Root/
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 - |
28 | * Documentation/RCU |
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> |
38 | #include <linux/nmi.h> |
39 | #include <linux/atomic.h> |
40 | #include <linux/bitops.h> |
41 | #include <linux/export.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> |
49 | #include <linux/kernel_stat.h> |
50 | #include <linux/wait.h> |
51 | #include <linux/kthread.h> |
52 | #include <linux/prefetch.h> |
53 | |
54 | #include "rcutree.h" |
55 | #include <trace/events/rcu.h> |
56 | |
57 | #include "rcu.h" |
58 | |
59 | /* Data structures. */ |
60 | |
61 | static struct lock_class_key rcu_node_class[NUM_RCU_LVLS]; |
62 | |
63 | #define RCU_STATE_INITIALIZER(structname) { \ |
64 | .level = { &structname##_state.node[0] }, \ |
65 | .levelcnt = { \ |
66 | NUM_RCU_LVL_0, /* root of hierarchy. */ \ |
67 | NUM_RCU_LVL_1, \ |
68 | NUM_RCU_LVL_2, \ |
69 | NUM_RCU_LVL_3, \ |
70 | NUM_RCU_LVL_4, /* == MAX_RCU_LVLS */ \ |
71 | }, \ |
72 | .signaled = RCU_GP_IDLE, \ |
73 | .gpnum = -300, \ |
74 | .completed = -300, \ |
75 | .onofflock = __RAW_SPIN_LOCK_UNLOCKED(&structname##_state.onofflock), \ |
76 | .fqslock = __RAW_SPIN_LOCK_UNLOCKED(&structname##_state.fqslock), \ |
77 | .n_force_qs = 0, \ |
78 | .n_force_qs_ngp = 0, \ |
79 | .name = #structname, \ |
80 | } |
81 | |
82 | struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched); |
83 | DEFINE_PER_CPU(struct rcu_data, rcu_sched_data); |
84 | |
85 | struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh); |
86 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); |
87 | |
88 | static struct rcu_state *rcu_state; |
89 | |
90 | /* |
91 | * The rcu_scheduler_active variable transitions from zero to one just |
92 | * before the first task is spawned. So when this variable is zero, RCU |
93 | * can assume that there is but one task, allowing RCU to (for example) |
94 | * optimized synchronize_sched() to a simple barrier(). When this variable |
95 | * is one, RCU must actually do all the hard work required to detect real |
96 | * grace periods. This variable is also used to suppress boot-time false |
97 | * positives from lockdep-RCU error checking. |
98 | */ |
99 | int rcu_scheduler_active __read_mostly; |
100 | EXPORT_SYMBOL_GPL(rcu_scheduler_active); |
101 | |
102 | /* |
103 | * The rcu_scheduler_fully_active variable transitions from zero to one |
104 | * during the early_initcall() processing, which is after the scheduler |
105 | * is capable of creating new tasks. So RCU processing (for example, |
106 | * creating tasks for RCU priority boosting) must be delayed until after |
107 | * rcu_scheduler_fully_active transitions from zero to one. We also |
108 | * currently delay invocation of any RCU callbacks until after this point. |
109 | * |
110 | * It might later prove better for people registering RCU callbacks during |
111 | * early boot to take responsibility for these callbacks, but one step at |
112 | * a time. |
113 | */ |
114 | static int rcu_scheduler_fully_active __read_mostly; |
115 | |
116 | #ifdef CONFIG_RCU_BOOST |
117 | |
118 | /* |
119 | * Control variables for per-CPU and per-rcu_node kthreads. These |
120 | * handle all flavors of RCU. |
121 | */ |
122 | static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task); |
123 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status); |
124 | DEFINE_PER_CPU(int, rcu_cpu_kthread_cpu); |
125 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops); |
126 | DEFINE_PER_CPU(char, rcu_cpu_has_work); |
127 | |
128 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
129 | |
130 | static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); |
131 | static void invoke_rcu_core(void); |
132 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp); |
133 | |
134 | /* |
135 | * Track the rcutorture test sequence number and the update version |
136 | * number within a given test. The rcutorture_testseq is incremented |
137 | * on every rcutorture module load and unload, so has an odd value |
138 | * when a test is running. The rcutorture_vernum is set to zero |
139 | * when rcutorture starts and is incremented on each rcutorture update. |
140 | * These variables enable correlating rcutorture output with the |
141 | * RCU tracing information. |
142 | */ |
143 | unsigned long rcutorture_testseq; |
144 | unsigned long rcutorture_vernum; |
145 | |
146 | /* |
147 | * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s |
148 | * permit this function to be invoked without holding the root rcu_node |
149 | * structure's ->lock, but of course results can be subject to change. |
150 | */ |
151 | static int rcu_gp_in_progress(struct rcu_state *rsp) |
152 | { |
153 | return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum); |
154 | } |
155 | |
156 | /* |
157 | * Note a quiescent state. Because we do not need to know |
158 | * how many quiescent states passed, just if there was at least |
159 | * one since the start of the grace period, this just sets a flag. |
160 | * The caller must have disabled preemption. |
161 | */ |
162 | void rcu_sched_qs(int cpu) |
163 | { |
164 | struct rcu_data *rdp = &per_cpu(rcu_sched_data, cpu); |
165 | |
166 | rdp->passed_quiesce_gpnum = rdp->gpnum; |
167 | barrier(); |
168 | if (rdp->passed_quiesce == 0) |
169 | trace_rcu_grace_period("rcu_sched", rdp->gpnum, "cpuqs"); |
170 | rdp->passed_quiesce = 1; |
171 | } |
172 | |
173 | void rcu_bh_qs(int cpu) |
174 | { |
175 | struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu); |
176 | |
177 | rdp->passed_quiesce_gpnum = rdp->gpnum; |
178 | barrier(); |
179 | if (rdp->passed_quiesce == 0) |
180 | trace_rcu_grace_period("rcu_bh", rdp->gpnum, "cpuqs"); |
181 | rdp->passed_quiesce = 1; |
182 | } |
183 | |
184 | /* |
185 | * Note a context switch. This is a quiescent state for RCU-sched, |
186 | * and requires special handling for preemptible RCU. |
187 | * The caller must have disabled preemption. |
188 | */ |
189 | void rcu_note_context_switch(int cpu) |
190 | { |
191 | trace_rcu_utilization("Start context switch"); |
192 | rcu_sched_qs(cpu); |
193 | rcu_preempt_note_context_switch(cpu); |
194 | trace_rcu_utilization("End context switch"); |
195 | } |
196 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
197 | |
198 | #ifdef CONFIG_NO_HZ |
199 | DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { |
200 | .dynticks_nesting = 1, |
201 | .dynticks = ATOMIC_INIT(1), |
202 | }; |
203 | #endif /* #ifdef CONFIG_NO_HZ */ |
204 | |
205 | static int blimit = 10; /* Maximum callbacks per rcu_do_batch. */ |
206 | static int qhimark = 10000; /* If this many pending, ignore blimit. */ |
207 | static int qlowmark = 100; /* Once only this many pending, use blimit. */ |
208 | |
209 | module_param(blimit, int, 0); |
210 | module_param(qhimark, int, 0); |
211 | module_param(qlowmark, int, 0); |
212 | |
213 | int rcu_cpu_stall_suppress __read_mostly; |
214 | module_param(rcu_cpu_stall_suppress, int, 0644); |
215 | |
216 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed); |
217 | static int rcu_pending(int cpu); |
218 | |
219 | /* |
220 | * Return the number of RCU-sched batches processed thus far for debug & stats. |
221 | */ |
222 | long rcu_batches_completed_sched(void) |
223 | { |
224 | return rcu_sched_state.completed; |
225 | } |
226 | EXPORT_SYMBOL_GPL(rcu_batches_completed_sched); |
227 | |
228 | /* |
229 | * Return the number of RCU BH batches processed thus far for debug & stats. |
230 | */ |
231 | long rcu_batches_completed_bh(void) |
232 | { |
233 | return rcu_bh_state.completed; |
234 | } |
235 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); |
236 | |
237 | /* |
238 | * Force a quiescent state for RCU BH. |
239 | */ |
240 | void rcu_bh_force_quiescent_state(void) |
241 | { |
242 | force_quiescent_state(&rcu_bh_state, 0); |
243 | } |
244 | EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state); |
245 | |
246 | /* |
247 | * Record the number of times rcutorture tests have been initiated and |
248 | * terminated. This information allows the debugfs tracing stats to be |
249 | * correlated to the rcutorture messages, even when the rcutorture module |
250 | * is being repeatedly loaded and unloaded. In other words, we cannot |
251 | * store this state in rcutorture itself. |
252 | */ |
253 | void rcutorture_record_test_transition(void) |
254 | { |
255 | rcutorture_testseq++; |
256 | rcutorture_vernum = 0; |
257 | } |
258 | EXPORT_SYMBOL_GPL(rcutorture_record_test_transition); |
259 | |
260 | /* |
261 | * Record the number of writer passes through the current rcutorture test. |
262 | * This is also used to correlate debugfs tracing stats with the rcutorture |
263 | * messages. |
264 | */ |
265 | void rcutorture_record_progress(unsigned long vernum) |
266 | { |
267 | rcutorture_vernum++; |
268 | } |
269 | EXPORT_SYMBOL_GPL(rcutorture_record_progress); |
270 | |
271 | /* |
272 | * Force a quiescent state for RCU-sched. |
273 | */ |
274 | void rcu_sched_force_quiescent_state(void) |
275 | { |
276 | force_quiescent_state(&rcu_sched_state, 0); |
277 | } |
278 | EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state); |
279 | |
280 | /* |
281 | * Does the CPU have callbacks ready to be invoked? |
282 | */ |
283 | static int |
284 | cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) |
285 | { |
286 | return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]; |
287 | } |
288 | |
289 | /* |
290 | * Does the current CPU require a yet-as-unscheduled grace period? |
291 | */ |
292 | static int |
293 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) |
294 | { |
295 | return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp); |
296 | } |
297 | |
298 | /* |
299 | * Return the root node of the specified rcu_state structure. |
300 | */ |
301 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) |
302 | { |
303 | return &rsp->node[0]; |
304 | } |
305 | |
306 | #ifdef CONFIG_SMP |
307 | |
308 | /* |
309 | * If the specified CPU is offline, tell the caller that it is in |
310 | * a quiescent state. Otherwise, whack it with a reschedule IPI. |
311 | * Grace periods can end up waiting on an offline CPU when that |
312 | * CPU is in the process of coming online -- it will be added to the |
313 | * rcu_node bitmasks before it actually makes it online. The same thing |
314 | * can happen while a CPU is in the process of coming online. Because this |
315 | * race is quite rare, we check for it after detecting that the grace |
316 | * period has been delayed rather than checking each and every CPU |
317 | * each and every time we start a new grace period. |
318 | */ |
319 | static int rcu_implicit_offline_qs(struct rcu_data *rdp) |
320 | { |
321 | /* |
322 | * If the CPU is offline, it is in a quiescent state. We can |
323 | * trust its state not to change because interrupts are disabled. |
324 | */ |
325 | if (cpu_is_offline(rdp->cpu)) { |
326 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, "ofl"); |
327 | rdp->offline_fqs++; |
328 | return 1; |
329 | } |
330 | |
331 | /* If preemptible RCU, no point in sending reschedule IPI. */ |
332 | if (rdp->preemptible) |
333 | return 0; |
334 | |
335 | /* The CPU is online, so send it a reschedule IPI. */ |
336 | if (rdp->cpu != smp_processor_id()) |
337 | smp_send_reschedule(rdp->cpu); |
338 | else |
339 | set_need_resched(); |
340 | rdp->resched_ipi++; |
341 | return 0; |
342 | } |
343 | |
344 | #endif /* #ifdef CONFIG_SMP */ |
345 | |
346 | #ifdef CONFIG_NO_HZ |
347 | |
348 | /** |
349 | * rcu_enter_nohz - inform RCU that current CPU is entering nohz |
350 | * |
351 | * Enter nohz mode, in other words, -leave- the mode in which RCU |
352 | * read-side critical sections can occur. (Though RCU read-side |
353 | * critical sections can occur in irq handlers in nohz mode, a possibility |
354 | * handled by rcu_irq_enter() and rcu_irq_exit()). |
355 | */ |
356 | void rcu_enter_nohz(void) |
357 | { |
358 | unsigned long flags; |
359 | struct rcu_dynticks *rdtp; |
360 | |
361 | local_irq_save(flags); |
362 | rdtp = &__get_cpu_var(rcu_dynticks); |
363 | if (--rdtp->dynticks_nesting) { |
364 | local_irq_restore(flags); |
365 | return; |
366 | } |
367 | trace_rcu_dyntick("Start"); |
368 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ |
369 | smp_mb__before_atomic_inc(); /* See above. */ |
370 | atomic_inc(&rdtp->dynticks); |
371 | smp_mb__after_atomic_inc(); /* Force ordering with next sojourn. */ |
372 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); |
373 | local_irq_restore(flags); |
374 | } |
375 | |
376 | /* |
377 | * rcu_exit_nohz - inform RCU that current CPU is leaving nohz |
378 | * |
379 | * Exit nohz mode, in other words, -enter- the mode in which RCU |
380 | * read-side critical sections normally occur. |
381 | */ |
382 | void rcu_exit_nohz(void) |
383 | { |
384 | unsigned long flags; |
385 | struct rcu_dynticks *rdtp; |
386 | |
387 | local_irq_save(flags); |
388 | rdtp = &__get_cpu_var(rcu_dynticks); |
389 | if (rdtp->dynticks_nesting++) { |
390 | local_irq_restore(flags); |
391 | return; |
392 | } |
393 | smp_mb__before_atomic_inc(); /* Force ordering w/previous sojourn. */ |
394 | atomic_inc(&rdtp->dynticks); |
395 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ |
396 | smp_mb__after_atomic_inc(); /* See above. */ |
397 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); |
398 | trace_rcu_dyntick("End"); |
399 | local_irq_restore(flags); |
400 | } |
401 | |
402 | /** |
403 | * rcu_nmi_enter - inform RCU of entry to NMI context |
404 | * |
405 | * If the CPU was idle with dynamic ticks active, and there is no |
406 | * irq handler running, this updates rdtp->dynticks_nmi to let the |
407 | * RCU grace-period handling know that the CPU is active. |
408 | */ |
409 | void rcu_nmi_enter(void) |
410 | { |
411 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); |
412 | |
413 | if (rdtp->dynticks_nmi_nesting == 0 && |
414 | (atomic_read(&rdtp->dynticks) & 0x1)) |
415 | return; |
416 | rdtp->dynticks_nmi_nesting++; |
417 | smp_mb__before_atomic_inc(); /* Force delay from prior write. */ |
418 | atomic_inc(&rdtp->dynticks); |
419 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ |
420 | smp_mb__after_atomic_inc(); /* See above. */ |
421 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); |
422 | } |
423 | |
424 | /** |
425 | * rcu_nmi_exit - inform RCU of exit from NMI context |
426 | * |
427 | * If the CPU was idle with dynamic ticks active, and there is no |
428 | * irq handler running, this updates rdtp->dynticks_nmi to let the |
429 | * RCU grace-period handling know that the CPU is no longer active. |
430 | */ |
431 | void rcu_nmi_exit(void) |
432 | { |
433 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); |
434 | |
435 | if (rdtp->dynticks_nmi_nesting == 0 || |
436 | --rdtp->dynticks_nmi_nesting != 0) |
437 | return; |
438 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ |
439 | smp_mb__before_atomic_inc(); /* See above. */ |
440 | atomic_inc(&rdtp->dynticks); |
441 | smp_mb__after_atomic_inc(); /* Force delay to next write. */ |
442 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); |
443 | } |
444 | |
445 | /** |
446 | * rcu_irq_enter - inform RCU of entry to hard irq context |
447 | * |
448 | * If the CPU was idle with dynamic ticks active, this updates the |
449 | * rdtp->dynticks to let the RCU handling know that the CPU is active. |
450 | */ |
451 | void rcu_irq_enter(void) |
452 | { |
453 | rcu_exit_nohz(); |
454 | } |
455 | |
456 | /** |
457 | * rcu_irq_exit - inform RCU of exit from hard irq context |
458 | * |
459 | * If the CPU was idle with dynamic ticks active, update the rdp->dynticks |
460 | * to put let the RCU handling be aware that the CPU is going back to idle |
461 | * with no ticks. |
462 | */ |
463 | void rcu_irq_exit(void) |
464 | { |
465 | rcu_enter_nohz(); |
466 | } |
467 | |
468 | #ifdef CONFIG_SMP |
469 | |
470 | /* |
471 | * Snapshot the specified CPU's dynticks counter so that we can later |
472 | * credit them with an implicit quiescent state. Return 1 if this CPU |
473 | * is in dynticks idle mode, which is an extended quiescent state. |
474 | */ |
475 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
476 | { |
477 | rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks); |
478 | return 0; |
479 | } |
480 | |
481 | /* |
482 | * Return true if the specified CPU has passed through a quiescent |
483 | * state by virtue of being in or having passed through an dynticks |
484 | * idle state since the last call to dyntick_save_progress_counter() |
485 | * for this same CPU. |
486 | */ |
487 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) |
488 | { |
489 | unsigned int curr; |
490 | unsigned int snap; |
491 | |
492 | curr = (unsigned int)atomic_add_return(0, &rdp->dynticks->dynticks); |
493 | snap = (unsigned int)rdp->dynticks_snap; |
494 | |
495 | /* |
496 | * If the CPU passed through or entered a dynticks idle phase with |
497 | * no active irq/NMI handlers, then we can safely pretend that the CPU |
498 | * already acknowledged the request to pass through a quiescent |
499 | * state. Either way, that CPU cannot possibly be in an RCU |
500 | * read-side critical section that started before the beginning |
501 | * of the current RCU grace period. |
502 | */ |
503 | if ((curr & 0x1) == 0 || UINT_CMP_GE(curr, snap + 2)) { |
504 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, "dti"); |
505 | rdp->dynticks_fqs++; |
506 | return 1; |
507 | } |
508 | |
509 | /* Go check for the CPU being offline. */ |
510 | return rcu_implicit_offline_qs(rdp); |
511 | } |
512 | |
513 | #endif /* #ifdef CONFIG_SMP */ |
514 | |
515 | #else /* #ifdef CONFIG_NO_HZ */ |
516 | |
517 | #ifdef CONFIG_SMP |
518 | |
519 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
520 | { |
521 | return 0; |
522 | } |
523 | |
524 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) |
525 | { |
526 | return rcu_implicit_offline_qs(rdp); |
527 | } |
528 | |
529 | #endif /* #ifdef CONFIG_SMP */ |
530 | |
531 | #endif /* #else #ifdef CONFIG_NO_HZ */ |
532 | |
533 | int rcu_cpu_stall_suppress __read_mostly; |
534 | |
535 | static void record_gp_stall_check_time(struct rcu_state *rsp) |
536 | { |
537 | rsp->gp_start = jiffies; |
538 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK; |
539 | } |
540 | |
541 | static void print_other_cpu_stall(struct rcu_state *rsp) |
542 | { |
543 | int cpu; |
544 | long delta; |
545 | unsigned long flags; |
546 | int ndetected; |
547 | struct rcu_node *rnp = rcu_get_root(rsp); |
548 | |
549 | /* Only let one CPU complain about others per time interval. */ |
550 | |
551 | raw_spin_lock_irqsave(&rnp->lock, flags); |
552 | delta = jiffies - rsp->jiffies_stall; |
553 | if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) { |
554 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
555 | return; |
556 | } |
557 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK; |
558 | |
559 | /* |
560 | * Now rat on any tasks that got kicked up to the root rcu_node |
561 | * due to CPU offlining. |
562 | */ |
563 | ndetected = rcu_print_task_stall(rnp); |
564 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
565 | |
566 | /* |
567 | * OK, time to rat on our buddy... |
568 | * See Documentation/RCU/stallwarn.txt for info on how to debug |
569 | * RCU CPU stall warnings. |
570 | */ |
571 | printk(KERN_ERR "INFO: %s detected stalls on CPUs/tasks: {", |
572 | rsp->name); |
573 | rcu_for_each_leaf_node(rsp, rnp) { |
574 | raw_spin_lock_irqsave(&rnp->lock, flags); |
575 | ndetected += rcu_print_task_stall(rnp); |
576 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
577 | if (rnp->qsmask == 0) |
578 | continue; |
579 | for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++) |
580 | if (rnp->qsmask & (1UL << cpu)) { |
581 | printk(" %d", rnp->grplo + cpu); |
582 | ndetected++; |
583 | } |
584 | } |
585 | printk("} (detected by %d, t=%ld jiffies)\n", |
586 | smp_processor_id(), (long)(jiffies - rsp->gp_start)); |
587 | if (ndetected == 0) |
588 | printk(KERN_ERR "INFO: Stall ended before state dump start\n"); |
589 | else if (!trigger_all_cpu_backtrace()) |
590 | dump_stack(); |
591 | |
592 | /* If so configured, complain about tasks blocking the grace period. */ |
593 | |
594 | rcu_print_detail_task_stall(rsp); |
595 | |
596 | force_quiescent_state(rsp, 0); /* Kick them all. */ |
597 | } |
598 | |
599 | static void print_cpu_stall(struct rcu_state *rsp) |
600 | { |
601 | unsigned long flags; |
602 | struct rcu_node *rnp = rcu_get_root(rsp); |
603 | |
604 | /* |
605 | * OK, time to rat on ourselves... |
606 | * See Documentation/RCU/stallwarn.txt for info on how to debug |
607 | * RCU CPU stall warnings. |
608 | */ |
609 | printk(KERN_ERR "INFO: %s detected stall on CPU %d (t=%lu jiffies)\n", |
610 | rsp->name, smp_processor_id(), jiffies - rsp->gp_start); |
611 | if (!trigger_all_cpu_backtrace()) |
612 | dump_stack(); |
613 | |
614 | raw_spin_lock_irqsave(&rnp->lock, flags); |
615 | if (ULONG_CMP_GE(jiffies, rsp->jiffies_stall)) |
616 | rsp->jiffies_stall = |
617 | jiffies + RCU_SECONDS_TILL_STALL_RECHECK; |
618 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
619 | |
620 | set_need_resched(); /* kick ourselves to get things going. */ |
621 | } |
622 | |
623 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) |
624 | { |
625 | unsigned long j; |
626 | unsigned long js; |
627 | struct rcu_node *rnp; |
628 | |
629 | if (rcu_cpu_stall_suppress) |
630 | return; |
631 | j = ACCESS_ONCE(jiffies); |
632 | js = ACCESS_ONCE(rsp->jiffies_stall); |
633 | rnp = rdp->mynode; |
634 | if ((ACCESS_ONCE(rnp->qsmask) & rdp->grpmask) && ULONG_CMP_GE(j, js)) { |
635 | |
636 | /* We haven't checked in, so go dump stack. */ |
637 | print_cpu_stall(rsp); |
638 | |
639 | } else if (rcu_gp_in_progress(rsp) && |
640 | ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) { |
641 | |
642 | /* They had a few time units to dump stack, so complain. */ |
643 | print_other_cpu_stall(rsp); |
644 | } |
645 | } |
646 | |
647 | static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr) |
648 | { |
649 | rcu_cpu_stall_suppress = 1; |
650 | return NOTIFY_DONE; |
651 | } |
652 | |
653 | /** |
654 | * rcu_cpu_stall_reset - prevent further stall warnings in current grace period |
655 | * |
656 | * Set the stall-warning timeout way off into the future, thus preventing |
657 | * any RCU CPU stall-warning messages from appearing in the current set of |
658 | * RCU grace periods. |
659 | * |
660 | * The caller must disable hard irqs. |
661 | */ |
662 | void rcu_cpu_stall_reset(void) |
663 | { |
664 | rcu_sched_state.jiffies_stall = jiffies + ULONG_MAX / 2; |
665 | rcu_bh_state.jiffies_stall = jiffies + ULONG_MAX / 2; |
666 | rcu_preempt_stall_reset(); |
667 | } |
668 | |
669 | static struct notifier_block rcu_panic_block = { |
670 | .notifier_call = rcu_panic, |
671 | }; |
672 | |
673 | static void __init check_cpu_stall_init(void) |
674 | { |
675 | atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block); |
676 | } |
677 | |
678 | /* |
679 | * Update CPU-local rcu_data state to record the newly noticed grace period. |
680 | * This is used both when we started the grace period and when we notice |
681 | * that someone else started the grace period. The caller must hold the |
682 | * ->lock of the leaf rcu_node structure corresponding to the current CPU, |
683 | * and must have irqs disabled. |
684 | */ |
685 | static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) |
686 | { |
687 | if (rdp->gpnum != rnp->gpnum) { |
688 | /* |
689 | * If the current grace period is waiting for this CPU, |
690 | * set up to detect a quiescent state, otherwise don't |
691 | * go looking for one. |
692 | */ |
693 | rdp->gpnum = rnp->gpnum; |
694 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpustart"); |
695 | if (rnp->qsmask & rdp->grpmask) { |
696 | rdp->qs_pending = 1; |
697 | rdp->passed_quiesce = 0; |
698 | } else |
699 | rdp->qs_pending = 0; |
700 | } |
701 | } |
702 | |
703 | static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp) |
704 | { |
705 | unsigned long flags; |
706 | struct rcu_node *rnp; |
707 | |
708 | local_irq_save(flags); |
709 | rnp = rdp->mynode; |
710 | if (rdp->gpnum == ACCESS_ONCE(rnp->gpnum) || /* outside lock. */ |
711 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
712 | local_irq_restore(flags); |
713 | return; |
714 | } |
715 | __note_new_gpnum(rsp, rnp, rdp); |
716 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
717 | } |
718 | |
719 | /* |
720 | * Did someone else start a new RCU grace period start since we last |
721 | * checked? Update local state appropriately if so. Must be called |
722 | * on the CPU corresponding to rdp. |
723 | */ |
724 | static int |
725 | check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp) |
726 | { |
727 | unsigned long flags; |
728 | int ret = 0; |
729 | |
730 | local_irq_save(flags); |
731 | if (rdp->gpnum != rsp->gpnum) { |
732 | note_new_gpnum(rsp, rdp); |
733 | ret = 1; |
734 | } |
735 | local_irq_restore(flags); |
736 | return ret; |
737 | } |
738 | |
739 | /* |
740 | * Advance this CPU's callbacks, but only if the current grace period |
741 | * has ended. This may be called only from the CPU to whom the rdp |
742 | * belongs. In addition, the corresponding leaf rcu_node structure's |
743 | * ->lock must be held by the caller, with irqs disabled. |
744 | */ |
745 | static void |
746 | __rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) |
747 | { |
748 | /* Did another grace period end? */ |
749 | if (rdp->completed != rnp->completed) { |
750 | |
751 | /* Advance callbacks. No harm if list empty. */ |
752 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL]; |
753 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL]; |
754 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
755 | |
756 | /* Remember that we saw this grace-period completion. */ |
757 | rdp->completed = rnp->completed; |
758 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuend"); |
759 | |
760 | /* |
761 | * If we were in an extended quiescent state, we may have |
762 | * missed some grace periods that others CPUs handled on |
763 | * our behalf. Catch up with this state to avoid noting |
764 | * spurious new grace periods. If another grace period |
765 | * has started, then rnp->gpnum will have advanced, so |
766 | * we will detect this later on. |
767 | */ |
768 | if (ULONG_CMP_LT(rdp->gpnum, rdp->completed)) |
769 | rdp->gpnum = rdp->completed; |
770 | |
771 | /* |
772 | * If RCU does not need a quiescent state from this CPU, |
773 | * then make sure that this CPU doesn't go looking for one. |
774 | */ |
775 | if ((rnp->qsmask & rdp->grpmask) == 0) |
776 | rdp->qs_pending = 0; |
777 | } |
778 | } |
779 | |
780 | /* |
781 | * Advance this CPU's callbacks, but only if the current grace period |
782 | * has ended. This may be called only from the CPU to whom the rdp |
783 | * belongs. |
784 | */ |
785 | static void |
786 | rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp) |
787 | { |
788 | unsigned long flags; |
789 | struct rcu_node *rnp; |
790 | |
791 | local_irq_save(flags); |
792 | rnp = rdp->mynode; |
793 | if (rdp->completed == ACCESS_ONCE(rnp->completed) || /* outside lock. */ |
794 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
795 | local_irq_restore(flags); |
796 | return; |
797 | } |
798 | __rcu_process_gp_end(rsp, rnp, rdp); |
799 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
800 | } |
801 | |
802 | /* |
803 | * Do per-CPU grace-period initialization for running CPU. The caller |
804 | * must hold the lock of the leaf rcu_node structure corresponding to |
805 | * this CPU. |
806 | */ |
807 | static void |
808 | rcu_start_gp_per_cpu(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) |
809 | { |
810 | /* Prior grace period ended, so advance callbacks for current CPU. */ |
811 | __rcu_process_gp_end(rsp, rnp, rdp); |
812 | |
813 | /* |
814 | * Because this CPU just now started the new grace period, we know |
815 | * that all of its callbacks will be covered by this upcoming grace |
816 | * period, even the ones that were registered arbitrarily recently. |
817 | * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL. |
818 | * |
819 | * Other CPUs cannot be sure exactly when the grace period started. |
820 | * Therefore, their recently registered callbacks must pass through |
821 | * an additional RCU_NEXT_READY stage, so that they will be handled |
822 | * by the next RCU grace period. |
823 | */ |
824 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
825 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
826 | |
827 | /* Set state so that this CPU will detect the next quiescent state. */ |
828 | __note_new_gpnum(rsp, rnp, rdp); |
829 | } |
830 | |
831 | /* |
832 | * Start a new RCU grace period if warranted, re-initializing the hierarchy |
833 | * in preparation for detecting the next grace period. The caller must hold |
834 | * the root node's ->lock, which is released before return. Hard irqs must |
835 | * be disabled. |
836 | */ |
837 | static void |
838 | rcu_start_gp(struct rcu_state *rsp, unsigned long flags) |
839 | __releases(rcu_get_root(rsp)->lock) |
840 | { |
841 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
842 | struct rcu_node *rnp = rcu_get_root(rsp); |
843 | |
844 | if (!rcu_scheduler_fully_active || |
845 | !cpu_needs_another_gp(rsp, rdp)) { |
846 | /* |
847 | * Either the scheduler hasn't yet spawned the first |
848 | * non-idle task or this CPU does not need another |
849 | * grace period. Either way, don't start a new grace |
850 | * period. |
851 | */ |
852 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
853 | return; |
854 | } |
855 | |
856 | if (rsp->fqs_active) { |
857 | /* |
858 | * This CPU needs a grace period, but force_quiescent_state() |
859 | * is running. Tell it to start one on this CPU's behalf. |
860 | */ |
861 | rsp->fqs_need_gp = 1; |
862 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
863 | return; |
864 | } |
865 | |
866 | /* Advance to a new grace period and initialize state. */ |
867 | rsp->gpnum++; |
868 | trace_rcu_grace_period(rsp->name, rsp->gpnum, "start"); |
869 | WARN_ON_ONCE(rsp->signaled == RCU_GP_INIT); |
870 | rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */ |
871 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
872 | record_gp_stall_check_time(rsp); |
873 | |
874 | /* Special-case the common single-level case. */ |
875 | if (NUM_RCU_NODES == 1) { |
876 | rcu_preempt_check_blocked_tasks(rnp); |
877 | rnp->qsmask = rnp->qsmaskinit; |
878 | rnp->gpnum = rsp->gpnum; |
879 | rnp->completed = rsp->completed; |
880 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */ |
881 | rcu_start_gp_per_cpu(rsp, rnp, rdp); |
882 | rcu_preempt_boost_start_gp(rnp); |
883 | trace_rcu_grace_period_init(rsp->name, rnp->gpnum, |
884 | rnp->level, rnp->grplo, |
885 | rnp->grphi, rnp->qsmask); |
886 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
887 | return; |
888 | } |
889 | |
890 | raw_spin_unlock(&rnp->lock); /* leave irqs disabled. */ |
891 | |
892 | |
893 | /* Exclude any concurrent CPU-hotplug operations. */ |
894 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
895 | |
896 | /* |
897 | * Set the quiescent-state-needed bits in all the rcu_node |
898 | * structures for all currently online CPUs in breadth-first |
899 | * order, starting from the root rcu_node structure. This |
900 | * operation relies on the layout of the hierarchy within the |
901 | * rsp->node[] array. Note that other CPUs will access only |
902 | * the leaves of the hierarchy, which still indicate that no |
903 | * grace period is in progress, at least until the corresponding |
904 | * leaf node has been initialized. In addition, we have excluded |
905 | * CPU-hotplug operations. |
906 | * |
907 | * Note that the grace period cannot complete until we finish |
908 | * the initialization process, as there will be at least one |
909 | * qsmask bit set in the root node until that time, namely the |
910 | * one corresponding to this CPU, due to the fact that we have |
911 | * irqs disabled. |
912 | */ |
913 | rcu_for_each_node_breadth_first(rsp, rnp) { |
914 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
915 | rcu_preempt_check_blocked_tasks(rnp); |
916 | rnp->qsmask = rnp->qsmaskinit; |
917 | rnp->gpnum = rsp->gpnum; |
918 | rnp->completed = rsp->completed; |
919 | if (rnp == rdp->mynode) |
920 | rcu_start_gp_per_cpu(rsp, rnp, rdp); |
921 | rcu_preempt_boost_start_gp(rnp); |
922 | trace_rcu_grace_period_init(rsp->name, rnp->gpnum, |
923 | rnp->level, rnp->grplo, |
924 | rnp->grphi, rnp->qsmask); |
925 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
926 | } |
927 | |
928 | rnp = rcu_get_root(rsp); |
929 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
930 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */ |
931 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
932 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); |
933 | } |
934 | |
935 | /* |
936 | * Report a full set of quiescent states to the specified rcu_state |
937 | * data structure. This involves cleaning up after the prior grace |
938 | * period and letting rcu_start_gp() start up the next grace period |
939 | * if one is needed. Note that the caller must hold rnp->lock, as |
940 | * required by rcu_start_gp(), which will release it. |
941 | */ |
942 | static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags) |
943 | __releases(rcu_get_root(rsp)->lock) |
944 | { |
945 | unsigned long gp_duration; |
946 | struct rcu_node *rnp = rcu_get_root(rsp); |
947 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
948 | |
949 | WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); |
950 | |
951 | /* |
952 | * Ensure that all grace-period and pre-grace-period activity |
953 | * is seen before the assignment to rsp->completed. |
954 | */ |
955 | smp_mb(); /* See above block comment. */ |
956 | gp_duration = jiffies - rsp->gp_start; |
957 | if (gp_duration > rsp->gp_max) |
958 | rsp->gp_max = gp_duration; |
959 | |
960 | /* |
961 | * We know the grace period is complete, but to everyone else |
962 | * it appears to still be ongoing. But it is also the case |
963 | * that to everyone else it looks like there is nothing that |
964 | * they can do to advance the grace period. It is therefore |
965 | * safe for us to drop the lock in order to mark the grace |
966 | * period as completed in all of the rcu_node structures. |
967 | * |
968 | * But if this CPU needs another grace period, it will take |
969 | * care of this while initializing the next grace period. |
970 | * We use RCU_WAIT_TAIL instead of the usual RCU_DONE_TAIL |
971 | * because the callbacks have not yet been advanced: Those |
972 | * callbacks are waiting on the grace period that just now |
973 | * completed. |
974 | */ |
975 | if (*rdp->nxttail[RCU_WAIT_TAIL] == NULL) { |
976 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
977 | |
978 | /* |
979 | * Propagate new ->completed value to rcu_node structures |
980 | * so that other CPUs don't have to wait until the start |
981 | * of the next grace period to process their callbacks. |
982 | */ |
983 | rcu_for_each_node_breadth_first(rsp, rnp) { |
984 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
985 | rnp->completed = rsp->gpnum; |
986 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
987 | } |
988 | rnp = rcu_get_root(rsp); |
989 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
990 | } |
991 | |
992 | rsp->completed = rsp->gpnum; /* Declare the grace period complete. */ |
993 | trace_rcu_grace_period(rsp->name, rsp->completed, "end"); |
994 | rsp->signaled = RCU_GP_IDLE; |
995 | rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */ |
996 | } |
997 | |
998 | /* |
999 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
1000 | * Allows quiescent states for a group of CPUs to be reported at one go |
1001 | * to the specified rcu_node structure, though all the CPUs in the group |
1002 | * must be represented by the same rcu_node structure (which need not be |
1003 | * a leaf rcu_node structure, though it often will be). That structure's |
1004 | * lock must be held upon entry, and it is released before return. |
1005 | */ |
1006 | static void |
1007 | rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp, |
1008 | struct rcu_node *rnp, unsigned long flags) |
1009 | __releases(rnp->lock) |
1010 | { |
1011 | struct rcu_node *rnp_c; |
1012 | |
1013 | /* Walk up the rcu_node hierarchy. */ |
1014 | for (;;) { |
1015 | if (!(rnp->qsmask & mask)) { |
1016 | |
1017 | /* Our bit has already been cleared, so done. */ |
1018 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1019 | return; |
1020 | } |
1021 | rnp->qsmask &= ~mask; |
1022 | trace_rcu_quiescent_state_report(rsp->name, rnp->gpnum, |
1023 | mask, rnp->qsmask, rnp->level, |
1024 | rnp->grplo, rnp->grphi, |
1025 | !!rnp->gp_tasks); |
1026 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
1027 | |
1028 | /* Other bits still set at this level, so done. */ |
1029 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1030 | return; |
1031 | } |
1032 | mask = rnp->grpmask; |
1033 | if (rnp->parent == NULL) { |
1034 | |
1035 | /* No more levels. Exit loop holding root lock. */ |
1036 | |
1037 | break; |
1038 | } |
1039 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1040 | rnp_c = rnp; |
1041 | rnp = rnp->parent; |
1042 | raw_spin_lock_irqsave(&rnp->lock, flags); |
1043 | WARN_ON_ONCE(rnp_c->qsmask); |
1044 | } |
1045 | |
1046 | /* |
1047 | * Get here if we are the last CPU to pass through a quiescent |
1048 | * state for this grace period. Invoke rcu_report_qs_rsp() |
1049 | * to clean up and start the next grace period if one is needed. |
1050 | */ |
1051 | rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */ |
1052 | } |
1053 | |
1054 | /* |
1055 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
1056 | * structure. This must be either called from the specified CPU, or |
1057 | * called when the specified CPU is known to be offline (and when it is |
1058 | * also known that no other CPU is concurrently trying to help the offline |
1059 | * CPU). The lastcomp argument is used to make sure we are still in the |
1060 | * grace period of interest. We don't want to end the current grace period |
1061 | * based on quiescent states detected in an earlier grace period! |
1062 | */ |
1063 | static void |
1064 | rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastgp) |
1065 | { |
1066 | unsigned long flags; |
1067 | unsigned long mask; |
1068 | struct rcu_node *rnp; |
1069 | |
1070 | rnp = rdp->mynode; |
1071 | raw_spin_lock_irqsave(&rnp->lock, flags); |
1072 | if (lastgp != rnp->gpnum || rnp->completed == rnp->gpnum) { |
1073 | |
1074 | /* |
1075 | * The grace period in which this quiescent state was |
1076 | * recorded has ended, so don't report it upwards. |
1077 | * We will instead need a new quiescent state that lies |
1078 | * within the current grace period. |
1079 | */ |
1080 | rdp->passed_quiesce = 0; /* need qs for new gp. */ |
1081 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1082 | return; |
1083 | } |
1084 | mask = rdp->grpmask; |
1085 | if ((rnp->qsmask & mask) == 0) { |
1086 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1087 | } else { |
1088 | rdp->qs_pending = 0; |
1089 | |
1090 | /* |
1091 | * This GP can't end until cpu checks in, so all of our |
1092 | * callbacks can be processed during the next GP. |
1093 | */ |
1094 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
1095 | |
1096 | rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */ |
1097 | } |
1098 | } |
1099 | |
1100 | /* |
1101 | * Check to see if there is a new grace period of which this CPU |
1102 | * is not yet aware, and if so, set up local rcu_data state for it. |
1103 | * Otherwise, see if this CPU has just passed through its first |
1104 | * quiescent state for this grace period, and record that fact if so. |
1105 | */ |
1106 | static void |
1107 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) |
1108 | { |
1109 | /* If there is now a new grace period, record and return. */ |
1110 | if (check_for_new_grace_period(rsp, rdp)) |
1111 | return; |
1112 | |
1113 | /* |
1114 | * Does this CPU still need to do its part for current grace period? |
1115 | * If no, return and let the other CPUs do their part as well. |
1116 | */ |
1117 | if (!rdp->qs_pending) |
1118 | return; |
1119 | |
1120 | /* |
1121 | * Was there a quiescent state since the beginning of the grace |
1122 | * period? If no, then exit and wait for the next call. |
1123 | */ |
1124 | if (!rdp->passed_quiesce) |
1125 | return; |
1126 | |
1127 | /* |
1128 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the |
1129 | * judge of that). |
1130 | */ |
1131 | rcu_report_qs_rdp(rdp->cpu, rsp, rdp, rdp->passed_quiesce_gpnum); |
1132 | } |
1133 | |
1134 | #ifdef CONFIG_HOTPLUG_CPU |
1135 | |
1136 | /* |
1137 | * Move a dying CPU's RCU callbacks to online CPU's callback list. |
1138 | * Synchronization is not required because this function executes |
1139 | * in stop_machine() context. |
1140 | */ |
1141 | static void rcu_send_cbs_to_online(struct rcu_state *rsp) |
1142 | { |
1143 | int i; |
1144 | /* current DYING CPU is cleared in the cpu_online_mask */ |
1145 | int receive_cpu = cpumask_any(cpu_online_mask); |
1146 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
1147 | struct rcu_data *receive_rdp = per_cpu_ptr(rsp->rda, receive_cpu); |
1148 | |
1149 | if (rdp->nxtlist == NULL) |
1150 | return; /* irqs disabled, so comparison is stable. */ |
1151 | |
1152 | *receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist; |
1153 | receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
1154 | receive_rdp->qlen += rdp->qlen; |
1155 | receive_rdp->n_cbs_adopted += rdp->qlen; |
1156 | rdp->n_cbs_orphaned += rdp->qlen; |
1157 | |
1158 | rdp->nxtlist = NULL; |
1159 | for (i = 0; i < RCU_NEXT_SIZE; i++) |
1160 | rdp->nxttail[i] = &rdp->nxtlist; |
1161 | rdp->qlen = 0; |
1162 | } |
1163 | |
1164 | /* |
1165 | * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy |
1166 | * and move all callbacks from the outgoing CPU to the current one. |
1167 | * There can only be one CPU hotplug operation at a time, so no other |
1168 | * CPU can be attempting to update rcu_cpu_kthread_task. |
1169 | */ |
1170 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp) |
1171 | { |
1172 | unsigned long flags; |
1173 | unsigned long mask; |
1174 | int need_report = 0; |
1175 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
1176 | struct rcu_node *rnp; |
1177 | |
1178 | rcu_stop_cpu_kthread(cpu); |
1179 | |
1180 | /* Exclude any attempts to start a new grace period. */ |
1181 | raw_spin_lock_irqsave(&rsp->onofflock, flags); |
1182 | |
1183 | /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */ |
1184 | rnp = rdp->mynode; /* this is the outgoing CPU's rnp. */ |
1185 | mask = rdp->grpmask; /* rnp->grplo is constant. */ |
1186 | do { |
1187 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
1188 | rnp->qsmaskinit &= ~mask; |
1189 | if (rnp->qsmaskinit != 0) { |
1190 | if (rnp != rdp->mynode) |
1191 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
1192 | else |
1193 | trace_rcu_grace_period(rsp->name, |
1194 | rnp->gpnum + 1 - |
1195 | !!(rnp->qsmask & mask), |
1196 | "cpuofl"); |
1197 | break; |
1198 | } |
1199 | if (rnp == rdp->mynode) { |
1200 | trace_rcu_grace_period(rsp->name, |
1201 | rnp->gpnum + 1 - |
1202 | !!(rnp->qsmask & mask), |
1203 | "cpuofl"); |
1204 | need_report = rcu_preempt_offline_tasks(rsp, rnp, rdp); |
1205 | } else |
1206 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
1207 | mask = rnp->grpmask; |
1208 | rnp = rnp->parent; |
1209 | } while (rnp != NULL); |
1210 | |
1211 | /* |
1212 | * We still hold the leaf rcu_node structure lock here, and |
1213 | * irqs are still disabled. The reason for this subterfuge is |
1214 | * because invoking rcu_report_unblock_qs_rnp() with ->onofflock |
1215 | * held leads to deadlock. |
1216 | */ |
1217 | raw_spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ |
1218 | rnp = rdp->mynode; |
1219 | if (need_report & RCU_OFL_TASKS_NORM_GP) |
1220 | rcu_report_unblock_qs_rnp(rnp, flags); |
1221 | else |
1222 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1223 | if (need_report & RCU_OFL_TASKS_EXP_GP) |
1224 | rcu_report_exp_rnp(rsp, rnp); |
1225 | rcu_node_kthread_setaffinity(rnp, -1); |
1226 | } |
1227 | |
1228 | /* |
1229 | * Remove the specified CPU from the RCU hierarchy and move any pending |
1230 | * callbacks that it might have to the current CPU. This code assumes |
1231 | * that at least one CPU in the system will remain running at all times. |
1232 | * Any attempt to offline -all- CPUs is likely to strand RCU callbacks. |
1233 | */ |
1234 | static void rcu_offline_cpu(int cpu) |
1235 | { |
1236 | __rcu_offline_cpu(cpu, &rcu_sched_state); |
1237 | __rcu_offline_cpu(cpu, &rcu_bh_state); |
1238 | rcu_preempt_offline_cpu(cpu); |
1239 | } |
1240 | |
1241 | #else /* #ifdef CONFIG_HOTPLUG_CPU */ |
1242 | |
1243 | static void rcu_send_cbs_to_online(struct rcu_state *rsp) |
1244 | { |
1245 | } |
1246 | |
1247 | static void rcu_offline_cpu(int cpu) |
1248 | { |
1249 | } |
1250 | |
1251 | #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */ |
1252 | |
1253 | /* |
1254 | * Invoke any RCU callbacks that have made it to the end of their grace |
1255 | * period. Thottle as specified by rdp->blimit. |
1256 | */ |
1257 | static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp) |
1258 | { |
1259 | unsigned long flags; |
1260 | struct rcu_head *next, *list, **tail; |
1261 | int bl, count; |
1262 | |
1263 | /* If no callbacks are ready, just return.*/ |
1264 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) { |
1265 | trace_rcu_batch_start(rsp->name, 0, 0); |
1266 | trace_rcu_batch_end(rsp->name, 0); |
1267 | return; |
1268 | } |
1269 | |
1270 | /* |
1271 | * Extract the list of ready callbacks, disabling to prevent |
1272 | * races with call_rcu() from interrupt handlers. |
1273 | */ |
1274 | local_irq_save(flags); |
1275 | bl = rdp->blimit; |
1276 | trace_rcu_batch_start(rsp->name, rdp->qlen, bl); |
1277 | list = rdp->nxtlist; |
1278 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; |
1279 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; |
1280 | tail = rdp->nxttail[RCU_DONE_TAIL]; |
1281 | for (count = RCU_NEXT_SIZE - 1; count >= 0; count--) |
1282 | if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL]) |
1283 | rdp->nxttail[count] = &rdp->nxtlist; |
1284 | local_irq_restore(flags); |
1285 | |
1286 | /* Invoke callbacks. */ |
1287 | count = 0; |
1288 | while (list) { |
1289 | next = list->next; |
1290 | prefetch(next); |
1291 | debug_rcu_head_unqueue(list); |
1292 | __rcu_reclaim(rsp->name, list); |
1293 | list = next; |
1294 | if (++count >= bl) |
1295 | break; |
1296 | } |
1297 | |
1298 | local_irq_save(flags); |
1299 | trace_rcu_batch_end(rsp->name, count); |
1300 | |
1301 | /* Update count, and requeue any remaining callbacks. */ |
1302 | rdp->qlen -= count; |
1303 | rdp->n_cbs_invoked += count; |
1304 | if (list != NULL) { |
1305 | *tail = rdp->nxtlist; |
1306 | rdp->nxtlist = list; |
1307 | for (count = 0; count < RCU_NEXT_SIZE; count++) |
1308 | if (&rdp->nxtlist == rdp->nxttail[count]) |
1309 | rdp->nxttail[count] = tail; |
1310 | else |
1311 | break; |
1312 | } |
1313 | |
1314 | /* Reinstate batch limit if we have worked down the excess. */ |
1315 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) |
1316 | rdp->blimit = blimit; |
1317 | |
1318 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
1319 | if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) { |
1320 | rdp->qlen_last_fqs_check = 0; |
1321 | rdp->n_force_qs_snap = rsp->n_force_qs; |
1322 | } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark) |
1323 | rdp->qlen_last_fqs_check = rdp->qlen; |
1324 | |
1325 | local_irq_restore(flags); |
1326 | |
1327 | /* Re-invoke RCU core processing if there are callbacks remaining. */ |
1328 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
1329 | invoke_rcu_core(); |
1330 | } |
1331 | |
1332 | /* |
1333 | * Check to see if this CPU is in a non-context-switch quiescent state |
1334 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). |
1335 | * Also schedule RCU core processing. |
1336 | * |
1337 | * This function must be called with hardirqs disabled. It is normally |
1338 | * invoked from the scheduling-clock interrupt. If rcu_pending returns |
1339 | * false, there is no point in invoking rcu_check_callbacks(). |
1340 | */ |
1341 | void rcu_check_callbacks(int cpu, int user) |
1342 | { |
1343 | trace_rcu_utilization("Start scheduler-tick"); |
1344 | if (user || |
1345 | (idle_cpu(cpu) && rcu_scheduler_active && |
1346 | !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) { |
1347 | |
1348 | /* |
1349 | * Get here if this CPU took its interrupt from user |
1350 | * mode or from the idle loop, and if this is not a |
1351 | * nested interrupt. In this case, the CPU is in |
1352 | * a quiescent state, so note it. |
1353 | * |
1354 | * No memory barrier is required here because both |
1355 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
1356 | * variables that other CPUs neither access nor modify, |
1357 | * at least not while the corresponding CPU is online. |
1358 | */ |
1359 | |
1360 | rcu_sched_qs(cpu); |
1361 | rcu_bh_qs(cpu); |
1362 | |
1363 | } else if (!in_softirq()) { |
1364 | |
1365 | /* |
1366 | * Get here if this CPU did not take its interrupt from |
1367 | * softirq, in other words, if it is not interrupting |
1368 | * a rcu_bh read-side critical section. This is an _bh |
1369 | * critical section, so note it. |
1370 | */ |
1371 | |
1372 | rcu_bh_qs(cpu); |
1373 | } |
1374 | rcu_preempt_check_callbacks(cpu); |
1375 | if (rcu_pending(cpu)) |
1376 | invoke_rcu_core(); |
1377 | trace_rcu_utilization("End scheduler-tick"); |
1378 | } |
1379 | |
1380 | #ifdef CONFIG_SMP |
1381 | |
1382 | /* |
1383 | * Scan the leaf rcu_node structures, processing dyntick state for any that |
1384 | * have not yet encountered a quiescent state, using the function specified. |
1385 | * Also initiate boosting for any threads blocked on the root rcu_node. |
1386 | * |
1387 | * The caller must have suppressed start of new grace periods. |
1388 | */ |
1389 | static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *)) |
1390 | { |
1391 | unsigned long bit; |
1392 | int cpu; |
1393 | unsigned long flags; |
1394 | unsigned long mask; |
1395 | struct rcu_node *rnp; |
1396 | |
1397 | rcu_for_each_leaf_node(rsp, rnp) { |
1398 | mask = 0; |
1399 | raw_spin_lock_irqsave(&rnp->lock, flags); |
1400 | if (!rcu_gp_in_progress(rsp)) { |
1401 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1402 | return; |
1403 | } |
1404 | if (rnp->qsmask == 0) { |
1405 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock */ |
1406 | continue; |
1407 | } |
1408 | cpu = rnp->grplo; |
1409 | bit = 1; |
1410 | for (; cpu <= rnp->grphi; cpu++, bit <<= 1) { |
1411 | if ((rnp->qsmask & bit) != 0 && |
1412 | f(per_cpu_ptr(rsp->rda, cpu))) |
1413 | mask |= bit; |
1414 | } |
1415 | if (mask != 0) { |
1416 | |
1417 | /* rcu_report_qs_rnp() releases rnp->lock. */ |
1418 | rcu_report_qs_rnp(mask, rsp, rnp, flags); |
1419 | continue; |
1420 | } |
1421 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1422 | } |
1423 | rnp = rcu_get_root(rsp); |
1424 | if (rnp->qsmask == 0) { |
1425 | raw_spin_lock_irqsave(&rnp->lock, flags); |
1426 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */ |
1427 | } |
1428 | } |
1429 | |
1430 | /* |
1431 | * Force quiescent states on reluctant CPUs, and also detect which |
1432 | * CPUs are in dyntick-idle mode. |
1433 | */ |
1434 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) |
1435 | { |
1436 | unsigned long flags; |
1437 | struct rcu_node *rnp = rcu_get_root(rsp); |
1438 | |
1439 | trace_rcu_utilization("Start fqs"); |
1440 | if (!rcu_gp_in_progress(rsp)) { |
1441 | trace_rcu_utilization("End fqs"); |
1442 | return; /* No grace period in progress, nothing to force. */ |
1443 | } |
1444 | if (!raw_spin_trylock_irqsave(&rsp->fqslock, flags)) { |
1445 | rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ |
1446 | trace_rcu_utilization("End fqs"); |
1447 | return; /* Someone else is already on the job. */ |
1448 | } |
1449 | if (relaxed && ULONG_CMP_GE(rsp->jiffies_force_qs, jiffies)) |
1450 | goto unlock_fqs_ret; /* no emergency and done recently. */ |
1451 | rsp->n_force_qs++; |
1452 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
1453 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
1454 | if(!rcu_gp_in_progress(rsp)) { |
1455 | rsp->n_force_qs_ngp++; |
1456 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
1457 | goto unlock_fqs_ret; /* no GP in progress, time updated. */ |
1458 | } |
1459 | rsp->fqs_active = 1; |
1460 | switch (rsp->signaled) { |
1461 | case RCU_GP_IDLE: |
1462 | case RCU_GP_INIT: |
1463 | |
1464 | break; /* grace period idle or initializing, ignore. */ |
1465 | |
1466 | case RCU_SAVE_DYNTICK: |
1467 | if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK) |
1468 | break; /* So gcc recognizes the dead code. */ |
1469 | |
1470 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
1471 | |
1472 | /* Record dyntick-idle state. */ |
1473 | force_qs_rnp(rsp, dyntick_save_progress_counter); |
1474 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
1475 | if (rcu_gp_in_progress(rsp)) |
1476 | rsp->signaled = RCU_FORCE_QS; |
1477 | break; |
1478 | |
1479 | case RCU_FORCE_QS: |
1480 | |
1481 | /* Check dyntick-idle state, send IPI to laggarts. */ |
1482 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
1483 | force_qs_rnp(rsp, rcu_implicit_dynticks_qs); |
1484 | |
1485 | /* Leave state in case more forcing is required. */ |
1486 | |
1487 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
1488 | break; |
1489 | } |
1490 | rsp->fqs_active = 0; |
1491 | if (rsp->fqs_need_gp) { |
1492 | raw_spin_unlock(&rsp->fqslock); /* irqs remain disabled */ |
1493 | rsp->fqs_need_gp = 0; |
1494 | rcu_start_gp(rsp, flags); /* releases rnp->lock */ |
1495 | trace_rcu_utilization("End fqs"); |
1496 | return; |
1497 | } |
1498 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
1499 | unlock_fqs_ret: |
1500 | raw_spin_unlock_irqrestore(&rsp->fqslock, flags); |
1501 | trace_rcu_utilization("End fqs"); |
1502 | } |
1503 | |
1504 | #else /* #ifdef CONFIG_SMP */ |
1505 | |
1506 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) |
1507 | { |
1508 | set_need_resched(); |
1509 | } |
1510 | |
1511 | #endif /* #else #ifdef CONFIG_SMP */ |
1512 | |
1513 | /* |
1514 | * This does the RCU core processing work for the specified rcu_state |
1515 | * and rcu_data structures. This may be called only from the CPU to |
1516 | * whom the rdp belongs. |
1517 | */ |
1518 | static void |
1519 | __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) |
1520 | { |
1521 | unsigned long flags; |
1522 | |
1523 | WARN_ON_ONCE(rdp->beenonline == 0); |
1524 | |
1525 | /* |
1526 | * If an RCU GP has gone long enough, go check for dyntick |
1527 | * idle CPUs and, if needed, send resched IPIs. |
1528 | */ |
1529 | if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
1530 | force_quiescent_state(rsp, 1); |
1531 | |
1532 | /* |
1533 | * Advance callbacks in response to end of earlier grace |
1534 | * period that some other CPU ended. |
1535 | */ |
1536 | rcu_process_gp_end(rsp, rdp); |
1537 | |
1538 | /* Update RCU state based on any recent quiescent states. */ |
1539 | rcu_check_quiescent_state(rsp, rdp); |
1540 | |
1541 | /* Does this CPU require a not-yet-started grace period? */ |
1542 | if (cpu_needs_another_gp(rsp, rdp)) { |
1543 | raw_spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags); |
1544 | rcu_start_gp(rsp, flags); /* releases above lock */ |
1545 | } |
1546 | |
1547 | /* If there are callbacks ready, invoke them. */ |
1548 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
1549 | invoke_rcu_callbacks(rsp, rdp); |
1550 | } |
1551 | |
1552 | /* |
1553 | * Do RCU core processing for the current CPU. |
1554 | */ |
1555 | static void rcu_process_callbacks(struct softirq_action *unused) |
1556 | { |
1557 | trace_rcu_utilization("Start RCU core"); |
1558 | __rcu_process_callbacks(&rcu_sched_state, |
1559 | &__get_cpu_var(rcu_sched_data)); |
1560 | __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); |
1561 | rcu_preempt_process_callbacks(); |
1562 | trace_rcu_utilization("End RCU core"); |
1563 | } |
1564 | |
1565 | /* |
1566 | * Schedule RCU callback invocation. If the specified type of RCU |
1567 | * does not support RCU priority boosting, just do a direct call, |
1568 | * otherwise wake up the per-CPU kernel kthread. Note that because we |
1569 | * are running on the current CPU with interrupts disabled, the |
1570 | * rcu_cpu_kthread_task cannot disappear out from under us. |
1571 | */ |
1572 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) |
1573 | { |
1574 | if (unlikely(!ACCESS_ONCE(rcu_scheduler_fully_active))) |
1575 | return; |
1576 | if (likely(!rsp->boost)) { |
1577 | rcu_do_batch(rsp, rdp); |
1578 | return; |
1579 | } |
1580 | invoke_rcu_callbacks_kthread(); |
1581 | } |
1582 | |
1583 | static void invoke_rcu_core(void) |
1584 | { |
1585 | raise_softirq(RCU_SOFTIRQ); |
1586 | } |
1587 | |
1588 | static void |
1589 | __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu), |
1590 | struct rcu_state *rsp) |
1591 | { |
1592 | unsigned long flags; |
1593 | struct rcu_data *rdp; |
1594 | |
1595 | debug_rcu_head_queue(head); |
1596 | head->func = func; |
1597 | head->next = NULL; |
1598 | |
1599 | smp_mb(); /* Ensure RCU update seen before callback registry. */ |
1600 | |
1601 | /* |
1602 | * Opportunistically note grace-period endings and beginnings. |
1603 | * Note that we might see a beginning right after we see an |
1604 | * end, but never vice versa, since this CPU has to pass through |
1605 | * a quiescent state betweentimes. |
1606 | */ |
1607 | local_irq_save(flags); |
1608 | rdp = this_cpu_ptr(rsp->rda); |
1609 | |
1610 | /* Add the callback to our list. */ |
1611 | *rdp->nxttail[RCU_NEXT_TAIL] = head; |
1612 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; |
1613 | rdp->qlen++; |
1614 | |
1615 | if (__is_kfree_rcu_offset((unsigned long)func)) |
1616 | trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func, |
1617 | rdp->qlen); |
1618 | else |
1619 | trace_rcu_callback(rsp->name, head, rdp->qlen); |
1620 | |
1621 | /* If interrupts were disabled, don't dive into RCU core. */ |
1622 | if (irqs_disabled_flags(flags)) { |
1623 | local_irq_restore(flags); |
1624 | return; |
1625 | } |
1626 | |
1627 | /* |
1628 | * Force the grace period if too many callbacks or too long waiting. |
1629 | * Enforce hysteresis, and don't invoke force_quiescent_state() |
1630 | * if some other CPU has recently done so. Also, don't bother |
1631 | * invoking force_quiescent_state() if the newly enqueued callback |
1632 | * is the only one waiting for a grace period to complete. |
1633 | */ |
1634 | if (unlikely(rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) { |
1635 | |
1636 | /* Are we ignoring a completed grace period? */ |
1637 | rcu_process_gp_end(rsp, rdp); |
1638 | check_for_new_grace_period(rsp, rdp); |
1639 | |
1640 | /* Start a new grace period if one not already started. */ |
1641 | if (!rcu_gp_in_progress(rsp)) { |
1642 | unsigned long nestflag; |
1643 | struct rcu_node *rnp_root = rcu_get_root(rsp); |
1644 | |
1645 | raw_spin_lock_irqsave(&rnp_root->lock, nestflag); |
1646 | rcu_start_gp(rsp, nestflag); /* rlses rnp_root->lock */ |
1647 | } else { |
1648 | /* Give the grace period a kick. */ |
1649 | rdp->blimit = LONG_MAX; |
1650 | if (rsp->n_force_qs == rdp->n_force_qs_snap && |
1651 | *rdp->nxttail[RCU_DONE_TAIL] != head) |
1652 | force_quiescent_state(rsp, 0); |
1653 | rdp->n_force_qs_snap = rsp->n_force_qs; |
1654 | rdp->qlen_last_fqs_check = rdp->qlen; |
1655 | } |
1656 | } else if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
1657 | force_quiescent_state(rsp, 1); |
1658 | local_irq_restore(flags); |
1659 | } |
1660 | |
1661 | /* |
1662 | * Queue an RCU-sched callback for invocation after a grace period. |
1663 | */ |
1664 | void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) |
1665 | { |
1666 | __call_rcu(head, func, &rcu_sched_state); |
1667 | } |
1668 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
1669 | |
1670 | /* |
1671 | * Queue an RCU for invocation after a quicker grace period. |
1672 | */ |
1673 | void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) |
1674 | { |
1675 | __call_rcu(head, func, &rcu_bh_state); |
1676 | } |
1677 | EXPORT_SYMBOL_GPL(call_rcu_bh); |
1678 | |
1679 | /** |
1680 | * synchronize_sched - wait until an rcu-sched grace period has elapsed. |
1681 | * |
1682 | * Control will return to the caller some time after a full rcu-sched |
1683 | * grace period has elapsed, in other words after all currently executing |
1684 | * rcu-sched read-side critical sections have completed. These read-side |
1685 | * critical sections are delimited by rcu_read_lock_sched() and |
1686 | * rcu_read_unlock_sched(), and may be nested. Note that preempt_disable(), |
1687 | * local_irq_disable(), and so on may be used in place of |
1688 | * rcu_read_lock_sched(). |
1689 | * |
1690 | * This means that all preempt_disable code sequences, including NMI and |
1691 | * hardware-interrupt handlers, in progress on entry will have completed |
1692 | * before this primitive returns. However, this does not guarantee that |
1693 | * softirq handlers will have completed, since in some kernels, these |
1694 | * handlers can run in process context, and can block. |
1695 | * |
1696 | * This primitive provides the guarantees made by the (now removed) |
1697 | * synchronize_kernel() API. In contrast, synchronize_rcu() only |
1698 | * guarantees that rcu_read_lock() sections will have completed. |
1699 | * In "classic RCU", these two guarantees happen to be one and |
1700 | * the same, but can differ in realtime RCU implementations. |
1701 | */ |
1702 | void synchronize_sched(void) |
1703 | { |
1704 | if (rcu_blocking_is_gp()) |
1705 | return; |
1706 | wait_rcu_gp(call_rcu_sched); |
1707 | } |
1708 | EXPORT_SYMBOL_GPL(synchronize_sched); |
1709 | |
1710 | /** |
1711 | * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed. |
1712 | * |
1713 | * Control will return to the caller some time after a full rcu_bh grace |
1714 | * period has elapsed, in other words after all currently executing rcu_bh |
1715 | * read-side critical sections have completed. RCU read-side critical |
1716 | * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(), |
1717 | * and may be nested. |
1718 | */ |
1719 | void synchronize_rcu_bh(void) |
1720 | { |
1721 | if (rcu_blocking_is_gp()) |
1722 | return; |
1723 | wait_rcu_gp(call_rcu_bh); |
1724 | } |
1725 | EXPORT_SYMBOL_GPL(synchronize_rcu_bh); |
1726 | |
1727 | /* |
1728 | * Check to see if there is any immediate RCU-related work to be done |
1729 | * by the current CPU, for the specified type of RCU, returning 1 if so. |
1730 | * The checks are in order of increasing expense: checks that can be |
1731 | * carried out against CPU-local state are performed first. However, |
1732 | * we must check for CPU stalls first, else we might not get a chance. |
1733 | */ |
1734 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) |
1735 | { |
1736 | struct rcu_node *rnp = rdp->mynode; |
1737 | |
1738 | rdp->n_rcu_pending++; |
1739 | |
1740 | /* Check for CPU stalls, if enabled. */ |
1741 | check_cpu_stall(rsp, rdp); |
1742 | |
1743 | /* Is the RCU core waiting for a quiescent state from this CPU? */ |
1744 | if (rcu_scheduler_fully_active && |
1745 | rdp->qs_pending && !rdp->passed_quiesce) { |
1746 | |
1747 | /* |
1748 | * If force_quiescent_state() coming soon and this CPU |
1749 | * needs a quiescent state, and this is either RCU-sched |
1750 | * or RCU-bh, force a local reschedule. |
1751 | */ |
1752 | rdp->n_rp_qs_pending++; |
1753 | if (!rdp->preemptible && |
1754 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs) - 1, |
1755 | jiffies)) |
1756 | set_need_resched(); |
1757 | } else if (rdp->qs_pending && rdp->passed_quiesce) { |
1758 | rdp->n_rp_report_qs++; |
1759 | return 1; |
1760 | } |
1761 | |
1762 | /* Does this CPU have callbacks ready to invoke? */ |
1763 | if (cpu_has_callbacks_ready_to_invoke(rdp)) { |
1764 | rdp->n_rp_cb_ready++; |
1765 | return 1; |
1766 | } |
1767 | |
1768 | /* Has RCU gone idle with this CPU needing another grace period? */ |
1769 | if (cpu_needs_another_gp(rsp, rdp)) { |
1770 | rdp->n_rp_cpu_needs_gp++; |
1771 | return 1; |
1772 | } |
1773 | |
1774 | /* Has another RCU grace period completed? */ |
1775 | if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */ |
1776 | rdp->n_rp_gp_completed++; |
1777 | return 1; |
1778 | } |
1779 | |
1780 | /* Has a new RCU grace period started? */ |
1781 | if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */ |
1782 | rdp->n_rp_gp_started++; |
1783 | return 1; |
1784 | } |
1785 | |
1786 | /* Has an RCU GP gone long enough to send resched IPIs &c? */ |
1787 | if (rcu_gp_in_progress(rsp) && |
1788 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) { |
1789 | rdp->n_rp_need_fqs++; |
1790 | return 1; |
1791 | } |
1792 | |
1793 | /* nothing to do */ |
1794 | rdp->n_rp_need_nothing++; |
1795 | return 0; |
1796 | } |
1797 | |
1798 | /* |
1799 | * Check to see if there is any immediate RCU-related work to be done |
1800 | * by the current CPU, returning 1 if so. This function is part of the |
1801 | * RCU implementation; it is -not- an exported member of the RCU API. |
1802 | */ |
1803 | static int rcu_pending(int cpu) |
1804 | { |
1805 | return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) || |
1806 | __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) || |
1807 | rcu_preempt_pending(cpu); |
1808 | } |
1809 | |
1810 | /* |
1811 | * Check to see if any future RCU-related work will need to be done |
1812 | * by the current CPU, even if none need be done immediately, returning |
1813 | * 1 if so. |
1814 | */ |
1815 | static int rcu_needs_cpu_quick_check(int cpu) |
1816 | { |
1817 | /* RCU callbacks either ready or pending? */ |
1818 | return per_cpu(rcu_sched_data, cpu).nxtlist || |
1819 | per_cpu(rcu_bh_data, cpu).nxtlist || |
1820 | rcu_preempt_needs_cpu(cpu); |
1821 | } |
1822 | |
1823 | static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL}; |
1824 | static atomic_t rcu_barrier_cpu_count; |
1825 | static DEFINE_MUTEX(rcu_barrier_mutex); |
1826 | static struct completion rcu_barrier_completion; |
1827 | |
1828 | static void rcu_barrier_callback(struct rcu_head *notused) |
1829 | { |
1830 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) |
1831 | complete(&rcu_barrier_completion); |
1832 | } |
1833 | |
1834 | /* |
1835 | * Called with preemption disabled, and from cross-cpu IRQ context. |
1836 | */ |
1837 | static void rcu_barrier_func(void *type) |
1838 | { |
1839 | int cpu = smp_processor_id(); |
1840 | struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu); |
1841 | void (*call_rcu_func)(struct rcu_head *head, |
1842 | void (*func)(struct rcu_head *head)); |
1843 | |
1844 | atomic_inc(&rcu_barrier_cpu_count); |
1845 | call_rcu_func = type; |
1846 | call_rcu_func(head, rcu_barrier_callback); |
1847 | } |
1848 | |
1849 | /* |
1850 | * Orchestrate the specified type of RCU barrier, waiting for all |
1851 | * RCU callbacks of the specified type to complete. |
1852 | */ |
1853 | static void _rcu_barrier(struct rcu_state *rsp, |
1854 | void (*call_rcu_func)(struct rcu_head *head, |
1855 | void (*func)(struct rcu_head *head))) |
1856 | { |
1857 | BUG_ON(in_interrupt()); |
1858 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
1859 | mutex_lock(&rcu_barrier_mutex); |
1860 | init_completion(&rcu_barrier_completion); |
1861 | /* |
1862 | * Initialize rcu_barrier_cpu_count to 1, then invoke |
1863 | * rcu_barrier_func() on each CPU, so that each CPU also has |
1864 | * incremented rcu_barrier_cpu_count. Only then is it safe to |
1865 | * decrement rcu_barrier_cpu_count -- otherwise the first CPU |
1866 | * might complete its grace period before all of the other CPUs |
1867 | * did their increment, causing this function to return too |
1868 | * early. Note that on_each_cpu() disables irqs, which prevents |
1869 | * any CPUs from coming online or going offline until each online |
1870 | * CPU has queued its RCU-barrier callback. |
1871 | */ |
1872 | atomic_set(&rcu_barrier_cpu_count, 1); |
1873 | on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1); |
1874 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) |
1875 | complete(&rcu_barrier_completion); |
1876 | wait_for_completion(&rcu_barrier_completion); |
1877 | mutex_unlock(&rcu_barrier_mutex); |
1878 | } |
1879 | |
1880 | /** |
1881 | * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete. |
1882 | */ |
1883 | void rcu_barrier_bh(void) |
1884 | { |
1885 | _rcu_barrier(&rcu_bh_state, call_rcu_bh); |
1886 | } |
1887 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); |
1888 | |
1889 | /** |
1890 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. |
1891 | */ |
1892 | void rcu_barrier_sched(void) |
1893 | { |
1894 | _rcu_barrier(&rcu_sched_state, call_rcu_sched); |
1895 | } |
1896 | EXPORT_SYMBOL_GPL(rcu_barrier_sched); |
1897 | |
1898 | /* |
1899 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
1900 | */ |
1901 | static void __init |
1902 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) |
1903 | { |
1904 | unsigned long flags; |
1905 | int i; |
1906 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
1907 | struct rcu_node *rnp = rcu_get_root(rsp); |
1908 | |
1909 | /* Set up local state, ensuring consistent view of global state. */ |
1910 | raw_spin_lock_irqsave(&rnp->lock, flags); |
1911 | rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo); |
1912 | rdp->nxtlist = NULL; |
1913 | for (i = 0; i < RCU_NEXT_SIZE; i++) |
1914 | rdp->nxttail[i] = &rdp->nxtlist; |
1915 | rdp->qlen = 0; |
1916 | #ifdef CONFIG_NO_HZ |
1917 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); |
1918 | #endif /* #ifdef CONFIG_NO_HZ */ |
1919 | rdp->cpu = cpu; |
1920 | rdp->rsp = rsp; |
1921 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
1922 | } |
1923 | |
1924 | /* |
1925 | * Initialize a CPU's per-CPU RCU data. Note that only one online or |
1926 | * offline event can be happening at a given time. Note also that we |
1927 | * can accept some slop in the rsp->completed access due to the fact |
1928 | * that this CPU cannot possibly have any RCU callbacks in flight yet. |
1929 | */ |
1930 | static void __cpuinit |
1931 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptible) |
1932 | { |
1933 | unsigned long flags; |
1934 | unsigned long mask; |
1935 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
1936 | struct rcu_node *rnp = rcu_get_root(rsp); |
1937 | |
1938 | /* Set up local state, ensuring consistent view of global state. */ |
1939 | raw_spin_lock_irqsave(&rnp->lock, flags); |
1940 | rdp->beenonline = 1; /* We have now been online. */ |
1941 | rdp->preemptible = preemptible; |
1942 | rdp->qlen_last_fqs_check = 0; |
1943 | rdp->n_force_qs_snap = rsp->n_force_qs; |
1944 | rdp->blimit = blimit; |
1945 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
1946 | |
1947 | /* |
1948 | * A new grace period might start here. If so, we won't be part |
1949 | * of it, but that is OK, as we are currently in a quiescent state. |
1950 | */ |
1951 | |
1952 | /* Exclude any attempts to start a new GP on large systems. */ |
1953 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
1954 | |
1955 | /* Add CPU to rcu_node bitmasks. */ |
1956 | rnp = rdp->mynode; |
1957 | mask = rdp->grpmask; |
1958 | do { |
1959 | /* Exclude any attempts to start a new GP on small systems. */ |
1960 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
1961 | rnp->qsmaskinit |= mask; |
1962 | mask = rnp->grpmask; |
1963 | if (rnp == rdp->mynode) { |
1964 | /* |
1965 | * If there is a grace period in progress, we will |
1966 | * set up to wait for it next time we run the |
1967 | * RCU core code. |
1968 | */ |
1969 | rdp->gpnum = rnp->completed; |
1970 | rdp->completed = rnp->completed; |
1971 | rdp->passed_quiesce = 0; |
1972 | rdp->qs_pending = 0; |
1973 | rdp->passed_quiesce_gpnum = rnp->gpnum - 1; |
1974 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuonl"); |
1975 | } |
1976 | raw_spin_unlock(&rnp->lock); /* irqs already disabled. */ |
1977 | rnp = rnp->parent; |
1978 | } while (rnp != NULL && !(rnp->qsmaskinit & mask)); |
1979 | |
1980 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); |
1981 | } |
1982 | |
1983 | static void __cpuinit rcu_prepare_cpu(int cpu) |
1984 | { |
1985 | rcu_init_percpu_data(cpu, &rcu_sched_state, 0); |
1986 | rcu_init_percpu_data(cpu, &rcu_bh_state, 0); |
1987 | rcu_preempt_init_percpu_data(cpu); |
1988 | } |
1989 | |
1990 | /* |
1991 | * Handle CPU online/offline notification events. |
1992 | */ |
1993 | static int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
1994 | unsigned long action, void *hcpu) |
1995 | { |
1996 | long cpu = (long)hcpu; |
1997 | struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu); |
1998 | struct rcu_node *rnp = rdp->mynode; |
1999 | |
2000 | trace_rcu_utilization("Start CPU hotplug"); |
2001 | switch (action) { |
2002 | case CPU_UP_PREPARE: |
2003 | case CPU_UP_PREPARE_FROZEN: |
2004 | rcu_prepare_cpu(cpu); |
2005 | rcu_prepare_kthreads(cpu); |
2006 | break; |
2007 | case CPU_ONLINE: |
2008 | case CPU_DOWN_FAILED: |
2009 | rcu_node_kthread_setaffinity(rnp, -1); |
2010 | rcu_cpu_kthread_setrt(cpu, 1); |
2011 | break; |
2012 | case CPU_DOWN_PREPARE: |
2013 | rcu_node_kthread_setaffinity(rnp, cpu); |
2014 | rcu_cpu_kthread_setrt(cpu, 0); |
2015 | break; |
2016 | case CPU_DYING: |
2017 | case CPU_DYING_FROZEN: |
2018 | /* |
2019 | * The whole machine is "stopped" except this CPU, so we can |
2020 | * touch any data without introducing corruption. We send the |
2021 | * dying CPU's callbacks to an arbitrarily chosen online CPU. |
2022 | */ |
2023 | rcu_send_cbs_to_online(&rcu_bh_state); |
2024 | rcu_send_cbs_to_online(&rcu_sched_state); |
2025 | rcu_preempt_send_cbs_to_online(); |
2026 | break; |
2027 | case CPU_DEAD: |
2028 | case CPU_DEAD_FROZEN: |
2029 | case CPU_UP_CANCELED: |
2030 | case CPU_UP_CANCELED_FROZEN: |
2031 | rcu_offline_cpu(cpu); |
2032 | break; |
2033 | default: |
2034 | break; |
2035 | } |
2036 | trace_rcu_utilization("End CPU hotplug"); |
2037 | return NOTIFY_OK; |
2038 | } |
2039 | |
2040 | /* |
2041 | * This function is invoked towards the end of the scheduler's initialization |
2042 | * process. Before this is called, the idle task might contain |
2043 | * RCU read-side critical sections (during which time, this idle |
2044 | * task is booting the system). After this function is called, the |
2045 | * idle tasks are prohibited from containing RCU read-side critical |
2046 | * sections. This function also enables RCU lockdep checking. |
2047 | */ |
2048 | void rcu_scheduler_starting(void) |
2049 | { |
2050 | WARN_ON(num_online_cpus() != 1); |
2051 | WARN_ON(nr_context_switches() > 0); |
2052 | rcu_scheduler_active = 1; |
2053 | } |
2054 | |
2055 | /* |
2056 | * Compute the per-level fanout, either using the exact fanout specified |
2057 | * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT. |
2058 | */ |
2059 | #ifdef CONFIG_RCU_FANOUT_EXACT |
2060 | static void __init rcu_init_levelspread(struct rcu_state *rsp) |
2061 | { |
2062 | int i; |
2063 | |
2064 | for (i = NUM_RCU_LVLS - 1; i > 0; i--) |
2065 | rsp->levelspread[i] = CONFIG_RCU_FANOUT; |
2066 | rsp->levelspread[0] = RCU_FANOUT_LEAF; |
2067 | } |
2068 | #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */ |
2069 | static void __init rcu_init_levelspread(struct rcu_state *rsp) |
2070 | { |
2071 | int ccur; |
2072 | int cprv; |
2073 | int i; |
2074 | |
2075 | cprv = NR_CPUS; |
2076 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { |
2077 | ccur = rsp->levelcnt[i]; |
2078 | rsp->levelspread[i] = (cprv + ccur - 1) / ccur; |
2079 | cprv = ccur; |
2080 | } |
2081 | } |
2082 | #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */ |
2083 | |
2084 | /* |
2085 | * Helper function for rcu_init() that initializes one rcu_state structure. |
2086 | */ |
2087 | static void __init rcu_init_one(struct rcu_state *rsp, |
2088 | struct rcu_data __percpu *rda) |
2089 | { |
2090 | static char *buf[] = { "rcu_node_level_0", |
2091 | "rcu_node_level_1", |
2092 | "rcu_node_level_2", |
2093 | "rcu_node_level_3" }; /* Match MAX_RCU_LVLS */ |
2094 | int cpustride = 1; |
2095 | int i; |
2096 | int j; |
2097 | struct rcu_node *rnp; |
2098 | |
2099 | BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
2100 | |
2101 | /* Initialize the level-tracking arrays. */ |
2102 | |
2103 | for (i = 1; i < NUM_RCU_LVLS; i++) |
2104 | rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1]; |
2105 | rcu_init_levelspread(rsp); |
2106 | |
2107 | /* Initialize the elements themselves, starting from the leaves. */ |
2108 | |
2109 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { |
2110 | cpustride *= rsp->levelspread[i]; |
2111 | rnp = rsp->level[i]; |
2112 | for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { |
2113 | raw_spin_lock_init(&rnp->lock); |
2114 | lockdep_set_class_and_name(&rnp->lock, |
2115 | &rcu_node_class[i], buf[i]); |
2116 | rnp->gpnum = 0; |
2117 | rnp->qsmask = 0; |
2118 | rnp->qsmaskinit = 0; |
2119 | rnp->grplo = j * cpustride; |
2120 | rnp->grphi = (j + 1) * cpustride - 1; |
2121 | if (rnp->grphi >= NR_CPUS) |
2122 | rnp->grphi = NR_CPUS - 1; |
2123 | if (i == 0) { |
2124 | rnp->grpnum = 0; |
2125 | rnp->grpmask = 0; |
2126 | rnp->parent = NULL; |
2127 | } else { |
2128 | rnp->grpnum = j % rsp->levelspread[i - 1]; |
2129 | rnp->grpmask = 1UL << rnp->grpnum; |
2130 | rnp->parent = rsp->level[i - 1] + |
2131 | j / rsp->levelspread[i - 1]; |
2132 | } |
2133 | rnp->level = i; |
2134 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
2135 | } |
2136 | } |
2137 | |
2138 | rsp->rda = rda; |
2139 | rnp = rsp->level[NUM_RCU_LVLS - 1]; |
2140 | for_each_possible_cpu(i) { |
2141 | while (i > rnp->grphi) |
2142 | rnp++; |
2143 | per_cpu_ptr(rsp->rda, i)->mynode = rnp; |
2144 | rcu_boot_init_percpu_data(i, rsp); |
2145 | } |
2146 | } |
2147 | |
2148 | void __init rcu_init(void) |
2149 | { |
2150 | int cpu; |
2151 | |
2152 | rcu_bootup_announce(); |
2153 | rcu_init_one(&rcu_sched_state, &rcu_sched_data); |
2154 | rcu_init_one(&rcu_bh_state, &rcu_bh_data); |
2155 | __rcu_init_preempt(); |
2156 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
2157 | |
2158 | /* |
2159 | * We don't need protection against CPU-hotplug here because |
2160 | * this is called early in boot, before either interrupts |
2161 | * or the scheduler are operational. |
2162 | */ |
2163 | cpu_notifier(rcu_cpu_notify, 0); |
2164 | for_each_online_cpu(cpu) |
2165 | rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu); |
2166 | check_cpu_stall_init(); |
2167 | } |
2168 | |
2169 | #include "rcutree_plugin.h" |
2170 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9