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 | |
15 | * along with this program; if not, write to the Free Software | |
16 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
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> | |
44 | #include <linux/percpu.h> | |
45 | #include <linux/notifier.h> | |
46 | #include <linux/cpu.h> | |
47 | #include <linux/mutex.h> | |
48 | #include <linux/time.h> | |
bbad9379 | 49 | #include <linux/kernel_stat.h> |
a26ac245 PM |
50 | #include <linux/wait.h> |
51 | #include <linux/kthread.h> | |
268bb0ce | 52 | #include <linux/prefetch.h> |
64db4cff | 53 | |
9f77da9f | 54 | #include "rcutree.h" |
29c00b4a PM |
55 | #include <trace/events/rcu.h> |
56 | ||
57 | #include "rcu.h" | |
9f77da9f | 58 | |
64db4cff PM |
59 | /* Data structures. */ |
60 | ||
b668c9cf | 61 | static struct lock_class_key rcu_node_class[NUM_RCU_LVLS]; |
88b91c7c | 62 | |
4300aa64 | 63 | #define RCU_STATE_INITIALIZER(structname) { \ |
e99033c5 | 64 | .level = { &structname##_state.node[0] }, \ |
64db4cff PM |
65 | .levelcnt = { \ |
66 | NUM_RCU_LVL_0, /* root of hierarchy. */ \ | |
67 | NUM_RCU_LVL_1, \ | |
68 | NUM_RCU_LVL_2, \ | |
cf244dc0 PM |
69 | NUM_RCU_LVL_3, \ |
70 | NUM_RCU_LVL_4, /* == MAX_RCU_LVLS */ \ | |
64db4cff | 71 | }, \ |
af446b70 | 72 | .fqs_state = RCU_GP_IDLE, \ |
64db4cff PM |
73 | .gpnum = -300, \ |
74 | .completed = -300, \ | |
e99033c5 PM |
75 | .onofflock = __RAW_SPIN_LOCK_UNLOCKED(&structname##_state.onofflock), \ |
76 | .fqslock = __RAW_SPIN_LOCK_UNLOCKED(&structname##_state.fqslock), \ | |
64db4cff PM |
77 | .n_force_qs = 0, \ |
78 | .n_force_qs_ngp = 0, \ | |
4300aa64 | 79 | .name = #structname, \ |
64db4cff PM |
80 | } |
81 | ||
e99033c5 | 82 | struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched); |
d6714c22 | 83 | DEFINE_PER_CPU(struct rcu_data, rcu_sched_data); |
64db4cff | 84 | |
e99033c5 | 85 | struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh); |
6258c4fb | 86 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); |
b1f77b05 | 87 | |
27f4d280 PM |
88 | static struct rcu_state *rcu_state; |
89 | ||
b0d30417 PM |
90 | /* |
91 | * The rcu_scheduler_active variable transitions from zero to one just | |
92 | * before the first task is spawned. So when this variable is zero, RCU | |
93 | * can assume that there is but one task, allowing RCU to (for example) | |
94 | * optimized synchronize_sched() to a simple barrier(). When this variable | |
95 | * is one, RCU must actually do all the hard work required to detect real | |
96 | * grace periods. This variable is also used to suppress boot-time false | |
97 | * positives from lockdep-RCU error checking. | |
98 | */ | |
bbad9379 PM |
99 | int rcu_scheduler_active __read_mostly; |
100 | EXPORT_SYMBOL_GPL(rcu_scheduler_active); | |
101 | ||
b0d30417 PM |
102 | /* |
103 | * The rcu_scheduler_fully_active variable transitions from zero to one | |
104 | * during the early_initcall() processing, which is after the scheduler | |
105 | * is capable of creating new tasks. So RCU processing (for example, | |
106 | * creating tasks for RCU priority boosting) must be delayed until after | |
107 | * rcu_scheduler_fully_active transitions from zero to one. We also | |
108 | * currently delay invocation of any RCU callbacks until after this point. | |
109 | * | |
110 | * It might later prove better for people registering RCU callbacks during | |
111 | * early boot to take responsibility for these callbacks, but one step at | |
112 | * a time. | |
113 | */ | |
114 | static int rcu_scheduler_fully_active __read_mostly; | |
115 | ||
a46e0899 PM |
116 | #ifdef CONFIG_RCU_BOOST |
117 | ||
a26ac245 PM |
118 | /* |
119 | * Control variables for per-CPU and per-rcu_node kthreads. These | |
120 | * handle all flavors of RCU. | |
121 | */ | |
122 | static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task); | |
d71df90e | 123 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status); |
15ba0ba8 | 124 | DEFINE_PER_CPU(int, rcu_cpu_kthread_cpu); |
5ece5bab | 125 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops); |
d71df90e | 126 | DEFINE_PER_CPU(char, rcu_cpu_has_work); |
a26ac245 | 127 | |
a46e0899 PM |
128 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
129 | ||
0f962a5e | 130 | static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); |
a46e0899 PM |
131 | static void invoke_rcu_core(void); |
132 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp); | |
a26ac245 | 133 | |
4a298656 PM |
134 | /* |
135 | * Track the rcutorture test sequence number and the update version | |
136 | * number within a given test. The rcutorture_testseq is incremented | |
137 | * on every rcutorture module load and unload, so has an odd value | |
138 | * when a test is running. The rcutorture_vernum is set to zero | |
139 | * when rcutorture starts and is incremented on each rcutorture update. | |
140 | * These variables enable correlating rcutorture output with the | |
141 | * RCU tracing information. | |
142 | */ | |
143 | unsigned long rcutorture_testseq; | |
144 | unsigned long rcutorture_vernum; | |
145 | ||
fc2219d4 PM |
146 | /* |
147 | * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s | |
148 | * permit this function to be invoked without holding the root rcu_node | |
149 | * structure's ->lock, but of course results can be subject to change. | |
150 | */ | |
151 | static int rcu_gp_in_progress(struct rcu_state *rsp) | |
152 | { | |
153 | return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum); | |
154 | } | |
155 | ||
b1f77b05 | 156 | /* |
d6714c22 | 157 | * Note a quiescent state. Because we do not need to know |
b1f77b05 | 158 | * how many quiescent states passed, just if there was at least |
d6714c22 | 159 | * one since the start of the grace period, this just sets a flag. |
e4cc1f22 | 160 | * The caller must have disabled preemption. |
b1f77b05 | 161 | */ |
d6714c22 | 162 | void rcu_sched_qs(int cpu) |
b1f77b05 | 163 | { |
25502a6c | 164 | struct rcu_data *rdp = &per_cpu(rcu_sched_data, cpu); |
f41d911f | 165 | |
e4cc1f22 | 166 | rdp->passed_quiesce_gpnum = rdp->gpnum; |
c3422bea | 167 | barrier(); |
e4cc1f22 | 168 | if (rdp->passed_quiesce == 0) |
d4c08f2a | 169 | trace_rcu_grace_period("rcu_sched", rdp->gpnum, "cpuqs"); |
e4cc1f22 | 170 | rdp->passed_quiesce = 1; |
b1f77b05 IM |
171 | } |
172 | ||
d6714c22 | 173 | void rcu_bh_qs(int cpu) |
b1f77b05 | 174 | { |
25502a6c | 175 | struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu); |
f41d911f | 176 | |
e4cc1f22 | 177 | rdp->passed_quiesce_gpnum = rdp->gpnum; |
c3422bea | 178 | barrier(); |
e4cc1f22 | 179 | if (rdp->passed_quiesce == 0) |
d4c08f2a | 180 | trace_rcu_grace_period("rcu_bh", rdp->gpnum, "cpuqs"); |
e4cc1f22 | 181 | rdp->passed_quiesce = 1; |
b1f77b05 | 182 | } |
64db4cff | 183 | |
25502a6c PM |
184 | /* |
185 | * Note a context switch. This is a quiescent state for RCU-sched, | |
186 | * and requires special handling for preemptible RCU. | |
e4cc1f22 | 187 | * The caller must have disabled preemption. |
25502a6c PM |
188 | */ |
189 | void rcu_note_context_switch(int cpu) | |
190 | { | |
300df91c | 191 | trace_rcu_utilization("Start context switch"); |
25502a6c PM |
192 | rcu_sched_qs(cpu); |
193 | rcu_preempt_note_context_switch(cpu); | |
300df91c | 194 | trace_rcu_utilization("End context switch"); |
25502a6c | 195 | } |
29ce8310 | 196 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
25502a6c | 197 | |
90a4d2c0 | 198 | DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { |
4145fa7f | 199 | .dynticks_nesting = DYNTICK_TASK_NESTING, |
23b5c8fa | 200 | .dynticks = ATOMIC_INIT(1), |
90a4d2c0 | 201 | }; |
64db4cff | 202 | |
e0f23060 | 203 | static int blimit = 10; /* Maximum callbacks per rcu_do_batch. */ |
64db4cff PM |
204 | static int qhimark = 10000; /* If this many pending, ignore blimit. */ |
205 | static int qlowmark = 100; /* Once only this many pending, use blimit. */ | |
206 | ||
3d76c082 PM |
207 | module_param(blimit, int, 0); |
208 | module_param(qhimark, int, 0); | |
209 | module_param(qlowmark, int, 0); | |
210 | ||
a00e0d71 | 211 | int rcu_cpu_stall_suppress __read_mostly; |
f2e0dd70 | 212 | module_param(rcu_cpu_stall_suppress, int, 0644); |
742734ee | 213 | |
64db4cff | 214 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed); |
a157229c | 215 | static int rcu_pending(int cpu); |
64db4cff PM |
216 | |
217 | /* | |
d6714c22 | 218 | * Return the number of RCU-sched batches processed thus far for debug & stats. |
64db4cff | 219 | */ |
d6714c22 | 220 | long rcu_batches_completed_sched(void) |
64db4cff | 221 | { |
d6714c22 | 222 | return rcu_sched_state.completed; |
64db4cff | 223 | } |
d6714c22 | 224 | EXPORT_SYMBOL_GPL(rcu_batches_completed_sched); |
64db4cff PM |
225 | |
226 | /* | |
227 | * Return the number of RCU BH batches processed thus far for debug & stats. | |
228 | */ | |
229 | long rcu_batches_completed_bh(void) | |
230 | { | |
231 | return rcu_bh_state.completed; | |
232 | } | |
233 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); | |
234 | ||
bf66f18e PM |
235 | /* |
236 | * Force a quiescent state for RCU BH. | |
237 | */ | |
238 | void rcu_bh_force_quiescent_state(void) | |
239 | { | |
240 | force_quiescent_state(&rcu_bh_state, 0); | |
241 | } | |
242 | EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state); | |
243 | ||
4a298656 PM |
244 | /* |
245 | * Record the number of times rcutorture tests have been initiated and | |
246 | * terminated. This information allows the debugfs tracing stats to be | |
247 | * correlated to the rcutorture messages, even when the rcutorture module | |
248 | * is being repeatedly loaded and unloaded. In other words, we cannot | |
249 | * store this state in rcutorture itself. | |
250 | */ | |
251 | void rcutorture_record_test_transition(void) | |
252 | { | |
253 | rcutorture_testseq++; | |
254 | rcutorture_vernum = 0; | |
255 | } | |
256 | EXPORT_SYMBOL_GPL(rcutorture_record_test_transition); | |
257 | ||
258 | /* | |
259 | * Record the number of writer passes through the current rcutorture test. | |
260 | * This is also used to correlate debugfs tracing stats with the rcutorture | |
261 | * messages. | |
262 | */ | |
263 | void rcutorture_record_progress(unsigned long vernum) | |
264 | { | |
265 | rcutorture_vernum++; | |
266 | } | |
267 | EXPORT_SYMBOL_GPL(rcutorture_record_progress); | |
268 | ||
bf66f18e PM |
269 | /* |
270 | * Force a quiescent state for RCU-sched. | |
271 | */ | |
272 | void rcu_sched_force_quiescent_state(void) | |
273 | { | |
274 | force_quiescent_state(&rcu_sched_state, 0); | |
275 | } | |
276 | EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state); | |
277 | ||
64db4cff PM |
278 | /* |
279 | * Does the CPU have callbacks ready to be invoked? | |
280 | */ | |
281 | static int | |
282 | cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) | |
283 | { | |
284 | return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]; | |
285 | } | |
286 | ||
287 | /* | |
288 | * Does the current CPU require a yet-as-unscheduled grace period? | |
289 | */ | |
290 | static int | |
291 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) | |
292 | { | |
fc2219d4 | 293 | return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp); |
64db4cff PM |
294 | } |
295 | ||
296 | /* | |
297 | * Return the root node of the specified rcu_state structure. | |
298 | */ | |
299 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) | |
300 | { | |
301 | return &rsp->node[0]; | |
302 | } | |
303 | ||
304 | #ifdef CONFIG_SMP | |
305 | ||
306 | /* | |
307 | * If the specified CPU is offline, tell the caller that it is in | |
308 | * a quiescent state. Otherwise, whack it with a reschedule IPI. | |
309 | * Grace periods can end up waiting on an offline CPU when that | |
310 | * CPU is in the process of coming online -- it will be added to the | |
311 | * rcu_node bitmasks before it actually makes it online. The same thing | |
312 | * can happen while a CPU is in the process of coming online. Because this | |
313 | * race is quite rare, we check for it after detecting that the grace | |
314 | * period has been delayed rather than checking each and every CPU | |
315 | * each and every time we start a new grace period. | |
316 | */ | |
317 | static int rcu_implicit_offline_qs(struct rcu_data *rdp) | |
318 | { | |
319 | /* | |
320 | * If the CPU is offline, it is in a quiescent state. We can | |
321 | * trust its state not to change because interrupts are disabled. | |
322 | */ | |
323 | if (cpu_is_offline(rdp->cpu)) { | |
d4c08f2a | 324 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, "ofl"); |
64db4cff PM |
325 | rdp->offline_fqs++; |
326 | return 1; | |
327 | } | |
328 | ||
9b2e4f18 PM |
329 | /* |
330 | * The CPU is online, so send it a reschedule IPI. This forces | |
331 | * it through the scheduler, and (inefficiently) also handles cases | |
332 | * where idle loops fail to inform RCU about the CPU being idle. | |
333 | */ | |
64db4cff PM |
334 | if (rdp->cpu != smp_processor_id()) |
335 | smp_send_reschedule(rdp->cpu); | |
336 | else | |
337 | set_need_resched(); | |
338 | rdp->resched_ipi++; | |
339 | return 0; | |
340 | } | |
341 | ||
342 | #endif /* #ifdef CONFIG_SMP */ | |
343 | ||
9b2e4f18 PM |
344 | /* |
345 | * rcu_idle_enter_common - inform RCU that current CPU is moving towards idle | |
346 | * | |
347 | * If the new value of the ->dynticks_nesting counter now is zero, | |
348 | * we really have entered idle, and must do the appropriate accounting. | |
349 | * The caller must have disabled interrupts. | |
350 | */ | |
4145fa7f | 351 | static void rcu_idle_enter_common(struct rcu_dynticks *rdtp, long long oldval) |
9b2e4f18 PM |
352 | { |
353 | if (rdtp->dynticks_nesting) { | |
4145fa7f | 354 | trace_rcu_dyntick("--=", oldval, rdtp->dynticks_nesting); |
9b2e4f18 PM |
355 | return; |
356 | } | |
4145fa7f | 357 | trace_rcu_dyntick("Start", oldval, rdtp->dynticks_nesting); |
9b2e4f18 PM |
358 | if (!idle_cpu(smp_processor_id())) { |
359 | WARN_ON_ONCE(1); /* must be idle task! */ | |
360 | trace_rcu_dyntick("Error on entry: not idle task", | |
4145fa7f | 361 | oldval, rdtp->dynticks_nesting); |
9b2e4f18 PM |
362 | ftrace_dump(DUMP_ALL); |
363 | } | |
364 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ | |
365 | smp_mb__before_atomic_inc(); /* See above. */ | |
366 | atomic_inc(&rdtp->dynticks); | |
367 | smp_mb__after_atomic_inc(); /* Force ordering with next sojourn. */ | |
368 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); | |
369 | } | |
64db4cff PM |
370 | |
371 | /** | |
9b2e4f18 | 372 | * rcu_idle_enter - inform RCU that current CPU is entering idle |
64db4cff | 373 | * |
9b2e4f18 | 374 | * Enter idle mode, in other words, -leave- the mode in which RCU |
64db4cff | 375 | * read-side critical sections can occur. (Though RCU read-side |
9b2e4f18 PM |
376 | * critical sections can occur in irq handlers in idle, a possibility |
377 | * handled by irq_enter() and irq_exit().) | |
378 | * | |
379 | * We crowbar the ->dynticks_nesting field to zero to allow for | |
380 | * the possibility of usermode upcalls having messed up our count | |
381 | * of interrupt nesting level during the prior busy period. | |
64db4cff | 382 | */ |
9b2e4f18 | 383 | void rcu_idle_enter(void) |
64db4cff PM |
384 | { |
385 | unsigned long flags; | |
4145fa7f | 386 | long long oldval; |
64db4cff PM |
387 | struct rcu_dynticks *rdtp; |
388 | ||
64db4cff PM |
389 | local_irq_save(flags); |
390 | rdtp = &__get_cpu_var(rcu_dynticks); | |
4145fa7f | 391 | oldval = rdtp->dynticks_nesting; |
9b2e4f18 | 392 | rdtp->dynticks_nesting = 0; |
4145fa7f | 393 | rcu_idle_enter_common(rdtp, oldval); |
64db4cff PM |
394 | local_irq_restore(flags); |
395 | } | |
396 | ||
9b2e4f18 PM |
397 | /** |
398 | * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle | |
399 | * | |
400 | * Exit from an interrupt handler, which might possibly result in entering | |
401 | * idle mode, in other words, leaving the mode in which read-side critical | |
402 | * sections can occur. | |
64db4cff | 403 | * |
9b2e4f18 PM |
404 | * This code assumes that the idle loop never does anything that might |
405 | * result in unbalanced calls to irq_enter() and irq_exit(). If your | |
406 | * architecture violates this assumption, RCU will give you what you | |
407 | * deserve, good and hard. But very infrequently and irreproducibly. | |
408 | * | |
409 | * Use things like work queues to work around this limitation. | |
410 | * | |
411 | * You have been warned. | |
64db4cff | 412 | */ |
9b2e4f18 | 413 | void rcu_irq_exit(void) |
64db4cff PM |
414 | { |
415 | unsigned long flags; | |
4145fa7f | 416 | long long oldval; |
64db4cff PM |
417 | struct rcu_dynticks *rdtp; |
418 | ||
419 | local_irq_save(flags); | |
420 | rdtp = &__get_cpu_var(rcu_dynticks); | |
4145fa7f | 421 | oldval = rdtp->dynticks_nesting; |
9b2e4f18 PM |
422 | rdtp->dynticks_nesting--; |
423 | WARN_ON_ONCE(rdtp->dynticks_nesting < 0); | |
4145fa7f | 424 | rcu_idle_enter_common(rdtp, oldval); |
9b2e4f18 PM |
425 | local_irq_restore(flags); |
426 | } | |
427 | ||
428 | /* | |
429 | * rcu_idle_exit_common - inform RCU that current CPU is moving away from idle | |
430 | * | |
431 | * If the new value of the ->dynticks_nesting counter was previously zero, | |
432 | * we really have exited idle, and must do the appropriate accounting. | |
433 | * The caller must have disabled interrupts. | |
434 | */ | |
435 | static void rcu_idle_exit_common(struct rcu_dynticks *rdtp, long long oldval) | |
436 | { | |
437 | if (oldval) { | |
4145fa7f | 438 | trace_rcu_dyntick("++=", oldval, rdtp->dynticks_nesting); |
23b5c8fa PM |
439 | return; |
440 | } | |
441 | smp_mb__before_atomic_inc(); /* Force ordering w/previous sojourn. */ | |
442 | atomic_inc(&rdtp->dynticks); | |
443 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ | |
444 | smp_mb__after_atomic_inc(); /* See above. */ | |
445 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); | |
4145fa7f | 446 | trace_rcu_dyntick("End", oldval, rdtp->dynticks_nesting); |
9b2e4f18 PM |
447 | if (!idle_cpu(smp_processor_id())) { |
448 | WARN_ON_ONCE(1); /* must be idle task! */ | |
4145fa7f PM |
449 | trace_rcu_dyntick("Error on exit: not idle task", |
450 | oldval, rdtp->dynticks_nesting); | |
9b2e4f18 PM |
451 | ftrace_dump(DUMP_ALL); |
452 | } | |
453 | } | |
454 | ||
455 | /** | |
456 | * rcu_idle_exit - inform RCU that current CPU is leaving idle | |
457 | * | |
458 | * Exit idle mode, in other words, -enter- the mode in which RCU | |
459 | * read-side critical sections can occur. | |
460 | * | |
4145fa7f PM |
461 | * We crowbar the ->dynticks_nesting field to DYNTICK_TASK_NESTING to |
462 | * allow for the possibility of usermode upcalls messing up our count | |
9b2e4f18 PM |
463 | * of interrupt nesting level during the busy period that is just |
464 | * now starting. | |
465 | */ | |
466 | void rcu_idle_exit(void) | |
467 | { | |
468 | unsigned long flags; | |
469 | struct rcu_dynticks *rdtp; | |
470 | long long oldval; | |
471 | ||
472 | local_irq_save(flags); | |
473 | rdtp = &__get_cpu_var(rcu_dynticks); | |
474 | oldval = rdtp->dynticks_nesting; | |
475 | WARN_ON_ONCE(oldval != 0); | |
4145fa7f | 476 | rdtp->dynticks_nesting = DYNTICK_TASK_NESTING; |
9b2e4f18 PM |
477 | rcu_idle_exit_common(rdtp, oldval); |
478 | local_irq_restore(flags); | |
479 | } | |
480 | ||
481 | /** | |
482 | * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle | |
483 | * | |
484 | * Enter an interrupt handler, which might possibly result in exiting | |
485 | * idle mode, in other words, entering the mode in which read-side critical | |
486 | * sections can occur. | |
487 | * | |
488 | * Note that the Linux kernel is fully capable of entering an interrupt | |
489 | * handler that it never exits, for example when doing upcalls to | |
490 | * user mode! This code assumes that the idle loop never does upcalls to | |
491 | * user mode. If your architecture does do upcalls from the idle loop (or | |
492 | * does anything else that results in unbalanced calls to the irq_enter() | |
493 | * and irq_exit() functions), RCU will give you what you deserve, good | |
494 | * and hard. But very infrequently and irreproducibly. | |
495 | * | |
496 | * Use things like work queues to work around this limitation. | |
497 | * | |
498 | * You have been warned. | |
499 | */ | |
500 | void rcu_irq_enter(void) | |
501 | { | |
502 | unsigned long flags; | |
503 | struct rcu_dynticks *rdtp; | |
504 | long long oldval; | |
505 | ||
506 | local_irq_save(flags); | |
507 | rdtp = &__get_cpu_var(rcu_dynticks); | |
508 | oldval = rdtp->dynticks_nesting; | |
509 | rdtp->dynticks_nesting++; | |
510 | WARN_ON_ONCE(rdtp->dynticks_nesting == 0); | |
511 | rcu_idle_exit_common(rdtp, oldval); | |
64db4cff | 512 | local_irq_restore(flags); |
64db4cff PM |
513 | } |
514 | ||
515 | /** | |
516 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
517 | * | |
518 | * If the CPU was idle with dynamic ticks active, and there is no | |
519 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
520 | * RCU grace-period handling know that the CPU is active. | |
521 | */ | |
522 | void rcu_nmi_enter(void) | |
523 | { | |
524 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
525 | ||
23b5c8fa PM |
526 | if (rdtp->dynticks_nmi_nesting == 0 && |
527 | (atomic_read(&rdtp->dynticks) & 0x1)) | |
64db4cff | 528 | return; |
23b5c8fa PM |
529 | rdtp->dynticks_nmi_nesting++; |
530 | smp_mb__before_atomic_inc(); /* Force delay from prior write. */ | |
531 | atomic_inc(&rdtp->dynticks); | |
532 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ | |
533 | smp_mb__after_atomic_inc(); /* See above. */ | |
534 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); | |
64db4cff PM |
535 | } |
536 | ||
537 | /** | |
538 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
539 | * | |
540 | * If the CPU was idle with dynamic ticks active, and there is no | |
541 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
542 | * RCU grace-period handling know that the CPU is no longer active. | |
543 | */ | |
544 | void rcu_nmi_exit(void) | |
545 | { | |
546 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
547 | ||
23b5c8fa PM |
548 | if (rdtp->dynticks_nmi_nesting == 0 || |
549 | --rdtp->dynticks_nmi_nesting != 0) | |
64db4cff | 550 | return; |
23b5c8fa PM |
551 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ |
552 | smp_mb__before_atomic_inc(); /* See above. */ | |
553 | atomic_inc(&rdtp->dynticks); | |
554 | smp_mb__after_atomic_inc(); /* Force delay to next write. */ | |
555 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); | |
64db4cff PM |
556 | } |
557 | ||
9b2e4f18 PM |
558 | #ifdef CONFIG_PROVE_RCU |
559 | ||
64db4cff | 560 | /** |
9b2e4f18 | 561 | * rcu_is_cpu_idle - see if RCU thinks that the current CPU is idle |
64db4cff | 562 | * |
9b2e4f18 | 563 | * If the current CPU is in its idle loop and is neither in an interrupt |
34240697 | 564 | * or NMI handler, return true. |
64db4cff | 565 | */ |
9b2e4f18 | 566 | int rcu_is_cpu_idle(void) |
64db4cff | 567 | { |
34240697 PM |
568 | int ret; |
569 | ||
570 | preempt_disable(); | |
571 | ret = (atomic_read(&__get_cpu_var(rcu_dynticks).dynticks) & 0x1) == 0; | |
572 | preempt_enable(); | |
573 | return ret; | |
64db4cff | 574 | } |
e6b80a3b | 575 | EXPORT_SYMBOL(rcu_is_cpu_idle); |
64db4cff | 576 | |
9b2e4f18 PM |
577 | #endif /* #ifdef CONFIG_PROVE_RCU */ |
578 | ||
64db4cff | 579 | /** |
9b2e4f18 | 580 | * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle |
64db4cff | 581 | * |
9b2e4f18 PM |
582 | * If the current CPU is idle or running at a first-level (not nested) |
583 | * interrupt from idle, return true. The caller must have at least | |
584 | * disabled preemption. | |
64db4cff | 585 | */ |
9b2e4f18 | 586 | int rcu_is_cpu_rrupt_from_idle(void) |
64db4cff | 587 | { |
9b2e4f18 | 588 | return __get_cpu_var(rcu_dynticks).dynticks_nesting <= 1; |
64db4cff PM |
589 | } |
590 | ||
64db4cff PM |
591 | #ifdef CONFIG_SMP |
592 | ||
64db4cff PM |
593 | /* |
594 | * Snapshot the specified CPU's dynticks counter so that we can later | |
595 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 596 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff PM |
597 | */ |
598 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | |
599 | { | |
23b5c8fa PM |
600 | rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks); |
601 | return 0; | |
64db4cff PM |
602 | } |
603 | ||
604 | /* | |
605 | * Return true if the specified CPU has passed through a quiescent | |
606 | * state by virtue of being in or having passed through an dynticks | |
607 | * idle state since the last call to dyntick_save_progress_counter() | |
608 | * for this same CPU. | |
609 | */ | |
610 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
611 | { | |
7eb4f455 PM |
612 | unsigned int curr; |
613 | unsigned int snap; | |
64db4cff | 614 | |
7eb4f455 PM |
615 | curr = (unsigned int)atomic_add_return(0, &rdp->dynticks->dynticks); |
616 | snap = (unsigned int)rdp->dynticks_snap; | |
64db4cff PM |
617 | |
618 | /* | |
619 | * If the CPU passed through or entered a dynticks idle phase with | |
620 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
621 | * already acknowledged the request to pass through a quiescent | |
622 | * state. Either way, that CPU cannot possibly be in an RCU | |
623 | * read-side critical section that started before the beginning | |
624 | * of the current RCU grace period. | |
625 | */ | |
7eb4f455 | 626 | if ((curr & 0x1) == 0 || UINT_CMP_GE(curr, snap + 2)) { |
d4c08f2a | 627 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, "dti"); |
64db4cff PM |
628 | rdp->dynticks_fqs++; |
629 | return 1; | |
630 | } | |
631 | ||
632 | /* Go check for the CPU being offline. */ | |
633 | return rcu_implicit_offline_qs(rdp); | |
634 | } | |
635 | ||
636 | #endif /* #ifdef CONFIG_SMP */ | |
637 | ||
742734ee | 638 | int rcu_cpu_stall_suppress __read_mostly; |
c68de209 | 639 | |
64db4cff PM |
640 | static void record_gp_stall_check_time(struct rcu_state *rsp) |
641 | { | |
642 | rsp->gp_start = jiffies; | |
643 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK; | |
644 | } | |
645 | ||
646 | static void print_other_cpu_stall(struct rcu_state *rsp) | |
647 | { | |
648 | int cpu; | |
649 | long delta; | |
650 | unsigned long flags; | |
9bc8b558 | 651 | int ndetected; |
64db4cff | 652 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff PM |
653 | |
654 | /* Only let one CPU complain about others per time interval. */ | |
655 | ||
1304afb2 | 656 | raw_spin_lock_irqsave(&rnp->lock, flags); |
64db4cff | 657 | delta = jiffies - rsp->jiffies_stall; |
fc2219d4 | 658 | if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) { |
1304afb2 | 659 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
660 | return; |
661 | } | |
662 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
a0b6c9a7 PM |
663 | |
664 | /* | |
665 | * Now rat on any tasks that got kicked up to the root rcu_node | |
666 | * due to CPU offlining. | |
667 | */ | |
9bc8b558 | 668 | ndetected = rcu_print_task_stall(rnp); |
1304afb2 | 669 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 670 | |
8cdd32a9 PM |
671 | /* |
672 | * OK, time to rat on our buddy... | |
673 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
674 | * RCU CPU stall warnings. | |
675 | */ | |
4300aa64 PM |
676 | printk(KERN_ERR "INFO: %s detected stalls on CPUs/tasks: {", |
677 | rsp->name); | |
a0b6c9a7 | 678 | rcu_for_each_leaf_node(rsp, rnp) { |
3acd9eb3 | 679 | raw_spin_lock_irqsave(&rnp->lock, flags); |
9bc8b558 | 680 | ndetected += rcu_print_task_stall(rnp); |
3acd9eb3 | 681 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
a0b6c9a7 | 682 | if (rnp->qsmask == 0) |
64db4cff | 683 | continue; |
a0b6c9a7 | 684 | for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++) |
9bc8b558 | 685 | if (rnp->qsmask & (1UL << cpu)) { |
a0b6c9a7 | 686 | printk(" %d", rnp->grplo + cpu); |
9bc8b558 PM |
687 | ndetected++; |
688 | } | |
64db4cff | 689 | } |
4300aa64 | 690 | printk("} (detected by %d, t=%ld jiffies)\n", |
64db4cff | 691 | smp_processor_id(), (long)(jiffies - rsp->gp_start)); |
9bc8b558 PM |
692 | if (ndetected == 0) |
693 | printk(KERN_ERR "INFO: Stall ended before state dump start\n"); | |
694 | else if (!trigger_all_cpu_backtrace()) | |
4627e240 | 695 | dump_stack(); |
c1dc0b9c | 696 | |
1ed509a2 PM |
697 | /* If so configured, complain about tasks blocking the grace period. */ |
698 | ||
699 | rcu_print_detail_task_stall(rsp); | |
700 | ||
64db4cff PM |
701 | force_quiescent_state(rsp, 0); /* Kick them all. */ |
702 | } | |
703 | ||
704 | static void print_cpu_stall(struct rcu_state *rsp) | |
705 | { | |
706 | unsigned long flags; | |
707 | struct rcu_node *rnp = rcu_get_root(rsp); | |
708 | ||
8cdd32a9 PM |
709 | /* |
710 | * OK, time to rat on ourselves... | |
711 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
712 | * RCU CPU stall warnings. | |
713 | */ | |
4300aa64 PM |
714 | printk(KERN_ERR "INFO: %s detected stall on CPU %d (t=%lu jiffies)\n", |
715 | rsp->name, smp_processor_id(), jiffies - rsp->gp_start); | |
4627e240 PM |
716 | if (!trigger_all_cpu_backtrace()) |
717 | dump_stack(); | |
c1dc0b9c | 718 | |
1304afb2 | 719 | raw_spin_lock_irqsave(&rnp->lock, flags); |
20133cfc | 720 | if (ULONG_CMP_GE(jiffies, rsp->jiffies_stall)) |
64db4cff PM |
721 | rsp->jiffies_stall = |
722 | jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
1304afb2 | 723 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
c1dc0b9c | 724 | |
64db4cff PM |
725 | set_need_resched(); /* kick ourselves to get things going. */ |
726 | } | |
727 | ||
728 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
729 | { | |
bad6e139 PM |
730 | unsigned long j; |
731 | unsigned long js; | |
64db4cff PM |
732 | struct rcu_node *rnp; |
733 | ||
742734ee | 734 | if (rcu_cpu_stall_suppress) |
c68de209 | 735 | return; |
bad6e139 PM |
736 | j = ACCESS_ONCE(jiffies); |
737 | js = ACCESS_ONCE(rsp->jiffies_stall); | |
64db4cff | 738 | rnp = rdp->mynode; |
bad6e139 | 739 | if ((ACCESS_ONCE(rnp->qsmask) & rdp->grpmask) && ULONG_CMP_GE(j, js)) { |
64db4cff PM |
740 | |
741 | /* We haven't checked in, so go dump stack. */ | |
742 | print_cpu_stall(rsp); | |
743 | ||
bad6e139 PM |
744 | } else if (rcu_gp_in_progress(rsp) && |
745 | ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) { | |
64db4cff | 746 | |
bad6e139 | 747 | /* They had a few time units to dump stack, so complain. */ |
64db4cff PM |
748 | print_other_cpu_stall(rsp); |
749 | } | |
750 | } | |
751 | ||
c68de209 PM |
752 | static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr) |
753 | { | |
742734ee | 754 | rcu_cpu_stall_suppress = 1; |
c68de209 PM |
755 | return NOTIFY_DONE; |
756 | } | |
757 | ||
53d84e00 PM |
758 | /** |
759 | * rcu_cpu_stall_reset - prevent further stall warnings in current grace period | |
760 | * | |
761 | * Set the stall-warning timeout way off into the future, thus preventing | |
762 | * any RCU CPU stall-warning messages from appearing in the current set of | |
763 | * RCU grace periods. | |
764 | * | |
765 | * The caller must disable hard irqs. | |
766 | */ | |
767 | void rcu_cpu_stall_reset(void) | |
768 | { | |
769 | rcu_sched_state.jiffies_stall = jiffies + ULONG_MAX / 2; | |
770 | rcu_bh_state.jiffies_stall = jiffies + ULONG_MAX / 2; | |
771 | rcu_preempt_stall_reset(); | |
772 | } | |
773 | ||
c68de209 PM |
774 | static struct notifier_block rcu_panic_block = { |
775 | .notifier_call = rcu_panic, | |
776 | }; | |
777 | ||
778 | static void __init check_cpu_stall_init(void) | |
779 | { | |
780 | atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block); | |
781 | } | |
782 | ||
64db4cff PM |
783 | /* |
784 | * Update CPU-local rcu_data state to record the newly noticed grace period. | |
785 | * This is used both when we started the grace period and when we notice | |
9160306e PM |
786 | * that someone else started the grace period. The caller must hold the |
787 | * ->lock of the leaf rcu_node structure corresponding to the current CPU, | |
788 | * and must have irqs disabled. | |
64db4cff | 789 | */ |
9160306e PM |
790 | static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) |
791 | { | |
792 | if (rdp->gpnum != rnp->gpnum) { | |
121dfc4b PM |
793 | /* |
794 | * If the current grace period is waiting for this CPU, | |
795 | * set up to detect a quiescent state, otherwise don't | |
796 | * go looking for one. | |
797 | */ | |
9160306e | 798 | rdp->gpnum = rnp->gpnum; |
d4c08f2a | 799 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpustart"); |
121dfc4b PM |
800 | if (rnp->qsmask & rdp->grpmask) { |
801 | rdp->qs_pending = 1; | |
e4cc1f22 | 802 | rdp->passed_quiesce = 0; |
121dfc4b PM |
803 | } else |
804 | rdp->qs_pending = 0; | |
9160306e PM |
805 | } |
806 | } | |
807 | ||
64db4cff PM |
808 | static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp) |
809 | { | |
9160306e PM |
810 | unsigned long flags; |
811 | struct rcu_node *rnp; | |
812 | ||
813 | local_irq_save(flags); | |
814 | rnp = rdp->mynode; | |
815 | if (rdp->gpnum == ACCESS_ONCE(rnp->gpnum) || /* outside lock. */ | |
1304afb2 | 816 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
9160306e PM |
817 | local_irq_restore(flags); |
818 | return; | |
819 | } | |
820 | __note_new_gpnum(rsp, rnp, rdp); | |
1304afb2 | 821 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
822 | } |
823 | ||
824 | /* | |
825 | * Did someone else start a new RCU grace period start since we last | |
826 | * checked? Update local state appropriately if so. Must be called | |
827 | * on the CPU corresponding to rdp. | |
828 | */ | |
829 | static int | |
830 | check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp) | |
831 | { | |
832 | unsigned long flags; | |
833 | int ret = 0; | |
834 | ||
835 | local_irq_save(flags); | |
836 | if (rdp->gpnum != rsp->gpnum) { | |
837 | note_new_gpnum(rsp, rdp); | |
838 | ret = 1; | |
839 | } | |
840 | local_irq_restore(flags); | |
841 | return ret; | |
842 | } | |
843 | ||
d09b62df PM |
844 | /* |
845 | * Advance this CPU's callbacks, but only if the current grace period | |
846 | * has ended. This may be called only from the CPU to whom the rdp | |
847 | * belongs. In addition, the corresponding leaf rcu_node structure's | |
848 | * ->lock must be held by the caller, with irqs disabled. | |
849 | */ | |
850 | static void | |
851 | __rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
852 | { | |
853 | /* Did another grace period end? */ | |
854 | if (rdp->completed != rnp->completed) { | |
855 | ||
856 | /* Advance callbacks. No harm if list empty. */ | |
857 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL]; | |
858 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL]; | |
859 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
860 | ||
861 | /* Remember that we saw this grace-period completion. */ | |
862 | rdp->completed = rnp->completed; | |
d4c08f2a | 863 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuend"); |
20377f32 | 864 | |
5ff8e6f0 FW |
865 | /* |
866 | * If we were in an extended quiescent state, we may have | |
121dfc4b | 867 | * missed some grace periods that others CPUs handled on |
5ff8e6f0 | 868 | * our behalf. Catch up with this state to avoid noting |
121dfc4b PM |
869 | * spurious new grace periods. If another grace period |
870 | * has started, then rnp->gpnum will have advanced, so | |
871 | * we will detect this later on. | |
5ff8e6f0 | 872 | */ |
121dfc4b | 873 | if (ULONG_CMP_LT(rdp->gpnum, rdp->completed)) |
5ff8e6f0 FW |
874 | rdp->gpnum = rdp->completed; |
875 | ||
20377f32 | 876 | /* |
121dfc4b PM |
877 | * If RCU does not need a quiescent state from this CPU, |
878 | * then make sure that this CPU doesn't go looking for one. | |
20377f32 | 879 | */ |
121dfc4b | 880 | if ((rnp->qsmask & rdp->grpmask) == 0) |
20377f32 | 881 | rdp->qs_pending = 0; |
d09b62df PM |
882 | } |
883 | } | |
884 | ||
885 | /* | |
886 | * Advance this CPU's callbacks, but only if the current grace period | |
887 | * has ended. This may be called only from the CPU to whom the rdp | |
888 | * belongs. | |
889 | */ | |
890 | static void | |
891 | rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp) | |
892 | { | |
893 | unsigned long flags; | |
894 | struct rcu_node *rnp; | |
895 | ||
896 | local_irq_save(flags); | |
897 | rnp = rdp->mynode; | |
898 | if (rdp->completed == ACCESS_ONCE(rnp->completed) || /* outside lock. */ | |
1304afb2 | 899 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
d09b62df PM |
900 | local_irq_restore(flags); |
901 | return; | |
902 | } | |
903 | __rcu_process_gp_end(rsp, rnp, rdp); | |
1304afb2 | 904 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
d09b62df PM |
905 | } |
906 | ||
907 | /* | |
908 | * Do per-CPU grace-period initialization for running CPU. The caller | |
909 | * must hold the lock of the leaf rcu_node structure corresponding to | |
910 | * this CPU. | |
911 | */ | |
912 | static void | |
913 | rcu_start_gp_per_cpu(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
914 | { | |
915 | /* Prior grace period ended, so advance callbacks for current CPU. */ | |
916 | __rcu_process_gp_end(rsp, rnp, rdp); | |
917 | ||
918 | /* | |
919 | * Because this CPU just now started the new grace period, we know | |
920 | * that all of its callbacks will be covered by this upcoming grace | |
921 | * period, even the ones that were registered arbitrarily recently. | |
922 | * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL. | |
923 | * | |
924 | * Other CPUs cannot be sure exactly when the grace period started. | |
925 | * Therefore, their recently registered callbacks must pass through | |
926 | * an additional RCU_NEXT_READY stage, so that they will be handled | |
927 | * by the next RCU grace period. | |
928 | */ | |
929 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
930 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
9160306e PM |
931 | |
932 | /* Set state so that this CPU will detect the next quiescent state. */ | |
933 | __note_new_gpnum(rsp, rnp, rdp); | |
d09b62df PM |
934 | } |
935 | ||
64db4cff PM |
936 | /* |
937 | * Start a new RCU grace period if warranted, re-initializing the hierarchy | |
938 | * in preparation for detecting the next grace period. The caller must hold | |
939 | * the root node's ->lock, which is released before return. Hard irqs must | |
940 | * be disabled. | |
941 | */ | |
942 | static void | |
943 | rcu_start_gp(struct rcu_state *rsp, unsigned long flags) | |
944 | __releases(rcu_get_root(rsp)->lock) | |
945 | { | |
394f99a9 | 946 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
64db4cff | 947 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff | 948 | |
037067a1 | 949 | if (!rcu_scheduler_fully_active || |
afe24b12 PM |
950 | !cpu_needs_another_gp(rsp, rdp)) { |
951 | /* | |
952 | * Either the scheduler hasn't yet spawned the first | |
953 | * non-idle task or this CPU does not need another | |
954 | * grace period. Either way, don't start a new grace | |
955 | * period. | |
956 | */ | |
957 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
958 | return; | |
959 | } | |
b32e9eb6 | 960 | |
afe24b12 | 961 | if (rsp->fqs_active) { |
b32e9eb6 | 962 | /* |
afe24b12 PM |
963 | * This CPU needs a grace period, but force_quiescent_state() |
964 | * is running. Tell it to start one on this CPU's behalf. | |
b32e9eb6 | 965 | */ |
afe24b12 PM |
966 | rsp->fqs_need_gp = 1; |
967 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
64db4cff PM |
968 | return; |
969 | } | |
970 | ||
971 | /* Advance to a new grace period and initialize state. */ | |
972 | rsp->gpnum++; | |
d4c08f2a | 973 | trace_rcu_grace_period(rsp->name, rsp->gpnum, "start"); |
af446b70 PM |
974 | WARN_ON_ONCE(rsp->fqs_state == RCU_GP_INIT); |
975 | rsp->fqs_state = RCU_GP_INIT; /* Hold off force_quiescent_state. */ | |
64db4cff | 976 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
64db4cff | 977 | record_gp_stall_check_time(rsp); |
64db4cff | 978 | |
64db4cff PM |
979 | /* Special-case the common single-level case. */ |
980 | if (NUM_RCU_NODES == 1) { | |
b0e165c0 | 981 | rcu_preempt_check_blocked_tasks(rnp); |
28ecd580 | 982 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 983 | rnp->gpnum = rsp->gpnum; |
d09b62df | 984 | rnp->completed = rsp->completed; |
af446b70 | 985 | rsp->fqs_state = RCU_SIGNAL_INIT; /* force_quiescent_state OK */ |
d09b62df | 986 | rcu_start_gp_per_cpu(rsp, rnp, rdp); |
27f4d280 | 987 | rcu_preempt_boost_start_gp(rnp); |
d4c08f2a PM |
988 | trace_rcu_grace_period_init(rsp->name, rnp->gpnum, |
989 | rnp->level, rnp->grplo, | |
990 | rnp->grphi, rnp->qsmask); | |
1304afb2 | 991 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
992 | return; |
993 | } | |
994 | ||
1304afb2 | 995 | raw_spin_unlock(&rnp->lock); /* leave irqs disabled. */ |
64db4cff PM |
996 | |
997 | ||
998 | /* Exclude any concurrent CPU-hotplug operations. */ | |
1304afb2 | 999 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
64db4cff PM |
1000 | |
1001 | /* | |
b835db1f PM |
1002 | * Set the quiescent-state-needed bits in all the rcu_node |
1003 | * structures for all currently online CPUs in breadth-first | |
1004 | * order, starting from the root rcu_node structure. This | |
1005 | * operation relies on the layout of the hierarchy within the | |
1006 | * rsp->node[] array. Note that other CPUs will access only | |
1007 | * the leaves of the hierarchy, which still indicate that no | |
1008 | * grace period is in progress, at least until the corresponding | |
1009 | * leaf node has been initialized. In addition, we have excluded | |
1010 | * CPU-hotplug operations. | |
64db4cff PM |
1011 | * |
1012 | * Note that the grace period cannot complete until we finish | |
1013 | * the initialization process, as there will be at least one | |
1014 | * qsmask bit set in the root node until that time, namely the | |
b835db1f PM |
1015 | * one corresponding to this CPU, due to the fact that we have |
1016 | * irqs disabled. | |
64db4cff | 1017 | */ |
a0b6c9a7 | 1018 | rcu_for_each_node_breadth_first(rsp, rnp) { |
1304afb2 | 1019 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
b0e165c0 | 1020 | rcu_preempt_check_blocked_tasks(rnp); |
49e29126 | 1021 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 1022 | rnp->gpnum = rsp->gpnum; |
d09b62df PM |
1023 | rnp->completed = rsp->completed; |
1024 | if (rnp == rdp->mynode) | |
1025 | rcu_start_gp_per_cpu(rsp, rnp, rdp); | |
27f4d280 | 1026 | rcu_preempt_boost_start_gp(rnp); |
d4c08f2a PM |
1027 | trace_rcu_grace_period_init(rsp->name, rnp->gpnum, |
1028 | rnp->level, rnp->grplo, | |
1029 | rnp->grphi, rnp->qsmask); | |
1304afb2 | 1030 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
1031 | } |
1032 | ||
83f5b01f | 1033 | rnp = rcu_get_root(rsp); |
1304afb2 | 1034 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
af446b70 | 1035 | rsp->fqs_state = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */ |
1304afb2 PM |
1036 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
1037 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); | |
64db4cff PM |
1038 | } |
1039 | ||
f41d911f | 1040 | /* |
d3f6bad3 PM |
1041 | * Report a full set of quiescent states to the specified rcu_state |
1042 | * data structure. This involves cleaning up after the prior grace | |
1043 | * period and letting rcu_start_gp() start up the next grace period | |
1044 | * if one is needed. Note that the caller must hold rnp->lock, as | |
1045 | * required by rcu_start_gp(), which will release it. | |
f41d911f | 1046 | */ |
d3f6bad3 | 1047 | static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags) |
fc2219d4 | 1048 | __releases(rcu_get_root(rsp)->lock) |
f41d911f | 1049 | { |
15ba0ba8 | 1050 | unsigned long gp_duration; |
afe24b12 PM |
1051 | struct rcu_node *rnp = rcu_get_root(rsp); |
1052 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); | |
15ba0ba8 | 1053 | |
fc2219d4 | 1054 | WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); |
0bbcc529 PM |
1055 | |
1056 | /* | |
1057 | * Ensure that all grace-period and pre-grace-period activity | |
1058 | * is seen before the assignment to rsp->completed. | |
1059 | */ | |
1060 | smp_mb(); /* See above block comment. */ | |
15ba0ba8 PM |
1061 | gp_duration = jiffies - rsp->gp_start; |
1062 | if (gp_duration > rsp->gp_max) | |
1063 | rsp->gp_max = gp_duration; | |
afe24b12 PM |
1064 | |
1065 | /* | |
1066 | * We know the grace period is complete, but to everyone else | |
1067 | * it appears to still be ongoing. But it is also the case | |
1068 | * that to everyone else it looks like there is nothing that | |
1069 | * they can do to advance the grace period. It is therefore | |
1070 | * safe for us to drop the lock in order to mark the grace | |
1071 | * period as completed in all of the rcu_node structures. | |
1072 | * | |
1073 | * But if this CPU needs another grace period, it will take | |
1074 | * care of this while initializing the next grace period. | |
1075 | * We use RCU_WAIT_TAIL instead of the usual RCU_DONE_TAIL | |
1076 | * because the callbacks have not yet been advanced: Those | |
1077 | * callbacks are waiting on the grace period that just now | |
1078 | * completed. | |
1079 | */ | |
1080 | if (*rdp->nxttail[RCU_WAIT_TAIL] == NULL) { | |
1081 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ | |
1082 | ||
1083 | /* | |
1084 | * Propagate new ->completed value to rcu_node structures | |
1085 | * so that other CPUs don't have to wait until the start | |
1086 | * of the next grace period to process their callbacks. | |
1087 | */ | |
1088 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
1089 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ | |
1090 | rnp->completed = rsp->gpnum; | |
1091 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ | |
1092 | } | |
1093 | rnp = rcu_get_root(rsp); | |
1094 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ | |
1095 | } | |
1096 | ||
1097 | rsp->completed = rsp->gpnum; /* Declare the grace period complete. */ | |
d4c08f2a | 1098 | trace_rcu_grace_period(rsp->name, rsp->completed, "end"); |
af446b70 | 1099 | rsp->fqs_state = RCU_GP_IDLE; |
f41d911f PM |
1100 | rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */ |
1101 | } | |
1102 | ||
64db4cff | 1103 | /* |
d3f6bad3 PM |
1104 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
1105 | * Allows quiescent states for a group of CPUs to be reported at one go | |
1106 | * to the specified rcu_node structure, though all the CPUs in the group | |
1107 | * must be represented by the same rcu_node structure (which need not be | |
1108 | * a leaf rcu_node structure, though it often will be). That structure's | |
1109 | * lock must be held upon entry, and it is released before return. | |
64db4cff PM |
1110 | */ |
1111 | static void | |
d3f6bad3 PM |
1112 | rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp, |
1113 | struct rcu_node *rnp, unsigned long flags) | |
64db4cff PM |
1114 | __releases(rnp->lock) |
1115 | { | |
28ecd580 PM |
1116 | struct rcu_node *rnp_c; |
1117 | ||
64db4cff PM |
1118 | /* Walk up the rcu_node hierarchy. */ |
1119 | for (;;) { | |
1120 | if (!(rnp->qsmask & mask)) { | |
1121 | ||
1122 | /* Our bit has already been cleared, so done. */ | |
1304afb2 | 1123 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1124 | return; |
1125 | } | |
1126 | rnp->qsmask &= ~mask; | |
d4c08f2a PM |
1127 | trace_rcu_quiescent_state_report(rsp->name, rnp->gpnum, |
1128 | mask, rnp->qsmask, rnp->level, | |
1129 | rnp->grplo, rnp->grphi, | |
1130 | !!rnp->gp_tasks); | |
27f4d280 | 1131 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
64db4cff PM |
1132 | |
1133 | /* Other bits still set at this level, so done. */ | |
1304afb2 | 1134 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1135 | return; |
1136 | } | |
1137 | mask = rnp->grpmask; | |
1138 | if (rnp->parent == NULL) { | |
1139 | ||
1140 | /* No more levels. Exit loop holding root lock. */ | |
1141 | ||
1142 | break; | |
1143 | } | |
1304afb2 | 1144 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
28ecd580 | 1145 | rnp_c = rnp; |
64db4cff | 1146 | rnp = rnp->parent; |
1304afb2 | 1147 | raw_spin_lock_irqsave(&rnp->lock, flags); |
28ecd580 | 1148 | WARN_ON_ONCE(rnp_c->qsmask); |
64db4cff PM |
1149 | } |
1150 | ||
1151 | /* | |
1152 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 1153 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 1154 | * to clean up and start the next grace period if one is needed. |
64db4cff | 1155 | */ |
d3f6bad3 | 1156 | rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */ |
64db4cff PM |
1157 | } |
1158 | ||
1159 | /* | |
d3f6bad3 PM |
1160 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
1161 | * structure. This must be either called from the specified CPU, or | |
1162 | * called when the specified CPU is known to be offline (and when it is | |
1163 | * also known that no other CPU is concurrently trying to help the offline | |
1164 | * CPU). The lastcomp argument is used to make sure we are still in the | |
1165 | * grace period of interest. We don't want to end the current grace period | |
1166 | * based on quiescent states detected in an earlier grace period! | |
64db4cff PM |
1167 | */ |
1168 | static void | |
e4cc1f22 | 1169 | rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastgp) |
64db4cff PM |
1170 | { |
1171 | unsigned long flags; | |
1172 | unsigned long mask; | |
1173 | struct rcu_node *rnp; | |
1174 | ||
1175 | rnp = rdp->mynode; | |
1304afb2 | 1176 | raw_spin_lock_irqsave(&rnp->lock, flags); |
e4cc1f22 | 1177 | if (lastgp != rnp->gpnum || rnp->completed == rnp->gpnum) { |
64db4cff PM |
1178 | |
1179 | /* | |
e4cc1f22 PM |
1180 | * The grace period in which this quiescent state was |
1181 | * recorded has ended, so don't report it upwards. | |
1182 | * We will instead need a new quiescent state that lies | |
1183 | * within the current grace period. | |
64db4cff | 1184 | */ |
e4cc1f22 | 1185 | rdp->passed_quiesce = 0; /* need qs for new gp. */ |
1304afb2 | 1186 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1187 | return; |
1188 | } | |
1189 | mask = rdp->grpmask; | |
1190 | if ((rnp->qsmask & mask) == 0) { | |
1304afb2 | 1191 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1192 | } else { |
1193 | rdp->qs_pending = 0; | |
1194 | ||
1195 | /* | |
1196 | * This GP can't end until cpu checks in, so all of our | |
1197 | * callbacks can be processed during the next GP. | |
1198 | */ | |
64db4cff PM |
1199 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
1200 | ||
d3f6bad3 | 1201 | rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */ |
64db4cff PM |
1202 | } |
1203 | } | |
1204 | ||
1205 | /* | |
1206 | * Check to see if there is a new grace period of which this CPU | |
1207 | * is not yet aware, and if so, set up local rcu_data state for it. | |
1208 | * Otherwise, see if this CPU has just passed through its first | |
1209 | * quiescent state for this grace period, and record that fact if so. | |
1210 | */ | |
1211 | static void | |
1212 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | |
1213 | { | |
1214 | /* If there is now a new grace period, record and return. */ | |
1215 | if (check_for_new_grace_period(rsp, rdp)) | |
1216 | return; | |
1217 | ||
1218 | /* | |
1219 | * Does this CPU still need to do its part for current grace period? | |
1220 | * If no, return and let the other CPUs do their part as well. | |
1221 | */ | |
1222 | if (!rdp->qs_pending) | |
1223 | return; | |
1224 | ||
1225 | /* | |
1226 | * Was there a quiescent state since the beginning of the grace | |
1227 | * period? If no, then exit and wait for the next call. | |
1228 | */ | |
e4cc1f22 | 1229 | if (!rdp->passed_quiesce) |
64db4cff PM |
1230 | return; |
1231 | ||
d3f6bad3 PM |
1232 | /* |
1233 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
1234 | * judge of that). | |
1235 | */ | |
e4cc1f22 | 1236 | rcu_report_qs_rdp(rdp->cpu, rsp, rdp, rdp->passed_quiesce_gpnum); |
64db4cff PM |
1237 | } |
1238 | ||
1239 | #ifdef CONFIG_HOTPLUG_CPU | |
1240 | ||
e74f4c45 | 1241 | /* |
29494be7 LJ |
1242 | * Move a dying CPU's RCU callbacks to online CPU's callback list. |
1243 | * Synchronization is not required because this function executes | |
1244 | * in stop_machine() context. | |
e74f4c45 | 1245 | */ |
29494be7 | 1246 | static void rcu_send_cbs_to_online(struct rcu_state *rsp) |
e74f4c45 PM |
1247 | { |
1248 | int i; | |
29494be7 LJ |
1249 | /* current DYING CPU is cleared in the cpu_online_mask */ |
1250 | int receive_cpu = cpumask_any(cpu_online_mask); | |
394f99a9 | 1251 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
29494be7 | 1252 | struct rcu_data *receive_rdp = per_cpu_ptr(rsp->rda, receive_cpu); |
e74f4c45 PM |
1253 | |
1254 | if (rdp->nxtlist == NULL) | |
1255 | return; /* irqs disabled, so comparison is stable. */ | |
29494be7 LJ |
1256 | |
1257 | *receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist; | |
1258 | receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
1259 | receive_rdp->qlen += rdp->qlen; | |
1260 | receive_rdp->n_cbs_adopted += rdp->qlen; | |
1261 | rdp->n_cbs_orphaned += rdp->qlen; | |
1262 | ||
e74f4c45 PM |
1263 | rdp->nxtlist = NULL; |
1264 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
1265 | rdp->nxttail[i] = &rdp->nxtlist; | |
e74f4c45 | 1266 | rdp->qlen = 0; |
e74f4c45 PM |
1267 | } |
1268 | ||
64db4cff PM |
1269 | /* |
1270 | * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy | |
1271 | * and move all callbacks from the outgoing CPU to the current one. | |
a26ac245 PM |
1272 | * There can only be one CPU hotplug operation at a time, so no other |
1273 | * CPU can be attempting to update rcu_cpu_kthread_task. | |
64db4cff PM |
1274 | */ |
1275 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp) | |
1276 | { | |
64db4cff | 1277 | unsigned long flags; |
64db4cff | 1278 | unsigned long mask; |
d9a3da06 | 1279 | int need_report = 0; |
394f99a9 | 1280 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff | 1281 | struct rcu_node *rnp; |
a26ac245 | 1282 | |
f8b7fc6b | 1283 | rcu_stop_cpu_kthread(cpu); |
64db4cff PM |
1284 | |
1285 | /* Exclude any attempts to start a new grace period. */ | |
1304afb2 | 1286 | raw_spin_lock_irqsave(&rsp->onofflock, flags); |
64db4cff PM |
1287 | |
1288 | /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */ | |
28ecd580 | 1289 | rnp = rdp->mynode; /* this is the outgoing CPU's rnp. */ |
64db4cff PM |
1290 | mask = rdp->grpmask; /* rnp->grplo is constant. */ |
1291 | do { | |
1304afb2 | 1292 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
1293 | rnp->qsmaskinit &= ~mask; |
1294 | if (rnp->qsmaskinit != 0) { | |
b668c9cf | 1295 | if (rnp != rdp->mynode) |
1304afb2 | 1296 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
d4c08f2a PM |
1297 | else |
1298 | trace_rcu_grace_period(rsp->name, | |
1299 | rnp->gpnum + 1 - | |
1300 | !!(rnp->qsmask & mask), | |
1301 | "cpuofl"); | |
64db4cff PM |
1302 | break; |
1303 | } | |
d4c08f2a PM |
1304 | if (rnp == rdp->mynode) { |
1305 | trace_rcu_grace_period(rsp->name, | |
1306 | rnp->gpnum + 1 - | |
1307 | !!(rnp->qsmask & mask), | |
1308 | "cpuofl"); | |
d9a3da06 | 1309 | need_report = rcu_preempt_offline_tasks(rsp, rnp, rdp); |
d4c08f2a | 1310 | } else |
1304afb2 | 1311 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff | 1312 | mask = rnp->grpmask; |
64db4cff PM |
1313 | rnp = rnp->parent; |
1314 | } while (rnp != NULL); | |
64db4cff | 1315 | |
b668c9cf PM |
1316 | /* |
1317 | * We still hold the leaf rcu_node structure lock here, and | |
1318 | * irqs are still disabled. The reason for this subterfuge is | |
d3f6bad3 PM |
1319 | * because invoking rcu_report_unblock_qs_rnp() with ->onofflock |
1320 | * held leads to deadlock. | |
b668c9cf | 1321 | */ |
1304afb2 | 1322 | raw_spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ |
b668c9cf | 1323 | rnp = rdp->mynode; |
d9a3da06 | 1324 | if (need_report & RCU_OFL_TASKS_NORM_GP) |
d3f6bad3 | 1325 | rcu_report_unblock_qs_rnp(rnp, flags); |
b668c9cf | 1326 | else |
1304afb2 | 1327 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
d9a3da06 | 1328 | if (need_report & RCU_OFL_TASKS_EXP_GP) |
b40d293e | 1329 | rcu_report_exp_rnp(rsp, rnp, true); |
1217ed1b | 1330 | rcu_node_kthread_setaffinity(rnp, -1); |
64db4cff PM |
1331 | } |
1332 | ||
1333 | /* | |
1334 | * Remove the specified CPU from the RCU hierarchy and move any pending | |
1335 | * callbacks that it might have to the current CPU. This code assumes | |
1336 | * that at least one CPU in the system will remain running at all times. | |
1337 | * Any attempt to offline -all- CPUs is likely to strand RCU callbacks. | |
1338 | */ | |
1339 | static void rcu_offline_cpu(int cpu) | |
1340 | { | |
d6714c22 | 1341 | __rcu_offline_cpu(cpu, &rcu_sched_state); |
64db4cff | 1342 | __rcu_offline_cpu(cpu, &rcu_bh_state); |
33f76148 | 1343 | rcu_preempt_offline_cpu(cpu); |
64db4cff PM |
1344 | } |
1345 | ||
1346 | #else /* #ifdef CONFIG_HOTPLUG_CPU */ | |
1347 | ||
29494be7 | 1348 | static void rcu_send_cbs_to_online(struct rcu_state *rsp) |
e74f4c45 PM |
1349 | { |
1350 | } | |
1351 | ||
64db4cff PM |
1352 | static void rcu_offline_cpu(int cpu) |
1353 | { | |
1354 | } | |
1355 | ||
1356 | #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */ | |
1357 | ||
1358 | /* | |
1359 | * Invoke any RCU callbacks that have made it to the end of their grace | |
1360 | * period. Thottle as specified by rdp->blimit. | |
1361 | */ | |
37c72e56 | 1362 | static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
1363 | { |
1364 | unsigned long flags; | |
1365 | struct rcu_head *next, *list, **tail; | |
29c00b4a | 1366 | int bl, count; |
64db4cff PM |
1367 | |
1368 | /* If no callbacks are ready, just return.*/ | |
29c00b4a | 1369 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) { |
72fe701b PM |
1370 | trace_rcu_batch_start(rsp->name, 0, 0); |
1371 | trace_rcu_batch_end(rsp->name, 0); | |
64db4cff | 1372 | return; |
29c00b4a | 1373 | } |
64db4cff PM |
1374 | |
1375 | /* | |
1376 | * Extract the list of ready callbacks, disabling to prevent | |
1377 | * races with call_rcu() from interrupt handlers. | |
1378 | */ | |
1379 | local_irq_save(flags); | |
29c00b4a | 1380 | bl = rdp->blimit; |
72fe701b | 1381 | trace_rcu_batch_start(rsp->name, rdp->qlen, bl); |
64db4cff PM |
1382 | list = rdp->nxtlist; |
1383 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; | |
1384 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | |
1385 | tail = rdp->nxttail[RCU_DONE_TAIL]; | |
1386 | for (count = RCU_NEXT_SIZE - 1; count >= 0; count--) | |
1387 | if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL]) | |
1388 | rdp->nxttail[count] = &rdp->nxtlist; | |
1389 | local_irq_restore(flags); | |
1390 | ||
1391 | /* Invoke callbacks. */ | |
1392 | count = 0; | |
1393 | while (list) { | |
1394 | next = list->next; | |
1395 | prefetch(next); | |
551d55a9 | 1396 | debug_rcu_head_unqueue(list); |
d4c08f2a | 1397 | __rcu_reclaim(rsp->name, list); |
64db4cff | 1398 | list = next; |
29c00b4a | 1399 | if (++count >= bl) |
64db4cff PM |
1400 | break; |
1401 | } | |
1402 | ||
1403 | local_irq_save(flags); | |
72fe701b | 1404 | trace_rcu_batch_end(rsp->name, count); |
64db4cff PM |
1405 | |
1406 | /* Update count, and requeue any remaining callbacks. */ | |
1407 | rdp->qlen -= count; | |
269dcc1c | 1408 | rdp->n_cbs_invoked += count; |
64db4cff PM |
1409 | if (list != NULL) { |
1410 | *tail = rdp->nxtlist; | |
1411 | rdp->nxtlist = list; | |
1412 | for (count = 0; count < RCU_NEXT_SIZE; count++) | |
1413 | if (&rdp->nxtlist == rdp->nxttail[count]) | |
1414 | rdp->nxttail[count] = tail; | |
1415 | else | |
1416 | break; | |
1417 | } | |
1418 | ||
1419 | /* Reinstate batch limit if we have worked down the excess. */ | |
1420 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) | |
1421 | rdp->blimit = blimit; | |
1422 | ||
37c72e56 PM |
1423 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
1424 | if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) { | |
1425 | rdp->qlen_last_fqs_check = 0; | |
1426 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1427 | } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark) | |
1428 | rdp->qlen_last_fqs_check = rdp->qlen; | |
1429 | ||
64db4cff PM |
1430 | local_irq_restore(flags); |
1431 | ||
e0f23060 | 1432 | /* Re-invoke RCU core processing if there are callbacks remaining. */ |
64db4cff | 1433 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
a46e0899 | 1434 | invoke_rcu_core(); |
64db4cff PM |
1435 | } |
1436 | ||
1437 | /* | |
1438 | * Check to see if this CPU is in a non-context-switch quiescent state | |
1439 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
e0f23060 | 1440 | * Also schedule RCU core processing. |
64db4cff | 1441 | * |
9b2e4f18 | 1442 | * This function must be called from hardirq context. It is normally |
64db4cff PM |
1443 | * invoked from the scheduling-clock interrupt. If rcu_pending returns |
1444 | * false, there is no point in invoking rcu_check_callbacks(). | |
1445 | */ | |
1446 | void rcu_check_callbacks(int cpu, int user) | |
1447 | { | |
300df91c | 1448 | trace_rcu_utilization("Start scheduler-tick"); |
9b2e4f18 | 1449 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
64db4cff PM |
1450 | |
1451 | /* | |
1452 | * Get here if this CPU took its interrupt from user | |
1453 | * mode or from the idle loop, and if this is not a | |
1454 | * nested interrupt. In this case, the CPU is in | |
d6714c22 | 1455 | * a quiescent state, so note it. |
64db4cff PM |
1456 | * |
1457 | * No memory barrier is required here because both | |
d6714c22 PM |
1458 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
1459 | * variables that other CPUs neither access nor modify, | |
1460 | * at least not while the corresponding CPU is online. | |
64db4cff PM |
1461 | */ |
1462 | ||
d6714c22 PM |
1463 | rcu_sched_qs(cpu); |
1464 | rcu_bh_qs(cpu); | |
64db4cff PM |
1465 | |
1466 | } else if (!in_softirq()) { | |
1467 | ||
1468 | /* | |
1469 | * Get here if this CPU did not take its interrupt from | |
1470 | * softirq, in other words, if it is not interrupting | |
1471 | * a rcu_bh read-side critical section. This is an _bh | |
d6714c22 | 1472 | * critical section, so note it. |
64db4cff PM |
1473 | */ |
1474 | ||
d6714c22 | 1475 | rcu_bh_qs(cpu); |
64db4cff | 1476 | } |
f41d911f | 1477 | rcu_preempt_check_callbacks(cpu); |
d21670ac | 1478 | if (rcu_pending(cpu)) |
a46e0899 | 1479 | invoke_rcu_core(); |
300df91c | 1480 | trace_rcu_utilization("End scheduler-tick"); |
64db4cff PM |
1481 | } |
1482 | ||
1483 | #ifdef CONFIG_SMP | |
1484 | ||
1485 | /* | |
1486 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
1487 | * have not yet encountered a quiescent state, using the function specified. | |
27f4d280 PM |
1488 | * Also initiate boosting for any threads blocked on the root rcu_node. |
1489 | * | |
ee47eb9f | 1490 | * The caller must have suppressed start of new grace periods. |
64db4cff | 1491 | */ |
45f014c5 | 1492 | static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *)) |
64db4cff PM |
1493 | { |
1494 | unsigned long bit; | |
1495 | int cpu; | |
1496 | unsigned long flags; | |
1497 | unsigned long mask; | |
a0b6c9a7 | 1498 | struct rcu_node *rnp; |
64db4cff | 1499 | |
a0b6c9a7 | 1500 | rcu_for_each_leaf_node(rsp, rnp) { |
64db4cff | 1501 | mask = 0; |
1304afb2 | 1502 | raw_spin_lock_irqsave(&rnp->lock, flags); |
ee47eb9f | 1503 | if (!rcu_gp_in_progress(rsp)) { |
1304afb2 | 1504 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
0f10dc82 | 1505 | return; |
64db4cff | 1506 | } |
a0b6c9a7 | 1507 | if (rnp->qsmask == 0) { |
1217ed1b | 1508 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock */ |
64db4cff PM |
1509 | continue; |
1510 | } | |
a0b6c9a7 | 1511 | cpu = rnp->grplo; |
64db4cff | 1512 | bit = 1; |
a0b6c9a7 | 1513 | for (; cpu <= rnp->grphi; cpu++, bit <<= 1) { |
394f99a9 LJ |
1514 | if ((rnp->qsmask & bit) != 0 && |
1515 | f(per_cpu_ptr(rsp->rda, cpu))) | |
64db4cff PM |
1516 | mask |= bit; |
1517 | } | |
45f014c5 | 1518 | if (mask != 0) { |
64db4cff | 1519 | |
d3f6bad3 PM |
1520 | /* rcu_report_qs_rnp() releases rnp->lock. */ |
1521 | rcu_report_qs_rnp(mask, rsp, rnp, flags); | |
64db4cff PM |
1522 | continue; |
1523 | } | |
1304afb2 | 1524 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 1525 | } |
27f4d280 | 1526 | rnp = rcu_get_root(rsp); |
1217ed1b PM |
1527 | if (rnp->qsmask == 0) { |
1528 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
1529 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */ | |
1530 | } | |
64db4cff PM |
1531 | } |
1532 | ||
1533 | /* | |
1534 | * Force quiescent states on reluctant CPUs, and also detect which | |
1535 | * CPUs are in dyntick-idle mode. | |
1536 | */ | |
1537 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1538 | { | |
1539 | unsigned long flags; | |
64db4cff | 1540 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff | 1541 | |
300df91c PM |
1542 | trace_rcu_utilization("Start fqs"); |
1543 | if (!rcu_gp_in_progress(rsp)) { | |
1544 | trace_rcu_utilization("End fqs"); | |
64db4cff | 1545 | return; /* No grace period in progress, nothing to force. */ |
300df91c | 1546 | } |
1304afb2 | 1547 | if (!raw_spin_trylock_irqsave(&rsp->fqslock, flags)) { |
64db4cff | 1548 | rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ |
300df91c | 1549 | trace_rcu_utilization("End fqs"); |
64db4cff PM |
1550 | return; /* Someone else is already on the job. */ |
1551 | } | |
20133cfc | 1552 | if (relaxed && ULONG_CMP_GE(rsp->jiffies_force_qs, jiffies)) |
f96e9232 | 1553 | goto unlock_fqs_ret; /* no emergency and done recently. */ |
64db4cff | 1554 | rsp->n_force_qs++; |
1304afb2 | 1555 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
64db4cff | 1556 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
560d4bc0 | 1557 | if(!rcu_gp_in_progress(rsp)) { |
64db4cff | 1558 | rsp->n_force_qs_ngp++; |
1304afb2 | 1559 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1560 | goto unlock_fqs_ret; /* no GP in progress, time updated. */ |
64db4cff | 1561 | } |
07079d53 | 1562 | rsp->fqs_active = 1; |
af446b70 | 1563 | switch (rsp->fqs_state) { |
83f5b01f | 1564 | case RCU_GP_IDLE: |
64db4cff PM |
1565 | case RCU_GP_INIT: |
1566 | ||
83f5b01f | 1567 | break; /* grace period idle or initializing, ignore. */ |
64db4cff PM |
1568 | |
1569 | case RCU_SAVE_DYNTICK: | |
64db4cff PM |
1570 | if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK) |
1571 | break; /* So gcc recognizes the dead code. */ | |
1572 | ||
f261414f LJ |
1573 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
1574 | ||
64db4cff | 1575 | /* Record dyntick-idle state. */ |
45f014c5 | 1576 | force_qs_rnp(rsp, dyntick_save_progress_counter); |
1304afb2 | 1577 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
ee47eb9f | 1578 | if (rcu_gp_in_progress(rsp)) |
af446b70 | 1579 | rsp->fqs_state = RCU_FORCE_QS; |
ee47eb9f | 1580 | break; |
64db4cff PM |
1581 | |
1582 | case RCU_FORCE_QS: | |
1583 | ||
1584 | /* Check dyntick-idle state, send IPI to laggarts. */ | |
1304afb2 | 1585 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
45f014c5 | 1586 | force_qs_rnp(rsp, rcu_implicit_dynticks_qs); |
64db4cff PM |
1587 | |
1588 | /* Leave state in case more forcing is required. */ | |
1589 | ||
1304afb2 | 1590 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
f96e9232 | 1591 | break; |
64db4cff | 1592 | } |
07079d53 | 1593 | rsp->fqs_active = 0; |
46a1e34e | 1594 | if (rsp->fqs_need_gp) { |
1304afb2 | 1595 | raw_spin_unlock(&rsp->fqslock); /* irqs remain disabled */ |
46a1e34e PM |
1596 | rsp->fqs_need_gp = 0; |
1597 | rcu_start_gp(rsp, flags); /* releases rnp->lock */ | |
300df91c | 1598 | trace_rcu_utilization("End fqs"); |
46a1e34e PM |
1599 | return; |
1600 | } | |
1304afb2 | 1601 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1602 | unlock_fqs_ret: |
1304afb2 | 1603 | raw_spin_unlock_irqrestore(&rsp->fqslock, flags); |
300df91c | 1604 | trace_rcu_utilization("End fqs"); |
64db4cff PM |
1605 | } |
1606 | ||
1607 | #else /* #ifdef CONFIG_SMP */ | |
1608 | ||
1609 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1610 | { | |
1611 | set_need_resched(); | |
1612 | } | |
1613 | ||
1614 | #endif /* #else #ifdef CONFIG_SMP */ | |
1615 | ||
1616 | /* | |
e0f23060 PM |
1617 | * This does the RCU core processing work for the specified rcu_state |
1618 | * and rcu_data structures. This may be called only from the CPU to | |
1619 | * whom the rdp belongs. | |
64db4cff PM |
1620 | */ |
1621 | static void | |
1622 | __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) | |
1623 | { | |
1624 | unsigned long flags; | |
1625 | ||
2e597558 PM |
1626 | WARN_ON_ONCE(rdp->beenonline == 0); |
1627 | ||
64db4cff PM |
1628 | /* |
1629 | * If an RCU GP has gone long enough, go check for dyntick | |
1630 | * idle CPUs and, if needed, send resched IPIs. | |
1631 | */ | |
20133cfc | 1632 | if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
64db4cff PM |
1633 | force_quiescent_state(rsp, 1); |
1634 | ||
1635 | /* | |
1636 | * Advance callbacks in response to end of earlier grace | |
1637 | * period that some other CPU ended. | |
1638 | */ | |
1639 | rcu_process_gp_end(rsp, rdp); | |
1640 | ||
1641 | /* Update RCU state based on any recent quiescent states. */ | |
1642 | rcu_check_quiescent_state(rsp, rdp); | |
1643 | ||
1644 | /* Does this CPU require a not-yet-started grace period? */ | |
1645 | if (cpu_needs_another_gp(rsp, rdp)) { | |
1304afb2 | 1646 | raw_spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags); |
64db4cff PM |
1647 | rcu_start_gp(rsp, flags); /* releases above lock */ |
1648 | } | |
1649 | ||
1650 | /* If there are callbacks ready, invoke them. */ | |
09223371 | 1651 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
a46e0899 | 1652 | invoke_rcu_callbacks(rsp, rdp); |
09223371 SL |
1653 | } |
1654 | ||
64db4cff | 1655 | /* |
e0f23060 | 1656 | * Do RCU core processing for the current CPU. |
64db4cff | 1657 | */ |
09223371 | 1658 | static void rcu_process_callbacks(struct softirq_action *unused) |
64db4cff | 1659 | { |
300df91c | 1660 | trace_rcu_utilization("Start RCU core"); |
d6714c22 PM |
1661 | __rcu_process_callbacks(&rcu_sched_state, |
1662 | &__get_cpu_var(rcu_sched_data)); | |
64db4cff | 1663 | __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); |
f41d911f | 1664 | rcu_preempt_process_callbacks(); |
300df91c | 1665 | trace_rcu_utilization("End RCU core"); |
64db4cff PM |
1666 | } |
1667 | ||
a26ac245 | 1668 | /* |
e0f23060 PM |
1669 | * Schedule RCU callback invocation. If the specified type of RCU |
1670 | * does not support RCU priority boosting, just do a direct call, | |
1671 | * otherwise wake up the per-CPU kernel kthread. Note that because we | |
1672 | * are running on the current CPU with interrupts disabled, the | |
1673 | * rcu_cpu_kthread_task cannot disappear out from under us. | |
a26ac245 | 1674 | */ |
a46e0899 | 1675 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) |
a26ac245 | 1676 | { |
b0d30417 PM |
1677 | if (unlikely(!ACCESS_ONCE(rcu_scheduler_fully_active))) |
1678 | return; | |
a46e0899 PM |
1679 | if (likely(!rsp->boost)) { |
1680 | rcu_do_batch(rsp, rdp); | |
a26ac245 PM |
1681 | return; |
1682 | } | |
a46e0899 | 1683 | invoke_rcu_callbacks_kthread(); |
a26ac245 PM |
1684 | } |
1685 | ||
a46e0899 | 1686 | static void invoke_rcu_core(void) |
09223371 SL |
1687 | { |
1688 | raise_softirq(RCU_SOFTIRQ); | |
1689 | } | |
1690 | ||
64db4cff PM |
1691 | static void |
1692 | __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu), | |
1693 | struct rcu_state *rsp) | |
1694 | { | |
1695 | unsigned long flags; | |
1696 | struct rcu_data *rdp; | |
1697 | ||
551d55a9 | 1698 | debug_rcu_head_queue(head); |
64db4cff PM |
1699 | head->func = func; |
1700 | head->next = NULL; | |
1701 | ||
1702 | smp_mb(); /* Ensure RCU update seen before callback registry. */ | |
1703 | ||
1704 | /* | |
1705 | * Opportunistically note grace-period endings and beginnings. | |
1706 | * Note that we might see a beginning right after we see an | |
1707 | * end, but never vice versa, since this CPU has to pass through | |
1708 | * a quiescent state betweentimes. | |
1709 | */ | |
1710 | local_irq_save(flags); | |
394f99a9 | 1711 | rdp = this_cpu_ptr(rsp->rda); |
64db4cff PM |
1712 | |
1713 | /* Add the callback to our list. */ | |
1714 | *rdp->nxttail[RCU_NEXT_TAIL] = head; | |
1715 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; | |
2655d57e PM |
1716 | rdp->qlen++; |
1717 | ||
d4c08f2a PM |
1718 | if (__is_kfree_rcu_offset((unsigned long)func)) |
1719 | trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func, | |
1720 | rdp->qlen); | |
1721 | else | |
1722 | trace_rcu_callback(rsp->name, head, rdp->qlen); | |
1723 | ||
2655d57e PM |
1724 | /* If interrupts were disabled, don't dive into RCU core. */ |
1725 | if (irqs_disabled_flags(flags)) { | |
1726 | local_irq_restore(flags); | |
1727 | return; | |
1728 | } | |
64db4cff | 1729 | |
37c72e56 PM |
1730 | /* |
1731 | * Force the grace period if too many callbacks or too long waiting. | |
1732 | * Enforce hysteresis, and don't invoke force_quiescent_state() | |
1733 | * if some other CPU has recently done so. Also, don't bother | |
1734 | * invoking force_quiescent_state() if the newly enqueued callback | |
1735 | * is the only one waiting for a grace period to complete. | |
1736 | */ | |
2655d57e | 1737 | if (unlikely(rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) { |
b52573d2 PM |
1738 | |
1739 | /* Are we ignoring a completed grace period? */ | |
1740 | rcu_process_gp_end(rsp, rdp); | |
1741 | check_for_new_grace_period(rsp, rdp); | |
1742 | ||
1743 | /* Start a new grace period if one not already started. */ | |
1744 | if (!rcu_gp_in_progress(rsp)) { | |
1745 | unsigned long nestflag; | |
1746 | struct rcu_node *rnp_root = rcu_get_root(rsp); | |
1747 | ||
1748 | raw_spin_lock_irqsave(&rnp_root->lock, nestflag); | |
1749 | rcu_start_gp(rsp, nestflag); /* rlses rnp_root->lock */ | |
1750 | } else { | |
1751 | /* Give the grace period a kick. */ | |
1752 | rdp->blimit = LONG_MAX; | |
1753 | if (rsp->n_force_qs == rdp->n_force_qs_snap && | |
1754 | *rdp->nxttail[RCU_DONE_TAIL] != head) | |
1755 | force_quiescent_state(rsp, 0); | |
1756 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1757 | rdp->qlen_last_fqs_check = rdp->qlen; | |
1758 | } | |
20133cfc | 1759 | } else if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
64db4cff PM |
1760 | force_quiescent_state(rsp, 1); |
1761 | local_irq_restore(flags); | |
1762 | } | |
1763 | ||
1764 | /* | |
d6714c22 | 1765 | * Queue an RCU-sched callback for invocation after a grace period. |
64db4cff | 1766 | */ |
d6714c22 | 1767 | void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) |
64db4cff | 1768 | { |
d6714c22 | 1769 | __call_rcu(head, func, &rcu_sched_state); |
64db4cff | 1770 | } |
d6714c22 | 1771 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
64db4cff PM |
1772 | |
1773 | /* | |
1774 | * Queue an RCU for invocation after a quicker grace period. | |
1775 | */ | |
1776 | void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
1777 | { | |
1778 | __call_rcu(head, func, &rcu_bh_state); | |
1779 | } | |
1780 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
1781 | ||
6ebb237b PM |
1782 | /** |
1783 | * synchronize_sched - wait until an rcu-sched grace period has elapsed. | |
1784 | * | |
1785 | * Control will return to the caller some time after a full rcu-sched | |
1786 | * grace period has elapsed, in other words after all currently executing | |
1787 | * rcu-sched read-side critical sections have completed. These read-side | |
1788 | * critical sections are delimited by rcu_read_lock_sched() and | |
1789 | * rcu_read_unlock_sched(), and may be nested. Note that preempt_disable(), | |
1790 | * local_irq_disable(), and so on may be used in place of | |
1791 | * rcu_read_lock_sched(). | |
1792 | * | |
1793 | * This means that all preempt_disable code sequences, including NMI and | |
1794 | * hardware-interrupt handlers, in progress on entry will have completed | |
1795 | * before this primitive returns. However, this does not guarantee that | |
1796 | * softirq handlers will have completed, since in some kernels, these | |
1797 | * handlers can run in process context, and can block. | |
1798 | * | |
1799 | * This primitive provides the guarantees made by the (now removed) | |
1800 | * synchronize_kernel() API. In contrast, synchronize_rcu() only | |
1801 | * guarantees that rcu_read_lock() sections will have completed. | |
1802 | * In "classic RCU", these two guarantees happen to be one and | |
1803 | * the same, but can differ in realtime RCU implementations. | |
1804 | */ | |
1805 | void synchronize_sched(void) | |
1806 | { | |
6ebb237b PM |
1807 | if (rcu_blocking_is_gp()) |
1808 | return; | |
2c42818e | 1809 | wait_rcu_gp(call_rcu_sched); |
6ebb237b PM |
1810 | } |
1811 | EXPORT_SYMBOL_GPL(synchronize_sched); | |
1812 | ||
1813 | /** | |
1814 | * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed. | |
1815 | * | |
1816 | * Control will return to the caller some time after a full rcu_bh grace | |
1817 | * period has elapsed, in other words after all currently executing rcu_bh | |
1818 | * read-side critical sections have completed. RCU read-side critical | |
1819 | * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(), | |
1820 | * and may be nested. | |
1821 | */ | |
1822 | void synchronize_rcu_bh(void) | |
1823 | { | |
6ebb237b PM |
1824 | if (rcu_blocking_is_gp()) |
1825 | return; | |
2c42818e | 1826 | wait_rcu_gp(call_rcu_bh); |
6ebb237b PM |
1827 | } |
1828 | EXPORT_SYMBOL_GPL(synchronize_rcu_bh); | |
1829 | ||
64db4cff PM |
1830 | /* |
1831 | * Check to see if there is any immediate RCU-related work to be done | |
1832 | * by the current CPU, for the specified type of RCU, returning 1 if so. | |
1833 | * The checks are in order of increasing expense: checks that can be | |
1834 | * carried out against CPU-local state are performed first. However, | |
1835 | * we must check for CPU stalls first, else we might not get a chance. | |
1836 | */ | |
1837 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | |
1838 | { | |
2f51f988 PM |
1839 | struct rcu_node *rnp = rdp->mynode; |
1840 | ||
64db4cff PM |
1841 | rdp->n_rcu_pending++; |
1842 | ||
1843 | /* Check for CPU stalls, if enabled. */ | |
1844 | check_cpu_stall(rsp, rdp); | |
1845 | ||
1846 | /* Is the RCU core waiting for a quiescent state from this CPU? */ | |
5c51dd73 PM |
1847 | if (rcu_scheduler_fully_active && |
1848 | rdp->qs_pending && !rdp->passed_quiesce) { | |
d25eb944 PM |
1849 | |
1850 | /* | |
1851 | * If force_quiescent_state() coming soon and this CPU | |
1852 | * needs a quiescent state, and this is either RCU-sched | |
1853 | * or RCU-bh, force a local reschedule. | |
1854 | */ | |
d21670ac | 1855 | rdp->n_rp_qs_pending++; |
6cc68793 | 1856 | if (!rdp->preemptible && |
d25eb944 PM |
1857 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs) - 1, |
1858 | jiffies)) | |
1859 | set_need_resched(); | |
e4cc1f22 | 1860 | } else if (rdp->qs_pending && rdp->passed_quiesce) { |
d21670ac | 1861 | rdp->n_rp_report_qs++; |
64db4cff | 1862 | return 1; |
7ba5c840 | 1863 | } |
64db4cff PM |
1864 | |
1865 | /* Does this CPU have callbacks ready to invoke? */ | |
7ba5c840 PM |
1866 | if (cpu_has_callbacks_ready_to_invoke(rdp)) { |
1867 | rdp->n_rp_cb_ready++; | |
64db4cff | 1868 | return 1; |
7ba5c840 | 1869 | } |
64db4cff PM |
1870 | |
1871 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
7ba5c840 PM |
1872 | if (cpu_needs_another_gp(rsp, rdp)) { |
1873 | rdp->n_rp_cpu_needs_gp++; | |
64db4cff | 1874 | return 1; |
7ba5c840 | 1875 | } |
64db4cff PM |
1876 | |
1877 | /* Has another RCU grace period completed? */ | |
2f51f988 | 1878 | if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */ |
7ba5c840 | 1879 | rdp->n_rp_gp_completed++; |
64db4cff | 1880 | return 1; |
7ba5c840 | 1881 | } |
64db4cff PM |
1882 | |
1883 | /* Has a new RCU grace period started? */ | |
2f51f988 | 1884 | if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */ |
7ba5c840 | 1885 | rdp->n_rp_gp_started++; |
64db4cff | 1886 | return 1; |
7ba5c840 | 1887 | } |
64db4cff PM |
1888 | |
1889 | /* Has an RCU GP gone long enough to send resched IPIs &c? */ | |
fc2219d4 | 1890 | if (rcu_gp_in_progress(rsp) && |
20133cfc | 1891 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) { |
7ba5c840 | 1892 | rdp->n_rp_need_fqs++; |
64db4cff | 1893 | return 1; |
7ba5c840 | 1894 | } |
64db4cff PM |
1895 | |
1896 | /* nothing to do */ | |
7ba5c840 | 1897 | rdp->n_rp_need_nothing++; |
64db4cff PM |
1898 | return 0; |
1899 | } | |
1900 | ||
1901 | /* | |
1902 | * Check to see if there is any immediate RCU-related work to be done | |
1903 | * by the current CPU, returning 1 if so. This function is part of the | |
1904 | * RCU implementation; it is -not- an exported member of the RCU API. | |
1905 | */ | |
a157229c | 1906 | static int rcu_pending(int cpu) |
64db4cff | 1907 | { |
d6714c22 | 1908 | return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) || |
f41d911f PM |
1909 | __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) || |
1910 | rcu_preempt_pending(cpu); | |
64db4cff PM |
1911 | } |
1912 | ||
1913 | /* | |
1914 | * Check to see if any future RCU-related work will need to be done | |
1915 | * by the current CPU, even if none need be done immediately, returning | |
8bd93a2c | 1916 | * 1 if so. |
64db4cff | 1917 | */ |
8bd93a2c | 1918 | static int rcu_needs_cpu_quick_check(int cpu) |
64db4cff PM |
1919 | { |
1920 | /* RCU callbacks either ready or pending? */ | |
d6714c22 | 1921 | return per_cpu(rcu_sched_data, cpu).nxtlist || |
f41d911f PM |
1922 | per_cpu(rcu_bh_data, cpu).nxtlist || |
1923 | rcu_preempt_needs_cpu(cpu); | |
64db4cff PM |
1924 | } |
1925 | ||
d0ec774c PM |
1926 | static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL}; |
1927 | static atomic_t rcu_barrier_cpu_count; | |
1928 | static DEFINE_MUTEX(rcu_barrier_mutex); | |
1929 | static struct completion rcu_barrier_completion; | |
d0ec774c PM |
1930 | |
1931 | static void rcu_barrier_callback(struct rcu_head *notused) | |
1932 | { | |
1933 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
1934 | complete(&rcu_barrier_completion); | |
1935 | } | |
1936 | ||
1937 | /* | |
1938 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
1939 | */ | |
1940 | static void rcu_barrier_func(void *type) | |
1941 | { | |
1942 | int cpu = smp_processor_id(); | |
1943 | struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu); | |
1944 | void (*call_rcu_func)(struct rcu_head *head, | |
1945 | void (*func)(struct rcu_head *head)); | |
1946 | ||
1947 | atomic_inc(&rcu_barrier_cpu_count); | |
1948 | call_rcu_func = type; | |
1949 | call_rcu_func(head, rcu_barrier_callback); | |
1950 | } | |
1951 | ||
d0ec774c PM |
1952 | /* |
1953 | * Orchestrate the specified type of RCU barrier, waiting for all | |
1954 | * RCU callbacks of the specified type to complete. | |
1955 | */ | |
e74f4c45 PM |
1956 | static void _rcu_barrier(struct rcu_state *rsp, |
1957 | void (*call_rcu_func)(struct rcu_head *head, | |
d0ec774c PM |
1958 | void (*func)(struct rcu_head *head))) |
1959 | { | |
1960 | BUG_ON(in_interrupt()); | |
e74f4c45 | 1961 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
d0ec774c PM |
1962 | mutex_lock(&rcu_barrier_mutex); |
1963 | init_completion(&rcu_barrier_completion); | |
1964 | /* | |
1965 | * Initialize rcu_barrier_cpu_count to 1, then invoke | |
1966 | * rcu_barrier_func() on each CPU, so that each CPU also has | |
1967 | * incremented rcu_barrier_cpu_count. Only then is it safe to | |
1968 | * decrement rcu_barrier_cpu_count -- otherwise the first CPU | |
1969 | * might complete its grace period before all of the other CPUs | |
1970 | * did their increment, causing this function to return too | |
2d999e03 PM |
1971 | * early. Note that on_each_cpu() disables irqs, which prevents |
1972 | * any CPUs from coming online or going offline until each online | |
1973 | * CPU has queued its RCU-barrier callback. | |
d0ec774c PM |
1974 | */ |
1975 | atomic_set(&rcu_barrier_cpu_count, 1); | |
1976 | on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1); | |
1977 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
1978 | complete(&rcu_barrier_completion); | |
1979 | wait_for_completion(&rcu_barrier_completion); | |
1980 | mutex_unlock(&rcu_barrier_mutex); | |
d0ec774c | 1981 | } |
d0ec774c PM |
1982 | |
1983 | /** | |
1984 | * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete. | |
1985 | */ | |
1986 | void rcu_barrier_bh(void) | |
1987 | { | |
e74f4c45 | 1988 | _rcu_barrier(&rcu_bh_state, call_rcu_bh); |
d0ec774c PM |
1989 | } |
1990 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); | |
1991 | ||
1992 | /** | |
1993 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. | |
1994 | */ | |
1995 | void rcu_barrier_sched(void) | |
1996 | { | |
e74f4c45 | 1997 | _rcu_barrier(&rcu_sched_state, call_rcu_sched); |
d0ec774c PM |
1998 | } |
1999 | EXPORT_SYMBOL_GPL(rcu_barrier_sched); | |
2000 | ||
64db4cff | 2001 | /* |
27569620 | 2002 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 2003 | */ |
27569620 PM |
2004 | static void __init |
2005 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) | |
64db4cff PM |
2006 | { |
2007 | unsigned long flags; | |
2008 | int i; | |
394f99a9 | 2009 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
27569620 PM |
2010 | struct rcu_node *rnp = rcu_get_root(rsp); |
2011 | ||
2012 | /* Set up local state, ensuring consistent view of global state. */ | |
1304afb2 | 2013 | raw_spin_lock_irqsave(&rnp->lock, flags); |
27569620 PM |
2014 | rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo); |
2015 | rdp->nxtlist = NULL; | |
2016 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
2017 | rdp->nxttail[i] = &rdp->nxtlist; | |
2018 | rdp->qlen = 0; | |
27569620 | 2019 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); |
4145fa7f | 2020 | WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != DYNTICK_TASK_NESTING); |
9b2e4f18 | 2021 | WARN_ON_ONCE(atomic_read(&rdp->dynticks->dynticks) != 1); |
27569620 | 2022 | rdp->cpu = cpu; |
d4c08f2a | 2023 | rdp->rsp = rsp; |
1304afb2 | 2024 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
27569620 PM |
2025 | } |
2026 | ||
2027 | /* | |
2028 | * Initialize a CPU's per-CPU RCU data. Note that only one online or | |
2029 | * offline event can be happening at a given time. Note also that we | |
2030 | * can accept some slop in the rsp->completed access due to the fact | |
2031 | * that this CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 2032 | */ |
e4fa4c97 | 2033 | static void __cpuinit |
6cc68793 | 2034 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptible) |
64db4cff PM |
2035 | { |
2036 | unsigned long flags; | |
64db4cff | 2037 | unsigned long mask; |
394f99a9 | 2038 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff PM |
2039 | struct rcu_node *rnp = rcu_get_root(rsp); |
2040 | ||
2041 | /* Set up local state, ensuring consistent view of global state. */ | |
1304afb2 | 2042 | raw_spin_lock_irqsave(&rnp->lock, flags); |
64db4cff | 2043 | rdp->beenonline = 1; /* We have now been online. */ |
6cc68793 | 2044 | rdp->preemptible = preemptible; |
37c72e56 PM |
2045 | rdp->qlen_last_fqs_check = 0; |
2046 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
64db4cff | 2047 | rdp->blimit = blimit; |
4145fa7f | 2048 | WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != DYNTICK_TASK_NESTING); |
9b2e4f18 | 2049 | WARN_ON_ONCE((atomic_read(&rdp->dynticks->dynticks) & 0x1) != 1); |
1304afb2 | 2050 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
2051 | |
2052 | /* | |
2053 | * A new grace period might start here. If so, we won't be part | |
2054 | * of it, but that is OK, as we are currently in a quiescent state. | |
2055 | */ | |
2056 | ||
2057 | /* Exclude any attempts to start a new GP on large systems. */ | |
1304afb2 | 2058 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
64db4cff PM |
2059 | |
2060 | /* Add CPU to rcu_node bitmasks. */ | |
2061 | rnp = rdp->mynode; | |
2062 | mask = rdp->grpmask; | |
2063 | do { | |
2064 | /* Exclude any attempts to start a new GP on small systems. */ | |
1304afb2 | 2065 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
2066 | rnp->qsmaskinit |= mask; |
2067 | mask = rnp->grpmask; | |
d09b62df | 2068 | if (rnp == rdp->mynode) { |
06ae115a PM |
2069 | /* |
2070 | * If there is a grace period in progress, we will | |
2071 | * set up to wait for it next time we run the | |
2072 | * RCU core code. | |
2073 | */ | |
2074 | rdp->gpnum = rnp->completed; | |
d09b62df | 2075 | rdp->completed = rnp->completed; |
06ae115a PM |
2076 | rdp->passed_quiesce = 0; |
2077 | rdp->qs_pending = 0; | |
e4cc1f22 | 2078 | rdp->passed_quiesce_gpnum = rnp->gpnum - 1; |
d4c08f2a | 2079 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuonl"); |
d09b62df | 2080 | } |
1304afb2 | 2081 | raw_spin_unlock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
2082 | rnp = rnp->parent; |
2083 | } while (rnp != NULL && !(rnp->qsmaskinit & mask)); | |
2084 | ||
1304afb2 | 2085 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); |
64db4cff PM |
2086 | } |
2087 | ||
d72bce0e | 2088 | static void __cpuinit rcu_prepare_cpu(int cpu) |
64db4cff | 2089 | { |
f41d911f PM |
2090 | rcu_init_percpu_data(cpu, &rcu_sched_state, 0); |
2091 | rcu_init_percpu_data(cpu, &rcu_bh_state, 0); | |
2092 | rcu_preempt_init_percpu_data(cpu); | |
64db4cff PM |
2093 | } |
2094 | ||
2095 | /* | |
f41d911f | 2096 | * Handle CPU online/offline notification events. |
64db4cff | 2097 | */ |
9f680ab4 PM |
2098 | static int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
2099 | unsigned long action, void *hcpu) | |
64db4cff PM |
2100 | { |
2101 | long cpu = (long)hcpu; | |
27f4d280 | 2102 | struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu); |
a26ac245 | 2103 | struct rcu_node *rnp = rdp->mynode; |
64db4cff | 2104 | |
300df91c | 2105 | trace_rcu_utilization("Start CPU hotplug"); |
64db4cff PM |
2106 | switch (action) { |
2107 | case CPU_UP_PREPARE: | |
2108 | case CPU_UP_PREPARE_FROZEN: | |
d72bce0e PZ |
2109 | rcu_prepare_cpu(cpu); |
2110 | rcu_prepare_kthreads(cpu); | |
a26ac245 PM |
2111 | break; |
2112 | case CPU_ONLINE: | |
0f962a5e PM |
2113 | case CPU_DOWN_FAILED: |
2114 | rcu_node_kthread_setaffinity(rnp, -1); | |
e3995a25 | 2115 | rcu_cpu_kthread_setrt(cpu, 1); |
0f962a5e PM |
2116 | break; |
2117 | case CPU_DOWN_PREPARE: | |
2118 | rcu_node_kthread_setaffinity(rnp, cpu); | |
e3995a25 | 2119 | rcu_cpu_kthread_setrt(cpu, 0); |
64db4cff | 2120 | break; |
d0ec774c PM |
2121 | case CPU_DYING: |
2122 | case CPU_DYING_FROZEN: | |
2123 | /* | |
2d999e03 PM |
2124 | * The whole machine is "stopped" except this CPU, so we can |
2125 | * touch any data without introducing corruption. We send the | |
2126 | * dying CPU's callbacks to an arbitrarily chosen online CPU. | |
d0ec774c | 2127 | */ |
29494be7 LJ |
2128 | rcu_send_cbs_to_online(&rcu_bh_state); |
2129 | rcu_send_cbs_to_online(&rcu_sched_state); | |
2130 | rcu_preempt_send_cbs_to_online(); | |
d0ec774c | 2131 | break; |
64db4cff PM |
2132 | case CPU_DEAD: |
2133 | case CPU_DEAD_FROZEN: | |
2134 | case CPU_UP_CANCELED: | |
2135 | case CPU_UP_CANCELED_FROZEN: | |
2136 | rcu_offline_cpu(cpu); | |
2137 | break; | |
2138 | default: | |
2139 | break; | |
2140 | } | |
300df91c | 2141 | trace_rcu_utilization("End CPU hotplug"); |
64db4cff PM |
2142 | return NOTIFY_OK; |
2143 | } | |
2144 | ||
bbad9379 PM |
2145 | /* |
2146 | * This function is invoked towards the end of the scheduler's initialization | |
2147 | * process. Before this is called, the idle task might contain | |
2148 | * RCU read-side critical sections (during which time, this idle | |
2149 | * task is booting the system). After this function is called, the | |
2150 | * idle tasks are prohibited from containing RCU read-side critical | |
2151 | * sections. This function also enables RCU lockdep checking. | |
2152 | */ | |
2153 | void rcu_scheduler_starting(void) | |
2154 | { | |
2155 | WARN_ON(num_online_cpus() != 1); | |
2156 | WARN_ON(nr_context_switches() > 0); | |
2157 | rcu_scheduler_active = 1; | |
2158 | } | |
2159 | ||
64db4cff PM |
2160 | /* |
2161 | * Compute the per-level fanout, either using the exact fanout specified | |
2162 | * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT. | |
2163 | */ | |
2164 | #ifdef CONFIG_RCU_FANOUT_EXACT | |
2165 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
2166 | { | |
2167 | int i; | |
2168 | ||
0209f649 | 2169 | for (i = NUM_RCU_LVLS - 1; i > 0; i--) |
64db4cff | 2170 | rsp->levelspread[i] = CONFIG_RCU_FANOUT; |
0209f649 | 2171 | rsp->levelspread[0] = RCU_FANOUT_LEAF; |
64db4cff PM |
2172 | } |
2173 | #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
2174 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
2175 | { | |
2176 | int ccur; | |
2177 | int cprv; | |
2178 | int i; | |
2179 | ||
2180 | cprv = NR_CPUS; | |
2181 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
2182 | ccur = rsp->levelcnt[i]; | |
2183 | rsp->levelspread[i] = (cprv + ccur - 1) / ccur; | |
2184 | cprv = ccur; | |
2185 | } | |
2186 | } | |
2187 | #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
2188 | ||
2189 | /* | |
2190 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
2191 | */ | |
394f99a9 LJ |
2192 | static void __init rcu_init_one(struct rcu_state *rsp, |
2193 | struct rcu_data __percpu *rda) | |
64db4cff | 2194 | { |
b6407e86 PM |
2195 | static char *buf[] = { "rcu_node_level_0", |
2196 | "rcu_node_level_1", | |
2197 | "rcu_node_level_2", | |
2198 | "rcu_node_level_3" }; /* Match MAX_RCU_LVLS */ | |
64db4cff PM |
2199 | int cpustride = 1; |
2200 | int i; | |
2201 | int j; | |
2202 | struct rcu_node *rnp; | |
2203 | ||
b6407e86 PM |
2204 | BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
2205 | ||
64db4cff PM |
2206 | /* Initialize the level-tracking arrays. */ |
2207 | ||
2208 | for (i = 1; i < NUM_RCU_LVLS; i++) | |
2209 | rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1]; | |
2210 | rcu_init_levelspread(rsp); | |
2211 | ||
2212 | /* Initialize the elements themselves, starting from the leaves. */ | |
2213 | ||
2214 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
2215 | cpustride *= rsp->levelspread[i]; | |
2216 | rnp = rsp->level[i]; | |
2217 | for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { | |
1304afb2 | 2218 | raw_spin_lock_init(&rnp->lock); |
b6407e86 PM |
2219 | lockdep_set_class_and_name(&rnp->lock, |
2220 | &rcu_node_class[i], buf[i]); | |
f41d911f | 2221 | rnp->gpnum = 0; |
64db4cff PM |
2222 | rnp->qsmask = 0; |
2223 | rnp->qsmaskinit = 0; | |
2224 | rnp->grplo = j * cpustride; | |
2225 | rnp->grphi = (j + 1) * cpustride - 1; | |
2226 | if (rnp->grphi >= NR_CPUS) | |
2227 | rnp->grphi = NR_CPUS - 1; | |
2228 | if (i == 0) { | |
2229 | rnp->grpnum = 0; | |
2230 | rnp->grpmask = 0; | |
2231 | rnp->parent = NULL; | |
2232 | } else { | |
2233 | rnp->grpnum = j % rsp->levelspread[i - 1]; | |
2234 | rnp->grpmask = 1UL << rnp->grpnum; | |
2235 | rnp->parent = rsp->level[i - 1] + | |
2236 | j / rsp->levelspread[i - 1]; | |
2237 | } | |
2238 | rnp->level = i; | |
12f5f524 | 2239 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
64db4cff PM |
2240 | } |
2241 | } | |
0c34029a | 2242 | |
394f99a9 | 2243 | rsp->rda = rda; |
0c34029a LJ |
2244 | rnp = rsp->level[NUM_RCU_LVLS - 1]; |
2245 | for_each_possible_cpu(i) { | |
4a90a068 | 2246 | while (i > rnp->grphi) |
0c34029a | 2247 | rnp++; |
394f99a9 | 2248 | per_cpu_ptr(rsp->rda, i)->mynode = rnp; |
0c34029a LJ |
2249 | rcu_boot_init_percpu_data(i, rsp); |
2250 | } | |
64db4cff PM |
2251 | } |
2252 | ||
9f680ab4 | 2253 | void __init rcu_init(void) |
64db4cff | 2254 | { |
017c4261 | 2255 | int cpu; |
9f680ab4 | 2256 | |
f41d911f | 2257 | rcu_bootup_announce(); |
394f99a9 LJ |
2258 | rcu_init_one(&rcu_sched_state, &rcu_sched_data); |
2259 | rcu_init_one(&rcu_bh_state, &rcu_bh_data); | |
f41d911f | 2260 | __rcu_init_preempt(); |
09223371 | 2261 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
9f680ab4 PM |
2262 | |
2263 | /* | |
2264 | * We don't need protection against CPU-hotplug here because | |
2265 | * this is called early in boot, before either interrupts | |
2266 | * or the scheduler are operational. | |
2267 | */ | |
2268 | cpu_notifier(rcu_cpu_notify, 0); | |
017c4261 PM |
2269 | for_each_online_cpu(cpu) |
2270 | rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu); | |
c68de209 | 2271 | check_cpu_stall_init(); |
64db4cff PM |
2272 | } |
2273 | ||
1eba8f84 | 2274 | #include "rcutree_plugin.h" |