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> |
78634061 | 32 | #include <linux/sched/isolation.h> |
ae7e81c0 | 33 | #include <uapi/linux/sched/types.h> |
4102adab | 34 | #include "../time/tick-internal.h" |
f41d911f | 35 | |
5b61b0ba | 36 | #ifdef CONFIG_RCU_BOOST |
61cfd097 | 37 | |
abaa93d9 | 38 | #include "../locking/rtmutex_common.h" |
21871d7e | 39 | |
61cfd097 PM |
40 | /* |
41 | * Control variables for per-CPU and per-rcu_node kthreads. These | |
42 | * handle all flavors of RCU. | |
43 | */ | |
44 | static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task); | |
45 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status); | |
46 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops); | |
47 | DEFINE_PER_CPU(char, rcu_cpu_has_work); | |
48 | ||
727b705b PM |
49 | #else /* #ifdef CONFIG_RCU_BOOST */ |
50 | ||
51 | /* | |
52 | * Some architectures do not define rt_mutexes, but if !CONFIG_RCU_BOOST, | |
53 | * all uses are in dead code. Provide a definition to keep the compiler | |
54 | * happy, but add WARN_ON_ONCE() to complain if used in the wrong place. | |
55 | * This probably needs to be excluded from -rt builds. | |
56 | */ | |
57 | #define rt_mutex_owner(a) ({ WARN_ON_ONCE(1); NULL; }) | |
b8869781 | 58 | #define rt_mutex_futex_unlock(x) WARN_ON_ONCE(1) |
727b705b PM |
59 | |
60 | #endif /* #else #ifdef CONFIG_RCU_BOOST */ | |
5b61b0ba | 61 | |
3fbfbf7a PM |
62 | #ifdef CONFIG_RCU_NOCB_CPU |
63 | static cpumask_var_t rcu_nocb_mask; /* CPUs to have callbacks offloaded. */ | |
64 | static bool have_rcu_nocb_mask; /* Was rcu_nocb_mask allocated? */ | |
1b0048a4 | 65 | static bool __read_mostly rcu_nocb_poll; /* Offload kthread are to poll. */ |
3fbfbf7a PM |
66 | #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ |
67 | ||
26845c28 PM |
68 | /* |
69 | * Check the RCU kernel configuration parameters and print informative | |
699d4035 | 70 | * messages about anything out of the ordinary. |
26845c28 PM |
71 | */ |
72 | static void __init rcu_bootup_announce_oddness(void) | |
73 | { | |
ab6f5bd6 | 74 | if (IS_ENABLED(CONFIG_RCU_TRACE)) |
ae91aa0a | 75 | pr_info("\tRCU event tracing is enabled.\n"); |
05c5df31 PM |
76 | if ((IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 64) || |
77 | (!IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 32)) | |
ab6f5bd6 | 78 | pr_info("\tCONFIG_RCU_FANOUT set to non-default value of %d\n", |
05c5df31 | 79 | RCU_FANOUT); |
7fa27001 | 80 | if (rcu_fanout_exact) |
ab6f5bd6 PM |
81 | pr_info("\tHierarchical RCU autobalancing is disabled.\n"); |
82 | if (IS_ENABLED(CONFIG_RCU_FAST_NO_HZ)) | |
83 | pr_info("\tRCU dyntick-idle grace-period acceleration is enabled.\n"); | |
c4a09ff7 | 84 | if (IS_ENABLED(CONFIG_PROVE_RCU)) |
ab6f5bd6 | 85 | pr_info("\tRCU lockdep checking is enabled.\n"); |
42621697 AG |
86 | if (RCU_NUM_LVLS >= 4) |
87 | pr_info("\tFour(or more)-level hierarchy is enabled.\n"); | |
47d631af | 88 | if (RCU_FANOUT_LEAF != 16) |
a3bd2c09 | 89 | pr_info("\tBuild-time adjustment of leaf fanout to %d.\n", |
47d631af PM |
90 | RCU_FANOUT_LEAF); |
91 | if (rcu_fanout_leaf != RCU_FANOUT_LEAF) | |
9a5739d7 | 92 | pr_info("\tBoot-time adjustment of leaf fanout to %d.\n", rcu_fanout_leaf); |
cca6f393 | 93 | if (nr_cpu_ids != NR_CPUS) |
9b130ad5 | 94 | pr_info("\tRCU restricting CPUs from NR_CPUS=%d to nr_cpu_ids=%u.\n", NR_CPUS, nr_cpu_ids); |
17c7798b PM |
95 | #ifdef CONFIG_RCU_BOOST |
96 | pr_info("\tRCU priority boosting: priority %d delay %d ms.\n", kthread_prio, CONFIG_RCU_BOOST_DELAY); | |
97 | #endif | |
98 | if (blimit != DEFAULT_RCU_BLIMIT) | |
99 | pr_info("\tBoot-time adjustment of callback invocation limit to %ld.\n", blimit); | |
100 | if (qhimark != DEFAULT_RCU_QHIMARK) | |
101 | pr_info("\tBoot-time adjustment of callback high-water mark to %ld.\n", qhimark); | |
102 | if (qlowmark != DEFAULT_RCU_QLOMARK) | |
103 | pr_info("\tBoot-time adjustment of callback low-water mark to %ld.\n", qlowmark); | |
104 | if (jiffies_till_first_fqs != ULONG_MAX) | |
105 | pr_info("\tBoot-time adjustment of first FQS scan delay to %ld jiffies.\n", jiffies_till_first_fqs); | |
106 | if (jiffies_till_next_fqs != ULONG_MAX) | |
107 | pr_info("\tBoot-time adjustment of subsequent FQS scan delay to %ld jiffies.\n", jiffies_till_next_fqs); | |
108 | if (rcu_kick_kthreads) | |
109 | pr_info("\tKick kthreads if too-long grace period.\n"); | |
110 | if (IS_ENABLED(CONFIG_DEBUG_OBJECTS_RCU_HEAD)) | |
111 | pr_info("\tRCU callback double-/use-after-free debug enabled.\n"); | |
90040c9e | 112 | if (gp_preinit_delay) |
17c7798b | 113 | pr_info("\tRCU debug GP pre-init slowdown %d jiffies.\n", gp_preinit_delay); |
90040c9e | 114 | if (gp_init_delay) |
17c7798b | 115 | pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_init_delay); |
90040c9e | 116 | if (gp_cleanup_delay) |
17c7798b PM |
117 | pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_cleanup_delay); |
118 | if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG)) | |
119 | pr_info("\tRCU debug extended QS entry/exit.\n"); | |
59d80fd8 | 120 | rcupdate_announce_bootup_oddness(); |
26845c28 PM |
121 | } |
122 | ||
28f6569a | 123 | #ifdef CONFIG_PREEMPT_RCU |
f41d911f | 124 | |
a41bfeb2 | 125 | RCU_STATE_INITIALIZER(rcu_preempt, 'p', call_rcu); |
b28a7c01 | 126 | static struct rcu_state *const rcu_state_p = &rcu_preempt_state; |
2927a689 | 127 | static struct rcu_data __percpu *const rcu_data_p = &rcu_preempt_data; |
f41d911f | 128 | |
d19fb8d1 PM |
129 | static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp, |
130 | bool wake); | |
d9a3da06 | 131 | |
f41d911f PM |
132 | /* |
133 | * Tell them what RCU they are running. | |
134 | */ | |
0e0fc1c2 | 135 | static void __init rcu_bootup_announce(void) |
f41d911f | 136 | { |
efc151c3 | 137 | pr_info("Preemptible hierarchical RCU implementation.\n"); |
26845c28 | 138 | rcu_bootup_announce_oddness(); |
f41d911f PM |
139 | } |
140 | ||
8203d6d0 PM |
141 | /* Flags for rcu_preempt_ctxt_queue() decision table. */ |
142 | #define RCU_GP_TASKS 0x8 | |
143 | #define RCU_EXP_TASKS 0x4 | |
144 | #define RCU_GP_BLKD 0x2 | |
145 | #define RCU_EXP_BLKD 0x1 | |
146 | ||
147 | /* | |
148 | * Queues a task preempted within an RCU-preempt read-side critical | |
149 | * section into the appropriate location within the ->blkd_tasks list, | |
150 | * depending on the states of any ongoing normal and expedited grace | |
151 | * periods. The ->gp_tasks pointer indicates which element the normal | |
152 | * grace period is waiting on (NULL if none), and the ->exp_tasks pointer | |
153 | * indicates which element the expedited grace period is waiting on (again, | |
154 | * NULL if none). If a grace period is waiting on a given element in the | |
155 | * ->blkd_tasks list, it also waits on all subsequent elements. Thus, | |
156 | * adding a task to the tail of the list blocks any grace period that is | |
157 | * already waiting on one of the elements. In contrast, adding a task | |
158 | * to the head of the list won't block any grace period that is already | |
159 | * waiting on one of the elements. | |
160 | * | |
161 | * This queuing is imprecise, and can sometimes make an ongoing grace | |
162 | * period wait for a task that is not strictly speaking blocking it. | |
163 | * Given the choice, we needlessly block a normal grace period rather than | |
164 | * blocking an expedited grace period. | |
165 | * | |
166 | * Note that an endless sequence of expedited grace periods still cannot | |
167 | * indefinitely postpone a normal grace period. Eventually, all of the | |
168 | * fixed number of preempted tasks blocking the normal grace period that are | |
169 | * not also blocking the expedited grace period will resume and complete | |
170 | * their RCU read-side critical sections. At that point, the ->gp_tasks | |
171 | * pointer will equal the ->exp_tasks pointer, at which point the end of | |
172 | * the corresponding expedited grace period will also be the end of the | |
173 | * normal grace period. | |
174 | */ | |
46a5d164 PM |
175 | static void rcu_preempt_ctxt_queue(struct rcu_node *rnp, struct rcu_data *rdp) |
176 | __releases(rnp->lock) /* But leaves rrupts disabled. */ | |
8203d6d0 PM |
177 | { |
178 | int blkd_state = (rnp->gp_tasks ? RCU_GP_TASKS : 0) + | |
179 | (rnp->exp_tasks ? RCU_EXP_TASKS : 0) + | |
180 | (rnp->qsmask & rdp->grpmask ? RCU_GP_BLKD : 0) + | |
181 | (rnp->expmask & rdp->grpmask ? RCU_EXP_BLKD : 0); | |
182 | struct task_struct *t = current; | |
183 | ||
ea9b0c8a | 184 | lockdep_assert_held(&rnp->lock); |
2dee9404 PM |
185 | WARN_ON_ONCE(rdp->mynode != rnp); |
186 | WARN_ON_ONCE(rnp->level != rcu_num_lvls - 1); | |
ea9b0c8a | 187 | |
8203d6d0 PM |
188 | /* |
189 | * Decide where to queue the newly blocked task. In theory, | |
190 | * this could be an if-statement. In practice, when I tried | |
191 | * that, it was quite messy. | |
192 | */ | |
193 | switch (blkd_state) { | |
194 | case 0: | |
195 | case RCU_EXP_TASKS: | |
196 | case RCU_EXP_TASKS + RCU_GP_BLKD: | |
197 | case RCU_GP_TASKS: | |
198 | case RCU_GP_TASKS + RCU_EXP_TASKS: | |
199 | ||
200 | /* | |
201 | * Blocking neither GP, or first task blocking the normal | |
202 | * GP but not blocking the already-waiting expedited GP. | |
203 | * Queue at the head of the list to avoid unnecessarily | |
204 | * blocking the already-waiting GPs. | |
205 | */ | |
206 | list_add(&t->rcu_node_entry, &rnp->blkd_tasks); | |
207 | break; | |
208 | ||
209 | case RCU_EXP_BLKD: | |
210 | case RCU_GP_BLKD: | |
211 | case RCU_GP_BLKD + RCU_EXP_BLKD: | |
212 | case RCU_GP_TASKS + RCU_EXP_BLKD: | |
213 | case RCU_GP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
214 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
215 | ||
216 | /* | |
217 | * First task arriving that blocks either GP, or first task | |
218 | * arriving that blocks the expedited GP (with the normal | |
219 | * GP already waiting), or a task arriving that blocks | |
220 | * both GPs with both GPs already waiting. Queue at the | |
221 | * tail of the list to avoid any GP waiting on any of the | |
222 | * already queued tasks that are not blocking it. | |
223 | */ | |
224 | list_add_tail(&t->rcu_node_entry, &rnp->blkd_tasks); | |
225 | break; | |
226 | ||
227 | case RCU_EXP_TASKS + RCU_EXP_BLKD: | |
228 | case RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
229 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_EXP_BLKD: | |
230 | ||
231 | /* | |
232 | * Second or subsequent task blocking the expedited GP. | |
233 | * The task either does not block the normal GP, or is the | |
234 | * first task blocking the normal GP. Queue just after | |
235 | * the first task blocking the expedited GP. | |
236 | */ | |
237 | list_add(&t->rcu_node_entry, rnp->exp_tasks); | |
238 | break; | |
239 | ||
240 | case RCU_GP_TASKS + RCU_GP_BLKD: | |
241 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD: | |
242 | ||
243 | /* | |
244 | * Second or subsequent task blocking the normal GP. | |
245 | * The task does not block the expedited GP. Queue just | |
246 | * after the first task blocking the normal GP. | |
247 | */ | |
248 | list_add(&t->rcu_node_entry, rnp->gp_tasks); | |
249 | break; | |
250 | ||
251 | default: | |
252 | ||
253 | /* Yet another exercise in excessive paranoia. */ | |
254 | WARN_ON_ONCE(1); | |
255 | break; | |
256 | } | |
257 | ||
258 | /* | |
259 | * We have now queued the task. If it was the first one to | |
260 | * block either grace period, update the ->gp_tasks and/or | |
261 | * ->exp_tasks pointers, respectively, to reference the newly | |
262 | * blocked tasks. | |
263 | */ | |
264 | if (!rnp->gp_tasks && (blkd_state & RCU_GP_BLKD)) | |
265 | rnp->gp_tasks = &t->rcu_node_entry; | |
266 | if (!rnp->exp_tasks && (blkd_state & RCU_EXP_BLKD)) | |
267 | rnp->exp_tasks = &t->rcu_node_entry; | |
2dee9404 PM |
268 | WARN_ON_ONCE(!(blkd_state & RCU_GP_BLKD) != |
269 | !(rnp->qsmask & rdp->grpmask)); | |
270 | WARN_ON_ONCE(!(blkd_state & RCU_EXP_BLKD) != | |
271 | !(rnp->expmask & rdp->grpmask)); | |
67c583a7 | 272 | raw_spin_unlock_rcu_node(rnp); /* interrupts remain disabled. */ |
8203d6d0 PM |
273 | |
274 | /* | |
275 | * Report the quiescent state for the expedited GP. This expedited | |
276 | * GP should not be able to end until we report, so there should be | |
277 | * no need to check for a subsequent expedited GP. (Though we are | |
278 | * still in a quiescent state in any case.) | |
279 | */ | |
280 | if (blkd_state & RCU_EXP_BLKD && | |
281 | t->rcu_read_unlock_special.b.exp_need_qs) { | |
282 | t->rcu_read_unlock_special.b.exp_need_qs = false; | |
283 | rcu_report_exp_rdp(rdp->rsp, rdp, true); | |
284 | } else { | |
285 | WARN_ON_ONCE(t->rcu_read_unlock_special.b.exp_need_qs); | |
286 | } | |
8203d6d0 PM |
287 | } |
288 | ||
f41d911f | 289 | /* |
6cc68793 | 290 | * Record a preemptible-RCU quiescent state for the specified CPU. Note |
f41d911f PM |
291 | * that this just means that the task currently running on the CPU is |
292 | * not in a quiescent state. There might be any number of tasks blocked | |
293 | * while in an RCU read-side critical section. | |
25502a6c | 294 | * |
1d082fd0 PM |
295 | * As with the other rcu_*_qs() functions, callers to this function |
296 | * must disable preemption. | |
f41d911f | 297 | */ |
284a8c93 | 298 | static void rcu_preempt_qs(void) |
f41d911f | 299 | { |
ea9b0c8a | 300 | RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_qs() invoked with preemption enabled!!!\n"); |
5b74c458 | 301 | if (__this_cpu_read(rcu_data_p->cpu_no_qs.s)) { |
284a8c93 | 302 | trace_rcu_grace_period(TPS("rcu_preempt"), |
2927a689 | 303 | __this_cpu_read(rcu_data_p->gpnum), |
284a8c93 | 304 | TPS("cpuqs")); |
5b74c458 | 305 | __this_cpu_write(rcu_data_p->cpu_no_qs.b.norm, false); |
284a8c93 PM |
306 | barrier(); /* Coordinate with rcu_preempt_check_callbacks(). */ |
307 | current->rcu_read_unlock_special.b.need_qs = false; | |
308 | } | |
f41d911f PM |
309 | } |
310 | ||
311 | /* | |
c3422bea PM |
312 | * We have entered the scheduler, and the current task might soon be |
313 | * context-switched away from. If this task is in an RCU read-side | |
314 | * critical section, we will no longer be able to rely on the CPU to | |
12f5f524 PM |
315 | * record that fact, so we enqueue the task on the blkd_tasks list. |
316 | * The task will dequeue itself when it exits the outermost enclosing | |
317 | * RCU read-side critical section. Therefore, the current grace period | |
318 | * cannot be permitted to complete until the blkd_tasks list entries | |
319 | * predating the current grace period drain, in other words, until | |
320 | * rnp->gp_tasks becomes NULL. | |
c3422bea | 321 | * |
46a5d164 | 322 | * Caller must disable interrupts. |
f41d911f | 323 | */ |
5b72f964 | 324 | static void rcu_preempt_note_context_switch(bool preempt) |
f41d911f PM |
325 | { |
326 | struct task_struct *t = current; | |
f41d911f PM |
327 | struct rcu_data *rdp; |
328 | struct rcu_node *rnp; | |
329 | ||
b04db8e1 | 330 | lockdep_assert_irqs_disabled(); |
5b72f964 | 331 | WARN_ON_ONCE(!preempt && t->rcu_read_lock_nesting > 0); |
10f39bb1 | 332 | if (t->rcu_read_lock_nesting > 0 && |
1d082fd0 | 333 | !t->rcu_read_unlock_special.b.blocked) { |
f41d911f PM |
334 | |
335 | /* Possibly blocking in an RCU read-side critical section. */ | |
e63c887c | 336 | rdp = this_cpu_ptr(rcu_state_p->rda); |
f41d911f | 337 | rnp = rdp->mynode; |
46a5d164 | 338 | raw_spin_lock_rcu_node(rnp); |
1d082fd0 | 339 | t->rcu_read_unlock_special.b.blocked = true; |
86848966 | 340 | t->rcu_blocked_node = rnp; |
f41d911f PM |
341 | |
342 | /* | |
8203d6d0 PM |
343 | * Verify the CPU's sanity, trace the preemption, and |
344 | * then queue the task as required based on the states | |
345 | * of any ongoing and expedited grace periods. | |
f41d911f | 346 | */ |
0aa04b05 | 347 | WARN_ON_ONCE((rdp->grpmask & rcu_rnp_online_cpus(rnp)) == 0); |
e7d8842e | 348 | WARN_ON_ONCE(!list_empty(&t->rcu_node_entry)); |
d4c08f2a PM |
349 | trace_rcu_preempt_task(rdp->rsp->name, |
350 | t->pid, | |
351 | (rnp->qsmask & rdp->grpmask) | |
352 | ? rnp->gpnum | |
353 | : rnp->gpnum + 1); | |
46a5d164 | 354 | rcu_preempt_ctxt_queue(rnp, rdp); |
10f39bb1 | 355 | } else if (t->rcu_read_lock_nesting < 0 && |
1d082fd0 | 356 | t->rcu_read_unlock_special.s) { |
10f39bb1 PM |
357 | |
358 | /* | |
359 | * Complete exit from RCU read-side critical section on | |
360 | * behalf of preempted instance of __rcu_read_unlock(). | |
361 | */ | |
362 | rcu_read_unlock_special(t); | |
f41d911f PM |
363 | } |
364 | ||
365 | /* | |
366 | * Either we were not in an RCU read-side critical section to | |
367 | * begin with, or we have now recorded that critical section | |
368 | * globally. Either way, we can now note a quiescent state | |
369 | * for this CPU. Again, if we were in an RCU read-side critical | |
370 | * section, and if that critical section was blocking the current | |
371 | * grace period, then the fact that the task has been enqueued | |
372 | * means that we continue to block the current grace period. | |
373 | */ | |
284a8c93 | 374 | rcu_preempt_qs(); |
f41d911f PM |
375 | } |
376 | ||
fc2219d4 PM |
377 | /* |
378 | * Check for preempted RCU readers blocking the current grace period | |
379 | * for the specified rcu_node structure. If the caller needs a reliable | |
380 | * answer, it must hold the rcu_node's ->lock. | |
381 | */ | |
27f4d280 | 382 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 | 383 | { |
12f5f524 | 384 | return rnp->gp_tasks != NULL; |
fc2219d4 PM |
385 | } |
386 | ||
12f5f524 PM |
387 | /* |
388 | * Advance a ->blkd_tasks-list pointer to the next entry, instead | |
389 | * returning NULL if at the end of the list. | |
390 | */ | |
391 | static struct list_head *rcu_next_node_entry(struct task_struct *t, | |
392 | struct rcu_node *rnp) | |
393 | { | |
394 | struct list_head *np; | |
395 | ||
396 | np = t->rcu_node_entry.next; | |
397 | if (np == &rnp->blkd_tasks) | |
398 | np = NULL; | |
399 | return np; | |
400 | } | |
401 | ||
8af3a5e7 PM |
402 | /* |
403 | * Return true if the specified rcu_node structure has tasks that were | |
404 | * preempted within an RCU read-side critical section. | |
405 | */ | |
406 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
407 | { | |
408 | return !list_empty(&rnp->blkd_tasks); | |
409 | } | |
410 | ||
b668c9cf PM |
411 | /* |
412 | * Handle special cases during rcu_read_unlock(), such as needing to | |
413 | * notify RCU core processing or task having blocked during the RCU | |
414 | * read-side critical section. | |
415 | */ | |
2a3fa843 | 416 | void rcu_read_unlock_special(struct task_struct *t) |
f41d911f | 417 | { |
b6a932d1 PM |
418 | bool empty_exp; |
419 | bool empty_norm; | |
420 | bool empty_exp_now; | |
f41d911f | 421 | unsigned long flags; |
12f5f524 | 422 | struct list_head *np; |
abaa93d9 | 423 | bool drop_boost_mutex = false; |
8203d6d0 | 424 | struct rcu_data *rdp; |
f41d911f | 425 | struct rcu_node *rnp; |
1d082fd0 | 426 | union rcu_special special; |
f41d911f PM |
427 | |
428 | /* NMI handlers cannot block and cannot safely manipulate state. */ | |
429 | if (in_nmi()) | |
430 | return; | |
431 | ||
432 | local_irq_save(flags); | |
433 | ||
434 | /* | |
8203d6d0 PM |
435 | * If RCU core is waiting for this CPU to exit its critical section, |
436 | * report the fact that it has exited. Because irqs are disabled, | |
1d082fd0 | 437 | * t->rcu_read_unlock_special cannot change. |
f41d911f PM |
438 | */ |
439 | special = t->rcu_read_unlock_special; | |
1d082fd0 | 440 | if (special.b.need_qs) { |
284a8c93 | 441 | rcu_preempt_qs(); |
c0135d07 | 442 | t->rcu_read_unlock_special.b.need_qs = false; |
1d082fd0 | 443 | if (!t->rcu_read_unlock_special.s) { |
79a62f95 LJ |
444 | local_irq_restore(flags); |
445 | return; | |
446 | } | |
f41d911f PM |
447 | } |
448 | ||
8203d6d0 PM |
449 | /* |
450 | * Respond to a request for an expedited grace period, but only if | |
451 | * we were not preempted, meaning that we were running on the same | |
452 | * CPU throughout. If we were preempted, the exp_need_qs flag | |
453 | * would have been cleared at the time of the first preemption, | |
454 | * and the quiescent state would be reported when we were dequeued. | |
455 | */ | |
456 | if (special.b.exp_need_qs) { | |
457 | WARN_ON_ONCE(special.b.blocked); | |
458 | t->rcu_read_unlock_special.b.exp_need_qs = false; | |
459 | rdp = this_cpu_ptr(rcu_state_p->rda); | |
460 | rcu_report_exp_rdp(rcu_state_p, rdp, true); | |
461 | if (!t->rcu_read_unlock_special.s) { | |
462 | local_irq_restore(flags); | |
463 | return; | |
464 | } | |
465 | } | |
466 | ||
79a62f95 | 467 | /* Hardware IRQ handlers cannot block, complain if they get here. */ |
d24209bb PM |
468 | if (in_irq() || in_serving_softirq()) { |
469 | lockdep_rcu_suspicious(__FILE__, __LINE__, | |
470 | "rcu_read_unlock() from irq or softirq with blocking in critical section!!!\n"); | |
8203d6d0 | 471 | pr_alert("->rcu_read_unlock_special: %#x (b: %d, enq: %d nq: %d)\n", |
d24209bb PM |
472 | t->rcu_read_unlock_special.s, |
473 | t->rcu_read_unlock_special.b.blocked, | |
8203d6d0 | 474 | t->rcu_read_unlock_special.b.exp_need_qs, |
d24209bb | 475 | t->rcu_read_unlock_special.b.need_qs); |
f41d911f PM |
476 | local_irq_restore(flags); |
477 | return; | |
478 | } | |
479 | ||
480 | /* Clean up if blocked during RCU read-side critical section. */ | |
1d082fd0 PM |
481 | if (special.b.blocked) { |
482 | t->rcu_read_unlock_special.b.blocked = false; | |
f41d911f | 483 | |
dd5d19ba | 484 | /* |
0a0ba1c9 | 485 | * Remove this task from the list it blocked on. The task |
8ba9153b PM |
486 | * now remains queued on the rcu_node corresponding to the |
487 | * CPU it first blocked on, so there is no longer any need | |
488 | * to loop. Retain a WARN_ON_ONCE() out of sheer paranoia. | |
dd5d19ba | 489 | */ |
8ba9153b PM |
490 | rnp = t->rcu_blocked_node; |
491 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
492 | WARN_ON_ONCE(rnp != t->rcu_blocked_node); | |
2dee9404 | 493 | WARN_ON_ONCE(rnp->level != rcu_num_lvls - 1); |
74e871ac | 494 | empty_norm = !rcu_preempt_blocked_readers_cgp(rnp); |
8203d6d0 | 495 | empty_exp = sync_rcu_preempt_exp_done(rnp); |
d9a3da06 | 496 | smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */ |
12f5f524 | 497 | np = rcu_next_node_entry(t, rnp); |
f41d911f | 498 | list_del_init(&t->rcu_node_entry); |
82e78d80 | 499 | t->rcu_blocked_node = NULL; |
f7f7bac9 | 500 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt"), |
d4c08f2a | 501 | rnp->gpnum, t->pid); |
12f5f524 PM |
502 | if (&t->rcu_node_entry == rnp->gp_tasks) |
503 | rnp->gp_tasks = np; | |
504 | if (&t->rcu_node_entry == rnp->exp_tasks) | |
505 | rnp->exp_tasks = np; | |
727b705b | 506 | if (IS_ENABLED(CONFIG_RCU_BOOST)) { |
727b705b PM |
507 | /* Snapshot ->boost_mtx ownership w/rnp->lock held. */ |
508 | drop_boost_mutex = rt_mutex_owner(&rnp->boost_mtx) == t; | |
2dee9404 PM |
509 | if (&t->rcu_node_entry == rnp->boost_tasks) |
510 | rnp->boost_tasks = np; | |
727b705b | 511 | } |
f41d911f PM |
512 | |
513 | /* | |
514 | * If this was the last task on the current list, and if | |
515 | * we aren't waiting on any CPUs, report the quiescent state. | |
389abd48 PM |
516 | * Note that rcu_report_unblock_qs_rnp() releases rnp->lock, |
517 | * so we must take a snapshot of the expedited state. | |
f41d911f | 518 | */ |
8203d6d0 | 519 | empty_exp_now = sync_rcu_preempt_exp_done(rnp); |
74e871ac | 520 | if (!empty_norm && !rcu_preempt_blocked_readers_cgp(rnp)) { |
f7f7bac9 | 521 | trace_rcu_quiescent_state_report(TPS("preempt_rcu"), |
d4c08f2a PM |
522 | rnp->gpnum, |
523 | 0, rnp->qsmask, | |
524 | rnp->level, | |
525 | rnp->grplo, | |
526 | rnp->grphi, | |
527 | !!rnp->gp_tasks); | |
e63c887c | 528 | rcu_report_unblock_qs_rnp(rcu_state_p, rnp, flags); |
c701d5d9 | 529 | } else { |
67c583a7 | 530 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
c701d5d9 | 531 | } |
d9a3da06 | 532 | |
27f4d280 | 533 | /* Unboost if we were boosted. */ |
727b705b | 534 | if (IS_ENABLED(CONFIG_RCU_BOOST) && drop_boost_mutex) |
02a7c234 | 535 | rt_mutex_futex_unlock(&rnp->boost_mtx); |
27f4d280 | 536 | |
d9a3da06 PM |
537 | /* |
538 | * If this was the last task on the expedited lists, | |
539 | * then we need to report up the rcu_node hierarchy. | |
540 | */ | |
389abd48 | 541 | if (!empty_exp && empty_exp_now) |
e63c887c | 542 | rcu_report_exp_rnp(rcu_state_p, rnp, true); |
b668c9cf PM |
543 | } else { |
544 | local_irq_restore(flags); | |
f41d911f | 545 | } |
f41d911f PM |
546 | } |
547 | ||
1ed509a2 PM |
548 | /* |
549 | * Dump detailed information for all tasks blocking the current RCU | |
550 | * grace period on the specified rcu_node structure. | |
551 | */ | |
552 | static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp) | |
553 | { | |
554 | unsigned long flags; | |
1ed509a2 PM |
555 | struct task_struct *t; |
556 | ||
6cf10081 | 557 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
5fd4dc06 | 558 | if (!rcu_preempt_blocked_readers_cgp(rnp)) { |
67c583a7 | 559 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
5fd4dc06 PM |
560 | return; |
561 | } | |
82efed06 | 562 | t = list_entry(rnp->gp_tasks->prev, |
12f5f524 PM |
563 | struct task_struct, rcu_node_entry); |
564 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) | |
565 | sched_show_task(t); | |
67c583a7 | 566 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
1ed509a2 PM |
567 | } |
568 | ||
569 | /* | |
570 | * Dump detailed information for all tasks blocking the current RCU | |
571 | * grace period. | |
572 | */ | |
573 | static void rcu_print_detail_task_stall(struct rcu_state *rsp) | |
574 | { | |
575 | struct rcu_node *rnp = rcu_get_root(rsp); | |
576 | ||
577 | rcu_print_detail_task_stall_rnp(rnp); | |
578 | rcu_for_each_leaf_node(rsp, rnp) | |
579 | rcu_print_detail_task_stall_rnp(rnp); | |
580 | } | |
581 | ||
a858af28 PM |
582 | static void rcu_print_task_stall_begin(struct rcu_node *rnp) |
583 | { | |
efc151c3 | 584 | pr_err("\tTasks blocked on level-%d rcu_node (CPUs %d-%d):", |
a858af28 PM |
585 | rnp->level, rnp->grplo, rnp->grphi); |
586 | } | |
587 | ||
588 | static void rcu_print_task_stall_end(void) | |
589 | { | |
efc151c3 | 590 | pr_cont("\n"); |
a858af28 PM |
591 | } |
592 | ||
f41d911f PM |
593 | /* |
594 | * Scan the current list of tasks blocked within RCU read-side critical | |
595 | * sections, printing out the tid of each. | |
596 | */ | |
9bc8b558 | 597 | static int rcu_print_task_stall(struct rcu_node *rnp) |
f41d911f | 598 | { |
f41d911f | 599 | struct task_struct *t; |
9bc8b558 | 600 | int ndetected = 0; |
f41d911f | 601 | |
27f4d280 | 602 | if (!rcu_preempt_blocked_readers_cgp(rnp)) |
9bc8b558 | 603 | return 0; |
a858af28 | 604 | rcu_print_task_stall_begin(rnp); |
82efed06 | 605 | t = list_entry(rnp->gp_tasks->prev, |
12f5f524 | 606 | struct task_struct, rcu_node_entry); |
9bc8b558 | 607 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) { |
efc151c3 | 608 | pr_cont(" P%d", t->pid); |
9bc8b558 PM |
609 | ndetected++; |
610 | } | |
a858af28 | 611 | rcu_print_task_stall_end(); |
9bc8b558 | 612 | return ndetected; |
f41d911f PM |
613 | } |
614 | ||
74611ecb PM |
615 | /* |
616 | * Scan the current list of tasks blocked within RCU read-side critical | |
617 | * sections, printing out the tid of each that is blocking the current | |
618 | * expedited grace period. | |
619 | */ | |
620 | static int rcu_print_task_exp_stall(struct rcu_node *rnp) | |
621 | { | |
622 | struct task_struct *t; | |
623 | int ndetected = 0; | |
624 | ||
625 | if (!rnp->exp_tasks) | |
626 | return 0; | |
627 | t = list_entry(rnp->exp_tasks->prev, | |
628 | struct task_struct, rcu_node_entry); | |
629 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) { | |
630 | pr_cont(" P%d", t->pid); | |
631 | ndetected++; | |
632 | } | |
633 | return ndetected; | |
634 | } | |
635 | ||
b0e165c0 PM |
636 | /* |
637 | * Check that the list of blocked tasks for the newly completed grace | |
638 | * period is in fact empty. It is a serious bug to complete a grace | |
639 | * period that still has RCU readers blocked! This function must be | |
640 | * invoked -before- updating this rnp's ->gpnum, and the rnp's ->lock | |
641 | * must be held by the caller. | |
12f5f524 PM |
642 | * |
643 | * Also, if there are blocked tasks on the list, they automatically | |
644 | * block the newly created grace period, so set up ->gp_tasks accordingly. | |
b0e165c0 PM |
645 | */ |
646 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) | |
647 | { | |
c5ebe66c PM |
648 | struct task_struct *t; |
649 | ||
ea9b0c8a | 650 | RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_check_blocked_tasks() invoked with preemption enabled!!!\n"); |
27f4d280 | 651 | WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)); |
c5ebe66c | 652 | if (rcu_preempt_has_tasks(rnp)) { |
12f5f524 | 653 | rnp->gp_tasks = rnp->blkd_tasks.next; |
c5ebe66c PM |
654 | t = container_of(rnp->gp_tasks, struct task_struct, |
655 | rcu_node_entry); | |
656 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt-GPS"), | |
657 | rnp->gpnum, t->pid); | |
658 | } | |
28ecd580 | 659 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
660 | } |
661 | ||
f41d911f PM |
662 | /* |
663 | * Check for a quiescent state from the current CPU. When a task blocks, | |
664 | * the task is recorded in the corresponding CPU's rcu_node structure, | |
665 | * which is checked elsewhere. | |
666 | * | |
667 | * Caller must disable hard irqs. | |
668 | */ | |
86aea0e6 | 669 | static void rcu_preempt_check_callbacks(void) |
f41d911f PM |
670 | { |
671 | struct task_struct *t = current; | |
672 | ||
673 | if (t->rcu_read_lock_nesting == 0) { | |
284a8c93 | 674 | rcu_preempt_qs(); |
f41d911f PM |
675 | return; |
676 | } | |
10f39bb1 | 677 | if (t->rcu_read_lock_nesting > 0 && |
97c668b8 | 678 | __this_cpu_read(rcu_data_p->core_needs_qs) && |
5b74c458 | 679 | __this_cpu_read(rcu_data_p->cpu_no_qs.b.norm)) |
1d082fd0 | 680 | t->rcu_read_unlock_special.b.need_qs = true; |
f41d911f PM |
681 | } |
682 | ||
a46e0899 PM |
683 | #ifdef CONFIG_RCU_BOOST |
684 | ||
09223371 SL |
685 | static void rcu_preempt_do_callbacks(void) |
686 | { | |
2927a689 | 687 | rcu_do_batch(rcu_state_p, this_cpu_ptr(rcu_data_p)); |
09223371 SL |
688 | } |
689 | ||
a46e0899 PM |
690 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
691 | ||
a68a2bb2 PM |
692 | /** |
693 | * call_rcu() - Queue an RCU callback for invocation after a grace period. | |
694 | * @head: structure to be used for queueing the RCU updates. | |
695 | * @func: actual callback function to be invoked after the grace period | |
696 | * | |
697 | * The callback function will be invoked some time after a full grace | |
698 | * period elapses, in other words after all pre-existing RCU read-side | |
699 | * critical sections have completed. However, the callback function | |
700 | * might well execute concurrently with RCU read-side critical sections | |
701 | * that started after call_rcu() was invoked. RCU read-side critical | |
702 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), | |
703 | * and may be nested. | |
704 | * | |
705 | * Note that all CPUs must agree that the grace period extended beyond | |
706 | * all pre-existing RCU read-side critical section. On systems with more | |
707 | * than one CPU, this means that when "func()" is invoked, each CPU is | |
708 | * guaranteed to have executed a full memory barrier since the end of its | |
709 | * last RCU read-side critical section whose beginning preceded the call | |
710 | * to call_rcu(). It also means that each CPU executing an RCU read-side | |
711 | * critical section that continues beyond the start of "func()" must have | |
712 | * executed a memory barrier after the call_rcu() but before the beginning | |
713 | * of that RCU read-side critical section. Note that these guarantees | |
714 | * include CPUs that are offline, idle, or executing in user mode, as | |
715 | * well as CPUs that are executing in the kernel. | |
716 | * | |
717 | * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the | |
718 | * resulting RCU callback function "func()", then both CPU A and CPU B are | |
719 | * guaranteed to execute a full memory barrier during the time interval | |
720 | * between the call to call_rcu() and the invocation of "func()" -- even | |
721 | * if CPU A and CPU B are the same CPU (but again only if the system has | |
722 | * more than one CPU). | |
f41d911f | 723 | */ |
b6a4ae76 | 724 | void call_rcu(struct rcu_head *head, rcu_callback_t func) |
f41d911f | 725 | { |
e63c887c | 726 | __call_rcu(head, func, rcu_state_p, -1, 0); |
f41d911f PM |
727 | } |
728 | EXPORT_SYMBOL_GPL(call_rcu); | |
729 | ||
6ebb237b PM |
730 | /** |
731 | * synchronize_rcu - wait until a grace period has elapsed. | |
732 | * | |
733 | * Control will return to the caller some time after a full grace | |
734 | * period has elapsed, in other words after all currently executing RCU | |
77d8485a PM |
735 | * read-side critical sections have completed. Note, however, that |
736 | * upon return from synchronize_rcu(), the caller might well be executing | |
737 | * concurrently with new RCU read-side critical sections that began while | |
738 | * synchronize_rcu() was waiting. RCU read-side critical sections are | |
739 | * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested. | |
f0a0e6f2 | 740 | * |
e28371c8 PM |
741 | * See the description of synchronize_sched() for more detailed |
742 | * information on memory-ordering guarantees. However, please note | |
743 | * that -only- the memory-ordering guarantees apply. For example, | |
744 | * synchronize_rcu() is -not- guaranteed to wait on things like code | |
745 | * protected by preempt_disable(), instead, synchronize_rcu() is -only- | |
746 | * guaranteed to wait on RCU read-side critical sections, that is, sections | |
747 | * of code protected by rcu_read_lock(). | |
6ebb237b PM |
748 | */ |
749 | void synchronize_rcu(void) | |
750 | { | |
f78f5b90 PM |
751 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) || |
752 | lock_is_held(&rcu_lock_map) || | |
753 | lock_is_held(&rcu_sched_lock_map), | |
754 | "Illegal synchronize_rcu() in RCU read-side critical section"); | |
52d7e48b | 755 | if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE) |
6ebb237b | 756 | return; |
5afff48b | 757 | if (rcu_gp_is_expedited()) |
3705b88d AM |
758 | synchronize_rcu_expedited(); |
759 | else | |
760 | wait_rcu_gp(call_rcu); | |
6ebb237b PM |
761 | } |
762 | EXPORT_SYMBOL_GPL(synchronize_rcu); | |
763 | ||
e74f4c45 PM |
764 | /** |
765 | * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete. | |
f0a0e6f2 PM |
766 | * |
767 | * Note that this primitive does not necessarily wait for an RCU grace period | |
768 | * to complete. For example, if there are no RCU callbacks queued anywhere | |
769 | * in the system, then rcu_barrier() is within its rights to return | |
770 | * immediately, without waiting for anything, much less an RCU grace period. | |
e74f4c45 PM |
771 | */ |
772 | void rcu_barrier(void) | |
773 | { | |
e63c887c | 774 | _rcu_barrier(rcu_state_p); |
e74f4c45 PM |
775 | } |
776 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
777 | ||
1eba8f84 | 778 | /* |
6cc68793 | 779 | * Initialize preemptible RCU's state structures. |
1eba8f84 PM |
780 | */ |
781 | static void __init __rcu_init_preempt(void) | |
782 | { | |
a87f203e | 783 | rcu_init_one(rcu_state_p); |
1eba8f84 PM |
784 | } |
785 | ||
2439b696 PM |
786 | /* |
787 | * Check for a task exiting while in a preemptible-RCU read-side | |
788 | * critical section, clean up if so. No need to issue warnings, | |
789 | * as debug_check_no_locks_held() already does this if lockdep | |
790 | * is enabled. | |
791 | */ | |
792 | void exit_rcu(void) | |
793 | { | |
794 | struct task_struct *t = current; | |
795 | ||
796 | if (likely(list_empty(¤t->rcu_node_entry))) | |
797 | return; | |
798 | t->rcu_read_lock_nesting = 1; | |
799 | barrier(); | |
1d082fd0 | 800 | t->rcu_read_unlock_special.b.blocked = true; |
2439b696 PM |
801 | __rcu_read_unlock(); |
802 | } | |
803 | ||
28f6569a | 804 | #else /* #ifdef CONFIG_PREEMPT_RCU */ |
f41d911f | 805 | |
b28a7c01 | 806 | static struct rcu_state *const rcu_state_p = &rcu_sched_state; |
27f4d280 | 807 | |
f41d911f PM |
808 | /* |
809 | * Tell them what RCU they are running. | |
810 | */ | |
0e0fc1c2 | 811 | static void __init rcu_bootup_announce(void) |
f41d911f | 812 | { |
efc151c3 | 813 | pr_info("Hierarchical RCU implementation.\n"); |
26845c28 | 814 | rcu_bootup_announce_oddness(); |
f41d911f PM |
815 | } |
816 | ||
cba6d0d6 PM |
817 | /* |
818 | * Because preemptible RCU does not exist, we never have to check for | |
819 | * CPUs being in quiescent states. | |
820 | */ | |
5b72f964 | 821 | static void rcu_preempt_note_context_switch(bool preempt) |
cba6d0d6 PM |
822 | { |
823 | } | |
824 | ||
fc2219d4 | 825 | /* |
6cc68793 | 826 | * Because preemptible RCU does not exist, there are never any preempted |
fc2219d4 PM |
827 | * RCU readers. |
828 | */ | |
27f4d280 | 829 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 PM |
830 | { |
831 | return 0; | |
832 | } | |
833 | ||
8af3a5e7 PM |
834 | /* |
835 | * Because there is no preemptible RCU, there can be no readers blocked. | |
836 | */ | |
837 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
b668c9cf | 838 | { |
8af3a5e7 | 839 | return false; |
b668c9cf PM |
840 | } |
841 | ||
1ed509a2 | 842 | /* |
6cc68793 | 843 | * Because preemptible RCU does not exist, we never have to check for |
1ed509a2 PM |
844 | * tasks blocked within RCU read-side critical sections. |
845 | */ | |
846 | static void rcu_print_detail_task_stall(struct rcu_state *rsp) | |
847 | { | |
848 | } | |
849 | ||
f41d911f | 850 | /* |
6cc68793 | 851 | * Because preemptible RCU does not exist, we never have to check for |
f41d911f PM |
852 | * tasks blocked within RCU read-side critical sections. |
853 | */ | |
9bc8b558 | 854 | static int rcu_print_task_stall(struct rcu_node *rnp) |
f41d911f | 855 | { |
9bc8b558 | 856 | return 0; |
f41d911f PM |
857 | } |
858 | ||
74611ecb PM |
859 | /* |
860 | * Because preemptible RCU does not exist, we never have to check for | |
861 | * tasks blocked within RCU read-side critical sections that are | |
862 | * blocking the current expedited grace period. | |
863 | */ | |
864 | static int rcu_print_task_exp_stall(struct rcu_node *rnp) | |
865 | { | |
866 | return 0; | |
867 | } | |
868 | ||
b0e165c0 | 869 | /* |
6cc68793 | 870 | * Because there is no preemptible RCU, there can be no readers blocked, |
49e29126 PM |
871 | * so there is no need to check for blocked tasks. So check only for |
872 | * bogus qsmask values. | |
b0e165c0 PM |
873 | */ |
874 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) | |
875 | { | |
49e29126 | 876 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
877 | } |
878 | ||
f41d911f | 879 | /* |
6cc68793 | 880 | * Because preemptible RCU does not exist, it never has any callbacks |
f41d911f PM |
881 | * to check. |
882 | */ | |
86aea0e6 | 883 | static void rcu_preempt_check_callbacks(void) |
f41d911f PM |
884 | { |
885 | } | |
886 | ||
e74f4c45 | 887 | /* |
6cc68793 | 888 | * Because preemptible RCU does not exist, rcu_barrier() is just |
e74f4c45 PM |
889 | * another name for rcu_barrier_sched(). |
890 | */ | |
891 | void rcu_barrier(void) | |
892 | { | |
893 | rcu_barrier_sched(); | |
894 | } | |
895 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
896 | ||
1eba8f84 | 897 | /* |
6cc68793 | 898 | * Because preemptible RCU does not exist, it need not be initialized. |
1eba8f84 PM |
899 | */ |
900 | static void __init __rcu_init_preempt(void) | |
901 | { | |
902 | } | |
903 | ||
2439b696 PM |
904 | /* |
905 | * Because preemptible RCU does not exist, tasks cannot possibly exit | |
906 | * while in preemptible RCU read-side critical sections. | |
907 | */ | |
908 | void exit_rcu(void) | |
909 | { | |
910 | } | |
911 | ||
28f6569a | 912 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ |
8bd93a2c | 913 | |
27f4d280 PM |
914 | #ifdef CONFIG_RCU_BOOST |
915 | ||
5d01bbd1 TG |
916 | static void rcu_wake_cond(struct task_struct *t, int status) |
917 | { | |
918 | /* | |
919 | * If the thread is yielding, only wake it when this | |
920 | * is invoked from idle | |
921 | */ | |
922 | if (status != RCU_KTHREAD_YIELDING || is_idle_task(current)) | |
923 | wake_up_process(t); | |
924 | } | |
925 | ||
27f4d280 PM |
926 | /* |
927 | * Carry out RCU priority boosting on the task indicated by ->exp_tasks | |
928 | * or ->boost_tasks, advancing the pointer to the next task in the | |
929 | * ->blkd_tasks list. | |
930 | * | |
931 | * Note that irqs must be enabled: boosting the task can block. | |
932 | * Returns 1 if there are more tasks needing to be boosted. | |
933 | */ | |
934 | static int rcu_boost(struct rcu_node *rnp) | |
935 | { | |
936 | unsigned long flags; | |
27f4d280 PM |
937 | struct task_struct *t; |
938 | struct list_head *tb; | |
939 | ||
7d0ae808 PM |
940 | if (READ_ONCE(rnp->exp_tasks) == NULL && |
941 | READ_ONCE(rnp->boost_tasks) == NULL) | |
27f4d280 PM |
942 | return 0; /* Nothing left to boost. */ |
943 | ||
2a67e741 | 944 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 PM |
945 | |
946 | /* | |
947 | * Recheck under the lock: all tasks in need of boosting | |
948 | * might exit their RCU read-side critical sections on their own. | |
949 | */ | |
950 | if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) { | |
67c583a7 | 951 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
952 | return 0; |
953 | } | |
954 | ||
955 | /* | |
956 | * Preferentially boost tasks blocking expedited grace periods. | |
957 | * This cannot starve the normal grace periods because a second | |
958 | * expedited grace period must boost all blocked tasks, including | |
959 | * those blocking the pre-existing normal grace period. | |
960 | */ | |
0ea1f2eb | 961 | if (rnp->exp_tasks != NULL) { |
27f4d280 | 962 | tb = rnp->exp_tasks; |
0ea1f2eb PM |
963 | rnp->n_exp_boosts++; |
964 | } else { | |
27f4d280 | 965 | tb = rnp->boost_tasks; |
0ea1f2eb PM |
966 | rnp->n_normal_boosts++; |
967 | } | |
968 | rnp->n_tasks_boosted++; | |
27f4d280 PM |
969 | |
970 | /* | |
971 | * We boost task t by manufacturing an rt_mutex that appears to | |
972 | * be held by task t. We leave a pointer to that rt_mutex where | |
973 | * task t can find it, and task t will release the mutex when it | |
974 | * exits its outermost RCU read-side critical section. Then | |
975 | * simply acquiring this artificial rt_mutex will boost task | |
976 | * t's priority. (Thanks to tglx for suggesting this approach!) | |
977 | * | |
978 | * Note that task t must acquire rnp->lock to remove itself from | |
979 | * the ->blkd_tasks list, which it will do from exit() if from | |
980 | * nowhere else. We therefore are guaranteed that task t will | |
981 | * stay around at least until we drop rnp->lock. Note that | |
982 | * rnp->lock also resolves races between our priority boosting | |
983 | * and task t's exiting its outermost RCU read-side critical | |
984 | * section. | |
985 | */ | |
986 | t = container_of(tb, struct task_struct, rcu_node_entry); | |
abaa93d9 | 987 | rt_mutex_init_proxy_locked(&rnp->boost_mtx, t); |
67c583a7 | 988 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
abaa93d9 PM |
989 | /* Lock only for side effect: boosts task t's priority. */ |
990 | rt_mutex_lock(&rnp->boost_mtx); | |
991 | rt_mutex_unlock(&rnp->boost_mtx); /* Then keep lockdep happy. */ | |
27f4d280 | 992 | |
7d0ae808 PM |
993 | return READ_ONCE(rnp->exp_tasks) != NULL || |
994 | READ_ONCE(rnp->boost_tasks) != NULL; | |
27f4d280 PM |
995 | } |
996 | ||
27f4d280 | 997 | /* |
bc17ea10 | 998 | * Priority-boosting kthread, one per leaf rcu_node. |
27f4d280 PM |
999 | */ |
1000 | static int rcu_boost_kthread(void *arg) | |
1001 | { | |
1002 | struct rcu_node *rnp = (struct rcu_node *)arg; | |
1003 | int spincnt = 0; | |
1004 | int more2boost; | |
1005 | ||
f7f7bac9 | 1006 | trace_rcu_utilization(TPS("Start boost kthread@init")); |
27f4d280 | 1007 | for (;;) { |
d71df90e | 1008 | rnp->boost_kthread_status = RCU_KTHREAD_WAITING; |
f7f7bac9 | 1009 | trace_rcu_utilization(TPS("End boost kthread@rcu_wait")); |
08bca60a | 1010 | rcu_wait(rnp->boost_tasks || rnp->exp_tasks); |
f7f7bac9 | 1011 | trace_rcu_utilization(TPS("Start boost kthread@rcu_wait")); |
d71df90e | 1012 | rnp->boost_kthread_status = RCU_KTHREAD_RUNNING; |
27f4d280 PM |
1013 | more2boost = rcu_boost(rnp); |
1014 | if (more2boost) | |
1015 | spincnt++; | |
1016 | else | |
1017 | spincnt = 0; | |
1018 | if (spincnt > 10) { | |
5d01bbd1 | 1019 | rnp->boost_kthread_status = RCU_KTHREAD_YIELDING; |
f7f7bac9 | 1020 | trace_rcu_utilization(TPS("End boost kthread@rcu_yield")); |
5d01bbd1 | 1021 | schedule_timeout_interruptible(2); |
f7f7bac9 | 1022 | trace_rcu_utilization(TPS("Start boost kthread@rcu_yield")); |
27f4d280 PM |
1023 | spincnt = 0; |
1024 | } | |
1025 | } | |
1217ed1b | 1026 | /* NOTREACHED */ |
f7f7bac9 | 1027 | trace_rcu_utilization(TPS("End boost kthread@notreached")); |
27f4d280 PM |
1028 | return 0; |
1029 | } | |
1030 | ||
1031 | /* | |
1032 | * Check to see if it is time to start boosting RCU readers that are | |
1033 | * blocking the current grace period, and, if so, tell the per-rcu_node | |
1034 | * kthread to start boosting them. If there is an expedited grace | |
1035 | * period in progress, it is always time to boost. | |
1036 | * | |
b065a853 PM |
1037 | * The caller must hold rnp->lock, which this function releases. |
1038 | * The ->boost_kthread_task is immortal, so we don't need to worry | |
1039 | * about it going away. | |
27f4d280 | 1040 | */ |
1217ed1b | 1041 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1042 | __releases(rnp->lock) |
27f4d280 PM |
1043 | { |
1044 | struct task_struct *t; | |
1045 | ||
ea9b0c8a | 1046 | lockdep_assert_held(&rnp->lock); |
0ea1f2eb | 1047 | if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) { |
67c583a7 | 1048 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 | 1049 | return; |
0ea1f2eb | 1050 | } |
27f4d280 PM |
1051 | if (rnp->exp_tasks != NULL || |
1052 | (rnp->gp_tasks != NULL && | |
1053 | rnp->boost_tasks == NULL && | |
1054 | rnp->qsmask == 0 && | |
1055 | ULONG_CMP_GE(jiffies, rnp->boost_time))) { | |
1056 | if (rnp->exp_tasks == NULL) | |
1057 | rnp->boost_tasks = rnp->gp_tasks; | |
67c583a7 | 1058 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 | 1059 | t = rnp->boost_kthread_task; |
5d01bbd1 TG |
1060 | if (t) |
1061 | rcu_wake_cond(t, rnp->boost_kthread_status); | |
1217ed1b | 1062 | } else { |
67c583a7 | 1063 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
1217ed1b | 1064 | } |
27f4d280 PM |
1065 | } |
1066 | ||
a46e0899 PM |
1067 | /* |
1068 | * Wake up the per-CPU kthread to invoke RCU callbacks. | |
1069 | */ | |
1070 | static void invoke_rcu_callbacks_kthread(void) | |
1071 | { | |
1072 | unsigned long flags; | |
1073 | ||
1074 | local_irq_save(flags); | |
1075 | __this_cpu_write(rcu_cpu_has_work, 1); | |
1eb52121 | 1076 | if (__this_cpu_read(rcu_cpu_kthread_task) != NULL && |
5d01bbd1 TG |
1077 | current != __this_cpu_read(rcu_cpu_kthread_task)) { |
1078 | rcu_wake_cond(__this_cpu_read(rcu_cpu_kthread_task), | |
1079 | __this_cpu_read(rcu_cpu_kthread_status)); | |
1080 | } | |
a46e0899 PM |
1081 | local_irq_restore(flags); |
1082 | } | |
1083 | ||
dff1672d PM |
1084 | /* |
1085 | * Is the current CPU running the RCU-callbacks kthread? | |
1086 | * Caller must have preemption disabled. | |
1087 | */ | |
1088 | static bool rcu_is_callbacks_kthread(void) | |
1089 | { | |
c9d4b0af | 1090 | return __this_cpu_read(rcu_cpu_kthread_task) == current; |
dff1672d PM |
1091 | } |
1092 | ||
27f4d280 PM |
1093 | #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000) |
1094 | ||
1095 | /* | |
1096 | * Do priority-boost accounting for the start of a new grace period. | |
1097 | */ | |
1098 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) | |
1099 | { | |
1100 | rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES; | |
1101 | } | |
1102 | ||
27f4d280 PM |
1103 | /* |
1104 | * Create an RCU-boost kthread for the specified node if one does not | |
1105 | * already exist. We only create this kthread for preemptible RCU. | |
1106 | * Returns zero if all is well, a negated errno otherwise. | |
1107 | */ | |
49fb4c62 | 1108 | static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp, |
0aa04b05 | 1109 | struct rcu_node *rnp) |
27f4d280 | 1110 | { |
5d01bbd1 | 1111 | int rnp_index = rnp - &rsp->node[0]; |
27f4d280 PM |
1112 | unsigned long flags; |
1113 | struct sched_param sp; | |
1114 | struct task_struct *t; | |
1115 | ||
e63c887c | 1116 | if (rcu_state_p != rsp) |
27f4d280 | 1117 | return 0; |
5d01bbd1 | 1118 | |
0aa04b05 | 1119 | if (!rcu_scheduler_fully_active || rcu_rnp_online_cpus(rnp) == 0) |
5d01bbd1 TG |
1120 | return 0; |
1121 | ||
a46e0899 | 1122 | rsp->boost = 1; |
27f4d280 PM |
1123 | if (rnp->boost_kthread_task != NULL) |
1124 | return 0; | |
1125 | t = kthread_create(rcu_boost_kthread, (void *)rnp, | |
5b61b0ba | 1126 | "rcub/%d", rnp_index); |
27f4d280 PM |
1127 | if (IS_ERR(t)) |
1128 | return PTR_ERR(t); | |
2a67e741 | 1129 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 | 1130 | rnp->boost_kthread_task = t; |
67c583a7 | 1131 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
21871d7e | 1132 | sp.sched_priority = kthread_prio; |
27f4d280 | 1133 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); |
9a432736 | 1134 | wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */ |
27f4d280 PM |
1135 | return 0; |
1136 | } | |
1137 | ||
f8b7fc6b PM |
1138 | static void rcu_kthread_do_work(void) |
1139 | { | |
c9d4b0af CL |
1140 | rcu_do_batch(&rcu_sched_state, this_cpu_ptr(&rcu_sched_data)); |
1141 | rcu_do_batch(&rcu_bh_state, this_cpu_ptr(&rcu_bh_data)); | |
f8b7fc6b PM |
1142 | rcu_preempt_do_callbacks(); |
1143 | } | |
1144 | ||
62ab7072 | 1145 | static void rcu_cpu_kthread_setup(unsigned int cpu) |
f8b7fc6b | 1146 | { |
f8b7fc6b | 1147 | struct sched_param sp; |
f8b7fc6b | 1148 | |
21871d7e | 1149 | sp.sched_priority = kthread_prio; |
62ab7072 | 1150 | sched_setscheduler_nocheck(current, SCHED_FIFO, &sp); |
f8b7fc6b PM |
1151 | } |
1152 | ||
62ab7072 | 1153 | static void rcu_cpu_kthread_park(unsigned int cpu) |
f8b7fc6b | 1154 | { |
62ab7072 | 1155 | per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU; |
f8b7fc6b PM |
1156 | } |
1157 | ||
62ab7072 | 1158 | static int rcu_cpu_kthread_should_run(unsigned int cpu) |
f8b7fc6b | 1159 | { |
c9d4b0af | 1160 | return __this_cpu_read(rcu_cpu_has_work); |
f8b7fc6b PM |
1161 | } |
1162 | ||
1163 | /* | |
1164 | * Per-CPU kernel thread that invokes RCU callbacks. This replaces the | |
e0f23060 PM |
1165 | * RCU softirq used in flavors and configurations of RCU that do not |
1166 | * support RCU priority boosting. | |
f8b7fc6b | 1167 | */ |
62ab7072 | 1168 | static void rcu_cpu_kthread(unsigned int cpu) |
f8b7fc6b | 1169 | { |
c9d4b0af CL |
1170 | unsigned int *statusp = this_cpu_ptr(&rcu_cpu_kthread_status); |
1171 | char work, *workp = this_cpu_ptr(&rcu_cpu_has_work); | |
62ab7072 | 1172 | int spincnt; |
f8b7fc6b | 1173 | |
62ab7072 | 1174 | for (spincnt = 0; spincnt < 10; spincnt++) { |
f7f7bac9 | 1175 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_wait")); |
f8b7fc6b | 1176 | local_bh_disable(); |
f8b7fc6b | 1177 | *statusp = RCU_KTHREAD_RUNNING; |
62ab7072 PM |
1178 | this_cpu_inc(rcu_cpu_kthread_loops); |
1179 | local_irq_disable(); | |
f8b7fc6b PM |
1180 | work = *workp; |
1181 | *workp = 0; | |
62ab7072 | 1182 | local_irq_enable(); |
f8b7fc6b PM |
1183 | if (work) |
1184 | rcu_kthread_do_work(); | |
1185 | local_bh_enable(); | |
62ab7072 | 1186 | if (*workp == 0) { |
f7f7bac9 | 1187 | trace_rcu_utilization(TPS("End CPU kthread@rcu_wait")); |
62ab7072 PM |
1188 | *statusp = RCU_KTHREAD_WAITING; |
1189 | return; | |
f8b7fc6b PM |
1190 | } |
1191 | } | |
62ab7072 | 1192 | *statusp = RCU_KTHREAD_YIELDING; |
f7f7bac9 | 1193 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_yield")); |
62ab7072 | 1194 | schedule_timeout_interruptible(2); |
f7f7bac9 | 1195 | trace_rcu_utilization(TPS("End CPU kthread@rcu_yield")); |
62ab7072 | 1196 | *statusp = RCU_KTHREAD_WAITING; |
f8b7fc6b PM |
1197 | } |
1198 | ||
1199 | /* | |
1200 | * Set the per-rcu_node kthread's affinity to cover all CPUs that are | |
1201 | * served by the rcu_node in question. The CPU hotplug lock is still | |
1202 | * held, so the value of rnp->qsmaskinit will be stable. | |
1203 | * | |
1204 | * We don't include outgoingcpu in the affinity set, use -1 if there is | |
1205 | * no outgoing CPU. If there are no CPUs left in the affinity set, | |
1206 | * this function allows the kthread to execute on any CPU. | |
1207 | */ | |
5d01bbd1 | 1208 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b | 1209 | { |
5d01bbd1 | 1210 | struct task_struct *t = rnp->boost_kthread_task; |
0aa04b05 | 1211 | unsigned long mask = rcu_rnp_online_cpus(rnp); |
f8b7fc6b PM |
1212 | cpumask_var_t cm; |
1213 | int cpu; | |
f8b7fc6b | 1214 | |
5d01bbd1 | 1215 | if (!t) |
f8b7fc6b | 1216 | return; |
5d01bbd1 | 1217 | if (!zalloc_cpumask_var(&cm, GFP_KERNEL)) |
f8b7fc6b | 1218 | return; |
bc75e999 MR |
1219 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1220 | if ((mask & leaf_node_cpu_bit(rnp, cpu)) && | |
1221 | cpu != outgoingcpu) | |
f8b7fc6b | 1222 | cpumask_set_cpu(cpu, cm); |
5d0b0249 | 1223 | if (cpumask_weight(cm) == 0) |
f8b7fc6b | 1224 | cpumask_setall(cm); |
5d01bbd1 | 1225 | set_cpus_allowed_ptr(t, cm); |
f8b7fc6b PM |
1226 | free_cpumask_var(cm); |
1227 | } | |
1228 | ||
62ab7072 PM |
1229 | static struct smp_hotplug_thread rcu_cpu_thread_spec = { |
1230 | .store = &rcu_cpu_kthread_task, | |
1231 | .thread_should_run = rcu_cpu_kthread_should_run, | |
1232 | .thread_fn = rcu_cpu_kthread, | |
1233 | .thread_comm = "rcuc/%u", | |
1234 | .setup = rcu_cpu_kthread_setup, | |
1235 | .park = rcu_cpu_kthread_park, | |
1236 | }; | |
f8b7fc6b PM |
1237 | |
1238 | /* | |
9386c0b7 | 1239 | * Spawn boost kthreads -- called as soon as the scheduler is running. |
f8b7fc6b | 1240 | */ |
9386c0b7 | 1241 | static void __init rcu_spawn_boost_kthreads(void) |
f8b7fc6b | 1242 | { |
f8b7fc6b | 1243 | struct rcu_node *rnp; |
5d01bbd1 | 1244 | int cpu; |
f8b7fc6b | 1245 | |
62ab7072 | 1246 | for_each_possible_cpu(cpu) |
f8b7fc6b | 1247 | per_cpu(rcu_cpu_has_work, cpu) = 0; |
62ab7072 | 1248 | BUG_ON(smpboot_register_percpu_thread(&rcu_cpu_thread_spec)); |
3e9f5c70 PM |
1249 | rcu_for_each_leaf_node(rcu_state_p, rnp) |
1250 | (void)rcu_spawn_one_boost_kthread(rcu_state_p, rnp); | |
f8b7fc6b | 1251 | } |
f8b7fc6b | 1252 | |
49fb4c62 | 1253 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b | 1254 | { |
e534165b | 1255 | struct rcu_data *rdp = per_cpu_ptr(rcu_state_p->rda, cpu); |
f8b7fc6b PM |
1256 | struct rcu_node *rnp = rdp->mynode; |
1257 | ||
1258 | /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */ | |
62ab7072 | 1259 | if (rcu_scheduler_fully_active) |
e534165b | 1260 | (void)rcu_spawn_one_boost_kthread(rcu_state_p, rnp); |
f8b7fc6b PM |
1261 | } |
1262 | ||
27f4d280 PM |
1263 | #else /* #ifdef CONFIG_RCU_BOOST */ |
1264 | ||
1217ed1b | 1265 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1266 | __releases(rnp->lock) |
27f4d280 | 1267 | { |
67c583a7 | 1268 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
1269 | } |
1270 | ||
a46e0899 | 1271 | static void invoke_rcu_callbacks_kthread(void) |
27f4d280 | 1272 | { |
a46e0899 | 1273 | WARN_ON_ONCE(1); |
27f4d280 PM |
1274 | } |
1275 | ||
dff1672d PM |
1276 | static bool rcu_is_callbacks_kthread(void) |
1277 | { | |
1278 | return false; | |
1279 | } | |
1280 | ||
27f4d280 PM |
1281 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) |
1282 | { | |
1283 | } | |
1284 | ||
5d01bbd1 | 1285 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b PM |
1286 | { |
1287 | } | |
1288 | ||
9386c0b7 | 1289 | static void __init rcu_spawn_boost_kthreads(void) |
b0d30417 | 1290 | { |
b0d30417 | 1291 | } |
b0d30417 | 1292 | |
49fb4c62 | 1293 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b PM |
1294 | { |
1295 | } | |
1296 | ||
27f4d280 PM |
1297 | #endif /* #else #ifdef CONFIG_RCU_BOOST */ |
1298 | ||
8bd93a2c PM |
1299 | #if !defined(CONFIG_RCU_FAST_NO_HZ) |
1300 | ||
1301 | /* | |
1302 | * Check to see if any future RCU-related work will need to be done | |
1303 | * by the current CPU, even if none need be done immediately, returning | |
1304 | * 1 if so. This function is part of the RCU implementation; it is -not- | |
1305 | * an exported member of the RCU API. | |
1306 | * | |
7cb92499 PM |
1307 | * Because we not have RCU_FAST_NO_HZ, just check whether this CPU needs |
1308 | * any flavor of RCU. | |
8bd93a2c | 1309 | */ |
c1ad348b | 1310 | int rcu_needs_cpu(u64 basemono, u64 *nextevt) |
8bd93a2c | 1311 | { |
c1ad348b | 1312 | *nextevt = KTIME_MAX; |
44c65ff2 | 1313 | return rcu_cpu_has_callbacks(NULL); |
7cb92499 PM |
1314 | } |
1315 | ||
1316 | /* | |
1317 | * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up | |
1318 | * after it. | |
1319 | */ | |
8fa7845d | 1320 | static void rcu_cleanup_after_idle(void) |
7cb92499 PM |
1321 | { |
1322 | } | |
1323 | ||
aea1b35e | 1324 | /* |
a858af28 | 1325 | * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n, |
aea1b35e PM |
1326 | * is nothing. |
1327 | */ | |
198bbf81 | 1328 | static void rcu_prepare_for_idle(void) |
aea1b35e PM |
1329 | { |
1330 | } | |
1331 | ||
c57afe80 PM |
1332 | /* |
1333 | * Don't bother keeping a running count of the number of RCU callbacks | |
1334 | * posted because CONFIG_RCU_FAST_NO_HZ=n. | |
1335 | */ | |
1336 | static void rcu_idle_count_callbacks_posted(void) | |
1337 | { | |
1338 | } | |
1339 | ||
8bd93a2c PM |
1340 | #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
1341 | ||
f23f7fa1 PM |
1342 | /* |
1343 | * This code is invoked when a CPU goes idle, at which point we want | |
1344 | * to have the CPU do everything required for RCU so that it can enter | |
1345 | * the energy-efficient dyntick-idle mode. This is handled by a | |
1346 | * state machine implemented by rcu_prepare_for_idle() below. | |
1347 | * | |
1348 | * The following three proprocessor symbols control this state machine: | |
1349 | * | |
f23f7fa1 PM |
1350 | * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted |
1351 | * to sleep in dyntick-idle mode with RCU callbacks pending. This | |
1352 | * is sized to be roughly one RCU grace period. Those energy-efficiency | |
1353 | * benchmarkers who might otherwise be tempted to set this to a large | |
1354 | * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your | |
1355 | * system. And if you are -that- concerned about energy efficiency, | |
1356 | * just power the system down and be done with it! | |
778d250a PM |
1357 | * RCU_IDLE_LAZY_GP_DELAY gives the number of jiffies that a CPU is |
1358 | * permitted to sleep in dyntick-idle mode with only lazy RCU | |
1359 | * callbacks pending. Setting this too high can OOM your system. | |
f23f7fa1 PM |
1360 | * |
1361 | * The values below work well in practice. If future workloads require | |
1362 | * adjustment, they can be converted into kernel config parameters, though | |
1363 | * making the state machine smarter might be a better option. | |
1364 | */ | |
e84c48ae | 1365 | #define RCU_IDLE_GP_DELAY 4 /* Roughly one grace period. */ |
778d250a | 1366 | #define RCU_IDLE_LAZY_GP_DELAY (6 * HZ) /* Roughly six seconds. */ |
f23f7fa1 | 1367 | |
5e44ce35 PM |
1368 | static int rcu_idle_gp_delay = RCU_IDLE_GP_DELAY; |
1369 | module_param(rcu_idle_gp_delay, int, 0644); | |
1370 | static int rcu_idle_lazy_gp_delay = RCU_IDLE_LAZY_GP_DELAY; | |
1371 | module_param(rcu_idle_lazy_gp_delay, int, 0644); | |
486e2593 | 1372 | |
486e2593 | 1373 | /* |
c229828c PM |
1374 | * Try to advance callbacks for all flavors of RCU on the current CPU, but |
1375 | * only if it has been awhile since the last time we did so. Afterwards, | |
1376 | * if there are any callbacks ready for immediate invocation, return true. | |
486e2593 | 1377 | */ |
f1f399d1 | 1378 | static bool __maybe_unused rcu_try_advance_all_cbs(void) |
486e2593 | 1379 | { |
c0f4dfd4 PM |
1380 | bool cbs_ready = false; |
1381 | struct rcu_data *rdp; | |
c229828c | 1382 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
c0f4dfd4 PM |
1383 | struct rcu_node *rnp; |
1384 | struct rcu_state *rsp; | |
486e2593 | 1385 | |
c229828c PM |
1386 | /* Exit early if we advanced recently. */ |
1387 | if (jiffies == rdtp->last_advance_all) | |
d0bc90fd | 1388 | return false; |
c229828c PM |
1389 | rdtp->last_advance_all = jiffies; |
1390 | ||
c0f4dfd4 PM |
1391 | for_each_rcu_flavor(rsp) { |
1392 | rdp = this_cpu_ptr(rsp->rda); | |
1393 | rnp = rdp->mynode; | |
486e2593 | 1394 | |
c0f4dfd4 PM |
1395 | /* |
1396 | * Don't bother checking unless a grace period has | |
1397 | * completed since we last checked and there are | |
1398 | * callbacks not yet ready to invoke. | |
1399 | */ | |
e3663b10 | 1400 | if ((rdp->completed != rnp->completed || |
7d0ae808 | 1401 | unlikely(READ_ONCE(rdp->gpwrap))) && |
15fecf89 | 1402 | rcu_segcblist_pend_cbs(&rdp->cblist)) |
470716fc | 1403 | note_gp_changes(rsp, rdp); |
486e2593 | 1404 | |
15fecf89 | 1405 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
c0f4dfd4 PM |
1406 | cbs_ready = true; |
1407 | } | |
1408 | return cbs_ready; | |
486e2593 PM |
1409 | } |
1410 | ||
aa9b1630 | 1411 | /* |
c0f4dfd4 PM |
1412 | * Allow the CPU to enter dyntick-idle mode unless it has callbacks ready |
1413 | * to invoke. If the CPU has callbacks, try to advance them. Tell the | |
1414 | * caller to set the timeout based on whether or not there are non-lazy | |
1415 | * callbacks. | |
aa9b1630 | 1416 | * |
c0f4dfd4 | 1417 | * The caller must have disabled interrupts. |
aa9b1630 | 1418 | */ |
c1ad348b | 1419 | int rcu_needs_cpu(u64 basemono, u64 *nextevt) |
aa9b1630 | 1420 | { |
aa6da514 | 1421 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
c1ad348b | 1422 | unsigned long dj; |
aa9b1630 | 1423 | |
b04db8e1 | 1424 | lockdep_assert_irqs_disabled(); |
3382adbc | 1425 | |
c0f4dfd4 PM |
1426 | /* Snapshot to detect later posting of non-lazy callback. */ |
1427 | rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted; | |
1428 | ||
aa9b1630 | 1429 | /* If no callbacks, RCU doesn't need the CPU. */ |
aa6da514 | 1430 | if (!rcu_cpu_has_callbacks(&rdtp->all_lazy)) { |
c1ad348b | 1431 | *nextevt = KTIME_MAX; |
aa9b1630 PM |
1432 | return 0; |
1433 | } | |
c0f4dfd4 PM |
1434 | |
1435 | /* Attempt to advance callbacks. */ | |
1436 | if (rcu_try_advance_all_cbs()) { | |
1437 | /* Some ready to invoke, so initiate later invocation. */ | |
1438 | invoke_rcu_core(); | |
aa9b1630 PM |
1439 | return 1; |
1440 | } | |
c0f4dfd4 PM |
1441 | rdtp->last_accelerate = jiffies; |
1442 | ||
1443 | /* Request timer delay depending on laziness, and round. */ | |
6faf7283 | 1444 | if (!rdtp->all_lazy) { |
c1ad348b | 1445 | dj = round_up(rcu_idle_gp_delay + jiffies, |
c0f4dfd4 | 1446 | rcu_idle_gp_delay) - jiffies; |
e84c48ae | 1447 | } else { |
c1ad348b | 1448 | dj = round_jiffies(rcu_idle_lazy_gp_delay + jiffies) - jiffies; |
e84c48ae | 1449 | } |
c1ad348b | 1450 | *nextevt = basemono + dj * TICK_NSEC; |
aa9b1630 PM |
1451 | return 0; |
1452 | } | |
1453 | ||
21e52e15 | 1454 | /* |
c0f4dfd4 PM |
1455 | * Prepare a CPU for idle from an RCU perspective. The first major task |
1456 | * is to sense whether nohz mode has been enabled or disabled via sysfs. | |
1457 | * The second major task is to check to see if a non-lazy callback has | |
1458 | * arrived at a CPU that previously had only lazy callbacks. The third | |
1459 | * major task is to accelerate (that is, assign grace-period numbers to) | |
1460 | * any recently arrived callbacks. | |
aea1b35e PM |
1461 | * |
1462 | * The caller must have disabled interrupts. | |
8bd93a2c | 1463 | */ |
198bbf81 | 1464 | static void rcu_prepare_for_idle(void) |
8bd93a2c | 1465 | { |
48a7639c | 1466 | bool needwake; |
c0f4dfd4 | 1467 | struct rcu_data *rdp; |
198bbf81 | 1468 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
c0f4dfd4 PM |
1469 | struct rcu_node *rnp; |
1470 | struct rcu_state *rsp; | |
9d2ad243 PM |
1471 | int tne; |
1472 | ||
b04db8e1 | 1473 | lockdep_assert_irqs_disabled(); |
44c65ff2 | 1474 | if (rcu_is_nocb_cpu(smp_processor_id())) |
3382adbc PM |
1475 | return; |
1476 | ||
9d2ad243 | 1477 | /* Handle nohz enablement switches conservatively. */ |
7d0ae808 | 1478 | tne = READ_ONCE(tick_nohz_active); |
9d2ad243 | 1479 | if (tne != rdtp->tick_nohz_enabled_snap) { |
aa6da514 | 1480 | if (rcu_cpu_has_callbacks(NULL)) |
9d2ad243 PM |
1481 | invoke_rcu_core(); /* force nohz to see update. */ |
1482 | rdtp->tick_nohz_enabled_snap = tne; | |
1483 | return; | |
1484 | } | |
1485 | if (!tne) | |
1486 | return; | |
f511fc62 | 1487 | |
c57afe80 | 1488 | /* |
c0f4dfd4 PM |
1489 | * If a non-lazy callback arrived at a CPU having only lazy |
1490 | * callbacks, invoke RCU core for the side-effect of recalculating | |
1491 | * idle duration on re-entry to idle. | |
c57afe80 | 1492 | */ |
c0f4dfd4 PM |
1493 | if (rdtp->all_lazy && |
1494 | rdtp->nonlazy_posted != rdtp->nonlazy_posted_snap) { | |
c337f8f5 PM |
1495 | rdtp->all_lazy = false; |
1496 | rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted; | |
c0f4dfd4 | 1497 | invoke_rcu_core(); |
c57afe80 PM |
1498 | return; |
1499 | } | |
c57afe80 | 1500 | |
3084f2f8 | 1501 | /* |
c0f4dfd4 PM |
1502 | * If we have not yet accelerated this jiffy, accelerate all |
1503 | * callbacks on this CPU. | |
3084f2f8 | 1504 | */ |
c0f4dfd4 | 1505 | if (rdtp->last_accelerate == jiffies) |
aea1b35e | 1506 | return; |
c0f4dfd4 PM |
1507 | rdtp->last_accelerate = jiffies; |
1508 | for_each_rcu_flavor(rsp) { | |
198bbf81 | 1509 | rdp = this_cpu_ptr(rsp->rda); |
135bd1a2 | 1510 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) |
c0f4dfd4 PM |
1511 | continue; |
1512 | rnp = rdp->mynode; | |
2a67e741 | 1513 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
48a7639c | 1514 | needwake = rcu_accelerate_cbs(rsp, rnp, rdp); |
67c583a7 | 1515 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
48a7639c PM |
1516 | if (needwake) |
1517 | rcu_gp_kthread_wake(rsp); | |
77e38ed3 | 1518 | } |
c0f4dfd4 | 1519 | } |
3084f2f8 | 1520 | |
c0f4dfd4 PM |
1521 | /* |
1522 | * Clean up for exit from idle. Attempt to advance callbacks based on | |
1523 | * any grace periods that elapsed while the CPU was idle, and if any | |
1524 | * callbacks are now ready to invoke, initiate invocation. | |
1525 | */ | |
8fa7845d | 1526 | static void rcu_cleanup_after_idle(void) |
c0f4dfd4 | 1527 | { |
b04db8e1 | 1528 | lockdep_assert_irqs_disabled(); |
44c65ff2 | 1529 | if (rcu_is_nocb_cpu(smp_processor_id())) |
aea1b35e | 1530 | return; |
7a497c96 PM |
1531 | if (rcu_try_advance_all_cbs()) |
1532 | invoke_rcu_core(); | |
8bd93a2c PM |
1533 | } |
1534 | ||
c57afe80 | 1535 | /* |
98248a0e PM |
1536 | * Keep a running count of the number of non-lazy callbacks posted |
1537 | * on this CPU. This running counter (which is never decremented) allows | |
1538 | * rcu_prepare_for_idle() to detect when something out of the idle loop | |
1539 | * posts a callback, even if an equal number of callbacks are invoked. | |
1540 | * Of course, callbacks should only be posted from within a trace event | |
1541 | * designed to be called from idle or from within RCU_NONIDLE(). | |
c57afe80 PM |
1542 | */ |
1543 | static void rcu_idle_count_callbacks_posted(void) | |
1544 | { | |
5955f7ee | 1545 | __this_cpu_add(rcu_dynticks.nonlazy_posted, 1); |
c57afe80 PM |
1546 | } |
1547 | ||
b626c1b6 PM |
1548 | /* |
1549 | * Data for flushing lazy RCU callbacks at OOM time. | |
1550 | */ | |
1551 | static atomic_t oom_callback_count; | |
1552 | static DECLARE_WAIT_QUEUE_HEAD(oom_callback_wq); | |
1553 | ||
1554 | /* | |
1555 | * RCU OOM callback -- decrement the outstanding count and deliver the | |
1556 | * wake-up if we are the last one. | |
1557 | */ | |
1558 | static void rcu_oom_callback(struct rcu_head *rhp) | |
1559 | { | |
1560 | if (atomic_dec_and_test(&oom_callback_count)) | |
1561 | wake_up(&oom_callback_wq); | |
1562 | } | |
1563 | ||
1564 | /* | |
1565 | * Post an rcu_oom_notify callback on the current CPU if it has at | |
1566 | * least one lazy callback. This will unnecessarily post callbacks | |
1567 | * to CPUs that already have a non-lazy callback at the end of their | |
1568 | * callback list, but this is an infrequent operation, so accept some | |
1569 | * extra overhead to keep things simple. | |
1570 | */ | |
1571 | static void rcu_oom_notify_cpu(void *unused) | |
1572 | { | |
1573 | struct rcu_state *rsp; | |
1574 | struct rcu_data *rdp; | |
1575 | ||
1576 | for_each_rcu_flavor(rsp) { | |
fa07a58f | 1577 | rdp = raw_cpu_ptr(rsp->rda); |
15fecf89 | 1578 | if (rcu_segcblist_n_lazy_cbs(&rdp->cblist)) { |
b626c1b6 PM |
1579 | atomic_inc(&oom_callback_count); |
1580 | rsp->call(&rdp->oom_head, rcu_oom_callback); | |
1581 | } | |
1582 | } | |
1583 | } | |
1584 | ||
1585 | /* | |
1586 | * If low on memory, ensure that each CPU has a non-lazy callback. | |
1587 | * This will wake up CPUs that have only lazy callbacks, in turn | |
1588 | * ensuring that they free up the corresponding memory in a timely manner. | |
1589 | * Because an uncertain amount of memory will be freed in some uncertain | |
1590 | * timeframe, we do not claim to have freed anything. | |
1591 | */ | |
1592 | static int rcu_oom_notify(struct notifier_block *self, | |
1593 | unsigned long notused, void *nfreed) | |
1594 | { | |
1595 | int cpu; | |
1596 | ||
1597 | /* Wait for callbacks from earlier instance to complete. */ | |
1598 | wait_event(oom_callback_wq, atomic_read(&oom_callback_count) == 0); | |
78e4bc34 | 1599 | smp_mb(); /* Ensure callback reuse happens after callback invocation. */ |
b626c1b6 PM |
1600 | |
1601 | /* | |
1602 | * Prevent premature wakeup: ensure that all increments happen | |
1603 | * before there is a chance of the counter reaching zero. | |
1604 | */ | |
1605 | atomic_set(&oom_callback_count, 1); | |
1606 | ||
b626c1b6 PM |
1607 | for_each_online_cpu(cpu) { |
1608 | smp_call_function_single(cpu, rcu_oom_notify_cpu, NULL, 1); | |
bde6c3aa | 1609 | cond_resched_rcu_qs(); |
b626c1b6 | 1610 | } |
b626c1b6 PM |
1611 | |
1612 | /* Unconditionally decrement: no need to wake ourselves up. */ | |
1613 | atomic_dec(&oom_callback_count); | |
1614 | ||
1615 | return NOTIFY_OK; | |
1616 | } | |
1617 | ||
1618 | static struct notifier_block rcu_oom_nb = { | |
1619 | .notifier_call = rcu_oom_notify | |
1620 | }; | |
1621 | ||
1622 | static int __init rcu_register_oom_notifier(void) | |
1623 | { | |
1624 | register_oom_notifier(&rcu_oom_nb); | |
1625 | return 0; | |
1626 | } | |
1627 | early_initcall(rcu_register_oom_notifier); | |
1628 | ||
8bd93a2c | 1629 | #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
a858af28 | 1630 | |
a858af28 PM |
1631 | #ifdef CONFIG_RCU_FAST_NO_HZ |
1632 | ||
1633 | static void print_cpu_stall_fast_no_hz(char *cp, int cpu) | |
1634 | { | |
5955f7ee | 1635 | struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); |
c0f4dfd4 | 1636 | unsigned long nlpd = rdtp->nonlazy_posted - rdtp->nonlazy_posted_snap; |
a858af28 | 1637 | |
c0f4dfd4 PM |
1638 | sprintf(cp, "last_accelerate: %04lx/%04lx, nonlazy_posted: %ld, %c%c", |
1639 | rdtp->last_accelerate & 0xffff, jiffies & 0xffff, | |
1640 | ulong2long(nlpd), | |
1641 | rdtp->all_lazy ? 'L' : '.', | |
1642 | rdtp->tick_nohz_enabled_snap ? '.' : 'D'); | |
a858af28 PM |
1643 | } |
1644 | ||
1645 | #else /* #ifdef CONFIG_RCU_FAST_NO_HZ */ | |
1646 | ||
1647 | static void print_cpu_stall_fast_no_hz(char *cp, int cpu) | |
1648 | { | |
1c17e4d4 | 1649 | *cp = '\0'; |
a858af28 PM |
1650 | } |
1651 | ||
1652 | #endif /* #else #ifdef CONFIG_RCU_FAST_NO_HZ */ | |
1653 | ||
1654 | /* Initiate the stall-info list. */ | |
1655 | static void print_cpu_stall_info_begin(void) | |
1656 | { | |
efc151c3 | 1657 | pr_cont("\n"); |
a858af28 PM |
1658 | } |
1659 | ||
1660 | /* | |
1661 | * Print out diagnostic information for the specified stalled CPU. | |
1662 | * | |
1663 | * If the specified CPU is aware of the current RCU grace period | |
1664 | * (flavor specified by rsp), then print the number of scheduling | |
1665 | * clock interrupts the CPU has taken during the time that it has | |
1666 | * been aware. Otherwise, print the number of RCU grace periods | |
1667 | * that this CPU is ignorant of, for example, "1" if the CPU was | |
1668 | * aware of the previous grace period. | |
1669 | * | |
1670 | * Also print out idle and (if CONFIG_RCU_FAST_NO_HZ) idle-entry info. | |
1671 | */ | |
1672 | static void print_cpu_stall_info(struct rcu_state *rsp, int cpu) | |
1673 | { | |
9b9500da | 1674 | unsigned long delta; |
a858af28 PM |
1675 | char fast_no_hz[72]; |
1676 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); | |
1677 | struct rcu_dynticks *rdtp = rdp->dynticks; | |
1678 | char *ticks_title; | |
1679 | unsigned long ticks_value; | |
1680 | ||
1681 | if (rsp->gpnum == rdp->gpnum) { | |
1682 | ticks_title = "ticks this GP"; | |
1683 | ticks_value = rdp->ticks_this_gp; | |
1684 | } else { | |
1685 | ticks_title = "GPs behind"; | |
1686 | ticks_value = rsp->gpnum - rdp->gpnum; | |
1687 | } | |
1688 | print_cpu_stall_fast_no_hz(fast_no_hz, cpu); | |
9b9500da PM |
1689 | delta = rdp->mynode->gpnum - rdp->rcu_iw_gpnum; |
1690 | pr_err("\t%d-%c%c%c%c: (%lu %s) idle=%03x/%llx/%d softirq=%u/%u fqs=%ld %s\n", | |
7f21aeef PM |
1691 | cpu, |
1692 | "O."[!!cpu_online(cpu)], | |
1693 | "o."[!!(rdp->grpmask & rdp->mynode->qsmaskinit)], | |
1694 | "N."[!!(rdp->grpmask & rdp->mynode->qsmaskinitnext)], | |
9b9500da PM |
1695 | !IS_ENABLED(CONFIG_IRQ_WORK) ? '?' : |
1696 | rdp->rcu_iw_pending ? (int)min(delta, 9UL) + '0' : | |
1697 | "!."[!delta], | |
7f21aeef | 1698 | ticks_value, ticks_title, |
02a5c550 | 1699 | rcu_dynticks_snap(rdtp) & 0xfff, |
a858af28 | 1700 | rdtp->dynticks_nesting, rdtp->dynticks_nmi_nesting, |
6231069b | 1701 | rdp->softirq_snap, kstat_softirqs_cpu(RCU_SOFTIRQ, cpu), |
7d0ae808 | 1702 | READ_ONCE(rsp->n_force_qs) - rsp->n_force_qs_gpstart, |
a858af28 PM |
1703 | fast_no_hz); |
1704 | } | |
1705 | ||
1706 | /* Terminate the stall-info list. */ | |
1707 | static void print_cpu_stall_info_end(void) | |
1708 | { | |
efc151c3 | 1709 | pr_err("\t"); |
a858af28 PM |
1710 | } |
1711 | ||
1712 | /* Zero ->ticks_this_gp for all flavors of RCU. */ | |
1713 | static void zero_cpu_stall_ticks(struct rcu_data *rdp) | |
1714 | { | |
1715 | rdp->ticks_this_gp = 0; | |
6231069b | 1716 | rdp->softirq_snap = kstat_softirqs_cpu(RCU_SOFTIRQ, smp_processor_id()); |
a858af28 PM |
1717 | } |
1718 | ||
1719 | /* Increment ->ticks_this_gp for all flavors of RCU. */ | |
1720 | static void increment_cpu_stall_ticks(void) | |
1721 | { | |
115f7a7c PM |
1722 | struct rcu_state *rsp; |
1723 | ||
1724 | for_each_rcu_flavor(rsp) | |
fa07a58f | 1725 | raw_cpu_inc(rsp->rda->ticks_this_gp); |
a858af28 PM |
1726 | } |
1727 | ||
3fbfbf7a PM |
1728 | #ifdef CONFIG_RCU_NOCB_CPU |
1729 | ||
1730 | /* | |
1731 | * Offload callback processing from the boot-time-specified set of CPUs | |
1732 | * specified by rcu_nocb_mask. For each CPU in the set, there is a | |
1733 | * kthread created that pulls the callbacks from the corresponding CPU, | |
1734 | * waits for a grace period to elapse, and invokes the callbacks. | |
1735 | * The no-CBs CPUs do a wake_up() on their kthread when they insert | |
1736 | * a callback into any empty list, unless the rcu_nocb_poll boot parameter | |
1737 | * has been specified, in which case each kthread actively polls its | |
1738 | * CPU. (Which isn't so great for energy efficiency, but which does | |
1739 | * reduce RCU's overhead on that CPU.) | |
1740 | * | |
1741 | * This is intended to be used in conjunction with Frederic Weisbecker's | |
1742 | * adaptive-idle work, which would seriously reduce OS jitter on CPUs | |
1743 | * running CPU-bound user-mode computations. | |
1744 | * | |
1745 | * Offloading of callback processing could also in theory be used as | |
1746 | * an energy-efficiency measure because CPUs with no RCU callbacks | |
1747 | * queued are more aggressive about entering dyntick-idle mode. | |
1748 | */ | |
1749 | ||
1750 | ||
1751 | /* Parse the boot-time rcu_nocb_mask CPU list from the kernel parameters. */ | |
1752 | static int __init rcu_nocb_setup(char *str) | |
1753 | { | |
1754 | alloc_bootmem_cpumask_var(&rcu_nocb_mask); | |
1755 | have_rcu_nocb_mask = true; | |
1756 | cpulist_parse(str, rcu_nocb_mask); | |
1757 | return 1; | |
1758 | } | |
1759 | __setup("rcu_nocbs=", rcu_nocb_setup); | |
1760 | ||
1b0048a4 PG |
1761 | static int __init parse_rcu_nocb_poll(char *arg) |
1762 | { | |
5455a7f6 | 1763 | rcu_nocb_poll = true; |
1b0048a4 PG |
1764 | return 0; |
1765 | } | |
1766 | early_param("rcu_nocb_poll", parse_rcu_nocb_poll); | |
1767 | ||
dae6e64d | 1768 | /* |
0446be48 PM |
1769 | * Wake up any no-CBs CPUs' kthreads that were waiting on the just-ended |
1770 | * grace period. | |
dae6e64d | 1771 | */ |
abedf8e2 | 1772 | static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq) |
dae6e64d | 1773 | { |
abedf8e2 | 1774 | swake_up_all(sq); |
dae6e64d PM |
1775 | } |
1776 | ||
1777 | /* | |
8b425aa8 | 1778 | * Set the root rcu_node structure's ->need_future_gp field |
dae6e64d PM |
1779 | * based on the sum of those of all rcu_node structures. This does |
1780 | * double-count the root rcu_node structure's requests, but this | |
1781 | * is necessary to handle the possibility of a rcu_nocb_kthread() | |
1782 | * having awakened during the time that the rcu_node structures | |
1783 | * were being updated for the end of the previous grace period. | |
34ed6246 | 1784 | */ |
dae6e64d PM |
1785 | static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq) |
1786 | { | |
8b425aa8 | 1787 | rnp->need_future_gp[(rnp->completed + 1) & 0x1] += nrq; |
dae6e64d PM |
1788 | } |
1789 | ||
abedf8e2 | 1790 | static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp) |
065bb78c DW |
1791 | { |
1792 | return &rnp->nocb_gp_wq[rnp->completed & 0x1]; | |
1793 | } | |
1794 | ||
dae6e64d | 1795 | static void rcu_init_one_nocb(struct rcu_node *rnp) |
34ed6246 | 1796 | { |
abedf8e2 PG |
1797 | init_swait_queue_head(&rnp->nocb_gp_wq[0]); |
1798 | init_swait_queue_head(&rnp->nocb_gp_wq[1]); | |
34ed6246 PM |
1799 | } |
1800 | ||
24342c96 | 1801 | /* Is the specified CPU a no-CBs CPU? */ |
d1e43fa5 | 1802 | bool rcu_is_nocb_cpu(int cpu) |
3fbfbf7a PM |
1803 | { |
1804 | if (have_rcu_nocb_mask) | |
1805 | return cpumask_test_cpu(cpu, rcu_nocb_mask); | |
1806 | return false; | |
1807 | } | |
1808 | ||
fbce7497 | 1809 | /* |
8be6e1b1 PM |
1810 | * Kick the leader kthread for this NOCB group. Caller holds ->nocb_lock |
1811 | * and this function releases it. | |
fbce7497 | 1812 | */ |
8be6e1b1 PM |
1813 | static void __wake_nocb_leader(struct rcu_data *rdp, bool force, |
1814 | unsigned long flags) | |
1815 | __releases(rdp->nocb_lock) | |
fbce7497 PM |
1816 | { |
1817 | struct rcu_data *rdp_leader = rdp->nocb_leader; | |
1818 | ||
8be6e1b1 PM |
1819 | lockdep_assert_held(&rdp->nocb_lock); |
1820 | if (!READ_ONCE(rdp_leader->nocb_kthread)) { | |
1821 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
fbce7497 | 1822 | return; |
8be6e1b1 PM |
1823 | } |
1824 | if (rdp_leader->nocb_leader_sleep || force) { | |
39953dfd | 1825 | /* Prior smp_mb__after_atomic() orders against prior enqueue. */ |
7d0ae808 | 1826 | WRITE_ONCE(rdp_leader->nocb_leader_sleep, false); |
8be6e1b1 PM |
1827 | del_timer(&rdp->nocb_timer); |
1828 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
6b5fc3a1 | 1829 | smp_mb(); /* ->nocb_leader_sleep before swake_up(). */ |
abedf8e2 | 1830 | swake_up(&rdp_leader->nocb_wq); |
8be6e1b1 PM |
1831 | } else { |
1832 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
fbce7497 PM |
1833 | } |
1834 | } | |
1835 | ||
8be6e1b1 PM |
1836 | /* |
1837 | * Kick the leader kthread for this NOCB group, but caller has not | |
1838 | * acquired locks. | |
1839 | */ | |
1840 | static void wake_nocb_leader(struct rcu_data *rdp, bool force) | |
1841 | { | |
1842 | unsigned long flags; | |
1843 | ||
1844 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); | |
1845 | __wake_nocb_leader(rdp, force, flags); | |
1846 | } | |
1847 | ||
1848 | /* | |
1849 | * Arrange to wake the leader kthread for this NOCB group at some | |
1850 | * future time when it is safe to do so. | |
1851 | */ | |
1852 | static void wake_nocb_leader_defer(struct rcu_data *rdp, int waketype, | |
1853 | const char *reason) | |
1854 | { | |
1855 | unsigned long flags; | |
1856 | ||
1857 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); | |
1858 | if (rdp->nocb_defer_wakeup == RCU_NOCB_WAKE_NOT) | |
1859 | mod_timer(&rdp->nocb_timer, jiffies + 1); | |
1860 | WRITE_ONCE(rdp->nocb_defer_wakeup, waketype); | |
1861 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, reason); | |
1862 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
1863 | } | |
1864 | ||
d7e29933 PM |
1865 | /* |
1866 | * Does the specified CPU need an RCU callback for the specified flavor | |
1867 | * of rcu_barrier()? | |
1868 | */ | |
1869 | static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu) | |
1870 | { | |
1871 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); | |
41050a00 PM |
1872 | unsigned long ret; |
1873 | #ifdef CONFIG_PROVE_RCU | |
d7e29933 | 1874 | struct rcu_head *rhp; |
41050a00 | 1875 | #endif /* #ifdef CONFIG_PROVE_RCU */ |
d7e29933 | 1876 | |
41050a00 PM |
1877 | /* |
1878 | * Check count of all no-CBs callbacks awaiting invocation. | |
1879 | * There needs to be a barrier before this function is called, | |
1880 | * but associated with a prior determination that no more | |
1881 | * callbacks would be posted. In the worst case, the first | |
1882 | * barrier in _rcu_barrier() suffices (but the caller cannot | |
1883 | * necessarily rely on this, not a substitute for the caller | |
1884 | * getting the concurrency design right!). There must also be | |
1885 | * a barrier between the following load an posting of a callback | |
1886 | * (if a callback is in fact needed). This is associated with an | |
1887 | * atomic_inc() in the caller. | |
1888 | */ | |
1889 | ret = atomic_long_read(&rdp->nocb_q_count); | |
d7e29933 | 1890 | |
41050a00 | 1891 | #ifdef CONFIG_PROVE_RCU |
7d0ae808 | 1892 | rhp = READ_ONCE(rdp->nocb_head); |
d7e29933 | 1893 | if (!rhp) |
7d0ae808 | 1894 | rhp = READ_ONCE(rdp->nocb_gp_head); |
d7e29933 | 1895 | if (!rhp) |
7d0ae808 | 1896 | rhp = READ_ONCE(rdp->nocb_follower_head); |
d7e29933 PM |
1897 | |
1898 | /* Having no rcuo kthread but CBs after scheduler starts is bad! */ | |
7d0ae808 | 1899 | if (!READ_ONCE(rdp->nocb_kthread) && rhp && |
59f792d1 | 1900 | rcu_scheduler_fully_active) { |
d7e29933 PM |
1901 | /* RCU callback enqueued before CPU first came online??? */ |
1902 | pr_err("RCU: Never-onlined no-CBs CPU %d has CB %p\n", | |
1903 | cpu, rhp->func); | |
1904 | WARN_ON_ONCE(1); | |
1905 | } | |
41050a00 | 1906 | #endif /* #ifdef CONFIG_PROVE_RCU */ |
d7e29933 | 1907 | |
41050a00 | 1908 | return !!ret; |
d7e29933 PM |
1909 | } |
1910 | ||
3fbfbf7a PM |
1911 | /* |
1912 | * Enqueue the specified string of rcu_head structures onto the specified | |
1913 | * CPU's no-CBs lists. The CPU is specified by rdp, the head of the | |
1914 | * string by rhp, and the tail of the string by rhtp. The non-lazy/lazy | |
1915 | * counts are supplied by rhcount and rhcount_lazy. | |
1916 | * | |
1917 | * If warranted, also wake up the kthread servicing this CPUs queues. | |
1918 | */ | |
1919 | static void __call_rcu_nocb_enqueue(struct rcu_data *rdp, | |
1920 | struct rcu_head *rhp, | |
1921 | struct rcu_head **rhtp, | |
96d3fd0d PM |
1922 | int rhcount, int rhcount_lazy, |
1923 | unsigned long flags) | |
3fbfbf7a PM |
1924 | { |
1925 | int len; | |
1926 | struct rcu_head **old_rhpp; | |
1927 | struct task_struct *t; | |
1928 | ||
1929 | /* Enqueue the callback on the nocb list and update counts. */ | |
41050a00 PM |
1930 | atomic_long_add(rhcount, &rdp->nocb_q_count); |
1931 | /* rcu_barrier() relies on ->nocb_q_count add before xchg. */ | |
3fbfbf7a | 1932 | old_rhpp = xchg(&rdp->nocb_tail, rhtp); |
7d0ae808 | 1933 | WRITE_ONCE(*old_rhpp, rhp); |
3fbfbf7a | 1934 | atomic_long_add(rhcount_lazy, &rdp->nocb_q_count_lazy); |
39953dfd | 1935 | smp_mb__after_atomic(); /* Store *old_rhpp before _wake test. */ |
3fbfbf7a PM |
1936 | |
1937 | /* If we are not being polled and there is a kthread, awaken it ... */ | |
7d0ae808 | 1938 | t = READ_ONCE(rdp->nocb_kthread); |
25e03a74 | 1939 | if (rcu_nocb_poll || !t) { |
9261dd0d PM |
1940 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
1941 | TPS("WakeNotPoll")); | |
3fbfbf7a | 1942 | return; |
9261dd0d | 1943 | } |
3fbfbf7a PM |
1944 | len = atomic_long_read(&rdp->nocb_q_count); |
1945 | if (old_rhpp == &rdp->nocb_head) { | |
96d3fd0d | 1946 | if (!irqs_disabled_flags(flags)) { |
fbce7497 PM |
1947 | /* ... if queue was empty ... */ |
1948 | wake_nocb_leader(rdp, false); | |
96d3fd0d PM |
1949 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
1950 | TPS("WakeEmpty")); | |
1951 | } else { | |
8be6e1b1 PM |
1952 | wake_nocb_leader_defer(rdp, RCU_NOCB_WAKE, |
1953 | TPS("WakeEmptyIsDeferred")); | |
96d3fd0d | 1954 | } |
3fbfbf7a PM |
1955 | rdp->qlen_last_fqs_check = 0; |
1956 | } else if (len > rdp->qlen_last_fqs_check + qhimark) { | |
fbce7497 | 1957 | /* ... or if many callbacks queued. */ |
9fdd3bc9 PM |
1958 | if (!irqs_disabled_flags(flags)) { |
1959 | wake_nocb_leader(rdp, true); | |
1960 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, | |
1961 | TPS("WakeOvf")); | |
1962 | } else { | |
8be6e1b1 PM |
1963 | wake_nocb_leader_defer(rdp, RCU_NOCB_WAKE, |
1964 | TPS("WakeOvfIsDeferred")); | |
9fdd3bc9 | 1965 | } |
3fbfbf7a | 1966 | rdp->qlen_last_fqs_check = LONG_MAX / 2; |
9261dd0d PM |
1967 | } else { |
1968 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("WakeNot")); | |
3fbfbf7a PM |
1969 | } |
1970 | return; | |
1971 | } | |
1972 | ||
1973 | /* | |
1974 | * This is a helper for __call_rcu(), which invokes this when the normal | |
1975 | * callback queue is inoperable. If this is not a no-CBs CPU, this | |
1976 | * function returns failure back to __call_rcu(), which can complain | |
1977 | * appropriately. | |
1978 | * | |
1979 | * Otherwise, this function queues the callback where the corresponding | |
1980 | * "rcuo" kthread can find it. | |
1981 | */ | |
1982 | static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, | |
96d3fd0d | 1983 | bool lazy, unsigned long flags) |
3fbfbf7a PM |
1984 | { |
1985 | ||
d1e43fa5 | 1986 | if (!rcu_is_nocb_cpu(rdp->cpu)) |
c271d3a9 | 1987 | return false; |
96d3fd0d | 1988 | __call_rcu_nocb_enqueue(rdp, rhp, &rhp->next, 1, lazy, flags); |
21e7a608 PM |
1989 | if (__is_kfree_rcu_offset((unsigned long)rhp->func)) |
1990 | trace_rcu_kfree_callback(rdp->rsp->name, rhp, | |
1991 | (unsigned long)rhp->func, | |
756cbf6b PM |
1992 | -atomic_long_read(&rdp->nocb_q_count_lazy), |
1993 | -atomic_long_read(&rdp->nocb_q_count)); | |
21e7a608 PM |
1994 | else |
1995 | trace_rcu_callback(rdp->rsp->name, rhp, | |
756cbf6b PM |
1996 | -atomic_long_read(&rdp->nocb_q_count_lazy), |
1997 | -atomic_long_read(&rdp->nocb_q_count)); | |
1772947b PM |
1998 | |
1999 | /* | |
2000 | * If called from an extended quiescent state with interrupts | |
2001 | * disabled, invoke the RCU core in order to allow the idle-entry | |
2002 | * deferred-wakeup check to function. | |
2003 | */ | |
2004 | if (irqs_disabled_flags(flags) && | |
2005 | !rcu_is_watching() && | |
2006 | cpu_online(smp_processor_id())) | |
2007 | invoke_rcu_core(); | |
2008 | ||
c271d3a9 | 2009 | return true; |
3fbfbf7a PM |
2010 | } |
2011 | ||
2012 | /* | |
2013 | * Adopt orphaned callbacks on a no-CBs CPU, or return 0 if this is | |
2014 | * not a no-CBs CPU. | |
2015 | */ | |
b1a2d79f | 2016 | static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_data *my_rdp, |
96d3fd0d PM |
2017 | struct rcu_data *rdp, |
2018 | unsigned long flags) | |
3fbfbf7a | 2019 | { |
b04db8e1 | 2020 | lockdep_assert_irqs_disabled(); |
d1e43fa5 | 2021 | if (!rcu_is_nocb_cpu(smp_processor_id())) |
b1a2d79f PM |
2022 | return false; /* Not NOCBs CPU, caller must migrate CBs. */ |
2023 | __call_rcu_nocb_enqueue(my_rdp, rcu_segcblist_head(&rdp->cblist), | |
2024 | rcu_segcblist_tail(&rdp->cblist), | |
2025 | rcu_segcblist_n_cbs(&rdp->cblist), | |
2026 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), flags); | |
2027 | rcu_segcblist_init(&rdp->cblist); | |
2028 | rcu_segcblist_disable(&rdp->cblist); | |
0a9e1e11 | 2029 | return true; |
3fbfbf7a PM |
2030 | } |
2031 | ||
2032 | /* | |
34ed6246 PM |
2033 | * If necessary, kick off a new grace period, and either way wait |
2034 | * for a subsequent grace period to complete. | |
3fbfbf7a | 2035 | */ |
34ed6246 | 2036 | static void rcu_nocb_wait_gp(struct rcu_data *rdp) |
3fbfbf7a | 2037 | { |
34ed6246 | 2038 | unsigned long c; |
dae6e64d | 2039 | bool d; |
34ed6246 | 2040 | unsigned long flags; |
48a7639c | 2041 | bool needwake; |
34ed6246 PM |
2042 | struct rcu_node *rnp = rdp->mynode; |
2043 | ||
2a67e741 | 2044 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
48a7639c | 2045 | needwake = rcu_start_future_gp(rnp, rdp, &c); |
67c583a7 | 2046 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
48a7639c PM |
2047 | if (needwake) |
2048 | rcu_gp_kthread_wake(rdp->rsp); | |
3fbfbf7a PM |
2049 | |
2050 | /* | |
34ed6246 PM |
2051 | * Wait for the grace period. Do so interruptibly to avoid messing |
2052 | * up the load average. | |
3fbfbf7a | 2053 | */ |
f7f7bac9 | 2054 | trace_rcu_future_gp(rnp, rdp, c, TPS("StartWait")); |
34ed6246 | 2055 | for (;;) { |
abedf8e2 | 2056 | swait_event_interruptible( |
dae6e64d | 2057 | rnp->nocb_gp_wq[c & 0x1], |
7d0ae808 | 2058 | (d = ULONG_CMP_GE(READ_ONCE(rnp->completed), c))); |
dae6e64d | 2059 | if (likely(d)) |
34ed6246 | 2060 | break; |
73a860cd | 2061 | WARN_ON(signal_pending(current)); |
f7f7bac9 | 2062 | trace_rcu_future_gp(rnp, rdp, c, TPS("ResumeWait")); |
34ed6246 | 2063 | } |
f7f7bac9 | 2064 | trace_rcu_future_gp(rnp, rdp, c, TPS("EndWait")); |
34ed6246 | 2065 | smp_mb(); /* Ensure that CB invocation happens after GP end. */ |
3fbfbf7a PM |
2066 | } |
2067 | ||
fbce7497 PM |
2068 | /* |
2069 | * Leaders come here to wait for additional callbacks to show up. | |
2070 | * This function does not return until callbacks appear. | |
2071 | */ | |
2072 | static void nocb_leader_wait(struct rcu_data *my_rdp) | |
2073 | { | |
2074 | bool firsttime = true; | |
8be6e1b1 | 2075 | unsigned long flags; |
fbce7497 PM |
2076 | bool gotcbs; |
2077 | struct rcu_data *rdp; | |
2078 | struct rcu_head **tail; | |
2079 | ||
2080 | wait_again: | |
2081 | ||
2082 | /* Wait for callbacks to appear. */ | |
2083 | if (!rcu_nocb_poll) { | |
bedbb648 | 2084 | trace_rcu_nocb_wake(my_rdp->rsp->name, my_rdp->cpu, TPS("Sleep")); |
abedf8e2 | 2085 | swait_event_interruptible(my_rdp->nocb_wq, |
7d0ae808 | 2086 | !READ_ONCE(my_rdp->nocb_leader_sleep)); |
8be6e1b1 PM |
2087 | raw_spin_lock_irqsave(&my_rdp->nocb_lock, flags); |
2088 | my_rdp->nocb_leader_sleep = true; | |
2089 | WRITE_ONCE(my_rdp->nocb_defer_wakeup, RCU_NOCB_WAKE_NOT); | |
2090 | del_timer(&my_rdp->nocb_timer); | |
2091 | raw_spin_unlock_irqrestore(&my_rdp->nocb_lock, flags); | |
fbce7497 PM |
2092 | } else if (firsttime) { |
2093 | firsttime = false; /* Don't drown trace log with "Poll"! */ | |
bedbb648 | 2094 | trace_rcu_nocb_wake(my_rdp->rsp->name, my_rdp->cpu, TPS("Poll")); |
fbce7497 PM |
2095 | } |
2096 | ||
2097 | /* | |
2098 | * Each pass through the following loop checks a follower for CBs. | |
2099 | * We are our own first follower. Any CBs found are moved to | |
2100 | * nocb_gp_head, where they await a grace period. | |
2101 | */ | |
2102 | gotcbs = false; | |
8be6e1b1 | 2103 | smp_mb(); /* wakeup and _sleep before ->nocb_head reads. */ |
fbce7497 | 2104 | for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) { |
7d0ae808 | 2105 | rdp->nocb_gp_head = READ_ONCE(rdp->nocb_head); |
fbce7497 PM |
2106 | if (!rdp->nocb_gp_head) |
2107 | continue; /* No CBs here, try next follower. */ | |
2108 | ||
2109 | /* Move callbacks to wait-for-GP list, which is empty. */ | |
7d0ae808 | 2110 | WRITE_ONCE(rdp->nocb_head, NULL); |
fbce7497 | 2111 | rdp->nocb_gp_tail = xchg(&rdp->nocb_tail, &rdp->nocb_head); |
fbce7497 PM |
2112 | gotcbs = true; |
2113 | } | |
2114 | ||
8be6e1b1 | 2115 | /* No callbacks? Sleep a bit if polling, and go retry. */ |
fbce7497 | 2116 | if (unlikely(!gotcbs)) { |
73a860cd | 2117 | WARN_ON(signal_pending(current)); |
8be6e1b1 PM |
2118 | if (rcu_nocb_poll) { |
2119 | schedule_timeout_interruptible(1); | |
2120 | } else { | |
fbce7497 | 2121 | trace_rcu_nocb_wake(my_rdp->rsp->name, my_rdp->cpu, |
bedbb648 | 2122 | TPS("WokeEmpty")); |
8be6e1b1 | 2123 | } |
fbce7497 PM |
2124 | goto wait_again; |
2125 | } | |
2126 | ||
2127 | /* Wait for one grace period. */ | |
2128 | rcu_nocb_wait_gp(my_rdp); | |
2129 | ||
fbce7497 PM |
2130 | /* Each pass through the following loop wakes a follower, if needed. */ |
2131 | for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) { | |
8be6e1b1 PM |
2132 | if (!rcu_nocb_poll && |
2133 | READ_ONCE(rdp->nocb_head) && | |
2134 | READ_ONCE(my_rdp->nocb_leader_sleep)) { | |
2135 | raw_spin_lock_irqsave(&my_rdp->nocb_lock, flags); | |
11ed7f93 | 2136 | my_rdp->nocb_leader_sleep = false;/* No need to sleep.*/ |
8be6e1b1 PM |
2137 | raw_spin_unlock_irqrestore(&my_rdp->nocb_lock, flags); |
2138 | } | |
fbce7497 PM |
2139 | if (!rdp->nocb_gp_head) |
2140 | continue; /* No CBs, so no need to wake follower. */ | |
2141 | ||
2142 | /* Append callbacks to follower's "done" list. */ | |
8be6e1b1 PM |
2143 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); |
2144 | tail = rdp->nocb_follower_tail; | |
2145 | rdp->nocb_follower_tail = rdp->nocb_gp_tail; | |
fbce7497 | 2146 | *tail = rdp->nocb_gp_head; |
8be6e1b1 | 2147 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); |
fbce7497 | 2148 | if (rdp != my_rdp && tail == &rdp->nocb_follower_head) { |
8be6e1b1 | 2149 | /* List was empty, so wake up the follower. */ |
abedf8e2 | 2150 | swake_up(&rdp->nocb_wq); |
fbce7497 PM |
2151 | } |
2152 | } | |
2153 | ||
2154 | /* If we (the leader) don't have CBs, go wait some more. */ | |
2155 | if (!my_rdp->nocb_follower_head) | |
2156 | goto wait_again; | |
2157 | } | |
2158 | ||
2159 | /* | |
2160 | * Followers come here to wait for additional callbacks to show up. | |
2161 | * This function does not return until callbacks appear. | |
2162 | */ | |
2163 | static void nocb_follower_wait(struct rcu_data *rdp) | |
2164 | { | |
fbce7497 | 2165 | for (;;) { |
bedbb648 | 2166 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("FollowerSleep")); |
8be6e1b1 PM |
2167 | swait_event_interruptible(rdp->nocb_wq, |
2168 | READ_ONCE(rdp->nocb_follower_head)); | |
fbce7497 PM |
2169 | if (smp_load_acquire(&rdp->nocb_follower_head)) { |
2170 | /* ^^^ Ensure CB invocation follows _head test. */ | |
2171 | return; | |
2172 | } | |
73a860cd | 2173 | WARN_ON(signal_pending(current)); |
bedbb648 | 2174 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("WokeEmpty")); |
fbce7497 PM |
2175 | } |
2176 | } | |
2177 | ||
3fbfbf7a PM |
2178 | /* |
2179 | * Per-rcu_data kthread, but only for no-CBs CPUs. Each kthread invokes | |
fbce7497 PM |
2180 | * callbacks queued by the corresponding no-CBs CPU, however, there is |
2181 | * an optional leader-follower relationship so that the grace-period | |
2182 | * kthreads don't have to do quite so many wakeups. | |
3fbfbf7a PM |
2183 | */ |
2184 | static int rcu_nocb_kthread(void *arg) | |
2185 | { | |
2186 | int c, cl; | |
8be6e1b1 | 2187 | unsigned long flags; |
3fbfbf7a PM |
2188 | struct rcu_head *list; |
2189 | struct rcu_head *next; | |
2190 | struct rcu_head **tail; | |
2191 | struct rcu_data *rdp = arg; | |
2192 | ||
2193 | /* Each pass through this loop invokes one batch of callbacks */ | |
2194 | for (;;) { | |
fbce7497 PM |
2195 | /* Wait for callbacks. */ |
2196 | if (rdp->nocb_leader == rdp) | |
2197 | nocb_leader_wait(rdp); | |
2198 | else | |
2199 | nocb_follower_wait(rdp); | |
2200 | ||
2201 | /* Pull the ready-to-invoke callbacks onto local list. */ | |
8be6e1b1 PM |
2202 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); |
2203 | list = rdp->nocb_follower_head; | |
2204 | rdp->nocb_follower_head = NULL; | |
2205 | tail = rdp->nocb_follower_tail; | |
2206 | rdp->nocb_follower_tail = &rdp->nocb_follower_head; | |
2207 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
fbce7497 | 2208 | BUG_ON(!list); |
bedbb648 | 2209 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("WokeNonEmpty")); |
3fbfbf7a PM |
2210 | |
2211 | /* Each pass through the following loop invokes a callback. */ | |
41050a00 PM |
2212 | trace_rcu_batch_start(rdp->rsp->name, |
2213 | atomic_long_read(&rdp->nocb_q_count_lazy), | |
2214 | atomic_long_read(&rdp->nocb_q_count), -1); | |
3fbfbf7a PM |
2215 | c = cl = 0; |
2216 | while (list) { | |
2217 | next = list->next; | |
2218 | /* Wait for enqueuing to complete, if needed. */ | |
2219 | while (next == NULL && &list->next != tail) { | |
69a79bb1 PM |
2220 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
2221 | TPS("WaitQueue")); | |
3fbfbf7a | 2222 | schedule_timeout_interruptible(1); |
69a79bb1 PM |
2223 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
2224 | TPS("WokeQueue")); | |
3fbfbf7a PM |
2225 | next = list->next; |
2226 | } | |
2227 | debug_rcu_head_unqueue(list); | |
2228 | local_bh_disable(); | |
2229 | if (__rcu_reclaim(rdp->rsp->name, list)) | |
2230 | cl++; | |
2231 | c++; | |
2232 | local_bh_enable(); | |
bedc1969 | 2233 | cond_resched_rcu_qs(); |
3fbfbf7a PM |
2234 | list = next; |
2235 | } | |
2236 | trace_rcu_batch_end(rdp->rsp->name, c, !!list, 0, 0, 1); | |
41050a00 PM |
2237 | smp_mb__before_atomic(); /* _add after CB invocation. */ |
2238 | atomic_long_add(-c, &rdp->nocb_q_count); | |
2239 | atomic_long_add(-cl, &rdp->nocb_q_count_lazy); | |
c635a4e1 | 2240 | rdp->n_nocbs_invoked += c; |
3fbfbf7a PM |
2241 | } |
2242 | return 0; | |
2243 | } | |
2244 | ||
96d3fd0d | 2245 | /* Is a deferred wakeup of rcu_nocb_kthread() required? */ |
9fdd3bc9 | 2246 | static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp) |
96d3fd0d | 2247 | { |
7d0ae808 | 2248 | return READ_ONCE(rdp->nocb_defer_wakeup); |
96d3fd0d PM |
2249 | } |
2250 | ||
2251 | /* Do a deferred wakeup of rcu_nocb_kthread(). */ | |
8be6e1b1 | 2252 | static void do_nocb_deferred_wakeup_common(struct rcu_data *rdp) |
96d3fd0d | 2253 | { |
8be6e1b1 | 2254 | unsigned long flags; |
9fdd3bc9 PM |
2255 | int ndw; |
2256 | ||
8be6e1b1 PM |
2257 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); |
2258 | if (!rcu_nocb_need_deferred_wakeup(rdp)) { | |
2259 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
96d3fd0d | 2260 | return; |
8be6e1b1 | 2261 | } |
7d0ae808 | 2262 | ndw = READ_ONCE(rdp->nocb_defer_wakeup); |
511324e4 | 2263 | WRITE_ONCE(rdp->nocb_defer_wakeup, RCU_NOCB_WAKE_NOT); |
8be6e1b1 | 2264 | __wake_nocb_leader(rdp, ndw == RCU_NOCB_WAKE_FORCE, flags); |
9fdd3bc9 | 2265 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("DeferredWake")); |
96d3fd0d PM |
2266 | } |
2267 | ||
8be6e1b1 | 2268 | /* Do a deferred wakeup of rcu_nocb_kthread() from a timer handler. */ |
fd30b717 | 2269 | static void do_nocb_deferred_wakeup_timer(struct timer_list *t) |
8be6e1b1 | 2270 | { |
fd30b717 KC |
2271 | struct rcu_data *rdp = from_timer(rdp, t, nocb_timer); |
2272 | ||
2273 | do_nocb_deferred_wakeup_common(rdp); | |
8be6e1b1 PM |
2274 | } |
2275 | ||
2276 | /* | |
2277 | * Do a deferred wakeup of rcu_nocb_kthread() from fastpath. | |
2278 | * This means we do an inexact common-case check. Note that if | |
2279 | * we miss, ->nocb_timer will eventually clean things up. | |
2280 | */ | |
2281 | static void do_nocb_deferred_wakeup(struct rcu_data *rdp) | |
2282 | { | |
2283 | if (rcu_nocb_need_deferred_wakeup(rdp)) | |
2284 | do_nocb_deferred_wakeup_common(rdp); | |
2285 | } | |
2286 | ||
f4579fc5 PM |
2287 | void __init rcu_init_nohz(void) |
2288 | { | |
2289 | int cpu; | |
2290 | bool need_rcu_nocb_mask = true; | |
2291 | struct rcu_state *rsp; | |
2292 | ||
f4579fc5 PM |
2293 | #if defined(CONFIG_NO_HZ_FULL) |
2294 | if (tick_nohz_full_running && cpumask_weight(tick_nohz_full_mask)) | |
2295 | need_rcu_nocb_mask = true; | |
2296 | #endif /* #if defined(CONFIG_NO_HZ_FULL) */ | |
2297 | ||
2298 | if (!have_rcu_nocb_mask && need_rcu_nocb_mask) { | |
949cccdb PK |
2299 | if (!zalloc_cpumask_var(&rcu_nocb_mask, GFP_KERNEL)) { |
2300 | pr_info("rcu_nocb_mask allocation failed, callback offloading disabled.\n"); | |
2301 | return; | |
2302 | } | |
f4579fc5 PM |
2303 | have_rcu_nocb_mask = true; |
2304 | } | |
2305 | if (!have_rcu_nocb_mask) | |
2306 | return; | |
2307 | ||
f4579fc5 PM |
2308 | #if defined(CONFIG_NO_HZ_FULL) |
2309 | if (tick_nohz_full_running) | |
2310 | cpumask_or(rcu_nocb_mask, rcu_nocb_mask, tick_nohz_full_mask); | |
2311 | #endif /* #if defined(CONFIG_NO_HZ_FULL) */ | |
2312 | ||
2313 | if (!cpumask_subset(rcu_nocb_mask, cpu_possible_mask)) { | |
2314 | pr_info("\tNote: kernel parameter 'rcu_nocbs=' contains nonexistent CPUs.\n"); | |
2315 | cpumask_and(rcu_nocb_mask, cpu_possible_mask, | |
2316 | rcu_nocb_mask); | |
2317 | } | |
ad853b48 TH |
2318 | pr_info("\tOffload RCU callbacks from CPUs: %*pbl.\n", |
2319 | cpumask_pr_args(rcu_nocb_mask)); | |
f4579fc5 PM |
2320 | if (rcu_nocb_poll) |
2321 | pr_info("\tPoll for callbacks from no-CBs CPUs.\n"); | |
2322 | ||
2323 | for_each_rcu_flavor(rsp) { | |
34404ca8 PM |
2324 | for_each_cpu(cpu, rcu_nocb_mask) |
2325 | init_nocb_callback_list(per_cpu_ptr(rsp->rda, cpu)); | |
35ce7f29 | 2326 | rcu_organize_nocb_kthreads(rsp); |
f4579fc5 | 2327 | } |
96d3fd0d PM |
2328 | } |
2329 | ||
3fbfbf7a PM |
2330 | /* Initialize per-rcu_data variables for no-CBs CPUs. */ |
2331 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) | |
2332 | { | |
2333 | rdp->nocb_tail = &rdp->nocb_head; | |
abedf8e2 | 2334 | init_swait_queue_head(&rdp->nocb_wq); |
fbce7497 | 2335 | rdp->nocb_follower_tail = &rdp->nocb_follower_head; |
8be6e1b1 | 2336 | raw_spin_lock_init(&rdp->nocb_lock); |
fd30b717 | 2337 | timer_setup(&rdp->nocb_timer, do_nocb_deferred_wakeup_timer, 0); |
3fbfbf7a PM |
2338 | } |
2339 | ||
35ce7f29 PM |
2340 | /* |
2341 | * If the specified CPU is a no-CBs CPU that does not already have its | |
2342 | * rcuo kthread for the specified RCU flavor, spawn it. If the CPUs are | |
2343 | * brought online out of order, this can require re-organizing the | |
2344 | * leader-follower relationships. | |
2345 | */ | |
2346 | static void rcu_spawn_one_nocb_kthread(struct rcu_state *rsp, int cpu) | |
2347 | { | |
2348 | struct rcu_data *rdp; | |
2349 | struct rcu_data *rdp_last; | |
2350 | struct rcu_data *rdp_old_leader; | |
2351 | struct rcu_data *rdp_spawn = per_cpu_ptr(rsp->rda, cpu); | |
2352 | struct task_struct *t; | |
2353 | ||
2354 | /* | |
2355 | * If this isn't a no-CBs CPU or if it already has an rcuo kthread, | |
2356 | * then nothing to do. | |
2357 | */ | |
2358 | if (!rcu_is_nocb_cpu(cpu) || rdp_spawn->nocb_kthread) | |
2359 | return; | |
2360 | ||
2361 | /* If we didn't spawn the leader first, reorganize! */ | |
2362 | rdp_old_leader = rdp_spawn->nocb_leader; | |
2363 | if (rdp_old_leader != rdp_spawn && !rdp_old_leader->nocb_kthread) { | |
2364 | rdp_last = NULL; | |
2365 | rdp = rdp_old_leader; | |
2366 | do { | |
2367 | rdp->nocb_leader = rdp_spawn; | |
2368 | if (rdp_last && rdp != rdp_spawn) | |
2369 | rdp_last->nocb_next_follower = rdp; | |
bbe5d7a9 PM |
2370 | if (rdp == rdp_spawn) { |
2371 | rdp = rdp->nocb_next_follower; | |
2372 | } else { | |
2373 | rdp_last = rdp; | |
2374 | rdp = rdp->nocb_next_follower; | |
2375 | rdp_last->nocb_next_follower = NULL; | |
2376 | } | |
35ce7f29 PM |
2377 | } while (rdp); |
2378 | rdp_spawn->nocb_next_follower = rdp_old_leader; | |
2379 | } | |
2380 | ||
2381 | /* Spawn the kthread for this CPU and RCU flavor. */ | |
2382 | t = kthread_run(rcu_nocb_kthread, rdp_spawn, | |
2383 | "rcuo%c/%d", rsp->abbr, cpu); | |
2384 | BUG_ON(IS_ERR(t)); | |
7d0ae808 | 2385 | WRITE_ONCE(rdp_spawn->nocb_kthread, t); |
35ce7f29 PM |
2386 | } |
2387 | ||
2388 | /* | |
2389 | * If the specified CPU is a no-CBs CPU that does not already have its | |
2390 | * rcuo kthreads, spawn them. | |
2391 | */ | |
2392 | static void rcu_spawn_all_nocb_kthreads(int cpu) | |
2393 | { | |
2394 | struct rcu_state *rsp; | |
2395 | ||
2396 | if (rcu_scheduler_fully_active) | |
2397 | for_each_rcu_flavor(rsp) | |
2398 | rcu_spawn_one_nocb_kthread(rsp, cpu); | |
2399 | } | |
2400 | ||
2401 | /* | |
2402 | * Once the scheduler is running, spawn rcuo kthreads for all online | |
2403 | * no-CBs CPUs. This assumes that the early_initcall()s happen before | |
2404 | * non-boot CPUs come online -- if this changes, we will need to add | |
2405 | * some mutual exclusion. | |
2406 | */ | |
2407 | static void __init rcu_spawn_nocb_kthreads(void) | |
2408 | { | |
2409 | int cpu; | |
2410 | ||
2411 | for_each_online_cpu(cpu) | |
2412 | rcu_spawn_all_nocb_kthreads(cpu); | |
2413 | } | |
2414 | ||
fbce7497 PM |
2415 | /* How many follower CPU IDs per leader? Default of -1 for sqrt(nr_cpu_ids). */ |
2416 | static int rcu_nocb_leader_stride = -1; | |
2417 | module_param(rcu_nocb_leader_stride, int, 0444); | |
2418 | ||
2419 | /* | |
35ce7f29 | 2420 | * Initialize leader-follower relationships for all no-CBs CPU. |
fbce7497 | 2421 | */ |
35ce7f29 | 2422 | static void __init rcu_organize_nocb_kthreads(struct rcu_state *rsp) |
3fbfbf7a PM |
2423 | { |
2424 | int cpu; | |
fbce7497 PM |
2425 | int ls = rcu_nocb_leader_stride; |
2426 | int nl = 0; /* Next leader. */ | |
3fbfbf7a | 2427 | struct rcu_data *rdp; |
fbce7497 PM |
2428 | struct rcu_data *rdp_leader = NULL; /* Suppress misguided gcc warn. */ |
2429 | struct rcu_data *rdp_prev = NULL; | |
3fbfbf7a | 2430 | |
22c2f669 | 2431 | if (!have_rcu_nocb_mask) |
3fbfbf7a | 2432 | return; |
fbce7497 PM |
2433 | if (ls == -1) { |
2434 | ls = int_sqrt(nr_cpu_ids); | |
2435 | rcu_nocb_leader_stride = ls; | |
2436 | } | |
2437 | ||
2438 | /* | |
9831ce3b PM |
2439 | * Each pass through this loop sets up one rcu_data structure. |
2440 | * Should the corresponding CPU come online in the future, then | |
2441 | * we will spawn the needed set of rcu_nocb_kthread() kthreads. | |
fbce7497 | 2442 | */ |
3fbfbf7a PM |
2443 | for_each_cpu(cpu, rcu_nocb_mask) { |
2444 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
fbce7497 PM |
2445 | if (rdp->cpu >= nl) { |
2446 | /* New leader, set up for followers & next leader. */ | |
2447 | nl = DIV_ROUND_UP(rdp->cpu + 1, ls) * ls; | |
2448 | rdp->nocb_leader = rdp; | |
2449 | rdp_leader = rdp; | |
2450 | } else { | |
2451 | /* Another follower, link to previous leader. */ | |
2452 | rdp->nocb_leader = rdp_leader; | |
2453 | rdp_prev->nocb_next_follower = rdp; | |
2454 | } | |
2455 | rdp_prev = rdp; | |
3fbfbf7a PM |
2456 | } |
2457 | } | |
2458 | ||
2459 | /* Prevent __call_rcu() from enqueuing callbacks on no-CBs CPUs */ | |
34ed6246 | 2460 | static bool init_nocb_callback_list(struct rcu_data *rdp) |
3fbfbf7a | 2461 | { |
22c2f669 | 2462 | if (!rcu_is_nocb_cpu(rdp->cpu)) |
34ed6246 | 2463 | return false; |
22c2f669 | 2464 | |
34404ca8 | 2465 | /* If there are early-boot callbacks, move them to nocb lists. */ |
15fecf89 PM |
2466 | if (!rcu_segcblist_empty(&rdp->cblist)) { |
2467 | rdp->nocb_head = rcu_segcblist_head(&rdp->cblist); | |
2468 | rdp->nocb_tail = rcu_segcblist_tail(&rdp->cblist); | |
2469 | atomic_long_set(&rdp->nocb_q_count, | |
2470 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
2471 | atomic_long_set(&rdp->nocb_q_count_lazy, | |
2472 | rcu_segcblist_n_lazy_cbs(&rdp->cblist)); | |
2473 | rcu_segcblist_init(&rdp->cblist); | |
34404ca8 | 2474 | } |
15fecf89 | 2475 | rcu_segcblist_disable(&rdp->cblist); |
34ed6246 | 2476 | return true; |
3fbfbf7a PM |
2477 | } |
2478 | ||
34ed6246 PM |
2479 | #else /* #ifdef CONFIG_RCU_NOCB_CPU */ |
2480 | ||
d7e29933 PM |
2481 | static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu) |
2482 | { | |
2483 | WARN_ON_ONCE(1); /* Should be dead code. */ | |
2484 | return false; | |
2485 | } | |
2486 | ||
abedf8e2 | 2487 | static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq) |
3fbfbf7a | 2488 | { |
3fbfbf7a PM |
2489 | } |
2490 | ||
dae6e64d PM |
2491 | static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq) |
2492 | { | |
2493 | } | |
2494 | ||
abedf8e2 | 2495 | static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp) |
065bb78c DW |
2496 | { |
2497 | return NULL; | |
2498 | } | |
2499 | ||
dae6e64d PM |
2500 | static void rcu_init_one_nocb(struct rcu_node *rnp) |
2501 | { | |
2502 | } | |
3fbfbf7a | 2503 | |
3fbfbf7a | 2504 | static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, |
96d3fd0d | 2505 | bool lazy, unsigned long flags) |
3fbfbf7a | 2506 | { |
4afc7e26 | 2507 | return false; |
3fbfbf7a PM |
2508 | } |
2509 | ||
b1a2d79f | 2510 | static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_data *my_rdp, |
96d3fd0d PM |
2511 | struct rcu_data *rdp, |
2512 | unsigned long flags) | |
3fbfbf7a | 2513 | { |
f4aa84ba | 2514 | return false; |
3fbfbf7a PM |
2515 | } |
2516 | ||
3fbfbf7a PM |
2517 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) |
2518 | { | |
2519 | } | |
2520 | ||
9fdd3bc9 | 2521 | static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp) |
96d3fd0d PM |
2522 | { |
2523 | return false; | |
2524 | } | |
2525 | ||
2526 | static void do_nocb_deferred_wakeup(struct rcu_data *rdp) | |
2527 | { | |
2528 | } | |
2529 | ||
35ce7f29 PM |
2530 | static void rcu_spawn_all_nocb_kthreads(int cpu) |
2531 | { | |
2532 | } | |
2533 | ||
2534 | static void __init rcu_spawn_nocb_kthreads(void) | |
3fbfbf7a PM |
2535 | { |
2536 | } | |
2537 | ||
34ed6246 | 2538 | static bool init_nocb_callback_list(struct rcu_data *rdp) |
3fbfbf7a | 2539 | { |
34ed6246 | 2540 | return false; |
3fbfbf7a PM |
2541 | } |
2542 | ||
2543 | #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ | |
65d798f0 PM |
2544 | |
2545 | /* | |
2546 | * An adaptive-ticks CPU can potentially execute in kernel mode for an | |
2547 | * arbitrarily long period of time with the scheduling-clock tick turned | |
2548 | * off. RCU will be paying attention to this CPU because it is in the | |
2549 | * kernel, but the CPU cannot be guaranteed to be executing the RCU state | |
2550 | * machine because the scheduling-clock tick has been disabled. Therefore, | |
2551 | * if an adaptive-ticks CPU is failing to respond to the current grace | |
2552 | * period and has not be idle from an RCU perspective, kick it. | |
2553 | */ | |
4a81e832 | 2554 | static void __maybe_unused rcu_kick_nohz_cpu(int cpu) |
65d798f0 PM |
2555 | { |
2556 | #ifdef CONFIG_NO_HZ_FULL | |
2557 | if (tick_nohz_full_cpu(cpu)) | |
2558 | smp_send_reschedule(cpu); | |
2559 | #endif /* #ifdef CONFIG_NO_HZ_FULL */ | |
2560 | } | |
2333210b | 2561 | |
a096932f PM |
2562 | /* |
2563 | * Is this CPU a NO_HZ_FULL CPU that should ignore RCU so that the | |
2564 | * grace-period kthread will do force_quiescent_state() processing? | |
2565 | * The idea is to avoid waking up RCU core processing on such a | |
2566 | * CPU unless the grace period has extended for too long. | |
2567 | * | |
2568 | * This code relies on the fact that all NO_HZ_FULL CPUs are also | |
52e2bb95 | 2569 | * CONFIG_RCU_NOCB_CPU CPUs. |
a096932f PM |
2570 | */ |
2571 | static bool rcu_nohz_full_cpu(struct rcu_state *rsp) | |
2572 | { | |
2573 | #ifdef CONFIG_NO_HZ_FULL | |
2574 | if (tick_nohz_full_cpu(smp_processor_id()) && | |
2575 | (!rcu_gp_in_progress(rsp) || | |
7d0ae808 | 2576 | ULONG_CMP_LT(jiffies, READ_ONCE(rsp->gp_start) + HZ))) |
5ce035fb | 2577 | return true; |
a096932f | 2578 | #endif /* #ifdef CONFIG_NO_HZ_FULL */ |
5ce035fb | 2579 | return false; |
a096932f | 2580 | } |
5057f55e PM |
2581 | |
2582 | /* | |
2583 | * Bind the grace-period kthread for the sysidle flavor of RCU to the | |
2584 | * timekeeping CPU. | |
2585 | */ | |
2586 | static void rcu_bind_gp_kthread(void) | |
2587 | { | |
c0f489d2 | 2588 | int __maybe_unused cpu; |
5057f55e | 2589 | |
c0f489d2 | 2590 | if (!tick_nohz_full_enabled()) |
5057f55e | 2591 | return; |
de201559 | 2592 | housekeeping_affine(current, HK_FLAG_RCU); |
5057f55e | 2593 | } |
176f8f7a PM |
2594 | |
2595 | /* Record the current task on dyntick-idle entry. */ | |
2596 | static void rcu_dynticks_task_enter(void) | |
2597 | { | |
2598 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 2599 | WRITE_ONCE(current->rcu_tasks_idle_cpu, smp_processor_id()); |
176f8f7a PM |
2600 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
2601 | } | |
2602 | ||
2603 | /* Record no current task on dyntick-idle exit. */ | |
2604 | static void rcu_dynticks_task_exit(void) | |
2605 | { | |
2606 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 2607 | WRITE_ONCE(current->rcu_tasks_idle_cpu, -1); |
176f8f7a PM |
2608 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
2609 | } |