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. */ | |
1b0048a4 | 64 | static bool __read_mostly rcu_nocb_poll; /* Offload kthread are to poll. */ |
3fbfbf7a PM |
65 | #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ |
66 | ||
26845c28 PM |
67 | /* |
68 | * Check the RCU kernel configuration parameters and print informative | |
699d4035 | 69 | * messages about anything out of the ordinary. |
26845c28 PM |
70 | */ |
71 | static void __init rcu_bootup_announce_oddness(void) | |
72 | { | |
ab6f5bd6 | 73 | if (IS_ENABLED(CONFIG_RCU_TRACE)) |
ae91aa0a | 74 | pr_info("\tRCU event tracing is enabled.\n"); |
05c5df31 PM |
75 | if ((IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 64) || |
76 | (!IS_ENABLED(CONFIG_64BIT) && RCU_FANOUT != 32)) | |
ab6f5bd6 | 77 | pr_info("\tCONFIG_RCU_FANOUT set to non-default value of %d\n", |
05c5df31 | 78 | RCU_FANOUT); |
7fa27001 | 79 | if (rcu_fanout_exact) |
ab6f5bd6 PM |
80 | pr_info("\tHierarchical RCU autobalancing is disabled.\n"); |
81 | if (IS_ENABLED(CONFIG_RCU_FAST_NO_HZ)) | |
82 | pr_info("\tRCU dyntick-idle grace-period acceleration is enabled.\n"); | |
c4a09ff7 | 83 | if (IS_ENABLED(CONFIG_PROVE_RCU)) |
ab6f5bd6 | 84 | pr_info("\tRCU lockdep checking is enabled.\n"); |
42621697 AG |
85 | if (RCU_NUM_LVLS >= 4) |
86 | pr_info("\tFour(or more)-level hierarchy is enabled.\n"); | |
47d631af | 87 | if (RCU_FANOUT_LEAF != 16) |
a3bd2c09 | 88 | pr_info("\tBuild-time adjustment of leaf fanout to %d.\n", |
47d631af PM |
89 | RCU_FANOUT_LEAF); |
90 | if (rcu_fanout_leaf != RCU_FANOUT_LEAF) | |
9a5739d7 | 91 | pr_info("\tBoot-time adjustment of leaf fanout to %d.\n", rcu_fanout_leaf); |
cca6f393 | 92 | if (nr_cpu_ids != NR_CPUS) |
9b130ad5 | 93 | pr_info("\tRCU restricting CPUs from NR_CPUS=%d to nr_cpu_ids=%u.\n", NR_CPUS, nr_cpu_ids); |
17c7798b PM |
94 | #ifdef CONFIG_RCU_BOOST |
95 | pr_info("\tRCU priority boosting: priority %d delay %d ms.\n", kthread_prio, CONFIG_RCU_BOOST_DELAY); | |
96 | #endif | |
97 | if (blimit != DEFAULT_RCU_BLIMIT) | |
98 | pr_info("\tBoot-time adjustment of callback invocation limit to %ld.\n", blimit); | |
99 | if (qhimark != DEFAULT_RCU_QHIMARK) | |
100 | pr_info("\tBoot-time adjustment of callback high-water mark to %ld.\n", qhimark); | |
101 | if (qlowmark != DEFAULT_RCU_QLOMARK) | |
102 | pr_info("\tBoot-time adjustment of callback low-water mark to %ld.\n", qlowmark); | |
103 | if (jiffies_till_first_fqs != ULONG_MAX) | |
104 | pr_info("\tBoot-time adjustment of first FQS scan delay to %ld jiffies.\n", jiffies_till_first_fqs); | |
105 | if (jiffies_till_next_fqs != ULONG_MAX) | |
106 | pr_info("\tBoot-time adjustment of subsequent FQS scan delay to %ld jiffies.\n", jiffies_till_next_fqs); | |
107 | if (rcu_kick_kthreads) | |
108 | pr_info("\tKick kthreads if too-long grace period.\n"); | |
109 | if (IS_ENABLED(CONFIG_DEBUG_OBJECTS_RCU_HEAD)) | |
110 | pr_info("\tRCU callback double-/use-after-free debug enabled.\n"); | |
90040c9e | 111 | if (gp_preinit_delay) |
17c7798b | 112 | pr_info("\tRCU debug GP pre-init slowdown %d jiffies.\n", gp_preinit_delay); |
90040c9e | 113 | if (gp_init_delay) |
17c7798b | 114 | pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_init_delay); |
90040c9e | 115 | if (gp_cleanup_delay) |
17c7798b PM |
116 | pr_info("\tRCU debug GP init slowdown %d jiffies.\n", gp_cleanup_delay); |
117 | if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG)) | |
118 | pr_info("\tRCU debug extended QS entry/exit.\n"); | |
59d80fd8 | 119 | rcupdate_announce_bootup_oddness(); |
26845c28 PM |
120 | } |
121 | ||
28f6569a | 122 | #ifdef CONFIG_PREEMPT_RCU |
f41d911f | 123 | |
a41bfeb2 | 124 | RCU_STATE_INITIALIZER(rcu_preempt, 'p', call_rcu); |
b28a7c01 | 125 | static struct rcu_state *const rcu_state_p = &rcu_preempt_state; |
2927a689 | 126 | static struct rcu_data __percpu *const rcu_data_p = &rcu_preempt_data; |
f41d911f | 127 | |
d19fb8d1 PM |
128 | static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp, |
129 | bool wake); | |
3949fa9b | 130 | static void rcu_read_unlock_special(struct task_struct *t); |
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 | ||
a32e01ee | 184 | raw_lockdep_assert_held_rcu_node(rnp); |
2dee9404 | 185 | WARN_ON_ONCE(rdp->mynode != rnp); |
5b4c11d5 | 186 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp)); |
1f3e5f51 PM |
187 | /* RCU better not be waiting on newly onlined CPUs! */ |
188 | WARN_ON_ONCE(rnp->qsmaskinitnext & ~rnp->qsmaskinit & rnp->qsmask & | |
189 | rdp->grpmask); | |
ea9b0c8a | 190 | |
8203d6d0 PM |
191 | /* |
192 | * Decide where to queue the newly blocked task. In theory, | |
193 | * this could be an if-statement. In practice, when I tried | |
194 | * that, it was quite messy. | |
195 | */ | |
196 | switch (blkd_state) { | |
197 | case 0: | |
198 | case RCU_EXP_TASKS: | |
199 | case RCU_EXP_TASKS + RCU_GP_BLKD: | |
200 | case RCU_GP_TASKS: | |
201 | case RCU_GP_TASKS + RCU_EXP_TASKS: | |
202 | ||
203 | /* | |
204 | * Blocking neither GP, or first task blocking the normal | |
205 | * GP but not blocking the already-waiting expedited GP. | |
206 | * Queue at the head of the list to avoid unnecessarily | |
207 | * blocking the already-waiting GPs. | |
208 | */ | |
209 | list_add(&t->rcu_node_entry, &rnp->blkd_tasks); | |
210 | break; | |
211 | ||
212 | case RCU_EXP_BLKD: | |
213 | case RCU_GP_BLKD: | |
214 | case RCU_GP_BLKD + RCU_EXP_BLKD: | |
215 | case RCU_GP_TASKS + RCU_EXP_BLKD: | |
216 | case RCU_GP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
217 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
218 | ||
219 | /* | |
220 | * First task arriving that blocks either GP, or first task | |
221 | * arriving that blocks the expedited GP (with the normal | |
222 | * GP already waiting), or a task arriving that blocks | |
223 | * both GPs with both GPs already waiting. Queue at the | |
224 | * tail of the list to avoid any GP waiting on any of the | |
225 | * already queued tasks that are not blocking it. | |
226 | */ | |
227 | list_add_tail(&t->rcu_node_entry, &rnp->blkd_tasks); | |
228 | break; | |
229 | ||
230 | case RCU_EXP_TASKS + RCU_EXP_BLKD: | |
231 | case RCU_EXP_TASKS + RCU_GP_BLKD + RCU_EXP_BLKD: | |
232 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_EXP_BLKD: | |
233 | ||
234 | /* | |
235 | * Second or subsequent task blocking the expedited GP. | |
236 | * The task either does not block the normal GP, or is the | |
237 | * first task blocking the normal GP. Queue just after | |
238 | * the first task blocking the expedited GP. | |
239 | */ | |
240 | list_add(&t->rcu_node_entry, rnp->exp_tasks); | |
241 | break; | |
242 | ||
243 | case RCU_GP_TASKS + RCU_GP_BLKD: | |
244 | case RCU_GP_TASKS + RCU_EXP_TASKS + RCU_GP_BLKD: | |
245 | ||
246 | /* | |
247 | * Second or subsequent task blocking the normal GP. | |
248 | * The task does not block the expedited GP. Queue just | |
249 | * after the first task blocking the normal GP. | |
250 | */ | |
251 | list_add(&t->rcu_node_entry, rnp->gp_tasks); | |
252 | break; | |
253 | ||
254 | default: | |
255 | ||
256 | /* Yet another exercise in excessive paranoia. */ | |
257 | WARN_ON_ONCE(1); | |
258 | break; | |
259 | } | |
260 | ||
261 | /* | |
262 | * We have now queued the task. If it was the first one to | |
263 | * block either grace period, update the ->gp_tasks and/or | |
264 | * ->exp_tasks pointers, respectively, to reference the newly | |
265 | * blocked tasks. | |
266 | */ | |
4bc8d555 | 267 | if (!rnp->gp_tasks && (blkd_state & RCU_GP_BLKD)) { |
8203d6d0 | 268 | rnp->gp_tasks = &t->rcu_node_entry; |
d43a5d32 | 269 | WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq); |
4bc8d555 | 270 | } |
8203d6d0 PM |
271 | if (!rnp->exp_tasks && (blkd_state & RCU_EXP_BLKD)) |
272 | rnp->exp_tasks = &t->rcu_node_entry; | |
2dee9404 PM |
273 | WARN_ON_ONCE(!(blkd_state & RCU_GP_BLKD) != |
274 | !(rnp->qsmask & rdp->grpmask)); | |
275 | WARN_ON_ONCE(!(blkd_state & RCU_EXP_BLKD) != | |
276 | !(rnp->expmask & rdp->grpmask)); | |
67c583a7 | 277 | raw_spin_unlock_rcu_node(rnp); /* interrupts remain disabled. */ |
8203d6d0 PM |
278 | |
279 | /* | |
280 | * Report the quiescent state for the expedited GP. This expedited | |
281 | * GP should not be able to end until we report, so there should be | |
282 | * no need to check for a subsequent expedited GP. (Though we are | |
283 | * still in a quiescent state in any case.) | |
284 | */ | |
285 | if (blkd_state & RCU_EXP_BLKD && | |
286 | t->rcu_read_unlock_special.b.exp_need_qs) { | |
287 | t->rcu_read_unlock_special.b.exp_need_qs = false; | |
288 | rcu_report_exp_rdp(rdp->rsp, rdp, true); | |
289 | } else { | |
290 | WARN_ON_ONCE(t->rcu_read_unlock_special.b.exp_need_qs); | |
291 | } | |
8203d6d0 PM |
292 | } |
293 | ||
f41d911f | 294 | /* |
6cc68793 | 295 | * Record a preemptible-RCU quiescent state for the specified CPU. Note |
f41d911f PM |
296 | * that this just means that the task currently running on the CPU is |
297 | * not in a quiescent state. There might be any number of tasks blocked | |
298 | * while in an RCU read-side critical section. | |
25502a6c | 299 | * |
1d082fd0 PM |
300 | * As with the other rcu_*_qs() functions, callers to this function |
301 | * must disable preemption. | |
f41d911f | 302 | */ |
284a8c93 | 303 | static void rcu_preempt_qs(void) |
f41d911f | 304 | { |
ea9b0c8a | 305 | RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_qs() invoked with preemption enabled!!!\n"); |
5b74c458 | 306 | if (__this_cpu_read(rcu_data_p->cpu_no_qs.s)) { |
284a8c93 | 307 | trace_rcu_grace_period(TPS("rcu_preempt"), |
477351f7 | 308 | __this_cpu_read(rcu_data_p->gp_seq), |
284a8c93 | 309 | TPS("cpuqs")); |
5b74c458 | 310 | __this_cpu_write(rcu_data_p->cpu_no_qs.b.norm, false); |
284a8c93 PM |
311 | barrier(); /* Coordinate with rcu_preempt_check_callbacks(). */ |
312 | current->rcu_read_unlock_special.b.need_qs = false; | |
313 | } | |
f41d911f PM |
314 | } |
315 | ||
316 | /* | |
c3422bea PM |
317 | * We have entered the scheduler, and the current task might soon be |
318 | * context-switched away from. If this task is in an RCU read-side | |
319 | * critical section, we will no longer be able to rely on the CPU to | |
12f5f524 PM |
320 | * record that fact, so we enqueue the task on the blkd_tasks list. |
321 | * The task will dequeue itself when it exits the outermost enclosing | |
322 | * RCU read-side critical section. Therefore, the current grace period | |
323 | * cannot be permitted to complete until the blkd_tasks list entries | |
324 | * predating the current grace period drain, in other words, until | |
325 | * rnp->gp_tasks becomes NULL. | |
c3422bea | 326 | * |
46a5d164 | 327 | * Caller must disable interrupts. |
f41d911f | 328 | */ |
5b72f964 | 329 | static void rcu_preempt_note_context_switch(bool preempt) |
f41d911f PM |
330 | { |
331 | struct task_struct *t = current; | |
f41d911f PM |
332 | struct rcu_data *rdp; |
333 | struct rcu_node *rnp; | |
334 | ||
b04db8e1 | 335 | lockdep_assert_irqs_disabled(); |
5b72f964 | 336 | WARN_ON_ONCE(!preempt && t->rcu_read_lock_nesting > 0); |
10f39bb1 | 337 | if (t->rcu_read_lock_nesting > 0 && |
1d082fd0 | 338 | !t->rcu_read_unlock_special.b.blocked) { |
f41d911f PM |
339 | |
340 | /* Possibly blocking in an RCU read-side critical section. */ | |
e63c887c | 341 | rdp = this_cpu_ptr(rcu_state_p->rda); |
f41d911f | 342 | rnp = rdp->mynode; |
46a5d164 | 343 | raw_spin_lock_rcu_node(rnp); |
1d082fd0 | 344 | t->rcu_read_unlock_special.b.blocked = true; |
86848966 | 345 | t->rcu_blocked_node = rnp; |
f41d911f PM |
346 | |
347 | /* | |
8203d6d0 PM |
348 | * Verify the CPU's sanity, trace the preemption, and |
349 | * then queue the task as required based on the states | |
350 | * of any ongoing and expedited grace periods. | |
f41d911f | 351 | */ |
0aa04b05 | 352 | WARN_ON_ONCE((rdp->grpmask & rcu_rnp_online_cpus(rnp)) == 0); |
e7d8842e | 353 | WARN_ON_ONCE(!list_empty(&t->rcu_node_entry)); |
d4c08f2a PM |
354 | trace_rcu_preempt_task(rdp->rsp->name, |
355 | t->pid, | |
356 | (rnp->qsmask & rdp->grpmask) | |
598ce094 PM |
357 | ? rnp->gp_seq |
358 | : rcu_seq_snap(&rnp->gp_seq)); | |
46a5d164 | 359 | rcu_preempt_ctxt_queue(rnp, rdp); |
10f39bb1 | 360 | } else if (t->rcu_read_lock_nesting < 0 && |
1d082fd0 | 361 | t->rcu_read_unlock_special.s) { |
10f39bb1 PM |
362 | |
363 | /* | |
364 | * Complete exit from RCU read-side critical section on | |
365 | * behalf of preempted instance of __rcu_read_unlock(). | |
366 | */ | |
367 | rcu_read_unlock_special(t); | |
f41d911f PM |
368 | } |
369 | ||
370 | /* | |
371 | * Either we were not in an RCU read-side critical section to | |
372 | * begin with, or we have now recorded that critical section | |
373 | * globally. Either way, we can now note a quiescent state | |
374 | * for this CPU. Again, if we were in an RCU read-side critical | |
375 | * section, and if that critical section was blocking the current | |
376 | * grace period, then the fact that the task has been enqueued | |
377 | * means that we continue to block the current grace period. | |
378 | */ | |
284a8c93 | 379 | rcu_preempt_qs(); |
f41d911f PM |
380 | } |
381 | ||
fc2219d4 PM |
382 | /* |
383 | * Check for preempted RCU readers blocking the current grace period | |
384 | * for the specified rcu_node structure. If the caller needs a reliable | |
385 | * answer, it must hold the rcu_node's ->lock. | |
386 | */ | |
27f4d280 | 387 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 | 388 | { |
12f5f524 | 389 | return rnp->gp_tasks != NULL; |
fc2219d4 PM |
390 | } |
391 | ||
0e5da22e PM |
392 | /* |
393 | * Preemptible RCU implementation for rcu_read_lock(). | |
394 | * Just increment ->rcu_read_lock_nesting, shared state will be updated | |
395 | * if we block. | |
396 | */ | |
397 | void __rcu_read_lock(void) | |
398 | { | |
399 | current->rcu_read_lock_nesting++; | |
400 | barrier(); /* critical section after entry code. */ | |
401 | } | |
402 | EXPORT_SYMBOL_GPL(__rcu_read_lock); | |
403 | ||
404 | /* | |
405 | * Preemptible RCU implementation for rcu_read_unlock(). | |
406 | * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost | |
407 | * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then | |
408 | * invoke rcu_read_unlock_special() to clean up after a context switch | |
409 | * in an RCU read-side critical section and other special cases. | |
410 | */ | |
411 | void __rcu_read_unlock(void) | |
412 | { | |
413 | struct task_struct *t = current; | |
414 | ||
415 | if (t->rcu_read_lock_nesting != 1) { | |
416 | --t->rcu_read_lock_nesting; | |
417 | } else { | |
418 | barrier(); /* critical section before exit code. */ | |
419 | t->rcu_read_lock_nesting = INT_MIN; | |
420 | barrier(); /* assign before ->rcu_read_unlock_special load */ | |
421 | if (unlikely(READ_ONCE(t->rcu_read_unlock_special.s))) | |
422 | rcu_read_unlock_special(t); | |
423 | barrier(); /* ->rcu_read_unlock_special load before assign */ | |
424 | t->rcu_read_lock_nesting = 0; | |
425 | } | |
426 | #ifdef CONFIG_PROVE_LOCKING | |
427 | { | |
428 | int rrln = READ_ONCE(t->rcu_read_lock_nesting); | |
429 | ||
430 | WARN_ON_ONCE(rrln < 0 && rrln > INT_MIN / 2); | |
431 | } | |
432 | #endif /* #ifdef CONFIG_PROVE_LOCKING */ | |
433 | } | |
434 | EXPORT_SYMBOL_GPL(__rcu_read_unlock); | |
435 | ||
12f5f524 PM |
436 | /* |
437 | * Advance a ->blkd_tasks-list pointer to the next entry, instead | |
438 | * returning NULL if at the end of the list. | |
439 | */ | |
440 | static struct list_head *rcu_next_node_entry(struct task_struct *t, | |
441 | struct rcu_node *rnp) | |
442 | { | |
443 | struct list_head *np; | |
444 | ||
445 | np = t->rcu_node_entry.next; | |
446 | if (np == &rnp->blkd_tasks) | |
447 | np = NULL; | |
448 | return np; | |
449 | } | |
450 | ||
8af3a5e7 PM |
451 | /* |
452 | * Return true if the specified rcu_node structure has tasks that were | |
453 | * preempted within an RCU read-side critical section. | |
454 | */ | |
455 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
456 | { | |
457 | return !list_empty(&rnp->blkd_tasks); | |
458 | } | |
459 | ||
b668c9cf PM |
460 | /* |
461 | * Handle special cases during rcu_read_unlock(), such as needing to | |
462 | * notify RCU core processing or task having blocked during the RCU | |
463 | * read-side critical section. | |
464 | */ | |
3949fa9b | 465 | static void rcu_read_unlock_special(struct task_struct *t) |
f41d911f | 466 | { |
b6a932d1 PM |
467 | bool empty_exp; |
468 | bool empty_norm; | |
469 | bool empty_exp_now; | |
f41d911f | 470 | unsigned long flags; |
12f5f524 | 471 | struct list_head *np; |
abaa93d9 | 472 | bool drop_boost_mutex = false; |
8203d6d0 | 473 | struct rcu_data *rdp; |
f41d911f | 474 | struct rcu_node *rnp; |
1d082fd0 | 475 | union rcu_special special; |
f41d911f PM |
476 | |
477 | /* NMI handlers cannot block and cannot safely manipulate state. */ | |
478 | if (in_nmi()) | |
479 | return; | |
480 | ||
481 | local_irq_save(flags); | |
482 | ||
483 | /* | |
8203d6d0 PM |
484 | * If RCU core is waiting for this CPU to exit its critical section, |
485 | * report the fact that it has exited. Because irqs are disabled, | |
1d082fd0 | 486 | * t->rcu_read_unlock_special cannot change. |
f41d911f PM |
487 | */ |
488 | special = t->rcu_read_unlock_special; | |
1d082fd0 | 489 | if (special.b.need_qs) { |
284a8c93 | 490 | rcu_preempt_qs(); |
c0135d07 | 491 | t->rcu_read_unlock_special.b.need_qs = false; |
1d082fd0 | 492 | if (!t->rcu_read_unlock_special.s) { |
79a62f95 LJ |
493 | local_irq_restore(flags); |
494 | return; | |
495 | } | |
f41d911f PM |
496 | } |
497 | ||
8203d6d0 PM |
498 | /* |
499 | * Respond to a request for an expedited grace period, but only if | |
500 | * we were not preempted, meaning that we were running on the same | |
501 | * CPU throughout. If we were preempted, the exp_need_qs flag | |
502 | * would have been cleared at the time of the first preemption, | |
503 | * and the quiescent state would be reported when we were dequeued. | |
504 | */ | |
505 | if (special.b.exp_need_qs) { | |
506 | WARN_ON_ONCE(special.b.blocked); | |
507 | t->rcu_read_unlock_special.b.exp_need_qs = false; | |
508 | rdp = this_cpu_ptr(rcu_state_p->rda); | |
509 | rcu_report_exp_rdp(rcu_state_p, rdp, true); | |
510 | if (!t->rcu_read_unlock_special.s) { | |
511 | local_irq_restore(flags); | |
512 | return; | |
513 | } | |
514 | } | |
515 | ||
79a62f95 | 516 | /* Hardware IRQ handlers cannot block, complain if they get here. */ |
d24209bb PM |
517 | if (in_irq() || in_serving_softirq()) { |
518 | lockdep_rcu_suspicious(__FILE__, __LINE__, | |
519 | "rcu_read_unlock() from irq or softirq with blocking in critical section!!!\n"); | |
8203d6d0 | 520 | pr_alert("->rcu_read_unlock_special: %#x (b: %d, enq: %d nq: %d)\n", |
d24209bb PM |
521 | t->rcu_read_unlock_special.s, |
522 | t->rcu_read_unlock_special.b.blocked, | |
8203d6d0 | 523 | t->rcu_read_unlock_special.b.exp_need_qs, |
d24209bb | 524 | t->rcu_read_unlock_special.b.need_qs); |
f41d911f PM |
525 | local_irq_restore(flags); |
526 | return; | |
527 | } | |
528 | ||
529 | /* Clean up if blocked during RCU read-side critical section. */ | |
1d082fd0 PM |
530 | if (special.b.blocked) { |
531 | t->rcu_read_unlock_special.b.blocked = false; | |
f41d911f | 532 | |
dd5d19ba | 533 | /* |
0a0ba1c9 | 534 | * Remove this task from the list it blocked on. The task |
8ba9153b PM |
535 | * now remains queued on the rcu_node corresponding to the |
536 | * CPU it first blocked on, so there is no longer any need | |
537 | * to loop. Retain a WARN_ON_ONCE() out of sheer paranoia. | |
dd5d19ba | 538 | */ |
8ba9153b PM |
539 | rnp = t->rcu_blocked_node; |
540 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
541 | WARN_ON_ONCE(rnp != t->rcu_blocked_node); | |
5b4c11d5 | 542 | WARN_ON_ONCE(!rcu_is_leaf_node(rnp)); |
74e871ac | 543 | empty_norm = !rcu_preempt_blocked_readers_cgp(rnp); |
d43a5d32 | 544 | WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq && |
4bc8d555 | 545 | (!empty_norm || rnp->qsmask)); |
8203d6d0 | 546 | empty_exp = sync_rcu_preempt_exp_done(rnp); |
d9a3da06 | 547 | smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */ |
12f5f524 | 548 | np = rcu_next_node_entry(t, rnp); |
f41d911f | 549 | list_del_init(&t->rcu_node_entry); |
82e78d80 | 550 | t->rcu_blocked_node = NULL; |
f7f7bac9 | 551 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt"), |
865aa1e0 | 552 | rnp->gp_seq, t->pid); |
12f5f524 PM |
553 | if (&t->rcu_node_entry == rnp->gp_tasks) |
554 | rnp->gp_tasks = np; | |
555 | if (&t->rcu_node_entry == rnp->exp_tasks) | |
556 | rnp->exp_tasks = np; | |
727b705b | 557 | if (IS_ENABLED(CONFIG_RCU_BOOST)) { |
727b705b PM |
558 | /* Snapshot ->boost_mtx ownership w/rnp->lock held. */ |
559 | drop_boost_mutex = rt_mutex_owner(&rnp->boost_mtx) == t; | |
2dee9404 PM |
560 | if (&t->rcu_node_entry == rnp->boost_tasks) |
561 | rnp->boost_tasks = np; | |
727b705b | 562 | } |
f41d911f PM |
563 | |
564 | /* | |
565 | * If this was the last task on the current list, and if | |
566 | * we aren't waiting on any CPUs, report the quiescent state. | |
389abd48 PM |
567 | * Note that rcu_report_unblock_qs_rnp() releases rnp->lock, |
568 | * so we must take a snapshot of the expedited state. | |
f41d911f | 569 | */ |
8203d6d0 | 570 | empty_exp_now = sync_rcu_preempt_exp_done(rnp); |
74e871ac | 571 | if (!empty_norm && !rcu_preempt_blocked_readers_cgp(rnp)) { |
f7f7bac9 | 572 | trace_rcu_quiescent_state_report(TPS("preempt_rcu"), |
db023296 | 573 | rnp->gp_seq, |
d4c08f2a PM |
574 | 0, rnp->qsmask, |
575 | rnp->level, | |
576 | rnp->grplo, | |
577 | rnp->grphi, | |
578 | !!rnp->gp_tasks); | |
e63c887c | 579 | rcu_report_unblock_qs_rnp(rcu_state_p, rnp, flags); |
c701d5d9 | 580 | } else { |
67c583a7 | 581 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
c701d5d9 | 582 | } |
d9a3da06 | 583 | |
27f4d280 | 584 | /* Unboost if we were boosted. */ |
727b705b | 585 | if (IS_ENABLED(CONFIG_RCU_BOOST) && drop_boost_mutex) |
02a7c234 | 586 | rt_mutex_futex_unlock(&rnp->boost_mtx); |
27f4d280 | 587 | |
d9a3da06 PM |
588 | /* |
589 | * If this was the last task on the expedited lists, | |
590 | * then we need to report up the rcu_node hierarchy. | |
591 | */ | |
389abd48 | 592 | if (!empty_exp && empty_exp_now) |
e63c887c | 593 | rcu_report_exp_rnp(rcu_state_p, rnp, true); |
b668c9cf PM |
594 | } else { |
595 | local_irq_restore(flags); | |
f41d911f | 596 | } |
f41d911f PM |
597 | } |
598 | ||
1ed509a2 PM |
599 | /* |
600 | * Dump detailed information for all tasks blocking the current RCU | |
601 | * grace period on the specified rcu_node structure. | |
602 | */ | |
603 | static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp) | |
604 | { | |
605 | unsigned long flags; | |
1ed509a2 PM |
606 | struct task_struct *t; |
607 | ||
6cf10081 | 608 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
5fd4dc06 | 609 | if (!rcu_preempt_blocked_readers_cgp(rnp)) { |
67c583a7 | 610 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
5fd4dc06 PM |
611 | return; |
612 | } | |
82efed06 | 613 | t = list_entry(rnp->gp_tasks->prev, |
12f5f524 | 614 | struct task_struct, rcu_node_entry); |
3caa973b TH |
615 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) { |
616 | /* | |
617 | * We could be printing a lot while holding a spinlock. | |
618 | * Avoid triggering hard lockup. | |
619 | */ | |
620 | touch_nmi_watchdog(); | |
12f5f524 | 621 | sched_show_task(t); |
3caa973b | 622 | } |
67c583a7 | 623 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
1ed509a2 PM |
624 | } |
625 | ||
626 | /* | |
627 | * Dump detailed information for all tasks blocking the current RCU | |
628 | * grace period. | |
629 | */ | |
630 | static void rcu_print_detail_task_stall(struct rcu_state *rsp) | |
631 | { | |
632 | struct rcu_node *rnp = rcu_get_root(rsp); | |
633 | ||
634 | rcu_print_detail_task_stall_rnp(rnp); | |
635 | rcu_for_each_leaf_node(rsp, rnp) | |
636 | rcu_print_detail_task_stall_rnp(rnp); | |
637 | } | |
638 | ||
a858af28 PM |
639 | static void rcu_print_task_stall_begin(struct rcu_node *rnp) |
640 | { | |
efc151c3 | 641 | pr_err("\tTasks blocked on level-%d rcu_node (CPUs %d-%d):", |
a858af28 PM |
642 | rnp->level, rnp->grplo, rnp->grphi); |
643 | } | |
644 | ||
645 | static void rcu_print_task_stall_end(void) | |
646 | { | |
efc151c3 | 647 | pr_cont("\n"); |
a858af28 PM |
648 | } |
649 | ||
f41d911f PM |
650 | /* |
651 | * Scan the current list of tasks blocked within RCU read-side critical | |
652 | * sections, printing out the tid of each. | |
653 | */ | |
9bc8b558 | 654 | static int rcu_print_task_stall(struct rcu_node *rnp) |
f41d911f | 655 | { |
f41d911f | 656 | struct task_struct *t; |
9bc8b558 | 657 | int ndetected = 0; |
f41d911f | 658 | |
27f4d280 | 659 | if (!rcu_preempt_blocked_readers_cgp(rnp)) |
9bc8b558 | 660 | return 0; |
a858af28 | 661 | rcu_print_task_stall_begin(rnp); |
82efed06 | 662 | t = list_entry(rnp->gp_tasks->prev, |
12f5f524 | 663 | struct task_struct, rcu_node_entry); |
9bc8b558 | 664 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) { |
efc151c3 | 665 | pr_cont(" P%d", t->pid); |
9bc8b558 PM |
666 | ndetected++; |
667 | } | |
a858af28 | 668 | rcu_print_task_stall_end(); |
9bc8b558 | 669 | return ndetected; |
f41d911f PM |
670 | } |
671 | ||
74611ecb PM |
672 | /* |
673 | * Scan the current list of tasks blocked within RCU read-side critical | |
674 | * sections, printing out the tid of each that is blocking the current | |
675 | * expedited grace period. | |
676 | */ | |
677 | static int rcu_print_task_exp_stall(struct rcu_node *rnp) | |
678 | { | |
679 | struct task_struct *t; | |
680 | int ndetected = 0; | |
681 | ||
682 | if (!rnp->exp_tasks) | |
683 | return 0; | |
684 | t = list_entry(rnp->exp_tasks->prev, | |
685 | struct task_struct, rcu_node_entry); | |
686 | list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) { | |
687 | pr_cont(" P%d", t->pid); | |
688 | ndetected++; | |
689 | } | |
690 | return ndetected; | |
691 | } | |
692 | ||
b0e165c0 PM |
693 | /* |
694 | * Check that the list of blocked tasks for the newly completed grace | |
695 | * period is in fact empty. It is a serious bug to complete a grace | |
696 | * period that still has RCU readers blocked! This function must be | |
ff3bb6f4 | 697 | * invoked -before- updating this rnp's ->gp_seq, and the rnp's ->lock |
b0e165c0 | 698 | * must be held by the caller. |
12f5f524 PM |
699 | * |
700 | * Also, if there are blocked tasks on the list, they automatically | |
701 | * block the newly created grace period, so set up ->gp_tasks accordingly. | |
b0e165c0 | 702 | */ |
57738942 PM |
703 | static void |
704 | rcu_preempt_check_blocked_tasks(struct rcu_state *rsp, struct rcu_node *rnp) | |
b0e165c0 | 705 | { |
c5ebe66c PM |
706 | struct task_struct *t; |
707 | ||
ea9b0c8a | 708 | RCU_LOCKDEP_WARN(preemptible(), "rcu_preempt_check_blocked_tasks() invoked with preemption enabled!!!\n"); |
4bc8d555 | 709 | if (WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp))) |
57738942 | 710 | dump_blkd_tasks(rsp, rnp, 10); |
0b107d24 PM |
711 | if (rcu_preempt_has_tasks(rnp) && |
712 | (rnp->qsmaskinit || rnp->wait_blkd_tasks)) { | |
12f5f524 | 713 | rnp->gp_tasks = rnp->blkd_tasks.next; |
c5ebe66c PM |
714 | t = container_of(rnp->gp_tasks, struct task_struct, |
715 | rcu_node_entry); | |
716 | trace_rcu_unlock_preempted_task(TPS("rcu_preempt-GPS"), | |
865aa1e0 | 717 | rnp->gp_seq, t->pid); |
c5ebe66c | 718 | } |
28ecd580 | 719 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
720 | } |
721 | ||
f41d911f PM |
722 | /* |
723 | * Check for a quiescent state from the current CPU. When a task blocks, | |
724 | * the task is recorded in the corresponding CPU's rcu_node structure, | |
725 | * which is checked elsewhere. | |
726 | * | |
727 | * Caller must disable hard irqs. | |
728 | */ | |
86aea0e6 | 729 | static void rcu_preempt_check_callbacks(void) |
f41d911f PM |
730 | { |
731 | struct task_struct *t = current; | |
732 | ||
733 | if (t->rcu_read_lock_nesting == 0) { | |
284a8c93 | 734 | rcu_preempt_qs(); |
f41d911f PM |
735 | return; |
736 | } | |
10f39bb1 | 737 | if (t->rcu_read_lock_nesting > 0 && |
97c668b8 | 738 | __this_cpu_read(rcu_data_p->core_needs_qs) && |
5b74c458 | 739 | __this_cpu_read(rcu_data_p->cpu_no_qs.b.norm)) |
1d082fd0 | 740 | t->rcu_read_unlock_special.b.need_qs = true; |
f41d911f PM |
741 | } |
742 | ||
a68a2bb2 PM |
743 | /** |
744 | * call_rcu() - Queue an RCU callback for invocation after a grace period. | |
745 | * @head: structure to be used for queueing the RCU updates. | |
746 | * @func: actual callback function to be invoked after the grace period | |
747 | * | |
748 | * The callback function will be invoked some time after a full grace | |
749 | * period elapses, in other words after all pre-existing RCU read-side | |
750 | * critical sections have completed. However, the callback function | |
751 | * might well execute concurrently with RCU read-side critical sections | |
752 | * that started after call_rcu() was invoked. RCU read-side critical | |
753 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), | |
754 | * and may be nested. | |
755 | * | |
756 | * Note that all CPUs must agree that the grace period extended beyond | |
757 | * all pre-existing RCU read-side critical section. On systems with more | |
758 | * than one CPU, this means that when "func()" is invoked, each CPU is | |
759 | * guaranteed to have executed a full memory barrier since the end of its | |
760 | * last RCU read-side critical section whose beginning preceded the call | |
761 | * to call_rcu(). It also means that each CPU executing an RCU read-side | |
762 | * critical section that continues beyond the start of "func()" must have | |
763 | * executed a memory barrier after the call_rcu() but before the beginning | |
764 | * of that RCU read-side critical section. Note that these guarantees | |
765 | * include CPUs that are offline, idle, or executing in user mode, as | |
766 | * well as CPUs that are executing in the kernel. | |
767 | * | |
768 | * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the | |
769 | * resulting RCU callback function "func()", then both CPU A and CPU B are | |
770 | * guaranteed to execute a full memory barrier during the time interval | |
771 | * between the call to call_rcu() and the invocation of "func()" -- even | |
772 | * if CPU A and CPU B are the same CPU (but again only if the system has | |
773 | * more than one CPU). | |
f41d911f | 774 | */ |
b6a4ae76 | 775 | void call_rcu(struct rcu_head *head, rcu_callback_t func) |
f41d911f | 776 | { |
e63c887c | 777 | __call_rcu(head, func, rcu_state_p, -1, 0); |
f41d911f PM |
778 | } |
779 | EXPORT_SYMBOL_GPL(call_rcu); | |
780 | ||
6ebb237b PM |
781 | /** |
782 | * synchronize_rcu - wait until a grace period has elapsed. | |
783 | * | |
784 | * Control will return to the caller some time after a full grace | |
785 | * period has elapsed, in other words after all currently executing RCU | |
77d8485a PM |
786 | * read-side critical sections have completed. Note, however, that |
787 | * upon return from synchronize_rcu(), the caller might well be executing | |
788 | * concurrently with new RCU read-side critical sections that began while | |
789 | * synchronize_rcu() was waiting. RCU read-side critical sections are | |
790 | * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested. | |
f0a0e6f2 | 791 | * |
e28371c8 PM |
792 | * See the description of synchronize_sched() for more detailed |
793 | * information on memory-ordering guarantees. However, please note | |
794 | * that -only- the memory-ordering guarantees apply. For example, | |
795 | * synchronize_rcu() is -not- guaranteed to wait on things like code | |
796 | * protected by preempt_disable(), instead, synchronize_rcu() is -only- | |
797 | * guaranteed to wait on RCU read-side critical sections, that is, sections | |
798 | * of code protected by rcu_read_lock(). | |
6ebb237b PM |
799 | */ |
800 | void synchronize_rcu(void) | |
801 | { | |
f78f5b90 PM |
802 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) || |
803 | lock_is_held(&rcu_lock_map) || | |
804 | lock_is_held(&rcu_sched_lock_map), | |
805 | "Illegal synchronize_rcu() in RCU read-side critical section"); | |
52d7e48b | 806 | if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE) |
6ebb237b | 807 | return; |
5afff48b | 808 | if (rcu_gp_is_expedited()) |
3705b88d AM |
809 | synchronize_rcu_expedited(); |
810 | else | |
811 | wait_rcu_gp(call_rcu); | |
6ebb237b PM |
812 | } |
813 | EXPORT_SYMBOL_GPL(synchronize_rcu); | |
814 | ||
e74f4c45 PM |
815 | /** |
816 | * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete. | |
f0a0e6f2 PM |
817 | * |
818 | * Note that this primitive does not necessarily wait for an RCU grace period | |
819 | * to complete. For example, if there are no RCU callbacks queued anywhere | |
820 | * in the system, then rcu_barrier() is within its rights to return | |
821 | * immediately, without waiting for anything, much less an RCU grace period. | |
e74f4c45 PM |
822 | */ |
823 | void rcu_barrier(void) | |
824 | { | |
e63c887c | 825 | _rcu_barrier(rcu_state_p); |
e74f4c45 PM |
826 | } |
827 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
828 | ||
1eba8f84 | 829 | /* |
6cc68793 | 830 | * Initialize preemptible RCU's state structures. |
1eba8f84 PM |
831 | */ |
832 | static void __init __rcu_init_preempt(void) | |
833 | { | |
a87f203e | 834 | rcu_init_one(rcu_state_p); |
1eba8f84 PM |
835 | } |
836 | ||
2439b696 PM |
837 | /* |
838 | * Check for a task exiting while in a preemptible-RCU read-side | |
839 | * critical section, clean up if so. No need to issue warnings, | |
840 | * as debug_check_no_locks_held() already does this if lockdep | |
841 | * is enabled. | |
842 | */ | |
843 | void exit_rcu(void) | |
844 | { | |
845 | struct task_struct *t = current; | |
846 | ||
847 | if (likely(list_empty(¤t->rcu_node_entry))) | |
848 | return; | |
849 | t->rcu_read_lock_nesting = 1; | |
850 | barrier(); | |
1d082fd0 | 851 | t->rcu_read_unlock_special.b.blocked = true; |
2439b696 PM |
852 | __rcu_read_unlock(); |
853 | } | |
854 | ||
4bc8d555 PM |
855 | /* |
856 | * Dump the blocked-tasks state, but limit the list dump to the | |
857 | * specified number of elements. | |
858 | */ | |
57738942 PM |
859 | static void |
860 | dump_blkd_tasks(struct rcu_state *rsp, struct rcu_node *rnp, int ncheck) | |
4bc8d555 | 861 | { |
57738942 | 862 | int cpu; |
4bc8d555 PM |
863 | int i; |
864 | struct list_head *lhp; | |
57738942 PM |
865 | bool onl; |
866 | struct rcu_data *rdp; | |
ff3cee39 | 867 | struct rcu_node *rnp1; |
4bc8d555 | 868 | |
ce11fae8 | 869 | raw_lockdep_assert_held_rcu_node(rnp); |
ff3cee39 | 870 | pr_info("%s: grp: %d-%d level: %d ->gp_seq %ld ->completedqs %ld\n", |
77cfc7bf | 871 | __func__, rnp->grplo, rnp->grphi, rnp->level, |
ff3cee39 PM |
872 | (long)rnp->gp_seq, (long)rnp->completedqs); |
873 | for (rnp1 = rnp; rnp1; rnp1 = rnp1->parent) | |
874 | pr_info("%s: %d:%d ->qsmask %#lx ->qsmaskinit %#lx ->qsmaskinitnext %#lx\n", | |
875 | __func__, rnp1->grplo, rnp1->grphi, rnp1->qsmask, rnp1->qsmaskinit, rnp1->qsmaskinitnext); | |
77cfc7bf PM |
876 | pr_info("%s: ->gp_tasks %p ->boost_tasks %p ->exp_tasks %p\n", |
877 | __func__, rnp->gp_tasks, rnp->boost_tasks, rnp->exp_tasks); | |
878 | pr_info("%s: ->blkd_tasks", __func__); | |
4bc8d555 PM |
879 | i = 0; |
880 | list_for_each(lhp, &rnp->blkd_tasks) { | |
881 | pr_cont(" %p", lhp); | |
882 | if (++i >= 10) | |
883 | break; | |
884 | } | |
885 | pr_cont("\n"); | |
57738942 PM |
886 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++) { |
887 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
888 | onl = !!(rdp->grpmask & rcu_rnp_online_cpus(rnp)); | |
889 | pr_info("\t%d: %c online: %ld(%d) offline: %ld(%d)\n", | |
890 | cpu, ".o"[onl], | |
891 | (long)rdp->rcu_onl_gp_seq, rdp->rcu_onl_gp_flags, | |
892 | (long)rdp->rcu_ofl_gp_seq, rdp->rcu_ofl_gp_flags); | |
893 | } | |
4bc8d555 PM |
894 | } |
895 | ||
28f6569a | 896 | #else /* #ifdef CONFIG_PREEMPT_RCU */ |
f41d911f | 897 | |
b28a7c01 | 898 | static struct rcu_state *const rcu_state_p = &rcu_sched_state; |
27f4d280 | 899 | |
f41d911f PM |
900 | /* |
901 | * Tell them what RCU they are running. | |
902 | */ | |
0e0fc1c2 | 903 | static void __init rcu_bootup_announce(void) |
f41d911f | 904 | { |
efc151c3 | 905 | pr_info("Hierarchical RCU implementation.\n"); |
26845c28 | 906 | rcu_bootup_announce_oddness(); |
f41d911f PM |
907 | } |
908 | ||
cba6d0d6 PM |
909 | /* |
910 | * Because preemptible RCU does not exist, we never have to check for | |
911 | * CPUs being in quiescent states. | |
912 | */ | |
5b72f964 | 913 | static void rcu_preempt_note_context_switch(bool preempt) |
cba6d0d6 PM |
914 | { |
915 | } | |
916 | ||
fc2219d4 | 917 | /* |
6cc68793 | 918 | * Because preemptible RCU does not exist, there are never any preempted |
fc2219d4 PM |
919 | * RCU readers. |
920 | */ | |
27f4d280 | 921 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp) |
fc2219d4 PM |
922 | { |
923 | return 0; | |
924 | } | |
925 | ||
8af3a5e7 PM |
926 | /* |
927 | * Because there is no preemptible RCU, there can be no readers blocked. | |
928 | */ | |
929 | static bool rcu_preempt_has_tasks(struct rcu_node *rnp) | |
b668c9cf | 930 | { |
8af3a5e7 | 931 | return false; |
b668c9cf PM |
932 | } |
933 | ||
1ed509a2 | 934 | /* |
6cc68793 | 935 | * Because preemptible RCU does not exist, we never have to check for |
1ed509a2 PM |
936 | * tasks blocked within RCU read-side critical sections. |
937 | */ | |
938 | static void rcu_print_detail_task_stall(struct rcu_state *rsp) | |
939 | { | |
940 | } | |
941 | ||
f41d911f | 942 | /* |
6cc68793 | 943 | * Because preemptible RCU does not exist, we never have to check for |
f41d911f PM |
944 | * tasks blocked within RCU read-side critical sections. |
945 | */ | |
9bc8b558 | 946 | static int rcu_print_task_stall(struct rcu_node *rnp) |
f41d911f | 947 | { |
9bc8b558 | 948 | return 0; |
f41d911f PM |
949 | } |
950 | ||
74611ecb PM |
951 | /* |
952 | * Because preemptible RCU does not exist, we never have to check for | |
953 | * tasks blocked within RCU read-side critical sections that are | |
954 | * blocking the current expedited grace period. | |
955 | */ | |
956 | static int rcu_print_task_exp_stall(struct rcu_node *rnp) | |
957 | { | |
958 | return 0; | |
959 | } | |
960 | ||
b0e165c0 | 961 | /* |
6cc68793 | 962 | * Because there is no preemptible RCU, there can be no readers blocked, |
49e29126 PM |
963 | * so there is no need to check for blocked tasks. So check only for |
964 | * bogus qsmask values. | |
b0e165c0 | 965 | */ |
57738942 PM |
966 | static void |
967 | rcu_preempt_check_blocked_tasks(struct rcu_state *rsp, struct rcu_node *rnp) | |
b0e165c0 | 968 | { |
49e29126 | 969 | WARN_ON_ONCE(rnp->qsmask); |
b0e165c0 PM |
970 | } |
971 | ||
f41d911f | 972 | /* |
6cc68793 | 973 | * Because preemptible RCU does not exist, it never has any callbacks |
f41d911f PM |
974 | * to check. |
975 | */ | |
86aea0e6 | 976 | static void rcu_preempt_check_callbacks(void) |
f41d911f PM |
977 | { |
978 | } | |
979 | ||
e74f4c45 | 980 | /* |
6cc68793 | 981 | * Because preemptible RCU does not exist, rcu_barrier() is just |
e74f4c45 PM |
982 | * another name for rcu_barrier_sched(). |
983 | */ | |
984 | void rcu_barrier(void) | |
985 | { | |
986 | rcu_barrier_sched(); | |
987 | } | |
988 | EXPORT_SYMBOL_GPL(rcu_barrier); | |
989 | ||
1eba8f84 | 990 | /* |
6cc68793 | 991 | * Because preemptible RCU does not exist, it need not be initialized. |
1eba8f84 PM |
992 | */ |
993 | static void __init __rcu_init_preempt(void) | |
994 | { | |
995 | } | |
996 | ||
2439b696 PM |
997 | /* |
998 | * Because preemptible RCU does not exist, tasks cannot possibly exit | |
999 | * while in preemptible RCU read-side critical sections. | |
1000 | */ | |
1001 | void exit_rcu(void) | |
1002 | { | |
1003 | } | |
1004 | ||
4bc8d555 PM |
1005 | /* |
1006 | * Dump the guaranteed-empty blocked-tasks state. Trust but verify. | |
1007 | */ | |
57738942 PM |
1008 | static void |
1009 | dump_blkd_tasks(struct rcu_state *rsp, struct rcu_node *rnp, int ncheck) | |
4bc8d555 PM |
1010 | { |
1011 | WARN_ON_ONCE(!list_empty(&rnp->blkd_tasks)); | |
1012 | } | |
1013 | ||
28f6569a | 1014 | #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ |
8bd93a2c | 1015 | |
27f4d280 PM |
1016 | #ifdef CONFIG_RCU_BOOST |
1017 | ||
5d01bbd1 TG |
1018 | static void rcu_wake_cond(struct task_struct *t, int status) |
1019 | { | |
1020 | /* | |
1021 | * If the thread is yielding, only wake it when this | |
1022 | * is invoked from idle | |
1023 | */ | |
1024 | if (status != RCU_KTHREAD_YIELDING || is_idle_task(current)) | |
1025 | wake_up_process(t); | |
1026 | } | |
1027 | ||
27f4d280 PM |
1028 | /* |
1029 | * Carry out RCU priority boosting on the task indicated by ->exp_tasks | |
1030 | * or ->boost_tasks, advancing the pointer to the next task in the | |
1031 | * ->blkd_tasks list. | |
1032 | * | |
1033 | * Note that irqs must be enabled: boosting the task can block. | |
1034 | * Returns 1 if there are more tasks needing to be boosted. | |
1035 | */ | |
1036 | static int rcu_boost(struct rcu_node *rnp) | |
1037 | { | |
1038 | unsigned long flags; | |
27f4d280 PM |
1039 | struct task_struct *t; |
1040 | struct list_head *tb; | |
1041 | ||
7d0ae808 PM |
1042 | if (READ_ONCE(rnp->exp_tasks) == NULL && |
1043 | READ_ONCE(rnp->boost_tasks) == NULL) | |
27f4d280 PM |
1044 | return 0; /* Nothing left to boost. */ |
1045 | ||
2a67e741 | 1046 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 PM |
1047 | |
1048 | /* | |
1049 | * Recheck under the lock: all tasks in need of boosting | |
1050 | * might exit their RCU read-side critical sections on their own. | |
1051 | */ | |
1052 | if (rnp->exp_tasks == NULL && rnp->boost_tasks == NULL) { | |
67c583a7 | 1053 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
1054 | return 0; |
1055 | } | |
1056 | ||
1057 | /* | |
1058 | * Preferentially boost tasks blocking expedited grace periods. | |
1059 | * This cannot starve the normal grace periods because a second | |
1060 | * expedited grace period must boost all blocked tasks, including | |
1061 | * those blocking the pre-existing normal grace period. | |
1062 | */ | |
bec06785 | 1063 | if (rnp->exp_tasks != NULL) |
27f4d280 | 1064 | tb = rnp->exp_tasks; |
bec06785 | 1065 | else |
27f4d280 PM |
1066 | tb = rnp->boost_tasks; |
1067 | ||
1068 | /* | |
1069 | * We boost task t by manufacturing an rt_mutex that appears to | |
1070 | * be held by task t. We leave a pointer to that rt_mutex where | |
1071 | * task t can find it, and task t will release the mutex when it | |
1072 | * exits its outermost RCU read-side critical section. Then | |
1073 | * simply acquiring this artificial rt_mutex will boost task | |
1074 | * t's priority. (Thanks to tglx for suggesting this approach!) | |
1075 | * | |
1076 | * Note that task t must acquire rnp->lock to remove itself from | |
1077 | * the ->blkd_tasks list, which it will do from exit() if from | |
1078 | * nowhere else. We therefore are guaranteed that task t will | |
1079 | * stay around at least until we drop rnp->lock. Note that | |
1080 | * rnp->lock also resolves races between our priority boosting | |
1081 | * and task t's exiting its outermost RCU read-side critical | |
1082 | * section. | |
1083 | */ | |
1084 | t = container_of(tb, struct task_struct, rcu_node_entry); | |
abaa93d9 | 1085 | rt_mutex_init_proxy_locked(&rnp->boost_mtx, t); |
67c583a7 | 1086 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
abaa93d9 PM |
1087 | /* Lock only for side effect: boosts task t's priority. */ |
1088 | rt_mutex_lock(&rnp->boost_mtx); | |
1089 | rt_mutex_unlock(&rnp->boost_mtx); /* Then keep lockdep happy. */ | |
27f4d280 | 1090 | |
7d0ae808 PM |
1091 | return READ_ONCE(rnp->exp_tasks) != NULL || |
1092 | READ_ONCE(rnp->boost_tasks) != NULL; | |
27f4d280 PM |
1093 | } |
1094 | ||
27f4d280 | 1095 | /* |
bc17ea10 | 1096 | * Priority-boosting kthread, one per leaf rcu_node. |
27f4d280 PM |
1097 | */ |
1098 | static int rcu_boost_kthread(void *arg) | |
1099 | { | |
1100 | struct rcu_node *rnp = (struct rcu_node *)arg; | |
1101 | int spincnt = 0; | |
1102 | int more2boost; | |
1103 | ||
f7f7bac9 | 1104 | trace_rcu_utilization(TPS("Start boost kthread@init")); |
27f4d280 | 1105 | for (;;) { |
d71df90e | 1106 | rnp->boost_kthread_status = RCU_KTHREAD_WAITING; |
f7f7bac9 | 1107 | trace_rcu_utilization(TPS("End boost kthread@rcu_wait")); |
08bca60a | 1108 | rcu_wait(rnp->boost_tasks || rnp->exp_tasks); |
f7f7bac9 | 1109 | trace_rcu_utilization(TPS("Start boost kthread@rcu_wait")); |
d71df90e | 1110 | rnp->boost_kthread_status = RCU_KTHREAD_RUNNING; |
27f4d280 PM |
1111 | more2boost = rcu_boost(rnp); |
1112 | if (more2boost) | |
1113 | spincnt++; | |
1114 | else | |
1115 | spincnt = 0; | |
1116 | if (spincnt > 10) { | |
5d01bbd1 | 1117 | rnp->boost_kthread_status = RCU_KTHREAD_YIELDING; |
f7f7bac9 | 1118 | trace_rcu_utilization(TPS("End boost kthread@rcu_yield")); |
5d01bbd1 | 1119 | schedule_timeout_interruptible(2); |
f7f7bac9 | 1120 | trace_rcu_utilization(TPS("Start boost kthread@rcu_yield")); |
27f4d280 PM |
1121 | spincnt = 0; |
1122 | } | |
1123 | } | |
1217ed1b | 1124 | /* NOTREACHED */ |
f7f7bac9 | 1125 | trace_rcu_utilization(TPS("End boost kthread@notreached")); |
27f4d280 PM |
1126 | return 0; |
1127 | } | |
1128 | ||
1129 | /* | |
1130 | * Check to see if it is time to start boosting RCU readers that are | |
1131 | * blocking the current grace period, and, if so, tell the per-rcu_node | |
1132 | * kthread to start boosting them. If there is an expedited grace | |
1133 | * period in progress, it is always time to boost. | |
1134 | * | |
b065a853 PM |
1135 | * The caller must hold rnp->lock, which this function releases. |
1136 | * The ->boost_kthread_task is immortal, so we don't need to worry | |
1137 | * about it going away. | |
27f4d280 | 1138 | */ |
1217ed1b | 1139 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1140 | __releases(rnp->lock) |
27f4d280 PM |
1141 | { |
1142 | struct task_struct *t; | |
1143 | ||
a32e01ee | 1144 | raw_lockdep_assert_held_rcu_node(rnp); |
0ea1f2eb | 1145 | if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) { |
67c583a7 | 1146 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 | 1147 | return; |
0ea1f2eb | 1148 | } |
27f4d280 PM |
1149 | if (rnp->exp_tasks != NULL || |
1150 | (rnp->gp_tasks != NULL && | |
1151 | rnp->boost_tasks == NULL && | |
1152 | rnp->qsmask == 0 && | |
1153 | ULONG_CMP_GE(jiffies, rnp->boost_time))) { | |
1154 | if (rnp->exp_tasks == NULL) | |
1155 | rnp->boost_tasks = rnp->gp_tasks; | |
67c583a7 | 1156 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 | 1157 | t = rnp->boost_kthread_task; |
5d01bbd1 TG |
1158 | if (t) |
1159 | rcu_wake_cond(t, rnp->boost_kthread_status); | |
1217ed1b | 1160 | } else { |
67c583a7 | 1161 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
1217ed1b | 1162 | } |
27f4d280 PM |
1163 | } |
1164 | ||
a46e0899 PM |
1165 | /* |
1166 | * Wake up the per-CPU kthread to invoke RCU callbacks. | |
1167 | */ | |
1168 | static void invoke_rcu_callbacks_kthread(void) | |
1169 | { | |
1170 | unsigned long flags; | |
1171 | ||
1172 | local_irq_save(flags); | |
1173 | __this_cpu_write(rcu_cpu_has_work, 1); | |
1eb52121 | 1174 | if (__this_cpu_read(rcu_cpu_kthread_task) != NULL && |
5d01bbd1 TG |
1175 | current != __this_cpu_read(rcu_cpu_kthread_task)) { |
1176 | rcu_wake_cond(__this_cpu_read(rcu_cpu_kthread_task), | |
1177 | __this_cpu_read(rcu_cpu_kthread_status)); | |
1178 | } | |
a46e0899 PM |
1179 | local_irq_restore(flags); |
1180 | } | |
1181 | ||
dff1672d PM |
1182 | /* |
1183 | * Is the current CPU running the RCU-callbacks kthread? | |
1184 | * Caller must have preemption disabled. | |
1185 | */ | |
1186 | static bool rcu_is_callbacks_kthread(void) | |
1187 | { | |
c9d4b0af | 1188 | return __this_cpu_read(rcu_cpu_kthread_task) == current; |
dff1672d PM |
1189 | } |
1190 | ||
27f4d280 PM |
1191 | #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000) |
1192 | ||
1193 | /* | |
1194 | * Do priority-boost accounting for the start of a new grace period. | |
1195 | */ | |
1196 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) | |
1197 | { | |
1198 | rnp->boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES; | |
1199 | } | |
1200 | ||
27f4d280 PM |
1201 | /* |
1202 | * Create an RCU-boost kthread for the specified node if one does not | |
1203 | * already exist. We only create this kthread for preemptible RCU. | |
1204 | * Returns zero if all is well, a negated errno otherwise. | |
1205 | */ | |
49fb4c62 | 1206 | static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp, |
0aa04b05 | 1207 | struct rcu_node *rnp) |
27f4d280 | 1208 | { |
5d01bbd1 | 1209 | int rnp_index = rnp - &rsp->node[0]; |
27f4d280 PM |
1210 | unsigned long flags; |
1211 | struct sched_param sp; | |
1212 | struct task_struct *t; | |
1213 | ||
e63c887c | 1214 | if (rcu_state_p != rsp) |
27f4d280 | 1215 | return 0; |
5d01bbd1 | 1216 | |
0aa04b05 | 1217 | if (!rcu_scheduler_fully_active || rcu_rnp_online_cpus(rnp) == 0) |
5d01bbd1 TG |
1218 | return 0; |
1219 | ||
a46e0899 | 1220 | rsp->boost = 1; |
27f4d280 PM |
1221 | if (rnp->boost_kthread_task != NULL) |
1222 | return 0; | |
1223 | t = kthread_create(rcu_boost_kthread, (void *)rnp, | |
5b61b0ba | 1224 | "rcub/%d", rnp_index); |
27f4d280 PM |
1225 | if (IS_ERR(t)) |
1226 | return PTR_ERR(t); | |
2a67e741 | 1227 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27f4d280 | 1228 | rnp->boost_kthread_task = t; |
67c583a7 | 1229 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
21871d7e | 1230 | sp.sched_priority = kthread_prio; |
27f4d280 | 1231 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); |
9a432736 | 1232 | wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */ |
27f4d280 PM |
1233 | return 0; |
1234 | } | |
1235 | ||
f8b7fc6b PM |
1236 | static void rcu_kthread_do_work(void) |
1237 | { | |
c9d4b0af CL |
1238 | rcu_do_batch(&rcu_sched_state, this_cpu_ptr(&rcu_sched_data)); |
1239 | rcu_do_batch(&rcu_bh_state, this_cpu_ptr(&rcu_bh_data)); | |
be01b4ca | 1240 | rcu_do_batch(&rcu_preempt_state, this_cpu_ptr(&rcu_preempt_data)); |
f8b7fc6b PM |
1241 | } |
1242 | ||
62ab7072 | 1243 | static void rcu_cpu_kthread_setup(unsigned int cpu) |
f8b7fc6b | 1244 | { |
f8b7fc6b | 1245 | struct sched_param sp; |
f8b7fc6b | 1246 | |
21871d7e | 1247 | sp.sched_priority = kthread_prio; |
62ab7072 | 1248 | sched_setscheduler_nocheck(current, SCHED_FIFO, &sp); |
f8b7fc6b PM |
1249 | } |
1250 | ||
62ab7072 | 1251 | static void rcu_cpu_kthread_park(unsigned int cpu) |
f8b7fc6b | 1252 | { |
62ab7072 | 1253 | per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU; |
f8b7fc6b PM |
1254 | } |
1255 | ||
62ab7072 | 1256 | static int rcu_cpu_kthread_should_run(unsigned int cpu) |
f8b7fc6b | 1257 | { |
c9d4b0af | 1258 | return __this_cpu_read(rcu_cpu_has_work); |
f8b7fc6b PM |
1259 | } |
1260 | ||
1261 | /* | |
1262 | * Per-CPU kernel thread that invokes RCU callbacks. This replaces the | |
e0f23060 PM |
1263 | * RCU softirq used in flavors and configurations of RCU that do not |
1264 | * support RCU priority boosting. | |
f8b7fc6b | 1265 | */ |
62ab7072 | 1266 | static void rcu_cpu_kthread(unsigned int cpu) |
f8b7fc6b | 1267 | { |
c9d4b0af CL |
1268 | unsigned int *statusp = this_cpu_ptr(&rcu_cpu_kthread_status); |
1269 | char work, *workp = this_cpu_ptr(&rcu_cpu_has_work); | |
62ab7072 | 1270 | int spincnt; |
f8b7fc6b | 1271 | |
62ab7072 | 1272 | for (spincnt = 0; spincnt < 10; spincnt++) { |
f7f7bac9 | 1273 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_wait")); |
f8b7fc6b | 1274 | local_bh_disable(); |
f8b7fc6b | 1275 | *statusp = RCU_KTHREAD_RUNNING; |
62ab7072 PM |
1276 | this_cpu_inc(rcu_cpu_kthread_loops); |
1277 | local_irq_disable(); | |
f8b7fc6b PM |
1278 | work = *workp; |
1279 | *workp = 0; | |
62ab7072 | 1280 | local_irq_enable(); |
f8b7fc6b PM |
1281 | if (work) |
1282 | rcu_kthread_do_work(); | |
1283 | local_bh_enable(); | |
62ab7072 | 1284 | if (*workp == 0) { |
f7f7bac9 | 1285 | trace_rcu_utilization(TPS("End CPU kthread@rcu_wait")); |
62ab7072 PM |
1286 | *statusp = RCU_KTHREAD_WAITING; |
1287 | return; | |
f8b7fc6b PM |
1288 | } |
1289 | } | |
62ab7072 | 1290 | *statusp = RCU_KTHREAD_YIELDING; |
f7f7bac9 | 1291 | trace_rcu_utilization(TPS("Start CPU kthread@rcu_yield")); |
62ab7072 | 1292 | schedule_timeout_interruptible(2); |
f7f7bac9 | 1293 | trace_rcu_utilization(TPS("End CPU kthread@rcu_yield")); |
62ab7072 | 1294 | *statusp = RCU_KTHREAD_WAITING; |
f8b7fc6b PM |
1295 | } |
1296 | ||
1297 | /* | |
1298 | * Set the per-rcu_node kthread's affinity to cover all CPUs that are | |
1299 | * served by the rcu_node in question. The CPU hotplug lock is still | |
1300 | * held, so the value of rnp->qsmaskinit will be stable. | |
1301 | * | |
1302 | * We don't include outgoingcpu in the affinity set, use -1 if there is | |
1303 | * no outgoing CPU. If there are no CPUs left in the affinity set, | |
1304 | * this function allows the kthread to execute on any CPU. | |
1305 | */ | |
5d01bbd1 | 1306 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b | 1307 | { |
5d01bbd1 | 1308 | struct task_struct *t = rnp->boost_kthread_task; |
0aa04b05 | 1309 | unsigned long mask = rcu_rnp_online_cpus(rnp); |
f8b7fc6b PM |
1310 | cpumask_var_t cm; |
1311 | int cpu; | |
f8b7fc6b | 1312 | |
5d01bbd1 | 1313 | if (!t) |
f8b7fc6b | 1314 | return; |
5d01bbd1 | 1315 | if (!zalloc_cpumask_var(&cm, GFP_KERNEL)) |
f8b7fc6b | 1316 | return; |
bc75e999 MR |
1317 | for_each_leaf_node_possible_cpu(rnp, cpu) |
1318 | if ((mask & leaf_node_cpu_bit(rnp, cpu)) && | |
1319 | cpu != outgoingcpu) | |
f8b7fc6b | 1320 | cpumask_set_cpu(cpu, cm); |
5d0b0249 | 1321 | if (cpumask_weight(cm) == 0) |
f8b7fc6b | 1322 | cpumask_setall(cm); |
5d01bbd1 | 1323 | set_cpus_allowed_ptr(t, cm); |
f8b7fc6b PM |
1324 | free_cpumask_var(cm); |
1325 | } | |
1326 | ||
62ab7072 PM |
1327 | static struct smp_hotplug_thread rcu_cpu_thread_spec = { |
1328 | .store = &rcu_cpu_kthread_task, | |
1329 | .thread_should_run = rcu_cpu_kthread_should_run, | |
1330 | .thread_fn = rcu_cpu_kthread, | |
1331 | .thread_comm = "rcuc/%u", | |
1332 | .setup = rcu_cpu_kthread_setup, | |
1333 | .park = rcu_cpu_kthread_park, | |
1334 | }; | |
f8b7fc6b PM |
1335 | |
1336 | /* | |
9386c0b7 | 1337 | * Spawn boost kthreads -- called as soon as the scheduler is running. |
f8b7fc6b | 1338 | */ |
9386c0b7 | 1339 | static void __init rcu_spawn_boost_kthreads(void) |
f8b7fc6b | 1340 | { |
f8b7fc6b | 1341 | struct rcu_node *rnp; |
5d01bbd1 | 1342 | int cpu; |
f8b7fc6b | 1343 | |
62ab7072 | 1344 | for_each_possible_cpu(cpu) |
f8b7fc6b | 1345 | per_cpu(rcu_cpu_has_work, cpu) = 0; |
62ab7072 | 1346 | BUG_ON(smpboot_register_percpu_thread(&rcu_cpu_thread_spec)); |
3e9f5c70 PM |
1347 | rcu_for_each_leaf_node(rcu_state_p, rnp) |
1348 | (void)rcu_spawn_one_boost_kthread(rcu_state_p, rnp); | |
f8b7fc6b | 1349 | } |
f8b7fc6b | 1350 | |
49fb4c62 | 1351 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b | 1352 | { |
e534165b | 1353 | struct rcu_data *rdp = per_cpu_ptr(rcu_state_p->rda, cpu); |
f8b7fc6b PM |
1354 | struct rcu_node *rnp = rdp->mynode; |
1355 | ||
1356 | /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */ | |
62ab7072 | 1357 | if (rcu_scheduler_fully_active) |
e534165b | 1358 | (void)rcu_spawn_one_boost_kthread(rcu_state_p, rnp); |
f8b7fc6b PM |
1359 | } |
1360 | ||
27f4d280 PM |
1361 | #else /* #ifdef CONFIG_RCU_BOOST */ |
1362 | ||
1217ed1b | 1363 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) |
615e41c6 | 1364 | __releases(rnp->lock) |
27f4d280 | 1365 | { |
67c583a7 | 1366 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27f4d280 PM |
1367 | } |
1368 | ||
a46e0899 | 1369 | static void invoke_rcu_callbacks_kthread(void) |
27f4d280 | 1370 | { |
a46e0899 | 1371 | WARN_ON_ONCE(1); |
27f4d280 PM |
1372 | } |
1373 | ||
dff1672d PM |
1374 | static bool rcu_is_callbacks_kthread(void) |
1375 | { | |
1376 | return false; | |
1377 | } | |
1378 | ||
27f4d280 PM |
1379 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp) |
1380 | { | |
1381 | } | |
1382 | ||
5d01bbd1 | 1383 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
f8b7fc6b PM |
1384 | { |
1385 | } | |
1386 | ||
9386c0b7 | 1387 | static void __init rcu_spawn_boost_kthreads(void) |
b0d30417 | 1388 | { |
b0d30417 | 1389 | } |
b0d30417 | 1390 | |
49fb4c62 | 1391 | static void rcu_prepare_kthreads(int cpu) |
f8b7fc6b PM |
1392 | { |
1393 | } | |
1394 | ||
27f4d280 PM |
1395 | #endif /* #else #ifdef CONFIG_RCU_BOOST */ |
1396 | ||
8bd93a2c PM |
1397 | #if !defined(CONFIG_RCU_FAST_NO_HZ) |
1398 | ||
1399 | /* | |
1400 | * Check to see if any future RCU-related work will need to be done | |
1401 | * by the current CPU, even if none need be done immediately, returning | |
1402 | * 1 if so. This function is part of the RCU implementation; it is -not- | |
1403 | * an exported member of the RCU API. | |
1404 | * | |
7cb92499 PM |
1405 | * Because we not have RCU_FAST_NO_HZ, just check whether this CPU needs |
1406 | * any flavor of RCU. | |
8bd93a2c | 1407 | */ |
c1ad348b | 1408 | int rcu_needs_cpu(u64 basemono, u64 *nextevt) |
8bd93a2c | 1409 | { |
c1ad348b | 1410 | *nextevt = KTIME_MAX; |
44c65ff2 | 1411 | return rcu_cpu_has_callbacks(NULL); |
7cb92499 PM |
1412 | } |
1413 | ||
1414 | /* | |
1415 | * Because we do not have RCU_FAST_NO_HZ, don't bother cleaning up | |
1416 | * after it. | |
1417 | */ | |
8fa7845d | 1418 | static void rcu_cleanup_after_idle(void) |
7cb92499 PM |
1419 | { |
1420 | } | |
1421 | ||
aea1b35e | 1422 | /* |
a858af28 | 1423 | * Do the idle-entry grace-period work, which, because CONFIG_RCU_FAST_NO_HZ=n, |
aea1b35e PM |
1424 | * is nothing. |
1425 | */ | |
198bbf81 | 1426 | static void rcu_prepare_for_idle(void) |
aea1b35e PM |
1427 | { |
1428 | } | |
1429 | ||
c57afe80 PM |
1430 | /* |
1431 | * Don't bother keeping a running count of the number of RCU callbacks | |
1432 | * posted because CONFIG_RCU_FAST_NO_HZ=n. | |
1433 | */ | |
1434 | static void rcu_idle_count_callbacks_posted(void) | |
1435 | { | |
1436 | } | |
1437 | ||
8bd93a2c PM |
1438 | #else /* #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
1439 | ||
f23f7fa1 PM |
1440 | /* |
1441 | * This code is invoked when a CPU goes idle, at which point we want | |
1442 | * to have the CPU do everything required for RCU so that it can enter | |
1443 | * the energy-efficient dyntick-idle mode. This is handled by a | |
1444 | * state machine implemented by rcu_prepare_for_idle() below. | |
1445 | * | |
1446 | * The following three proprocessor symbols control this state machine: | |
1447 | * | |
f23f7fa1 PM |
1448 | * RCU_IDLE_GP_DELAY gives the number of jiffies that a CPU is permitted |
1449 | * to sleep in dyntick-idle mode with RCU callbacks pending. This | |
1450 | * is sized to be roughly one RCU grace period. Those energy-efficiency | |
1451 | * benchmarkers who might otherwise be tempted to set this to a large | |
1452 | * number, be warned: Setting RCU_IDLE_GP_DELAY too high can hang your | |
1453 | * system. And if you are -that- concerned about energy efficiency, | |
1454 | * just power the system down and be done with it! | |
778d250a PM |
1455 | * RCU_IDLE_LAZY_GP_DELAY gives the number of jiffies that a CPU is |
1456 | * permitted to sleep in dyntick-idle mode with only lazy RCU | |
1457 | * callbacks pending. Setting this too high can OOM your system. | |
f23f7fa1 PM |
1458 | * |
1459 | * The values below work well in practice. If future workloads require | |
1460 | * adjustment, they can be converted into kernel config parameters, though | |
1461 | * making the state machine smarter might be a better option. | |
1462 | */ | |
e84c48ae | 1463 | #define RCU_IDLE_GP_DELAY 4 /* Roughly one grace period. */ |
778d250a | 1464 | #define RCU_IDLE_LAZY_GP_DELAY (6 * HZ) /* Roughly six seconds. */ |
f23f7fa1 | 1465 | |
5e44ce35 PM |
1466 | static int rcu_idle_gp_delay = RCU_IDLE_GP_DELAY; |
1467 | module_param(rcu_idle_gp_delay, int, 0644); | |
1468 | static int rcu_idle_lazy_gp_delay = RCU_IDLE_LAZY_GP_DELAY; | |
1469 | module_param(rcu_idle_lazy_gp_delay, int, 0644); | |
486e2593 | 1470 | |
486e2593 | 1471 | /* |
c229828c PM |
1472 | * Try to advance callbacks for all flavors of RCU on the current CPU, but |
1473 | * only if it has been awhile since the last time we did so. Afterwards, | |
1474 | * if there are any callbacks ready for immediate invocation, return true. | |
486e2593 | 1475 | */ |
f1f399d1 | 1476 | static bool __maybe_unused rcu_try_advance_all_cbs(void) |
486e2593 | 1477 | { |
c0f4dfd4 PM |
1478 | bool cbs_ready = false; |
1479 | struct rcu_data *rdp; | |
c229828c | 1480 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
c0f4dfd4 PM |
1481 | struct rcu_node *rnp; |
1482 | struct rcu_state *rsp; | |
486e2593 | 1483 | |
c229828c PM |
1484 | /* Exit early if we advanced recently. */ |
1485 | if (jiffies == rdtp->last_advance_all) | |
d0bc90fd | 1486 | return false; |
c229828c PM |
1487 | rdtp->last_advance_all = jiffies; |
1488 | ||
c0f4dfd4 PM |
1489 | for_each_rcu_flavor(rsp) { |
1490 | rdp = this_cpu_ptr(rsp->rda); | |
1491 | rnp = rdp->mynode; | |
486e2593 | 1492 | |
c0f4dfd4 PM |
1493 | /* |
1494 | * Don't bother checking unless a grace period has | |
1495 | * completed since we last checked and there are | |
1496 | * callbacks not yet ready to invoke. | |
1497 | */ | |
03c8cb76 PM |
1498 | if ((rcu_seq_completed_gp(rdp->gp_seq, |
1499 | rcu_seq_current(&rnp->gp_seq)) || | |
7d0ae808 | 1500 | unlikely(READ_ONCE(rdp->gpwrap))) && |
15fecf89 | 1501 | rcu_segcblist_pend_cbs(&rdp->cblist)) |
470716fc | 1502 | note_gp_changes(rsp, rdp); |
486e2593 | 1503 | |
15fecf89 | 1504 | if (rcu_segcblist_ready_cbs(&rdp->cblist)) |
c0f4dfd4 PM |
1505 | cbs_ready = true; |
1506 | } | |
1507 | return cbs_ready; | |
486e2593 PM |
1508 | } |
1509 | ||
aa9b1630 | 1510 | /* |
c0f4dfd4 PM |
1511 | * Allow the CPU to enter dyntick-idle mode unless it has callbacks ready |
1512 | * to invoke. If the CPU has callbacks, try to advance them. Tell the | |
1513 | * caller to set the timeout based on whether or not there are non-lazy | |
1514 | * callbacks. | |
aa9b1630 | 1515 | * |
c0f4dfd4 | 1516 | * The caller must have disabled interrupts. |
aa9b1630 | 1517 | */ |
c1ad348b | 1518 | int rcu_needs_cpu(u64 basemono, u64 *nextevt) |
aa9b1630 | 1519 | { |
aa6da514 | 1520 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
c1ad348b | 1521 | unsigned long dj; |
aa9b1630 | 1522 | |
b04db8e1 | 1523 | lockdep_assert_irqs_disabled(); |
3382adbc | 1524 | |
c0f4dfd4 PM |
1525 | /* Snapshot to detect later posting of non-lazy callback. */ |
1526 | rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted; | |
1527 | ||
aa9b1630 | 1528 | /* If no callbacks, RCU doesn't need the CPU. */ |
aa6da514 | 1529 | if (!rcu_cpu_has_callbacks(&rdtp->all_lazy)) { |
c1ad348b | 1530 | *nextevt = KTIME_MAX; |
aa9b1630 PM |
1531 | return 0; |
1532 | } | |
c0f4dfd4 PM |
1533 | |
1534 | /* Attempt to advance callbacks. */ | |
1535 | if (rcu_try_advance_all_cbs()) { | |
1536 | /* Some ready to invoke, so initiate later invocation. */ | |
1537 | invoke_rcu_core(); | |
aa9b1630 PM |
1538 | return 1; |
1539 | } | |
c0f4dfd4 PM |
1540 | rdtp->last_accelerate = jiffies; |
1541 | ||
1542 | /* Request timer delay depending on laziness, and round. */ | |
6faf7283 | 1543 | if (!rdtp->all_lazy) { |
c1ad348b | 1544 | dj = round_up(rcu_idle_gp_delay + jiffies, |
c0f4dfd4 | 1545 | rcu_idle_gp_delay) - jiffies; |
e84c48ae | 1546 | } else { |
c1ad348b | 1547 | dj = round_jiffies(rcu_idle_lazy_gp_delay + jiffies) - jiffies; |
e84c48ae | 1548 | } |
c1ad348b | 1549 | *nextevt = basemono + dj * TICK_NSEC; |
aa9b1630 PM |
1550 | return 0; |
1551 | } | |
1552 | ||
21e52e15 | 1553 | /* |
c0f4dfd4 PM |
1554 | * Prepare a CPU for idle from an RCU perspective. The first major task |
1555 | * is to sense whether nohz mode has been enabled or disabled via sysfs. | |
1556 | * The second major task is to check to see if a non-lazy callback has | |
1557 | * arrived at a CPU that previously had only lazy callbacks. The third | |
1558 | * major task is to accelerate (that is, assign grace-period numbers to) | |
1559 | * any recently arrived callbacks. | |
aea1b35e PM |
1560 | * |
1561 | * The caller must have disabled interrupts. | |
8bd93a2c | 1562 | */ |
198bbf81 | 1563 | static void rcu_prepare_for_idle(void) |
8bd93a2c | 1564 | { |
48a7639c | 1565 | bool needwake; |
c0f4dfd4 | 1566 | struct rcu_data *rdp; |
198bbf81 | 1567 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
c0f4dfd4 PM |
1568 | struct rcu_node *rnp; |
1569 | struct rcu_state *rsp; | |
9d2ad243 PM |
1570 | int tne; |
1571 | ||
b04db8e1 | 1572 | lockdep_assert_irqs_disabled(); |
44c65ff2 | 1573 | if (rcu_is_nocb_cpu(smp_processor_id())) |
3382adbc PM |
1574 | return; |
1575 | ||
9d2ad243 | 1576 | /* Handle nohz enablement switches conservatively. */ |
7d0ae808 | 1577 | tne = READ_ONCE(tick_nohz_active); |
9d2ad243 | 1578 | if (tne != rdtp->tick_nohz_enabled_snap) { |
aa6da514 | 1579 | if (rcu_cpu_has_callbacks(NULL)) |
9d2ad243 PM |
1580 | invoke_rcu_core(); /* force nohz to see update. */ |
1581 | rdtp->tick_nohz_enabled_snap = tne; | |
1582 | return; | |
1583 | } | |
1584 | if (!tne) | |
1585 | return; | |
f511fc62 | 1586 | |
c57afe80 | 1587 | /* |
c0f4dfd4 PM |
1588 | * If a non-lazy callback arrived at a CPU having only lazy |
1589 | * callbacks, invoke RCU core for the side-effect of recalculating | |
1590 | * idle duration on re-entry to idle. | |
c57afe80 | 1591 | */ |
c0f4dfd4 PM |
1592 | if (rdtp->all_lazy && |
1593 | rdtp->nonlazy_posted != rdtp->nonlazy_posted_snap) { | |
c337f8f5 PM |
1594 | rdtp->all_lazy = false; |
1595 | rdtp->nonlazy_posted_snap = rdtp->nonlazy_posted; | |
c0f4dfd4 | 1596 | invoke_rcu_core(); |
c57afe80 PM |
1597 | return; |
1598 | } | |
c57afe80 | 1599 | |
3084f2f8 | 1600 | /* |
c0f4dfd4 PM |
1601 | * If we have not yet accelerated this jiffy, accelerate all |
1602 | * callbacks on this CPU. | |
3084f2f8 | 1603 | */ |
c0f4dfd4 | 1604 | if (rdtp->last_accelerate == jiffies) |
aea1b35e | 1605 | return; |
c0f4dfd4 PM |
1606 | rdtp->last_accelerate = jiffies; |
1607 | for_each_rcu_flavor(rsp) { | |
198bbf81 | 1608 | rdp = this_cpu_ptr(rsp->rda); |
135bd1a2 | 1609 | if (!rcu_segcblist_pend_cbs(&rdp->cblist)) |
c0f4dfd4 PM |
1610 | continue; |
1611 | rnp = rdp->mynode; | |
2a67e741 | 1612 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
48a7639c | 1613 | needwake = rcu_accelerate_cbs(rsp, rnp, rdp); |
67c583a7 | 1614 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
48a7639c PM |
1615 | if (needwake) |
1616 | rcu_gp_kthread_wake(rsp); | |
77e38ed3 | 1617 | } |
c0f4dfd4 | 1618 | } |
3084f2f8 | 1619 | |
c0f4dfd4 PM |
1620 | /* |
1621 | * Clean up for exit from idle. Attempt to advance callbacks based on | |
1622 | * any grace periods that elapsed while the CPU was idle, and if any | |
1623 | * callbacks are now ready to invoke, initiate invocation. | |
1624 | */ | |
8fa7845d | 1625 | static void rcu_cleanup_after_idle(void) |
c0f4dfd4 | 1626 | { |
b04db8e1 | 1627 | lockdep_assert_irqs_disabled(); |
44c65ff2 | 1628 | if (rcu_is_nocb_cpu(smp_processor_id())) |
aea1b35e | 1629 | return; |
7a497c96 PM |
1630 | if (rcu_try_advance_all_cbs()) |
1631 | invoke_rcu_core(); | |
8bd93a2c PM |
1632 | } |
1633 | ||
c57afe80 | 1634 | /* |
98248a0e PM |
1635 | * Keep a running count of the number of non-lazy callbacks posted |
1636 | * on this CPU. This running counter (which is never decremented) allows | |
1637 | * rcu_prepare_for_idle() to detect when something out of the idle loop | |
1638 | * posts a callback, even if an equal number of callbacks are invoked. | |
1639 | * Of course, callbacks should only be posted from within a trace event | |
1640 | * designed to be called from idle or from within RCU_NONIDLE(). | |
c57afe80 PM |
1641 | */ |
1642 | static void rcu_idle_count_callbacks_posted(void) | |
1643 | { | |
5955f7ee | 1644 | __this_cpu_add(rcu_dynticks.nonlazy_posted, 1); |
c57afe80 PM |
1645 | } |
1646 | ||
b626c1b6 PM |
1647 | /* |
1648 | * Data for flushing lazy RCU callbacks at OOM time. | |
1649 | */ | |
1650 | static atomic_t oom_callback_count; | |
1651 | static DECLARE_WAIT_QUEUE_HEAD(oom_callback_wq); | |
1652 | ||
1653 | /* | |
1654 | * RCU OOM callback -- decrement the outstanding count and deliver the | |
1655 | * wake-up if we are the last one. | |
1656 | */ | |
1657 | static void rcu_oom_callback(struct rcu_head *rhp) | |
1658 | { | |
1659 | if (atomic_dec_and_test(&oom_callback_count)) | |
1660 | wake_up(&oom_callback_wq); | |
1661 | } | |
1662 | ||
1663 | /* | |
1664 | * Post an rcu_oom_notify callback on the current CPU if it has at | |
1665 | * least one lazy callback. This will unnecessarily post callbacks | |
1666 | * to CPUs that already have a non-lazy callback at the end of their | |
1667 | * callback list, but this is an infrequent operation, so accept some | |
1668 | * extra overhead to keep things simple. | |
1669 | */ | |
1670 | static void rcu_oom_notify_cpu(void *unused) | |
1671 | { | |
1672 | struct rcu_state *rsp; | |
1673 | struct rcu_data *rdp; | |
1674 | ||
1675 | for_each_rcu_flavor(rsp) { | |
fa07a58f | 1676 | rdp = raw_cpu_ptr(rsp->rda); |
15fecf89 | 1677 | if (rcu_segcblist_n_lazy_cbs(&rdp->cblist)) { |
b626c1b6 PM |
1678 | atomic_inc(&oom_callback_count); |
1679 | rsp->call(&rdp->oom_head, rcu_oom_callback); | |
1680 | } | |
1681 | } | |
1682 | } | |
1683 | ||
1684 | /* | |
1685 | * If low on memory, ensure that each CPU has a non-lazy callback. | |
1686 | * This will wake up CPUs that have only lazy callbacks, in turn | |
1687 | * ensuring that they free up the corresponding memory in a timely manner. | |
1688 | * Because an uncertain amount of memory will be freed in some uncertain | |
1689 | * timeframe, we do not claim to have freed anything. | |
1690 | */ | |
1691 | static int rcu_oom_notify(struct notifier_block *self, | |
1692 | unsigned long notused, void *nfreed) | |
1693 | { | |
1694 | int cpu; | |
1695 | ||
1696 | /* Wait for callbacks from earlier instance to complete. */ | |
1697 | wait_event(oom_callback_wq, atomic_read(&oom_callback_count) == 0); | |
78e4bc34 | 1698 | smp_mb(); /* Ensure callback reuse happens after callback invocation. */ |
b626c1b6 PM |
1699 | |
1700 | /* | |
1701 | * Prevent premature wakeup: ensure that all increments happen | |
1702 | * before there is a chance of the counter reaching zero. | |
1703 | */ | |
1704 | atomic_set(&oom_callback_count, 1); | |
1705 | ||
b626c1b6 PM |
1706 | for_each_online_cpu(cpu) { |
1707 | smp_call_function_single(cpu, rcu_oom_notify_cpu, NULL, 1); | |
cee43939 | 1708 | cond_resched_tasks_rcu_qs(); |
b626c1b6 | 1709 | } |
b626c1b6 PM |
1710 | |
1711 | /* Unconditionally decrement: no need to wake ourselves up. */ | |
1712 | atomic_dec(&oom_callback_count); | |
1713 | ||
1714 | return NOTIFY_OK; | |
1715 | } | |
1716 | ||
1717 | static struct notifier_block rcu_oom_nb = { | |
1718 | .notifier_call = rcu_oom_notify | |
1719 | }; | |
1720 | ||
1721 | static int __init rcu_register_oom_notifier(void) | |
1722 | { | |
1723 | register_oom_notifier(&rcu_oom_nb); | |
1724 | return 0; | |
1725 | } | |
1726 | early_initcall(rcu_register_oom_notifier); | |
1727 | ||
8bd93a2c | 1728 | #endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */ |
a858af28 | 1729 | |
a858af28 PM |
1730 | #ifdef CONFIG_RCU_FAST_NO_HZ |
1731 | ||
1732 | static void print_cpu_stall_fast_no_hz(char *cp, int cpu) | |
1733 | { | |
5955f7ee | 1734 | struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); |
c0f4dfd4 | 1735 | unsigned long nlpd = rdtp->nonlazy_posted - rdtp->nonlazy_posted_snap; |
a858af28 | 1736 | |
c0f4dfd4 PM |
1737 | sprintf(cp, "last_accelerate: %04lx/%04lx, nonlazy_posted: %ld, %c%c", |
1738 | rdtp->last_accelerate & 0xffff, jiffies & 0xffff, | |
1739 | ulong2long(nlpd), | |
1740 | rdtp->all_lazy ? 'L' : '.', | |
1741 | rdtp->tick_nohz_enabled_snap ? '.' : 'D'); | |
a858af28 PM |
1742 | } |
1743 | ||
1744 | #else /* #ifdef CONFIG_RCU_FAST_NO_HZ */ | |
1745 | ||
1746 | static void print_cpu_stall_fast_no_hz(char *cp, int cpu) | |
1747 | { | |
1c17e4d4 | 1748 | *cp = '\0'; |
a858af28 PM |
1749 | } |
1750 | ||
1751 | #endif /* #else #ifdef CONFIG_RCU_FAST_NO_HZ */ | |
1752 | ||
1753 | /* Initiate the stall-info list. */ | |
1754 | static void print_cpu_stall_info_begin(void) | |
1755 | { | |
efc151c3 | 1756 | pr_cont("\n"); |
a858af28 PM |
1757 | } |
1758 | ||
1759 | /* | |
1760 | * Print out diagnostic information for the specified stalled CPU. | |
1761 | * | |
1762 | * If the specified CPU is aware of the current RCU grace period | |
1763 | * (flavor specified by rsp), then print the number of scheduling | |
1764 | * clock interrupts the CPU has taken during the time that it has | |
1765 | * been aware. Otherwise, print the number of RCU grace periods | |
1766 | * that this CPU is ignorant of, for example, "1" if the CPU was | |
1767 | * aware of the previous grace period. | |
1768 | * | |
1769 | * Also print out idle and (if CONFIG_RCU_FAST_NO_HZ) idle-entry info. | |
1770 | */ | |
1771 | static void print_cpu_stall_info(struct rcu_state *rsp, int cpu) | |
1772 | { | |
9b9500da | 1773 | unsigned long delta; |
a858af28 PM |
1774 | char fast_no_hz[72]; |
1775 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); | |
1776 | struct rcu_dynticks *rdtp = rdp->dynticks; | |
1777 | char *ticks_title; | |
1778 | unsigned long ticks_value; | |
1779 | ||
3caa973b TH |
1780 | /* |
1781 | * We could be printing a lot while holding a spinlock. Avoid | |
1782 | * triggering hard lockup. | |
1783 | */ | |
1784 | touch_nmi_watchdog(); | |
1785 | ||
471f87c3 PM |
1786 | ticks_value = rcu_seq_ctr(rsp->gp_seq - rdp->gp_seq); |
1787 | if (ticks_value) { | |
1788 | ticks_title = "GPs behind"; | |
1789 | } else { | |
a858af28 PM |
1790 | ticks_title = "ticks this GP"; |
1791 | ticks_value = rdp->ticks_this_gp; | |
a858af28 PM |
1792 | } |
1793 | print_cpu_stall_fast_no_hz(fast_no_hz, cpu); | |
8aa670cd | 1794 | delta = rcu_seq_ctr(rdp->mynode->gp_seq - rdp->rcu_iw_gp_seq); |
84585aa8 | 1795 | pr_err("\t%d-%c%c%c%c: (%lu %s) idle=%03x/%ld/%ld softirq=%u/%u fqs=%ld %s\n", |
7f21aeef PM |
1796 | cpu, |
1797 | "O."[!!cpu_online(cpu)], | |
1798 | "o."[!!(rdp->grpmask & rdp->mynode->qsmaskinit)], | |
1799 | "N."[!!(rdp->grpmask & rdp->mynode->qsmaskinitnext)], | |
9b9500da PM |
1800 | !IS_ENABLED(CONFIG_IRQ_WORK) ? '?' : |
1801 | rdp->rcu_iw_pending ? (int)min(delta, 9UL) + '0' : | |
1802 | "!."[!delta], | |
7f21aeef | 1803 | ticks_value, ticks_title, |
02a5c550 | 1804 | rcu_dynticks_snap(rdtp) & 0xfff, |
a858af28 | 1805 | rdtp->dynticks_nesting, rdtp->dynticks_nmi_nesting, |
6231069b | 1806 | rdp->softirq_snap, kstat_softirqs_cpu(RCU_SOFTIRQ, cpu), |
7d0ae808 | 1807 | READ_ONCE(rsp->n_force_qs) - rsp->n_force_qs_gpstart, |
a858af28 PM |
1808 | fast_no_hz); |
1809 | } | |
1810 | ||
1811 | /* Terminate the stall-info list. */ | |
1812 | static void print_cpu_stall_info_end(void) | |
1813 | { | |
efc151c3 | 1814 | pr_err("\t"); |
a858af28 PM |
1815 | } |
1816 | ||
1817 | /* Zero ->ticks_this_gp for all flavors of RCU. */ | |
1818 | static void zero_cpu_stall_ticks(struct rcu_data *rdp) | |
1819 | { | |
1820 | rdp->ticks_this_gp = 0; | |
6231069b | 1821 | rdp->softirq_snap = kstat_softirqs_cpu(RCU_SOFTIRQ, smp_processor_id()); |
a858af28 PM |
1822 | } |
1823 | ||
1824 | /* Increment ->ticks_this_gp for all flavors of RCU. */ | |
1825 | static void increment_cpu_stall_ticks(void) | |
1826 | { | |
115f7a7c PM |
1827 | struct rcu_state *rsp; |
1828 | ||
1829 | for_each_rcu_flavor(rsp) | |
fa07a58f | 1830 | raw_cpu_inc(rsp->rda->ticks_this_gp); |
a858af28 PM |
1831 | } |
1832 | ||
3fbfbf7a PM |
1833 | #ifdef CONFIG_RCU_NOCB_CPU |
1834 | ||
1835 | /* | |
1836 | * Offload callback processing from the boot-time-specified set of CPUs | |
1837 | * specified by rcu_nocb_mask. For each CPU in the set, there is a | |
1838 | * kthread created that pulls the callbacks from the corresponding CPU, | |
1839 | * waits for a grace period to elapse, and invokes the callbacks. | |
1840 | * The no-CBs CPUs do a wake_up() on their kthread when they insert | |
1841 | * a callback into any empty list, unless the rcu_nocb_poll boot parameter | |
1842 | * has been specified, in which case each kthread actively polls its | |
1843 | * CPU. (Which isn't so great for energy efficiency, but which does | |
1844 | * reduce RCU's overhead on that CPU.) | |
1845 | * | |
1846 | * This is intended to be used in conjunction with Frederic Weisbecker's | |
1847 | * adaptive-idle work, which would seriously reduce OS jitter on CPUs | |
1848 | * running CPU-bound user-mode computations. | |
1849 | * | |
1850 | * Offloading of callback processing could also in theory be used as | |
1851 | * an energy-efficiency measure because CPUs with no RCU callbacks | |
1852 | * queued are more aggressive about entering dyntick-idle mode. | |
1853 | */ | |
1854 | ||
1855 | ||
1856 | /* Parse the boot-time rcu_nocb_mask CPU list from the kernel parameters. */ | |
1857 | static int __init rcu_nocb_setup(char *str) | |
1858 | { | |
1859 | alloc_bootmem_cpumask_var(&rcu_nocb_mask); | |
3fbfbf7a PM |
1860 | cpulist_parse(str, rcu_nocb_mask); |
1861 | return 1; | |
1862 | } | |
1863 | __setup("rcu_nocbs=", rcu_nocb_setup); | |
1864 | ||
1b0048a4 PG |
1865 | static int __init parse_rcu_nocb_poll(char *arg) |
1866 | { | |
5455a7f6 | 1867 | rcu_nocb_poll = true; |
1b0048a4 PG |
1868 | return 0; |
1869 | } | |
1870 | early_param("rcu_nocb_poll", parse_rcu_nocb_poll); | |
1871 | ||
dae6e64d | 1872 | /* |
0446be48 PM |
1873 | * Wake up any no-CBs CPUs' kthreads that were waiting on the just-ended |
1874 | * grace period. | |
dae6e64d | 1875 | */ |
abedf8e2 | 1876 | static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq) |
dae6e64d | 1877 | { |
abedf8e2 | 1878 | swake_up_all(sq); |
dae6e64d PM |
1879 | } |
1880 | ||
abedf8e2 | 1881 | static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp) |
065bb78c | 1882 | { |
e0da2374 | 1883 | return &rnp->nocb_gp_wq[rcu_seq_ctr(rnp->gp_seq) & 0x1]; |
065bb78c DW |
1884 | } |
1885 | ||
dae6e64d | 1886 | static void rcu_init_one_nocb(struct rcu_node *rnp) |
34ed6246 | 1887 | { |
abedf8e2 PG |
1888 | init_swait_queue_head(&rnp->nocb_gp_wq[0]); |
1889 | init_swait_queue_head(&rnp->nocb_gp_wq[1]); | |
34ed6246 PM |
1890 | } |
1891 | ||
24342c96 | 1892 | /* Is the specified CPU a no-CBs CPU? */ |
d1e43fa5 | 1893 | bool rcu_is_nocb_cpu(int cpu) |
3fbfbf7a | 1894 | { |
84b12b75 | 1895 | if (cpumask_available(rcu_nocb_mask)) |
3fbfbf7a PM |
1896 | return cpumask_test_cpu(cpu, rcu_nocb_mask); |
1897 | return false; | |
1898 | } | |
1899 | ||
fbce7497 | 1900 | /* |
8be6e1b1 PM |
1901 | * Kick the leader kthread for this NOCB group. Caller holds ->nocb_lock |
1902 | * and this function releases it. | |
fbce7497 | 1903 | */ |
8be6e1b1 PM |
1904 | static void __wake_nocb_leader(struct rcu_data *rdp, bool force, |
1905 | unsigned long flags) | |
1906 | __releases(rdp->nocb_lock) | |
fbce7497 PM |
1907 | { |
1908 | struct rcu_data *rdp_leader = rdp->nocb_leader; | |
1909 | ||
8be6e1b1 PM |
1910 | lockdep_assert_held(&rdp->nocb_lock); |
1911 | if (!READ_ONCE(rdp_leader->nocb_kthread)) { | |
1912 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
fbce7497 | 1913 | return; |
8be6e1b1 PM |
1914 | } |
1915 | if (rdp_leader->nocb_leader_sleep || force) { | |
39953dfd | 1916 | /* Prior smp_mb__after_atomic() orders against prior enqueue. */ |
7d0ae808 | 1917 | WRITE_ONCE(rdp_leader->nocb_leader_sleep, false); |
8be6e1b1 PM |
1918 | del_timer(&rdp->nocb_timer); |
1919 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
6b5fc3a1 | 1920 | smp_mb(); /* ->nocb_leader_sleep before swake_up(). */ |
abedf8e2 | 1921 | swake_up(&rdp_leader->nocb_wq); |
8be6e1b1 PM |
1922 | } else { |
1923 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
fbce7497 PM |
1924 | } |
1925 | } | |
1926 | ||
8be6e1b1 PM |
1927 | /* |
1928 | * Kick the leader kthread for this NOCB group, but caller has not | |
1929 | * acquired locks. | |
1930 | */ | |
1931 | static void wake_nocb_leader(struct rcu_data *rdp, bool force) | |
1932 | { | |
1933 | unsigned long flags; | |
1934 | ||
1935 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); | |
1936 | __wake_nocb_leader(rdp, force, flags); | |
1937 | } | |
1938 | ||
1939 | /* | |
1940 | * Arrange to wake the leader kthread for this NOCB group at some | |
1941 | * future time when it is safe to do so. | |
1942 | */ | |
1943 | static void wake_nocb_leader_defer(struct rcu_data *rdp, int waketype, | |
1944 | const char *reason) | |
1945 | { | |
1946 | unsigned long flags; | |
1947 | ||
1948 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); | |
1949 | if (rdp->nocb_defer_wakeup == RCU_NOCB_WAKE_NOT) | |
1950 | mod_timer(&rdp->nocb_timer, jiffies + 1); | |
1951 | WRITE_ONCE(rdp->nocb_defer_wakeup, waketype); | |
1952 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, reason); | |
1953 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
1954 | } | |
1955 | ||
d7e29933 PM |
1956 | /* |
1957 | * Does the specified CPU need an RCU callback for the specified flavor | |
1958 | * of rcu_barrier()? | |
1959 | */ | |
1960 | static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu) | |
1961 | { | |
1962 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); | |
41050a00 PM |
1963 | unsigned long ret; |
1964 | #ifdef CONFIG_PROVE_RCU | |
d7e29933 | 1965 | struct rcu_head *rhp; |
41050a00 | 1966 | #endif /* #ifdef CONFIG_PROVE_RCU */ |
d7e29933 | 1967 | |
41050a00 PM |
1968 | /* |
1969 | * Check count of all no-CBs callbacks awaiting invocation. | |
1970 | * There needs to be a barrier before this function is called, | |
1971 | * but associated with a prior determination that no more | |
1972 | * callbacks would be posted. In the worst case, the first | |
1973 | * barrier in _rcu_barrier() suffices (but the caller cannot | |
1974 | * necessarily rely on this, not a substitute for the caller | |
1975 | * getting the concurrency design right!). There must also be | |
1976 | * a barrier between the following load an posting of a callback | |
1977 | * (if a callback is in fact needed). This is associated with an | |
1978 | * atomic_inc() in the caller. | |
1979 | */ | |
1980 | ret = atomic_long_read(&rdp->nocb_q_count); | |
d7e29933 | 1981 | |
41050a00 | 1982 | #ifdef CONFIG_PROVE_RCU |
7d0ae808 | 1983 | rhp = READ_ONCE(rdp->nocb_head); |
d7e29933 | 1984 | if (!rhp) |
7d0ae808 | 1985 | rhp = READ_ONCE(rdp->nocb_gp_head); |
d7e29933 | 1986 | if (!rhp) |
7d0ae808 | 1987 | rhp = READ_ONCE(rdp->nocb_follower_head); |
d7e29933 PM |
1988 | |
1989 | /* Having no rcuo kthread but CBs after scheduler starts is bad! */ | |
7d0ae808 | 1990 | if (!READ_ONCE(rdp->nocb_kthread) && rhp && |
59f792d1 | 1991 | rcu_scheduler_fully_active) { |
d7e29933 PM |
1992 | /* RCU callback enqueued before CPU first came online??? */ |
1993 | pr_err("RCU: Never-onlined no-CBs CPU %d has CB %p\n", | |
1994 | cpu, rhp->func); | |
1995 | WARN_ON_ONCE(1); | |
1996 | } | |
41050a00 | 1997 | #endif /* #ifdef CONFIG_PROVE_RCU */ |
d7e29933 | 1998 | |
41050a00 | 1999 | return !!ret; |
d7e29933 PM |
2000 | } |
2001 | ||
3fbfbf7a PM |
2002 | /* |
2003 | * Enqueue the specified string of rcu_head structures onto the specified | |
2004 | * CPU's no-CBs lists. The CPU is specified by rdp, the head of the | |
2005 | * string by rhp, and the tail of the string by rhtp. The non-lazy/lazy | |
2006 | * counts are supplied by rhcount and rhcount_lazy. | |
2007 | * | |
2008 | * If warranted, also wake up the kthread servicing this CPUs queues. | |
2009 | */ | |
2010 | static void __call_rcu_nocb_enqueue(struct rcu_data *rdp, | |
2011 | struct rcu_head *rhp, | |
2012 | struct rcu_head **rhtp, | |
96d3fd0d PM |
2013 | int rhcount, int rhcount_lazy, |
2014 | unsigned long flags) | |
3fbfbf7a PM |
2015 | { |
2016 | int len; | |
2017 | struct rcu_head **old_rhpp; | |
2018 | struct task_struct *t; | |
2019 | ||
2020 | /* Enqueue the callback on the nocb list and update counts. */ | |
41050a00 PM |
2021 | atomic_long_add(rhcount, &rdp->nocb_q_count); |
2022 | /* rcu_barrier() relies on ->nocb_q_count add before xchg. */ | |
3fbfbf7a | 2023 | old_rhpp = xchg(&rdp->nocb_tail, rhtp); |
7d0ae808 | 2024 | WRITE_ONCE(*old_rhpp, rhp); |
3fbfbf7a | 2025 | atomic_long_add(rhcount_lazy, &rdp->nocb_q_count_lazy); |
39953dfd | 2026 | smp_mb__after_atomic(); /* Store *old_rhpp before _wake test. */ |
3fbfbf7a PM |
2027 | |
2028 | /* If we are not being polled and there is a kthread, awaken it ... */ | |
7d0ae808 | 2029 | t = READ_ONCE(rdp->nocb_kthread); |
25e03a74 | 2030 | if (rcu_nocb_poll || !t) { |
9261dd0d PM |
2031 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
2032 | TPS("WakeNotPoll")); | |
3fbfbf7a | 2033 | return; |
9261dd0d | 2034 | } |
3fbfbf7a PM |
2035 | len = atomic_long_read(&rdp->nocb_q_count); |
2036 | if (old_rhpp == &rdp->nocb_head) { | |
96d3fd0d | 2037 | if (!irqs_disabled_flags(flags)) { |
fbce7497 PM |
2038 | /* ... if queue was empty ... */ |
2039 | wake_nocb_leader(rdp, false); | |
96d3fd0d PM |
2040 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
2041 | TPS("WakeEmpty")); | |
2042 | } else { | |
8be6e1b1 PM |
2043 | wake_nocb_leader_defer(rdp, RCU_NOCB_WAKE, |
2044 | TPS("WakeEmptyIsDeferred")); | |
96d3fd0d | 2045 | } |
3fbfbf7a PM |
2046 | rdp->qlen_last_fqs_check = 0; |
2047 | } else if (len > rdp->qlen_last_fqs_check + qhimark) { | |
fbce7497 | 2048 | /* ... or if many callbacks queued. */ |
9fdd3bc9 PM |
2049 | if (!irqs_disabled_flags(flags)) { |
2050 | wake_nocb_leader(rdp, true); | |
2051 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, | |
2052 | TPS("WakeOvf")); | |
2053 | } else { | |
efcd2d54 | 2054 | wake_nocb_leader_defer(rdp, RCU_NOCB_WAKE_FORCE, |
8be6e1b1 | 2055 | TPS("WakeOvfIsDeferred")); |
9fdd3bc9 | 2056 | } |
3fbfbf7a | 2057 | rdp->qlen_last_fqs_check = LONG_MAX / 2; |
9261dd0d PM |
2058 | } else { |
2059 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("WakeNot")); | |
3fbfbf7a PM |
2060 | } |
2061 | return; | |
2062 | } | |
2063 | ||
2064 | /* | |
2065 | * This is a helper for __call_rcu(), which invokes this when the normal | |
2066 | * callback queue is inoperable. If this is not a no-CBs CPU, this | |
2067 | * function returns failure back to __call_rcu(), which can complain | |
2068 | * appropriately. | |
2069 | * | |
2070 | * Otherwise, this function queues the callback where the corresponding | |
2071 | * "rcuo" kthread can find it. | |
2072 | */ | |
2073 | static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, | |
96d3fd0d | 2074 | bool lazy, unsigned long flags) |
3fbfbf7a PM |
2075 | { |
2076 | ||
d1e43fa5 | 2077 | if (!rcu_is_nocb_cpu(rdp->cpu)) |
c271d3a9 | 2078 | return false; |
96d3fd0d | 2079 | __call_rcu_nocb_enqueue(rdp, rhp, &rhp->next, 1, lazy, flags); |
21e7a608 PM |
2080 | if (__is_kfree_rcu_offset((unsigned long)rhp->func)) |
2081 | trace_rcu_kfree_callback(rdp->rsp->name, rhp, | |
2082 | (unsigned long)rhp->func, | |
756cbf6b PM |
2083 | -atomic_long_read(&rdp->nocb_q_count_lazy), |
2084 | -atomic_long_read(&rdp->nocb_q_count)); | |
21e7a608 PM |
2085 | else |
2086 | trace_rcu_callback(rdp->rsp->name, rhp, | |
756cbf6b PM |
2087 | -atomic_long_read(&rdp->nocb_q_count_lazy), |
2088 | -atomic_long_read(&rdp->nocb_q_count)); | |
1772947b PM |
2089 | |
2090 | /* | |
2091 | * If called from an extended quiescent state with interrupts | |
2092 | * disabled, invoke the RCU core in order to allow the idle-entry | |
2093 | * deferred-wakeup check to function. | |
2094 | */ | |
2095 | if (irqs_disabled_flags(flags) && | |
2096 | !rcu_is_watching() && | |
2097 | cpu_online(smp_processor_id())) | |
2098 | invoke_rcu_core(); | |
2099 | ||
c271d3a9 | 2100 | return true; |
3fbfbf7a PM |
2101 | } |
2102 | ||
2103 | /* | |
2104 | * Adopt orphaned callbacks on a no-CBs CPU, or return 0 if this is | |
2105 | * not a no-CBs CPU. | |
2106 | */ | |
b1a2d79f | 2107 | static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_data *my_rdp, |
96d3fd0d PM |
2108 | struct rcu_data *rdp, |
2109 | unsigned long flags) | |
3fbfbf7a | 2110 | { |
b04db8e1 | 2111 | lockdep_assert_irqs_disabled(); |
d1e43fa5 | 2112 | if (!rcu_is_nocb_cpu(smp_processor_id())) |
b1a2d79f PM |
2113 | return false; /* Not NOCBs CPU, caller must migrate CBs. */ |
2114 | __call_rcu_nocb_enqueue(my_rdp, rcu_segcblist_head(&rdp->cblist), | |
2115 | rcu_segcblist_tail(&rdp->cblist), | |
2116 | rcu_segcblist_n_cbs(&rdp->cblist), | |
2117 | rcu_segcblist_n_lazy_cbs(&rdp->cblist), flags); | |
2118 | rcu_segcblist_init(&rdp->cblist); | |
2119 | rcu_segcblist_disable(&rdp->cblist); | |
0a9e1e11 | 2120 | return true; |
3fbfbf7a PM |
2121 | } |
2122 | ||
2123 | /* | |
34ed6246 PM |
2124 | * If necessary, kick off a new grace period, and either way wait |
2125 | * for a subsequent grace period to complete. | |
3fbfbf7a | 2126 | */ |
34ed6246 | 2127 | static void rcu_nocb_wait_gp(struct rcu_data *rdp) |
3fbfbf7a | 2128 | { |
34ed6246 | 2129 | unsigned long c; |
dae6e64d | 2130 | bool d; |
34ed6246 | 2131 | unsigned long flags; |
48a7639c | 2132 | bool needwake; |
34ed6246 PM |
2133 | struct rcu_node *rnp = rdp->mynode; |
2134 | ||
ab5e869c | 2135 | local_irq_save(flags); |
29365e56 | 2136 | c = rcu_seq_snap(&rdp->rsp->gp_seq); |
ab5e869c PM |
2137 | if (!rdp->gpwrap && ULONG_CMP_GE(rdp->gp_seq_needed, c)) { |
2138 | local_irq_restore(flags); | |
2139 | } else { | |
2140 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ | |
2141 | needwake = rcu_start_this_gp(rnp, rdp, c); | |
2142 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); | |
2143 | if (needwake) | |
2144 | rcu_gp_kthread_wake(rdp->rsp); | |
2145 | } | |
3fbfbf7a PM |
2146 | |
2147 | /* | |
34ed6246 PM |
2148 | * Wait for the grace period. Do so interruptibly to avoid messing |
2149 | * up the load average. | |
3fbfbf7a | 2150 | */ |
41e80595 | 2151 | trace_rcu_this_gp(rnp, rdp, c, TPS("StartWait")); |
34ed6246 | 2152 | for (;;) { |
abedf8e2 | 2153 | swait_event_interruptible( |
29365e56 PM |
2154 | rnp->nocb_gp_wq[rcu_seq_ctr(c) & 0x1], |
2155 | (d = rcu_seq_done(&rnp->gp_seq, c))); | |
dae6e64d | 2156 | if (likely(d)) |
34ed6246 | 2157 | break; |
73a860cd | 2158 | WARN_ON(signal_pending(current)); |
41e80595 | 2159 | trace_rcu_this_gp(rnp, rdp, c, TPS("ResumeWait")); |
34ed6246 | 2160 | } |
41e80595 | 2161 | trace_rcu_this_gp(rnp, rdp, c, TPS("EndWait")); |
34ed6246 | 2162 | smp_mb(); /* Ensure that CB invocation happens after GP end. */ |
3fbfbf7a PM |
2163 | } |
2164 | ||
fbce7497 PM |
2165 | /* |
2166 | * Leaders come here to wait for additional callbacks to show up. | |
2167 | * This function does not return until callbacks appear. | |
2168 | */ | |
2169 | static void nocb_leader_wait(struct rcu_data *my_rdp) | |
2170 | { | |
2171 | bool firsttime = true; | |
8be6e1b1 | 2172 | unsigned long flags; |
fbce7497 PM |
2173 | bool gotcbs; |
2174 | struct rcu_data *rdp; | |
2175 | struct rcu_head **tail; | |
2176 | ||
2177 | wait_again: | |
2178 | ||
2179 | /* Wait for callbacks to appear. */ | |
2180 | if (!rcu_nocb_poll) { | |
bedbb648 | 2181 | trace_rcu_nocb_wake(my_rdp->rsp->name, my_rdp->cpu, TPS("Sleep")); |
abedf8e2 | 2182 | swait_event_interruptible(my_rdp->nocb_wq, |
7d0ae808 | 2183 | !READ_ONCE(my_rdp->nocb_leader_sleep)); |
8be6e1b1 PM |
2184 | raw_spin_lock_irqsave(&my_rdp->nocb_lock, flags); |
2185 | my_rdp->nocb_leader_sleep = true; | |
2186 | WRITE_ONCE(my_rdp->nocb_defer_wakeup, RCU_NOCB_WAKE_NOT); | |
2187 | del_timer(&my_rdp->nocb_timer); | |
2188 | raw_spin_unlock_irqrestore(&my_rdp->nocb_lock, flags); | |
fbce7497 PM |
2189 | } else if (firsttime) { |
2190 | firsttime = false; /* Don't drown trace log with "Poll"! */ | |
bedbb648 | 2191 | trace_rcu_nocb_wake(my_rdp->rsp->name, my_rdp->cpu, TPS("Poll")); |
fbce7497 PM |
2192 | } |
2193 | ||
2194 | /* | |
2195 | * Each pass through the following loop checks a follower for CBs. | |
2196 | * We are our own first follower. Any CBs found are moved to | |
2197 | * nocb_gp_head, where they await a grace period. | |
2198 | */ | |
2199 | gotcbs = false; | |
8be6e1b1 | 2200 | smp_mb(); /* wakeup and _sleep before ->nocb_head reads. */ |
fbce7497 | 2201 | for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) { |
7d0ae808 | 2202 | rdp->nocb_gp_head = READ_ONCE(rdp->nocb_head); |
fbce7497 PM |
2203 | if (!rdp->nocb_gp_head) |
2204 | continue; /* No CBs here, try next follower. */ | |
2205 | ||
2206 | /* Move callbacks to wait-for-GP list, which is empty. */ | |
7d0ae808 | 2207 | WRITE_ONCE(rdp->nocb_head, NULL); |
fbce7497 | 2208 | rdp->nocb_gp_tail = xchg(&rdp->nocb_tail, &rdp->nocb_head); |
fbce7497 PM |
2209 | gotcbs = true; |
2210 | } | |
2211 | ||
8be6e1b1 | 2212 | /* No callbacks? Sleep a bit if polling, and go retry. */ |
fbce7497 | 2213 | if (unlikely(!gotcbs)) { |
73a860cd | 2214 | WARN_ON(signal_pending(current)); |
8be6e1b1 PM |
2215 | if (rcu_nocb_poll) { |
2216 | schedule_timeout_interruptible(1); | |
2217 | } else { | |
fbce7497 | 2218 | trace_rcu_nocb_wake(my_rdp->rsp->name, my_rdp->cpu, |
bedbb648 | 2219 | TPS("WokeEmpty")); |
8be6e1b1 | 2220 | } |
fbce7497 PM |
2221 | goto wait_again; |
2222 | } | |
2223 | ||
2224 | /* Wait for one grace period. */ | |
2225 | rcu_nocb_wait_gp(my_rdp); | |
2226 | ||
fbce7497 PM |
2227 | /* Each pass through the following loop wakes a follower, if needed. */ |
2228 | for (rdp = my_rdp; rdp; rdp = rdp->nocb_next_follower) { | |
8be6e1b1 PM |
2229 | if (!rcu_nocb_poll && |
2230 | READ_ONCE(rdp->nocb_head) && | |
2231 | READ_ONCE(my_rdp->nocb_leader_sleep)) { | |
2232 | raw_spin_lock_irqsave(&my_rdp->nocb_lock, flags); | |
11ed7f93 | 2233 | my_rdp->nocb_leader_sleep = false;/* No need to sleep.*/ |
8be6e1b1 PM |
2234 | raw_spin_unlock_irqrestore(&my_rdp->nocb_lock, flags); |
2235 | } | |
fbce7497 PM |
2236 | if (!rdp->nocb_gp_head) |
2237 | continue; /* No CBs, so no need to wake follower. */ | |
2238 | ||
2239 | /* Append callbacks to follower's "done" list. */ | |
8be6e1b1 PM |
2240 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); |
2241 | tail = rdp->nocb_follower_tail; | |
2242 | rdp->nocb_follower_tail = rdp->nocb_gp_tail; | |
fbce7497 | 2243 | *tail = rdp->nocb_gp_head; |
8be6e1b1 | 2244 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); |
fbce7497 | 2245 | if (rdp != my_rdp && tail == &rdp->nocb_follower_head) { |
8be6e1b1 | 2246 | /* List was empty, so wake up the follower. */ |
abedf8e2 | 2247 | swake_up(&rdp->nocb_wq); |
fbce7497 PM |
2248 | } |
2249 | } | |
2250 | ||
2251 | /* If we (the leader) don't have CBs, go wait some more. */ | |
2252 | if (!my_rdp->nocb_follower_head) | |
2253 | goto wait_again; | |
2254 | } | |
2255 | ||
2256 | /* | |
2257 | * Followers come here to wait for additional callbacks to show up. | |
2258 | * This function does not return until callbacks appear. | |
2259 | */ | |
2260 | static void nocb_follower_wait(struct rcu_data *rdp) | |
2261 | { | |
fbce7497 | 2262 | for (;;) { |
bedbb648 | 2263 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("FollowerSleep")); |
8be6e1b1 PM |
2264 | swait_event_interruptible(rdp->nocb_wq, |
2265 | READ_ONCE(rdp->nocb_follower_head)); | |
fbce7497 PM |
2266 | if (smp_load_acquire(&rdp->nocb_follower_head)) { |
2267 | /* ^^^ Ensure CB invocation follows _head test. */ | |
2268 | return; | |
2269 | } | |
73a860cd | 2270 | WARN_ON(signal_pending(current)); |
bedbb648 | 2271 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("WokeEmpty")); |
fbce7497 PM |
2272 | } |
2273 | } | |
2274 | ||
3fbfbf7a PM |
2275 | /* |
2276 | * Per-rcu_data kthread, but only for no-CBs CPUs. Each kthread invokes | |
fbce7497 PM |
2277 | * callbacks queued by the corresponding no-CBs CPU, however, there is |
2278 | * an optional leader-follower relationship so that the grace-period | |
2279 | * kthreads don't have to do quite so many wakeups. | |
3fbfbf7a PM |
2280 | */ |
2281 | static int rcu_nocb_kthread(void *arg) | |
2282 | { | |
2283 | int c, cl; | |
8be6e1b1 | 2284 | unsigned long flags; |
3fbfbf7a PM |
2285 | struct rcu_head *list; |
2286 | struct rcu_head *next; | |
2287 | struct rcu_head **tail; | |
2288 | struct rcu_data *rdp = arg; | |
2289 | ||
2290 | /* Each pass through this loop invokes one batch of callbacks */ | |
2291 | for (;;) { | |
fbce7497 PM |
2292 | /* Wait for callbacks. */ |
2293 | if (rdp->nocb_leader == rdp) | |
2294 | nocb_leader_wait(rdp); | |
2295 | else | |
2296 | nocb_follower_wait(rdp); | |
2297 | ||
2298 | /* Pull the ready-to-invoke callbacks onto local list. */ | |
8be6e1b1 PM |
2299 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); |
2300 | list = rdp->nocb_follower_head; | |
2301 | rdp->nocb_follower_head = NULL; | |
2302 | tail = rdp->nocb_follower_tail; | |
2303 | rdp->nocb_follower_tail = &rdp->nocb_follower_head; | |
2304 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
fbce7497 | 2305 | BUG_ON(!list); |
bedbb648 | 2306 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("WokeNonEmpty")); |
3fbfbf7a PM |
2307 | |
2308 | /* Each pass through the following loop invokes a callback. */ | |
41050a00 PM |
2309 | trace_rcu_batch_start(rdp->rsp->name, |
2310 | atomic_long_read(&rdp->nocb_q_count_lazy), | |
2311 | atomic_long_read(&rdp->nocb_q_count), -1); | |
3fbfbf7a PM |
2312 | c = cl = 0; |
2313 | while (list) { | |
2314 | next = list->next; | |
2315 | /* Wait for enqueuing to complete, if needed. */ | |
2316 | while (next == NULL && &list->next != tail) { | |
69a79bb1 PM |
2317 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
2318 | TPS("WaitQueue")); | |
3fbfbf7a | 2319 | schedule_timeout_interruptible(1); |
69a79bb1 PM |
2320 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, |
2321 | TPS("WokeQueue")); | |
3fbfbf7a PM |
2322 | next = list->next; |
2323 | } | |
2324 | debug_rcu_head_unqueue(list); | |
2325 | local_bh_disable(); | |
2326 | if (__rcu_reclaim(rdp->rsp->name, list)) | |
2327 | cl++; | |
2328 | c++; | |
2329 | local_bh_enable(); | |
cee43939 | 2330 | cond_resched_tasks_rcu_qs(); |
3fbfbf7a PM |
2331 | list = next; |
2332 | } | |
2333 | trace_rcu_batch_end(rdp->rsp->name, c, !!list, 0, 0, 1); | |
41050a00 PM |
2334 | smp_mb__before_atomic(); /* _add after CB invocation. */ |
2335 | atomic_long_add(-c, &rdp->nocb_q_count); | |
2336 | atomic_long_add(-cl, &rdp->nocb_q_count_lazy); | |
3fbfbf7a PM |
2337 | } |
2338 | return 0; | |
2339 | } | |
2340 | ||
96d3fd0d | 2341 | /* Is a deferred wakeup of rcu_nocb_kthread() required? */ |
9fdd3bc9 | 2342 | static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp) |
96d3fd0d | 2343 | { |
7d0ae808 | 2344 | return READ_ONCE(rdp->nocb_defer_wakeup); |
96d3fd0d PM |
2345 | } |
2346 | ||
2347 | /* Do a deferred wakeup of rcu_nocb_kthread(). */ | |
8be6e1b1 | 2348 | static void do_nocb_deferred_wakeup_common(struct rcu_data *rdp) |
96d3fd0d | 2349 | { |
8be6e1b1 | 2350 | unsigned long flags; |
9fdd3bc9 PM |
2351 | int ndw; |
2352 | ||
8be6e1b1 PM |
2353 | raw_spin_lock_irqsave(&rdp->nocb_lock, flags); |
2354 | if (!rcu_nocb_need_deferred_wakeup(rdp)) { | |
2355 | raw_spin_unlock_irqrestore(&rdp->nocb_lock, flags); | |
96d3fd0d | 2356 | return; |
8be6e1b1 | 2357 | } |
7d0ae808 | 2358 | ndw = READ_ONCE(rdp->nocb_defer_wakeup); |
511324e4 | 2359 | WRITE_ONCE(rdp->nocb_defer_wakeup, RCU_NOCB_WAKE_NOT); |
8be6e1b1 | 2360 | __wake_nocb_leader(rdp, ndw == RCU_NOCB_WAKE_FORCE, flags); |
9fdd3bc9 | 2361 | trace_rcu_nocb_wake(rdp->rsp->name, rdp->cpu, TPS("DeferredWake")); |
96d3fd0d PM |
2362 | } |
2363 | ||
8be6e1b1 | 2364 | /* Do a deferred wakeup of rcu_nocb_kthread() from a timer handler. */ |
fd30b717 | 2365 | static void do_nocb_deferred_wakeup_timer(struct timer_list *t) |
8be6e1b1 | 2366 | { |
fd30b717 KC |
2367 | struct rcu_data *rdp = from_timer(rdp, t, nocb_timer); |
2368 | ||
2369 | do_nocb_deferred_wakeup_common(rdp); | |
8be6e1b1 PM |
2370 | } |
2371 | ||
2372 | /* | |
2373 | * Do a deferred wakeup of rcu_nocb_kthread() from fastpath. | |
2374 | * This means we do an inexact common-case check. Note that if | |
2375 | * we miss, ->nocb_timer will eventually clean things up. | |
2376 | */ | |
2377 | static void do_nocb_deferred_wakeup(struct rcu_data *rdp) | |
2378 | { | |
2379 | if (rcu_nocb_need_deferred_wakeup(rdp)) | |
2380 | do_nocb_deferred_wakeup_common(rdp); | |
2381 | } | |
2382 | ||
f4579fc5 PM |
2383 | void __init rcu_init_nohz(void) |
2384 | { | |
2385 | int cpu; | |
ef126206 | 2386 | bool need_rcu_nocb_mask = false; |
f4579fc5 PM |
2387 | struct rcu_state *rsp; |
2388 | ||
f4579fc5 PM |
2389 | #if defined(CONFIG_NO_HZ_FULL) |
2390 | if (tick_nohz_full_running && cpumask_weight(tick_nohz_full_mask)) | |
2391 | need_rcu_nocb_mask = true; | |
2392 | #endif /* #if defined(CONFIG_NO_HZ_FULL) */ | |
2393 | ||
84b12b75 | 2394 | if (!cpumask_available(rcu_nocb_mask) && need_rcu_nocb_mask) { |
949cccdb PK |
2395 | if (!zalloc_cpumask_var(&rcu_nocb_mask, GFP_KERNEL)) { |
2396 | pr_info("rcu_nocb_mask allocation failed, callback offloading disabled.\n"); | |
2397 | return; | |
2398 | } | |
f4579fc5 | 2399 | } |
84b12b75 | 2400 | if (!cpumask_available(rcu_nocb_mask)) |
f4579fc5 PM |
2401 | return; |
2402 | ||
f4579fc5 PM |
2403 | #if defined(CONFIG_NO_HZ_FULL) |
2404 | if (tick_nohz_full_running) | |
2405 | cpumask_or(rcu_nocb_mask, rcu_nocb_mask, tick_nohz_full_mask); | |
2406 | #endif /* #if defined(CONFIG_NO_HZ_FULL) */ | |
2407 | ||
2408 | if (!cpumask_subset(rcu_nocb_mask, cpu_possible_mask)) { | |
ef126206 | 2409 | pr_info("\tNote: kernel parameter 'rcu_nocbs=', 'nohz_full', or 'isolcpus=' contains nonexistent CPUs.\n"); |
f4579fc5 PM |
2410 | cpumask_and(rcu_nocb_mask, cpu_possible_mask, |
2411 | rcu_nocb_mask); | |
2412 | } | |
3016611e PM |
2413 | if (cpumask_empty(rcu_nocb_mask)) |
2414 | pr_info("\tOffload RCU callbacks from CPUs: (none).\n"); | |
2415 | else | |
2416 | pr_info("\tOffload RCU callbacks from CPUs: %*pbl.\n", | |
2417 | cpumask_pr_args(rcu_nocb_mask)); | |
f4579fc5 PM |
2418 | if (rcu_nocb_poll) |
2419 | pr_info("\tPoll for callbacks from no-CBs CPUs.\n"); | |
2420 | ||
2421 | for_each_rcu_flavor(rsp) { | |
34404ca8 PM |
2422 | for_each_cpu(cpu, rcu_nocb_mask) |
2423 | init_nocb_callback_list(per_cpu_ptr(rsp->rda, cpu)); | |
35ce7f29 | 2424 | rcu_organize_nocb_kthreads(rsp); |
f4579fc5 | 2425 | } |
96d3fd0d PM |
2426 | } |
2427 | ||
3fbfbf7a PM |
2428 | /* Initialize per-rcu_data variables for no-CBs CPUs. */ |
2429 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) | |
2430 | { | |
2431 | rdp->nocb_tail = &rdp->nocb_head; | |
abedf8e2 | 2432 | init_swait_queue_head(&rdp->nocb_wq); |
fbce7497 | 2433 | rdp->nocb_follower_tail = &rdp->nocb_follower_head; |
8be6e1b1 | 2434 | raw_spin_lock_init(&rdp->nocb_lock); |
fd30b717 | 2435 | timer_setup(&rdp->nocb_timer, do_nocb_deferred_wakeup_timer, 0); |
3fbfbf7a PM |
2436 | } |
2437 | ||
35ce7f29 PM |
2438 | /* |
2439 | * If the specified CPU is a no-CBs CPU that does not already have its | |
2440 | * rcuo kthread for the specified RCU flavor, spawn it. If the CPUs are | |
2441 | * brought online out of order, this can require re-organizing the | |
2442 | * leader-follower relationships. | |
2443 | */ | |
2444 | static void rcu_spawn_one_nocb_kthread(struct rcu_state *rsp, int cpu) | |
2445 | { | |
2446 | struct rcu_data *rdp; | |
2447 | struct rcu_data *rdp_last; | |
2448 | struct rcu_data *rdp_old_leader; | |
2449 | struct rcu_data *rdp_spawn = per_cpu_ptr(rsp->rda, cpu); | |
2450 | struct task_struct *t; | |
2451 | ||
2452 | /* | |
2453 | * If this isn't a no-CBs CPU or if it already has an rcuo kthread, | |
2454 | * then nothing to do. | |
2455 | */ | |
2456 | if (!rcu_is_nocb_cpu(cpu) || rdp_spawn->nocb_kthread) | |
2457 | return; | |
2458 | ||
2459 | /* If we didn't spawn the leader first, reorganize! */ | |
2460 | rdp_old_leader = rdp_spawn->nocb_leader; | |
2461 | if (rdp_old_leader != rdp_spawn && !rdp_old_leader->nocb_kthread) { | |
2462 | rdp_last = NULL; | |
2463 | rdp = rdp_old_leader; | |
2464 | do { | |
2465 | rdp->nocb_leader = rdp_spawn; | |
2466 | if (rdp_last && rdp != rdp_spawn) | |
2467 | rdp_last->nocb_next_follower = rdp; | |
bbe5d7a9 PM |
2468 | if (rdp == rdp_spawn) { |
2469 | rdp = rdp->nocb_next_follower; | |
2470 | } else { | |
2471 | rdp_last = rdp; | |
2472 | rdp = rdp->nocb_next_follower; | |
2473 | rdp_last->nocb_next_follower = NULL; | |
2474 | } | |
35ce7f29 PM |
2475 | } while (rdp); |
2476 | rdp_spawn->nocb_next_follower = rdp_old_leader; | |
2477 | } | |
2478 | ||
2479 | /* Spawn the kthread for this CPU and RCU flavor. */ | |
2480 | t = kthread_run(rcu_nocb_kthread, rdp_spawn, | |
2481 | "rcuo%c/%d", rsp->abbr, cpu); | |
2482 | BUG_ON(IS_ERR(t)); | |
7d0ae808 | 2483 | WRITE_ONCE(rdp_spawn->nocb_kthread, t); |
35ce7f29 PM |
2484 | } |
2485 | ||
2486 | /* | |
2487 | * If the specified CPU is a no-CBs CPU that does not already have its | |
2488 | * rcuo kthreads, spawn them. | |
2489 | */ | |
2490 | static void rcu_spawn_all_nocb_kthreads(int cpu) | |
2491 | { | |
2492 | struct rcu_state *rsp; | |
2493 | ||
2494 | if (rcu_scheduler_fully_active) | |
2495 | for_each_rcu_flavor(rsp) | |
2496 | rcu_spawn_one_nocb_kthread(rsp, cpu); | |
2497 | } | |
2498 | ||
2499 | /* | |
2500 | * Once the scheduler is running, spawn rcuo kthreads for all online | |
2501 | * no-CBs CPUs. This assumes that the early_initcall()s happen before | |
2502 | * non-boot CPUs come online -- if this changes, we will need to add | |
2503 | * some mutual exclusion. | |
2504 | */ | |
2505 | static void __init rcu_spawn_nocb_kthreads(void) | |
2506 | { | |
2507 | int cpu; | |
2508 | ||
2509 | for_each_online_cpu(cpu) | |
2510 | rcu_spawn_all_nocb_kthreads(cpu); | |
2511 | } | |
2512 | ||
fbce7497 PM |
2513 | /* How many follower CPU IDs per leader? Default of -1 for sqrt(nr_cpu_ids). */ |
2514 | static int rcu_nocb_leader_stride = -1; | |
2515 | module_param(rcu_nocb_leader_stride, int, 0444); | |
2516 | ||
2517 | /* | |
35ce7f29 | 2518 | * Initialize leader-follower relationships for all no-CBs CPU. |
fbce7497 | 2519 | */ |
35ce7f29 | 2520 | static void __init rcu_organize_nocb_kthreads(struct rcu_state *rsp) |
3fbfbf7a PM |
2521 | { |
2522 | int cpu; | |
fbce7497 PM |
2523 | int ls = rcu_nocb_leader_stride; |
2524 | int nl = 0; /* Next leader. */ | |
3fbfbf7a | 2525 | struct rcu_data *rdp; |
fbce7497 PM |
2526 | struct rcu_data *rdp_leader = NULL; /* Suppress misguided gcc warn. */ |
2527 | struct rcu_data *rdp_prev = NULL; | |
3fbfbf7a | 2528 | |
84b12b75 | 2529 | if (!cpumask_available(rcu_nocb_mask)) |
3fbfbf7a | 2530 | return; |
fbce7497 PM |
2531 | if (ls == -1) { |
2532 | ls = int_sqrt(nr_cpu_ids); | |
2533 | rcu_nocb_leader_stride = ls; | |
2534 | } | |
2535 | ||
2536 | /* | |
9831ce3b PM |
2537 | * Each pass through this loop sets up one rcu_data structure. |
2538 | * Should the corresponding CPU come online in the future, then | |
2539 | * we will spawn the needed set of rcu_nocb_kthread() kthreads. | |
fbce7497 | 2540 | */ |
3fbfbf7a PM |
2541 | for_each_cpu(cpu, rcu_nocb_mask) { |
2542 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
fbce7497 PM |
2543 | if (rdp->cpu >= nl) { |
2544 | /* New leader, set up for followers & next leader. */ | |
2545 | nl = DIV_ROUND_UP(rdp->cpu + 1, ls) * ls; | |
2546 | rdp->nocb_leader = rdp; | |
2547 | rdp_leader = rdp; | |
2548 | } else { | |
2549 | /* Another follower, link to previous leader. */ | |
2550 | rdp->nocb_leader = rdp_leader; | |
2551 | rdp_prev->nocb_next_follower = rdp; | |
2552 | } | |
2553 | rdp_prev = rdp; | |
3fbfbf7a PM |
2554 | } |
2555 | } | |
2556 | ||
2557 | /* Prevent __call_rcu() from enqueuing callbacks on no-CBs CPUs */ | |
34ed6246 | 2558 | static bool init_nocb_callback_list(struct rcu_data *rdp) |
3fbfbf7a | 2559 | { |
22c2f669 | 2560 | if (!rcu_is_nocb_cpu(rdp->cpu)) |
34ed6246 | 2561 | return false; |
22c2f669 | 2562 | |
34404ca8 | 2563 | /* If there are early-boot callbacks, move them to nocb lists. */ |
15fecf89 PM |
2564 | if (!rcu_segcblist_empty(&rdp->cblist)) { |
2565 | rdp->nocb_head = rcu_segcblist_head(&rdp->cblist); | |
2566 | rdp->nocb_tail = rcu_segcblist_tail(&rdp->cblist); | |
2567 | atomic_long_set(&rdp->nocb_q_count, | |
2568 | rcu_segcblist_n_cbs(&rdp->cblist)); | |
2569 | atomic_long_set(&rdp->nocb_q_count_lazy, | |
2570 | rcu_segcblist_n_lazy_cbs(&rdp->cblist)); | |
2571 | rcu_segcblist_init(&rdp->cblist); | |
34404ca8 | 2572 | } |
15fecf89 | 2573 | rcu_segcblist_disable(&rdp->cblist); |
34ed6246 | 2574 | return true; |
3fbfbf7a PM |
2575 | } |
2576 | ||
34ed6246 PM |
2577 | #else /* #ifdef CONFIG_RCU_NOCB_CPU */ |
2578 | ||
d7e29933 PM |
2579 | static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu) |
2580 | { | |
2581 | WARN_ON_ONCE(1); /* Should be dead code. */ | |
2582 | return false; | |
2583 | } | |
2584 | ||
abedf8e2 | 2585 | static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq) |
3fbfbf7a | 2586 | { |
3fbfbf7a PM |
2587 | } |
2588 | ||
abedf8e2 | 2589 | static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp) |
065bb78c DW |
2590 | { |
2591 | return NULL; | |
2592 | } | |
2593 | ||
dae6e64d PM |
2594 | static void rcu_init_one_nocb(struct rcu_node *rnp) |
2595 | { | |
2596 | } | |
3fbfbf7a | 2597 | |
3fbfbf7a | 2598 | static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, |
96d3fd0d | 2599 | bool lazy, unsigned long flags) |
3fbfbf7a | 2600 | { |
4afc7e26 | 2601 | return false; |
3fbfbf7a PM |
2602 | } |
2603 | ||
b1a2d79f | 2604 | static bool __maybe_unused rcu_nocb_adopt_orphan_cbs(struct rcu_data *my_rdp, |
96d3fd0d PM |
2605 | struct rcu_data *rdp, |
2606 | unsigned long flags) | |
3fbfbf7a | 2607 | { |
f4aa84ba | 2608 | return false; |
3fbfbf7a PM |
2609 | } |
2610 | ||
3fbfbf7a PM |
2611 | static void __init rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp) |
2612 | { | |
2613 | } | |
2614 | ||
9fdd3bc9 | 2615 | static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp) |
96d3fd0d PM |
2616 | { |
2617 | return false; | |
2618 | } | |
2619 | ||
2620 | static void do_nocb_deferred_wakeup(struct rcu_data *rdp) | |
2621 | { | |
2622 | } | |
2623 | ||
35ce7f29 PM |
2624 | static void rcu_spawn_all_nocb_kthreads(int cpu) |
2625 | { | |
2626 | } | |
2627 | ||
2628 | static void __init rcu_spawn_nocb_kthreads(void) | |
3fbfbf7a PM |
2629 | { |
2630 | } | |
2631 | ||
34ed6246 | 2632 | static bool init_nocb_callback_list(struct rcu_data *rdp) |
3fbfbf7a | 2633 | { |
34ed6246 | 2634 | return false; |
3fbfbf7a PM |
2635 | } |
2636 | ||
2637 | #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ | |
65d798f0 PM |
2638 | |
2639 | /* | |
2640 | * An adaptive-ticks CPU can potentially execute in kernel mode for an | |
2641 | * arbitrarily long period of time with the scheduling-clock tick turned | |
2642 | * off. RCU will be paying attention to this CPU because it is in the | |
2643 | * kernel, but the CPU cannot be guaranteed to be executing the RCU state | |
2644 | * machine because the scheduling-clock tick has been disabled. Therefore, | |
2645 | * if an adaptive-ticks CPU is failing to respond to the current grace | |
2646 | * period and has not be idle from an RCU perspective, kick it. | |
2647 | */ | |
4a81e832 | 2648 | static void __maybe_unused rcu_kick_nohz_cpu(int cpu) |
65d798f0 PM |
2649 | { |
2650 | #ifdef CONFIG_NO_HZ_FULL | |
2651 | if (tick_nohz_full_cpu(cpu)) | |
2652 | smp_send_reschedule(cpu); | |
2653 | #endif /* #ifdef CONFIG_NO_HZ_FULL */ | |
2654 | } | |
2333210b | 2655 | |
a096932f PM |
2656 | /* |
2657 | * Is this CPU a NO_HZ_FULL CPU that should ignore RCU so that the | |
2658 | * grace-period kthread will do force_quiescent_state() processing? | |
2659 | * The idea is to avoid waking up RCU core processing on such a | |
2660 | * CPU unless the grace period has extended for too long. | |
2661 | * | |
2662 | * This code relies on the fact that all NO_HZ_FULL CPUs are also | |
52e2bb95 | 2663 | * CONFIG_RCU_NOCB_CPU CPUs. |
a096932f PM |
2664 | */ |
2665 | static bool rcu_nohz_full_cpu(struct rcu_state *rsp) | |
2666 | { | |
2667 | #ifdef CONFIG_NO_HZ_FULL | |
2668 | if (tick_nohz_full_cpu(smp_processor_id()) && | |
2669 | (!rcu_gp_in_progress(rsp) || | |
7d0ae808 | 2670 | ULONG_CMP_LT(jiffies, READ_ONCE(rsp->gp_start) + HZ))) |
5ce035fb | 2671 | return true; |
a096932f | 2672 | #endif /* #ifdef CONFIG_NO_HZ_FULL */ |
5ce035fb | 2673 | return false; |
a096932f | 2674 | } |
5057f55e PM |
2675 | |
2676 | /* | |
265f5f28 | 2677 | * Bind the RCU grace-period kthreads to the housekeeping CPU. |
5057f55e PM |
2678 | */ |
2679 | static void rcu_bind_gp_kthread(void) | |
2680 | { | |
c0f489d2 | 2681 | int __maybe_unused cpu; |
5057f55e | 2682 | |
c0f489d2 | 2683 | if (!tick_nohz_full_enabled()) |
5057f55e | 2684 | return; |
de201559 | 2685 | housekeeping_affine(current, HK_FLAG_RCU); |
5057f55e | 2686 | } |
176f8f7a PM |
2687 | |
2688 | /* Record the current task on dyntick-idle entry. */ | |
2689 | static void rcu_dynticks_task_enter(void) | |
2690 | { | |
2691 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 2692 | WRITE_ONCE(current->rcu_tasks_idle_cpu, smp_processor_id()); |
176f8f7a PM |
2693 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
2694 | } | |
2695 | ||
2696 | /* Record no current task on dyntick-idle exit. */ | |
2697 | static void rcu_dynticks_task_exit(void) | |
2698 | { | |
2699 | #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) | |
7d0ae808 | 2700 | WRITE_ONCE(current->rcu_tasks_idle_cpu, -1); |
176f8f7a PM |
2701 | #endif /* #if defined(CONFIG_TASKS_RCU) && defined(CONFIG_NO_HZ_FULL) */ |
2702 | } |