Commit | Line | Data |
---|---|---|
64db4cff PM |
1 | /* |
2 | * Read-Copy Update mechanism for mutual exclusion | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License as published by | |
6 | * the Free Software Foundation; either version 2 of the License, or | |
7 | * (at your option) any later version. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
87de1cfd PM |
15 | * along with this program; if not, you can access it online at |
16 | * http://www.gnu.org/licenses/gpl-2.0.html. | |
64db4cff PM |
17 | * |
18 | * Copyright IBM Corporation, 2008 | |
19 | * | |
20 | * Authors: Dipankar Sarma <dipankar@in.ibm.com> | |
21 | * Manfred Spraul <manfred@colorfullife.com> | |
22 | * Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical version | |
23 | * | |
24 | * Based on the original work by Paul McKenney <paulmck@us.ibm.com> | |
25 | * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen. | |
26 | * | |
27 | * For detailed explanation of Read-Copy Update mechanism see - | |
a71fca58 | 28 | * Documentation/RCU |
64db4cff PM |
29 | */ |
30 | #include <linux/types.h> | |
31 | #include <linux/kernel.h> | |
32 | #include <linux/init.h> | |
33 | #include <linux/spinlock.h> | |
34 | #include <linux/smp.h> | |
35 | #include <linux/rcupdate.h> | |
36 | #include <linux/interrupt.h> | |
37 | #include <linux/sched.h> | |
c1dc0b9c | 38 | #include <linux/nmi.h> |
8826f3b0 | 39 | #include <linux/atomic.h> |
64db4cff | 40 | #include <linux/bitops.h> |
9984de1a | 41 | #include <linux/export.h> |
64db4cff PM |
42 | #include <linux/completion.h> |
43 | #include <linux/moduleparam.h> | |
4102adab | 44 | #include <linux/module.h> |
64db4cff PM |
45 | #include <linux/percpu.h> |
46 | #include <linux/notifier.h> | |
47 | #include <linux/cpu.h> | |
48 | #include <linux/mutex.h> | |
49 | #include <linux/time.h> | |
bbad9379 | 50 | #include <linux/kernel_stat.h> |
a26ac245 PM |
51 | #include <linux/wait.h> |
52 | #include <linux/kthread.h> | |
268bb0ce | 53 | #include <linux/prefetch.h> |
3d3b7db0 PM |
54 | #include <linux/delay.h> |
55 | #include <linux/stop_machine.h> | |
661a85dc | 56 | #include <linux/random.h> |
af658dca | 57 | #include <linux/trace_events.h> |
d1d74d14 | 58 | #include <linux/suspend.h> |
64db4cff | 59 | |
4102adab | 60 | #include "tree.h" |
29c00b4a | 61 | #include "rcu.h" |
9f77da9f | 62 | |
4102adab PM |
63 | MODULE_ALIAS("rcutree"); |
64 | #ifdef MODULE_PARAM_PREFIX | |
65 | #undef MODULE_PARAM_PREFIX | |
66 | #endif | |
67 | #define MODULE_PARAM_PREFIX "rcutree." | |
68 | ||
64db4cff PM |
69 | /* Data structures. */ |
70 | ||
f7f7bac9 SRRH |
71 | /* |
72 | * In order to export the rcu_state name to the tracing tools, it | |
73 | * needs to be added in the __tracepoint_string section. | |
74 | * This requires defining a separate variable tp_<sname>_varname | |
75 | * that points to the string being used, and this will allow | |
76 | * the tracing userspace tools to be able to decipher the string | |
77 | * address to the matching string. | |
78 | */ | |
a8a29b3b AB |
79 | #ifdef CONFIG_TRACING |
80 | # define DEFINE_RCU_TPS(sname) \ | |
f7f7bac9 | 81 | static char sname##_varname[] = #sname; \ |
a8a29b3b AB |
82 | static const char *tp_##sname##_varname __used __tracepoint_string = sname##_varname; |
83 | # define RCU_STATE_NAME(sname) sname##_varname | |
84 | #else | |
85 | # define DEFINE_RCU_TPS(sname) | |
86 | # define RCU_STATE_NAME(sname) __stringify(sname) | |
87 | #endif | |
88 | ||
89 | #define RCU_STATE_INITIALIZER(sname, sabbr, cr) \ | |
90 | DEFINE_RCU_TPS(sname) \ | |
c92fb057 | 91 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rcu_data, sname##_data); \ |
a41bfeb2 | 92 | struct rcu_state sname##_state = { \ |
6c90cc7b | 93 | .level = { &sname##_state.node[0] }, \ |
2723249a | 94 | .rda = &sname##_data, \ |
037b64ed | 95 | .call = cr, \ |
77f81fe0 | 96 | .gp_state = RCU_GP_IDLE, \ |
42c3533e PM |
97 | .gpnum = 0UL - 300UL, \ |
98 | .completed = 0UL - 300UL, \ | |
7b2e6011 | 99 | .orphan_lock = __RAW_SPIN_LOCK_UNLOCKED(&sname##_state.orphan_lock), \ |
6c90cc7b PM |
100 | .orphan_nxttail = &sname##_state.orphan_nxtlist, \ |
101 | .orphan_donetail = &sname##_state.orphan_donelist, \ | |
7be7f0be | 102 | .barrier_mutex = __MUTEX_INITIALIZER(sname##_state.barrier_mutex), \ |
a8a29b3b | 103 | .name = RCU_STATE_NAME(sname), \ |
a4889858 | 104 | .abbr = sabbr, \ |
f6a12f34 | 105 | .exp_mutex = __MUTEX_INITIALIZER(sname##_state.exp_mutex), \ |
3b5f668e | 106 | .exp_wake_mutex = __MUTEX_INITIALIZER(sname##_state.exp_wake_mutex), \ |
2723249a | 107 | } |
64db4cff | 108 | |
a41bfeb2 SRRH |
109 | RCU_STATE_INITIALIZER(rcu_sched, 's', call_rcu_sched); |
110 | RCU_STATE_INITIALIZER(rcu_bh, 'b', call_rcu_bh); | |
b1f77b05 | 111 | |
b28a7c01 | 112 | static struct rcu_state *const rcu_state_p; |
6ce75a23 | 113 | LIST_HEAD(rcu_struct_flavors); |
27f4d280 | 114 | |
a3dc2948 PM |
115 | /* Dump rcu_node combining tree at boot to verify correct setup. */ |
116 | static bool dump_tree; | |
117 | module_param(dump_tree, bool, 0444); | |
7fa27001 PM |
118 | /* Control rcu_node-tree auto-balancing at boot time. */ |
119 | static bool rcu_fanout_exact; | |
120 | module_param(rcu_fanout_exact, bool, 0444); | |
47d631af PM |
121 | /* Increase (but not decrease) the RCU_FANOUT_LEAF at boot time. */ |
122 | static int rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
7e5c2dfb | 123 | module_param(rcu_fanout_leaf, int, 0444); |
f885b7f2 | 124 | int rcu_num_lvls __read_mostly = RCU_NUM_LVLS; |
cb007102 AG |
125 | /* Number of rcu_nodes at specified level. */ |
126 | static int num_rcu_lvl[] = NUM_RCU_LVL_INIT; | |
f885b7f2 PM |
127 | int rcu_num_nodes __read_mostly = NUM_RCU_NODES; /* Total # rcu_nodes in use. */ |
128 | ||
b0d30417 PM |
129 | /* |
130 | * The rcu_scheduler_active variable transitions from zero to one just | |
131 | * before the first task is spawned. So when this variable is zero, RCU | |
132 | * can assume that there is but one task, allowing RCU to (for example) | |
b44f6656 | 133 | * optimize synchronize_sched() to a simple barrier(). When this variable |
b0d30417 PM |
134 | * is one, RCU must actually do all the hard work required to detect real |
135 | * grace periods. This variable is also used to suppress boot-time false | |
136 | * positives from lockdep-RCU error checking. | |
137 | */ | |
bbad9379 PM |
138 | int rcu_scheduler_active __read_mostly; |
139 | EXPORT_SYMBOL_GPL(rcu_scheduler_active); | |
140 | ||
b0d30417 PM |
141 | /* |
142 | * The rcu_scheduler_fully_active variable transitions from zero to one | |
143 | * during the early_initcall() processing, which is after the scheduler | |
144 | * is capable of creating new tasks. So RCU processing (for example, | |
145 | * creating tasks for RCU priority boosting) must be delayed until after | |
146 | * rcu_scheduler_fully_active transitions from zero to one. We also | |
147 | * currently delay invocation of any RCU callbacks until after this point. | |
148 | * | |
149 | * It might later prove better for people registering RCU callbacks during | |
150 | * early boot to take responsibility for these callbacks, but one step at | |
151 | * a time. | |
152 | */ | |
153 | static int rcu_scheduler_fully_active __read_mostly; | |
154 | ||
0aa04b05 PM |
155 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf); |
156 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf); | |
5d01bbd1 | 157 | static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); |
a46e0899 PM |
158 | static void invoke_rcu_core(void); |
159 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp); | |
6587a23b PM |
160 | static void rcu_report_exp_rdp(struct rcu_state *rsp, |
161 | struct rcu_data *rdp, bool wake); | |
a26ac245 | 162 | |
a94844b2 | 163 | /* rcuc/rcub kthread realtime priority */ |
26730f55 | 164 | #ifdef CONFIG_RCU_KTHREAD_PRIO |
a94844b2 | 165 | static int kthread_prio = CONFIG_RCU_KTHREAD_PRIO; |
26730f55 PM |
166 | #else /* #ifdef CONFIG_RCU_KTHREAD_PRIO */ |
167 | static int kthread_prio = IS_ENABLED(CONFIG_RCU_BOOST) ? 1 : 0; | |
168 | #endif /* #else #ifdef CONFIG_RCU_KTHREAD_PRIO */ | |
a94844b2 PM |
169 | module_param(kthread_prio, int, 0644); |
170 | ||
8d7dc928 | 171 | /* Delay in jiffies for grace-period initialization delays, debug only. */ |
0f41c0dd PM |
172 | |
173 | #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_PREINIT | |
174 | static int gp_preinit_delay = CONFIG_RCU_TORTURE_TEST_SLOW_PREINIT_DELAY; | |
175 | module_param(gp_preinit_delay, int, 0644); | |
176 | #else /* #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_PREINIT */ | |
177 | static const int gp_preinit_delay; | |
178 | #endif /* #else #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_PREINIT */ | |
179 | ||
8d7dc928 PM |
180 | #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_INIT |
181 | static int gp_init_delay = CONFIG_RCU_TORTURE_TEST_SLOW_INIT_DELAY; | |
37745d28 | 182 | module_param(gp_init_delay, int, 0644); |
8d7dc928 PM |
183 | #else /* #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_INIT */ |
184 | static const int gp_init_delay; | |
185 | #endif /* #else #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_INIT */ | |
eab128e8 | 186 | |
0f41c0dd PM |
187 | #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_CLEANUP |
188 | static int gp_cleanup_delay = CONFIG_RCU_TORTURE_TEST_SLOW_CLEANUP_DELAY; | |
189 | module_param(gp_cleanup_delay, int, 0644); | |
190 | #else /* #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_CLEANUP */ | |
191 | static const int gp_cleanup_delay; | |
192 | #endif /* #else #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_CLEANUP */ | |
193 | ||
eab128e8 PM |
194 | /* |
195 | * Number of grace periods between delays, normalized by the duration of | |
196 | * the delay. The longer the the delay, the more the grace periods between | |
197 | * each delay. The reason for this normalization is that it means that, | |
198 | * for non-zero delays, the overall slowdown of grace periods is constant | |
199 | * regardless of the duration of the delay. This arrangement balances | |
200 | * the need for long delays to increase some race probabilities with the | |
201 | * need for fast grace periods to increase other race probabilities. | |
202 | */ | |
203 | #define PER_RCU_NODE_PERIOD 3 /* Number of grace periods between delays. */ | |
37745d28 | 204 | |
4a298656 PM |
205 | /* |
206 | * Track the rcutorture test sequence number and the update version | |
207 | * number within a given test. The rcutorture_testseq is incremented | |
208 | * on every rcutorture module load and unload, so has an odd value | |
209 | * when a test is running. The rcutorture_vernum is set to zero | |
210 | * when rcutorture starts and is incremented on each rcutorture update. | |
211 | * These variables enable correlating rcutorture output with the | |
212 | * RCU tracing information. | |
213 | */ | |
214 | unsigned long rcutorture_testseq; | |
215 | unsigned long rcutorture_vernum; | |
216 | ||
0aa04b05 PM |
217 | /* |
218 | * Compute the mask of online CPUs for the specified rcu_node structure. | |
219 | * This will not be stable unless the rcu_node structure's ->lock is | |
220 | * held, but the bit corresponding to the current CPU will be stable | |
221 | * in most contexts. | |
222 | */ | |
223 | unsigned long rcu_rnp_online_cpus(struct rcu_node *rnp) | |
224 | { | |
7d0ae808 | 225 | return READ_ONCE(rnp->qsmaskinitnext); |
0aa04b05 PM |
226 | } |
227 | ||
fc2219d4 | 228 | /* |
7d0ae808 | 229 | * Return true if an RCU grace period is in progress. The READ_ONCE()s |
fc2219d4 PM |
230 | * permit this function to be invoked without holding the root rcu_node |
231 | * structure's ->lock, but of course results can be subject to change. | |
232 | */ | |
233 | static int rcu_gp_in_progress(struct rcu_state *rsp) | |
234 | { | |
7d0ae808 | 235 | return READ_ONCE(rsp->completed) != READ_ONCE(rsp->gpnum); |
fc2219d4 PM |
236 | } |
237 | ||
b1f77b05 | 238 | /* |
d6714c22 | 239 | * Note a quiescent state. Because we do not need to know |
b1f77b05 | 240 | * how many quiescent states passed, just if there was at least |
d6714c22 | 241 | * one since the start of the grace period, this just sets a flag. |
e4cc1f22 | 242 | * The caller must have disabled preemption. |
b1f77b05 | 243 | */ |
284a8c93 | 244 | void rcu_sched_qs(void) |
b1f77b05 | 245 | { |
fecbf6f0 PM |
246 | if (!__this_cpu_read(rcu_sched_data.cpu_no_qs.s)) |
247 | return; | |
248 | trace_rcu_grace_period(TPS("rcu_sched"), | |
249 | __this_cpu_read(rcu_sched_data.gpnum), | |
250 | TPS("cpuqs")); | |
251 | __this_cpu_write(rcu_sched_data.cpu_no_qs.b.norm, false); | |
252 | if (!__this_cpu_read(rcu_sched_data.cpu_no_qs.b.exp)) | |
253 | return; | |
46a5d164 PM |
254 | __this_cpu_write(rcu_sched_data.cpu_no_qs.b.exp, false); |
255 | rcu_report_exp_rdp(&rcu_sched_state, | |
256 | this_cpu_ptr(&rcu_sched_data), true); | |
b1f77b05 IM |
257 | } |
258 | ||
284a8c93 | 259 | void rcu_bh_qs(void) |
b1f77b05 | 260 | { |
5b74c458 | 261 | if (__this_cpu_read(rcu_bh_data.cpu_no_qs.s)) { |
284a8c93 PM |
262 | trace_rcu_grace_period(TPS("rcu_bh"), |
263 | __this_cpu_read(rcu_bh_data.gpnum), | |
264 | TPS("cpuqs")); | |
5b74c458 | 265 | __this_cpu_write(rcu_bh_data.cpu_no_qs.b.norm, false); |
284a8c93 | 266 | } |
b1f77b05 | 267 | } |
64db4cff | 268 | |
4a81e832 PM |
269 | static DEFINE_PER_CPU(int, rcu_sched_qs_mask); |
270 | ||
271 | static DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { | |
272 | .dynticks_nesting = DYNTICK_TASK_EXIT_IDLE, | |
273 | .dynticks = ATOMIC_INIT(1), | |
274 | #ifdef CONFIG_NO_HZ_FULL_SYSIDLE | |
275 | .dynticks_idle_nesting = DYNTICK_TASK_NEST_VALUE, | |
276 | .dynticks_idle = ATOMIC_INIT(1), | |
277 | #endif /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ | |
278 | }; | |
279 | ||
5cd37193 PM |
280 | DEFINE_PER_CPU_SHARED_ALIGNED(unsigned long, rcu_qs_ctr); |
281 | EXPORT_PER_CPU_SYMBOL_GPL(rcu_qs_ctr); | |
282 | ||
4a81e832 PM |
283 | /* |
284 | * Let the RCU core know that this CPU has gone through the scheduler, | |
285 | * which is a quiescent state. This is called when the need for a | |
286 | * quiescent state is urgent, so we burn an atomic operation and full | |
287 | * memory barriers to let the RCU core know about it, regardless of what | |
288 | * this CPU might (or might not) do in the near future. | |
289 | * | |
290 | * We inform the RCU core by emulating a zero-duration dyntick-idle | |
291 | * period, which we in turn do by incrementing the ->dynticks counter | |
292 | * by two. | |
46a5d164 PM |
293 | * |
294 | * The caller must have disabled interrupts. | |
4a81e832 PM |
295 | */ |
296 | static void rcu_momentary_dyntick_idle(void) | |
297 | { | |
4a81e832 PM |
298 | struct rcu_data *rdp; |
299 | struct rcu_dynticks *rdtp; | |
300 | int resched_mask; | |
301 | struct rcu_state *rsp; | |
302 | ||
4a81e832 PM |
303 | /* |
304 | * Yes, we can lose flag-setting operations. This is OK, because | |
305 | * the flag will be set again after some delay. | |
306 | */ | |
307 | resched_mask = raw_cpu_read(rcu_sched_qs_mask); | |
308 | raw_cpu_write(rcu_sched_qs_mask, 0); | |
309 | ||
310 | /* Find the flavor that needs a quiescent state. */ | |
311 | for_each_rcu_flavor(rsp) { | |
312 | rdp = raw_cpu_ptr(rsp->rda); | |
313 | if (!(resched_mask & rsp->flavor_mask)) | |
314 | continue; | |
315 | smp_mb(); /* rcu_sched_qs_mask before cond_resched_completed. */ | |
7d0ae808 PM |
316 | if (READ_ONCE(rdp->mynode->completed) != |
317 | READ_ONCE(rdp->cond_resched_completed)) | |
4a81e832 PM |
318 | continue; |
319 | ||
320 | /* | |
321 | * Pretend to be momentarily idle for the quiescent state. | |
322 | * This allows the grace-period kthread to record the | |
323 | * quiescent state, with no need for this CPU to do anything | |
324 | * further. | |
325 | */ | |
326 | rdtp = this_cpu_ptr(&rcu_dynticks); | |
327 | smp_mb__before_atomic(); /* Earlier stuff before QS. */ | |
328 | atomic_add(2, &rdtp->dynticks); /* QS. */ | |
329 | smp_mb__after_atomic(); /* Later stuff after QS. */ | |
330 | break; | |
331 | } | |
4a81e832 PM |
332 | } |
333 | ||
25502a6c PM |
334 | /* |
335 | * Note a context switch. This is a quiescent state for RCU-sched, | |
336 | * and requires special handling for preemptible RCU. | |
46a5d164 | 337 | * The caller must have disabled interrupts. |
25502a6c | 338 | */ |
38200cf2 | 339 | void rcu_note_context_switch(void) |
25502a6c | 340 | { |
bb73c52b | 341 | barrier(); /* Avoid RCU read-side critical sections leaking down. */ |
f7f7bac9 | 342 | trace_rcu_utilization(TPS("Start context switch")); |
284a8c93 | 343 | rcu_sched_qs(); |
38200cf2 | 344 | rcu_preempt_note_context_switch(); |
4a81e832 PM |
345 | if (unlikely(raw_cpu_read(rcu_sched_qs_mask))) |
346 | rcu_momentary_dyntick_idle(); | |
f7f7bac9 | 347 | trace_rcu_utilization(TPS("End context switch")); |
bb73c52b | 348 | barrier(); /* Avoid RCU read-side critical sections leaking up. */ |
25502a6c | 349 | } |
29ce8310 | 350 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
25502a6c | 351 | |
5cd37193 | 352 | /* |
1925d196 | 353 | * Register a quiescent state for all RCU flavors. If there is an |
5cd37193 PM |
354 | * emergency, invoke rcu_momentary_dyntick_idle() to do a heavy-weight |
355 | * dyntick-idle quiescent state visible to other CPUs (but only for those | |
1925d196 | 356 | * RCU flavors in desperate need of a quiescent state, which will normally |
5cd37193 PM |
357 | * be none of them). Either way, do a lightweight quiescent state for |
358 | * all RCU flavors. | |
bb73c52b BF |
359 | * |
360 | * The barrier() calls are redundant in the common case when this is | |
361 | * called externally, but just in case this is called from within this | |
362 | * file. | |
363 | * | |
5cd37193 PM |
364 | */ |
365 | void rcu_all_qs(void) | |
366 | { | |
46a5d164 PM |
367 | unsigned long flags; |
368 | ||
bb73c52b | 369 | barrier(); /* Avoid RCU read-side critical sections leaking down. */ |
46a5d164 PM |
370 | if (unlikely(raw_cpu_read(rcu_sched_qs_mask))) { |
371 | local_irq_save(flags); | |
5cd37193 | 372 | rcu_momentary_dyntick_idle(); |
46a5d164 PM |
373 | local_irq_restore(flags); |
374 | } | |
a1e12248 PM |
375 | if (unlikely(raw_cpu_read(rcu_sched_data.cpu_no_qs.b.exp))) { |
376 | /* | |
377 | * Yes, we just checked a per-CPU variable with preemption | |
378 | * enabled, so we might be migrated to some other CPU at | |
379 | * this point. That is OK because in that case, the | |
380 | * migration will supply the needed quiescent state. | |
381 | * We might end up needlessly disabling preemption and | |
382 | * invoking rcu_sched_qs() on the destination CPU, but | |
383 | * the probability and cost are both quite low, so this | |
384 | * should not be a problem in practice. | |
385 | */ | |
386 | preempt_disable(); | |
387 | rcu_sched_qs(); | |
388 | preempt_enable(); | |
389 | } | |
5cd37193 | 390 | this_cpu_inc(rcu_qs_ctr); |
bb73c52b | 391 | barrier(); /* Avoid RCU read-side critical sections leaking up. */ |
5cd37193 PM |
392 | } |
393 | EXPORT_SYMBOL_GPL(rcu_all_qs); | |
394 | ||
878d7439 ED |
395 | static long blimit = 10; /* Maximum callbacks per rcu_do_batch. */ |
396 | static long qhimark = 10000; /* If this many pending, ignore blimit. */ | |
397 | static long qlowmark = 100; /* Once only this many pending, use blimit. */ | |
64db4cff | 398 | |
878d7439 ED |
399 | module_param(blimit, long, 0444); |
400 | module_param(qhimark, long, 0444); | |
401 | module_param(qlowmark, long, 0444); | |
3d76c082 | 402 | |
026ad283 PM |
403 | static ulong jiffies_till_first_fqs = ULONG_MAX; |
404 | static ulong jiffies_till_next_fqs = ULONG_MAX; | |
8c7c4829 | 405 | static bool rcu_kick_kthreads; |
d40011f6 PM |
406 | |
407 | module_param(jiffies_till_first_fqs, ulong, 0644); | |
408 | module_param(jiffies_till_next_fqs, ulong, 0644); | |
8c7c4829 | 409 | module_param(rcu_kick_kthreads, bool, 0644); |
d40011f6 | 410 | |
4a81e832 PM |
411 | /* |
412 | * How long the grace period must be before we start recruiting | |
413 | * quiescent-state help from rcu_note_context_switch(). | |
414 | */ | |
415 | static ulong jiffies_till_sched_qs = HZ / 20; | |
416 | module_param(jiffies_till_sched_qs, ulong, 0644); | |
417 | ||
48a7639c | 418 | static bool rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp, |
910ee45d | 419 | struct rcu_data *rdp); |
217af2a2 PM |
420 | static void force_qs_rnp(struct rcu_state *rsp, |
421 | int (*f)(struct rcu_data *rsp, bool *isidle, | |
422 | unsigned long *maxj), | |
423 | bool *isidle, unsigned long *maxj); | |
4cdfc175 | 424 | static void force_quiescent_state(struct rcu_state *rsp); |
e3950ecd | 425 | static int rcu_pending(void); |
64db4cff PM |
426 | |
427 | /* | |
917963d0 | 428 | * Return the number of RCU batches started thus far for debug & stats. |
64db4cff | 429 | */ |
917963d0 PM |
430 | unsigned long rcu_batches_started(void) |
431 | { | |
432 | return rcu_state_p->gpnum; | |
433 | } | |
434 | EXPORT_SYMBOL_GPL(rcu_batches_started); | |
435 | ||
436 | /* | |
437 | * Return the number of RCU-sched batches started thus far for debug & stats. | |
64db4cff | 438 | */ |
917963d0 PM |
439 | unsigned long rcu_batches_started_sched(void) |
440 | { | |
441 | return rcu_sched_state.gpnum; | |
442 | } | |
443 | EXPORT_SYMBOL_GPL(rcu_batches_started_sched); | |
444 | ||
445 | /* | |
446 | * Return the number of RCU BH batches started thus far for debug & stats. | |
447 | */ | |
448 | unsigned long rcu_batches_started_bh(void) | |
449 | { | |
450 | return rcu_bh_state.gpnum; | |
451 | } | |
452 | EXPORT_SYMBOL_GPL(rcu_batches_started_bh); | |
453 | ||
454 | /* | |
455 | * Return the number of RCU batches completed thus far for debug & stats. | |
456 | */ | |
457 | unsigned long rcu_batches_completed(void) | |
458 | { | |
459 | return rcu_state_p->completed; | |
460 | } | |
461 | EXPORT_SYMBOL_GPL(rcu_batches_completed); | |
462 | ||
463 | /* | |
464 | * Return the number of RCU-sched batches completed thus far for debug & stats. | |
64db4cff | 465 | */ |
9733e4f0 | 466 | unsigned long rcu_batches_completed_sched(void) |
64db4cff | 467 | { |
d6714c22 | 468 | return rcu_sched_state.completed; |
64db4cff | 469 | } |
d6714c22 | 470 | EXPORT_SYMBOL_GPL(rcu_batches_completed_sched); |
64db4cff PM |
471 | |
472 | /* | |
917963d0 | 473 | * Return the number of RCU BH batches completed thus far for debug & stats. |
64db4cff | 474 | */ |
9733e4f0 | 475 | unsigned long rcu_batches_completed_bh(void) |
64db4cff PM |
476 | { |
477 | return rcu_bh_state.completed; | |
478 | } | |
479 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); | |
480 | ||
291783b8 PM |
481 | /* |
482 | * Return the number of RCU expedited batches completed thus far for | |
483 | * debug & stats. Odd numbers mean that a batch is in progress, even | |
484 | * numbers mean idle. The value returned will thus be roughly double | |
485 | * the cumulative batches since boot. | |
486 | */ | |
487 | unsigned long rcu_exp_batches_completed(void) | |
488 | { | |
489 | return rcu_state_p->expedited_sequence; | |
490 | } | |
491 | EXPORT_SYMBOL_GPL(rcu_exp_batches_completed); | |
492 | ||
493 | /* | |
494 | * Return the number of RCU-sched expedited batches completed thus far | |
495 | * for debug & stats. Similar to rcu_exp_batches_completed(). | |
496 | */ | |
497 | unsigned long rcu_exp_batches_completed_sched(void) | |
498 | { | |
499 | return rcu_sched_state.expedited_sequence; | |
500 | } | |
501 | EXPORT_SYMBOL_GPL(rcu_exp_batches_completed_sched); | |
502 | ||
a381d757 ACB |
503 | /* |
504 | * Force a quiescent state. | |
505 | */ | |
506 | void rcu_force_quiescent_state(void) | |
507 | { | |
e534165b | 508 | force_quiescent_state(rcu_state_p); |
a381d757 ACB |
509 | } |
510 | EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); | |
511 | ||
bf66f18e PM |
512 | /* |
513 | * Force a quiescent state for RCU BH. | |
514 | */ | |
515 | void rcu_bh_force_quiescent_state(void) | |
516 | { | |
4cdfc175 | 517 | force_quiescent_state(&rcu_bh_state); |
bf66f18e PM |
518 | } |
519 | EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state); | |
520 | ||
e7580f33 PM |
521 | /* |
522 | * Force a quiescent state for RCU-sched. | |
523 | */ | |
524 | void rcu_sched_force_quiescent_state(void) | |
525 | { | |
526 | force_quiescent_state(&rcu_sched_state); | |
527 | } | |
528 | EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state); | |
529 | ||
afea227f PM |
530 | /* |
531 | * Show the state of the grace-period kthreads. | |
532 | */ | |
533 | void show_rcu_gp_kthreads(void) | |
534 | { | |
535 | struct rcu_state *rsp; | |
536 | ||
537 | for_each_rcu_flavor(rsp) { | |
538 | pr_info("%s: wait state: %d ->state: %#lx\n", | |
539 | rsp->name, rsp->gp_state, rsp->gp_kthread->state); | |
540 | /* sched_show_task(rsp->gp_kthread); */ | |
541 | } | |
542 | } | |
543 | EXPORT_SYMBOL_GPL(show_rcu_gp_kthreads); | |
544 | ||
4a298656 PM |
545 | /* |
546 | * Record the number of times rcutorture tests have been initiated and | |
547 | * terminated. This information allows the debugfs tracing stats to be | |
548 | * correlated to the rcutorture messages, even when the rcutorture module | |
549 | * is being repeatedly loaded and unloaded. In other words, we cannot | |
550 | * store this state in rcutorture itself. | |
551 | */ | |
552 | void rcutorture_record_test_transition(void) | |
553 | { | |
554 | rcutorture_testseq++; | |
555 | rcutorture_vernum = 0; | |
556 | } | |
557 | EXPORT_SYMBOL_GPL(rcutorture_record_test_transition); | |
558 | ||
ad0dc7f9 PM |
559 | /* |
560 | * Send along grace-period-related data for rcutorture diagnostics. | |
561 | */ | |
562 | void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags, | |
563 | unsigned long *gpnum, unsigned long *completed) | |
564 | { | |
565 | struct rcu_state *rsp = NULL; | |
566 | ||
567 | switch (test_type) { | |
568 | case RCU_FLAVOR: | |
e534165b | 569 | rsp = rcu_state_p; |
ad0dc7f9 PM |
570 | break; |
571 | case RCU_BH_FLAVOR: | |
572 | rsp = &rcu_bh_state; | |
573 | break; | |
574 | case RCU_SCHED_FLAVOR: | |
575 | rsp = &rcu_sched_state; | |
576 | break; | |
577 | default: | |
578 | break; | |
579 | } | |
580 | if (rsp != NULL) { | |
7d0ae808 PM |
581 | *flags = READ_ONCE(rsp->gp_flags); |
582 | *gpnum = READ_ONCE(rsp->gpnum); | |
583 | *completed = READ_ONCE(rsp->completed); | |
ad0dc7f9 PM |
584 | return; |
585 | } | |
586 | *flags = 0; | |
587 | *gpnum = 0; | |
588 | *completed = 0; | |
589 | } | |
590 | EXPORT_SYMBOL_GPL(rcutorture_get_gp_data); | |
591 | ||
4a298656 PM |
592 | /* |
593 | * Record the number of writer passes through the current rcutorture test. | |
594 | * This is also used to correlate debugfs tracing stats with the rcutorture | |
595 | * messages. | |
596 | */ | |
597 | void rcutorture_record_progress(unsigned long vernum) | |
598 | { | |
599 | rcutorture_vernum++; | |
600 | } | |
601 | EXPORT_SYMBOL_GPL(rcutorture_record_progress); | |
602 | ||
64db4cff PM |
603 | /* |
604 | * Does the CPU have callbacks ready to be invoked? | |
605 | */ | |
606 | static int | |
607 | cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) | |
608 | { | |
3fbfbf7a PM |
609 | return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL] && |
610 | rdp->nxttail[RCU_DONE_TAIL] != NULL; | |
64db4cff PM |
611 | } |
612 | ||
365187fb PM |
613 | /* |
614 | * Return the root node of the specified rcu_state structure. | |
615 | */ | |
616 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) | |
617 | { | |
618 | return &rsp->node[0]; | |
619 | } | |
620 | ||
621 | /* | |
622 | * Is there any need for future grace periods? | |
623 | * Interrupts must be disabled. If the caller does not hold the root | |
624 | * rnp_node structure's ->lock, the results are advisory only. | |
625 | */ | |
626 | static int rcu_future_needs_gp(struct rcu_state *rsp) | |
627 | { | |
628 | struct rcu_node *rnp = rcu_get_root(rsp); | |
7d0ae808 | 629 | int idx = (READ_ONCE(rnp->completed) + 1) & 0x1; |
365187fb PM |
630 | int *fp = &rnp->need_future_gp[idx]; |
631 | ||
7d0ae808 | 632 | return READ_ONCE(*fp); |
365187fb PM |
633 | } |
634 | ||
64db4cff | 635 | /* |
dc35c893 PM |
636 | * Does the current CPU require a not-yet-started grace period? |
637 | * The caller must have disabled interrupts to prevent races with | |
638 | * normal callback registry. | |
64db4cff | 639 | */ |
d117c8aa | 640 | static bool |
64db4cff PM |
641 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) |
642 | { | |
dc35c893 | 643 | int i; |
3fbfbf7a | 644 | |
dc35c893 | 645 | if (rcu_gp_in_progress(rsp)) |
d117c8aa | 646 | return false; /* No, a grace period is already in progress. */ |
365187fb | 647 | if (rcu_future_needs_gp(rsp)) |
d117c8aa | 648 | return true; /* Yes, a no-CBs CPU needs one. */ |
dc35c893 | 649 | if (!rdp->nxttail[RCU_NEXT_TAIL]) |
d117c8aa | 650 | return false; /* No, this is a no-CBs (or offline) CPU. */ |
dc35c893 | 651 | if (*rdp->nxttail[RCU_NEXT_READY_TAIL]) |
d117c8aa | 652 | return true; /* Yes, CPU has newly registered callbacks. */ |
dc35c893 PM |
653 | for (i = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++) |
654 | if (rdp->nxttail[i - 1] != rdp->nxttail[i] && | |
7d0ae808 | 655 | ULONG_CMP_LT(READ_ONCE(rsp->completed), |
dc35c893 | 656 | rdp->nxtcompleted[i])) |
d117c8aa PM |
657 | return true; /* Yes, CBs for future grace period. */ |
658 | return false; /* No grace period needed. */ | |
64db4cff PM |
659 | } |
660 | ||
9b2e4f18 | 661 | /* |
adf5091e | 662 | * rcu_eqs_enter_common - current CPU is moving towards extended quiescent state |
9b2e4f18 PM |
663 | * |
664 | * If the new value of the ->dynticks_nesting counter now is zero, | |
665 | * we really have entered idle, and must do the appropriate accounting. | |
666 | * The caller must have disabled interrupts. | |
667 | */ | |
28ced795 | 668 | static void rcu_eqs_enter_common(long long oldval, bool user) |
9b2e4f18 | 669 | { |
96d3fd0d PM |
670 | struct rcu_state *rsp; |
671 | struct rcu_data *rdp; | |
28ced795 | 672 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
96d3fd0d | 673 | |
f7f7bac9 | 674 | trace_rcu_dyntick(TPS("Start"), oldval, rdtp->dynticks_nesting); |
1ce46ee5 PM |
675 | if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
676 | !user && !is_idle_task(current)) { | |
289828e6 PM |
677 | struct task_struct *idle __maybe_unused = |
678 | idle_task(smp_processor_id()); | |
0989cb46 | 679 | |
f7f7bac9 | 680 | trace_rcu_dyntick(TPS("Error on entry: not idle task"), oldval, 0); |
274529ba | 681 | rcu_ftrace_dump(DUMP_ORIG); |
0989cb46 PM |
682 | WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s", |
683 | current->pid, current->comm, | |
684 | idle->pid, idle->comm); /* must be idle task! */ | |
9b2e4f18 | 685 | } |
96d3fd0d PM |
686 | for_each_rcu_flavor(rsp) { |
687 | rdp = this_cpu_ptr(rsp->rda); | |
688 | do_nocb_deferred_wakeup(rdp); | |
689 | } | |
198bbf81 | 690 | rcu_prepare_for_idle(); |
9b2e4f18 | 691 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ |
4e857c58 | 692 | smp_mb__before_atomic(); /* See above. */ |
9b2e4f18 | 693 | atomic_inc(&rdtp->dynticks); |
4e857c58 | 694 | smp_mb__after_atomic(); /* Force ordering with next sojourn. */ |
1ce46ee5 PM |
695 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
696 | atomic_read(&rdtp->dynticks) & 0x1); | |
176f8f7a | 697 | rcu_dynticks_task_enter(); |
c44e2cdd PM |
698 | |
699 | /* | |
adf5091e | 700 | * It is illegal to enter an extended quiescent state while |
c44e2cdd PM |
701 | * in an RCU read-side critical section. |
702 | */ | |
f78f5b90 PM |
703 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map), |
704 | "Illegal idle entry in RCU read-side critical section."); | |
705 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), | |
706 | "Illegal idle entry in RCU-bh read-side critical section."); | |
707 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), | |
708 | "Illegal idle entry in RCU-sched read-side critical section."); | |
9b2e4f18 | 709 | } |
64db4cff | 710 | |
adf5091e FW |
711 | /* |
712 | * Enter an RCU extended quiescent state, which can be either the | |
713 | * idle loop or adaptive-tickless usermode execution. | |
64db4cff | 714 | */ |
adf5091e | 715 | static void rcu_eqs_enter(bool user) |
64db4cff | 716 | { |
4145fa7f | 717 | long long oldval; |
64db4cff PM |
718 | struct rcu_dynticks *rdtp; |
719 | ||
c9d4b0af | 720 | rdtp = this_cpu_ptr(&rcu_dynticks); |
4145fa7f | 721 | oldval = rdtp->dynticks_nesting; |
1ce46ee5 PM |
722 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
723 | (oldval & DYNTICK_TASK_NEST_MASK) == 0); | |
3a592405 | 724 | if ((oldval & DYNTICK_TASK_NEST_MASK) == DYNTICK_TASK_NEST_VALUE) { |
29e37d81 | 725 | rdtp->dynticks_nesting = 0; |
28ced795 | 726 | rcu_eqs_enter_common(oldval, user); |
3a592405 | 727 | } else { |
29e37d81 | 728 | rdtp->dynticks_nesting -= DYNTICK_TASK_NEST_VALUE; |
3a592405 | 729 | } |
64db4cff | 730 | } |
adf5091e FW |
731 | |
732 | /** | |
733 | * rcu_idle_enter - inform RCU that current CPU is entering idle | |
734 | * | |
735 | * Enter idle mode, in other words, -leave- the mode in which RCU | |
736 | * read-side critical sections can occur. (Though RCU read-side | |
737 | * critical sections can occur in irq handlers in idle, a possibility | |
738 | * handled by irq_enter() and irq_exit().) | |
739 | * | |
740 | * We crowbar the ->dynticks_nesting field to zero to allow for | |
741 | * the possibility of usermode upcalls having messed up our count | |
742 | * of interrupt nesting level during the prior busy period. | |
743 | */ | |
744 | void rcu_idle_enter(void) | |
745 | { | |
c5d900bf FW |
746 | unsigned long flags; |
747 | ||
748 | local_irq_save(flags); | |
cb349ca9 | 749 | rcu_eqs_enter(false); |
28ced795 | 750 | rcu_sysidle_enter(0); |
c5d900bf | 751 | local_irq_restore(flags); |
adf5091e | 752 | } |
8a2ecf47 | 753 | EXPORT_SYMBOL_GPL(rcu_idle_enter); |
64db4cff | 754 | |
d1ec4c34 | 755 | #ifdef CONFIG_NO_HZ_FULL |
adf5091e FW |
756 | /** |
757 | * rcu_user_enter - inform RCU that we are resuming userspace. | |
758 | * | |
759 | * Enter RCU idle mode right before resuming userspace. No use of RCU | |
760 | * is permitted between this call and rcu_user_exit(). This way the | |
761 | * CPU doesn't need to maintain the tick for RCU maintenance purposes | |
762 | * when the CPU runs in userspace. | |
763 | */ | |
764 | void rcu_user_enter(void) | |
765 | { | |
91d1aa43 | 766 | rcu_eqs_enter(1); |
adf5091e | 767 | } |
d1ec4c34 | 768 | #endif /* CONFIG_NO_HZ_FULL */ |
19dd1591 | 769 | |
9b2e4f18 PM |
770 | /** |
771 | * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle | |
772 | * | |
773 | * Exit from an interrupt handler, which might possibly result in entering | |
774 | * idle mode, in other words, leaving the mode in which read-side critical | |
7c9906ca | 775 | * sections can occur. The caller must have disabled interrupts. |
64db4cff | 776 | * |
9b2e4f18 PM |
777 | * This code assumes that the idle loop never does anything that might |
778 | * result in unbalanced calls to irq_enter() and irq_exit(). If your | |
779 | * architecture violates this assumption, RCU will give you what you | |
780 | * deserve, good and hard. But very infrequently and irreproducibly. | |
781 | * | |
782 | * Use things like work queues to work around this limitation. | |
783 | * | |
784 | * You have been warned. | |
64db4cff | 785 | */ |
9b2e4f18 | 786 | void rcu_irq_exit(void) |
64db4cff | 787 | { |
4145fa7f | 788 | long long oldval; |
64db4cff PM |
789 | struct rcu_dynticks *rdtp; |
790 | ||
7c9906ca | 791 | RCU_LOCKDEP_WARN(!irqs_disabled(), "rcu_irq_exit() invoked with irqs enabled!!!"); |
c9d4b0af | 792 | rdtp = this_cpu_ptr(&rcu_dynticks); |
4145fa7f | 793 | oldval = rdtp->dynticks_nesting; |
9b2e4f18 | 794 | rdtp->dynticks_nesting--; |
1ce46ee5 PM |
795 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
796 | rdtp->dynticks_nesting < 0); | |
b6fc6020 | 797 | if (rdtp->dynticks_nesting) |
f7f7bac9 | 798 | trace_rcu_dyntick(TPS("--="), oldval, rdtp->dynticks_nesting); |
b6fc6020 | 799 | else |
28ced795 CL |
800 | rcu_eqs_enter_common(oldval, true); |
801 | rcu_sysidle_enter(1); | |
7c9906ca PM |
802 | } |
803 | ||
804 | /* | |
805 | * Wrapper for rcu_irq_exit() where interrupts are enabled. | |
806 | */ | |
807 | void rcu_irq_exit_irqson(void) | |
808 | { | |
809 | unsigned long flags; | |
810 | ||
811 | local_irq_save(flags); | |
812 | rcu_irq_exit(); | |
9b2e4f18 PM |
813 | local_irq_restore(flags); |
814 | } | |
815 | ||
816 | /* | |
adf5091e | 817 | * rcu_eqs_exit_common - current CPU moving away from extended quiescent state |
9b2e4f18 PM |
818 | * |
819 | * If the new value of the ->dynticks_nesting counter was previously zero, | |
820 | * we really have exited idle, and must do the appropriate accounting. | |
821 | * The caller must have disabled interrupts. | |
822 | */ | |
28ced795 | 823 | static void rcu_eqs_exit_common(long long oldval, int user) |
9b2e4f18 | 824 | { |
28ced795 CL |
825 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
826 | ||
176f8f7a | 827 | rcu_dynticks_task_exit(); |
4e857c58 | 828 | smp_mb__before_atomic(); /* Force ordering w/previous sojourn. */ |
23b5c8fa PM |
829 | atomic_inc(&rdtp->dynticks); |
830 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ | |
4e857c58 | 831 | smp_mb__after_atomic(); /* See above. */ |
1ce46ee5 PM |
832 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
833 | !(atomic_read(&rdtp->dynticks) & 0x1)); | |
8fa7845d | 834 | rcu_cleanup_after_idle(); |
f7f7bac9 | 835 | trace_rcu_dyntick(TPS("End"), oldval, rdtp->dynticks_nesting); |
1ce46ee5 PM |
836 | if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
837 | !user && !is_idle_task(current)) { | |
289828e6 PM |
838 | struct task_struct *idle __maybe_unused = |
839 | idle_task(smp_processor_id()); | |
0989cb46 | 840 | |
f7f7bac9 | 841 | trace_rcu_dyntick(TPS("Error on exit: not idle task"), |
4145fa7f | 842 | oldval, rdtp->dynticks_nesting); |
274529ba | 843 | rcu_ftrace_dump(DUMP_ORIG); |
0989cb46 PM |
844 | WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s", |
845 | current->pid, current->comm, | |
846 | idle->pid, idle->comm); /* must be idle task! */ | |
9b2e4f18 PM |
847 | } |
848 | } | |
849 | ||
adf5091e FW |
850 | /* |
851 | * Exit an RCU extended quiescent state, which can be either the | |
852 | * idle loop or adaptive-tickless usermode execution. | |
9b2e4f18 | 853 | */ |
adf5091e | 854 | static void rcu_eqs_exit(bool user) |
9b2e4f18 | 855 | { |
9b2e4f18 PM |
856 | struct rcu_dynticks *rdtp; |
857 | long long oldval; | |
858 | ||
c9d4b0af | 859 | rdtp = this_cpu_ptr(&rcu_dynticks); |
9b2e4f18 | 860 | oldval = rdtp->dynticks_nesting; |
1ce46ee5 | 861 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && oldval < 0); |
3a592405 | 862 | if (oldval & DYNTICK_TASK_NEST_MASK) { |
29e37d81 | 863 | rdtp->dynticks_nesting += DYNTICK_TASK_NEST_VALUE; |
3a592405 | 864 | } else { |
29e37d81 | 865 | rdtp->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE; |
28ced795 | 866 | rcu_eqs_exit_common(oldval, user); |
3a592405 | 867 | } |
9b2e4f18 | 868 | } |
adf5091e FW |
869 | |
870 | /** | |
871 | * rcu_idle_exit - inform RCU that current CPU is leaving idle | |
872 | * | |
873 | * Exit idle mode, in other words, -enter- the mode in which RCU | |
874 | * read-side critical sections can occur. | |
875 | * | |
876 | * We crowbar the ->dynticks_nesting field to DYNTICK_TASK_NEST to | |
877 | * allow for the possibility of usermode upcalls messing up our count | |
878 | * of interrupt nesting level during the busy period that is just | |
879 | * now starting. | |
880 | */ | |
881 | void rcu_idle_exit(void) | |
882 | { | |
c5d900bf FW |
883 | unsigned long flags; |
884 | ||
885 | local_irq_save(flags); | |
cb349ca9 | 886 | rcu_eqs_exit(false); |
28ced795 | 887 | rcu_sysidle_exit(0); |
c5d900bf | 888 | local_irq_restore(flags); |
adf5091e | 889 | } |
8a2ecf47 | 890 | EXPORT_SYMBOL_GPL(rcu_idle_exit); |
9b2e4f18 | 891 | |
d1ec4c34 | 892 | #ifdef CONFIG_NO_HZ_FULL |
adf5091e FW |
893 | /** |
894 | * rcu_user_exit - inform RCU that we are exiting userspace. | |
895 | * | |
896 | * Exit RCU idle mode while entering the kernel because it can | |
897 | * run a RCU read side critical section anytime. | |
898 | */ | |
899 | void rcu_user_exit(void) | |
900 | { | |
91d1aa43 | 901 | rcu_eqs_exit(1); |
adf5091e | 902 | } |
d1ec4c34 | 903 | #endif /* CONFIG_NO_HZ_FULL */ |
19dd1591 | 904 | |
9b2e4f18 PM |
905 | /** |
906 | * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle | |
907 | * | |
908 | * Enter an interrupt handler, which might possibly result in exiting | |
909 | * idle mode, in other words, entering the mode in which read-side critical | |
7c9906ca | 910 | * sections can occur. The caller must have disabled interrupts. |
9b2e4f18 PM |
911 | * |
912 | * Note that the Linux kernel is fully capable of entering an interrupt | |
913 | * handler that it never exits, for example when doing upcalls to | |
914 | * user mode! This code assumes that the idle loop never does upcalls to | |
915 | * user mode. If your architecture does do upcalls from the idle loop (or | |
916 | * does anything else that results in unbalanced calls to the irq_enter() | |
917 | * and irq_exit() functions), RCU will give you what you deserve, good | |
918 | * and hard. But very infrequently and irreproducibly. | |
919 | * | |
920 | * Use things like work queues to work around this limitation. | |
921 | * | |
922 | * You have been warned. | |
923 | */ | |
924 | void rcu_irq_enter(void) | |
925 | { | |
9b2e4f18 PM |
926 | struct rcu_dynticks *rdtp; |
927 | long long oldval; | |
928 | ||
7c9906ca | 929 | RCU_LOCKDEP_WARN(!irqs_disabled(), "rcu_irq_enter() invoked with irqs enabled!!!"); |
c9d4b0af | 930 | rdtp = this_cpu_ptr(&rcu_dynticks); |
9b2e4f18 PM |
931 | oldval = rdtp->dynticks_nesting; |
932 | rdtp->dynticks_nesting++; | |
1ce46ee5 PM |
933 | WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && |
934 | rdtp->dynticks_nesting == 0); | |
b6fc6020 | 935 | if (oldval) |
f7f7bac9 | 936 | trace_rcu_dyntick(TPS("++="), oldval, rdtp->dynticks_nesting); |
b6fc6020 | 937 | else |
28ced795 CL |
938 | rcu_eqs_exit_common(oldval, true); |
939 | rcu_sysidle_exit(1); | |
7c9906ca PM |
940 | } |
941 | ||
942 | /* | |
943 | * Wrapper for rcu_irq_enter() where interrupts are enabled. | |
944 | */ | |
945 | void rcu_irq_enter_irqson(void) | |
946 | { | |
947 | unsigned long flags; | |
948 | ||
949 | local_irq_save(flags); | |
950 | rcu_irq_enter(); | |
64db4cff | 951 | local_irq_restore(flags); |
64db4cff PM |
952 | } |
953 | ||
954 | /** | |
955 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
956 | * | |
734d1680 PM |
957 | * If the CPU was idle from RCU's viewpoint, update rdtp->dynticks and |
958 | * rdtp->dynticks_nmi_nesting to let the RCU grace-period handling know | |
959 | * that the CPU is active. This implementation permits nested NMIs, as | |
960 | * long as the nesting level does not overflow an int. (You will probably | |
961 | * run out of stack space first.) | |
64db4cff PM |
962 | */ |
963 | void rcu_nmi_enter(void) | |
964 | { | |
c9d4b0af | 965 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
734d1680 | 966 | int incby = 2; |
64db4cff | 967 | |
734d1680 PM |
968 | /* Complain about underflow. */ |
969 | WARN_ON_ONCE(rdtp->dynticks_nmi_nesting < 0); | |
970 | ||
971 | /* | |
972 | * If idle from RCU viewpoint, atomically increment ->dynticks | |
973 | * to mark non-idle and increment ->dynticks_nmi_nesting by one. | |
974 | * Otherwise, increment ->dynticks_nmi_nesting by two. This means | |
975 | * if ->dynticks_nmi_nesting is equal to one, we are guaranteed | |
976 | * to be in the outermost NMI handler that interrupted an RCU-idle | |
977 | * period (observation due to Andy Lutomirski). | |
978 | */ | |
979 | if (!(atomic_read(&rdtp->dynticks) & 0x1)) { | |
980 | smp_mb__before_atomic(); /* Force delay from prior write. */ | |
981 | atomic_inc(&rdtp->dynticks); | |
982 | /* atomic_inc() before later RCU read-side crit sects */ | |
983 | smp_mb__after_atomic(); /* See above. */ | |
984 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); | |
985 | incby = 1; | |
986 | } | |
987 | rdtp->dynticks_nmi_nesting += incby; | |
988 | barrier(); | |
64db4cff PM |
989 | } |
990 | ||
991 | /** | |
992 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
993 | * | |
734d1680 PM |
994 | * If we are returning from the outermost NMI handler that interrupted an |
995 | * RCU-idle period, update rdtp->dynticks and rdtp->dynticks_nmi_nesting | |
996 | * to let the RCU grace-period handling know that the CPU is back to | |
997 | * being RCU-idle. | |
64db4cff PM |
998 | */ |
999 | void rcu_nmi_exit(void) | |
1000 | { | |
c9d4b0af | 1001 | struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks); |
64db4cff | 1002 | |
734d1680 PM |
1003 | /* |
1004 | * Check for ->dynticks_nmi_nesting underflow and bad ->dynticks. | |
1005 | * (We are exiting an NMI handler, so RCU better be paying attention | |
1006 | * to us!) | |
1007 | */ | |
1008 | WARN_ON_ONCE(rdtp->dynticks_nmi_nesting <= 0); | |
1009 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); | |
1010 | ||
1011 | /* | |
1012 | * If the nesting level is not 1, the CPU wasn't RCU-idle, so | |
1013 | * leave it in non-RCU-idle state. | |
1014 | */ | |
1015 | if (rdtp->dynticks_nmi_nesting != 1) { | |
1016 | rdtp->dynticks_nmi_nesting -= 2; | |
64db4cff | 1017 | return; |
734d1680 PM |
1018 | } |
1019 | ||
1020 | /* This NMI interrupted an RCU-idle CPU, restore RCU-idleness. */ | |
1021 | rdtp->dynticks_nmi_nesting = 0; | |
23b5c8fa | 1022 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ |
4e857c58 | 1023 | smp_mb__before_atomic(); /* See above. */ |
23b5c8fa | 1024 | atomic_inc(&rdtp->dynticks); |
4e857c58 | 1025 | smp_mb__after_atomic(); /* Force delay to next write. */ |
23b5c8fa | 1026 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); |
64db4cff PM |
1027 | } |
1028 | ||
1029 | /** | |
5c173eb8 PM |
1030 | * __rcu_is_watching - are RCU read-side critical sections safe? |
1031 | * | |
1032 | * Return true if RCU is watching the running CPU, which means that | |
1033 | * this CPU can safely enter RCU read-side critical sections. Unlike | |
1034 | * rcu_is_watching(), the caller of __rcu_is_watching() must have at | |
1035 | * least disabled preemption. | |
1036 | */ | |
9418fb20 | 1037 | bool notrace __rcu_is_watching(void) |
5c173eb8 PM |
1038 | { |
1039 | return atomic_read(this_cpu_ptr(&rcu_dynticks.dynticks)) & 0x1; | |
1040 | } | |
1041 | ||
1042 | /** | |
1043 | * rcu_is_watching - see if RCU thinks that the current CPU is idle | |
64db4cff | 1044 | * |
9b2e4f18 | 1045 | * If the current CPU is in its idle loop and is neither in an interrupt |
34240697 | 1046 | * or NMI handler, return true. |
64db4cff | 1047 | */ |
9418fb20 | 1048 | bool notrace rcu_is_watching(void) |
64db4cff | 1049 | { |
f534ed1f | 1050 | bool ret; |
34240697 | 1051 | |
46f00d18 | 1052 | preempt_disable_notrace(); |
5c173eb8 | 1053 | ret = __rcu_is_watching(); |
46f00d18 | 1054 | preempt_enable_notrace(); |
34240697 | 1055 | return ret; |
64db4cff | 1056 | } |
5c173eb8 | 1057 | EXPORT_SYMBOL_GPL(rcu_is_watching); |
64db4cff | 1058 | |
62fde6ed | 1059 | #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) |
c0d6d01b PM |
1060 | |
1061 | /* | |
1062 | * Is the current CPU online? Disable preemption to avoid false positives | |
1063 | * that could otherwise happen due to the current CPU number being sampled, | |
1064 | * this task being preempted, its old CPU being taken offline, resuming | |
1065 | * on some other CPU, then determining that its old CPU is now offline. | |
1066 | * It is OK to use RCU on an offline processor during initial boot, hence | |
2036d94a PM |
1067 | * the check for rcu_scheduler_fully_active. Note also that it is OK |
1068 | * for a CPU coming online to use RCU for one jiffy prior to marking itself | |
1069 | * online in the cpu_online_mask. Similarly, it is OK for a CPU going | |
1070 | * offline to continue to use RCU for one jiffy after marking itself | |
1071 | * offline in the cpu_online_mask. This leniency is necessary given the | |
1072 | * non-atomic nature of the online and offline processing, for example, | |
1073 | * the fact that a CPU enters the scheduler after completing the CPU_DYING | |
1074 | * notifiers. | |
1075 | * | |
1076 | * This is also why RCU internally marks CPUs online during the | |
1077 | * CPU_UP_PREPARE phase and offline during the CPU_DEAD phase. | |
c0d6d01b PM |
1078 | * |
1079 | * Disable checking if in an NMI handler because we cannot safely report | |
1080 | * errors from NMI handlers anyway. | |
1081 | */ | |
1082 | bool rcu_lockdep_current_cpu_online(void) | |
1083 | { | |
2036d94a PM |
1084 | struct rcu_data *rdp; |
1085 | struct rcu_node *rnp; | |
c0d6d01b PM |
1086 | bool ret; |
1087 | ||
1088 | if (in_nmi()) | |
f6f7ee9a | 1089 | return true; |
c0d6d01b | 1090 | preempt_disable(); |
c9d4b0af | 1091 | rdp = this_cpu_ptr(&rcu_sched_data); |
2036d94a | 1092 | rnp = rdp->mynode; |
0aa04b05 | 1093 | ret = (rdp->grpmask & rcu_rnp_online_cpus(rnp)) || |
c0d6d01b PM |
1094 | !rcu_scheduler_fully_active; |
1095 | preempt_enable(); | |
1096 | return ret; | |
1097 | } | |
1098 | EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online); | |
1099 | ||
62fde6ed | 1100 | #endif /* #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) */ |
9b2e4f18 | 1101 | |
64db4cff | 1102 | /** |
9b2e4f18 | 1103 | * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle |
64db4cff | 1104 | * |
9b2e4f18 PM |
1105 | * If the current CPU is idle or running at a first-level (not nested) |
1106 | * interrupt from idle, return true. The caller must have at least | |
1107 | * disabled preemption. | |
64db4cff | 1108 | */ |
62e3cb14 | 1109 | static int rcu_is_cpu_rrupt_from_idle(void) |
64db4cff | 1110 | { |
c9d4b0af | 1111 | return __this_cpu_read(rcu_dynticks.dynticks_nesting) <= 1; |
64db4cff PM |
1112 | } |
1113 | ||
64db4cff PM |
1114 | /* |
1115 | * Snapshot the specified CPU's dynticks counter so that we can later | |
1116 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 1117 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff | 1118 | */ |
217af2a2 PM |
1119 | static int dyntick_save_progress_counter(struct rcu_data *rdp, |
1120 | bool *isidle, unsigned long *maxj) | |
64db4cff | 1121 | { |
23b5c8fa | 1122 | rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks); |
0edd1b17 | 1123 | rcu_sysidle_check_cpu(rdp, isidle, maxj); |
7941dbde ACB |
1124 | if ((rdp->dynticks_snap & 0x1) == 0) { |
1125 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("dti")); | |
7d0ae808 | 1126 | if (ULONG_CMP_LT(READ_ONCE(rdp->gpnum) + ULONG_MAX / 4, |
e3663b10 | 1127 | rdp->mynode->gpnum)) |
7d0ae808 | 1128 | WRITE_ONCE(rdp->gpwrap, true); |
23a9bacd | 1129 | return 1; |
7941dbde | 1130 | } |
23a9bacd | 1131 | return 0; |
64db4cff PM |
1132 | } |
1133 | ||
1134 | /* | |
1135 | * Return true if the specified CPU has passed through a quiescent | |
1136 | * state by virtue of being in or having passed through an dynticks | |
1137 | * idle state since the last call to dyntick_save_progress_counter() | |
a82dcc76 | 1138 | * for this same CPU, or by virtue of having been offline. |
64db4cff | 1139 | */ |
217af2a2 PM |
1140 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp, |
1141 | bool *isidle, unsigned long *maxj) | |
64db4cff | 1142 | { |
7eb4f455 | 1143 | unsigned int curr; |
4a81e832 | 1144 | int *rcrmp; |
7eb4f455 | 1145 | unsigned int snap; |
64db4cff | 1146 | |
7eb4f455 PM |
1147 | curr = (unsigned int)atomic_add_return(0, &rdp->dynticks->dynticks); |
1148 | snap = (unsigned int)rdp->dynticks_snap; | |
64db4cff PM |
1149 | |
1150 | /* | |
1151 | * If the CPU passed through or entered a dynticks idle phase with | |
1152 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
1153 | * already acknowledged the request to pass through a quiescent | |
1154 | * state. Either way, that CPU cannot possibly be in an RCU | |
1155 | * read-side critical section that started before the beginning | |
1156 | * of the current RCU grace period. | |
1157 | */ | |
7eb4f455 | 1158 | if ((curr & 0x1) == 0 || UINT_CMP_GE(curr, snap + 2)) { |
f7f7bac9 | 1159 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("dti")); |
64db4cff PM |
1160 | rdp->dynticks_fqs++; |
1161 | return 1; | |
1162 | } | |
1163 | ||
a82dcc76 PM |
1164 | /* |
1165 | * Check for the CPU being offline, but only if the grace period | |
1166 | * is old enough. We don't need to worry about the CPU changing | |
1167 | * state: If we see it offline even once, it has been through a | |
1168 | * quiescent state. | |
1169 | * | |
1170 | * The reason for insisting that the grace period be at least | |
1171 | * one jiffy old is that CPUs that are not quite online and that | |
1172 | * have just gone offline can still execute RCU read-side critical | |
1173 | * sections. | |
1174 | */ | |
1175 | if (ULONG_CMP_GE(rdp->rsp->gp_start + 2, jiffies)) | |
1176 | return 0; /* Grace period is not old enough. */ | |
1177 | barrier(); | |
1178 | if (cpu_is_offline(rdp->cpu)) { | |
f7f7bac9 | 1179 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("ofl")); |
a82dcc76 PM |
1180 | rdp->offline_fqs++; |
1181 | return 1; | |
1182 | } | |
65d798f0 PM |
1183 | |
1184 | /* | |
4a81e832 PM |
1185 | * A CPU running for an extended time within the kernel can |
1186 | * delay RCU grace periods. When the CPU is in NO_HZ_FULL mode, | |
1187 | * even context-switching back and forth between a pair of | |
1188 | * in-kernel CPU-bound tasks cannot advance grace periods. | |
1189 | * So if the grace period is old enough, make the CPU pay attention. | |
1190 | * Note that the unsynchronized assignments to the per-CPU | |
1191 | * rcu_sched_qs_mask variable are safe. Yes, setting of | |
1192 | * bits can be lost, but they will be set again on the next | |
1193 | * force-quiescent-state pass. So lost bit sets do not result | |
1194 | * in incorrect behavior, merely in a grace period lasting | |
1195 | * a few jiffies longer than it might otherwise. Because | |
1196 | * there are at most four threads involved, and because the | |
1197 | * updates are only once every few jiffies, the probability of | |
1198 | * lossage (and thus of slight grace-period extension) is | |
1199 | * quite low. | |
1200 | * | |
1201 | * Note that if the jiffies_till_sched_qs boot/sysfs parameter | |
1202 | * is set too high, we override with half of the RCU CPU stall | |
1203 | * warning delay. | |
6193c76a | 1204 | */ |
4a81e832 PM |
1205 | rcrmp = &per_cpu(rcu_sched_qs_mask, rdp->cpu); |
1206 | if (ULONG_CMP_GE(jiffies, | |
1207 | rdp->rsp->gp_start + jiffies_till_sched_qs) || | |
cb1e78cf | 1208 | ULONG_CMP_GE(jiffies, rdp->rsp->jiffies_resched)) { |
7d0ae808 PM |
1209 | if (!(READ_ONCE(*rcrmp) & rdp->rsp->flavor_mask)) { |
1210 | WRITE_ONCE(rdp->cond_resched_completed, | |
1211 | READ_ONCE(rdp->mynode->completed)); | |
4a81e832 | 1212 | smp_mb(); /* ->cond_resched_completed before *rcrmp. */ |
7d0ae808 PM |
1213 | WRITE_ONCE(*rcrmp, |
1214 | READ_ONCE(*rcrmp) + rdp->rsp->flavor_mask); | |
4a81e832 | 1215 | } |
4914950a | 1216 | rdp->rsp->jiffies_resched += 5; /* Re-enable beating. */ |
6193c76a PM |
1217 | } |
1218 | ||
4914950a PM |
1219 | /* And if it has been a really long time, kick the CPU as well. */ |
1220 | if (ULONG_CMP_GE(jiffies, | |
1221 | rdp->rsp->gp_start + 2 * jiffies_till_sched_qs) || | |
1222 | ULONG_CMP_GE(jiffies, rdp->rsp->gp_start + jiffies_till_sched_qs)) | |
1223 | resched_cpu(rdp->cpu); /* Force CPU into scheduler. */ | |
1224 | ||
a82dcc76 | 1225 | return 0; |
64db4cff PM |
1226 | } |
1227 | ||
64db4cff PM |
1228 | static void record_gp_stall_check_time(struct rcu_state *rsp) |
1229 | { | |
cb1e78cf | 1230 | unsigned long j = jiffies; |
6193c76a | 1231 | unsigned long j1; |
26cdfedf PM |
1232 | |
1233 | rsp->gp_start = j; | |
1234 | smp_wmb(); /* Record start time before stall time. */ | |
6193c76a | 1235 | j1 = rcu_jiffies_till_stall_check(); |
7d0ae808 | 1236 | WRITE_ONCE(rsp->jiffies_stall, j + j1); |
6193c76a | 1237 | rsp->jiffies_resched = j + j1 / 2; |
7d0ae808 | 1238 | rsp->n_force_qs_gpstart = READ_ONCE(rsp->n_force_qs); |
64db4cff PM |
1239 | } |
1240 | ||
6b50e119 PM |
1241 | /* |
1242 | * Convert a ->gp_state value to a character string. | |
1243 | */ | |
1244 | static const char *gp_state_getname(short gs) | |
1245 | { | |
1246 | if (gs < 0 || gs >= ARRAY_SIZE(gp_state_names)) | |
1247 | return "???"; | |
1248 | return gp_state_names[gs]; | |
1249 | } | |
1250 | ||
fb81a44b PM |
1251 | /* |
1252 | * Complain about starvation of grace-period kthread. | |
1253 | */ | |
1254 | static void rcu_check_gp_kthread_starvation(struct rcu_state *rsp) | |
1255 | { | |
1256 | unsigned long gpa; | |
1257 | unsigned long j; | |
1258 | ||
1259 | j = jiffies; | |
7d0ae808 | 1260 | gpa = READ_ONCE(rsp->gp_activity); |
b1adb3e2 | 1261 | if (j - gpa > 2 * HZ) { |
6b50e119 | 1262 | pr_err("%s kthread starved for %ld jiffies! g%lu c%lu f%#x %s(%d) ->state=%#lx\n", |
81e701e4 | 1263 | rsp->name, j - gpa, |
319362c9 | 1264 | rsp->gpnum, rsp->completed, |
6b50e119 PM |
1265 | rsp->gp_flags, |
1266 | gp_state_getname(rsp->gp_state), rsp->gp_state, | |
a0e3a3aa | 1267 | rsp->gp_kthread ? rsp->gp_kthread->state : ~0); |
86057b80 | 1268 | if (rsp->gp_kthread) { |
b1adb3e2 | 1269 | sched_show_task(rsp->gp_kthread); |
86057b80 PM |
1270 | wake_up_process(rsp->gp_kthread); |
1271 | } | |
b1adb3e2 | 1272 | } |
64db4cff PM |
1273 | } |
1274 | ||
b637a328 | 1275 | /* |
bc1dce51 | 1276 | * Dump stacks of all tasks running on stalled CPUs. |
b637a328 PM |
1277 | */ |
1278 | static void rcu_dump_cpu_stacks(struct rcu_state *rsp) | |
1279 | { | |
1280 | int cpu; | |
1281 | unsigned long flags; | |
1282 | struct rcu_node *rnp; | |
1283 | ||
1284 | rcu_for_each_leaf_node(rsp, rnp) { | |
6cf10081 | 1285 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
b637a328 PM |
1286 | if (rnp->qsmask != 0) { |
1287 | for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++) | |
1288 | if (rnp->qsmask & (1UL << cpu)) | |
1289 | dump_cpu_task(rnp->grplo + cpu); | |
1290 | } | |
67c583a7 | 1291 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
b637a328 PM |
1292 | } |
1293 | } | |
1294 | ||
8c7c4829 PM |
1295 | /* |
1296 | * If too much time has passed in the current grace period, and if | |
1297 | * so configured, go kick the relevant kthreads. | |
1298 | */ | |
1299 | static void rcu_stall_kick_kthreads(struct rcu_state *rsp) | |
1300 | { | |
1301 | unsigned long j; | |
1302 | ||
1303 | if (!rcu_kick_kthreads) | |
1304 | return; | |
1305 | j = READ_ONCE(rsp->jiffies_kick_kthreads); | |
1306 | if (time_after(jiffies, j) && rsp->gp_kthread) { | |
1307 | WARN_ONCE(1, "Kicking %s grace-period kthread\n", rsp->name); | |
5dffed1e | 1308 | rcu_ftrace_dump(DUMP_ALL); |
8c7c4829 PM |
1309 | wake_up_process(rsp->gp_kthread); |
1310 | WRITE_ONCE(rsp->jiffies_kick_kthreads, j + HZ); | |
1311 | } | |
1312 | } | |
1313 | ||
6ccd2ecd | 1314 | static void print_other_cpu_stall(struct rcu_state *rsp, unsigned long gpnum) |
64db4cff PM |
1315 | { |
1316 | int cpu; | |
1317 | long delta; | |
1318 | unsigned long flags; | |
6ccd2ecd PM |
1319 | unsigned long gpa; |
1320 | unsigned long j; | |
285fe294 | 1321 | int ndetected = 0; |
64db4cff | 1322 | struct rcu_node *rnp = rcu_get_root(rsp); |
53bb857c | 1323 | long totqlen = 0; |
64db4cff | 1324 | |
8c7c4829 PM |
1325 | /* Kick and suppress, if so configured. */ |
1326 | rcu_stall_kick_kthreads(rsp); | |
1327 | if (rcu_cpu_stall_suppress) | |
1328 | return; | |
1329 | ||
64db4cff PM |
1330 | /* Only let one CPU complain about others per time interval. */ |
1331 | ||
6cf10081 | 1332 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
7d0ae808 | 1333 | delta = jiffies - READ_ONCE(rsp->jiffies_stall); |
fc2219d4 | 1334 | if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) { |
67c583a7 | 1335 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
1336 | return; |
1337 | } | |
7d0ae808 PM |
1338 | WRITE_ONCE(rsp->jiffies_stall, |
1339 | jiffies + 3 * rcu_jiffies_till_stall_check() + 3); | |
67c583a7 | 1340 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 1341 | |
8cdd32a9 PM |
1342 | /* |
1343 | * OK, time to rat on our buddy... | |
1344 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
1345 | * RCU CPU stall warnings. | |
1346 | */ | |
d7f3e207 | 1347 | pr_err("INFO: %s detected stalls on CPUs/tasks:", |
4300aa64 | 1348 | rsp->name); |
a858af28 | 1349 | print_cpu_stall_info_begin(); |
a0b6c9a7 | 1350 | rcu_for_each_leaf_node(rsp, rnp) { |
6cf10081 | 1351 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
9bc8b558 | 1352 | ndetected += rcu_print_task_stall(rnp); |
c8020a67 PM |
1353 | if (rnp->qsmask != 0) { |
1354 | for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++) | |
1355 | if (rnp->qsmask & (1UL << cpu)) { | |
1356 | print_cpu_stall_info(rsp, | |
1357 | rnp->grplo + cpu); | |
1358 | ndetected++; | |
1359 | } | |
1360 | } | |
67c583a7 | 1361 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 1362 | } |
a858af28 | 1363 | |
a858af28 | 1364 | print_cpu_stall_info_end(); |
53bb857c PM |
1365 | for_each_possible_cpu(cpu) |
1366 | totqlen += per_cpu_ptr(rsp->rda, cpu)->qlen; | |
83ebe63e | 1367 | pr_cont("(detected by %d, t=%ld jiffies, g=%ld, c=%ld, q=%lu)\n", |
eee05882 | 1368 | smp_processor_id(), (long)(jiffies - rsp->gp_start), |
83ebe63e | 1369 | (long)rsp->gpnum, (long)rsp->completed, totqlen); |
6ccd2ecd | 1370 | if (ndetected) { |
b637a328 | 1371 | rcu_dump_cpu_stacks(rsp); |
6ccd2ecd | 1372 | } else { |
7d0ae808 PM |
1373 | if (READ_ONCE(rsp->gpnum) != gpnum || |
1374 | READ_ONCE(rsp->completed) == gpnum) { | |
6ccd2ecd PM |
1375 | pr_err("INFO: Stall ended before state dump start\n"); |
1376 | } else { | |
1377 | j = jiffies; | |
7d0ae808 | 1378 | gpa = READ_ONCE(rsp->gp_activity); |
237a0f21 | 1379 | pr_err("All QSes seen, last %s kthread activity %ld (%ld-%ld), jiffies_till_next_fqs=%ld, root ->qsmask %#lx\n", |
6ccd2ecd | 1380 | rsp->name, j - gpa, j, gpa, |
237a0f21 PM |
1381 | jiffies_till_next_fqs, |
1382 | rcu_get_root(rsp)->qsmask); | |
6ccd2ecd PM |
1383 | /* In this case, the current CPU might be at fault. */ |
1384 | sched_show_task(current); | |
1385 | } | |
1386 | } | |
c1dc0b9c | 1387 | |
4cdfc175 | 1388 | /* Complain about tasks blocking the grace period. */ |
1ed509a2 PM |
1389 | rcu_print_detail_task_stall(rsp); |
1390 | ||
fb81a44b PM |
1391 | rcu_check_gp_kthread_starvation(rsp); |
1392 | ||
4cdfc175 | 1393 | force_quiescent_state(rsp); /* Kick them all. */ |
64db4cff PM |
1394 | } |
1395 | ||
1396 | static void print_cpu_stall(struct rcu_state *rsp) | |
1397 | { | |
53bb857c | 1398 | int cpu; |
64db4cff PM |
1399 | unsigned long flags; |
1400 | struct rcu_node *rnp = rcu_get_root(rsp); | |
53bb857c | 1401 | long totqlen = 0; |
64db4cff | 1402 | |
8c7c4829 PM |
1403 | /* Kick and suppress, if so configured. */ |
1404 | rcu_stall_kick_kthreads(rsp); | |
1405 | if (rcu_cpu_stall_suppress) | |
1406 | return; | |
1407 | ||
8cdd32a9 PM |
1408 | /* |
1409 | * OK, time to rat on ourselves... | |
1410 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
1411 | * RCU CPU stall warnings. | |
1412 | */ | |
d7f3e207 | 1413 | pr_err("INFO: %s self-detected stall on CPU", rsp->name); |
a858af28 PM |
1414 | print_cpu_stall_info_begin(); |
1415 | print_cpu_stall_info(rsp, smp_processor_id()); | |
1416 | print_cpu_stall_info_end(); | |
53bb857c PM |
1417 | for_each_possible_cpu(cpu) |
1418 | totqlen += per_cpu_ptr(rsp->rda, cpu)->qlen; | |
83ebe63e PM |
1419 | pr_cont(" (t=%lu jiffies g=%ld c=%ld q=%lu)\n", |
1420 | jiffies - rsp->gp_start, | |
1421 | (long)rsp->gpnum, (long)rsp->completed, totqlen); | |
fb81a44b PM |
1422 | |
1423 | rcu_check_gp_kthread_starvation(rsp); | |
1424 | ||
bc1dce51 | 1425 | rcu_dump_cpu_stacks(rsp); |
c1dc0b9c | 1426 | |
6cf10081 | 1427 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
7d0ae808 PM |
1428 | if (ULONG_CMP_GE(jiffies, READ_ONCE(rsp->jiffies_stall))) |
1429 | WRITE_ONCE(rsp->jiffies_stall, | |
1430 | jiffies + 3 * rcu_jiffies_till_stall_check() + 3); | |
67c583a7 | 1431 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
c1dc0b9c | 1432 | |
b021fe3e PZ |
1433 | /* |
1434 | * Attempt to revive the RCU machinery by forcing a context switch. | |
1435 | * | |
1436 | * A context switch would normally allow the RCU state machine to make | |
1437 | * progress and it could be we're stuck in kernel space without context | |
1438 | * switches for an entirely unreasonable amount of time. | |
1439 | */ | |
1440 | resched_cpu(smp_processor_id()); | |
64db4cff PM |
1441 | } |
1442 | ||
1443 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
1444 | { | |
26cdfedf PM |
1445 | unsigned long completed; |
1446 | unsigned long gpnum; | |
1447 | unsigned long gps; | |
bad6e139 PM |
1448 | unsigned long j; |
1449 | unsigned long js; | |
64db4cff PM |
1450 | struct rcu_node *rnp; |
1451 | ||
8c7c4829 PM |
1452 | if ((rcu_cpu_stall_suppress && !rcu_kick_kthreads) || |
1453 | !rcu_gp_in_progress(rsp)) | |
c68de209 | 1454 | return; |
8c7c4829 | 1455 | rcu_stall_kick_kthreads(rsp); |
cb1e78cf | 1456 | j = jiffies; |
26cdfedf PM |
1457 | |
1458 | /* | |
1459 | * Lots of memory barriers to reject false positives. | |
1460 | * | |
1461 | * The idea is to pick up rsp->gpnum, then rsp->jiffies_stall, | |
1462 | * then rsp->gp_start, and finally rsp->completed. These values | |
1463 | * are updated in the opposite order with memory barriers (or | |
1464 | * equivalent) during grace-period initialization and cleanup. | |
1465 | * Now, a false positive can occur if we get an new value of | |
1466 | * rsp->gp_start and a old value of rsp->jiffies_stall. But given | |
1467 | * the memory barriers, the only way that this can happen is if one | |
1468 | * grace period ends and another starts between these two fetches. | |
1469 | * Detect this by comparing rsp->completed with the previous fetch | |
1470 | * from rsp->gpnum. | |
1471 | * | |
1472 | * Given this check, comparisons of jiffies, rsp->jiffies_stall, | |
1473 | * and rsp->gp_start suffice to forestall false positives. | |
1474 | */ | |
7d0ae808 | 1475 | gpnum = READ_ONCE(rsp->gpnum); |
26cdfedf | 1476 | smp_rmb(); /* Pick up ->gpnum first... */ |
7d0ae808 | 1477 | js = READ_ONCE(rsp->jiffies_stall); |
26cdfedf | 1478 | smp_rmb(); /* ...then ->jiffies_stall before the rest... */ |
7d0ae808 | 1479 | gps = READ_ONCE(rsp->gp_start); |
26cdfedf | 1480 | smp_rmb(); /* ...and finally ->gp_start before ->completed. */ |
7d0ae808 | 1481 | completed = READ_ONCE(rsp->completed); |
26cdfedf PM |
1482 | if (ULONG_CMP_GE(completed, gpnum) || |
1483 | ULONG_CMP_LT(j, js) || | |
1484 | ULONG_CMP_GE(gps, js)) | |
1485 | return; /* No stall or GP completed since entering function. */ | |
64db4cff | 1486 | rnp = rdp->mynode; |
c96ea7cf | 1487 | if (rcu_gp_in_progress(rsp) && |
7d0ae808 | 1488 | (READ_ONCE(rnp->qsmask) & rdp->grpmask)) { |
64db4cff PM |
1489 | |
1490 | /* We haven't checked in, so go dump stack. */ | |
1491 | print_cpu_stall(rsp); | |
1492 | ||
bad6e139 PM |
1493 | } else if (rcu_gp_in_progress(rsp) && |
1494 | ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) { | |
64db4cff | 1495 | |
bad6e139 | 1496 | /* They had a few time units to dump stack, so complain. */ |
6ccd2ecd | 1497 | print_other_cpu_stall(rsp, gpnum); |
64db4cff PM |
1498 | } |
1499 | } | |
1500 | ||
53d84e00 PM |
1501 | /** |
1502 | * rcu_cpu_stall_reset - prevent further stall warnings in current grace period | |
1503 | * | |
1504 | * Set the stall-warning timeout way off into the future, thus preventing | |
1505 | * any RCU CPU stall-warning messages from appearing in the current set of | |
1506 | * RCU grace periods. | |
1507 | * | |
1508 | * The caller must disable hard irqs. | |
1509 | */ | |
1510 | void rcu_cpu_stall_reset(void) | |
1511 | { | |
6ce75a23 PM |
1512 | struct rcu_state *rsp; |
1513 | ||
1514 | for_each_rcu_flavor(rsp) | |
7d0ae808 | 1515 | WRITE_ONCE(rsp->jiffies_stall, jiffies + ULONG_MAX / 2); |
53d84e00 PM |
1516 | } |
1517 | ||
3f5d3ea6 | 1518 | /* |
d3f3f3f2 PM |
1519 | * Initialize the specified rcu_data structure's default callback list |
1520 | * to empty. The default callback list is the one that is not used by | |
1521 | * no-callbacks CPUs. | |
3f5d3ea6 | 1522 | */ |
d3f3f3f2 | 1523 | static void init_default_callback_list(struct rcu_data *rdp) |
3f5d3ea6 PM |
1524 | { |
1525 | int i; | |
1526 | ||
1527 | rdp->nxtlist = NULL; | |
1528 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
1529 | rdp->nxttail[i] = &rdp->nxtlist; | |
1530 | } | |
1531 | ||
d3f3f3f2 PM |
1532 | /* |
1533 | * Initialize the specified rcu_data structure's callback list to empty. | |
1534 | */ | |
1535 | static void init_callback_list(struct rcu_data *rdp) | |
1536 | { | |
1537 | if (init_nocb_callback_list(rdp)) | |
1538 | return; | |
1539 | init_default_callback_list(rdp); | |
1540 | } | |
1541 | ||
dc35c893 PM |
1542 | /* |
1543 | * Determine the value that ->completed will have at the end of the | |
1544 | * next subsequent grace period. This is used to tag callbacks so that | |
1545 | * a CPU can invoke callbacks in a timely fashion even if that CPU has | |
1546 | * been dyntick-idle for an extended period with callbacks under the | |
1547 | * influence of RCU_FAST_NO_HZ. | |
1548 | * | |
1549 | * The caller must hold rnp->lock with interrupts disabled. | |
1550 | */ | |
1551 | static unsigned long rcu_cbs_completed(struct rcu_state *rsp, | |
1552 | struct rcu_node *rnp) | |
1553 | { | |
1554 | /* | |
1555 | * If RCU is idle, we just wait for the next grace period. | |
1556 | * But we can only be sure that RCU is idle if we are looking | |
1557 | * at the root rcu_node structure -- otherwise, a new grace | |
1558 | * period might have started, but just not yet gotten around | |
1559 | * to initializing the current non-root rcu_node structure. | |
1560 | */ | |
1561 | if (rcu_get_root(rsp) == rnp && rnp->gpnum == rnp->completed) | |
1562 | return rnp->completed + 1; | |
1563 | ||
1564 | /* | |
1565 | * Otherwise, wait for a possible partial grace period and | |
1566 | * then the subsequent full grace period. | |
1567 | */ | |
1568 | return rnp->completed + 2; | |
1569 | } | |
1570 | ||
0446be48 PM |
1571 | /* |
1572 | * Trace-event helper function for rcu_start_future_gp() and | |
1573 | * rcu_nocb_wait_gp(). | |
1574 | */ | |
1575 | static void trace_rcu_future_gp(struct rcu_node *rnp, struct rcu_data *rdp, | |
e66c33d5 | 1576 | unsigned long c, const char *s) |
0446be48 PM |
1577 | { |
1578 | trace_rcu_future_grace_period(rdp->rsp->name, rnp->gpnum, | |
1579 | rnp->completed, c, rnp->level, | |
1580 | rnp->grplo, rnp->grphi, s); | |
1581 | } | |
1582 | ||
1583 | /* | |
1584 | * Start some future grace period, as needed to handle newly arrived | |
1585 | * callbacks. The required future grace periods are recorded in each | |
48a7639c PM |
1586 | * rcu_node structure's ->need_future_gp field. Returns true if there |
1587 | * is reason to awaken the grace-period kthread. | |
0446be48 PM |
1588 | * |
1589 | * The caller must hold the specified rcu_node structure's ->lock. | |
1590 | */ | |
48a7639c PM |
1591 | static bool __maybe_unused |
1592 | rcu_start_future_gp(struct rcu_node *rnp, struct rcu_data *rdp, | |
1593 | unsigned long *c_out) | |
0446be48 PM |
1594 | { |
1595 | unsigned long c; | |
1596 | int i; | |
48a7639c | 1597 | bool ret = false; |
0446be48 PM |
1598 | struct rcu_node *rnp_root = rcu_get_root(rdp->rsp); |
1599 | ||
1600 | /* | |
1601 | * Pick up grace-period number for new callbacks. If this | |
1602 | * grace period is already marked as needed, return to the caller. | |
1603 | */ | |
1604 | c = rcu_cbs_completed(rdp->rsp, rnp); | |
f7f7bac9 | 1605 | trace_rcu_future_gp(rnp, rdp, c, TPS("Startleaf")); |
0446be48 | 1606 | if (rnp->need_future_gp[c & 0x1]) { |
f7f7bac9 | 1607 | trace_rcu_future_gp(rnp, rdp, c, TPS("Prestartleaf")); |
48a7639c | 1608 | goto out; |
0446be48 PM |
1609 | } |
1610 | ||
1611 | /* | |
1612 | * If either this rcu_node structure or the root rcu_node structure | |
1613 | * believe that a grace period is in progress, then we must wait | |
1614 | * for the one following, which is in "c". Because our request | |
1615 | * will be noticed at the end of the current grace period, we don't | |
48bd8e9b PK |
1616 | * need to explicitly start one. We only do the lockless check |
1617 | * of rnp_root's fields if the current rcu_node structure thinks | |
1618 | * there is no grace period in flight, and because we hold rnp->lock, | |
1619 | * the only possible change is when rnp_root's two fields are | |
1620 | * equal, in which case rnp_root->gpnum might be concurrently | |
1621 | * incremented. But that is OK, as it will just result in our | |
1622 | * doing some extra useless work. | |
0446be48 PM |
1623 | */ |
1624 | if (rnp->gpnum != rnp->completed || | |
7d0ae808 | 1625 | READ_ONCE(rnp_root->gpnum) != READ_ONCE(rnp_root->completed)) { |
0446be48 | 1626 | rnp->need_future_gp[c & 0x1]++; |
f7f7bac9 | 1627 | trace_rcu_future_gp(rnp, rdp, c, TPS("Startedleaf")); |
48a7639c | 1628 | goto out; |
0446be48 PM |
1629 | } |
1630 | ||
1631 | /* | |
1632 | * There might be no grace period in progress. If we don't already | |
1633 | * hold it, acquire the root rcu_node structure's lock in order to | |
1634 | * start one (if needed). | |
1635 | */ | |
2a67e741 PZ |
1636 | if (rnp != rnp_root) |
1637 | raw_spin_lock_rcu_node(rnp_root); | |
0446be48 PM |
1638 | |
1639 | /* | |
1640 | * Get a new grace-period number. If there really is no grace | |
1641 | * period in progress, it will be smaller than the one we obtained | |
1642 | * earlier. Adjust callbacks as needed. Note that even no-CBs | |
1643 | * CPUs have a ->nxtcompleted[] array, so no no-CBs checks needed. | |
1644 | */ | |
1645 | c = rcu_cbs_completed(rdp->rsp, rnp_root); | |
1646 | for (i = RCU_DONE_TAIL; i < RCU_NEXT_TAIL; i++) | |
1647 | if (ULONG_CMP_LT(c, rdp->nxtcompleted[i])) | |
1648 | rdp->nxtcompleted[i] = c; | |
1649 | ||
1650 | /* | |
1651 | * If the needed for the required grace period is already | |
1652 | * recorded, trace and leave. | |
1653 | */ | |
1654 | if (rnp_root->need_future_gp[c & 0x1]) { | |
f7f7bac9 | 1655 | trace_rcu_future_gp(rnp, rdp, c, TPS("Prestartedroot")); |
0446be48 PM |
1656 | goto unlock_out; |
1657 | } | |
1658 | ||
1659 | /* Record the need for the future grace period. */ | |
1660 | rnp_root->need_future_gp[c & 0x1]++; | |
1661 | ||
1662 | /* If a grace period is not already in progress, start one. */ | |
1663 | if (rnp_root->gpnum != rnp_root->completed) { | |
f7f7bac9 | 1664 | trace_rcu_future_gp(rnp, rdp, c, TPS("Startedleafroot")); |
0446be48 | 1665 | } else { |
f7f7bac9 | 1666 | trace_rcu_future_gp(rnp, rdp, c, TPS("Startedroot")); |
48a7639c | 1667 | ret = rcu_start_gp_advanced(rdp->rsp, rnp_root, rdp); |
0446be48 PM |
1668 | } |
1669 | unlock_out: | |
1670 | if (rnp != rnp_root) | |
67c583a7 | 1671 | raw_spin_unlock_rcu_node(rnp_root); |
48a7639c PM |
1672 | out: |
1673 | if (c_out != NULL) | |
1674 | *c_out = c; | |
1675 | return ret; | |
0446be48 PM |
1676 | } |
1677 | ||
1678 | /* | |
1679 | * Clean up any old requests for the just-ended grace period. Also return | |
1680 | * whether any additional grace periods have been requested. Also invoke | |
1681 | * rcu_nocb_gp_cleanup() in order to wake up any no-callbacks kthreads | |
1682 | * waiting for this grace period to complete. | |
1683 | */ | |
1684 | static int rcu_future_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp) | |
1685 | { | |
1686 | int c = rnp->completed; | |
1687 | int needmore; | |
1688 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); | |
1689 | ||
0446be48 PM |
1690 | rnp->need_future_gp[c & 0x1] = 0; |
1691 | needmore = rnp->need_future_gp[(c + 1) & 0x1]; | |
f7f7bac9 SRRH |
1692 | trace_rcu_future_gp(rnp, rdp, c, |
1693 | needmore ? TPS("CleanupMore") : TPS("Cleanup")); | |
0446be48 PM |
1694 | return needmore; |
1695 | } | |
1696 | ||
48a7639c PM |
1697 | /* |
1698 | * Awaken the grace-period kthread for the specified flavor of RCU. | |
1699 | * Don't do a self-awaken, and don't bother awakening when there is | |
1700 | * nothing for the grace-period kthread to do (as in several CPUs | |
1701 | * raced to awaken, and we lost), and finally don't try to awaken | |
1702 | * a kthread that has not yet been created. | |
1703 | */ | |
1704 | static void rcu_gp_kthread_wake(struct rcu_state *rsp) | |
1705 | { | |
1706 | if (current == rsp->gp_kthread || | |
7d0ae808 | 1707 | !READ_ONCE(rsp->gp_flags) || |
48a7639c PM |
1708 | !rsp->gp_kthread) |
1709 | return; | |
abedf8e2 | 1710 | swake_up(&rsp->gp_wq); |
48a7639c PM |
1711 | } |
1712 | ||
dc35c893 PM |
1713 | /* |
1714 | * If there is room, assign a ->completed number to any callbacks on | |
1715 | * this CPU that have not already been assigned. Also accelerate any | |
1716 | * callbacks that were previously assigned a ->completed number that has | |
1717 | * since proven to be too conservative, which can happen if callbacks get | |
1718 | * assigned a ->completed number while RCU is idle, but with reference to | |
1719 | * a non-root rcu_node structure. This function is idempotent, so it does | |
48a7639c PM |
1720 | * not hurt to call it repeatedly. Returns an flag saying that we should |
1721 | * awaken the RCU grace-period kthread. | |
dc35c893 PM |
1722 | * |
1723 | * The caller must hold rnp->lock with interrupts disabled. | |
1724 | */ | |
48a7639c | 1725 | static bool rcu_accelerate_cbs(struct rcu_state *rsp, struct rcu_node *rnp, |
dc35c893 PM |
1726 | struct rcu_data *rdp) |
1727 | { | |
1728 | unsigned long c; | |
1729 | int i; | |
48a7639c | 1730 | bool ret; |
dc35c893 PM |
1731 | |
1732 | /* If the CPU has no callbacks, nothing to do. */ | |
1733 | if (!rdp->nxttail[RCU_NEXT_TAIL] || !*rdp->nxttail[RCU_DONE_TAIL]) | |
48a7639c | 1734 | return false; |
dc35c893 PM |
1735 | |
1736 | /* | |
1737 | * Starting from the sublist containing the callbacks most | |
1738 | * recently assigned a ->completed number and working down, find the | |
1739 | * first sublist that is not assignable to an upcoming grace period. | |
1740 | * Such a sublist has something in it (first two tests) and has | |
1741 | * a ->completed number assigned that will complete sooner than | |
1742 | * the ->completed number for newly arrived callbacks (last test). | |
1743 | * | |
1744 | * The key point is that any later sublist can be assigned the | |
1745 | * same ->completed number as the newly arrived callbacks, which | |
1746 | * means that the callbacks in any of these later sublist can be | |
1747 | * grouped into a single sublist, whether or not they have already | |
1748 | * been assigned a ->completed number. | |
1749 | */ | |
1750 | c = rcu_cbs_completed(rsp, rnp); | |
1751 | for (i = RCU_NEXT_TAIL - 1; i > RCU_DONE_TAIL; i--) | |
1752 | if (rdp->nxttail[i] != rdp->nxttail[i - 1] && | |
1753 | !ULONG_CMP_GE(rdp->nxtcompleted[i], c)) | |
1754 | break; | |
1755 | ||
1756 | /* | |
1757 | * If there are no sublist for unassigned callbacks, leave. | |
1758 | * At the same time, advance "i" one sublist, so that "i" will | |
1759 | * index into the sublist where all the remaining callbacks should | |
1760 | * be grouped into. | |
1761 | */ | |
1762 | if (++i >= RCU_NEXT_TAIL) | |
48a7639c | 1763 | return false; |
dc35c893 PM |
1764 | |
1765 | /* | |
1766 | * Assign all subsequent callbacks' ->completed number to the next | |
1767 | * full grace period and group them all in the sublist initially | |
1768 | * indexed by "i". | |
1769 | */ | |
1770 | for (; i <= RCU_NEXT_TAIL; i++) { | |
1771 | rdp->nxttail[i] = rdp->nxttail[RCU_NEXT_TAIL]; | |
1772 | rdp->nxtcompleted[i] = c; | |
1773 | } | |
910ee45d | 1774 | /* Record any needed additional grace periods. */ |
48a7639c | 1775 | ret = rcu_start_future_gp(rnp, rdp, NULL); |
6d4b418c PM |
1776 | |
1777 | /* Trace depending on how much we were able to accelerate. */ | |
1778 | if (!*rdp->nxttail[RCU_WAIT_TAIL]) | |
f7f7bac9 | 1779 | trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("AccWaitCB")); |
6d4b418c | 1780 | else |
f7f7bac9 | 1781 | trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("AccReadyCB")); |
48a7639c | 1782 | return ret; |
dc35c893 PM |
1783 | } |
1784 | ||
1785 | /* | |
1786 | * Move any callbacks whose grace period has completed to the | |
1787 | * RCU_DONE_TAIL sublist, then compact the remaining sublists and | |
1788 | * assign ->completed numbers to any callbacks in the RCU_NEXT_TAIL | |
1789 | * sublist. This function is idempotent, so it does not hurt to | |
1790 | * invoke it repeatedly. As long as it is not invoked -too- often... | |
48a7639c | 1791 | * Returns true if the RCU grace-period kthread needs to be awakened. |
dc35c893 PM |
1792 | * |
1793 | * The caller must hold rnp->lock with interrupts disabled. | |
1794 | */ | |
48a7639c | 1795 | static bool rcu_advance_cbs(struct rcu_state *rsp, struct rcu_node *rnp, |
dc35c893 PM |
1796 | struct rcu_data *rdp) |
1797 | { | |
1798 | int i, j; | |
1799 | ||
1800 | /* If the CPU has no callbacks, nothing to do. */ | |
1801 | if (!rdp->nxttail[RCU_NEXT_TAIL] || !*rdp->nxttail[RCU_DONE_TAIL]) | |
48a7639c | 1802 | return false; |
dc35c893 PM |
1803 | |
1804 | /* | |
1805 | * Find all callbacks whose ->completed numbers indicate that they | |
1806 | * are ready to invoke, and put them into the RCU_DONE_TAIL sublist. | |
1807 | */ | |
1808 | for (i = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++) { | |
1809 | if (ULONG_CMP_LT(rnp->completed, rdp->nxtcompleted[i])) | |
1810 | break; | |
1811 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[i]; | |
1812 | } | |
1813 | /* Clean up any sublist tail pointers that were misordered above. */ | |
1814 | for (j = RCU_WAIT_TAIL; j < i; j++) | |
1815 | rdp->nxttail[j] = rdp->nxttail[RCU_DONE_TAIL]; | |
1816 | ||
1817 | /* Copy down callbacks to fill in empty sublists. */ | |
1818 | for (j = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++, j++) { | |
1819 | if (rdp->nxttail[j] == rdp->nxttail[RCU_NEXT_TAIL]) | |
1820 | break; | |
1821 | rdp->nxttail[j] = rdp->nxttail[i]; | |
1822 | rdp->nxtcompleted[j] = rdp->nxtcompleted[i]; | |
1823 | } | |
1824 | ||
1825 | /* Classify any remaining callbacks. */ | |
48a7639c | 1826 | return rcu_accelerate_cbs(rsp, rnp, rdp); |
dc35c893 PM |
1827 | } |
1828 | ||
d09b62df | 1829 | /* |
ba9fbe95 PM |
1830 | * Update CPU-local rcu_data state to record the beginnings and ends of |
1831 | * grace periods. The caller must hold the ->lock of the leaf rcu_node | |
1832 | * structure corresponding to the current CPU, and must have irqs disabled. | |
48a7639c | 1833 | * Returns true if the grace-period kthread needs to be awakened. |
d09b62df | 1834 | */ |
48a7639c PM |
1835 | static bool __note_gp_changes(struct rcu_state *rsp, struct rcu_node *rnp, |
1836 | struct rcu_data *rdp) | |
d09b62df | 1837 | { |
48a7639c PM |
1838 | bool ret; |
1839 | ||
ba9fbe95 | 1840 | /* Handle the ends of any preceding grace periods first. */ |
e3663b10 | 1841 | if (rdp->completed == rnp->completed && |
7d0ae808 | 1842 | !unlikely(READ_ONCE(rdp->gpwrap))) { |
d09b62df | 1843 | |
ba9fbe95 | 1844 | /* No grace period end, so just accelerate recent callbacks. */ |
48a7639c | 1845 | ret = rcu_accelerate_cbs(rsp, rnp, rdp); |
d09b62df | 1846 | |
dc35c893 PM |
1847 | } else { |
1848 | ||
1849 | /* Advance callbacks. */ | |
48a7639c | 1850 | ret = rcu_advance_cbs(rsp, rnp, rdp); |
d09b62df PM |
1851 | |
1852 | /* Remember that we saw this grace-period completion. */ | |
1853 | rdp->completed = rnp->completed; | |
f7f7bac9 | 1854 | trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuend")); |
d09b62df | 1855 | } |
398ebe60 | 1856 | |
7d0ae808 | 1857 | if (rdp->gpnum != rnp->gpnum || unlikely(READ_ONCE(rdp->gpwrap))) { |
6eaef633 PM |
1858 | /* |
1859 | * If the current grace period is waiting for this CPU, | |
1860 | * set up to detect a quiescent state, otherwise don't | |
1861 | * go looking for one. | |
1862 | */ | |
1863 | rdp->gpnum = rnp->gpnum; | |
f7f7bac9 | 1864 | trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpustart")); |
5b74c458 | 1865 | rdp->cpu_no_qs.b.norm = true; |
5cd37193 | 1866 | rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_qs_ctr); |
97c668b8 | 1867 | rdp->core_needs_qs = !!(rnp->qsmask & rdp->grpmask); |
6eaef633 | 1868 | zero_cpu_stall_ticks(rdp); |
7d0ae808 | 1869 | WRITE_ONCE(rdp->gpwrap, false); |
6eaef633 | 1870 | } |
48a7639c | 1871 | return ret; |
6eaef633 PM |
1872 | } |
1873 | ||
d34ea322 | 1874 | static void note_gp_changes(struct rcu_state *rsp, struct rcu_data *rdp) |
6eaef633 PM |
1875 | { |
1876 | unsigned long flags; | |
48a7639c | 1877 | bool needwake; |
6eaef633 PM |
1878 | struct rcu_node *rnp; |
1879 | ||
1880 | local_irq_save(flags); | |
1881 | rnp = rdp->mynode; | |
7d0ae808 PM |
1882 | if ((rdp->gpnum == READ_ONCE(rnp->gpnum) && |
1883 | rdp->completed == READ_ONCE(rnp->completed) && | |
1884 | !unlikely(READ_ONCE(rdp->gpwrap))) || /* w/out lock. */ | |
2a67e741 | 1885 | !raw_spin_trylock_rcu_node(rnp)) { /* irqs already off, so later. */ |
6eaef633 PM |
1886 | local_irq_restore(flags); |
1887 | return; | |
1888 | } | |
48a7639c | 1889 | needwake = __note_gp_changes(rsp, rnp, rdp); |
67c583a7 | 1890 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
48a7639c PM |
1891 | if (needwake) |
1892 | rcu_gp_kthread_wake(rsp); | |
6eaef633 PM |
1893 | } |
1894 | ||
0f41c0dd PM |
1895 | static void rcu_gp_slow(struct rcu_state *rsp, int delay) |
1896 | { | |
1897 | if (delay > 0 && | |
1898 | !(rsp->gpnum % (rcu_num_nodes * PER_RCU_NODE_PERIOD * delay))) | |
1899 | schedule_timeout_uninterruptible(delay); | |
1900 | } | |
1901 | ||
b3dbec76 | 1902 | /* |
45fed3e7 | 1903 | * Initialize a new grace period. Return false if no grace period required. |
b3dbec76 | 1904 | */ |
45fed3e7 | 1905 | static bool rcu_gp_init(struct rcu_state *rsp) |
b3dbec76 | 1906 | { |
0aa04b05 | 1907 | unsigned long oldmask; |
b3dbec76 | 1908 | struct rcu_data *rdp; |
7fdefc10 | 1909 | struct rcu_node *rnp = rcu_get_root(rsp); |
b3dbec76 | 1910 | |
7d0ae808 | 1911 | WRITE_ONCE(rsp->gp_activity, jiffies); |
2a67e741 | 1912 | raw_spin_lock_irq_rcu_node(rnp); |
7d0ae808 | 1913 | if (!READ_ONCE(rsp->gp_flags)) { |
f7be8209 | 1914 | /* Spurious wakeup, tell caller to go back to sleep. */ |
67c583a7 | 1915 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 1916 | return false; |
f7be8209 | 1917 | } |
7d0ae808 | 1918 | WRITE_ONCE(rsp->gp_flags, 0); /* Clear all flags: New grace period. */ |
b3dbec76 | 1919 | |
f7be8209 PM |
1920 | if (WARN_ON_ONCE(rcu_gp_in_progress(rsp))) { |
1921 | /* | |
1922 | * Grace period already in progress, don't start another. | |
1923 | * Not supposed to be able to happen. | |
1924 | */ | |
67c583a7 | 1925 | raw_spin_unlock_irq_rcu_node(rnp); |
45fed3e7 | 1926 | return false; |
7fdefc10 PM |
1927 | } |
1928 | ||
7fdefc10 | 1929 | /* Advance to a new grace period and initialize state. */ |
26cdfedf | 1930 | record_gp_stall_check_time(rsp); |
765a3f4f PM |
1931 | /* Record GP times before starting GP, hence smp_store_release(). */ |
1932 | smp_store_release(&rsp->gpnum, rsp->gpnum + 1); | |
f7f7bac9 | 1933 | trace_rcu_grace_period(rsp->name, rsp->gpnum, TPS("start")); |
67c583a7 | 1934 | raw_spin_unlock_irq_rcu_node(rnp); |
7fdefc10 | 1935 | |
0aa04b05 PM |
1936 | /* |
1937 | * Apply per-leaf buffered online and offline operations to the | |
1938 | * rcu_node tree. Note that this new grace period need not wait | |
1939 | * for subsequent online CPUs, and that quiescent-state forcing | |
1940 | * will handle subsequent offline CPUs. | |
1941 | */ | |
1942 | rcu_for_each_leaf_node(rsp, rnp) { | |
0f41c0dd | 1943 | rcu_gp_slow(rsp, gp_preinit_delay); |
2a67e741 | 1944 | raw_spin_lock_irq_rcu_node(rnp); |
0aa04b05 PM |
1945 | if (rnp->qsmaskinit == rnp->qsmaskinitnext && |
1946 | !rnp->wait_blkd_tasks) { | |
1947 | /* Nothing to do on this leaf rcu_node structure. */ | |
67c583a7 | 1948 | raw_spin_unlock_irq_rcu_node(rnp); |
0aa04b05 PM |
1949 | continue; |
1950 | } | |
1951 | ||
1952 | /* Record old state, apply changes to ->qsmaskinit field. */ | |
1953 | oldmask = rnp->qsmaskinit; | |
1954 | rnp->qsmaskinit = rnp->qsmaskinitnext; | |
1955 | ||
1956 | /* If zero-ness of ->qsmaskinit changed, propagate up tree. */ | |
1957 | if (!oldmask != !rnp->qsmaskinit) { | |
1958 | if (!oldmask) /* First online CPU for this rcu_node. */ | |
1959 | rcu_init_new_rnp(rnp); | |
1960 | else if (rcu_preempt_has_tasks(rnp)) /* blocked tasks */ | |
1961 | rnp->wait_blkd_tasks = true; | |
1962 | else /* Last offline CPU and can propagate. */ | |
1963 | rcu_cleanup_dead_rnp(rnp); | |
1964 | } | |
1965 | ||
1966 | /* | |
1967 | * If all waited-on tasks from prior grace period are | |
1968 | * done, and if all this rcu_node structure's CPUs are | |
1969 | * still offline, propagate up the rcu_node tree and | |
1970 | * clear ->wait_blkd_tasks. Otherwise, if one of this | |
1971 | * rcu_node structure's CPUs has since come back online, | |
1972 | * simply clear ->wait_blkd_tasks (but rcu_cleanup_dead_rnp() | |
1973 | * checks for this, so just call it unconditionally). | |
1974 | */ | |
1975 | if (rnp->wait_blkd_tasks && | |
1976 | (!rcu_preempt_has_tasks(rnp) || | |
1977 | rnp->qsmaskinit)) { | |
1978 | rnp->wait_blkd_tasks = false; | |
1979 | rcu_cleanup_dead_rnp(rnp); | |
1980 | } | |
1981 | ||
67c583a7 | 1982 | raw_spin_unlock_irq_rcu_node(rnp); |
0aa04b05 | 1983 | } |
7fdefc10 PM |
1984 | |
1985 | /* | |
1986 | * Set the quiescent-state-needed bits in all the rcu_node | |
1987 | * structures for all currently online CPUs in breadth-first order, | |
1988 | * starting from the root rcu_node structure, relying on the layout | |
1989 | * of the tree within the rsp->node[] array. Note that other CPUs | |
1990 | * will access only the leaves of the hierarchy, thus seeing that no | |
1991 | * grace period is in progress, at least until the corresponding | |
1992 | * leaf node has been initialized. In addition, we have excluded | |
1993 | * CPU-hotplug operations. | |
1994 | * | |
1995 | * The grace period cannot complete until the initialization | |
1996 | * process finishes, because this kthread handles both. | |
1997 | */ | |
1998 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
0f41c0dd | 1999 | rcu_gp_slow(rsp, gp_init_delay); |
2a67e741 | 2000 | raw_spin_lock_irq_rcu_node(rnp); |
b3dbec76 | 2001 | rdp = this_cpu_ptr(rsp->rda); |
7fdefc10 PM |
2002 | rcu_preempt_check_blocked_tasks(rnp); |
2003 | rnp->qsmask = rnp->qsmaskinit; | |
7d0ae808 | 2004 | WRITE_ONCE(rnp->gpnum, rsp->gpnum); |
3f47da0f | 2005 | if (WARN_ON_ONCE(rnp->completed != rsp->completed)) |
7d0ae808 | 2006 | WRITE_ONCE(rnp->completed, rsp->completed); |
7fdefc10 | 2007 | if (rnp == rdp->mynode) |
48a7639c | 2008 | (void)__note_gp_changes(rsp, rnp, rdp); |
7fdefc10 PM |
2009 | rcu_preempt_boost_start_gp(rnp); |
2010 | trace_rcu_grace_period_init(rsp->name, rnp->gpnum, | |
2011 | rnp->level, rnp->grplo, | |
2012 | rnp->grphi, rnp->qsmask); | |
67c583a7 | 2013 | raw_spin_unlock_irq_rcu_node(rnp); |
bde6c3aa | 2014 | cond_resched_rcu_qs(); |
7d0ae808 | 2015 | WRITE_ONCE(rsp->gp_activity, jiffies); |
7fdefc10 | 2016 | } |
b3dbec76 | 2017 | |
45fed3e7 | 2018 | return true; |
7fdefc10 | 2019 | } |
b3dbec76 | 2020 | |
b9a425cf PM |
2021 | /* |
2022 | * Helper function for wait_event_interruptible_timeout() wakeup | |
2023 | * at force-quiescent-state time. | |
2024 | */ | |
2025 | static bool rcu_gp_fqs_check_wake(struct rcu_state *rsp, int *gfp) | |
2026 | { | |
2027 | struct rcu_node *rnp = rcu_get_root(rsp); | |
2028 | ||
2029 | /* Someone like call_rcu() requested a force-quiescent-state scan. */ | |
2030 | *gfp = READ_ONCE(rsp->gp_flags); | |
2031 | if (*gfp & RCU_GP_FLAG_FQS) | |
2032 | return true; | |
2033 | ||
2034 | /* The current grace period has completed. */ | |
2035 | if (!READ_ONCE(rnp->qsmask) && !rcu_preempt_blocked_readers_cgp(rnp)) | |
2036 | return true; | |
2037 | ||
2038 | return false; | |
2039 | } | |
2040 | ||
4cdfc175 PM |
2041 | /* |
2042 | * Do one round of quiescent-state forcing. | |
2043 | */ | |
77f81fe0 | 2044 | static void rcu_gp_fqs(struct rcu_state *rsp, bool first_time) |
4cdfc175 | 2045 | { |
217af2a2 PM |
2046 | bool isidle = false; |
2047 | unsigned long maxj; | |
4cdfc175 PM |
2048 | struct rcu_node *rnp = rcu_get_root(rsp); |
2049 | ||
7d0ae808 | 2050 | WRITE_ONCE(rsp->gp_activity, jiffies); |
4cdfc175 | 2051 | rsp->n_force_qs++; |
77f81fe0 | 2052 | if (first_time) { |
4cdfc175 | 2053 | /* Collect dyntick-idle snapshots. */ |
0edd1b17 | 2054 | if (is_sysidle_rcu_state(rsp)) { |
e02b2edf | 2055 | isidle = true; |
0edd1b17 PM |
2056 | maxj = jiffies - ULONG_MAX / 4; |
2057 | } | |
217af2a2 PM |
2058 | force_qs_rnp(rsp, dyntick_save_progress_counter, |
2059 | &isidle, &maxj); | |
0edd1b17 | 2060 | rcu_sysidle_report_gp(rsp, isidle, maxj); |
4cdfc175 PM |
2061 | } else { |
2062 | /* Handle dyntick-idle and offline CPUs. */ | |
675da67f | 2063 | isidle = true; |
217af2a2 | 2064 | force_qs_rnp(rsp, rcu_implicit_dynticks_qs, &isidle, &maxj); |
4cdfc175 PM |
2065 | } |
2066 | /* Clear flag to prevent immediate re-entry. */ | |
7d0ae808 | 2067 | if (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) { |
2a67e741 | 2068 | raw_spin_lock_irq_rcu_node(rnp); |
7d0ae808 PM |
2069 | WRITE_ONCE(rsp->gp_flags, |
2070 | READ_ONCE(rsp->gp_flags) & ~RCU_GP_FLAG_FQS); | |
67c583a7 | 2071 | raw_spin_unlock_irq_rcu_node(rnp); |
4cdfc175 | 2072 | } |
4cdfc175 PM |
2073 | } |
2074 | ||
7fdefc10 PM |
2075 | /* |
2076 | * Clean up after the old grace period. | |
2077 | */ | |
4cdfc175 | 2078 | static void rcu_gp_cleanup(struct rcu_state *rsp) |
7fdefc10 PM |
2079 | { |
2080 | unsigned long gp_duration; | |
48a7639c | 2081 | bool needgp = false; |
dae6e64d | 2082 | int nocb = 0; |
7fdefc10 PM |
2083 | struct rcu_data *rdp; |
2084 | struct rcu_node *rnp = rcu_get_root(rsp); | |
abedf8e2 | 2085 | struct swait_queue_head *sq; |
b3dbec76 | 2086 | |
7d0ae808 | 2087 | WRITE_ONCE(rsp->gp_activity, jiffies); |
2a67e741 | 2088 | raw_spin_lock_irq_rcu_node(rnp); |
7fdefc10 PM |
2089 | gp_duration = jiffies - rsp->gp_start; |
2090 | if (gp_duration > rsp->gp_max) | |
2091 | rsp->gp_max = gp_duration; | |
b3dbec76 | 2092 | |
7fdefc10 PM |
2093 | /* |
2094 | * We know the grace period is complete, but to everyone else | |
2095 | * it appears to still be ongoing. But it is also the case | |
2096 | * that to everyone else it looks like there is nothing that | |
2097 | * they can do to advance the grace period. It is therefore | |
2098 | * safe for us to drop the lock in order to mark the grace | |
2099 | * period as completed in all of the rcu_node structures. | |
7fdefc10 | 2100 | */ |
67c583a7 | 2101 | raw_spin_unlock_irq_rcu_node(rnp); |
b3dbec76 | 2102 | |
5d4b8659 PM |
2103 | /* |
2104 | * Propagate new ->completed value to rcu_node structures so | |
2105 | * that other CPUs don't have to wait until the start of the next | |
2106 | * grace period to process their callbacks. This also avoids | |
2107 | * some nasty RCU grace-period initialization races by forcing | |
2108 | * the end of the current grace period to be completely recorded in | |
2109 | * all of the rcu_node structures before the beginning of the next | |
2110 | * grace period is recorded in any of the rcu_node structures. | |
2111 | */ | |
2112 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
2a67e741 | 2113 | raw_spin_lock_irq_rcu_node(rnp); |
5c60d25f PM |
2114 | WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp)); |
2115 | WARN_ON_ONCE(rnp->qsmask); | |
7d0ae808 | 2116 | WRITE_ONCE(rnp->completed, rsp->gpnum); |
b11cc576 PM |
2117 | rdp = this_cpu_ptr(rsp->rda); |
2118 | if (rnp == rdp->mynode) | |
48a7639c | 2119 | needgp = __note_gp_changes(rsp, rnp, rdp) || needgp; |
78e4bc34 | 2120 | /* smp_mb() provided by prior unlock-lock pair. */ |
0446be48 | 2121 | nocb += rcu_future_gp_cleanup(rsp, rnp); |
065bb78c | 2122 | sq = rcu_nocb_gp_get(rnp); |
67c583a7 | 2123 | raw_spin_unlock_irq_rcu_node(rnp); |
065bb78c | 2124 | rcu_nocb_gp_cleanup(sq); |
bde6c3aa | 2125 | cond_resched_rcu_qs(); |
7d0ae808 | 2126 | WRITE_ONCE(rsp->gp_activity, jiffies); |
0f41c0dd | 2127 | rcu_gp_slow(rsp, gp_cleanup_delay); |
7fdefc10 | 2128 | } |
5d4b8659 | 2129 | rnp = rcu_get_root(rsp); |
2a67e741 | 2130 | raw_spin_lock_irq_rcu_node(rnp); /* Order GP before ->completed update. */ |
dae6e64d | 2131 | rcu_nocb_gp_set(rnp, nocb); |
7fdefc10 | 2132 | |
765a3f4f | 2133 | /* Declare grace period done. */ |
7d0ae808 | 2134 | WRITE_ONCE(rsp->completed, rsp->gpnum); |
f7f7bac9 | 2135 | trace_rcu_grace_period(rsp->name, rsp->completed, TPS("end")); |
77f81fe0 | 2136 | rsp->gp_state = RCU_GP_IDLE; |
5d4b8659 | 2137 | rdp = this_cpu_ptr(rsp->rda); |
48a7639c PM |
2138 | /* Advance CBs to reduce false positives below. */ |
2139 | needgp = rcu_advance_cbs(rsp, rnp, rdp) || needgp; | |
2140 | if (needgp || cpu_needs_another_gp(rsp, rdp)) { | |
7d0ae808 | 2141 | WRITE_ONCE(rsp->gp_flags, RCU_GP_FLAG_INIT); |
bb311ecc | 2142 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2143 | READ_ONCE(rsp->gpnum), |
bb311ecc PM |
2144 | TPS("newreq")); |
2145 | } | |
67c583a7 | 2146 | raw_spin_unlock_irq_rcu_node(rnp); |
7fdefc10 PM |
2147 | } |
2148 | ||
2149 | /* | |
2150 | * Body of kthread that handles grace periods. | |
2151 | */ | |
2152 | static int __noreturn rcu_gp_kthread(void *arg) | |
2153 | { | |
77f81fe0 | 2154 | bool first_gp_fqs; |
88d6df61 | 2155 | int gf; |
d40011f6 | 2156 | unsigned long j; |
4cdfc175 | 2157 | int ret; |
7fdefc10 PM |
2158 | struct rcu_state *rsp = arg; |
2159 | struct rcu_node *rnp = rcu_get_root(rsp); | |
2160 | ||
5871968d | 2161 | rcu_bind_gp_kthread(); |
7fdefc10 PM |
2162 | for (;;) { |
2163 | ||
2164 | /* Handle grace-period start. */ | |
2165 | for (;;) { | |
63c4db78 | 2166 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2167 | READ_ONCE(rsp->gpnum), |
63c4db78 | 2168 | TPS("reqwait")); |
afea227f | 2169 | rsp->gp_state = RCU_GP_WAIT_GPS; |
abedf8e2 | 2170 | swait_event_interruptible(rsp->gp_wq, |
7d0ae808 | 2171 | READ_ONCE(rsp->gp_flags) & |
4cdfc175 | 2172 | RCU_GP_FLAG_INIT); |
319362c9 | 2173 | rsp->gp_state = RCU_GP_DONE_GPS; |
78e4bc34 | 2174 | /* Locking provides needed memory barrier. */ |
f7be8209 | 2175 | if (rcu_gp_init(rsp)) |
7fdefc10 | 2176 | break; |
bde6c3aa | 2177 | cond_resched_rcu_qs(); |
7d0ae808 | 2178 | WRITE_ONCE(rsp->gp_activity, jiffies); |
73a860cd | 2179 | WARN_ON(signal_pending(current)); |
63c4db78 | 2180 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2181 | READ_ONCE(rsp->gpnum), |
63c4db78 | 2182 | TPS("reqwaitsig")); |
7fdefc10 | 2183 | } |
cabc49c1 | 2184 | |
4cdfc175 | 2185 | /* Handle quiescent-state forcing. */ |
77f81fe0 | 2186 | first_gp_fqs = true; |
d40011f6 PM |
2187 | j = jiffies_till_first_fqs; |
2188 | if (j > HZ) { | |
2189 | j = HZ; | |
2190 | jiffies_till_first_fqs = HZ; | |
2191 | } | |
88d6df61 | 2192 | ret = 0; |
cabc49c1 | 2193 | for (;;) { |
8c7c4829 | 2194 | if (!ret) { |
88d6df61 | 2195 | rsp->jiffies_force_qs = jiffies + j; |
8c7c4829 PM |
2196 | WRITE_ONCE(rsp->jiffies_kick_kthreads, |
2197 | jiffies + 3 * j); | |
2198 | } | |
63c4db78 | 2199 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2200 | READ_ONCE(rsp->gpnum), |
63c4db78 | 2201 | TPS("fqswait")); |
afea227f | 2202 | rsp->gp_state = RCU_GP_WAIT_FQS; |
abedf8e2 | 2203 | ret = swait_event_interruptible_timeout(rsp->gp_wq, |
b9a425cf | 2204 | rcu_gp_fqs_check_wake(rsp, &gf), j); |
32bb1c79 | 2205 | rsp->gp_state = RCU_GP_DOING_FQS; |
78e4bc34 | 2206 | /* Locking provides needed memory barriers. */ |
4cdfc175 | 2207 | /* If grace period done, leave loop. */ |
7d0ae808 | 2208 | if (!READ_ONCE(rnp->qsmask) && |
4cdfc175 | 2209 | !rcu_preempt_blocked_readers_cgp(rnp)) |
cabc49c1 | 2210 | break; |
4cdfc175 | 2211 | /* If time for quiescent-state forcing, do it. */ |
88d6df61 PM |
2212 | if (ULONG_CMP_GE(jiffies, rsp->jiffies_force_qs) || |
2213 | (gf & RCU_GP_FLAG_FQS)) { | |
63c4db78 | 2214 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2215 | READ_ONCE(rsp->gpnum), |
63c4db78 | 2216 | TPS("fqsstart")); |
77f81fe0 PM |
2217 | rcu_gp_fqs(rsp, first_gp_fqs); |
2218 | first_gp_fqs = false; | |
63c4db78 | 2219 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2220 | READ_ONCE(rsp->gpnum), |
63c4db78 | 2221 | TPS("fqsend")); |
bde6c3aa | 2222 | cond_resched_rcu_qs(); |
7d0ae808 | 2223 | WRITE_ONCE(rsp->gp_activity, jiffies); |
fcfd0a23 PM |
2224 | ret = 0; /* Force full wait till next FQS. */ |
2225 | j = jiffies_till_next_fqs; | |
2226 | if (j > HZ) { | |
2227 | j = HZ; | |
2228 | jiffies_till_next_fqs = HZ; | |
2229 | } else if (j < 1) { | |
2230 | j = 1; | |
2231 | jiffies_till_next_fqs = 1; | |
2232 | } | |
4cdfc175 PM |
2233 | } else { |
2234 | /* Deal with stray signal. */ | |
bde6c3aa | 2235 | cond_resched_rcu_qs(); |
7d0ae808 | 2236 | WRITE_ONCE(rsp->gp_activity, jiffies); |
73a860cd | 2237 | WARN_ON(signal_pending(current)); |
63c4db78 | 2238 | trace_rcu_grace_period(rsp->name, |
7d0ae808 | 2239 | READ_ONCE(rsp->gpnum), |
63c4db78 | 2240 | TPS("fqswaitsig")); |
fcfd0a23 PM |
2241 | ret = 1; /* Keep old FQS timing. */ |
2242 | j = jiffies; | |
2243 | if (time_after(jiffies, rsp->jiffies_force_qs)) | |
2244 | j = 1; | |
2245 | else | |
2246 | j = rsp->jiffies_force_qs - j; | |
d40011f6 | 2247 | } |
cabc49c1 | 2248 | } |
4cdfc175 PM |
2249 | |
2250 | /* Handle grace-period end. */ | |
319362c9 | 2251 | rsp->gp_state = RCU_GP_CLEANUP; |
4cdfc175 | 2252 | rcu_gp_cleanup(rsp); |
319362c9 | 2253 | rsp->gp_state = RCU_GP_CLEANED; |
b3dbec76 | 2254 | } |
b3dbec76 PM |
2255 | } |
2256 | ||
64db4cff PM |
2257 | /* |
2258 | * Start a new RCU grace period if warranted, re-initializing the hierarchy | |
2259 | * in preparation for detecting the next grace period. The caller must hold | |
b8462084 | 2260 | * the root node's ->lock and hard irqs must be disabled. |
e5601400 PM |
2261 | * |
2262 | * Note that it is legal for a dying CPU (which is marked as offline) to | |
2263 | * invoke this function. This can happen when the dying CPU reports its | |
2264 | * quiescent state. | |
48a7639c PM |
2265 | * |
2266 | * Returns true if the grace-period kthread must be awakened. | |
64db4cff | 2267 | */ |
48a7639c | 2268 | static bool |
910ee45d PM |
2269 | rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp, |
2270 | struct rcu_data *rdp) | |
64db4cff | 2271 | { |
b8462084 | 2272 | if (!rsp->gp_kthread || !cpu_needs_another_gp(rsp, rdp)) { |
afe24b12 | 2273 | /* |
b3dbec76 | 2274 | * Either we have not yet spawned the grace-period |
62da1921 PM |
2275 | * task, this CPU does not need another grace period, |
2276 | * or a grace period is already in progress. | |
b3dbec76 | 2277 | * Either way, don't start a new grace period. |
afe24b12 | 2278 | */ |
48a7639c | 2279 | return false; |
afe24b12 | 2280 | } |
7d0ae808 PM |
2281 | WRITE_ONCE(rsp->gp_flags, RCU_GP_FLAG_INIT); |
2282 | trace_rcu_grace_period(rsp->name, READ_ONCE(rsp->gpnum), | |
bb311ecc | 2283 | TPS("newreq")); |
62da1921 | 2284 | |
016a8d5b SR |
2285 | /* |
2286 | * We can't do wakeups while holding the rnp->lock, as that | |
1eafd31c | 2287 | * could cause possible deadlocks with the rq->lock. Defer |
48a7639c | 2288 | * the wakeup to our caller. |
016a8d5b | 2289 | */ |
48a7639c | 2290 | return true; |
64db4cff PM |
2291 | } |
2292 | ||
910ee45d PM |
2293 | /* |
2294 | * Similar to rcu_start_gp_advanced(), but also advance the calling CPU's | |
2295 | * callbacks. Note that rcu_start_gp_advanced() cannot do this because it | |
2296 | * is invoked indirectly from rcu_advance_cbs(), which would result in | |
2297 | * endless recursion -- or would do so if it wasn't for the self-deadlock | |
2298 | * that is encountered beforehand. | |
48a7639c PM |
2299 | * |
2300 | * Returns true if the grace-period kthread needs to be awakened. | |
910ee45d | 2301 | */ |
48a7639c | 2302 | static bool rcu_start_gp(struct rcu_state *rsp) |
910ee45d PM |
2303 | { |
2304 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); | |
2305 | struct rcu_node *rnp = rcu_get_root(rsp); | |
48a7639c | 2306 | bool ret = false; |
910ee45d PM |
2307 | |
2308 | /* | |
2309 | * If there is no grace period in progress right now, any | |
2310 | * callbacks we have up to this point will be satisfied by the | |
2311 | * next grace period. Also, advancing the callbacks reduces the | |
2312 | * probability of false positives from cpu_needs_another_gp() | |
2313 | * resulting in pointless grace periods. So, advance callbacks | |
2314 | * then start the grace period! | |
2315 | */ | |
48a7639c PM |
2316 | ret = rcu_advance_cbs(rsp, rnp, rdp) || ret; |
2317 | ret = rcu_start_gp_advanced(rsp, rnp, rdp) || ret; | |
2318 | return ret; | |
910ee45d PM |
2319 | } |
2320 | ||
f41d911f | 2321 | /* |
8994515c PM |
2322 | * Report a full set of quiescent states to the specified rcu_state data |
2323 | * structure. Invoke rcu_gp_kthread_wake() to awaken the grace-period | |
2324 | * kthread if another grace period is required. Whether we wake | |
2325 | * the grace-period kthread or it awakens itself for the next round | |
2326 | * of quiescent-state forcing, that kthread will clean up after the | |
2327 | * just-completed grace period. Note that the caller must hold rnp->lock, | |
2328 | * which is released before return. | |
f41d911f | 2329 | */ |
d3f6bad3 | 2330 | static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags) |
fc2219d4 | 2331 | __releases(rcu_get_root(rsp)->lock) |
f41d911f | 2332 | { |
fc2219d4 | 2333 | WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); |
cd73ca21 | 2334 | WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS); |
67c583a7 | 2335 | raw_spin_unlock_irqrestore_rcu_node(rcu_get_root(rsp), flags); |
abedf8e2 | 2336 | swake_up(&rsp->gp_wq); /* Memory barrier implied by swake_up() path. */ |
f41d911f PM |
2337 | } |
2338 | ||
64db4cff | 2339 | /* |
d3f6bad3 PM |
2340 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
2341 | * Allows quiescent states for a group of CPUs to be reported at one go | |
2342 | * to the specified rcu_node structure, though all the CPUs in the group | |
654e9533 PM |
2343 | * must be represented by the same rcu_node structure (which need not be a |
2344 | * leaf rcu_node structure, though it often will be). The gps parameter | |
2345 | * is the grace-period snapshot, which means that the quiescent states | |
2346 | * are valid only if rnp->gpnum is equal to gps. That structure's lock | |
2347 | * must be held upon entry, and it is released before return. | |
64db4cff PM |
2348 | */ |
2349 | static void | |
d3f6bad3 | 2350 | rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp, |
654e9533 | 2351 | struct rcu_node *rnp, unsigned long gps, unsigned long flags) |
64db4cff PM |
2352 | __releases(rnp->lock) |
2353 | { | |
654e9533 | 2354 | unsigned long oldmask = 0; |
28ecd580 PM |
2355 | struct rcu_node *rnp_c; |
2356 | ||
64db4cff PM |
2357 | /* Walk up the rcu_node hierarchy. */ |
2358 | for (;;) { | |
654e9533 | 2359 | if (!(rnp->qsmask & mask) || rnp->gpnum != gps) { |
64db4cff | 2360 | |
654e9533 PM |
2361 | /* |
2362 | * Our bit has already been cleared, or the | |
2363 | * relevant grace period is already over, so done. | |
2364 | */ | |
67c583a7 | 2365 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2366 | return; |
2367 | } | |
654e9533 | 2368 | WARN_ON_ONCE(oldmask); /* Any child must be all zeroed! */ |
64db4cff | 2369 | rnp->qsmask &= ~mask; |
d4c08f2a PM |
2370 | trace_rcu_quiescent_state_report(rsp->name, rnp->gpnum, |
2371 | mask, rnp->qsmask, rnp->level, | |
2372 | rnp->grplo, rnp->grphi, | |
2373 | !!rnp->gp_tasks); | |
27f4d280 | 2374 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
64db4cff PM |
2375 | |
2376 | /* Other bits still set at this level, so done. */ | |
67c583a7 | 2377 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2378 | return; |
2379 | } | |
2380 | mask = rnp->grpmask; | |
2381 | if (rnp->parent == NULL) { | |
2382 | ||
2383 | /* No more levels. Exit loop holding root lock. */ | |
2384 | ||
2385 | break; | |
2386 | } | |
67c583a7 | 2387 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
28ecd580 | 2388 | rnp_c = rnp; |
64db4cff | 2389 | rnp = rnp->parent; |
2a67e741 | 2390 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
654e9533 | 2391 | oldmask = rnp_c->qsmask; |
64db4cff PM |
2392 | } |
2393 | ||
2394 | /* | |
2395 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 2396 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 2397 | * to clean up and start the next grace period if one is needed. |
64db4cff | 2398 | */ |
d3f6bad3 | 2399 | rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */ |
64db4cff PM |
2400 | } |
2401 | ||
cc99a310 PM |
2402 | /* |
2403 | * Record a quiescent state for all tasks that were previously queued | |
2404 | * on the specified rcu_node structure and that were blocking the current | |
2405 | * RCU grace period. The caller must hold the specified rnp->lock with | |
2406 | * irqs disabled, and this lock is released upon return, but irqs remain | |
2407 | * disabled. | |
2408 | */ | |
0aa04b05 | 2409 | static void rcu_report_unblock_qs_rnp(struct rcu_state *rsp, |
cc99a310 PM |
2410 | struct rcu_node *rnp, unsigned long flags) |
2411 | __releases(rnp->lock) | |
2412 | { | |
654e9533 | 2413 | unsigned long gps; |
cc99a310 PM |
2414 | unsigned long mask; |
2415 | struct rcu_node *rnp_p; | |
2416 | ||
a77da14c PM |
2417 | if (rcu_state_p == &rcu_sched_state || rsp != rcu_state_p || |
2418 | rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { | |
67c583a7 | 2419 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
cc99a310 PM |
2420 | return; /* Still need more quiescent states! */ |
2421 | } | |
2422 | ||
2423 | rnp_p = rnp->parent; | |
2424 | if (rnp_p == NULL) { | |
2425 | /* | |
a77da14c PM |
2426 | * Only one rcu_node structure in the tree, so don't |
2427 | * try to report up to its nonexistent parent! | |
cc99a310 PM |
2428 | */ |
2429 | rcu_report_qs_rsp(rsp, flags); | |
2430 | return; | |
2431 | } | |
2432 | ||
654e9533 PM |
2433 | /* Report up the rest of the hierarchy, tracking current ->gpnum. */ |
2434 | gps = rnp->gpnum; | |
cc99a310 | 2435 | mask = rnp->grpmask; |
67c583a7 | 2436 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
2a67e741 | 2437 | raw_spin_lock_rcu_node(rnp_p); /* irqs already disabled. */ |
654e9533 | 2438 | rcu_report_qs_rnp(mask, rsp, rnp_p, gps, flags); |
cc99a310 PM |
2439 | } |
2440 | ||
64db4cff | 2441 | /* |
d3f6bad3 | 2442 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
4b455dc3 | 2443 | * structure. This must be called from the specified CPU. |
64db4cff PM |
2444 | */ |
2445 | static void | |
d7d6a11e | 2446 | rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
2447 | { |
2448 | unsigned long flags; | |
2449 | unsigned long mask; | |
48a7639c | 2450 | bool needwake; |
64db4cff PM |
2451 | struct rcu_node *rnp; |
2452 | ||
2453 | rnp = rdp->mynode; | |
2a67e741 | 2454 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
5b74c458 | 2455 | if ((rdp->cpu_no_qs.b.norm && |
5cd37193 PM |
2456 | rdp->rcu_qs_ctr_snap == __this_cpu_read(rcu_qs_ctr)) || |
2457 | rdp->gpnum != rnp->gpnum || rnp->completed == rnp->gpnum || | |
2458 | rdp->gpwrap) { | |
64db4cff PM |
2459 | |
2460 | /* | |
e4cc1f22 PM |
2461 | * The grace period in which this quiescent state was |
2462 | * recorded has ended, so don't report it upwards. | |
2463 | * We will instead need a new quiescent state that lies | |
2464 | * within the current grace period. | |
64db4cff | 2465 | */ |
5b74c458 | 2466 | rdp->cpu_no_qs.b.norm = true; /* need qs for new gp. */ |
5cd37193 | 2467 | rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_qs_ctr); |
67c583a7 | 2468 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
2469 | return; |
2470 | } | |
2471 | mask = rdp->grpmask; | |
2472 | if ((rnp->qsmask & mask) == 0) { | |
67c583a7 | 2473 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 2474 | } else { |
bb53e416 | 2475 | rdp->core_needs_qs = false; |
64db4cff PM |
2476 | |
2477 | /* | |
2478 | * This GP can't end until cpu checks in, so all of our | |
2479 | * callbacks can be processed during the next GP. | |
2480 | */ | |
48a7639c | 2481 | needwake = rcu_accelerate_cbs(rsp, rnp, rdp); |
64db4cff | 2482 | |
654e9533 PM |
2483 | rcu_report_qs_rnp(mask, rsp, rnp, rnp->gpnum, flags); |
2484 | /* ^^^ Released rnp->lock */ | |
48a7639c PM |
2485 | if (needwake) |
2486 | rcu_gp_kthread_wake(rsp); | |
64db4cff PM |
2487 | } |
2488 | } | |
2489 | ||
2490 | /* | |
2491 | * Check to see if there is a new grace period of which this CPU | |
2492 | * is not yet aware, and if so, set up local rcu_data state for it. | |
2493 | * Otherwise, see if this CPU has just passed through its first | |
2494 | * quiescent state for this grace period, and record that fact if so. | |
2495 | */ | |
2496 | static void | |
2497 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | |
2498 | { | |
05eb552b PM |
2499 | /* Check for grace-period ends and beginnings. */ |
2500 | note_gp_changes(rsp, rdp); | |
64db4cff PM |
2501 | |
2502 | /* | |
2503 | * Does this CPU still need to do its part for current grace period? | |
2504 | * If no, return and let the other CPUs do their part as well. | |
2505 | */ | |
97c668b8 | 2506 | if (!rdp->core_needs_qs) |
64db4cff PM |
2507 | return; |
2508 | ||
2509 | /* | |
2510 | * Was there a quiescent state since the beginning of the grace | |
2511 | * period? If no, then exit and wait for the next call. | |
2512 | */ | |
5b74c458 | 2513 | if (rdp->cpu_no_qs.b.norm && |
5cd37193 | 2514 | rdp->rcu_qs_ctr_snap == __this_cpu_read(rcu_qs_ctr)) |
64db4cff PM |
2515 | return; |
2516 | ||
d3f6bad3 PM |
2517 | /* |
2518 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
2519 | * judge of that). | |
2520 | */ | |
d7d6a11e | 2521 | rcu_report_qs_rdp(rdp->cpu, rsp, rdp); |
64db4cff PM |
2522 | } |
2523 | ||
e74f4c45 | 2524 | /* |
b1420f1c PM |
2525 | * Send the specified CPU's RCU callbacks to the orphanage. The |
2526 | * specified CPU must be offline, and the caller must hold the | |
7b2e6011 | 2527 | * ->orphan_lock. |
e74f4c45 | 2528 | */ |
b1420f1c PM |
2529 | static void |
2530 | rcu_send_cbs_to_orphanage(int cpu, struct rcu_state *rsp, | |
2531 | struct rcu_node *rnp, struct rcu_data *rdp) | |
e74f4c45 | 2532 | { |
3fbfbf7a | 2533 | /* No-CBs CPUs do not have orphanable callbacks. */ |
ea46351c | 2534 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU) || rcu_is_nocb_cpu(rdp->cpu)) |
3fbfbf7a PM |
2535 | return; |
2536 | ||
b1420f1c PM |
2537 | /* |
2538 | * Orphan the callbacks. First adjust the counts. This is safe | |
abfd6e58 PM |
2539 | * because _rcu_barrier() excludes CPU-hotplug operations, so it |
2540 | * cannot be running now. Thus no memory barrier is required. | |
b1420f1c | 2541 | */ |
a50c3af9 | 2542 | if (rdp->nxtlist != NULL) { |
b1420f1c PM |
2543 | rsp->qlen_lazy += rdp->qlen_lazy; |
2544 | rsp->qlen += rdp->qlen; | |
2545 | rdp->n_cbs_orphaned += rdp->qlen; | |
a50c3af9 | 2546 | rdp->qlen_lazy = 0; |
7d0ae808 | 2547 | WRITE_ONCE(rdp->qlen, 0); |
a50c3af9 PM |
2548 | } |
2549 | ||
2550 | /* | |
b1420f1c PM |
2551 | * Next, move those callbacks still needing a grace period to |
2552 | * the orphanage, where some other CPU will pick them up. | |
2553 | * Some of the callbacks might have gone partway through a grace | |
2554 | * period, but that is too bad. They get to start over because we | |
2555 | * cannot assume that grace periods are synchronized across CPUs. | |
2556 | * We don't bother updating the ->nxttail[] array yet, instead | |
2557 | * we just reset the whole thing later on. | |
a50c3af9 | 2558 | */ |
b1420f1c PM |
2559 | if (*rdp->nxttail[RCU_DONE_TAIL] != NULL) { |
2560 | *rsp->orphan_nxttail = *rdp->nxttail[RCU_DONE_TAIL]; | |
2561 | rsp->orphan_nxttail = rdp->nxttail[RCU_NEXT_TAIL]; | |
2562 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | |
a50c3af9 PM |
2563 | } |
2564 | ||
2565 | /* | |
b1420f1c PM |
2566 | * Then move the ready-to-invoke callbacks to the orphanage, |
2567 | * where some other CPU will pick them up. These will not be | |
2568 | * required to pass though another grace period: They are done. | |
a50c3af9 | 2569 | */ |
e5601400 | 2570 | if (rdp->nxtlist != NULL) { |
b1420f1c PM |
2571 | *rsp->orphan_donetail = rdp->nxtlist; |
2572 | rsp->orphan_donetail = rdp->nxttail[RCU_DONE_TAIL]; | |
e5601400 | 2573 | } |
e74f4c45 | 2574 | |
b33078b6 PM |
2575 | /* |
2576 | * Finally, initialize the rcu_data structure's list to empty and | |
2577 | * disallow further callbacks on this CPU. | |
2578 | */ | |
3f5d3ea6 | 2579 | init_callback_list(rdp); |
b33078b6 | 2580 | rdp->nxttail[RCU_NEXT_TAIL] = NULL; |
b1420f1c PM |
2581 | } |
2582 | ||
2583 | /* | |
2584 | * Adopt the RCU callbacks from the specified rcu_state structure's | |
7b2e6011 | 2585 | * orphanage. The caller must hold the ->orphan_lock. |
b1420f1c | 2586 | */ |
96d3fd0d | 2587 | static void rcu_adopt_orphan_cbs(struct rcu_state *rsp, unsigned long flags) |
b1420f1c PM |
2588 | { |
2589 | int i; | |
fa07a58f | 2590 | struct rcu_data *rdp = raw_cpu_ptr(rsp->rda); |
b1420f1c | 2591 | |
3fbfbf7a | 2592 | /* No-CBs CPUs are handled specially. */ |
ea46351c PM |
2593 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU) || |
2594 | rcu_nocb_adopt_orphan_cbs(rsp, rdp, flags)) | |
3fbfbf7a PM |
2595 | return; |
2596 | ||
b1420f1c PM |
2597 | /* Do the accounting first. */ |
2598 | rdp->qlen_lazy += rsp->qlen_lazy; | |
2599 | rdp->qlen += rsp->qlen; | |
2600 | rdp->n_cbs_adopted += rsp->qlen; | |
8f5af6f1 PM |
2601 | if (rsp->qlen_lazy != rsp->qlen) |
2602 | rcu_idle_count_callbacks_posted(); | |
b1420f1c PM |
2603 | rsp->qlen_lazy = 0; |
2604 | rsp->qlen = 0; | |
2605 | ||
2606 | /* | |
2607 | * We do not need a memory barrier here because the only way we | |
2608 | * can get here if there is an rcu_barrier() in flight is if | |
2609 | * we are the task doing the rcu_barrier(). | |
2610 | */ | |
2611 | ||
2612 | /* First adopt the ready-to-invoke callbacks. */ | |
2613 | if (rsp->orphan_donelist != NULL) { | |
2614 | *rsp->orphan_donetail = *rdp->nxttail[RCU_DONE_TAIL]; | |
2615 | *rdp->nxttail[RCU_DONE_TAIL] = rsp->orphan_donelist; | |
2616 | for (i = RCU_NEXT_SIZE - 1; i >= RCU_DONE_TAIL; i--) | |
2617 | if (rdp->nxttail[i] == rdp->nxttail[RCU_DONE_TAIL]) | |
2618 | rdp->nxttail[i] = rsp->orphan_donetail; | |
2619 | rsp->orphan_donelist = NULL; | |
2620 | rsp->orphan_donetail = &rsp->orphan_donelist; | |
2621 | } | |
2622 | ||
2623 | /* And then adopt the callbacks that still need a grace period. */ | |
2624 | if (rsp->orphan_nxtlist != NULL) { | |
2625 | *rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_nxtlist; | |
2626 | rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_nxttail; | |
2627 | rsp->orphan_nxtlist = NULL; | |
2628 | rsp->orphan_nxttail = &rsp->orphan_nxtlist; | |
2629 | } | |
2630 | } | |
2631 | ||
2632 | /* | |
2633 | * Trace the fact that this CPU is going offline. | |
2634 | */ | |
2635 | static void rcu_cleanup_dying_cpu(struct rcu_state *rsp) | |
2636 | { | |
2637 | RCU_TRACE(unsigned long mask); | |
2638 | RCU_TRACE(struct rcu_data *rdp = this_cpu_ptr(rsp->rda)); | |
2639 | RCU_TRACE(struct rcu_node *rnp = rdp->mynode); | |
2640 | ||
ea46351c PM |
2641 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
2642 | return; | |
2643 | ||
b1420f1c | 2644 | RCU_TRACE(mask = rdp->grpmask); |
e5601400 PM |
2645 | trace_rcu_grace_period(rsp->name, |
2646 | rnp->gpnum + 1 - !!(rnp->qsmask & mask), | |
f7f7bac9 | 2647 | TPS("cpuofl")); |
64db4cff PM |
2648 | } |
2649 | ||
8af3a5e7 PM |
2650 | /* |
2651 | * All CPUs for the specified rcu_node structure have gone offline, | |
2652 | * and all tasks that were preempted within an RCU read-side critical | |
2653 | * section while running on one of those CPUs have since exited their RCU | |
2654 | * read-side critical section. Some other CPU is reporting this fact with | |
2655 | * the specified rcu_node structure's ->lock held and interrupts disabled. | |
2656 | * This function therefore goes up the tree of rcu_node structures, | |
2657 | * clearing the corresponding bits in the ->qsmaskinit fields. Note that | |
2658 | * the leaf rcu_node structure's ->qsmaskinit field has already been | |
2659 | * updated | |
2660 | * | |
2661 | * This function does check that the specified rcu_node structure has | |
2662 | * all CPUs offline and no blocked tasks, so it is OK to invoke it | |
2663 | * prematurely. That said, invoking it after the fact will cost you | |
2664 | * a needless lock acquisition. So once it has done its work, don't | |
2665 | * invoke it again. | |
2666 | */ | |
2667 | static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf) | |
2668 | { | |
2669 | long mask; | |
2670 | struct rcu_node *rnp = rnp_leaf; | |
2671 | ||
ea46351c PM |
2672 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU) || |
2673 | rnp->qsmaskinit || rcu_preempt_has_tasks(rnp)) | |
8af3a5e7 PM |
2674 | return; |
2675 | for (;;) { | |
2676 | mask = rnp->grpmask; | |
2677 | rnp = rnp->parent; | |
2678 | if (!rnp) | |
2679 | break; | |
2a67e741 | 2680 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
8af3a5e7 | 2681 | rnp->qsmaskinit &= ~mask; |
0aa04b05 | 2682 | rnp->qsmask &= ~mask; |
8af3a5e7 | 2683 | if (rnp->qsmaskinit) { |
67c583a7 BF |
2684 | raw_spin_unlock_rcu_node(rnp); |
2685 | /* irqs remain disabled. */ | |
8af3a5e7 PM |
2686 | return; |
2687 | } | |
67c583a7 | 2688 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
8af3a5e7 PM |
2689 | } |
2690 | } | |
2691 | ||
64db4cff | 2692 | /* |
e5601400 | 2693 | * The CPU has been completely removed, and some other CPU is reporting |
b1420f1c PM |
2694 | * this fact from process context. Do the remainder of the cleanup, |
2695 | * including orphaning the outgoing CPU's RCU callbacks, and also | |
1331e7a1 PM |
2696 | * adopting them. There can only be one CPU hotplug operation at a time, |
2697 | * so no other CPU can be attempting to update rcu_cpu_kthread_task. | |
64db4cff | 2698 | */ |
e5601400 | 2699 | static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp) |
64db4cff | 2700 | { |
2036d94a | 2701 | unsigned long flags; |
e5601400 | 2702 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
b1420f1c | 2703 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ |
e5601400 | 2704 | |
ea46351c PM |
2705 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) |
2706 | return; | |
2707 | ||
2036d94a | 2708 | /* Adjust any no-longer-needed kthreads. */ |
5d01bbd1 | 2709 | rcu_boost_kthread_setaffinity(rnp, -1); |
2036d94a | 2710 | |
b1420f1c | 2711 | /* Orphan the dead CPU's callbacks, and adopt them if appropriate. */ |
78043c46 | 2712 | raw_spin_lock_irqsave(&rsp->orphan_lock, flags); |
b1420f1c | 2713 | rcu_send_cbs_to_orphanage(cpu, rsp, rnp, rdp); |
96d3fd0d | 2714 | rcu_adopt_orphan_cbs(rsp, flags); |
a8f4cbad | 2715 | raw_spin_unlock_irqrestore(&rsp->orphan_lock, flags); |
b1420f1c | 2716 | |
cf01537e PM |
2717 | WARN_ONCE(rdp->qlen != 0 || rdp->nxtlist != NULL, |
2718 | "rcu_cleanup_dead_cpu: Callbacks on offline CPU %d: qlen=%lu, nxtlist=%p\n", | |
2719 | cpu, rdp->qlen, rdp->nxtlist); | |
64db4cff PM |
2720 | } |
2721 | ||
64db4cff PM |
2722 | /* |
2723 | * Invoke any RCU callbacks that have made it to the end of their grace | |
2724 | * period. Thottle as specified by rdp->blimit. | |
2725 | */ | |
37c72e56 | 2726 | static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
2727 | { |
2728 | unsigned long flags; | |
2729 | struct rcu_head *next, *list, **tail; | |
878d7439 ED |
2730 | long bl, count, count_lazy; |
2731 | int i; | |
64db4cff | 2732 | |
dc35c893 | 2733 | /* If no callbacks are ready, just return. */ |
29c00b4a | 2734 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) { |
486e2593 | 2735 | trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, 0); |
7d0ae808 | 2736 | trace_rcu_batch_end(rsp->name, 0, !!READ_ONCE(rdp->nxtlist), |
4968c300 PM |
2737 | need_resched(), is_idle_task(current), |
2738 | rcu_is_callbacks_kthread()); | |
64db4cff | 2739 | return; |
29c00b4a | 2740 | } |
64db4cff PM |
2741 | |
2742 | /* | |
2743 | * Extract the list of ready callbacks, disabling to prevent | |
2744 | * races with call_rcu() from interrupt handlers. | |
2745 | */ | |
2746 | local_irq_save(flags); | |
8146c4e2 | 2747 | WARN_ON_ONCE(cpu_is_offline(smp_processor_id())); |
29c00b4a | 2748 | bl = rdp->blimit; |
486e2593 | 2749 | trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, bl); |
64db4cff PM |
2750 | list = rdp->nxtlist; |
2751 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; | |
2752 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | |
2753 | tail = rdp->nxttail[RCU_DONE_TAIL]; | |
b41772ab PM |
2754 | for (i = RCU_NEXT_SIZE - 1; i >= 0; i--) |
2755 | if (rdp->nxttail[i] == rdp->nxttail[RCU_DONE_TAIL]) | |
2756 | rdp->nxttail[i] = &rdp->nxtlist; | |
64db4cff PM |
2757 | local_irq_restore(flags); |
2758 | ||
2759 | /* Invoke callbacks. */ | |
486e2593 | 2760 | count = count_lazy = 0; |
64db4cff PM |
2761 | while (list) { |
2762 | next = list->next; | |
2763 | prefetch(next); | |
551d55a9 | 2764 | debug_rcu_head_unqueue(list); |
486e2593 PM |
2765 | if (__rcu_reclaim(rsp->name, list)) |
2766 | count_lazy++; | |
64db4cff | 2767 | list = next; |
dff1672d PM |
2768 | /* Stop only if limit reached and CPU has something to do. */ |
2769 | if (++count >= bl && | |
2770 | (need_resched() || | |
2771 | (!is_idle_task(current) && !rcu_is_callbacks_kthread()))) | |
64db4cff PM |
2772 | break; |
2773 | } | |
2774 | ||
2775 | local_irq_save(flags); | |
4968c300 PM |
2776 | trace_rcu_batch_end(rsp->name, count, !!list, need_resched(), |
2777 | is_idle_task(current), | |
2778 | rcu_is_callbacks_kthread()); | |
64db4cff PM |
2779 | |
2780 | /* Update count, and requeue any remaining callbacks. */ | |
64db4cff PM |
2781 | if (list != NULL) { |
2782 | *tail = rdp->nxtlist; | |
2783 | rdp->nxtlist = list; | |
b41772ab PM |
2784 | for (i = 0; i < RCU_NEXT_SIZE; i++) |
2785 | if (&rdp->nxtlist == rdp->nxttail[i]) | |
2786 | rdp->nxttail[i] = tail; | |
64db4cff PM |
2787 | else |
2788 | break; | |
2789 | } | |
b1420f1c PM |
2790 | smp_mb(); /* List handling before counting for rcu_barrier(). */ |
2791 | rdp->qlen_lazy -= count_lazy; | |
7d0ae808 | 2792 | WRITE_ONCE(rdp->qlen, rdp->qlen - count); |
b1420f1c | 2793 | rdp->n_cbs_invoked += count; |
64db4cff PM |
2794 | |
2795 | /* Reinstate batch limit if we have worked down the excess. */ | |
2796 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) | |
2797 | rdp->blimit = blimit; | |
2798 | ||
37c72e56 PM |
2799 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
2800 | if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) { | |
2801 | rdp->qlen_last_fqs_check = 0; | |
2802 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
2803 | } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark) | |
2804 | rdp->qlen_last_fqs_check = rdp->qlen; | |
cfca9279 | 2805 | WARN_ON_ONCE((rdp->nxtlist == NULL) != (rdp->qlen == 0)); |
37c72e56 | 2806 | |
64db4cff PM |
2807 | local_irq_restore(flags); |
2808 | ||
e0f23060 | 2809 | /* Re-invoke RCU core processing if there are callbacks remaining. */ |
64db4cff | 2810 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
a46e0899 | 2811 | invoke_rcu_core(); |
64db4cff PM |
2812 | } |
2813 | ||
2814 | /* | |
2815 | * Check to see if this CPU is in a non-context-switch quiescent state | |
2816 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
e0f23060 | 2817 | * Also schedule RCU core processing. |
64db4cff | 2818 | * |
9b2e4f18 | 2819 | * This function must be called from hardirq context. It is normally |
64db4cff PM |
2820 | * invoked from the scheduling-clock interrupt. If rcu_pending returns |
2821 | * false, there is no point in invoking rcu_check_callbacks(). | |
2822 | */ | |
c3377c2d | 2823 | void rcu_check_callbacks(int user) |
64db4cff | 2824 | { |
f7f7bac9 | 2825 | trace_rcu_utilization(TPS("Start scheduler-tick")); |
a858af28 | 2826 | increment_cpu_stall_ticks(); |
9b2e4f18 | 2827 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
64db4cff PM |
2828 | |
2829 | /* | |
2830 | * Get here if this CPU took its interrupt from user | |
2831 | * mode or from the idle loop, and if this is not a | |
2832 | * nested interrupt. In this case, the CPU is in | |
d6714c22 | 2833 | * a quiescent state, so note it. |
64db4cff PM |
2834 | * |
2835 | * No memory barrier is required here because both | |
d6714c22 PM |
2836 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
2837 | * variables that other CPUs neither access nor modify, | |
2838 | * at least not while the corresponding CPU is online. | |
64db4cff PM |
2839 | */ |
2840 | ||
284a8c93 PM |
2841 | rcu_sched_qs(); |
2842 | rcu_bh_qs(); | |
64db4cff PM |
2843 | |
2844 | } else if (!in_softirq()) { | |
2845 | ||
2846 | /* | |
2847 | * Get here if this CPU did not take its interrupt from | |
2848 | * softirq, in other words, if it is not interrupting | |
2849 | * a rcu_bh read-side critical section. This is an _bh | |
d6714c22 | 2850 | * critical section, so note it. |
64db4cff PM |
2851 | */ |
2852 | ||
284a8c93 | 2853 | rcu_bh_qs(); |
64db4cff | 2854 | } |
86aea0e6 | 2855 | rcu_preempt_check_callbacks(); |
e3950ecd | 2856 | if (rcu_pending()) |
a46e0899 | 2857 | invoke_rcu_core(); |
8315f422 PM |
2858 | if (user) |
2859 | rcu_note_voluntary_context_switch(current); | |
f7f7bac9 | 2860 | trace_rcu_utilization(TPS("End scheduler-tick")); |
64db4cff PM |
2861 | } |
2862 | ||
64db4cff PM |
2863 | /* |
2864 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
2865 | * have not yet encountered a quiescent state, using the function specified. | |
27f4d280 PM |
2866 | * Also initiate boosting for any threads blocked on the root rcu_node. |
2867 | * | |
ee47eb9f | 2868 | * The caller must have suppressed start of new grace periods. |
64db4cff | 2869 | */ |
217af2a2 PM |
2870 | static void force_qs_rnp(struct rcu_state *rsp, |
2871 | int (*f)(struct rcu_data *rsp, bool *isidle, | |
2872 | unsigned long *maxj), | |
2873 | bool *isidle, unsigned long *maxj) | |
64db4cff PM |
2874 | { |
2875 | unsigned long bit; | |
2876 | int cpu; | |
2877 | unsigned long flags; | |
2878 | unsigned long mask; | |
a0b6c9a7 | 2879 | struct rcu_node *rnp; |
64db4cff | 2880 | |
a0b6c9a7 | 2881 | rcu_for_each_leaf_node(rsp, rnp) { |
bde6c3aa | 2882 | cond_resched_rcu_qs(); |
64db4cff | 2883 | mask = 0; |
2a67e741 | 2884 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
a0b6c9a7 | 2885 | if (rnp->qsmask == 0) { |
a77da14c PM |
2886 | if (rcu_state_p == &rcu_sched_state || |
2887 | rsp != rcu_state_p || | |
2888 | rcu_preempt_blocked_readers_cgp(rnp)) { | |
2889 | /* | |
2890 | * No point in scanning bits because they | |
2891 | * are all zero. But we might need to | |
2892 | * priority-boost blocked readers. | |
2893 | */ | |
2894 | rcu_initiate_boost(rnp, flags); | |
2895 | /* rcu_initiate_boost() releases rnp->lock */ | |
2896 | continue; | |
2897 | } | |
2898 | if (rnp->parent && | |
2899 | (rnp->parent->qsmask & rnp->grpmask)) { | |
2900 | /* | |
2901 | * Race between grace-period | |
2902 | * initialization and task exiting RCU | |
2903 | * read-side critical section: Report. | |
2904 | */ | |
2905 | rcu_report_unblock_qs_rnp(rsp, rnp, flags); | |
2906 | /* rcu_report_unblock_qs_rnp() rlses ->lock */ | |
2907 | continue; | |
2908 | } | |
64db4cff | 2909 | } |
a0b6c9a7 | 2910 | cpu = rnp->grplo; |
64db4cff | 2911 | bit = 1; |
a0b6c9a7 | 2912 | for (; cpu <= rnp->grphi; cpu++, bit <<= 1) { |
0edd1b17 | 2913 | if ((rnp->qsmask & bit) != 0) { |
0edd1b17 PM |
2914 | if (f(per_cpu_ptr(rsp->rda, cpu), isidle, maxj)) |
2915 | mask |= bit; | |
2916 | } | |
64db4cff | 2917 | } |
45f014c5 | 2918 | if (mask != 0) { |
654e9533 PM |
2919 | /* Idle/offline CPUs, report (releases rnp->lock. */ |
2920 | rcu_report_qs_rnp(mask, rsp, rnp, rnp->gpnum, flags); | |
0aa04b05 PM |
2921 | } else { |
2922 | /* Nothing to do here, so just drop the lock. */ | |
67c583a7 | 2923 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff | 2924 | } |
64db4cff | 2925 | } |
64db4cff PM |
2926 | } |
2927 | ||
2928 | /* | |
2929 | * Force quiescent states on reluctant CPUs, and also detect which | |
2930 | * CPUs are in dyntick-idle mode. | |
2931 | */ | |
4cdfc175 | 2932 | static void force_quiescent_state(struct rcu_state *rsp) |
64db4cff PM |
2933 | { |
2934 | unsigned long flags; | |
394f2769 PM |
2935 | bool ret; |
2936 | struct rcu_node *rnp; | |
2937 | struct rcu_node *rnp_old = NULL; | |
2938 | ||
2939 | /* Funnel through hierarchy to reduce memory contention. */ | |
d860d403 | 2940 | rnp = __this_cpu_read(rsp->rda->mynode); |
394f2769 | 2941 | for (; rnp != NULL; rnp = rnp->parent) { |
7d0ae808 | 2942 | ret = (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) || |
394f2769 PM |
2943 | !raw_spin_trylock(&rnp->fqslock); |
2944 | if (rnp_old != NULL) | |
2945 | raw_spin_unlock(&rnp_old->fqslock); | |
2946 | if (ret) { | |
a792563b | 2947 | rsp->n_force_qs_lh++; |
394f2769 PM |
2948 | return; |
2949 | } | |
2950 | rnp_old = rnp; | |
2951 | } | |
2952 | /* rnp_old == rcu_get_root(rsp), rnp == NULL. */ | |
64db4cff | 2953 | |
394f2769 | 2954 | /* Reached the root of the rcu_node tree, acquire lock. */ |
2a67e741 | 2955 | raw_spin_lock_irqsave_rcu_node(rnp_old, flags); |
394f2769 | 2956 | raw_spin_unlock(&rnp_old->fqslock); |
7d0ae808 | 2957 | if (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) { |
a792563b | 2958 | rsp->n_force_qs_lh++; |
67c583a7 | 2959 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
4cdfc175 | 2960 | return; /* Someone beat us to it. */ |
46a1e34e | 2961 | } |
7d0ae808 | 2962 | WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS); |
67c583a7 | 2963 | raw_spin_unlock_irqrestore_rcu_node(rnp_old, flags); |
abedf8e2 | 2964 | swake_up(&rsp->gp_wq); /* Memory barrier implied by swake_up() path. */ |
64db4cff PM |
2965 | } |
2966 | ||
64db4cff | 2967 | /* |
e0f23060 PM |
2968 | * This does the RCU core processing work for the specified rcu_state |
2969 | * and rcu_data structures. This may be called only from the CPU to | |
2970 | * whom the rdp belongs. | |
64db4cff PM |
2971 | */ |
2972 | static void | |
1bca8cf1 | 2973 | __rcu_process_callbacks(struct rcu_state *rsp) |
64db4cff PM |
2974 | { |
2975 | unsigned long flags; | |
48a7639c | 2976 | bool needwake; |
fa07a58f | 2977 | struct rcu_data *rdp = raw_cpu_ptr(rsp->rda); |
64db4cff | 2978 | |
2e597558 PM |
2979 | WARN_ON_ONCE(rdp->beenonline == 0); |
2980 | ||
64db4cff PM |
2981 | /* Update RCU state based on any recent quiescent states. */ |
2982 | rcu_check_quiescent_state(rsp, rdp); | |
2983 | ||
2984 | /* Does this CPU require a not-yet-started grace period? */ | |
dc35c893 | 2985 | local_irq_save(flags); |
64db4cff | 2986 | if (cpu_needs_another_gp(rsp, rdp)) { |
6cf10081 | 2987 | raw_spin_lock_rcu_node(rcu_get_root(rsp)); /* irqs disabled. */ |
48a7639c | 2988 | needwake = rcu_start_gp(rsp); |
67c583a7 | 2989 | raw_spin_unlock_irqrestore_rcu_node(rcu_get_root(rsp), flags); |
48a7639c PM |
2990 | if (needwake) |
2991 | rcu_gp_kthread_wake(rsp); | |
dc35c893 PM |
2992 | } else { |
2993 | local_irq_restore(flags); | |
64db4cff PM |
2994 | } |
2995 | ||
2996 | /* If there are callbacks ready, invoke them. */ | |
09223371 | 2997 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
a46e0899 | 2998 | invoke_rcu_callbacks(rsp, rdp); |
96d3fd0d PM |
2999 | |
3000 | /* Do any needed deferred wakeups of rcuo kthreads. */ | |
3001 | do_nocb_deferred_wakeup(rdp); | |
09223371 SL |
3002 | } |
3003 | ||
64db4cff | 3004 | /* |
e0f23060 | 3005 | * Do RCU core processing for the current CPU. |
64db4cff | 3006 | */ |
09223371 | 3007 | static void rcu_process_callbacks(struct softirq_action *unused) |
64db4cff | 3008 | { |
6ce75a23 PM |
3009 | struct rcu_state *rsp; |
3010 | ||
bfa00b4c PM |
3011 | if (cpu_is_offline(smp_processor_id())) |
3012 | return; | |
f7f7bac9 | 3013 | trace_rcu_utilization(TPS("Start RCU core")); |
6ce75a23 PM |
3014 | for_each_rcu_flavor(rsp) |
3015 | __rcu_process_callbacks(rsp); | |
f7f7bac9 | 3016 | trace_rcu_utilization(TPS("End RCU core")); |
64db4cff PM |
3017 | } |
3018 | ||
a26ac245 | 3019 | /* |
e0f23060 PM |
3020 | * Schedule RCU callback invocation. If the specified type of RCU |
3021 | * does not support RCU priority boosting, just do a direct call, | |
3022 | * otherwise wake up the per-CPU kernel kthread. Note that because we | |
924df8a0 | 3023 | * are running on the current CPU with softirqs disabled, the |
e0f23060 | 3024 | * rcu_cpu_kthread_task cannot disappear out from under us. |
a26ac245 | 3025 | */ |
a46e0899 | 3026 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) |
a26ac245 | 3027 | { |
7d0ae808 | 3028 | if (unlikely(!READ_ONCE(rcu_scheduler_fully_active))) |
b0d30417 | 3029 | return; |
a46e0899 PM |
3030 | if (likely(!rsp->boost)) { |
3031 | rcu_do_batch(rsp, rdp); | |
a26ac245 PM |
3032 | return; |
3033 | } | |
a46e0899 | 3034 | invoke_rcu_callbacks_kthread(); |
a26ac245 PM |
3035 | } |
3036 | ||
a46e0899 | 3037 | static void invoke_rcu_core(void) |
09223371 | 3038 | { |
b0f74036 PM |
3039 | if (cpu_online(smp_processor_id())) |
3040 | raise_softirq(RCU_SOFTIRQ); | |
09223371 SL |
3041 | } |
3042 | ||
29154c57 PM |
3043 | /* |
3044 | * Handle any core-RCU processing required by a call_rcu() invocation. | |
3045 | */ | |
3046 | static void __call_rcu_core(struct rcu_state *rsp, struct rcu_data *rdp, | |
3047 | struct rcu_head *head, unsigned long flags) | |
64db4cff | 3048 | { |
48a7639c PM |
3049 | bool needwake; |
3050 | ||
62fde6ed PM |
3051 | /* |
3052 | * If called from an extended quiescent state, invoke the RCU | |
3053 | * core in order to force a re-evaluation of RCU's idleness. | |
3054 | */ | |
9910affa | 3055 | if (!rcu_is_watching()) |
62fde6ed PM |
3056 | invoke_rcu_core(); |
3057 | ||
a16b7a69 | 3058 | /* If interrupts were disabled or CPU offline, don't invoke RCU core. */ |
29154c57 | 3059 | if (irqs_disabled_flags(flags) || cpu_is_offline(smp_processor_id())) |
2655d57e | 3060 | return; |
64db4cff | 3061 | |
37c72e56 PM |
3062 | /* |
3063 | * Force the grace period if too many callbacks or too long waiting. | |
3064 | * Enforce hysteresis, and don't invoke force_quiescent_state() | |
3065 | * if some other CPU has recently done so. Also, don't bother | |
3066 | * invoking force_quiescent_state() if the newly enqueued callback | |
3067 | * is the only one waiting for a grace period to complete. | |
3068 | */ | |
2655d57e | 3069 | if (unlikely(rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) { |
b52573d2 PM |
3070 | |
3071 | /* Are we ignoring a completed grace period? */ | |
470716fc | 3072 | note_gp_changes(rsp, rdp); |
b52573d2 PM |
3073 | |
3074 | /* Start a new grace period if one not already started. */ | |
3075 | if (!rcu_gp_in_progress(rsp)) { | |
b52573d2 PM |
3076 | struct rcu_node *rnp_root = rcu_get_root(rsp); |
3077 | ||
2a67e741 | 3078 | raw_spin_lock_rcu_node(rnp_root); |
48a7639c | 3079 | needwake = rcu_start_gp(rsp); |
67c583a7 | 3080 | raw_spin_unlock_rcu_node(rnp_root); |
48a7639c PM |
3081 | if (needwake) |
3082 | rcu_gp_kthread_wake(rsp); | |
b52573d2 PM |
3083 | } else { |
3084 | /* Give the grace period a kick. */ | |
3085 | rdp->blimit = LONG_MAX; | |
3086 | if (rsp->n_force_qs == rdp->n_force_qs_snap && | |
3087 | *rdp->nxttail[RCU_DONE_TAIL] != head) | |
4cdfc175 | 3088 | force_quiescent_state(rsp); |
b52573d2 PM |
3089 | rdp->n_force_qs_snap = rsp->n_force_qs; |
3090 | rdp->qlen_last_fqs_check = rdp->qlen; | |
3091 | } | |
4cdfc175 | 3092 | } |
29154c57 PM |
3093 | } |
3094 | ||
ae150184 PM |
3095 | /* |
3096 | * RCU callback function to leak a callback. | |
3097 | */ | |
3098 | static void rcu_leak_callback(struct rcu_head *rhp) | |
3099 | { | |
3100 | } | |
3101 | ||
3fbfbf7a PM |
3102 | /* |
3103 | * Helper function for call_rcu() and friends. The cpu argument will | |
3104 | * normally be -1, indicating "currently running CPU". It may specify | |
3105 | * a CPU only if that CPU is a no-CBs CPU. Currently, only _rcu_barrier() | |
3106 | * is expected to specify a CPU. | |
3107 | */ | |
64db4cff | 3108 | static void |
b6a4ae76 | 3109 | __call_rcu(struct rcu_head *head, rcu_callback_t func, |
3fbfbf7a | 3110 | struct rcu_state *rsp, int cpu, bool lazy) |
64db4cff PM |
3111 | { |
3112 | unsigned long flags; | |
3113 | struct rcu_data *rdp; | |
3114 | ||
1146edcb | 3115 | WARN_ON_ONCE((unsigned long)head & 0x1); /* Misaligned rcu_head! */ |
ae150184 PM |
3116 | if (debug_rcu_head_queue(head)) { |
3117 | /* Probable double call_rcu(), so leak the callback. */ | |
7d0ae808 | 3118 | WRITE_ONCE(head->func, rcu_leak_callback); |
ae150184 PM |
3119 | WARN_ONCE(1, "__call_rcu(): Leaked duplicate callback\n"); |
3120 | return; | |
3121 | } | |
64db4cff PM |
3122 | head->func = func; |
3123 | head->next = NULL; | |
3124 | ||
64db4cff PM |
3125 | /* |
3126 | * Opportunistically note grace-period endings and beginnings. | |
3127 | * Note that we might see a beginning right after we see an | |
3128 | * end, but never vice versa, since this CPU has to pass through | |
3129 | * a quiescent state betweentimes. | |
3130 | */ | |
3131 | local_irq_save(flags); | |
394f99a9 | 3132 | rdp = this_cpu_ptr(rsp->rda); |
64db4cff PM |
3133 | |
3134 | /* Add the callback to our list. */ | |
3fbfbf7a PM |
3135 | if (unlikely(rdp->nxttail[RCU_NEXT_TAIL] == NULL) || cpu != -1) { |
3136 | int offline; | |
3137 | ||
3138 | if (cpu != -1) | |
3139 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
143da9c2 PM |
3140 | if (likely(rdp->mynode)) { |
3141 | /* Post-boot, so this should be for a no-CBs CPU. */ | |
3142 | offline = !__call_rcu_nocb(rdp, head, lazy, flags); | |
3143 | WARN_ON_ONCE(offline); | |
3144 | /* Offline CPU, _call_rcu() illegal, leak callback. */ | |
3145 | local_irq_restore(flags); | |
3146 | return; | |
3147 | } | |
3148 | /* | |
3149 | * Very early boot, before rcu_init(). Initialize if needed | |
3150 | * and then drop through to queue the callback. | |
3151 | */ | |
3152 | BUG_ON(cpu != -1); | |
34404ca8 | 3153 | WARN_ON_ONCE(!rcu_is_watching()); |
143da9c2 PM |
3154 | if (!likely(rdp->nxtlist)) |
3155 | init_default_callback_list(rdp); | |
0d8ee37e | 3156 | } |
7d0ae808 | 3157 | WRITE_ONCE(rdp->qlen, rdp->qlen + 1); |
486e2593 PM |
3158 | if (lazy) |
3159 | rdp->qlen_lazy++; | |
c57afe80 PM |
3160 | else |
3161 | rcu_idle_count_callbacks_posted(); | |
b1420f1c PM |
3162 | smp_mb(); /* Count before adding callback for rcu_barrier(). */ |
3163 | *rdp->nxttail[RCU_NEXT_TAIL] = head; | |
3164 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; | |
2655d57e | 3165 | |
d4c08f2a PM |
3166 | if (__is_kfree_rcu_offset((unsigned long)func)) |
3167 | trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func, | |
486e2593 | 3168 | rdp->qlen_lazy, rdp->qlen); |
d4c08f2a | 3169 | else |
486e2593 | 3170 | trace_rcu_callback(rsp->name, head, rdp->qlen_lazy, rdp->qlen); |
d4c08f2a | 3171 | |
29154c57 PM |
3172 | /* Go handle any RCU core processing required. */ |
3173 | __call_rcu_core(rsp, rdp, head, flags); | |
64db4cff PM |
3174 | local_irq_restore(flags); |
3175 | } | |
3176 | ||
3177 | /* | |
d6714c22 | 3178 | * Queue an RCU-sched callback for invocation after a grace period. |
64db4cff | 3179 | */ |
b6a4ae76 | 3180 | void call_rcu_sched(struct rcu_head *head, rcu_callback_t func) |
64db4cff | 3181 | { |
3fbfbf7a | 3182 | __call_rcu(head, func, &rcu_sched_state, -1, 0); |
64db4cff | 3183 | } |
d6714c22 | 3184 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
64db4cff PM |
3185 | |
3186 | /* | |
486e2593 | 3187 | * Queue an RCU callback for invocation after a quicker grace period. |
64db4cff | 3188 | */ |
b6a4ae76 | 3189 | void call_rcu_bh(struct rcu_head *head, rcu_callback_t func) |
64db4cff | 3190 | { |
3fbfbf7a | 3191 | __call_rcu(head, func, &rcu_bh_state, -1, 0); |
64db4cff PM |
3192 | } |
3193 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
3194 | ||
495aa969 ACB |
3195 | /* |
3196 | * Queue an RCU callback for lazy invocation after a grace period. | |
3197 | * This will likely be later named something like "call_rcu_lazy()", | |
3198 | * but this change will require some way of tagging the lazy RCU | |
3199 | * callbacks in the list of pending callbacks. Until then, this | |
3200 | * function may only be called from __kfree_rcu(). | |
3201 | */ | |
3202 | void kfree_call_rcu(struct rcu_head *head, | |
b6a4ae76 | 3203 | rcu_callback_t func) |
495aa969 | 3204 | { |
e534165b | 3205 | __call_rcu(head, func, rcu_state_p, -1, 1); |
495aa969 ACB |
3206 | } |
3207 | EXPORT_SYMBOL_GPL(kfree_call_rcu); | |
3208 | ||
6d813391 PM |
3209 | /* |
3210 | * Because a context switch is a grace period for RCU-sched and RCU-bh, | |
3211 | * any blocking grace-period wait automatically implies a grace period | |
3212 | * if there is only one CPU online at any point time during execution | |
3213 | * of either synchronize_sched() or synchronize_rcu_bh(). It is OK to | |
3214 | * occasionally incorrectly indicate that there are multiple CPUs online | |
3215 | * when there was in fact only one the whole time, as this just adds | |
3216 | * some overhead: RCU still operates correctly. | |
6d813391 PM |
3217 | */ |
3218 | static inline int rcu_blocking_is_gp(void) | |
3219 | { | |
95f0c1de PM |
3220 | int ret; |
3221 | ||
6d813391 | 3222 | might_sleep(); /* Check for RCU read-side critical section. */ |
95f0c1de PM |
3223 | preempt_disable(); |
3224 | ret = num_online_cpus() <= 1; | |
3225 | preempt_enable(); | |
3226 | return ret; | |
6d813391 PM |
3227 | } |
3228 | ||
6ebb237b PM |
3229 | /** |
3230 | * synchronize_sched - wait until an rcu-sched grace period has elapsed. | |
3231 | * | |
3232 | * Control will return to the caller some time after a full rcu-sched | |
3233 | * grace period has elapsed, in other words after all currently executing | |
3234 | * rcu-sched read-side critical sections have completed. These read-side | |
3235 | * critical sections are delimited by rcu_read_lock_sched() and | |
3236 | * rcu_read_unlock_sched(), and may be nested. Note that preempt_disable(), | |
3237 | * local_irq_disable(), and so on may be used in place of | |
3238 | * rcu_read_lock_sched(). | |
3239 | * | |
3240 | * This means that all preempt_disable code sequences, including NMI and | |
f0a0e6f2 PM |
3241 | * non-threaded hardware-interrupt handlers, in progress on entry will |
3242 | * have completed before this primitive returns. However, this does not | |
3243 | * guarantee that softirq handlers will have completed, since in some | |
3244 | * kernels, these handlers can run in process context, and can block. | |
3245 | * | |
3246 | * Note that this guarantee implies further memory-ordering guarantees. | |
3247 | * On systems with more than one CPU, when synchronize_sched() returns, | |
3248 | * each CPU is guaranteed to have executed a full memory barrier since the | |
3249 | * end of its last RCU-sched read-side critical section whose beginning | |
3250 | * preceded the call to synchronize_sched(). In addition, each CPU having | |
3251 | * an RCU read-side critical section that extends beyond the return from | |
3252 | * synchronize_sched() is guaranteed to have executed a full memory barrier | |
3253 | * after the beginning of synchronize_sched() and before the beginning of | |
3254 | * that RCU read-side critical section. Note that these guarantees include | |
3255 | * CPUs that are offline, idle, or executing in user mode, as well as CPUs | |
3256 | * that are executing in the kernel. | |
3257 | * | |
3258 | * Furthermore, if CPU A invoked synchronize_sched(), which returned | |
3259 | * to its caller on CPU B, then both CPU A and CPU B are guaranteed | |
3260 | * to have executed a full memory barrier during the execution of | |
3261 | * synchronize_sched() -- even if CPU A and CPU B are the same CPU (but | |
3262 | * again only if the system has more than one CPU). | |
6ebb237b PM |
3263 | * |
3264 | * This primitive provides the guarantees made by the (now removed) | |
3265 | * synchronize_kernel() API. In contrast, synchronize_rcu() only | |
3266 | * guarantees that rcu_read_lock() sections will have completed. | |
3267 | * In "classic RCU", these two guarantees happen to be one and | |
3268 | * the same, but can differ in realtime RCU implementations. | |
3269 | */ | |
3270 | void synchronize_sched(void) | |
3271 | { | |
f78f5b90 PM |
3272 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) || |
3273 | lock_is_held(&rcu_lock_map) || | |
3274 | lock_is_held(&rcu_sched_lock_map), | |
3275 | "Illegal synchronize_sched() in RCU-sched read-side critical section"); | |
6ebb237b PM |
3276 | if (rcu_blocking_is_gp()) |
3277 | return; | |
5afff48b | 3278 | if (rcu_gp_is_expedited()) |
3705b88d AM |
3279 | synchronize_sched_expedited(); |
3280 | else | |
3281 | wait_rcu_gp(call_rcu_sched); | |
6ebb237b PM |
3282 | } |
3283 | EXPORT_SYMBOL_GPL(synchronize_sched); | |
3284 | ||
3285 | /** | |
3286 | * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed. | |
3287 | * | |
3288 | * Control will return to the caller some time after a full rcu_bh grace | |
3289 | * period has elapsed, in other words after all currently executing rcu_bh | |
3290 | * read-side critical sections have completed. RCU read-side critical | |
3291 | * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(), | |
3292 | * and may be nested. | |
f0a0e6f2 PM |
3293 | * |
3294 | * See the description of synchronize_sched() for more detailed information | |
3295 | * on memory ordering guarantees. | |
6ebb237b PM |
3296 | */ |
3297 | void synchronize_rcu_bh(void) | |
3298 | { | |
f78f5b90 PM |
3299 | RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) || |
3300 | lock_is_held(&rcu_lock_map) || | |
3301 | lock_is_held(&rcu_sched_lock_map), | |
3302 | "Illegal synchronize_rcu_bh() in RCU-bh read-side critical section"); | |
6ebb237b PM |
3303 | if (rcu_blocking_is_gp()) |
3304 | return; | |
5afff48b | 3305 | if (rcu_gp_is_expedited()) |
3705b88d AM |
3306 | synchronize_rcu_bh_expedited(); |
3307 | else | |
3308 | wait_rcu_gp(call_rcu_bh); | |
6ebb237b PM |
3309 | } |
3310 | EXPORT_SYMBOL_GPL(synchronize_rcu_bh); | |
3311 | ||
765a3f4f PM |
3312 | /** |
3313 | * get_state_synchronize_rcu - Snapshot current RCU state | |
3314 | * | |
3315 | * Returns a cookie that is used by a later call to cond_synchronize_rcu() | |
3316 | * to determine whether or not a full grace period has elapsed in the | |
3317 | * meantime. | |
3318 | */ | |
3319 | unsigned long get_state_synchronize_rcu(void) | |
3320 | { | |
3321 | /* | |
3322 | * Any prior manipulation of RCU-protected data must happen | |
3323 | * before the load from ->gpnum. | |
3324 | */ | |
3325 | smp_mb(); /* ^^^ */ | |
3326 | ||
3327 | /* | |
3328 | * Make sure this load happens before the purportedly | |
3329 | * time-consuming work between get_state_synchronize_rcu() | |
3330 | * and cond_synchronize_rcu(). | |
3331 | */ | |
e534165b | 3332 | return smp_load_acquire(&rcu_state_p->gpnum); |
765a3f4f PM |
3333 | } |
3334 | EXPORT_SYMBOL_GPL(get_state_synchronize_rcu); | |
3335 | ||
3336 | /** | |
3337 | * cond_synchronize_rcu - Conditionally wait for an RCU grace period | |
3338 | * | |
3339 | * @oldstate: return value from earlier call to get_state_synchronize_rcu() | |
3340 | * | |
3341 | * If a full RCU grace period has elapsed since the earlier call to | |
3342 | * get_state_synchronize_rcu(), just return. Otherwise, invoke | |
3343 | * synchronize_rcu() to wait for a full grace period. | |
3344 | * | |
3345 | * Yes, this function does not take counter wrap into account. But | |
3346 | * counter wrap is harmless. If the counter wraps, we have waited for | |
3347 | * more than 2 billion grace periods (and way more on a 64-bit system!), | |
3348 | * so waiting for one additional grace period should be just fine. | |
3349 | */ | |
3350 | void cond_synchronize_rcu(unsigned long oldstate) | |
3351 | { | |
3352 | unsigned long newstate; | |
3353 | ||
3354 | /* | |
3355 | * Ensure that this load happens before any RCU-destructive | |
3356 | * actions the caller might carry out after we return. | |
3357 | */ | |
e534165b | 3358 | newstate = smp_load_acquire(&rcu_state_p->completed); |
765a3f4f PM |
3359 | if (ULONG_CMP_GE(oldstate, newstate)) |
3360 | synchronize_rcu(); | |
3361 | } | |
3362 | EXPORT_SYMBOL_GPL(cond_synchronize_rcu); | |
3363 | ||
24560056 PM |
3364 | /** |
3365 | * get_state_synchronize_sched - Snapshot current RCU-sched state | |
3366 | * | |
3367 | * Returns a cookie that is used by a later call to cond_synchronize_sched() | |
3368 | * to determine whether or not a full grace period has elapsed in the | |
3369 | * meantime. | |
3370 | */ | |
3371 | unsigned long get_state_synchronize_sched(void) | |
3372 | { | |
3373 | /* | |
3374 | * Any prior manipulation of RCU-protected data must happen | |
3375 | * before the load from ->gpnum. | |
3376 | */ | |
3377 | smp_mb(); /* ^^^ */ | |
3378 | ||
3379 | /* | |
3380 | * Make sure this load happens before the purportedly | |
3381 | * time-consuming work between get_state_synchronize_sched() | |
3382 | * and cond_synchronize_sched(). | |
3383 | */ | |
3384 | return smp_load_acquire(&rcu_sched_state.gpnum); | |
3385 | } | |
3386 | EXPORT_SYMBOL_GPL(get_state_synchronize_sched); | |
3387 | ||
3388 | /** | |
3389 | * cond_synchronize_sched - Conditionally wait for an RCU-sched grace period | |
3390 | * | |
3391 | * @oldstate: return value from earlier call to get_state_synchronize_sched() | |
3392 | * | |
3393 | * If a full RCU-sched grace period has elapsed since the earlier call to | |
3394 | * get_state_synchronize_sched(), just return. Otherwise, invoke | |
3395 | * synchronize_sched() to wait for a full grace period. | |
3396 | * | |
3397 | * Yes, this function does not take counter wrap into account. But | |
3398 | * counter wrap is harmless. If the counter wraps, we have waited for | |
3399 | * more than 2 billion grace periods (and way more on a 64-bit system!), | |
3400 | * so waiting for one additional grace period should be just fine. | |
3401 | */ | |
3402 | void cond_synchronize_sched(unsigned long oldstate) | |
3403 | { | |
3404 | unsigned long newstate; | |
3405 | ||
3406 | /* | |
3407 | * Ensure that this load happens before any RCU-destructive | |
3408 | * actions the caller might carry out after we return. | |
3409 | */ | |
3410 | newstate = smp_load_acquire(&rcu_sched_state.completed); | |
3411 | if (ULONG_CMP_GE(oldstate, newstate)) | |
3412 | synchronize_sched(); | |
3413 | } | |
3414 | EXPORT_SYMBOL_GPL(cond_synchronize_sched); | |
3415 | ||
28f00767 PM |
3416 | /* Adjust sequence number for start of update-side operation. */ |
3417 | static void rcu_seq_start(unsigned long *sp) | |
3418 | { | |
3419 | WRITE_ONCE(*sp, *sp + 1); | |
3420 | smp_mb(); /* Ensure update-side operation after counter increment. */ | |
3421 | WARN_ON_ONCE(!(*sp & 0x1)); | |
3422 | } | |
3423 | ||
3424 | /* Adjust sequence number for end of update-side operation. */ | |
3425 | static void rcu_seq_end(unsigned long *sp) | |
3426 | { | |
3427 | smp_mb(); /* Ensure update-side operation before counter increment. */ | |
3428 | WRITE_ONCE(*sp, *sp + 1); | |
3429 | WARN_ON_ONCE(*sp & 0x1); | |
3430 | } | |
3431 | ||
3432 | /* Take a snapshot of the update side's sequence number. */ | |
3433 | static unsigned long rcu_seq_snap(unsigned long *sp) | |
3434 | { | |
3435 | unsigned long s; | |
3436 | ||
28f00767 PM |
3437 | s = (READ_ONCE(*sp) + 3) & ~0x1; |
3438 | smp_mb(); /* Above access must not bleed into critical section. */ | |
3439 | return s; | |
3440 | } | |
3441 | ||
3442 | /* | |
3443 | * Given a snapshot from rcu_seq_snap(), determine whether or not a | |
3444 | * full update-side operation has occurred. | |
3445 | */ | |
3446 | static bool rcu_seq_done(unsigned long *sp, unsigned long s) | |
3447 | { | |
3448 | return ULONG_CMP_GE(READ_ONCE(*sp), s); | |
3449 | } | |
3450 | ||
3451 | /* Wrapper functions for expedited grace periods. */ | |
3452 | static void rcu_exp_gp_seq_start(struct rcu_state *rsp) | |
3453 | { | |
3454 | rcu_seq_start(&rsp->expedited_sequence); | |
3455 | } | |
3456 | static void rcu_exp_gp_seq_end(struct rcu_state *rsp) | |
3457 | { | |
3458 | rcu_seq_end(&rsp->expedited_sequence); | |
704dd435 | 3459 | smp_mb(); /* Ensure that consecutive grace periods serialize. */ |
28f00767 PM |
3460 | } |
3461 | static unsigned long rcu_exp_gp_seq_snap(struct rcu_state *rsp) | |
3462 | { | |
179e5dcd PM |
3463 | unsigned long s; |
3464 | ||
886ef5a1 | 3465 | smp_mb(); /* Caller's modifications seen first by other CPUs. */ |
179e5dcd PM |
3466 | s = rcu_seq_snap(&rsp->expedited_sequence); |
3467 | trace_rcu_exp_grace_period(rsp->name, s, TPS("snap")); | |
3468 | return s; | |
28f00767 PM |
3469 | } |
3470 | static bool rcu_exp_gp_seq_done(struct rcu_state *rsp, unsigned long s) | |
3471 | { | |
3472 | return rcu_seq_done(&rsp->expedited_sequence, s); | |
3473 | } | |
3474 | ||
b9585e94 PM |
3475 | /* |
3476 | * Reset the ->expmaskinit values in the rcu_node tree to reflect any | |
3477 | * recent CPU-online activity. Note that these masks are not cleared | |
3478 | * when CPUs go offline, so they reflect the union of all CPUs that have | |
3479 | * ever been online. This means that this function normally takes its | |
3480 | * no-work-to-do fastpath. | |
3481 | */ | |
3482 | static void sync_exp_reset_tree_hotplug(struct rcu_state *rsp) | |
3483 | { | |
3484 | bool done; | |
3485 | unsigned long flags; | |
3486 | unsigned long mask; | |
3487 | unsigned long oldmask; | |
3488 | int ncpus = READ_ONCE(rsp->ncpus); | |
3489 | struct rcu_node *rnp; | |
3490 | struct rcu_node *rnp_up; | |
3491 | ||
3492 | /* If no new CPUs onlined since last time, nothing to do. */ | |
3493 | if (likely(ncpus == rsp->ncpus_snap)) | |
3494 | return; | |
3495 | rsp->ncpus_snap = ncpus; | |
3496 | ||
3497 | /* | |
3498 | * Each pass through the following loop propagates newly onlined | |
3499 | * CPUs for the current rcu_node structure up the rcu_node tree. | |
3500 | */ | |
3501 | rcu_for_each_leaf_node(rsp, rnp) { | |
2a67e741 | 3502 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
b9585e94 | 3503 | if (rnp->expmaskinit == rnp->expmaskinitnext) { |
67c583a7 | 3504 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
b9585e94 PM |
3505 | continue; /* No new CPUs, nothing to do. */ |
3506 | } | |
3507 | ||
3508 | /* Update this node's mask, track old value for propagation. */ | |
3509 | oldmask = rnp->expmaskinit; | |
3510 | rnp->expmaskinit = rnp->expmaskinitnext; | |
67c583a7 | 3511 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
b9585e94 PM |
3512 | |
3513 | /* If was already nonzero, nothing to propagate. */ | |
3514 | if (oldmask) | |
3515 | continue; | |
3516 | ||
3517 | /* Propagate the new CPU up the tree. */ | |
3518 | mask = rnp->grpmask; | |
3519 | rnp_up = rnp->parent; | |
3520 | done = false; | |
3521 | while (rnp_up) { | |
2a67e741 | 3522 | raw_spin_lock_irqsave_rcu_node(rnp_up, flags); |
b9585e94 PM |
3523 | if (rnp_up->expmaskinit) |
3524 | done = true; | |
3525 | rnp_up->expmaskinit |= mask; | |
67c583a7 | 3526 | raw_spin_unlock_irqrestore_rcu_node(rnp_up, flags); |
b9585e94 PM |
3527 | if (done) |
3528 | break; | |
3529 | mask = rnp_up->grpmask; | |
3530 | rnp_up = rnp_up->parent; | |
3531 | } | |
3532 | } | |
3533 | } | |
3534 | ||
3535 | /* | |
3536 | * Reset the ->expmask values in the rcu_node tree in preparation for | |
3537 | * a new expedited grace period. | |
3538 | */ | |
3539 | static void __maybe_unused sync_exp_reset_tree(struct rcu_state *rsp) | |
3540 | { | |
3541 | unsigned long flags; | |
3542 | struct rcu_node *rnp; | |
3543 | ||
3544 | sync_exp_reset_tree_hotplug(rsp); | |
3545 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
2a67e741 | 3546 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
b9585e94 PM |
3547 | WARN_ON_ONCE(rnp->expmask); |
3548 | rnp->expmask = rnp->expmaskinit; | |
67c583a7 | 3549 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
b9585e94 PM |
3550 | } |
3551 | } | |
3552 | ||
7922cd0e | 3553 | /* |
8203d6d0 | 3554 | * Return non-zero if there is no RCU expedited grace period in progress |
7922cd0e PM |
3555 | * for the specified rcu_node structure, in other words, if all CPUs and |
3556 | * tasks covered by the specified rcu_node structure have done their bit | |
3557 | * for the current expedited grace period. Works only for preemptible | |
3558 | * RCU -- other RCU implementation use other means. | |
3559 | * | |
f6a12f34 | 3560 | * Caller must hold the rcu_state's exp_mutex. |
7922cd0e PM |
3561 | */ |
3562 | static int sync_rcu_preempt_exp_done(struct rcu_node *rnp) | |
3563 | { | |
8203d6d0 | 3564 | return rnp->exp_tasks == NULL && |
7922cd0e PM |
3565 | READ_ONCE(rnp->expmask) == 0; |
3566 | } | |
3567 | ||
3568 | /* | |
3569 | * Report the exit from RCU read-side critical section for the last task | |
3570 | * that queued itself during or before the current expedited preemptible-RCU | |
3571 | * grace period. This event is reported either to the rcu_node structure on | |
3572 | * which the task was queued or to one of that rcu_node structure's ancestors, | |
3573 | * recursively up the tree. (Calm down, calm down, we do the recursion | |
3574 | * iteratively!) | |
3575 | * | |
f6a12f34 PM |
3576 | * Caller must hold the rcu_state's exp_mutex and the specified rcu_node |
3577 | * structure's ->lock. | |
7922cd0e | 3578 | */ |
8203d6d0 PM |
3579 | static void __rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp, |
3580 | bool wake, unsigned long flags) | |
3581 | __releases(rnp->lock) | |
7922cd0e | 3582 | { |
7922cd0e PM |
3583 | unsigned long mask; |
3584 | ||
7922cd0e PM |
3585 | for (;;) { |
3586 | if (!sync_rcu_preempt_exp_done(rnp)) { | |
8203d6d0 PM |
3587 | if (!rnp->expmask) |
3588 | rcu_initiate_boost(rnp, flags); | |
3589 | else | |
67c583a7 | 3590 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
7922cd0e PM |
3591 | break; |
3592 | } | |
3593 | if (rnp->parent == NULL) { | |
67c583a7 | 3594 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
7922cd0e PM |
3595 | if (wake) { |
3596 | smp_mb(); /* EGP done before wake_up(). */ | |
abedf8e2 | 3597 | swake_up(&rsp->expedited_wq); |
7922cd0e PM |
3598 | } |
3599 | break; | |
3600 | } | |
3601 | mask = rnp->grpmask; | |
67c583a7 | 3602 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled */ |
7922cd0e | 3603 | rnp = rnp->parent; |
2a67e741 | 3604 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled */ |
8203d6d0 | 3605 | WARN_ON_ONCE(!(rnp->expmask & mask)); |
7922cd0e PM |
3606 | rnp->expmask &= ~mask; |
3607 | } | |
3608 | } | |
3609 | ||
8203d6d0 PM |
3610 | /* |
3611 | * Report expedited quiescent state for specified node. This is a | |
3612 | * lock-acquisition wrapper function for __rcu_report_exp_rnp(). | |
3613 | * | |
f6a12f34 | 3614 | * Caller must hold the rcu_state's exp_mutex. |
8203d6d0 PM |
3615 | */ |
3616 | static void __maybe_unused rcu_report_exp_rnp(struct rcu_state *rsp, | |
3617 | struct rcu_node *rnp, bool wake) | |
3618 | { | |
3619 | unsigned long flags; | |
3620 | ||
2a67e741 | 3621 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
8203d6d0 PM |
3622 | __rcu_report_exp_rnp(rsp, rnp, wake, flags); |
3623 | } | |
3624 | ||
3625 | /* | |
3626 | * Report expedited quiescent state for multiple CPUs, all covered by the | |
f6a12f34 PM |
3627 | * specified leaf rcu_node structure. Caller must hold the rcu_state's |
3628 | * exp_mutex. | |
8203d6d0 PM |
3629 | */ |
3630 | static void rcu_report_exp_cpu_mult(struct rcu_state *rsp, struct rcu_node *rnp, | |
3631 | unsigned long mask, bool wake) | |
3632 | { | |
3633 | unsigned long flags; | |
3634 | ||
2a67e741 | 3635 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
338b0f76 | 3636 | if (!(rnp->expmask & mask)) { |
67c583a7 | 3637 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
338b0f76 PM |
3638 | return; |
3639 | } | |
8203d6d0 PM |
3640 | rnp->expmask &= ~mask; |
3641 | __rcu_report_exp_rnp(rsp, rnp, wake, flags); /* Releases rnp->lock. */ | |
3642 | } | |
3643 | ||
3644 | /* | |
3645 | * Report expedited quiescent state for specified rcu_data (CPU). | |
8203d6d0 | 3646 | */ |
6587a23b PM |
3647 | static void rcu_report_exp_rdp(struct rcu_state *rsp, struct rcu_data *rdp, |
3648 | bool wake) | |
8203d6d0 PM |
3649 | { |
3650 | rcu_report_exp_cpu_mult(rsp, rdp->mynode, rdp->grpmask, wake); | |
3651 | } | |
3652 | ||
29fd9309 | 3653 | /* Common code for synchronize_{rcu,sched}_expedited() work-done checking. */ |
f6a12f34 PM |
3654 | static bool sync_exp_work_done(struct rcu_state *rsp, atomic_long_t *stat, |
3655 | unsigned long s) | |
3d3b7db0 | 3656 | { |
28f00767 | 3657 | if (rcu_exp_gp_seq_done(rsp, s)) { |
4f415302 | 3658 | trace_rcu_exp_grace_period(rsp->name, s, TPS("done")); |
385b73c0 PM |
3659 | /* Ensure test happens before caller kfree(). */ |
3660 | smp_mb__before_atomic(); /* ^^^ */ | |
3661 | atomic_long_inc(stat); | |
385b73c0 PM |
3662 | return true; |
3663 | } | |
3664 | return false; | |
3665 | } | |
3666 | ||
b09e5f86 | 3667 | /* |
f6a12f34 PM |
3668 | * Funnel-lock acquisition for expedited grace periods. Returns true |
3669 | * if some other task completed an expedited grace period that this task | |
3670 | * can piggy-back on, and with no mutex held. Otherwise, returns false | |
3671 | * with the mutex held, indicating that the caller must actually do the | |
3672 | * expedited grace period. | |
b09e5f86 | 3673 | */ |
f6a12f34 | 3674 | static bool exp_funnel_lock(struct rcu_state *rsp, unsigned long s) |
b09e5f86 | 3675 | { |
df5bd514 | 3676 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, raw_smp_processor_id()); |
f6a12f34 | 3677 | struct rcu_node *rnp = rdp->mynode; |
356051e1 PM |
3678 | struct rcu_node *rnp_root = rcu_get_root(rsp); |
3679 | ||
3680 | /* Low-contention fastpath. */ | |
3681 | if (ULONG_CMP_LT(READ_ONCE(rnp->exp_seq_rq), s) && | |
3682 | (rnp == rnp_root || | |
3683 | ULONG_CMP_LT(READ_ONCE(rnp_root->exp_seq_rq), s)) && | |
3684 | !mutex_is_locked(&rsp->exp_mutex) && | |
3685 | mutex_trylock(&rsp->exp_mutex)) | |
3686 | goto fastpath; | |
b09e5f86 | 3687 | |
3d3b7db0 | 3688 | /* |
f6a12f34 PM |
3689 | * Each pass through the following loop works its way up |
3690 | * the rcu_node tree, returning if others have done the work or | |
3691 | * otherwise falls through to acquire rsp->exp_mutex. The mapping | |
3692 | * from CPU to rcu_node structure can be inexact, as it is just | |
3693 | * promoting locality and is not strictly needed for correctness. | |
3d3b7db0 | 3694 | */ |
f6a12f34 PM |
3695 | for (; rnp != NULL; rnp = rnp->parent) { |
3696 | if (sync_exp_work_done(rsp, &rdp->exp_workdone1, s)) | |
3697 | return true; | |
3698 | ||
3699 | /* Work not done, either wait here or go up. */ | |
3700 | spin_lock(&rnp->exp_lock); | |
3701 | if (ULONG_CMP_GE(rnp->exp_seq_rq, s)) { | |
3702 | ||
3703 | /* Someone else doing GP, so wait for them. */ | |
3704 | spin_unlock(&rnp->exp_lock); | |
3705 | trace_rcu_exp_funnel_lock(rsp->name, rnp->level, | |
3706 | rnp->grplo, rnp->grphi, | |
3707 | TPS("wait")); | |
3b5f668e | 3708 | wait_event(rnp->exp_wq[(s >> 1) & 0x3], |
f6a12f34 PM |
3709 | sync_exp_work_done(rsp, |
3710 | &rdp->exp_workdone2, s)); | |
3711 | return true; | |
cdacbe1f | 3712 | } |
f6a12f34 PM |
3713 | rnp->exp_seq_rq = s; /* Followers can wait on us. */ |
3714 | spin_unlock(&rnp->exp_lock); | |
3715 | trace_rcu_exp_funnel_lock(rsp->name, rnp->level, rnp->grplo, | |
3716 | rnp->grphi, TPS("nxtlvl")); | |
cdacbe1f | 3717 | } |
f6a12f34 | 3718 | mutex_lock(&rsp->exp_mutex); |
356051e1 | 3719 | fastpath: |
f6a12f34 PM |
3720 | if (sync_exp_work_done(rsp, &rdp->exp_workdone3, s)) { |
3721 | mutex_unlock(&rsp->exp_mutex); | |
3722 | return true; | |
b09e5f86 | 3723 | } |
aff12cdf PM |
3724 | rcu_exp_gp_seq_start(rsp); |
3725 | trace_rcu_exp_grace_period(rsp->name, s, TPS("start")); | |
f6a12f34 | 3726 | return false; |
b09e5f86 PM |
3727 | } |
3728 | ||
cf3620a6 | 3729 | /* Invoked on each online non-idle CPU for expedited quiescent state. */ |
338b0f76 | 3730 | static void sync_sched_exp_handler(void *data) |
b09e5f86 | 3731 | { |
338b0f76 PM |
3732 | struct rcu_data *rdp; |
3733 | struct rcu_node *rnp; | |
3734 | struct rcu_state *rsp = data; | |
b09e5f86 | 3735 | |
338b0f76 PM |
3736 | rdp = this_cpu_ptr(rsp->rda); |
3737 | rnp = rdp->mynode; | |
3738 | if (!(READ_ONCE(rnp->expmask) & rdp->grpmask) || | |
3739 | __this_cpu_read(rcu_sched_data.cpu_no_qs.b.exp)) | |
3740 | return; | |
28728dd3 PM |
3741 | if (rcu_is_cpu_rrupt_from_idle()) { |
3742 | rcu_report_exp_rdp(&rcu_sched_state, | |
3743 | this_cpu_ptr(&rcu_sched_data), true); | |
3744 | return; | |
3745 | } | |
6587a23b PM |
3746 | __this_cpu_write(rcu_sched_data.cpu_no_qs.b.exp, true); |
3747 | resched_cpu(smp_processor_id()); | |
3d3b7db0 PM |
3748 | } |
3749 | ||
338b0f76 PM |
3750 | /* Send IPI for expedited cleanup if needed at end of CPU-hotplug operation. */ |
3751 | static void sync_sched_exp_online_cleanup(int cpu) | |
3752 | { | |
3753 | struct rcu_data *rdp; | |
3754 | int ret; | |
3755 | struct rcu_node *rnp; | |
3756 | struct rcu_state *rsp = &rcu_sched_state; | |
3757 | ||
3758 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
3759 | rnp = rdp->mynode; | |
3760 | if (!(READ_ONCE(rnp->expmask) & rdp->grpmask)) | |
3761 | return; | |
3762 | ret = smp_call_function_single(cpu, sync_sched_exp_handler, rsp, 0); | |
3763 | WARN_ON_ONCE(ret); | |
3764 | } | |
3765 | ||
bce5fa12 PM |
3766 | /* |
3767 | * Select the nodes that the upcoming expedited grace period needs | |
3768 | * to wait for. | |
3769 | */ | |
dcdb8807 PM |
3770 | static void sync_rcu_exp_select_cpus(struct rcu_state *rsp, |
3771 | smp_call_func_t func) | |
bce5fa12 PM |
3772 | { |
3773 | int cpu; | |
3774 | unsigned long flags; | |
3775 | unsigned long mask; | |
3776 | unsigned long mask_ofl_test; | |
3777 | unsigned long mask_ofl_ipi; | |
6587a23b | 3778 | int ret; |
bce5fa12 PM |
3779 | struct rcu_node *rnp; |
3780 | ||
3781 | sync_exp_reset_tree(rsp); | |
3782 | rcu_for_each_leaf_node(rsp, rnp) { | |
2a67e741 | 3783 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
bce5fa12 PM |
3784 | |
3785 | /* Each pass checks a CPU for identity, offline, and idle. */ | |
3786 | mask_ofl_test = 0; | |
3787 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++) { | |
3788 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); | |
3789 | struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); | |
3790 | ||
3791 | if (raw_smp_processor_id() == cpu || | |
bce5fa12 PM |
3792 | !(atomic_add_return(0, &rdtp->dynticks) & 0x1)) |
3793 | mask_ofl_test |= rdp->grpmask; | |
3794 | } | |
3795 | mask_ofl_ipi = rnp->expmask & ~mask_ofl_test; | |
3796 | ||
3797 | /* | |
3798 | * Need to wait for any blocked tasks as well. Note that | |
3799 | * additional blocking tasks will also block the expedited | |
3800 | * GP until such time as the ->expmask bits are cleared. | |
3801 | */ | |
3802 | if (rcu_preempt_has_tasks(rnp)) | |
3803 | rnp->exp_tasks = rnp->blkd_tasks.next; | |
67c583a7 | 3804 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
bce5fa12 PM |
3805 | |
3806 | /* IPI the remaining CPUs for expedited quiescent state. */ | |
3807 | mask = 1; | |
3808 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask <<= 1) { | |
3809 | if (!(mask_ofl_ipi & mask)) | |
3810 | continue; | |
338b0f76 | 3811 | retry_ipi: |
dcdb8807 | 3812 | ret = smp_call_function_single(cpu, func, rsp, 0); |
338b0f76 | 3813 | if (!ret) { |
6587a23b | 3814 | mask_ofl_ipi &= ~mask; |
1307f214 PM |
3815 | continue; |
3816 | } | |
3817 | /* Failed, raced with offline. */ | |
3818 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
3819 | if (cpu_online(cpu) && | |
3820 | (rnp->expmask & mask)) { | |
67c583a7 | 3821 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
1307f214 PM |
3822 | schedule_timeout_uninterruptible(1); |
3823 | if (cpu_online(cpu) && | |
3824 | (rnp->expmask & mask)) | |
3825 | goto retry_ipi; | |
3826 | raw_spin_lock_irqsave_rcu_node(rnp, flags); | |
338b0f76 | 3827 | } |
1307f214 PM |
3828 | if (!(rnp->expmask & mask)) |
3829 | mask_ofl_ipi &= ~mask; | |
67c583a7 | 3830 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
bce5fa12 PM |
3831 | } |
3832 | /* Report quiescent states for those that went offline. */ | |
3833 | mask_ofl_test |= mask_ofl_ipi; | |
3834 | if (mask_ofl_test) | |
3835 | rcu_report_exp_cpu_mult(rsp, rnp, mask_ofl_test, false); | |
3836 | } | |
3d3b7db0 PM |
3837 | } |
3838 | ||
cf3620a6 PM |
3839 | static void synchronize_sched_expedited_wait(struct rcu_state *rsp) |
3840 | { | |
3841 | int cpu; | |
3842 | unsigned long jiffies_stall; | |
3843 | unsigned long jiffies_start; | |
bce5fa12 | 3844 | unsigned long mask; |
72611ab9 | 3845 | int ndetected; |
bce5fa12 PM |
3846 | struct rcu_node *rnp; |
3847 | struct rcu_node *rnp_root = rcu_get_root(rsp); | |
cf3620a6 PM |
3848 | int ret; |
3849 | ||
3850 | jiffies_stall = rcu_jiffies_till_stall_check(); | |
3851 | jiffies_start = jiffies; | |
3852 | ||
3853 | for (;;) { | |
abedf8e2 | 3854 | ret = swait_event_timeout( |
cf3620a6 | 3855 | rsp->expedited_wq, |
bce5fa12 | 3856 | sync_rcu_preempt_exp_done(rnp_root), |
cf3620a6 | 3857 | jiffies_stall); |
73f36f9d | 3858 | if (ret > 0 || sync_rcu_preempt_exp_done(rnp_root)) |
cf3620a6 PM |
3859 | return; |
3860 | if (ret < 0) { | |
3861 | /* Hit a signal, disable CPU stall warnings. */ | |
abedf8e2 | 3862 | swait_event(rsp->expedited_wq, |
bce5fa12 | 3863 | sync_rcu_preempt_exp_done(rnp_root)); |
cf3620a6 PM |
3864 | return; |
3865 | } | |
c5865638 | 3866 | pr_err("INFO: %s detected expedited stalls on CPUs/tasks: {", |
cf3620a6 | 3867 | rsp->name); |
72611ab9 | 3868 | ndetected = 0; |
bce5fa12 | 3869 | rcu_for_each_leaf_node(rsp, rnp) { |
251c617c | 3870 | ndetected += rcu_print_task_exp_stall(rnp); |
bce5fa12 PM |
3871 | mask = 1; |
3872 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask <<= 1) { | |
74611ecb PM |
3873 | struct rcu_data *rdp; |
3874 | ||
bce5fa12 PM |
3875 | if (!(rnp->expmask & mask)) |
3876 | continue; | |
72611ab9 | 3877 | ndetected++; |
74611ecb PM |
3878 | rdp = per_cpu_ptr(rsp->rda, cpu); |
3879 | pr_cont(" %d-%c%c%c", cpu, | |
ec3833ed | 3880 | "O."[!!cpu_online(cpu)], |
74611ecb PM |
3881 | "o."[!!(rdp->grpmask & rnp->expmaskinit)], |
3882 | "N."[!!(rdp->grpmask & rnp->expmaskinitnext)]); | |
bce5fa12 PM |
3883 | } |
3884 | mask <<= 1; | |
cf3620a6 | 3885 | } |
72611ab9 PM |
3886 | pr_cont(" } %lu jiffies s: %lu root: %#lx/%c\n", |
3887 | jiffies - jiffies_start, rsp->expedited_sequence, | |
3888 | rnp_root->expmask, ".T"[!!rnp_root->exp_tasks]); | |
251c617c | 3889 | if (ndetected) { |
72611ab9 PM |
3890 | pr_err("blocking rcu_node structures:"); |
3891 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
3892 | if (rnp == rnp_root) | |
3893 | continue; /* printed unconditionally */ | |
3894 | if (sync_rcu_preempt_exp_done(rnp)) | |
3895 | continue; | |
3896 | pr_cont(" l=%u:%d-%d:%#lx/%c", | |
3897 | rnp->level, rnp->grplo, rnp->grphi, | |
3898 | rnp->expmask, | |
3899 | ".T"[!!rnp->exp_tasks]); | |
3900 | } | |
3901 | pr_cont("\n"); | |
3902 | } | |
bce5fa12 PM |
3903 | rcu_for_each_leaf_node(rsp, rnp) { |
3904 | mask = 1; | |
3905 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask <<= 1) { | |
3906 | if (!(rnp->expmask & mask)) | |
3907 | continue; | |
3908 | dump_cpu_task(cpu); | |
3909 | } | |
cf3620a6 PM |
3910 | } |
3911 | jiffies_stall = 3 * rcu_jiffies_till_stall_check() + 3; | |
3912 | } | |
3913 | } | |
3914 | ||
f6a12f34 | 3915 | /* |
4ea3e85b PM |
3916 | * Wait for the current expedited grace period to complete, and then |
3917 | * wake up everyone who piggybacked on the just-completed expedited | |
f6a12f34 PM |
3918 | * grace period. Also update all the ->exp_seq_rq counters as needed |
3919 | * in order to avoid counter-wrap problems. | |
3920 | */ | |
4ea3e85b | 3921 | static void rcu_exp_wait_wake(struct rcu_state *rsp, unsigned long s) |
f6a12f34 PM |
3922 | { |
3923 | struct rcu_node *rnp; | |
3924 | ||
4ea3e85b PM |
3925 | synchronize_sched_expedited_wait(rsp); |
3926 | rcu_exp_gp_seq_end(rsp); | |
3927 | trace_rcu_exp_grace_period(rsp->name, s, TPS("end")); | |
3b5f668e PM |
3928 | |
3929 | /* | |
3930 | * Switch over to wakeup mode, allowing the next GP, but -only- the | |
3931 | * next GP, to proceed. | |
3932 | */ | |
3933 | mutex_lock(&rsp->exp_wake_mutex); | |
3934 | mutex_unlock(&rsp->exp_mutex); | |
3935 | ||
f6a12f34 PM |
3936 | rcu_for_each_node_breadth_first(rsp, rnp) { |
3937 | if (ULONG_CMP_LT(READ_ONCE(rnp->exp_seq_rq), s)) { | |
3938 | spin_lock(&rnp->exp_lock); | |
3939 | /* Recheck, avoid hang in case someone just arrived. */ | |
3940 | if (ULONG_CMP_LT(rnp->exp_seq_rq, s)) | |
3941 | rnp->exp_seq_rq = s; | |
3942 | spin_unlock(&rnp->exp_lock); | |
3943 | } | |
3b5f668e | 3944 | wake_up_all(&rnp->exp_wq[(rsp->expedited_sequence >> 1) & 0x3]); |
f6a12f34 | 3945 | } |
4ea3e85b | 3946 | trace_rcu_exp_grace_period(rsp->name, s, TPS("endwake")); |
3b5f668e | 3947 | mutex_unlock(&rsp->exp_wake_mutex); |
f6a12f34 PM |
3948 | } |
3949 | ||
236fefaf PM |
3950 | /** |
3951 | * synchronize_sched_expedited - Brute-force RCU-sched grace period | |
3952 | * | |
3953 | * Wait for an RCU-sched grace period to elapse, but use a "big hammer" | |
3954 | * approach to force the grace period to end quickly. This consumes | |
3955 | * significant time on all CPUs and is unfriendly to real-time workloads, | |
3956 | * so is thus not recommended for any sort of common-case code. In fact, | |
3957 | * if you are using synchronize_sched_expedited() in a loop, please | |
3958 | * restructure your code to batch your updates, and then use a single | |
3959 | * synchronize_sched() instead. | |
3d3b7db0 | 3960 | * |
d6ada2cf PM |
3961 | * This implementation can be thought of as an application of sequence |
3962 | * locking to expedited grace periods, but using the sequence counter to | |
3963 | * determine when someone else has already done the work instead of for | |
385b73c0 | 3964 | * retrying readers. |
3d3b7db0 PM |
3965 | */ |
3966 | void synchronize_sched_expedited(void) | |
3967 | { | |
7fd0ddc5 | 3968 | unsigned long s; |
40694d66 | 3969 | struct rcu_state *rsp = &rcu_sched_state; |
3d3b7db0 | 3970 | |
06f60de1 PM |
3971 | /* If only one CPU, this is automatically a grace period. */ |
3972 | if (rcu_blocking_is_gp()) | |
3973 | return; | |
3974 | ||
5a9be7c6 PM |
3975 | /* If expedited grace periods are prohibited, fall back to normal. */ |
3976 | if (rcu_gp_is_normal()) { | |
3977 | wait_rcu_gp(call_rcu_sched); | |
3978 | return; | |
3979 | } | |
3980 | ||
d6ada2cf | 3981 | /* Take a snapshot of the sequence number. */ |
28f00767 | 3982 | s = rcu_exp_gp_seq_snap(rsp); |
f6a12f34 | 3983 | if (exp_funnel_lock(rsp, s)) |
b09e5f86 | 3984 | return; /* Someone else did our work for us. */ |
e0775cef | 3985 | |
f6a12f34 | 3986 | /* Initialize the rcu_node tree in preparation for the wait. */ |
338b0f76 | 3987 | sync_rcu_exp_select_cpus(rsp, sync_sched_exp_handler); |
e0775cef | 3988 | |
f6a12f34 | 3989 | /* Wait and clean up, including waking everyone. */ |
4ea3e85b | 3990 | rcu_exp_wait_wake(rsp, s); |
3d3b7db0 PM |
3991 | } |
3992 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
3993 | ||
64db4cff PM |
3994 | /* |
3995 | * Check to see if there is any immediate RCU-related work to be done | |
3996 | * by the current CPU, for the specified type of RCU, returning 1 if so. | |
3997 | * The checks are in order of increasing expense: checks that can be | |
3998 | * carried out against CPU-local state are performed first. However, | |
3999 | * we must check for CPU stalls first, else we might not get a chance. | |
4000 | */ | |
4001 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | |
4002 | { | |
2f51f988 PM |
4003 | struct rcu_node *rnp = rdp->mynode; |
4004 | ||
64db4cff PM |
4005 | rdp->n_rcu_pending++; |
4006 | ||
4007 | /* Check for CPU stalls, if enabled. */ | |
4008 | check_cpu_stall(rsp, rdp); | |
4009 | ||
a096932f PM |
4010 | /* Is this CPU a NO_HZ_FULL CPU that should ignore RCU? */ |
4011 | if (rcu_nohz_full_cpu(rsp)) | |
4012 | return 0; | |
4013 | ||
64db4cff | 4014 | /* Is the RCU core waiting for a quiescent state from this CPU? */ |
5c51dd73 | 4015 | if (rcu_scheduler_fully_active && |
5b74c458 | 4016 | rdp->core_needs_qs && rdp->cpu_no_qs.b.norm && |
5cd37193 | 4017 | rdp->rcu_qs_ctr_snap == __this_cpu_read(rcu_qs_ctr)) { |
97c668b8 PM |
4018 | rdp->n_rp_core_needs_qs++; |
4019 | } else if (rdp->core_needs_qs && | |
5b74c458 | 4020 | (!rdp->cpu_no_qs.b.norm || |
5cd37193 | 4021 | rdp->rcu_qs_ctr_snap != __this_cpu_read(rcu_qs_ctr))) { |
d21670ac | 4022 | rdp->n_rp_report_qs++; |
64db4cff | 4023 | return 1; |
7ba5c840 | 4024 | } |
64db4cff PM |
4025 | |
4026 | /* Does this CPU have callbacks ready to invoke? */ | |
7ba5c840 PM |
4027 | if (cpu_has_callbacks_ready_to_invoke(rdp)) { |
4028 | rdp->n_rp_cb_ready++; | |
64db4cff | 4029 | return 1; |
7ba5c840 | 4030 | } |
64db4cff PM |
4031 | |
4032 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
7ba5c840 PM |
4033 | if (cpu_needs_another_gp(rsp, rdp)) { |
4034 | rdp->n_rp_cpu_needs_gp++; | |
64db4cff | 4035 | return 1; |
7ba5c840 | 4036 | } |
64db4cff PM |
4037 | |
4038 | /* Has another RCU grace period completed? */ | |
7d0ae808 | 4039 | if (READ_ONCE(rnp->completed) != rdp->completed) { /* outside lock */ |
7ba5c840 | 4040 | rdp->n_rp_gp_completed++; |
64db4cff | 4041 | return 1; |
7ba5c840 | 4042 | } |
64db4cff PM |
4043 | |
4044 | /* Has a new RCU grace period started? */ | |
7d0ae808 PM |
4045 | if (READ_ONCE(rnp->gpnum) != rdp->gpnum || |
4046 | unlikely(READ_ONCE(rdp->gpwrap))) { /* outside lock */ | |
7ba5c840 | 4047 | rdp->n_rp_gp_started++; |
64db4cff | 4048 | return 1; |
7ba5c840 | 4049 | } |
64db4cff | 4050 | |
96d3fd0d PM |
4051 | /* Does this CPU need a deferred NOCB wakeup? */ |
4052 | if (rcu_nocb_need_deferred_wakeup(rdp)) { | |
4053 | rdp->n_rp_nocb_defer_wakeup++; | |
4054 | return 1; | |
4055 | } | |
4056 | ||
64db4cff | 4057 | /* nothing to do */ |
7ba5c840 | 4058 | rdp->n_rp_need_nothing++; |
64db4cff PM |
4059 | return 0; |
4060 | } | |
4061 | ||
4062 | /* | |
4063 | * Check to see if there is any immediate RCU-related work to be done | |
4064 | * by the current CPU, returning 1 if so. This function is part of the | |
4065 | * RCU implementation; it is -not- an exported member of the RCU API. | |
4066 | */ | |
e3950ecd | 4067 | static int rcu_pending(void) |
64db4cff | 4068 | { |
6ce75a23 PM |
4069 | struct rcu_state *rsp; |
4070 | ||
4071 | for_each_rcu_flavor(rsp) | |
e3950ecd | 4072 | if (__rcu_pending(rsp, this_cpu_ptr(rsp->rda))) |
6ce75a23 PM |
4073 | return 1; |
4074 | return 0; | |
64db4cff PM |
4075 | } |
4076 | ||
4077 | /* | |
c0f4dfd4 PM |
4078 | * Return true if the specified CPU has any callback. If all_lazy is |
4079 | * non-NULL, store an indication of whether all callbacks are lazy. | |
4080 | * (If there are no callbacks, all of them are deemed to be lazy.) | |
64db4cff | 4081 | */ |
82072c4f | 4082 | static bool __maybe_unused rcu_cpu_has_callbacks(bool *all_lazy) |
64db4cff | 4083 | { |
c0f4dfd4 PM |
4084 | bool al = true; |
4085 | bool hc = false; | |
4086 | struct rcu_data *rdp; | |
6ce75a23 PM |
4087 | struct rcu_state *rsp; |
4088 | ||
c0f4dfd4 | 4089 | for_each_rcu_flavor(rsp) { |
aa6da514 | 4090 | rdp = this_cpu_ptr(rsp->rda); |
69c8d28c PM |
4091 | if (!rdp->nxtlist) |
4092 | continue; | |
4093 | hc = true; | |
4094 | if (rdp->qlen != rdp->qlen_lazy || !all_lazy) { | |
c0f4dfd4 | 4095 | al = false; |
69c8d28c PM |
4096 | break; |
4097 | } | |
c0f4dfd4 PM |
4098 | } |
4099 | if (all_lazy) | |
4100 | *all_lazy = al; | |
4101 | return hc; | |
64db4cff PM |
4102 | } |
4103 | ||
a83eff0a PM |
4104 | /* |
4105 | * Helper function for _rcu_barrier() tracing. If tracing is disabled, | |
4106 | * the compiler is expected to optimize this away. | |
4107 | */ | |
e66c33d5 | 4108 | static void _rcu_barrier_trace(struct rcu_state *rsp, const char *s, |
a83eff0a PM |
4109 | int cpu, unsigned long done) |
4110 | { | |
4111 | trace_rcu_barrier(rsp->name, s, cpu, | |
4112 | atomic_read(&rsp->barrier_cpu_count), done); | |
4113 | } | |
4114 | ||
b1420f1c PM |
4115 | /* |
4116 | * RCU callback function for _rcu_barrier(). If we are last, wake | |
4117 | * up the task executing _rcu_barrier(). | |
4118 | */ | |
24ebbca8 | 4119 | static void rcu_barrier_callback(struct rcu_head *rhp) |
d0ec774c | 4120 | { |
24ebbca8 PM |
4121 | struct rcu_data *rdp = container_of(rhp, struct rcu_data, barrier_head); |
4122 | struct rcu_state *rsp = rdp->rsp; | |
4123 | ||
a83eff0a | 4124 | if (atomic_dec_and_test(&rsp->barrier_cpu_count)) { |
4f525a52 | 4125 | _rcu_barrier_trace(rsp, "LastCB", -1, rsp->barrier_sequence); |
7db74df8 | 4126 | complete(&rsp->barrier_completion); |
a83eff0a | 4127 | } else { |
4f525a52 | 4128 | _rcu_barrier_trace(rsp, "CB", -1, rsp->barrier_sequence); |
a83eff0a | 4129 | } |
d0ec774c PM |
4130 | } |
4131 | ||
4132 | /* | |
4133 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
4134 | */ | |
4135 | static void rcu_barrier_func(void *type) | |
4136 | { | |
037b64ed | 4137 | struct rcu_state *rsp = type; |
fa07a58f | 4138 | struct rcu_data *rdp = raw_cpu_ptr(rsp->rda); |
d0ec774c | 4139 | |
4f525a52 | 4140 | _rcu_barrier_trace(rsp, "IRQ", -1, rsp->barrier_sequence); |
24ebbca8 | 4141 | atomic_inc(&rsp->barrier_cpu_count); |
06668efa | 4142 | rsp->call(&rdp->barrier_head, rcu_barrier_callback); |
d0ec774c PM |
4143 | } |
4144 | ||
d0ec774c PM |
4145 | /* |
4146 | * Orchestrate the specified type of RCU barrier, waiting for all | |
4147 | * RCU callbacks of the specified type to complete. | |
4148 | */ | |
037b64ed | 4149 | static void _rcu_barrier(struct rcu_state *rsp) |
d0ec774c | 4150 | { |
b1420f1c | 4151 | int cpu; |
b1420f1c | 4152 | struct rcu_data *rdp; |
4f525a52 | 4153 | unsigned long s = rcu_seq_snap(&rsp->barrier_sequence); |
b1420f1c | 4154 | |
4f525a52 | 4155 | _rcu_barrier_trace(rsp, "Begin", -1, s); |
b1420f1c | 4156 | |
e74f4c45 | 4157 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
7be7f0be | 4158 | mutex_lock(&rsp->barrier_mutex); |
b1420f1c | 4159 | |
4f525a52 PM |
4160 | /* Did someone else do our work for us? */ |
4161 | if (rcu_seq_done(&rsp->barrier_sequence, s)) { | |
4162 | _rcu_barrier_trace(rsp, "EarlyExit", -1, rsp->barrier_sequence); | |
cf3a9c48 PM |
4163 | smp_mb(); /* caller's subsequent code after above check. */ |
4164 | mutex_unlock(&rsp->barrier_mutex); | |
4165 | return; | |
4166 | } | |
4167 | ||
4f525a52 PM |
4168 | /* Mark the start of the barrier operation. */ |
4169 | rcu_seq_start(&rsp->barrier_sequence); | |
4170 | _rcu_barrier_trace(rsp, "Inc1", -1, rsp->barrier_sequence); | |
b1420f1c | 4171 | |
d0ec774c | 4172 | /* |
b1420f1c PM |
4173 | * Initialize the count to one rather than to zero in order to |
4174 | * avoid a too-soon return to zero in case of a short grace period | |
1331e7a1 PM |
4175 | * (or preemption of this task). Exclude CPU-hotplug operations |
4176 | * to ensure that no offline CPU has callbacks queued. | |
d0ec774c | 4177 | */ |
7db74df8 | 4178 | init_completion(&rsp->barrier_completion); |
24ebbca8 | 4179 | atomic_set(&rsp->barrier_cpu_count, 1); |
1331e7a1 | 4180 | get_online_cpus(); |
b1420f1c PM |
4181 | |
4182 | /* | |
1331e7a1 PM |
4183 | * Force each CPU with callbacks to register a new callback. |
4184 | * When that callback is invoked, we will know that all of the | |
4185 | * corresponding CPU's preceding callbacks have been invoked. | |
b1420f1c | 4186 | */ |
3fbfbf7a | 4187 | for_each_possible_cpu(cpu) { |
d1e43fa5 | 4188 | if (!cpu_online(cpu) && !rcu_is_nocb_cpu(cpu)) |
3fbfbf7a | 4189 | continue; |
b1420f1c | 4190 | rdp = per_cpu_ptr(rsp->rda, cpu); |
d1e43fa5 | 4191 | if (rcu_is_nocb_cpu(cpu)) { |
d7e29933 PM |
4192 | if (!rcu_nocb_cpu_needs_barrier(rsp, cpu)) { |
4193 | _rcu_barrier_trace(rsp, "OfflineNoCB", cpu, | |
4f525a52 | 4194 | rsp->barrier_sequence); |
d7e29933 PM |
4195 | } else { |
4196 | _rcu_barrier_trace(rsp, "OnlineNoCB", cpu, | |
4f525a52 | 4197 | rsp->barrier_sequence); |
41050a00 | 4198 | smp_mb__before_atomic(); |
d7e29933 PM |
4199 | atomic_inc(&rsp->barrier_cpu_count); |
4200 | __call_rcu(&rdp->barrier_head, | |
4201 | rcu_barrier_callback, rsp, cpu, 0); | |
4202 | } | |
7d0ae808 | 4203 | } else if (READ_ONCE(rdp->qlen)) { |
a83eff0a | 4204 | _rcu_barrier_trace(rsp, "OnlineQ", cpu, |
4f525a52 | 4205 | rsp->barrier_sequence); |
037b64ed | 4206 | smp_call_function_single(cpu, rcu_barrier_func, rsp, 1); |
b1420f1c | 4207 | } else { |
a83eff0a | 4208 | _rcu_barrier_trace(rsp, "OnlineNQ", cpu, |
4f525a52 | 4209 | rsp->barrier_sequence); |
b1420f1c PM |
4210 | } |
4211 | } | |
1331e7a1 | 4212 | put_online_cpus(); |
b1420f1c PM |
4213 | |
4214 | /* | |
4215 | * Now that we have an rcu_barrier_callback() callback on each | |
4216 | * CPU, and thus each counted, remove the initial count. | |
4217 | */ | |
24ebbca8 | 4218 | if (atomic_dec_and_test(&rsp->barrier_cpu_count)) |
7db74df8 | 4219 | complete(&rsp->barrier_completion); |
b1420f1c PM |
4220 | |
4221 | /* Wait for all rcu_barrier_callback() callbacks to be invoked. */ | |
7db74df8 | 4222 | wait_for_completion(&rsp->barrier_completion); |
b1420f1c | 4223 | |
4f525a52 PM |
4224 | /* Mark the end of the barrier operation. */ |
4225 | _rcu_barrier_trace(rsp, "Inc2", -1, rsp->barrier_sequence); | |
4226 | rcu_seq_end(&rsp->barrier_sequence); | |
4227 | ||
b1420f1c | 4228 | /* Other rcu_barrier() invocations can now safely proceed. */ |
7be7f0be | 4229 | mutex_unlock(&rsp->barrier_mutex); |
d0ec774c | 4230 | } |
d0ec774c PM |
4231 | |
4232 | /** | |
4233 | * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete. | |
4234 | */ | |
4235 | void rcu_barrier_bh(void) | |
4236 | { | |
037b64ed | 4237 | _rcu_barrier(&rcu_bh_state); |
d0ec774c PM |
4238 | } |
4239 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); | |
4240 | ||
4241 | /** | |
4242 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. | |
4243 | */ | |
4244 | void rcu_barrier_sched(void) | |
4245 | { | |
037b64ed | 4246 | _rcu_barrier(&rcu_sched_state); |
d0ec774c PM |
4247 | } |
4248 | EXPORT_SYMBOL_GPL(rcu_barrier_sched); | |
4249 | ||
0aa04b05 PM |
4250 | /* |
4251 | * Propagate ->qsinitmask bits up the rcu_node tree to account for the | |
4252 | * first CPU in a given leaf rcu_node structure coming online. The caller | |
4253 | * must hold the corresponding leaf rcu_node ->lock with interrrupts | |
4254 | * disabled. | |
4255 | */ | |
4256 | static void rcu_init_new_rnp(struct rcu_node *rnp_leaf) | |
4257 | { | |
4258 | long mask; | |
4259 | struct rcu_node *rnp = rnp_leaf; | |
4260 | ||
4261 | for (;;) { | |
4262 | mask = rnp->grpmask; | |
4263 | rnp = rnp->parent; | |
4264 | if (rnp == NULL) | |
4265 | return; | |
6cf10081 | 4266 | raw_spin_lock_rcu_node(rnp); /* Interrupts already disabled. */ |
0aa04b05 | 4267 | rnp->qsmaskinit |= mask; |
67c583a7 | 4268 | raw_spin_unlock_rcu_node(rnp); /* Interrupts remain disabled. */ |
0aa04b05 PM |
4269 | } |
4270 | } | |
4271 | ||
64db4cff | 4272 | /* |
27569620 | 4273 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 4274 | */ |
27569620 PM |
4275 | static void __init |
4276 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) | |
64db4cff PM |
4277 | { |
4278 | unsigned long flags; | |
394f99a9 | 4279 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
27569620 PM |
4280 | struct rcu_node *rnp = rcu_get_root(rsp); |
4281 | ||
4282 | /* Set up local state, ensuring consistent view of global state. */ | |
6cf10081 | 4283 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
27569620 | 4284 | rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo); |
27569620 | 4285 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); |
29e37d81 | 4286 | WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != DYNTICK_TASK_EXIT_IDLE); |
9b2e4f18 | 4287 | WARN_ON_ONCE(atomic_read(&rdp->dynticks->dynticks) != 1); |
27569620 | 4288 | rdp->cpu = cpu; |
d4c08f2a | 4289 | rdp->rsp = rsp; |
3fbfbf7a | 4290 | rcu_boot_init_nocb_percpu_data(rdp); |
67c583a7 | 4291 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27569620 PM |
4292 | } |
4293 | ||
4294 | /* | |
4295 | * Initialize a CPU's per-CPU RCU data. Note that only one online or | |
4296 | * offline event can be happening at a given time. Note also that we | |
4297 | * can accept some slop in the rsp->completed access due to the fact | |
4298 | * that this CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 4299 | */ |
49fb4c62 | 4300 | static void |
9b67122a | 4301 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp) |
64db4cff PM |
4302 | { |
4303 | unsigned long flags; | |
64db4cff | 4304 | unsigned long mask; |
394f99a9 | 4305 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff PM |
4306 | struct rcu_node *rnp = rcu_get_root(rsp); |
4307 | ||
4308 | /* Set up local state, ensuring consistent view of global state. */ | |
6cf10081 | 4309 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
37c72e56 PM |
4310 | rdp->qlen_last_fqs_check = 0; |
4311 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
64db4cff | 4312 | rdp->blimit = blimit; |
39c8d313 PM |
4313 | if (!rdp->nxtlist) |
4314 | init_callback_list(rdp); /* Re-enable callbacks on this CPU. */ | |
29e37d81 | 4315 | rdp->dynticks->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE; |
2333210b | 4316 | rcu_sysidle_init_percpu_data(rdp->dynticks); |
c92b131b PM |
4317 | atomic_set(&rdp->dynticks->dynticks, |
4318 | (atomic_read(&rdp->dynticks->dynticks) & ~0x1) + 1); | |
67c583a7 | 4319 | raw_spin_unlock_rcu_node(rnp); /* irqs remain disabled. */ |
64db4cff | 4320 | |
0aa04b05 PM |
4321 | /* |
4322 | * Add CPU to leaf rcu_node pending-online bitmask. Any needed | |
4323 | * propagation up the rcu_node tree will happen at the beginning | |
4324 | * of the next grace period. | |
4325 | */ | |
64db4cff PM |
4326 | rnp = rdp->mynode; |
4327 | mask = rdp->grpmask; | |
2a67e741 | 4328 | raw_spin_lock_rcu_node(rnp); /* irqs already disabled. */ |
0aa04b05 | 4329 | rnp->qsmaskinitnext |= mask; |
b9585e94 PM |
4330 | rnp->expmaskinitnext |= mask; |
4331 | if (!rdp->beenonline) | |
4332 | WRITE_ONCE(rsp->ncpus, READ_ONCE(rsp->ncpus) + 1); | |
4333 | rdp->beenonline = true; /* We have now been online. */ | |
0aa04b05 PM |
4334 | rdp->gpnum = rnp->completed; /* Make CPU later note any new GP. */ |
4335 | rdp->completed = rnp->completed; | |
5b74c458 | 4336 | rdp->cpu_no_qs.b.norm = true; |
a738eec6 | 4337 | rdp->rcu_qs_ctr_snap = per_cpu(rcu_qs_ctr, cpu); |
97c668b8 | 4338 | rdp->core_needs_qs = false; |
0aa04b05 | 4339 | trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuonl")); |
67c583a7 | 4340 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
64db4cff PM |
4341 | } |
4342 | ||
49fb4c62 | 4343 | static void rcu_prepare_cpu(int cpu) |
64db4cff | 4344 | { |
6ce75a23 PM |
4345 | struct rcu_state *rsp; |
4346 | ||
4347 | for_each_rcu_flavor(rsp) | |
9b67122a | 4348 | rcu_init_percpu_data(cpu, rsp); |
64db4cff PM |
4349 | } |
4350 | ||
27d50c7e TG |
4351 | #ifdef CONFIG_HOTPLUG_CPU |
4352 | /* | |
710d60cb LT |
4353 | * The CPU is exiting the idle loop into the arch_cpu_idle_dead() |
4354 | * function. We now remove it from the rcu_node tree's ->qsmaskinit | |
4355 | * bit masks. | |
27d50c7e TG |
4356 | * The CPU is exiting the idle loop into the arch_cpu_idle_dead() |
4357 | * function. We now remove it from the rcu_node tree's ->qsmaskinit | |
4358 | * bit masks. | |
4359 | */ | |
4360 | static void rcu_cleanup_dying_idle_cpu(int cpu, struct rcu_state *rsp) | |
4361 | { | |
4362 | unsigned long flags; | |
4363 | unsigned long mask; | |
4364 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); | |
4365 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ | |
4366 | ||
4367 | if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) | |
4368 | return; | |
4369 | ||
4370 | /* Remove outgoing CPU from mask in the leaf rcu_node structure. */ | |
4371 | mask = rdp->grpmask; | |
4372 | raw_spin_lock_irqsave_rcu_node(rnp, flags); /* Enforce GP memory-order guarantee. */ | |
4373 | rnp->qsmaskinitnext &= ~mask; | |
710d60cb | 4374 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
27d50c7e TG |
4375 | } |
4376 | ||
4377 | void rcu_report_dead(unsigned int cpu) | |
4378 | { | |
4379 | struct rcu_state *rsp; | |
4380 | ||
4381 | /* QS for any half-done expedited RCU-sched GP. */ | |
4382 | preempt_disable(); | |
4383 | rcu_report_exp_rdp(&rcu_sched_state, | |
4384 | this_cpu_ptr(rcu_sched_state.rda), true); | |
4385 | preempt_enable(); | |
4386 | for_each_rcu_flavor(rsp) | |
4387 | rcu_cleanup_dying_idle_cpu(cpu, rsp); | |
4388 | } | |
4389 | #endif | |
4390 | ||
64db4cff | 4391 | /* |
f41d911f | 4392 | * Handle CPU online/offline notification events. |
64db4cff | 4393 | */ |
88428cc5 PM |
4394 | int rcu_cpu_notify(struct notifier_block *self, |
4395 | unsigned long action, void *hcpu) | |
64db4cff PM |
4396 | { |
4397 | long cpu = (long)hcpu; | |
e534165b | 4398 | struct rcu_data *rdp = per_cpu_ptr(rcu_state_p->rda, cpu); |
a26ac245 | 4399 | struct rcu_node *rnp = rdp->mynode; |
6ce75a23 | 4400 | struct rcu_state *rsp; |
64db4cff PM |
4401 | |
4402 | switch (action) { | |
4403 | case CPU_UP_PREPARE: | |
4404 | case CPU_UP_PREPARE_FROZEN: | |
d72bce0e PZ |
4405 | rcu_prepare_cpu(cpu); |
4406 | rcu_prepare_kthreads(cpu); | |
35ce7f29 | 4407 | rcu_spawn_all_nocb_kthreads(cpu); |
a26ac245 PM |
4408 | break; |
4409 | case CPU_ONLINE: | |
0f962a5e | 4410 | case CPU_DOWN_FAILED: |
338b0f76 | 4411 | sync_sched_exp_online_cleanup(cpu); |
5d01bbd1 | 4412 | rcu_boost_kthread_setaffinity(rnp, -1); |
0f962a5e PM |
4413 | break; |
4414 | case CPU_DOWN_PREPARE: | |
34ed6246 | 4415 | rcu_boost_kthread_setaffinity(rnp, cpu); |
64db4cff | 4416 | break; |
d0ec774c PM |
4417 | case CPU_DYING: |
4418 | case CPU_DYING_FROZEN: | |
6ce75a23 PM |
4419 | for_each_rcu_flavor(rsp) |
4420 | rcu_cleanup_dying_cpu(rsp); | |
d0ec774c | 4421 | break; |
64db4cff PM |
4422 | case CPU_DEAD: |
4423 | case CPU_DEAD_FROZEN: | |
4424 | case CPU_UP_CANCELED: | |
4425 | case CPU_UP_CANCELED_FROZEN: | |
776d6807 | 4426 | for_each_rcu_flavor(rsp) { |
6ce75a23 | 4427 | rcu_cleanup_dead_cpu(cpu, rsp); |
776d6807 PM |
4428 | do_nocb_deferred_wakeup(per_cpu_ptr(rsp->rda, cpu)); |
4429 | } | |
64db4cff PM |
4430 | break; |
4431 | default: | |
4432 | break; | |
4433 | } | |
34ed6246 | 4434 | return NOTIFY_OK; |
64db4cff PM |
4435 | } |
4436 | ||
d1d74d14 BP |
4437 | static int rcu_pm_notify(struct notifier_block *self, |
4438 | unsigned long action, void *hcpu) | |
4439 | { | |
4440 | switch (action) { | |
4441 | case PM_HIBERNATION_PREPARE: | |
4442 | case PM_SUSPEND_PREPARE: | |
4443 | if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */ | |
5afff48b | 4444 | rcu_expedite_gp(); |
d1d74d14 BP |
4445 | break; |
4446 | case PM_POST_HIBERNATION: | |
4447 | case PM_POST_SUSPEND: | |
5afff48b PM |
4448 | if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */ |
4449 | rcu_unexpedite_gp(); | |
d1d74d14 BP |
4450 | break; |
4451 | default: | |
4452 | break; | |
4453 | } | |
4454 | return NOTIFY_OK; | |
4455 | } | |
4456 | ||
b3dbec76 | 4457 | /* |
9386c0b7 | 4458 | * Spawn the kthreads that handle each RCU flavor's grace periods. |
b3dbec76 PM |
4459 | */ |
4460 | static int __init rcu_spawn_gp_kthread(void) | |
4461 | { | |
4462 | unsigned long flags; | |
a94844b2 | 4463 | int kthread_prio_in = kthread_prio; |
b3dbec76 PM |
4464 | struct rcu_node *rnp; |
4465 | struct rcu_state *rsp; | |
a94844b2 | 4466 | struct sched_param sp; |
b3dbec76 PM |
4467 | struct task_struct *t; |
4468 | ||
a94844b2 PM |
4469 | /* Force priority into range. */ |
4470 | if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 1) | |
4471 | kthread_prio = 1; | |
4472 | else if (kthread_prio < 0) | |
4473 | kthread_prio = 0; | |
4474 | else if (kthread_prio > 99) | |
4475 | kthread_prio = 99; | |
4476 | if (kthread_prio != kthread_prio_in) | |
4477 | pr_alert("rcu_spawn_gp_kthread(): Limited prio to %d from %d\n", | |
4478 | kthread_prio, kthread_prio_in); | |
4479 | ||
9386c0b7 | 4480 | rcu_scheduler_fully_active = 1; |
b3dbec76 | 4481 | for_each_rcu_flavor(rsp) { |
a94844b2 | 4482 | t = kthread_create(rcu_gp_kthread, rsp, "%s", rsp->name); |
b3dbec76 PM |
4483 | BUG_ON(IS_ERR(t)); |
4484 | rnp = rcu_get_root(rsp); | |
6cf10081 | 4485 | raw_spin_lock_irqsave_rcu_node(rnp, flags); |
b3dbec76 | 4486 | rsp->gp_kthread = t; |
a94844b2 PM |
4487 | if (kthread_prio) { |
4488 | sp.sched_priority = kthread_prio; | |
4489 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); | |
4490 | } | |
67c583a7 | 4491 | raw_spin_unlock_irqrestore_rcu_node(rnp, flags); |
e11f1335 | 4492 | wake_up_process(t); |
b3dbec76 | 4493 | } |
35ce7f29 | 4494 | rcu_spawn_nocb_kthreads(); |
9386c0b7 | 4495 | rcu_spawn_boost_kthreads(); |
b3dbec76 PM |
4496 | return 0; |
4497 | } | |
4498 | early_initcall(rcu_spawn_gp_kthread); | |
4499 | ||
bbad9379 PM |
4500 | /* |
4501 | * This function is invoked towards the end of the scheduler's initialization | |
4502 | * process. Before this is called, the idle task might contain | |
4503 | * RCU read-side critical sections (during which time, this idle | |
4504 | * task is booting the system). After this function is called, the | |
4505 | * idle tasks are prohibited from containing RCU read-side critical | |
4506 | * sections. This function also enables RCU lockdep checking. | |
4507 | */ | |
4508 | void rcu_scheduler_starting(void) | |
4509 | { | |
4510 | WARN_ON(num_online_cpus() != 1); | |
4511 | WARN_ON(nr_context_switches() > 0); | |
4512 | rcu_scheduler_active = 1; | |
4513 | } | |
4514 | ||
64db4cff PM |
4515 | /* |
4516 | * Compute the per-level fanout, either using the exact fanout specified | |
7fa27001 | 4517 | * or balancing the tree, depending on the rcu_fanout_exact boot parameter. |
64db4cff | 4518 | */ |
199977bf | 4519 | static void __init rcu_init_levelspread(int *levelspread, const int *levelcnt) |
64db4cff | 4520 | { |
64db4cff PM |
4521 | int i; |
4522 | ||
7fa27001 | 4523 | if (rcu_fanout_exact) { |
199977bf | 4524 | levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf; |
66292405 | 4525 | for (i = rcu_num_lvls - 2; i >= 0; i--) |
199977bf | 4526 | levelspread[i] = RCU_FANOUT; |
66292405 PM |
4527 | } else { |
4528 | int ccur; | |
4529 | int cprv; | |
4530 | ||
4531 | cprv = nr_cpu_ids; | |
4532 | for (i = rcu_num_lvls - 1; i >= 0; i--) { | |
199977bf AG |
4533 | ccur = levelcnt[i]; |
4534 | levelspread[i] = (cprv + ccur - 1) / ccur; | |
66292405 PM |
4535 | cprv = ccur; |
4536 | } | |
64db4cff PM |
4537 | } |
4538 | } | |
64db4cff PM |
4539 | |
4540 | /* | |
4541 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
4542 | */ | |
a87f203e | 4543 | static void __init rcu_init_one(struct rcu_state *rsp) |
64db4cff | 4544 | { |
cb007102 AG |
4545 | static const char * const buf[] = RCU_NODE_NAME_INIT; |
4546 | static const char * const fqs[] = RCU_FQS_NAME_INIT; | |
3dc5dbe9 PM |
4547 | static struct lock_class_key rcu_node_class[RCU_NUM_LVLS]; |
4548 | static struct lock_class_key rcu_fqs_class[RCU_NUM_LVLS]; | |
4a81e832 | 4549 | static u8 fl_mask = 0x1; |
199977bf AG |
4550 | |
4551 | int levelcnt[RCU_NUM_LVLS]; /* # nodes in each level. */ | |
4552 | int levelspread[RCU_NUM_LVLS]; /* kids/node in each level. */ | |
64db4cff PM |
4553 | int cpustride = 1; |
4554 | int i; | |
4555 | int j; | |
4556 | struct rcu_node *rnp; | |
4557 | ||
05b84aec | 4558 | BUILD_BUG_ON(RCU_NUM_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
b6407e86 | 4559 | |
3eaaaf6c PM |
4560 | /* Silence gcc 4.8 false positive about array index out of range. */ |
4561 | if (rcu_num_lvls <= 0 || rcu_num_lvls > RCU_NUM_LVLS) | |
4562 | panic("rcu_init_one: rcu_num_lvls out of range"); | |
4930521a | 4563 | |
64db4cff PM |
4564 | /* Initialize the level-tracking arrays. */ |
4565 | ||
f885b7f2 | 4566 | for (i = 0; i < rcu_num_lvls; i++) |
199977bf | 4567 | levelcnt[i] = num_rcu_lvl[i]; |
f885b7f2 | 4568 | for (i = 1; i < rcu_num_lvls; i++) |
199977bf AG |
4569 | rsp->level[i] = rsp->level[i - 1] + levelcnt[i - 1]; |
4570 | rcu_init_levelspread(levelspread, levelcnt); | |
4a81e832 PM |
4571 | rsp->flavor_mask = fl_mask; |
4572 | fl_mask <<= 1; | |
64db4cff PM |
4573 | |
4574 | /* Initialize the elements themselves, starting from the leaves. */ | |
4575 | ||
f885b7f2 | 4576 | for (i = rcu_num_lvls - 1; i >= 0; i--) { |
199977bf | 4577 | cpustride *= levelspread[i]; |
64db4cff | 4578 | rnp = rsp->level[i]; |
199977bf | 4579 | for (j = 0; j < levelcnt[i]; j++, rnp++) { |
67c583a7 BF |
4580 | raw_spin_lock_init(&ACCESS_PRIVATE(rnp, lock)); |
4581 | lockdep_set_class_and_name(&ACCESS_PRIVATE(rnp, lock), | |
b6407e86 | 4582 | &rcu_node_class[i], buf[i]); |
394f2769 PM |
4583 | raw_spin_lock_init(&rnp->fqslock); |
4584 | lockdep_set_class_and_name(&rnp->fqslock, | |
4585 | &rcu_fqs_class[i], fqs[i]); | |
25d30cf4 PM |
4586 | rnp->gpnum = rsp->gpnum; |
4587 | rnp->completed = rsp->completed; | |
64db4cff PM |
4588 | rnp->qsmask = 0; |
4589 | rnp->qsmaskinit = 0; | |
4590 | rnp->grplo = j * cpustride; | |
4591 | rnp->grphi = (j + 1) * cpustride - 1; | |
595f3900 HS |
4592 | if (rnp->grphi >= nr_cpu_ids) |
4593 | rnp->grphi = nr_cpu_ids - 1; | |
64db4cff PM |
4594 | if (i == 0) { |
4595 | rnp->grpnum = 0; | |
4596 | rnp->grpmask = 0; | |
4597 | rnp->parent = NULL; | |
4598 | } else { | |
199977bf | 4599 | rnp->grpnum = j % levelspread[i - 1]; |
64db4cff PM |
4600 | rnp->grpmask = 1UL << rnp->grpnum; |
4601 | rnp->parent = rsp->level[i - 1] + | |
199977bf | 4602 | j / levelspread[i - 1]; |
64db4cff PM |
4603 | } |
4604 | rnp->level = i; | |
12f5f524 | 4605 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
dae6e64d | 4606 | rcu_init_one_nocb(rnp); |
f6a12f34 PM |
4607 | init_waitqueue_head(&rnp->exp_wq[0]); |
4608 | init_waitqueue_head(&rnp->exp_wq[1]); | |
3b5f668e PM |
4609 | init_waitqueue_head(&rnp->exp_wq[2]); |
4610 | init_waitqueue_head(&rnp->exp_wq[3]); | |
f6a12f34 | 4611 | spin_lock_init(&rnp->exp_lock); |
64db4cff PM |
4612 | } |
4613 | } | |
0c34029a | 4614 | |
abedf8e2 PG |
4615 | init_swait_queue_head(&rsp->gp_wq); |
4616 | init_swait_queue_head(&rsp->expedited_wq); | |
f885b7f2 | 4617 | rnp = rsp->level[rcu_num_lvls - 1]; |
0c34029a | 4618 | for_each_possible_cpu(i) { |
4a90a068 | 4619 | while (i > rnp->grphi) |
0c34029a | 4620 | rnp++; |
394f99a9 | 4621 | per_cpu_ptr(rsp->rda, i)->mynode = rnp; |
0c34029a LJ |
4622 | rcu_boot_init_percpu_data(i, rsp); |
4623 | } | |
6ce75a23 | 4624 | list_add(&rsp->flavors, &rcu_struct_flavors); |
64db4cff PM |
4625 | } |
4626 | ||
f885b7f2 PM |
4627 | /* |
4628 | * Compute the rcu_node tree geometry from kernel parameters. This cannot | |
4102adab | 4629 | * replace the definitions in tree.h because those are needed to size |
f885b7f2 PM |
4630 | * the ->node array in the rcu_state structure. |
4631 | */ | |
4632 | static void __init rcu_init_geometry(void) | |
4633 | { | |
026ad283 | 4634 | ulong d; |
f885b7f2 | 4635 | int i; |
05b84aec | 4636 | int rcu_capacity[RCU_NUM_LVLS]; |
f885b7f2 | 4637 | |
026ad283 PM |
4638 | /* |
4639 | * Initialize any unspecified boot parameters. | |
4640 | * The default values of jiffies_till_first_fqs and | |
4641 | * jiffies_till_next_fqs are set to the RCU_JIFFIES_TILL_FORCE_QS | |
4642 | * value, which is a function of HZ, then adding one for each | |
4643 | * RCU_JIFFIES_FQS_DIV CPUs that might be on the system. | |
4644 | */ | |
4645 | d = RCU_JIFFIES_TILL_FORCE_QS + nr_cpu_ids / RCU_JIFFIES_FQS_DIV; | |
4646 | if (jiffies_till_first_fqs == ULONG_MAX) | |
4647 | jiffies_till_first_fqs = d; | |
4648 | if (jiffies_till_next_fqs == ULONG_MAX) | |
4649 | jiffies_till_next_fqs = d; | |
4650 | ||
f885b7f2 | 4651 | /* If the compile-time values are accurate, just leave. */ |
47d631af | 4652 | if (rcu_fanout_leaf == RCU_FANOUT_LEAF && |
b17c7035 | 4653 | nr_cpu_ids == NR_CPUS) |
f885b7f2 | 4654 | return; |
39479098 PM |
4655 | pr_info("RCU: Adjusting geometry for rcu_fanout_leaf=%d, nr_cpu_ids=%d\n", |
4656 | rcu_fanout_leaf, nr_cpu_ids); | |
f885b7f2 | 4657 | |
f885b7f2 | 4658 | /* |
ee968ac6 PM |
4659 | * The boot-time rcu_fanout_leaf parameter must be at least two |
4660 | * and cannot exceed the number of bits in the rcu_node masks. | |
4661 | * Complain and fall back to the compile-time values if this | |
4662 | * limit is exceeded. | |
f885b7f2 | 4663 | */ |
ee968ac6 | 4664 | if (rcu_fanout_leaf < 2 || |
75cf15a4 | 4665 | rcu_fanout_leaf > sizeof(unsigned long) * 8) { |
13bd6494 | 4666 | rcu_fanout_leaf = RCU_FANOUT_LEAF; |
f885b7f2 PM |
4667 | WARN_ON(1); |
4668 | return; | |
4669 | } | |
4670 | ||
f885b7f2 PM |
4671 | /* |
4672 | * Compute number of nodes that can be handled an rcu_node tree | |
9618138b | 4673 | * with the given number of levels. |
f885b7f2 | 4674 | */ |
9618138b | 4675 | rcu_capacity[0] = rcu_fanout_leaf; |
05b84aec | 4676 | for (i = 1; i < RCU_NUM_LVLS; i++) |
05c5df31 | 4677 | rcu_capacity[i] = rcu_capacity[i - 1] * RCU_FANOUT; |
f885b7f2 PM |
4678 | |
4679 | /* | |
75cf15a4 | 4680 | * The tree must be able to accommodate the configured number of CPUs. |
ee968ac6 | 4681 | * If this limit is exceeded, fall back to the compile-time values. |
f885b7f2 | 4682 | */ |
ee968ac6 PM |
4683 | if (nr_cpu_ids > rcu_capacity[RCU_NUM_LVLS - 1]) { |
4684 | rcu_fanout_leaf = RCU_FANOUT_LEAF; | |
4685 | WARN_ON(1); | |
4686 | return; | |
4687 | } | |
f885b7f2 | 4688 | |
679f9858 | 4689 | /* Calculate the number of levels in the tree. */ |
9618138b | 4690 | for (i = 0; nr_cpu_ids > rcu_capacity[i]; i++) { |
679f9858 | 4691 | } |
9618138b | 4692 | rcu_num_lvls = i + 1; |
679f9858 | 4693 | |
f885b7f2 | 4694 | /* Calculate the number of rcu_nodes at each level of the tree. */ |
679f9858 | 4695 | for (i = 0; i < rcu_num_lvls; i++) { |
9618138b | 4696 | int cap = rcu_capacity[(rcu_num_lvls - 1) - i]; |
679f9858 AG |
4697 | num_rcu_lvl[i] = DIV_ROUND_UP(nr_cpu_ids, cap); |
4698 | } | |
f885b7f2 PM |
4699 | |
4700 | /* Calculate the total number of rcu_node structures. */ | |
4701 | rcu_num_nodes = 0; | |
679f9858 | 4702 | for (i = 0; i < rcu_num_lvls; i++) |
f885b7f2 | 4703 | rcu_num_nodes += num_rcu_lvl[i]; |
f885b7f2 PM |
4704 | } |
4705 | ||
a3dc2948 PM |
4706 | /* |
4707 | * Dump out the structure of the rcu_node combining tree associated | |
4708 | * with the rcu_state structure referenced by rsp. | |
4709 | */ | |
4710 | static void __init rcu_dump_rcu_node_tree(struct rcu_state *rsp) | |
4711 | { | |
4712 | int level = 0; | |
4713 | struct rcu_node *rnp; | |
4714 | ||
4715 | pr_info("rcu_node tree layout dump\n"); | |
4716 | pr_info(" "); | |
4717 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
4718 | if (rnp->level != level) { | |
4719 | pr_cont("\n"); | |
4720 | pr_info(" "); | |
4721 | level = rnp->level; | |
4722 | } | |
4723 | pr_cont("%d:%d ^%d ", rnp->grplo, rnp->grphi, rnp->grpnum); | |
4724 | } | |
4725 | pr_cont("\n"); | |
4726 | } | |
4727 | ||
9f680ab4 | 4728 | void __init rcu_init(void) |
64db4cff | 4729 | { |
017c4261 | 4730 | int cpu; |
9f680ab4 | 4731 | |
47627678 PM |
4732 | rcu_early_boot_tests(); |
4733 | ||
f41d911f | 4734 | rcu_bootup_announce(); |
f885b7f2 | 4735 | rcu_init_geometry(); |
a87f203e PM |
4736 | rcu_init_one(&rcu_bh_state); |
4737 | rcu_init_one(&rcu_sched_state); | |
a3dc2948 PM |
4738 | if (dump_tree) |
4739 | rcu_dump_rcu_node_tree(&rcu_sched_state); | |
f41d911f | 4740 | __rcu_init_preempt(); |
b5b39360 | 4741 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
9f680ab4 PM |
4742 | |
4743 | /* | |
4744 | * We don't need protection against CPU-hotplug here because | |
4745 | * this is called early in boot, before either interrupts | |
4746 | * or the scheduler are operational. | |
4747 | */ | |
4748 | cpu_notifier(rcu_cpu_notify, 0); | |
d1d74d14 | 4749 | pm_notifier(rcu_pm_notify, 0); |
017c4261 PM |
4750 | for_each_online_cpu(cpu) |
4751 | rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu); | |
64db4cff PM |
4752 | } |
4753 | ||
4102adab | 4754 | #include "tree_plugin.h" |