]>
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> |
64db4cff PM |
39 | #include <asm/atomic.h> |
40 | #include <linux/bitops.h> | |
41 | #include <linux/module.h> | |
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 PM |
54 | #include "rcutree.h" |
55 | ||
64db4cff PM |
56 | /* Data structures. */ |
57 | ||
b668c9cf | 58 | static struct lock_class_key rcu_node_class[NUM_RCU_LVLS]; |
88b91c7c | 59 | |
4300aa64 PM |
60 | #define RCU_STATE_INITIALIZER(structname) { \ |
61 | .level = { &structname.node[0] }, \ | |
64db4cff PM |
62 | .levelcnt = { \ |
63 | NUM_RCU_LVL_0, /* root of hierarchy. */ \ | |
64 | NUM_RCU_LVL_1, \ | |
65 | NUM_RCU_LVL_2, \ | |
cf244dc0 PM |
66 | NUM_RCU_LVL_3, \ |
67 | NUM_RCU_LVL_4, /* == MAX_RCU_LVLS */ \ | |
64db4cff | 68 | }, \ |
83f5b01f | 69 | .signaled = RCU_GP_IDLE, \ |
64db4cff PM |
70 | .gpnum = -300, \ |
71 | .completed = -300, \ | |
4300aa64 | 72 | .onofflock = __RAW_SPIN_LOCK_UNLOCKED(&structname.onofflock), \ |
4300aa64 | 73 | .fqslock = __RAW_SPIN_LOCK_UNLOCKED(&structname.fqslock), \ |
64db4cff PM |
74 | .n_force_qs = 0, \ |
75 | .n_force_qs_ngp = 0, \ | |
4300aa64 | 76 | .name = #structname, \ |
64db4cff PM |
77 | } |
78 | ||
d6714c22 PM |
79 | struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched_state); |
80 | DEFINE_PER_CPU(struct rcu_data, rcu_sched_data); | |
64db4cff | 81 | |
6258c4fb IM |
82 | struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state); |
83 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); | |
b1f77b05 | 84 | |
27f4d280 PM |
85 | static struct rcu_state *rcu_state; |
86 | ||
bbad9379 PM |
87 | int rcu_scheduler_active __read_mostly; |
88 | EXPORT_SYMBOL_GPL(rcu_scheduler_active); | |
89 | ||
a26ac245 PM |
90 | /* |
91 | * Control variables for per-CPU and per-rcu_node kthreads. These | |
92 | * handle all flavors of RCU. | |
93 | */ | |
94 | static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task); | |
d71df90e | 95 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status); |
15ba0ba8 | 96 | DEFINE_PER_CPU(int, rcu_cpu_kthread_cpu); |
5ece5bab | 97 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops); |
a26ac245 | 98 | static DEFINE_PER_CPU(wait_queue_head_t, rcu_cpu_wq); |
d71df90e | 99 | DEFINE_PER_CPU(char, rcu_cpu_has_work); |
a26ac245 PM |
100 | static char rcu_kthreads_spawnable; |
101 | ||
0f962a5e | 102 | static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); |
27f4d280 | 103 | static void invoke_rcu_cpu_kthread(void); |
a26ac245 PM |
104 | |
105 | #define RCU_KTHREAD_PRIO 1 /* RT priority for per-CPU kthreads. */ | |
106 | ||
4a298656 PM |
107 | /* |
108 | * Track the rcutorture test sequence number and the update version | |
109 | * number within a given test. The rcutorture_testseq is incremented | |
110 | * on every rcutorture module load and unload, so has an odd value | |
111 | * when a test is running. The rcutorture_vernum is set to zero | |
112 | * when rcutorture starts and is incremented on each rcutorture update. | |
113 | * These variables enable correlating rcutorture output with the | |
114 | * RCU tracing information. | |
115 | */ | |
116 | unsigned long rcutorture_testseq; | |
117 | unsigned long rcutorture_vernum; | |
118 | ||
fc2219d4 PM |
119 | /* |
120 | * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s | |
121 | * permit this function to be invoked without holding the root rcu_node | |
122 | * structure's ->lock, but of course results can be subject to change. | |
123 | */ | |
124 | static int rcu_gp_in_progress(struct rcu_state *rsp) | |
125 | { | |
126 | return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum); | |
127 | } | |
128 | ||
b1f77b05 | 129 | /* |
d6714c22 | 130 | * Note a quiescent state. Because we do not need to know |
b1f77b05 | 131 | * how many quiescent states passed, just if there was at least |
d6714c22 | 132 | * one since the start of the grace period, this just sets a flag. |
b1f77b05 | 133 | */ |
d6714c22 | 134 | void rcu_sched_qs(int cpu) |
b1f77b05 | 135 | { |
25502a6c | 136 | struct rcu_data *rdp = &per_cpu(rcu_sched_data, cpu); |
f41d911f | 137 | |
c64ac3ce | 138 | rdp->passed_quiesc_completed = rdp->gpnum - 1; |
c3422bea PM |
139 | barrier(); |
140 | rdp->passed_quiesc = 1; | |
b1f77b05 IM |
141 | } |
142 | ||
d6714c22 | 143 | void rcu_bh_qs(int cpu) |
b1f77b05 | 144 | { |
25502a6c | 145 | struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu); |
f41d911f | 146 | |
c64ac3ce | 147 | rdp->passed_quiesc_completed = rdp->gpnum - 1; |
c3422bea PM |
148 | barrier(); |
149 | rdp->passed_quiesc = 1; | |
b1f77b05 | 150 | } |
64db4cff | 151 | |
25502a6c PM |
152 | /* |
153 | * Note a context switch. This is a quiescent state for RCU-sched, | |
154 | * and requires special handling for preemptible RCU. | |
155 | */ | |
156 | void rcu_note_context_switch(int cpu) | |
157 | { | |
158 | rcu_sched_qs(cpu); | |
159 | rcu_preempt_note_context_switch(cpu); | |
160 | } | |
29ce8310 | 161 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
25502a6c | 162 | |
64db4cff | 163 | #ifdef CONFIG_NO_HZ |
90a4d2c0 PM |
164 | DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { |
165 | .dynticks_nesting = 1, | |
23b5c8fa | 166 | .dynticks = ATOMIC_INIT(1), |
90a4d2c0 | 167 | }; |
64db4cff PM |
168 | #endif /* #ifdef CONFIG_NO_HZ */ |
169 | ||
170 | static int blimit = 10; /* Maximum callbacks per softirq. */ | |
171 | static int qhimark = 10000; /* If this many pending, ignore blimit. */ | |
172 | static int qlowmark = 100; /* Once only this many pending, use blimit. */ | |
173 | ||
3d76c082 PM |
174 | module_param(blimit, int, 0); |
175 | module_param(qhimark, int, 0); | |
176 | module_param(qlowmark, int, 0); | |
177 | ||
a00e0d71 | 178 | int rcu_cpu_stall_suppress __read_mostly; |
f2e0dd70 | 179 | module_param(rcu_cpu_stall_suppress, int, 0644); |
742734ee | 180 | |
64db4cff | 181 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed); |
a157229c | 182 | static int rcu_pending(int cpu); |
64db4cff PM |
183 | |
184 | /* | |
d6714c22 | 185 | * Return the number of RCU-sched batches processed thus far for debug & stats. |
64db4cff | 186 | */ |
d6714c22 | 187 | long rcu_batches_completed_sched(void) |
64db4cff | 188 | { |
d6714c22 | 189 | return rcu_sched_state.completed; |
64db4cff | 190 | } |
d6714c22 | 191 | EXPORT_SYMBOL_GPL(rcu_batches_completed_sched); |
64db4cff PM |
192 | |
193 | /* | |
194 | * Return the number of RCU BH batches processed thus far for debug & stats. | |
195 | */ | |
196 | long rcu_batches_completed_bh(void) | |
197 | { | |
198 | return rcu_bh_state.completed; | |
199 | } | |
200 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); | |
201 | ||
bf66f18e PM |
202 | /* |
203 | * Force a quiescent state for RCU BH. | |
204 | */ | |
205 | void rcu_bh_force_quiescent_state(void) | |
206 | { | |
207 | force_quiescent_state(&rcu_bh_state, 0); | |
208 | } | |
209 | EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state); | |
210 | ||
4a298656 PM |
211 | /* |
212 | * Record the number of times rcutorture tests have been initiated and | |
213 | * terminated. This information allows the debugfs tracing stats to be | |
214 | * correlated to the rcutorture messages, even when the rcutorture module | |
215 | * is being repeatedly loaded and unloaded. In other words, we cannot | |
216 | * store this state in rcutorture itself. | |
217 | */ | |
218 | void rcutorture_record_test_transition(void) | |
219 | { | |
220 | rcutorture_testseq++; | |
221 | rcutorture_vernum = 0; | |
222 | } | |
223 | EXPORT_SYMBOL_GPL(rcutorture_record_test_transition); | |
224 | ||
225 | /* | |
226 | * Record the number of writer passes through the current rcutorture test. | |
227 | * This is also used to correlate debugfs tracing stats with the rcutorture | |
228 | * messages. | |
229 | */ | |
230 | void rcutorture_record_progress(unsigned long vernum) | |
231 | { | |
232 | rcutorture_vernum++; | |
233 | } | |
234 | EXPORT_SYMBOL_GPL(rcutorture_record_progress); | |
235 | ||
bf66f18e PM |
236 | /* |
237 | * Force a quiescent state for RCU-sched. | |
238 | */ | |
239 | void rcu_sched_force_quiescent_state(void) | |
240 | { | |
241 | force_quiescent_state(&rcu_sched_state, 0); | |
242 | } | |
243 | EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state); | |
244 | ||
64db4cff PM |
245 | /* |
246 | * Does the CPU have callbacks ready to be invoked? | |
247 | */ | |
248 | static int | |
249 | cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) | |
250 | { | |
251 | return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]; | |
252 | } | |
253 | ||
254 | /* | |
255 | * Does the current CPU require a yet-as-unscheduled grace period? | |
256 | */ | |
257 | static int | |
258 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) | |
259 | { | |
fc2219d4 | 260 | return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp); |
64db4cff PM |
261 | } |
262 | ||
263 | /* | |
264 | * Return the root node of the specified rcu_state structure. | |
265 | */ | |
266 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) | |
267 | { | |
268 | return &rsp->node[0]; | |
269 | } | |
270 | ||
271 | #ifdef CONFIG_SMP | |
272 | ||
273 | /* | |
274 | * If the specified CPU is offline, tell the caller that it is in | |
275 | * a quiescent state. Otherwise, whack it with a reschedule IPI. | |
276 | * Grace periods can end up waiting on an offline CPU when that | |
277 | * CPU is in the process of coming online -- it will be added to the | |
278 | * rcu_node bitmasks before it actually makes it online. The same thing | |
279 | * can happen while a CPU is in the process of coming online. Because this | |
280 | * race is quite rare, we check for it after detecting that the grace | |
281 | * period has been delayed rather than checking each and every CPU | |
282 | * each and every time we start a new grace period. | |
283 | */ | |
284 | static int rcu_implicit_offline_qs(struct rcu_data *rdp) | |
285 | { | |
286 | /* | |
287 | * If the CPU is offline, it is in a quiescent state. We can | |
288 | * trust its state not to change because interrupts are disabled. | |
289 | */ | |
290 | if (cpu_is_offline(rdp->cpu)) { | |
291 | rdp->offline_fqs++; | |
292 | return 1; | |
293 | } | |
294 | ||
6cc68793 PM |
295 | /* If preemptible RCU, no point in sending reschedule IPI. */ |
296 | if (rdp->preemptible) | |
f41d911f PM |
297 | return 0; |
298 | ||
64db4cff PM |
299 | /* The CPU is online, so send it a reschedule IPI. */ |
300 | if (rdp->cpu != smp_processor_id()) | |
301 | smp_send_reschedule(rdp->cpu); | |
302 | else | |
303 | set_need_resched(); | |
304 | rdp->resched_ipi++; | |
305 | return 0; | |
306 | } | |
307 | ||
308 | #endif /* #ifdef CONFIG_SMP */ | |
309 | ||
310 | #ifdef CONFIG_NO_HZ | |
64db4cff PM |
311 | |
312 | /** | |
313 | * rcu_enter_nohz - inform RCU that current CPU is entering nohz | |
314 | * | |
315 | * Enter nohz mode, in other words, -leave- the mode in which RCU | |
316 | * read-side critical sections can occur. (Though RCU read-side | |
317 | * critical sections can occur in irq handlers in nohz mode, a possibility | |
318 | * handled by rcu_irq_enter() and rcu_irq_exit()). | |
319 | */ | |
320 | void rcu_enter_nohz(void) | |
321 | { | |
322 | unsigned long flags; | |
323 | struct rcu_dynticks *rdtp; | |
324 | ||
64db4cff PM |
325 | local_irq_save(flags); |
326 | rdtp = &__get_cpu_var(rcu_dynticks); | |
23b5c8fa PM |
327 | if (--rdtp->dynticks_nesting) { |
328 | local_irq_restore(flags); | |
329 | return; | |
330 | } | |
331 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ | |
332 | smp_mb__before_atomic_inc(); /* See above. */ | |
333 | atomic_inc(&rdtp->dynticks); | |
334 | smp_mb__after_atomic_inc(); /* Force ordering with next sojourn. */ | |
335 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); | |
64db4cff | 336 | local_irq_restore(flags); |
23b5c8fa PM |
337 | |
338 | /* If the interrupt queued a callback, get out of dyntick mode. */ | |
339 | if (in_irq() && | |
340 | (__get_cpu_var(rcu_sched_data).nxtlist || | |
341 | __get_cpu_var(rcu_bh_data).nxtlist || | |
342 | rcu_preempt_needs_cpu(smp_processor_id()))) | |
343 | set_need_resched(); | |
64db4cff PM |
344 | } |
345 | ||
346 | /* | |
347 | * rcu_exit_nohz - inform RCU that current CPU is leaving nohz | |
348 | * | |
349 | * Exit nohz mode, in other words, -enter- the mode in which RCU | |
350 | * read-side critical sections normally occur. | |
351 | */ | |
352 | void rcu_exit_nohz(void) | |
353 | { | |
354 | unsigned long flags; | |
355 | struct rcu_dynticks *rdtp; | |
356 | ||
357 | local_irq_save(flags); | |
358 | rdtp = &__get_cpu_var(rcu_dynticks); | |
23b5c8fa PM |
359 | if (rdtp->dynticks_nesting++) { |
360 | local_irq_restore(flags); | |
361 | return; | |
362 | } | |
363 | smp_mb__before_atomic_inc(); /* Force ordering w/previous sojourn. */ | |
364 | atomic_inc(&rdtp->dynticks); | |
365 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ | |
366 | smp_mb__after_atomic_inc(); /* See above. */ | |
367 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); | |
64db4cff | 368 | local_irq_restore(flags); |
64db4cff PM |
369 | } |
370 | ||
371 | /** | |
372 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
373 | * | |
374 | * If the CPU was idle with dynamic ticks active, and there is no | |
375 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
376 | * RCU grace-period handling know that the CPU is active. | |
377 | */ | |
378 | void rcu_nmi_enter(void) | |
379 | { | |
380 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
381 | ||
23b5c8fa PM |
382 | if (rdtp->dynticks_nmi_nesting == 0 && |
383 | (atomic_read(&rdtp->dynticks) & 0x1)) | |
64db4cff | 384 | return; |
23b5c8fa PM |
385 | rdtp->dynticks_nmi_nesting++; |
386 | smp_mb__before_atomic_inc(); /* Force delay from prior write. */ | |
387 | atomic_inc(&rdtp->dynticks); | |
388 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ | |
389 | smp_mb__after_atomic_inc(); /* See above. */ | |
390 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); | |
64db4cff PM |
391 | } |
392 | ||
393 | /** | |
394 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
395 | * | |
396 | * If the CPU was idle with dynamic ticks active, and there is no | |
397 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
398 | * RCU grace-period handling know that the CPU is no longer active. | |
399 | */ | |
400 | void rcu_nmi_exit(void) | |
401 | { | |
402 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
403 | ||
23b5c8fa PM |
404 | if (rdtp->dynticks_nmi_nesting == 0 || |
405 | --rdtp->dynticks_nmi_nesting != 0) | |
64db4cff | 406 | return; |
23b5c8fa PM |
407 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ |
408 | smp_mb__before_atomic_inc(); /* See above. */ | |
409 | atomic_inc(&rdtp->dynticks); | |
410 | smp_mb__after_atomic_inc(); /* Force delay to next write. */ | |
411 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); | |
64db4cff PM |
412 | } |
413 | ||
414 | /** | |
415 | * rcu_irq_enter - inform RCU of entry to hard irq context | |
416 | * | |
417 | * If the CPU was idle with dynamic ticks active, this updates the | |
418 | * rdtp->dynticks to let the RCU handling know that the CPU is active. | |
419 | */ | |
420 | void rcu_irq_enter(void) | |
421 | { | |
23b5c8fa | 422 | rcu_exit_nohz(); |
64db4cff PM |
423 | } |
424 | ||
425 | /** | |
426 | * rcu_irq_exit - inform RCU of exit from hard irq context | |
427 | * | |
428 | * If the CPU was idle with dynamic ticks active, update the rdp->dynticks | |
429 | * to put let the RCU handling be aware that the CPU is going back to idle | |
430 | * with no ticks. | |
431 | */ | |
432 | void rcu_irq_exit(void) | |
433 | { | |
23b5c8fa | 434 | rcu_enter_nohz(); |
64db4cff PM |
435 | } |
436 | ||
64db4cff PM |
437 | #ifdef CONFIG_SMP |
438 | ||
64db4cff PM |
439 | /* |
440 | * Snapshot the specified CPU's dynticks counter so that we can later | |
441 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 442 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff PM |
443 | */ |
444 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | |
445 | { | |
23b5c8fa PM |
446 | rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks); |
447 | return 0; | |
64db4cff PM |
448 | } |
449 | ||
450 | /* | |
451 | * Return true if the specified CPU has passed through a quiescent | |
452 | * state by virtue of being in or having passed through an dynticks | |
453 | * idle state since the last call to dyntick_save_progress_counter() | |
454 | * for this same CPU. | |
455 | */ | |
456 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
457 | { | |
23b5c8fa PM |
458 | unsigned long curr; |
459 | unsigned long snap; | |
64db4cff | 460 | |
23b5c8fa PM |
461 | curr = (unsigned long)atomic_add_return(0, &rdp->dynticks->dynticks); |
462 | snap = (unsigned long)rdp->dynticks_snap; | |
64db4cff PM |
463 | |
464 | /* | |
465 | * If the CPU passed through or entered a dynticks idle phase with | |
466 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
467 | * already acknowledged the request to pass through a quiescent | |
468 | * state. Either way, that CPU cannot possibly be in an RCU | |
469 | * read-side critical section that started before the beginning | |
470 | * of the current RCU grace period. | |
471 | */ | |
23b5c8fa | 472 | if ((curr & 0x1) == 0 || ULONG_CMP_GE(curr, snap + 2)) { |
64db4cff PM |
473 | rdp->dynticks_fqs++; |
474 | return 1; | |
475 | } | |
476 | ||
477 | /* Go check for the CPU being offline. */ | |
478 | return rcu_implicit_offline_qs(rdp); | |
479 | } | |
480 | ||
481 | #endif /* #ifdef CONFIG_SMP */ | |
482 | ||
483 | #else /* #ifdef CONFIG_NO_HZ */ | |
484 | ||
64db4cff PM |
485 | #ifdef CONFIG_SMP |
486 | ||
64db4cff PM |
487 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
488 | { | |
489 | return 0; | |
490 | } | |
491 | ||
492 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
493 | { | |
494 | return rcu_implicit_offline_qs(rdp); | |
495 | } | |
496 | ||
497 | #endif /* #ifdef CONFIG_SMP */ | |
498 | ||
499 | #endif /* #else #ifdef CONFIG_NO_HZ */ | |
500 | ||
742734ee | 501 | int rcu_cpu_stall_suppress __read_mostly; |
c68de209 | 502 | |
64db4cff PM |
503 | static void record_gp_stall_check_time(struct rcu_state *rsp) |
504 | { | |
505 | rsp->gp_start = jiffies; | |
506 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK; | |
507 | } | |
508 | ||
509 | static void print_other_cpu_stall(struct rcu_state *rsp) | |
510 | { | |
511 | int cpu; | |
512 | long delta; | |
513 | unsigned long flags; | |
514 | struct rcu_node *rnp = rcu_get_root(rsp); | |
64db4cff PM |
515 | |
516 | /* Only let one CPU complain about others per time interval. */ | |
517 | ||
1304afb2 | 518 | raw_spin_lock_irqsave(&rnp->lock, flags); |
64db4cff | 519 | delta = jiffies - rsp->jiffies_stall; |
fc2219d4 | 520 | if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) { |
1304afb2 | 521 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
522 | return; |
523 | } | |
524 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
a0b6c9a7 PM |
525 | |
526 | /* | |
527 | * Now rat on any tasks that got kicked up to the root rcu_node | |
528 | * due to CPU offlining. | |
529 | */ | |
530 | rcu_print_task_stall(rnp); | |
1304afb2 | 531 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 532 | |
8cdd32a9 PM |
533 | /* |
534 | * OK, time to rat on our buddy... | |
535 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
536 | * RCU CPU stall warnings. | |
537 | */ | |
4300aa64 PM |
538 | printk(KERN_ERR "INFO: %s detected stalls on CPUs/tasks: {", |
539 | rsp->name); | |
a0b6c9a7 | 540 | rcu_for_each_leaf_node(rsp, rnp) { |
3acd9eb3 | 541 | raw_spin_lock_irqsave(&rnp->lock, flags); |
f41d911f | 542 | rcu_print_task_stall(rnp); |
3acd9eb3 | 543 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
a0b6c9a7 | 544 | if (rnp->qsmask == 0) |
64db4cff | 545 | continue; |
a0b6c9a7 PM |
546 | for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++) |
547 | if (rnp->qsmask & (1UL << cpu)) | |
548 | printk(" %d", rnp->grplo + cpu); | |
64db4cff | 549 | } |
4300aa64 | 550 | printk("} (detected by %d, t=%ld jiffies)\n", |
64db4cff | 551 | smp_processor_id(), (long)(jiffies - rsp->gp_start)); |
c1dc0b9c IM |
552 | trigger_all_cpu_backtrace(); |
553 | ||
1ed509a2 PM |
554 | /* If so configured, complain about tasks blocking the grace period. */ |
555 | ||
556 | rcu_print_detail_task_stall(rsp); | |
557 | ||
64db4cff PM |
558 | force_quiescent_state(rsp, 0); /* Kick them all. */ |
559 | } | |
560 | ||
561 | static void print_cpu_stall(struct rcu_state *rsp) | |
562 | { | |
563 | unsigned long flags; | |
564 | struct rcu_node *rnp = rcu_get_root(rsp); | |
565 | ||
8cdd32a9 PM |
566 | /* |
567 | * OK, time to rat on ourselves... | |
568 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
569 | * RCU CPU stall warnings. | |
570 | */ | |
4300aa64 PM |
571 | printk(KERN_ERR "INFO: %s detected stall on CPU %d (t=%lu jiffies)\n", |
572 | rsp->name, smp_processor_id(), jiffies - rsp->gp_start); | |
c1dc0b9c IM |
573 | trigger_all_cpu_backtrace(); |
574 | ||
1304afb2 | 575 | raw_spin_lock_irqsave(&rnp->lock, flags); |
20133cfc | 576 | if (ULONG_CMP_GE(jiffies, rsp->jiffies_stall)) |
64db4cff PM |
577 | rsp->jiffies_stall = |
578 | jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
1304afb2 | 579 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
c1dc0b9c | 580 | |
64db4cff PM |
581 | set_need_resched(); /* kick ourselves to get things going. */ |
582 | } | |
583 | ||
584 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
585 | { | |
bad6e139 PM |
586 | unsigned long j; |
587 | unsigned long js; | |
64db4cff PM |
588 | struct rcu_node *rnp; |
589 | ||
742734ee | 590 | if (rcu_cpu_stall_suppress) |
c68de209 | 591 | return; |
bad6e139 PM |
592 | j = ACCESS_ONCE(jiffies); |
593 | js = ACCESS_ONCE(rsp->jiffies_stall); | |
64db4cff | 594 | rnp = rdp->mynode; |
bad6e139 | 595 | if ((ACCESS_ONCE(rnp->qsmask) & rdp->grpmask) && ULONG_CMP_GE(j, js)) { |
64db4cff PM |
596 | |
597 | /* We haven't checked in, so go dump stack. */ | |
598 | print_cpu_stall(rsp); | |
599 | ||
bad6e139 PM |
600 | } else if (rcu_gp_in_progress(rsp) && |
601 | ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) { | |
64db4cff | 602 | |
bad6e139 | 603 | /* They had a few time units to dump stack, so complain. */ |
64db4cff PM |
604 | print_other_cpu_stall(rsp); |
605 | } | |
606 | } | |
607 | ||
c68de209 PM |
608 | static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr) |
609 | { | |
742734ee | 610 | rcu_cpu_stall_suppress = 1; |
c68de209 PM |
611 | return NOTIFY_DONE; |
612 | } | |
613 | ||
53d84e00 PM |
614 | /** |
615 | * rcu_cpu_stall_reset - prevent further stall warnings in current grace period | |
616 | * | |
617 | * Set the stall-warning timeout way off into the future, thus preventing | |
618 | * any RCU CPU stall-warning messages from appearing in the current set of | |
619 | * RCU grace periods. | |
620 | * | |
621 | * The caller must disable hard irqs. | |
622 | */ | |
623 | void rcu_cpu_stall_reset(void) | |
624 | { | |
625 | rcu_sched_state.jiffies_stall = jiffies + ULONG_MAX / 2; | |
626 | rcu_bh_state.jiffies_stall = jiffies + ULONG_MAX / 2; | |
627 | rcu_preempt_stall_reset(); | |
628 | } | |
629 | ||
c68de209 PM |
630 | static struct notifier_block rcu_panic_block = { |
631 | .notifier_call = rcu_panic, | |
632 | }; | |
633 | ||
634 | static void __init check_cpu_stall_init(void) | |
635 | { | |
636 | atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block); | |
637 | } | |
638 | ||
64db4cff PM |
639 | /* |
640 | * Update CPU-local rcu_data state to record the newly noticed grace period. | |
641 | * This is used both when we started the grace period and when we notice | |
9160306e PM |
642 | * that someone else started the grace period. The caller must hold the |
643 | * ->lock of the leaf rcu_node structure corresponding to the current CPU, | |
644 | * and must have irqs disabled. | |
64db4cff | 645 | */ |
9160306e PM |
646 | static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) |
647 | { | |
648 | if (rdp->gpnum != rnp->gpnum) { | |
121dfc4b PM |
649 | /* |
650 | * If the current grace period is waiting for this CPU, | |
651 | * set up to detect a quiescent state, otherwise don't | |
652 | * go looking for one. | |
653 | */ | |
9160306e | 654 | rdp->gpnum = rnp->gpnum; |
121dfc4b PM |
655 | if (rnp->qsmask & rdp->grpmask) { |
656 | rdp->qs_pending = 1; | |
657 | rdp->passed_quiesc = 0; | |
658 | } else | |
659 | rdp->qs_pending = 0; | |
9160306e PM |
660 | } |
661 | } | |
662 | ||
64db4cff PM |
663 | static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp) |
664 | { | |
9160306e PM |
665 | unsigned long flags; |
666 | struct rcu_node *rnp; | |
667 | ||
668 | local_irq_save(flags); | |
669 | rnp = rdp->mynode; | |
670 | if (rdp->gpnum == ACCESS_ONCE(rnp->gpnum) || /* outside lock. */ | |
1304afb2 | 671 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
9160306e PM |
672 | local_irq_restore(flags); |
673 | return; | |
674 | } | |
675 | __note_new_gpnum(rsp, rnp, rdp); | |
1304afb2 | 676 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
677 | } |
678 | ||
679 | /* | |
680 | * Did someone else start a new RCU grace period start since we last | |
681 | * checked? Update local state appropriately if so. Must be called | |
682 | * on the CPU corresponding to rdp. | |
683 | */ | |
684 | static int | |
685 | check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp) | |
686 | { | |
687 | unsigned long flags; | |
688 | int ret = 0; | |
689 | ||
690 | local_irq_save(flags); | |
691 | if (rdp->gpnum != rsp->gpnum) { | |
692 | note_new_gpnum(rsp, rdp); | |
693 | ret = 1; | |
694 | } | |
695 | local_irq_restore(flags); | |
696 | return ret; | |
697 | } | |
698 | ||
d09b62df PM |
699 | /* |
700 | * Advance this CPU's callbacks, but only if the current grace period | |
701 | * has ended. This may be called only from the CPU to whom the rdp | |
702 | * belongs. In addition, the corresponding leaf rcu_node structure's | |
703 | * ->lock must be held by the caller, with irqs disabled. | |
704 | */ | |
705 | static void | |
706 | __rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
707 | { | |
708 | /* Did another grace period end? */ | |
709 | if (rdp->completed != rnp->completed) { | |
710 | ||
711 | /* Advance callbacks. No harm if list empty. */ | |
712 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL]; | |
713 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL]; | |
714 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
715 | ||
716 | /* Remember that we saw this grace-period completion. */ | |
717 | rdp->completed = rnp->completed; | |
20377f32 | 718 | |
5ff8e6f0 FW |
719 | /* |
720 | * If we were in an extended quiescent state, we may have | |
121dfc4b | 721 | * missed some grace periods that others CPUs handled on |
5ff8e6f0 | 722 | * our behalf. Catch up with this state to avoid noting |
121dfc4b PM |
723 | * spurious new grace periods. If another grace period |
724 | * has started, then rnp->gpnum will have advanced, so | |
725 | * we will detect this later on. | |
5ff8e6f0 | 726 | */ |
121dfc4b | 727 | if (ULONG_CMP_LT(rdp->gpnum, rdp->completed)) |
5ff8e6f0 FW |
728 | rdp->gpnum = rdp->completed; |
729 | ||
20377f32 | 730 | /* |
121dfc4b PM |
731 | * If RCU does not need a quiescent state from this CPU, |
732 | * then make sure that this CPU doesn't go looking for one. | |
20377f32 | 733 | */ |
121dfc4b | 734 | if ((rnp->qsmask & rdp->grpmask) == 0) |
20377f32 | 735 | rdp->qs_pending = 0; |
d09b62df PM |
736 | } |
737 | } | |
738 | ||
739 | /* | |
740 | * Advance this CPU's callbacks, but only if the current grace period | |
741 | * has ended. This may be called only from the CPU to whom the rdp | |
742 | * belongs. | |
743 | */ | |
744 | static void | |
745 | rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp) | |
746 | { | |
747 | unsigned long flags; | |
748 | struct rcu_node *rnp; | |
749 | ||
750 | local_irq_save(flags); | |
751 | rnp = rdp->mynode; | |
752 | if (rdp->completed == ACCESS_ONCE(rnp->completed) || /* outside lock. */ | |
1304afb2 | 753 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
d09b62df PM |
754 | local_irq_restore(flags); |
755 | return; | |
756 | } | |
757 | __rcu_process_gp_end(rsp, rnp, rdp); | |
1304afb2 | 758 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
d09b62df PM |
759 | } |
760 | ||
761 | /* | |
762 | * Do per-CPU grace-period initialization for running CPU. The caller | |
763 | * must hold the lock of the leaf rcu_node structure corresponding to | |
764 | * this CPU. | |
765 | */ | |
766 | static void | |
767 | rcu_start_gp_per_cpu(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
768 | { | |
769 | /* Prior grace period ended, so advance callbacks for current CPU. */ | |
770 | __rcu_process_gp_end(rsp, rnp, rdp); | |
771 | ||
772 | /* | |
773 | * Because this CPU just now started the new grace period, we know | |
774 | * that all of its callbacks will be covered by this upcoming grace | |
775 | * period, even the ones that were registered arbitrarily recently. | |
776 | * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL. | |
777 | * | |
778 | * Other CPUs cannot be sure exactly when the grace period started. | |
779 | * Therefore, their recently registered callbacks must pass through | |
780 | * an additional RCU_NEXT_READY stage, so that they will be handled | |
781 | * by the next RCU grace period. | |
782 | */ | |
783 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
784 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
9160306e PM |
785 | |
786 | /* Set state so that this CPU will detect the next quiescent state. */ | |
787 | __note_new_gpnum(rsp, rnp, rdp); | |
d09b62df PM |
788 | } |
789 | ||
64db4cff PM |
790 | /* |
791 | * Start a new RCU grace period if warranted, re-initializing the hierarchy | |
792 | * in preparation for detecting the next grace period. The caller must hold | |
793 | * the root node's ->lock, which is released before return. Hard irqs must | |
794 | * be disabled. | |
795 | */ | |
796 | static void | |
797 | rcu_start_gp(struct rcu_state *rsp, unsigned long flags) | |
798 | __releases(rcu_get_root(rsp)->lock) | |
799 | { | |
394f99a9 | 800 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
64db4cff | 801 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff | 802 | |
07079d53 | 803 | if (!cpu_needs_another_gp(rsp, rdp) || rsp->fqs_active) { |
46a1e34e PM |
804 | if (cpu_needs_another_gp(rsp, rdp)) |
805 | rsp->fqs_need_gp = 1; | |
b32e9eb6 | 806 | if (rnp->completed == rsp->completed) { |
1304afb2 | 807 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
b32e9eb6 PM |
808 | return; |
809 | } | |
1304afb2 | 810 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
b32e9eb6 PM |
811 | |
812 | /* | |
813 | * Propagate new ->completed value to rcu_node structures | |
814 | * so that other CPUs don't have to wait until the start | |
815 | * of the next grace period to process their callbacks. | |
816 | */ | |
817 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
1304afb2 | 818 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
b32e9eb6 | 819 | rnp->completed = rsp->completed; |
1304afb2 | 820 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
b32e9eb6 PM |
821 | } |
822 | local_irq_restore(flags); | |
64db4cff PM |
823 | return; |
824 | } | |
825 | ||
826 | /* Advance to a new grace period and initialize state. */ | |
827 | rsp->gpnum++; | |
c3422bea | 828 | WARN_ON_ONCE(rsp->signaled == RCU_GP_INIT); |
64db4cff PM |
829 | rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */ |
830 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; | |
64db4cff | 831 | record_gp_stall_check_time(rsp); |
64db4cff | 832 | |
64db4cff PM |
833 | /* Special-case the common single-level case. */ |
834 | if (NUM_RCU_NODES == 1) { | |
b0e165c0 | 835 | rcu_preempt_check_blocked_tasks(rnp); |
28ecd580 | 836 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 837 | rnp->gpnum = rsp->gpnum; |
d09b62df | 838 | rnp->completed = rsp->completed; |
c12172c0 | 839 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */ |
d09b62df | 840 | rcu_start_gp_per_cpu(rsp, rnp, rdp); |
27f4d280 | 841 | rcu_preempt_boost_start_gp(rnp); |
1304afb2 | 842 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
843 | return; |
844 | } | |
845 | ||
1304afb2 | 846 | raw_spin_unlock(&rnp->lock); /* leave irqs disabled. */ |
64db4cff PM |
847 | |
848 | ||
849 | /* Exclude any concurrent CPU-hotplug operations. */ | |
1304afb2 | 850 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
64db4cff PM |
851 | |
852 | /* | |
b835db1f PM |
853 | * Set the quiescent-state-needed bits in all the rcu_node |
854 | * structures for all currently online CPUs in breadth-first | |
855 | * order, starting from the root rcu_node structure. This | |
856 | * operation relies on the layout of the hierarchy within the | |
857 | * rsp->node[] array. Note that other CPUs will access only | |
858 | * the leaves of the hierarchy, which still indicate that no | |
859 | * grace period is in progress, at least until the corresponding | |
860 | * leaf node has been initialized. In addition, we have excluded | |
861 | * CPU-hotplug operations. | |
64db4cff PM |
862 | * |
863 | * Note that the grace period cannot complete until we finish | |
864 | * the initialization process, as there will be at least one | |
865 | * qsmask bit set in the root node until that time, namely the | |
b835db1f PM |
866 | * one corresponding to this CPU, due to the fact that we have |
867 | * irqs disabled. | |
64db4cff | 868 | */ |
a0b6c9a7 | 869 | rcu_for_each_node_breadth_first(rsp, rnp) { |
1304afb2 | 870 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
b0e165c0 | 871 | rcu_preempt_check_blocked_tasks(rnp); |
49e29126 | 872 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 873 | rnp->gpnum = rsp->gpnum; |
d09b62df PM |
874 | rnp->completed = rsp->completed; |
875 | if (rnp == rdp->mynode) | |
876 | rcu_start_gp_per_cpu(rsp, rnp, rdp); | |
27f4d280 | 877 | rcu_preempt_boost_start_gp(rnp); |
1304afb2 | 878 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
879 | } |
880 | ||
83f5b01f | 881 | rnp = rcu_get_root(rsp); |
1304afb2 | 882 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff | 883 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */ |
1304afb2 PM |
884 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
885 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); | |
64db4cff PM |
886 | } |
887 | ||
f41d911f | 888 | /* |
d3f6bad3 PM |
889 | * Report a full set of quiescent states to the specified rcu_state |
890 | * data structure. This involves cleaning up after the prior grace | |
891 | * period and letting rcu_start_gp() start up the next grace period | |
892 | * if one is needed. Note that the caller must hold rnp->lock, as | |
893 | * required by rcu_start_gp(), which will release it. | |
f41d911f | 894 | */ |
d3f6bad3 | 895 | static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags) |
fc2219d4 | 896 | __releases(rcu_get_root(rsp)->lock) |
f41d911f | 897 | { |
15ba0ba8 PM |
898 | unsigned long gp_duration; |
899 | ||
fc2219d4 | 900 | WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); |
0bbcc529 PM |
901 | |
902 | /* | |
903 | * Ensure that all grace-period and pre-grace-period activity | |
904 | * is seen before the assignment to rsp->completed. | |
905 | */ | |
906 | smp_mb(); /* See above block comment. */ | |
15ba0ba8 PM |
907 | gp_duration = jiffies - rsp->gp_start; |
908 | if (gp_duration > rsp->gp_max) | |
909 | rsp->gp_max = gp_duration; | |
f41d911f | 910 | rsp->completed = rsp->gpnum; |
83f5b01f | 911 | rsp->signaled = RCU_GP_IDLE; |
f41d911f PM |
912 | rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */ |
913 | } | |
914 | ||
64db4cff | 915 | /* |
d3f6bad3 PM |
916 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
917 | * Allows quiescent states for a group of CPUs to be reported at one go | |
918 | * to the specified rcu_node structure, though all the CPUs in the group | |
919 | * must be represented by the same rcu_node structure (which need not be | |
920 | * a leaf rcu_node structure, though it often will be). That structure's | |
921 | * lock must be held upon entry, and it is released before return. | |
64db4cff PM |
922 | */ |
923 | static void | |
d3f6bad3 PM |
924 | rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp, |
925 | struct rcu_node *rnp, unsigned long flags) | |
64db4cff PM |
926 | __releases(rnp->lock) |
927 | { | |
28ecd580 PM |
928 | struct rcu_node *rnp_c; |
929 | ||
64db4cff PM |
930 | /* Walk up the rcu_node hierarchy. */ |
931 | for (;;) { | |
932 | if (!(rnp->qsmask & mask)) { | |
933 | ||
934 | /* Our bit has already been cleared, so done. */ | |
1304afb2 | 935 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
936 | return; |
937 | } | |
938 | rnp->qsmask &= ~mask; | |
27f4d280 | 939 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
64db4cff PM |
940 | |
941 | /* Other bits still set at this level, so done. */ | |
1304afb2 | 942 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
943 | return; |
944 | } | |
945 | mask = rnp->grpmask; | |
946 | if (rnp->parent == NULL) { | |
947 | ||
948 | /* No more levels. Exit loop holding root lock. */ | |
949 | ||
950 | break; | |
951 | } | |
1304afb2 | 952 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
28ecd580 | 953 | rnp_c = rnp; |
64db4cff | 954 | rnp = rnp->parent; |
1304afb2 | 955 | raw_spin_lock_irqsave(&rnp->lock, flags); |
28ecd580 | 956 | WARN_ON_ONCE(rnp_c->qsmask); |
64db4cff PM |
957 | } |
958 | ||
959 | /* | |
960 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 961 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 962 | * to clean up and start the next grace period if one is needed. |
64db4cff | 963 | */ |
d3f6bad3 | 964 | rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */ |
64db4cff PM |
965 | } |
966 | ||
967 | /* | |
d3f6bad3 PM |
968 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
969 | * structure. This must be either called from the specified CPU, or | |
970 | * called when the specified CPU is known to be offline (and when it is | |
971 | * also known that no other CPU is concurrently trying to help the offline | |
972 | * CPU). The lastcomp argument is used to make sure we are still in the | |
973 | * grace period of interest. We don't want to end the current grace period | |
974 | * based on quiescent states detected in an earlier grace period! | |
64db4cff PM |
975 | */ |
976 | static void | |
d3f6bad3 | 977 | rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp) |
64db4cff PM |
978 | { |
979 | unsigned long flags; | |
980 | unsigned long mask; | |
981 | struct rcu_node *rnp; | |
982 | ||
983 | rnp = rdp->mynode; | |
1304afb2 | 984 | raw_spin_lock_irqsave(&rnp->lock, flags); |
560d4bc0 | 985 | if (lastcomp != rnp->completed) { |
64db4cff PM |
986 | |
987 | /* | |
988 | * Someone beat us to it for this grace period, so leave. | |
989 | * The race with GP start is resolved by the fact that we | |
990 | * hold the leaf rcu_node lock, so that the per-CPU bits | |
991 | * cannot yet be initialized -- so we would simply find our | |
d3f6bad3 PM |
992 | * CPU's bit already cleared in rcu_report_qs_rnp() if this |
993 | * race occurred. | |
64db4cff PM |
994 | */ |
995 | rdp->passed_quiesc = 0; /* try again later! */ | |
1304afb2 | 996 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
997 | return; |
998 | } | |
999 | mask = rdp->grpmask; | |
1000 | if ((rnp->qsmask & mask) == 0) { | |
1304afb2 | 1001 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1002 | } else { |
1003 | rdp->qs_pending = 0; | |
1004 | ||
1005 | /* | |
1006 | * This GP can't end until cpu checks in, so all of our | |
1007 | * callbacks can be processed during the next GP. | |
1008 | */ | |
64db4cff PM |
1009 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
1010 | ||
d3f6bad3 | 1011 | rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */ |
64db4cff PM |
1012 | } |
1013 | } | |
1014 | ||
1015 | /* | |
1016 | * Check to see if there is a new grace period of which this CPU | |
1017 | * is not yet aware, and if so, set up local rcu_data state for it. | |
1018 | * Otherwise, see if this CPU has just passed through its first | |
1019 | * quiescent state for this grace period, and record that fact if so. | |
1020 | */ | |
1021 | static void | |
1022 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | |
1023 | { | |
1024 | /* If there is now a new grace period, record and return. */ | |
1025 | if (check_for_new_grace_period(rsp, rdp)) | |
1026 | return; | |
1027 | ||
1028 | /* | |
1029 | * Does this CPU still need to do its part for current grace period? | |
1030 | * If no, return and let the other CPUs do their part as well. | |
1031 | */ | |
1032 | if (!rdp->qs_pending) | |
1033 | return; | |
1034 | ||
1035 | /* | |
1036 | * Was there a quiescent state since the beginning of the grace | |
1037 | * period? If no, then exit and wait for the next call. | |
1038 | */ | |
1039 | if (!rdp->passed_quiesc) | |
1040 | return; | |
1041 | ||
d3f6bad3 PM |
1042 | /* |
1043 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
1044 | * judge of that). | |
1045 | */ | |
1046 | rcu_report_qs_rdp(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed); | |
64db4cff PM |
1047 | } |
1048 | ||
1049 | #ifdef CONFIG_HOTPLUG_CPU | |
1050 | ||
e74f4c45 | 1051 | /* |
29494be7 LJ |
1052 | * Move a dying CPU's RCU callbacks to online CPU's callback list. |
1053 | * Synchronization is not required because this function executes | |
1054 | * in stop_machine() context. | |
e74f4c45 | 1055 | */ |
29494be7 | 1056 | static void rcu_send_cbs_to_online(struct rcu_state *rsp) |
e74f4c45 PM |
1057 | { |
1058 | int i; | |
29494be7 LJ |
1059 | /* current DYING CPU is cleared in the cpu_online_mask */ |
1060 | int receive_cpu = cpumask_any(cpu_online_mask); | |
394f99a9 | 1061 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
29494be7 | 1062 | struct rcu_data *receive_rdp = per_cpu_ptr(rsp->rda, receive_cpu); |
e74f4c45 PM |
1063 | |
1064 | if (rdp->nxtlist == NULL) | |
1065 | return; /* irqs disabled, so comparison is stable. */ | |
29494be7 LJ |
1066 | |
1067 | *receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist; | |
1068 | receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
1069 | receive_rdp->qlen += rdp->qlen; | |
1070 | receive_rdp->n_cbs_adopted += rdp->qlen; | |
1071 | rdp->n_cbs_orphaned += rdp->qlen; | |
1072 | ||
e74f4c45 PM |
1073 | rdp->nxtlist = NULL; |
1074 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
1075 | rdp->nxttail[i] = &rdp->nxtlist; | |
e74f4c45 | 1076 | rdp->qlen = 0; |
e74f4c45 PM |
1077 | } |
1078 | ||
64db4cff PM |
1079 | /* |
1080 | * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy | |
1081 | * and move all callbacks from the outgoing CPU to the current one. | |
a26ac245 PM |
1082 | * There can only be one CPU hotplug operation at a time, so no other |
1083 | * CPU can be attempting to update rcu_cpu_kthread_task. | |
64db4cff PM |
1084 | */ |
1085 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp) | |
1086 | { | |
64db4cff | 1087 | unsigned long flags; |
64db4cff | 1088 | unsigned long mask; |
d9a3da06 | 1089 | int need_report = 0; |
394f99a9 | 1090 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff | 1091 | struct rcu_node *rnp; |
a26ac245 PM |
1092 | struct task_struct *t; |
1093 | ||
1094 | /* Stop the CPU's kthread. */ | |
1095 | t = per_cpu(rcu_cpu_kthread_task, cpu); | |
1096 | if (t != NULL) { | |
1097 | per_cpu(rcu_cpu_kthread_task, cpu) = NULL; | |
1098 | kthread_stop(t); | |
1099 | } | |
64db4cff PM |
1100 | |
1101 | /* Exclude any attempts to start a new grace period. */ | |
1304afb2 | 1102 | raw_spin_lock_irqsave(&rsp->onofflock, flags); |
64db4cff PM |
1103 | |
1104 | /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */ | |
28ecd580 | 1105 | rnp = rdp->mynode; /* this is the outgoing CPU's rnp. */ |
64db4cff PM |
1106 | mask = rdp->grpmask; /* rnp->grplo is constant. */ |
1107 | do { | |
1304afb2 | 1108 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
1109 | rnp->qsmaskinit &= ~mask; |
1110 | if (rnp->qsmaskinit != 0) { | |
b668c9cf | 1111 | if (rnp != rdp->mynode) |
1304afb2 | 1112 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
1113 | break; |
1114 | } | |
b668c9cf | 1115 | if (rnp == rdp->mynode) |
d9a3da06 | 1116 | need_report = rcu_preempt_offline_tasks(rsp, rnp, rdp); |
b668c9cf | 1117 | else |
1304afb2 | 1118 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff | 1119 | mask = rnp->grpmask; |
64db4cff PM |
1120 | rnp = rnp->parent; |
1121 | } while (rnp != NULL); | |
64db4cff | 1122 | |
b668c9cf PM |
1123 | /* |
1124 | * We still hold the leaf rcu_node structure lock here, and | |
1125 | * irqs are still disabled. The reason for this subterfuge is | |
d3f6bad3 PM |
1126 | * because invoking rcu_report_unblock_qs_rnp() with ->onofflock |
1127 | * held leads to deadlock. | |
b668c9cf | 1128 | */ |
1304afb2 | 1129 | raw_spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ |
b668c9cf | 1130 | rnp = rdp->mynode; |
d9a3da06 | 1131 | if (need_report & RCU_OFL_TASKS_NORM_GP) |
d3f6bad3 | 1132 | rcu_report_unblock_qs_rnp(rnp, flags); |
b668c9cf | 1133 | else |
1304afb2 | 1134 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
d9a3da06 PM |
1135 | if (need_report & RCU_OFL_TASKS_EXP_GP) |
1136 | rcu_report_exp_rnp(rsp, rnp); | |
1217ed1b | 1137 | rcu_node_kthread_setaffinity(rnp, -1); |
64db4cff PM |
1138 | } |
1139 | ||
1140 | /* | |
1141 | * Remove the specified CPU from the RCU hierarchy and move any pending | |
1142 | * callbacks that it might have to the current CPU. This code assumes | |
1143 | * that at least one CPU in the system will remain running at all times. | |
1144 | * Any attempt to offline -all- CPUs is likely to strand RCU callbacks. | |
1145 | */ | |
1146 | static void rcu_offline_cpu(int cpu) | |
1147 | { | |
d6714c22 | 1148 | __rcu_offline_cpu(cpu, &rcu_sched_state); |
64db4cff | 1149 | __rcu_offline_cpu(cpu, &rcu_bh_state); |
33f76148 | 1150 | rcu_preempt_offline_cpu(cpu); |
64db4cff PM |
1151 | } |
1152 | ||
1153 | #else /* #ifdef CONFIG_HOTPLUG_CPU */ | |
1154 | ||
29494be7 | 1155 | static void rcu_send_cbs_to_online(struct rcu_state *rsp) |
e74f4c45 PM |
1156 | { |
1157 | } | |
1158 | ||
64db4cff PM |
1159 | static void rcu_offline_cpu(int cpu) |
1160 | { | |
1161 | } | |
1162 | ||
1163 | #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */ | |
1164 | ||
1165 | /* | |
1166 | * Invoke any RCU callbacks that have made it to the end of their grace | |
1167 | * period. Thottle as specified by rdp->blimit. | |
1168 | */ | |
37c72e56 | 1169 | static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
1170 | { |
1171 | unsigned long flags; | |
1172 | struct rcu_head *next, *list, **tail; | |
1173 | int count; | |
1174 | ||
1175 | /* If no callbacks are ready, just return.*/ | |
1176 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) | |
1177 | return; | |
1178 | ||
1179 | /* | |
1180 | * Extract the list of ready callbacks, disabling to prevent | |
1181 | * races with call_rcu() from interrupt handlers. | |
1182 | */ | |
1183 | local_irq_save(flags); | |
1184 | list = rdp->nxtlist; | |
1185 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; | |
1186 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | |
1187 | tail = rdp->nxttail[RCU_DONE_TAIL]; | |
1188 | for (count = RCU_NEXT_SIZE - 1; count >= 0; count--) | |
1189 | if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL]) | |
1190 | rdp->nxttail[count] = &rdp->nxtlist; | |
1191 | local_irq_restore(flags); | |
1192 | ||
1193 | /* Invoke callbacks. */ | |
1194 | count = 0; | |
1195 | while (list) { | |
1196 | next = list->next; | |
1197 | prefetch(next); | |
551d55a9 | 1198 | debug_rcu_head_unqueue(list); |
9ab1544e | 1199 | __rcu_reclaim(list); |
64db4cff PM |
1200 | list = next; |
1201 | if (++count >= rdp->blimit) | |
1202 | break; | |
1203 | } | |
1204 | ||
1205 | local_irq_save(flags); | |
1206 | ||
1207 | /* Update count, and requeue any remaining callbacks. */ | |
1208 | rdp->qlen -= count; | |
269dcc1c | 1209 | rdp->n_cbs_invoked += count; |
64db4cff PM |
1210 | if (list != NULL) { |
1211 | *tail = rdp->nxtlist; | |
1212 | rdp->nxtlist = list; | |
1213 | for (count = 0; count < RCU_NEXT_SIZE; count++) | |
1214 | if (&rdp->nxtlist == rdp->nxttail[count]) | |
1215 | rdp->nxttail[count] = tail; | |
1216 | else | |
1217 | break; | |
1218 | } | |
1219 | ||
1220 | /* Reinstate batch limit if we have worked down the excess. */ | |
1221 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) | |
1222 | rdp->blimit = blimit; | |
1223 | ||
37c72e56 PM |
1224 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
1225 | if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) { | |
1226 | rdp->qlen_last_fqs_check = 0; | |
1227 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1228 | } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark) | |
1229 | rdp->qlen_last_fqs_check = rdp->qlen; | |
1230 | ||
64db4cff PM |
1231 | local_irq_restore(flags); |
1232 | ||
1233 | /* Re-raise the RCU softirq if there are callbacks remaining. */ | |
1234 | if (cpu_has_callbacks_ready_to_invoke(rdp)) | |
27f4d280 | 1235 | invoke_rcu_cpu_kthread(); |
64db4cff PM |
1236 | } |
1237 | ||
1238 | /* | |
1239 | * Check to see if this CPU is in a non-context-switch quiescent state | |
1240 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
1241 | * Also schedule the RCU softirq handler. | |
1242 | * | |
1243 | * This function must be called with hardirqs disabled. It is normally | |
1244 | * invoked from the scheduling-clock interrupt. If rcu_pending returns | |
1245 | * false, there is no point in invoking rcu_check_callbacks(). | |
1246 | */ | |
1247 | void rcu_check_callbacks(int cpu, int user) | |
1248 | { | |
1249 | if (user || | |
a6826048 PM |
1250 | (idle_cpu(cpu) && rcu_scheduler_active && |
1251 | !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) { | |
64db4cff PM |
1252 | |
1253 | /* | |
1254 | * Get here if this CPU took its interrupt from user | |
1255 | * mode or from the idle loop, and if this is not a | |
1256 | * nested interrupt. In this case, the CPU is in | |
d6714c22 | 1257 | * a quiescent state, so note it. |
64db4cff PM |
1258 | * |
1259 | * No memory barrier is required here because both | |
d6714c22 PM |
1260 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
1261 | * variables that other CPUs neither access nor modify, | |
1262 | * at least not while the corresponding CPU is online. | |
64db4cff PM |
1263 | */ |
1264 | ||
d6714c22 PM |
1265 | rcu_sched_qs(cpu); |
1266 | rcu_bh_qs(cpu); | |
64db4cff PM |
1267 | |
1268 | } else if (!in_softirq()) { | |
1269 | ||
1270 | /* | |
1271 | * Get here if this CPU did not take its interrupt from | |
1272 | * softirq, in other words, if it is not interrupting | |
1273 | * a rcu_bh read-side critical section. This is an _bh | |
d6714c22 | 1274 | * critical section, so note it. |
64db4cff PM |
1275 | */ |
1276 | ||
d6714c22 | 1277 | rcu_bh_qs(cpu); |
64db4cff | 1278 | } |
f41d911f | 1279 | rcu_preempt_check_callbacks(cpu); |
d21670ac | 1280 | if (rcu_pending(cpu)) |
27f4d280 | 1281 | invoke_rcu_cpu_kthread(); |
64db4cff PM |
1282 | } |
1283 | ||
1284 | #ifdef CONFIG_SMP | |
1285 | ||
1286 | /* | |
1287 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
1288 | * have not yet encountered a quiescent state, using the function specified. | |
27f4d280 PM |
1289 | * Also initiate boosting for any threads blocked on the root rcu_node. |
1290 | * | |
ee47eb9f | 1291 | * The caller must have suppressed start of new grace periods. |
64db4cff | 1292 | */ |
45f014c5 | 1293 | static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *)) |
64db4cff PM |
1294 | { |
1295 | unsigned long bit; | |
1296 | int cpu; | |
1297 | unsigned long flags; | |
1298 | unsigned long mask; | |
a0b6c9a7 | 1299 | struct rcu_node *rnp; |
64db4cff | 1300 | |
a0b6c9a7 | 1301 | rcu_for_each_leaf_node(rsp, rnp) { |
64db4cff | 1302 | mask = 0; |
1304afb2 | 1303 | raw_spin_lock_irqsave(&rnp->lock, flags); |
ee47eb9f | 1304 | if (!rcu_gp_in_progress(rsp)) { |
1304afb2 | 1305 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
0f10dc82 | 1306 | return; |
64db4cff | 1307 | } |
a0b6c9a7 | 1308 | if (rnp->qsmask == 0) { |
1217ed1b | 1309 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock */ |
64db4cff PM |
1310 | continue; |
1311 | } | |
a0b6c9a7 | 1312 | cpu = rnp->grplo; |
64db4cff | 1313 | bit = 1; |
a0b6c9a7 | 1314 | for (; cpu <= rnp->grphi; cpu++, bit <<= 1) { |
394f99a9 LJ |
1315 | if ((rnp->qsmask & bit) != 0 && |
1316 | f(per_cpu_ptr(rsp->rda, cpu))) | |
64db4cff PM |
1317 | mask |= bit; |
1318 | } | |
45f014c5 | 1319 | if (mask != 0) { |
64db4cff | 1320 | |
d3f6bad3 PM |
1321 | /* rcu_report_qs_rnp() releases rnp->lock. */ |
1322 | rcu_report_qs_rnp(mask, rsp, rnp, flags); | |
64db4cff PM |
1323 | continue; |
1324 | } | |
1304afb2 | 1325 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 1326 | } |
27f4d280 | 1327 | rnp = rcu_get_root(rsp); |
1217ed1b PM |
1328 | if (rnp->qsmask == 0) { |
1329 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
1330 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */ | |
1331 | } | |
64db4cff PM |
1332 | } |
1333 | ||
1334 | /* | |
1335 | * Force quiescent states on reluctant CPUs, and also detect which | |
1336 | * CPUs are in dyntick-idle mode. | |
1337 | */ | |
1338 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1339 | { | |
1340 | unsigned long flags; | |
64db4cff | 1341 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff | 1342 | |
fc2219d4 | 1343 | if (!rcu_gp_in_progress(rsp)) |
64db4cff | 1344 | return; /* No grace period in progress, nothing to force. */ |
1304afb2 | 1345 | if (!raw_spin_trylock_irqsave(&rsp->fqslock, flags)) { |
64db4cff PM |
1346 | rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ |
1347 | return; /* Someone else is already on the job. */ | |
1348 | } | |
20133cfc | 1349 | if (relaxed && ULONG_CMP_GE(rsp->jiffies_force_qs, jiffies)) |
f96e9232 | 1350 | goto unlock_fqs_ret; /* no emergency and done recently. */ |
64db4cff | 1351 | rsp->n_force_qs++; |
1304afb2 | 1352 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
64db4cff | 1353 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
560d4bc0 | 1354 | if(!rcu_gp_in_progress(rsp)) { |
64db4cff | 1355 | rsp->n_force_qs_ngp++; |
1304afb2 | 1356 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1357 | goto unlock_fqs_ret; /* no GP in progress, time updated. */ |
64db4cff | 1358 | } |
07079d53 | 1359 | rsp->fqs_active = 1; |
f3a8b5c6 | 1360 | switch (rsp->signaled) { |
83f5b01f | 1361 | case RCU_GP_IDLE: |
64db4cff PM |
1362 | case RCU_GP_INIT: |
1363 | ||
83f5b01f | 1364 | break; /* grace period idle or initializing, ignore. */ |
64db4cff PM |
1365 | |
1366 | case RCU_SAVE_DYNTICK: | |
64db4cff PM |
1367 | if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK) |
1368 | break; /* So gcc recognizes the dead code. */ | |
1369 | ||
f261414f LJ |
1370 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
1371 | ||
64db4cff | 1372 | /* Record dyntick-idle state. */ |
45f014c5 | 1373 | force_qs_rnp(rsp, dyntick_save_progress_counter); |
1304afb2 | 1374 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
ee47eb9f | 1375 | if (rcu_gp_in_progress(rsp)) |
64db4cff | 1376 | rsp->signaled = RCU_FORCE_QS; |
ee47eb9f | 1377 | break; |
64db4cff PM |
1378 | |
1379 | case RCU_FORCE_QS: | |
1380 | ||
1381 | /* Check dyntick-idle state, send IPI to laggarts. */ | |
1304afb2 | 1382 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
45f014c5 | 1383 | force_qs_rnp(rsp, rcu_implicit_dynticks_qs); |
64db4cff PM |
1384 | |
1385 | /* Leave state in case more forcing is required. */ | |
1386 | ||
1304afb2 | 1387 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
f96e9232 | 1388 | break; |
64db4cff | 1389 | } |
07079d53 | 1390 | rsp->fqs_active = 0; |
46a1e34e | 1391 | if (rsp->fqs_need_gp) { |
1304afb2 | 1392 | raw_spin_unlock(&rsp->fqslock); /* irqs remain disabled */ |
46a1e34e PM |
1393 | rsp->fqs_need_gp = 0; |
1394 | rcu_start_gp(rsp, flags); /* releases rnp->lock */ | |
1395 | return; | |
1396 | } | |
1304afb2 | 1397 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1398 | unlock_fqs_ret: |
1304afb2 | 1399 | raw_spin_unlock_irqrestore(&rsp->fqslock, flags); |
64db4cff PM |
1400 | } |
1401 | ||
1402 | #else /* #ifdef CONFIG_SMP */ | |
1403 | ||
1404 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1405 | { | |
1406 | set_need_resched(); | |
1407 | } | |
1408 | ||
1409 | #endif /* #else #ifdef CONFIG_SMP */ | |
1410 | ||
1411 | /* | |
1412 | * This does the RCU processing work from softirq context for the | |
1413 | * specified rcu_state and rcu_data structures. This may be called | |
1414 | * only from the CPU to whom the rdp belongs. | |
1415 | */ | |
1416 | static void | |
1417 | __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) | |
1418 | { | |
1419 | unsigned long flags; | |
1420 | ||
2e597558 PM |
1421 | WARN_ON_ONCE(rdp->beenonline == 0); |
1422 | ||
64db4cff PM |
1423 | /* |
1424 | * If an RCU GP has gone long enough, go check for dyntick | |
1425 | * idle CPUs and, if needed, send resched IPIs. | |
1426 | */ | |
20133cfc | 1427 | if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
64db4cff PM |
1428 | force_quiescent_state(rsp, 1); |
1429 | ||
1430 | /* | |
1431 | * Advance callbacks in response to end of earlier grace | |
1432 | * period that some other CPU ended. | |
1433 | */ | |
1434 | rcu_process_gp_end(rsp, rdp); | |
1435 | ||
1436 | /* Update RCU state based on any recent quiescent states. */ | |
1437 | rcu_check_quiescent_state(rsp, rdp); | |
1438 | ||
1439 | /* Does this CPU require a not-yet-started grace period? */ | |
1440 | if (cpu_needs_another_gp(rsp, rdp)) { | |
1304afb2 | 1441 | raw_spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags); |
64db4cff PM |
1442 | rcu_start_gp(rsp, flags); /* releases above lock */ |
1443 | } | |
1444 | ||
1445 | /* If there are callbacks ready, invoke them. */ | |
37c72e56 | 1446 | rcu_do_batch(rsp, rdp); |
64db4cff PM |
1447 | } |
1448 | ||
1449 | /* | |
1450 | * Do softirq processing for the current CPU. | |
1451 | */ | |
a26ac245 | 1452 | static void rcu_process_callbacks(void) |
64db4cff | 1453 | { |
d6714c22 PM |
1454 | __rcu_process_callbacks(&rcu_sched_state, |
1455 | &__get_cpu_var(rcu_sched_data)); | |
64db4cff | 1456 | __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); |
f41d911f | 1457 | rcu_preempt_process_callbacks(); |
a47cd880 PM |
1458 | |
1459 | /* If we are last CPU on way to dyntick-idle mode, accelerate it. */ | |
1460 | rcu_needs_cpu_flush(); | |
64db4cff PM |
1461 | } |
1462 | ||
a26ac245 PM |
1463 | /* |
1464 | * Wake up the current CPU's kthread. This replaces raise_softirq() | |
1465 | * in earlier versions of RCU. Note that because we are running on | |
1466 | * the current CPU with interrupts disabled, the rcu_cpu_kthread_task | |
1467 | * cannot disappear out from under us. | |
1468 | */ | |
27f4d280 | 1469 | static void invoke_rcu_cpu_kthread(void) |
a26ac245 PM |
1470 | { |
1471 | unsigned long flags; | |
a26ac245 PM |
1472 | |
1473 | local_irq_save(flags); | |
f0a07aea PM |
1474 | __this_cpu_write(rcu_cpu_has_work, 1); |
1475 | if (__this_cpu_read(rcu_cpu_kthread_task) == NULL) { | |
a26ac245 PM |
1476 | local_irq_restore(flags); |
1477 | return; | |
1478 | } | |
f0a07aea | 1479 | wake_up(&__get_cpu_var(rcu_cpu_wq)); |
a26ac245 PM |
1480 | local_irq_restore(flags); |
1481 | } | |
1482 | ||
27f4d280 PM |
1483 | /* |
1484 | * Wake up the specified per-rcu_node-structure kthread. | |
1217ed1b PM |
1485 | * Because the per-rcu_node kthreads are immortal, we don't need |
1486 | * to do anything to keep them alive. | |
27f4d280 PM |
1487 | */ |
1488 | static void invoke_rcu_node_kthread(struct rcu_node *rnp) | |
1489 | { | |
1490 | struct task_struct *t; | |
1491 | ||
1492 | t = rnp->node_kthread_task; | |
1493 | if (t != NULL) | |
1494 | wake_up_process(t); | |
1495 | } | |
1496 | ||
e3995a25 PM |
1497 | /* |
1498 | * Set the specified CPU's kthread to run RT or not, as specified by | |
1499 | * the to_rt argument. The CPU-hotplug locks are held, so the task | |
1500 | * is not going away. | |
1501 | */ | |
1502 | static void rcu_cpu_kthread_setrt(int cpu, int to_rt) | |
1503 | { | |
1504 | int policy; | |
1505 | struct sched_param sp; | |
1506 | struct task_struct *t; | |
1507 | ||
1508 | t = per_cpu(rcu_cpu_kthread_task, cpu); | |
1509 | if (t == NULL) | |
1510 | return; | |
1511 | if (to_rt) { | |
1512 | policy = SCHED_FIFO; | |
1513 | sp.sched_priority = RCU_KTHREAD_PRIO; | |
1514 | } else { | |
1515 | policy = SCHED_NORMAL; | |
1516 | sp.sched_priority = 0; | |
1517 | } | |
1518 | sched_setscheduler_nocheck(t, policy, &sp); | |
1519 | } | |
1520 | ||
a26ac245 PM |
1521 | /* |
1522 | * Timer handler to initiate the waking up of per-CPU kthreads that | |
1523 | * have yielded the CPU due to excess numbers of RCU callbacks. | |
27f4d280 PM |
1524 | * We wake up the per-rcu_node kthread, which in turn will wake up |
1525 | * the booster kthread. | |
a26ac245 PM |
1526 | */ |
1527 | static void rcu_cpu_kthread_timer(unsigned long arg) | |
1528 | { | |
1529 | unsigned long flags; | |
27f4d280 | 1530 | struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, arg); |
a26ac245 | 1531 | struct rcu_node *rnp = rdp->mynode; |
a26ac245 PM |
1532 | |
1533 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
1534 | rnp->wakemask |= rdp->grpmask; | |
a26ac245 | 1535 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1217ed1b | 1536 | invoke_rcu_node_kthread(rnp); |
a26ac245 PM |
1537 | } |
1538 | ||
1539 | /* | |
1540 | * Drop to non-real-time priority and yield, but only after posting a | |
1541 | * timer that will cause us to regain our real-time priority if we | |
1542 | * remain preempted. Either way, we restore our real-time priority | |
1543 | * before returning. | |
1544 | */ | |
27f4d280 | 1545 | static void rcu_yield(void (*f)(unsigned long), unsigned long arg) |
a26ac245 | 1546 | { |
a26ac245 PM |
1547 | struct sched_param sp; |
1548 | struct timer_list yield_timer; | |
1549 | ||
27f4d280 | 1550 | setup_timer_on_stack(&yield_timer, f, arg); |
a26ac245 PM |
1551 | mod_timer(&yield_timer, jiffies + 2); |
1552 | sp.sched_priority = 0; | |
1553 | sched_setscheduler_nocheck(current, SCHED_NORMAL, &sp); | |
baa1ae0c | 1554 | set_user_nice(current, 19); |
a26ac245 PM |
1555 | schedule(); |
1556 | sp.sched_priority = RCU_KTHREAD_PRIO; | |
1557 | sched_setscheduler_nocheck(current, SCHED_FIFO, &sp); | |
1558 | del_timer(&yield_timer); | |
1559 | } | |
1560 | ||
1561 | /* | |
1562 | * Handle cases where the rcu_cpu_kthread() ends up on the wrong CPU. | |
1563 | * This can happen while the corresponding CPU is either coming online | |
1564 | * or going offline. We cannot wait until the CPU is fully online | |
1565 | * before starting the kthread, because the various notifier functions | |
1566 | * can wait for RCU grace periods. So we park rcu_cpu_kthread() until | |
1567 | * the corresponding CPU is online. | |
1568 | * | |
1569 | * Return 1 if the kthread needs to stop, 0 otherwise. | |
1570 | * | |
1571 | * Caller must disable bh. This function can momentarily enable it. | |
1572 | */ | |
1573 | static int rcu_cpu_kthread_should_stop(int cpu) | |
1574 | { | |
1575 | while (cpu_is_offline(cpu) || | |
1576 | !cpumask_equal(¤t->cpus_allowed, cpumask_of(cpu)) || | |
1577 | smp_processor_id() != cpu) { | |
1578 | if (kthread_should_stop()) | |
1579 | return 1; | |
15ba0ba8 PM |
1580 | per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU; |
1581 | per_cpu(rcu_cpu_kthread_cpu, cpu) = raw_smp_processor_id(); | |
a26ac245 PM |
1582 | local_bh_enable(); |
1583 | schedule_timeout_uninterruptible(1); | |
1584 | if (!cpumask_equal(¤t->cpus_allowed, cpumask_of(cpu))) | |
1585 | set_cpus_allowed_ptr(current, cpumask_of(cpu)); | |
1586 | local_bh_disable(); | |
1587 | } | |
15ba0ba8 | 1588 | per_cpu(rcu_cpu_kthread_cpu, cpu) = cpu; |
a26ac245 PM |
1589 | return 0; |
1590 | } | |
1591 | ||
1592 | /* | |
1593 | * Per-CPU kernel thread that invokes RCU callbacks. This replaces the | |
1594 | * earlier RCU softirq. | |
1595 | */ | |
1596 | static int rcu_cpu_kthread(void *arg) | |
1597 | { | |
1598 | int cpu = (int)(long)arg; | |
1599 | unsigned long flags; | |
1600 | int spincnt = 0; | |
d71df90e | 1601 | unsigned int *statusp = &per_cpu(rcu_cpu_kthread_status, cpu); |
a26ac245 PM |
1602 | wait_queue_head_t *wqp = &per_cpu(rcu_cpu_wq, cpu); |
1603 | char work; | |
1604 | char *workp = &per_cpu(rcu_cpu_has_work, cpu); | |
1605 | ||
1606 | for (;;) { | |
d71df90e | 1607 | *statusp = RCU_KTHREAD_WAITING; |
a26ac245 PM |
1608 | wait_event_interruptible(*wqp, |
1609 | *workp != 0 || kthread_should_stop()); | |
1610 | local_bh_disable(); | |
1611 | if (rcu_cpu_kthread_should_stop(cpu)) { | |
1612 | local_bh_enable(); | |
1613 | break; | |
1614 | } | |
d71df90e | 1615 | *statusp = RCU_KTHREAD_RUNNING; |
5ece5bab | 1616 | per_cpu(rcu_cpu_kthread_loops, cpu)++; |
a26ac245 PM |
1617 | local_irq_save(flags); |
1618 | work = *workp; | |
1619 | *workp = 0; | |
1620 | local_irq_restore(flags); | |
1621 | if (work) | |
1622 | rcu_process_callbacks(); | |
1623 | local_bh_enable(); | |
1624 | if (*workp != 0) | |
1625 | spincnt++; | |
1626 | else | |
1627 | spincnt = 0; | |
1628 | if (spincnt > 10) { | |
d71df90e | 1629 | *statusp = RCU_KTHREAD_YIELDING; |
27f4d280 | 1630 | rcu_yield(rcu_cpu_kthread_timer, (unsigned long)cpu); |
a26ac245 PM |
1631 | spincnt = 0; |
1632 | } | |
1633 | } | |
d71df90e | 1634 | *statusp = RCU_KTHREAD_STOPPED; |
a26ac245 PM |
1635 | return 0; |
1636 | } | |
1637 | ||
1638 | /* | |
1639 | * Spawn a per-CPU kthread, setting up affinity and priority. | |
1640 | * Because the CPU hotplug lock is held, no other CPU will be attempting | |
1641 | * to manipulate rcu_cpu_kthread_task. There might be another CPU | |
1642 | * attempting to access it during boot, but the locking in kthread_bind() | |
1643 | * will enforce sufficient ordering. | |
1644 | */ | |
1645 | static int __cpuinit rcu_spawn_one_cpu_kthread(int cpu) | |
1646 | { | |
1647 | struct sched_param sp; | |
1648 | struct task_struct *t; | |
1649 | ||
1650 | if (!rcu_kthreads_spawnable || | |
1651 | per_cpu(rcu_cpu_kthread_task, cpu) != NULL) | |
1652 | return 0; | |
1653 | t = kthread_create(rcu_cpu_kthread, (void *)(long)cpu, "rcuc%d", cpu); | |
1654 | if (IS_ERR(t)) | |
1655 | return PTR_ERR(t); | |
1656 | kthread_bind(t, cpu); | |
15ba0ba8 | 1657 | per_cpu(rcu_cpu_kthread_cpu, cpu) = cpu; |
a26ac245 PM |
1658 | WARN_ON_ONCE(per_cpu(rcu_cpu_kthread_task, cpu) != NULL); |
1659 | per_cpu(rcu_cpu_kthread_task, cpu) = t; | |
1660 | wake_up_process(t); | |
1661 | sp.sched_priority = RCU_KTHREAD_PRIO; | |
1662 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); | |
1663 | return 0; | |
1664 | } | |
1665 | ||
1666 | /* | |
1667 | * Per-rcu_node kthread, which is in charge of waking up the per-CPU | |
1668 | * kthreads when needed. We ignore requests to wake up kthreads | |
1669 | * for offline CPUs, which is OK because force_quiescent_state() | |
1670 | * takes care of this case. | |
1671 | */ | |
1672 | static int rcu_node_kthread(void *arg) | |
1673 | { | |
1674 | int cpu; | |
1675 | unsigned long flags; | |
1676 | unsigned long mask; | |
1677 | struct rcu_node *rnp = (struct rcu_node *)arg; | |
1678 | struct sched_param sp; | |
1679 | struct task_struct *t; | |
1680 | ||
1681 | for (;;) { | |
d71df90e | 1682 | rnp->node_kthread_status = RCU_KTHREAD_WAITING; |
1217ed1b | 1683 | wait_event_interruptible(rnp->node_wq, rnp->wakemask != 0); |
d71df90e | 1684 | rnp->node_kthread_status = RCU_KTHREAD_RUNNING; |
a26ac245 PM |
1685 | raw_spin_lock_irqsave(&rnp->lock, flags); |
1686 | mask = rnp->wakemask; | |
1687 | rnp->wakemask = 0; | |
1217ed1b | 1688 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */ |
a26ac245 PM |
1689 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1) { |
1690 | if ((mask & 0x1) == 0) | |
1691 | continue; | |
1692 | preempt_disable(); | |
1693 | t = per_cpu(rcu_cpu_kthread_task, cpu); | |
1694 | if (!cpu_online(cpu) || t == NULL) { | |
1695 | preempt_enable(); | |
1696 | continue; | |
1697 | } | |
1698 | per_cpu(rcu_cpu_has_work, cpu) = 1; | |
1699 | sp.sched_priority = RCU_KTHREAD_PRIO; | |
1700 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); | |
1701 | preempt_enable(); | |
1702 | } | |
1703 | } | |
1217ed1b | 1704 | /* NOTREACHED */ |
d71df90e | 1705 | rnp->node_kthread_status = RCU_KTHREAD_STOPPED; |
a26ac245 PM |
1706 | return 0; |
1707 | } | |
1708 | ||
1709 | /* | |
1710 | * Set the per-rcu_node kthread's affinity to cover all CPUs that are | |
27f4d280 PM |
1711 | * served by the rcu_node in question. The CPU hotplug lock is still |
1712 | * held, so the value of rnp->qsmaskinit will be stable. | |
0f962a5e PM |
1713 | * |
1714 | * We don't include outgoingcpu in the affinity set, use -1 if there is | |
1715 | * no outgoing CPU. If there are no CPUs left in the affinity set, | |
1716 | * this function allows the kthread to execute on any CPU. | |
a26ac245 | 1717 | */ |
0f962a5e | 1718 | static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
a26ac245 PM |
1719 | { |
1720 | cpumask_var_t cm; | |
1721 | int cpu; | |
1722 | unsigned long mask = rnp->qsmaskinit; | |
1723 | ||
1217ed1b | 1724 | if (rnp->node_kthread_task == NULL) |
a26ac245 PM |
1725 | return; |
1726 | if (!alloc_cpumask_var(&cm, GFP_KERNEL)) | |
1727 | return; | |
1728 | cpumask_clear(cm); | |
1729 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1) | |
0f962a5e | 1730 | if ((mask & 0x1) && cpu != outgoingcpu) |
a26ac245 | 1731 | cpumask_set_cpu(cpu, cm); |
0f962a5e PM |
1732 | if (cpumask_weight(cm) == 0) { |
1733 | cpumask_setall(cm); | |
1734 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++) | |
1735 | cpumask_clear_cpu(cpu, cm); | |
1736 | WARN_ON_ONCE(cpumask_weight(cm) == 0); | |
1737 | } | |
a26ac245 | 1738 | set_cpus_allowed_ptr(rnp->node_kthread_task, cm); |
27f4d280 | 1739 | rcu_boost_kthread_setaffinity(rnp, cm); |
a26ac245 PM |
1740 | free_cpumask_var(cm); |
1741 | } | |
1742 | ||
1743 | /* | |
1744 | * Spawn a per-rcu_node kthread, setting priority and affinity. | |
27f4d280 PM |
1745 | * Called during boot before online/offline can happen, or, if |
1746 | * during runtime, with the main CPU-hotplug locks held. So only | |
1747 | * one of these can be executing at a time. | |
a26ac245 PM |
1748 | */ |
1749 | static int __cpuinit rcu_spawn_one_node_kthread(struct rcu_state *rsp, | |
1750 | struct rcu_node *rnp) | |
1751 | { | |
27f4d280 | 1752 | unsigned long flags; |
a26ac245 PM |
1753 | int rnp_index = rnp - &rsp->node[0]; |
1754 | struct sched_param sp; | |
1755 | struct task_struct *t; | |
1756 | ||
1757 | if (!rcu_kthreads_spawnable || | |
27f4d280 | 1758 | rnp->qsmaskinit == 0) |
a26ac245 | 1759 | return 0; |
27f4d280 PM |
1760 | if (rnp->node_kthread_task == NULL) { |
1761 | t = kthread_create(rcu_node_kthread, (void *)rnp, | |
1762 | "rcun%d", rnp_index); | |
1763 | if (IS_ERR(t)) | |
1764 | return PTR_ERR(t); | |
1765 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
1766 | rnp->node_kthread_task = t; | |
1767 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
1768 | wake_up_process(t); | |
1769 | sp.sched_priority = 99; | |
1770 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); | |
1771 | } | |
1772 | return rcu_spawn_one_boost_kthread(rsp, rnp, rnp_index); | |
a26ac245 PM |
1773 | } |
1774 | ||
1775 | /* | |
1776 | * Spawn all kthreads -- called as soon as the scheduler is running. | |
1777 | */ | |
1778 | static int __init rcu_spawn_kthreads(void) | |
1779 | { | |
1780 | int cpu; | |
1781 | struct rcu_node *rnp; | |
1782 | ||
1783 | rcu_kthreads_spawnable = 1; | |
1784 | for_each_possible_cpu(cpu) { | |
1785 | init_waitqueue_head(&per_cpu(rcu_cpu_wq, cpu)); | |
1786 | per_cpu(rcu_cpu_has_work, cpu) = 0; | |
1787 | if (cpu_online(cpu)) | |
1788 | (void)rcu_spawn_one_cpu_kthread(cpu); | |
1789 | } | |
27f4d280 PM |
1790 | rnp = rcu_get_root(rcu_state); |
1791 | init_waitqueue_head(&rnp->node_wq); | |
1792 | rcu_init_boost_waitqueue(rnp); | |
1793 | (void)rcu_spawn_one_node_kthread(rcu_state, rnp); | |
1794 | if (NUM_RCU_NODES > 1) | |
1795 | rcu_for_each_leaf_node(rcu_state, rnp) { | |
1796 | init_waitqueue_head(&rnp->node_wq); | |
1797 | rcu_init_boost_waitqueue(rnp); | |
1798 | (void)rcu_spawn_one_node_kthread(rcu_state, rnp); | |
1799 | } | |
a26ac245 PM |
1800 | return 0; |
1801 | } | |
1802 | early_initcall(rcu_spawn_kthreads); | |
1803 | ||
64db4cff PM |
1804 | static void |
1805 | __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu), | |
1806 | struct rcu_state *rsp) | |
1807 | { | |
1808 | unsigned long flags; | |
1809 | struct rcu_data *rdp; | |
1810 | ||
551d55a9 | 1811 | debug_rcu_head_queue(head); |
64db4cff PM |
1812 | head->func = func; |
1813 | head->next = NULL; | |
1814 | ||
1815 | smp_mb(); /* Ensure RCU update seen before callback registry. */ | |
1816 | ||
1817 | /* | |
1818 | * Opportunistically note grace-period endings and beginnings. | |
1819 | * Note that we might see a beginning right after we see an | |
1820 | * end, but never vice versa, since this CPU has to pass through | |
1821 | * a quiescent state betweentimes. | |
1822 | */ | |
1823 | local_irq_save(flags); | |
394f99a9 | 1824 | rdp = this_cpu_ptr(rsp->rda); |
64db4cff PM |
1825 | |
1826 | /* Add the callback to our list. */ | |
1827 | *rdp->nxttail[RCU_NEXT_TAIL] = head; | |
1828 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; | |
2655d57e PM |
1829 | rdp->qlen++; |
1830 | ||
1831 | /* If interrupts were disabled, don't dive into RCU core. */ | |
1832 | if (irqs_disabled_flags(flags)) { | |
1833 | local_irq_restore(flags); | |
1834 | return; | |
1835 | } | |
64db4cff | 1836 | |
37c72e56 PM |
1837 | /* |
1838 | * Force the grace period if too many callbacks or too long waiting. | |
1839 | * Enforce hysteresis, and don't invoke force_quiescent_state() | |
1840 | * if some other CPU has recently done so. Also, don't bother | |
1841 | * invoking force_quiescent_state() if the newly enqueued callback | |
1842 | * is the only one waiting for a grace period to complete. | |
1843 | */ | |
2655d57e | 1844 | if (unlikely(rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) { |
b52573d2 PM |
1845 | |
1846 | /* Are we ignoring a completed grace period? */ | |
1847 | rcu_process_gp_end(rsp, rdp); | |
1848 | check_for_new_grace_period(rsp, rdp); | |
1849 | ||
1850 | /* Start a new grace period if one not already started. */ | |
1851 | if (!rcu_gp_in_progress(rsp)) { | |
1852 | unsigned long nestflag; | |
1853 | struct rcu_node *rnp_root = rcu_get_root(rsp); | |
1854 | ||
1855 | raw_spin_lock_irqsave(&rnp_root->lock, nestflag); | |
1856 | rcu_start_gp(rsp, nestflag); /* rlses rnp_root->lock */ | |
1857 | } else { | |
1858 | /* Give the grace period a kick. */ | |
1859 | rdp->blimit = LONG_MAX; | |
1860 | if (rsp->n_force_qs == rdp->n_force_qs_snap && | |
1861 | *rdp->nxttail[RCU_DONE_TAIL] != head) | |
1862 | force_quiescent_state(rsp, 0); | |
1863 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1864 | rdp->qlen_last_fqs_check = rdp->qlen; | |
1865 | } | |
20133cfc | 1866 | } else if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
64db4cff PM |
1867 | force_quiescent_state(rsp, 1); |
1868 | local_irq_restore(flags); | |
1869 | } | |
1870 | ||
1871 | /* | |
d6714c22 | 1872 | * Queue an RCU-sched callback for invocation after a grace period. |
64db4cff | 1873 | */ |
d6714c22 | 1874 | void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) |
64db4cff | 1875 | { |
d6714c22 | 1876 | __call_rcu(head, func, &rcu_sched_state); |
64db4cff | 1877 | } |
d6714c22 | 1878 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
64db4cff PM |
1879 | |
1880 | /* | |
1881 | * Queue an RCU for invocation after a quicker grace period. | |
1882 | */ | |
1883 | void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
1884 | { | |
1885 | __call_rcu(head, func, &rcu_bh_state); | |
1886 | } | |
1887 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
1888 | ||
6ebb237b PM |
1889 | /** |
1890 | * synchronize_sched - wait until an rcu-sched grace period has elapsed. | |
1891 | * | |
1892 | * Control will return to the caller some time after a full rcu-sched | |
1893 | * grace period has elapsed, in other words after all currently executing | |
1894 | * rcu-sched read-side critical sections have completed. These read-side | |
1895 | * critical sections are delimited by rcu_read_lock_sched() and | |
1896 | * rcu_read_unlock_sched(), and may be nested. Note that preempt_disable(), | |
1897 | * local_irq_disable(), and so on may be used in place of | |
1898 | * rcu_read_lock_sched(). | |
1899 | * | |
1900 | * This means that all preempt_disable code sequences, including NMI and | |
1901 | * hardware-interrupt handlers, in progress on entry will have completed | |
1902 | * before this primitive returns. However, this does not guarantee that | |
1903 | * softirq handlers will have completed, since in some kernels, these | |
1904 | * handlers can run in process context, and can block. | |
1905 | * | |
1906 | * This primitive provides the guarantees made by the (now removed) | |
1907 | * synchronize_kernel() API. In contrast, synchronize_rcu() only | |
1908 | * guarantees that rcu_read_lock() sections will have completed. | |
1909 | * In "classic RCU", these two guarantees happen to be one and | |
1910 | * the same, but can differ in realtime RCU implementations. | |
1911 | */ | |
1912 | void synchronize_sched(void) | |
1913 | { | |
1914 | struct rcu_synchronize rcu; | |
1915 | ||
1916 | if (rcu_blocking_is_gp()) | |
1917 | return; | |
1918 | ||
72d5a9f7 | 1919 | init_rcu_head_on_stack(&rcu.head); |
6ebb237b PM |
1920 | init_completion(&rcu.completion); |
1921 | /* Will wake me after RCU finished. */ | |
1922 | call_rcu_sched(&rcu.head, wakeme_after_rcu); | |
1923 | /* Wait for it. */ | |
1924 | wait_for_completion(&rcu.completion); | |
72d5a9f7 | 1925 | destroy_rcu_head_on_stack(&rcu.head); |
6ebb237b PM |
1926 | } |
1927 | EXPORT_SYMBOL_GPL(synchronize_sched); | |
1928 | ||
1929 | /** | |
1930 | * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed. | |
1931 | * | |
1932 | * Control will return to the caller some time after a full rcu_bh grace | |
1933 | * period has elapsed, in other words after all currently executing rcu_bh | |
1934 | * read-side critical sections have completed. RCU read-side critical | |
1935 | * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(), | |
1936 | * and may be nested. | |
1937 | */ | |
1938 | void synchronize_rcu_bh(void) | |
1939 | { | |
1940 | struct rcu_synchronize rcu; | |
1941 | ||
1942 | if (rcu_blocking_is_gp()) | |
1943 | return; | |
1944 | ||
72d5a9f7 | 1945 | init_rcu_head_on_stack(&rcu.head); |
6ebb237b PM |
1946 | init_completion(&rcu.completion); |
1947 | /* Will wake me after RCU finished. */ | |
1948 | call_rcu_bh(&rcu.head, wakeme_after_rcu); | |
1949 | /* Wait for it. */ | |
1950 | wait_for_completion(&rcu.completion); | |
72d5a9f7 | 1951 | destroy_rcu_head_on_stack(&rcu.head); |
6ebb237b PM |
1952 | } |
1953 | EXPORT_SYMBOL_GPL(synchronize_rcu_bh); | |
1954 | ||
64db4cff PM |
1955 | /* |
1956 | * Check to see if there is any immediate RCU-related work to be done | |
1957 | * by the current CPU, for the specified type of RCU, returning 1 if so. | |
1958 | * The checks are in order of increasing expense: checks that can be | |
1959 | * carried out against CPU-local state are performed first. However, | |
1960 | * we must check for CPU stalls first, else we might not get a chance. | |
1961 | */ | |
1962 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | |
1963 | { | |
2f51f988 PM |
1964 | struct rcu_node *rnp = rdp->mynode; |
1965 | ||
64db4cff PM |
1966 | rdp->n_rcu_pending++; |
1967 | ||
1968 | /* Check for CPU stalls, if enabled. */ | |
1969 | check_cpu_stall(rsp, rdp); | |
1970 | ||
1971 | /* Is the RCU core waiting for a quiescent state from this CPU? */ | |
d21670ac | 1972 | if (rdp->qs_pending && !rdp->passed_quiesc) { |
d25eb944 PM |
1973 | |
1974 | /* | |
1975 | * If force_quiescent_state() coming soon and this CPU | |
1976 | * needs a quiescent state, and this is either RCU-sched | |
1977 | * or RCU-bh, force a local reschedule. | |
1978 | */ | |
d21670ac | 1979 | rdp->n_rp_qs_pending++; |
6cc68793 | 1980 | if (!rdp->preemptible && |
d25eb944 PM |
1981 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs) - 1, |
1982 | jiffies)) | |
1983 | set_need_resched(); | |
d21670ac PM |
1984 | } else if (rdp->qs_pending && rdp->passed_quiesc) { |
1985 | rdp->n_rp_report_qs++; | |
64db4cff | 1986 | return 1; |
7ba5c840 | 1987 | } |
64db4cff PM |
1988 | |
1989 | /* Does this CPU have callbacks ready to invoke? */ | |
7ba5c840 PM |
1990 | if (cpu_has_callbacks_ready_to_invoke(rdp)) { |
1991 | rdp->n_rp_cb_ready++; | |
64db4cff | 1992 | return 1; |
7ba5c840 | 1993 | } |
64db4cff PM |
1994 | |
1995 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
7ba5c840 PM |
1996 | if (cpu_needs_another_gp(rsp, rdp)) { |
1997 | rdp->n_rp_cpu_needs_gp++; | |
64db4cff | 1998 | return 1; |
7ba5c840 | 1999 | } |
64db4cff PM |
2000 | |
2001 | /* Has another RCU grace period completed? */ | |
2f51f988 | 2002 | if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */ |
7ba5c840 | 2003 | rdp->n_rp_gp_completed++; |
64db4cff | 2004 | return 1; |
7ba5c840 | 2005 | } |
64db4cff PM |
2006 | |
2007 | /* Has a new RCU grace period started? */ | |
2f51f988 | 2008 | if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */ |
7ba5c840 | 2009 | rdp->n_rp_gp_started++; |
64db4cff | 2010 | return 1; |
7ba5c840 | 2011 | } |
64db4cff PM |
2012 | |
2013 | /* Has an RCU GP gone long enough to send resched IPIs &c? */ | |
fc2219d4 | 2014 | if (rcu_gp_in_progress(rsp) && |
20133cfc | 2015 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) { |
7ba5c840 | 2016 | rdp->n_rp_need_fqs++; |
64db4cff | 2017 | return 1; |
7ba5c840 | 2018 | } |
64db4cff PM |
2019 | |
2020 | /* nothing to do */ | |
7ba5c840 | 2021 | rdp->n_rp_need_nothing++; |
64db4cff PM |
2022 | return 0; |
2023 | } | |
2024 | ||
2025 | /* | |
2026 | * Check to see if there is any immediate RCU-related work to be done | |
2027 | * by the current CPU, returning 1 if so. This function is part of the | |
2028 | * RCU implementation; it is -not- an exported member of the RCU API. | |
2029 | */ | |
a157229c | 2030 | static int rcu_pending(int cpu) |
64db4cff | 2031 | { |
d6714c22 | 2032 | return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) || |
f41d911f PM |
2033 | __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) || |
2034 | rcu_preempt_pending(cpu); | |
64db4cff PM |
2035 | } |
2036 | ||
2037 | /* | |
2038 | * Check to see if any future RCU-related work will need to be done | |
2039 | * by the current CPU, even if none need be done immediately, returning | |
8bd93a2c | 2040 | * 1 if so. |
64db4cff | 2041 | */ |
8bd93a2c | 2042 | static int rcu_needs_cpu_quick_check(int cpu) |
64db4cff PM |
2043 | { |
2044 | /* RCU callbacks either ready or pending? */ | |
d6714c22 | 2045 | return per_cpu(rcu_sched_data, cpu).nxtlist || |
f41d911f PM |
2046 | per_cpu(rcu_bh_data, cpu).nxtlist || |
2047 | rcu_preempt_needs_cpu(cpu); | |
64db4cff PM |
2048 | } |
2049 | ||
d0ec774c PM |
2050 | static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL}; |
2051 | static atomic_t rcu_barrier_cpu_count; | |
2052 | static DEFINE_MUTEX(rcu_barrier_mutex); | |
2053 | static struct completion rcu_barrier_completion; | |
d0ec774c PM |
2054 | |
2055 | static void rcu_barrier_callback(struct rcu_head *notused) | |
2056 | { | |
2057 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
2058 | complete(&rcu_barrier_completion); | |
2059 | } | |
2060 | ||
2061 | /* | |
2062 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
2063 | */ | |
2064 | static void rcu_barrier_func(void *type) | |
2065 | { | |
2066 | int cpu = smp_processor_id(); | |
2067 | struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu); | |
2068 | void (*call_rcu_func)(struct rcu_head *head, | |
2069 | void (*func)(struct rcu_head *head)); | |
2070 | ||
2071 | atomic_inc(&rcu_barrier_cpu_count); | |
2072 | call_rcu_func = type; | |
2073 | call_rcu_func(head, rcu_barrier_callback); | |
2074 | } | |
2075 | ||
d0ec774c PM |
2076 | /* |
2077 | * Orchestrate the specified type of RCU barrier, waiting for all | |
2078 | * RCU callbacks of the specified type to complete. | |
2079 | */ | |
e74f4c45 PM |
2080 | static void _rcu_barrier(struct rcu_state *rsp, |
2081 | void (*call_rcu_func)(struct rcu_head *head, | |
d0ec774c PM |
2082 | void (*func)(struct rcu_head *head))) |
2083 | { | |
2084 | BUG_ON(in_interrupt()); | |
e74f4c45 | 2085 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
d0ec774c PM |
2086 | mutex_lock(&rcu_barrier_mutex); |
2087 | init_completion(&rcu_barrier_completion); | |
2088 | /* | |
2089 | * Initialize rcu_barrier_cpu_count to 1, then invoke | |
2090 | * rcu_barrier_func() on each CPU, so that each CPU also has | |
2091 | * incremented rcu_barrier_cpu_count. Only then is it safe to | |
2092 | * decrement rcu_barrier_cpu_count -- otherwise the first CPU | |
2093 | * might complete its grace period before all of the other CPUs | |
2094 | * did their increment, causing this function to return too | |
2d999e03 PM |
2095 | * early. Note that on_each_cpu() disables irqs, which prevents |
2096 | * any CPUs from coming online or going offline until each online | |
2097 | * CPU has queued its RCU-barrier callback. | |
d0ec774c PM |
2098 | */ |
2099 | atomic_set(&rcu_barrier_cpu_count, 1); | |
2100 | on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1); | |
2101 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
2102 | complete(&rcu_barrier_completion); | |
2103 | wait_for_completion(&rcu_barrier_completion); | |
2104 | mutex_unlock(&rcu_barrier_mutex); | |
d0ec774c | 2105 | } |
d0ec774c PM |
2106 | |
2107 | /** | |
2108 | * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete. | |
2109 | */ | |
2110 | void rcu_barrier_bh(void) | |
2111 | { | |
e74f4c45 | 2112 | _rcu_barrier(&rcu_bh_state, call_rcu_bh); |
d0ec774c PM |
2113 | } |
2114 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); | |
2115 | ||
2116 | /** | |
2117 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. | |
2118 | */ | |
2119 | void rcu_barrier_sched(void) | |
2120 | { | |
e74f4c45 | 2121 | _rcu_barrier(&rcu_sched_state, call_rcu_sched); |
d0ec774c PM |
2122 | } |
2123 | EXPORT_SYMBOL_GPL(rcu_barrier_sched); | |
2124 | ||
64db4cff | 2125 | /* |
27569620 | 2126 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 2127 | */ |
27569620 PM |
2128 | static void __init |
2129 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) | |
64db4cff PM |
2130 | { |
2131 | unsigned long flags; | |
2132 | int i; | |
394f99a9 | 2133 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
27569620 PM |
2134 | struct rcu_node *rnp = rcu_get_root(rsp); |
2135 | ||
2136 | /* Set up local state, ensuring consistent view of global state. */ | |
1304afb2 | 2137 | raw_spin_lock_irqsave(&rnp->lock, flags); |
27569620 PM |
2138 | rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo); |
2139 | rdp->nxtlist = NULL; | |
2140 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
2141 | rdp->nxttail[i] = &rdp->nxtlist; | |
2142 | rdp->qlen = 0; | |
2143 | #ifdef CONFIG_NO_HZ | |
2144 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); | |
2145 | #endif /* #ifdef CONFIG_NO_HZ */ | |
2146 | rdp->cpu = cpu; | |
1304afb2 | 2147 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
27569620 PM |
2148 | } |
2149 | ||
2150 | /* | |
2151 | * Initialize a CPU's per-CPU RCU data. Note that only one online or | |
2152 | * offline event can be happening at a given time. Note also that we | |
2153 | * can accept some slop in the rsp->completed access due to the fact | |
2154 | * that this CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 2155 | */ |
e4fa4c97 | 2156 | static void __cpuinit |
6cc68793 | 2157 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptible) |
64db4cff PM |
2158 | { |
2159 | unsigned long flags; | |
64db4cff | 2160 | unsigned long mask; |
394f99a9 | 2161 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff PM |
2162 | struct rcu_node *rnp = rcu_get_root(rsp); |
2163 | ||
2164 | /* Set up local state, ensuring consistent view of global state. */ | |
1304afb2 | 2165 | raw_spin_lock_irqsave(&rnp->lock, flags); |
64db4cff PM |
2166 | rdp->passed_quiesc = 0; /* We could be racing with new GP, */ |
2167 | rdp->qs_pending = 1; /* so set up to respond to current GP. */ | |
2168 | rdp->beenonline = 1; /* We have now been online. */ | |
6cc68793 | 2169 | rdp->preemptible = preemptible; |
37c72e56 PM |
2170 | rdp->qlen_last_fqs_check = 0; |
2171 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
64db4cff | 2172 | rdp->blimit = blimit; |
1304afb2 | 2173 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
2174 | |
2175 | /* | |
2176 | * A new grace period might start here. If so, we won't be part | |
2177 | * of it, but that is OK, as we are currently in a quiescent state. | |
2178 | */ | |
2179 | ||
2180 | /* Exclude any attempts to start a new GP on large systems. */ | |
1304afb2 | 2181 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
64db4cff PM |
2182 | |
2183 | /* Add CPU to rcu_node bitmasks. */ | |
2184 | rnp = rdp->mynode; | |
2185 | mask = rdp->grpmask; | |
2186 | do { | |
2187 | /* Exclude any attempts to start a new GP on small systems. */ | |
1304afb2 | 2188 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
2189 | rnp->qsmaskinit |= mask; |
2190 | mask = rnp->grpmask; | |
d09b62df PM |
2191 | if (rnp == rdp->mynode) { |
2192 | rdp->gpnum = rnp->completed; /* if GP in progress... */ | |
2193 | rdp->completed = rnp->completed; | |
2194 | rdp->passed_quiesc_completed = rnp->completed - 1; | |
2195 | } | |
1304afb2 | 2196 | raw_spin_unlock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
2197 | rnp = rnp->parent; |
2198 | } while (rnp != NULL && !(rnp->qsmaskinit & mask)); | |
2199 | ||
1304afb2 | 2200 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); |
64db4cff PM |
2201 | } |
2202 | ||
2203 | static void __cpuinit rcu_online_cpu(int cpu) | |
2204 | { | |
f41d911f PM |
2205 | rcu_init_percpu_data(cpu, &rcu_sched_state, 0); |
2206 | rcu_init_percpu_data(cpu, &rcu_bh_state, 0); | |
2207 | rcu_preempt_init_percpu_data(cpu); | |
64db4cff PM |
2208 | } |
2209 | ||
a26ac245 PM |
2210 | static void __cpuinit rcu_online_kthreads(int cpu) |
2211 | { | |
27f4d280 | 2212 | struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu); |
a26ac245 PM |
2213 | struct rcu_node *rnp = rdp->mynode; |
2214 | ||
2215 | /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */ | |
2216 | if (rcu_kthreads_spawnable) { | |
2217 | (void)rcu_spawn_one_cpu_kthread(cpu); | |
2218 | if (rnp->node_kthread_task == NULL) | |
27f4d280 | 2219 | (void)rcu_spawn_one_node_kthread(rcu_state, rnp); |
a26ac245 PM |
2220 | } |
2221 | } | |
2222 | ||
64db4cff | 2223 | /* |
f41d911f | 2224 | * Handle CPU online/offline notification events. |
64db4cff | 2225 | */ |
9f680ab4 PM |
2226 | static int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
2227 | unsigned long action, void *hcpu) | |
64db4cff PM |
2228 | { |
2229 | long cpu = (long)hcpu; | |
27f4d280 | 2230 | struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu); |
a26ac245 | 2231 | struct rcu_node *rnp = rdp->mynode; |
64db4cff PM |
2232 | |
2233 | switch (action) { | |
2234 | case CPU_UP_PREPARE: | |
2235 | case CPU_UP_PREPARE_FROZEN: | |
2236 | rcu_online_cpu(cpu); | |
a26ac245 PM |
2237 | rcu_online_kthreads(cpu); |
2238 | break; | |
2239 | case CPU_ONLINE: | |
0f962a5e PM |
2240 | case CPU_DOWN_FAILED: |
2241 | rcu_node_kthread_setaffinity(rnp, -1); | |
e3995a25 | 2242 | rcu_cpu_kthread_setrt(cpu, 1); |
0f962a5e PM |
2243 | break; |
2244 | case CPU_DOWN_PREPARE: | |
2245 | rcu_node_kthread_setaffinity(rnp, cpu); | |
e3995a25 | 2246 | rcu_cpu_kthread_setrt(cpu, 0); |
64db4cff | 2247 | break; |
d0ec774c PM |
2248 | case CPU_DYING: |
2249 | case CPU_DYING_FROZEN: | |
2250 | /* | |
2d999e03 PM |
2251 | * The whole machine is "stopped" except this CPU, so we can |
2252 | * touch any data without introducing corruption. We send the | |
2253 | * dying CPU's callbacks to an arbitrarily chosen online CPU. | |
d0ec774c | 2254 | */ |
29494be7 LJ |
2255 | rcu_send_cbs_to_online(&rcu_bh_state); |
2256 | rcu_send_cbs_to_online(&rcu_sched_state); | |
2257 | rcu_preempt_send_cbs_to_online(); | |
d0ec774c | 2258 | break; |
64db4cff PM |
2259 | case CPU_DEAD: |
2260 | case CPU_DEAD_FROZEN: | |
2261 | case CPU_UP_CANCELED: | |
2262 | case CPU_UP_CANCELED_FROZEN: | |
2263 | rcu_offline_cpu(cpu); | |
2264 | break; | |
2265 | default: | |
2266 | break; | |
2267 | } | |
2268 | return NOTIFY_OK; | |
2269 | } | |
2270 | ||
bbad9379 PM |
2271 | /* |
2272 | * This function is invoked towards the end of the scheduler's initialization | |
2273 | * process. Before this is called, the idle task might contain | |
2274 | * RCU read-side critical sections (during which time, this idle | |
2275 | * task is booting the system). After this function is called, the | |
2276 | * idle tasks are prohibited from containing RCU read-side critical | |
2277 | * sections. This function also enables RCU lockdep checking. | |
2278 | */ | |
2279 | void rcu_scheduler_starting(void) | |
2280 | { | |
2281 | WARN_ON(num_online_cpus() != 1); | |
2282 | WARN_ON(nr_context_switches() > 0); | |
2283 | rcu_scheduler_active = 1; | |
2284 | } | |
2285 | ||
64db4cff PM |
2286 | /* |
2287 | * Compute the per-level fanout, either using the exact fanout specified | |
2288 | * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT. | |
2289 | */ | |
2290 | #ifdef CONFIG_RCU_FANOUT_EXACT | |
2291 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
2292 | { | |
2293 | int i; | |
2294 | ||
0209f649 | 2295 | for (i = NUM_RCU_LVLS - 1; i > 0; i--) |
64db4cff | 2296 | rsp->levelspread[i] = CONFIG_RCU_FANOUT; |
0209f649 | 2297 | rsp->levelspread[0] = RCU_FANOUT_LEAF; |
64db4cff PM |
2298 | } |
2299 | #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
2300 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
2301 | { | |
2302 | int ccur; | |
2303 | int cprv; | |
2304 | int i; | |
2305 | ||
2306 | cprv = NR_CPUS; | |
2307 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
2308 | ccur = rsp->levelcnt[i]; | |
2309 | rsp->levelspread[i] = (cprv + ccur - 1) / ccur; | |
2310 | cprv = ccur; | |
2311 | } | |
2312 | } | |
2313 | #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
2314 | ||
2315 | /* | |
2316 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
2317 | */ | |
394f99a9 LJ |
2318 | static void __init rcu_init_one(struct rcu_state *rsp, |
2319 | struct rcu_data __percpu *rda) | |
64db4cff | 2320 | { |
b6407e86 PM |
2321 | static char *buf[] = { "rcu_node_level_0", |
2322 | "rcu_node_level_1", | |
2323 | "rcu_node_level_2", | |
2324 | "rcu_node_level_3" }; /* Match MAX_RCU_LVLS */ | |
64db4cff PM |
2325 | int cpustride = 1; |
2326 | int i; | |
2327 | int j; | |
2328 | struct rcu_node *rnp; | |
2329 | ||
b6407e86 PM |
2330 | BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
2331 | ||
64db4cff PM |
2332 | /* Initialize the level-tracking arrays. */ |
2333 | ||
2334 | for (i = 1; i < NUM_RCU_LVLS; i++) | |
2335 | rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1]; | |
2336 | rcu_init_levelspread(rsp); | |
2337 | ||
2338 | /* Initialize the elements themselves, starting from the leaves. */ | |
2339 | ||
2340 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
2341 | cpustride *= rsp->levelspread[i]; | |
2342 | rnp = rsp->level[i]; | |
2343 | for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { | |
1304afb2 | 2344 | raw_spin_lock_init(&rnp->lock); |
b6407e86 PM |
2345 | lockdep_set_class_and_name(&rnp->lock, |
2346 | &rcu_node_class[i], buf[i]); | |
f41d911f | 2347 | rnp->gpnum = 0; |
64db4cff PM |
2348 | rnp->qsmask = 0; |
2349 | rnp->qsmaskinit = 0; | |
2350 | rnp->grplo = j * cpustride; | |
2351 | rnp->grphi = (j + 1) * cpustride - 1; | |
2352 | if (rnp->grphi >= NR_CPUS) | |
2353 | rnp->grphi = NR_CPUS - 1; | |
2354 | if (i == 0) { | |
2355 | rnp->grpnum = 0; | |
2356 | rnp->grpmask = 0; | |
2357 | rnp->parent = NULL; | |
2358 | } else { | |
2359 | rnp->grpnum = j % rsp->levelspread[i - 1]; | |
2360 | rnp->grpmask = 1UL << rnp->grpnum; | |
2361 | rnp->parent = rsp->level[i - 1] + | |
2362 | j / rsp->levelspread[i - 1]; | |
2363 | } | |
2364 | rnp->level = i; | |
12f5f524 | 2365 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
64db4cff PM |
2366 | } |
2367 | } | |
0c34029a | 2368 | |
394f99a9 | 2369 | rsp->rda = rda; |
0c34029a LJ |
2370 | rnp = rsp->level[NUM_RCU_LVLS - 1]; |
2371 | for_each_possible_cpu(i) { | |
4a90a068 | 2372 | while (i > rnp->grphi) |
0c34029a | 2373 | rnp++; |
394f99a9 | 2374 | per_cpu_ptr(rsp->rda, i)->mynode = rnp; |
0c34029a LJ |
2375 | rcu_boot_init_percpu_data(i, rsp); |
2376 | } | |
64db4cff PM |
2377 | } |
2378 | ||
9f680ab4 | 2379 | void __init rcu_init(void) |
64db4cff | 2380 | { |
017c4261 | 2381 | int cpu; |
9f680ab4 | 2382 | |
f41d911f | 2383 | rcu_bootup_announce(); |
394f99a9 LJ |
2384 | rcu_init_one(&rcu_sched_state, &rcu_sched_data); |
2385 | rcu_init_one(&rcu_bh_state, &rcu_bh_data); | |
f41d911f | 2386 | __rcu_init_preempt(); |
9f680ab4 PM |
2387 | |
2388 | /* | |
2389 | * We don't need protection against CPU-hotplug here because | |
2390 | * this is called early in boot, before either interrupts | |
2391 | * or the scheduler are operational. | |
2392 | */ | |
2393 | cpu_notifier(rcu_cpu_notify, 0); | |
017c4261 PM |
2394 | for_each_online_cpu(cpu) |
2395 | rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu); | |
c68de209 | 2396 | check_cpu_stall_init(); |
64db4cff PM |
2397 | } |
2398 | ||
1eba8f84 | 2399 | #include "rcutree_plugin.h" |