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> | |
64db4cff | 52 | |
9f77da9f PM |
53 | #include "rcutree.h" |
54 | ||
64db4cff PM |
55 | /* Data structures. */ |
56 | ||
b668c9cf | 57 | static struct lock_class_key rcu_node_class[NUM_RCU_LVLS]; |
88b91c7c | 58 | |
4300aa64 PM |
59 | #define RCU_STATE_INITIALIZER(structname) { \ |
60 | .level = { &structname.node[0] }, \ | |
64db4cff PM |
61 | .levelcnt = { \ |
62 | NUM_RCU_LVL_0, /* root of hierarchy. */ \ | |
63 | NUM_RCU_LVL_1, \ | |
64 | NUM_RCU_LVL_2, \ | |
cf244dc0 PM |
65 | NUM_RCU_LVL_3, \ |
66 | NUM_RCU_LVL_4, /* == MAX_RCU_LVLS */ \ | |
64db4cff | 67 | }, \ |
83f5b01f | 68 | .signaled = RCU_GP_IDLE, \ |
64db4cff PM |
69 | .gpnum = -300, \ |
70 | .completed = -300, \ | |
4300aa64 | 71 | .onofflock = __RAW_SPIN_LOCK_UNLOCKED(&structname.onofflock), \ |
4300aa64 | 72 | .fqslock = __RAW_SPIN_LOCK_UNLOCKED(&structname.fqslock), \ |
64db4cff PM |
73 | .n_force_qs = 0, \ |
74 | .n_force_qs_ngp = 0, \ | |
4300aa64 | 75 | .name = #structname, \ |
64db4cff PM |
76 | } |
77 | ||
d6714c22 PM |
78 | struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched_state); |
79 | DEFINE_PER_CPU(struct rcu_data, rcu_sched_data); | |
64db4cff | 80 | |
6258c4fb IM |
81 | struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state); |
82 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); | |
b1f77b05 | 83 | |
27f4d280 PM |
84 | static struct rcu_state *rcu_state; |
85 | ||
bbad9379 PM |
86 | int rcu_scheduler_active __read_mostly; |
87 | EXPORT_SYMBOL_GPL(rcu_scheduler_active); | |
88 | ||
a26ac245 PM |
89 | /* |
90 | * Control variables for per-CPU and per-rcu_node kthreads. These | |
91 | * handle all flavors of RCU. | |
92 | */ | |
93 | static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task); | |
d71df90e | 94 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status); |
15ba0ba8 | 95 | DEFINE_PER_CPU(int, rcu_cpu_kthread_cpu); |
5ece5bab | 96 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops); |
a26ac245 | 97 | static DEFINE_PER_CPU(wait_queue_head_t, rcu_cpu_wq); |
d71df90e | 98 | DEFINE_PER_CPU(char, rcu_cpu_has_work); |
a26ac245 PM |
99 | static char rcu_kthreads_spawnable; |
100 | ||
0f962a5e | 101 | static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); |
27f4d280 | 102 | static void invoke_rcu_cpu_kthread(void); |
a26ac245 PM |
103 | |
104 | #define RCU_KTHREAD_PRIO 1 /* RT priority for per-CPU kthreads. */ | |
105 | ||
4a298656 PM |
106 | /* |
107 | * Track the rcutorture test sequence number and the update version | |
108 | * number within a given test. The rcutorture_testseq is incremented | |
109 | * on every rcutorture module load and unload, so has an odd value | |
110 | * when a test is running. The rcutorture_vernum is set to zero | |
111 | * when rcutorture starts and is incremented on each rcutorture update. | |
112 | * These variables enable correlating rcutorture output with the | |
113 | * RCU tracing information. | |
114 | */ | |
115 | unsigned long rcutorture_testseq; | |
116 | unsigned long rcutorture_vernum; | |
117 | ||
fc2219d4 PM |
118 | /* |
119 | * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s | |
120 | * permit this function to be invoked without holding the root rcu_node | |
121 | * structure's ->lock, but of course results can be subject to change. | |
122 | */ | |
123 | static int rcu_gp_in_progress(struct rcu_state *rsp) | |
124 | { | |
125 | return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum); | |
126 | } | |
127 | ||
b1f77b05 | 128 | /* |
d6714c22 | 129 | * Note a quiescent state. Because we do not need to know |
b1f77b05 | 130 | * how many quiescent states passed, just if there was at least |
d6714c22 | 131 | * one since the start of the grace period, this just sets a flag. |
b1f77b05 | 132 | */ |
d6714c22 | 133 | void rcu_sched_qs(int cpu) |
b1f77b05 | 134 | { |
25502a6c | 135 | struct rcu_data *rdp = &per_cpu(rcu_sched_data, cpu); |
f41d911f | 136 | |
c64ac3ce | 137 | rdp->passed_quiesc_completed = rdp->gpnum - 1; |
c3422bea PM |
138 | barrier(); |
139 | rdp->passed_quiesc = 1; | |
b1f77b05 IM |
140 | } |
141 | ||
d6714c22 | 142 | void rcu_bh_qs(int cpu) |
b1f77b05 | 143 | { |
25502a6c | 144 | struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu); |
f41d911f | 145 | |
c64ac3ce | 146 | rdp->passed_quiesc_completed = rdp->gpnum - 1; |
c3422bea PM |
147 | barrier(); |
148 | rdp->passed_quiesc = 1; | |
b1f77b05 | 149 | } |
64db4cff | 150 | |
25502a6c PM |
151 | /* |
152 | * Note a context switch. This is a quiescent state for RCU-sched, | |
153 | * and requires special handling for preemptible RCU. | |
154 | */ | |
155 | void rcu_note_context_switch(int cpu) | |
156 | { | |
157 | rcu_sched_qs(cpu); | |
158 | rcu_preempt_note_context_switch(cpu); | |
159 | } | |
160 | ||
64db4cff | 161 | #ifdef CONFIG_NO_HZ |
90a4d2c0 PM |
162 | DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { |
163 | .dynticks_nesting = 1, | |
e59fb312 | 164 | .dynticks = ATOMIC_INIT(1), |
90a4d2c0 | 165 | }; |
64db4cff PM |
166 | #endif /* #ifdef CONFIG_NO_HZ */ |
167 | ||
168 | static int blimit = 10; /* Maximum callbacks per softirq. */ | |
169 | static int qhimark = 10000; /* If this many pending, ignore blimit. */ | |
170 | static int qlowmark = 100; /* Once only this many pending, use blimit. */ | |
171 | ||
3d76c082 PM |
172 | module_param(blimit, int, 0); |
173 | module_param(qhimark, int, 0); | |
174 | module_param(qlowmark, int, 0); | |
175 | ||
a00e0d71 | 176 | int rcu_cpu_stall_suppress __read_mostly; |
f2e0dd70 | 177 | module_param(rcu_cpu_stall_suppress, int, 0644); |
742734ee | 178 | |
64db4cff | 179 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed); |
a157229c | 180 | static int rcu_pending(int cpu); |
64db4cff PM |
181 | |
182 | /* | |
d6714c22 | 183 | * Return the number of RCU-sched batches processed thus far for debug & stats. |
64db4cff | 184 | */ |
d6714c22 | 185 | long rcu_batches_completed_sched(void) |
64db4cff | 186 | { |
d6714c22 | 187 | return rcu_sched_state.completed; |
64db4cff | 188 | } |
d6714c22 | 189 | EXPORT_SYMBOL_GPL(rcu_batches_completed_sched); |
64db4cff PM |
190 | |
191 | /* | |
192 | * Return the number of RCU BH batches processed thus far for debug & stats. | |
193 | */ | |
194 | long rcu_batches_completed_bh(void) | |
195 | { | |
196 | return rcu_bh_state.completed; | |
197 | } | |
198 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); | |
199 | ||
bf66f18e PM |
200 | /* |
201 | * Force a quiescent state for RCU BH. | |
202 | */ | |
203 | void rcu_bh_force_quiescent_state(void) | |
204 | { | |
205 | force_quiescent_state(&rcu_bh_state, 0); | |
206 | } | |
207 | EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state); | |
208 | ||
4a298656 PM |
209 | /* |
210 | * Record the number of times rcutorture tests have been initiated and | |
211 | * terminated. This information allows the debugfs tracing stats to be | |
212 | * correlated to the rcutorture messages, even when the rcutorture module | |
213 | * is being repeatedly loaded and unloaded. In other words, we cannot | |
214 | * store this state in rcutorture itself. | |
215 | */ | |
216 | void rcutorture_record_test_transition(void) | |
217 | { | |
218 | rcutorture_testseq++; | |
219 | rcutorture_vernum = 0; | |
220 | } | |
221 | EXPORT_SYMBOL_GPL(rcutorture_record_test_transition); | |
222 | ||
223 | /* | |
224 | * Record the number of writer passes through the current rcutorture test. | |
225 | * This is also used to correlate debugfs tracing stats with the rcutorture | |
226 | * messages. | |
227 | */ | |
228 | void rcutorture_record_progress(unsigned long vernum) | |
229 | { | |
230 | rcutorture_vernum++; | |
231 | } | |
232 | EXPORT_SYMBOL_GPL(rcutorture_record_progress); | |
233 | ||
bf66f18e PM |
234 | /* |
235 | * Force a quiescent state for RCU-sched. | |
236 | */ | |
237 | void rcu_sched_force_quiescent_state(void) | |
238 | { | |
239 | force_quiescent_state(&rcu_sched_state, 0); | |
240 | } | |
241 | EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state); | |
242 | ||
64db4cff PM |
243 | /* |
244 | * Does the CPU have callbacks ready to be invoked? | |
245 | */ | |
246 | static int | |
247 | cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) | |
248 | { | |
249 | return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]; | |
250 | } | |
251 | ||
252 | /* | |
253 | * Does the current CPU require a yet-as-unscheduled grace period? | |
254 | */ | |
255 | static int | |
256 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) | |
257 | { | |
fc2219d4 | 258 | return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp); |
64db4cff PM |
259 | } |
260 | ||
261 | /* | |
262 | * Return the root node of the specified rcu_state structure. | |
263 | */ | |
264 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) | |
265 | { | |
266 | return &rsp->node[0]; | |
267 | } | |
268 | ||
269 | #ifdef CONFIG_SMP | |
270 | ||
271 | /* | |
272 | * If the specified CPU is offline, tell the caller that it is in | |
273 | * a quiescent state. Otherwise, whack it with a reschedule IPI. | |
274 | * Grace periods can end up waiting on an offline CPU when that | |
275 | * CPU is in the process of coming online -- it will be added to the | |
276 | * rcu_node bitmasks before it actually makes it online. The same thing | |
277 | * can happen while a CPU is in the process of coming online. Because this | |
278 | * race is quite rare, we check for it after detecting that the grace | |
279 | * period has been delayed rather than checking each and every CPU | |
280 | * each and every time we start a new grace period. | |
281 | */ | |
282 | static int rcu_implicit_offline_qs(struct rcu_data *rdp) | |
283 | { | |
284 | /* | |
285 | * If the CPU is offline, it is in a quiescent state. We can | |
286 | * trust its state not to change because interrupts are disabled. | |
287 | */ | |
288 | if (cpu_is_offline(rdp->cpu)) { | |
289 | rdp->offline_fqs++; | |
290 | return 1; | |
291 | } | |
292 | ||
6cc68793 PM |
293 | /* If preemptible RCU, no point in sending reschedule IPI. */ |
294 | if (rdp->preemptible) | |
f41d911f PM |
295 | return 0; |
296 | ||
64db4cff PM |
297 | /* The CPU is online, so send it a reschedule IPI. */ |
298 | if (rdp->cpu != smp_processor_id()) | |
299 | smp_send_reschedule(rdp->cpu); | |
300 | else | |
301 | set_need_resched(); | |
302 | rdp->resched_ipi++; | |
303 | return 0; | |
304 | } | |
305 | ||
306 | #endif /* #ifdef CONFIG_SMP */ | |
307 | ||
308 | #ifdef CONFIG_NO_HZ | |
64db4cff PM |
309 | |
310 | /** | |
311 | * rcu_enter_nohz - inform RCU that current CPU is entering nohz | |
312 | * | |
313 | * Enter nohz mode, in other words, -leave- the mode in which RCU | |
314 | * read-side critical sections can occur. (Though RCU read-side | |
315 | * critical sections can occur in irq handlers in nohz mode, a possibility | |
316 | * handled by rcu_irq_enter() and rcu_irq_exit()). | |
317 | */ | |
318 | void rcu_enter_nohz(void) | |
319 | { | |
320 | unsigned long flags; | |
321 | struct rcu_dynticks *rdtp; | |
322 | ||
64db4cff PM |
323 | local_irq_save(flags); |
324 | rdtp = &__get_cpu_var(rcu_dynticks); | |
e59fb312 PM |
325 | if (--rdtp->dynticks_nesting) { |
326 | local_irq_restore(flags); | |
327 | return; | |
328 | } | |
329 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ | |
330 | smp_mb__before_atomic_inc(); /* See above. */ | |
331 | atomic_inc(&rdtp->dynticks); | |
332 | smp_mb__after_atomic_inc(); /* Force ordering with next sojourn. */ | |
333 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); | |
64db4cff | 334 | local_irq_restore(flags); |
e59fb312 PM |
335 | |
336 | /* If the interrupt queued a callback, get out of dyntick mode. */ | |
337 | if (in_irq() && | |
338 | (__get_cpu_var(rcu_sched_data).nxtlist || | |
339 | __get_cpu_var(rcu_bh_data).nxtlist || | |
340 | rcu_preempt_needs_cpu(smp_processor_id()))) | |
341 | set_need_resched(); | |
64db4cff PM |
342 | } |
343 | ||
344 | /* | |
345 | * rcu_exit_nohz - inform RCU that current CPU is leaving nohz | |
346 | * | |
347 | * Exit nohz mode, in other words, -enter- the mode in which RCU | |
348 | * read-side critical sections normally occur. | |
349 | */ | |
350 | void rcu_exit_nohz(void) | |
351 | { | |
352 | unsigned long flags; | |
353 | struct rcu_dynticks *rdtp; | |
354 | ||
355 | local_irq_save(flags); | |
356 | rdtp = &__get_cpu_var(rcu_dynticks); | |
e59fb312 PM |
357 | if (rdtp->dynticks_nesting++) { |
358 | local_irq_restore(flags); | |
359 | return; | |
360 | } | |
361 | smp_mb__before_atomic_inc(); /* Force ordering w/previous sojourn. */ | |
362 | atomic_inc(&rdtp->dynticks); | |
363 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ | |
364 | smp_mb__after_atomic_inc(); /* See above. */ | |
365 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); | |
64db4cff | 366 | local_irq_restore(flags); |
64db4cff PM |
367 | } |
368 | ||
369 | /** | |
370 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
371 | * | |
372 | * If the CPU was idle with dynamic ticks active, and there is no | |
373 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
374 | * RCU grace-period handling know that the CPU is active. | |
375 | */ | |
376 | void rcu_nmi_enter(void) | |
377 | { | |
378 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
379 | ||
e59fb312 PM |
380 | if (rdtp->dynticks_nmi_nesting == 0 && |
381 | (atomic_read(&rdtp->dynticks) & 0x1)) | |
64db4cff | 382 | return; |
e59fb312 PM |
383 | rdtp->dynticks_nmi_nesting++; |
384 | smp_mb__before_atomic_inc(); /* Force delay from prior write. */ | |
385 | atomic_inc(&rdtp->dynticks); | |
386 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ | |
387 | smp_mb__after_atomic_inc(); /* See above. */ | |
388 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); | |
64db4cff PM |
389 | } |
390 | ||
391 | /** | |
392 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
393 | * | |
394 | * If the CPU was idle with dynamic ticks active, and there is no | |
395 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
396 | * RCU grace-period handling know that the CPU is no longer active. | |
397 | */ | |
398 | void rcu_nmi_exit(void) | |
399 | { | |
400 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
401 | ||
e59fb312 PM |
402 | if (rdtp->dynticks_nmi_nesting == 0 || |
403 | --rdtp->dynticks_nmi_nesting != 0) | |
64db4cff | 404 | return; |
e59fb312 PM |
405 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ |
406 | smp_mb__before_atomic_inc(); /* See above. */ | |
407 | atomic_inc(&rdtp->dynticks); | |
408 | smp_mb__after_atomic_inc(); /* Force delay to next write. */ | |
409 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); | |
64db4cff PM |
410 | } |
411 | ||
412 | /** | |
413 | * rcu_irq_enter - inform RCU of entry to hard irq context | |
414 | * | |
415 | * If the CPU was idle with dynamic ticks active, this updates the | |
416 | * rdtp->dynticks to let the RCU handling know that the CPU is active. | |
417 | */ | |
418 | void rcu_irq_enter(void) | |
419 | { | |
e59fb312 | 420 | rcu_exit_nohz(); |
64db4cff PM |
421 | } |
422 | ||
423 | /** | |
424 | * rcu_irq_exit - inform RCU of exit from hard irq context | |
425 | * | |
426 | * If the CPU was idle with dynamic ticks active, update the rdp->dynticks | |
427 | * to put let the RCU handling be aware that the CPU is going back to idle | |
428 | * with no ticks. | |
429 | */ | |
430 | void rcu_irq_exit(void) | |
431 | { | |
e59fb312 | 432 | rcu_enter_nohz(); |
64db4cff PM |
433 | } |
434 | ||
64db4cff PM |
435 | #ifdef CONFIG_SMP |
436 | ||
64db4cff PM |
437 | /* |
438 | * Snapshot the specified CPU's dynticks counter so that we can later | |
439 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 440 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff PM |
441 | */ |
442 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | |
443 | { | |
e59fb312 PM |
444 | rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks); |
445 | return 0; | |
64db4cff PM |
446 | } |
447 | ||
448 | /* | |
449 | * Return true if the specified CPU has passed through a quiescent | |
450 | * state by virtue of being in or having passed through an dynticks | |
451 | * idle state since the last call to dyntick_save_progress_counter() | |
452 | * for this same CPU. | |
453 | */ | |
454 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
455 | { | |
e59fb312 PM |
456 | unsigned long curr; |
457 | unsigned long snap; | |
64db4cff | 458 | |
e59fb312 PM |
459 | curr = (unsigned long)atomic_add_return(0, &rdp->dynticks->dynticks); |
460 | snap = (unsigned long)rdp->dynticks_snap; | |
64db4cff PM |
461 | |
462 | /* | |
463 | * If the CPU passed through or entered a dynticks idle phase with | |
464 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
465 | * already acknowledged the request to pass through a quiescent | |
466 | * state. Either way, that CPU cannot possibly be in an RCU | |
467 | * read-side critical section that started before the beginning | |
468 | * of the current RCU grace period. | |
469 | */ | |
e59fb312 | 470 | if ((curr & 0x1) == 0 || ULONG_CMP_GE(curr, snap + 2)) { |
64db4cff PM |
471 | rdp->dynticks_fqs++; |
472 | return 1; | |
473 | } | |
474 | ||
475 | /* Go check for the CPU being offline. */ | |
476 | return rcu_implicit_offline_qs(rdp); | |
477 | } | |
478 | ||
479 | #endif /* #ifdef CONFIG_SMP */ | |
480 | ||
481 | #else /* #ifdef CONFIG_NO_HZ */ | |
482 | ||
64db4cff PM |
483 | #ifdef CONFIG_SMP |
484 | ||
64db4cff PM |
485 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
486 | { | |
487 | return 0; | |
488 | } | |
489 | ||
490 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
491 | { | |
492 | return rcu_implicit_offline_qs(rdp); | |
493 | } | |
494 | ||
495 | #endif /* #ifdef CONFIG_SMP */ | |
496 | ||
497 | #endif /* #else #ifdef CONFIG_NO_HZ */ | |
498 | ||
742734ee | 499 | int rcu_cpu_stall_suppress __read_mostly; |
c68de209 | 500 | |
64db4cff PM |
501 | static void record_gp_stall_check_time(struct rcu_state *rsp) |
502 | { | |
503 | rsp->gp_start = jiffies; | |
504 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK; | |
505 | } | |
506 | ||
507 | static void print_other_cpu_stall(struct rcu_state *rsp) | |
508 | { | |
509 | int cpu; | |
510 | long delta; | |
511 | unsigned long flags; | |
512 | struct rcu_node *rnp = rcu_get_root(rsp); | |
64db4cff PM |
513 | |
514 | /* Only let one CPU complain about others per time interval. */ | |
515 | ||
1304afb2 | 516 | raw_spin_lock_irqsave(&rnp->lock, flags); |
64db4cff | 517 | delta = jiffies - rsp->jiffies_stall; |
fc2219d4 | 518 | if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) { |
1304afb2 | 519 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
520 | return; |
521 | } | |
522 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
a0b6c9a7 PM |
523 | |
524 | /* | |
525 | * Now rat on any tasks that got kicked up to the root rcu_node | |
526 | * due to CPU offlining. | |
527 | */ | |
528 | rcu_print_task_stall(rnp); | |
1304afb2 | 529 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 530 | |
8cdd32a9 PM |
531 | /* |
532 | * OK, time to rat on our buddy... | |
533 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
534 | * RCU CPU stall warnings. | |
535 | */ | |
4300aa64 PM |
536 | printk(KERN_ERR "INFO: %s detected stalls on CPUs/tasks: {", |
537 | rsp->name); | |
a0b6c9a7 | 538 | rcu_for_each_leaf_node(rsp, rnp) { |
3acd9eb3 | 539 | raw_spin_lock_irqsave(&rnp->lock, flags); |
f41d911f | 540 | rcu_print_task_stall(rnp); |
3acd9eb3 | 541 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
a0b6c9a7 | 542 | if (rnp->qsmask == 0) |
64db4cff | 543 | continue; |
a0b6c9a7 PM |
544 | for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++) |
545 | if (rnp->qsmask & (1UL << cpu)) | |
546 | printk(" %d", rnp->grplo + cpu); | |
64db4cff | 547 | } |
4300aa64 | 548 | printk("} (detected by %d, t=%ld jiffies)\n", |
64db4cff | 549 | smp_processor_id(), (long)(jiffies - rsp->gp_start)); |
c1dc0b9c IM |
550 | trigger_all_cpu_backtrace(); |
551 | ||
1ed509a2 PM |
552 | /* If so configured, complain about tasks blocking the grace period. */ |
553 | ||
554 | rcu_print_detail_task_stall(rsp); | |
555 | ||
64db4cff PM |
556 | force_quiescent_state(rsp, 0); /* Kick them all. */ |
557 | } | |
558 | ||
559 | static void print_cpu_stall(struct rcu_state *rsp) | |
560 | { | |
561 | unsigned long flags; | |
562 | struct rcu_node *rnp = rcu_get_root(rsp); | |
563 | ||
8cdd32a9 PM |
564 | /* |
565 | * OK, time to rat on ourselves... | |
566 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
567 | * RCU CPU stall warnings. | |
568 | */ | |
4300aa64 PM |
569 | printk(KERN_ERR "INFO: %s detected stall on CPU %d (t=%lu jiffies)\n", |
570 | rsp->name, smp_processor_id(), jiffies - rsp->gp_start); | |
c1dc0b9c IM |
571 | trigger_all_cpu_backtrace(); |
572 | ||
1304afb2 | 573 | raw_spin_lock_irqsave(&rnp->lock, flags); |
20133cfc | 574 | if (ULONG_CMP_GE(jiffies, rsp->jiffies_stall)) |
64db4cff PM |
575 | rsp->jiffies_stall = |
576 | jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
1304afb2 | 577 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
c1dc0b9c | 578 | |
64db4cff PM |
579 | set_need_resched(); /* kick ourselves to get things going. */ |
580 | } | |
581 | ||
582 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
583 | { | |
bad6e139 PM |
584 | unsigned long j; |
585 | unsigned long js; | |
64db4cff PM |
586 | struct rcu_node *rnp; |
587 | ||
742734ee | 588 | if (rcu_cpu_stall_suppress) |
c68de209 | 589 | return; |
bad6e139 PM |
590 | j = ACCESS_ONCE(jiffies); |
591 | js = ACCESS_ONCE(rsp->jiffies_stall); | |
64db4cff | 592 | rnp = rdp->mynode; |
bad6e139 | 593 | if ((ACCESS_ONCE(rnp->qsmask) & rdp->grpmask) && ULONG_CMP_GE(j, js)) { |
64db4cff PM |
594 | |
595 | /* We haven't checked in, so go dump stack. */ | |
596 | print_cpu_stall(rsp); | |
597 | ||
bad6e139 PM |
598 | } else if (rcu_gp_in_progress(rsp) && |
599 | ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) { | |
64db4cff | 600 | |
bad6e139 | 601 | /* They had a few time units to dump stack, so complain. */ |
64db4cff PM |
602 | print_other_cpu_stall(rsp); |
603 | } | |
604 | } | |
605 | ||
c68de209 PM |
606 | static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr) |
607 | { | |
742734ee | 608 | rcu_cpu_stall_suppress = 1; |
c68de209 PM |
609 | return NOTIFY_DONE; |
610 | } | |
611 | ||
53d84e00 PM |
612 | /** |
613 | * rcu_cpu_stall_reset - prevent further stall warnings in current grace period | |
614 | * | |
615 | * Set the stall-warning timeout way off into the future, thus preventing | |
616 | * any RCU CPU stall-warning messages from appearing in the current set of | |
617 | * RCU grace periods. | |
618 | * | |
619 | * The caller must disable hard irqs. | |
620 | */ | |
621 | void rcu_cpu_stall_reset(void) | |
622 | { | |
623 | rcu_sched_state.jiffies_stall = jiffies + ULONG_MAX / 2; | |
624 | rcu_bh_state.jiffies_stall = jiffies + ULONG_MAX / 2; | |
625 | rcu_preempt_stall_reset(); | |
626 | } | |
627 | ||
c68de209 PM |
628 | static struct notifier_block rcu_panic_block = { |
629 | .notifier_call = rcu_panic, | |
630 | }; | |
631 | ||
632 | static void __init check_cpu_stall_init(void) | |
633 | { | |
634 | atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block); | |
635 | } | |
636 | ||
64db4cff PM |
637 | /* |
638 | * Update CPU-local rcu_data state to record the newly noticed grace period. | |
639 | * This is used both when we started the grace period and when we notice | |
9160306e PM |
640 | * that someone else started the grace period. The caller must hold the |
641 | * ->lock of the leaf rcu_node structure corresponding to the current CPU, | |
642 | * and must have irqs disabled. | |
64db4cff | 643 | */ |
9160306e PM |
644 | static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) |
645 | { | |
646 | if (rdp->gpnum != rnp->gpnum) { | |
121dfc4b PM |
647 | /* |
648 | * If the current grace period is waiting for this CPU, | |
649 | * set up to detect a quiescent state, otherwise don't | |
650 | * go looking for one. | |
651 | */ | |
9160306e | 652 | rdp->gpnum = rnp->gpnum; |
121dfc4b PM |
653 | if (rnp->qsmask & rdp->grpmask) { |
654 | rdp->qs_pending = 1; | |
655 | rdp->passed_quiesc = 0; | |
656 | } else | |
657 | rdp->qs_pending = 0; | |
9160306e PM |
658 | } |
659 | } | |
660 | ||
64db4cff PM |
661 | static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp) |
662 | { | |
9160306e PM |
663 | unsigned long flags; |
664 | struct rcu_node *rnp; | |
665 | ||
666 | local_irq_save(flags); | |
667 | rnp = rdp->mynode; | |
668 | if (rdp->gpnum == ACCESS_ONCE(rnp->gpnum) || /* outside lock. */ | |
1304afb2 | 669 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
9160306e PM |
670 | local_irq_restore(flags); |
671 | return; | |
672 | } | |
673 | __note_new_gpnum(rsp, rnp, rdp); | |
1304afb2 | 674 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
675 | } |
676 | ||
677 | /* | |
678 | * Did someone else start a new RCU grace period start since we last | |
679 | * checked? Update local state appropriately if so. Must be called | |
680 | * on the CPU corresponding to rdp. | |
681 | */ | |
682 | static int | |
683 | check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp) | |
684 | { | |
685 | unsigned long flags; | |
686 | int ret = 0; | |
687 | ||
688 | local_irq_save(flags); | |
689 | if (rdp->gpnum != rsp->gpnum) { | |
690 | note_new_gpnum(rsp, rdp); | |
691 | ret = 1; | |
692 | } | |
693 | local_irq_restore(flags); | |
694 | return ret; | |
695 | } | |
696 | ||
d09b62df PM |
697 | /* |
698 | * Advance this CPU's callbacks, but only if the current grace period | |
699 | * has ended. This may be called only from the CPU to whom the rdp | |
700 | * belongs. In addition, the corresponding leaf rcu_node structure's | |
701 | * ->lock must be held by the caller, with irqs disabled. | |
702 | */ | |
703 | static void | |
704 | __rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
705 | { | |
706 | /* Did another grace period end? */ | |
707 | if (rdp->completed != rnp->completed) { | |
708 | ||
709 | /* Advance callbacks. No harm if list empty. */ | |
710 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL]; | |
711 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL]; | |
712 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
713 | ||
714 | /* Remember that we saw this grace-period completion. */ | |
715 | rdp->completed = rnp->completed; | |
20377f32 | 716 | |
5ff8e6f0 FW |
717 | /* |
718 | * If we were in an extended quiescent state, we may have | |
121dfc4b | 719 | * missed some grace periods that others CPUs handled on |
5ff8e6f0 | 720 | * our behalf. Catch up with this state to avoid noting |
121dfc4b PM |
721 | * spurious new grace periods. If another grace period |
722 | * has started, then rnp->gpnum will have advanced, so | |
723 | * we will detect this later on. | |
5ff8e6f0 | 724 | */ |
121dfc4b | 725 | if (ULONG_CMP_LT(rdp->gpnum, rdp->completed)) |
5ff8e6f0 FW |
726 | rdp->gpnum = rdp->completed; |
727 | ||
20377f32 | 728 | /* |
121dfc4b PM |
729 | * If RCU does not need a quiescent state from this CPU, |
730 | * then make sure that this CPU doesn't go looking for one. | |
20377f32 | 731 | */ |
121dfc4b | 732 | if ((rnp->qsmask & rdp->grpmask) == 0) |
20377f32 | 733 | rdp->qs_pending = 0; |
d09b62df PM |
734 | } |
735 | } | |
736 | ||
737 | /* | |
738 | * Advance this CPU's callbacks, but only if the current grace period | |
739 | * has ended. This may be called only from the CPU to whom the rdp | |
740 | * belongs. | |
741 | */ | |
742 | static void | |
743 | rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp) | |
744 | { | |
745 | unsigned long flags; | |
746 | struct rcu_node *rnp; | |
747 | ||
748 | local_irq_save(flags); | |
749 | rnp = rdp->mynode; | |
750 | if (rdp->completed == ACCESS_ONCE(rnp->completed) || /* outside lock. */ | |
1304afb2 | 751 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
d09b62df PM |
752 | local_irq_restore(flags); |
753 | return; | |
754 | } | |
755 | __rcu_process_gp_end(rsp, rnp, rdp); | |
1304afb2 | 756 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
d09b62df PM |
757 | } |
758 | ||
759 | /* | |
760 | * Do per-CPU grace-period initialization for running CPU. The caller | |
761 | * must hold the lock of the leaf rcu_node structure corresponding to | |
762 | * this CPU. | |
763 | */ | |
764 | static void | |
765 | rcu_start_gp_per_cpu(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
766 | { | |
767 | /* Prior grace period ended, so advance callbacks for current CPU. */ | |
768 | __rcu_process_gp_end(rsp, rnp, rdp); | |
769 | ||
770 | /* | |
771 | * Because this CPU just now started the new grace period, we know | |
772 | * that all of its callbacks will be covered by this upcoming grace | |
773 | * period, even the ones that were registered arbitrarily recently. | |
774 | * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL. | |
775 | * | |
776 | * Other CPUs cannot be sure exactly when the grace period started. | |
777 | * Therefore, their recently registered callbacks must pass through | |
778 | * an additional RCU_NEXT_READY stage, so that they will be handled | |
779 | * by the next RCU grace period. | |
780 | */ | |
781 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
782 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
9160306e PM |
783 | |
784 | /* Set state so that this CPU will detect the next quiescent state. */ | |
785 | __note_new_gpnum(rsp, rnp, rdp); | |
d09b62df PM |
786 | } |
787 | ||
64db4cff PM |
788 | /* |
789 | * Start a new RCU grace period if warranted, re-initializing the hierarchy | |
790 | * in preparation for detecting the next grace period. The caller must hold | |
791 | * the root node's ->lock, which is released before return. Hard irqs must | |
792 | * be disabled. | |
793 | */ | |
794 | static void | |
795 | rcu_start_gp(struct rcu_state *rsp, unsigned long flags) | |
796 | __releases(rcu_get_root(rsp)->lock) | |
797 | { | |
394f99a9 | 798 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
64db4cff | 799 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff | 800 | |
07079d53 | 801 | if (!cpu_needs_another_gp(rsp, rdp) || rsp->fqs_active) { |
46a1e34e PM |
802 | if (cpu_needs_another_gp(rsp, rdp)) |
803 | rsp->fqs_need_gp = 1; | |
b32e9eb6 | 804 | if (rnp->completed == rsp->completed) { |
1304afb2 | 805 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
b32e9eb6 PM |
806 | return; |
807 | } | |
1304afb2 | 808 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
b32e9eb6 PM |
809 | |
810 | /* | |
811 | * Propagate new ->completed value to rcu_node structures | |
812 | * so that other CPUs don't have to wait until the start | |
813 | * of the next grace period to process their callbacks. | |
814 | */ | |
815 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
1304afb2 | 816 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
b32e9eb6 | 817 | rnp->completed = rsp->completed; |
1304afb2 | 818 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
b32e9eb6 PM |
819 | } |
820 | local_irq_restore(flags); | |
64db4cff PM |
821 | return; |
822 | } | |
823 | ||
824 | /* Advance to a new grace period and initialize state. */ | |
825 | rsp->gpnum++; | |
c3422bea | 826 | WARN_ON_ONCE(rsp->signaled == RCU_GP_INIT); |
64db4cff PM |
827 | rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */ |
828 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; | |
64db4cff | 829 | record_gp_stall_check_time(rsp); |
64db4cff | 830 | |
64db4cff PM |
831 | /* Special-case the common single-level case. */ |
832 | if (NUM_RCU_NODES == 1) { | |
b0e165c0 | 833 | rcu_preempt_check_blocked_tasks(rnp); |
28ecd580 | 834 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 835 | rnp->gpnum = rsp->gpnum; |
d09b62df | 836 | rnp->completed = rsp->completed; |
c12172c0 | 837 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */ |
d09b62df | 838 | rcu_start_gp_per_cpu(rsp, rnp, rdp); |
27f4d280 | 839 | rcu_preempt_boost_start_gp(rnp); |
1304afb2 | 840 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
841 | return; |
842 | } | |
843 | ||
1304afb2 | 844 | raw_spin_unlock(&rnp->lock); /* leave irqs disabled. */ |
64db4cff PM |
845 | |
846 | ||
847 | /* Exclude any concurrent CPU-hotplug operations. */ | |
1304afb2 | 848 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
64db4cff PM |
849 | |
850 | /* | |
b835db1f PM |
851 | * Set the quiescent-state-needed bits in all the rcu_node |
852 | * structures for all currently online CPUs in breadth-first | |
853 | * order, starting from the root rcu_node structure. This | |
854 | * operation relies on the layout of the hierarchy within the | |
855 | * rsp->node[] array. Note that other CPUs will access only | |
856 | * the leaves of the hierarchy, which still indicate that no | |
857 | * grace period is in progress, at least until the corresponding | |
858 | * leaf node has been initialized. In addition, we have excluded | |
859 | * CPU-hotplug operations. | |
64db4cff PM |
860 | * |
861 | * Note that the grace period cannot complete until we finish | |
862 | * the initialization process, as there will be at least one | |
863 | * qsmask bit set in the root node until that time, namely the | |
b835db1f PM |
864 | * one corresponding to this CPU, due to the fact that we have |
865 | * irqs disabled. | |
64db4cff | 866 | */ |
a0b6c9a7 | 867 | rcu_for_each_node_breadth_first(rsp, rnp) { |
1304afb2 | 868 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
b0e165c0 | 869 | rcu_preempt_check_blocked_tasks(rnp); |
49e29126 | 870 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 871 | rnp->gpnum = rsp->gpnum; |
d09b62df PM |
872 | rnp->completed = rsp->completed; |
873 | if (rnp == rdp->mynode) | |
874 | rcu_start_gp_per_cpu(rsp, rnp, rdp); | |
27f4d280 | 875 | rcu_preempt_boost_start_gp(rnp); |
1304afb2 | 876 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
877 | } |
878 | ||
83f5b01f | 879 | rnp = rcu_get_root(rsp); |
1304afb2 | 880 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff | 881 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */ |
1304afb2 PM |
882 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
883 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); | |
64db4cff PM |
884 | } |
885 | ||
f41d911f | 886 | /* |
d3f6bad3 PM |
887 | * Report a full set of quiescent states to the specified rcu_state |
888 | * data structure. This involves cleaning up after the prior grace | |
889 | * period and letting rcu_start_gp() start up the next grace period | |
890 | * if one is needed. Note that the caller must hold rnp->lock, as | |
891 | * required by rcu_start_gp(), which will release it. | |
f41d911f | 892 | */ |
d3f6bad3 | 893 | static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags) |
fc2219d4 | 894 | __releases(rcu_get_root(rsp)->lock) |
f41d911f | 895 | { |
15ba0ba8 PM |
896 | unsigned long gp_duration; |
897 | ||
fc2219d4 | 898 | WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); |
e59fb312 PM |
899 | |
900 | /* | |
901 | * Ensure that all grace-period and pre-grace-period activity | |
902 | * is seen before the assignment to rsp->completed. | |
903 | */ | |
904 | smp_mb(); /* See above block comment. */ | |
15ba0ba8 PM |
905 | gp_duration = jiffies - rsp->gp_start; |
906 | if (gp_duration > rsp->gp_max) | |
907 | rsp->gp_max = gp_duration; | |
f41d911f | 908 | rsp->completed = rsp->gpnum; |
83f5b01f | 909 | rsp->signaled = RCU_GP_IDLE; |
f41d911f PM |
910 | rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */ |
911 | } | |
912 | ||
64db4cff | 913 | /* |
d3f6bad3 PM |
914 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
915 | * Allows quiescent states for a group of CPUs to be reported at one go | |
916 | * to the specified rcu_node structure, though all the CPUs in the group | |
917 | * must be represented by the same rcu_node structure (which need not be | |
918 | * a leaf rcu_node structure, though it often will be). That structure's | |
919 | * lock must be held upon entry, and it is released before return. | |
64db4cff PM |
920 | */ |
921 | static void | |
d3f6bad3 PM |
922 | rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp, |
923 | struct rcu_node *rnp, unsigned long flags) | |
64db4cff PM |
924 | __releases(rnp->lock) |
925 | { | |
28ecd580 PM |
926 | struct rcu_node *rnp_c; |
927 | ||
64db4cff PM |
928 | /* Walk up the rcu_node hierarchy. */ |
929 | for (;;) { | |
930 | if (!(rnp->qsmask & mask)) { | |
931 | ||
932 | /* Our bit has already been cleared, so done. */ | |
1304afb2 | 933 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
934 | return; |
935 | } | |
936 | rnp->qsmask &= ~mask; | |
27f4d280 | 937 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
64db4cff PM |
938 | |
939 | /* Other bits still set at this level, so done. */ | |
1304afb2 | 940 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
941 | return; |
942 | } | |
943 | mask = rnp->grpmask; | |
944 | if (rnp->parent == NULL) { | |
945 | ||
946 | /* No more levels. Exit loop holding root lock. */ | |
947 | ||
948 | break; | |
949 | } | |
1304afb2 | 950 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
28ecd580 | 951 | rnp_c = rnp; |
64db4cff | 952 | rnp = rnp->parent; |
1304afb2 | 953 | raw_spin_lock_irqsave(&rnp->lock, flags); |
28ecd580 | 954 | WARN_ON_ONCE(rnp_c->qsmask); |
64db4cff PM |
955 | } |
956 | ||
957 | /* | |
958 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 959 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 960 | * to clean up and start the next grace period if one is needed. |
64db4cff | 961 | */ |
d3f6bad3 | 962 | rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */ |
64db4cff PM |
963 | } |
964 | ||
965 | /* | |
d3f6bad3 PM |
966 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
967 | * structure. This must be either called from the specified CPU, or | |
968 | * called when the specified CPU is known to be offline (and when it is | |
969 | * also known that no other CPU is concurrently trying to help the offline | |
970 | * CPU). The lastcomp argument is used to make sure we are still in the | |
971 | * grace period of interest. We don't want to end the current grace period | |
972 | * based on quiescent states detected in an earlier grace period! | |
64db4cff PM |
973 | */ |
974 | static void | |
d3f6bad3 | 975 | rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp) |
64db4cff PM |
976 | { |
977 | unsigned long flags; | |
978 | unsigned long mask; | |
979 | struct rcu_node *rnp; | |
980 | ||
981 | rnp = rdp->mynode; | |
1304afb2 | 982 | raw_spin_lock_irqsave(&rnp->lock, flags); |
560d4bc0 | 983 | if (lastcomp != rnp->completed) { |
64db4cff PM |
984 | |
985 | /* | |
986 | * Someone beat us to it for this grace period, so leave. | |
987 | * The race with GP start is resolved by the fact that we | |
988 | * hold the leaf rcu_node lock, so that the per-CPU bits | |
989 | * cannot yet be initialized -- so we would simply find our | |
d3f6bad3 PM |
990 | * CPU's bit already cleared in rcu_report_qs_rnp() if this |
991 | * race occurred. | |
64db4cff PM |
992 | */ |
993 | rdp->passed_quiesc = 0; /* try again later! */ | |
1304afb2 | 994 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
995 | return; |
996 | } | |
997 | mask = rdp->grpmask; | |
998 | if ((rnp->qsmask & mask) == 0) { | |
1304afb2 | 999 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1000 | } else { |
1001 | rdp->qs_pending = 0; | |
1002 | ||
1003 | /* | |
1004 | * This GP can't end until cpu checks in, so all of our | |
1005 | * callbacks can be processed during the next GP. | |
1006 | */ | |
64db4cff PM |
1007 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
1008 | ||
d3f6bad3 | 1009 | rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */ |
64db4cff PM |
1010 | } |
1011 | } | |
1012 | ||
1013 | /* | |
1014 | * Check to see if there is a new grace period of which this CPU | |
1015 | * is not yet aware, and if so, set up local rcu_data state for it. | |
1016 | * Otherwise, see if this CPU has just passed through its first | |
1017 | * quiescent state for this grace period, and record that fact if so. | |
1018 | */ | |
1019 | static void | |
1020 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | |
1021 | { | |
1022 | /* If there is now a new grace period, record and return. */ | |
1023 | if (check_for_new_grace_period(rsp, rdp)) | |
1024 | return; | |
1025 | ||
1026 | /* | |
1027 | * Does this CPU still need to do its part for current grace period? | |
1028 | * If no, return and let the other CPUs do their part as well. | |
1029 | */ | |
1030 | if (!rdp->qs_pending) | |
1031 | return; | |
1032 | ||
1033 | /* | |
1034 | * Was there a quiescent state since the beginning of the grace | |
1035 | * period? If no, then exit and wait for the next call. | |
1036 | */ | |
1037 | if (!rdp->passed_quiesc) | |
1038 | return; | |
1039 | ||
d3f6bad3 PM |
1040 | /* |
1041 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
1042 | * judge of that). | |
1043 | */ | |
1044 | rcu_report_qs_rdp(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed); | |
64db4cff PM |
1045 | } |
1046 | ||
1047 | #ifdef CONFIG_HOTPLUG_CPU | |
1048 | ||
e74f4c45 | 1049 | /* |
29494be7 LJ |
1050 | * Move a dying CPU's RCU callbacks to online CPU's callback list. |
1051 | * Synchronization is not required because this function executes | |
1052 | * in stop_machine() context. | |
e74f4c45 | 1053 | */ |
29494be7 | 1054 | static void rcu_send_cbs_to_online(struct rcu_state *rsp) |
e74f4c45 PM |
1055 | { |
1056 | int i; | |
29494be7 LJ |
1057 | /* current DYING CPU is cleared in the cpu_online_mask */ |
1058 | int receive_cpu = cpumask_any(cpu_online_mask); | |
394f99a9 | 1059 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
29494be7 | 1060 | struct rcu_data *receive_rdp = per_cpu_ptr(rsp->rda, receive_cpu); |
e74f4c45 PM |
1061 | |
1062 | if (rdp->nxtlist == NULL) | |
1063 | return; /* irqs disabled, so comparison is stable. */ | |
29494be7 LJ |
1064 | |
1065 | *receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist; | |
1066 | receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
1067 | receive_rdp->qlen += rdp->qlen; | |
1068 | receive_rdp->n_cbs_adopted += rdp->qlen; | |
1069 | rdp->n_cbs_orphaned += rdp->qlen; | |
1070 | ||
e74f4c45 PM |
1071 | rdp->nxtlist = NULL; |
1072 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
1073 | rdp->nxttail[i] = &rdp->nxtlist; | |
e74f4c45 | 1074 | rdp->qlen = 0; |
e74f4c45 PM |
1075 | } |
1076 | ||
64db4cff PM |
1077 | /* |
1078 | * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy | |
1079 | * and move all callbacks from the outgoing CPU to the current one. | |
a26ac245 PM |
1080 | * There can only be one CPU hotplug operation at a time, so no other |
1081 | * CPU can be attempting to update rcu_cpu_kthread_task. | |
64db4cff PM |
1082 | */ |
1083 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp) | |
1084 | { | |
64db4cff | 1085 | unsigned long flags; |
64db4cff | 1086 | unsigned long mask; |
d9a3da06 | 1087 | int need_report = 0; |
394f99a9 | 1088 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff | 1089 | struct rcu_node *rnp; |
a26ac245 PM |
1090 | struct task_struct *t; |
1091 | ||
1092 | /* Stop the CPU's kthread. */ | |
1093 | t = per_cpu(rcu_cpu_kthread_task, cpu); | |
1094 | if (t != NULL) { | |
1095 | per_cpu(rcu_cpu_kthread_task, cpu) = NULL; | |
1096 | kthread_stop(t); | |
1097 | } | |
64db4cff PM |
1098 | |
1099 | /* Exclude any attempts to start a new grace period. */ | |
1304afb2 | 1100 | raw_spin_lock_irqsave(&rsp->onofflock, flags); |
64db4cff PM |
1101 | |
1102 | /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */ | |
28ecd580 | 1103 | rnp = rdp->mynode; /* this is the outgoing CPU's rnp. */ |
64db4cff PM |
1104 | mask = rdp->grpmask; /* rnp->grplo is constant. */ |
1105 | do { | |
1304afb2 | 1106 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
1107 | rnp->qsmaskinit &= ~mask; |
1108 | if (rnp->qsmaskinit != 0) { | |
b668c9cf | 1109 | if (rnp != rdp->mynode) |
1304afb2 | 1110 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
1111 | break; |
1112 | } | |
b668c9cf | 1113 | if (rnp == rdp->mynode) |
d9a3da06 | 1114 | need_report = rcu_preempt_offline_tasks(rsp, rnp, rdp); |
b668c9cf | 1115 | else |
1304afb2 | 1116 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff | 1117 | mask = rnp->grpmask; |
64db4cff PM |
1118 | rnp = rnp->parent; |
1119 | } while (rnp != NULL); | |
64db4cff | 1120 | |
b668c9cf PM |
1121 | /* |
1122 | * We still hold the leaf rcu_node structure lock here, and | |
1123 | * irqs are still disabled. The reason for this subterfuge is | |
d3f6bad3 PM |
1124 | * because invoking rcu_report_unblock_qs_rnp() with ->onofflock |
1125 | * held leads to deadlock. | |
b668c9cf | 1126 | */ |
1304afb2 | 1127 | raw_spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ |
b668c9cf | 1128 | rnp = rdp->mynode; |
d9a3da06 | 1129 | if (need_report & RCU_OFL_TASKS_NORM_GP) |
d3f6bad3 | 1130 | rcu_report_unblock_qs_rnp(rnp, flags); |
b668c9cf | 1131 | else |
1304afb2 | 1132 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
d9a3da06 PM |
1133 | if (need_report & RCU_OFL_TASKS_EXP_GP) |
1134 | rcu_report_exp_rnp(rsp, rnp); | |
a26ac245 PM |
1135 | |
1136 | /* | |
1137 | * If there are no more online CPUs for this rcu_node structure, | |
1138 | * kill the rcu_node structure's kthread. Otherwise, adjust its | |
1139 | * affinity. | |
1140 | */ | |
1141 | t = rnp->node_kthread_task; | |
1142 | if (t != NULL && | |
1143 | rnp->qsmaskinit == 0) { | |
27f4d280 | 1144 | raw_spin_lock_irqsave(&rnp->lock, flags); |
a26ac245 | 1145 | rnp->node_kthread_task = NULL; |
27f4d280 PM |
1146 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1147 | kthread_stop(t); | |
1148 | rcu_stop_boost_kthread(rnp); | |
a26ac245 | 1149 | } else |
0f962a5e | 1150 | rcu_node_kthread_setaffinity(rnp, -1); |
64db4cff PM |
1151 | } |
1152 | ||
1153 | /* | |
1154 | * Remove the specified CPU from the RCU hierarchy and move any pending | |
1155 | * callbacks that it might have to the current CPU. This code assumes | |
1156 | * that at least one CPU in the system will remain running at all times. | |
1157 | * Any attempt to offline -all- CPUs is likely to strand RCU callbacks. | |
1158 | */ | |
1159 | static void rcu_offline_cpu(int cpu) | |
1160 | { | |
d6714c22 | 1161 | __rcu_offline_cpu(cpu, &rcu_sched_state); |
64db4cff | 1162 | __rcu_offline_cpu(cpu, &rcu_bh_state); |
33f76148 | 1163 | rcu_preempt_offline_cpu(cpu); |
64db4cff PM |
1164 | } |
1165 | ||
1166 | #else /* #ifdef CONFIG_HOTPLUG_CPU */ | |
1167 | ||
29494be7 | 1168 | static void rcu_send_cbs_to_online(struct rcu_state *rsp) |
e74f4c45 PM |
1169 | { |
1170 | } | |
1171 | ||
64db4cff PM |
1172 | static void rcu_offline_cpu(int cpu) |
1173 | { | |
1174 | } | |
1175 | ||
1176 | #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */ | |
1177 | ||
1178 | /* | |
1179 | * Invoke any RCU callbacks that have made it to the end of their grace | |
1180 | * period. Thottle as specified by rdp->blimit. | |
1181 | */ | |
37c72e56 | 1182 | static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
1183 | { |
1184 | unsigned long flags; | |
1185 | struct rcu_head *next, *list, **tail; | |
1186 | int count; | |
1187 | ||
1188 | /* If no callbacks are ready, just return.*/ | |
1189 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) | |
1190 | return; | |
1191 | ||
1192 | /* | |
1193 | * Extract the list of ready callbacks, disabling to prevent | |
1194 | * races with call_rcu() from interrupt handlers. | |
1195 | */ | |
1196 | local_irq_save(flags); | |
1197 | list = rdp->nxtlist; | |
1198 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; | |
1199 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | |
1200 | tail = rdp->nxttail[RCU_DONE_TAIL]; | |
1201 | for (count = RCU_NEXT_SIZE - 1; count >= 0; count--) | |
1202 | if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL]) | |
1203 | rdp->nxttail[count] = &rdp->nxtlist; | |
1204 | local_irq_restore(flags); | |
1205 | ||
1206 | /* Invoke callbacks. */ | |
1207 | count = 0; | |
1208 | while (list) { | |
1209 | next = list->next; | |
1210 | prefetch(next); | |
551d55a9 | 1211 | debug_rcu_head_unqueue(list); |
9ab1544e | 1212 | __rcu_reclaim(list); |
64db4cff PM |
1213 | list = next; |
1214 | if (++count >= rdp->blimit) | |
1215 | break; | |
1216 | } | |
1217 | ||
1218 | local_irq_save(flags); | |
1219 | ||
1220 | /* Update count, and requeue any remaining callbacks. */ | |
1221 | rdp->qlen -= count; | |
269dcc1c | 1222 | rdp->n_cbs_invoked += count; |
64db4cff PM |
1223 | if (list != NULL) { |
1224 | *tail = rdp->nxtlist; | |
1225 | rdp->nxtlist = list; | |
1226 | for (count = 0; count < RCU_NEXT_SIZE; count++) | |
1227 | if (&rdp->nxtlist == rdp->nxttail[count]) | |
1228 | rdp->nxttail[count] = tail; | |
1229 | else | |
1230 | break; | |
1231 | } | |
1232 | ||
1233 | /* Reinstate batch limit if we have worked down the excess. */ | |
1234 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) | |
1235 | rdp->blimit = blimit; | |
1236 | ||
37c72e56 PM |
1237 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
1238 | if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) { | |
1239 | rdp->qlen_last_fqs_check = 0; | |
1240 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1241 | } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark) | |
1242 | rdp->qlen_last_fqs_check = rdp->qlen; | |
1243 | ||
64db4cff PM |
1244 | local_irq_restore(flags); |
1245 | ||
1246 | /* Re-raise the RCU softirq if there are callbacks remaining. */ | |
1247 | if (cpu_has_callbacks_ready_to_invoke(rdp)) | |
27f4d280 | 1248 | invoke_rcu_cpu_kthread(); |
64db4cff PM |
1249 | } |
1250 | ||
1251 | /* | |
1252 | * Check to see if this CPU is in a non-context-switch quiescent state | |
1253 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
1254 | * Also schedule the RCU softirq handler. | |
1255 | * | |
1256 | * This function must be called with hardirqs disabled. It is normally | |
1257 | * invoked from the scheduling-clock interrupt. If rcu_pending returns | |
1258 | * false, there is no point in invoking rcu_check_callbacks(). | |
1259 | */ | |
1260 | void rcu_check_callbacks(int cpu, int user) | |
1261 | { | |
1262 | if (user || | |
a6826048 PM |
1263 | (idle_cpu(cpu) && rcu_scheduler_active && |
1264 | !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) { | |
64db4cff PM |
1265 | |
1266 | /* | |
1267 | * Get here if this CPU took its interrupt from user | |
1268 | * mode or from the idle loop, and if this is not a | |
1269 | * nested interrupt. In this case, the CPU is in | |
d6714c22 | 1270 | * a quiescent state, so note it. |
64db4cff PM |
1271 | * |
1272 | * No memory barrier is required here because both | |
d6714c22 PM |
1273 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
1274 | * variables that other CPUs neither access nor modify, | |
1275 | * at least not while the corresponding CPU is online. | |
64db4cff PM |
1276 | */ |
1277 | ||
d6714c22 PM |
1278 | rcu_sched_qs(cpu); |
1279 | rcu_bh_qs(cpu); | |
64db4cff PM |
1280 | |
1281 | } else if (!in_softirq()) { | |
1282 | ||
1283 | /* | |
1284 | * Get here if this CPU did not take its interrupt from | |
1285 | * softirq, in other words, if it is not interrupting | |
1286 | * a rcu_bh read-side critical section. This is an _bh | |
d6714c22 | 1287 | * critical section, so note it. |
64db4cff PM |
1288 | */ |
1289 | ||
d6714c22 | 1290 | rcu_bh_qs(cpu); |
64db4cff | 1291 | } |
f41d911f | 1292 | rcu_preempt_check_callbacks(cpu); |
d21670ac | 1293 | if (rcu_pending(cpu)) |
27f4d280 | 1294 | invoke_rcu_cpu_kthread(); |
64db4cff PM |
1295 | } |
1296 | ||
1297 | #ifdef CONFIG_SMP | |
1298 | ||
1299 | /* | |
1300 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
1301 | * have not yet encountered a quiescent state, using the function specified. | |
27f4d280 PM |
1302 | * Also initiate boosting for any threads blocked on the root rcu_node. |
1303 | * | |
ee47eb9f | 1304 | * The caller must have suppressed start of new grace periods. |
64db4cff | 1305 | */ |
45f014c5 | 1306 | static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *)) |
64db4cff PM |
1307 | { |
1308 | unsigned long bit; | |
1309 | int cpu; | |
1310 | unsigned long flags; | |
1311 | unsigned long mask; | |
a0b6c9a7 | 1312 | struct rcu_node *rnp; |
64db4cff | 1313 | |
a0b6c9a7 | 1314 | rcu_for_each_leaf_node(rsp, rnp) { |
64db4cff | 1315 | mask = 0; |
1304afb2 | 1316 | raw_spin_lock_irqsave(&rnp->lock, flags); |
ee47eb9f | 1317 | if (!rcu_gp_in_progress(rsp)) { |
1304afb2 | 1318 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
0f10dc82 | 1319 | return; |
64db4cff | 1320 | } |
a0b6c9a7 | 1321 | if (rnp->qsmask == 0) { |
27f4d280 | 1322 | rcu_initiate_boost(rnp); |
1304afb2 | 1323 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1324 | continue; |
1325 | } | |
a0b6c9a7 | 1326 | cpu = rnp->grplo; |
64db4cff | 1327 | bit = 1; |
a0b6c9a7 | 1328 | for (; cpu <= rnp->grphi; cpu++, bit <<= 1) { |
394f99a9 LJ |
1329 | if ((rnp->qsmask & bit) != 0 && |
1330 | f(per_cpu_ptr(rsp->rda, cpu))) | |
64db4cff PM |
1331 | mask |= bit; |
1332 | } | |
45f014c5 | 1333 | if (mask != 0) { |
64db4cff | 1334 | |
d3f6bad3 PM |
1335 | /* rcu_report_qs_rnp() releases rnp->lock. */ |
1336 | rcu_report_qs_rnp(mask, rsp, rnp, flags); | |
64db4cff PM |
1337 | continue; |
1338 | } | |
1304afb2 | 1339 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 1340 | } |
27f4d280 PM |
1341 | rnp = rcu_get_root(rsp); |
1342 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
1343 | if (rnp->qsmask == 0) | |
1344 | rcu_initiate_boost(rnp); | |
1345 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
64db4cff PM |
1346 | } |
1347 | ||
1348 | /* | |
1349 | * Force quiescent states on reluctant CPUs, and also detect which | |
1350 | * CPUs are in dyntick-idle mode. | |
1351 | */ | |
1352 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1353 | { | |
1354 | unsigned long flags; | |
64db4cff | 1355 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff | 1356 | |
fc2219d4 | 1357 | if (!rcu_gp_in_progress(rsp)) |
64db4cff | 1358 | return; /* No grace period in progress, nothing to force. */ |
1304afb2 | 1359 | if (!raw_spin_trylock_irqsave(&rsp->fqslock, flags)) { |
64db4cff PM |
1360 | rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ |
1361 | return; /* Someone else is already on the job. */ | |
1362 | } | |
20133cfc | 1363 | if (relaxed && ULONG_CMP_GE(rsp->jiffies_force_qs, jiffies)) |
f96e9232 | 1364 | goto unlock_fqs_ret; /* no emergency and done recently. */ |
64db4cff | 1365 | rsp->n_force_qs++; |
1304afb2 | 1366 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
64db4cff | 1367 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
560d4bc0 | 1368 | if(!rcu_gp_in_progress(rsp)) { |
64db4cff | 1369 | rsp->n_force_qs_ngp++; |
1304afb2 | 1370 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1371 | goto unlock_fqs_ret; /* no GP in progress, time updated. */ |
64db4cff | 1372 | } |
07079d53 | 1373 | rsp->fqs_active = 1; |
f3a8b5c6 | 1374 | switch (rsp->signaled) { |
83f5b01f | 1375 | case RCU_GP_IDLE: |
64db4cff PM |
1376 | case RCU_GP_INIT: |
1377 | ||
83f5b01f | 1378 | break; /* grace period idle or initializing, ignore. */ |
64db4cff PM |
1379 | |
1380 | case RCU_SAVE_DYNTICK: | |
64db4cff PM |
1381 | if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK) |
1382 | break; /* So gcc recognizes the dead code. */ | |
1383 | ||
f261414f LJ |
1384 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
1385 | ||
64db4cff | 1386 | /* Record dyntick-idle state. */ |
45f014c5 | 1387 | force_qs_rnp(rsp, dyntick_save_progress_counter); |
1304afb2 | 1388 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
ee47eb9f | 1389 | if (rcu_gp_in_progress(rsp)) |
64db4cff | 1390 | rsp->signaled = RCU_FORCE_QS; |
ee47eb9f | 1391 | break; |
64db4cff PM |
1392 | |
1393 | case RCU_FORCE_QS: | |
1394 | ||
1395 | /* Check dyntick-idle state, send IPI to laggarts. */ | |
1304afb2 | 1396 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
45f014c5 | 1397 | force_qs_rnp(rsp, rcu_implicit_dynticks_qs); |
64db4cff PM |
1398 | |
1399 | /* Leave state in case more forcing is required. */ | |
1400 | ||
1304afb2 | 1401 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
f96e9232 | 1402 | break; |
64db4cff | 1403 | } |
07079d53 | 1404 | rsp->fqs_active = 0; |
46a1e34e | 1405 | if (rsp->fqs_need_gp) { |
1304afb2 | 1406 | raw_spin_unlock(&rsp->fqslock); /* irqs remain disabled */ |
46a1e34e PM |
1407 | rsp->fqs_need_gp = 0; |
1408 | rcu_start_gp(rsp, flags); /* releases rnp->lock */ | |
1409 | return; | |
1410 | } | |
1304afb2 | 1411 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1412 | unlock_fqs_ret: |
1304afb2 | 1413 | raw_spin_unlock_irqrestore(&rsp->fqslock, flags); |
64db4cff PM |
1414 | } |
1415 | ||
1416 | #else /* #ifdef CONFIG_SMP */ | |
1417 | ||
1418 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1419 | { | |
1420 | set_need_resched(); | |
1421 | } | |
1422 | ||
1423 | #endif /* #else #ifdef CONFIG_SMP */ | |
1424 | ||
1425 | /* | |
1426 | * This does the RCU processing work from softirq context for the | |
1427 | * specified rcu_state and rcu_data structures. This may be called | |
1428 | * only from the CPU to whom the rdp belongs. | |
1429 | */ | |
1430 | static void | |
1431 | __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) | |
1432 | { | |
1433 | unsigned long flags; | |
1434 | ||
2e597558 PM |
1435 | WARN_ON_ONCE(rdp->beenonline == 0); |
1436 | ||
64db4cff PM |
1437 | /* |
1438 | * If an RCU GP has gone long enough, go check for dyntick | |
1439 | * idle CPUs and, if needed, send resched IPIs. | |
1440 | */ | |
20133cfc | 1441 | if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
64db4cff PM |
1442 | force_quiescent_state(rsp, 1); |
1443 | ||
1444 | /* | |
1445 | * Advance callbacks in response to end of earlier grace | |
1446 | * period that some other CPU ended. | |
1447 | */ | |
1448 | rcu_process_gp_end(rsp, rdp); | |
1449 | ||
1450 | /* Update RCU state based on any recent quiescent states. */ | |
1451 | rcu_check_quiescent_state(rsp, rdp); | |
1452 | ||
1453 | /* Does this CPU require a not-yet-started grace period? */ | |
1454 | if (cpu_needs_another_gp(rsp, rdp)) { | |
1304afb2 | 1455 | raw_spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags); |
64db4cff PM |
1456 | rcu_start_gp(rsp, flags); /* releases above lock */ |
1457 | } | |
1458 | ||
1459 | /* If there are callbacks ready, invoke them. */ | |
37c72e56 | 1460 | rcu_do_batch(rsp, rdp); |
64db4cff PM |
1461 | } |
1462 | ||
1463 | /* | |
1464 | * Do softirq processing for the current CPU. | |
1465 | */ | |
a26ac245 | 1466 | static void rcu_process_callbacks(void) |
64db4cff | 1467 | { |
d6714c22 PM |
1468 | __rcu_process_callbacks(&rcu_sched_state, |
1469 | &__get_cpu_var(rcu_sched_data)); | |
64db4cff | 1470 | __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); |
f41d911f | 1471 | rcu_preempt_process_callbacks(); |
a47cd880 PM |
1472 | |
1473 | /* If we are last CPU on way to dyntick-idle mode, accelerate it. */ | |
1474 | rcu_needs_cpu_flush(); | |
64db4cff PM |
1475 | } |
1476 | ||
a26ac245 PM |
1477 | /* |
1478 | * Wake up the current CPU's kthread. This replaces raise_softirq() | |
1479 | * in earlier versions of RCU. Note that because we are running on | |
1480 | * the current CPU with interrupts disabled, the rcu_cpu_kthread_task | |
1481 | * cannot disappear out from under us. | |
1482 | */ | |
27f4d280 | 1483 | static void invoke_rcu_cpu_kthread(void) |
a26ac245 PM |
1484 | { |
1485 | unsigned long flags; | |
a26ac245 PM |
1486 | |
1487 | local_irq_save(flags); | |
f0a07aea PM |
1488 | __this_cpu_write(rcu_cpu_has_work, 1); |
1489 | if (__this_cpu_read(rcu_cpu_kthread_task) == NULL) { | |
a26ac245 PM |
1490 | local_irq_restore(flags); |
1491 | return; | |
1492 | } | |
f0a07aea | 1493 | wake_up(&__get_cpu_var(rcu_cpu_wq)); |
a26ac245 PM |
1494 | local_irq_restore(flags); |
1495 | } | |
1496 | ||
27f4d280 PM |
1497 | /* |
1498 | * Wake up the specified per-rcu_node-structure kthread. | |
1499 | * The caller must hold ->lock. | |
1500 | */ | |
1501 | static void invoke_rcu_node_kthread(struct rcu_node *rnp) | |
1502 | { | |
1503 | struct task_struct *t; | |
1504 | ||
1505 | t = rnp->node_kthread_task; | |
1506 | if (t != NULL) | |
1507 | wake_up_process(t); | |
1508 | } | |
1509 | ||
e3995a25 PM |
1510 | /* |
1511 | * Set the specified CPU's kthread to run RT or not, as specified by | |
1512 | * the to_rt argument. The CPU-hotplug locks are held, so the task | |
1513 | * is not going away. | |
1514 | */ | |
1515 | static void rcu_cpu_kthread_setrt(int cpu, int to_rt) | |
1516 | { | |
1517 | int policy; | |
1518 | struct sched_param sp; | |
1519 | struct task_struct *t; | |
1520 | ||
1521 | t = per_cpu(rcu_cpu_kthread_task, cpu); | |
1522 | if (t == NULL) | |
1523 | return; | |
1524 | if (to_rt) { | |
1525 | policy = SCHED_FIFO; | |
1526 | sp.sched_priority = RCU_KTHREAD_PRIO; | |
1527 | } else { | |
1528 | policy = SCHED_NORMAL; | |
1529 | sp.sched_priority = 0; | |
1530 | } | |
1531 | sched_setscheduler_nocheck(t, policy, &sp); | |
1532 | } | |
1533 | ||
a26ac245 PM |
1534 | /* |
1535 | * Timer handler to initiate the waking up of per-CPU kthreads that | |
1536 | * have yielded the CPU due to excess numbers of RCU callbacks. | |
27f4d280 PM |
1537 | * We wake up the per-rcu_node kthread, which in turn will wake up |
1538 | * the booster kthread. | |
a26ac245 PM |
1539 | */ |
1540 | static void rcu_cpu_kthread_timer(unsigned long arg) | |
1541 | { | |
1542 | unsigned long flags; | |
27f4d280 | 1543 | struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, arg); |
a26ac245 | 1544 | struct rcu_node *rnp = rdp->mynode; |
a26ac245 PM |
1545 | |
1546 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
1547 | rnp->wakemask |= rdp->grpmask; | |
27f4d280 | 1548 | invoke_rcu_node_kthread(rnp); |
a26ac245 PM |
1549 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1550 | } | |
1551 | ||
1552 | /* | |
1553 | * Drop to non-real-time priority and yield, but only after posting a | |
1554 | * timer that will cause us to regain our real-time priority if we | |
1555 | * remain preempted. Either way, we restore our real-time priority | |
1556 | * before returning. | |
1557 | */ | |
27f4d280 | 1558 | static void rcu_yield(void (*f)(unsigned long), unsigned long arg) |
a26ac245 | 1559 | { |
a26ac245 PM |
1560 | struct sched_param sp; |
1561 | struct timer_list yield_timer; | |
1562 | ||
27f4d280 | 1563 | setup_timer_on_stack(&yield_timer, f, arg); |
a26ac245 PM |
1564 | mod_timer(&yield_timer, jiffies + 2); |
1565 | sp.sched_priority = 0; | |
1566 | sched_setscheduler_nocheck(current, SCHED_NORMAL, &sp); | |
baa1ae0c | 1567 | set_user_nice(current, 19); |
a26ac245 PM |
1568 | schedule(); |
1569 | sp.sched_priority = RCU_KTHREAD_PRIO; | |
1570 | sched_setscheduler_nocheck(current, SCHED_FIFO, &sp); | |
1571 | del_timer(&yield_timer); | |
1572 | } | |
1573 | ||
1574 | /* | |
1575 | * Handle cases where the rcu_cpu_kthread() ends up on the wrong CPU. | |
1576 | * This can happen while the corresponding CPU is either coming online | |
1577 | * or going offline. We cannot wait until the CPU is fully online | |
1578 | * before starting the kthread, because the various notifier functions | |
1579 | * can wait for RCU grace periods. So we park rcu_cpu_kthread() until | |
1580 | * the corresponding CPU is online. | |
1581 | * | |
1582 | * Return 1 if the kthread needs to stop, 0 otherwise. | |
1583 | * | |
1584 | * Caller must disable bh. This function can momentarily enable it. | |
1585 | */ | |
1586 | static int rcu_cpu_kthread_should_stop(int cpu) | |
1587 | { | |
1588 | while (cpu_is_offline(cpu) || | |
1589 | !cpumask_equal(¤t->cpus_allowed, cpumask_of(cpu)) || | |
1590 | smp_processor_id() != cpu) { | |
1591 | if (kthread_should_stop()) | |
1592 | return 1; | |
15ba0ba8 PM |
1593 | per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU; |
1594 | per_cpu(rcu_cpu_kthread_cpu, cpu) = raw_smp_processor_id(); | |
a26ac245 PM |
1595 | local_bh_enable(); |
1596 | schedule_timeout_uninterruptible(1); | |
1597 | if (!cpumask_equal(¤t->cpus_allowed, cpumask_of(cpu))) | |
1598 | set_cpus_allowed_ptr(current, cpumask_of(cpu)); | |
1599 | local_bh_disable(); | |
1600 | } | |
15ba0ba8 | 1601 | per_cpu(rcu_cpu_kthread_cpu, cpu) = cpu; |
a26ac245 PM |
1602 | return 0; |
1603 | } | |
1604 | ||
1605 | /* | |
1606 | * Per-CPU kernel thread that invokes RCU callbacks. This replaces the | |
1607 | * earlier RCU softirq. | |
1608 | */ | |
1609 | static int rcu_cpu_kthread(void *arg) | |
1610 | { | |
1611 | int cpu = (int)(long)arg; | |
1612 | unsigned long flags; | |
1613 | int spincnt = 0; | |
d71df90e | 1614 | unsigned int *statusp = &per_cpu(rcu_cpu_kthread_status, cpu); |
a26ac245 PM |
1615 | wait_queue_head_t *wqp = &per_cpu(rcu_cpu_wq, cpu); |
1616 | char work; | |
1617 | char *workp = &per_cpu(rcu_cpu_has_work, cpu); | |
1618 | ||
1619 | for (;;) { | |
d71df90e | 1620 | *statusp = RCU_KTHREAD_WAITING; |
a26ac245 PM |
1621 | wait_event_interruptible(*wqp, |
1622 | *workp != 0 || kthread_should_stop()); | |
1623 | local_bh_disable(); | |
1624 | if (rcu_cpu_kthread_should_stop(cpu)) { | |
1625 | local_bh_enable(); | |
1626 | break; | |
1627 | } | |
d71df90e | 1628 | *statusp = RCU_KTHREAD_RUNNING; |
5ece5bab | 1629 | per_cpu(rcu_cpu_kthread_loops, cpu)++; |
a26ac245 PM |
1630 | local_irq_save(flags); |
1631 | work = *workp; | |
1632 | *workp = 0; | |
1633 | local_irq_restore(flags); | |
1634 | if (work) | |
1635 | rcu_process_callbacks(); | |
1636 | local_bh_enable(); | |
1637 | if (*workp != 0) | |
1638 | spincnt++; | |
1639 | else | |
1640 | spincnt = 0; | |
1641 | if (spincnt > 10) { | |
d71df90e | 1642 | *statusp = RCU_KTHREAD_YIELDING; |
27f4d280 | 1643 | rcu_yield(rcu_cpu_kthread_timer, (unsigned long)cpu); |
a26ac245 PM |
1644 | spincnt = 0; |
1645 | } | |
1646 | } | |
d71df90e | 1647 | *statusp = RCU_KTHREAD_STOPPED; |
a26ac245 PM |
1648 | return 0; |
1649 | } | |
1650 | ||
1651 | /* | |
1652 | * Spawn a per-CPU kthread, setting up affinity and priority. | |
1653 | * Because the CPU hotplug lock is held, no other CPU will be attempting | |
1654 | * to manipulate rcu_cpu_kthread_task. There might be another CPU | |
1655 | * attempting to access it during boot, but the locking in kthread_bind() | |
1656 | * will enforce sufficient ordering. | |
1657 | */ | |
1658 | static int __cpuinit rcu_spawn_one_cpu_kthread(int cpu) | |
1659 | { | |
1660 | struct sched_param sp; | |
1661 | struct task_struct *t; | |
1662 | ||
1663 | if (!rcu_kthreads_spawnable || | |
1664 | per_cpu(rcu_cpu_kthread_task, cpu) != NULL) | |
1665 | return 0; | |
1666 | t = kthread_create(rcu_cpu_kthread, (void *)(long)cpu, "rcuc%d", cpu); | |
1667 | if (IS_ERR(t)) | |
1668 | return PTR_ERR(t); | |
1669 | kthread_bind(t, cpu); | |
15ba0ba8 | 1670 | per_cpu(rcu_cpu_kthread_cpu, cpu) = cpu; |
a26ac245 PM |
1671 | WARN_ON_ONCE(per_cpu(rcu_cpu_kthread_task, cpu) != NULL); |
1672 | per_cpu(rcu_cpu_kthread_task, cpu) = t; | |
1673 | wake_up_process(t); | |
1674 | sp.sched_priority = RCU_KTHREAD_PRIO; | |
1675 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); | |
1676 | return 0; | |
1677 | } | |
1678 | ||
1679 | /* | |
1680 | * Per-rcu_node kthread, which is in charge of waking up the per-CPU | |
1681 | * kthreads when needed. We ignore requests to wake up kthreads | |
1682 | * for offline CPUs, which is OK because force_quiescent_state() | |
1683 | * takes care of this case. | |
1684 | */ | |
1685 | static int rcu_node_kthread(void *arg) | |
1686 | { | |
1687 | int cpu; | |
1688 | unsigned long flags; | |
1689 | unsigned long mask; | |
1690 | struct rcu_node *rnp = (struct rcu_node *)arg; | |
1691 | struct sched_param sp; | |
1692 | struct task_struct *t; | |
1693 | ||
1694 | for (;;) { | |
d71df90e | 1695 | rnp->node_kthread_status = RCU_KTHREAD_WAITING; |
a26ac245 PM |
1696 | wait_event_interruptible(rnp->node_wq, rnp->wakemask != 0 || |
1697 | kthread_should_stop()); | |
1698 | if (kthread_should_stop()) | |
1699 | break; | |
d71df90e | 1700 | rnp->node_kthread_status = RCU_KTHREAD_RUNNING; |
a26ac245 PM |
1701 | raw_spin_lock_irqsave(&rnp->lock, flags); |
1702 | mask = rnp->wakemask; | |
1703 | rnp->wakemask = 0; | |
27f4d280 | 1704 | rcu_initiate_boost(rnp); |
a26ac245 PM |
1705 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1706 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1) { | |
1707 | if ((mask & 0x1) == 0) | |
1708 | continue; | |
1709 | preempt_disable(); | |
1710 | t = per_cpu(rcu_cpu_kthread_task, cpu); | |
1711 | if (!cpu_online(cpu) || t == NULL) { | |
1712 | preempt_enable(); | |
1713 | continue; | |
1714 | } | |
1715 | per_cpu(rcu_cpu_has_work, cpu) = 1; | |
1716 | sp.sched_priority = RCU_KTHREAD_PRIO; | |
1717 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); | |
1718 | preempt_enable(); | |
1719 | } | |
1720 | } | |
d71df90e | 1721 | rnp->node_kthread_status = RCU_KTHREAD_STOPPED; |
a26ac245 PM |
1722 | return 0; |
1723 | } | |
1724 | ||
1725 | /* | |
1726 | * Set the per-rcu_node kthread's affinity to cover all CPUs that are | |
27f4d280 PM |
1727 | * served by the rcu_node in question. The CPU hotplug lock is still |
1728 | * held, so the value of rnp->qsmaskinit will be stable. | |
0f962a5e PM |
1729 | * |
1730 | * We don't include outgoingcpu in the affinity set, use -1 if there is | |
1731 | * no outgoing CPU. If there are no CPUs left in the affinity set, | |
1732 | * this function allows the kthread to execute on any CPU. | |
a26ac245 | 1733 | */ |
0f962a5e | 1734 | static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
a26ac245 PM |
1735 | { |
1736 | cpumask_var_t cm; | |
1737 | int cpu; | |
1738 | unsigned long mask = rnp->qsmaskinit; | |
1739 | ||
27f4d280 | 1740 | if (rnp->node_kthread_task == NULL || mask == 0) |
a26ac245 PM |
1741 | return; |
1742 | if (!alloc_cpumask_var(&cm, GFP_KERNEL)) | |
1743 | return; | |
1744 | cpumask_clear(cm); | |
1745 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1) | |
0f962a5e | 1746 | if ((mask & 0x1) && cpu != outgoingcpu) |
a26ac245 | 1747 | cpumask_set_cpu(cpu, cm); |
0f962a5e PM |
1748 | if (cpumask_weight(cm) == 0) { |
1749 | cpumask_setall(cm); | |
1750 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++) | |
1751 | cpumask_clear_cpu(cpu, cm); | |
1752 | WARN_ON_ONCE(cpumask_weight(cm) == 0); | |
1753 | } | |
a26ac245 | 1754 | set_cpus_allowed_ptr(rnp->node_kthread_task, cm); |
27f4d280 | 1755 | rcu_boost_kthread_setaffinity(rnp, cm); |
a26ac245 PM |
1756 | free_cpumask_var(cm); |
1757 | } | |
1758 | ||
1759 | /* | |
1760 | * Spawn a per-rcu_node kthread, setting priority and affinity. | |
27f4d280 PM |
1761 | * Called during boot before online/offline can happen, or, if |
1762 | * during runtime, with the main CPU-hotplug locks held. So only | |
1763 | * one of these can be executing at a time. | |
a26ac245 PM |
1764 | */ |
1765 | static int __cpuinit rcu_spawn_one_node_kthread(struct rcu_state *rsp, | |
1766 | struct rcu_node *rnp) | |
1767 | { | |
27f4d280 | 1768 | unsigned long flags; |
a26ac245 PM |
1769 | int rnp_index = rnp - &rsp->node[0]; |
1770 | struct sched_param sp; | |
1771 | struct task_struct *t; | |
1772 | ||
1773 | if (!rcu_kthreads_spawnable || | |
27f4d280 | 1774 | rnp->qsmaskinit == 0) |
a26ac245 | 1775 | return 0; |
27f4d280 PM |
1776 | if (rnp->node_kthread_task == NULL) { |
1777 | t = kthread_create(rcu_node_kthread, (void *)rnp, | |
1778 | "rcun%d", rnp_index); | |
1779 | if (IS_ERR(t)) | |
1780 | return PTR_ERR(t); | |
1781 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
1782 | rnp->node_kthread_task = t; | |
1783 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
1784 | wake_up_process(t); | |
1785 | sp.sched_priority = 99; | |
1786 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); | |
1787 | } | |
1788 | return rcu_spawn_one_boost_kthread(rsp, rnp, rnp_index); | |
a26ac245 PM |
1789 | } |
1790 | ||
1791 | /* | |
1792 | * Spawn all kthreads -- called as soon as the scheduler is running. | |
1793 | */ | |
1794 | static int __init rcu_spawn_kthreads(void) | |
1795 | { | |
1796 | int cpu; | |
1797 | struct rcu_node *rnp; | |
1798 | ||
1799 | rcu_kthreads_spawnable = 1; | |
1800 | for_each_possible_cpu(cpu) { | |
1801 | init_waitqueue_head(&per_cpu(rcu_cpu_wq, cpu)); | |
1802 | per_cpu(rcu_cpu_has_work, cpu) = 0; | |
1803 | if (cpu_online(cpu)) | |
1804 | (void)rcu_spawn_one_cpu_kthread(cpu); | |
1805 | } | |
27f4d280 PM |
1806 | rnp = rcu_get_root(rcu_state); |
1807 | init_waitqueue_head(&rnp->node_wq); | |
1808 | rcu_init_boost_waitqueue(rnp); | |
1809 | (void)rcu_spawn_one_node_kthread(rcu_state, rnp); | |
1810 | if (NUM_RCU_NODES > 1) | |
1811 | rcu_for_each_leaf_node(rcu_state, rnp) { | |
1812 | init_waitqueue_head(&rnp->node_wq); | |
1813 | rcu_init_boost_waitqueue(rnp); | |
1814 | (void)rcu_spawn_one_node_kthread(rcu_state, rnp); | |
1815 | } | |
a26ac245 PM |
1816 | return 0; |
1817 | } | |
1818 | early_initcall(rcu_spawn_kthreads); | |
1819 | ||
64db4cff PM |
1820 | static void |
1821 | __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu), | |
1822 | struct rcu_state *rsp) | |
1823 | { | |
1824 | unsigned long flags; | |
1825 | struct rcu_data *rdp; | |
1826 | ||
551d55a9 | 1827 | debug_rcu_head_queue(head); |
64db4cff PM |
1828 | head->func = func; |
1829 | head->next = NULL; | |
1830 | ||
1831 | smp_mb(); /* Ensure RCU update seen before callback registry. */ | |
1832 | ||
1833 | /* | |
1834 | * Opportunistically note grace-period endings and beginnings. | |
1835 | * Note that we might see a beginning right after we see an | |
1836 | * end, but never vice versa, since this CPU has to pass through | |
1837 | * a quiescent state betweentimes. | |
1838 | */ | |
1839 | local_irq_save(flags); | |
394f99a9 | 1840 | rdp = this_cpu_ptr(rsp->rda); |
64db4cff PM |
1841 | |
1842 | /* Add the callback to our list. */ | |
1843 | *rdp->nxttail[RCU_NEXT_TAIL] = head; | |
1844 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; | |
2655d57e PM |
1845 | rdp->qlen++; |
1846 | ||
1847 | /* If interrupts were disabled, don't dive into RCU core. */ | |
1848 | if (irqs_disabled_flags(flags)) { | |
1849 | local_irq_restore(flags); | |
1850 | return; | |
1851 | } | |
64db4cff | 1852 | |
37c72e56 PM |
1853 | /* |
1854 | * Force the grace period if too many callbacks or too long waiting. | |
1855 | * Enforce hysteresis, and don't invoke force_quiescent_state() | |
1856 | * if some other CPU has recently done so. Also, don't bother | |
1857 | * invoking force_quiescent_state() if the newly enqueued callback | |
1858 | * is the only one waiting for a grace period to complete. | |
1859 | */ | |
2655d57e | 1860 | if (unlikely(rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) { |
b52573d2 PM |
1861 | |
1862 | /* Are we ignoring a completed grace period? */ | |
1863 | rcu_process_gp_end(rsp, rdp); | |
1864 | check_for_new_grace_period(rsp, rdp); | |
1865 | ||
1866 | /* Start a new grace period if one not already started. */ | |
1867 | if (!rcu_gp_in_progress(rsp)) { | |
1868 | unsigned long nestflag; | |
1869 | struct rcu_node *rnp_root = rcu_get_root(rsp); | |
1870 | ||
1871 | raw_spin_lock_irqsave(&rnp_root->lock, nestflag); | |
1872 | rcu_start_gp(rsp, nestflag); /* rlses rnp_root->lock */ | |
1873 | } else { | |
1874 | /* Give the grace period a kick. */ | |
1875 | rdp->blimit = LONG_MAX; | |
1876 | if (rsp->n_force_qs == rdp->n_force_qs_snap && | |
1877 | *rdp->nxttail[RCU_DONE_TAIL] != head) | |
1878 | force_quiescent_state(rsp, 0); | |
1879 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1880 | rdp->qlen_last_fqs_check = rdp->qlen; | |
1881 | } | |
20133cfc | 1882 | } else if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
64db4cff PM |
1883 | force_quiescent_state(rsp, 1); |
1884 | local_irq_restore(flags); | |
1885 | } | |
1886 | ||
1887 | /* | |
d6714c22 | 1888 | * Queue an RCU-sched callback for invocation after a grace period. |
64db4cff | 1889 | */ |
d6714c22 | 1890 | void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) |
64db4cff | 1891 | { |
d6714c22 | 1892 | __call_rcu(head, func, &rcu_sched_state); |
64db4cff | 1893 | } |
d6714c22 | 1894 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
64db4cff PM |
1895 | |
1896 | /* | |
1897 | * Queue an RCU for invocation after a quicker grace period. | |
1898 | */ | |
1899 | void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
1900 | { | |
1901 | __call_rcu(head, func, &rcu_bh_state); | |
1902 | } | |
1903 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
1904 | ||
6ebb237b PM |
1905 | /** |
1906 | * synchronize_sched - wait until an rcu-sched grace period has elapsed. | |
1907 | * | |
1908 | * Control will return to the caller some time after a full rcu-sched | |
1909 | * grace period has elapsed, in other words after all currently executing | |
1910 | * rcu-sched read-side critical sections have completed. These read-side | |
1911 | * critical sections are delimited by rcu_read_lock_sched() and | |
1912 | * rcu_read_unlock_sched(), and may be nested. Note that preempt_disable(), | |
1913 | * local_irq_disable(), and so on may be used in place of | |
1914 | * rcu_read_lock_sched(). | |
1915 | * | |
1916 | * This means that all preempt_disable code sequences, including NMI and | |
1917 | * hardware-interrupt handlers, in progress on entry will have completed | |
1918 | * before this primitive returns. However, this does not guarantee that | |
1919 | * softirq handlers will have completed, since in some kernels, these | |
1920 | * handlers can run in process context, and can block. | |
1921 | * | |
1922 | * This primitive provides the guarantees made by the (now removed) | |
1923 | * synchronize_kernel() API. In contrast, synchronize_rcu() only | |
1924 | * guarantees that rcu_read_lock() sections will have completed. | |
1925 | * In "classic RCU", these two guarantees happen to be one and | |
1926 | * the same, but can differ in realtime RCU implementations. | |
1927 | */ | |
1928 | void synchronize_sched(void) | |
1929 | { | |
1930 | struct rcu_synchronize rcu; | |
1931 | ||
1932 | if (rcu_blocking_is_gp()) | |
1933 | return; | |
1934 | ||
72d5a9f7 | 1935 | init_rcu_head_on_stack(&rcu.head); |
6ebb237b PM |
1936 | init_completion(&rcu.completion); |
1937 | /* Will wake me after RCU finished. */ | |
1938 | call_rcu_sched(&rcu.head, wakeme_after_rcu); | |
1939 | /* Wait for it. */ | |
1940 | wait_for_completion(&rcu.completion); | |
72d5a9f7 | 1941 | destroy_rcu_head_on_stack(&rcu.head); |
6ebb237b PM |
1942 | } |
1943 | EXPORT_SYMBOL_GPL(synchronize_sched); | |
1944 | ||
1945 | /** | |
1946 | * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed. | |
1947 | * | |
1948 | * Control will return to the caller some time after a full rcu_bh grace | |
1949 | * period has elapsed, in other words after all currently executing rcu_bh | |
1950 | * read-side critical sections have completed. RCU read-side critical | |
1951 | * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(), | |
1952 | * and may be nested. | |
1953 | */ | |
1954 | void synchronize_rcu_bh(void) | |
1955 | { | |
1956 | struct rcu_synchronize rcu; | |
1957 | ||
1958 | if (rcu_blocking_is_gp()) | |
1959 | return; | |
1960 | ||
72d5a9f7 | 1961 | init_rcu_head_on_stack(&rcu.head); |
6ebb237b PM |
1962 | init_completion(&rcu.completion); |
1963 | /* Will wake me after RCU finished. */ | |
1964 | call_rcu_bh(&rcu.head, wakeme_after_rcu); | |
1965 | /* Wait for it. */ | |
1966 | wait_for_completion(&rcu.completion); | |
72d5a9f7 | 1967 | destroy_rcu_head_on_stack(&rcu.head); |
6ebb237b PM |
1968 | } |
1969 | EXPORT_SYMBOL_GPL(synchronize_rcu_bh); | |
1970 | ||
64db4cff PM |
1971 | /* |
1972 | * Check to see if there is any immediate RCU-related work to be done | |
1973 | * by the current CPU, for the specified type of RCU, returning 1 if so. | |
1974 | * The checks are in order of increasing expense: checks that can be | |
1975 | * carried out against CPU-local state are performed first. However, | |
1976 | * we must check for CPU stalls first, else we might not get a chance. | |
1977 | */ | |
1978 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | |
1979 | { | |
2f51f988 PM |
1980 | struct rcu_node *rnp = rdp->mynode; |
1981 | ||
64db4cff PM |
1982 | rdp->n_rcu_pending++; |
1983 | ||
1984 | /* Check for CPU stalls, if enabled. */ | |
1985 | check_cpu_stall(rsp, rdp); | |
1986 | ||
1987 | /* Is the RCU core waiting for a quiescent state from this CPU? */ | |
d21670ac | 1988 | if (rdp->qs_pending && !rdp->passed_quiesc) { |
d25eb944 PM |
1989 | |
1990 | /* | |
1991 | * If force_quiescent_state() coming soon and this CPU | |
1992 | * needs a quiescent state, and this is either RCU-sched | |
1993 | * or RCU-bh, force a local reschedule. | |
1994 | */ | |
d21670ac | 1995 | rdp->n_rp_qs_pending++; |
6cc68793 | 1996 | if (!rdp->preemptible && |
d25eb944 PM |
1997 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs) - 1, |
1998 | jiffies)) | |
1999 | set_need_resched(); | |
d21670ac PM |
2000 | } else if (rdp->qs_pending && rdp->passed_quiesc) { |
2001 | rdp->n_rp_report_qs++; | |
64db4cff | 2002 | return 1; |
7ba5c840 | 2003 | } |
64db4cff PM |
2004 | |
2005 | /* Does this CPU have callbacks ready to invoke? */ | |
7ba5c840 PM |
2006 | if (cpu_has_callbacks_ready_to_invoke(rdp)) { |
2007 | rdp->n_rp_cb_ready++; | |
64db4cff | 2008 | return 1; |
7ba5c840 | 2009 | } |
64db4cff PM |
2010 | |
2011 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
7ba5c840 PM |
2012 | if (cpu_needs_another_gp(rsp, rdp)) { |
2013 | rdp->n_rp_cpu_needs_gp++; | |
64db4cff | 2014 | return 1; |
7ba5c840 | 2015 | } |
64db4cff PM |
2016 | |
2017 | /* Has another RCU grace period completed? */ | |
2f51f988 | 2018 | if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */ |
7ba5c840 | 2019 | rdp->n_rp_gp_completed++; |
64db4cff | 2020 | return 1; |
7ba5c840 | 2021 | } |
64db4cff PM |
2022 | |
2023 | /* Has a new RCU grace period started? */ | |
2f51f988 | 2024 | if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */ |
7ba5c840 | 2025 | rdp->n_rp_gp_started++; |
64db4cff | 2026 | return 1; |
7ba5c840 | 2027 | } |
64db4cff PM |
2028 | |
2029 | /* Has an RCU GP gone long enough to send resched IPIs &c? */ | |
fc2219d4 | 2030 | if (rcu_gp_in_progress(rsp) && |
20133cfc | 2031 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) { |
7ba5c840 | 2032 | rdp->n_rp_need_fqs++; |
64db4cff | 2033 | return 1; |
7ba5c840 | 2034 | } |
64db4cff PM |
2035 | |
2036 | /* nothing to do */ | |
7ba5c840 | 2037 | rdp->n_rp_need_nothing++; |
64db4cff PM |
2038 | return 0; |
2039 | } | |
2040 | ||
2041 | /* | |
2042 | * Check to see if there is any immediate RCU-related work to be done | |
2043 | * by the current CPU, returning 1 if so. This function is part of the | |
2044 | * RCU implementation; it is -not- an exported member of the RCU API. | |
2045 | */ | |
a157229c | 2046 | static int rcu_pending(int cpu) |
64db4cff | 2047 | { |
d6714c22 | 2048 | return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) || |
f41d911f PM |
2049 | __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) || |
2050 | rcu_preempt_pending(cpu); | |
64db4cff PM |
2051 | } |
2052 | ||
2053 | /* | |
2054 | * Check to see if any future RCU-related work will need to be done | |
2055 | * by the current CPU, even if none need be done immediately, returning | |
8bd93a2c | 2056 | * 1 if so. |
64db4cff | 2057 | */ |
8bd93a2c | 2058 | static int rcu_needs_cpu_quick_check(int cpu) |
64db4cff PM |
2059 | { |
2060 | /* RCU callbacks either ready or pending? */ | |
d6714c22 | 2061 | return per_cpu(rcu_sched_data, cpu).nxtlist || |
f41d911f PM |
2062 | per_cpu(rcu_bh_data, cpu).nxtlist || |
2063 | rcu_preempt_needs_cpu(cpu); | |
64db4cff PM |
2064 | } |
2065 | ||
d0ec774c PM |
2066 | static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL}; |
2067 | static atomic_t rcu_barrier_cpu_count; | |
2068 | static DEFINE_MUTEX(rcu_barrier_mutex); | |
2069 | static struct completion rcu_barrier_completion; | |
d0ec774c PM |
2070 | |
2071 | static void rcu_barrier_callback(struct rcu_head *notused) | |
2072 | { | |
2073 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
2074 | complete(&rcu_barrier_completion); | |
2075 | } | |
2076 | ||
2077 | /* | |
2078 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
2079 | */ | |
2080 | static void rcu_barrier_func(void *type) | |
2081 | { | |
2082 | int cpu = smp_processor_id(); | |
2083 | struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu); | |
2084 | void (*call_rcu_func)(struct rcu_head *head, | |
2085 | void (*func)(struct rcu_head *head)); | |
2086 | ||
2087 | atomic_inc(&rcu_barrier_cpu_count); | |
2088 | call_rcu_func = type; | |
2089 | call_rcu_func(head, rcu_barrier_callback); | |
2090 | } | |
2091 | ||
d0ec774c PM |
2092 | /* |
2093 | * Orchestrate the specified type of RCU barrier, waiting for all | |
2094 | * RCU callbacks of the specified type to complete. | |
2095 | */ | |
e74f4c45 PM |
2096 | static void _rcu_barrier(struct rcu_state *rsp, |
2097 | void (*call_rcu_func)(struct rcu_head *head, | |
d0ec774c PM |
2098 | void (*func)(struct rcu_head *head))) |
2099 | { | |
2100 | BUG_ON(in_interrupt()); | |
e74f4c45 | 2101 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
d0ec774c PM |
2102 | mutex_lock(&rcu_barrier_mutex); |
2103 | init_completion(&rcu_barrier_completion); | |
2104 | /* | |
2105 | * Initialize rcu_barrier_cpu_count to 1, then invoke | |
2106 | * rcu_barrier_func() on each CPU, so that each CPU also has | |
2107 | * incremented rcu_barrier_cpu_count. Only then is it safe to | |
2108 | * decrement rcu_barrier_cpu_count -- otherwise the first CPU | |
2109 | * might complete its grace period before all of the other CPUs | |
2110 | * did their increment, causing this function to return too | |
2d999e03 PM |
2111 | * early. Note that on_each_cpu() disables irqs, which prevents |
2112 | * any CPUs from coming online or going offline until each online | |
2113 | * CPU has queued its RCU-barrier callback. | |
d0ec774c PM |
2114 | */ |
2115 | atomic_set(&rcu_barrier_cpu_count, 1); | |
2116 | on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1); | |
2117 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
2118 | complete(&rcu_barrier_completion); | |
2119 | wait_for_completion(&rcu_barrier_completion); | |
2120 | mutex_unlock(&rcu_barrier_mutex); | |
d0ec774c | 2121 | } |
d0ec774c PM |
2122 | |
2123 | /** | |
2124 | * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete. | |
2125 | */ | |
2126 | void rcu_barrier_bh(void) | |
2127 | { | |
e74f4c45 | 2128 | _rcu_barrier(&rcu_bh_state, call_rcu_bh); |
d0ec774c PM |
2129 | } |
2130 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); | |
2131 | ||
2132 | /** | |
2133 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. | |
2134 | */ | |
2135 | void rcu_barrier_sched(void) | |
2136 | { | |
e74f4c45 | 2137 | _rcu_barrier(&rcu_sched_state, call_rcu_sched); |
d0ec774c PM |
2138 | } |
2139 | EXPORT_SYMBOL_GPL(rcu_barrier_sched); | |
2140 | ||
64db4cff | 2141 | /* |
27569620 | 2142 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 2143 | */ |
27569620 PM |
2144 | static void __init |
2145 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) | |
64db4cff PM |
2146 | { |
2147 | unsigned long flags; | |
2148 | int i; | |
394f99a9 | 2149 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
27569620 PM |
2150 | struct rcu_node *rnp = rcu_get_root(rsp); |
2151 | ||
2152 | /* Set up local state, ensuring consistent view of global state. */ | |
1304afb2 | 2153 | raw_spin_lock_irqsave(&rnp->lock, flags); |
27569620 PM |
2154 | rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo); |
2155 | rdp->nxtlist = NULL; | |
2156 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
2157 | rdp->nxttail[i] = &rdp->nxtlist; | |
2158 | rdp->qlen = 0; | |
2159 | #ifdef CONFIG_NO_HZ | |
2160 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); | |
2161 | #endif /* #ifdef CONFIG_NO_HZ */ | |
2162 | rdp->cpu = cpu; | |
1304afb2 | 2163 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
27569620 PM |
2164 | } |
2165 | ||
2166 | /* | |
2167 | * Initialize a CPU's per-CPU RCU data. Note that only one online or | |
2168 | * offline event can be happening at a given time. Note also that we | |
2169 | * can accept some slop in the rsp->completed access due to the fact | |
2170 | * that this CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 2171 | */ |
e4fa4c97 | 2172 | static void __cpuinit |
6cc68793 | 2173 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptible) |
64db4cff PM |
2174 | { |
2175 | unsigned long flags; | |
64db4cff | 2176 | unsigned long mask; |
394f99a9 | 2177 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff PM |
2178 | struct rcu_node *rnp = rcu_get_root(rsp); |
2179 | ||
2180 | /* Set up local state, ensuring consistent view of global state. */ | |
1304afb2 | 2181 | raw_spin_lock_irqsave(&rnp->lock, flags); |
64db4cff PM |
2182 | rdp->passed_quiesc = 0; /* We could be racing with new GP, */ |
2183 | rdp->qs_pending = 1; /* so set up to respond to current GP. */ | |
2184 | rdp->beenonline = 1; /* We have now been online. */ | |
6cc68793 | 2185 | rdp->preemptible = preemptible; |
37c72e56 PM |
2186 | rdp->qlen_last_fqs_check = 0; |
2187 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
64db4cff | 2188 | rdp->blimit = blimit; |
1304afb2 | 2189 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
2190 | |
2191 | /* | |
2192 | * A new grace period might start here. If so, we won't be part | |
2193 | * of it, but that is OK, as we are currently in a quiescent state. | |
2194 | */ | |
2195 | ||
2196 | /* Exclude any attempts to start a new GP on large systems. */ | |
1304afb2 | 2197 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
64db4cff PM |
2198 | |
2199 | /* Add CPU to rcu_node bitmasks. */ | |
2200 | rnp = rdp->mynode; | |
2201 | mask = rdp->grpmask; | |
2202 | do { | |
2203 | /* Exclude any attempts to start a new GP on small systems. */ | |
1304afb2 | 2204 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
2205 | rnp->qsmaskinit |= mask; |
2206 | mask = rnp->grpmask; | |
d09b62df PM |
2207 | if (rnp == rdp->mynode) { |
2208 | rdp->gpnum = rnp->completed; /* if GP in progress... */ | |
2209 | rdp->completed = rnp->completed; | |
2210 | rdp->passed_quiesc_completed = rnp->completed - 1; | |
2211 | } | |
1304afb2 | 2212 | raw_spin_unlock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
2213 | rnp = rnp->parent; |
2214 | } while (rnp != NULL && !(rnp->qsmaskinit & mask)); | |
2215 | ||
1304afb2 | 2216 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); |
64db4cff PM |
2217 | } |
2218 | ||
2219 | static void __cpuinit rcu_online_cpu(int cpu) | |
2220 | { | |
f41d911f PM |
2221 | rcu_init_percpu_data(cpu, &rcu_sched_state, 0); |
2222 | rcu_init_percpu_data(cpu, &rcu_bh_state, 0); | |
2223 | rcu_preempt_init_percpu_data(cpu); | |
64db4cff PM |
2224 | } |
2225 | ||
a26ac245 PM |
2226 | static void __cpuinit rcu_online_kthreads(int cpu) |
2227 | { | |
27f4d280 | 2228 | struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu); |
a26ac245 PM |
2229 | struct rcu_node *rnp = rdp->mynode; |
2230 | ||
2231 | /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */ | |
2232 | if (rcu_kthreads_spawnable) { | |
2233 | (void)rcu_spawn_one_cpu_kthread(cpu); | |
2234 | if (rnp->node_kthread_task == NULL) | |
27f4d280 | 2235 | (void)rcu_spawn_one_node_kthread(rcu_state, rnp); |
a26ac245 PM |
2236 | } |
2237 | } | |
2238 | ||
64db4cff | 2239 | /* |
f41d911f | 2240 | * Handle CPU online/offline notification events. |
64db4cff | 2241 | */ |
9f680ab4 PM |
2242 | static int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
2243 | unsigned long action, void *hcpu) | |
64db4cff PM |
2244 | { |
2245 | long cpu = (long)hcpu; | |
27f4d280 | 2246 | struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu); |
a26ac245 | 2247 | struct rcu_node *rnp = rdp->mynode; |
64db4cff PM |
2248 | |
2249 | switch (action) { | |
2250 | case CPU_UP_PREPARE: | |
2251 | case CPU_UP_PREPARE_FROZEN: | |
2252 | rcu_online_cpu(cpu); | |
a26ac245 PM |
2253 | rcu_online_kthreads(cpu); |
2254 | break; | |
2255 | case CPU_ONLINE: | |
0f962a5e PM |
2256 | case CPU_DOWN_FAILED: |
2257 | rcu_node_kthread_setaffinity(rnp, -1); | |
e3995a25 | 2258 | rcu_cpu_kthread_setrt(cpu, 1); |
0f962a5e PM |
2259 | break; |
2260 | case CPU_DOWN_PREPARE: | |
2261 | rcu_node_kthread_setaffinity(rnp, cpu); | |
e3995a25 | 2262 | rcu_cpu_kthread_setrt(cpu, 0); |
64db4cff | 2263 | break; |
d0ec774c PM |
2264 | case CPU_DYING: |
2265 | case CPU_DYING_FROZEN: | |
2266 | /* | |
2d999e03 PM |
2267 | * The whole machine is "stopped" except this CPU, so we can |
2268 | * touch any data without introducing corruption. We send the | |
2269 | * dying CPU's callbacks to an arbitrarily chosen online CPU. | |
d0ec774c | 2270 | */ |
29494be7 LJ |
2271 | rcu_send_cbs_to_online(&rcu_bh_state); |
2272 | rcu_send_cbs_to_online(&rcu_sched_state); | |
2273 | rcu_preempt_send_cbs_to_online(); | |
d0ec774c | 2274 | break; |
64db4cff PM |
2275 | case CPU_DEAD: |
2276 | case CPU_DEAD_FROZEN: | |
2277 | case CPU_UP_CANCELED: | |
2278 | case CPU_UP_CANCELED_FROZEN: | |
2279 | rcu_offline_cpu(cpu); | |
2280 | break; | |
2281 | default: | |
2282 | break; | |
2283 | } | |
2284 | return NOTIFY_OK; | |
2285 | } | |
2286 | ||
bbad9379 PM |
2287 | /* |
2288 | * This function is invoked towards the end of the scheduler's initialization | |
2289 | * process. Before this is called, the idle task might contain | |
2290 | * RCU read-side critical sections (during which time, this idle | |
2291 | * task is booting the system). After this function is called, the | |
2292 | * idle tasks are prohibited from containing RCU read-side critical | |
2293 | * sections. This function also enables RCU lockdep checking. | |
2294 | */ | |
2295 | void rcu_scheduler_starting(void) | |
2296 | { | |
2297 | WARN_ON(num_online_cpus() != 1); | |
2298 | WARN_ON(nr_context_switches() > 0); | |
2299 | rcu_scheduler_active = 1; | |
2300 | } | |
2301 | ||
64db4cff PM |
2302 | /* |
2303 | * Compute the per-level fanout, either using the exact fanout specified | |
2304 | * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT. | |
2305 | */ | |
2306 | #ifdef CONFIG_RCU_FANOUT_EXACT | |
2307 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
2308 | { | |
2309 | int i; | |
2310 | ||
0209f649 | 2311 | for (i = NUM_RCU_LVLS - 1; i > 0; i--) |
64db4cff | 2312 | rsp->levelspread[i] = CONFIG_RCU_FANOUT; |
0209f649 | 2313 | rsp->levelspread[0] = RCU_FANOUT_LEAF; |
64db4cff PM |
2314 | } |
2315 | #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
2316 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
2317 | { | |
2318 | int ccur; | |
2319 | int cprv; | |
2320 | int i; | |
2321 | ||
2322 | cprv = NR_CPUS; | |
2323 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
2324 | ccur = rsp->levelcnt[i]; | |
2325 | rsp->levelspread[i] = (cprv + ccur - 1) / ccur; | |
2326 | cprv = ccur; | |
2327 | } | |
2328 | } | |
2329 | #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
2330 | ||
2331 | /* | |
2332 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
2333 | */ | |
394f99a9 LJ |
2334 | static void __init rcu_init_one(struct rcu_state *rsp, |
2335 | struct rcu_data __percpu *rda) | |
64db4cff | 2336 | { |
b6407e86 PM |
2337 | static char *buf[] = { "rcu_node_level_0", |
2338 | "rcu_node_level_1", | |
2339 | "rcu_node_level_2", | |
2340 | "rcu_node_level_3" }; /* Match MAX_RCU_LVLS */ | |
64db4cff PM |
2341 | int cpustride = 1; |
2342 | int i; | |
2343 | int j; | |
2344 | struct rcu_node *rnp; | |
2345 | ||
b6407e86 PM |
2346 | BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
2347 | ||
64db4cff PM |
2348 | /* Initialize the level-tracking arrays. */ |
2349 | ||
2350 | for (i = 1; i < NUM_RCU_LVLS; i++) | |
2351 | rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1]; | |
2352 | rcu_init_levelspread(rsp); | |
2353 | ||
2354 | /* Initialize the elements themselves, starting from the leaves. */ | |
2355 | ||
2356 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
2357 | cpustride *= rsp->levelspread[i]; | |
2358 | rnp = rsp->level[i]; | |
2359 | for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { | |
1304afb2 | 2360 | raw_spin_lock_init(&rnp->lock); |
b6407e86 PM |
2361 | lockdep_set_class_and_name(&rnp->lock, |
2362 | &rcu_node_class[i], buf[i]); | |
f41d911f | 2363 | rnp->gpnum = 0; |
64db4cff PM |
2364 | rnp->qsmask = 0; |
2365 | rnp->qsmaskinit = 0; | |
2366 | rnp->grplo = j * cpustride; | |
2367 | rnp->grphi = (j + 1) * cpustride - 1; | |
2368 | if (rnp->grphi >= NR_CPUS) | |
2369 | rnp->grphi = NR_CPUS - 1; | |
2370 | if (i == 0) { | |
2371 | rnp->grpnum = 0; | |
2372 | rnp->grpmask = 0; | |
2373 | rnp->parent = NULL; | |
2374 | } else { | |
2375 | rnp->grpnum = j % rsp->levelspread[i - 1]; | |
2376 | rnp->grpmask = 1UL << rnp->grpnum; | |
2377 | rnp->parent = rsp->level[i - 1] + | |
2378 | j / rsp->levelspread[i - 1]; | |
2379 | } | |
2380 | rnp->level = i; | |
12f5f524 | 2381 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
64db4cff PM |
2382 | } |
2383 | } | |
0c34029a | 2384 | |
394f99a9 | 2385 | rsp->rda = rda; |
0c34029a LJ |
2386 | rnp = rsp->level[NUM_RCU_LVLS - 1]; |
2387 | for_each_possible_cpu(i) { | |
4a90a068 | 2388 | while (i > rnp->grphi) |
0c34029a | 2389 | rnp++; |
394f99a9 | 2390 | per_cpu_ptr(rsp->rda, i)->mynode = rnp; |
0c34029a LJ |
2391 | rcu_boot_init_percpu_data(i, rsp); |
2392 | } | |
64db4cff PM |
2393 | } |
2394 | ||
9f680ab4 | 2395 | void __init rcu_init(void) |
64db4cff | 2396 | { |
017c4261 | 2397 | int cpu; |
9f680ab4 | 2398 | |
f41d911f | 2399 | rcu_bootup_announce(); |
394f99a9 LJ |
2400 | rcu_init_one(&rcu_sched_state, &rcu_sched_data); |
2401 | rcu_init_one(&rcu_bh_state, &rcu_bh_data); | |
f41d911f | 2402 | __rcu_init_preempt(); |
9f680ab4 PM |
2403 | |
2404 | /* | |
2405 | * We don't need protection against CPU-hotplug here because | |
2406 | * this is called early in boot, before either interrupts | |
2407 | * or the scheduler are operational. | |
2408 | */ | |
2409 | cpu_notifier(rcu_cpu_notify, 0); | |
017c4261 PM |
2410 | for_each_online_cpu(cpu) |
2411 | rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu); | |
c68de209 | 2412 | check_cpu_stall_init(); |
64db4cff PM |
2413 | } |
2414 | ||
1eba8f84 | 2415 | #include "rcutree_plugin.h" |