |
681 | ||
486e2593 PM |
682 | /* |
683 | * Queue an RCU callback for lazy invocation after a grace period. | |
684 | * This will likely be later named something like "call_rcu_lazy()", | |
685 | * but this change will require some way of tagging the lazy RCU | |
686 | * callbacks in the list of pending callbacks. Until then, this | |
687 | * function may only be called from __kfree_rcu(). | |
688 | */ | |
689 | void kfree_call_rcu(struct rcu_head *head, | |
690 | void (*func)(struct rcu_head *rcu)) | |
691 | { | |
3fbfbf7a | 692 | __call_rcu(head, func, &rcu_preempt_state, -1, 1); |
486e2593 PM |
693 | } |
694 | EXPORT_SYMBOL_GPL(kfree_call_rcu); | |
695 | ||
6ebb237b PM |
696 | /** |
697 | * synchronize_rcu - wait until a grace period has elapsed. | |
698 | * | |
699 | * Control will return to the caller some time after a full grace | |
700 | * period has elapsed, in other words after all currently executing RCU | |
77d8485a PM |
701 | * read-side critical sections have completed. Note, however, that |
702 | * upon return from synchronize_rcu(), the caller might well be executing | |
703 | * concurrently with new RCU read-side critical sections that began while | |
704 | * synchronize_rcu() was waiting. RCU read-side critical sections are | |
705 | * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested. | |
f0a0e6f2 PM |
706 | * |
707 | * See the description of synchronize_sched() for more detailed information | |
708 | * on memory ordering guarantees. | |
6ebb237b PM |
709 | */ |
710 | void synchronize_rcu(void) | |
711 | { | |
fe15d706 PM |
712 | rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map) && |
713 | !lock_is_held(&rcu_lock_map) && | |
714 | !lock_is_held(&rcu_sched_lock_map), | |
715 | "Illegal synchronize_rcu() in RCU read-side critical section"); | |
6ebb237b PM |
716 | if (!rcu_scheduler_active) |
717 | return; | |
3705b88d AM |
718 | if (rcu_expedited) |
719 | synchronize_rcu_expedited(); | |
720 | else | |
721 | wait_rcu_gp(call_rcu); | |
6ebb237b PM |
722 | } |
723 | EXPORT_SYMBOL_GPL(synchronize_rcu); | |
724 | ||
d9a3da06 | 725 | static DECLARE_WAIT_QUEUE_HEAD(sync_rcu_preempt_exp_wq); |
bcfa57ce | 726 | static unsigned long sync_rcu_preempt_exp_count; |
d9a3da06 PM |
727 | static DEFINE_MUTEX(sync_rcu_preempt_exp_mutex); |
728 | ||
729 | /* | |
730 | * Return non-zero if there are any tasks in RCU read-side critical | |
731 | * sections blocking the current preemptible-RCU expedited grace period. | |
732 | * If there is no preemptible-RCU expedited grace period currently in | |
733 | * progress, returns zero unconditionally. | |
734 | */ | |
735 | static int rcu_preempted_readers_exp(struct rcu_node *rnp) | |
736 | { | |
12f5f524 | 737 | return rnp->exp_tasks != NULL; |
d9a3da06 PM |
738 | } |
739 | ||
740 | /* | |
741 | * return non-zero if there is no RCU expedited grace period in progress | |
742 | * for the specified rcu_node structure, in other words, if all CPUs and | |
743 | * tasks covered by the specified rcu_node structure have done their bit | |
744 | * for the current expedited grace period. Works only for preemptible | |
745 | * RCU -- other RCU implementation use other means. | |
746 | * | |
747 | * Caller must hold sync_rcu_preempt_exp_mutex. | |
748 | */ | |
749 | static int sync_rcu_preempt_exp_done(struct rcu_node *rnp) | |
750 | { | |
751 | return !rcu_preempted_readers_exp(rnp) && | |
752 | ACCESS_ONCE(rnp->expmask) == 0; | |
753 | } | |
754 | ||
755 | /* | |
756 | * Report the exit from RCU read-side critical section for the last task | |
757 | * that queued itself during or before the current expedited preemptible-RCU | |
758 | * grace period. This event is reported either to the rcu_node structure on | |
759 | * which the task was queued or to one of that rcu_node structure's ancestors, | |
760 | * recursively up the tree. (Calm down, calm down, we do the recursion | |
761 | * iteratively!) | |
762 | * | |
b40d293e TG |
763 | * Most callers will set the "wake" flag, but the task initiating the |
764 | * expedited grace period need not wake itself. | |
765 | * | |
d9a3da06 PM |
766 | * Caller must hold sync_rcu_preempt_exp_mutex. |
767 | */ | |
b40d293e TG |
768 | static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp, |
769 | bool wake) | |
d9a3da06 PM |
770 | { |
771 | unsigned long flags; | |
772 | unsigned long mask; | |
773 | ||
1304afb2 | 774 | raw_spin_lock_irqsave(&rnp->lock, flags); |
d9a3da06 | 775 | for (;;) { |
131906b0 PM |
776 | if (!sync_rcu_preempt_exp_done(rnp)) { |
777 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
d9a3da06 | 778 | break; |
131906b0 | 779 | } |
d9a3da06 | 780 | if (rnp->parent == NULL) { |
131906b0 | 781 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
b40d293e TG |
782 | if (wake) |
783 | wake_up(&sync_rcu_preempt_exp_wq); | |
d9a3da06 PM |
784 | break; |
785 | } | |
786 | mask = rnp->grpmask; | |
1304afb2 | 787 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
d9a3da06 | 788 | rnp = rnp->parent; |
1304afb2 | 789 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
d9a3da06 PM |
790 | rnp->expmask &= ~mask; |
791 | } | |
d9a3da06 PM |
792 | } |
793 | ||
794 | /* | |
795 | * Snapshot the tasks blocking the newly started preemptible-RCU expedited | |
796 | * grace period for the specified rcu_node structure. If there are no such | |
797 | * tasks, report it up the rcu_node hierarchy. | |
798 | * | |
7b2e6011 PM |
799 | * Caller must hold sync_rcu_preempt_exp_mutex and must exclude |
800 | * CPU hotplug operations. | |
d9a3da06 PM |
801 | */ |
802 | static void | |
803 | sync_rcu_preempt_exp_init(struct rcu_state *rsp, struct rcu_node *rnp) | |
804 | { | |
1217ed1b | 805 | unsigned long flags; |
12f5f524 | 806 | int must_wait = 0; |
d9a3da06 | 807 | |
1217ed1b | 808 | raw_spin_lock_irqsave(&rnp->lock, flags); |
c701d5d9 | 809 | if (list_empty(&rnp->blkd_tasks)) { |
1217ed1b | 810 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
c701d5d9 | 811 | } else { |
12f5f524 | 812 | rnp->exp_tasks = rnp->blkd_tasks.next; |
1217ed1b | 813 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock */ |
12f5f524 PM |
814 | must_wait = 1; |
815 | } | |
d9a3da06 | 816 | if (!must_wait) |
b40d293e | 817 | rcu_report_exp_rnp(rsp, rnp, false); /* Don't wake self. */ |
d9a3da06 PM |
818 | } |
819 | ||
236fefaf PM |
820 | /** |
821 | * synchronize_rcu_expedited - Brute-force RCU grace period | |
822 | * | |
823 | * Wait for an RCU-preempt grace period, but expedite it. The basic | |
824 | * idea is to invoke synchronize_sched_expedited() to push all the tasks to | |
825 | * the ->blkd_tasks lists and wait for this list to drain. This consumes | |
826 | * significant time on all CPUs and is unfriendly to real-time workloads, | |
827 | * so is thus not recommended for any sort of common-case code. | |
828 | * In fact, if you are using synchronize_rcu_expedited() in a loop, | |
829 | * please restructure your code to batch your updates, and then Use a | |
830 | * single synchronize_rcu() instead. | |
831 | * | |
832 | * Note that it is illegal to call this function while holding any lock | |
833 | * that is acquired by a CPU-hotplug notifier. And yes, it is also illegal | |
834 | * to call this function from a CPU-hotplug notifier. Failing to observe | |
835 | * these restriction will result in deadlock. | |
019129d5 PM |
836 | */ |
837 | void synchronize_rcu_expedited(void) | |
838 | { | |
d9a3da06 PM |
839 | unsigned long flags; |
840 | struct rcu_node *rnp; | |
841 | struct rcu_state *rsp = &rcu_preempt_state; | |
bcfa57ce | 842 | unsigned long snap; |
d9a3da06 PM |
843 | int trycount = 0; |
844 | ||
845 | smp_mb(); /* Caller's modifications seen first by other CPUs. */ | |
846 | snap = ACCESS_ONCE(sync_rcu_preempt_exp_count) + 1; | |
847 | smp_mb(); /* Above access cannot bleed into critical section. */ | |
848 | ||
1943c89d PM |
849 | /* |
850 | * Block CPU-hotplug operations. This means that any CPU-hotplug | |
851 | * operation that finds an rcu_node structure with tasks in the | |
852 | * process of being boosted will know that all tasks blocking | |
853 | * this expedited grace period will already be in the process of | |
854 | * being boosted. This simplifies the process of moving tasks | |
855 | * from leaf to root rcu_node structures. | |
856 | */ | |
857 | get_online_cpus(); | |
858 | ||
d9a3da06 PM |
859 | /* |
860 | * Acquire lock, falling back to synchronize_rcu() if too many | |
861 | * lock-acquisition failures. Of course, if someone does the | |
862 | * expedited grace period for us, just leave. | |
863 | */ | |
864 | while (!mutex_trylock(&sync_rcu_preempt_exp_mutex)) { | |
1943c89d PM |
865 | if (ULONG_CMP_LT(snap, |
866 | ACCESS_ONCE(sync_rcu_preempt_exp_count))) { | |
867 | put_online_cpus(); | |
868 | goto mb_ret; /* Others did our work for us. */ | |
869 | } | |
c701d5d9 | 870 | if (trycount++ < 10) { |
d9a3da06 | 871 | udelay(trycount * num_online_cpus()); |
c701d5d9 | 872 | } else { |
1943c89d | 873 | put_online_cpus(); |
3705b88d | 874 | wait_rcu_gp(call_rcu); |
d9a3da06 PM |
875 | return; |
876 | } | |
d9a3da06 | 877 | } |
1943c89d PM |
878 | if (ULONG_CMP_LT(snap, ACCESS_ONCE(sync_rcu_preempt_exp_count))) { |
879 | put_online_cpus(); | |
d9a3da06 | 880 | goto unlock_mb_ret; /* Others did our work for us. */ |
1943c89d | 881 | } |
d9a3da06 | 882 | |
12f5f524 | 883 | /* force all RCU readers onto ->blkd_tasks lists. */ |
d9a3da06 PM |
884 | synchronize_sched_expedited(); |
885 | ||
d9a3da06 PM |
886 | /* Initialize ->expmask for all non-leaf rcu_node structures. */ |
887 | rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) { | |
1943c89d | 888 | raw_spin_lock_irqsave(&rnp->lock, flags); |
d9a3da06 | 889 | rnp->expmask = rnp->qsmaskinit; |
1943c89d | 890 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
d9a3da06 PM |
891 | } |
892 | ||
12f5f524 | 893 | /* Snapshot current state of ->blkd_tasks lists. */ |
d9a3da06 PM |
894 | rcu_for_each_leaf_node(rsp, rnp) |
895 | sync_rcu_preempt_exp_init(rsp, rnp); | |
896 | if (NUM_RCU_NODES > 1) | |
897 | sync_rcu_preempt_exp_init(rsp, rcu_get_root(rsp)); | |
898 | ||
1943c89d | 899 | put_online_cpus(); |
d9a3da06 | 900 | |
12f5f524 | 901 | /* Wait for snapshotted ->blkd_tasks lists to drain. */ |
d9a3da06 PM |
902 | rnp = rcu_get_root(rsp); |
903 | wait_event(sync_rcu_preempt_exp_wq, | |
904 | sync_rcu_preempt_exp_done(rnp)); | |
905 | ||
906 | /* Clean up and exit. */ | |
907 | smp_mb(); /* ensure expedited GP seen before counter increment. */ | |
908 | ACCESS_ONCE(sync_rcu_preempt_exp_count)++; | |
909 | unlock_mb_ret: | |
910 | mutex_unlock(&sync_rcu_preempt_exp_mutex); | |
911 | mb_ret: | |
912 | smp_mb(); /* ensure subsequent action seen after grace period. */ | |
019129d5 PM |
913 | } |
914 | EXPORT_SYMBOL_GPL(synchronize_rcu_expedited); | |
915 | ||
e74f4c45 PM |
916 | /** |
917 | * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete. | |
f0a0e6f2 PM |
918 | * |
919 | * Note that this primitive does not necessarily wait for an RCU grace period | |
920 | * to complete. For example, if there are no RCU callbacks queued anywhere | |
921 | * in the system, then rcu_barrier() is within its rights to return | |
922 | * immediately, without waiting for anything, much less an RCU grace period. | |
e74f4c45 PM |
923 | */ |
924 | void rcu_barrier(void) | |
925 | { | |
037b64ed | 926 | _rcu_barrier(&rcu_preempt_state); |
e74f4c45 PM |
927 | } |
928 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
929 | ||
1eba8f84 | 930 | /* |
6cc68793 | 931 | * Initialize preemptible RCU's state structures. |
1eba8f84 PM |
932 | */ |
933 | static void __init __rcu_init_preempt(void) | |
934 | { | |
394f99a9 | 935 | rcu_init_one(&rcu_preempt_state, &rcu_preempt_data); |
1eba8f84 PM |
936 | } |
937 | ||
2439b696 PM |
938 | /* |
939 | * Check for a task exiting while in a preemptible-RCU read-side | |
940 | * critical section, clean up if so. No need to issue warnings, | |
941 | * as debug_check_no_locks_held() already does this if lockdep | |
942 | * is enabled. | |
943 | */ | |
944 | void exit_rcu(void) | |
945 | { | |
946 | struct task_struct *t = current; | |
947 | ||
948 | if (likely(list_empty(¤t->rcu_node_entry))) | |
949 | return; | |
950 | t->rcu_read_lock_nesting = 1; | |
951 | barrier(); | |
952 | t->rcu_read_unlock_special = RCU_READ_UNLOCK_BLOCKED; | |
953 | __rcu_read_unlock(); | |
954 | } | |
955 | ||
f41d911f PM |
956 | #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */ |
957 | ||
27f4d280 PM |
958 | static struct rcu_state *rcu_state = &rcu_sched_state; |
959 | ||
f41d911f PM |
960 | /* |
961 | * Tell them what RCU they are running. | |
962 | */ | |
0e0fc1c2 | 963 | static void __init rcu_bootup_announce(void) |
f41d911f | 964 | { |
efc151c3 | 965 | pr_info("Hierarchical RCU implementation.\n"); |
26845c28 | 966 | rcu_bootup_announce_oddness(); |
f41d911f PM |
967 | } |
968 | ||
969 | /* | |
970 | * Return the number of RCU batches processed thus far for debug & stats. | |
971 | */ | |
972 | long rcu_batches_completed(void) | |
973 | { | |
974 | return rcu_batches_completed_sched(); | |
975 | } | |
976 | EXPORT_SYMBOL_GPL(rcu_batches_completed); | |
977 | ||
bf66f18e PM |
978 | /* |
979 | * Force a quiescent state for RCU, which, because there is no preemptible | |
980 | * RCU, becomes the same as rcu-sched. | |
981 | */ | |
982 | void rcu_force_quiescent_state(void) | |
983 | { | |
984 | rcu_sched_force_quiescent_state(); | |
985 | } | |
986 | EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); | |
987 | ||
cba6d0d6 PM |
988 | /* |
989 | * Because preemptible RCU does not exist, we never have to check for | |
990 | * CPUs being in quiescent states. | |
991 | */ | |
992 | static void rcu_preempt_note_context_switch(int cpu) | |
993 | { | |
994 | } | |
995 | ||
fc2219d4 | 996 | /* |
6cc68793 | 997 | * Because preemptible RCU does not exist, there are never any preempted |
fc2219d4 PM |
998 | * RCU readers. |
999 | */ | |
27f4d280 | 1000 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 PM |
1001 | { |
1002 | return 0; | |
1003 | } | |
1004 | ||
b668c9cf PM |
1005 | #ifdef CONFIG_HOTPLUG_CPU |
1006 | ||
1007 | /* Because preemptible RCU does not exist, no quieting of tasks. */ | |
d3f6bad3 | 1008 | static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, unsigned long flags) |
b668c9cf | 1009 | { |
1304afb2 | 1010 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
b668c9cf PM |
1011 | } |
1012 | ||
1013 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | |
1014 | ||
1ed509a2 | 1015 | /* |
6cc68793 | 1016 | * Because preemptible RCU does not exist, we never have to check for |
1ed509a2 PM |
1017 | * tasks blocked within RCU read-side critical sections. |
1018 | */ | |
1019 | static void rcu_print_detail_task_stall(struct rcu_state *rsp) | |
1020 | { | |
1021 | } | |
1022 | ||
f41d911f | 1023 | /* |
6cc68793 | 1024 | * Because preemptible RCU does not exist, we never have to check for |
f41d911f PM |
1025 | * tasks blocked within RCU read-side critical sections. |
1026 | */ | |
9bc8b558 | 1027 | static int rcu_print_task_stall(struct rcu_node *rnp) |
f41d911f | 1028 | { |
9bc8b558 | 1029 | return 0; |
f41d911f PM |
1030 | } |
1031 | ||
b0e165c0 | 1032 | /* |
6cc68793 | 1033 | * Because there is no preemptible RCU, there can be no readers blocked, |
49e29126 PM |
1034 | * so there is no need to check for blocked tasks. So check only for |
1035 | * bogus qsmask values. | |
b0e165c0 PM |
1036 | */ |
1037 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) | |
1038 | { | |
49e29126 | 1039 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
1040 | } |
1041 | ||
33f76148 PM |
1042 | #ifdef CONFIG_HOTPLUG_CPU |
1043 | ||
dd5d19ba | 1044 | /* |
6cc68793 | 1045 | * Because preemptible RCU does not exist, it never needs to migrate |
237c80c5 PM |
1046 | * tasks that were blocked within RCU read-side critical sections, and |
1047 | * such non-existent tasks cannot possibly have been blocking the current | |
1048 | * grace period. | |
dd5d19ba | 1049 | */ |
237c80c5 PM |
1050 | static int rcu_preempt_offline_tasks(struct rcu_state *rsp, |
1051 | struct rcu_node *rnp, | |
1052 | struct rcu_data *rdp) | |
dd5d19ba | 1053 | { |
237c80c5 | 1054 | return 0; |
dd5d19ba PM |
1055 | } |
1056 | ||
e5601400 PM |
1057 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ |
1058 | ||
f41d911f | 1059 | /* |
6cc68793 | 1060 | * Because preemptible RCU does not exist, it never has any callbacks |
f41d911f PM |
1061 | * to check. |
1062 | */ | |
1eba8f84 | 1063 | static void rcu_preempt_check_callbacks(int cpu) |
f41d911f PM |
1064 | { |
1065 | } | |
1066 | ||
486e2593 PM |
1067 | /* |
1068 | * Queue an RCU callback for lazy invocation after a grace period. | |
1069 | * This will likely be later named something like "call_rcu_lazy()", | |
1070 | * but this change will require some way of tagging the lazy RCU | |
1071 | * callbacks in the list of pending callbacks. Until then, this | |
1072 | * function may only be called from __kfree_rcu(). | |
1073 | * | |
1074 | * Because there is no preemptible RCU, we use RCU-sched instead. | |
1075 | */ | |
1076 | void kfree_call_rcu(struct rcu_head *head, | |
1077 | void (*func)(struct rcu_head *rcu)) | |
1078 | { | |
3fbfbf7a | 1079 | __call_rcu(head, func, &rcu_sched_state, -1, 1); |
486e2593 PM |
1080 | } |
1081 | EXPORT_SYMBOL_GPL(kfree_call_rcu); | |
1082 | ||
019129d5 PM |
1083 | /* |
1084 | * Wait for an rcu-preempt grace period, but make it happen quickly. | |
6cc68793 | 1085 | * But because preemptible RCU does not exist, map to rcu-sched. |
019129d5 PM |
1086 | */ |
1087 | void synchronize_rcu_expedited(void) | |
1088 | { | |
1089 | synchronize_sched_expedited(); | |
1090 | } | |
1091 | EXPORT_SYMBOL_GPL(synchronize_rcu_expedited); | |
1092 | ||
d9a3da06 PM |
1093 | #ifdef CONFIG_HOTPLUG_CPU |
1094 | ||
1095 | /* | |
6cc68793 | 1096 | * Because preemptible RCU does not exist, there is never any need to |
d9a3da06 PM |
1097 | * report on tasks preempted in RCU read-side critical sections during |
1098 | * expedited RCU grace periods. | |
1099 | */ | |
b40d293e TG |
1100 | static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp, |
1101 | bool wake) | |
d9a3da06 | 1102 | { |
d9a3da06 PM |
1103 | } |
1104 | ||
1105 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | |
1106 | ||
e74f4c45 | 1107 | /* |
6cc68793 | 1108 | * Because preemptible RCU does not exist, rcu_barrier() is just |
e74f4c45 PM |
1109 | * another name for rcu_barrier_sched(). |
1110 | */ | |
1111 | void rcu_barrier(void) | |
1112 | { | |
1113 | rcu_barrier_sched(); | |
1114 | } | |
1115 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
1116 | ||
1eba8f84 | 1117 | /* |
6cc68793 | 1118 | * Because preemptible RCU does not exist, it need not be initialized. |
1eba8f84 PM |
1119 | */ |
1120 | static void __init __rcu_init_preempt(void) | |
1121 | { | |
1122 | } | |
1123 | ||
2439b696 PM |
1124 | /* |
1125 | * Because preemptible RCU does not exist, tasks cannot possibly exit | |
1126 | * while in preemptible RCU read-side critical sections. | |
1127 | */ | |
1128 | void exit_rcu(void) | |
1129 | { | |
1130 | } | |
1131 | ||
f41d911f | 1132 | #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */ |
8bd93a2c | 1133 | |
27f4d280 PM |
1134 | #ifdef CONFIG_RCU_BOOST |
1135 | ||
1136 | #include "rtmutex_common.h" | |
1137 | ||
0ea1f2eb PM |
1138 | #ifdef CONFIG_RCU_TRACE |
1139 | ||
1140 | static void rcu_initiate_boost_trace(struct rcu_node *rnp) | |
1141 | { | |
1142 | if (list_empty(&rnp->blkd_tasks)) | |
1143 | rnp->n_balk_blkd_tasks++; | |
1144 | else if (rnp->exp_tasks == NULL && rnp->gp_tasks == NULL) | |
1145 | rnp->n_balk_exp_gp_tasks++; | |
1146 | else if (rnp->gp_tasks != NULL && rnp->boost_tasks != NULL) | |
1147 | rnp->n_balk_boost_tasks++; | |
1148 | else if (rnp->gp_tasks != NULL && rnp->qsmask != 0) | |
1149 | rnp->n_balk_notblocked++; | |
1150 | else if (rnp->gp_tasks != NULL && | |
a9f4793d | 1151 | ULONG_CMP_LT(jiffies, rnp->boost_time)) |
0ea1f2eb PM |
1152 | rnp->n_balk_notyet++; |
1153 | else | |
1154 | rnp->n_balk_nos++; | |
1155 | } | |
1156 | ||
1157 | #else /* #ifdef CONFIG_RCU_TRACE */ | |
1158 | ||
1159 | static void rcu_initiate_boost_trace(struct rcu_node *rnp) | |
1160 | { | |
1161 | } | |
1162 | ||
1163 | #endif /* #else #ifdef CONFIG_RCU_TRACE */ | |
1164 | ||
5d01bbd1 TG |
1165 | static void rcu_wake_cond(struct task_struct *t, int status) |
1166 | { | |
1167 | /* | |
1168 | * If the thread is yielding, only wake it when this | |
1169 | * is invoked from idle | |
1170 | */ | |
1171 | if (status != RCU_KTHREAD_YIELDING || is_idle_task(current)) | |
1172 | wake_up_process(t); | |
1173 | } | |
1174 | ||
27f4d280 PM |
1175 | /* |
1176 | * Carry out RCU priority boosting on the task indicated by ->exp_tasks | |
1177 | * or ->boost_tasks, advancing the pointer to the next task in the | |
1178 | * ->blkd_tasks list. | |
1179 | * | |
1180 | * Note that irqs must be enabled: boosting the task can block. | |
1181 | * Returns 1 if there are more tasks needing to be boosted. | |
1182 | */ | |
1183 | static int rcu_boost(struct rcu_node *rnp) | |
1184 | { | |
1185 | unsigned long flags; | |
1186 | struct rt_mutex mtx; | |
1187 | struct task_struct *t; | |
1188 | struct list_head *tb; | |
1189 | ||
1190 | if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) | |
1191 | return 0; /* Nothing left to boost. */ | |
1192 | ||
1193 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
1194 | ||
1195 | /* | |
1196 | * Recheck under the lock: all tasks in need of boosting | |
1197 | * might exit their RCU read-side critical sections on their own. | |
1198 | */ | |
1199 | if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) { | |
1200 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
1201 | return 0; | |
1202 | } | |
1203 | ||
1204 | /* | |
1205 | * Preferentially boost tasks blocking expedited grace periods. | |
1206 | * This cannot starve the normal grace periods because a second | |
1207 | * expedited grace period must boost all blocked tasks, including | |
1208 | * those blocking the pre-existing normal grace period. | |
1209 | */ | |
0ea1f2eb | 1210 | if (rnp->exp_tasks != NULL) { |
27f4d280 | 1211 | tb = rnp->exp_tasks; |
0ea1f2eb PM |
1212 | rnp->n_exp_boosts++; |
1213 | } else { | |
27f4d280 | 1214 | tb = rnp->boost_tasks; |
0ea1f2eb PM |
1215 | rnp->n_normal_boosts++; |
1216 | } | |
1217 | rnp->n_tasks_boosted++; | |
27f4d280 PM |
1218 | |
1219 | /* | |
1220 | * We boost task t by manufacturing an rt_mutex that appears to | |
1221 | * be held by task t. We leave a pointer to that rt_mutex where | |
1222 | * task t can find it, and task t will release the mutex when it | |
1223 | * exits its outermost RCU read-side critical section. Then | |
1224 | * simply acquiring this artificial rt_mutex will boost task | |
1225 | * t's priority. (Thanks to tglx for suggesting this approach!) | |
1226 | * | |
1227 | * Note that task t must acquire rnp->lock to remove itself from | |
1228 | * the ->blkd_tasks list, which it will do from exit() if from | |
1229 | * nowhere else. We therefore are guaranteed that task t will | |
1230 | * stay around at least until we drop rnp->lock. Note that | |
1231 | * rnp->lock also resolves races between our priority boosting | |
1232 | * and task t's exiting its outermost RCU read-side critical | |
1233 | * section. | |
1234 | */ | |
1235 | t = container_of(tb, struct task_struct, rcu_node_entry); | |
1236 | rt_mutex_init_proxy_locked(&mtx, t); | |
1237 | t->rcu_boost_mutex = &mtx; | |
27f4d280 PM |
1238 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1239 | rt_mutex_lock(&mtx); /* Side effect: boosts task t's priority. */ | |
1240 | rt_mutex_unlock(&mtx); /* Keep lockdep happy. */ | |
1241 | ||
4f89b336 PM |
1242 | return ACCESS_ONCE(rnp->exp_tasks) != NULL || |
1243 | ACCESS_ONCE(rnp->boost_tasks) != NULL; | |
27f4d280 PM |
1244 | } |
1245 | ||
27f4d280 PM |
1246 | /* |
1247 | * Priority-boosting kthread. One per leaf rcu_node and one for the | |
1248 | * root rcu_node. | |
1249 | */ | |
1250 | static int rcu_boost_kthread(void *arg) | |
1251 | { | |
1252 | struct rcu_node *rnp = (struct rcu_node *)arg; | |
1253 | int spincnt = 0; | |
1254 | int more2boost; | |
1255 | ||
f7f7bac9 | 1256 | trace_rcu_utilization(TPS("Start boost kthread@init")); |
27f4d280 | 1257 | for (;;) { |
d71df90e | 1258 | rnp->boost_kthread_status = RCU_KTHREAD_WAITING; |
f7f7bac9 | 1259 | trace_rcu_utilization(TPS("End boost kthread@rcu_wait")); |
08bca60a | 1260 | rcu_wait(rnp->boost_tasks || rnp->exp_tasks); |
f7f7bac9 | 1261 | trace_rcu_utilization(TPS("Start boost kthread@rcu_wait")); |
d71df90e | 1262 | rnp->boost_kthread_status = RCU_KTHREAD_RUNNING; |
27f4d280 PM |
1263 | more2boost = rcu_boost(rnp); |
1264 | if (more2boost) | |
1265 | spincnt++; | |
1266 | else | |
1267 | spincnt = 0; | |
1268 | if (spincnt > 10) { | |
5d01bbd1 | 1269 | rnp->boost_kthread_status = RCU_KTHREAD_YIELDING; |
f7f7bac9 | 1270 | trace_rcu_utilization(TPS("End boost kthread@rcu_yield")); |
5d01bbd1 | 1271 | schedule_timeout_interruptible(2); |
f7f7bac9 | 1272 | trace_rcu_utilization(TPS("Start boost kthread@rcu_yield")); |
27f4d280 PM |
1273 | spincnt = 0; |
1274 | } | |
1275 | } | |
1217ed1b | 1276 | /* NOTREACHED */ |
f7f7bac9 | 1277 | trace_rcu_utilization(TPS("End boost kthread@notreached")); |
27f4d280 PM |
1278 | return 0; |
1279 | } | |
1280 | ||
1281 | /* | |
1282 | * Check to see if it is time to start boosting RCU readers that are | |
1283 | * blocking the current grace period, and, if so, tell the per-rcu_node | |
1284 | * kthread to start boosting them. If there is an expedited grace | |
1285 | * period in progress, it is always time to boost. | |
1286 | * | |
b065a853 PM |
1287 | * The caller must hold rnp->lock, which this function releases. |
1288 | * The ->boost_kthread_task is immortal, so we don't need to worry | |
1289 | * about it going away. | |
27f4d280 | 1290 | */ |
1217ed1b | 1291 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
27f4d280 PM |
1292 | { |
1293 | struct task_struct *t; | |
1294 | ||
0ea1f2eb PM |
1295 | if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) { |
1296 | rnp->n_balk_exp_gp_tasks++; | |
1217ed1b | 1297 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
27f4d280 | 1298 | return; |
0ea1f2eb | 1299 | } |
27f4d280 PM |
1300 | if (rnp->exp_tasks != NULL || |
1301 | (rnp->gp_tasks != NULL && | |
1302 | rnp->boost_tasks == NULL && | |
1303 | rnp->qsmask == 0 && | |
1304 | ULONG_CMP_GE(jiffies, rnp->boost_time))) { | |
1305 | if (rnp->exp_tasks == NULL) | |
1306 | rnp->boost_tasks = rnp->gp_tasks; | |
1217ed1b | 1307 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
27f4d280 | 1308 | t = rnp->boost_kthread_task; |
5d01bbd1 TG |
1309 | if (t) |
1310 | rcu_wake_cond(t, rnp->boost_kthread_status); | |
1217ed1b | 1311 | } else { |
0ea1f2eb | 1312 | rcu_initiate_boost_trace(rnp); |
1217ed1b PM |
1313 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1314 | } | |
27f4d280 PM |
1315 | } |
1316 | ||
a46e0899 PM |
1317 | /* |
1318 | * Wake up the per-CPU kthread to invoke RCU callbacks. | |
1319 | */ | |
1320 | static void invoke_rcu_callbacks_kthread(void) | |
1321 | { | |
1322 | unsigned long flags; | |
1323 | ||
1324 | local_irq_save(flags); | |
1325 | __this_cpu_write(rcu_cpu_has_work, 1); | |
1eb52121 | 1326 | if (__this_cpu_read(rcu_cpu_kthread_task) != NULL && |
5d01bbd1 TG |
1327 | current != __this_cpu_read(rcu_cpu_kthread_task)) { |
1328 | rcu_wake_cond(__this_cpu_read(rcu_cpu_kthread_task), | |
1329 | __this_cpu_read(rcu_cpu_kthread_status)); | |
1330 | } | |
a46e0899 PM |
1331 | local_irq_restore(flags); |
1332 | } | |
1333 | ||
dff1672d PM |
1334 | /* |
1335 | * Is the current CPU running the RCU-callbacks kthread? | |
1336 | * Caller must have preemption disabled. | |
1337 | */ | |
1338 | static bool rcu_is_callbacks_kthread(void) | |
1339 | { | |
c9d4b0af | 1340 | return __this_cpu_read(rcu_cpu_kthread_task) == current; |
dff1672d PM |
1341 | } |
1342 | ||
27f4d280 PM |
1343 | #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000) |
1344 | ||
1345 | /* | |
1346 | * Do priority-boost accounting for the start of a new grace period. | |
1347 | */ | |
1348 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) | |
1349 | { | |
1350 | rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES; | |
1351 | } | |
1352 | ||
27f4d280 PM |
1353 | /* |
1354 | * Create an RCU-boost kthread for the specified node if one does not | |
1355 | * already exist. We only create this kthread for preemptible RCU. | |
1356 | * Returns zero if all is well, a negated errno otherwise. | |
1357 | */ | |
49fb4c62 | 1358 | static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp, |
5d01bbd1 | 1359 | struct rcu_node *rnp) |
27f4d280 | 1360 | { |
5d01bbd1 | 1361 | int rnp_index = rnp - &rsp->node[0]; |
27f4d280 PM |
1362 | unsigned long flags; |
1363 | struct sched_param sp; | |
1364 | struct task_struct *t; | |
1365 | ||
1366 | if (&rcu_preempt_state != rsp) | |
1367 | return 0; | |
5d01bbd1 TG |
1368 | |
1369 | if (!rcu_scheduler_fully_active || rnp->qsmaskinit == 0) | |
1370 | return 0; | |
1371 | ||
a46e0899 | 1372 | rsp->boost = 1; |
27f4d280 PM |
1373 | if (rnp->boost_kthread_task != NULL) |
1374 | return 0; | |
1375 | t = kthread_create(rcu_boost_kthread, (void *)rnp, | |
5b61b0ba | 1376 | "rcub/%d", rnp_index); |
27f4d280 PM |
1377 | if (IS_ERR(t)) |
1378 | return PTR_ERR(t); | |
1379 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
1380 | rnp->boost_kthread_task = t; | |
1381 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
5b61b0ba | 1382 | sp.sched_priority = RCU_BOOST_PRIO; |
27f4d280 | 1383 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); |
9a432736 | 1384 | wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */ |
27f4d280 PM |
1385 | return 0; |
1386 | } | |
1387 | ||
f8b7fc6b PM |
1388 | static void rcu_kthread_do_work(void) |
1389 | { | |
c9d4b0af CL |
1390 | rcu_do_batch(&rcu_sched_state, this_cpu_ptr(&rcu_sched_data)); |
1391 | rcu_do_batch(&rcu_bh_state, this_cpu_ptr(&rcu_bh_data)); | |
f8b7fc6b PM |
1392 | rcu_preempt_do_callbacks(); |
1393 | } | |
1394 | ||
62ab7072 | 1395 | static void rcu_cpu_kthread_setup(unsigned int cpu) |
f8b7fc6b | 1396 | { |
f8b7fc6b | 1397 | struct sched_param sp; |
f8b7fc6b | 1398 | |
62ab7072 PM |
1399 | sp.sched_priority = RCU_KTHREAD_PRIO; |
1400 | sched_setscheduler_nocheck(current, SCHED_FIFO, &sp); | |
f8b7fc6b PM |
1401 | } |
1402 | ||
62ab7072 | 1403 | static void rcu_cpu_kthread_park(unsigned int cpu) |
f8b7fc6b | 1404 | { |
62ab7072 | 1405 | per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU; |
f8b7fc6b PM |
1406 | } |
1407 | ||
62ab7072 | 1408 | static int rcu_cpu_kthread_should_run(unsigned int cpu) |
f8b7fc6b | 1409 | { |
c9d4b0af | 1410 | return __this_cpu_read(rcu_cpu_has_work); |
f8b7fc6b PM |
1411 | } |
1412 | ||
1413 | /* | |
1414 | * Per-CPU kernel thread that invokes RCU callbacks. This replaces the | |
e0f23060 PM |
1415 | * RCU softirq used in flavors and configurations of RCU that do not |
1416 | * support RCU priority boosting. | |
f8b7fc6b | 1417 | */ |
62ab7072 | 1418 | static void rcu_cpu_kthread(unsigned int cpu) |
f8b7fc6b | 1419 | { |
c9d4b0af CL |
1420 | unsigned int *statusp = this_cpu_ptr(&rcu_cpu_kthread_status); |
1421 | char work, *workp = this_cpu_ptr(&rcu_cpu_has_work); | |
62ab7072 | 1422 | int spincnt; |
f8b7fc6b | 1423 | |
62ab7072 | 1424 | for (spincnt = 0; spincnt < 10; spincnt++) { |
f7f7bac9 | 1425 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_wait")); |
f8b7fc6b | 1426 | local_bh_disable(); |
f8b7fc6b | 1427 | *statusp = RCU_KTHREAD_RUNNING; |
62ab7072 PM |
1428 | this_cpu_inc(rcu_cpu_kthread_loops); |
1429 | local_irq_disable(); | |
f8b7fc6b PM |
1430 | work = *workp; |
1431 | *workp = 0; | |
62ab7072 | 1432 | local_irq_enable(); |
f8b7fc6b PM |
1433 | if (work) |
1434 | rcu_kthread_do_work(); | |
1435 | local_bh_enable(); | |
62ab7072 | 1436 | if (*workp == 0) { |
f7f7bac9 | 1437 | trace_rcu_utilization(TPS("End CPU kthread@rcu_wait")); |
62ab7072 PM |
1438 | *statusp = RCU_KTHREAD_WAITING; |
1439 | return; | |
f8b7fc6b PM |
1440 | } |
1441 | } | |
62ab7072 | 1442 | *statusp = RCU_KTHREAD_YIELDING; |
f7f7bac9 | 1443 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_yield")); |
62ab7072 | 1444 | schedule_timeout_interruptible(2); |
f7f7bac9 | 1445 | trace_rcu_utilization(TPS("End CPU kthread@rcu_yield")); |
62ab7072 | 1446 | *statusp = RCU_KTHREAD_WAITING; |
f8b7fc6b PM |
1447 | } |
1448 | ||
1449 | /* | |
1450 | * Set the per-rcu_node kthread's affinity to cover all CPUs that are | |
1451 | * served by the rcu_node in question. The CPU hotplug lock is still | |
1452 | * held, so the value of rnp->qsmaskinit will be stable. | |
1453 | * | |
1454 | * We don't include outgoingcpu in the affinity set, use -1 if there is | |
1455 | * no outgoing CPU. If there are no CPUs left in the affinity set, | |
1456 | * this function allows the kthread to execute on any CPU. | |
1457 | */ | |
5d01bbd1 | 1458 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b | 1459 | { |
5d01bbd1 TG |
1460 | struct task_struct *t = rnp->boost_kthread_task; |
1461 | unsigned long mask = rnp->qsmaskinit; | |
f8b7fc6b PM |
1462 | cpumask_var_t cm; |
1463 | int cpu; | |
f8b7fc6b | 1464 | |
5d01bbd1 | 1465 | if (!t) |
f8b7fc6b | 1466 | return; |
5d01bbd1 | 1467 | if (!zalloc_cpumask_var(&cm, GFP_KERNEL)) |
f8b7fc6b | 1468 | return; |
f8b7fc6b PM |
1469 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1) |
1470 | if ((mask & 0x1) && cpu != outgoingcpu) | |
1471 | cpumask_set_cpu(cpu, cm); | |
1472 | if (cpumask_weight(cm) == 0) { | |
1473 | cpumask_setall(cm); | |
1474 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++) | |
1475 | cpumask_clear_cpu(cpu, cm); | |
1476 | WARN_ON_ONCE(cpumask_weight(cm) == 0); | |
1477 | } | |
5d01bbd1 | 1478 | set_cpus_allowed_ptr(t, cm); |
f8b7fc6b PM |
1479 | free_cpumask_var(cm); |
1480 | } | |
1481 | ||
62ab7072 PM |
1482 | static struct smp_hotplug_thread rcu_cpu_thread_spec = { |
1483 | .store = &rcu_cpu_kthread_task, | |
1484 | .thread_should_run = rcu_cpu_kthread_should_run, | |
1485 | .thread_fn = rcu_cpu_kthread, | |
1486 | .thread_comm = "rcuc/%u", | |
1487 | .setup = rcu_cpu_kthread_setup, | |
1488 | .park = rcu_cpu_kthread_park, | |
1489 | }; | |
f8b7fc6b PM |
1490 | |
1491 | /* | |
1492 | * Spawn all kthreads -- called as soon as the scheduler is running. | |
1493 | */ | |
1494 | static int __init rcu_spawn_kthreads(void) | |
1495 | { | |
f8b7fc6b | 1496 | struct rcu_node *rnp; |
5d01bbd1 | 1497 | int cpu; |
f8b7fc6b | 1498 | |
b0d30417 | 1499 | rcu_scheduler_fully_active = 1; |
62ab7072 | 1500 | for_each_possible_cpu(cpu) |
f8b7fc6b | 1501 | per_cpu(rcu_cpu_has_work, cpu) = 0; |
62ab7072 | 1502 | BUG_ON(smpboot_register_percpu_thread(&rcu_cpu_thread_spec)); |
f8b7fc6b | 1503 | rnp = rcu_get_root(rcu_state); |
5d01bbd1 | 1504 | (void)rcu_spawn_one_boost_kthread(rcu_state, rnp); |
f8b7fc6b PM |
1505 | if (NUM_RCU_NODES > 1) { |
1506 | rcu_for_each_leaf_node(rcu_state, rnp) | |
5d01bbd1 | 1507 | (void)rcu_spawn_one_boost_kthread(rcu_state, rnp); |
f8b7fc6b PM |
1508 | } |
1509 | return 0; | |
1510 | } | |
1511 | early_initcall(rcu_spawn_kthreads); | |
1512 | ||
49fb4c62 | 1513 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b PM |
1514 | { |
1515 | struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu); | |
1516 | struct rcu_node *rnp = rdp->mynode; | |
1517 | ||
1518 | /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */ | |
62ab7072 | 1519 | if (rcu_scheduler_fully_active) |
5d01bbd1 | 1520 | (void)rcu_spawn_one_boost_kthread(rcu_state, rnp); |
f8b7fc6b PM |
1521 | } |
1522 | ||
27f4d280 PM |
1523 | #else /* #ifdef CONFIG_RCU_BOOST */ |
1524 | ||
1217ed1b | 1525 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
27f4d280 | 1526 | { |
1217ed1b | 1527 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
27f4d280 PM |
1528 | } |
1529 | ||
a46e0899 | 1530 | static void invoke_rcu_callbacks_kthread(void) |
27f4d280 | 1531 | { |
a46e0899 | 1532 | WARN_ON_ONCE(1); |
27f4d280 PM |
1533 | } |
1534 | ||
dff1672d PM |
1535 | static bool rcu_is_callbacks_kthread(void) |
1536 | { | |
1537 | return false; | |
1538 | } | |
1539 | ||
27f4d280 PM |
1540 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) |
1541 | { | |
1542 | } | |
1543 | ||
5d01bbd1 | 1544 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b PM |
1545 | { |
1546 | } | |
1547 | ||
b0d30417 PM |
1548 | static int __init rcu_scheduler_really_started(void) |
1549 | { | |
1550 | rcu_scheduler_fully_active = 1; | |
1551 | return 0; | |
1552 | } | |
1553 | early_initcall(rcu_scheduler_really_started); | |
1554 | ||
49fb4c62 | 1555 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b PM |
1556 | { |
1557 | } | |
1558 | ||
27f4d280 PM |
1559 | #endif /* #else #ifdef CONFIG_RCU_BOOST */ |
1560 | ||
8bd93a2c PM |
1561 | #if !defined(CONFIG_RCU_FAST_NO_HZ) |
1562 | ||
1563 | /* | |
1564 | * Check to see if any future RCU-related work will need to be done | |
1565 | * by the current CPU, even if none need be done immediately, returning | |
1566 | * 1 if so. This function is part of the RCU implementation; it is -not- | |
1567 | * an exported member of the RCU API. | |
1568 | * | |
7cb92499 PM |
1569 | * Because we not have RCU_FAST_NO_HZ, just check whether this CPU needs |
1570 | * any flavor of RCU. | |
8bd93a2c | 1571 | */ |
aa9b1630 | 1572 | int rcu_needs_cpu(int cpu, unsigned long *delta_jiffies) |
8bd93a2c | 1573 | { |
aa9b1630 | 1574 | *delta_jiffies = ULONG_MAX; |
c0f4dfd4 | 1575 | return rcu_cpu_has_callbacks(cpu, NULL); |
7cb92499 PM |
1576 | } |
1577 | ||
1578 | /* | |
1579 | * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up | |
1580 | * after it. | |
1581 | */ | |
1582 | static void rcu_cleanup_after_idle(int cpu) | |
1583 | { | |
1584 | } | |
1585 | ||
aea1b35e | 1586 | /* |
a858af28 | 1587 | * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n, |
aea1b35e PM |
1588 | * is nothing. |
1589 | */ | |
1590 | static void rcu_prepare_for_idle(int cpu) | |
1591 | { | |
1592 | } | |
1593 | ||
c57afe80 PM |
1594 | /* |
1595 | * Don't bother keeping a running count of the number of RCU callbacks | |
1596 | * posted because CONFIG_RCU_FAST_NO_HZ=n. | |
1597 | */ | |
1598 | static void rcu_idle_count_callbacks_posted(void) | |
1599 | { | |
1600 | } | |
1601 | ||
8bd93a2c PM |
1602 | #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
1603 | ||
f23f7fa1 PM |
1604 | /* |
1605 | * This code is invoked when a CPU goes idle, at which point we want | |
1606 | * to have the CPU do everything required for RCU so that it can enter | |
1607 | * the energy-efficient dyntick-idle mode. This is handled by a | |
1608 | * state machine implemented by rcu_prepare_for_idle() below. | |
1609 | * | |
1610 | * The following three proprocessor symbols control this state machine: | |
1611 | * | |
f23f7fa1 PM |
1612 | * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted |
1613 | * to sleep in dyntick-idle mode with RCU callbacks pending. This | |
1614 | * is sized to be roughly one RCU grace period. Those energy-efficiency | |
1615 | * benchmarkers who might otherwise be tempted to set this to a large | |
1616 | * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your | |
1617 | * system. And if you are -that- concerned about energy efficiency, | |
1618 | * just power the system down and be done with it! | |
778d250a PM |
1619 | * RCU_IDLE_LAZY_GP_DELAY gives the number of jiffies that a CPU is |
1620 | * permitted to sleep in dyntick-idle mode with only lazy RCU | |
1621 | * callbacks pending. Setting this too high can OOM your system. | |
f23f7fa1 PM |
1622 | * |
1623 | * The values below work well in practice. If future workloads require | |
1624 | * adjustment, they can be converted into kernel config parameters, though | |
1625 | * making the state machine smarter might be a better option. | |
1626 | */ | |
e84c48ae | 1627 | #define RCU_IDLE_GP_DELAY 4 /* Roughly one grace period. */ |
778d250a | 1628 | #define RCU_IDLE_LAZY_GP_DELAY (6 * HZ) /* Roughly six seconds. */ |
f23f7fa1 | 1629 | |
5e44ce35 PM |
1630 | static int rcu_idle_gp_delay = RCU_IDLE_GP_DELAY; |
1631 | module_param(rcu_idle_gp_delay, int, 0644); | |
1632 | static int rcu_idle_lazy_gp_delay = RCU_IDLE_LAZY_GP_DELAY; | |
1633 | module_param(rcu_idle_lazy_gp_delay, int, 0644); | |
486e2593 | 1634 | |
9d2ad243 | 1635 | extern int tick_nohz_enabled; |
486e2593 PM |
1636 | |
1637 | /* | |
c0f4dfd4 PM |
1638 | * Try to advance callbacks for all flavors of RCU on the current CPU. |
1639 | * Afterwards, if there are any callbacks ready for immediate invocation, | |
1640 | * return true. | |
486e2593 | 1641 | */ |
c0f4dfd4 | 1642 | static bool rcu_try_advance_all_cbs(void) |
486e2593 | 1643 | { |
c0f4dfd4 PM |
1644 | bool cbs_ready = false; |
1645 | struct rcu_data *rdp; | |
1646 | struct rcu_node *rnp; | |
1647 | struct rcu_state *rsp; | |
486e2593 | 1648 | |
c0f4dfd4 PM |
1649 | for_each_rcu_flavor(rsp) { |
1650 | rdp = this_cpu_ptr(rsp->rda); | |
1651 | rnp = rdp->mynode; | |
486e2593 | 1652 | |
c0f4dfd4 PM |
1653 | /* |
1654 | * Don't bother checking unless a grace period has | |
1655 | * completed since we last checked and there are | |
1656 | * callbacks not yet ready to invoke. | |
1657 | */ | |
1658 | if (rdp->completed != rnp->completed && | |
1659 | rdp->nxttail[RCU_DONE_TAIL] != rdp->nxttail[RCU_NEXT_TAIL]) | |
470716fc | 1660 | note_gp_changes(rsp, rdp); |
486e2593 | 1661 | |
c0f4dfd4 PM |
1662 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
1663 | cbs_ready = true; | |
1664 | } | |
1665 | return cbs_ready; | |
486e2593 PM |
1666 | } |
1667 | ||
aa9b1630 | 1668 | /* |
c0f4dfd4 PM |
1669 | * Allow the CPU to enter dyntick-idle mode unless it has callbacks ready |
1670 | * to invoke. If the CPU has callbacks, try to advance them. Tell the | |
1671 | * caller to set the timeout based on whether or not there are non-lazy | |
1672 | * callbacks. | |
aa9b1630 | 1673 | * |
c0f4dfd4 | 1674 | * The caller must have disabled interrupts. |
aa9b1630 | 1675 | */ |
c0f4dfd4 | 1676 | int rcu_needs_cpu(int cpu, unsigned long *dj) |
aa9b1630 PM |
1677 | { |
1678 | struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); | |
1679 | ||
c0f4dfd4 PM |
1680 | /* Snapshot to detect later posting of non-lazy callback. */ |
1681 | rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted; | |
1682 | ||
aa9b1630 | 1683 | /* If no callbacks, RCU doesn't need the CPU. */ |
c0f4dfd4 PM |
1684 | if (!rcu_cpu_has_callbacks(cpu, &rdtp->all_lazy)) { |
1685 | *dj = ULONG_MAX; | |
aa9b1630 PM |
1686 | return 0; |
1687 | } | |
c0f4dfd4 PM |
1688 | |
1689 | /* Attempt to advance callbacks. */ | |
1690 | if (rcu_try_advance_all_cbs()) { | |
1691 | /* Some ready to invoke, so initiate later invocation. */ | |
1692 | invoke_rcu_core(); | |
aa9b1630 PM |
1693 | return 1; |
1694 | } | |
c0f4dfd4 PM |
1695 | rdtp->last_accelerate = jiffies; |
1696 | ||
1697 | /* Request timer delay depending on laziness, and round. */ | |
6faf7283 | 1698 | if (!rdtp->all_lazy) { |
c0f4dfd4 PM |
1699 | *dj = round_up(rcu_idle_gp_delay + jiffies, |
1700 | rcu_idle_gp_delay) - jiffies; | |
e84c48ae | 1701 | } else { |
c0f4dfd4 | 1702 | *dj = round_jiffies(rcu_idle_lazy_gp_delay + jiffies) - jiffies; |
e84c48ae | 1703 | } |
aa9b1630 PM |
1704 | return 0; |
1705 | } | |
1706 | ||
21e52e15 | 1707 | /* |
c0f4dfd4 PM |
1708 | * Prepare a CPU for idle from an RCU perspective. The first major task |
1709 | * is to sense whether nohz mode has been enabled or disabled via sysfs. | |
1710 | * The second major task is to check to see if a non-lazy callback has | |
1711 | * arrived at a CPU that previously had only lazy callbacks. The third | |
1712 | * major task is to accelerate (that is, assign grace-period numbers to) | |
1713 | * any recently arrived callbacks. | |
aea1b35e PM |
1714 | * |
1715 | * The caller must have disabled interrupts. | |
8bd93a2c | 1716 | */ |
aea1b35e | 1717 | static void rcu_prepare_for_idle(int cpu) |
8bd93a2c | 1718 | { |
c0f4dfd4 | 1719 | struct rcu_data *rdp; |
5955f7ee | 1720 | struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); |
c0f4dfd4 PM |
1721 | struct rcu_node *rnp; |
1722 | struct rcu_state *rsp; | |
9d2ad243 PM |
1723 | int tne; |
1724 | ||
1725 | /* Handle nohz enablement switches conservatively. */ | |
1726 | tne = ACCESS_ONCE(tick_nohz_enabled); | |
1727 | if (tne != rdtp->tick_nohz_enabled_snap) { | |
c0f4dfd4 | 1728 | if (rcu_cpu_has_callbacks(cpu, NULL)) |
9d2ad243 PM |
1729 | invoke_rcu_core(); /* force nohz to see update. */ |
1730 | rdtp->tick_nohz_enabled_snap = tne; | |
1731 | return; | |
1732 | } | |
1733 | if (!tne) | |
1734 | return; | |
f511fc62 | 1735 | |
c0f4dfd4 | 1736 | /* If this is a no-CBs CPU, no callbacks, just return. */ |
534c97b0 | 1737 | if (rcu_is_nocb_cpu(cpu)) |
9a0c6fef | 1738 | return; |
9a0c6fef | 1739 | |
c57afe80 | 1740 | /* |
c0f4dfd4 PM |
1741 | * If a non-lazy callback arrived at a CPU having only lazy |
1742 | * callbacks, invoke RCU core for the side-effect of recalculating | |
1743 | * idle duration on re-entry to idle. | |
c57afe80 | 1744 | */ |
c0f4dfd4 PM |
1745 | if (rdtp->all_lazy && |
1746 | rdtp->nonlazy_posted != rdtp->nonlazy_posted_snap) { | |
1747 | invoke_rcu_core(); | |
c57afe80 PM |
1748 | return; |
1749 | } | |
c57afe80 | 1750 | |
3084f2f8 | 1751 | /* |
c0f4dfd4 PM |
1752 | * If we have not yet accelerated this jiffy, accelerate all |
1753 | * callbacks on this CPU. | |
3084f2f8 | 1754 | */ |
c0f4dfd4 | 1755 | if (rdtp->last_accelerate == jiffies) |
aea1b35e | 1756 | return; |
c0f4dfd4 PM |
1757 | rdtp->last_accelerate = jiffies; |
1758 | for_each_rcu_flavor(rsp) { | |
1759 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
1760 | if (!*rdp->nxttail[RCU_DONE_TAIL]) | |
1761 | continue; | |
1762 | rnp = rdp->mynode; | |
1763 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ | |
1764 | rcu_accelerate_cbs(rsp, rnp, rdp); | |
1765 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ | |
77e38ed3 | 1766 | } |
c0f4dfd4 | 1767 | } |
3084f2f8 | 1768 | |
c0f4dfd4 PM |
1769 | /* |
1770 | * Clean up for exit from idle. Attempt to advance callbacks based on | |
1771 | * any grace periods that elapsed while the CPU was idle, and if any | |
1772 | * callbacks are now ready to invoke, initiate invocation. | |
1773 | */ | |
1774 | static void rcu_cleanup_after_idle(int cpu) | |
1775 | { | |
1776 | struct rcu_data *rdp; | |
1777 | struct rcu_state *rsp; | |
a47cd880 | 1778 | |
534c97b0 | 1779 | if (rcu_is_nocb_cpu(cpu)) |
aea1b35e | 1780 | return; |
c0f4dfd4 PM |
1781 | rcu_try_advance_all_cbs(); |
1782 | for_each_rcu_flavor(rsp) { | |
1783 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
1784 | if (cpu_has_callbacks_ready_to_invoke(rdp)) | |
1785 | invoke_rcu_core(); | |
c701d5d9 | 1786 | } |
8bd93a2c PM |
1787 | } |
1788 | ||
c57afe80 | 1789 | /* |
98248a0e PM |
1790 | * Keep a running count of the number of non-lazy callbacks posted |
1791 | * on this CPU. This running counter (which is never decremented) allows | |
1792 | * rcu_prepare_for_idle() to detect when something out of the idle loop | |
1793 | * posts a callback, even if an equal number of callbacks are invoked. | |
1794 | * Of course, callbacks should only be posted from within a trace event | |
1795 | * designed to be called from idle or from within RCU_NONIDLE(). | |
c57afe80 PM |
1796 | */ |
1797 | static void rcu_idle_count_callbacks_posted(void) | |
1798 | { | |
5955f7ee | 1799 | __this_cpu_add(rcu_dynticks.nonlazy_posted, 1); |
c57afe80 PM |
1800 | } |
1801 | ||
b626c1b6 PM |
1802 | /* |
1803 | * Data for flushing lazy RCU callbacks at OOM time. | |
1804 | */ | |
1805 | static atomic_t oom_callback_count; | |
1806 | static DECLARE_WAIT_QUEUE_HEAD(oom_callback_wq); | |
1807 | ||
1808 | /* | |
1809 | * RCU OOM callback -- decrement the outstanding count and deliver the | |
1810 | * wake-up if we are the last one. | |
1811 | */ | |
1812 | static void rcu_oom_callback(struct rcu_head *rhp) | |
1813 | { | |
1814 | if (atomic_dec_and_test(&oom_callback_count)) | |
1815 | wake_up(&oom_callback_wq); | |
1816 | } | |
1817 | ||
1818 | /* | |
1819 | * Post an rcu_oom_notify callback on the current CPU if it has at | |
1820 | * least one lazy callback. This will unnecessarily post callbacks | |
1821 | * to CPUs that already have a non-lazy callback at the end of their | |
1822 | * callback list, but this is an infrequent operation, so accept some | |
1823 | * extra overhead to keep things simple. | |
1824 | */ | |
1825 | static void rcu_oom_notify_cpu(void *unused) | |
1826 | { | |
1827 | struct rcu_state *rsp; | |
1828 | struct rcu_data *rdp; | |
1829 | ||
1830 | for_each_rcu_flavor(rsp) { | |
1831 | rdp = __this_cpu_ptr(rsp->rda); | |
1832 | if (rdp->qlen_lazy != 0) { | |
1833 | atomic_inc(&oom_callback_count); | |
1834 | rsp->call(&rdp->oom_head, rcu_oom_callback); | |
1835 | } | |
1836 | } | |
1837 | } | |
1838 | ||
1839 | /* | |
1840 | * If low on memory, ensure that each CPU has a non-lazy callback. | |
1841 | * This will wake up CPUs that have only lazy callbacks, in turn | |
1842 | * ensuring that they free up the corresponding memory in a timely manner. | |
1843 | * Because an uncertain amount of memory will be freed in some uncertain | |
1844 | * timeframe, we do not claim to have freed anything. | |
1845 | */ | |
1846 | static int rcu_oom_notify(struct notifier_block *self, | |
1847 | unsigned long notused, void *nfreed) | |
1848 | { | |
1849 | int cpu; | |
1850 | ||
1851 | /* Wait for callbacks from earlier instance to complete. */ | |
1852 | wait_event(oom_callback_wq, atomic_read(&oom_callback_count) == 0); | |
1853 | ||
1854 | /* | |
1855 | * Prevent premature wakeup: ensure that all increments happen | |
1856 | * before there is a chance of the counter reaching zero. | |
1857 | */ | |
1858 | atomic_set(&oom_callback_count, 1); | |
1859 | ||
1860 | get_online_cpus(); | |
1861 | for_each_online_cpu(cpu) { | |
1862 | smp_call_function_single(cpu, rcu_oom_notify_cpu, NULL, 1); | |
1863 | cond_resched(); | |
1864 | } | |
1865 | put_online_cpus(); | |
1866 | ||
1867 | /* Unconditionally decrement: no need to wake ourselves up. */ | |
1868 | atomic_dec(&oom_callback_count); | |
1869 | ||
1870 | return NOTIFY_OK; | |
1871 | } | |
1872 | ||
1873 | static struct notifier_block rcu_oom_nb = { | |
1874 | .notifier_call = rcu_oom_notify | |
1875 | }; | |
1876 | ||
1877 | static int __init rcu_register_oom_notifier(void) | |
1878 | { | |
1879 | register_oom_notifier(&rcu_oom_nb); | |
1880 | return 0; | |
1881 | } | |
1882 | early_initcall(rcu_register_oom_notifier); | |
1883 | ||
8bd93a2c | 1884 | #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
a858af28 PM |
1885 | |
1886 | #ifdef CONFIG_RCU_CPU_STALL_INFO | |
1887 | ||
1888 | #ifdef CONFIG_RCU_FAST_NO_HZ | |
1889 | ||
1890 | static void print_cpu_stall_fast_no_hz(char *cp, int cpu) | |
1891 | { | |
5955f7ee | 1892 | struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); |
c0f4dfd4 | 1893 | unsigned long nlpd = rdtp->nonlazy_posted - rdtp->nonlazy_posted_snap; |
a858af28 | 1894 | |
c0f4dfd4 PM |
1895 | sprintf(cp, "last_accelerate: %04lx/%04lx, nonlazy_posted: %ld, %c%c", |
1896 | rdtp->last_accelerate & 0xffff, jiffies & 0xffff, | |
1897 | ulong2long(nlpd), | |
1898 | rdtp->all_lazy ? 'L' : '.', | |
1899 | rdtp->tick_nohz_enabled_snap ? '.' : 'D'); | |
a858af28 PM |
1900 | } |
1901 | ||
1902 | #else /* #ifdef CONFIG_RCU_FAST_NO_HZ */ | |
1903 | ||
1904 | static void print_cpu_stall_fast_no_hz(char *cp, int cpu) | |
1905 | { | |
1c17e4d4 | 1906 | *cp = '\0'; |
a858af28 PM |
1907 | } |
1908 | ||
1909 | #endif /* #else #ifdef CONFIG_RCU_FAST_NO_HZ */ | |
1910 | ||
1911 | /* Initiate the stall-info list. */ | |
1912 | static void print_cpu_stall_info_begin(void) | |
1913 | { | |
efc151c3 | 1914 | pr_cont("\n"); |
a858af28 PM |
1915 | } |
1916 | ||
1917 | /* | |
1918 | * Print out diagnostic information for the specified stalled CPU. | |
1919 | * | |
1920 | * If the specified CPU is aware of the current RCU grace period | |
1921 | * (flavor specified by rsp), then print the number of scheduling | |
1922 | * clock interrupts the CPU has taken during the time that it has | |
1923 | * been aware. Otherwise, print the number of RCU grace periods | |
1924 | * that this CPU is ignorant of, for example, "1" if the CPU was | |
1925 | * aware of the previous grace period. | |
1926 | * | |
1927 | * Also print out idle and (if CONFIG_RCU_FAST_NO_HZ) idle-entry info. | |
1928 | */ | |
1929 | static void print_cpu_stall_info(struct rcu_state *rsp, int cpu) | |
1930 | { | |
1931 | char fast_no_hz[72]; | |
1932 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); | |
1933 | struct rcu_dynticks *rdtp = rdp->dynticks; | |
1934 | char *ticks_title; | |
1935 | unsigned long ticks_value; | |
1936 | ||
1937 | if (rsp->gpnum == rdp->gpnum) { | |
1938 | ticks_title = "ticks this GP"; | |
1939 | ticks_value = rdp->ticks_this_gp; | |
1940 | } else { | |
1941 | ticks_title = "GPs behind"; | |
1942 | ticks_value = rsp->gpnum - rdp->gpnum; | |
1943 | } | |
1944 | print_cpu_stall_fast_no_hz(fast_no_hz, cpu); | |
efc151c3 | 1945 | pr_err("\t%d: (%lu %s) idle=%03x/%llx/%d softirq=%u/%u %s\n", |
a858af28 PM |
1946 | cpu, ticks_value, ticks_title, |
1947 | atomic_read(&rdtp->dynticks) & 0xfff, | |
1948 | rdtp->dynticks_nesting, rdtp->dynticks_nmi_nesting, | |
6231069b | 1949 | rdp->softirq_snap, kstat_softirqs_cpu(RCU_SOFTIRQ, cpu), |
a858af28 PM |
1950 | fast_no_hz); |
1951 | } | |
1952 | ||
1953 | /* Terminate the stall-info list. */ | |
1954 | static void print_cpu_stall_info_end(void) | |
1955 | { | |
efc151c3 | 1956 | pr_err("\t"); |
a858af28 PM |
1957 | } |
1958 | ||
1959 | /* Zero ->ticks_this_gp for all flavors of RCU. */ | |
1960 | static void zero_cpu_stall_ticks(struct rcu_data *rdp) | |
1961 | { | |
1962 | rdp->ticks_this_gp = 0; | |
6231069b | 1963 | rdp->softirq_snap = kstat_softirqs_cpu(RCU_SOFTIRQ, smp_processor_id()); |
a858af28 PM |
1964 | } |
1965 | ||
1966 | /* Increment ->ticks_this_gp for all flavors of RCU. */ | |
1967 | static void increment_cpu_stall_ticks(void) | |
1968 | { | |
115f7a7c PM |
1969 | struct rcu_state *rsp; |
1970 | ||
1971 | for_each_rcu_flavor(rsp) | |
1972 | __this_cpu_ptr(rsp->rda)->ticks_this_gp++; | |
a858af28 PM |
1973 | } |
1974 | ||
1975 | #else /* #ifdef CONFIG_RCU_CPU_STALL_INFO */ | |
1976 | ||
1977 | static void print_cpu_stall_info_begin(void) | |
1978 | { | |
efc151c3 | 1979 | pr_cont(" {"); |
a858af28 PM |
1980 | } |
1981 | ||
1982 | static void print_cpu_stall_info(struct rcu_state *rsp, int cpu) | |
1983 | { | |
efc151c3 | 1984 | pr_cont(" %d", cpu); |
a858af28 PM |
1985 | } |
1986 | ||
1987 | static void print_cpu_stall_info_end(void) | |
1988 | { | |
efc151c3 | 1989 | pr_cont("} "); |
a858af28 PM |
1990 | } |
1991 | ||
1992 | static void zero_cpu_stall_ticks(struct rcu_data *rdp) | |
1993 | { | |
1994 | } | |
1995 | ||
1996 | static void increment_cpu_stall_ticks(void) | |
1997 | { | |
1998 | } | |
1999 | ||
2000 | #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_INFO */ | |
3fbfbf7a PM |
2001 | |
2002 | #ifdef CONFIG_RCU_NOCB_CPU | |
2003 | ||
2004 | /* | |
2005 | * Offload callback processing from the boot-time-specified set of CPUs | |
2006 | * specified by rcu_nocb_mask. For each CPU in the set, there is a | |
2007 | * kthread created that pulls the callbacks from the corresponding CPU, | |
2008 | * waits for a grace period to elapse, and invokes the callbacks. | |
2009 | * The no-CBs CPUs do a wake_up() on their kthread when they insert | |
2010 | * a callback into any empty list, unless the rcu_nocb_poll boot parameter | |
2011 | * has been specified, in which case each kthread actively polls its | |
2012 | * CPU. (Which isn't so great for energy efficiency, but which does | |
2013 | * reduce RCU's overhead on that CPU.) | |
2014 | * | |
2015 | * This is intended to be used in conjunction with Frederic Weisbecker's | |
2016 | * adaptive-idle work, which would seriously reduce OS jitter on CPUs | |
2017 | * running CPU-bound user-mode computations. | |
2018 | * | |
2019 | * Offloading of callback processing could also in theory be used as | |
2020 | * an energy-efficiency measure because CPUs with no RCU callbacks | |
2021 | * queued are more aggressive about entering dyntick-idle mode. | |
2022 | */ | |
2023 | ||
2024 | ||
2025 | /* Parse the boot-time rcu_nocb_mask CPU list from the kernel parameters. */ | |
2026 | static int __init rcu_nocb_setup(char *str) | |
2027 | { | |
2028 | alloc_bootmem_cpumask_var(&rcu_nocb_mask); | |
2029 | have_rcu_nocb_mask = true; | |
2030 | cpulist_parse(str, rcu_nocb_mask); | |
2031 | return 1; | |
2032 | } | |
2033 | __setup("rcu_nocbs=", rcu_nocb_setup); | |
2034 | ||
1b0048a4 PG |
2035 | static int __init parse_rcu_nocb_poll(char *arg) |
2036 | { | |
2037 | rcu_nocb_poll = 1; | |
2038 | return 0; | |
2039 | } | |
2040 | early_param("rcu_nocb_poll", parse_rcu_nocb_poll); | |
2041 | ||
34ed6246 | 2042 | /* |
dae6e64d PM |
2043 | * Do any no-CBs CPUs need another grace period? |
2044 | * | |
2045 | * Interrupts must be disabled. If the caller does not hold the root | |
2046 | * rnp_node structure's ->lock, the results are advisory only. | |
2047 | */ | |
2048 | static int rcu_nocb_needs_gp(struct rcu_state *rsp) | |
2049 | { | |
2050 | struct rcu_node *rnp = rcu_get_root(rsp); | |
2051 | ||
8b425aa8 | 2052 | return rnp->need_future_gp[(ACCESS_ONCE(rnp->completed) + 1) & 0x1]; |
dae6e64d PM |
2053 | } |
2054 | ||
2055 | /* | |
0446be48 PM |
2056 | * Wake up any no-CBs CPUs' kthreads that were waiting on the just-ended |
2057 | * grace period. | |
dae6e64d | 2058 | */ |
0446be48 | 2059 | static void rcu_nocb_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp) |
dae6e64d | 2060 | { |
0446be48 | 2061 | wake_up_all(&rnp->nocb_gp_wq[rnp->completed & 0x1]); |
dae6e64d PM |
2062 | } |
2063 | ||
2064 | /* | |
8b425aa8 | 2065 | * Set the root rcu_node structure's ->need_future_gp field |
dae6e64d PM |
2066 | * based on the sum of those of all rcu_node structures. This does |
2067 | * double-count the root rcu_node structure's requests, but this | |
2068 | * is necessary to handle the possibility of a rcu_nocb_kthread() | |
2069 | * having awakened during the time that the rcu_node structures | |
2070 | * were being updated for the end of the previous grace period. | |
34ed6246 | 2071 | */ |
dae6e64d PM |
2072 | static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq) |
2073 | { | |
8b425aa8 | 2074 | rnp->need_future_gp[(rnp->completed + 1) & 0x1] += nrq; |
dae6e64d PM |
2075 | } |
2076 | ||
2077 | static void rcu_init_one_nocb(struct rcu_node *rnp) | |
34ed6246 | 2078 | { |
dae6e64d PM |
2079 | init_waitqueue_head(&rnp->nocb_gp_wq[0]); |
2080 | init_waitqueue_head(&rnp->nocb_gp_wq[1]); | |
34ed6246 PM |
2081 | } |
2082 | ||
3fbfbf7a | 2083 | /* Is the specified CPU a no-CPUs CPU? */ |
d1e43fa5 | 2084 | bool rcu_is_nocb_cpu(int cpu) |
3fbfbf7a PM |
2085 | { |
2086 | if (have_rcu_nocb_mask) | |
2087 | return cpumask_test_cpu(cpu, rcu_nocb_mask); | |
2088 | return false; | |
2089 | } | |
2090 | ||
2091 | /* | |
2092 | * Enqueue the specified string of rcu_head structures onto the specified | |
2093 | * CPU's no-CBs lists. The CPU is specified by rdp, the head of the | |
2094 | * string by rhp, and the tail of the string by rhtp. The non-lazy/lazy | |
2095 | * counts are supplied by rhcount and rhcount_lazy. | |
2096 | * | |
2097 | * If warranted, also wake up the kthread servicing this CPUs queues. | |
2098 | */ | |
2099 | static void __call_rcu_nocb_enqueue(struct rcu_data *rdp, | |
2100 | struct rcu_head *rhp, | |
2101 | struct rcu_head **rhtp, | |
2102 | int rhcount, int rhcount_lazy) | |
2103 | { | |
2104 | int len; | |
2105 | struct rcu_head **old_rhpp; | |
2106 | struct task_struct *t; | |
2107 | ||
2108 | /* Enqueue the callback on the nocb list and update counts. */ | |
2109 | old_rhpp = xchg(&rdp->nocb_tail, rhtp); | |
2110 | ACCESS_ONCE(*old_rhpp) = rhp; | |
2111 | atomic_long_add(rhcount, &rdp->nocb_q_count); | |
2112 | atomic_long_add(rhcount_lazy, &rdp->nocb_q_count_lazy); | |
2113 | ||
2114 | /* If we are not being polled and there is a kthread, awaken it ... */ | |
2115 | t = ACCESS_ONCE(rdp->nocb_kthread); | |
829511d8 | 2116 | if (rcu_nocb_poll || !t) |
3fbfbf7a PM |
2117 | return; |
2118 | len = atomic_long_read(&rdp->nocb_q_count); | |
2119 | if (old_rhpp == &rdp->nocb_head) { | |
2120 | wake_up(&rdp->nocb_wq); /* ... only if queue was empty ... */ | |
2121 | rdp->qlen_last_fqs_check = 0; | |
2122 | } else if (len > rdp->qlen_last_fqs_check + qhimark) { | |
2123 | wake_up_process(t); /* ... or if many callbacks queued. */ | |
2124 | rdp->qlen_last_fqs_check = LONG_MAX / 2; | |
2125 | } | |
2126 | return; | |
2127 | } | |
2128 | ||
2129 | /* | |
2130 | * This is a helper for __call_rcu(), which invokes this when the normal | |
2131 | * callback queue is inoperable. If this is not a no-CBs CPU, this | |
2132 | * function returns failure back to __call_rcu(), which can complain | |
2133 | * appropriately. | |
2134 | * | |
2135 | * Otherwise, this function queues the callback where the corresponding | |
2136 | * "rcuo" kthread can find it. | |
2137 | */ | |
2138 | static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, | |
2139 | bool lazy) | |
2140 | { | |
2141 | ||
d1e43fa5 | 2142 | if (!rcu_is_nocb_cpu(rdp->cpu)) |
3fbfbf7a PM |
2143 | return 0; |
2144 | __call_rcu_nocb_enqueue(rdp, rhp, &rhp->next, 1, lazy); | |
21e7a608 PM |
2145 | if (__is_kfree_rcu_offset((unsigned long)rhp->func)) |
2146 | trace_rcu_kfree_callback(rdp->rsp->name, rhp, | |
2147 | (unsigned long)rhp->func, | |
2148 | rdp->qlen_lazy, rdp->qlen); | |
2149 | else | |
2150 | trace_rcu_callback(rdp->rsp->name, rhp, | |
2151 | rdp->qlen_lazy, rdp->qlen); | |
3fbfbf7a PM |
2152 | return 1; |
2153 | } | |
2154 | ||
2155 | /* | |
2156 | * Adopt orphaned callbacks on a no-CBs CPU, or return 0 if this is | |
2157 | * not a no-CBs CPU. | |
2158 | */ | |
2159 | static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp, | |
2160 | struct rcu_data *rdp) | |
2161 | { | |
2162 | long ql = rsp->qlen; | |
2163 | long qll = rsp->qlen_lazy; | |
2164 | ||
2165 | /* If this is not a no-CBs CPU, tell the caller to do it the old way. */ | |
d1e43fa5 | 2166 | if (!rcu_is_nocb_cpu(smp_processor_id())) |
3fbfbf7a PM |
2167 | return 0; |
2168 | rsp->qlen = 0; | |
2169 | rsp->qlen_lazy = 0; | |
2170 | ||
2171 | /* First, enqueue the donelist, if any. This preserves CB ordering. */ | |
2172 | if (rsp->orphan_donelist != NULL) { | |
2173 | __call_rcu_nocb_enqueue(rdp, rsp->orphan_donelist, | |
2174 | rsp->orphan_donetail, ql, qll); | |
2175 | ql = qll = 0; | |
2176 | rsp->orphan_donelist = NULL; | |
2177 | rsp->orphan_donetail = &rsp->orphan_donelist; | |
2178 | } | |
2179 | if (rsp->orphan_nxtlist != NULL) { | |
2180 | __call_rcu_nocb_enqueue(rdp, rsp->orphan_nxtlist, | |
2181 | rsp->orphan_nxttail, ql, qll); | |
2182 | ql = qll = 0; | |
2183 | rsp->orphan_nxtlist = NULL; | |
2184 | rsp->orphan_nxttail = &rsp->orphan_nxtlist; | |
2185 | } | |
2186 | return 1; | |
2187 | } | |
2188 | ||
2189 | /* | |
34ed6246 PM |
2190 | * If necessary, kick off a new grace period, and either way wait |
2191 | * for a subsequent grace period to complete. | |
3fbfbf7a | 2192 | */ |
34ed6246 | 2193 | static void rcu_nocb_wait_gp(struct rcu_data *rdp) |
3fbfbf7a | 2194 | { |
34ed6246 | 2195 | unsigned long c; |
dae6e64d | 2196 | bool d; |
34ed6246 | 2197 | unsigned long flags; |
34ed6246 PM |
2198 | struct rcu_node *rnp = rdp->mynode; |
2199 | ||
2200 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
0446be48 PM |
2201 | c = rcu_start_future_gp(rnp, rdp); |
2202 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
3fbfbf7a PM |
2203 | |
2204 | /* | |
34ed6246 PM |
2205 | * Wait for the grace period. Do so interruptibly to avoid messing |
2206 | * up the load average. | |
3fbfbf7a | 2207 | */ |
f7f7bac9 | 2208 | trace_rcu_future_gp(rnp, rdp, c, TPS("StartWait")); |
34ed6246 | 2209 | for (;;) { |
dae6e64d PM |
2210 | wait_event_interruptible( |
2211 | rnp->nocb_gp_wq[c & 0x1], | |
2212 | (d = ULONG_CMP_GE(ACCESS_ONCE(rnp->completed), c))); | |
2213 | if (likely(d)) | |
34ed6246 | 2214 | break; |
dae6e64d | 2215 | flush_signals(current); |
f7f7bac9 | 2216 | trace_rcu_future_gp(rnp, rdp, c, TPS("ResumeWait")); |
34ed6246 | 2217 | } |
f7f7bac9 | 2218 | trace_rcu_future_gp(rnp, rdp, c, TPS("EndWait")); |
34ed6246 | 2219 | smp_mb(); /* Ensure that CB invocation happens after GP end. */ |
3fbfbf7a PM |
2220 | } |
2221 | ||
2222 | /* | |
2223 | * Per-rcu_data kthread, but only for no-CBs CPUs. Each kthread invokes | |
2224 | * callbacks queued by the corresponding no-CBs CPU. | |
2225 | */ | |
2226 | static int rcu_nocb_kthread(void *arg) | |
2227 | { | |
2228 | int c, cl; | |
2229 | struct rcu_head *list; | |
2230 | struct rcu_head *next; | |
2231 | struct rcu_head **tail; | |
2232 | struct rcu_data *rdp = arg; | |
2233 | ||
2234 | /* Each pass through this loop invokes one batch of callbacks */ | |
2235 | for (;;) { | |
2236 | /* If not polling, wait for next batch of callbacks. */ | |
2237 | if (!rcu_nocb_poll) | |
353af9c9 | 2238 | wait_event_interruptible(rdp->nocb_wq, rdp->nocb_head); |
3fbfbf7a PM |
2239 | list = ACCESS_ONCE(rdp->nocb_head); |
2240 | if (!list) { | |
2241 | schedule_timeout_interruptible(1); | |
353af9c9 | 2242 | flush_signals(current); |
3fbfbf7a PM |
2243 | continue; |
2244 | } | |
2245 | ||
2246 | /* | |
2247 | * Extract queued callbacks, update counts, and wait | |
2248 | * for a grace period to elapse. | |
2249 | */ | |
2250 | ACCESS_ONCE(rdp->nocb_head) = NULL; | |
2251 | tail = xchg(&rdp->nocb_tail, &rdp->nocb_head); | |
2252 | c = atomic_long_xchg(&rdp->nocb_q_count, 0); | |
2253 | cl = atomic_long_xchg(&rdp->nocb_q_count_lazy, 0); | |
2254 | ACCESS_ONCE(rdp->nocb_p_count) += c; | |
2255 | ACCESS_ONCE(rdp->nocb_p_count_lazy) += cl; | |
34ed6246 | 2256 | rcu_nocb_wait_gp(rdp); |
3fbfbf7a PM |
2257 | |
2258 | /* Each pass through the following loop invokes a callback. */ | |
2259 | trace_rcu_batch_start(rdp->rsp->name, cl, c, -1); | |
2260 | c = cl = 0; | |
2261 | while (list) { | |
2262 | next = list->next; | |
2263 | /* Wait for enqueuing to complete, if needed. */ | |
2264 | while (next == NULL && &list->next != tail) { | |
2265 | schedule_timeout_interruptible(1); | |
2266 | next = list->next; | |
2267 | } | |
2268 | debug_rcu_head_unqueue(list); | |
2269 | local_bh_disable(); | |
2270 | if (__rcu_reclaim(rdp->rsp->name, list)) | |
2271 | cl++; | |
2272 | c++; | |
2273 | local_bh_enable(); | |
2274 | list = next; | |
2275 | } | |
2276 | trace_rcu_batch_end(rdp->rsp->name, c, !!list, 0, 0, 1); | |
2277 | ACCESS_ONCE(rdp->nocb_p_count) -= c; | |
2278 | ACCESS_ONCE(rdp->nocb_p_count_lazy) -= cl; | |
c635a4e1 | 2279 | rdp->n_nocbs_invoked += c; |
3fbfbf7a PM |
2280 | } |
2281 | return 0; | |
2282 | } | |
2283 | ||
2284 | /* Initialize per-rcu_data variables for no-CBs CPUs. */ | |
2285 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) | |
2286 | { | |
2287 | rdp->nocb_tail = &rdp->nocb_head; | |
2288 | init_waitqueue_head(&rdp->nocb_wq); | |
2289 | } | |
2290 | ||
2291 | /* Create a kthread for each RCU flavor for each no-CBs CPU. */ | |
2292 | static void __init rcu_spawn_nocb_kthreads(struct rcu_state *rsp) | |
2293 | { | |
2294 | int cpu; | |
2295 | struct rcu_data *rdp; | |
2296 | struct task_struct *t; | |
2297 | ||
2298 | if (rcu_nocb_mask == NULL) | |
2299 | return; | |
2300 | for_each_cpu(cpu, rcu_nocb_mask) { | |
2301 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
a4889858 PM |
2302 | t = kthread_run(rcu_nocb_kthread, rdp, |
2303 | "rcuo%c/%d", rsp->abbr, cpu); | |
3fbfbf7a PM |
2304 | BUG_ON(IS_ERR(t)); |
2305 | ACCESS_ONCE(rdp->nocb_kthread) = t; | |
2306 | } | |
2307 | } | |
2308 | ||
2309 | /* Prevent __call_rcu() from enqueuing callbacks on no-CBs CPUs */ | |
34ed6246 | 2310 | static bool init_nocb_callback_list(struct rcu_data *rdp) |
3fbfbf7a PM |
2311 | { |
2312 | if (rcu_nocb_mask == NULL || | |
2313 | !cpumask_test_cpu(rdp->cpu, rcu_nocb_mask)) | |
34ed6246 | 2314 | return false; |
3fbfbf7a | 2315 | rdp->nxttail[RCU_NEXT_TAIL] = NULL; |
34ed6246 | 2316 | return true; |
3fbfbf7a PM |
2317 | } |
2318 | ||
34ed6246 PM |
2319 | #else /* #ifdef CONFIG_RCU_NOCB_CPU */ |
2320 | ||
dae6e64d PM |
2321 | static int rcu_nocb_needs_gp(struct rcu_state *rsp) |
2322 | { | |
2323 | return 0; | |
3fbfbf7a PM |
2324 | } |
2325 | ||
0446be48 | 2326 | static void rcu_nocb_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp) |
3fbfbf7a | 2327 | { |
3fbfbf7a PM |
2328 | } |
2329 | ||
dae6e64d PM |
2330 | static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq) |
2331 | { | |
2332 | } | |
2333 | ||
2334 | static void rcu_init_one_nocb(struct rcu_node *rnp) | |
2335 | { | |
2336 | } | |
3fbfbf7a | 2337 | |
3fbfbf7a PM |
2338 | static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, |
2339 | bool lazy) | |
2340 | { | |
2341 | return 0; | |
2342 | } | |
2343 | ||
2344 | static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp, | |
2345 | struct rcu_data *rdp) | |
2346 | { | |
2347 | return 0; | |
2348 | } | |
2349 | ||
3fbfbf7a PM |
2350 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) |
2351 | { | |
2352 | } | |
2353 | ||
2354 | static void __init rcu_spawn_nocb_kthreads(struct rcu_state *rsp) | |
2355 | { | |
2356 | } | |
2357 | ||
34ed6246 | 2358 | static bool init_nocb_callback_list(struct rcu_data *rdp) |
3fbfbf7a | 2359 | { |
34ed6246 | 2360 | return false; |
3fbfbf7a PM |
2361 | } |
2362 | ||
2363 | #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ | |
65d798f0 PM |
2364 | |
2365 | /* | |
2366 | * An adaptive-ticks CPU can potentially execute in kernel mode for an | |
2367 | * arbitrarily long period of time with the scheduling-clock tick turned | |
2368 | * off. RCU will be paying attention to this CPU because it is in the | |
2369 | * kernel, but the CPU cannot be guaranteed to be executing the RCU state | |
2370 | * machine because the scheduling-clock tick has been disabled. Therefore, | |
2371 | * if an adaptive-ticks CPU is failing to respond to the current grace | |
2372 | * period and has not be idle from an RCU perspective, kick it. | |
2373 | */ | |
2374 | static void rcu_kick_nohz_cpu(int cpu) | |
2375 | { | |
2376 | #ifdef CONFIG_NO_HZ_FULL | |
2377 | if (tick_nohz_full_cpu(cpu)) | |
2378 | smp_send_reschedule(cpu); | |
2379 | #endif /* #ifdef CONFIG_NO_HZ_FULL */ | |
2380 | } | |
2333210b PM |
2381 | |
2382 | ||
2383 | #ifdef CONFIG_NO_HZ_FULL_SYSIDLE | |
2384 | ||
d4bd54fb PM |
2385 | /* |
2386 | * Define RCU flavor that holds sysidle state. This needs to be the | |
2387 | * most active flavor of RCU. | |
2388 | */ | |
2389 | #ifdef CONFIG_PREEMPT_RCU | |
0edd1b17 | 2390 | static struct rcu_state *rcu_sysidle_state = &rcu_preempt_state; |
d4bd54fb | 2391 | #else /* #ifdef CONFIG_PREEMPT_RCU */ |
0edd1b17 | 2392 | static struct rcu_state *rcu_sysidle_state = &rcu_sched_state; |
d4bd54fb PM |
2393 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ |
2394 | ||
0edd1b17 | 2395 | static int full_sysidle_state; /* Current system-idle state. */ |
d4bd54fb PM |
2396 | #define RCU_SYSIDLE_NOT 0 /* Some CPU is not idle. */ |
2397 | #define RCU_SYSIDLE_SHORT 1 /* All CPUs idle for brief period. */ | |
2398 | #define RCU_SYSIDLE_LONG 2 /* All CPUs idle for long enough. */ | |
2399 | #define RCU_SYSIDLE_FULL 3 /* All CPUs idle, ready for sysidle. */ | |
2400 | #define RCU_SYSIDLE_FULL_NOTED 4 /* Actually entered sysidle state. */ | |
2401 | ||
eb348b89 PM |
2402 | /* |
2403 | * Invoked to note exit from irq or task transition to idle. Note that | |
2404 | * usermode execution does -not- count as idle here! After all, we want | |
2405 | * to detect full-system idle states, not RCU quiescent states and grace | |
2406 | * periods. The caller must have disabled interrupts. | |
2407 | */ | |
2408 | static void rcu_sysidle_enter(struct rcu_dynticks *rdtp, int irq) | |
2409 | { | |
2410 | unsigned long j; | |
2411 | ||
2412 | /* Adjust nesting, check for fully idle. */ | |
2413 | if (irq) { | |
2414 | rdtp->dynticks_idle_nesting--; | |
2415 | WARN_ON_ONCE(rdtp->dynticks_idle_nesting < 0); | |
2416 | if (rdtp->dynticks_idle_nesting != 0) | |
2417 | return; /* Still not fully idle. */ | |
2418 | } else { | |
2419 | if ((rdtp->dynticks_idle_nesting & DYNTICK_TASK_NEST_MASK) == | |
2420 | DYNTICK_TASK_NEST_VALUE) { | |
2421 | rdtp->dynticks_idle_nesting = 0; | |
2422 | } else { | |
2423 | rdtp->dynticks_idle_nesting -= DYNTICK_TASK_NEST_VALUE; | |
2424 | WARN_ON_ONCE(rdtp->dynticks_idle_nesting < 0); | |
2425 | return; /* Still not fully idle. */ | |
2426 | } | |
2427 | } | |
2428 | ||
2429 | /* Record start of fully idle period. */ | |
2430 | j = jiffies; | |
2431 | ACCESS_ONCE(rdtp->dynticks_idle_jiffies) = j; | |
2432 | smp_mb__before_atomic_inc(); | |
2433 | atomic_inc(&rdtp->dynticks_idle); | |
2434 | smp_mb__after_atomic_inc(); | |
2435 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks_idle) & 0x1); | |
2436 | } | |
2437 | ||
0edd1b17 PM |
2438 | /* |
2439 | * Unconditionally force exit from full system-idle state. This is | |
2440 | * invoked when a normal CPU exits idle, but must be called separately | |
2441 | * for the timekeeping CPU (tick_do_timer_cpu). The reason for this | |
2442 | * is that the timekeeping CPU is permitted to take scheduling-clock | |
2443 | * interrupts while the system is in system-idle state, and of course | |
2444 | * rcu_sysidle_exit() has no way of distinguishing a scheduling-clock | |
2445 | * interrupt from any other type of interrupt. | |
2446 | */ | |
2447 | void rcu_sysidle_force_exit(void) | |
2448 | { | |
2449 | int oldstate = ACCESS_ONCE(full_sysidle_state); | |
2450 | int newoldstate; | |
2451 | ||
2452 | /* | |
2453 | * Each pass through the following loop attempts to exit full | |
2454 | * system-idle state. If contention proves to be a problem, | |
2455 | * a trylock-based contention tree could be used here. | |
2456 | */ | |
2457 | while (oldstate > RCU_SYSIDLE_SHORT) { | |
2458 | newoldstate = cmpxchg(&full_sysidle_state, | |
2459 | oldstate, RCU_SYSIDLE_NOT); | |
2460 | if (oldstate == newoldstate && | |
2461 | oldstate == RCU_SYSIDLE_FULL_NOTED) { | |
2462 | rcu_kick_nohz_cpu(tick_do_timer_cpu); | |
2463 | return; /* We cleared it, done! */ | |
2464 | } | |
2465 | oldstate = newoldstate; | |
2466 | } | |
2467 | smp_mb(); /* Order initial oldstate fetch vs. later non-idle work. */ | |
2468 | } | |
2469 | ||
eb348b89 PM |
2470 | /* |
2471 | * Invoked to note entry to irq or task transition from idle. Note that | |
2472 | * usermode execution does -not- count as idle here! The caller must | |
2473 | * have disabled interrupts. | |
2474 | */ | |
2475 | static void rcu_sysidle_exit(struct rcu_dynticks *rdtp, int irq) | |
2476 | { | |
2477 | /* Adjust nesting, check for already non-idle. */ | |
2478 | if (irq) { | |
2479 | rdtp->dynticks_idle_nesting++; | |
2480 | WARN_ON_ONCE(rdtp->dynticks_idle_nesting <= 0); | |
2481 | if (rdtp->dynticks_idle_nesting != 1) | |
2482 | return; /* Already non-idle. */ | |
2483 | } else { | |
2484 | /* | |
2485 | * Allow for irq misnesting. Yes, it really is possible | |
2486 | * to enter an irq handler then never leave it, and maybe | |
2487 | * also vice versa. Handle both possibilities. | |
2488 | */ | |
2489 | if (rdtp->dynticks_idle_nesting & DYNTICK_TASK_NEST_MASK) { | |
2490 | rdtp->dynticks_idle_nesting += DYNTICK_TASK_NEST_VALUE; | |
2491 | WARN_ON_ONCE(rdtp->dynticks_idle_nesting <= 0); | |
2492 | return; /* Already non-idle. */ | |
2493 | } else { | |
2494 | rdtp->dynticks_idle_nesting = DYNTICK_TASK_EXIT_IDLE; | |
2495 | } | |
2496 | } | |
2497 | ||
2498 | /* Record end of idle period. */ | |
2499 | smp_mb__before_atomic_inc(); | |
2500 | atomic_inc(&rdtp->dynticks_idle); | |
2501 | smp_mb__after_atomic_inc(); | |
2502 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks_idle) & 0x1)); | |
0edd1b17 PM |
2503 | |
2504 | /* | |
2505 | * If we are the timekeeping CPU, we are permitted to be non-idle | |
2506 | * during a system-idle state. This must be the case, because | |
2507 | * the timekeeping CPU has to take scheduling-clock interrupts | |
2508 | * during the time that the system is transitioning to full | |
2509 | * system-idle state. This means that the timekeeping CPU must | |
2510 | * invoke rcu_sysidle_force_exit() directly if it does anything | |
2511 | * more than take a scheduling-clock interrupt. | |
2512 | */ | |
2513 | if (smp_processor_id() == tick_do_timer_cpu) | |
2514 | return; | |
2515 | ||
2516 | /* Update system-idle state: We are clearly no longer fully idle! */ | |
2517 | rcu_sysidle_force_exit(); | |
2518 | } | |
2519 | ||
2520 | /* | |
2521 | * Check to see if the current CPU is idle. Note that usermode execution | |
2522 | * does not count as idle. The caller must have disabled interrupts. | |
2523 | */ | |
2524 | static void rcu_sysidle_check_cpu(struct rcu_data *rdp, bool *isidle, | |
2525 | unsigned long *maxj) | |
2526 | { | |
2527 | int cur; | |
2528 | unsigned long j; | |
2529 | struct rcu_dynticks *rdtp = rdp->dynticks; | |
2530 | ||
2531 | /* | |
2532 | * If some other CPU has already reported non-idle, if this is | |
2533 | * not the flavor of RCU that tracks sysidle state, or if this | |
2534 | * is an offline or the timekeeping CPU, nothing to do. | |
2535 | */ | |
2536 | if (!*isidle || rdp->rsp != rcu_sysidle_state || | |
2537 | cpu_is_offline(rdp->cpu) || rdp->cpu == tick_do_timer_cpu) | |
2538 | return; | |
eb75767b PM |
2539 | if (rcu_gp_in_progress(rdp->rsp)) |
2540 | WARN_ON_ONCE(smp_processor_id() != tick_do_timer_cpu); | |
0edd1b17 PM |
2541 | |
2542 | /* Pick up current idle and NMI-nesting counter and check. */ | |
2543 | cur = atomic_read(&rdtp->dynticks_idle); | |
2544 | if (cur & 0x1) { | |
2545 | *isidle = false; /* We are not idle! */ | |
2546 | return; | |
2547 | } | |
2548 | smp_mb(); /* Read counters before timestamps. */ | |
2549 | ||
2550 | /* Pick up timestamps. */ | |
2551 | j = ACCESS_ONCE(rdtp->dynticks_idle_jiffies); | |
2552 | /* If this CPU entered idle more recently, update maxj timestamp. */ | |
2553 | if (ULONG_CMP_LT(*maxj, j)) | |
2554 | *maxj = j; | |
2555 | } | |
2556 | ||
2557 | /* | |
2558 | * Is this the flavor of RCU that is handling full-system idle? | |
2559 | */ | |
2560 | static bool is_sysidle_rcu_state(struct rcu_state *rsp) | |
2561 | { | |
2562 | return rsp == rcu_sysidle_state; | |
2563 | } | |
2564 | ||
eb75767b PM |
2565 | /* |
2566 | * Bind the grace-period kthread for the sysidle flavor of RCU to the | |
2567 | * timekeeping CPU. | |
2568 | */ | |
2569 | static void rcu_bind_gp_kthread(void) | |
2570 | { | |
2571 | int cpu = ACCESS_ONCE(tick_do_timer_cpu); | |
2572 | ||
2573 | if (cpu < 0 || cpu >= nr_cpu_ids) | |
2574 | return; | |
2575 | if (raw_smp_processor_id() != cpu) | |
2576 | set_cpus_allowed_ptr(current, cpumask_of(cpu)); | |
2577 | } | |
2578 | ||
0edd1b17 PM |
2579 | /* |
2580 | * Return a delay in jiffies based on the number of CPUs, rcu_node | |
2581 | * leaf fanout, and jiffies tick rate. The idea is to allow larger | |
2582 | * systems more time to transition to full-idle state in order to | |
2583 | * avoid the cache thrashing that otherwise occur on the state variable. | |
2584 | * Really small systems (less than a couple of tens of CPUs) should | |
2585 | * instead use a single global atomically incremented counter, and later | |
2586 | * versions of this will automatically reconfigure themselves accordingly. | |
2587 | */ | |
2588 | static unsigned long rcu_sysidle_delay(void) | |
2589 | { | |
2590 | if (nr_cpu_ids <= CONFIG_NO_HZ_FULL_SYSIDLE_SMALL) | |
2591 | return 0; | |
2592 | return DIV_ROUND_UP(nr_cpu_ids * HZ, rcu_fanout_leaf * 1000); | |
2593 | } | |
2594 | ||
2595 | /* | |
2596 | * Advance the full-system-idle state. This is invoked when all of | |
2597 | * the non-timekeeping CPUs are idle. | |
2598 | */ | |
2599 | static void rcu_sysidle(unsigned long j) | |
2600 | { | |
2601 | /* Check the current state. */ | |
2602 | switch (ACCESS_ONCE(full_sysidle_state)) { | |
2603 | case RCU_SYSIDLE_NOT: | |
2604 | ||
2605 | /* First time all are idle, so note a short idle period. */ | |
2606 | ACCESS_ONCE(full_sysidle_state) = RCU_SYSIDLE_SHORT; | |
2607 | break; | |
2608 | ||
2609 | case RCU_SYSIDLE_SHORT: | |
2610 | ||
2611 | /* | |
2612 | * Idle for a bit, time to advance to next state? | |
2613 | * cmpxchg failure means race with non-idle, let them win. | |
2614 | */ | |
2615 | if (ULONG_CMP_GE(jiffies, j + rcu_sysidle_delay())) | |
2616 | (void)cmpxchg(&full_sysidle_state, | |
2617 | RCU_SYSIDLE_SHORT, RCU_SYSIDLE_LONG); | |
2618 | break; | |
2619 | ||
2620 | case RCU_SYSIDLE_LONG: | |
2621 | ||
2622 | /* | |
2623 | * Do an additional check pass before advancing to full. | |
2624 | * cmpxchg failure means race with non-idle, let them win. | |
2625 | */ | |
2626 | if (ULONG_CMP_GE(jiffies, j + rcu_sysidle_delay())) | |
2627 | (void)cmpxchg(&full_sysidle_state, | |
2628 | RCU_SYSIDLE_LONG, RCU_SYSIDLE_FULL); | |
2629 | break; | |
2630 | ||
2631 | default: | |
2632 | break; | |
2633 | } | |
2634 | } | |
2635 | ||
2636 | /* | |
2637 | * Found a non-idle non-timekeeping CPU, so kick the system-idle state | |
2638 | * back to the beginning. | |
2639 | */ | |
2640 | static void rcu_sysidle_cancel(void) | |
2641 | { | |
2642 | smp_mb(); | |
2643 | ACCESS_ONCE(full_sysidle_state) = RCU_SYSIDLE_NOT; | |
2644 | } | |
2645 | ||
2646 | /* | |
2647 | * Update the sysidle state based on the results of a force-quiescent-state | |
2648 | * scan of the CPUs' dyntick-idle state. | |
2649 | */ | |
2650 | static void rcu_sysidle_report(struct rcu_state *rsp, int isidle, | |
2651 | unsigned long maxj, bool gpkt) | |
2652 | { | |
2653 | if (rsp != rcu_sysidle_state) | |
2654 | return; /* Wrong flavor, ignore. */ | |
2655 | if (gpkt && nr_cpu_ids <= CONFIG_NO_HZ_FULL_SYSIDLE_SMALL) | |
2656 | return; /* Running state machine from timekeeping CPU. */ | |
2657 | if (isidle) | |
2658 | rcu_sysidle(maxj); /* More idle! */ | |
2659 | else | |
2660 | rcu_sysidle_cancel(); /* Idle is over. */ | |
2661 | } | |
2662 | ||
2663 | /* | |
2664 | * Wrapper for rcu_sysidle_report() when called from the grace-period | |
2665 | * kthread's context. | |
2666 | */ | |
2667 | static void rcu_sysidle_report_gp(struct rcu_state *rsp, int isidle, | |
2668 | unsigned long maxj) | |
2669 | { | |
2670 | rcu_sysidle_report(rsp, isidle, maxj, true); | |
2671 | } | |
2672 | ||
2673 | /* Callback and function for forcing an RCU grace period. */ | |
2674 | struct rcu_sysidle_head { | |
2675 | struct rcu_head rh; | |
2676 | int inuse; | |
2677 | }; | |
2678 | ||
2679 | static void rcu_sysidle_cb(struct rcu_head *rhp) | |
2680 | { | |
2681 | struct rcu_sysidle_head *rshp; | |
2682 | ||
2683 | /* | |
2684 | * The following memory barrier is needed to replace the | |
2685 | * memory barriers that would normally be in the memory | |
2686 | * allocator. | |
2687 | */ | |
2688 | smp_mb(); /* grace period precedes setting inuse. */ | |
2689 | ||
2690 | rshp = container_of(rhp, struct rcu_sysidle_head, rh); | |
2691 | ACCESS_ONCE(rshp->inuse) = 0; | |
2692 | } | |
2693 | ||
2694 | /* | |
2695 | * Check to see if the system is fully idle, other than the timekeeping CPU. | |
2696 | * The caller must have disabled interrupts. | |
2697 | */ | |
2698 | bool rcu_sys_is_idle(void) | |
2699 | { | |
2700 | static struct rcu_sysidle_head rsh; | |
2701 | int rss = ACCESS_ONCE(full_sysidle_state); | |
2702 | ||
2703 | if (WARN_ON_ONCE(smp_processor_id() != tick_do_timer_cpu)) | |
2704 | return false; | |
2705 | ||
2706 | /* Handle small-system case by doing a full scan of CPUs. */ | |
2707 | if (nr_cpu_ids <= CONFIG_NO_HZ_FULL_SYSIDLE_SMALL) { | |
2708 | int oldrss = rss - 1; | |
2709 | ||
2710 | /* | |
2711 | * One pass to advance to each state up to _FULL. | |
2712 | * Give up if any pass fails to advance the state. | |
2713 | */ | |
2714 | while (rss < RCU_SYSIDLE_FULL && oldrss < rss) { | |
2715 | int cpu; | |
2716 | bool isidle = true; | |
2717 | unsigned long maxj = jiffies - ULONG_MAX / 4; | |
2718 | struct rcu_data *rdp; | |
2719 | ||
2720 | /* Scan all the CPUs looking for nonidle CPUs. */ | |
2721 | for_each_possible_cpu(cpu) { | |
2722 | rdp = per_cpu_ptr(rcu_sysidle_state->rda, cpu); | |
2723 | rcu_sysidle_check_cpu(rdp, &isidle, &maxj); | |
2724 | if (!isidle) | |
2725 | break; | |
2726 | } | |
2727 | rcu_sysidle_report(rcu_sysidle_state, | |
2728 | isidle, maxj, false); | |
2729 | oldrss = rss; | |
2730 | rss = ACCESS_ONCE(full_sysidle_state); | |
2731 | } | |
2732 | } | |
2733 | ||
2734 | /* If this is the first observation of an idle period, record it. */ | |
2735 | if (rss == RCU_SYSIDLE_FULL) { | |
2736 | rss = cmpxchg(&full_sysidle_state, | |
2737 | RCU_SYSIDLE_FULL, RCU_SYSIDLE_FULL_NOTED); | |
2738 | return rss == RCU_SYSIDLE_FULL; | |
2739 | } | |
2740 | ||
2741 | smp_mb(); /* ensure rss load happens before later caller actions. */ | |
2742 | ||
2743 | /* If already fully idle, tell the caller (in case of races). */ | |
2744 | if (rss == RCU_SYSIDLE_FULL_NOTED) | |
2745 | return true; | |
2746 | ||
2747 | /* | |
2748 | * If we aren't there yet, and a grace period is not in flight, | |
2749 | * initiate a grace period. Either way, tell the caller that | |
2750 | * we are not there yet. We use an xchg() rather than an assignment | |
2751 | * to make up for the memory barriers that would otherwise be | |
2752 | * provided by the memory allocator. | |
2753 | */ | |
2754 | if (nr_cpu_ids > CONFIG_NO_HZ_FULL_SYSIDLE_SMALL && | |
2755 | !rcu_gp_in_progress(rcu_sysidle_state) && | |
2756 | !rsh.inuse && xchg(&rsh.inuse, 1) == 0) | |
2757 | call_rcu(&rsh.rh, rcu_sysidle_cb); | |
2758 | return false; | |
eb348b89 PM |
2759 | } |
2760 | ||
2333210b PM |
2761 | /* |
2762 | * Initialize dynticks sysidle state for CPUs coming online. | |
2763 | */ | |
2764 | static void rcu_sysidle_init_percpu_data(struct rcu_dynticks *rdtp) | |
2765 | { | |
2766 | rdtp->dynticks_idle_nesting = DYNTICK_TASK_NEST_VALUE; | |
2767 | } | |
2768 | ||
2769 | #else /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ | |
2770 | ||
eb348b89 PM |
2771 | static void rcu_sysidle_enter(struct rcu_dynticks *rdtp, int irq) |
2772 | { | |
2773 | } | |
2774 | ||
2775 | static void rcu_sysidle_exit(struct rcu_dynticks *rdtp, int irq) | |
2776 | { | |
2777 | } | |
2778 | ||
0edd1b17 PM |
2779 | static void rcu_sysidle_check_cpu(struct rcu_data *rdp, bool *isidle, |
2780 | unsigned long *maxj) | |
2781 | { | |
2782 | } | |
2783 | ||
2784 | static bool is_sysidle_rcu_state(struct rcu_state *rsp) | |
2785 | { | |
2786 | return false; | |
2787 | } | |
2788 | ||
eb75767b PM |
2789 | static void rcu_bind_gp_kthread(void) |
2790 | { | |
2791 | } | |
2792 | ||
0edd1b17 PM |
2793 | static void rcu_sysidle_report_gp(struct rcu_state *rsp, int isidle, |
2794 | unsigned long maxj) | |
2795 | { | |
2796 | } | |
2797 | ||
2333210b PM |
2798 | static void rcu_sysidle_init_percpu_data(struct rcu_dynticks *rdtp) |
2799 | { | |
2800 | } | |
2801 | ||
2802 | #endif /* #else #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ |