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); | |
94 | static DEFINE_PER_CPU(wait_queue_head_t, rcu_cpu_wq); | |
95 | static DEFINE_PER_CPU(char, rcu_cpu_has_work); | |
96 | static char rcu_kthreads_spawnable; | |
97 | ||
0f962a5e | 98 | static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); |
27f4d280 | 99 | static void invoke_rcu_cpu_kthread(void); |
a26ac245 PM |
100 | |
101 | #define RCU_KTHREAD_PRIO 1 /* RT priority for per-CPU kthreads. */ | |
102 | ||
fc2219d4 PM |
103 | /* |
104 | * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s | |
105 | * permit this function to be invoked without holding the root rcu_node | |
106 | * structure's ->lock, but of course results can be subject to change. | |
107 | */ | |
108 | static int rcu_gp_in_progress(struct rcu_state *rsp) | |
109 | { | |
110 | return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum); | |
111 | } | |
112 | ||
b1f77b05 | 113 | /* |
d6714c22 | 114 | * Note a quiescent state. Because we do not need to know |
b1f77b05 | 115 | * how many quiescent states passed, just if there was at least |
d6714c22 | 116 | * one since the start of the grace period, this just sets a flag. |
b1f77b05 | 117 | */ |
d6714c22 | 118 | void rcu_sched_qs(int cpu) |
b1f77b05 | 119 | { |
25502a6c | 120 | struct rcu_data *rdp = &per_cpu(rcu_sched_data, cpu); |
f41d911f | 121 | |
c64ac3ce | 122 | rdp->passed_quiesc_completed = rdp->gpnum - 1; |
c3422bea PM |
123 | barrier(); |
124 | rdp->passed_quiesc = 1; | |
b1f77b05 IM |
125 | } |
126 | ||
d6714c22 | 127 | void rcu_bh_qs(int cpu) |
b1f77b05 | 128 | { |
25502a6c | 129 | struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu); |
f41d911f | 130 | |
c64ac3ce | 131 | rdp->passed_quiesc_completed = rdp->gpnum - 1; |
c3422bea PM |
132 | barrier(); |
133 | rdp->passed_quiesc = 1; | |
b1f77b05 | 134 | } |
64db4cff | 135 | |
25502a6c PM |
136 | /* |
137 | * Note a context switch. This is a quiescent state for RCU-sched, | |
138 | * and requires special handling for preemptible RCU. | |
139 | */ | |
140 | void rcu_note_context_switch(int cpu) | |
141 | { | |
142 | rcu_sched_qs(cpu); | |
143 | rcu_preempt_note_context_switch(cpu); | |
144 | } | |
145 | ||
64db4cff | 146 | #ifdef CONFIG_NO_HZ |
90a4d2c0 PM |
147 | DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { |
148 | .dynticks_nesting = 1, | |
e59fb312 | 149 | .dynticks = ATOMIC_INIT(1), |
90a4d2c0 | 150 | }; |
64db4cff PM |
151 | #endif /* #ifdef CONFIG_NO_HZ */ |
152 | ||
153 | static int blimit = 10; /* Maximum callbacks per softirq. */ | |
154 | static int qhimark = 10000; /* If this many pending, ignore blimit. */ | |
155 | static int qlowmark = 100; /* Once only this many pending, use blimit. */ | |
156 | ||
3d76c082 PM |
157 | module_param(blimit, int, 0); |
158 | module_param(qhimark, int, 0); | |
159 | module_param(qlowmark, int, 0); | |
160 | ||
a00e0d71 | 161 | int rcu_cpu_stall_suppress __read_mostly; |
f2e0dd70 | 162 | module_param(rcu_cpu_stall_suppress, int, 0644); |
742734ee | 163 | |
64db4cff | 164 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed); |
a157229c | 165 | static int rcu_pending(int cpu); |
64db4cff PM |
166 | |
167 | /* | |
d6714c22 | 168 | * Return the number of RCU-sched batches processed thus far for debug & stats. |
64db4cff | 169 | */ |
d6714c22 | 170 | long rcu_batches_completed_sched(void) |
64db4cff | 171 | { |
d6714c22 | 172 | return rcu_sched_state.completed; |
64db4cff | 173 | } |
d6714c22 | 174 | EXPORT_SYMBOL_GPL(rcu_batches_completed_sched); |
64db4cff PM |
175 | |
176 | /* | |
177 | * Return the number of RCU BH batches processed thus far for debug & stats. | |
178 | */ | |
179 | long rcu_batches_completed_bh(void) | |
180 | { | |
181 | return rcu_bh_state.completed; | |
182 | } | |
183 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); | |
184 | ||
bf66f18e PM |
185 | /* |
186 | * Force a quiescent state for RCU BH. | |
187 | */ | |
188 | void rcu_bh_force_quiescent_state(void) | |
189 | { | |
190 | force_quiescent_state(&rcu_bh_state, 0); | |
191 | } | |
192 | EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state); | |
193 | ||
194 | /* | |
195 | * Force a quiescent state for RCU-sched. | |
196 | */ | |
197 | void rcu_sched_force_quiescent_state(void) | |
198 | { | |
199 | force_quiescent_state(&rcu_sched_state, 0); | |
200 | } | |
201 | EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state); | |
202 | ||
64db4cff PM |
203 | /* |
204 | * Does the CPU have callbacks ready to be invoked? | |
205 | */ | |
206 | static int | |
207 | cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) | |
208 | { | |
209 | return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]; | |
210 | } | |
211 | ||
212 | /* | |
213 | * Does the current CPU require a yet-as-unscheduled grace period? | |
214 | */ | |
215 | static int | |
216 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) | |
217 | { | |
fc2219d4 | 218 | return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp); |
64db4cff PM |
219 | } |
220 | ||
221 | /* | |
222 | * Return the root node of the specified rcu_state structure. | |
223 | */ | |
224 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) | |
225 | { | |
226 | return &rsp->node[0]; | |
227 | } | |
228 | ||
229 | #ifdef CONFIG_SMP | |
230 | ||
231 | /* | |
232 | * If the specified CPU is offline, tell the caller that it is in | |
233 | * a quiescent state. Otherwise, whack it with a reschedule IPI. | |
234 | * Grace periods can end up waiting on an offline CPU when that | |
235 | * CPU is in the process of coming online -- it will be added to the | |
236 | * rcu_node bitmasks before it actually makes it online. The same thing | |
237 | * can happen while a CPU is in the process of coming online. Because this | |
238 | * race is quite rare, we check for it after detecting that the grace | |
239 | * period has been delayed rather than checking each and every CPU | |
240 | * each and every time we start a new grace period. | |
241 | */ | |
242 | static int rcu_implicit_offline_qs(struct rcu_data *rdp) | |
243 | { | |
244 | /* | |
245 | * If the CPU is offline, it is in a quiescent state. We can | |
246 | * trust its state not to change because interrupts are disabled. | |
247 | */ | |
248 | if (cpu_is_offline(rdp->cpu)) { | |
249 | rdp->offline_fqs++; | |
250 | return 1; | |
251 | } | |
252 | ||
f41d911f PM |
253 | /* If preemptable RCU, no point in sending reschedule IPI. */ |
254 | if (rdp->preemptable) | |
255 | return 0; | |
256 | ||
64db4cff PM |
257 | /* The CPU is online, so send it a reschedule IPI. */ |
258 | if (rdp->cpu != smp_processor_id()) | |
259 | smp_send_reschedule(rdp->cpu); | |
260 | else | |
261 | set_need_resched(); | |
262 | rdp->resched_ipi++; | |
263 | return 0; | |
264 | } | |
265 | ||
266 | #endif /* #ifdef CONFIG_SMP */ | |
267 | ||
268 | #ifdef CONFIG_NO_HZ | |
64db4cff PM |
269 | |
270 | /** | |
271 | * rcu_enter_nohz - inform RCU that current CPU is entering nohz | |
272 | * | |
273 | * Enter nohz mode, in other words, -leave- the mode in which RCU | |
274 | * read-side critical sections can occur. (Though RCU read-side | |
275 | * critical sections can occur in irq handlers in nohz mode, a possibility | |
276 | * handled by rcu_irq_enter() and rcu_irq_exit()). | |
277 | */ | |
278 | void rcu_enter_nohz(void) | |
279 | { | |
280 | unsigned long flags; | |
281 | struct rcu_dynticks *rdtp; | |
282 | ||
64db4cff PM |
283 | local_irq_save(flags); |
284 | rdtp = &__get_cpu_var(rcu_dynticks); | |
e59fb312 PM |
285 | if (--rdtp->dynticks_nesting) { |
286 | local_irq_restore(flags); | |
287 | return; | |
288 | } | |
289 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ | |
290 | smp_mb__before_atomic_inc(); /* See above. */ | |
291 | atomic_inc(&rdtp->dynticks); | |
292 | smp_mb__after_atomic_inc(); /* Force ordering with next sojourn. */ | |
293 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); | |
64db4cff | 294 | local_irq_restore(flags); |
e59fb312 PM |
295 | |
296 | /* If the interrupt queued a callback, get out of dyntick mode. */ | |
297 | if (in_irq() && | |
298 | (__get_cpu_var(rcu_sched_data).nxtlist || | |
299 | __get_cpu_var(rcu_bh_data).nxtlist || | |
300 | rcu_preempt_needs_cpu(smp_processor_id()))) | |
301 | set_need_resched(); | |
64db4cff PM |
302 | } |
303 | ||
304 | /* | |
305 | * rcu_exit_nohz - inform RCU that current CPU is leaving nohz | |
306 | * | |
307 | * Exit nohz mode, in other words, -enter- the mode in which RCU | |
308 | * read-side critical sections normally occur. | |
309 | */ | |
310 | void rcu_exit_nohz(void) | |
311 | { | |
312 | unsigned long flags; | |
313 | struct rcu_dynticks *rdtp; | |
314 | ||
315 | local_irq_save(flags); | |
316 | rdtp = &__get_cpu_var(rcu_dynticks); | |
e59fb312 PM |
317 | if (rdtp->dynticks_nesting++) { |
318 | local_irq_restore(flags); | |
319 | return; | |
320 | } | |
321 | smp_mb__before_atomic_inc(); /* Force ordering w/previous sojourn. */ | |
322 | atomic_inc(&rdtp->dynticks); | |
323 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ | |
324 | smp_mb__after_atomic_inc(); /* See above. */ | |
325 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); | |
64db4cff | 326 | local_irq_restore(flags); |
64db4cff PM |
327 | } |
328 | ||
329 | /** | |
330 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
331 | * | |
332 | * If the CPU was idle with dynamic ticks active, and there is no | |
333 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
334 | * RCU grace-period handling know that the CPU is active. | |
335 | */ | |
336 | void rcu_nmi_enter(void) | |
337 | { | |
338 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
339 | ||
e59fb312 PM |
340 | if (rdtp->dynticks_nmi_nesting == 0 && |
341 | (atomic_read(&rdtp->dynticks) & 0x1)) | |
64db4cff | 342 | return; |
e59fb312 PM |
343 | rdtp->dynticks_nmi_nesting++; |
344 | smp_mb__before_atomic_inc(); /* Force delay from prior write. */ | |
345 | atomic_inc(&rdtp->dynticks); | |
346 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ | |
347 | smp_mb__after_atomic_inc(); /* See above. */ | |
348 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); | |
64db4cff PM |
349 | } |
350 | ||
351 | /** | |
352 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
353 | * | |
354 | * If the CPU was idle with dynamic ticks active, and there is no | |
355 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
356 | * RCU grace-period handling know that the CPU is no longer active. | |
357 | */ | |
358 | void rcu_nmi_exit(void) | |
359 | { | |
360 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
361 | ||
e59fb312 PM |
362 | if (rdtp->dynticks_nmi_nesting == 0 || |
363 | --rdtp->dynticks_nmi_nesting != 0) | |
64db4cff | 364 | return; |
e59fb312 PM |
365 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ |
366 | smp_mb__before_atomic_inc(); /* See above. */ | |
367 | atomic_inc(&rdtp->dynticks); | |
368 | smp_mb__after_atomic_inc(); /* Force delay to next write. */ | |
369 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); | |
64db4cff PM |
370 | } |
371 | ||
372 | /** | |
373 | * rcu_irq_enter - inform RCU of entry to hard irq context | |
374 | * | |
375 | * If the CPU was idle with dynamic ticks active, this updates the | |
376 | * rdtp->dynticks to let the RCU handling know that the CPU is active. | |
377 | */ | |
378 | void rcu_irq_enter(void) | |
379 | { | |
e59fb312 | 380 | rcu_exit_nohz(); |
64db4cff PM |
381 | } |
382 | ||
383 | /** | |
384 | * rcu_irq_exit - inform RCU of exit from hard irq context | |
385 | * | |
386 | * If the CPU was idle with dynamic ticks active, update the rdp->dynticks | |
387 | * to put let the RCU handling be aware that the CPU is going back to idle | |
388 | * with no ticks. | |
389 | */ | |
390 | void rcu_irq_exit(void) | |
391 | { | |
e59fb312 | 392 | rcu_enter_nohz(); |
64db4cff PM |
393 | } |
394 | ||
64db4cff PM |
395 | #ifdef CONFIG_SMP |
396 | ||
64db4cff PM |
397 | /* |
398 | * Snapshot the specified CPU's dynticks counter so that we can later | |
399 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 400 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff PM |
401 | */ |
402 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | |
403 | { | |
e59fb312 PM |
404 | rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks); |
405 | return 0; | |
64db4cff PM |
406 | } |
407 | ||
408 | /* | |
409 | * Return true if the specified CPU has passed through a quiescent | |
410 | * state by virtue of being in or having passed through an dynticks | |
411 | * idle state since the last call to dyntick_save_progress_counter() | |
412 | * for this same CPU. | |
413 | */ | |
414 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
415 | { | |
e59fb312 PM |
416 | unsigned long curr; |
417 | unsigned long snap; | |
64db4cff | 418 | |
e59fb312 PM |
419 | curr = (unsigned long)atomic_add_return(0, &rdp->dynticks->dynticks); |
420 | snap = (unsigned long)rdp->dynticks_snap; | |
64db4cff PM |
421 | |
422 | /* | |
423 | * If the CPU passed through or entered a dynticks idle phase with | |
424 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
425 | * already acknowledged the request to pass through a quiescent | |
426 | * state. Either way, that CPU cannot possibly be in an RCU | |
427 | * read-side critical section that started before the beginning | |
428 | * of the current RCU grace period. | |
429 | */ | |
e59fb312 | 430 | if ((curr & 0x1) == 0 || ULONG_CMP_GE(curr, snap + 2)) { |
64db4cff PM |
431 | rdp->dynticks_fqs++; |
432 | return 1; | |
433 | } | |
434 | ||
435 | /* Go check for the CPU being offline. */ | |
436 | return rcu_implicit_offline_qs(rdp); | |
437 | } | |
438 | ||
439 | #endif /* #ifdef CONFIG_SMP */ | |
440 | ||
441 | #else /* #ifdef CONFIG_NO_HZ */ | |
442 | ||
64db4cff PM |
443 | #ifdef CONFIG_SMP |
444 | ||
64db4cff PM |
445 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
446 | { | |
447 | return 0; | |
448 | } | |
449 | ||
450 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
451 | { | |
452 | return rcu_implicit_offline_qs(rdp); | |
453 | } | |
454 | ||
455 | #endif /* #ifdef CONFIG_SMP */ | |
456 | ||
457 | #endif /* #else #ifdef CONFIG_NO_HZ */ | |
458 | ||
742734ee | 459 | int rcu_cpu_stall_suppress __read_mostly; |
c68de209 | 460 | |
64db4cff PM |
461 | static void record_gp_stall_check_time(struct rcu_state *rsp) |
462 | { | |
463 | rsp->gp_start = jiffies; | |
464 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK; | |
465 | } | |
466 | ||
467 | static void print_other_cpu_stall(struct rcu_state *rsp) | |
468 | { | |
469 | int cpu; | |
470 | long delta; | |
471 | unsigned long flags; | |
472 | struct rcu_node *rnp = rcu_get_root(rsp); | |
64db4cff PM |
473 | |
474 | /* Only let one CPU complain about others per time interval. */ | |
475 | ||
1304afb2 | 476 | raw_spin_lock_irqsave(&rnp->lock, flags); |
64db4cff | 477 | delta = jiffies - rsp->jiffies_stall; |
fc2219d4 | 478 | if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) { |
1304afb2 | 479 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
480 | return; |
481 | } | |
482 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
a0b6c9a7 PM |
483 | |
484 | /* | |
485 | * Now rat on any tasks that got kicked up to the root rcu_node | |
486 | * due to CPU offlining. | |
487 | */ | |
488 | rcu_print_task_stall(rnp); | |
1304afb2 | 489 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 490 | |
8cdd32a9 PM |
491 | /* |
492 | * OK, time to rat on our buddy... | |
493 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
494 | * RCU CPU stall warnings. | |
495 | */ | |
4300aa64 PM |
496 | printk(KERN_ERR "INFO: %s detected stalls on CPUs/tasks: {", |
497 | rsp->name); | |
a0b6c9a7 | 498 | rcu_for_each_leaf_node(rsp, rnp) { |
3acd9eb3 | 499 | raw_spin_lock_irqsave(&rnp->lock, flags); |
f41d911f | 500 | rcu_print_task_stall(rnp); |
3acd9eb3 | 501 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
a0b6c9a7 | 502 | if (rnp->qsmask == 0) |
64db4cff | 503 | continue; |
a0b6c9a7 PM |
504 | for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++) |
505 | if (rnp->qsmask & (1UL << cpu)) | |
506 | printk(" %d", rnp->grplo + cpu); | |
64db4cff | 507 | } |
4300aa64 | 508 | printk("} (detected by %d, t=%ld jiffies)\n", |
64db4cff | 509 | smp_processor_id(), (long)(jiffies - rsp->gp_start)); |
c1dc0b9c IM |
510 | trigger_all_cpu_backtrace(); |
511 | ||
1ed509a2 PM |
512 | /* If so configured, complain about tasks blocking the grace period. */ |
513 | ||
514 | rcu_print_detail_task_stall(rsp); | |
515 | ||
64db4cff PM |
516 | force_quiescent_state(rsp, 0); /* Kick them all. */ |
517 | } | |
518 | ||
519 | static void print_cpu_stall(struct rcu_state *rsp) | |
520 | { | |
521 | unsigned long flags; | |
522 | struct rcu_node *rnp = rcu_get_root(rsp); | |
523 | ||
8cdd32a9 PM |
524 | /* |
525 | * OK, time to rat on ourselves... | |
526 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
527 | * RCU CPU stall warnings. | |
528 | */ | |
4300aa64 PM |
529 | printk(KERN_ERR "INFO: %s detected stall on CPU %d (t=%lu jiffies)\n", |
530 | rsp->name, smp_processor_id(), jiffies - rsp->gp_start); | |
c1dc0b9c IM |
531 | trigger_all_cpu_backtrace(); |
532 | ||
1304afb2 | 533 | raw_spin_lock_irqsave(&rnp->lock, flags); |
20133cfc | 534 | if (ULONG_CMP_GE(jiffies, rsp->jiffies_stall)) |
64db4cff PM |
535 | rsp->jiffies_stall = |
536 | jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
1304afb2 | 537 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
c1dc0b9c | 538 | |
64db4cff PM |
539 | set_need_resched(); /* kick ourselves to get things going. */ |
540 | } | |
541 | ||
542 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
543 | { | |
544 | long delta; | |
545 | struct rcu_node *rnp; | |
546 | ||
742734ee | 547 | if (rcu_cpu_stall_suppress) |
c68de209 | 548 | return; |
4ee0a603 | 549 | delta = jiffies - ACCESS_ONCE(rsp->jiffies_stall); |
64db4cff | 550 | rnp = rdp->mynode; |
4ee0a603 | 551 | if ((ACCESS_ONCE(rnp->qsmask) & rdp->grpmask) && delta >= 0) { |
64db4cff PM |
552 | |
553 | /* We haven't checked in, so go dump stack. */ | |
554 | print_cpu_stall(rsp); | |
555 | ||
fc2219d4 | 556 | } else if (rcu_gp_in_progress(rsp) && delta >= RCU_STALL_RAT_DELAY) { |
64db4cff PM |
557 | |
558 | /* They had two time units to dump stack, so complain. */ | |
559 | print_other_cpu_stall(rsp); | |
560 | } | |
561 | } | |
562 | ||
c68de209 PM |
563 | static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr) |
564 | { | |
742734ee | 565 | rcu_cpu_stall_suppress = 1; |
c68de209 PM |
566 | return NOTIFY_DONE; |
567 | } | |
568 | ||
53d84e00 PM |
569 | /** |
570 | * rcu_cpu_stall_reset - prevent further stall warnings in current grace period | |
571 | * | |
572 | * Set the stall-warning timeout way off into the future, thus preventing | |
573 | * any RCU CPU stall-warning messages from appearing in the current set of | |
574 | * RCU grace periods. | |
575 | * | |
576 | * The caller must disable hard irqs. | |
577 | */ | |
578 | void rcu_cpu_stall_reset(void) | |
579 | { | |
580 | rcu_sched_state.jiffies_stall = jiffies + ULONG_MAX / 2; | |
581 | rcu_bh_state.jiffies_stall = jiffies + ULONG_MAX / 2; | |
582 | rcu_preempt_stall_reset(); | |
583 | } | |
584 | ||
c68de209 PM |
585 | static struct notifier_block rcu_panic_block = { |
586 | .notifier_call = rcu_panic, | |
587 | }; | |
588 | ||
589 | static void __init check_cpu_stall_init(void) | |
590 | { | |
591 | atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block); | |
592 | } | |
593 | ||
64db4cff PM |
594 | /* |
595 | * Update CPU-local rcu_data state to record the newly noticed grace period. | |
596 | * This is used both when we started the grace period and when we notice | |
9160306e PM |
597 | * that someone else started the grace period. The caller must hold the |
598 | * ->lock of the leaf rcu_node structure corresponding to the current CPU, | |
599 | * and must have irqs disabled. | |
64db4cff | 600 | */ |
9160306e PM |
601 | static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) |
602 | { | |
603 | if (rdp->gpnum != rnp->gpnum) { | |
121dfc4b PM |
604 | /* |
605 | * If the current grace period is waiting for this CPU, | |
606 | * set up to detect a quiescent state, otherwise don't | |
607 | * go looking for one. | |
608 | */ | |
9160306e | 609 | rdp->gpnum = rnp->gpnum; |
121dfc4b PM |
610 | if (rnp->qsmask & rdp->grpmask) { |
611 | rdp->qs_pending = 1; | |
612 | rdp->passed_quiesc = 0; | |
613 | } else | |
614 | rdp->qs_pending = 0; | |
9160306e PM |
615 | } |
616 | } | |
617 | ||
64db4cff PM |
618 | static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp) |
619 | { | |
9160306e PM |
620 | unsigned long flags; |
621 | struct rcu_node *rnp; | |
622 | ||
623 | local_irq_save(flags); | |
624 | rnp = rdp->mynode; | |
625 | if (rdp->gpnum == ACCESS_ONCE(rnp->gpnum) || /* outside lock. */ | |
1304afb2 | 626 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
9160306e PM |
627 | local_irq_restore(flags); |
628 | return; | |
629 | } | |
630 | __note_new_gpnum(rsp, rnp, rdp); | |
1304afb2 | 631 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
632 | } |
633 | ||
634 | /* | |
635 | * Did someone else start a new RCU grace period start since we last | |
636 | * checked? Update local state appropriately if so. Must be called | |
637 | * on the CPU corresponding to rdp. | |
638 | */ | |
639 | static int | |
640 | check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp) | |
641 | { | |
642 | unsigned long flags; | |
643 | int ret = 0; | |
644 | ||
645 | local_irq_save(flags); | |
646 | if (rdp->gpnum != rsp->gpnum) { | |
647 | note_new_gpnum(rsp, rdp); | |
648 | ret = 1; | |
649 | } | |
650 | local_irq_restore(flags); | |
651 | return ret; | |
652 | } | |
653 | ||
d09b62df PM |
654 | /* |
655 | * Advance this CPU's callbacks, but only if the current grace period | |
656 | * has ended. This may be called only from the CPU to whom the rdp | |
657 | * belongs. In addition, the corresponding leaf rcu_node structure's | |
658 | * ->lock must be held by the caller, with irqs disabled. | |
659 | */ | |
660 | static void | |
661 | __rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
662 | { | |
663 | /* Did another grace period end? */ | |
664 | if (rdp->completed != rnp->completed) { | |
665 | ||
666 | /* Advance callbacks. No harm if list empty. */ | |
667 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL]; | |
668 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL]; | |
669 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
670 | ||
671 | /* Remember that we saw this grace-period completion. */ | |
672 | rdp->completed = rnp->completed; | |
20377f32 | 673 | |
5ff8e6f0 FW |
674 | /* |
675 | * If we were in an extended quiescent state, we may have | |
121dfc4b | 676 | * missed some grace periods that others CPUs handled on |
5ff8e6f0 | 677 | * our behalf. Catch up with this state to avoid noting |
121dfc4b PM |
678 | * spurious new grace periods. If another grace period |
679 | * has started, then rnp->gpnum will have advanced, so | |
680 | * we will detect this later on. | |
5ff8e6f0 | 681 | */ |
121dfc4b | 682 | if (ULONG_CMP_LT(rdp->gpnum, rdp->completed)) |
5ff8e6f0 FW |
683 | rdp->gpnum = rdp->completed; |
684 | ||
20377f32 | 685 | /* |
121dfc4b PM |
686 | * If RCU does not need a quiescent state from this CPU, |
687 | * then make sure that this CPU doesn't go looking for one. | |
20377f32 | 688 | */ |
121dfc4b | 689 | if ((rnp->qsmask & rdp->grpmask) == 0) |
20377f32 | 690 | rdp->qs_pending = 0; |
d09b62df PM |
691 | } |
692 | } | |
693 | ||
694 | /* | |
695 | * Advance this CPU's callbacks, but only if the current grace period | |
696 | * has ended. This may be called only from the CPU to whom the rdp | |
697 | * belongs. | |
698 | */ | |
699 | static void | |
700 | rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp) | |
701 | { | |
702 | unsigned long flags; | |
703 | struct rcu_node *rnp; | |
704 | ||
705 | local_irq_save(flags); | |
706 | rnp = rdp->mynode; | |
707 | if (rdp->completed == ACCESS_ONCE(rnp->completed) || /* outside lock. */ | |
1304afb2 | 708 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
d09b62df PM |
709 | local_irq_restore(flags); |
710 | return; | |
711 | } | |
712 | __rcu_process_gp_end(rsp, rnp, rdp); | |
1304afb2 | 713 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
d09b62df PM |
714 | } |
715 | ||
716 | /* | |
717 | * Do per-CPU grace-period initialization for running CPU. The caller | |
718 | * must hold the lock of the leaf rcu_node structure corresponding to | |
719 | * this CPU. | |
720 | */ | |
721 | static void | |
722 | rcu_start_gp_per_cpu(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
723 | { | |
724 | /* Prior grace period ended, so advance callbacks for current CPU. */ | |
725 | __rcu_process_gp_end(rsp, rnp, rdp); | |
726 | ||
727 | /* | |
728 | * Because this CPU just now started the new grace period, we know | |
729 | * that all of its callbacks will be covered by this upcoming grace | |
730 | * period, even the ones that were registered arbitrarily recently. | |
731 | * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL. | |
732 | * | |
733 | * Other CPUs cannot be sure exactly when the grace period started. | |
734 | * Therefore, their recently registered callbacks must pass through | |
735 | * an additional RCU_NEXT_READY stage, so that they will be handled | |
736 | * by the next RCU grace period. | |
737 | */ | |
738 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
739 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
9160306e PM |
740 | |
741 | /* Set state so that this CPU will detect the next quiescent state. */ | |
742 | __note_new_gpnum(rsp, rnp, rdp); | |
d09b62df PM |
743 | } |
744 | ||
64db4cff PM |
745 | /* |
746 | * Start a new RCU grace period if warranted, re-initializing the hierarchy | |
747 | * in preparation for detecting the next grace period. The caller must hold | |
748 | * the root node's ->lock, which is released before return. Hard irqs must | |
749 | * be disabled. | |
750 | */ | |
751 | static void | |
752 | rcu_start_gp(struct rcu_state *rsp, unsigned long flags) | |
753 | __releases(rcu_get_root(rsp)->lock) | |
754 | { | |
394f99a9 | 755 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
64db4cff | 756 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff | 757 | |
07079d53 | 758 | if (!cpu_needs_another_gp(rsp, rdp) || rsp->fqs_active) { |
46a1e34e PM |
759 | if (cpu_needs_another_gp(rsp, rdp)) |
760 | rsp->fqs_need_gp = 1; | |
b32e9eb6 | 761 | if (rnp->completed == rsp->completed) { |
1304afb2 | 762 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
b32e9eb6 PM |
763 | return; |
764 | } | |
1304afb2 | 765 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
b32e9eb6 PM |
766 | |
767 | /* | |
768 | * Propagate new ->completed value to rcu_node structures | |
769 | * so that other CPUs don't have to wait until the start | |
770 | * of the next grace period to process their callbacks. | |
771 | */ | |
772 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
1304afb2 | 773 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
b32e9eb6 | 774 | rnp->completed = rsp->completed; |
1304afb2 | 775 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
b32e9eb6 PM |
776 | } |
777 | local_irq_restore(flags); | |
64db4cff PM |
778 | return; |
779 | } | |
780 | ||
781 | /* Advance to a new grace period and initialize state. */ | |
782 | rsp->gpnum++; | |
c3422bea | 783 | WARN_ON_ONCE(rsp->signaled == RCU_GP_INIT); |
64db4cff PM |
784 | rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */ |
785 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; | |
64db4cff | 786 | record_gp_stall_check_time(rsp); |
64db4cff | 787 | |
64db4cff PM |
788 | /* Special-case the common single-level case. */ |
789 | if (NUM_RCU_NODES == 1) { | |
b0e165c0 | 790 | rcu_preempt_check_blocked_tasks(rnp); |
28ecd580 | 791 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 792 | rnp->gpnum = rsp->gpnum; |
d09b62df | 793 | rnp->completed = rsp->completed; |
c12172c0 | 794 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */ |
d09b62df | 795 | rcu_start_gp_per_cpu(rsp, rnp, rdp); |
27f4d280 | 796 | rcu_preempt_boost_start_gp(rnp); |
1304afb2 | 797 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
798 | return; |
799 | } | |
800 | ||
1304afb2 | 801 | raw_spin_unlock(&rnp->lock); /* leave irqs disabled. */ |
64db4cff PM |
802 | |
803 | ||
804 | /* Exclude any concurrent CPU-hotplug operations. */ | |
1304afb2 | 805 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
64db4cff PM |
806 | |
807 | /* | |
b835db1f PM |
808 | * Set the quiescent-state-needed bits in all the rcu_node |
809 | * structures for all currently online CPUs in breadth-first | |
810 | * order, starting from the root rcu_node structure. This | |
811 | * operation relies on the layout of the hierarchy within the | |
812 | * rsp->node[] array. Note that other CPUs will access only | |
813 | * the leaves of the hierarchy, which still indicate that no | |
814 | * grace period is in progress, at least until the corresponding | |
815 | * leaf node has been initialized. In addition, we have excluded | |
816 | * CPU-hotplug operations. | |
64db4cff PM |
817 | * |
818 | * Note that the grace period cannot complete until we finish | |
819 | * the initialization process, as there will be at least one | |
820 | * qsmask bit set in the root node until that time, namely the | |
b835db1f PM |
821 | * one corresponding to this CPU, due to the fact that we have |
822 | * irqs disabled. | |
64db4cff | 823 | */ |
a0b6c9a7 | 824 | rcu_for_each_node_breadth_first(rsp, rnp) { |
1304afb2 | 825 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
b0e165c0 | 826 | rcu_preempt_check_blocked_tasks(rnp); |
49e29126 | 827 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 828 | rnp->gpnum = rsp->gpnum; |
d09b62df PM |
829 | rnp->completed = rsp->completed; |
830 | if (rnp == rdp->mynode) | |
831 | rcu_start_gp_per_cpu(rsp, rnp, rdp); | |
27f4d280 | 832 | rcu_preempt_boost_start_gp(rnp); |
1304afb2 | 833 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
834 | } |
835 | ||
83f5b01f | 836 | rnp = rcu_get_root(rsp); |
1304afb2 | 837 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff | 838 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */ |
1304afb2 PM |
839 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
840 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); | |
64db4cff PM |
841 | } |
842 | ||
f41d911f | 843 | /* |
d3f6bad3 PM |
844 | * Report a full set of quiescent states to the specified rcu_state |
845 | * data structure. This involves cleaning up after the prior grace | |
846 | * period and letting rcu_start_gp() start up the next grace period | |
847 | * if one is needed. Note that the caller must hold rnp->lock, as | |
848 | * required by rcu_start_gp(), which will release it. | |
f41d911f | 849 | */ |
d3f6bad3 | 850 | static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags) |
fc2219d4 | 851 | __releases(rcu_get_root(rsp)->lock) |
f41d911f | 852 | { |
fc2219d4 | 853 | WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); |
e59fb312 PM |
854 | |
855 | /* | |
856 | * Ensure that all grace-period and pre-grace-period activity | |
857 | * is seen before the assignment to rsp->completed. | |
858 | */ | |
859 | smp_mb(); /* See above block comment. */ | |
f41d911f | 860 | rsp->completed = rsp->gpnum; |
83f5b01f | 861 | rsp->signaled = RCU_GP_IDLE; |
f41d911f PM |
862 | rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */ |
863 | } | |
864 | ||
64db4cff | 865 | /* |
d3f6bad3 PM |
866 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
867 | * Allows quiescent states for a group of CPUs to be reported at one go | |
868 | * to the specified rcu_node structure, though all the CPUs in the group | |
869 | * must be represented by the same rcu_node structure (which need not be | |
870 | * a leaf rcu_node structure, though it often will be). That structure's | |
871 | * lock must be held upon entry, and it is released before return. | |
64db4cff PM |
872 | */ |
873 | static void | |
d3f6bad3 PM |
874 | rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp, |
875 | struct rcu_node *rnp, unsigned long flags) | |
64db4cff PM |
876 | __releases(rnp->lock) |
877 | { | |
28ecd580 PM |
878 | struct rcu_node *rnp_c; |
879 | ||
64db4cff PM |
880 | /* Walk up the rcu_node hierarchy. */ |
881 | for (;;) { | |
882 | if (!(rnp->qsmask & mask)) { | |
883 | ||
884 | /* Our bit has already been cleared, so done. */ | |
1304afb2 | 885 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
886 | return; |
887 | } | |
888 | rnp->qsmask &= ~mask; | |
27f4d280 | 889 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
64db4cff PM |
890 | |
891 | /* Other bits still set at this level, so done. */ | |
1304afb2 | 892 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
893 | return; |
894 | } | |
895 | mask = rnp->grpmask; | |
896 | if (rnp->parent == NULL) { | |
897 | ||
898 | /* No more levels. Exit loop holding root lock. */ | |
899 | ||
900 | break; | |
901 | } | |
1304afb2 | 902 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
28ecd580 | 903 | rnp_c = rnp; |
64db4cff | 904 | rnp = rnp->parent; |
1304afb2 | 905 | raw_spin_lock_irqsave(&rnp->lock, flags); |
28ecd580 | 906 | WARN_ON_ONCE(rnp_c->qsmask); |
64db4cff PM |
907 | } |
908 | ||
909 | /* | |
910 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 911 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 912 | * to clean up and start the next grace period if one is needed. |
64db4cff | 913 | */ |
d3f6bad3 | 914 | rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */ |
64db4cff PM |
915 | } |
916 | ||
917 | /* | |
d3f6bad3 PM |
918 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
919 | * structure. This must be either called from the specified CPU, or | |
920 | * called when the specified CPU is known to be offline (and when it is | |
921 | * also known that no other CPU is concurrently trying to help the offline | |
922 | * CPU). The lastcomp argument is used to make sure we are still in the | |
923 | * grace period of interest. We don't want to end the current grace period | |
924 | * based on quiescent states detected in an earlier grace period! | |
64db4cff PM |
925 | */ |
926 | static void | |
d3f6bad3 | 927 | rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp) |
64db4cff PM |
928 | { |
929 | unsigned long flags; | |
930 | unsigned long mask; | |
931 | struct rcu_node *rnp; | |
932 | ||
933 | rnp = rdp->mynode; | |
1304afb2 | 934 | raw_spin_lock_irqsave(&rnp->lock, flags); |
560d4bc0 | 935 | if (lastcomp != rnp->completed) { |
64db4cff PM |
936 | |
937 | /* | |
938 | * Someone beat us to it for this grace period, so leave. | |
939 | * The race with GP start is resolved by the fact that we | |
940 | * hold the leaf rcu_node lock, so that the per-CPU bits | |
941 | * cannot yet be initialized -- so we would simply find our | |
d3f6bad3 PM |
942 | * CPU's bit already cleared in rcu_report_qs_rnp() if this |
943 | * race occurred. | |
64db4cff PM |
944 | */ |
945 | rdp->passed_quiesc = 0; /* try again later! */ | |
1304afb2 | 946 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
947 | return; |
948 | } | |
949 | mask = rdp->grpmask; | |
950 | if ((rnp->qsmask & mask) == 0) { | |
1304afb2 | 951 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
952 | } else { |
953 | rdp->qs_pending = 0; | |
954 | ||
955 | /* | |
956 | * This GP can't end until cpu checks in, so all of our | |
957 | * callbacks can be processed during the next GP. | |
958 | */ | |
64db4cff PM |
959 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
960 | ||
d3f6bad3 | 961 | rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */ |
64db4cff PM |
962 | } |
963 | } | |
964 | ||
965 | /* | |
966 | * Check to see if there is a new grace period of which this CPU | |
967 | * is not yet aware, and if so, set up local rcu_data state for it. | |
968 | * Otherwise, see if this CPU has just passed through its first | |
969 | * quiescent state for this grace period, and record that fact if so. | |
970 | */ | |
971 | static void | |
972 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | |
973 | { | |
974 | /* If there is now a new grace period, record and return. */ | |
975 | if (check_for_new_grace_period(rsp, rdp)) | |
976 | return; | |
977 | ||
978 | /* | |
979 | * Does this CPU still need to do its part for current grace period? | |
980 | * If no, return and let the other CPUs do their part as well. | |
981 | */ | |
982 | if (!rdp->qs_pending) | |
983 | return; | |
984 | ||
985 | /* | |
986 | * Was there a quiescent state since the beginning of the grace | |
987 | * period? If no, then exit and wait for the next call. | |
988 | */ | |
989 | if (!rdp->passed_quiesc) | |
990 | return; | |
991 | ||
d3f6bad3 PM |
992 | /* |
993 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
994 | * judge of that). | |
995 | */ | |
996 | rcu_report_qs_rdp(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed); | |
64db4cff PM |
997 | } |
998 | ||
999 | #ifdef CONFIG_HOTPLUG_CPU | |
1000 | ||
e74f4c45 | 1001 | /* |
29494be7 LJ |
1002 | * Move a dying CPU's RCU callbacks to online CPU's callback list. |
1003 | * Synchronization is not required because this function executes | |
1004 | * in stop_machine() context. | |
e74f4c45 | 1005 | */ |
29494be7 | 1006 | static void rcu_send_cbs_to_online(struct rcu_state *rsp) |
e74f4c45 PM |
1007 | { |
1008 | int i; | |
29494be7 LJ |
1009 | /* current DYING CPU is cleared in the cpu_online_mask */ |
1010 | int receive_cpu = cpumask_any(cpu_online_mask); | |
394f99a9 | 1011 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
29494be7 | 1012 | struct rcu_data *receive_rdp = per_cpu_ptr(rsp->rda, receive_cpu); |
e74f4c45 PM |
1013 | |
1014 | if (rdp->nxtlist == NULL) | |
1015 | return; /* irqs disabled, so comparison is stable. */ | |
29494be7 LJ |
1016 | |
1017 | *receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist; | |
1018 | receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
1019 | receive_rdp->qlen += rdp->qlen; | |
1020 | receive_rdp->n_cbs_adopted += rdp->qlen; | |
1021 | rdp->n_cbs_orphaned += rdp->qlen; | |
1022 | ||
e74f4c45 PM |
1023 | rdp->nxtlist = NULL; |
1024 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
1025 | rdp->nxttail[i] = &rdp->nxtlist; | |
e74f4c45 | 1026 | rdp->qlen = 0; |
e74f4c45 PM |
1027 | } |
1028 | ||
64db4cff PM |
1029 | /* |
1030 | * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy | |
1031 | * and move all callbacks from the outgoing CPU to the current one. | |
a26ac245 PM |
1032 | * There can only be one CPU hotplug operation at a time, so no other |
1033 | * CPU can be attempting to update rcu_cpu_kthread_task. | |
64db4cff PM |
1034 | */ |
1035 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp) | |
1036 | { | |
64db4cff | 1037 | unsigned long flags; |
64db4cff | 1038 | unsigned long mask; |
d9a3da06 | 1039 | int need_report = 0; |
394f99a9 | 1040 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff | 1041 | struct rcu_node *rnp; |
a26ac245 PM |
1042 | struct task_struct *t; |
1043 | ||
1044 | /* Stop the CPU's kthread. */ | |
1045 | t = per_cpu(rcu_cpu_kthread_task, cpu); | |
1046 | if (t != NULL) { | |
1047 | per_cpu(rcu_cpu_kthread_task, cpu) = NULL; | |
1048 | kthread_stop(t); | |
1049 | } | |
64db4cff PM |
1050 | |
1051 | /* Exclude any attempts to start a new grace period. */ | |
1304afb2 | 1052 | raw_spin_lock_irqsave(&rsp->onofflock, flags); |
64db4cff PM |
1053 | |
1054 | /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */ | |
28ecd580 | 1055 | rnp = rdp->mynode; /* this is the outgoing CPU's rnp. */ |
64db4cff PM |
1056 | mask = rdp->grpmask; /* rnp->grplo is constant. */ |
1057 | do { | |
1304afb2 | 1058 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
1059 | rnp->qsmaskinit &= ~mask; |
1060 | if (rnp->qsmaskinit != 0) { | |
b668c9cf | 1061 | if (rnp != rdp->mynode) |
1304afb2 | 1062 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
1063 | break; |
1064 | } | |
b668c9cf | 1065 | if (rnp == rdp->mynode) |
d9a3da06 | 1066 | need_report = rcu_preempt_offline_tasks(rsp, rnp, rdp); |
b668c9cf | 1067 | else |
1304afb2 | 1068 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff | 1069 | mask = rnp->grpmask; |
64db4cff PM |
1070 | rnp = rnp->parent; |
1071 | } while (rnp != NULL); | |
64db4cff | 1072 | |
b668c9cf PM |
1073 | /* |
1074 | * We still hold the leaf rcu_node structure lock here, and | |
1075 | * irqs are still disabled. The reason for this subterfuge is | |
d3f6bad3 PM |
1076 | * because invoking rcu_report_unblock_qs_rnp() with ->onofflock |
1077 | * held leads to deadlock. | |
b668c9cf | 1078 | */ |
1304afb2 | 1079 | raw_spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ |
b668c9cf | 1080 | rnp = rdp->mynode; |
d9a3da06 | 1081 | if (need_report & RCU_OFL_TASKS_NORM_GP) |
d3f6bad3 | 1082 | rcu_report_unblock_qs_rnp(rnp, flags); |
b668c9cf | 1083 | else |
1304afb2 | 1084 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
d9a3da06 PM |
1085 | if (need_report & RCU_OFL_TASKS_EXP_GP) |
1086 | rcu_report_exp_rnp(rsp, rnp); | |
a26ac245 PM |
1087 | |
1088 | /* | |
1089 | * If there are no more online CPUs for this rcu_node structure, | |
1090 | * kill the rcu_node structure's kthread. Otherwise, adjust its | |
1091 | * affinity. | |
1092 | */ | |
1093 | t = rnp->node_kthread_task; | |
1094 | if (t != NULL && | |
1095 | rnp->qsmaskinit == 0) { | |
27f4d280 | 1096 | raw_spin_lock_irqsave(&rnp->lock, flags); |
a26ac245 | 1097 | rnp->node_kthread_task = NULL; |
27f4d280 PM |
1098 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1099 | kthread_stop(t); | |
1100 | rcu_stop_boost_kthread(rnp); | |
a26ac245 | 1101 | } else |
0f962a5e | 1102 | rcu_node_kthread_setaffinity(rnp, -1); |
64db4cff PM |
1103 | } |
1104 | ||
1105 | /* | |
1106 | * Remove the specified CPU from the RCU hierarchy and move any pending | |
1107 | * callbacks that it might have to the current CPU. This code assumes | |
1108 | * that at least one CPU in the system will remain running at all times. | |
1109 | * Any attempt to offline -all- CPUs is likely to strand RCU callbacks. | |
1110 | */ | |
1111 | static void rcu_offline_cpu(int cpu) | |
1112 | { | |
d6714c22 | 1113 | __rcu_offline_cpu(cpu, &rcu_sched_state); |
64db4cff | 1114 | __rcu_offline_cpu(cpu, &rcu_bh_state); |
33f76148 | 1115 | rcu_preempt_offline_cpu(cpu); |
64db4cff PM |
1116 | } |
1117 | ||
1118 | #else /* #ifdef CONFIG_HOTPLUG_CPU */ | |
1119 | ||
29494be7 | 1120 | static void rcu_send_cbs_to_online(struct rcu_state *rsp) |
e74f4c45 PM |
1121 | { |
1122 | } | |
1123 | ||
64db4cff PM |
1124 | static void rcu_offline_cpu(int cpu) |
1125 | { | |
1126 | } | |
1127 | ||
1128 | #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */ | |
1129 | ||
1130 | /* | |
1131 | * Invoke any RCU callbacks that have made it to the end of their grace | |
1132 | * period. Thottle as specified by rdp->blimit. | |
1133 | */ | |
37c72e56 | 1134 | static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
1135 | { |
1136 | unsigned long flags; | |
1137 | struct rcu_head *next, *list, **tail; | |
1138 | int count; | |
1139 | ||
1140 | /* If no callbacks are ready, just return.*/ | |
1141 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) | |
1142 | return; | |
1143 | ||
1144 | /* | |
1145 | * Extract the list of ready callbacks, disabling to prevent | |
1146 | * races with call_rcu() from interrupt handlers. | |
1147 | */ | |
1148 | local_irq_save(flags); | |
1149 | list = rdp->nxtlist; | |
1150 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; | |
1151 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | |
1152 | tail = rdp->nxttail[RCU_DONE_TAIL]; | |
1153 | for (count = RCU_NEXT_SIZE - 1; count >= 0; count--) | |
1154 | if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL]) | |
1155 | rdp->nxttail[count] = &rdp->nxtlist; | |
1156 | local_irq_restore(flags); | |
1157 | ||
1158 | /* Invoke callbacks. */ | |
1159 | count = 0; | |
1160 | while (list) { | |
1161 | next = list->next; | |
1162 | prefetch(next); | |
551d55a9 | 1163 | debug_rcu_head_unqueue(list); |
64db4cff PM |
1164 | list->func(list); |
1165 | list = next; | |
1166 | if (++count >= rdp->blimit) | |
1167 | break; | |
1168 | } | |
1169 | ||
1170 | local_irq_save(flags); | |
1171 | ||
1172 | /* Update count, and requeue any remaining callbacks. */ | |
1173 | rdp->qlen -= count; | |
269dcc1c | 1174 | rdp->n_cbs_invoked += count; |
64db4cff PM |
1175 | if (list != NULL) { |
1176 | *tail = rdp->nxtlist; | |
1177 | rdp->nxtlist = list; | |
1178 | for (count = 0; count < RCU_NEXT_SIZE; count++) | |
1179 | if (&rdp->nxtlist == rdp->nxttail[count]) | |
1180 | rdp->nxttail[count] = tail; | |
1181 | else | |
1182 | break; | |
1183 | } | |
1184 | ||
1185 | /* Reinstate batch limit if we have worked down the excess. */ | |
1186 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) | |
1187 | rdp->blimit = blimit; | |
1188 | ||
37c72e56 PM |
1189 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
1190 | if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) { | |
1191 | rdp->qlen_last_fqs_check = 0; | |
1192 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1193 | } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark) | |
1194 | rdp->qlen_last_fqs_check = rdp->qlen; | |
1195 | ||
64db4cff PM |
1196 | local_irq_restore(flags); |
1197 | ||
1198 | /* Re-raise the RCU softirq if there are callbacks remaining. */ | |
1199 | if (cpu_has_callbacks_ready_to_invoke(rdp)) | |
27f4d280 | 1200 | invoke_rcu_cpu_kthread(); |
64db4cff PM |
1201 | } |
1202 | ||
1203 | /* | |
1204 | * Check to see if this CPU is in a non-context-switch quiescent state | |
1205 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
1206 | * Also schedule the RCU softirq handler. | |
1207 | * | |
1208 | * This function must be called with hardirqs disabled. It is normally | |
1209 | * invoked from the scheduling-clock interrupt. If rcu_pending returns | |
1210 | * false, there is no point in invoking rcu_check_callbacks(). | |
1211 | */ | |
1212 | void rcu_check_callbacks(int cpu, int user) | |
1213 | { | |
1214 | if (user || | |
a6826048 PM |
1215 | (idle_cpu(cpu) && rcu_scheduler_active && |
1216 | !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) { | |
64db4cff PM |
1217 | |
1218 | /* | |
1219 | * Get here if this CPU took its interrupt from user | |
1220 | * mode or from the idle loop, and if this is not a | |
1221 | * nested interrupt. In this case, the CPU is in | |
d6714c22 | 1222 | * a quiescent state, so note it. |
64db4cff PM |
1223 | * |
1224 | * No memory barrier is required here because both | |
d6714c22 PM |
1225 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
1226 | * variables that other CPUs neither access nor modify, | |
1227 | * at least not while the corresponding CPU is online. | |
64db4cff PM |
1228 | */ |
1229 | ||
d6714c22 PM |
1230 | rcu_sched_qs(cpu); |
1231 | rcu_bh_qs(cpu); | |
64db4cff PM |
1232 | |
1233 | } else if (!in_softirq()) { | |
1234 | ||
1235 | /* | |
1236 | * Get here if this CPU did not take its interrupt from | |
1237 | * softirq, in other words, if it is not interrupting | |
1238 | * a rcu_bh read-side critical section. This is an _bh | |
d6714c22 | 1239 | * critical section, so note it. |
64db4cff PM |
1240 | */ |
1241 | ||
d6714c22 | 1242 | rcu_bh_qs(cpu); |
64db4cff | 1243 | } |
f41d911f | 1244 | rcu_preempt_check_callbacks(cpu); |
d21670ac | 1245 | if (rcu_pending(cpu)) |
27f4d280 | 1246 | invoke_rcu_cpu_kthread(); |
64db4cff PM |
1247 | } |
1248 | ||
1249 | #ifdef CONFIG_SMP | |
1250 | ||
1251 | /* | |
1252 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
1253 | * have not yet encountered a quiescent state, using the function specified. | |
27f4d280 PM |
1254 | * Also initiate boosting for any threads blocked on the root rcu_node. |
1255 | * | |
ee47eb9f | 1256 | * The caller must have suppressed start of new grace periods. |
64db4cff | 1257 | */ |
45f014c5 | 1258 | static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *)) |
64db4cff PM |
1259 | { |
1260 | unsigned long bit; | |
1261 | int cpu; | |
1262 | unsigned long flags; | |
1263 | unsigned long mask; | |
a0b6c9a7 | 1264 | struct rcu_node *rnp; |
64db4cff | 1265 | |
a0b6c9a7 | 1266 | rcu_for_each_leaf_node(rsp, rnp) { |
64db4cff | 1267 | mask = 0; |
1304afb2 | 1268 | raw_spin_lock_irqsave(&rnp->lock, flags); |
ee47eb9f | 1269 | if (!rcu_gp_in_progress(rsp)) { |
1304afb2 | 1270 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
0f10dc82 | 1271 | return; |
64db4cff | 1272 | } |
a0b6c9a7 | 1273 | if (rnp->qsmask == 0) { |
27f4d280 | 1274 | rcu_initiate_boost(rnp); |
1304afb2 | 1275 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1276 | continue; |
1277 | } | |
a0b6c9a7 | 1278 | cpu = rnp->grplo; |
64db4cff | 1279 | bit = 1; |
a0b6c9a7 | 1280 | for (; cpu <= rnp->grphi; cpu++, bit <<= 1) { |
394f99a9 LJ |
1281 | if ((rnp->qsmask & bit) != 0 && |
1282 | f(per_cpu_ptr(rsp->rda, cpu))) | |
64db4cff PM |
1283 | mask |= bit; |
1284 | } | |
45f014c5 | 1285 | if (mask != 0) { |
64db4cff | 1286 | |
d3f6bad3 PM |
1287 | /* rcu_report_qs_rnp() releases rnp->lock. */ |
1288 | rcu_report_qs_rnp(mask, rsp, rnp, flags); | |
64db4cff PM |
1289 | continue; |
1290 | } | |
1304afb2 | 1291 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 1292 | } |
27f4d280 PM |
1293 | rnp = rcu_get_root(rsp); |
1294 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
1295 | if (rnp->qsmask == 0) | |
1296 | rcu_initiate_boost(rnp); | |
1297 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
64db4cff PM |
1298 | } |
1299 | ||
1300 | /* | |
1301 | * Force quiescent states on reluctant CPUs, and also detect which | |
1302 | * CPUs are in dyntick-idle mode. | |
1303 | */ | |
1304 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1305 | { | |
1306 | unsigned long flags; | |
64db4cff | 1307 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff | 1308 | |
fc2219d4 | 1309 | if (!rcu_gp_in_progress(rsp)) |
64db4cff | 1310 | return; /* No grace period in progress, nothing to force. */ |
1304afb2 | 1311 | if (!raw_spin_trylock_irqsave(&rsp->fqslock, flags)) { |
64db4cff PM |
1312 | rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ |
1313 | return; /* Someone else is already on the job. */ | |
1314 | } | |
20133cfc | 1315 | if (relaxed && ULONG_CMP_GE(rsp->jiffies_force_qs, jiffies)) |
f96e9232 | 1316 | goto unlock_fqs_ret; /* no emergency and done recently. */ |
64db4cff | 1317 | rsp->n_force_qs++; |
1304afb2 | 1318 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
64db4cff | 1319 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
560d4bc0 | 1320 | if(!rcu_gp_in_progress(rsp)) { |
64db4cff | 1321 | rsp->n_force_qs_ngp++; |
1304afb2 | 1322 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1323 | goto unlock_fqs_ret; /* no GP in progress, time updated. */ |
64db4cff | 1324 | } |
07079d53 | 1325 | rsp->fqs_active = 1; |
f3a8b5c6 | 1326 | switch (rsp->signaled) { |
83f5b01f | 1327 | case RCU_GP_IDLE: |
64db4cff PM |
1328 | case RCU_GP_INIT: |
1329 | ||
83f5b01f | 1330 | break; /* grace period idle or initializing, ignore. */ |
64db4cff PM |
1331 | |
1332 | case RCU_SAVE_DYNTICK: | |
64db4cff PM |
1333 | if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK) |
1334 | break; /* So gcc recognizes the dead code. */ | |
1335 | ||
f261414f LJ |
1336 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
1337 | ||
64db4cff | 1338 | /* Record dyntick-idle state. */ |
45f014c5 | 1339 | force_qs_rnp(rsp, dyntick_save_progress_counter); |
1304afb2 | 1340 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
ee47eb9f | 1341 | if (rcu_gp_in_progress(rsp)) |
64db4cff | 1342 | rsp->signaled = RCU_FORCE_QS; |
ee47eb9f | 1343 | break; |
64db4cff PM |
1344 | |
1345 | case RCU_FORCE_QS: | |
1346 | ||
1347 | /* Check dyntick-idle state, send IPI to laggarts. */ | |
1304afb2 | 1348 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
45f014c5 | 1349 | force_qs_rnp(rsp, rcu_implicit_dynticks_qs); |
64db4cff PM |
1350 | |
1351 | /* Leave state in case more forcing is required. */ | |
1352 | ||
1304afb2 | 1353 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
f96e9232 | 1354 | break; |
64db4cff | 1355 | } |
07079d53 | 1356 | rsp->fqs_active = 0; |
46a1e34e | 1357 | if (rsp->fqs_need_gp) { |
1304afb2 | 1358 | raw_spin_unlock(&rsp->fqslock); /* irqs remain disabled */ |
46a1e34e PM |
1359 | rsp->fqs_need_gp = 0; |
1360 | rcu_start_gp(rsp, flags); /* releases rnp->lock */ | |
1361 | return; | |
1362 | } | |
1304afb2 | 1363 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1364 | unlock_fqs_ret: |
1304afb2 | 1365 | raw_spin_unlock_irqrestore(&rsp->fqslock, flags); |
64db4cff PM |
1366 | } |
1367 | ||
1368 | #else /* #ifdef CONFIG_SMP */ | |
1369 | ||
1370 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1371 | { | |
1372 | set_need_resched(); | |
1373 | } | |
1374 | ||
1375 | #endif /* #else #ifdef CONFIG_SMP */ | |
1376 | ||
1377 | /* | |
1378 | * This does the RCU processing work from softirq context for the | |
1379 | * specified rcu_state and rcu_data structures. This may be called | |
1380 | * only from the CPU to whom the rdp belongs. | |
1381 | */ | |
1382 | static void | |
1383 | __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) | |
1384 | { | |
1385 | unsigned long flags; | |
1386 | ||
2e597558 PM |
1387 | WARN_ON_ONCE(rdp->beenonline == 0); |
1388 | ||
64db4cff PM |
1389 | /* |
1390 | * If an RCU GP has gone long enough, go check for dyntick | |
1391 | * idle CPUs and, if needed, send resched IPIs. | |
1392 | */ | |
20133cfc | 1393 | if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
64db4cff PM |
1394 | force_quiescent_state(rsp, 1); |
1395 | ||
1396 | /* | |
1397 | * Advance callbacks in response to end of earlier grace | |
1398 | * period that some other CPU ended. | |
1399 | */ | |
1400 | rcu_process_gp_end(rsp, rdp); | |
1401 | ||
1402 | /* Update RCU state based on any recent quiescent states. */ | |
1403 | rcu_check_quiescent_state(rsp, rdp); | |
1404 | ||
1405 | /* Does this CPU require a not-yet-started grace period? */ | |
1406 | if (cpu_needs_another_gp(rsp, rdp)) { | |
1304afb2 | 1407 | raw_spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags); |
64db4cff PM |
1408 | rcu_start_gp(rsp, flags); /* releases above lock */ |
1409 | } | |
1410 | ||
1411 | /* If there are callbacks ready, invoke them. */ | |
37c72e56 | 1412 | rcu_do_batch(rsp, rdp); |
64db4cff PM |
1413 | } |
1414 | ||
1415 | /* | |
1416 | * Do softirq processing for the current CPU. | |
1417 | */ | |
a26ac245 | 1418 | static void rcu_process_callbacks(void) |
64db4cff | 1419 | { |
d6714c22 PM |
1420 | __rcu_process_callbacks(&rcu_sched_state, |
1421 | &__get_cpu_var(rcu_sched_data)); | |
64db4cff | 1422 | __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); |
f41d911f | 1423 | rcu_preempt_process_callbacks(); |
a47cd880 PM |
1424 | |
1425 | /* If we are last CPU on way to dyntick-idle mode, accelerate it. */ | |
1426 | rcu_needs_cpu_flush(); | |
64db4cff PM |
1427 | } |
1428 | ||
a26ac245 PM |
1429 | /* |
1430 | * Wake up the current CPU's kthread. This replaces raise_softirq() | |
1431 | * in earlier versions of RCU. Note that because we are running on | |
1432 | * the current CPU with interrupts disabled, the rcu_cpu_kthread_task | |
1433 | * cannot disappear out from under us. | |
1434 | */ | |
27f4d280 | 1435 | static void invoke_rcu_cpu_kthread(void) |
a26ac245 PM |
1436 | { |
1437 | unsigned long flags; | |
1438 | wait_queue_head_t *q; | |
1439 | int cpu; | |
1440 | ||
1441 | local_irq_save(flags); | |
1442 | cpu = smp_processor_id(); | |
1443 | per_cpu(rcu_cpu_has_work, cpu) = 1; | |
1444 | if (per_cpu(rcu_cpu_kthread_task, cpu) == NULL) { | |
1445 | local_irq_restore(flags); | |
1446 | return; | |
1447 | } | |
1448 | q = &per_cpu(rcu_cpu_wq, cpu); | |
1449 | wake_up(q); | |
1450 | local_irq_restore(flags); | |
1451 | } | |
1452 | ||
27f4d280 PM |
1453 | /* |
1454 | * Wake up the specified per-rcu_node-structure kthread. | |
1455 | * The caller must hold ->lock. | |
1456 | */ | |
1457 | static void invoke_rcu_node_kthread(struct rcu_node *rnp) | |
1458 | { | |
1459 | struct task_struct *t; | |
1460 | ||
1461 | t = rnp->node_kthread_task; | |
1462 | if (t != NULL) | |
1463 | wake_up_process(t); | |
1464 | } | |
1465 | ||
a26ac245 PM |
1466 | /* |
1467 | * Timer handler to initiate the waking up of per-CPU kthreads that | |
1468 | * have yielded the CPU due to excess numbers of RCU callbacks. | |
27f4d280 PM |
1469 | * We wake up the per-rcu_node kthread, which in turn will wake up |
1470 | * the booster kthread. | |
a26ac245 PM |
1471 | */ |
1472 | static void rcu_cpu_kthread_timer(unsigned long arg) | |
1473 | { | |
1474 | unsigned long flags; | |
27f4d280 | 1475 | struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, arg); |
a26ac245 | 1476 | struct rcu_node *rnp = rdp->mynode; |
a26ac245 PM |
1477 | |
1478 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
1479 | rnp->wakemask |= rdp->grpmask; | |
27f4d280 | 1480 | invoke_rcu_node_kthread(rnp); |
a26ac245 PM |
1481 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1482 | } | |
1483 | ||
1484 | /* | |
1485 | * Drop to non-real-time priority and yield, but only after posting a | |
1486 | * timer that will cause us to regain our real-time priority if we | |
1487 | * remain preempted. Either way, we restore our real-time priority | |
1488 | * before returning. | |
1489 | */ | |
27f4d280 | 1490 | static void rcu_yield(void (*f)(unsigned long), unsigned long arg) |
a26ac245 | 1491 | { |
a26ac245 PM |
1492 | struct sched_param sp; |
1493 | struct timer_list yield_timer; | |
1494 | ||
27f4d280 | 1495 | setup_timer_on_stack(&yield_timer, f, arg); |
a26ac245 PM |
1496 | mod_timer(&yield_timer, jiffies + 2); |
1497 | sp.sched_priority = 0; | |
1498 | sched_setscheduler_nocheck(current, SCHED_NORMAL, &sp); | |
1499 | schedule(); | |
1500 | sp.sched_priority = RCU_KTHREAD_PRIO; | |
1501 | sched_setscheduler_nocheck(current, SCHED_FIFO, &sp); | |
1502 | del_timer(&yield_timer); | |
1503 | } | |
1504 | ||
1505 | /* | |
1506 | * Handle cases where the rcu_cpu_kthread() ends up on the wrong CPU. | |
1507 | * This can happen while the corresponding CPU is either coming online | |
1508 | * or going offline. We cannot wait until the CPU is fully online | |
1509 | * before starting the kthread, because the various notifier functions | |
1510 | * can wait for RCU grace periods. So we park rcu_cpu_kthread() until | |
1511 | * the corresponding CPU is online. | |
1512 | * | |
1513 | * Return 1 if the kthread needs to stop, 0 otherwise. | |
1514 | * | |
1515 | * Caller must disable bh. This function can momentarily enable it. | |
1516 | */ | |
1517 | static int rcu_cpu_kthread_should_stop(int cpu) | |
1518 | { | |
1519 | while (cpu_is_offline(cpu) || | |
1520 | !cpumask_equal(¤t->cpus_allowed, cpumask_of(cpu)) || | |
1521 | smp_processor_id() != cpu) { | |
1522 | if (kthread_should_stop()) | |
1523 | return 1; | |
1524 | local_bh_enable(); | |
1525 | schedule_timeout_uninterruptible(1); | |
1526 | if (!cpumask_equal(¤t->cpus_allowed, cpumask_of(cpu))) | |
1527 | set_cpus_allowed_ptr(current, cpumask_of(cpu)); | |
1528 | local_bh_disable(); | |
1529 | } | |
1530 | return 0; | |
1531 | } | |
1532 | ||
1533 | /* | |
1534 | * Per-CPU kernel thread that invokes RCU callbacks. This replaces the | |
1535 | * earlier RCU softirq. | |
1536 | */ | |
1537 | static int rcu_cpu_kthread(void *arg) | |
1538 | { | |
1539 | int cpu = (int)(long)arg; | |
1540 | unsigned long flags; | |
1541 | int spincnt = 0; | |
1542 | wait_queue_head_t *wqp = &per_cpu(rcu_cpu_wq, cpu); | |
1543 | char work; | |
1544 | char *workp = &per_cpu(rcu_cpu_has_work, cpu); | |
1545 | ||
1546 | for (;;) { | |
1547 | wait_event_interruptible(*wqp, | |
1548 | *workp != 0 || kthread_should_stop()); | |
1549 | local_bh_disable(); | |
1550 | if (rcu_cpu_kthread_should_stop(cpu)) { | |
1551 | local_bh_enable(); | |
1552 | break; | |
1553 | } | |
1554 | local_irq_save(flags); | |
1555 | work = *workp; | |
1556 | *workp = 0; | |
1557 | local_irq_restore(flags); | |
1558 | if (work) | |
1559 | rcu_process_callbacks(); | |
1560 | local_bh_enable(); | |
1561 | if (*workp != 0) | |
1562 | spincnt++; | |
1563 | else | |
1564 | spincnt = 0; | |
1565 | if (spincnt > 10) { | |
27f4d280 | 1566 | rcu_yield(rcu_cpu_kthread_timer, (unsigned long)cpu); |
a26ac245 PM |
1567 | spincnt = 0; |
1568 | } | |
1569 | } | |
1570 | return 0; | |
1571 | } | |
1572 | ||
1573 | /* | |
1574 | * Spawn a per-CPU kthread, setting up affinity and priority. | |
1575 | * Because the CPU hotplug lock is held, no other CPU will be attempting | |
1576 | * to manipulate rcu_cpu_kthread_task. There might be another CPU | |
1577 | * attempting to access it during boot, but the locking in kthread_bind() | |
1578 | * will enforce sufficient ordering. | |
1579 | */ | |
1580 | static int __cpuinit rcu_spawn_one_cpu_kthread(int cpu) | |
1581 | { | |
1582 | struct sched_param sp; | |
1583 | struct task_struct *t; | |
1584 | ||
1585 | if (!rcu_kthreads_spawnable || | |
1586 | per_cpu(rcu_cpu_kthread_task, cpu) != NULL) | |
1587 | return 0; | |
1588 | t = kthread_create(rcu_cpu_kthread, (void *)(long)cpu, "rcuc%d", cpu); | |
1589 | if (IS_ERR(t)) | |
1590 | return PTR_ERR(t); | |
1591 | kthread_bind(t, cpu); | |
1592 | WARN_ON_ONCE(per_cpu(rcu_cpu_kthread_task, cpu) != NULL); | |
1593 | per_cpu(rcu_cpu_kthread_task, cpu) = t; | |
1594 | wake_up_process(t); | |
1595 | sp.sched_priority = RCU_KTHREAD_PRIO; | |
1596 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); | |
1597 | return 0; | |
1598 | } | |
1599 | ||
1600 | /* | |
1601 | * Per-rcu_node kthread, which is in charge of waking up the per-CPU | |
1602 | * kthreads when needed. We ignore requests to wake up kthreads | |
1603 | * for offline CPUs, which is OK because force_quiescent_state() | |
1604 | * takes care of this case. | |
1605 | */ | |
1606 | static int rcu_node_kthread(void *arg) | |
1607 | { | |
1608 | int cpu; | |
1609 | unsigned long flags; | |
1610 | unsigned long mask; | |
1611 | struct rcu_node *rnp = (struct rcu_node *)arg; | |
1612 | struct sched_param sp; | |
1613 | struct task_struct *t; | |
1614 | ||
1615 | for (;;) { | |
1616 | wait_event_interruptible(rnp->node_wq, rnp->wakemask != 0 || | |
1617 | kthread_should_stop()); | |
1618 | if (kthread_should_stop()) | |
1619 | break; | |
1620 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
1621 | mask = rnp->wakemask; | |
1622 | rnp->wakemask = 0; | |
27f4d280 | 1623 | rcu_initiate_boost(rnp); |
a26ac245 PM |
1624 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1625 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1) { | |
1626 | if ((mask & 0x1) == 0) | |
1627 | continue; | |
1628 | preempt_disable(); | |
1629 | t = per_cpu(rcu_cpu_kthread_task, cpu); | |
1630 | if (!cpu_online(cpu) || t == NULL) { | |
1631 | preempt_enable(); | |
1632 | continue; | |
1633 | } | |
1634 | per_cpu(rcu_cpu_has_work, cpu) = 1; | |
1635 | sp.sched_priority = RCU_KTHREAD_PRIO; | |
1636 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); | |
1637 | preempt_enable(); | |
1638 | } | |
1639 | } | |
1640 | return 0; | |
1641 | } | |
1642 | ||
1643 | /* | |
1644 | * Set the per-rcu_node kthread's affinity to cover all CPUs that are | |
27f4d280 PM |
1645 | * served by the rcu_node in question. The CPU hotplug lock is still |
1646 | * held, so the value of rnp->qsmaskinit will be stable. | |
0f962a5e PM |
1647 | * |
1648 | * We don't include outgoingcpu in the affinity set, use -1 if there is | |
1649 | * no outgoing CPU. If there are no CPUs left in the affinity set, | |
1650 | * this function allows the kthread to execute on any CPU. | |
a26ac245 | 1651 | */ |
0f962a5e | 1652 | static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) |
a26ac245 PM |
1653 | { |
1654 | cpumask_var_t cm; | |
1655 | int cpu; | |
1656 | unsigned long mask = rnp->qsmaskinit; | |
1657 | ||
27f4d280 | 1658 | if (rnp->node_kthread_task == NULL || mask == 0) |
a26ac245 PM |
1659 | return; |
1660 | if (!alloc_cpumask_var(&cm, GFP_KERNEL)) | |
1661 | return; | |
1662 | cpumask_clear(cm); | |
1663 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1) | |
0f962a5e | 1664 | if ((mask & 0x1) && cpu != outgoingcpu) |
a26ac245 | 1665 | cpumask_set_cpu(cpu, cm); |
0f962a5e PM |
1666 | if (cpumask_weight(cm) == 0) { |
1667 | cpumask_setall(cm); | |
1668 | for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++) | |
1669 | cpumask_clear_cpu(cpu, cm); | |
1670 | WARN_ON_ONCE(cpumask_weight(cm) == 0); | |
1671 | } | |
a26ac245 | 1672 | set_cpus_allowed_ptr(rnp->node_kthread_task, cm); |
27f4d280 | 1673 | rcu_boost_kthread_setaffinity(rnp, cm); |
a26ac245 PM |
1674 | free_cpumask_var(cm); |
1675 | } | |
1676 | ||
1677 | /* | |
1678 | * Spawn a per-rcu_node kthread, setting priority and affinity. | |
27f4d280 PM |
1679 | * Called during boot before online/offline can happen, or, if |
1680 | * during runtime, with the main CPU-hotplug locks held. So only | |
1681 | * one of these can be executing at a time. | |
a26ac245 PM |
1682 | */ |
1683 | static int __cpuinit rcu_spawn_one_node_kthread(struct rcu_state *rsp, | |
1684 | struct rcu_node *rnp) | |
1685 | { | |
27f4d280 | 1686 | unsigned long flags; |
a26ac245 PM |
1687 | int rnp_index = rnp - &rsp->node[0]; |
1688 | struct sched_param sp; | |
1689 | struct task_struct *t; | |
1690 | ||
1691 | if (!rcu_kthreads_spawnable || | |
27f4d280 | 1692 | rnp->qsmaskinit == 0) |
a26ac245 | 1693 | return 0; |
27f4d280 PM |
1694 | if (rnp->node_kthread_task == NULL) { |
1695 | t = kthread_create(rcu_node_kthread, (void *)rnp, | |
1696 | "rcun%d", rnp_index); | |
1697 | if (IS_ERR(t)) | |
1698 | return PTR_ERR(t); | |
1699 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
1700 | rnp->node_kthread_task = t; | |
1701 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
1702 | wake_up_process(t); | |
1703 | sp.sched_priority = 99; | |
1704 | sched_setscheduler_nocheck(t, SCHED_FIFO, &sp); | |
1705 | } | |
1706 | return rcu_spawn_one_boost_kthread(rsp, rnp, rnp_index); | |
a26ac245 PM |
1707 | } |
1708 | ||
1709 | /* | |
1710 | * Spawn all kthreads -- called as soon as the scheduler is running. | |
1711 | */ | |
1712 | static int __init rcu_spawn_kthreads(void) | |
1713 | { | |
1714 | int cpu; | |
1715 | struct rcu_node *rnp; | |
1716 | ||
1717 | rcu_kthreads_spawnable = 1; | |
1718 | for_each_possible_cpu(cpu) { | |
1719 | init_waitqueue_head(&per_cpu(rcu_cpu_wq, cpu)); | |
1720 | per_cpu(rcu_cpu_has_work, cpu) = 0; | |
1721 | if (cpu_online(cpu)) | |
1722 | (void)rcu_spawn_one_cpu_kthread(cpu); | |
1723 | } | |
27f4d280 PM |
1724 | rnp = rcu_get_root(rcu_state); |
1725 | init_waitqueue_head(&rnp->node_wq); | |
1726 | rcu_init_boost_waitqueue(rnp); | |
1727 | (void)rcu_spawn_one_node_kthread(rcu_state, rnp); | |
1728 | if (NUM_RCU_NODES > 1) | |
1729 | rcu_for_each_leaf_node(rcu_state, rnp) { | |
1730 | init_waitqueue_head(&rnp->node_wq); | |
1731 | rcu_init_boost_waitqueue(rnp); | |
1732 | (void)rcu_spawn_one_node_kthread(rcu_state, rnp); | |
1733 | } | |
a26ac245 PM |
1734 | return 0; |
1735 | } | |
1736 | early_initcall(rcu_spawn_kthreads); | |
1737 | ||
64db4cff PM |
1738 | static void |
1739 | __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu), | |
1740 | struct rcu_state *rsp) | |
1741 | { | |
1742 | unsigned long flags; | |
1743 | struct rcu_data *rdp; | |
1744 | ||
551d55a9 | 1745 | debug_rcu_head_queue(head); |
64db4cff PM |
1746 | head->func = func; |
1747 | head->next = NULL; | |
1748 | ||
1749 | smp_mb(); /* Ensure RCU update seen before callback registry. */ | |
1750 | ||
1751 | /* | |
1752 | * Opportunistically note grace-period endings and beginnings. | |
1753 | * Note that we might see a beginning right after we see an | |
1754 | * end, but never vice versa, since this CPU has to pass through | |
1755 | * a quiescent state betweentimes. | |
1756 | */ | |
1757 | local_irq_save(flags); | |
394f99a9 | 1758 | rdp = this_cpu_ptr(rsp->rda); |
64db4cff PM |
1759 | |
1760 | /* Add the callback to our list. */ | |
1761 | *rdp->nxttail[RCU_NEXT_TAIL] = head; | |
1762 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; | |
1763 | ||
37c72e56 PM |
1764 | /* |
1765 | * Force the grace period if too many callbacks or too long waiting. | |
1766 | * Enforce hysteresis, and don't invoke force_quiescent_state() | |
1767 | * if some other CPU has recently done so. Also, don't bother | |
1768 | * invoking force_quiescent_state() if the newly enqueued callback | |
1769 | * is the only one waiting for a grace period to complete. | |
1770 | */ | |
1771 | if (unlikely(++rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) { | |
b52573d2 PM |
1772 | |
1773 | /* Are we ignoring a completed grace period? */ | |
1774 | rcu_process_gp_end(rsp, rdp); | |
1775 | check_for_new_grace_period(rsp, rdp); | |
1776 | ||
1777 | /* Start a new grace period if one not already started. */ | |
1778 | if (!rcu_gp_in_progress(rsp)) { | |
1779 | unsigned long nestflag; | |
1780 | struct rcu_node *rnp_root = rcu_get_root(rsp); | |
1781 | ||
1782 | raw_spin_lock_irqsave(&rnp_root->lock, nestflag); | |
1783 | rcu_start_gp(rsp, nestflag); /* rlses rnp_root->lock */ | |
1784 | } else { | |
1785 | /* Give the grace period a kick. */ | |
1786 | rdp->blimit = LONG_MAX; | |
1787 | if (rsp->n_force_qs == rdp->n_force_qs_snap && | |
1788 | *rdp->nxttail[RCU_DONE_TAIL] != head) | |
1789 | force_quiescent_state(rsp, 0); | |
1790 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1791 | rdp->qlen_last_fqs_check = rdp->qlen; | |
1792 | } | |
20133cfc | 1793 | } else if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
64db4cff PM |
1794 | force_quiescent_state(rsp, 1); |
1795 | local_irq_restore(flags); | |
1796 | } | |
1797 | ||
1798 | /* | |
d6714c22 | 1799 | * Queue an RCU-sched callback for invocation after a grace period. |
64db4cff | 1800 | */ |
d6714c22 | 1801 | void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) |
64db4cff | 1802 | { |
d6714c22 | 1803 | __call_rcu(head, func, &rcu_sched_state); |
64db4cff | 1804 | } |
d6714c22 | 1805 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
64db4cff PM |
1806 | |
1807 | /* | |
1808 | * Queue an RCU for invocation after a quicker grace period. | |
1809 | */ | |
1810 | void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
1811 | { | |
1812 | __call_rcu(head, func, &rcu_bh_state); | |
1813 | } | |
1814 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
1815 | ||
6ebb237b PM |
1816 | /** |
1817 | * synchronize_sched - wait until an rcu-sched grace period has elapsed. | |
1818 | * | |
1819 | * Control will return to the caller some time after a full rcu-sched | |
1820 | * grace period has elapsed, in other words after all currently executing | |
1821 | * rcu-sched read-side critical sections have completed. These read-side | |
1822 | * critical sections are delimited by rcu_read_lock_sched() and | |
1823 | * rcu_read_unlock_sched(), and may be nested. Note that preempt_disable(), | |
1824 | * local_irq_disable(), and so on may be used in place of | |
1825 | * rcu_read_lock_sched(). | |
1826 | * | |
1827 | * This means that all preempt_disable code sequences, including NMI and | |
1828 | * hardware-interrupt handlers, in progress on entry will have completed | |
1829 | * before this primitive returns. However, this does not guarantee that | |
1830 | * softirq handlers will have completed, since in some kernels, these | |
1831 | * handlers can run in process context, and can block. | |
1832 | * | |
1833 | * This primitive provides the guarantees made by the (now removed) | |
1834 | * synchronize_kernel() API. In contrast, synchronize_rcu() only | |
1835 | * guarantees that rcu_read_lock() sections will have completed. | |
1836 | * In "classic RCU", these two guarantees happen to be one and | |
1837 | * the same, but can differ in realtime RCU implementations. | |
1838 | */ | |
1839 | void synchronize_sched(void) | |
1840 | { | |
1841 | struct rcu_synchronize rcu; | |
1842 | ||
1843 | if (rcu_blocking_is_gp()) | |
1844 | return; | |
1845 | ||
72d5a9f7 | 1846 | init_rcu_head_on_stack(&rcu.head); |
6ebb237b PM |
1847 | init_completion(&rcu.completion); |
1848 | /* Will wake me after RCU finished. */ | |
1849 | call_rcu_sched(&rcu.head, wakeme_after_rcu); | |
1850 | /* Wait for it. */ | |
1851 | wait_for_completion(&rcu.completion); | |
72d5a9f7 | 1852 | destroy_rcu_head_on_stack(&rcu.head); |
6ebb237b PM |
1853 | } |
1854 | EXPORT_SYMBOL_GPL(synchronize_sched); | |
1855 | ||
1856 | /** | |
1857 | * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed. | |
1858 | * | |
1859 | * Control will return to the caller some time after a full rcu_bh grace | |
1860 | * period has elapsed, in other words after all currently executing rcu_bh | |
1861 | * read-side critical sections have completed. RCU read-side critical | |
1862 | * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(), | |
1863 | * and may be nested. | |
1864 | */ | |
1865 | void synchronize_rcu_bh(void) | |
1866 | { | |
1867 | struct rcu_synchronize rcu; | |
1868 | ||
1869 | if (rcu_blocking_is_gp()) | |
1870 | return; | |
1871 | ||
72d5a9f7 | 1872 | init_rcu_head_on_stack(&rcu.head); |
6ebb237b PM |
1873 | init_completion(&rcu.completion); |
1874 | /* Will wake me after RCU finished. */ | |
1875 | call_rcu_bh(&rcu.head, wakeme_after_rcu); | |
1876 | /* Wait for it. */ | |
1877 | wait_for_completion(&rcu.completion); | |
72d5a9f7 | 1878 | destroy_rcu_head_on_stack(&rcu.head); |
6ebb237b PM |
1879 | } |
1880 | EXPORT_SYMBOL_GPL(synchronize_rcu_bh); | |
1881 | ||
64db4cff PM |
1882 | /* |
1883 | * Check to see if there is any immediate RCU-related work to be done | |
1884 | * by the current CPU, for the specified type of RCU, returning 1 if so. | |
1885 | * The checks are in order of increasing expense: checks that can be | |
1886 | * carried out against CPU-local state are performed first. However, | |
1887 | * we must check for CPU stalls first, else we might not get a chance. | |
1888 | */ | |
1889 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | |
1890 | { | |
2f51f988 PM |
1891 | struct rcu_node *rnp = rdp->mynode; |
1892 | ||
64db4cff PM |
1893 | rdp->n_rcu_pending++; |
1894 | ||
1895 | /* Check for CPU stalls, if enabled. */ | |
1896 | check_cpu_stall(rsp, rdp); | |
1897 | ||
1898 | /* Is the RCU core waiting for a quiescent state from this CPU? */ | |
d21670ac | 1899 | if (rdp->qs_pending && !rdp->passed_quiesc) { |
d25eb944 PM |
1900 | |
1901 | /* | |
1902 | * If force_quiescent_state() coming soon and this CPU | |
1903 | * needs a quiescent state, and this is either RCU-sched | |
1904 | * or RCU-bh, force a local reschedule. | |
1905 | */ | |
d21670ac | 1906 | rdp->n_rp_qs_pending++; |
d25eb944 PM |
1907 | if (!rdp->preemptable && |
1908 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs) - 1, | |
1909 | jiffies)) | |
1910 | set_need_resched(); | |
d21670ac PM |
1911 | } else if (rdp->qs_pending && rdp->passed_quiesc) { |
1912 | rdp->n_rp_report_qs++; | |
64db4cff | 1913 | return 1; |
7ba5c840 | 1914 | } |
64db4cff PM |
1915 | |
1916 | /* Does this CPU have callbacks ready to invoke? */ | |
7ba5c840 PM |
1917 | if (cpu_has_callbacks_ready_to_invoke(rdp)) { |
1918 | rdp->n_rp_cb_ready++; | |
64db4cff | 1919 | return 1; |
7ba5c840 | 1920 | } |
64db4cff PM |
1921 | |
1922 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
7ba5c840 PM |
1923 | if (cpu_needs_another_gp(rsp, rdp)) { |
1924 | rdp->n_rp_cpu_needs_gp++; | |
64db4cff | 1925 | return 1; |
7ba5c840 | 1926 | } |
64db4cff PM |
1927 | |
1928 | /* Has another RCU grace period completed? */ | |
2f51f988 | 1929 | if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */ |
7ba5c840 | 1930 | rdp->n_rp_gp_completed++; |
64db4cff | 1931 | return 1; |
7ba5c840 | 1932 | } |
64db4cff PM |
1933 | |
1934 | /* Has a new RCU grace period started? */ | |
2f51f988 | 1935 | if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */ |
7ba5c840 | 1936 | rdp->n_rp_gp_started++; |
64db4cff | 1937 | return 1; |
7ba5c840 | 1938 | } |
64db4cff PM |
1939 | |
1940 | /* Has an RCU GP gone long enough to send resched IPIs &c? */ | |
fc2219d4 | 1941 | if (rcu_gp_in_progress(rsp) && |
20133cfc | 1942 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) { |
7ba5c840 | 1943 | rdp->n_rp_need_fqs++; |
64db4cff | 1944 | return 1; |
7ba5c840 | 1945 | } |
64db4cff PM |
1946 | |
1947 | /* nothing to do */ | |
7ba5c840 | 1948 | rdp->n_rp_need_nothing++; |
64db4cff PM |
1949 | return 0; |
1950 | } | |
1951 | ||
1952 | /* | |
1953 | * Check to see if there is any immediate RCU-related work to be done | |
1954 | * by the current CPU, returning 1 if so. This function is part of the | |
1955 | * RCU implementation; it is -not- an exported member of the RCU API. | |
1956 | */ | |
a157229c | 1957 | static int rcu_pending(int cpu) |
64db4cff | 1958 | { |
d6714c22 | 1959 | return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) || |
f41d911f PM |
1960 | __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) || |
1961 | rcu_preempt_pending(cpu); | |
64db4cff PM |
1962 | } |
1963 | ||
1964 | /* | |
1965 | * Check to see if any future RCU-related work will need to be done | |
1966 | * by the current CPU, even if none need be done immediately, returning | |
8bd93a2c | 1967 | * 1 if so. |
64db4cff | 1968 | */ |
8bd93a2c | 1969 | static int rcu_needs_cpu_quick_check(int cpu) |
64db4cff PM |
1970 | { |
1971 | /* RCU callbacks either ready or pending? */ | |
d6714c22 | 1972 | return per_cpu(rcu_sched_data, cpu).nxtlist || |
f41d911f PM |
1973 | per_cpu(rcu_bh_data, cpu).nxtlist || |
1974 | rcu_preempt_needs_cpu(cpu); | |
64db4cff PM |
1975 | } |
1976 | ||
d0ec774c PM |
1977 | static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL}; |
1978 | static atomic_t rcu_barrier_cpu_count; | |
1979 | static DEFINE_MUTEX(rcu_barrier_mutex); | |
1980 | static struct completion rcu_barrier_completion; | |
d0ec774c PM |
1981 | |
1982 | static void rcu_barrier_callback(struct rcu_head *notused) | |
1983 | { | |
1984 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
1985 | complete(&rcu_barrier_completion); | |
1986 | } | |
1987 | ||
1988 | /* | |
1989 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
1990 | */ | |
1991 | static void rcu_barrier_func(void *type) | |
1992 | { | |
1993 | int cpu = smp_processor_id(); | |
1994 | struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu); | |
1995 | void (*call_rcu_func)(struct rcu_head *head, | |
1996 | void (*func)(struct rcu_head *head)); | |
1997 | ||
1998 | atomic_inc(&rcu_barrier_cpu_count); | |
1999 | call_rcu_func = type; | |
2000 | call_rcu_func(head, rcu_barrier_callback); | |
2001 | } | |
2002 | ||
d0ec774c PM |
2003 | /* |
2004 | * Orchestrate the specified type of RCU barrier, waiting for all | |
2005 | * RCU callbacks of the specified type to complete. | |
2006 | */ | |
e74f4c45 PM |
2007 | static void _rcu_barrier(struct rcu_state *rsp, |
2008 | void (*call_rcu_func)(struct rcu_head *head, | |
d0ec774c PM |
2009 | void (*func)(struct rcu_head *head))) |
2010 | { | |
2011 | BUG_ON(in_interrupt()); | |
e74f4c45 | 2012 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
d0ec774c PM |
2013 | mutex_lock(&rcu_barrier_mutex); |
2014 | init_completion(&rcu_barrier_completion); | |
2015 | /* | |
2016 | * Initialize rcu_barrier_cpu_count to 1, then invoke | |
2017 | * rcu_barrier_func() on each CPU, so that each CPU also has | |
2018 | * incremented rcu_barrier_cpu_count. Only then is it safe to | |
2019 | * decrement rcu_barrier_cpu_count -- otherwise the first CPU | |
2020 | * might complete its grace period before all of the other CPUs | |
2021 | * did their increment, causing this function to return too | |
2d999e03 PM |
2022 | * early. Note that on_each_cpu() disables irqs, which prevents |
2023 | * any CPUs from coming online or going offline until each online | |
2024 | * CPU has queued its RCU-barrier callback. | |
d0ec774c PM |
2025 | */ |
2026 | atomic_set(&rcu_barrier_cpu_count, 1); | |
2027 | on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1); | |
2028 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
2029 | complete(&rcu_barrier_completion); | |
2030 | wait_for_completion(&rcu_barrier_completion); | |
2031 | mutex_unlock(&rcu_barrier_mutex); | |
d0ec774c | 2032 | } |
d0ec774c PM |
2033 | |
2034 | /** | |
2035 | * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete. | |
2036 | */ | |
2037 | void rcu_barrier_bh(void) | |
2038 | { | |
e74f4c45 | 2039 | _rcu_barrier(&rcu_bh_state, call_rcu_bh); |
d0ec774c PM |
2040 | } |
2041 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); | |
2042 | ||
2043 | /** | |
2044 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. | |
2045 | */ | |
2046 | void rcu_barrier_sched(void) | |
2047 | { | |
e74f4c45 | 2048 | _rcu_barrier(&rcu_sched_state, call_rcu_sched); |
d0ec774c PM |
2049 | } |
2050 | EXPORT_SYMBOL_GPL(rcu_barrier_sched); | |
2051 | ||
64db4cff | 2052 | /* |
27569620 | 2053 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 2054 | */ |
27569620 PM |
2055 | static void __init |
2056 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) | |
64db4cff PM |
2057 | { |
2058 | unsigned long flags; | |
2059 | int i; | |
394f99a9 | 2060 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
27569620 PM |
2061 | struct rcu_node *rnp = rcu_get_root(rsp); |
2062 | ||
2063 | /* Set up local state, ensuring consistent view of global state. */ | |
1304afb2 | 2064 | raw_spin_lock_irqsave(&rnp->lock, flags); |
27569620 PM |
2065 | rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo); |
2066 | rdp->nxtlist = NULL; | |
2067 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
2068 | rdp->nxttail[i] = &rdp->nxtlist; | |
2069 | rdp->qlen = 0; | |
2070 | #ifdef CONFIG_NO_HZ | |
2071 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); | |
2072 | #endif /* #ifdef CONFIG_NO_HZ */ | |
2073 | rdp->cpu = cpu; | |
1304afb2 | 2074 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
27569620 PM |
2075 | } |
2076 | ||
2077 | /* | |
2078 | * Initialize a CPU's per-CPU RCU data. Note that only one online or | |
2079 | * offline event can be happening at a given time. Note also that we | |
2080 | * can accept some slop in the rsp->completed access due to the fact | |
2081 | * that this CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 2082 | */ |
e4fa4c97 | 2083 | static void __cpuinit |
f41d911f | 2084 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable) |
64db4cff PM |
2085 | { |
2086 | unsigned long flags; | |
64db4cff | 2087 | unsigned long mask; |
394f99a9 | 2088 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff PM |
2089 | struct rcu_node *rnp = rcu_get_root(rsp); |
2090 | ||
2091 | /* Set up local state, ensuring consistent view of global state. */ | |
1304afb2 | 2092 | raw_spin_lock_irqsave(&rnp->lock, flags); |
64db4cff PM |
2093 | rdp->passed_quiesc = 0; /* We could be racing with new GP, */ |
2094 | rdp->qs_pending = 1; /* so set up to respond to current GP. */ | |
2095 | rdp->beenonline = 1; /* We have now been online. */ | |
f41d911f | 2096 | rdp->preemptable = preemptable; |
37c72e56 PM |
2097 | rdp->qlen_last_fqs_check = 0; |
2098 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
64db4cff | 2099 | rdp->blimit = blimit; |
1304afb2 | 2100 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
2101 | |
2102 | /* | |
2103 | * A new grace period might start here. If so, we won't be part | |
2104 | * of it, but that is OK, as we are currently in a quiescent state. | |
2105 | */ | |
2106 | ||
2107 | /* Exclude any attempts to start a new GP on large systems. */ | |
1304afb2 | 2108 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
64db4cff PM |
2109 | |
2110 | /* Add CPU to rcu_node bitmasks. */ | |
2111 | rnp = rdp->mynode; | |
2112 | mask = rdp->grpmask; | |
2113 | do { | |
2114 | /* Exclude any attempts to start a new GP on small systems. */ | |
1304afb2 | 2115 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
2116 | rnp->qsmaskinit |= mask; |
2117 | mask = rnp->grpmask; | |
d09b62df PM |
2118 | if (rnp == rdp->mynode) { |
2119 | rdp->gpnum = rnp->completed; /* if GP in progress... */ | |
2120 | rdp->completed = rnp->completed; | |
2121 | rdp->passed_quiesc_completed = rnp->completed - 1; | |
2122 | } | |
1304afb2 | 2123 | raw_spin_unlock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
2124 | rnp = rnp->parent; |
2125 | } while (rnp != NULL && !(rnp->qsmaskinit & mask)); | |
2126 | ||
1304afb2 | 2127 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); |
64db4cff PM |
2128 | } |
2129 | ||
2130 | static void __cpuinit rcu_online_cpu(int cpu) | |
2131 | { | |
f41d911f PM |
2132 | rcu_init_percpu_data(cpu, &rcu_sched_state, 0); |
2133 | rcu_init_percpu_data(cpu, &rcu_bh_state, 0); | |
2134 | rcu_preempt_init_percpu_data(cpu); | |
64db4cff PM |
2135 | } |
2136 | ||
a26ac245 PM |
2137 | static void __cpuinit rcu_online_kthreads(int cpu) |
2138 | { | |
27f4d280 | 2139 | struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu); |
a26ac245 PM |
2140 | struct rcu_node *rnp = rdp->mynode; |
2141 | ||
2142 | /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */ | |
2143 | if (rcu_kthreads_spawnable) { | |
2144 | (void)rcu_spawn_one_cpu_kthread(cpu); | |
2145 | if (rnp->node_kthread_task == NULL) | |
27f4d280 | 2146 | (void)rcu_spawn_one_node_kthread(rcu_state, rnp); |
a26ac245 PM |
2147 | } |
2148 | } | |
2149 | ||
64db4cff | 2150 | /* |
f41d911f | 2151 | * Handle CPU online/offline notification events. |
64db4cff | 2152 | */ |
9f680ab4 PM |
2153 | static int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
2154 | unsigned long action, void *hcpu) | |
64db4cff PM |
2155 | { |
2156 | long cpu = (long)hcpu; | |
27f4d280 | 2157 | struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu); |
a26ac245 | 2158 | struct rcu_node *rnp = rdp->mynode; |
64db4cff PM |
2159 | |
2160 | switch (action) { | |
2161 | case CPU_UP_PREPARE: | |
2162 | case CPU_UP_PREPARE_FROZEN: | |
2163 | rcu_online_cpu(cpu); | |
a26ac245 PM |
2164 | rcu_online_kthreads(cpu); |
2165 | break; | |
2166 | case CPU_ONLINE: | |
0f962a5e PM |
2167 | case CPU_DOWN_FAILED: |
2168 | rcu_node_kthread_setaffinity(rnp, -1); | |
2169 | break; | |
2170 | case CPU_DOWN_PREPARE: | |
2171 | rcu_node_kthread_setaffinity(rnp, cpu); | |
64db4cff | 2172 | break; |
d0ec774c PM |
2173 | case CPU_DYING: |
2174 | case CPU_DYING_FROZEN: | |
2175 | /* | |
2d999e03 PM |
2176 | * The whole machine is "stopped" except this CPU, so we can |
2177 | * touch any data without introducing corruption. We send the | |
2178 | * dying CPU's callbacks to an arbitrarily chosen online CPU. | |
d0ec774c | 2179 | */ |
29494be7 LJ |
2180 | rcu_send_cbs_to_online(&rcu_bh_state); |
2181 | rcu_send_cbs_to_online(&rcu_sched_state); | |
2182 | rcu_preempt_send_cbs_to_online(); | |
d0ec774c | 2183 | break; |
64db4cff PM |
2184 | case CPU_DEAD: |
2185 | case CPU_DEAD_FROZEN: | |
2186 | case CPU_UP_CANCELED: | |
2187 | case CPU_UP_CANCELED_FROZEN: | |
2188 | rcu_offline_cpu(cpu); | |
2189 | break; | |
2190 | default: | |
2191 | break; | |
2192 | } | |
2193 | return NOTIFY_OK; | |
2194 | } | |
2195 | ||
bbad9379 PM |
2196 | /* |
2197 | * This function is invoked towards the end of the scheduler's initialization | |
2198 | * process. Before this is called, the idle task might contain | |
2199 | * RCU read-side critical sections (during which time, this idle | |
2200 | * task is booting the system). After this function is called, the | |
2201 | * idle tasks are prohibited from containing RCU read-side critical | |
2202 | * sections. This function also enables RCU lockdep checking. | |
2203 | */ | |
2204 | void rcu_scheduler_starting(void) | |
2205 | { | |
2206 | WARN_ON(num_online_cpus() != 1); | |
2207 | WARN_ON(nr_context_switches() > 0); | |
2208 | rcu_scheduler_active = 1; | |
2209 | } | |
2210 | ||
64db4cff PM |
2211 | /* |
2212 | * Compute the per-level fanout, either using the exact fanout specified | |
2213 | * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT. | |
2214 | */ | |
2215 | #ifdef CONFIG_RCU_FANOUT_EXACT | |
2216 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
2217 | { | |
2218 | int i; | |
2219 | ||
0209f649 | 2220 | for (i = NUM_RCU_LVLS - 1; i > 0; i--) |
64db4cff | 2221 | rsp->levelspread[i] = CONFIG_RCU_FANOUT; |
0209f649 | 2222 | rsp->levelspread[0] = RCU_FANOUT_LEAF; |
64db4cff PM |
2223 | } |
2224 | #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
2225 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
2226 | { | |
2227 | int ccur; | |
2228 | int cprv; | |
2229 | int i; | |
2230 | ||
2231 | cprv = NR_CPUS; | |
2232 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
2233 | ccur = rsp->levelcnt[i]; | |
2234 | rsp->levelspread[i] = (cprv + ccur - 1) / ccur; | |
2235 | cprv = ccur; | |
2236 | } | |
2237 | } | |
2238 | #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
2239 | ||
2240 | /* | |
2241 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
2242 | */ | |
394f99a9 LJ |
2243 | static void __init rcu_init_one(struct rcu_state *rsp, |
2244 | struct rcu_data __percpu *rda) | |
64db4cff | 2245 | { |
b6407e86 PM |
2246 | static char *buf[] = { "rcu_node_level_0", |
2247 | "rcu_node_level_1", | |
2248 | "rcu_node_level_2", | |
2249 | "rcu_node_level_3" }; /* Match MAX_RCU_LVLS */ | |
64db4cff PM |
2250 | int cpustride = 1; |
2251 | int i; | |
2252 | int j; | |
2253 | struct rcu_node *rnp; | |
2254 | ||
b6407e86 PM |
2255 | BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
2256 | ||
64db4cff PM |
2257 | /* Initialize the level-tracking arrays. */ |
2258 | ||
2259 | for (i = 1; i < NUM_RCU_LVLS; i++) | |
2260 | rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1]; | |
2261 | rcu_init_levelspread(rsp); | |
2262 | ||
2263 | /* Initialize the elements themselves, starting from the leaves. */ | |
2264 | ||
2265 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
2266 | cpustride *= rsp->levelspread[i]; | |
2267 | rnp = rsp->level[i]; | |
2268 | for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { | |
1304afb2 | 2269 | raw_spin_lock_init(&rnp->lock); |
b6407e86 PM |
2270 | lockdep_set_class_and_name(&rnp->lock, |
2271 | &rcu_node_class[i], buf[i]); | |
f41d911f | 2272 | rnp->gpnum = 0; |
64db4cff PM |
2273 | rnp->qsmask = 0; |
2274 | rnp->qsmaskinit = 0; | |
2275 | rnp->grplo = j * cpustride; | |
2276 | rnp->grphi = (j + 1) * cpustride - 1; | |
2277 | if (rnp->grphi >= NR_CPUS) | |
2278 | rnp->grphi = NR_CPUS - 1; | |
2279 | if (i == 0) { | |
2280 | rnp->grpnum = 0; | |
2281 | rnp->grpmask = 0; | |
2282 | rnp->parent = NULL; | |
2283 | } else { | |
2284 | rnp->grpnum = j % rsp->levelspread[i - 1]; | |
2285 | rnp->grpmask = 1UL << rnp->grpnum; | |
2286 | rnp->parent = rsp->level[i - 1] + | |
2287 | j / rsp->levelspread[i - 1]; | |
2288 | } | |
2289 | rnp->level = i; | |
12f5f524 | 2290 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
64db4cff PM |
2291 | } |
2292 | } | |
0c34029a | 2293 | |
394f99a9 | 2294 | rsp->rda = rda; |
0c34029a LJ |
2295 | rnp = rsp->level[NUM_RCU_LVLS - 1]; |
2296 | for_each_possible_cpu(i) { | |
4a90a068 | 2297 | while (i > rnp->grphi) |
0c34029a | 2298 | rnp++; |
394f99a9 | 2299 | per_cpu_ptr(rsp->rda, i)->mynode = rnp; |
0c34029a LJ |
2300 | rcu_boot_init_percpu_data(i, rsp); |
2301 | } | |
64db4cff PM |
2302 | } |
2303 | ||
9f680ab4 | 2304 | void __init rcu_init(void) |
64db4cff | 2305 | { |
017c4261 | 2306 | int cpu; |
9f680ab4 | 2307 | |
f41d911f | 2308 | rcu_bootup_announce(); |
394f99a9 LJ |
2309 | rcu_init_one(&rcu_sched_state, &rcu_sched_data); |
2310 | rcu_init_one(&rcu_bh_state, &rcu_bh_data); | |
f41d911f | 2311 | __rcu_init_preempt(); |
9f680ab4 PM |
2312 | |
2313 | /* | |
2314 | * We don't need protection against CPU-hotplug here because | |
2315 | * this is called early in boot, before either interrupts | |
2316 | * or the scheduler are operational. | |
2317 | */ | |
2318 | cpu_notifier(rcu_cpu_notify, 0); | |
017c4261 PM |
2319 | for_each_online_cpu(cpu) |
2320 | rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu); | |
c68de209 | 2321 | check_cpu_stall_init(); |
64db4cff PM |
2322 | } |
2323 | ||
1eba8f84 | 2324 | #include "rcutree_plugin.h" |