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 | |
87de1cfd PM |
17 | * along with this program; if not, you can access it online at |
18 | * http://www.gnu.org/licenses/gpl-2.0.html. | |
f41d911f PM |
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> |
b17b0153 | 30 | #include <linux/sched/debug.h> |
62ab7072 | 31 | #include <linux/smpboot.h> |
ae7e81c0 | 32 | #include <uapi/linux/sched/types.h> |
4102adab | 33 | #include "../time/tick-internal.h" |
f41d911f | 34 | |
5b61b0ba | 35 | #ifdef CONFIG_RCU_BOOST |
61cfd097 | 36 | |
abaa93d9 | 37 | #include "../locking/rtmutex_common.h" |
21871d7e | 38 | |
61cfd097 PM |
39 | /* |
40 | * Control variables for per-CPU and per-rcu_node kthreads. These | |
41 | * handle all flavors of RCU. | |
42 | */ | |
43 | static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task); | |
44 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status); | |
45 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops); | |
46 | DEFINE_PER_CPU(char, rcu_cpu_has_work); | |
47 | ||
727b705b PM |
48 | #else /* #ifdef CONFIG_RCU_BOOST */ |
49 | ||
50 | /* | |
51 | * Some architectures do not define rt_mutexes, but if !CONFIG_RCU_BOOST, | |
52 | * all uses are in dead code. Provide a definition to keep the compiler | |
53 | * happy, but add WARN_ON_ONCE() to complain if used in the wrong place. | |
54 | * This probably needs to be excluded from -rt builds. | |
55 | */ | |
56 | #define rt_mutex_owner(a) ({ WARN_ON_ONCE(1); NULL; }) | |
57 | ||
58 | #endif /* #else #ifdef CONFIG_RCU_BOOST */ | |
5b61b0ba | 59 | |
3fbfbf7a PM |
60 | #ifdef CONFIG_RCU_NOCB_CPU |
61 | static cpumask_var_t rcu_nocb_mask; /* CPUs to have callbacks offloaded. */ | |
62 | static bool have_rcu_nocb_mask; /* Was rcu_nocb_mask allocated? */ | |
1b0048a4 | 63 | static bool __read_mostly rcu_nocb_poll; /* Offload kthread are to poll. */ |
3fbfbf7a PM |
64 | #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ |
65 | ||
26845c28 PM |
66 | /* |
67 | * Check the RCU kernel configuration parameters and print informative | |
699d4035 | 68 | * messages about anything out of the ordinary. |
26845c28 PM |
69 | */ |
70 | static void __init rcu_bootup_announce_oddness(void) | |
71 | { | |
ab6f5bd6 PM |
72 | if (IS_ENABLED(CONFIG_RCU_TRACE)) |
73 | pr_info("\tRCU debugfs-based tracing is enabled.\n"); | |
05c5df31 PM |
74 | if ((IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 64) || |
75 | (!IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 32)) | |
ab6f5bd6 | 76 | pr_info("\tCONFIG_RCU_FANOUT set to non-default value of %d\n", |
05c5df31 | 77 | RCU_FANOUT); |
7fa27001 | 78 | if (rcu_fanout_exact) |
ab6f5bd6 PM |
79 | pr_info("\tHierarchical RCU autobalancing is disabled.\n"); |
80 | if (IS_ENABLED(CONFIG_RCU_FAST_NO_HZ)) | |
81 | pr_info("\tRCU dyntick-idle grace-period acceleration is enabled.\n"); | |
c4a09ff7 | 82 | if (IS_ENABLED(CONFIG_PROVE_RCU)) |
ab6f5bd6 | 83 | pr_info("\tRCU lockdep checking is enabled.\n"); |
42621697 AG |
84 | if (RCU_NUM_LVLS >= 4) |
85 | pr_info("\tFour(or more)-level hierarchy is enabled.\n"); | |
47d631af | 86 | if (RCU_FANOUT_LEAF != 16) |
a3bd2c09 | 87 | pr_info("\tBuild-time adjustment of leaf fanout to %d.\n", |
47d631af PM |
88 | RCU_FANOUT_LEAF); |
89 | if (rcu_fanout_leaf != RCU_FANOUT_LEAF) | |
9a5739d7 | 90 | pr_info("\tBoot-time adjustment of leaf fanout to %d.\n", rcu_fanout_leaf); |
cca6f393 | 91 | if (nr_cpu_ids != NR_CPUS) |
efc151c3 | 92 | pr_info("\tRCU restricting CPUs from NR_CPUS=%d to nr_cpu_ids=%d.\n", NR_CPUS, nr_cpu_ids); |
17c7798b PM |
93 | #ifdef CONFIG_RCU_BOOST |
94 | pr_info("\tRCU priority boosting: priority %d delay %d ms.\n", kthread_prio, CONFIG_RCU_BOOST_DELAY); | |
95 | #endif | |
96 | if (blimit != DEFAULT_RCU_BLIMIT) | |
97 | pr_info("\tBoot-time adjustment of callback invocation limit to %ld.\n", blimit); | |
98 | if (qhimark != DEFAULT_RCU_QHIMARK) | |
99 | pr_info("\tBoot-time adjustment of callback high-water mark to %ld.\n", qhimark); | |
100 | if (qlowmark != DEFAULT_RCU_QLOMARK) | |
101 | pr_info("\tBoot-time adjustment of callback low-water mark to %ld.\n", qlowmark); | |
102 | if (jiffies_till_first_fqs != ULONG_MAX) | |
103 | pr_info("\tBoot-time adjustment of first FQS scan delay to %ld jiffies.\n", jiffies_till_first_fqs); | |
104 | if (jiffies_till_next_fqs != ULONG_MAX) | |
105 | pr_info("\tBoot-time adjustment of subsequent FQS scan delay to %ld jiffies.\n", jiffies_till_next_fqs); | |
106 | if (rcu_kick_kthreads) | |
107 | pr_info("\tKick kthreads if too-long grace period.\n"); | |
108 | if (IS_ENABLED(CONFIG_DEBUG_OBJECTS_RCU_HEAD)) | |
109 | pr_info("\tRCU callback double-/use-after-free debug enabled.\n"); | |
90040c9e | 110 | if (gp_preinit_delay) |
17c7798b | 111 | pr_info("\tRCU debug GP pre-init slowdown %d jiffies.\n", gp_preinit_delay); |
90040c9e | 112 | if (gp_init_delay) |
17c7798b | 113 | pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_init_delay); |
90040c9e | 114 | if (gp_cleanup_delay) |
17c7798b PM |
115 | pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_cleanup_delay); |
116 | if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG)) | |
117 | pr_info("\tRCU debug extended QS entry/exit.\n"); | |
59d80fd8 | 118 | rcupdate_announce_bootup_oddness(); |
26845c28 PM |
119 | } |
120 | ||
28f6569a | 121 | #ifdef CONFIG_PREEMPT_RCU |
f41d911f | 122 | |
a41bfeb2 | 123 | RCU_STATE_INITIALIZER(rcu_preempt, 'p', call_rcu); |
b28a7c01 | 124 | static struct rcu_state *const rcu_state_p = &rcu_preempt_state; |
2927a689 | 125 | static struct rcu_data __percpu *const rcu_data_p = &rcu_preempt_data; |
f41d911f | 126 | |
d19fb8d1 PM |
127 | static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp, |
128 | bool wake); | |
d9a3da06 | 129 | |
f41d911f PM |
130 | /* |
131 | * Tell them what RCU they are running. | |
132 | */ | |
0e0fc1c2 | 133 | static void __init rcu_bootup_announce(void) |
f41d911f | 134 | { |
efc151c3 | 135 | pr_info("Preemptible hierarchical RCU implementation.\n"); |
26845c28 | 136 | rcu_bootup_announce_oddness(); |
f41d911f PM |
137 | } |
138 | ||
8203d6d0 PM |
139 | /* Flags for rcu_preempt_ctxt_queue() decision table. */ |
140 | #define RCU_GP_TASKS 0x8 | |
141 | #define RCU_EXP_TASKS 0x4 | |
142 | #define RCU_GP_BLKD 0x2 | |
143 | #define RCU_EXP_BLKD 0x1 | |
144 | ||
145 | /* | |
146 | * Queues a task preempted within an RCU-preempt read-side critical | |
147 | * section into the appropriate location within the ->blkd_tasks list, | |
148 | * depending on the states of any ongoing normal and expedited grace | |
149 | * periods. The ->gp_tasks pointer indicates which element the normal | |
150 | * grace period is waiting on (NULL if none), and the ->exp_tasks pointer | |
151 | * indicates which element the expedited grace period is waiting on (again, | |
152 | * NULL if none). If a grace period is waiting on a given element in the | |
153 | * ->blkd_tasks list, it also waits on all subsequent elements. Thus, | |
154 | * adding a task to the tail of the list blocks any grace period that is | |
155 | * already waiting on one of the elements. In contrast, adding a task | |
156 | * to the head of the list won't block any grace period that is already | |
157 | * waiting on one of the elements. | |
158 | * | |
159 | * This queuing is imprecise, and can sometimes make an ongoing grace | |
160 | * period wait for a task that is not strictly speaking blocking it. | |
161 | * Given the choice, we needlessly block a normal grace period rather than | |
162 | * blocking an expedited grace period. | |
163 | * | |
164 | * Note that an endless sequence of expedited grace periods still cannot | |
165 | * indefinitely postpone a normal grace period. Eventually, all of the | |
166 | * fixed number of preempted tasks blocking the normal grace period that are | |
167 | * not also blocking the expedited grace period will resume and complete | |
168 | * their RCU read-side critical sections. At that point, the ->gp_tasks | |
169 | * pointer will equal the ->exp_tasks pointer, at which point the end of | |
170 | * the corresponding expedited grace period will also be the end of the | |
171 | * normal grace period. | |
172 | */ | |
46a5d164 PM |
173 | static void rcu_preempt_ctxt_queue(struct rcu_node *rnp, struct rcu_data *rdp) |
174 | __releases(rnp->lock) /* But leaves rrupts disabled. */ | |
8203d6d0 PM |
175 | { |
176 | int blkd_state = (rnp->gp_tasks ? RCU_GP_TASKS : 0) + | |
177 | (rnp->exp_tasks ? RCU_EXP_TASKS : 0) + | |
178 | (rnp->qsmask & rdp->grpmask ? RCU_GP_BLKD : 0) + | |
179 | (rnp->expmask & rdp->grpmask ? RCU_EXP_BLKD : 0); | |
180 | struct task_struct *t = current; | |
181 | ||
ea9b0c8a PM |
182 | lockdep_assert_held(&rnp->lock); |
183 | ||
8203d6d0 PM |
184 | /* |
185 | * Decide where to queue the newly blocked task. In theory, | |
186 | * this could be an if-statement. In practice, when I tried | |
187 | * that, it was quite messy. | |
188 | */ | |
189 | switch (blkd_state) { | |
190 | case 0: | |
191 | case RCU_EXP_TASKS: | |
192 | case RCU_EXP_TASKS + RCU_GP_BLKD: | |
193 | case RCU_GP_TASKS: | |
194 | case RCU_GP_TASKS + RCU_EXP_TASKS: | |
195 | ||
196 | /* | |
197 | * Blocking neither GP, or first task blocking the normal | |
198 | * GP but not blocking the already-waiting expedited GP. | |
199 | * Queue at the head of the list to avoid unnecessarily | |
200 | * blocking the already-waiting GPs. | |
201 | */ | |
202 | list_add(&t->rcu_node_entry, &rnp->blkd_tasks); | |
203 | break; | |
204 | ||
205 | case RCU_EXP_BLKD: | |
206 | case RCU_GP_BLKD: | |
207 | case RCU_GP_BLKD + RCU_EXP_BLKD: | |
208 | case RCU_GP_TASKS + RCU_EXP_BLKD: | |
209 | case RCU_GP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
210 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
211 | ||
212 | /* | |
213 | * First task arriving that blocks either GP, or first task | |
214 | * arriving that blocks the expedited GP (with the normal | |
215 | * GP already waiting), or a task arriving that blocks | |
216 | * both GPs with both GPs already waiting. Queue at the | |
217 | * tail of the list to avoid any GP waiting on any of the | |
218 | * already queued tasks that are not blocking it. | |
219 | */ | |
220 | list_add_tail(&t->rcu_node_entry, &rnp->blkd_tasks); | |
221 | break; | |
222 | ||
223 | case RCU_EXP_TASKS + RCU_EXP_BLKD: | |
224 | case RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
225 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_EXP_BLKD: | |
226 | ||
227 | /* | |
228 | * Second or subsequent task blocking the expedited GP. | |
229 | * The task either does not block the normal GP, or is the | |
230 | * first task blocking the normal GP. Queue just after | |
231 | * the first task blocking the expedited GP. | |
232 | */ | |
233 | list_add(&t->rcu_node_entry, rnp->exp_tasks); | |
234 | break; | |
235 | ||
236 | case RCU_GP_TASKS + RCU_GP_BLKD: | |
237 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD: | |
238 | ||
239 | /* | |
240 | * Second or subsequent task blocking the normal GP. | |
241 | * The task does not block the expedited GP. Queue just | |
242 | * after the first task blocking the normal GP. | |
243 | */ | |
244 | list_add(&t->rcu_node_entry, rnp->gp_tasks); | |
245 | break; | |
246 | ||
247 | default: | |
248 | ||
249 | /* Yet another exercise in excessive paranoia. */ | |
250 | WARN_ON_ONCE(1); | |
251 | break; | |
252 | } | |
253 | ||
254 | /* | |
255 | * We have now queued the task. If it was the first one to | |
256 | * block either grace period, update the ->gp_tasks and/or | |
257 | * ->exp_tasks pointers, respectively, to reference the newly | |
258 | * blocked tasks. | |
259 | */ | |
260 | if (!rnp->gp_tasks && (blkd_state & RCU_GP_BLKD)) | |
261 | rnp->gp_tasks = &t->rcu_node_entry; | |
262 | if (!rnp->exp_tasks && (blkd_state & RCU_EXP_BLKD)) | |
263 | rnp->exp_tasks = &t->rcu_node_entry; | |
67c583a7 | 264 | raw_spin_unlock_rcu_node(rnp); /* interrupts remain disabled. */ |
8203d6d0 PM |
265 | |
266 | /* | |
267 | * Report the quiescent state for the expedited GP. This expedited | |
268 | * GP should not be able to end until we report, so there should be | |
269 | * no need to check for a subsequent expedited GP. (Though we are | |
270 | * still in a quiescent state in any case.) | |
271 | */ | |
272 | if (blkd_state & RCU_EXP_BLKD && | |
273 | t->rcu_read_unlock_special.b.exp_need_qs) { | |
274 | t->rcu_read_unlock_special.b.exp_need_qs = false; | |
275 | rcu_report_exp_rdp(rdp->rsp, rdp, true); | |
276 | } else { | |
277 | WARN_ON_ONCE(t->rcu_read_unlock_special.b.exp_need_qs); | |
278 | } | |
8203d6d0 PM |
279 | } |
280 | ||
f41d911f | 281 | /* |
6cc68793 | 282 | * Record a preemptible-RCU quiescent state for the specified CPU. Note |
f41d911f PM |
283 | * that this just means that the task currently running on the CPU is |
284 | * not in a quiescent state. There might be any number of tasks blocked | |
285 | * while in an RCU read-side critical section. | |
25502a6c | 286 | * |
1d082fd0 PM |
287 | * As with the other rcu_*_qs() functions, callers to this function |
288 | * must disable preemption. | |
f41d911f | 289 | */ |
284a8c93 | 290 | static void rcu_preempt_qs(void) |
f41d911f | 291 | { |
ea9b0c8a | 292 | RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_qs() invoked with preemption enabled!!!\n"); |
5b74c458 | 293 | if (__this_cpu_read(rcu_data_p->cpu_no_qs.s)) { |
284a8c93 | 294 | trace_rcu_grace_period(TPS("rcu_preempt"), |
2927a689 | 295 | __this_cpu_read(rcu_data_p->gpnum), |
284a8c93 | 296 | TPS("cpuqs")); |
5b74c458 | 297 | __this_cpu_write(rcu_data_p->cpu_no_qs.b.norm, false); |
284a8c93 PM |
298 | barrier(); /* Coordinate with rcu_preempt_check_callbacks(). */ |
299 | current->rcu_read_unlock_special.b.need_qs = false; | |
300 | } | |
f41d911f PM |
301 | } |
302 | ||
303 | /* | |
c3422bea PM |
304 | * We have entered the scheduler, and the current task might soon be |
305 | * context-switched away from. If this task is in an RCU read-side | |
306 | * critical section, we will no longer be able to rely on the CPU to | |
12f5f524 PM |
307 | * record that fact, so we enqueue the task on the blkd_tasks list. |
308 | * The task will dequeue itself when it exits the outermost enclosing | |
309 | * RCU read-side critical section. Therefore, the current grace period | |
310 | * cannot be permitted to complete until the blkd_tasks list entries | |
311 | * predating the current grace period drain, in other words, until | |
312 | * rnp->gp_tasks becomes NULL. | |
c3422bea | 313 | * |
46a5d164 | 314 | * Caller must disable interrupts. |
f41d911f | 315 | */ |
5b72f964 | 316 | static void rcu_preempt_note_context_switch(bool preempt) |
f41d911f PM |
317 | { |
318 | struct task_struct *t = current; | |
f41d911f PM |
319 | struct rcu_data *rdp; |
320 | struct rcu_node *rnp; | |
321 | ||
ea9b0c8a | 322 | RCU_LOCKDEP_WARN(!irqs_disabled(), "rcu_preempt_note_context_switch() invoked with interrupts enabled!!!\n"); |
5b72f964 | 323 | WARN_ON_ONCE(!preempt && t->rcu_read_lock_nesting > 0); |
10f39bb1 | 324 | if (t->rcu_read_lock_nesting > 0 && |
1d082fd0 | 325 | !t->rcu_read_unlock_special.b.blocked) { |
f41d911f PM |
326 | |
327 | /* Possibly blocking in an RCU read-side critical section. */ | |
e63c887c | 328 | rdp = this_cpu_ptr(rcu_state_p->rda); |
f41d911f | 329 | rnp = rdp->mynode; |
46a5d164 | 330 | raw_spin_lock_rcu_node(rnp); |
1d082fd0 | 331 | t->rcu_read_unlock_special.b.blocked = true; |
86848966 | 332 | t->rcu_blocked_node = rnp; |
f41d911f PM |
333 | |
334 | /* | |
8203d6d0 PM |
335 | * Verify the CPU's sanity, trace the preemption, and |
336 | * then queue the task as required based on the states | |
337 | * of any ongoing and expedited grace periods. | |
f41d911f | 338 | */ |
0aa04b05 | 339 | WARN_ON_ONCE((rdp->grpmask & rcu_rnp_online_cpus(rnp)) == 0); |
e7d8842e | 340 | WARN_ON_ONCE(!list_empty(&t->rcu_node_entry)); |
d4c08f2a PM |
341 | trace_rcu_preempt_task(rdp->rsp->name, |
342 | t->pid, | |
343 | (rnp->qsmask & rdp->grpmask) | |
344 | ? rnp->gpnum | |
345 | : rnp->gpnum + 1); | |
46a5d164 | 346 | rcu_preempt_ctxt_queue(rnp, rdp); |
10f39bb1 | 347 | } else if (t->rcu_read_lock_nesting < 0 && |
1d082fd0 | 348 | t->rcu_read_unlock_special.s) { |
10f39bb1 PM |
349 | |
350 | /* | |
351 | * Complete exit from RCU read-side critical section on | |
352 | * behalf of preempted instance of __rcu_read_unlock(). | |
353 | */ | |
354 | rcu_read_unlock_special(t); | |
f41d911f PM |
355 | } |
356 | ||
357 | /* | |
358 | * Either we were not in an RCU read-side critical section to | |
359 | * begin with, or we have now recorded that critical section | |
360 | * globally. Either way, we can now note a quiescent state | |
361 | * for this CPU. Again, if we were in an RCU read-side critical | |
362 | * section, and if that critical section was blocking the current | |
363 | * grace period, then the fact that the task has been enqueued | |
364 | * means that we continue to block the current grace period. | |
365 | */ | |
284a8c93 | 366 | rcu_preempt_qs(); |
f41d911f PM |
367 | } |
368 | ||
fc2219d4 PM |
369 | /* |
370 | * Check for preempted RCU readers blocking the current grace period | |
371 | * for the specified rcu_node structure. If the caller needs a reliable | |
372 | * answer, it must hold the rcu_node's ->lock. | |
373 | */ | |
27f4d280 | 374 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 | 375 | { |
12f5f524 | 376 | return rnp->gp_tasks != NULL; |
fc2219d4 PM |
377 | } |
378 | ||
12f5f524 PM |
379 | /* |
380 | * Advance a ->blkd_tasks-list pointer to the next entry, instead | |
381 | * returning NULL if at the end of the list. | |
382 | */ | |
383 | static struct list_head *rcu_next_node_entry(struct task_struct *t, | |
384 | struct rcu_node *rnp) | |
385 | { | |
386 | struct list_head *np; | |
387 | ||
388 | np = t->rcu_node_entry.next; | |
389 | if (np == &rnp->blkd_tasks) | |
390 | np = NULL; | |
391 | return np; | |
392 | } | |
393 | ||
8af3a5e7 PM |
394 | /* |
395 | * Return true if the specified rcu_node structure has tasks that were | |
396 | * preempted within an RCU read-side critical section. | |
397 | */ | |
398 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
399 | { | |
400 | return !list_empty(&rnp->blkd_tasks); | |
401 | } | |
402 | ||
b668c9cf PM |
403 | /* |
404 | * Handle special cases during rcu_read_unlock(), such as needing to | |
405 | * notify RCU core processing or task having blocked during the RCU | |
406 | * read-side critical section. | |
407 | */ | |
2a3fa843 | 408 | void rcu_read_unlock_special(struct task_struct *t) |
f41d911f | 409 | { |
b6a932d1 PM |
410 | bool empty_exp; |
411 | bool empty_norm; | |
412 | bool empty_exp_now; | |
f41d911f | 413 | unsigned long flags; |
12f5f524 | 414 | struct list_head *np; |
abaa93d9 | 415 | bool drop_boost_mutex = false; |
8203d6d0 | 416 | struct rcu_data *rdp; |
f41d911f | 417 | struct rcu_node *rnp; |
1d082fd0 | 418 | union rcu_special special; |
f41d911f PM |
419 | |
420 | /* NMI handlers cannot block and cannot safely manipulate state. */ | |
421 | if (in_nmi()) | |
422 | return; | |
423 | ||
424 | local_irq_save(flags); | |
425 | ||
426 | /* | |
8203d6d0 PM |
427 | * If RCU core is waiting for this CPU to exit its critical section, |
428 | * report the fact that it has exited. Because irqs are disabled, | |
1d082fd0 | 429 | * t->rcu_read_unlock_special cannot change. |
f41d911f PM |
430 | */ |
431 | special = t->rcu_read_unlock_special; | |
1d082fd0 | 432 | if (special.b.need_qs) { |
284a8c93 | 433 | rcu_preempt_qs(); |
c0135d07 | 434 | t->rcu_read_unlock_special.b.need_qs = false; |
1d082fd0 | 435 | if (!t->rcu_read_unlock_special.s) { |
79a62f95 LJ |
436 | local_irq_restore(flags); |
437 | return; | |
438 | } | |
f41d911f PM |
439 | } |
440 | ||
8203d6d0 PM |
441 | /* |
442 | * Respond to a request for an expedited grace period, but only if | |
443 | * we were not preempted, meaning that we were running on the same | |
444 | * CPU throughout. If we were preempted, the exp_need_qs flag | |
445 | * would have been cleared at the time of the first preemption, | |
446 | * and the quiescent state would be reported when we were dequeued. | |
447 | */ | |
448 | if (special.b.exp_need_qs) { | |
449 | WARN_ON_ONCE(special.b.blocked); | |
450 | t->rcu_read_unlock_special.b.exp_need_qs = false; | |
451 | rdp = this_cpu_ptr(rcu_state_p->rda); | |
452 | rcu_report_exp_rdp(rcu_state_p, rdp, true); | |
453 | if (!t->rcu_read_unlock_special.s) { | |
454 | local_irq_restore(flags); | |
455 | return; | |
456 | } | |
457 | } | |
458 | ||
79a62f95 | 459 | /* Hardware IRQ handlers cannot block, complain if they get here. */ |
d24209bb PM |
460 | if (in_irq() || in_serving_softirq()) { |
461 | lockdep_rcu_suspicious(__FILE__, __LINE__, | |
462 | "rcu_read_unlock() from irq or softirq with blocking in critical section!!!\n"); | |
8203d6d0 | 463 | pr_alert("->rcu_read_unlock_special: %#x (b: %d, enq: %d nq: %d)\n", |
d24209bb PM |
464 | t->rcu_read_unlock_special.s, |
465 | t->rcu_read_unlock_special.b.blocked, | |
8203d6d0 | 466 | t->rcu_read_unlock_special.b.exp_need_qs, |
d24209bb | 467 | t->rcu_read_unlock_special.b.need_qs); |
f41d911f PM |
468 | local_irq_restore(flags); |
469 | return; | |
470 | } | |
471 | ||
472 | /* Clean up if blocked during RCU read-side critical section. */ | |
1d082fd0 PM |
473 | if (special.b.blocked) { |
474 | t->rcu_read_unlock_special.b.blocked = false; | |
f41d911f | 475 | |
dd5d19ba | 476 | /* |
0a0ba1c9 | 477 | * Remove this task from the list it blocked on. The task |
8ba9153b PM |
478 | * now remains queued on the rcu_node corresponding to the |
479 | * CPU it first blocked on, so there is no longer any need | |
480 | * to loop. Retain a WARN_ON_ONCE() out of sheer paranoia. | |
dd5d19ba | 481 | */ |
8ba9153b PM |
482 | rnp = t->rcu_blocked_node; |
483 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
484 | WARN_ON_ONCE(rnp != t->rcu_blocked_node); | |
74e871ac | 485 | empty_norm = !rcu_preempt_blocked_readers_cgp(rnp); |
8203d6d0 | 486 | empty_exp = sync_rcu_preempt_exp_done(rnp); |
d9a3da06 | 487 | smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */ |
12f5f524 | 488 | np = rcu_next_node_entry(t, rnp); |
f41d911f | 489 | list_del_init(&t->rcu_node_entry); |
82e78d80 | 490 | t->rcu_blocked_node = NULL; |
f7f7bac9 | 491 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt"), |
d4c08f2a | 492 | rnp->gpnum, t->pid); |
12f5f524 PM |
493 | if (&t->rcu_node_entry == rnp->gp_tasks) |
494 | rnp->gp_tasks = np; | |
495 | if (&t->rcu_node_entry == rnp->exp_tasks) | |
496 | rnp->exp_tasks = np; | |
727b705b PM |
497 | if (IS_ENABLED(CONFIG_RCU_BOOST)) { |
498 | if (&t->rcu_node_entry == rnp->boost_tasks) | |
499 | rnp->boost_tasks = np; | |
500 | /* Snapshot ->boost_mtx ownership w/rnp->lock held. */ | |
501 | drop_boost_mutex = rt_mutex_owner(&rnp->boost_mtx) == t; | |
502 | } | |
f41d911f PM |
503 | |
504 | /* | |
505 | * If this was the last task on the current list, and if | |
506 | * we aren't waiting on any CPUs, report the quiescent state. | |
389abd48 PM |
507 | * Note that rcu_report_unblock_qs_rnp() releases rnp->lock, |
508 | * so we must take a snapshot of the expedited state. | |
f41d911f | 509 | */ |
8203d6d0 | 510 | empty_exp_now = sync_rcu_preempt_exp_done(rnp); |
74e871ac | 511 | if (!empty_norm && !rcu_preempt_blocked_readers_cgp(rnp)) { |
f7f7bac9 | 512 | trace_rcu_quiescent_state_report(TPS("preempt_rcu"), |
d4c08f2a PM |
513 | rnp->gpnum, |
514 | 0, rnp->qsmask, | |
515 | rnp->level, | |
516 | rnp->grplo, | |
517 | rnp->grphi, | |
518 | !!rnp->gp_tasks); | |
e63c887c | 519 | rcu_report_unblock_qs_rnp(rcu_state_p, rnp, flags); |
c701d5d9 | 520 | } else { |
67c583a7 | 521 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
c701d5d9 | 522 | } |
d9a3da06 | 523 | |
27f4d280 | 524 | /* Unboost if we were boosted. */ |
727b705b | 525 | if (IS_ENABLED(CONFIG_RCU_BOOST) && drop_boost_mutex) |
abaa93d9 | 526 | rt_mutex_unlock(&rnp->boost_mtx); |
27f4d280 | 527 | |
d9a3da06 PM |
528 | /* |
529 | * If this was the last task on the expedited lists, | |
530 | * then we need to report up the rcu_node hierarchy. | |
531 | */ | |
389abd48 | 532 | if (!empty_exp && empty_exp_now) |
e63c887c | 533 | rcu_report_exp_rnp(rcu_state_p, rnp, true); |
b668c9cf PM |
534 | } else { |
535 | local_irq_restore(flags); | |
f41d911f | 536 | } |
f41d911f PM |
537 | } |
538 | ||
1ed509a2 PM |
539 | /* |
540 | * Dump detailed information for all tasks blocking the current RCU | |
541 | * grace period on the specified rcu_node structure. | |
542 | */ | |
543 | static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp) | |
544 | { | |
545 | unsigned long flags; | |
1ed509a2 PM |
546 | struct task_struct *t; |
547 | ||
6cf10081 | 548 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
5fd4dc06 | 549 | if (!rcu_preempt_blocked_readers_cgp(rnp)) { |
67c583a7 | 550 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
5fd4dc06 PM |
551 | return; |
552 | } | |
82efed06 | 553 | t = list_entry(rnp->gp_tasks->prev, |
12f5f524 PM |
554 | struct task_struct, rcu_node_entry); |
555 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) | |
556 | sched_show_task(t); | |
67c583a7 | 557 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
1ed509a2 PM |
558 | } |
559 | ||
560 | /* | |
561 | * Dump detailed information for all tasks blocking the current RCU | |
562 | * grace period. | |
563 | */ | |
564 | static void rcu_print_detail_task_stall(struct rcu_state *rsp) | |
565 | { | |
566 | struct rcu_node *rnp = rcu_get_root(rsp); | |
567 | ||
568 | rcu_print_detail_task_stall_rnp(rnp); | |
569 | rcu_for_each_leaf_node(rsp, rnp) | |
570 | rcu_print_detail_task_stall_rnp(rnp); | |
571 | } | |
572 | ||
a858af28 PM |
573 | static void rcu_print_task_stall_begin(struct rcu_node *rnp) |
574 | { | |
efc151c3 | 575 | pr_err("\tTasks blocked on level-%d rcu_node (CPUs %d-%d):", |
a858af28 PM |
576 | rnp->level, rnp->grplo, rnp->grphi); |
577 | } | |
578 | ||
579 | static void rcu_print_task_stall_end(void) | |
580 | { | |
efc151c3 | 581 | pr_cont("\n"); |
a858af28 PM |
582 | } |
583 | ||
f41d911f PM |
584 | /* |
585 | * Scan the current list of tasks blocked within RCU read-side critical | |
586 | * sections, printing out the tid of each. | |
587 | */ | |
9bc8b558 | 588 | static int rcu_print_task_stall(struct rcu_node *rnp) |
f41d911f | 589 | { |
f41d911f | 590 | struct task_struct *t; |
9bc8b558 | 591 | int ndetected = 0; |
f41d911f | 592 | |
27f4d280 | 593 | if (!rcu_preempt_blocked_readers_cgp(rnp)) |
9bc8b558 | 594 | return 0; |
a858af28 | 595 | rcu_print_task_stall_begin(rnp); |
82efed06 | 596 | t = list_entry(rnp->gp_tasks->prev, |
12f5f524 | 597 | struct task_struct, rcu_node_entry); |
9bc8b558 | 598 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) { |
efc151c3 | 599 | pr_cont(" P%d", t->pid); |
9bc8b558 PM |
600 | ndetected++; |
601 | } | |
a858af28 | 602 | rcu_print_task_stall_end(); |
9bc8b558 | 603 | return ndetected; |
f41d911f PM |
604 | } |
605 | ||
74611ecb PM |
606 | /* |
607 | * Scan the current list of tasks blocked within RCU read-side critical | |
608 | * sections, printing out the tid of each that is blocking the current | |
609 | * expedited grace period. | |
610 | */ | |
611 | static int rcu_print_task_exp_stall(struct rcu_node *rnp) | |
612 | { | |
613 | struct task_struct *t; | |
614 | int ndetected = 0; | |
615 | ||
616 | if (!rnp->exp_tasks) | |
617 | return 0; | |
618 | t = list_entry(rnp->exp_tasks->prev, | |
619 | struct task_struct, rcu_node_entry); | |
620 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) { | |
621 | pr_cont(" P%d", t->pid); | |
622 | ndetected++; | |
623 | } | |
624 | return ndetected; | |
625 | } | |
626 | ||
b0e165c0 PM |
627 | /* |
628 | * Check that the list of blocked tasks for the newly completed grace | |
629 | * period is in fact empty. It is a serious bug to complete a grace | |
630 | * period that still has RCU readers blocked! This function must be | |
631 | * invoked -before- updating this rnp's ->gpnum, and the rnp's ->lock | |
632 | * must be held by the caller. | |
12f5f524 PM |
633 | * |
634 | * Also, if there are blocked tasks on the list, they automatically | |
635 | * block the newly created grace period, so set up ->gp_tasks accordingly. | |
b0e165c0 PM |
636 | */ |
637 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) | |
638 | { | |
ea9b0c8a | 639 | RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_check_blocked_tasks() invoked with preemption enabled!!!\n"); |
27f4d280 | 640 | WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)); |
96e92021 | 641 | if (rcu_preempt_has_tasks(rnp)) |
12f5f524 | 642 | rnp->gp_tasks = rnp->blkd_tasks.next; |
28ecd580 | 643 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
644 | } |
645 | ||
f41d911f PM |
646 | /* |
647 | * Check for a quiescent state from the current CPU. When a task blocks, | |
648 | * the task is recorded in the corresponding CPU's rcu_node structure, | |
649 | * which is checked elsewhere. | |
650 | * | |
651 | * Caller must disable hard irqs. | |
652 | */ | |
86aea0e6 | 653 | static void rcu_preempt_check_callbacks(void) |
f41d911f PM |
654 | { |
655 | struct task_struct *t = current; | |
656 | ||
657 | if (t->rcu_read_lock_nesting == 0) { | |
284a8c93 | 658 | rcu_preempt_qs(); |
f41d911f PM |
659 | return; |
660 | } | |
10f39bb1 | 661 | if (t->rcu_read_lock_nesting > 0 && |
97c668b8 | 662 | __this_cpu_read(rcu_data_p->core_needs_qs) && |
5b74c458 | 663 | __this_cpu_read(rcu_data_p->cpu_no_qs.b.norm)) |
1d082fd0 | 664 | t->rcu_read_unlock_special.b.need_qs = true; |
f41d911f PM |
665 | } |
666 | ||
a46e0899 PM |
667 | #ifdef CONFIG_RCU_BOOST |
668 | ||
09223371 SL |
669 | static void rcu_preempt_do_callbacks(void) |
670 | { | |
2927a689 | 671 | rcu_do_batch(rcu_state_p, this_cpu_ptr(rcu_data_p)); |
09223371 SL |
672 | } |
673 | ||
a46e0899 PM |
674 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
675 | ||
a68a2bb2 PM |
676 | /** |
677 | * call_rcu() - Queue an RCU callback for invocation after a grace period. | |
678 | * @head: structure to be used for queueing the RCU updates. | |
679 | * @func: actual callback function to be invoked after the grace period | |
680 | * | |
681 | * The callback function will be invoked some time after a full grace | |
682 | * period elapses, in other words after all pre-existing RCU read-side | |
683 | * critical sections have completed. However, the callback function | |
684 | * might well execute concurrently with RCU read-side critical sections | |
685 | * that started after call_rcu() was invoked. RCU read-side critical | |
686 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), | |
687 | * and may be nested. | |
688 | * | |
689 | * Note that all CPUs must agree that the grace period extended beyond | |
690 | * all pre-existing RCU read-side critical section. On systems with more | |
691 | * than one CPU, this means that when "func()" is invoked, each CPU is | |
692 | * guaranteed to have executed a full memory barrier since the end of its | |
693 | * last RCU read-side critical section whose beginning preceded the call | |
694 | * to call_rcu(). It also means that each CPU executing an RCU read-side | |
695 | * critical section that continues beyond the start of "func()" must have | |
696 | * executed a memory barrier after the call_rcu() but before the beginning | |
697 | * of that RCU read-side critical section. Note that these guarantees | |
698 | * include CPUs that are offline, idle, or executing in user mode, as | |
699 | * well as CPUs that are executing in the kernel. | |
700 | * | |
701 | * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the | |
702 | * resulting RCU callback function "func()", then both CPU A and CPU B are | |
703 | * guaranteed to execute a full memory barrier during the time interval | |
704 | * between the call to call_rcu() and the invocation of "func()" -- even | |
705 | * if CPU A and CPU B are the same CPU (but again only if the system has | |
706 | * more than one CPU). | |
f41d911f | 707 | */ |
b6a4ae76 | 708 | void call_rcu(struct rcu_head *head, rcu_callback_t func) |
f41d911f | 709 | { |
e63c887c | 710 | __call_rcu(head, func, rcu_state_p, -1, 0); |
f41d911f PM |
711 | } |
712 | EXPORT_SYMBOL_GPL(call_rcu); | |
713 | ||
6ebb237b PM |
714 | /** |
715 | * synchronize_rcu - wait until a grace period has elapsed. | |
716 | * | |
717 | * Control will return to the caller some time after a full grace | |
718 | * period has elapsed, in other words after all currently executing RCU | |
77d8485a PM |
719 | * read-side critical sections have completed. Note, however, that |
720 | * upon return from synchronize_rcu(), the caller might well be executing | |
721 | * concurrently with new RCU read-side critical sections that began while | |
722 | * synchronize_rcu() was waiting. RCU read-side critical sections are | |
723 | * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested. | |
f0a0e6f2 | 724 | * |
e28371c8 PM |
725 | * See the description of synchronize_sched() for more detailed |
726 | * information on memory-ordering guarantees. However, please note | |
727 | * that -only- the memory-ordering guarantees apply. For example, | |
728 | * synchronize_rcu() is -not- guaranteed to wait on things like code | |
729 | * protected by preempt_disable(), instead, synchronize_rcu() is -only- | |
730 | * guaranteed to wait on RCU read-side critical sections, that is, sections | |
731 | * of code protected by rcu_read_lock(). | |
6ebb237b PM |
732 | */ |
733 | void synchronize_rcu(void) | |
734 | { | |
f78f5b90 PM |
735 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) || |
736 | lock_is_held(&rcu_lock_map) || | |
737 | lock_is_held(&rcu_sched_lock_map), | |
738 | "Illegal synchronize_rcu() in RCU read-side critical section"); | |
52d7e48b | 739 | if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE) |
6ebb237b | 740 | return; |
5afff48b | 741 | if (rcu_gp_is_expedited()) |
3705b88d AM |
742 | synchronize_rcu_expedited(); |
743 | else | |
744 | wait_rcu_gp(call_rcu); | |
6ebb237b PM |
745 | } |
746 | EXPORT_SYMBOL_GPL(synchronize_rcu); | |
747 | ||
e74f4c45 PM |
748 | /** |
749 | * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete. | |
f0a0e6f2 PM |
750 | * |
751 | * Note that this primitive does not necessarily wait for an RCU grace period | |
752 | * to complete. For example, if there are no RCU callbacks queued anywhere | |
753 | * in the system, then rcu_barrier() is within its rights to return | |
754 | * immediately, without waiting for anything, much less an RCU grace period. | |
e74f4c45 PM |
755 | */ |
756 | void rcu_barrier(void) | |
757 | { | |
e63c887c | 758 | _rcu_barrier(rcu_state_p); |
e74f4c45 PM |
759 | } |
760 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
761 | ||
1eba8f84 | 762 | /* |
6cc68793 | 763 | * Initialize preemptible RCU's state structures. |
1eba8f84 PM |
764 | */ |
765 | static void __init __rcu_init_preempt(void) | |
766 | { | |
a87f203e | 767 | rcu_init_one(rcu_state_p); |
1eba8f84 PM |
768 | } |
769 | ||
2439b696 PM |
770 | /* |
771 | * Check for a task exiting while in a preemptible-RCU read-side | |
772 | * critical section, clean up if so. No need to issue warnings, | |
773 | * as debug_check_no_locks_held() already does this if lockdep | |
774 | * is enabled. | |
775 | */ | |
776 | void exit_rcu(void) | |
777 | { | |
778 | struct task_struct *t = current; | |
779 | ||
780 | if (likely(list_empty(¤t->rcu_node_entry))) | |
781 | return; | |
782 | t->rcu_read_lock_nesting = 1; | |
783 | barrier(); | |
1d082fd0 | 784 | t->rcu_read_unlock_special.b.blocked = true; |
2439b696 PM |
785 | __rcu_read_unlock(); |
786 | } | |
787 | ||
28f6569a | 788 | #else /* #ifdef CONFIG_PREEMPT_RCU */ |
f41d911f | 789 | |
b28a7c01 | 790 | static struct rcu_state *const rcu_state_p = &rcu_sched_state; |
27f4d280 | 791 | |
f41d911f PM |
792 | /* |
793 | * Tell them what RCU they are running. | |
794 | */ | |
0e0fc1c2 | 795 | static void __init rcu_bootup_announce(void) |
f41d911f | 796 | { |
efc151c3 | 797 | pr_info("Hierarchical RCU implementation.\n"); |
26845c28 | 798 | rcu_bootup_announce_oddness(); |
f41d911f PM |
799 | } |
800 | ||
cba6d0d6 PM |
801 | /* |
802 | * Because preemptible RCU does not exist, we never have to check for | |
803 | * CPUs being in quiescent states. | |
804 | */ | |
5b72f964 | 805 | static void rcu_preempt_note_context_switch(bool preempt) |
cba6d0d6 PM |
806 | { |
807 | } | |
808 | ||
fc2219d4 | 809 | /* |
6cc68793 | 810 | * Because preemptible RCU does not exist, there are never any preempted |
fc2219d4 PM |
811 | * RCU readers. |
812 | */ | |
27f4d280 | 813 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 PM |
814 | { |
815 | return 0; | |
816 | } | |
817 | ||
8af3a5e7 PM |
818 | /* |
819 | * Because there is no preemptible RCU, there can be no readers blocked. | |
820 | */ | |
821 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
b668c9cf | 822 | { |
8af3a5e7 | 823 | return false; |
b668c9cf PM |
824 | } |
825 | ||
1ed509a2 | 826 | /* |
6cc68793 | 827 | * Because preemptible RCU does not exist, we never have to check for |
1ed509a2 PM |
828 | * tasks blocked within RCU read-side critical sections. |
829 | */ | |
830 | static void rcu_print_detail_task_stall(struct rcu_state *rsp) | |
831 | { | |
832 | } | |
833 | ||
f41d911f | 834 | /* |
6cc68793 | 835 | * Because preemptible RCU does not exist, we never have to check for |
f41d911f PM |
836 | * tasks blocked within RCU read-side critical sections. |
837 | */ | |
9bc8b558 | 838 | static int rcu_print_task_stall(struct rcu_node *rnp) |
f41d911f | 839 | { |
9bc8b558 | 840 | return 0; |
f41d911f PM |
841 | } |
842 | ||
74611ecb PM |
843 | /* |
844 | * Because preemptible RCU does not exist, we never have to check for | |
845 | * tasks blocked within RCU read-side critical sections that are | |
846 | * blocking the current expedited grace period. | |
847 | */ | |
848 | static int rcu_print_task_exp_stall(struct rcu_node *rnp) | |
849 | { | |
850 | return 0; | |
851 | } | |
852 | ||
b0e165c0 | 853 | /* |
6cc68793 | 854 | * Because there is no preemptible RCU, there can be no readers blocked, |
49e29126 PM |
855 | * so there is no need to check for blocked tasks. So check only for |
856 | * bogus qsmask values. | |
b0e165c0 PM |
857 | */ |
858 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) | |
859 | { | |
49e29126 | 860 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
861 | } |
862 | ||
f41d911f | 863 | /* |
6cc68793 | 864 | * Because preemptible RCU does not exist, it never has any callbacks |
f41d911f PM |
865 | * to check. |
866 | */ | |
86aea0e6 | 867 | static void rcu_preempt_check_callbacks(void) |
f41d911f PM |
868 | { |
869 | } | |
870 | ||
e74f4c45 | 871 | /* |
6cc68793 | 872 | * Because preemptible RCU does not exist, rcu_barrier() is just |
e74f4c45 PM |
873 | * another name for rcu_barrier_sched(). |
874 | */ | |
875 | void rcu_barrier(void) | |
876 | { | |
877 | rcu_barrier_sched(); | |
878 | } | |
879 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
880 | ||
1eba8f84 | 881 | /* |
6cc68793 | 882 | * Because preemptible RCU does not exist, it need not be initialized. |
1eba8f84 PM |
883 | */ |
884 | static void __init __rcu_init_preempt(void) | |
885 | { | |
886 | } | |
887 | ||
2439b696 PM |
888 | /* |
889 | * Because preemptible RCU does not exist, tasks cannot possibly exit | |
890 | * while in preemptible RCU read-side critical sections. | |
891 | */ | |
892 | void exit_rcu(void) | |
893 | { | |
894 | } | |
895 | ||
28f6569a | 896 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ |
8bd93a2c | 897 | |
27f4d280 PM |
898 | #ifdef CONFIG_RCU_BOOST |
899 | ||
1696a8be | 900 | #include "../locking/rtmutex_common.h" |
27f4d280 | 901 | |
0ea1f2eb PM |
902 | #ifdef CONFIG_RCU_TRACE |
903 | ||
904 | static void rcu_initiate_boost_trace(struct rcu_node *rnp) | |
905 | { | |
96e92021 | 906 | if (!rcu_preempt_has_tasks(rnp)) |
0ea1f2eb PM |
907 | rnp->n_balk_blkd_tasks++; |
908 | else if (rnp->exp_tasks == NULL && rnp->gp_tasks == NULL) | |
909 | rnp->n_balk_exp_gp_tasks++; | |
910 | else if (rnp->gp_tasks != NULL && rnp->boost_tasks != NULL) | |
911 | rnp->n_balk_boost_tasks++; | |
912 | else if (rnp->gp_tasks != NULL && rnp->qsmask != 0) | |
913 | rnp->n_balk_notblocked++; | |
914 | else if (rnp->gp_tasks != NULL && | |
a9f4793d | 915 | ULONG_CMP_LT(jiffies, rnp->boost_time)) |
0ea1f2eb PM |
916 | rnp->n_balk_notyet++; |
917 | else | |
918 | rnp->n_balk_nos++; | |
919 | } | |
920 | ||
921 | #else /* #ifdef CONFIG_RCU_TRACE */ | |
922 | ||
923 | static void rcu_initiate_boost_trace(struct rcu_node *rnp) | |
924 | { | |
925 | } | |
926 | ||
927 | #endif /* #else #ifdef CONFIG_RCU_TRACE */ | |
928 | ||
5d01bbd1 TG |
929 | static void rcu_wake_cond(struct task_struct *t, int status) |
930 | { | |
931 | /* | |
932 | * If the thread is yielding, only wake it when this | |
933 | * is invoked from idle | |
934 | */ | |
935 | if (status != RCU_KTHREAD_YIELDING || is_idle_task(current)) | |
936 | wake_up_process(t); | |
937 | } | |
938 | ||
27f4d280 PM |
939 | /* |
940 | * Carry out RCU priority boosting on the task indicated by ->exp_tasks | |
941 | * or ->boost_tasks, advancing the pointer to the next task in the | |
942 | * ->blkd_tasks list. | |
943 | * | |
944 | * Note that irqs must be enabled: boosting the task can block. | |
945 | * Returns 1 if there are more tasks needing to be boosted. | |
946 | */ | |
947 | static int rcu_boost(struct rcu_node *rnp) | |
948 | { | |
949 | unsigned long flags; | |
27f4d280 PM |
950 | struct task_struct *t; |
951 | struct list_head *tb; | |
952 | ||
7d0ae808 PM |
953 | if (READ_ONCE(rnp->exp_tasks) == NULL && |
954 | READ_ONCE(rnp->boost_tasks) == NULL) | |
27f4d280 PM |
955 | return 0; /* Nothing left to boost. */ |
956 | ||
2a67e741 | 957 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 PM |
958 | |
959 | /* | |
960 | * Recheck under the lock: all tasks in need of boosting | |
961 | * might exit their RCU read-side critical sections on their own. | |
962 | */ | |
963 | if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) { | |
67c583a7 | 964 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
965 | return 0; |
966 | } | |
967 | ||
968 | /* | |
969 | * Preferentially boost tasks blocking expedited grace periods. | |
970 | * This cannot starve the normal grace periods because a second | |
971 | * expedited grace period must boost all blocked tasks, including | |
972 | * those blocking the pre-existing normal grace period. | |
973 | */ | |
0ea1f2eb | 974 | if (rnp->exp_tasks != NULL) { |
27f4d280 | 975 | tb = rnp->exp_tasks; |
0ea1f2eb PM |
976 | rnp->n_exp_boosts++; |
977 | } else { | |
27f4d280 | 978 | tb = rnp->boost_tasks; |
0ea1f2eb PM |
979 | rnp->n_normal_boosts++; |
980 | } | |
981 | rnp->n_tasks_boosted++; | |
27f4d280 PM |
982 | |
983 | /* | |
984 | * We boost task t by manufacturing an rt_mutex that appears to | |
985 | * be held by task t. We leave a pointer to that rt_mutex where | |
986 | * task t can find it, and task t will release the mutex when it | |
987 | * exits its outermost RCU read-side critical section. Then | |
988 | * simply acquiring this artificial rt_mutex will boost task | |
989 | * t's priority. (Thanks to tglx for suggesting this approach!) | |
990 | * | |
991 | * Note that task t must acquire rnp->lock to remove itself from | |
992 | * the ->blkd_tasks list, which it will do from exit() if from | |
993 | * nowhere else. We therefore are guaranteed that task t will | |
994 | * stay around at least until we drop rnp->lock. Note that | |
995 | * rnp->lock also resolves races between our priority boosting | |
996 | * and task t's exiting its outermost RCU read-side critical | |
997 | * section. | |
998 | */ | |
999 | t = container_of(tb, struct task_struct, rcu_node_entry); | |
abaa93d9 | 1000 | rt_mutex_init_proxy_locked(&rnp->boost_mtx, t); |
67c583a7 | 1001 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
abaa93d9 PM |
1002 | /* Lock only for side effect: boosts task t's priority. */ |
1003 | rt_mutex_lock(&rnp->boost_mtx); | |
1004 | rt_mutex_unlock(&rnp->boost_mtx); /* Then keep lockdep happy. */ | |
27f4d280 | 1005 | |
7d0ae808 PM |
1006 | return READ_ONCE(rnp->exp_tasks) != NULL || |
1007 | READ_ONCE(rnp->boost_tasks) != NULL; | |
27f4d280 PM |
1008 | } |
1009 | ||
27f4d280 | 1010 | /* |
bc17ea10 | 1011 | * Priority-boosting kthread, one per leaf rcu_node. |
27f4d280 PM |
1012 | */ |
1013 | static int rcu_boost_kthread(void *arg) | |
1014 | { | |
1015 | struct rcu_node *rnp = (struct rcu_node *)arg; | |
1016 | int spincnt = 0; | |
1017 | int more2boost; | |
1018 | ||
f7f7bac9 | 1019 | trace_rcu_utilization(TPS("Start boost kthread@init")); |
27f4d280 | 1020 | for (;;) { |
d71df90e | 1021 | rnp->boost_kthread_status = RCU_KTHREAD_WAITING; |
f7f7bac9 | 1022 | trace_rcu_utilization(TPS("End boost kthread@rcu_wait")); |
08bca60a | 1023 | rcu_wait(rnp->boost_tasks || rnp->exp_tasks); |
f7f7bac9 | 1024 | trace_rcu_utilization(TPS("Start boost kthread@rcu_wait")); |
d71df90e | 1025 | rnp->boost_kthread_status = RCU_KTHREAD_RUNNING; |
27f4d280 PM |
1026 | more2boost = rcu_boost(rnp); |
1027 | if (more2boost) | |
1028 | spincnt++; | |
1029 | else | |
1030 | spincnt = 0; | |
1031 | if (spincnt > 10) { | |
5d01bbd1 | 1032 | rnp->boost_kthread_status = RCU_KTHREAD_YIELDING; |
f7f7bac9 | 1033 | trace_rcu_utilization(TPS("End boost kthread@rcu_yield")); |
5d01bbd1 | 1034 | schedule_timeout_interruptible(2); |
f7f7bac9 | 1035 | trace_rcu_utilization(TPS("Start boost kthread@rcu_yield")); |
27f4d280 PM |
1036 | spincnt = 0; |
1037 | } | |
1038 | } | |
1217ed1b | 1039 | /* NOTREACHED */ |
f7f7bac9 | 1040 | trace_rcu_utilization(TPS("End boost kthread@notreached")); |
27f4d280 PM |
1041 | return 0; |
1042 | } | |
1043 | ||
1044 | /* | |
1045 | * Check to see if it is time to start boosting RCU readers that are | |
1046 | * blocking the current grace period, and, if so, tell the per-rcu_node | |
1047 | * kthread to start boosting them. If there is an expedited grace | |
1048 | * period in progress, it is always time to boost. | |
1049 | * | |
b065a853 PM |
1050 | * The caller must hold rnp->lock, which this function releases. |
1051 | * The ->boost_kthread_task is immortal, so we don't need to worry | |
1052 | * about it going away. | |
27f4d280 | 1053 | */ |
1217ed1b | 1054 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1055 | __releases(rnp->lock) |
27f4d280 PM |
1056 | { |
1057 | struct task_struct *t; | |
1058 | ||
ea9b0c8a | 1059 | lockdep_assert_held(&rnp->lock); |
0ea1f2eb PM |
1060 | if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) { |
1061 | rnp->n_balk_exp_gp_tasks++; | |
67c583a7 | 1062 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 | 1063 | return; |
0ea1f2eb | 1064 | } |
27f4d280 PM |
1065 | if (rnp->exp_tasks != NULL || |
1066 | (rnp->gp_tasks != NULL && | |
1067 | rnp->boost_tasks == NULL && | |
1068 | rnp->qsmask == 0 && | |
1069 | ULONG_CMP_GE(jiffies, rnp->boost_time))) { | |
1070 | if (rnp->exp_tasks == NULL) | |
1071 | rnp->boost_tasks = rnp->gp_tasks; | |
67c583a7 | 1072 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 | 1073 | t = rnp->boost_kthread_task; |
5d01bbd1 TG |
1074 | if (t) |
1075 | rcu_wake_cond(t, rnp->boost_kthread_status); | |
1217ed1b | 1076 | } else { |
0ea1f2eb | 1077 | rcu_initiate_boost_trace(rnp); |
67c583a7 | 1078 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
1217ed1b | 1079 | } |
27f4d280 PM |
1080 | } |
1081 | ||
a46e0899 PM |
1082 | /* |
1083 | * Wake up the per-CPU kthread to invoke RCU callbacks. | |
1084 | */ | |
1085 | static void invoke_rcu_callbacks_kthread(void) | |
1086 | { | |
1087 | unsigned long flags; | |
1088 | ||
1089 | local_irq_save(flags); | |
1090 | __this_cpu_write(rcu_cpu_has_work, 1); | |
1eb52121 | 1091 | if (__this_cpu_read(rcu_cpu_kthread_task) != NULL && |
5d01bbd1 TG |
1092 | current != __this_cpu_read(rcu_cpu_kthread_task)) { |
1093 | rcu_wake_cond(__this_cpu_read(rcu_cpu_kthread_task), | |
1094 | __this_cpu_read(rcu_cpu_kthread_status)); | |
1095 | } | |
a46e0899 PM |
1096 | local_irq_restore(flags); |
1097 | } | |
1098 | ||
dff1672d PM |
1099 | /* |
1100 | * Is the current CPU running the RCU-callbacks kthread? | |
1101 | * Caller must have preemption disabled. | |
1102 | */ | |
1103 | static bool rcu_is_callbacks_kthread(void) | |
1104 | { | |
c9d4b0af | 1105 | return __this_cpu_read(rcu_cpu_kthread_task) == current; |
dff1672d PM |
1106 | } |
1107 | ||
27f4d280 PM |
1108 | #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000) |
1109 | ||
1110 | /* | |
1111 | * Do priority-boost accounting for the start of a new grace period. | |
1112 | */ | |
1113 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) | |
1114 | { | |
1115 | rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES; | |
1116 | } | |
1117 | ||
27f4d280 PM |
1118 | /* |
1119 | * Create an RCU-boost kthread for the specified node if one does not | |
1120 | * already exist. We only create this kthread for preemptible RCU. | |
1121 | * Returns zero if all is well, a negated errno otherwise. | |
1122 | */ | |
49fb4c62 | 1123 | static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp, |
0aa04b05 | 1124 | struct rcu_node *rnp) |
27f4d280 | 1125 | { |
5d01bbd1 | 1126 | int rnp_index = rnp - &rsp->node[0]; |
27f4d280 PM |
1127 | unsigned long flags; |
1128 | struct sched_param sp; | |
1129 | struct task_struct *t; | |
1130 | ||
e63c887c | 1131 | if (rcu_state_p != rsp) |
27f4d280 | 1132 | return 0; |
5d01bbd1 | 1133 | |
0aa04b05 | 1134 | if (!rcu_scheduler_fully_active || rcu_rnp_online_cpus(rnp) == 0) |
5d01bbd1 TG |
1135 | return 0; |
1136 | ||
a46e0899 | 1137 | rsp->boost = 1; |
27f4d280 PM |
1138 | if (rnp->boost_kthread_task != NULL) |
1139 | return 0; | |
1140 | t = kthread_create(rcu_boost_kthread, (void *)rnp, | |
5b61b0ba | 1141 | "rcub/%d", rnp_index); |
27f4d280 PM |
1142 | if (IS_ERR(t)) |
1143 | return PTR_ERR(t); | |
2a67e741 | 1144 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 | 1145 | rnp->boost_kthread_task = t; |
67c583a7 | 1146 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
21871d7e | 1147 | sp.sched_priority = kthread_prio; |
27f4d280 | 1148 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); |
9a432736 | 1149 | wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */ |
27f4d280 PM |
1150 | return 0; |
1151 | } | |
1152 | ||
f8b7fc6b PM |
1153 | static void rcu_kthread_do_work(void) |
1154 | { | |
c9d4b0af CL |
1155 | rcu_do_batch(&rcu_sched_state, this_cpu_ptr(&rcu_sched_data)); |
1156 | rcu_do_batch(&rcu_bh_state, this_cpu_ptr(&rcu_bh_data)); | |
f8b7fc6b PM |
1157 | rcu_preempt_do_callbacks(); |
1158 | } | |
1159 | ||
62ab7072 | 1160 | static void rcu_cpu_kthread_setup(unsigned int cpu) |
f8b7fc6b | 1161 | { |
f8b7fc6b | 1162 | struct sched_param sp; |
f8b7fc6b | 1163 | |
21871d7e | 1164 | sp.sched_priority = kthread_prio; |
62ab7072 | 1165 | sched_setscheduler_nocheck(current, SCHED_FIFO, &sp); |
f8b7fc6b PM |
1166 | } |
1167 | ||
62ab7072 | 1168 | static void rcu_cpu_kthread_park(unsigned int cpu) |
f8b7fc6b | 1169 | { |
62ab7072 | 1170 | per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU; |
f8b7fc6b PM |
1171 | } |
1172 | ||
62ab7072 | 1173 | static int rcu_cpu_kthread_should_run(unsigned int cpu) |
f8b7fc6b | 1174 | { |
c9d4b0af | 1175 | return __this_cpu_read(rcu_cpu_has_work); |
f8b7fc6b PM |
1176 | } |
1177 | ||
1178 | /* | |
1179 | * Per-CPU kernel thread that invokes RCU callbacks. This replaces the | |
e0f23060 PM |
1180 | * RCU softirq used in flavors and configurations of RCU that do not |
1181 | * support RCU priority boosting. | |
f8b7fc6b | 1182 | */ |
62ab7072 | 1183 | static void rcu_cpu_kthread(unsigned int cpu) |
f8b7fc6b | 1184 | { |
c9d4b0af CL |
1185 | unsigned int *statusp = this_cpu_ptr(&rcu_cpu_kthread_status); |
1186 | char work, *workp = this_cpu_ptr(&rcu_cpu_has_work); | |
62ab7072 | 1187 | int spincnt; |
f8b7fc6b | 1188 | |
62ab7072 | 1189 | for (spincnt = 0; spincnt < 10; spincnt++) { |
f7f7bac9 | 1190 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_wait")); |
f8b7fc6b | 1191 | local_bh_disable(); |
f8b7fc6b | 1192 | *statusp = RCU_KTHREAD_RUNNING; |
62ab7072 PM |
1193 | this_cpu_inc(rcu_cpu_kthread_loops); |
1194 | local_irq_disable(); | |
f8b7fc6b PM |
1195 | work = *workp; |
1196 | *workp = 0; | |
62ab7072 | 1197 | local_irq_enable(); |
f8b7fc6b PM |
1198 | if (work) |
1199 | rcu_kthread_do_work(); | |
1200 | local_bh_enable(); | |
62ab7072 | 1201 | if (*workp == 0) { |
f7f7bac9 | 1202 | trace_rcu_utilization(TPS("End CPU kthread@rcu_wait")); |
62ab7072 PM |
1203 | *statusp = RCU_KTHREAD_WAITING; |
1204 | return; | |
f8b7fc6b PM |
1205 | } |
1206 | } | |
62ab7072 | 1207 | *statusp = RCU_KTHREAD_YIELDING; |
f7f7bac9 | 1208 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_yield")); |
62ab7072 | 1209 | schedule_timeout_interruptible(2); |
f7f7bac9 | 1210 | trace_rcu_utilization(TPS("End CPU kthread@rcu_yield")); |
62ab7072 | 1211 | *statusp = RCU_KTHREAD_WAITING; |
f8b7fc6b PM |
1212 | } |
1213 | ||
1214 | /* | |
1215 | * Set the per-rcu_node kthread's affinity to cover all CPUs that are | |
1216 | * served by the rcu_node in question. The CPU hotplug lock is still | |
1217 | * held, so the value of rnp->qsmaskinit will be stable. | |
1218 | * | |
1219 | * We don't include outgoingcpu in the affinity set, use -1 if there is | |
1220 | * no outgoing CPU. If there are no CPUs left in the affinity set, | |
1221 | * this function allows the kthread to execute on any CPU. | |
1222 | */ | |
5d01bbd1 | 1223 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b | 1224 | { |
5d01bbd1 | 1225 | struct task_struct *t = rnp->boost_kthread_task; |
0aa04b05 | 1226 | unsigned long mask = rcu_rnp_online_cpus(rnp); |
f8b7fc6b PM |
1227 | cpumask_var_t cm; |
1228 | int cpu; | |
f8b7fc6b | 1229 | |
5d01bbd1 | 1230 | if (!t) |
f8b7fc6b | 1231 | return; |
5d01bbd1 | 1232 | if (!zalloc_cpumask_var(&cm, GFP_KERNEL)) |
f8b7fc6b | 1233 | return; |
bc75e999 MR |
1234 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1235 | if ((mask & leaf_node_cpu_bit(rnp, cpu)) && | |
1236 | cpu != outgoingcpu) | |
f8b7fc6b | 1237 | cpumask_set_cpu(cpu, cm); |
5d0b0249 | 1238 | if (cpumask_weight(cm) == 0) |
f8b7fc6b | 1239 | cpumask_setall(cm); |
5d01bbd1 | 1240 | set_cpus_allowed_ptr(t, cm); |
f8b7fc6b PM |
1241 | free_cpumask_var(cm); |
1242 | } | |
1243 | ||
62ab7072 PM |
1244 | static struct smp_hotplug_thread rcu_cpu_thread_spec = { |
1245 | .store = &rcu_cpu_kthread_task, | |
1246 | .thread_should_run = rcu_cpu_kthread_should_run, | |
1247 | .thread_fn = rcu_cpu_kthread, | |
1248 | .thread_comm = "rcuc/%u", | |
1249 | .setup = rcu_cpu_kthread_setup, | |
1250 | .park = rcu_cpu_kthread_park, | |
1251 | }; | |
f8b7fc6b PM |
1252 | |
1253 | /* | |
9386c0b7 | 1254 | * Spawn boost kthreads -- called as soon as the scheduler is running. |
f8b7fc6b | 1255 | */ |
9386c0b7 | 1256 | static void __init rcu_spawn_boost_kthreads(void) |
f8b7fc6b | 1257 | { |
f8b7fc6b | 1258 | struct rcu_node *rnp; |
5d01bbd1 | 1259 | int cpu; |
f8b7fc6b | 1260 | |
62ab7072 | 1261 | for_each_possible_cpu(cpu) |
f8b7fc6b | 1262 | per_cpu(rcu_cpu_has_work, cpu) = 0; |
62ab7072 | 1263 | BUG_ON(smpboot_register_percpu_thread(&rcu_cpu_thread_spec)); |
3e9f5c70 PM |
1264 | rcu_for_each_leaf_node(rcu_state_p, rnp) |
1265 | (void)rcu_spawn_one_boost_kthread(rcu_state_p, rnp); | |
f8b7fc6b | 1266 | } |
f8b7fc6b | 1267 | |
49fb4c62 | 1268 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b | 1269 | { |
e534165b | 1270 | struct rcu_data *rdp = per_cpu_ptr(rcu_state_p->rda, cpu); |
f8b7fc6b PM |
1271 | struct rcu_node *rnp = rdp->mynode; |
1272 | ||
1273 | /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */ | |
62ab7072 | 1274 | if (rcu_scheduler_fully_active) |
e534165b | 1275 | (void)rcu_spawn_one_boost_kthread(rcu_state_p, rnp); |
f8b7fc6b PM |
1276 | } |
1277 | ||
27f4d280 PM |
1278 | #else /* #ifdef CONFIG_RCU_BOOST */ |
1279 | ||
1217ed1b | 1280 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1281 | __releases(rnp->lock) |
27f4d280 | 1282 | { |
67c583a7 | 1283 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
1284 | } |
1285 | ||
a46e0899 | 1286 | static void invoke_rcu_callbacks_kthread(void) |
27f4d280 | 1287 | { |
a46e0899 | 1288 | WARN_ON_ONCE(1); |
27f4d280 PM |
1289 | } |
1290 | ||
dff1672d PM |
1291 | static bool rcu_is_callbacks_kthread(void) |
1292 | { | |
1293 | return false; | |
1294 | } | |
1295 | ||
27f4d280 PM |
1296 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) |
1297 | { | |
1298 | } | |
1299 | ||
5d01bbd1 | 1300 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b PM |
1301 | { |
1302 | } | |
1303 | ||
9386c0b7 | 1304 | static void __init rcu_spawn_boost_kthreads(void) |
b0d30417 | 1305 | { |
b0d30417 | 1306 | } |
b0d30417 | 1307 | |
49fb4c62 | 1308 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b PM |
1309 | { |
1310 | } | |
1311 | ||
27f4d280 PM |
1312 | #endif /* #else #ifdef CONFIG_RCU_BOOST */ |
1313 | ||
8bd93a2c PM |
1314 | #if !defined(CONFIG_RCU_FAST_NO_HZ) |
1315 | ||
1316 | /* | |
1317 | * Check to see if any future RCU-related work will need to be done | |
1318 | * by the current CPU, even if none need be done immediately, returning | |
1319 | * 1 if so. This function is part of the RCU implementation; it is -not- | |
1320 | * an exported member of the RCU API. | |
1321 | * | |
7cb92499 PM |
1322 | * Because we not have RCU_FAST_NO_HZ, just check whether this CPU needs |
1323 | * any flavor of RCU. | |
8bd93a2c | 1324 | */ |
c1ad348b | 1325 | int rcu_needs_cpu(u64 basemono, u64 *nextevt) |
8bd93a2c | 1326 | { |
c1ad348b | 1327 | *nextevt = KTIME_MAX; |
3382adbc PM |
1328 | return IS_ENABLED(CONFIG_RCU_NOCB_CPU_ALL) |
1329 | ? 0 : rcu_cpu_has_callbacks(NULL); | |
7cb92499 PM |
1330 | } |
1331 | ||
1332 | /* | |
1333 | * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up | |
1334 | * after it. | |
1335 | */ | |
8fa7845d | 1336 | static void rcu_cleanup_after_idle(void) |
7cb92499 PM |
1337 | { |
1338 | } | |
1339 | ||
aea1b35e | 1340 | /* |
a858af28 | 1341 | * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n, |
aea1b35e PM |
1342 | * is nothing. |
1343 | */ | |
198bbf81 | 1344 | static void rcu_prepare_for_idle(void) |
aea1b35e PM |
1345 | { |
1346 | } | |
1347 | ||
c57afe80 PM |
1348 | /* |
1349 | * Don't bother keeping a running count of the number of RCU callbacks | |
1350 | * posted because CONFIG_RCU_FAST_NO_HZ=n. | |
1351 | */ | |
1352 | static void rcu_idle_count_callbacks_posted(void) | |
1353 | { | |
1354 | } | |
1355 | ||
8bd93a2c PM |
1356 | #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
1357 | ||
f23f7fa1 PM |
1358 | /* |
1359 | * This code is invoked when a CPU goes idle, at which point we want | |
1360 | * to have the CPU do everything required for RCU so that it can enter | |
1361 | * the energy-efficient dyntick-idle mode. This is handled by a | |
1362 | * state machine implemented by rcu_prepare_for_idle() below. | |
1363 | * | |
1364 | * The following three proprocessor symbols control this state machine: | |
1365 | * | |
f23f7fa1 PM |
1366 | * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted |
1367 | * to sleep in dyntick-idle mode with RCU callbacks pending. This | |
1368 | * is sized to be roughly one RCU grace period. Those energy-efficiency | |
1369 | * benchmarkers who might otherwise be tempted to set this to a large | |
1370 | * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your | |
1371 | * system. And if you are -that- concerned about energy efficiency, | |
1372 | * just power the system down and be done with it! | |
778d250a PM |
1373 | * RCU_IDLE_LAZY_GP_DELAY gives the number of jiffies that a CPU is |
1374 | * permitted to sleep in dyntick-idle mode with only lazy RCU | |
1375 | * callbacks pending. Setting this too high can OOM your system. | |
f23f7fa1 PM |
1376 | * |
1377 | * The values below work well in practice. If future workloads require | |
1378 | * adjustment, they can be converted into kernel config parameters, though | |
1379 | * making the state machine smarter might be a better option. | |
1380 | */ | |
e84c48ae | 1381 | #define RCU_IDLE_GP_DELAY 4 /* Roughly one grace period. */ |
778d250a | 1382 | #define RCU_IDLE_LAZY_GP_DELAY (6 * HZ) /* Roughly six seconds. */ |
f23f7fa1 | 1383 | |
5e44ce35 PM |
1384 | static int rcu_idle_gp_delay = RCU_IDLE_GP_DELAY; |
1385 | module_param(rcu_idle_gp_delay, int, 0644); | |
1386 | static int rcu_idle_lazy_gp_delay = RCU_IDLE_LAZY_GP_DELAY; | |
1387 | module_param(rcu_idle_lazy_gp_delay, int, 0644); | |
486e2593 | 1388 | |
486e2593 | 1389 | /* |
c229828c PM |
1390 | * Try to advance callbacks for all flavors of RCU on the current CPU, but |
1391 | * only if it has been awhile since the last time we did so. Afterwards, | |
1392 | * if there are any callbacks ready for immediate invocation, return true. | |
486e2593 | 1393 | */ |
f1f399d1 | 1394 | static bool __maybe_unused rcu_try_advance_all_cbs(void) |
486e2593 | 1395 | { |
c0f4dfd4 PM |
1396 | bool cbs_ready = false; |
1397 | struct rcu_data *rdp; | |
c229828c | 1398 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
c0f4dfd4 PM |
1399 | struct rcu_node *rnp; |
1400 | struct rcu_state *rsp; | |
486e2593 | 1401 | |
c229828c PM |
1402 | /* Exit early if we advanced recently. */ |
1403 | if (jiffies == rdtp->last_advance_all) | |
d0bc90fd | 1404 | return false; |
c229828c PM |
1405 | rdtp->last_advance_all = jiffies; |
1406 | ||
c0f4dfd4 PM |
1407 | for_each_rcu_flavor(rsp) { |
1408 | rdp = this_cpu_ptr(rsp->rda); | |
1409 | rnp = rdp->mynode; | |
486e2593 | 1410 | |
c0f4dfd4 PM |
1411 | /* |
1412 | * Don't bother checking unless a grace period has | |
1413 | * completed since we last checked and there are | |
1414 | * callbacks not yet ready to invoke. | |
1415 | */ | |
e3663b10 | 1416 | if ((rdp->completed != rnp->completed || |
7d0ae808 | 1417 | unlikely(READ_ONCE(rdp->gpwrap))) && |
15fecf89 | 1418 | rcu_segcblist_pend_cbs(&rdp->cblist)) |
470716fc | 1419 | note_gp_changes(rsp, rdp); |
486e2593 | 1420 | |
15fecf89 | 1421 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
c0f4dfd4 PM |
1422 | cbs_ready = true; |
1423 | } | |
1424 | return cbs_ready; | |
486e2593 PM |
1425 | } |
1426 | ||
aa9b1630 | 1427 | /* |
c0f4dfd4 PM |
1428 | * Allow the CPU to enter dyntick-idle mode unless it has callbacks ready |
1429 | * to invoke. If the CPU has callbacks, try to advance them. Tell the | |
1430 | * caller to set the timeout based on whether or not there are non-lazy | |
1431 | * callbacks. | |
aa9b1630 | 1432 | * |
c0f4dfd4 | 1433 | * The caller must have disabled interrupts. |
aa9b1630 | 1434 | */ |
c1ad348b | 1435 | int rcu_needs_cpu(u64 basemono, u64 *nextevt) |
aa9b1630 | 1436 | { |
aa6da514 | 1437 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
c1ad348b | 1438 | unsigned long dj; |
aa9b1630 | 1439 | |
ea9b0c8a | 1440 | RCU_LOCKDEP_WARN(!irqs_disabled(), "rcu_needs_cpu() invoked with irqs enabled!!!"); |
3382adbc | 1441 | if (IS_ENABLED(CONFIG_RCU_NOCB_CPU_ALL)) { |
43224b96 | 1442 | *nextevt = KTIME_MAX; |
3382adbc PM |
1443 | return 0; |
1444 | } | |
1445 | ||
c0f4dfd4 PM |
1446 | /* Snapshot to detect later posting of non-lazy callback. */ |
1447 | rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted; | |
1448 | ||
aa9b1630 | 1449 | /* If no callbacks, RCU doesn't need the CPU. */ |
aa6da514 | 1450 | if (!rcu_cpu_has_callbacks(&rdtp->all_lazy)) { |
c1ad348b | 1451 | *nextevt = KTIME_MAX; |
aa9b1630 PM |
1452 | return 0; |
1453 | } | |
c0f4dfd4 PM |
1454 | |
1455 | /* Attempt to advance callbacks. */ | |
1456 | if (rcu_try_advance_all_cbs()) { | |
1457 | /* Some ready to invoke, so initiate later invocation. */ | |
1458 | invoke_rcu_core(); | |
aa9b1630 PM |
1459 | return 1; |
1460 | } | |
c0f4dfd4 PM |
1461 | rdtp->last_accelerate = jiffies; |
1462 | ||
1463 | /* Request timer delay depending on laziness, and round. */ | |
6faf7283 | 1464 | if (!rdtp->all_lazy) { |
c1ad348b | 1465 | dj = round_up(rcu_idle_gp_delay + jiffies, |
c0f4dfd4 | 1466 | rcu_idle_gp_delay) - jiffies; |
e84c48ae | 1467 | } else { |
c1ad348b | 1468 | dj = round_jiffies(rcu_idle_lazy_gp_delay + jiffies) - jiffies; |
e84c48ae | 1469 | } |
c1ad348b | 1470 | *nextevt = basemono + dj * TICK_NSEC; |
aa9b1630 PM |
1471 | return 0; |
1472 | } | |
1473 | ||
21e52e15 | 1474 | /* |
c0f4dfd4 PM |
1475 | * Prepare a CPU for idle from an RCU perspective. The first major task |
1476 | * is to sense whether nohz mode has been enabled or disabled via sysfs. | |
1477 | * The second major task is to check to see if a non-lazy callback has | |
1478 | * arrived at a CPU that previously had only lazy callbacks. The third | |
1479 | * major task is to accelerate (that is, assign grace-period numbers to) | |
1480 | * any recently arrived callbacks. | |
aea1b35e PM |
1481 | * |
1482 | * The caller must have disabled interrupts. | |
8bd93a2c | 1483 | */ |
198bbf81 | 1484 | static void rcu_prepare_for_idle(void) |
8bd93a2c | 1485 | { |
48a7639c | 1486 | bool needwake; |
c0f4dfd4 | 1487 | struct rcu_data *rdp; |
198bbf81 | 1488 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
c0f4dfd4 PM |
1489 | struct rcu_node *rnp; |
1490 | struct rcu_state *rsp; | |
9d2ad243 PM |
1491 | int tne; |
1492 | ||
ea9b0c8a | 1493 | RCU_LOCKDEP_WARN(!irqs_disabled(), "rcu_prepare_for_idle() invoked with irqs enabled!!!"); |
f0f2e7d3 PM |
1494 | if (IS_ENABLED(CONFIG_RCU_NOCB_CPU_ALL) || |
1495 | rcu_is_nocb_cpu(smp_processor_id())) | |
3382adbc PM |
1496 | return; |
1497 | ||
9d2ad243 | 1498 | /* Handle nohz enablement switches conservatively. */ |
7d0ae808 | 1499 | tne = READ_ONCE(tick_nohz_active); |
9d2ad243 | 1500 | if (tne != rdtp->tick_nohz_enabled_snap) { |
aa6da514 | 1501 | if (rcu_cpu_has_callbacks(NULL)) |
9d2ad243 PM |
1502 | invoke_rcu_core(); /* force nohz to see update. */ |
1503 | rdtp->tick_nohz_enabled_snap = tne; | |
1504 | return; | |
1505 | } | |
1506 | if (!tne) | |
1507 | return; | |
f511fc62 | 1508 | |
c57afe80 | 1509 | /* |
c0f4dfd4 PM |
1510 | * If a non-lazy callback arrived at a CPU having only lazy |
1511 | * callbacks, invoke RCU core for the side-effect of recalculating | |
1512 | * idle duration on re-entry to idle. | |
c57afe80 | 1513 | */ |
c0f4dfd4 PM |
1514 | if (rdtp->all_lazy && |
1515 | rdtp->nonlazy_posted != rdtp->nonlazy_posted_snap) { | |
c337f8f5 PM |
1516 | rdtp->all_lazy = false; |
1517 | rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted; | |
c0f4dfd4 | 1518 | invoke_rcu_core(); |
c57afe80 PM |
1519 | return; |
1520 | } | |
c57afe80 | 1521 | |
3084f2f8 | 1522 | /* |
c0f4dfd4 PM |
1523 | * If we have not yet accelerated this jiffy, accelerate all |
1524 | * callbacks on this CPU. | |
3084f2f8 | 1525 | */ |
c0f4dfd4 | 1526 | if (rdtp->last_accelerate == jiffies) |
aea1b35e | 1527 | return; |
c0f4dfd4 PM |
1528 | rdtp->last_accelerate = jiffies; |
1529 | for_each_rcu_flavor(rsp) { | |
198bbf81 | 1530 | rdp = this_cpu_ptr(rsp->rda); |
15fecf89 | 1531 | if (rcu_segcblist_pend_cbs(&rdp->cblist)) |
c0f4dfd4 PM |
1532 | continue; |
1533 | rnp = rdp->mynode; | |
2a67e741 | 1534 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
48a7639c | 1535 | needwake = rcu_accelerate_cbs(rsp, rnp, rdp); |
67c583a7 | 1536 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
48a7639c PM |
1537 | if (needwake) |
1538 | rcu_gp_kthread_wake(rsp); | |
77e38ed3 | 1539 | } |
c0f4dfd4 | 1540 | } |
3084f2f8 | 1541 | |
c0f4dfd4 PM |
1542 | /* |
1543 | * Clean up for exit from idle. Attempt to advance callbacks based on | |
1544 | * any grace periods that elapsed while the CPU was idle, and if any | |
1545 | * callbacks are now ready to invoke, initiate invocation. | |
1546 | */ | |
8fa7845d | 1547 | static void rcu_cleanup_after_idle(void) |
c0f4dfd4 | 1548 | { |
ea9b0c8a | 1549 | RCU_LOCKDEP_WARN(!irqs_disabled(), "rcu_cleanup_after_idle() invoked with irqs enabled!!!"); |
3382adbc PM |
1550 | if (IS_ENABLED(CONFIG_RCU_NOCB_CPU_ALL) || |
1551 | rcu_is_nocb_cpu(smp_processor_id())) | |
aea1b35e | 1552 | return; |
7a497c96 PM |
1553 | if (rcu_try_advance_all_cbs()) |
1554 | invoke_rcu_core(); | |
8bd93a2c PM |
1555 | } |
1556 | ||
c57afe80 | 1557 | /* |
98248a0e PM |
1558 | * Keep a running count of the number of non-lazy callbacks posted |
1559 | * on this CPU. This running counter (which is never decremented) allows | |
1560 | * rcu_prepare_for_idle() to detect when something out of the idle loop | |
1561 | * posts a callback, even if an equal number of callbacks are invoked. | |
1562 | * Of course, callbacks should only be posted from within a trace event | |
1563 | * designed to be called from idle or from within RCU_NONIDLE(). | |
c57afe80 PM |
1564 | */ |
1565 | static void rcu_idle_count_callbacks_posted(void) | |
1566 | { | |
5955f7ee | 1567 | __this_cpu_add(rcu_dynticks.nonlazy_posted, 1); |
c57afe80 PM |
1568 | } |
1569 | ||
b626c1b6 PM |
1570 | /* |
1571 | * Data for flushing lazy RCU callbacks at OOM time. | |
1572 | */ | |
1573 | static atomic_t oom_callback_count; | |
1574 | static DECLARE_WAIT_QUEUE_HEAD(oom_callback_wq); | |
1575 | ||
1576 | /* | |
1577 | * RCU OOM callback -- decrement the outstanding count and deliver the | |
1578 | * wake-up if we are the last one. | |
1579 | */ | |
1580 | static void rcu_oom_callback(struct rcu_head *rhp) | |
1581 | { | |
1582 | if (atomic_dec_and_test(&oom_callback_count)) | |
1583 | wake_up(&oom_callback_wq); | |
1584 | } | |
1585 | ||
1586 | /* | |
1587 | * Post an rcu_oom_notify callback on the current CPU if it has at | |
1588 | * least one lazy callback. This will unnecessarily post callbacks | |
1589 | * to CPUs that already have a non-lazy callback at the end of their | |
1590 | * callback list, but this is an infrequent operation, so accept some | |
1591 | * extra overhead to keep things simple. | |
1592 | */ | |
1593 | static void rcu_oom_notify_cpu(void *unused) | |
1594 | { | |
1595 | struct rcu_state *rsp; | |
1596 | struct rcu_data *rdp; | |
1597 | ||
1598 | for_each_rcu_flavor(rsp) { | |
fa07a58f | 1599 | rdp = raw_cpu_ptr(rsp->rda); |
15fecf89 | 1600 | if (rcu_segcblist_n_lazy_cbs(&rdp->cblist)) { |
b626c1b6 PM |
1601 | atomic_inc(&oom_callback_count); |
1602 | rsp->call(&rdp->oom_head, rcu_oom_callback); | |
1603 | } | |
1604 | } | |
1605 | } | |
1606 | ||
1607 | /* | |
1608 | * If low on memory, ensure that each CPU has a non-lazy callback. | |
1609 | * This will wake up CPUs that have only lazy callbacks, in turn | |
1610 | * ensuring that they free up the corresponding memory in a timely manner. | |
1611 | * Because an uncertain amount of memory will be freed in some uncertain | |
1612 | * timeframe, we do not claim to have freed anything. | |
1613 | */ | |
1614 | static int rcu_oom_notify(struct notifier_block *self, | |
1615 | unsigned long notused, void *nfreed) | |
1616 | { | |
1617 | int cpu; | |
1618 | ||
1619 | /* Wait for callbacks from earlier instance to complete. */ | |
1620 | wait_event(oom_callback_wq, atomic_read(&oom_callback_count) == 0); | |
78e4bc34 | 1621 | smp_mb(); /* Ensure callback reuse happens after callback invocation. */ |
b626c1b6 PM |
1622 | |
1623 | /* | |
1624 | * Prevent premature wakeup: ensure that all increments happen | |
1625 | * before there is a chance of the counter reaching zero. | |
1626 | */ | |
1627 | atomic_set(&oom_callback_count, 1); | |
1628 | ||
b626c1b6 PM |
1629 | for_each_online_cpu(cpu) { |
1630 | smp_call_function_single(cpu, rcu_oom_notify_cpu, NULL, 1); | |
bde6c3aa | 1631 | cond_resched_rcu_qs(); |
b626c1b6 | 1632 | } |
b626c1b6 PM |
1633 | |
1634 | /* Unconditionally decrement: no need to wake ourselves up. */ | |
1635 | atomic_dec(&oom_callback_count); | |
1636 | ||
1637 | return NOTIFY_OK; | |
1638 | } | |
1639 | ||
1640 | static struct notifier_block rcu_oom_nb = { | |
1641 | .notifier_call = rcu_oom_notify | |
1642 | }; | |
1643 | ||
1644 | static int __init rcu_register_oom_notifier(void) | |
1645 | { | |
1646 | register_oom_notifier(&rcu_oom_nb); | |
1647 | return 0; | |
1648 | } | |
1649 | early_initcall(rcu_register_oom_notifier); | |
1650 | ||
8bd93a2c | 1651 | #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
a858af28 | 1652 | |
a858af28 PM |
1653 | #ifdef CONFIG_RCU_FAST_NO_HZ |
1654 | ||
1655 | static void print_cpu_stall_fast_no_hz(char *cp, int cpu) | |
1656 | { | |
5955f7ee | 1657 | struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); |
c0f4dfd4 | 1658 | unsigned long nlpd = rdtp->nonlazy_posted - rdtp->nonlazy_posted_snap; |
a858af28 | 1659 | |
c0f4dfd4 PM |
1660 | sprintf(cp, "last_accelerate: %04lx/%04lx, nonlazy_posted: %ld, %c%c", |
1661 | rdtp->last_accelerate & 0xffff, jiffies & 0xffff, | |
1662 | ulong2long(nlpd), | |
1663 | rdtp->all_lazy ? 'L' : '.', | |
1664 | rdtp->tick_nohz_enabled_snap ? '.' : 'D'); | |
a858af28 PM |
1665 | } |
1666 | ||
1667 | #else /* #ifdef CONFIG_RCU_FAST_NO_HZ */ | |
1668 | ||
1669 | static void print_cpu_stall_fast_no_hz(char *cp, int cpu) | |
1670 | { | |
1c17e4d4 | 1671 | *cp = '\0'; |
a858af28 PM |
1672 | } |
1673 | ||
1674 | #endif /* #else #ifdef CONFIG_RCU_FAST_NO_HZ */ | |
1675 | ||
1676 | /* Initiate the stall-info list. */ | |
1677 | static void print_cpu_stall_info_begin(void) | |
1678 | { | |
efc151c3 | 1679 | pr_cont("\n"); |
a858af28 PM |
1680 | } |
1681 | ||
1682 | /* | |
1683 | * Print out diagnostic information for the specified stalled CPU. | |
1684 | * | |
1685 | * If the specified CPU is aware of the current RCU grace period | |
1686 | * (flavor specified by rsp), then print the number of scheduling | |
1687 | * clock interrupts the CPU has taken during the time that it has | |
1688 | * been aware. Otherwise, print the number of RCU grace periods | |
1689 | * that this CPU is ignorant of, for example, "1" if the CPU was | |
1690 | * aware of the previous grace period. | |
1691 | * | |
1692 | * Also print out idle and (if CONFIG_RCU_FAST_NO_HZ) idle-entry info. | |
1693 | */ | |
1694 | static void print_cpu_stall_info(struct rcu_state *rsp, int cpu) | |
1695 | { | |
1696 | char fast_no_hz[72]; | |
1697 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); | |
1698 | struct rcu_dynticks *rdtp = rdp->dynticks; | |
1699 | char *ticks_title; | |
1700 | unsigned long ticks_value; | |
1701 | ||
1702 | if (rsp->gpnum == rdp->gpnum) { | |
1703 | ticks_title = "ticks this GP"; | |
1704 | ticks_value = rdp->ticks_this_gp; | |
1705 | } else { | |
1706 | ticks_title = "GPs behind"; | |
1707 | ticks_value = rsp->gpnum - rdp->gpnum; | |
1708 | } | |
1709 | print_cpu_stall_fast_no_hz(fast_no_hz, cpu); | |
7f21aeef PM |
1710 | pr_err("\t%d-%c%c%c: (%lu %s) idle=%03x/%llx/%d softirq=%u/%u fqs=%ld %s\n", |
1711 | cpu, | |
1712 | "O."[!!cpu_online(cpu)], | |
1713 | "o."[!!(rdp->grpmask & rdp->mynode->qsmaskinit)], | |
1714 | "N."[!!(rdp->grpmask & rdp->mynode->qsmaskinitnext)], | |
1715 | ticks_value, ticks_title, | |
02a5c550 | 1716 | rcu_dynticks_snap(rdtp) & 0xfff, |
a858af28 | 1717 | rdtp->dynticks_nesting, rdtp->dynticks_nmi_nesting, |
6231069b | 1718 | rdp->softirq_snap, kstat_softirqs_cpu(RCU_SOFTIRQ, cpu), |
7d0ae808 | 1719 | READ_ONCE(rsp->n_force_qs) - rsp->n_force_qs_gpstart, |
a858af28 PM |
1720 | fast_no_hz); |
1721 | } | |
1722 | ||
1723 | /* Terminate the stall-info list. */ | |
1724 | static void print_cpu_stall_info_end(void) | |
1725 | { | |
efc151c3 | 1726 | pr_err("\t"); |
a858af28 PM |
1727 | } |
1728 | ||
1729 | /* Zero ->ticks_this_gp for all flavors of RCU. */ | |
1730 | static void zero_cpu_stall_ticks(struct rcu_data *rdp) | |
1731 | { | |
1732 | rdp->ticks_this_gp = 0; | |
6231069b | 1733 | rdp->softirq_snap = kstat_softirqs_cpu(RCU_SOFTIRQ, smp_processor_id()); |
a858af28 PM |
1734 | } |
1735 | ||
1736 | /* Increment ->ticks_this_gp for all flavors of RCU. */ | |
1737 | static void increment_cpu_stall_ticks(void) | |
1738 | { | |
115f7a7c PM |
1739 | struct rcu_state *rsp; |
1740 | ||
1741 | for_each_rcu_flavor(rsp) | |
fa07a58f | 1742 | raw_cpu_inc(rsp->rda->ticks_this_gp); |
a858af28 PM |
1743 | } |
1744 | ||
3fbfbf7a PM |
1745 | #ifdef CONFIG_RCU_NOCB_CPU |
1746 | ||
1747 | /* | |
1748 | * Offload callback processing from the boot-time-specified set of CPUs | |
1749 | * specified by rcu_nocb_mask. For each CPU in the set, there is a | |
1750 | * kthread created that pulls the callbacks from the corresponding CPU, | |
1751 | * waits for a grace period to elapse, and invokes the callbacks. | |
1752 | * The no-CBs CPUs do a wake_up() on their kthread when they insert | |
1753 | * a callback into any empty list, unless the rcu_nocb_poll boot parameter | |
1754 | * has been specified, in which case each kthread actively polls its | |
1755 | * CPU. (Which isn't so great for energy efficiency, but which does | |
1756 | * reduce RCU's overhead on that CPU.) | |
1757 | * | |
1758 | * This is intended to be used in conjunction with Frederic Weisbecker's | |
1759 | * adaptive-idle work, which would seriously reduce OS jitter on CPUs | |
1760 | * running CPU-bound user-mode computations. | |
1761 | * | |
1762 | * Offloading of callback processing could also in theory be used as | |
1763 | * an energy-efficiency measure because CPUs with no RCU callbacks | |
1764 | * queued are more aggressive about entering dyntick-idle mode. | |
1765 | */ | |
1766 | ||
1767 | ||
1768 | /* Parse the boot-time rcu_nocb_mask CPU list from the kernel parameters. */ | |
1769 | static int __init rcu_nocb_setup(char *str) | |
1770 | { | |
1771 | alloc_bootmem_cpumask_var(&rcu_nocb_mask); | |
1772 | have_rcu_nocb_mask = true; | |
1773 | cpulist_parse(str, rcu_nocb_mask); | |
1774 | return 1; | |
1775 | } | |
1776 | __setup("rcu_nocbs=", rcu_nocb_setup); | |
1777 | ||
1b0048a4 PG |
1778 | static int __init parse_rcu_nocb_poll(char *arg) |
1779 | { | |
5455a7f6 | 1780 | rcu_nocb_poll = true; |
1b0048a4 PG |
1781 | return 0; |
1782 | } | |
1783 | early_param("rcu_nocb_poll", parse_rcu_nocb_poll); | |
1784 | ||
dae6e64d | 1785 | /* |
0446be48 PM |
1786 | * Wake up any no-CBs CPUs' kthreads that were waiting on the just-ended |
1787 | * grace period. | |
dae6e64d | 1788 | */ |
abedf8e2 | 1789 | static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq) |
dae6e64d | 1790 | { |
abedf8e2 | 1791 | swake_up_all(sq); |
dae6e64d PM |
1792 | } |
1793 | ||
1794 | /* | |
8b425aa8 | 1795 | * Set the root rcu_node structure's ->need_future_gp field |
dae6e64d PM |
1796 | * based on the sum of those of all rcu_node structures. This does |
1797 | * double-count the root rcu_node structure's requests, but this | |
1798 | * is necessary to handle the possibility of a rcu_nocb_kthread() | |
1799 | * having awakened during the time that the rcu_node structures | |
1800 | * were being updated for the end of the previous grace period. | |
34ed6246 | 1801 | */ |
dae6e64d PM |
1802 | static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq) |
1803 | { | |
8b425aa8 | 1804 | rnp->need_future_gp[(rnp->completed + 1) & 0x1] += nrq; |
dae6e64d PM |
1805 | } |
1806 | ||
abedf8e2 | 1807 | static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp) |
065bb78c DW |
1808 | { |
1809 | return &rnp->nocb_gp_wq[rnp->completed & 0x1]; | |
1810 | } | |
1811 | ||
dae6e64d | 1812 | static void rcu_init_one_nocb(struct rcu_node *rnp) |
34ed6246 | 1813 | { |
abedf8e2 PG |
1814 | init_swait_queue_head(&rnp->nocb_gp_wq[0]); |
1815 | init_swait_queue_head(&rnp->nocb_gp_wq[1]); | |
34ed6246 PM |
1816 | } |
1817 | ||
2f33b512 | 1818 | #ifndef CONFIG_RCU_NOCB_CPU_ALL |
24342c96 | 1819 | /* Is the specified CPU a no-CBs CPU? */ |
d1e43fa5 | 1820 | bool rcu_is_nocb_cpu(int cpu) |
3fbfbf7a PM |
1821 | { |
1822 | if (have_rcu_nocb_mask) | |
1823 | return cpumask_test_cpu(cpu, rcu_nocb_mask); | |
1824 | return false; | |
1825 | } | |
2f33b512 | 1826 | #endif /* #ifndef CONFIG_RCU_NOCB_CPU_ALL */ |
3fbfbf7a | 1827 | |
fbce7497 PM |
1828 | /* |
1829 | * Kick the leader kthread for this NOCB group. | |
1830 | */ | |
1831 | static void wake_nocb_leader(struct rcu_data *rdp, bool force) | |
1832 | { | |
1833 | struct rcu_data *rdp_leader = rdp->nocb_leader; | |
1834 | ||
7d0ae808 | 1835 | if (!READ_ONCE(rdp_leader->nocb_kthread)) |
fbce7497 | 1836 | return; |
7d0ae808 | 1837 | if (READ_ONCE(rdp_leader->nocb_leader_sleep) || force) { |
39953dfd | 1838 | /* Prior smp_mb__after_atomic() orders against prior enqueue. */ |
7d0ae808 | 1839 | WRITE_ONCE(rdp_leader->nocb_leader_sleep, false); |
6b5fc3a1 | 1840 | smp_mb(); /* ->nocb_leader_sleep before swake_up(). */ |
abedf8e2 | 1841 | swake_up(&rdp_leader->nocb_wq); |
fbce7497 PM |
1842 | } |
1843 | } | |
1844 | ||
d7e29933 PM |
1845 | /* |
1846 | * Does the specified CPU need an RCU callback for the specified flavor | |
1847 | * of rcu_barrier()? | |
1848 | */ | |
1849 | static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu) | |
1850 | { | |
1851 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); | |
41050a00 PM |
1852 | unsigned long ret; |
1853 | #ifdef CONFIG_PROVE_RCU | |
d7e29933 | 1854 | struct rcu_head *rhp; |
41050a00 | 1855 | #endif /* #ifdef CONFIG_PROVE_RCU */ |
d7e29933 | 1856 | |
41050a00 PM |
1857 | /* |
1858 | * Check count of all no-CBs callbacks awaiting invocation. | |
1859 | * There needs to be a barrier before this function is called, | |
1860 | * but associated with a prior determination that no more | |
1861 | * callbacks would be posted. In the worst case, the first | |
1862 | * barrier in _rcu_barrier() suffices (but the caller cannot | |
1863 | * necessarily rely on this, not a substitute for the caller | |
1864 | * getting the concurrency design right!). There must also be | |
1865 | * a barrier between the following load an posting of a callback | |
1866 | * (if a callback is in fact needed). This is associated with an | |
1867 | * atomic_inc() in the caller. | |
1868 | */ | |
1869 | ret = atomic_long_read(&rdp->nocb_q_count); | |
d7e29933 | 1870 | |
41050a00 | 1871 | #ifdef CONFIG_PROVE_RCU |
7d0ae808 | 1872 | rhp = READ_ONCE(rdp->nocb_head); |
d7e29933 | 1873 | if (!rhp) |
7d0ae808 | 1874 | rhp = READ_ONCE(rdp->nocb_gp_head); |
d7e29933 | 1875 | if (!rhp) |
7d0ae808 | 1876 | rhp = READ_ONCE(rdp->nocb_follower_head); |
d7e29933 PM |
1877 | |
1878 | /* Having no rcuo kthread but CBs after scheduler starts is bad! */ | |
7d0ae808 | 1879 | if (!READ_ONCE(rdp->nocb_kthread) && rhp && |
59f792d1 | 1880 | rcu_scheduler_fully_active) { |
d7e29933 PM |
1881 | /* RCU callback enqueued before CPU first came online??? */ |
1882 | pr_err("RCU: Never-onlined no-CBs CPU %d has CB %p\n", | |
1883 | cpu, rhp->func); | |
1884 | WARN_ON_ONCE(1); | |
1885 | } | |
41050a00 | 1886 | #endif /* #ifdef CONFIG_PROVE_RCU */ |
d7e29933 | 1887 | |
41050a00 | 1888 | return !!ret; |
d7e29933 PM |
1889 | } |
1890 | ||
3fbfbf7a PM |
1891 | /* |
1892 | * Enqueue the specified string of rcu_head structures onto the specified | |
1893 | * CPU's no-CBs lists. The CPU is specified by rdp, the head of the | |
1894 | * string by rhp, and the tail of the string by rhtp. The non-lazy/lazy | |
1895 | * counts are supplied by rhcount and rhcount_lazy. | |
1896 | * | |
1897 | * If warranted, also wake up the kthread servicing this CPUs queues. | |
1898 | */ | |
1899 | static void __call_rcu_nocb_enqueue(struct rcu_data *rdp, | |
1900 | struct rcu_head *rhp, | |
1901 | struct rcu_head **rhtp, | |
96d3fd0d PM |
1902 | int rhcount, int rhcount_lazy, |
1903 | unsigned long flags) | |
3fbfbf7a PM |
1904 | { |
1905 | int len; | |
1906 | struct rcu_head **old_rhpp; | |
1907 | struct task_struct *t; | |
1908 | ||
1909 | /* Enqueue the callback on the nocb list and update counts. */ | |
41050a00 PM |
1910 | atomic_long_add(rhcount, &rdp->nocb_q_count); |
1911 | /* rcu_barrier() relies on ->nocb_q_count add before xchg. */ | |
3fbfbf7a | 1912 | old_rhpp = xchg(&rdp->nocb_tail, rhtp); |
7d0ae808 | 1913 | WRITE_ONCE(*old_rhpp, rhp); |
3fbfbf7a | 1914 | atomic_long_add(rhcount_lazy, &rdp->nocb_q_count_lazy); |
39953dfd | 1915 | smp_mb__after_atomic(); /* Store *old_rhpp before _wake test. */ |
3fbfbf7a PM |
1916 | |
1917 | /* If we are not being polled and there is a kthread, awaken it ... */ | |
7d0ae808 | 1918 | t = READ_ONCE(rdp->nocb_kthread); |
25e03a74 | 1919 | if (rcu_nocb_poll || !t) { |
9261dd0d PM |
1920 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
1921 | TPS("WakeNotPoll")); | |
3fbfbf7a | 1922 | return; |
9261dd0d | 1923 | } |
3fbfbf7a PM |
1924 | len = atomic_long_read(&rdp->nocb_q_count); |
1925 | if (old_rhpp == &rdp->nocb_head) { | |
96d3fd0d | 1926 | if (!irqs_disabled_flags(flags)) { |
fbce7497 PM |
1927 | /* ... if queue was empty ... */ |
1928 | wake_nocb_leader(rdp, false); | |
96d3fd0d PM |
1929 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
1930 | TPS("WakeEmpty")); | |
1931 | } else { | |
511324e4 | 1932 | WRITE_ONCE(rdp->nocb_defer_wakeup, RCU_NOCB_WAKE); |
9226b10d PM |
1933 | /* Store ->nocb_defer_wakeup before ->rcu_urgent_qs. */ |
1934 | smp_store_release(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs), true); | |
96d3fd0d PM |
1935 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
1936 | TPS("WakeEmptyIsDeferred")); | |
1937 | } | |
3fbfbf7a PM |
1938 | rdp->qlen_last_fqs_check = 0; |
1939 | } else if (len > rdp->qlen_last_fqs_check + qhimark) { | |
fbce7497 | 1940 | /* ... or if many callbacks queued. */ |
9fdd3bc9 PM |
1941 | if (!irqs_disabled_flags(flags)) { |
1942 | wake_nocb_leader(rdp, true); | |
1943 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, | |
1944 | TPS("WakeOvf")); | |
1945 | } else { | |
511324e4 | 1946 | WRITE_ONCE(rdp->nocb_defer_wakeup, RCU_NOCB_WAKE_FORCE); |
9226b10d PM |
1947 | /* Store ->nocb_defer_wakeup before ->rcu_urgent_qs. */ |
1948 | smp_store_release(this_cpu_ptr(&rcu_dynticks.rcu_urgent_qs), true); | |
9fdd3bc9 PM |
1949 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
1950 | TPS("WakeOvfIsDeferred")); | |
1951 | } | |
3fbfbf7a | 1952 | rdp->qlen_last_fqs_check = LONG_MAX / 2; |
9261dd0d PM |
1953 | } else { |
1954 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("WakeNot")); | |
3fbfbf7a PM |
1955 | } |
1956 | return; | |
1957 | } | |
1958 | ||
1959 | /* | |
1960 | * This is a helper for __call_rcu(), which invokes this when the normal | |
1961 | * callback queue is inoperable. If this is not a no-CBs CPU, this | |
1962 | * function returns failure back to __call_rcu(), which can complain | |
1963 | * appropriately. | |
1964 | * | |
1965 | * Otherwise, this function queues the callback where the corresponding | |
1966 | * "rcuo" kthread can find it. | |
1967 | */ | |
1968 | static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, | |
96d3fd0d | 1969 | bool lazy, unsigned long flags) |
3fbfbf7a PM |
1970 | { |
1971 | ||
d1e43fa5 | 1972 | if (!rcu_is_nocb_cpu(rdp->cpu)) |
c271d3a9 | 1973 | return false; |
96d3fd0d | 1974 | __call_rcu_nocb_enqueue(rdp, rhp, &rhp->next, 1, lazy, flags); |
21e7a608 PM |
1975 | if (__is_kfree_rcu_offset((unsigned long)rhp->func)) |
1976 | trace_rcu_kfree_callback(rdp->rsp->name, rhp, | |
1977 | (unsigned long)rhp->func, | |
756cbf6b PM |
1978 | -atomic_long_read(&rdp->nocb_q_count_lazy), |
1979 | -atomic_long_read(&rdp->nocb_q_count)); | |
21e7a608 PM |
1980 | else |
1981 | trace_rcu_callback(rdp->rsp->name, rhp, | |
756cbf6b PM |
1982 | -atomic_long_read(&rdp->nocb_q_count_lazy), |
1983 | -atomic_long_read(&rdp->nocb_q_count)); | |
1772947b PM |
1984 | |
1985 | /* | |
1986 | * If called from an extended quiescent state with interrupts | |
1987 | * disabled, invoke the RCU core in order to allow the idle-entry | |
1988 | * deferred-wakeup check to function. | |
1989 | */ | |
1990 | if (irqs_disabled_flags(flags) && | |
1991 | !rcu_is_watching() && | |
1992 | cpu_online(smp_processor_id())) | |
1993 | invoke_rcu_core(); | |
1994 | ||
c271d3a9 | 1995 | return true; |
3fbfbf7a PM |
1996 | } |
1997 | ||
1998 | /* | |
1999 | * Adopt orphaned callbacks on a no-CBs CPU, or return 0 if this is | |
2000 | * not a no-CBs CPU. | |
2001 | */ | |
2002 | static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp, | |
96d3fd0d PM |
2003 | struct rcu_data *rdp, |
2004 | unsigned long flags) | |
3fbfbf7a | 2005 | { |
4b27f20b | 2006 | long ql = rsp->orphan_done.len; |
933dfbd7 | 2007 | long qll = rsp->orphan_done.len_lazy; |
3fbfbf7a PM |
2008 | |
2009 | /* If this is not a no-CBs CPU, tell the caller to do it the old way. */ | |
d1e43fa5 | 2010 | if (!rcu_is_nocb_cpu(smp_processor_id())) |
0a9e1e11 | 2011 | return false; |
3fbfbf7a PM |
2012 | |
2013 | /* First, enqueue the donelist, if any. This preserves CB ordering. */ | |
8ef0f37e | 2014 | if (rsp->orphan_done.head) { |
15fecf89 PM |
2015 | __call_rcu_nocb_enqueue(rdp, rcu_cblist_head(&rsp->orphan_done), |
2016 | rcu_cblist_tail(&rsp->orphan_done), | |
2017 | ql, qll, flags); | |
3fbfbf7a | 2018 | } |
8ef0f37e | 2019 | if (rsp->orphan_pend.head) { |
15fecf89 PM |
2020 | __call_rcu_nocb_enqueue(rdp, rcu_cblist_head(&rsp->orphan_pend), |
2021 | rcu_cblist_tail(&rsp->orphan_pend), | |
2022 | ql, qll, flags); | |
3fbfbf7a | 2023 | } |
15fecf89 PM |
2024 | rcu_cblist_init(&rsp->orphan_done); |
2025 | rcu_cblist_init(&rsp->orphan_pend); | |
0a9e1e11 | 2026 | return true; |
3fbfbf7a PM |
2027 | } |
2028 | ||
2029 | /* | |
34ed6246 PM |
2030 | * If necessary, kick off a new grace period, and either way wait |
2031 | * for a subsequent grace period to complete. | |
3fbfbf7a | 2032 | */ |
34ed6246 | 2033 | static void rcu_nocb_wait_gp(struct rcu_data *rdp) |
3fbfbf7a | 2034 | { |
34ed6246 | 2035 | unsigned long c; |
dae6e64d | 2036 | bool d; |
34ed6246 | 2037 | unsigned long flags; |
48a7639c | 2038 | bool needwake; |
34ed6246 PM |
2039 | struct rcu_node *rnp = rdp->mynode; |
2040 | ||
2a67e741 | 2041 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
48a7639c | 2042 | needwake = rcu_start_future_gp(rnp, rdp, &c); |
67c583a7 | 2043 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
48a7639c PM |
2044 | if (needwake) |
2045 | rcu_gp_kthread_wake(rdp->rsp); | |
3fbfbf7a PM |
2046 | |
2047 | /* | |
34ed6246 PM |
2048 | * Wait for the grace period. Do so interruptibly to avoid messing |
2049 | * up the load average. | |
3fbfbf7a | 2050 | */ |
f7f7bac9 | 2051 | trace_rcu_future_gp(rnp, rdp, c, TPS("StartWait")); |
34ed6246 | 2052 | for (;;) { |
abedf8e2 | 2053 | swait_event_interruptible( |
dae6e64d | 2054 | rnp->nocb_gp_wq[c & 0x1], |
7d0ae808 | 2055 | (d = ULONG_CMP_GE(READ_ONCE(rnp->completed), c))); |
dae6e64d | 2056 | if (likely(d)) |
34ed6246 | 2057 | break; |
73a860cd | 2058 | WARN_ON(signal_pending(current)); |
f7f7bac9 | 2059 | trace_rcu_future_gp(rnp, rdp, c, TPS("ResumeWait")); |
34ed6246 | 2060 | } |
f7f7bac9 | 2061 | trace_rcu_future_gp(rnp, rdp, c, TPS("EndWait")); |
34ed6246 | 2062 | smp_mb(); /* Ensure that CB invocation happens after GP end. */ |
3fbfbf7a PM |
2063 | } |
2064 | ||
fbce7497 PM |
2065 | /* |
2066 | * Leaders come here to wait for additional callbacks to show up. | |
2067 | * This function does not return until callbacks appear. | |
2068 | */ | |
2069 | static void nocb_leader_wait(struct rcu_data *my_rdp) | |
2070 | { | |
2071 | bool firsttime = true; | |
2072 | bool gotcbs; | |
2073 | struct rcu_data *rdp; | |
2074 | struct rcu_head **tail; | |
2075 | ||
2076 | wait_again: | |
2077 | ||
2078 | /* Wait for callbacks to appear. */ | |
2079 | if (!rcu_nocb_poll) { | |
2080 | trace_rcu_nocb_wake(my_rdp->rsp->name, my_rdp->cpu, "Sleep"); | |
abedf8e2 | 2081 | swait_event_interruptible(my_rdp->nocb_wq, |
7d0ae808 | 2082 | !READ_ONCE(my_rdp->nocb_leader_sleep)); |
fbce7497 PM |
2083 | /* Memory barrier handled by smp_mb() calls below and repoll. */ |
2084 | } else if (firsttime) { | |
2085 | firsttime = false; /* Don't drown trace log with "Poll"! */ | |
2086 | trace_rcu_nocb_wake(my_rdp->rsp->name, my_rdp->cpu, "Poll"); | |
2087 | } | |
2088 | ||
2089 | /* | |
2090 | * Each pass through the following loop checks a follower for CBs. | |
2091 | * We are our own first follower. Any CBs found are moved to | |
2092 | * nocb_gp_head, where they await a grace period. | |
2093 | */ | |
2094 | gotcbs = false; | |
6b5fc3a1 | 2095 | smp_mb(); /* wakeup before ->nocb_head reads. */ |
fbce7497 | 2096 | for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) { |
7d0ae808 | 2097 | rdp->nocb_gp_head = READ_ONCE(rdp->nocb_head); |
fbce7497 PM |
2098 | if (!rdp->nocb_gp_head) |
2099 | continue; /* No CBs here, try next follower. */ | |
2100 | ||
2101 | /* Move callbacks to wait-for-GP list, which is empty. */ | |
7d0ae808 | 2102 | WRITE_ONCE(rdp->nocb_head, NULL); |
fbce7497 | 2103 | rdp->nocb_gp_tail = xchg(&rdp->nocb_tail, &rdp->nocb_head); |
fbce7497 PM |
2104 | gotcbs = true; |
2105 | } | |
2106 | ||
2107 | /* | |
2108 | * If there were no callbacks, sleep a bit, rescan after a | |
2109 | * memory barrier, and go retry. | |
2110 | */ | |
2111 | if (unlikely(!gotcbs)) { | |
2112 | if (!rcu_nocb_poll) | |
2113 | trace_rcu_nocb_wake(my_rdp->rsp->name, my_rdp->cpu, | |
2114 | "WokeEmpty"); | |
73a860cd | 2115 | WARN_ON(signal_pending(current)); |
fbce7497 PM |
2116 | schedule_timeout_interruptible(1); |
2117 | ||
2118 | /* Rescan in case we were a victim of memory ordering. */ | |
11ed7f93 PK |
2119 | my_rdp->nocb_leader_sleep = true; |
2120 | smp_mb(); /* Ensure _sleep true before scan. */ | |
fbce7497 | 2121 | for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) |
7d0ae808 | 2122 | if (READ_ONCE(rdp->nocb_head)) { |
fbce7497 | 2123 | /* Found CB, so short-circuit next wait. */ |
11ed7f93 | 2124 | my_rdp->nocb_leader_sleep = false; |
fbce7497 PM |
2125 | break; |
2126 | } | |
2127 | goto wait_again; | |
2128 | } | |
2129 | ||
2130 | /* Wait for one grace period. */ | |
2131 | rcu_nocb_wait_gp(my_rdp); | |
2132 | ||
2133 | /* | |
11ed7f93 PK |
2134 | * We left ->nocb_leader_sleep unset to reduce cache thrashing. |
2135 | * We set it now, but recheck for new callbacks while | |
fbce7497 PM |
2136 | * traversing our follower list. |
2137 | */ | |
11ed7f93 PK |
2138 | my_rdp->nocb_leader_sleep = true; |
2139 | smp_mb(); /* Ensure _sleep true before scan of ->nocb_head. */ | |
fbce7497 PM |
2140 | |
2141 | /* Each pass through the following loop wakes a follower, if needed. */ | |
2142 | for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) { | |
7d0ae808 | 2143 | if (READ_ONCE(rdp->nocb_head)) |
11ed7f93 | 2144 | my_rdp->nocb_leader_sleep = false;/* No need to sleep.*/ |
fbce7497 PM |
2145 | if (!rdp->nocb_gp_head) |
2146 | continue; /* No CBs, so no need to wake follower. */ | |
2147 | ||
2148 | /* Append callbacks to follower's "done" list. */ | |
2149 | tail = xchg(&rdp->nocb_follower_tail, rdp->nocb_gp_tail); | |
2150 | *tail = rdp->nocb_gp_head; | |
c847f142 | 2151 | smp_mb__after_atomic(); /* Store *tail before wakeup. */ |
fbce7497 PM |
2152 | if (rdp != my_rdp && tail == &rdp->nocb_follower_head) { |
2153 | /* | |
2154 | * List was empty, wake up the follower. | |
2155 | * Memory barriers supplied by atomic_long_add(). | |
2156 | */ | |
abedf8e2 | 2157 | swake_up(&rdp->nocb_wq); |
fbce7497 PM |
2158 | } |
2159 | } | |
2160 | ||
2161 | /* If we (the leader) don't have CBs, go wait some more. */ | |
2162 | if (!my_rdp->nocb_follower_head) | |
2163 | goto wait_again; | |
2164 | } | |
2165 | ||
2166 | /* | |
2167 | * Followers come here to wait for additional callbacks to show up. | |
2168 | * This function does not return until callbacks appear. | |
2169 | */ | |
2170 | static void nocb_follower_wait(struct rcu_data *rdp) | |
2171 | { | |
2172 | bool firsttime = true; | |
2173 | ||
2174 | for (;;) { | |
2175 | if (!rcu_nocb_poll) { | |
2176 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, | |
2177 | "FollowerSleep"); | |
abedf8e2 | 2178 | swait_event_interruptible(rdp->nocb_wq, |
7d0ae808 | 2179 | READ_ONCE(rdp->nocb_follower_head)); |
fbce7497 PM |
2180 | } else if (firsttime) { |
2181 | /* Don't drown trace log with "Poll"! */ | |
2182 | firsttime = false; | |
2183 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, "Poll"); | |
2184 | } | |
2185 | if (smp_load_acquire(&rdp->nocb_follower_head)) { | |
2186 | /* ^^^ Ensure CB invocation follows _head test. */ | |
2187 | return; | |
2188 | } | |
2189 | if (!rcu_nocb_poll) | |
2190 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, | |
2191 | "WokeEmpty"); | |
73a860cd | 2192 | WARN_ON(signal_pending(current)); |
fbce7497 PM |
2193 | schedule_timeout_interruptible(1); |
2194 | } | |
2195 | } | |
2196 | ||
3fbfbf7a PM |
2197 | /* |
2198 | * Per-rcu_data kthread, but only for no-CBs CPUs. Each kthread invokes | |
fbce7497 PM |
2199 | * callbacks queued by the corresponding no-CBs CPU, however, there is |
2200 | * an optional leader-follower relationship so that the grace-period | |
2201 | * kthreads don't have to do quite so many wakeups. | |
3fbfbf7a PM |
2202 | */ |
2203 | static int rcu_nocb_kthread(void *arg) | |
2204 | { | |
2205 | int c, cl; | |
2206 | struct rcu_head *list; | |
2207 | struct rcu_head *next; | |
2208 | struct rcu_head **tail; | |
2209 | struct rcu_data *rdp = arg; | |
2210 | ||
2211 | /* Each pass through this loop invokes one batch of callbacks */ | |
2212 | for (;;) { | |
fbce7497 PM |
2213 | /* Wait for callbacks. */ |
2214 | if (rdp->nocb_leader == rdp) | |
2215 | nocb_leader_wait(rdp); | |
2216 | else | |
2217 | nocb_follower_wait(rdp); | |
2218 | ||
2219 | /* Pull the ready-to-invoke callbacks onto local list. */ | |
7d0ae808 | 2220 | list = READ_ONCE(rdp->nocb_follower_head); |
fbce7497 PM |
2221 | BUG_ON(!list); |
2222 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, "WokeNonEmpty"); | |
7d0ae808 | 2223 | WRITE_ONCE(rdp->nocb_follower_head, NULL); |
fbce7497 | 2224 | tail = xchg(&rdp->nocb_follower_tail, &rdp->nocb_follower_head); |
3fbfbf7a PM |
2225 | |
2226 | /* Each pass through the following loop invokes a callback. */ | |
41050a00 PM |
2227 | trace_rcu_batch_start(rdp->rsp->name, |
2228 | atomic_long_read(&rdp->nocb_q_count_lazy), | |
2229 | atomic_long_read(&rdp->nocb_q_count), -1); | |
3fbfbf7a PM |
2230 | c = cl = 0; |
2231 | while (list) { | |
2232 | next = list->next; | |
2233 | /* Wait for enqueuing to complete, if needed. */ | |
2234 | while (next == NULL && &list->next != tail) { | |
69a79bb1 PM |
2235 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
2236 | TPS("WaitQueue")); | |
3fbfbf7a | 2237 | schedule_timeout_interruptible(1); |
69a79bb1 PM |
2238 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
2239 | TPS("WokeQueue")); | |
3fbfbf7a PM |
2240 | next = list->next; |
2241 | } | |
2242 | debug_rcu_head_unqueue(list); | |
2243 | local_bh_disable(); | |
2244 | if (__rcu_reclaim(rdp->rsp->name, list)) | |
2245 | cl++; | |
2246 | c++; | |
2247 | local_bh_enable(); | |
bedc1969 | 2248 | cond_resched_rcu_qs(); |
3fbfbf7a PM |
2249 | list = next; |
2250 | } | |
2251 | trace_rcu_batch_end(rdp->rsp->name, c, !!list, 0, 0, 1); | |
41050a00 PM |
2252 | smp_mb__before_atomic(); /* _add after CB invocation. */ |
2253 | atomic_long_add(-c, &rdp->nocb_q_count); | |
2254 | atomic_long_add(-cl, &rdp->nocb_q_count_lazy); | |
c635a4e1 | 2255 | rdp->n_nocbs_invoked += c; |
3fbfbf7a PM |
2256 | } |
2257 | return 0; | |
2258 | } | |
2259 | ||
96d3fd0d | 2260 | /* Is a deferred wakeup of rcu_nocb_kthread() required? */ |
9fdd3bc9 | 2261 | static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp) |
96d3fd0d | 2262 | { |
7d0ae808 | 2263 | return READ_ONCE(rdp->nocb_defer_wakeup); |
96d3fd0d PM |
2264 | } |
2265 | ||
2266 | /* Do a deferred wakeup of rcu_nocb_kthread(). */ | |
2267 | static void do_nocb_deferred_wakeup(struct rcu_data *rdp) | |
2268 | { | |
9fdd3bc9 PM |
2269 | int ndw; |
2270 | ||
96d3fd0d PM |
2271 | if (!rcu_nocb_need_deferred_wakeup(rdp)) |
2272 | return; | |
7d0ae808 | 2273 | ndw = READ_ONCE(rdp->nocb_defer_wakeup); |
511324e4 PM |
2274 | WRITE_ONCE(rdp->nocb_defer_wakeup, RCU_NOCB_WAKE_NOT); |
2275 | wake_nocb_leader(rdp, ndw == RCU_NOCB_WAKE_FORCE); | |
9fdd3bc9 | 2276 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("DeferredWake")); |
96d3fd0d PM |
2277 | } |
2278 | ||
f4579fc5 PM |
2279 | void __init rcu_init_nohz(void) |
2280 | { | |
2281 | int cpu; | |
2282 | bool need_rcu_nocb_mask = true; | |
2283 | struct rcu_state *rsp; | |
2284 | ||
2285 | #ifdef CONFIG_RCU_NOCB_CPU_NONE | |
2286 | need_rcu_nocb_mask = false; | |
2287 | #endif /* #ifndef CONFIG_RCU_NOCB_CPU_NONE */ | |
2288 | ||
2289 | #if defined(CONFIG_NO_HZ_FULL) | |
2290 | if (tick_nohz_full_running && cpumask_weight(tick_nohz_full_mask)) | |
2291 | need_rcu_nocb_mask = true; | |
2292 | #endif /* #if defined(CONFIG_NO_HZ_FULL) */ | |
2293 | ||
2294 | if (!have_rcu_nocb_mask && need_rcu_nocb_mask) { | |
949cccdb PK |
2295 | if (!zalloc_cpumask_var(&rcu_nocb_mask, GFP_KERNEL)) { |
2296 | pr_info("rcu_nocb_mask allocation failed, callback offloading disabled.\n"); | |
2297 | return; | |
2298 | } | |
f4579fc5 PM |
2299 | have_rcu_nocb_mask = true; |
2300 | } | |
2301 | if (!have_rcu_nocb_mask) | |
2302 | return; | |
2303 | ||
2304 | #ifdef CONFIG_RCU_NOCB_CPU_ZERO | |
2305 | pr_info("\tOffload RCU callbacks from CPU 0\n"); | |
2306 | cpumask_set_cpu(0, rcu_nocb_mask); | |
2307 | #endif /* #ifdef CONFIG_RCU_NOCB_CPU_ZERO */ | |
2308 | #ifdef CONFIG_RCU_NOCB_CPU_ALL | |
2309 | pr_info("\tOffload RCU callbacks from all CPUs\n"); | |
2310 | cpumask_copy(rcu_nocb_mask, cpu_possible_mask); | |
2311 | #endif /* #ifdef CONFIG_RCU_NOCB_CPU_ALL */ | |
2312 | #if defined(CONFIG_NO_HZ_FULL) | |
2313 | if (tick_nohz_full_running) | |
2314 | cpumask_or(rcu_nocb_mask, rcu_nocb_mask, tick_nohz_full_mask); | |
2315 | #endif /* #if defined(CONFIG_NO_HZ_FULL) */ | |
2316 | ||
2317 | if (!cpumask_subset(rcu_nocb_mask, cpu_possible_mask)) { | |
2318 | pr_info("\tNote: kernel parameter 'rcu_nocbs=' contains nonexistent CPUs.\n"); | |
2319 | cpumask_and(rcu_nocb_mask, cpu_possible_mask, | |
2320 | rcu_nocb_mask); | |
2321 | } | |
ad853b48 TH |
2322 | pr_info("\tOffload RCU callbacks from CPUs: %*pbl.\n", |
2323 | cpumask_pr_args(rcu_nocb_mask)); | |
f4579fc5 PM |
2324 | if (rcu_nocb_poll) |
2325 | pr_info("\tPoll for callbacks from no-CBs CPUs.\n"); | |
2326 | ||
2327 | for_each_rcu_flavor(rsp) { | |
34404ca8 PM |
2328 | for_each_cpu(cpu, rcu_nocb_mask) |
2329 | init_nocb_callback_list(per_cpu_ptr(rsp->rda, cpu)); | |
35ce7f29 | 2330 | rcu_organize_nocb_kthreads(rsp); |
f4579fc5 | 2331 | } |
96d3fd0d PM |
2332 | } |
2333 | ||
3fbfbf7a PM |
2334 | /* Initialize per-rcu_data variables for no-CBs CPUs. */ |
2335 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) | |
2336 | { | |
2337 | rdp->nocb_tail = &rdp->nocb_head; | |
abedf8e2 | 2338 | init_swait_queue_head(&rdp->nocb_wq); |
fbce7497 | 2339 | rdp->nocb_follower_tail = &rdp->nocb_follower_head; |
3fbfbf7a PM |
2340 | } |
2341 | ||
35ce7f29 PM |
2342 | /* |
2343 | * If the specified CPU is a no-CBs CPU that does not already have its | |
2344 | * rcuo kthread for the specified RCU flavor, spawn it. If the CPUs are | |
2345 | * brought online out of order, this can require re-organizing the | |
2346 | * leader-follower relationships. | |
2347 | */ | |
2348 | static void rcu_spawn_one_nocb_kthread(struct rcu_state *rsp, int cpu) | |
2349 | { | |
2350 | struct rcu_data *rdp; | |
2351 | struct rcu_data *rdp_last; | |
2352 | struct rcu_data *rdp_old_leader; | |
2353 | struct rcu_data *rdp_spawn = per_cpu_ptr(rsp->rda, cpu); | |
2354 | struct task_struct *t; | |
2355 | ||
2356 | /* | |
2357 | * If this isn't a no-CBs CPU or if it already has an rcuo kthread, | |
2358 | * then nothing to do. | |
2359 | */ | |
2360 | if (!rcu_is_nocb_cpu(cpu) || rdp_spawn->nocb_kthread) | |
2361 | return; | |
2362 | ||
2363 | /* If we didn't spawn the leader first, reorganize! */ | |
2364 | rdp_old_leader = rdp_spawn->nocb_leader; | |
2365 | if (rdp_old_leader != rdp_spawn && !rdp_old_leader->nocb_kthread) { | |
2366 | rdp_last = NULL; | |
2367 | rdp = rdp_old_leader; | |
2368 | do { | |
2369 | rdp->nocb_leader = rdp_spawn; | |
2370 | if (rdp_last && rdp != rdp_spawn) | |
2371 | rdp_last->nocb_next_follower = rdp; | |
bbe5d7a9 PM |
2372 | if (rdp == rdp_spawn) { |
2373 | rdp = rdp->nocb_next_follower; | |
2374 | } else { | |
2375 | rdp_last = rdp; | |
2376 | rdp = rdp->nocb_next_follower; | |
2377 | rdp_last->nocb_next_follower = NULL; | |
2378 | } | |
35ce7f29 PM |
2379 | } while (rdp); |
2380 | rdp_spawn->nocb_next_follower = rdp_old_leader; | |
2381 | } | |
2382 | ||
2383 | /* Spawn the kthread for this CPU and RCU flavor. */ | |
2384 | t = kthread_run(rcu_nocb_kthread, rdp_spawn, | |
2385 | "rcuo%c/%d", rsp->abbr, cpu); | |
2386 | BUG_ON(IS_ERR(t)); | |
7d0ae808 | 2387 | WRITE_ONCE(rdp_spawn->nocb_kthread, t); |
35ce7f29 PM |
2388 | } |
2389 | ||
2390 | /* | |
2391 | * If the specified CPU is a no-CBs CPU that does not already have its | |
2392 | * rcuo kthreads, spawn them. | |
2393 | */ | |
2394 | static void rcu_spawn_all_nocb_kthreads(int cpu) | |
2395 | { | |
2396 | struct rcu_state *rsp; | |
2397 | ||
2398 | if (rcu_scheduler_fully_active) | |
2399 | for_each_rcu_flavor(rsp) | |
2400 | rcu_spawn_one_nocb_kthread(rsp, cpu); | |
2401 | } | |
2402 | ||
2403 | /* | |
2404 | * Once the scheduler is running, spawn rcuo kthreads for all online | |
2405 | * no-CBs CPUs. This assumes that the early_initcall()s happen before | |
2406 | * non-boot CPUs come online -- if this changes, we will need to add | |
2407 | * some mutual exclusion. | |
2408 | */ | |
2409 | static void __init rcu_spawn_nocb_kthreads(void) | |
2410 | { | |
2411 | int cpu; | |
2412 | ||
2413 | for_each_online_cpu(cpu) | |
2414 | rcu_spawn_all_nocb_kthreads(cpu); | |
2415 | } | |
2416 | ||
fbce7497 PM |
2417 | /* How many follower CPU IDs per leader? Default of -1 for sqrt(nr_cpu_ids). */ |
2418 | static int rcu_nocb_leader_stride = -1; | |
2419 | module_param(rcu_nocb_leader_stride, int, 0444); | |
2420 | ||
2421 | /* | |
35ce7f29 | 2422 | * Initialize leader-follower relationships for all no-CBs CPU. |
fbce7497 | 2423 | */ |
35ce7f29 | 2424 | static void __init rcu_organize_nocb_kthreads(struct rcu_state *rsp) |
3fbfbf7a PM |
2425 | { |
2426 | int cpu; | |
fbce7497 PM |
2427 | int ls = rcu_nocb_leader_stride; |
2428 | int nl = 0; /* Next leader. */ | |
3fbfbf7a | 2429 | struct rcu_data *rdp; |
fbce7497 PM |
2430 | struct rcu_data *rdp_leader = NULL; /* Suppress misguided gcc warn. */ |
2431 | struct rcu_data *rdp_prev = NULL; | |
3fbfbf7a | 2432 | |
22c2f669 | 2433 | if (!have_rcu_nocb_mask) |
3fbfbf7a | 2434 | return; |
fbce7497 PM |
2435 | if (ls == -1) { |
2436 | ls = int_sqrt(nr_cpu_ids); | |
2437 | rcu_nocb_leader_stride = ls; | |
2438 | } | |
2439 | ||
2440 | /* | |
9831ce3b PM |
2441 | * Each pass through this loop sets up one rcu_data structure. |
2442 | * Should the corresponding CPU come online in the future, then | |
2443 | * we will spawn the needed set of rcu_nocb_kthread() kthreads. | |
fbce7497 | 2444 | */ |
3fbfbf7a PM |
2445 | for_each_cpu(cpu, rcu_nocb_mask) { |
2446 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
fbce7497 PM |
2447 | if (rdp->cpu >= nl) { |
2448 | /* New leader, set up for followers & next leader. */ | |
2449 | nl = DIV_ROUND_UP(rdp->cpu + 1, ls) * ls; | |
2450 | rdp->nocb_leader = rdp; | |
2451 | rdp_leader = rdp; | |
2452 | } else { | |
2453 | /* Another follower, link to previous leader. */ | |
2454 | rdp->nocb_leader = rdp_leader; | |
2455 | rdp_prev->nocb_next_follower = rdp; | |
2456 | } | |
2457 | rdp_prev = rdp; | |
3fbfbf7a PM |
2458 | } |
2459 | } | |
2460 | ||
2461 | /* Prevent __call_rcu() from enqueuing callbacks on no-CBs CPUs */ | |
34ed6246 | 2462 | static bool init_nocb_callback_list(struct rcu_data *rdp) |
3fbfbf7a | 2463 | { |
22c2f669 | 2464 | if (!rcu_is_nocb_cpu(rdp->cpu)) |
34ed6246 | 2465 | return false; |
22c2f669 | 2466 | |
34404ca8 | 2467 | /* If there are early-boot callbacks, move them to nocb lists. */ |
15fecf89 PM |
2468 | if (!rcu_segcblist_empty(&rdp->cblist)) { |
2469 | rdp->nocb_head = rcu_segcblist_head(&rdp->cblist); | |
2470 | rdp->nocb_tail = rcu_segcblist_tail(&rdp->cblist); | |
2471 | atomic_long_set(&rdp->nocb_q_count, | |
2472 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
2473 | atomic_long_set(&rdp->nocb_q_count_lazy, | |
2474 | rcu_segcblist_n_lazy_cbs(&rdp->cblist)); | |
2475 | rcu_segcblist_init(&rdp->cblist); | |
34404ca8 | 2476 | } |
15fecf89 | 2477 | rcu_segcblist_disable(&rdp->cblist); |
34ed6246 | 2478 | return true; |
3fbfbf7a PM |
2479 | } |
2480 | ||
34ed6246 PM |
2481 | #else /* #ifdef CONFIG_RCU_NOCB_CPU */ |
2482 | ||
d7e29933 PM |
2483 | static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu) |
2484 | { | |
2485 | WARN_ON_ONCE(1); /* Should be dead code. */ | |
2486 | return false; | |
2487 | } | |
2488 | ||
abedf8e2 | 2489 | static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq) |
3fbfbf7a | 2490 | { |
3fbfbf7a PM |
2491 | } |
2492 | ||
dae6e64d PM |
2493 | static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq) |
2494 | { | |
2495 | } | |
2496 | ||
abedf8e2 | 2497 | static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp) |
065bb78c DW |
2498 | { |
2499 | return NULL; | |
2500 | } | |
2501 | ||
dae6e64d PM |
2502 | static void rcu_init_one_nocb(struct rcu_node *rnp) |
2503 | { | |
2504 | } | |
3fbfbf7a | 2505 | |
3fbfbf7a | 2506 | static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, |
96d3fd0d | 2507 | bool lazy, unsigned long flags) |
3fbfbf7a | 2508 | { |
4afc7e26 | 2509 | return false; |
3fbfbf7a PM |
2510 | } |
2511 | ||
2512 | static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp, | |
96d3fd0d PM |
2513 | struct rcu_data *rdp, |
2514 | unsigned long flags) | |
3fbfbf7a | 2515 | { |
f4aa84ba | 2516 | return false; |
3fbfbf7a PM |
2517 | } |
2518 | ||
3fbfbf7a PM |
2519 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) |
2520 | { | |
2521 | } | |
2522 | ||
9fdd3bc9 | 2523 | static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp) |
96d3fd0d PM |
2524 | { |
2525 | return false; | |
2526 | } | |
2527 | ||
2528 | static void do_nocb_deferred_wakeup(struct rcu_data *rdp) | |
2529 | { | |
2530 | } | |
2531 | ||
35ce7f29 PM |
2532 | static void rcu_spawn_all_nocb_kthreads(int cpu) |
2533 | { | |
2534 | } | |
2535 | ||
2536 | static void __init rcu_spawn_nocb_kthreads(void) | |
3fbfbf7a PM |
2537 | { |
2538 | } | |
2539 | ||
34ed6246 | 2540 | static bool init_nocb_callback_list(struct rcu_data *rdp) |
3fbfbf7a | 2541 | { |
34ed6246 | 2542 | return false; |
3fbfbf7a PM |
2543 | } |
2544 | ||
2545 | #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ | |
65d798f0 PM |
2546 | |
2547 | /* | |
2548 | * An adaptive-ticks CPU can potentially execute in kernel mode for an | |
2549 | * arbitrarily long period of time with the scheduling-clock tick turned | |
2550 | * off. RCU will be paying attention to this CPU because it is in the | |
2551 | * kernel, but the CPU cannot be guaranteed to be executing the RCU state | |
2552 | * machine because the scheduling-clock tick has been disabled. Therefore, | |
2553 | * if an adaptive-ticks CPU is failing to respond to the current grace | |
2554 | * period and has not be idle from an RCU perspective, kick it. | |
2555 | */ | |
4a81e832 | 2556 | static void __maybe_unused rcu_kick_nohz_cpu(int cpu) |
65d798f0 PM |
2557 | { |
2558 | #ifdef CONFIG_NO_HZ_FULL | |
2559 | if (tick_nohz_full_cpu(cpu)) | |
2560 | smp_send_reschedule(cpu); | |
2561 | #endif /* #ifdef CONFIG_NO_HZ_FULL */ | |
2562 | } | |
2333210b | 2563 | |
a096932f PM |
2564 | /* |
2565 | * Is this CPU a NO_HZ_FULL CPU that should ignore RCU so that the | |
2566 | * grace-period kthread will do force_quiescent_state() processing? | |
2567 | * The idea is to avoid waking up RCU core processing on such a | |
2568 | * CPU unless the grace period has extended for too long. | |
2569 | * | |
2570 | * This code relies on the fact that all NO_HZ_FULL CPUs are also | |
52e2bb95 | 2571 | * CONFIG_RCU_NOCB_CPU CPUs. |
a096932f PM |
2572 | */ |
2573 | static bool rcu_nohz_full_cpu(struct rcu_state *rsp) | |
2574 | { | |
2575 | #ifdef CONFIG_NO_HZ_FULL | |
2576 | if (tick_nohz_full_cpu(smp_processor_id()) && | |
2577 | (!rcu_gp_in_progress(rsp) || | |
7d0ae808 | 2578 | ULONG_CMP_LT(jiffies, READ_ONCE(rsp->gp_start) + HZ))) |
5ce035fb | 2579 | return true; |
a096932f | 2580 | #endif /* #ifdef CONFIG_NO_HZ_FULL */ |
5ce035fb | 2581 | return false; |
a096932f | 2582 | } |
5057f55e PM |
2583 | |
2584 | /* | |
2585 | * Bind the grace-period kthread for the sysidle flavor of RCU to the | |
2586 | * timekeeping CPU. | |
2587 | */ | |
2588 | static void rcu_bind_gp_kthread(void) | |
2589 | { | |
c0f489d2 | 2590 | int __maybe_unused cpu; |
5057f55e | 2591 | |
c0f489d2 | 2592 | if (!tick_nohz_full_enabled()) |
5057f55e | 2593 | return; |
5871968d | 2594 | housekeeping_affine(current); |
5057f55e | 2595 | } |
176f8f7a PM |
2596 | |
2597 | /* Record the current task on dyntick-idle entry. */ | |
2598 | static void rcu_dynticks_task_enter(void) | |
2599 | { | |
2600 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 2601 | WRITE_ONCE(current->rcu_tasks_idle_cpu, smp_processor_id()); |
176f8f7a PM |
2602 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
2603 | } | |
2604 | ||
2605 | /* Record no current task on dyntick-idle exit. */ | |
2606 | static void rcu_dynticks_task_exit(void) | |
2607 | { | |
2608 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 2609 | WRITE_ONCE(current->rcu_tasks_idle_cpu, -1); |
176f8f7a PM |
2610 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
2611 | } |