]>
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); |
99745b6a | 358 | if (!is_idle_task(current)) { |
0989cb46 PM |
359 | struct task_struct *idle = idle_task(smp_processor_id()); |
360 | ||
9b2e4f18 | 361 | trace_rcu_dyntick("Error on entry: not idle task", |
4145fa7f | 362 | oldval, rdtp->dynticks_nesting); |
9b2e4f18 | 363 | ftrace_dump(DUMP_ALL); |
0989cb46 PM |
364 | WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s", |
365 | current->pid, current->comm, | |
366 | idle->pid, idle->comm); /* must be idle task! */ | |
9b2e4f18 | 367 | } |
aea1b35e | 368 | rcu_prepare_for_idle(smp_processor_id()); |
9b2e4f18 PM |
369 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ |
370 | smp_mb__before_atomic_inc(); /* See above. */ | |
371 | atomic_inc(&rdtp->dynticks); | |
372 | smp_mb__after_atomic_inc(); /* Force ordering with next sojourn. */ | |
373 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); | |
374 | } | |
64db4cff PM |
375 | |
376 | /** | |
9b2e4f18 | 377 | * rcu_idle_enter - inform RCU that current CPU is entering idle |
64db4cff | 378 | * |
9b2e4f18 | 379 | * Enter idle mode, in other words, -leave- the mode in which RCU |
64db4cff | 380 | * read-side critical sections can occur. (Though RCU read-side |
9b2e4f18 PM |
381 | * critical sections can occur in irq handlers in idle, a possibility |
382 | * handled by irq_enter() and irq_exit().) | |
383 | * | |
384 | * We crowbar the ->dynticks_nesting field to zero to allow for | |
385 | * the possibility of usermode upcalls having messed up our count | |
386 | * of interrupt nesting level during the prior busy period. | |
64db4cff | 387 | */ |
9b2e4f18 | 388 | void rcu_idle_enter(void) |
64db4cff PM |
389 | { |
390 | unsigned long flags; | |
4145fa7f | 391 | long long oldval; |
64db4cff PM |
392 | struct rcu_dynticks *rdtp; |
393 | ||
64db4cff PM |
394 | local_irq_save(flags); |
395 | rdtp = &__get_cpu_var(rcu_dynticks); | |
4145fa7f | 396 | oldval = rdtp->dynticks_nesting; |
9b2e4f18 | 397 | rdtp->dynticks_nesting = 0; |
4145fa7f | 398 | rcu_idle_enter_common(rdtp, oldval); |
64db4cff PM |
399 | local_irq_restore(flags); |
400 | } | |
401 | ||
9b2e4f18 PM |
402 | /** |
403 | * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle | |
404 | * | |
405 | * Exit from an interrupt handler, which might possibly result in entering | |
406 | * idle mode, in other words, leaving the mode in which read-side critical | |
407 | * sections can occur. | |
64db4cff | 408 | * |
9b2e4f18 PM |
409 | * This code assumes that the idle loop never does anything that might |
410 | * result in unbalanced calls to irq_enter() and irq_exit(). If your | |
411 | * architecture violates this assumption, RCU will give you what you | |
412 | * deserve, good and hard. But very infrequently and irreproducibly. | |
413 | * | |
414 | * Use things like work queues to work around this limitation. | |
415 | * | |
416 | * You have been warned. | |
64db4cff | 417 | */ |
9b2e4f18 | 418 | void rcu_irq_exit(void) |
64db4cff PM |
419 | { |
420 | unsigned long flags; | |
4145fa7f | 421 | long long oldval; |
64db4cff PM |
422 | struct rcu_dynticks *rdtp; |
423 | ||
424 | local_irq_save(flags); | |
425 | rdtp = &__get_cpu_var(rcu_dynticks); | |
4145fa7f | 426 | oldval = rdtp->dynticks_nesting; |
9b2e4f18 PM |
427 | rdtp->dynticks_nesting--; |
428 | WARN_ON_ONCE(rdtp->dynticks_nesting < 0); | |
4145fa7f | 429 | rcu_idle_enter_common(rdtp, oldval); |
9b2e4f18 PM |
430 | local_irq_restore(flags); |
431 | } | |
432 | ||
433 | /* | |
434 | * rcu_idle_exit_common - inform RCU that current CPU is moving away from idle | |
435 | * | |
436 | * If the new value of the ->dynticks_nesting counter was previously zero, | |
437 | * we really have exited idle, and must do the appropriate accounting. | |
438 | * The caller must have disabled interrupts. | |
439 | */ | |
440 | static void rcu_idle_exit_common(struct rcu_dynticks *rdtp, long long oldval) | |
441 | { | |
442 | if (oldval) { | |
4145fa7f | 443 | trace_rcu_dyntick("++=", oldval, rdtp->dynticks_nesting); |
23b5c8fa PM |
444 | return; |
445 | } | |
446 | smp_mb__before_atomic_inc(); /* Force ordering w/previous sojourn. */ | |
447 | atomic_inc(&rdtp->dynticks); | |
448 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ | |
449 | smp_mb__after_atomic_inc(); /* See above. */ | |
450 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); | |
7cb92499 | 451 | rcu_cleanup_after_idle(smp_processor_id()); |
4145fa7f | 452 | trace_rcu_dyntick("End", oldval, rdtp->dynticks_nesting); |
99745b6a | 453 | if (!is_idle_task(current)) { |
0989cb46 PM |
454 | struct task_struct *idle = idle_task(smp_processor_id()); |
455 | ||
4145fa7f PM |
456 | trace_rcu_dyntick("Error on exit: not idle task", |
457 | oldval, rdtp->dynticks_nesting); | |
9b2e4f18 | 458 | ftrace_dump(DUMP_ALL); |
0989cb46 PM |
459 | WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s", |
460 | current->pid, current->comm, | |
461 | idle->pid, idle->comm); /* must be idle task! */ | |
9b2e4f18 PM |
462 | } |
463 | } | |
464 | ||
465 | /** | |
466 | * rcu_idle_exit - inform RCU that current CPU is leaving idle | |
467 | * | |
468 | * Exit idle mode, in other words, -enter- the mode in which RCU | |
469 | * read-side critical sections can occur. | |
470 | * | |
4145fa7f PM |
471 | * We crowbar the ->dynticks_nesting field to DYNTICK_TASK_NESTING to |
472 | * allow for the possibility of usermode upcalls messing up our count | |
9b2e4f18 PM |
473 | * of interrupt nesting level during the busy period that is just |
474 | * now starting. | |
475 | */ | |
476 | void rcu_idle_exit(void) | |
477 | { | |
478 | unsigned long flags; | |
479 | struct rcu_dynticks *rdtp; | |
480 | long long oldval; | |
481 | ||
482 | local_irq_save(flags); | |
483 | rdtp = &__get_cpu_var(rcu_dynticks); | |
484 | oldval = rdtp->dynticks_nesting; | |
485 | WARN_ON_ONCE(oldval != 0); | |
4145fa7f | 486 | rdtp->dynticks_nesting = DYNTICK_TASK_NESTING; |
9b2e4f18 PM |
487 | rcu_idle_exit_common(rdtp, oldval); |
488 | local_irq_restore(flags); | |
489 | } | |
490 | ||
491 | /** | |
492 | * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle | |
493 | * | |
494 | * Enter an interrupt handler, which might possibly result in exiting | |
495 | * idle mode, in other words, entering the mode in which read-side critical | |
496 | * sections can occur. | |
497 | * | |
498 | * Note that the Linux kernel is fully capable of entering an interrupt | |
499 | * handler that it never exits, for example when doing upcalls to | |
500 | * user mode! This code assumes that the idle loop never does upcalls to | |
501 | * user mode. If your architecture does do upcalls from the idle loop (or | |
502 | * does anything else that results in unbalanced calls to the irq_enter() | |
503 | * and irq_exit() functions), RCU will give you what you deserve, good | |
504 | * and hard. But very infrequently and irreproducibly. | |
505 | * | |
506 | * Use things like work queues to work around this limitation. | |
507 | * | |
508 | * You have been warned. | |
509 | */ | |
510 | void rcu_irq_enter(void) | |
511 | { | |
512 | unsigned long flags; | |
513 | struct rcu_dynticks *rdtp; | |
514 | long long oldval; | |
515 | ||
516 | local_irq_save(flags); | |
517 | rdtp = &__get_cpu_var(rcu_dynticks); | |
518 | oldval = rdtp->dynticks_nesting; | |
519 | rdtp->dynticks_nesting++; | |
520 | WARN_ON_ONCE(rdtp->dynticks_nesting == 0); | |
521 | rcu_idle_exit_common(rdtp, oldval); | |
64db4cff | 522 | local_irq_restore(flags); |
64db4cff PM |
523 | } |
524 | ||
525 | /** | |
526 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
527 | * | |
528 | * If the CPU was idle with dynamic ticks active, and there is no | |
529 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
530 | * RCU grace-period handling know that the CPU is active. | |
531 | */ | |
532 | void rcu_nmi_enter(void) | |
533 | { | |
534 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
535 | ||
23b5c8fa PM |
536 | if (rdtp->dynticks_nmi_nesting == 0 && |
537 | (atomic_read(&rdtp->dynticks) & 0x1)) | |
64db4cff | 538 | return; |
23b5c8fa PM |
539 | rdtp->dynticks_nmi_nesting++; |
540 | smp_mb__before_atomic_inc(); /* Force delay from prior write. */ | |
541 | atomic_inc(&rdtp->dynticks); | |
542 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ | |
543 | smp_mb__after_atomic_inc(); /* See above. */ | |
544 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); | |
64db4cff PM |
545 | } |
546 | ||
547 | /** | |
548 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
549 | * | |
550 | * If the CPU was idle with dynamic ticks active, and there is no | |
551 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
552 | * RCU grace-period handling know that the CPU is no longer active. | |
553 | */ | |
554 | void rcu_nmi_exit(void) | |
555 | { | |
556 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
557 | ||
23b5c8fa PM |
558 | if (rdtp->dynticks_nmi_nesting == 0 || |
559 | --rdtp->dynticks_nmi_nesting != 0) | |
64db4cff | 560 | return; |
23b5c8fa PM |
561 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ |
562 | smp_mb__before_atomic_inc(); /* See above. */ | |
563 | atomic_inc(&rdtp->dynticks); | |
564 | smp_mb__after_atomic_inc(); /* Force delay to next write. */ | |
565 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); | |
64db4cff PM |
566 | } |
567 | ||
9b2e4f18 PM |
568 | #ifdef CONFIG_PROVE_RCU |
569 | ||
64db4cff | 570 | /** |
9b2e4f18 | 571 | * rcu_is_cpu_idle - see if RCU thinks that the current CPU is idle |
64db4cff | 572 | * |
9b2e4f18 | 573 | * If the current CPU is in its idle loop and is neither in an interrupt |
34240697 | 574 | * or NMI handler, return true. |
64db4cff | 575 | */ |
9b2e4f18 | 576 | int rcu_is_cpu_idle(void) |
64db4cff | 577 | { |
34240697 PM |
578 | int ret; |
579 | ||
580 | preempt_disable(); | |
581 | ret = (atomic_read(&__get_cpu_var(rcu_dynticks).dynticks) & 0x1) == 0; | |
582 | preempt_enable(); | |
583 | return ret; | |
64db4cff | 584 | } |
e6b80a3b | 585 | EXPORT_SYMBOL(rcu_is_cpu_idle); |
64db4cff | 586 | |
9b2e4f18 PM |
587 | #endif /* #ifdef CONFIG_PROVE_RCU */ |
588 | ||
64db4cff | 589 | /** |
9b2e4f18 | 590 | * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle |
64db4cff | 591 | * |
9b2e4f18 PM |
592 | * If the current CPU is idle or running at a first-level (not nested) |
593 | * interrupt from idle, return true. The caller must have at least | |
594 | * disabled preemption. | |
64db4cff | 595 | */ |
9b2e4f18 | 596 | int rcu_is_cpu_rrupt_from_idle(void) |
64db4cff | 597 | { |
9b2e4f18 | 598 | return __get_cpu_var(rcu_dynticks).dynticks_nesting <= 1; |
64db4cff PM |
599 | } |
600 | ||
64db4cff PM |
601 | #ifdef CONFIG_SMP |
602 | ||
64db4cff PM |
603 | /* |
604 | * Snapshot the specified CPU's dynticks counter so that we can later | |
605 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 606 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff PM |
607 | */ |
608 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | |
609 | { | |
23b5c8fa | 610 | rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks); |
f0e7c19d | 611 | return (rdp->dynticks_snap & 0x1) == 0; |
64db4cff PM |
612 | } |
613 | ||
614 | /* | |
615 | * Return true if the specified CPU has passed through a quiescent | |
616 | * state by virtue of being in or having passed through an dynticks | |
617 | * idle state since the last call to dyntick_save_progress_counter() | |
618 | * for this same CPU. | |
619 | */ | |
620 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
621 | { | |
7eb4f455 PM |
622 | unsigned int curr; |
623 | unsigned int snap; | |
64db4cff | 624 | |
7eb4f455 PM |
625 | curr = (unsigned int)atomic_add_return(0, &rdp->dynticks->dynticks); |
626 | snap = (unsigned int)rdp->dynticks_snap; | |
64db4cff PM |
627 | |
628 | /* | |
629 | * If the CPU passed through or entered a dynticks idle phase with | |
630 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
631 | * already acknowledged the request to pass through a quiescent | |
632 | * state. Either way, that CPU cannot possibly be in an RCU | |
633 | * read-side critical section that started before the beginning | |
634 | * of the current RCU grace period. | |
635 | */ | |
7eb4f455 | 636 | if ((curr & 0x1) == 0 || UINT_CMP_GE(curr, snap + 2)) { |
d4c08f2a | 637 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, "dti"); |
64db4cff PM |
638 | rdp->dynticks_fqs++; |
639 | return 1; | |
640 | } | |
641 | ||
642 | /* Go check for the CPU being offline. */ | |
643 | return rcu_implicit_offline_qs(rdp); | |
644 | } | |
645 | ||
646 | #endif /* #ifdef CONFIG_SMP */ | |
647 | ||
742734ee | 648 | int rcu_cpu_stall_suppress __read_mostly; |
c68de209 | 649 | |
64db4cff PM |
650 | static void record_gp_stall_check_time(struct rcu_state *rsp) |
651 | { | |
652 | rsp->gp_start = jiffies; | |
653 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK; | |
654 | } | |
655 | ||
656 | static void print_other_cpu_stall(struct rcu_state *rsp) | |
657 | { | |
658 | int cpu; | |
659 | long delta; | |
660 | unsigned long flags; | |
9bc8b558 | 661 | int ndetected; |
64db4cff | 662 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff PM |
663 | |
664 | /* Only let one CPU complain about others per time interval. */ | |
665 | ||
1304afb2 | 666 | raw_spin_lock_irqsave(&rnp->lock, flags); |
64db4cff | 667 | delta = jiffies - rsp->jiffies_stall; |
fc2219d4 | 668 | if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) { |
1304afb2 | 669 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
670 | return; |
671 | } | |
672 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
a0b6c9a7 PM |
673 | |
674 | /* | |
675 | * Now rat on any tasks that got kicked up to the root rcu_node | |
676 | * due to CPU offlining. | |
677 | */ | |
9bc8b558 | 678 | ndetected = rcu_print_task_stall(rnp); |
1304afb2 | 679 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 680 | |
8cdd32a9 PM |
681 | /* |
682 | * OK, time to rat on our buddy... | |
683 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
684 | * RCU CPU stall warnings. | |
685 | */ | |
4300aa64 PM |
686 | printk(KERN_ERR "INFO: %s detected stalls on CPUs/tasks: {", |
687 | rsp->name); | |
a0b6c9a7 | 688 | rcu_for_each_leaf_node(rsp, rnp) { |
3acd9eb3 | 689 | raw_spin_lock_irqsave(&rnp->lock, flags); |
9bc8b558 | 690 | ndetected += rcu_print_task_stall(rnp); |
3acd9eb3 | 691 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
a0b6c9a7 | 692 | if (rnp->qsmask == 0) |
64db4cff | 693 | continue; |
a0b6c9a7 | 694 | for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++) |
9bc8b558 | 695 | if (rnp->qsmask & (1UL << cpu)) { |
a0b6c9a7 | 696 | printk(" %d", rnp->grplo + cpu); |
9bc8b558 PM |
697 | ndetected++; |
698 | } | |
64db4cff | 699 | } |
4300aa64 | 700 | printk("} (detected by %d, t=%ld jiffies)\n", |
64db4cff | 701 | smp_processor_id(), (long)(jiffies - rsp->gp_start)); |
9bc8b558 PM |
702 | if (ndetected == 0) |
703 | printk(KERN_ERR "INFO: Stall ended before state dump start\n"); | |
704 | else if (!trigger_all_cpu_backtrace()) | |
4627e240 | 705 | dump_stack(); |
c1dc0b9c | 706 | |
1ed509a2 PM |
707 | /* If so configured, complain about tasks blocking the grace period. */ |
708 | ||
709 | rcu_print_detail_task_stall(rsp); | |
710 | ||
64db4cff PM |
711 | force_quiescent_state(rsp, 0); /* Kick them all. */ |
712 | } | |
713 | ||
714 | static void print_cpu_stall(struct rcu_state *rsp) | |
715 | { | |
716 | unsigned long flags; | |
717 | struct rcu_node *rnp = rcu_get_root(rsp); | |
718 | ||
8cdd32a9 PM |
719 | /* |
720 | * OK, time to rat on ourselves... | |
721 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
722 | * RCU CPU stall warnings. | |
723 | */ | |
4300aa64 PM |
724 | printk(KERN_ERR "INFO: %s detected stall on CPU %d (t=%lu jiffies)\n", |
725 | rsp->name, smp_processor_id(), jiffies - rsp->gp_start); | |
4627e240 PM |
726 | if (!trigger_all_cpu_backtrace()) |
727 | dump_stack(); | |
c1dc0b9c | 728 | |
1304afb2 | 729 | raw_spin_lock_irqsave(&rnp->lock, flags); |
20133cfc | 730 | if (ULONG_CMP_GE(jiffies, rsp->jiffies_stall)) |
64db4cff PM |
731 | rsp->jiffies_stall = |
732 | jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
1304afb2 | 733 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
c1dc0b9c | 734 | |
64db4cff PM |
735 | set_need_resched(); /* kick ourselves to get things going. */ |
736 | } | |
737 | ||
738 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
739 | { | |
bad6e139 PM |
740 | unsigned long j; |
741 | unsigned long js; | |
64db4cff PM |
742 | struct rcu_node *rnp; |
743 | ||
742734ee | 744 | if (rcu_cpu_stall_suppress) |
c68de209 | 745 | return; |
bad6e139 PM |
746 | j = ACCESS_ONCE(jiffies); |
747 | js = ACCESS_ONCE(rsp->jiffies_stall); | |
64db4cff | 748 | rnp = rdp->mynode; |
bad6e139 | 749 | if ((ACCESS_ONCE(rnp->qsmask) & rdp->grpmask) && ULONG_CMP_GE(j, js)) { |
64db4cff PM |
750 | |
751 | /* We haven't checked in, so go dump stack. */ | |
752 | print_cpu_stall(rsp); | |
753 | ||
bad6e139 PM |
754 | } else if (rcu_gp_in_progress(rsp) && |
755 | ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) { | |
64db4cff | 756 | |
bad6e139 | 757 | /* They had a few time units to dump stack, so complain. */ |
64db4cff PM |
758 | print_other_cpu_stall(rsp); |
759 | } | |
760 | } | |
761 | ||
c68de209 PM |
762 | static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr) |
763 | { | |
742734ee | 764 | rcu_cpu_stall_suppress = 1; |
c68de209 PM |
765 | return NOTIFY_DONE; |
766 | } | |
767 | ||
53d84e00 PM |
768 | /** |
769 | * rcu_cpu_stall_reset - prevent further stall warnings in current grace period | |
770 | * | |
771 | * Set the stall-warning timeout way off into the future, thus preventing | |
772 | * any RCU CPU stall-warning messages from appearing in the current set of | |
773 | * RCU grace periods. | |
774 | * | |
775 | * The caller must disable hard irqs. | |
776 | */ | |
777 | void rcu_cpu_stall_reset(void) | |
778 | { | |
779 | rcu_sched_state.jiffies_stall = jiffies + ULONG_MAX / 2; | |
780 | rcu_bh_state.jiffies_stall = jiffies + ULONG_MAX / 2; | |
781 | rcu_preempt_stall_reset(); | |
782 | } | |
783 | ||
c68de209 PM |
784 | static struct notifier_block rcu_panic_block = { |
785 | .notifier_call = rcu_panic, | |
786 | }; | |
787 | ||
788 | static void __init check_cpu_stall_init(void) | |
789 | { | |
790 | atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block); | |
791 | } | |
792 | ||
64db4cff PM |
793 | /* |
794 | * Update CPU-local rcu_data state to record the newly noticed grace period. | |
795 | * This is used both when we started the grace period and when we notice | |
9160306e PM |
796 | * that someone else started the grace period. The caller must hold the |
797 | * ->lock of the leaf rcu_node structure corresponding to the current CPU, | |
798 | * and must have irqs disabled. | |
64db4cff | 799 | */ |
9160306e PM |
800 | static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) |
801 | { | |
802 | if (rdp->gpnum != rnp->gpnum) { | |
121dfc4b PM |
803 | /* |
804 | * If the current grace period is waiting for this CPU, | |
805 | * set up to detect a quiescent state, otherwise don't | |
806 | * go looking for one. | |
807 | */ | |
9160306e | 808 | rdp->gpnum = rnp->gpnum; |
d4c08f2a | 809 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpustart"); |
121dfc4b PM |
810 | if (rnp->qsmask & rdp->grpmask) { |
811 | rdp->qs_pending = 1; | |
e4cc1f22 | 812 | rdp->passed_quiesce = 0; |
121dfc4b PM |
813 | } else |
814 | rdp->qs_pending = 0; | |
9160306e PM |
815 | } |
816 | } | |
817 | ||
64db4cff PM |
818 | static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp) |
819 | { | |
9160306e PM |
820 | unsigned long flags; |
821 | struct rcu_node *rnp; | |
822 | ||
823 | local_irq_save(flags); | |
824 | rnp = rdp->mynode; | |
825 | if (rdp->gpnum == ACCESS_ONCE(rnp->gpnum) || /* outside lock. */ | |
1304afb2 | 826 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
9160306e PM |
827 | local_irq_restore(flags); |
828 | return; | |
829 | } | |
830 | __note_new_gpnum(rsp, rnp, rdp); | |
1304afb2 | 831 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
832 | } |
833 | ||
834 | /* | |
835 | * Did someone else start a new RCU grace period start since we last | |
836 | * checked? Update local state appropriately if so. Must be called | |
837 | * on the CPU corresponding to rdp. | |
838 | */ | |
839 | static int | |
840 | check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp) | |
841 | { | |
842 | unsigned long flags; | |
843 | int ret = 0; | |
844 | ||
845 | local_irq_save(flags); | |
846 | if (rdp->gpnum != rsp->gpnum) { | |
847 | note_new_gpnum(rsp, rdp); | |
848 | ret = 1; | |
849 | } | |
850 | local_irq_restore(flags); | |
851 | return ret; | |
852 | } | |
853 | ||
d09b62df PM |
854 | /* |
855 | * Advance this CPU's callbacks, but only if the current grace period | |
856 | * has ended. This may be called only from the CPU to whom the rdp | |
857 | * belongs. In addition, the corresponding leaf rcu_node structure's | |
858 | * ->lock must be held by the caller, with irqs disabled. | |
859 | */ | |
860 | static void | |
861 | __rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
862 | { | |
863 | /* Did another grace period end? */ | |
864 | if (rdp->completed != rnp->completed) { | |
865 | ||
866 | /* Advance callbacks. No harm if list empty. */ | |
867 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL]; | |
868 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL]; | |
869 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
870 | ||
871 | /* Remember that we saw this grace-period completion. */ | |
872 | rdp->completed = rnp->completed; | |
d4c08f2a | 873 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuend"); |
20377f32 | 874 | |
5ff8e6f0 FW |
875 | /* |
876 | * If we were in an extended quiescent state, we may have | |
121dfc4b | 877 | * missed some grace periods that others CPUs handled on |
5ff8e6f0 | 878 | * our behalf. Catch up with this state to avoid noting |
121dfc4b PM |
879 | * spurious new grace periods. If another grace period |
880 | * has started, then rnp->gpnum will have advanced, so | |
881 | * we will detect this later on. | |
5ff8e6f0 | 882 | */ |
121dfc4b | 883 | if (ULONG_CMP_LT(rdp->gpnum, rdp->completed)) |
5ff8e6f0 FW |
884 | rdp->gpnum = rdp->completed; |
885 | ||
20377f32 | 886 | /* |
121dfc4b PM |
887 | * If RCU does not need a quiescent state from this CPU, |
888 | * then make sure that this CPU doesn't go looking for one. | |
20377f32 | 889 | */ |
121dfc4b | 890 | if ((rnp->qsmask & rdp->grpmask) == 0) |
20377f32 | 891 | rdp->qs_pending = 0; |
d09b62df PM |
892 | } |
893 | } | |
894 | ||
895 | /* | |
896 | * Advance this CPU's callbacks, but only if the current grace period | |
897 | * has ended. This may be called only from the CPU to whom the rdp | |
898 | * belongs. | |
899 | */ | |
900 | static void | |
901 | rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp) | |
902 | { | |
903 | unsigned long flags; | |
904 | struct rcu_node *rnp; | |
905 | ||
906 | local_irq_save(flags); | |
907 | rnp = rdp->mynode; | |
908 | if (rdp->completed == ACCESS_ONCE(rnp->completed) || /* outside lock. */ | |
1304afb2 | 909 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
d09b62df PM |
910 | local_irq_restore(flags); |
911 | return; | |
912 | } | |
913 | __rcu_process_gp_end(rsp, rnp, rdp); | |
1304afb2 | 914 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
d09b62df PM |
915 | } |
916 | ||
917 | /* | |
918 | * Do per-CPU grace-period initialization for running CPU. The caller | |
919 | * must hold the lock of the leaf rcu_node structure corresponding to | |
920 | * this CPU. | |
921 | */ | |
922 | static void | |
923 | rcu_start_gp_per_cpu(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
924 | { | |
925 | /* Prior grace period ended, so advance callbacks for current CPU. */ | |
926 | __rcu_process_gp_end(rsp, rnp, rdp); | |
927 | ||
928 | /* | |
929 | * Because this CPU just now started the new grace period, we know | |
930 | * that all of its callbacks will be covered by this upcoming grace | |
931 | * period, even the ones that were registered arbitrarily recently. | |
932 | * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL. | |
933 | * | |
934 | * Other CPUs cannot be sure exactly when the grace period started. | |
935 | * Therefore, their recently registered callbacks must pass through | |
936 | * an additional RCU_NEXT_READY stage, so that they will be handled | |
937 | * by the next RCU grace period. | |
938 | */ | |
939 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
940 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
9160306e PM |
941 | |
942 | /* Set state so that this CPU will detect the next quiescent state. */ | |
943 | __note_new_gpnum(rsp, rnp, rdp); | |
d09b62df PM |
944 | } |
945 | ||
64db4cff PM |
946 | /* |
947 | * Start a new RCU grace period if warranted, re-initializing the hierarchy | |
948 | * in preparation for detecting the next grace period. The caller must hold | |
949 | * the root node's ->lock, which is released before return. Hard irqs must | |
950 | * be disabled. | |
951 | */ | |
952 | static void | |
953 | rcu_start_gp(struct rcu_state *rsp, unsigned long flags) | |
954 | __releases(rcu_get_root(rsp)->lock) | |
955 | { | |
394f99a9 | 956 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
64db4cff | 957 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff | 958 | |
037067a1 | 959 | if (!rcu_scheduler_fully_active || |
afe24b12 PM |
960 | !cpu_needs_another_gp(rsp, rdp)) { |
961 | /* | |
962 | * Either the scheduler hasn't yet spawned the first | |
963 | * non-idle task or this CPU does not need another | |
964 | * grace period. Either way, don't start a new grace | |
965 | * period. | |
966 | */ | |
967 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
968 | return; | |
969 | } | |
b32e9eb6 | 970 | |
afe24b12 | 971 | if (rsp->fqs_active) { |
b32e9eb6 | 972 | /* |
afe24b12 PM |
973 | * This CPU needs a grace period, but force_quiescent_state() |
974 | * is running. Tell it to start one on this CPU's behalf. | |
b32e9eb6 | 975 | */ |
afe24b12 PM |
976 | rsp->fqs_need_gp = 1; |
977 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
64db4cff PM |
978 | return; |
979 | } | |
980 | ||
981 | /* Advance to a new grace period and initialize state. */ | |
982 | rsp->gpnum++; | |
d4c08f2a | 983 | trace_rcu_grace_period(rsp->name, rsp->gpnum, "start"); |
af446b70 PM |
984 | WARN_ON_ONCE(rsp->fqs_state == RCU_GP_INIT); |
985 | rsp->fqs_state = RCU_GP_INIT; /* Hold off force_quiescent_state. */ | |
64db4cff | 986 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
64db4cff | 987 | record_gp_stall_check_time(rsp); |
64db4cff | 988 | |
64db4cff PM |
989 | /* Special-case the common single-level case. */ |
990 | if (NUM_RCU_NODES == 1) { | |
b0e165c0 | 991 | rcu_preempt_check_blocked_tasks(rnp); |
28ecd580 | 992 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 993 | rnp->gpnum = rsp->gpnum; |
d09b62df | 994 | rnp->completed = rsp->completed; |
af446b70 | 995 | rsp->fqs_state = RCU_SIGNAL_INIT; /* force_quiescent_state OK */ |
d09b62df | 996 | rcu_start_gp_per_cpu(rsp, rnp, rdp); |
27f4d280 | 997 | rcu_preempt_boost_start_gp(rnp); |
d4c08f2a PM |
998 | trace_rcu_grace_period_init(rsp->name, rnp->gpnum, |
999 | rnp->level, rnp->grplo, | |
1000 | rnp->grphi, rnp->qsmask); | |
1304afb2 | 1001 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1002 | return; |
1003 | } | |
1004 | ||
1304afb2 | 1005 | raw_spin_unlock(&rnp->lock); /* leave irqs disabled. */ |
64db4cff PM |
1006 | |
1007 | ||
1008 | /* Exclude any concurrent CPU-hotplug operations. */ | |
1304afb2 | 1009 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
64db4cff PM |
1010 | |
1011 | /* | |
b835db1f PM |
1012 | * Set the quiescent-state-needed bits in all the rcu_node |
1013 | * structures for all currently online CPUs in breadth-first | |
1014 | * order, starting from the root rcu_node structure. This | |
1015 | * operation relies on the layout of the hierarchy within the | |
1016 | * rsp->node[] array. Note that other CPUs will access only | |
1017 | * the leaves of the hierarchy, which still indicate that no | |
1018 | * grace period is in progress, at least until the corresponding | |
1019 | * leaf node has been initialized. In addition, we have excluded | |
1020 | * CPU-hotplug operations. | |
64db4cff PM |
1021 | * |
1022 | * Note that the grace period cannot complete until we finish | |
1023 | * the initialization process, as there will be at least one | |
1024 | * qsmask bit set in the root node until that time, namely the | |
b835db1f PM |
1025 | * one corresponding to this CPU, due to the fact that we have |
1026 | * irqs disabled. | |
64db4cff | 1027 | */ |
a0b6c9a7 | 1028 | rcu_for_each_node_breadth_first(rsp, rnp) { |
1304afb2 | 1029 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
b0e165c0 | 1030 | rcu_preempt_check_blocked_tasks(rnp); |
49e29126 | 1031 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 1032 | rnp->gpnum = rsp->gpnum; |
d09b62df PM |
1033 | rnp->completed = rsp->completed; |
1034 | if (rnp == rdp->mynode) | |
1035 | rcu_start_gp_per_cpu(rsp, rnp, rdp); | |
27f4d280 | 1036 | rcu_preempt_boost_start_gp(rnp); |
d4c08f2a PM |
1037 | trace_rcu_grace_period_init(rsp->name, rnp->gpnum, |
1038 | rnp->level, rnp->grplo, | |
1039 | rnp->grphi, rnp->qsmask); | |
1304afb2 | 1040 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
1041 | } |
1042 | ||
83f5b01f | 1043 | rnp = rcu_get_root(rsp); |
1304afb2 | 1044 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
af446b70 | 1045 | rsp->fqs_state = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */ |
1304afb2 PM |
1046 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
1047 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); | |
64db4cff PM |
1048 | } |
1049 | ||
f41d911f | 1050 | /* |
d3f6bad3 PM |
1051 | * Report a full set of quiescent states to the specified rcu_state |
1052 | * data structure. This involves cleaning up after the prior grace | |
1053 | * period and letting rcu_start_gp() start up the next grace period | |
1054 | * if one is needed. Note that the caller must hold rnp->lock, as | |
1055 | * required by rcu_start_gp(), which will release it. | |
f41d911f | 1056 | */ |
d3f6bad3 | 1057 | static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags) |
fc2219d4 | 1058 | __releases(rcu_get_root(rsp)->lock) |
f41d911f | 1059 | { |
15ba0ba8 | 1060 | unsigned long gp_duration; |
afe24b12 PM |
1061 | struct rcu_node *rnp = rcu_get_root(rsp); |
1062 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); | |
15ba0ba8 | 1063 | |
fc2219d4 | 1064 | WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); |
0bbcc529 PM |
1065 | |
1066 | /* | |
1067 | * Ensure that all grace-period and pre-grace-period activity | |
1068 | * is seen before the assignment to rsp->completed. | |
1069 | */ | |
1070 | smp_mb(); /* See above block comment. */ | |
15ba0ba8 PM |
1071 | gp_duration = jiffies - rsp->gp_start; |
1072 | if (gp_duration > rsp->gp_max) | |
1073 | rsp->gp_max = gp_duration; | |
afe24b12 PM |
1074 | |
1075 | /* | |
1076 | * We know the grace period is complete, but to everyone else | |
1077 | * it appears to still be ongoing. But it is also the case | |
1078 | * that to everyone else it looks like there is nothing that | |
1079 | * they can do to advance the grace period. It is therefore | |
1080 | * safe for us to drop the lock in order to mark the grace | |
1081 | * period as completed in all of the rcu_node structures. | |
1082 | * | |
1083 | * But if this CPU needs another grace period, it will take | |
1084 | * care of this while initializing the next grace period. | |
1085 | * We use RCU_WAIT_TAIL instead of the usual RCU_DONE_TAIL | |
1086 | * because the callbacks have not yet been advanced: Those | |
1087 | * callbacks are waiting on the grace period that just now | |
1088 | * completed. | |
1089 | */ | |
1090 | if (*rdp->nxttail[RCU_WAIT_TAIL] == NULL) { | |
1091 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ | |
1092 | ||
1093 | /* | |
1094 | * Propagate new ->completed value to rcu_node structures | |
1095 | * so that other CPUs don't have to wait until the start | |
1096 | * of the next grace period to process their callbacks. | |
1097 | */ | |
1098 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
1099 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ | |
1100 | rnp->completed = rsp->gpnum; | |
1101 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ | |
1102 | } | |
1103 | rnp = rcu_get_root(rsp); | |
1104 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ | |
1105 | } | |
1106 | ||
1107 | rsp->completed = rsp->gpnum; /* Declare the grace period complete. */ | |
d4c08f2a | 1108 | trace_rcu_grace_period(rsp->name, rsp->completed, "end"); |
af446b70 | 1109 | rsp->fqs_state = RCU_GP_IDLE; |
f41d911f PM |
1110 | rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */ |
1111 | } | |
1112 | ||
64db4cff | 1113 | /* |
d3f6bad3 PM |
1114 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
1115 | * Allows quiescent states for a group of CPUs to be reported at one go | |
1116 | * to the specified rcu_node structure, though all the CPUs in the group | |
1117 | * must be represented by the same rcu_node structure (which need not be | |
1118 | * a leaf rcu_node structure, though it often will be). That structure's | |
1119 | * lock must be held upon entry, and it is released before return. | |
64db4cff PM |
1120 | */ |
1121 | static void | |
d3f6bad3 PM |
1122 | rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp, |
1123 | struct rcu_node *rnp, unsigned long flags) | |
64db4cff PM |
1124 | __releases(rnp->lock) |
1125 | { | |
28ecd580 PM |
1126 | struct rcu_node *rnp_c; |
1127 | ||
64db4cff PM |
1128 | /* Walk up the rcu_node hierarchy. */ |
1129 | for (;;) { | |
1130 | if (!(rnp->qsmask & mask)) { | |
1131 | ||
1132 | /* Our bit has already been cleared, so done. */ | |
1304afb2 | 1133 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1134 | return; |
1135 | } | |
1136 | rnp->qsmask &= ~mask; | |
d4c08f2a PM |
1137 | trace_rcu_quiescent_state_report(rsp->name, rnp->gpnum, |
1138 | mask, rnp->qsmask, rnp->level, | |
1139 | rnp->grplo, rnp->grphi, | |
1140 | !!rnp->gp_tasks); | |
27f4d280 | 1141 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
64db4cff PM |
1142 | |
1143 | /* Other bits still set at this level, so done. */ | |
1304afb2 | 1144 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1145 | return; |
1146 | } | |
1147 | mask = rnp->grpmask; | |
1148 | if (rnp->parent == NULL) { | |
1149 | ||
1150 | /* No more levels. Exit loop holding root lock. */ | |
1151 | ||
1152 | break; | |
1153 | } | |
1304afb2 | 1154 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
28ecd580 | 1155 | rnp_c = rnp; |
64db4cff | 1156 | rnp = rnp->parent; |
1304afb2 | 1157 | raw_spin_lock_irqsave(&rnp->lock, flags); |
28ecd580 | 1158 | WARN_ON_ONCE(rnp_c->qsmask); |
64db4cff PM |
1159 | } |
1160 | ||
1161 | /* | |
1162 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 1163 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 1164 | * to clean up and start the next grace period if one is needed. |
64db4cff | 1165 | */ |
d3f6bad3 | 1166 | rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */ |
64db4cff PM |
1167 | } |
1168 | ||
1169 | /* | |
d3f6bad3 PM |
1170 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
1171 | * structure. This must be either called from the specified CPU, or | |
1172 | * called when the specified CPU is known to be offline (and when it is | |
1173 | * also known that no other CPU is concurrently trying to help the offline | |
1174 | * CPU). The lastcomp argument is used to make sure we are still in the | |
1175 | * grace period of interest. We don't want to end the current grace period | |
1176 | * based on quiescent states detected in an earlier grace period! | |
64db4cff PM |
1177 | */ |
1178 | static void | |
e4cc1f22 | 1179 | rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastgp) |
64db4cff PM |
1180 | { |
1181 | unsigned long flags; | |
1182 | unsigned long mask; | |
1183 | struct rcu_node *rnp; | |
1184 | ||
1185 | rnp = rdp->mynode; | |
1304afb2 | 1186 | raw_spin_lock_irqsave(&rnp->lock, flags); |
e4cc1f22 | 1187 | if (lastgp != rnp->gpnum || rnp->completed == rnp->gpnum) { |
64db4cff PM |
1188 | |
1189 | /* | |
e4cc1f22 PM |
1190 | * The grace period in which this quiescent state was |
1191 | * recorded has ended, so don't report it upwards. | |
1192 | * We will instead need a new quiescent state that lies | |
1193 | * within the current grace period. | |
64db4cff | 1194 | */ |
e4cc1f22 | 1195 | rdp->passed_quiesce = 0; /* need qs for new gp. */ |
1304afb2 | 1196 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1197 | return; |
1198 | } | |
1199 | mask = rdp->grpmask; | |
1200 | if ((rnp->qsmask & mask) == 0) { | |
1304afb2 | 1201 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1202 | } else { |
1203 | rdp->qs_pending = 0; | |
1204 | ||
1205 | /* | |
1206 | * This GP can't end until cpu checks in, so all of our | |
1207 | * callbacks can be processed during the next GP. | |
1208 | */ | |
64db4cff PM |
1209 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
1210 | ||
d3f6bad3 | 1211 | rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */ |
64db4cff PM |
1212 | } |
1213 | } | |
1214 | ||
1215 | /* | |
1216 | * Check to see if there is a new grace period of which this CPU | |
1217 | * is not yet aware, and if so, set up local rcu_data state for it. | |
1218 | * Otherwise, see if this CPU has just passed through its first | |
1219 | * quiescent state for this grace period, and record that fact if so. | |
1220 | */ | |
1221 | static void | |
1222 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | |
1223 | { | |
1224 | /* If there is now a new grace period, record and return. */ | |
1225 | if (check_for_new_grace_period(rsp, rdp)) | |
1226 | return; | |
1227 | ||
1228 | /* | |
1229 | * Does this CPU still need to do its part for current grace period? | |
1230 | * If no, return and let the other CPUs do their part as well. | |
1231 | */ | |
1232 | if (!rdp->qs_pending) | |
1233 | return; | |
1234 | ||
1235 | /* | |
1236 | * Was there a quiescent state since the beginning of the grace | |
1237 | * period? If no, then exit and wait for the next call. | |
1238 | */ | |
e4cc1f22 | 1239 | if (!rdp->passed_quiesce) |
64db4cff PM |
1240 | return; |
1241 | ||
d3f6bad3 PM |
1242 | /* |
1243 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
1244 | * judge of that). | |
1245 | */ | |
e4cc1f22 | 1246 | rcu_report_qs_rdp(rdp->cpu, rsp, rdp, rdp->passed_quiesce_gpnum); |
64db4cff PM |
1247 | } |
1248 | ||
1249 | #ifdef CONFIG_HOTPLUG_CPU | |
1250 | ||
e74f4c45 | 1251 | /* |
29494be7 LJ |
1252 | * Move a dying CPU's RCU callbacks to online CPU's callback list. |
1253 | * Synchronization is not required because this function executes | |
1254 | * in stop_machine() context. | |
e74f4c45 | 1255 | */ |
29494be7 | 1256 | static void rcu_send_cbs_to_online(struct rcu_state *rsp) |
e74f4c45 PM |
1257 | { |
1258 | int i; | |
29494be7 LJ |
1259 | /* current DYING CPU is cleared in the cpu_online_mask */ |
1260 | int receive_cpu = cpumask_any(cpu_online_mask); | |
394f99a9 | 1261 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
29494be7 | 1262 | struct rcu_data *receive_rdp = per_cpu_ptr(rsp->rda, receive_cpu); |
e74f4c45 PM |
1263 | |
1264 | if (rdp->nxtlist == NULL) | |
1265 | return; /* irqs disabled, so comparison is stable. */ | |
29494be7 LJ |
1266 | |
1267 | *receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist; | |
1268 | receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
1269 | receive_rdp->qlen += rdp->qlen; | |
1270 | receive_rdp->n_cbs_adopted += rdp->qlen; | |
1271 | rdp->n_cbs_orphaned += rdp->qlen; | |
1272 | ||
e74f4c45 PM |
1273 | rdp->nxtlist = NULL; |
1274 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
1275 | rdp->nxttail[i] = &rdp->nxtlist; | |
e74f4c45 | 1276 | rdp->qlen = 0; |
e74f4c45 PM |
1277 | } |
1278 | ||
64db4cff PM |
1279 | /* |
1280 | * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy | |
1281 | * and move all callbacks from the outgoing CPU to the current one. | |
a26ac245 PM |
1282 | * There can only be one CPU hotplug operation at a time, so no other |
1283 | * CPU can be attempting to update rcu_cpu_kthread_task. | |
64db4cff PM |
1284 | */ |
1285 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp) | |
1286 | { | |
64db4cff | 1287 | unsigned long flags; |
64db4cff | 1288 | unsigned long mask; |
d9a3da06 | 1289 | int need_report = 0; |
394f99a9 | 1290 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff | 1291 | struct rcu_node *rnp; |
a26ac245 | 1292 | |
f8b7fc6b | 1293 | rcu_stop_cpu_kthread(cpu); |
64db4cff PM |
1294 | |
1295 | /* Exclude any attempts to start a new grace period. */ | |
1304afb2 | 1296 | raw_spin_lock_irqsave(&rsp->onofflock, flags); |
64db4cff PM |
1297 | |
1298 | /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */ | |
28ecd580 | 1299 | rnp = rdp->mynode; /* this is the outgoing CPU's rnp. */ |
64db4cff PM |
1300 | mask = rdp->grpmask; /* rnp->grplo is constant. */ |
1301 | do { | |
1304afb2 | 1302 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
1303 | rnp->qsmaskinit &= ~mask; |
1304 | if (rnp->qsmaskinit != 0) { | |
b668c9cf | 1305 | if (rnp != rdp->mynode) |
1304afb2 | 1306 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
d4c08f2a PM |
1307 | else |
1308 | trace_rcu_grace_period(rsp->name, | |
1309 | rnp->gpnum + 1 - | |
1310 | !!(rnp->qsmask & mask), | |
1311 | "cpuofl"); | |
64db4cff PM |
1312 | break; |
1313 | } | |
d4c08f2a PM |
1314 | if (rnp == rdp->mynode) { |
1315 | trace_rcu_grace_period(rsp->name, | |
1316 | rnp->gpnum + 1 - | |
1317 | !!(rnp->qsmask & mask), | |
1318 | "cpuofl"); | |
d9a3da06 | 1319 | need_report = rcu_preempt_offline_tasks(rsp, rnp, rdp); |
d4c08f2a | 1320 | } else |
1304afb2 | 1321 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff | 1322 | mask = rnp->grpmask; |
64db4cff PM |
1323 | rnp = rnp->parent; |
1324 | } while (rnp != NULL); | |
64db4cff | 1325 | |
b668c9cf PM |
1326 | /* |
1327 | * We still hold the leaf rcu_node structure lock here, and | |
1328 | * irqs are still disabled. The reason for this subterfuge is | |
d3f6bad3 PM |
1329 | * because invoking rcu_report_unblock_qs_rnp() with ->onofflock |
1330 | * held leads to deadlock. | |
b668c9cf | 1331 | */ |
1304afb2 | 1332 | raw_spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ |
b668c9cf | 1333 | rnp = rdp->mynode; |
d9a3da06 | 1334 | if (need_report & RCU_OFL_TASKS_NORM_GP) |
d3f6bad3 | 1335 | rcu_report_unblock_qs_rnp(rnp, flags); |
b668c9cf | 1336 | else |
1304afb2 | 1337 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
d9a3da06 | 1338 | if (need_report & RCU_OFL_TASKS_EXP_GP) |
b40d293e | 1339 | rcu_report_exp_rnp(rsp, rnp, true); |
1217ed1b | 1340 | rcu_node_kthread_setaffinity(rnp, -1); |
64db4cff PM |
1341 | } |
1342 | ||
1343 | /* | |
1344 | * Remove the specified CPU from the RCU hierarchy and move any pending | |
1345 | * callbacks that it might have to the current CPU. This code assumes | |
1346 | * that at least one CPU in the system will remain running at all times. | |
1347 | * Any attempt to offline -all- CPUs is likely to strand RCU callbacks. | |
1348 | */ | |
1349 | static void rcu_offline_cpu(int cpu) | |
1350 | { | |
d6714c22 | 1351 | __rcu_offline_cpu(cpu, &rcu_sched_state); |
64db4cff | 1352 | __rcu_offline_cpu(cpu, &rcu_bh_state); |
33f76148 | 1353 | rcu_preempt_offline_cpu(cpu); |
64db4cff PM |
1354 | } |
1355 | ||
1356 | #else /* #ifdef CONFIG_HOTPLUG_CPU */ | |
1357 | ||
29494be7 | 1358 | static void rcu_send_cbs_to_online(struct rcu_state *rsp) |
e74f4c45 PM |
1359 | { |
1360 | } | |
1361 | ||
64db4cff PM |
1362 | static void rcu_offline_cpu(int cpu) |
1363 | { | |
1364 | } | |
1365 | ||
1366 | #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */ | |
1367 | ||
1368 | /* | |
1369 | * Invoke any RCU callbacks that have made it to the end of their grace | |
1370 | * period. Thottle as specified by rdp->blimit. | |
1371 | */ | |
37c72e56 | 1372 | static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
1373 | { |
1374 | unsigned long flags; | |
1375 | struct rcu_head *next, *list, **tail; | |
29c00b4a | 1376 | int bl, count; |
64db4cff PM |
1377 | |
1378 | /* If no callbacks are ready, just return.*/ | |
29c00b4a | 1379 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) { |
72fe701b PM |
1380 | trace_rcu_batch_start(rsp->name, 0, 0); |
1381 | trace_rcu_batch_end(rsp->name, 0); | |
64db4cff | 1382 | return; |
29c00b4a | 1383 | } |
64db4cff PM |
1384 | |
1385 | /* | |
1386 | * Extract the list of ready callbacks, disabling to prevent | |
1387 | * races with call_rcu() from interrupt handlers. | |
1388 | */ | |
1389 | local_irq_save(flags); | |
29c00b4a | 1390 | bl = rdp->blimit; |
72fe701b | 1391 | trace_rcu_batch_start(rsp->name, rdp->qlen, bl); |
64db4cff PM |
1392 | list = rdp->nxtlist; |
1393 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; | |
1394 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | |
1395 | tail = rdp->nxttail[RCU_DONE_TAIL]; | |
1396 | for (count = RCU_NEXT_SIZE - 1; count >= 0; count--) | |
1397 | if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL]) | |
1398 | rdp->nxttail[count] = &rdp->nxtlist; | |
1399 | local_irq_restore(flags); | |
1400 | ||
1401 | /* Invoke callbacks. */ | |
1402 | count = 0; | |
1403 | while (list) { | |
1404 | next = list->next; | |
1405 | prefetch(next); | |
551d55a9 | 1406 | debug_rcu_head_unqueue(list); |
d4c08f2a | 1407 | __rcu_reclaim(rsp->name, list); |
64db4cff | 1408 | list = next; |
29c00b4a | 1409 | if (++count >= bl) |
64db4cff PM |
1410 | break; |
1411 | } | |
1412 | ||
1413 | local_irq_save(flags); | |
72fe701b | 1414 | trace_rcu_batch_end(rsp->name, count); |
64db4cff PM |
1415 | |
1416 | /* Update count, and requeue any remaining callbacks. */ | |
1417 | rdp->qlen -= count; | |
269dcc1c | 1418 | rdp->n_cbs_invoked += count; |
64db4cff PM |
1419 | if (list != NULL) { |
1420 | *tail = rdp->nxtlist; | |
1421 | rdp->nxtlist = list; | |
1422 | for (count = 0; count < RCU_NEXT_SIZE; count++) | |
1423 | if (&rdp->nxtlist == rdp->nxttail[count]) | |
1424 | rdp->nxttail[count] = tail; | |
1425 | else | |
1426 | break; | |
1427 | } | |
1428 | ||
1429 | /* Reinstate batch limit if we have worked down the excess. */ | |
1430 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) | |
1431 | rdp->blimit = blimit; | |
1432 | ||
37c72e56 PM |
1433 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
1434 | if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) { | |
1435 | rdp->qlen_last_fqs_check = 0; | |
1436 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1437 | } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark) | |
1438 | rdp->qlen_last_fqs_check = rdp->qlen; | |
1439 | ||
64db4cff PM |
1440 | local_irq_restore(flags); |
1441 | ||
e0f23060 | 1442 | /* Re-invoke RCU core processing if there are callbacks remaining. */ |
64db4cff | 1443 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
a46e0899 | 1444 | invoke_rcu_core(); |
64db4cff PM |
1445 | } |
1446 | ||
1447 | /* | |
1448 | * Check to see if this CPU is in a non-context-switch quiescent state | |
1449 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
e0f23060 | 1450 | * Also schedule RCU core processing. |
64db4cff | 1451 | * |
9b2e4f18 | 1452 | * This function must be called from hardirq context. It is normally |
64db4cff PM |
1453 | * invoked from the scheduling-clock interrupt. If rcu_pending returns |
1454 | * false, there is no point in invoking rcu_check_callbacks(). | |
1455 | */ | |
1456 | void rcu_check_callbacks(int cpu, int user) | |
1457 | { | |
300df91c | 1458 | trace_rcu_utilization("Start scheduler-tick"); |
9b2e4f18 | 1459 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
64db4cff PM |
1460 | |
1461 | /* | |
1462 | * Get here if this CPU took its interrupt from user | |
1463 | * mode or from the idle loop, and if this is not a | |
1464 | * nested interrupt. In this case, the CPU is in | |
d6714c22 | 1465 | * a quiescent state, so note it. |
64db4cff PM |
1466 | * |
1467 | * No memory barrier is required here because both | |
d6714c22 PM |
1468 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
1469 | * variables that other CPUs neither access nor modify, | |
1470 | * at least not while the corresponding CPU is online. | |
64db4cff PM |
1471 | */ |
1472 | ||
d6714c22 PM |
1473 | rcu_sched_qs(cpu); |
1474 | rcu_bh_qs(cpu); | |
64db4cff PM |
1475 | |
1476 | } else if (!in_softirq()) { | |
1477 | ||
1478 | /* | |
1479 | * Get here if this CPU did not take its interrupt from | |
1480 | * softirq, in other words, if it is not interrupting | |
1481 | * a rcu_bh read-side critical section. This is an _bh | |
d6714c22 | 1482 | * critical section, so note it. |
64db4cff PM |
1483 | */ |
1484 | ||
d6714c22 | 1485 | rcu_bh_qs(cpu); |
64db4cff | 1486 | } |
f41d911f | 1487 | rcu_preempt_check_callbacks(cpu); |
d21670ac | 1488 | if (rcu_pending(cpu)) |
a46e0899 | 1489 | invoke_rcu_core(); |
300df91c | 1490 | trace_rcu_utilization("End scheduler-tick"); |
64db4cff PM |
1491 | } |
1492 | ||
1493 | #ifdef CONFIG_SMP | |
1494 | ||
1495 | /* | |
1496 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
1497 | * have not yet encountered a quiescent state, using the function specified. | |
27f4d280 PM |
1498 | * Also initiate boosting for any threads blocked on the root rcu_node. |
1499 | * | |
ee47eb9f | 1500 | * The caller must have suppressed start of new grace periods. |
64db4cff | 1501 | */ |
45f014c5 | 1502 | static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *)) |
64db4cff PM |
1503 | { |
1504 | unsigned long bit; | |
1505 | int cpu; | |
1506 | unsigned long flags; | |
1507 | unsigned long mask; | |
a0b6c9a7 | 1508 | struct rcu_node *rnp; |
64db4cff | 1509 | |
a0b6c9a7 | 1510 | rcu_for_each_leaf_node(rsp, rnp) { |
64db4cff | 1511 | mask = 0; |
1304afb2 | 1512 | raw_spin_lock_irqsave(&rnp->lock, flags); |
ee47eb9f | 1513 | if (!rcu_gp_in_progress(rsp)) { |
1304afb2 | 1514 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
0f10dc82 | 1515 | return; |
64db4cff | 1516 | } |
a0b6c9a7 | 1517 | if (rnp->qsmask == 0) { |
1217ed1b | 1518 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock */ |
64db4cff PM |
1519 | continue; |
1520 | } | |
a0b6c9a7 | 1521 | cpu = rnp->grplo; |
64db4cff | 1522 | bit = 1; |
a0b6c9a7 | 1523 | for (; cpu <= rnp->grphi; cpu++, bit <<= 1) { |
394f99a9 LJ |
1524 | if ((rnp->qsmask & bit) != 0 && |
1525 | f(per_cpu_ptr(rsp->rda, cpu))) | |
64db4cff PM |
1526 | mask |= bit; |
1527 | } | |
45f014c5 | 1528 | if (mask != 0) { |
64db4cff | 1529 | |
d3f6bad3 PM |
1530 | /* rcu_report_qs_rnp() releases rnp->lock. */ |
1531 | rcu_report_qs_rnp(mask, rsp, rnp, flags); | |
64db4cff PM |
1532 | continue; |
1533 | } | |
1304afb2 | 1534 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 1535 | } |
27f4d280 | 1536 | rnp = rcu_get_root(rsp); |
1217ed1b PM |
1537 | if (rnp->qsmask == 0) { |
1538 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
1539 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */ | |
1540 | } | |
64db4cff PM |
1541 | } |
1542 | ||
1543 | /* | |
1544 | * Force quiescent states on reluctant CPUs, and also detect which | |
1545 | * CPUs are in dyntick-idle mode. | |
1546 | */ | |
1547 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1548 | { | |
1549 | unsigned long flags; | |
64db4cff | 1550 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff | 1551 | |
300df91c PM |
1552 | trace_rcu_utilization("Start fqs"); |
1553 | if (!rcu_gp_in_progress(rsp)) { | |
1554 | trace_rcu_utilization("End fqs"); | |
64db4cff | 1555 | return; /* No grace period in progress, nothing to force. */ |
300df91c | 1556 | } |
1304afb2 | 1557 | if (!raw_spin_trylock_irqsave(&rsp->fqslock, flags)) { |
64db4cff | 1558 | rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ |
300df91c | 1559 | trace_rcu_utilization("End fqs"); |
64db4cff PM |
1560 | return; /* Someone else is already on the job. */ |
1561 | } | |
20133cfc | 1562 | if (relaxed && ULONG_CMP_GE(rsp->jiffies_force_qs, jiffies)) |
f96e9232 | 1563 | goto unlock_fqs_ret; /* no emergency and done recently. */ |
64db4cff | 1564 | rsp->n_force_qs++; |
1304afb2 | 1565 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
64db4cff | 1566 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
560d4bc0 | 1567 | if(!rcu_gp_in_progress(rsp)) { |
64db4cff | 1568 | rsp->n_force_qs_ngp++; |
1304afb2 | 1569 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1570 | goto unlock_fqs_ret; /* no GP in progress, time updated. */ |
64db4cff | 1571 | } |
07079d53 | 1572 | rsp->fqs_active = 1; |
af446b70 | 1573 | switch (rsp->fqs_state) { |
83f5b01f | 1574 | case RCU_GP_IDLE: |
64db4cff PM |
1575 | case RCU_GP_INIT: |
1576 | ||
83f5b01f | 1577 | break; /* grace period idle or initializing, ignore. */ |
64db4cff PM |
1578 | |
1579 | case RCU_SAVE_DYNTICK: | |
64db4cff PM |
1580 | if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK) |
1581 | break; /* So gcc recognizes the dead code. */ | |
1582 | ||
f261414f LJ |
1583 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
1584 | ||
64db4cff | 1585 | /* Record dyntick-idle state. */ |
45f014c5 | 1586 | force_qs_rnp(rsp, dyntick_save_progress_counter); |
1304afb2 | 1587 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
ee47eb9f | 1588 | if (rcu_gp_in_progress(rsp)) |
af446b70 | 1589 | rsp->fqs_state = RCU_FORCE_QS; |
ee47eb9f | 1590 | break; |
64db4cff PM |
1591 | |
1592 | case RCU_FORCE_QS: | |
1593 | ||
1594 | /* Check dyntick-idle state, send IPI to laggarts. */ | |
1304afb2 | 1595 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
45f014c5 | 1596 | force_qs_rnp(rsp, rcu_implicit_dynticks_qs); |
64db4cff PM |
1597 | |
1598 | /* Leave state in case more forcing is required. */ | |
1599 | ||
1304afb2 | 1600 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
f96e9232 | 1601 | break; |
64db4cff | 1602 | } |
07079d53 | 1603 | rsp->fqs_active = 0; |
46a1e34e | 1604 | if (rsp->fqs_need_gp) { |
1304afb2 | 1605 | raw_spin_unlock(&rsp->fqslock); /* irqs remain disabled */ |
46a1e34e PM |
1606 | rsp->fqs_need_gp = 0; |
1607 | rcu_start_gp(rsp, flags); /* releases rnp->lock */ | |
300df91c | 1608 | trace_rcu_utilization("End fqs"); |
46a1e34e PM |
1609 | return; |
1610 | } | |
1304afb2 | 1611 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1612 | unlock_fqs_ret: |
1304afb2 | 1613 | raw_spin_unlock_irqrestore(&rsp->fqslock, flags); |
300df91c | 1614 | trace_rcu_utilization("End fqs"); |
64db4cff PM |
1615 | } |
1616 | ||
1617 | #else /* #ifdef CONFIG_SMP */ | |
1618 | ||
1619 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1620 | { | |
1621 | set_need_resched(); | |
1622 | } | |
1623 | ||
1624 | #endif /* #else #ifdef CONFIG_SMP */ | |
1625 | ||
1626 | /* | |
e0f23060 PM |
1627 | * This does the RCU core processing work for the specified rcu_state |
1628 | * and rcu_data structures. This may be called only from the CPU to | |
1629 | * whom the rdp belongs. | |
64db4cff PM |
1630 | */ |
1631 | static void | |
1632 | __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) | |
1633 | { | |
1634 | unsigned long flags; | |
1635 | ||
2e597558 PM |
1636 | WARN_ON_ONCE(rdp->beenonline == 0); |
1637 | ||
64db4cff PM |
1638 | /* |
1639 | * If an RCU GP has gone long enough, go check for dyntick | |
1640 | * idle CPUs and, if needed, send resched IPIs. | |
1641 | */ | |
20133cfc | 1642 | if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
64db4cff PM |
1643 | force_quiescent_state(rsp, 1); |
1644 | ||
1645 | /* | |
1646 | * Advance callbacks in response to end of earlier grace | |
1647 | * period that some other CPU ended. | |
1648 | */ | |
1649 | rcu_process_gp_end(rsp, rdp); | |
1650 | ||
1651 | /* Update RCU state based on any recent quiescent states. */ | |
1652 | rcu_check_quiescent_state(rsp, rdp); | |
1653 | ||
1654 | /* Does this CPU require a not-yet-started grace period? */ | |
1655 | if (cpu_needs_another_gp(rsp, rdp)) { | |
1304afb2 | 1656 | raw_spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags); |
64db4cff PM |
1657 | rcu_start_gp(rsp, flags); /* releases above lock */ |
1658 | } | |
1659 | ||
1660 | /* If there are callbacks ready, invoke them. */ | |
09223371 | 1661 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
a46e0899 | 1662 | invoke_rcu_callbacks(rsp, rdp); |
09223371 SL |
1663 | } |
1664 | ||
64db4cff | 1665 | /* |
e0f23060 | 1666 | * Do RCU core processing for the current CPU. |
64db4cff | 1667 | */ |
09223371 | 1668 | static void rcu_process_callbacks(struct softirq_action *unused) |
64db4cff | 1669 | { |
300df91c | 1670 | trace_rcu_utilization("Start RCU core"); |
d6714c22 PM |
1671 | __rcu_process_callbacks(&rcu_sched_state, |
1672 | &__get_cpu_var(rcu_sched_data)); | |
64db4cff | 1673 | __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); |
f41d911f | 1674 | rcu_preempt_process_callbacks(); |
300df91c | 1675 | trace_rcu_utilization("End RCU core"); |
64db4cff PM |
1676 | } |
1677 | ||
a26ac245 | 1678 | /* |
e0f23060 PM |
1679 | * Schedule RCU callback invocation. If the specified type of RCU |
1680 | * does not support RCU priority boosting, just do a direct call, | |
1681 | * otherwise wake up the per-CPU kernel kthread. Note that because we | |
1682 | * are running on the current CPU with interrupts disabled, the | |
1683 | * rcu_cpu_kthread_task cannot disappear out from under us. | |
a26ac245 | 1684 | */ |
a46e0899 | 1685 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) |
a26ac245 | 1686 | { |
b0d30417 PM |
1687 | if (unlikely(!ACCESS_ONCE(rcu_scheduler_fully_active))) |
1688 | return; | |
a46e0899 PM |
1689 | if (likely(!rsp->boost)) { |
1690 | rcu_do_batch(rsp, rdp); | |
a26ac245 PM |
1691 | return; |
1692 | } | |
a46e0899 | 1693 | invoke_rcu_callbacks_kthread(); |
a26ac245 PM |
1694 | } |
1695 | ||
a46e0899 | 1696 | static void invoke_rcu_core(void) |
09223371 SL |
1697 | { |
1698 | raise_softirq(RCU_SOFTIRQ); | |
1699 | } | |
1700 | ||
64db4cff PM |
1701 | static void |
1702 | __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu), | |
1703 | struct rcu_state *rsp) | |
1704 | { | |
1705 | unsigned long flags; | |
1706 | struct rcu_data *rdp; | |
1707 | ||
551d55a9 | 1708 | debug_rcu_head_queue(head); |
64db4cff PM |
1709 | head->func = func; |
1710 | head->next = NULL; | |
1711 | ||
1712 | smp_mb(); /* Ensure RCU update seen before callback registry. */ | |
1713 | ||
1714 | /* | |
1715 | * Opportunistically note grace-period endings and beginnings. | |
1716 | * Note that we might see a beginning right after we see an | |
1717 | * end, but never vice versa, since this CPU has to pass through | |
1718 | * a quiescent state betweentimes. | |
1719 | */ | |
1720 | local_irq_save(flags); | |
394f99a9 | 1721 | rdp = this_cpu_ptr(rsp->rda); |
64db4cff PM |
1722 | |
1723 | /* Add the callback to our list. */ | |
1724 | *rdp->nxttail[RCU_NEXT_TAIL] = head; | |
1725 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; | |
2655d57e PM |
1726 | rdp->qlen++; |
1727 | ||
d4c08f2a PM |
1728 | if (__is_kfree_rcu_offset((unsigned long)func)) |
1729 | trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func, | |
1730 | rdp->qlen); | |
1731 | else | |
1732 | trace_rcu_callback(rsp->name, head, rdp->qlen); | |
1733 | ||
2655d57e PM |
1734 | /* If interrupts were disabled, don't dive into RCU core. */ |
1735 | if (irqs_disabled_flags(flags)) { | |
1736 | local_irq_restore(flags); | |
1737 | return; | |
1738 | } | |
64db4cff | 1739 | |
37c72e56 PM |
1740 | /* |
1741 | * Force the grace period if too many callbacks or too long waiting. | |
1742 | * Enforce hysteresis, and don't invoke force_quiescent_state() | |
1743 | * if some other CPU has recently done so. Also, don't bother | |
1744 | * invoking force_quiescent_state() if the newly enqueued callback | |
1745 | * is the only one waiting for a grace period to complete. | |
1746 | */ | |
2655d57e | 1747 | if (unlikely(rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) { |
b52573d2 PM |
1748 | |
1749 | /* Are we ignoring a completed grace period? */ | |
1750 | rcu_process_gp_end(rsp, rdp); | |
1751 | check_for_new_grace_period(rsp, rdp); | |
1752 | ||
1753 | /* Start a new grace period if one not already started. */ | |
1754 | if (!rcu_gp_in_progress(rsp)) { | |
1755 | unsigned long nestflag; | |
1756 | struct rcu_node *rnp_root = rcu_get_root(rsp); | |
1757 | ||
1758 | raw_spin_lock_irqsave(&rnp_root->lock, nestflag); | |
1759 | rcu_start_gp(rsp, nestflag); /* rlses rnp_root->lock */ | |
1760 | } else { | |
1761 | /* Give the grace period a kick. */ | |
1762 | rdp->blimit = LONG_MAX; | |
1763 | if (rsp->n_force_qs == rdp->n_force_qs_snap && | |
1764 | *rdp->nxttail[RCU_DONE_TAIL] != head) | |
1765 | force_quiescent_state(rsp, 0); | |
1766 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1767 | rdp->qlen_last_fqs_check = rdp->qlen; | |
1768 | } | |
20133cfc | 1769 | } else if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
64db4cff PM |
1770 | force_quiescent_state(rsp, 1); |
1771 | local_irq_restore(flags); | |
1772 | } | |
1773 | ||
1774 | /* | |
d6714c22 | 1775 | * Queue an RCU-sched callback for invocation after a grace period. |
64db4cff | 1776 | */ |
d6714c22 | 1777 | void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) |
64db4cff | 1778 | { |
d6714c22 | 1779 | __call_rcu(head, func, &rcu_sched_state); |
64db4cff | 1780 | } |
d6714c22 | 1781 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
64db4cff PM |
1782 | |
1783 | /* | |
1784 | * Queue an RCU for invocation after a quicker grace period. | |
1785 | */ | |
1786 | void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
1787 | { | |
1788 | __call_rcu(head, func, &rcu_bh_state); | |
1789 | } | |
1790 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
1791 | ||
6ebb237b PM |
1792 | /** |
1793 | * synchronize_sched - wait until an rcu-sched grace period has elapsed. | |
1794 | * | |
1795 | * Control will return to the caller some time after a full rcu-sched | |
1796 | * grace period has elapsed, in other words after all currently executing | |
1797 | * rcu-sched read-side critical sections have completed. These read-side | |
1798 | * critical sections are delimited by rcu_read_lock_sched() and | |
1799 | * rcu_read_unlock_sched(), and may be nested. Note that preempt_disable(), | |
1800 | * local_irq_disable(), and so on may be used in place of | |
1801 | * rcu_read_lock_sched(). | |
1802 | * | |
1803 | * This means that all preempt_disable code sequences, including NMI and | |
1804 | * hardware-interrupt handlers, in progress on entry will have completed | |
1805 | * before this primitive returns. However, this does not guarantee that | |
1806 | * softirq handlers will have completed, since in some kernels, these | |
1807 | * handlers can run in process context, and can block. | |
1808 | * | |
1809 | * This primitive provides the guarantees made by the (now removed) | |
1810 | * synchronize_kernel() API. In contrast, synchronize_rcu() only | |
1811 | * guarantees that rcu_read_lock() sections will have completed. | |
1812 | * In "classic RCU", these two guarantees happen to be one and | |
1813 | * the same, but can differ in realtime RCU implementations. | |
1814 | */ | |
1815 | void synchronize_sched(void) | |
1816 | { | |
6ebb237b PM |
1817 | if (rcu_blocking_is_gp()) |
1818 | return; | |
2c42818e | 1819 | wait_rcu_gp(call_rcu_sched); |
6ebb237b PM |
1820 | } |
1821 | EXPORT_SYMBOL_GPL(synchronize_sched); | |
1822 | ||
1823 | /** | |
1824 | * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed. | |
1825 | * | |
1826 | * Control will return to the caller some time after a full rcu_bh grace | |
1827 | * period has elapsed, in other words after all currently executing rcu_bh | |
1828 | * read-side critical sections have completed. RCU read-side critical | |
1829 | * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(), | |
1830 | * and may be nested. | |
1831 | */ | |
1832 | void synchronize_rcu_bh(void) | |
1833 | { | |
6ebb237b PM |
1834 | if (rcu_blocking_is_gp()) |
1835 | return; | |
2c42818e | 1836 | wait_rcu_gp(call_rcu_bh); |
6ebb237b PM |
1837 | } |
1838 | EXPORT_SYMBOL_GPL(synchronize_rcu_bh); | |
1839 | ||
64db4cff PM |
1840 | /* |
1841 | * Check to see if there is any immediate RCU-related work to be done | |
1842 | * by the current CPU, for the specified type of RCU, returning 1 if so. | |
1843 | * The checks are in order of increasing expense: checks that can be | |
1844 | * carried out against CPU-local state are performed first. However, | |
1845 | * we must check for CPU stalls first, else we might not get a chance. | |
1846 | */ | |
1847 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | |
1848 | { | |
2f51f988 PM |
1849 | struct rcu_node *rnp = rdp->mynode; |
1850 | ||
64db4cff PM |
1851 | rdp->n_rcu_pending++; |
1852 | ||
1853 | /* Check for CPU stalls, if enabled. */ | |
1854 | check_cpu_stall(rsp, rdp); | |
1855 | ||
1856 | /* Is the RCU core waiting for a quiescent state from this CPU? */ | |
5c51dd73 PM |
1857 | if (rcu_scheduler_fully_active && |
1858 | rdp->qs_pending && !rdp->passed_quiesce) { | |
d25eb944 PM |
1859 | |
1860 | /* | |
1861 | * If force_quiescent_state() coming soon and this CPU | |
1862 | * needs a quiescent state, and this is either RCU-sched | |
1863 | * or RCU-bh, force a local reschedule. | |
1864 | */ | |
d21670ac | 1865 | rdp->n_rp_qs_pending++; |
6cc68793 | 1866 | if (!rdp->preemptible && |
d25eb944 PM |
1867 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs) - 1, |
1868 | jiffies)) | |
1869 | set_need_resched(); | |
e4cc1f22 | 1870 | } else if (rdp->qs_pending && rdp->passed_quiesce) { |
d21670ac | 1871 | rdp->n_rp_report_qs++; |
64db4cff | 1872 | return 1; |
7ba5c840 | 1873 | } |
64db4cff PM |
1874 | |
1875 | /* Does this CPU have callbacks ready to invoke? */ | |
7ba5c840 PM |
1876 | if (cpu_has_callbacks_ready_to_invoke(rdp)) { |
1877 | rdp->n_rp_cb_ready++; | |
64db4cff | 1878 | return 1; |
7ba5c840 | 1879 | } |
64db4cff PM |
1880 | |
1881 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
7ba5c840 PM |
1882 | if (cpu_needs_another_gp(rsp, rdp)) { |
1883 | rdp->n_rp_cpu_needs_gp++; | |
64db4cff | 1884 | return 1; |
7ba5c840 | 1885 | } |
64db4cff PM |
1886 | |
1887 | /* Has another RCU grace period completed? */ | |
2f51f988 | 1888 | if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */ |
7ba5c840 | 1889 | rdp->n_rp_gp_completed++; |
64db4cff | 1890 | return 1; |
7ba5c840 | 1891 | } |
64db4cff PM |
1892 | |
1893 | /* Has a new RCU grace period started? */ | |
2f51f988 | 1894 | if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */ |
7ba5c840 | 1895 | rdp->n_rp_gp_started++; |
64db4cff | 1896 | return 1; |
7ba5c840 | 1897 | } |
64db4cff PM |
1898 | |
1899 | /* Has an RCU GP gone long enough to send resched IPIs &c? */ | |
fc2219d4 | 1900 | if (rcu_gp_in_progress(rsp) && |
20133cfc | 1901 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) { |
7ba5c840 | 1902 | rdp->n_rp_need_fqs++; |
64db4cff | 1903 | return 1; |
7ba5c840 | 1904 | } |
64db4cff PM |
1905 | |
1906 | /* nothing to do */ | |
7ba5c840 | 1907 | rdp->n_rp_need_nothing++; |
64db4cff PM |
1908 | return 0; |
1909 | } | |
1910 | ||
1911 | /* | |
1912 | * Check to see if there is any immediate RCU-related work to be done | |
1913 | * by the current CPU, returning 1 if so. This function is part of the | |
1914 | * RCU implementation; it is -not- an exported member of the RCU API. | |
1915 | */ | |
a157229c | 1916 | static int rcu_pending(int cpu) |
64db4cff | 1917 | { |
d6714c22 | 1918 | return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) || |
f41d911f PM |
1919 | __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) || |
1920 | rcu_preempt_pending(cpu); | |
64db4cff PM |
1921 | } |
1922 | ||
1923 | /* | |
1924 | * Check to see if any future RCU-related work will need to be done | |
1925 | * by the current CPU, even if none need be done immediately, returning | |
8bd93a2c | 1926 | * 1 if so. |
64db4cff | 1927 | */ |
aea1b35e | 1928 | static int rcu_cpu_has_callbacks(int cpu) |
64db4cff PM |
1929 | { |
1930 | /* RCU callbacks either ready or pending? */ | |
d6714c22 | 1931 | return per_cpu(rcu_sched_data, cpu).nxtlist || |
f41d911f PM |
1932 | per_cpu(rcu_bh_data, cpu).nxtlist || |
1933 | rcu_preempt_needs_cpu(cpu); | |
64db4cff PM |
1934 | } |
1935 | ||
d0ec774c PM |
1936 | static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL}; |
1937 | static atomic_t rcu_barrier_cpu_count; | |
1938 | static DEFINE_MUTEX(rcu_barrier_mutex); | |
1939 | static struct completion rcu_barrier_completion; | |
d0ec774c PM |
1940 | |
1941 | static void rcu_barrier_callback(struct rcu_head *notused) | |
1942 | { | |
1943 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
1944 | complete(&rcu_barrier_completion); | |
1945 | } | |
1946 | ||
1947 | /* | |
1948 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
1949 | */ | |
1950 | static void rcu_barrier_func(void *type) | |
1951 | { | |
1952 | int cpu = smp_processor_id(); | |
1953 | struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu); | |
1954 | void (*call_rcu_func)(struct rcu_head *head, | |
1955 | void (*func)(struct rcu_head *head)); | |
1956 | ||
1957 | atomic_inc(&rcu_barrier_cpu_count); | |
1958 | call_rcu_func = type; | |
1959 | call_rcu_func(head, rcu_barrier_callback); | |
1960 | } | |
1961 | ||
d0ec774c PM |
1962 | /* |
1963 | * Orchestrate the specified type of RCU barrier, waiting for all | |
1964 | * RCU callbacks of the specified type to complete. | |
1965 | */ | |
e74f4c45 PM |
1966 | static void _rcu_barrier(struct rcu_state *rsp, |
1967 | void (*call_rcu_func)(struct rcu_head *head, | |
d0ec774c PM |
1968 | void (*func)(struct rcu_head *head))) |
1969 | { | |
1970 | BUG_ON(in_interrupt()); | |
e74f4c45 | 1971 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
d0ec774c PM |
1972 | mutex_lock(&rcu_barrier_mutex); |
1973 | init_completion(&rcu_barrier_completion); | |
1974 | /* | |
1975 | * Initialize rcu_barrier_cpu_count to 1, then invoke | |
1976 | * rcu_barrier_func() on each CPU, so that each CPU also has | |
1977 | * incremented rcu_barrier_cpu_count. Only then is it safe to | |
1978 | * decrement rcu_barrier_cpu_count -- otherwise the first CPU | |
1979 | * might complete its grace period before all of the other CPUs | |
1980 | * did their increment, causing this function to return too | |
2d999e03 PM |
1981 | * early. Note that on_each_cpu() disables irqs, which prevents |
1982 | * any CPUs from coming online or going offline until each online | |
1983 | * CPU has queued its RCU-barrier callback. | |
d0ec774c PM |
1984 | */ |
1985 | atomic_set(&rcu_barrier_cpu_count, 1); | |
1986 | on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1); | |
1987 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
1988 | complete(&rcu_barrier_completion); | |
1989 | wait_for_completion(&rcu_barrier_completion); | |
1990 | mutex_unlock(&rcu_barrier_mutex); | |
d0ec774c | 1991 | } |
d0ec774c PM |
1992 | |
1993 | /** | |
1994 | * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete. | |
1995 | */ | |
1996 | void rcu_barrier_bh(void) | |
1997 | { | |
e74f4c45 | 1998 | _rcu_barrier(&rcu_bh_state, call_rcu_bh); |
d0ec774c PM |
1999 | } |
2000 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); | |
2001 | ||
2002 | /** | |
2003 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. | |
2004 | */ | |
2005 | void rcu_barrier_sched(void) | |
2006 | { | |
e74f4c45 | 2007 | _rcu_barrier(&rcu_sched_state, call_rcu_sched); |
d0ec774c PM |
2008 | } |
2009 | EXPORT_SYMBOL_GPL(rcu_barrier_sched); | |
2010 | ||
64db4cff | 2011 | /* |
27569620 | 2012 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 2013 | */ |
27569620 PM |
2014 | static void __init |
2015 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) | |
64db4cff PM |
2016 | { |
2017 | unsigned long flags; | |
2018 | int i; | |
394f99a9 | 2019 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
27569620 PM |
2020 | struct rcu_node *rnp = rcu_get_root(rsp); |
2021 | ||
2022 | /* Set up local state, ensuring consistent view of global state. */ | |
1304afb2 | 2023 | raw_spin_lock_irqsave(&rnp->lock, flags); |
27569620 PM |
2024 | rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo); |
2025 | rdp->nxtlist = NULL; | |
2026 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
2027 | rdp->nxttail[i] = &rdp->nxtlist; | |
2028 | rdp->qlen = 0; | |
27569620 | 2029 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); |
4145fa7f | 2030 | WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != DYNTICK_TASK_NESTING); |
9b2e4f18 | 2031 | WARN_ON_ONCE(atomic_read(&rdp->dynticks->dynticks) != 1); |
27569620 | 2032 | rdp->cpu = cpu; |
d4c08f2a | 2033 | rdp->rsp = rsp; |
1304afb2 | 2034 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
27569620 PM |
2035 | } |
2036 | ||
2037 | /* | |
2038 | * Initialize a CPU's per-CPU RCU data. Note that only one online or | |
2039 | * offline event can be happening at a given time. Note also that we | |
2040 | * can accept some slop in the rsp->completed access due to the fact | |
2041 | * that this CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 2042 | */ |
e4fa4c97 | 2043 | static void __cpuinit |
6cc68793 | 2044 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptible) |
64db4cff PM |
2045 | { |
2046 | unsigned long flags; | |
64db4cff | 2047 | unsigned long mask; |
394f99a9 | 2048 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff PM |
2049 | struct rcu_node *rnp = rcu_get_root(rsp); |
2050 | ||
2051 | /* Set up local state, ensuring consistent view of global state. */ | |
1304afb2 | 2052 | raw_spin_lock_irqsave(&rnp->lock, flags); |
64db4cff | 2053 | rdp->beenonline = 1; /* We have now been online. */ |
6cc68793 | 2054 | rdp->preemptible = preemptible; |
37c72e56 PM |
2055 | rdp->qlen_last_fqs_check = 0; |
2056 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
64db4cff | 2057 | rdp->blimit = blimit; |
c92b131b PM |
2058 | rdp->dynticks->dynticks_nesting = DYNTICK_TASK_NESTING; |
2059 | atomic_set(&rdp->dynticks->dynticks, | |
2060 | (atomic_read(&rdp->dynticks->dynticks) & ~0x1) + 1); | |
7cb92499 | 2061 | rcu_prepare_for_idle_init(cpu); |
1304afb2 | 2062 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
2063 | |
2064 | /* | |
2065 | * A new grace period might start here. If so, we won't be part | |
2066 | * of it, but that is OK, as we are currently in a quiescent state. | |
2067 | */ | |
2068 | ||
2069 | /* Exclude any attempts to start a new GP on large systems. */ | |
1304afb2 | 2070 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
64db4cff PM |
2071 | |
2072 | /* Add CPU to rcu_node bitmasks. */ | |
2073 | rnp = rdp->mynode; | |
2074 | mask = rdp->grpmask; | |
2075 | do { | |
2076 | /* Exclude any attempts to start a new GP on small systems. */ | |
1304afb2 | 2077 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
2078 | rnp->qsmaskinit |= mask; |
2079 | mask = rnp->grpmask; | |
d09b62df | 2080 | if (rnp == rdp->mynode) { |
06ae115a PM |
2081 | /* |
2082 | * If there is a grace period in progress, we will | |
2083 | * set up to wait for it next time we run the | |
2084 | * RCU core code. | |
2085 | */ | |
2086 | rdp->gpnum = rnp->completed; | |
d09b62df | 2087 | rdp->completed = rnp->completed; |
06ae115a PM |
2088 | rdp->passed_quiesce = 0; |
2089 | rdp->qs_pending = 0; | |
e4cc1f22 | 2090 | rdp->passed_quiesce_gpnum = rnp->gpnum - 1; |
d4c08f2a | 2091 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuonl"); |
d09b62df | 2092 | } |
1304afb2 | 2093 | raw_spin_unlock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
2094 | rnp = rnp->parent; |
2095 | } while (rnp != NULL && !(rnp->qsmaskinit & mask)); | |
2096 | ||
1304afb2 | 2097 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); |
64db4cff PM |
2098 | } |
2099 | ||
d72bce0e | 2100 | static void __cpuinit rcu_prepare_cpu(int cpu) |
64db4cff | 2101 | { |
f41d911f PM |
2102 | rcu_init_percpu_data(cpu, &rcu_sched_state, 0); |
2103 | rcu_init_percpu_data(cpu, &rcu_bh_state, 0); | |
2104 | rcu_preempt_init_percpu_data(cpu); | |
64db4cff PM |
2105 | } |
2106 | ||
2107 | /* | |
f41d911f | 2108 | * Handle CPU online/offline notification events. |
64db4cff | 2109 | */ |
9f680ab4 PM |
2110 | static int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
2111 | unsigned long action, void *hcpu) | |
64db4cff PM |
2112 | { |
2113 | long cpu = (long)hcpu; | |
27f4d280 | 2114 | struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu); |
a26ac245 | 2115 | struct rcu_node *rnp = rdp->mynode; |
64db4cff | 2116 | |
300df91c | 2117 | trace_rcu_utilization("Start CPU hotplug"); |
64db4cff PM |
2118 | switch (action) { |
2119 | case CPU_UP_PREPARE: | |
2120 | case CPU_UP_PREPARE_FROZEN: | |
d72bce0e PZ |
2121 | rcu_prepare_cpu(cpu); |
2122 | rcu_prepare_kthreads(cpu); | |
a26ac245 PM |
2123 | break; |
2124 | case CPU_ONLINE: | |
0f962a5e PM |
2125 | case CPU_DOWN_FAILED: |
2126 | rcu_node_kthread_setaffinity(rnp, -1); | |
e3995a25 | 2127 | rcu_cpu_kthread_setrt(cpu, 1); |
0f962a5e PM |
2128 | break; |
2129 | case CPU_DOWN_PREPARE: | |
2130 | rcu_node_kthread_setaffinity(rnp, cpu); | |
e3995a25 | 2131 | rcu_cpu_kthread_setrt(cpu, 0); |
64db4cff | 2132 | break; |
d0ec774c PM |
2133 | case CPU_DYING: |
2134 | case CPU_DYING_FROZEN: | |
2135 | /* | |
2d999e03 PM |
2136 | * The whole machine is "stopped" except this CPU, so we can |
2137 | * touch any data without introducing corruption. We send the | |
2138 | * dying CPU's callbacks to an arbitrarily chosen online CPU. | |
d0ec774c | 2139 | */ |
29494be7 LJ |
2140 | rcu_send_cbs_to_online(&rcu_bh_state); |
2141 | rcu_send_cbs_to_online(&rcu_sched_state); | |
2142 | rcu_preempt_send_cbs_to_online(); | |
7cb92499 | 2143 | rcu_cleanup_after_idle(cpu); |
d0ec774c | 2144 | break; |
64db4cff PM |
2145 | case CPU_DEAD: |
2146 | case CPU_DEAD_FROZEN: | |
2147 | case CPU_UP_CANCELED: | |
2148 | case CPU_UP_CANCELED_FROZEN: | |
2149 | rcu_offline_cpu(cpu); | |
2150 | break; | |
2151 | default: | |
2152 | break; | |
2153 | } | |
300df91c | 2154 | trace_rcu_utilization("End CPU hotplug"); |
64db4cff PM |
2155 | return NOTIFY_OK; |
2156 | } | |
2157 | ||
bbad9379 PM |
2158 | /* |
2159 | * This function is invoked towards the end of the scheduler's initialization | |
2160 | * process. Before this is called, the idle task might contain | |
2161 | * RCU read-side critical sections (during which time, this idle | |
2162 | * task is booting the system). After this function is called, the | |
2163 | * idle tasks are prohibited from containing RCU read-side critical | |
2164 | * sections. This function also enables RCU lockdep checking. | |
2165 | */ | |
2166 | void rcu_scheduler_starting(void) | |
2167 | { | |
2168 | WARN_ON(num_online_cpus() != 1); | |
2169 | WARN_ON(nr_context_switches() > 0); | |
2170 | rcu_scheduler_active = 1; | |
2171 | } | |
2172 | ||
64db4cff PM |
2173 | /* |
2174 | * Compute the per-level fanout, either using the exact fanout specified | |
2175 | * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT. | |
2176 | */ | |
2177 | #ifdef CONFIG_RCU_FANOUT_EXACT | |
2178 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
2179 | { | |
2180 | int i; | |
2181 | ||
0209f649 | 2182 | for (i = NUM_RCU_LVLS - 1; i > 0; i--) |
64db4cff | 2183 | rsp->levelspread[i] = CONFIG_RCU_FANOUT; |
0209f649 | 2184 | rsp->levelspread[0] = RCU_FANOUT_LEAF; |
64db4cff PM |
2185 | } |
2186 | #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
2187 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
2188 | { | |
2189 | int ccur; | |
2190 | int cprv; | |
2191 | int i; | |
2192 | ||
2193 | cprv = NR_CPUS; | |
2194 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
2195 | ccur = rsp->levelcnt[i]; | |
2196 | rsp->levelspread[i] = (cprv + ccur - 1) / ccur; | |
2197 | cprv = ccur; | |
2198 | } | |
2199 | } | |
2200 | #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
2201 | ||
2202 | /* | |
2203 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
2204 | */ | |
394f99a9 LJ |
2205 | static void __init rcu_init_one(struct rcu_state *rsp, |
2206 | struct rcu_data __percpu *rda) | |
64db4cff | 2207 | { |
b6407e86 PM |
2208 | static char *buf[] = { "rcu_node_level_0", |
2209 | "rcu_node_level_1", | |
2210 | "rcu_node_level_2", | |
2211 | "rcu_node_level_3" }; /* Match MAX_RCU_LVLS */ | |
64db4cff PM |
2212 | int cpustride = 1; |
2213 | int i; | |
2214 | int j; | |
2215 | struct rcu_node *rnp; | |
2216 | ||
b6407e86 PM |
2217 | BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
2218 | ||
64db4cff PM |
2219 | /* Initialize the level-tracking arrays. */ |
2220 | ||
2221 | for (i = 1; i < NUM_RCU_LVLS; i++) | |
2222 | rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1]; | |
2223 | rcu_init_levelspread(rsp); | |
2224 | ||
2225 | /* Initialize the elements themselves, starting from the leaves. */ | |
2226 | ||
2227 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
2228 | cpustride *= rsp->levelspread[i]; | |
2229 | rnp = rsp->level[i]; | |
2230 | for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { | |
1304afb2 | 2231 | raw_spin_lock_init(&rnp->lock); |
b6407e86 PM |
2232 | lockdep_set_class_and_name(&rnp->lock, |
2233 | &rcu_node_class[i], buf[i]); | |
f41d911f | 2234 | rnp->gpnum = 0; |
64db4cff PM |
2235 | rnp->qsmask = 0; |
2236 | rnp->qsmaskinit = 0; | |
2237 | rnp->grplo = j * cpustride; | |
2238 | rnp->grphi = (j + 1) * cpustride - 1; | |
2239 | if (rnp->grphi >= NR_CPUS) | |
2240 | rnp->grphi = NR_CPUS - 1; | |
2241 | if (i == 0) { | |
2242 | rnp->grpnum = 0; | |
2243 | rnp->grpmask = 0; | |
2244 | rnp->parent = NULL; | |
2245 | } else { | |
2246 | rnp->grpnum = j % rsp->levelspread[i - 1]; | |
2247 | rnp->grpmask = 1UL << rnp->grpnum; | |
2248 | rnp->parent = rsp->level[i - 1] + | |
2249 | j / rsp->levelspread[i - 1]; | |
2250 | } | |
2251 | rnp->level = i; | |
12f5f524 | 2252 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
64db4cff PM |
2253 | } |
2254 | } | |
0c34029a | 2255 | |
394f99a9 | 2256 | rsp->rda = rda; |
0c34029a LJ |
2257 | rnp = rsp->level[NUM_RCU_LVLS - 1]; |
2258 | for_each_possible_cpu(i) { | |
4a90a068 | 2259 | while (i > rnp->grphi) |
0c34029a | 2260 | rnp++; |
394f99a9 | 2261 | per_cpu_ptr(rsp->rda, i)->mynode = rnp; |
0c34029a LJ |
2262 | rcu_boot_init_percpu_data(i, rsp); |
2263 | } | |
64db4cff PM |
2264 | } |
2265 | ||
9f680ab4 | 2266 | void __init rcu_init(void) |
64db4cff | 2267 | { |
017c4261 | 2268 | int cpu; |
9f680ab4 | 2269 | |
f41d911f | 2270 | rcu_bootup_announce(); |
394f99a9 LJ |
2271 | rcu_init_one(&rcu_sched_state, &rcu_sched_data); |
2272 | rcu_init_one(&rcu_bh_state, &rcu_bh_data); | |
f41d911f | 2273 | __rcu_init_preempt(); |
09223371 | 2274 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
9f680ab4 PM |
2275 | |
2276 | /* | |
2277 | * We don't need protection against CPU-hotplug here because | |
2278 | * this is called early in boot, before either interrupts | |
2279 | * or the scheduler are operational. | |
2280 | */ | |
2281 | cpu_notifier(rcu_cpu_notify, 0); | |
017c4261 PM |
2282 | for_each_online_cpu(cpu) |
2283 | rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu); | |
c68de209 | 2284 | check_cpu_stall_init(); |
64db4cff PM |
2285 | } |
2286 | ||
1eba8f84 | 2287 | #include "rcutree_plugin.h" |