Commit | Line | Data |
---|---|---|
64db4cff PM |
1 | /* |
2 | * Read-Copy Update mechanism for mutual exclusion | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License as published by | |
6 | * the Free Software Foundation; either version 2 of the License, or | |
7 | * (at your option) any later version. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; if not, write to the Free Software | |
16 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
17 | * | |
18 | * Copyright IBM Corporation, 2008 | |
19 | * | |
20 | * Authors: Dipankar Sarma <dipankar@in.ibm.com> | |
21 | * Manfred Spraul <manfred@colorfullife.com> | |
22 | * Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical version | |
23 | * | |
24 | * Based on the original work by Paul McKenney <paulmck@us.ibm.com> | |
25 | * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen. | |
26 | * | |
27 | * For detailed explanation of Read-Copy Update mechanism see - | |
a71fca58 | 28 | * Documentation/RCU |
64db4cff PM |
29 | */ |
30 | #include <linux/types.h> | |
31 | #include <linux/kernel.h> | |
32 | #include <linux/init.h> | |
33 | #include <linux/spinlock.h> | |
34 | #include <linux/smp.h> | |
35 | #include <linux/rcupdate.h> | |
36 | #include <linux/interrupt.h> | |
37 | #include <linux/sched.h> | |
c1dc0b9c | 38 | #include <linux/nmi.h> |
8826f3b0 | 39 | #include <linux/atomic.h> |
64db4cff PM |
40 | #include <linux/bitops.h> |
41 | #include <linux/module.h> | |
42 | #include <linux/completion.h> | |
43 | #include <linux/moduleparam.h> | |
44 | #include <linux/percpu.h> | |
45 | #include <linux/notifier.h> | |
46 | #include <linux/cpu.h> | |
47 | #include <linux/mutex.h> | |
48 | #include <linux/time.h> | |
bbad9379 | 49 | #include <linux/kernel_stat.h> |
a26ac245 PM |
50 | #include <linux/wait.h> |
51 | #include <linux/kthread.h> | |
268bb0ce | 52 | #include <linux/prefetch.h> |
64db4cff | 53 | |
9f77da9f | 54 | #include "rcutree.h" |
29c00b4a PM |
55 | #include <trace/events/rcu.h> |
56 | ||
57 | #include "rcu.h" | |
9f77da9f | 58 | |
64db4cff PM |
59 | /* Data structures. */ |
60 | ||
b668c9cf | 61 | static struct lock_class_key rcu_node_class[NUM_RCU_LVLS]; |
88b91c7c | 62 | |
4300aa64 | 63 | #define RCU_STATE_INITIALIZER(structname) { \ |
e99033c5 | 64 | .level = { &structname##_state.node[0] }, \ |
64db4cff PM |
65 | .levelcnt = { \ |
66 | NUM_RCU_LVL_0, /* root of hierarchy. */ \ | |
67 | NUM_RCU_LVL_1, \ | |
68 | NUM_RCU_LVL_2, \ | |
cf244dc0 PM |
69 | NUM_RCU_LVL_3, \ |
70 | NUM_RCU_LVL_4, /* == MAX_RCU_LVLS */ \ | |
64db4cff | 71 | }, \ |
83f5b01f | 72 | .signaled = RCU_GP_IDLE, \ |
64db4cff PM |
73 | .gpnum = -300, \ |
74 | .completed = -300, \ | |
e99033c5 PM |
75 | .onofflock = __RAW_SPIN_LOCK_UNLOCKED(&structname##_state.onofflock), \ |
76 | .fqslock = __RAW_SPIN_LOCK_UNLOCKED(&structname##_state.fqslock), \ | |
64db4cff PM |
77 | .n_force_qs = 0, \ |
78 | .n_force_qs_ngp = 0, \ | |
4300aa64 | 79 | .name = #structname, \ |
64db4cff PM |
80 | } |
81 | ||
e99033c5 | 82 | struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched); |
d6714c22 | 83 | DEFINE_PER_CPU(struct rcu_data, rcu_sched_data); |
64db4cff | 84 | |
e99033c5 | 85 | struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh); |
6258c4fb | 86 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); |
b1f77b05 | 87 | |
27f4d280 PM |
88 | static struct rcu_state *rcu_state; |
89 | ||
b0d30417 PM |
90 | /* |
91 | * The rcu_scheduler_active variable transitions from zero to one just | |
92 | * before the first task is spawned. So when this variable is zero, RCU | |
93 | * can assume that there is but one task, allowing RCU to (for example) | |
94 | * optimized synchronize_sched() to a simple barrier(). When this variable | |
95 | * is one, RCU must actually do all the hard work required to detect real | |
96 | * grace periods. This variable is also used to suppress boot-time false | |
97 | * positives from lockdep-RCU error checking. | |
98 | */ | |
bbad9379 PM |
99 | int rcu_scheduler_active __read_mostly; |
100 | EXPORT_SYMBOL_GPL(rcu_scheduler_active); | |
101 | ||
b0d30417 PM |
102 | /* |
103 | * The rcu_scheduler_fully_active variable transitions from zero to one | |
104 | * during the early_initcall() processing, which is after the scheduler | |
105 | * is capable of creating new tasks. So RCU processing (for example, | |
106 | * creating tasks for RCU priority boosting) must be delayed until after | |
107 | * rcu_scheduler_fully_active transitions from zero to one. We also | |
108 | * currently delay invocation of any RCU callbacks until after this point. | |
109 | * | |
110 | * It might later prove better for people registering RCU callbacks during | |
111 | * early boot to take responsibility for these callbacks, but one step at | |
112 | * a time. | |
113 | */ | |
114 | static int rcu_scheduler_fully_active __read_mostly; | |
115 | ||
a46e0899 PM |
116 | #ifdef CONFIG_RCU_BOOST |
117 | ||
a26ac245 PM |
118 | /* |
119 | * Control variables for per-CPU and per-rcu_node kthreads. These | |
120 | * handle all flavors of RCU. | |
121 | */ | |
122 | static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task); | |
d71df90e | 123 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status); |
15ba0ba8 | 124 | DEFINE_PER_CPU(int, rcu_cpu_kthread_cpu); |
5ece5bab | 125 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops); |
d71df90e | 126 | DEFINE_PER_CPU(char, rcu_cpu_has_work); |
a26ac245 | 127 | |
a46e0899 PM |
128 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
129 | ||
0f962a5e | 130 | static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); |
a46e0899 PM |
131 | static void invoke_rcu_core(void); |
132 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp); | |
a26ac245 PM |
133 | |
134 | #define RCU_KTHREAD_PRIO 1 /* RT priority for per-CPU kthreads. */ | |
135 | ||
4a298656 PM |
136 | /* |
137 | * Track the rcutorture test sequence number and the update version | |
138 | * number within a given test. The rcutorture_testseq is incremented | |
139 | * on every rcutorture module load and unload, so has an odd value | |
140 | * when a test is running. The rcutorture_vernum is set to zero | |
141 | * when rcutorture starts and is incremented on each rcutorture update. | |
142 | * These variables enable correlating rcutorture output with the | |
143 | * RCU tracing information. | |
144 | */ | |
145 | unsigned long rcutorture_testseq; | |
146 | unsigned long rcutorture_vernum; | |
147 | ||
fc2219d4 PM |
148 | /* |
149 | * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s | |
150 | * permit this function to be invoked without holding the root rcu_node | |
151 | * structure's ->lock, but of course results can be subject to change. | |
152 | */ | |
153 | static int rcu_gp_in_progress(struct rcu_state *rsp) | |
154 | { | |
155 | return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum); | |
156 | } | |
157 | ||
b1f77b05 | 158 | /* |
d6714c22 | 159 | * Note a quiescent state. Because we do not need to know |
b1f77b05 | 160 | * how many quiescent states passed, just if there was at least |
d6714c22 | 161 | * one since the start of the grace period, this just sets a flag. |
e4cc1f22 | 162 | * The caller must have disabled preemption. |
b1f77b05 | 163 | */ |
d6714c22 | 164 | void rcu_sched_qs(int cpu) |
b1f77b05 | 165 | { |
25502a6c | 166 | struct rcu_data *rdp = &per_cpu(rcu_sched_data, cpu); |
f41d911f | 167 | |
e4cc1f22 | 168 | rdp->passed_quiesce_gpnum = rdp->gpnum; |
c3422bea | 169 | barrier(); |
e4cc1f22 | 170 | if (rdp->passed_quiesce == 0) |
d4c08f2a | 171 | trace_rcu_grace_period("rcu_sched", rdp->gpnum, "cpuqs"); |
e4cc1f22 | 172 | rdp->passed_quiesce = 1; |
b1f77b05 IM |
173 | } |
174 | ||
d6714c22 | 175 | void rcu_bh_qs(int cpu) |
b1f77b05 | 176 | { |
25502a6c | 177 | struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu); |
f41d911f | 178 | |
e4cc1f22 | 179 | rdp->passed_quiesce_gpnum = rdp->gpnum; |
c3422bea | 180 | barrier(); |
e4cc1f22 | 181 | if (rdp->passed_quiesce == 0) |
d4c08f2a | 182 | trace_rcu_grace_period("rcu_bh", rdp->gpnum, "cpuqs"); |
e4cc1f22 | 183 | rdp->passed_quiesce = 1; |
b1f77b05 | 184 | } |
64db4cff | 185 | |
25502a6c PM |
186 | /* |
187 | * Note a context switch. This is a quiescent state for RCU-sched, | |
188 | * and requires special handling for preemptible RCU. | |
e4cc1f22 | 189 | * The caller must have disabled preemption. |
25502a6c PM |
190 | */ |
191 | void rcu_note_context_switch(int cpu) | |
192 | { | |
300df91c | 193 | trace_rcu_utilization("Start context switch"); |
25502a6c PM |
194 | rcu_sched_qs(cpu); |
195 | rcu_preempt_note_context_switch(cpu); | |
300df91c | 196 | trace_rcu_utilization("End context switch"); |
25502a6c | 197 | } |
29ce8310 | 198 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
25502a6c | 199 | |
64db4cff | 200 | #ifdef CONFIG_NO_HZ |
90a4d2c0 PM |
201 | DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { |
202 | .dynticks_nesting = 1, | |
23b5c8fa | 203 | .dynticks = ATOMIC_INIT(1), |
90a4d2c0 | 204 | }; |
64db4cff PM |
205 | #endif /* #ifdef CONFIG_NO_HZ */ |
206 | ||
e0f23060 | 207 | static int blimit = 10; /* Maximum callbacks per rcu_do_batch. */ |
64db4cff PM |
208 | static int qhimark = 10000; /* If this many pending, ignore blimit. */ |
209 | static int qlowmark = 100; /* Once only this many pending, use blimit. */ | |
210 | ||
3d76c082 PM |
211 | module_param(blimit, int, 0); |
212 | module_param(qhimark, int, 0); | |
213 | module_param(qlowmark, int, 0); | |
214 | ||
a00e0d71 | 215 | int rcu_cpu_stall_suppress __read_mostly; |
f2e0dd70 | 216 | module_param(rcu_cpu_stall_suppress, int, 0644); |
742734ee | 217 | |
64db4cff | 218 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed); |
a157229c | 219 | static int rcu_pending(int cpu); |
64db4cff PM |
220 | |
221 | /* | |
d6714c22 | 222 | * Return the number of RCU-sched batches processed thus far for debug & stats. |
64db4cff | 223 | */ |
d6714c22 | 224 | long rcu_batches_completed_sched(void) |
64db4cff | 225 | { |
d6714c22 | 226 | return rcu_sched_state.completed; |
64db4cff | 227 | } |
d6714c22 | 228 | EXPORT_SYMBOL_GPL(rcu_batches_completed_sched); |
64db4cff PM |
229 | |
230 | /* | |
231 | * Return the number of RCU BH batches processed thus far for debug & stats. | |
232 | */ | |
233 | long rcu_batches_completed_bh(void) | |
234 | { | |
235 | return rcu_bh_state.completed; | |
236 | } | |
237 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); | |
238 | ||
bf66f18e PM |
239 | /* |
240 | * Force a quiescent state for RCU BH. | |
241 | */ | |
242 | void rcu_bh_force_quiescent_state(void) | |
243 | { | |
244 | force_quiescent_state(&rcu_bh_state, 0); | |
245 | } | |
246 | EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state); | |
247 | ||
4a298656 PM |
248 | /* |
249 | * Record the number of times rcutorture tests have been initiated and | |
250 | * terminated. This information allows the debugfs tracing stats to be | |
251 | * correlated to the rcutorture messages, even when the rcutorture module | |
252 | * is being repeatedly loaded and unloaded. In other words, we cannot | |
253 | * store this state in rcutorture itself. | |
254 | */ | |
255 | void rcutorture_record_test_transition(void) | |
256 | { | |
257 | rcutorture_testseq++; | |
258 | rcutorture_vernum = 0; | |
259 | } | |
260 | EXPORT_SYMBOL_GPL(rcutorture_record_test_transition); | |
261 | ||
262 | /* | |
263 | * Record the number of writer passes through the current rcutorture test. | |
264 | * This is also used to correlate debugfs tracing stats with the rcutorture | |
265 | * messages. | |
266 | */ | |
267 | void rcutorture_record_progress(unsigned long vernum) | |
268 | { | |
269 | rcutorture_vernum++; | |
270 | } | |
271 | EXPORT_SYMBOL_GPL(rcutorture_record_progress); | |
272 | ||
bf66f18e PM |
273 | /* |
274 | * Force a quiescent state for RCU-sched. | |
275 | */ | |
276 | void rcu_sched_force_quiescent_state(void) | |
277 | { | |
278 | force_quiescent_state(&rcu_sched_state, 0); | |
279 | } | |
280 | EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state); | |
281 | ||
64db4cff PM |
282 | /* |
283 | * Does the CPU have callbacks ready to be invoked? | |
284 | */ | |
285 | static int | |
286 | cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) | |
287 | { | |
288 | return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]; | |
289 | } | |
290 | ||
291 | /* | |
292 | * Does the current CPU require a yet-as-unscheduled grace period? | |
293 | */ | |
294 | static int | |
295 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) | |
296 | { | |
fc2219d4 | 297 | return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp); |
64db4cff PM |
298 | } |
299 | ||
300 | /* | |
301 | * Return the root node of the specified rcu_state structure. | |
302 | */ | |
303 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) | |
304 | { | |
305 | return &rsp->node[0]; | |
306 | } | |
307 | ||
308 | #ifdef CONFIG_SMP | |
309 | ||
310 | /* | |
311 | * If the specified CPU is offline, tell the caller that it is in | |
312 | * a quiescent state. Otherwise, whack it with a reschedule IPI. | |
313 | * Grace periods can end up waiting on an offline CPU when that | |
314 | * CPU is in the process of coming online -- it will be added to the | |
315 | * rcu_node bitmasks before it actually makes it online. The same thing | |
316 | * can happen while a CPU is in the process of coming online. Because this | |
317 | * race is quite rare, we check for it after detecting that the grace | |
318 | * period has been delayed rather than checking each and every CPU | |
319 | * each and every time we start a new grace period. | |
320 | */ | |
321 | static int rcu_implicit_offline_qs(struct rcu_data *rdp) | |
322 | { | |
323 | /* | |
324 | * If the CPU is offline, it is in a quiescent state. We can | |
325 | * trust its state not to change because interrupts are disabled. | |
326 | */ | |
327 | if (cpu_is_offline(rdp->cpu)) { | |
d4c08f2a | 328 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, "ofl"); |
64db4cff PM |
329 | rdp->offline_fqs++; |
330 | return 1; | |
331 | } | |
332 | ||
6cc68793 PM |
333 | /* If preemptible RCU, no point in sending reschedule IPI. */ |
334 | if (rdp->preemptible) | |
f41d911f PM |
335 | return 0; |
336 | ||
64db4cff PM |
337 | /* The CPU is online, so send it a reschedule IPI. */ |
338 | if (rdp->cpu != smp_processor_id()) | |
339 | smp_send_reschedule(rdp->cpu); | |
340 | else | |
341 | set_need_resched(); | |
342 | rdp->resched_ipi++; | |
343 | return 0; | |
344 | } | |
345 | ||
346 | #endif /* #ifdef CONFIG_SMP */ | |
347 | ||
348 | #ifdef CONFIG_NO_HZ | |
64db4cff PM |
349 | |
350 | /** | |
351 | * rcu_enter_nohz - inform RCU that current CPU is entering nohz | |
352 | * | |
353 | * Enter nohz mode, in other words, -leave- the mode in which RCU | |
354 | * read-side critical sections can occur. (Though RCU read-side | |
355 | * critical sections can occur in irq handlers in nohz mode, a possibility | |
356 | * handled by rcu_irq_enter() and rcu_irq_exit()). | |
357 | */ | |
358 | void rcu_enter_nohz(void) | |
359 | { | |
360 | unsigned long flags; | |
361 | struct rcu_dynticks *rdtp; | |
362 | ||
64db4cff PM |
363 | local_irq_save(flags); |
364 | rdtp = &__get_cpu_var(rcu_dynticks); | |
23b5c8fa PM |
365 | if (--rdtp->dynticks_nesting) { |
366 | local_irq_restore(flags); | |
367 | return; | |
368 | } | |
d4c08f2a | 369 | trace_rcu_dyntick("Start"); |
23b5c8fa PM |
370 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ |
371 | smp_mb__before_atomic_inc(); /* See above. */ | |
372 | atomic_inc(&rdtp->dynticks); | |
373 | smp_mb__after_atomic_inc(); /* Force ordering with next sojourn. */ | |
374 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); | |
64db4cff PM |
375 | local_irq_restore(flags); |
376 | } | |
377 | ||
378 | /* | |
379 | * rcu_exit_nohz - inform RCU that current CPU is leaving nohz | |
380 | * | |
381 | * Exit nohz mode, in other words, -enter- the mode in which RCU | |
382 | * read-side critical sections normally occur. | |
383 | */ | |
384 | void rcu_exit_nohz(void) | |
385 | { | |
386 | unsigned long flags; | |
387 | struct rcu_dynticks *rdtp; | |
388 | ||
389 | local_irq_save(flags); | |
390 | rdtp = &__get_cpu_var(rcu_dynticks); | |
23b5c8fa PM |
391 | if (rdtp->dynticks_nesting++) { |
392 | local_irq_restore(flags); | |
393 | return; | |
394 | } | |
395 | smp_mb__before_atomic_inc(); /* Force ordering w/previous sojourn. */ | |
396 | atomic_inc(&rdtp->dynticks); | |
397 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ | |
398 | smp_mb__after_atomic_inc(); /* See above. */ | |
399 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); | |
d4c08f2a | 400 | trace_rcu_dyntick("End"); |
64db4cff | 401 | local_irq_restore(flags); |
64db4cff PM |
402 | } |
403 | ||
404 | /** | |
405 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
406 | * | |
407 | * If the CPU was idle with dynamic ticks active, and there is no | |
408 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
409 | * RCU grace-period handling know that the CPU is active. | |
410 | */ | |
411 | void rcu_nmi_enter(void) | |
412 | { | |
413 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
414 | ||
23b5c8fa PM |
415 | if (rdtp->dynticks_nmi_nesting == 0 && |
416 | (atomic_read(&rdtp->dynticks) & 0x1)) | |
64db4cff | 417 | return; |
23b5c8fa PM |
418 | rdtp->dynticks_nmi_nesting++; |
419 | smp_mb__before_atomic_inc(); /* Force delay from prior write. */ | |
420 | atomic_inc(&rdtp->dynticks); | |
421 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ | |
422 | smp_mb__after_atomic_inc(); /* See above. */ | |
423 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); | |
64db4cff PM |
424 | } |
425 | ||
426 | /** | |
427 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
428 | * | |
429 | * If the CPU was idle with dynamic ticks active, and there is no | |
430 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
431 | * RCU grace-period handling know that the CPU is no longer active. | |
432 | */ | |
433 | void rcu_nmi_exit(void) | |
434 | { | |
435 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
436 | ||
23b5c8fa PM |
437 | if (rdtp->dynticks_nmi_nesting == 0 || |
438 | --rdtp->dynticks_nmi_nesting != 0) | |
64db4cff | 439 | return; |
23b5c8fa PM |
440 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ |
441 | smp_mb__before_atomic_inc(); /* See above. */ | |
442 | atomic_inc(&rdtp->dynticks); | |
443 | smp_mb__after_atomic_inc(); /* Force delay to next write. */ | |
444 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); | |
64db4cff PM |
445 | } |
446 | ||
447 | /** | |
448 | * rcu_irq_enter - inform RCU of entry to hard irq context | |
449 | * | |
450 | * If the CPU was idle with dynamic ticks active, this updates the | |
451 | * rdtp->dynticks to let the RCU handling know that the CPU is active. | |
452 | */ | |
453 | void rcu_irq_enter(void) | |
454 | { | |
23b5c8fa | 455 | rcu_exit_nohz(); |
64db4cff PM |
456 | } |
457 | ||
458 | /** | |
459 | * rcu_irq_exit - inform RCU of exit from hard irq context | |
460 | * | |
461 | * If the CPU was idle with dynamic ticks active, update the rdp->dynticks | |
462 | * to put let the RCU handling be aware that the CPU is going back to idle | |
463 | * with no ticks. | |
464 | */ | |
465 | void rcu_irq_exit(void) | |
466 | { | |
23b5c8fa | 467 | rcu_enter_nohz(); |
64db4cff PM |
468 | } |
469 | ||
64db4cff PM |
470 | #ifdef CONFIG_SMP |
471 | ||
64db4cff PM |
472 | /* |
473 | * Snapshot the specified CPU's dynticks counter so that we can later | |
474 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 475 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff PM |
476 | */ |
477 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | |
478 | { | |
23b5c8fa PM |
479 | rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks); |
480 | return 0; | |
64db4cff PM |
481 | } |
482 | ||
483 | /* | |
484 | * Return true if the specified CPU has passed through a quiescent | |
485 | * state by virtue of being in or having passed through an dynticks | |
486 | * idle state since the last call to dyntick_save_progress_counter() | |
487 | * for this same CPU. | |
488 | */ | |
489 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
490 | { | |
23b5c8fa PM |
491 | unsigned long curr; |
492 | unsigned long snap; | |
64db4cff | 493 | |
23b5c8fa PM |
494 | curr = (unsigned long)atomic_add_return(0, &rdp->dynticks->dynticks); |
495 | snap = (unsigned long)rdp->dynticks_snap; | |
64db4cff PM |
496 | |
497 | /* | |
498 | * If the CPU passed through or entered a dynticks idle phase with | |
499 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
500 | * already acknowledged the request to pass through a quiescent | |
501 | * state. Either way, that CPU cannot possibly be in an RCU | |
502 | * read-side critical section that started before the beginning | |
503 | * of the current RCU grace period. | |
504 | */ | |
23b5c8fa | 505 | if ((curr & 0x1) == 0 || ULONG_CMP_GE(curr, snap + 2)) { |
d4c08f2a | 506 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, "dti"); |
64db4cff PM |
507 | rdp->dynticks_fqs++; |
508 | return 1; | |
509 | } | |
510 | ||
511 | /* Go check for the CPU being offline. */ | |
512 | return rcu_implicit_offline_qs(rdp); | |
513 | } | |
514 | ||
515 | #endif /* #ifdef CONFIG_SMP */ | |
516 | ||
517 | #else /* #ifdef CONFIG_NO_HZ */ | |
518 | ||
64db4cff PM |
519 | #ifdef CONFIG_SMP |
520 | ||
64db4cff PM |
521 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
522 | { | |
523 | return 0; | |
524 | } | |
525 | ||
526 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
527 | { | |
528 | return rcu_implicit_offline_qs(rdp); | |
529 | } | |
530 | ||
531 | #endif /* #ifdef CONFIG_SMP */ | |
532 | ||
533 | #endif /* #else #ifdef CONFIG_NO_HZ */ | |
534 | ||
742734ee | 535 | int rcu_cpu_stall_suppress __read_mostly; |
c68de209 | 536 | |
64db4cff PM |
537 | static void record_gp_stall_check_time(struct rcu_state *rsp) |
538 | { | |
539 | rsp->gp_start = jiffies; | |
540 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK; | |
541 | } | |
542 | ||
543 | static void print_other_cpu_stall(struct rcu_state *rsp) | |
544 | { | |
545 | int cpu; | |
546 | long delta; | |
547 | unsigned long flags; | |
548 | struct rcu_node *rnp = rcu_get_root(rsp); | |
64db4cff PM |
549 | |
550 | /* Only let one CPU complain about others per time interval. */ | |
551 | ||
1304afb2 | 552 | raw_spin_lock_irqsave(&rnp->lock, flags); |
64db4cff | 553 | delta = jiffies - rsp->jiffies_stall; |
fc2219d4 | 554 | if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) { |
1304afb2 | 555 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
556 | return; |
557 | } | |
558 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
a0b6c9a7 PM |
559 | |
560 | /* | |
561 | * Now rat on any tasks that got kicked up to the root rcu_node | |
562 | * due to CPU offlining. | |
563 | */ | |
564 | rcu_print_task_stall(rnp); | |
1304afb2 | 565 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 566 | |
8cdd32a9 PM |
567 | /* |
568 | * OK, time to rat on our buddy... | |
569 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
570 | * RCU CPU stall warnings. | |
571 | */ | |
4300aa64 PM |
572 | printk(KERN_ERR "INFO: %s detected stalls on CPUs/tasks: {", |
573 | rsp->name); | |
a0b6c9a7 | 574 | rcu_for_each_leaf_node(rsp, rnp) { |
3acd9eb3 | 575 | raw_spin_lock_irqsave(&rnp->lock, flags); |
f41d911f | 576 | rcu_print_task_stall(rnp); |
3acd9eb3 | 577 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
a0b6c9a7 | 578 | if (rnp->qsmask == 0) |
64db4cff | 579 | continue; |
a0b6c9a7 PM |
580 | for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++) |
581 | if (rnp->qsmask & (1UL << cpu)) | |
582 | printk(" %d", rnp->grplo + cpu); | |
64db4cff | 583 | } |
4300aa64 | 584 | printk("} (detected by %d, t=%ld jiffies)\n", |
64db4cff | 585 | smp_processor_id(), (long)(jiffies - rsp->gp_start)); |
c1dc0b9c IM |
586 | trigger_all_cpu_backtrace(); |
587 | ||
1ed509a2 PM |
588 | /* If so configured, complain about tasks blocking the grace period. */ |
589 | ||
590 | rcu_print_detail_task_stall(rsp); | |
591 | ||
64db4cff PM |
592 | force_quiescent_state(rsp, 0); /* Kick them all. */ |
593 | } | |
594 | ||
595 | static void print_cpu_stall(struct rcu_state *rsp) | |
596 | { | |
597 | unsigned long flags; | |
598 | struct rcu_node *rnp = rcu_get_root(rsp); | |
599 | ||
8cdd32a9 PM |
600 | /* |
601 | * OK, time to rat on ourselves... | |
602 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
603 | * RCU CPU stall warnings. | |
604 | */ | |
4300aa64 PM |
605 | printk(KERN_ERR "INFO: %s detected stall on CPU %d (t=%lu jiffies)\n", |
606 | rsp->name, smp_processor_id(), jiffies - rsp->gp_start); | |
c1dc0b9c IM |
607 | trigger_all_cpu_backtrace(); |
608 | ||
1304afb2 | 609 | raw_spin_lock_irqsave(&rnp->lock, flags); |
20133cfc | 610 | if (ULONG_CMP_GE(jiffies, rsp->jiffies_stall)) |
64db4cff PM |
611 | rsp->jiffies_stall = |
612 | jiffies + RCU_SECONDS_TILL_STALL_RECHECK; | |
1304afb2 | 613 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
c1dc0b9c | 614 | |
64db4cff PM |
615 | set_need_resched(); /* kick ourselves to get things going. */ |
616 | } | |
617 | ||
618 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
619 | { | |
bad6e139 PM |
620 | unsigned long j; |
621 | unsigned long js; | |
64db4cff PM |
622 | struct rcu_node *rnp; |
623 | ||
742734ee | 624 | if (rcu_cpu_stall_suppress) |
c68de209 | 625 | return; |
bad6e139 PM |
626 | j = ACCESS_ONCE(jiffies); |
627 | js = ACCESS_ONCE(rsp->jiffies_stall); | |
64db4cff | 628 | rnp = rdp->mynode; |
bad6e139 | 629 | if ((ACCESS_ONCE(rnp->qsmask) & rdp->grpmask) && ULONG_CMP_GE(j, js)) { |
64db4cff PM |
630 | |
631 | /* We haven't checked in, so go dump stack. */ | |
632 | print_cpu_stall(rsp); | |
633 | ||
bad6e139 PM |
634 | } else if (rcu_gp_in_progress(rsp) && |
635 | ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) { | |
64db4cff | 636 | |
bad6e139 | 637 | /* They had a few time units to dump stack, so complain. */ |
64db4cff PM |
638 | print_other_cpu_stall(rsp); |
639 | } | |
640 | } | |
641 | ||
c68de209 PM |
642 | static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr) |
643 | { | |
742734ee | 644 | rcu_cpu_stall_suppress = 1; |
c68de209 PM |
645 | return NOTIFY_DONE; |
646 | } | |
647 | ||
53d84e00 PM |
648 | /** |
649 | * rcu_cpu_stall_reset - prevent further stall warnings in current grace period | |
650 | * | |
651 | * Set the stall-warning timeout way off into the future, thus preventing | |
652 | * any RCU CPU stall-warning messages from appearing in the current set of | |
653 | * RCU grace periods. | |
654 | * | |
655 | * The caller must disable hard irqs. | |
656 | */ | |
657 | void rcu_cpu_stall_reset(void) | |
658 | { | |
659 | rcu_sched_state.jiffies_stall = jiffies + ULONG_MAX / 2; | |
660 | rcu_bh_state.jiffies_stall = jiffies + ULONG_MAX / 2; | |
661 | rcu_preempt_stall_reset(); | |
662 | } | |
663 | ||
c68de209 PM |
664 | static struct notifier_block rcu_panic_block = { |
665 | .notifier_call = rcu_panic, | |
666 | }; | |
667 | ||
668 | static void __init check_cpu_stall_init(void) | |
669 | { | |
670 | atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block); | |
671 | } | |
672 | ||
64db4cff PM |
673 | /* |
674 | * Update CPU-local rcu_data state to record the newly noticed grace period. | |
675 | * This is used both when we started the grace period and when we notice | |
9160306e PM |
676 | * that someone else started the grace period. The caller must hold the |
677 | * ->lock of the leaf rcu_node structure corresponding to the current CPU, | |
678 | * and must have irqs disabled. | |
64db4cff | 679 | */ |
9160306e PM |
680 | static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) |
681 | { | |
682 | if (rdp->gpnum != rnp->gpnum) { | |
121dfc4b PM |
683 | /* |
684 | * If the current grace period is waiting for this CPU, | |
685 | * set up to detect a quiescent state, otherwise don't | |
686 | * go looking for one. | |
687 | */ | |
9160306e | 688 | rdp->gpnum = rnp->gpnum; |
d4c08f2a | 689 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpustart"); |
121dfc4b PM |
690 | if (rnp->qsmask & rdp->grpmask) { |
691 | rdp->qs_pending = 1; | |
e4cc1f22 | 692 | rdp->passed_quiesce = 0; |
121dfc4b PM |
693 | } else |
694 | rdp->qs_pending = 0; | |
9160306e PM |
695 | } |
696 | } | |
697 | ||
64db4cff PM |
698 | static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp) |
699 | { | |
9160306e PM |
700 | unsigned long flags; |
701 | struct rcu_node *rnp; | |
702 | ||
703 | local_irq_save(flags); | |
704 | rnp = rdp->mynode; | |
705 | if (rdp->gpnum == ACCESS_ONCE(rnp->gpnum) || /* outside lock. */ | |
1304afb2 | 706 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
9160306e PM |
707 | local_irq_restore(flags); |
708 | return; | |
709 | } | |
710 | __note_new_gpnum(rsp, rnp, rdp); | |
1304afb2 | 711 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
712 | } |
713 | ||
714 | /* | |
715 | * Did someone else start a new RCU grace period start since we last | |
716 | * checked? Update local state appropriately if so. Must be called | |
717 | * on the CPU corresponding to rdp. | |
718 | */ | |
719 | static int | |
720 | check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp) | |
721 | { | |
722 | unsigned long flags; | |
723 | int ret = 0; | |
724 | ||
725 | local_irq_save(flags); | |
726 | if (rdp->gpnum != rsp->gpnum) { | |
727 | note_new_gpnum(rsp, rdp); | |
728 | ret = 1; | |
729 | } | |
730 | local_irq_restore(flags); | |
731 | return ret; | |
732 | } | |
733 | ||
d09b62df PM |
734 | /* |
735 | * Advance this CPU's callbacks, but only if the current grace period | |
736 | * has ended. This may be called only from the CPU to whom the rdp | |
737 | * belongs. In addition, the corresponding leaf rcu_node structure's | |
738 | * ->lock must be held by the caller, with irqs disabled. | |
739 | */ | |
740 | static void | |
741 | __rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
742 | { | |
743 | /* Did another grace period end? */ | |
744 | if (rdp->completed != rnp->completed) { | |
745 | ||
746 | /* Advance callbacks. No harm if list empty. */ | |
747 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL]; | |
748 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL]; | |
749 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
750 | ||
751 | /* Remember that we saw this grace-period completion. */ | |
752 | rdp->completed = rnp->completed; | |
d4c08f2a | 753 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuend"); |
20377f32 | 754 | |
5ff8e6f0 FW |
755 | /* |
756 | * If we were in an extended quiescent state, we may have | |
121dfc4b | 757 | * missed some grace periods that others CPUs handled on |
5ff8e6f0 | 758 | * our behalf. Catch up with this state to avoid noting |
121dfc4b PM |
759 | * spurious new grace periods. If another grace period |
760 | * has started, then rnp->gpnum will have advanced, so | |
761 | * we will detect this later on. | |
5ff8e6f0 | 762 | */ |
121dfc4b | 763 | if (ULONG_CMP_LT(rdp->gpnum, rdp->completed)) |
5ff8e6f0 FW |
764 | rdp->gpnum = rdp->completed; |
765 | ||
20377f32 | 766 | /* |
121dfc4b PM |
767 | * If RCU does not need a quiescent state from this CPU, |
768 | * then make sure that this CPU doesn't go looking for one. | |
20377f32 | 769 | */ |
121dfc4b | 770 | if ((rnp->qsmask & rdp->grpmask) == 0) |
20377f32 | 771 | rdp->qs_pending = 0; |
d09b62df PM |
772 | } |
773 | } | |
774 | ||
775 | /* | |
776 | * Advance this CPU's callbacks, but only if the current grace period | |
777 | * has ended. This may be called only from the CPU to whom the rdp | |
778 | * belongs. | |
779 | */ | |
780 | static void | |
781 | rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp) | |
782 | { | |
783 | unsigned long flags; | |
784 | struct rcu_node *rnp; | |
785 | ||
786 | local_irq_save(flags); | |
787 | rnp = rdp->mynode; | |
788 | if (rdp->completed == ACCESS_ONCE(rnp->completed) || /* outside lock. */ | |
1304afb2 | 789 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
d09b62df PM |
790 | local_irq_restore(flags); |
791 | return; | |
792 | } | |
793 | __rcu_process_gp_end(rsp, rnp, rdp); | |
1304afb2 | 794 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
d09b62df PM |
795 | } |
796 | ||
797 | /* | |
798 | * Do per-CPU grace-period initialization for running CPU. The caller | |
799 | * must hold the lock of the leaf rcu_node structure corresponding to | |
800 | * this CPU. | |
801 | */ | |
802 | static void | |
803 | rcu_start_gp_per_cpu(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
804 | { | |
805 | /* Prior grace period ended, so advance callbacks for current CPU. */ | |
806 | __rcu_process_gp_end(rsp, rnp, rdp); | |
807 | ||
808 | /* | |
809 | * Because this CPU just now started the new grace period, we know | |
810 | * that all of its callbacks will be covered by this upcoming grace | |
811 | * period, even the ones that were registered arbitrarily recently. | |
812 | * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL. | |
813 | * | |
814 | * Other CPUs cannot be sure exactly when the grace period started. | |
815 | * Therefore, their recently registered callbacks must pass through | |
816 | * an additional RCU_NEXT_READY stage, so that they will be handled | |
817 | * by the next RCU grace period. | |
818 | */ | |
819 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
820 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
9160306e PM |
821 | |
822 | /* Set state so that this CPU will detect the next quiescent state. */ | |
823 | __note_new_gpnum(rsp, rnp, rdp); | |
d09b62df PM |
824 | } |
825 | ||
64db4cff PM |
826 | /* |
827 | * Start a new RCU grace period if warranted, re-initializing the hierarchy | |
828 | * in preparation for detecting the next grace period. The caller must hold | |
829 | * the root node's ->lock, which is released before return. Hard irqs must | |
830 | * be disabled. | |
831 | */ | |
832 | static void | |
833 | rcu_start_gp(struct rcu_state *rsp, unsigned long flags) | |
834 | __releases(rcu_get_root(rsp)->lock) | |
835 | { | |
394f99a9 | 836 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
64db4cff | 837 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff | 838 | |
07079d53 | 839 | if (!cpu_needs_another_gp(rsp, rdp) || rsp->fqs_active) { |
46a1e34e PM |
840 | if (cpu_needs_another_gp(rsp, rdp)) |
841 | rsp->fqs_need_gp = 1; | |
b32e9eb6 | 842 | if (rnp->completed == rsp->completed) { |
1304afb2 | 843 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
b32e9eb6 PM |
844 | return; |
845 | } | |
1304afb2 | 846 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
b32e9eb6 PM |
847 | |
848 | /* | |
849 | * Propagate new ->completed value to rcu_node structures | |
850 | * so that other CPUs don't have to wait until the start | |
851 | * of the next grace period to process their callbacks. | |
852 | */ | |
853 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
1304afb2 | 854 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
b32e9eb6 | 855 | rnp->completed = rsp->completed; |
1304afb2 | 856 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
b32e9eb6 PM |
857 | } |
858 | local_irq_restore(flags); | |
64db4cff PM |
859 | return; |
860 | } | |
861 | ||
862 | /* Advance to a new grace period and initialize state. */ | |
863 | rsp->gpnum++; | |
d4c08f2a | 864 | trace_rcu_grace_period(rsp->name, rsp->gpnum, "start"); |
c3422bea | 865 | WARN_ON_ONCE(rsp->signaled == RCU_GP_INIT); |
64db4cff PM |
866 | rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */ |
867 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; | |
64db4cff | 868 | record_gp_stall_check_time(rsp); |
64db4cff | 869 | |
64db4cff PM |
870 | /* Special-case the common single-level case. */ |
871 | if (NUM_RCU_NODES == 1) { | |
b0e165c0 | 872 | rcu_preempt_check_blocked_tasks(rnp); |
28ecd580 | 873 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 874 | rnp->gpnum = rsp->gpnum; |
d09b62df | 875 | rnp->completed = rsp->completed; |
c12172c0 | 876 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */ |
d09b62df | 877 | rcu_start_gp_per_cpu(rsp, rnp, rdp); |
27f4d280 | 878 | rcu_preempt_boost_start_gp(rnp); |
d4c08f2a PM |
879 | trace_rcu_grace_period_init(rsp->name, rnp->gpnum, |
880 | rnp->level, rnp->grplo, | |
881 | rnp->grphi, rnp->qsmask); | |
1304afb2 | 882 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
883 | return; |
884 | } | |
885 | ||
1304afb2 | 886 | raw_spin_unlock(&rnp->lock); /* leave irqs disabled. */ |
64db4cff PM |
887 | |
888 | ||
889 | /* Exclude any concurrent CPU-hotplug operations. */ | |
1304afb2 | 890 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
64db4cff PM |
891 | |
892 | /* | |
b835db1f PM |
893 | * Set the quiescent-state-needed bits in all the rcu_node |
894 | * structures for all currently online CPUs in breadth-first | |
895 | * order, starting from the root rcu_node structure. This | |
896 | * operation relies on the layout of the hierarchy within the | |
897 | * rsp->node[] array. Note that other CPUs will access only | |
898 | * the leaves of the hierarchy, which still indicate that no | |
899 | * grace period is in progress, at least until the corresponding | |
900 | * leaf node has been initialized. In addition, we have excluded | |
901 | * CPU-hotplug operations. | |
64db4cff PM |
902 | * |
903 | * Note that the grace period cannot complete until we finish | |
904 | * the initialization process, as there will be at least one | |
905 | * qsmask bit set in the root node until that time, namely the | |
b835db1f PM |
906 | * one corresponding to this CPU, due to the fact that we have |
907 | * irqs disabled. | |
64db4cff | 908 | */ |
a0b6c9a7 | 909 | rcu_for_each_node_breadth_first(rsp, rnp) { |
1304afb2 | 910 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
b0e165c0 | 911 | rcu_preempt_check_blocked_tasks(rnp); |
49e29126 | 912 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 913 | rnp->gpnum = rsp->gpnum; |
d09b62df PM |
914 | rnp->completed = rsp->completed; |
915 | if (rnp == rdp->mynode) | |
916 | rcu_start_gp_per_cpu(rsp, rnp, rdp); | |
27f4d280 | 917 | rcu_preempt_boost_start_gp(rnp); |
d4c08f2a PM |
918 | trace_rcu_grace_period_init(rsp->name, rnp->gpnum, |
919 | rnp->level, rnp->grplo, | |
920 | rnp->grphi, rnp->qsmask); | |
1304afb2 | 921 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
922 | } |
923 | ||
83f5b01f | 924 | rnp = rcu_get_root(rsp); |
1304afb2 | 925 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff | 926 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */ |
1304afb2 PM |
927 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
928 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); | |
64db4cff PM |
929 | } |
930 | ||
f41d911f | 931 | /* |
d3f6bad3 PM |
932 | * Report a full set of quiescent states to the specified rcu_state |
933 | * data structure. This involves cleaning up after the prior grace | |
934 | * period and letting rcu_start_gp() start up the next grace period | |
935 | * if one is needed. Note that the caller must hold rnp->lock, as | |
936 | * required by rcu_start_gp(), which will release it. | |
f41d911f | 937 | */ |
d3f6bad3 | 938 | static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags) |
fc2219d4 | 939 | __releases(rcu_get_root(rsp)->lock) |
f41d911f | 940 | { |
15ba0ba8 PM |
941 | unsigned long gp_duration; |
942 | ||
fc2219d4 | 943 | WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); |
0bbcc529 PM |
944 | |
945 | /* | |
946 | * Ensure that all grace-period and pre-grace-period activity | |
947 | * is seen before the assignment to rsp->completed. | |
948 | */ | |
949 | smp_mb(); /* See above block comment. */ | |
15ba0ba8 PM |
950 | gp_duration = jiffies - rsp->gp_start; |
951 | if (gp_duration > rsp->gp_max) | |
952 | rsp->gp_max = gp_duration; | |
f41d911f | 953 | rsp->completed = rsp->gpnum; |
d4c08f2a | 954 | trace_rcu_grace_period(rsp->name, rsp->completed, "end"); |
83f5b01f | 955 | rsp->signaled = RCU_GP_IDLE; |
f41d911f PM |
956 | rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */ |
957 | } | |
958 | ||
64db4cff | 959 | /* |
d3f6bad3 PM |
960 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
961 | * Allows quiescent states for a group of CPUs to be reported at one go | |
962 | * to the specified rcu_node structure, though all the CPUs in the group | |
963 | * must be represented by the same rcu_node structure (which need not be | |
964 | * a leaf rcu_node structure, though it often will be). That structure's | |
965 | * lock must be held upon entry, and it is released before return. | |
64db4cff PM |
966 | */ |
967 | static void | |
d3f6bad3 PM |
968 | rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp, |
969 | struct rcu_node *rnp, unsigned long flags) | |
64db4cff PM |
970 | __releases(rnp->lock) |
971 | { | |
28ecd580 PM |
972 | struct rcu_node *rnp_c; |
973 | ||
64db4cff PM |
974 | /* Walk up the rcu_node hierarchy. */ |
975 | for (;;) { | |
976 | if (!(rnp->qsmask & mask)) { | |
977 | ||
978 | /* Our bit has already been cleared, so done. */ | |
1304afb2 | 979 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
980 | return; |
981 | } | |
982 | rnp->qsmask &= ~mask; | |
d4c08f2a PM |
983 | trace_rcu_quiescent_state_report(rsp->name, rnp->gpnum, |
984 | mask, rnp->qsmask, rnp->level, | |
985 | rnp->grplo, rnp->grphi, | |
986 | !!rnp->gp_tasks); | |
27f4d280 | 987 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
64db4cff PM |
988 | |
989 | /* Other bits still set at this level, so done. */ | |
1304afb2 | 990 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
991 | return; |
992 | } | |
993 | mask = rnp->grpmask; | |
994 | if (rnp->parent == NULL) { | |
995 | ||
996 | /* No more levels. Exit loop holding root lock. */ | |
997 | ||
998 | break; | |
999 | } | |
1304afb2 | 1000 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
28ecd580 | 1001 | rnp_c = rnp; |
64db4cff | 1002 | rnp = rnp->parent; |
1304afb2 | 1003 | raw_spin_lock_irqsave(&rnp->lock, flags); |
28ecd580 | 1004 | WARN_ON_ONCE(rnp_c->qsmask); |
64db4cff PM |
1005 | } |
1006 | ||
1007 | /* | |
1008 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 1009 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 1010 | * to clean up and start the next grace period if one is needed. |
64db4cff | 1011 | */ |
d3f6bad3 | 1012 | rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */ |
64db4cff PM |
1013 | } |
1014 | ||
1015 | /* | |
d3f6bad3 PM |
1016 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
1017 | * structure. This must be either called from the specified CPU, or | |
1018 | * called when the specified CPU is known to be offline (and when it is | |
1019 | * also known that no other CPU is concurrently trying to help the offline | |
1020 | * CPU). The lastcomp argument is used to make sure we are still in the | |
1021 | * grace period of interest. We don't want to end the current grace period | |
1022 | * based on quiescent states detected in an earlier grace period! | |
64db4cff PM |
1023 | */ |
1024 | static void | |
e4cc1f22 | 1025 | rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastgp) |
64db4cff PM |
1026 | { |
1027 | unsigned long flags; | |
1028 | unsigned long mask; | |
1029 | struct rcu_node *rnp; | |
1030 | ||
1031 | rnp = rdp->mynode; | |
1304afb2 | 1032 | raw_spin_lock_irqsave(&rnp->lock, flags); |
e4cc1f22 | 1033 | if (lastgp != rnp->gpnum || rnp->completed == rnp->gpnum) { |
64db4cff PM |
1034 | |
1035 | /* | |
e4cc1f22 PM |
1036 | * The grace period in which this quiescent state was |
1037 | * recorded has ended, so don't report it upwards. | |
1038 | * We will instead need a new quiescent state that lies | |
1039 | * within the current grace period. | |
64db4cff | 1040 | */ |
e4cc1f22 | 1041 | rdp->passed_quiesce = 0; /* need qs for new gp. */ |
1304afb2 | 1042 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1043 | return; |
1044 | } | |
1045 | mask = rdp->grpmask; | |
1046 | if ((rnp->qsmask & mask) == 0) { | |
1304afb2 | 1047 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1048 | } else { |
1049 | rdp->qs_pending = 0; | |
1050 | ||
1051 | /* | |
1052 | * This GP can't end until cpu checks in, so all of our | |
1053 | * callbacks can be processed during the next GP. | |
1054 | */ | |
64db4cff PM |
1055 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
1056 | ||
d3f6bad3 | 1057 | rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */ |
64db4cff PM |
1058 | } |
1059 | } | |
1060 | ||
1061 | /* | |
1062 | * Check to see if there is a new grace period of which this CPU | |
1063 | * is not yet aware, and if so, set up local rcu_data state for it. | |
1064 | * Otherwise, see if this CPU has just passed through its first | |
1065 | * quiescent state for this grace period, and record that fact if so. | |
1066 | */ | |
1067 | static void | |
1068 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | |
1069 | { | |
1070 | /* If there is now a new grace period, record and return. */ | |
1071 | if (check_for_new_grace_period(rsp, rdp)) | |
1072 | return; | |
1073 | ||
1074 | /* | |
1075 | * Does this CPU still need to do its part for current grace period? | |
1076 | * If no, return and let the other CPUs do their part as well. | |
1077 | */ | |
1078 | if (!rdp->qs_pending) | |
1079 | return; | |
1080 | ||
1081 | /* | |
1082 | * Was there a quiescent state since the beginning of the grace | |
1083 | * period? If no, then exit and wait for the next call. | |
1084 | */ | |
e4cc1f22 | 1085 | if (!rdp->passed_quiesce) |
64db4cff PM |
1086 | return; |
1087 | ||
d3f6bad3 PM |
1088 | /* |
1089 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
1090 | * judge of that). | |
1091 | */ | |
e4cc1f22 | 1092 | rcu_report_qs_rdp(rdp->cpu, rsp, rdp, rdp->passed_quiesce_gpnum); |
64db4cff PM |
1093 | } |
1094 | ||
1095 | #ifdef CONFIG_HOTPLUG_CPU | |
1096 | ||
e74f4c45 | 1097 | /* |
29494be7 LJ |
1098 | * Move a dying CPU's RCU callbacks to online CPU's callback list. |
1099 | * Synchronization is not required because this function executes | |
1100 | * in stop_machine() context. | |
e74f4c45 | 1101 | */ |
29494be7 | 1102 | static void rcu_send_cbs_to_online(struct rcu_state *rsp) |
e74f4c45 PM |
1103 | { |
1104 | int i; | |
29494be7 LJ |
1105 | /* current DYING CPU is cleared in the cpu_online_mask */ |
1106 | int receive_cpu = cpumask_any(cpu_online_mask); | |
394f99a9 | 1107 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
29494be7 | 1108 | struct rcu_data *receive_rdp = per_cpu_ptr(rsp->rda, receive_cpu); |
e74f4c45 PM |
1109 | |
1110 | if (rdp->nxtlist == NULL) | |
1111 | return; /* irqs disabled, so comparison is stable. */ | |
29494be7 LJ |
1112 | |
1113 | *receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist; | |
1114 | receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
1115 | receive_rdp->qlen += rdp->qlen; | |
1116 | receive_rdp->n_cbs_adopted += rdp->qlen; | |
1117 | rdp->n_cbs_orphaned += rdp->qlen; | |
1118 | ||
e74f4c45 PM |
1119 | rdp->nxtlist = NULL; |
1120 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
1121 | rdp->nxttail[i] = &rdp->nxtlist; | |
e74f4c45 | 1122 | rdp->qlen = 0; |
e74f4c45 PM |
1123 | } |
1124 | ||
64db4cff PM |
1125 | /* |
1126 | * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy | |
1127 | * and move all callbacks from the outgoing CPU to the current one. | |
a26ac245 PM |
1128 | * There can only be one CPU hotplug operation at a time, so no other |
1129 | * CPU can be attempting to update rcu_cpu_kthread_task. | |
64db4cff PM |
1130 | */ |
1131 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp) | |
1132 | { | |
64db4cff | 1133 | unsigned long flags; |
64db4cff | 1134 | unsigned long mask; |
d9a3da06 | 1135 | int need_report = 0; |
394f99a9 | 1136 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff | 1137 | struct rcu_node *rnp; |
a26ac245 | 1138 | |
f8b7fc6b | 1139 | rcu_stop_cpu_kthread(cpu); |
64db4cff PM |
1140 | |
1141 | /* Exclude any attempts to start a new grace period. */ | |
1304afb2 | 1142 | raw_spin_lock_irqsave(&rsp->onofflock, flags); |
64db4cff PM |
1143 | |
1144 | /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */ | |
28ecd580 | 1145 | rnp = rdp->mynode; /* this is the outgoing CPU's rnp. */ |
64db4cff PM |
1146 | mask = rdp->grpmask; /* rnp->grplo is constant. */ |
1147 | do { | |
1304afb2 | 1148 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
1149 | rnp->qsmaskinit &= ~mask; |
1150 | if (rnp->qsmaskinit != 0) { | |
b668c9cf | 1151 | if (rnp != rdp->mynode) |
1304afb2 | 1152 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
d4c08f2a PM |
1153 | else |
1154 | trace_rcu_grace_period(rsp->name, | |
1155 | rnp->gpnum + 1 - | |
1156 | !!(rnp->qsmask & mask), | |
1157 | "cpuofl"); | |
64db4cff PM |
1158 | break; |
1159 | } | |
d4c08f2a PM |
1160 | if (rnp == rdp->mynode) { |
1161 | trace_rcu_grace_period(rsp->name, | |
1162 | rnp->gpnum + 1 - | |
1163 | !!(rnp->qsmask & mask), | |
1164 | "cpuofl"); | |
d9a3da06 | 1165 | need_report = rcu_preempt_offline_tasks(rsp, rnp, rdp); |
d4c08f2a | 1166 | } else |
1304afb2 | 1167 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff | 1168 | mask = rnp->grpmask; |
64db4cff PM |
1169 | rnp = rnp->parent; |
1170 | } while (rnp != NULL); | |
64db4cff | 1171 | |
b668c9cf PM |
1172 | /* |
1173 | * We still hold the leaf rcu_node structure lock here, and | |
1174 | * irqs are still disabled. The reason for this subterfuge is | |
d3f6bad3 PM |
1175 | * because invoking rcu_report_unblock_qs_rnp() with ->onofflock |
1176 | * held leads to deadlock. | |
b668c9cf | 1177 | */ |
1304afb2 | 1178 | raw_spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ |
b668c9cf | 1179 | rnp = rdp->mynode; |
d9a3da06 | 1180 | if (need_report & RCU_OFL_TASKS_NORM_GP) |
d3f6bad3 | 1181 | rcu_report_unblock_qs_rnp(rnp, flags); |
b668c9cf | 1182 | else |
1304afb2 | 1183 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
d9a3da06 PM |
1184 | if (need_report & RCU_OFL_TASKS_EXP_GP) |
1185 | rcu_report_exp_rnp(rsp, rnp); | |
1217ed1b | 1186 | rcu_node_kthread_setaffinity(rnp, -1); |
64db4cff PM |
1187 | } |
1188 | ||
1189 | /* | |
1190 | * Remove the specified CPU from the RCU hierarchy and move any pending | |
1191 | * callbacks that it might have to the current CPU. This code assumes | |
1192 | * that at least one CPU in the system will remain running at all times. | |
1193 | * Any attempt to offline -all- CPUs is likely to strand RCU callbacks. | |
1194 | */ | |
1195 | static void rcu_offline_cpu(int cpu) | |
1196 | { | |
d6714c22 | 1197 | __rcu_offline_cpu(cpu, &rcu_sched_state); |
64db4cff | 1198 | __rcu_offline_cpu(cpu, &rcu_bh_state); |
33f76148 | 1199 | rcu_preempt_offline_cpu(cpu); |
64db4cff PM |
1200 | } |
1201 | ||
1202 | #else /* #ifdef CONFIG_HOTPLUG_CPU */ | |
1203 | ||
29494be7 | 1204 | static void rcu_send_cbs_to_online(struct rcu_state *rsp) |
e74f4c45 PM |
1205 | { |
1206 | } | |
1207 | ||
64db4cff PM |
1208 | static void rcu_offline_cpu(int cpu) |
1209 | { | |
1210 | } | |
1211 | ||
1212 | #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */ | |
1213 | ||
1214 | /* | |
1215 | * Invoke any RCU callbacks that have made it to the end of their grace | |
1216 | * period. Thottle as specified by rdp->blimit. | |
1217 | */ | |
37c72e56 | 1218 | static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
1219 | { |
1220 | unsigned long flags; | |
1221 | struct rcu_head *next, *list, **tail; | |
29c00b4a | 1222 | int bl, count; |
64db4cff PM |
1223 | |
1224 | /* If no callbacks are ready, just return.*/ | |
29c00b4a | 1225 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) { |
72fe701b PM |
1226 | trace_rcu_batch_start(rsp->name, 0, 0); |
1227 | trace_rcu_batch_end(rsp->name, 0); | |
64db4cff | 1228 | return; |
29c00b4a | 1229 | } |
64db4cff PM |
1230 | |
1231 | /* | |
1232 | * Extract the list of ready callbacks, disabling to prevent | |
1233 | * races with call_rcu() from interrupt handlers. | |
1234 | */ | |
1235 | local_irq_save(flags); | |
29c00b4a | 1236 | bl = rdp->blimit; |
72fe701b | 1237 | trace_rcu_batch_start(rsp->name, rdp->qlen, bl); |
64db4cff PM |
1238 | list = rdp->nxtlist; |
1239 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; | |
1240 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | |
1241 | tail = rdp->nxttail[RCU_DONE_TAIL]; | |
1242 | for (count = RCU_NEXT_SIZE - 1; count >= 0; count--) | |
1243 | if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL]) | |
1244 | rdp->nxttail[count] = &rdp->nxtlist; | |
1245 | local_irq_restore(flags); | |
1246 | ||
1247 | /* Invoke callbacks. */ | |
1248 | count = 0; | |
1249 | while (list) { | |
1250 | next = list->next; | |
1251 | prefetch(next); | |
551d55a9 | 1252 | debug_rcu_head_unqueue(list); |
d4c08f2a | 1253 | __rcu_reclaim(rsp->name, list); |
64db4cff | 1254 | list = next; |
29c00b4a | 1255 | if (++count >= bl) |
64db4cff PM |
1256 | break; |
1257 | } | |
1258 | ||
1259 | local_irq_save(flags); | |
72fe701b | 1260 | trace_rcu_batch_end(rsp->name, count); |
64db4cff PM |
1261 | |
1262 | /* Update count, and requeue any remaining callbacks. */ | |
1263 | rdp->qlen -= count; | |
269dcc1c | 1264 | rdp->n_cbs_invoked += count; |
64db4cff PM |
1265 | if (list != NULL) { |
1266 | *tail = rdp->nxtlist; | |
1267 | rdp->nxtlist = list; | |
1268 | for (count = 0; count < RCU_NEXT_SIZE; count++) | |
1269 | if (&rdp->nxtlist == rdp->nxttail[count]) | |
1270 | rdp->nxttail[count] = tail; | |
1271 | else | |
1272 | break; | |
1273 | } | |
1274 | ||
1275 | /* Reinstate batch limit if we have worked down the excess. */ | |
1276 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) | |
1277 | rdp->blimit = blimit; | |
1278 | ||
37c72e56 PM |
1279 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
1280 | if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) { | |
1281 | rdp->qlen_last_fqs_check = 0; | |
1282 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1283 | } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark) | |
1284 | rdp->qlen_last_fqs_check = rdp->qlen; | |
1285 | ||
64db4cff PM |
1286 | local_irq_restore(flags); |
1287 | ||
e0f23060 | 1288 | /* Re-invoke RCU core processing if there are callbacks remaining. */ |
64db4cff | 1289 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
a46e0899 | 1290 | invoke_rcu_core(); |
64db4cff PM |
1291 | } |
1292 | ||
1293 | /* | |
1294 | * Check to see if this CPU is in a non-context-switch quiescent state | |
1295 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
e0f23060 | 1296 | * Also schedule RCU core processing. |
64db4cff PM |
1297 | * |
1298 | * This function must be called with hardirqs disabled. It is normally | |
1299 | * invoked from the scheduling-clock interrupt. If rcu_pending returns | |
1300 | * false, there is no point in invoking rcu_check_callbacks(). | |
1301 | */ | |
1302 | void rcu_check_callbacks(int cpu, int user) | |
1303 | { | |
300df91c | 1304 | trace_rcu_utilization("Start scheduler-tick"); |
64db4cff | 1305 | if (user || |
a6826048 PM |
1306 | (idle_cpu(cpu) && rcu_scheduler_active && |
1307 | !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) { | |
64db4cff PM |
1308 | |
1309 | /* | |
1310 | * Get here if this CPU took its interrupt from user | |
1311 | * mode or from the idle loop, and if this is not a | |
1312 | * nested interrupt. In this case, the CPU is in | |
d6714c22 | 1313 | * a quiescent state, so note it. |
64db4cff PM |
1314 | * |
1315 | * No memory barrier is required here because both | |
d6714c22 PM |
1316 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
1317 | * variables that other CPUs neither access nor modify, | |
1318 | * at least not while the corresponding CPU is online. | |
64db4cff PM |
1319 | */ |
1320 | ||
d6714c22 PM |
1321 | rcu_sched_qs(cpu); |
1322 | rcu_bh_qs(cpu); | |
64db4cff PM |
1323 | |
1324 | } else if (!in_softirq()) { | |
1325 | ||
1326 | /* | |
1327 | * Get here if this CPU did not take its interrupt from | |
1328 | * softirq, in other words, if it is not interrupting | |
1329 | * a rcu_bh read-side critical section. This is an _bh | |
d6714c22 | 1330 | * critical section, so note it. |
64db4cff PM |
1331 | */ |
1332 | ||
d6714c22 | 1333 | rcu_bh_qs(cpu); |
64db4cff | 1334 | } |
f41d911f | 1335 | rcu_preempt_check_callbacks(cpu); |
d21670ac | 1336 | if (rcu_pending(cpu)) |
a46e0899 | 1337 | invoke_rcu_core(); |
300df91c | 1338 | trace_rcu_utilization("End scheduler-tick"); |
64db4cff PM |
1339 | } |
1340 | ||
1341 | #ifdef CONFIG_SMP | |
1342 | ||
1343 | /* | |
1344 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
1345 | * have not yet encountered a quiescent state, using the function specified. | |
27f4d280 PM |
1346 | * Also initiate boosting for any threads blocked on the root rcu_node. |
1347 | * | |
ee47eb9f | 1348 | * The caller must have suppressed start of new grace periods. |
64db4cff | 1349 | */ |
45f014c5 | 1350 | static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *)) |
64db4cff PM |
1351 | { |
1352 | unsigned long bit; | |
1353 | int cpu; | |
1354 | unsigned long flags; | |
1355 | unsigned long mask; | |
a0b6c9a7 | 1356 | struct rcu_node *rnp; |
64db4cff | 1357 | |
a0b6c9a7 | 1358 | rcu_for_each_leaf_node(rsp, rnp) { |
64db4cff | 1359 | mask = 0; |
1304afb2 | 1360 | raw_spin_lock_irqsave(&rnp->lock, flags); |
ee47eb9f | 1361 | if (!rcu_gp_in_progress(rsp)) { |
1304afb2 | 1362 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
0f10dc82 | 1363 | return; |
64db4cff | 1364 | } |
a0b6c9a7 | 1365 | if (rnp->qsmask == 0) { |
1217ed1b | 1366 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock */ |
64db4cff PM |
1367 | continue; |
1368 | } | |
a0b6c9a7 | 1369 | cpu = rnp->grplo; |
64db4cff | 1370 | bit = 1; |
a0b6c9a7 | 1371 | for (; cpu <= rnp->grphi; cpu++, bit <<= 1) { |
394f99a9 LJ |
1372 | if ((rnp->qsmask & bit) != 0 && |
1373 | f(per_cpu_ptr(rsp->rda, cpu))) | |
64db4cff PM |
1374 | mask |= bit; |
1375 | } | |
45f014c5 | 1376 | if (mask != 0) { |
64db4cff | 1377 | |
d3f6bad3 PM |
1378 | /* rcu_report_qs_rnp() releases rnp->lock. */ |
1379 | rcu_report_qs_rnp(mask, rsp, rnp, flags); | |
64db4cff PM |
1380 | continue; |
1381 | } | |
1304afb2 | 1382 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 1383 | } |
27f4d280 | 1384 | rnp = rcu_get_root(rsp); |
1217ed1b PM |
1385 | if (rnp->qsmask == 0) { |
1386 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
1387 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */ | |
1388 | } | |
64db4cff PM |
1389 | } |
1390 | ||
1391 | /* | |
1392 | * Force quiescent states on reluctant CPUs, and also detect which | |
1393 | * CPUs are in dyntick-idle mode. | |
1394 | */ | |
1395 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1396 | { | |
1397 | unsigned long flags; | |
64db4cff | 1398 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff | 1399 | |
300df91c PM |
1400 | trace_rcu_utilization("Start fqs"); |
1401 | if (!rcu_gp_in_progress(rsp)) { | |
1402 | trace_rcu_utilization("End fqs"); | |
64db4cff | 1403 | return; /* No grace period in progress, nothing to force. */ |
300df91c | 1404 | } |
1304afb2 | 1405 | if (!raw_spin_trylock_irqsave(&rsp->fqslock, flags)) { |
64db4cff | 1406 | rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ |
300df91c | 1407 | trace_rcu_utilization("End fqs"); |
64db4cff PM |
1408 | return; /* Someone else is already on the job. */ |
1409 | } | |
20133cfc | 1410 | if (relaxed && ULONG_CMP_GE(rsp->jiffies_force_qs, jiffies)) |
f96e9232 | 1411 | goto unlock_fqs_ret; /* no emergency and done recently. */ |
64db4cff | 1412 | rsp->n_force_qs++; |
1304afb2 | 1413 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
64db4cff | 1414 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
560d4bc0 | 1415 | if(!rcu_gp_in_progress(rsp)) { |
64db4cff | 1416 | rsp->n_force_qs_ngp++; |
1304afb2 | 1417 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1418 | goto unlock_fqs_ret; /* no GP in progress, time updated. */ |
64db4cff | 1419 | } |
07079d53 | 1420 | rsp->fqs_active = 1; |
f3a8b5c6 | 1421 | switch (rsp->signaled) { |
83f5b01f | 1422 | case RCU_GP_IDLE: |
64db4cff PM |
1423 | case RCU_GP_INIT: |
1424 | ||
83f5b01f | 1425 | break; /* grace period idle or initializing, ignore. */ |
64db4cff PM |
1426 | |
1427 | case RCU_SAVE_DYNTICK: | |
64db4cff PM |
1428 | if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK) |
1429 | break; /* So gcc recognizes the dead code. */ | |
1430 | ||
f261414f LJ |
1431 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
1432 | ||
64db4cff | 1433 | /* Record dyntick-idle state. */ |
45f014c5 | 1434 | force_qs_rnp(rsp, dyntick_save_progress_counter); |
1304afb2 | 1435 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
ee47eb9f | 1436 | if (rcu_gp_in_progress(rsp)) |
64db4cff | 1437 | rsp->signaled = RCU_FORCE_QS; |
ee47eb9f | 1438 | break; |
64db4cff PM |
1439 | |
1440 | case RCU_FORCE_QS: | |
1441 | ||
1442 | /* Check dyntick-idle state, send IPI to laggarts. */ | |
1304afb2 | 1443 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
45f014c5 | 1444 | force_qs_rnp(rsp, rcu_implicit_dynticks_qs); |
64db4cff PM |
1445 | |
1446 | /* Leave state in case more forcing is required. */ | |
1447 | ||
1304afb2 | 1448 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
f96e9232 | 1449 | break; |
64db4cff | 1450 | } |
07079d53 | 1451 | rsp->fqs_active = 0; |
46a1e34e | 1452 | if (rsp->fqs_need_gp) { |
1304afb2 | 1453 | raw_spin_unlock(&rsp->fqslock); /* irqs remain disabled */ |
46a1e34e PM |
1454 | rsp->fqs_need_gp = 0; |
1455 | rcu_start_gp(rsp, flags); /* releases rnp->lock */ | |
300df91c | 1456 | trace_rcu_utilization("End fqs"); |
46a1e34e PM |
1457 | return; |
1458 | } | |
1304afb2 | 1459 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1460 | unlock_fqs_ret: |
1304afb2 | 1461 | raw_spin_unlock_irqrestore(&rsp->fqslock, flags); |
300df91c | 1462 | trace_rcu_utilization("End fqs"); |
64db4cff PM |
1463 | } |
1464 | ||
1465 | #else /* #ifdef CONFIG_SMP */ | |
1466 | ||
1467 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1468 | { | |
1469 | set_need_resched(); | |
1470 | } | |
1471 | ||
1472 | #endif /* #else #ifdef CONFIG_SMP */ | |
1473 | ||
1474 | /* | |
e0f23060 PM |
1475 | * This does the RCU core processing work for the specified rcu_state |
1476 | * and rcu_data structures. This may be called only from the CPU to | |
1477 | * whom the rdp belongs. | |
64db4cff PM |
1478 | */ |
1479 | static void | |
1480 | __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) | |
1481 | { | |
1482 | unsigned long flags; | |
1483 | ||
2e597558 PM |
1484 | WARN_ON_ONCE(rdp->beenonline == 0); |
1485 | ||
64db4cff PM |
1486 | /* |
1487 | * If an RCU GP has gone long enough, go check for dyntick | |
1488 | * idle CPUs and, if needed, send resched IPIs. | |
1489 | */ | |
20133cfc | 1490 | if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
64db4cff PM |
1491 | force_quiescent_state(rsp, 1); |
1492 | ||
1493 | /* | |
1494 | * Advance callbacks in response to end of earlier grace | |
1495 | * period that some other CPU ended. | |
1496 | */ | |
1497 | rcu_process_gp_end(rsp, rdp); | |
1498 | ||
1499 | /* Update RCU state based on any recent quiescent states. */ | |
1500 | rcu_check_quiescent_state(rsp, rdp); | |
1501 | ||
1502 | /* Does this CPU require a not-yet-started grace period? */ | |
1503 | if (cpu_needs_another_gp(rsp, rdp)) { | |
1304afb2 | 1504 | raw_spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags); |
64db4cff PM |
1505 | rcu_start_gp(rsp, flags); /* releases above lock */ |
1506 | } | |
1507 | ||
1508 | /* If there are callbacks ready, invoke them. */ | |
09223371 | 1509 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
a46e0899 | 1510 | invoke_rcu_callbacks(rsp, rdp); |
09223371 SL |
1511 | } |
1512 | ||
64db4cff | 1513 | /* |
e0f23060 | 1514 | * Do RCU core processing for the current CPU. |
64db4cff | 1515 | */ |
09223371 | 1516 | static void rcu_process_callbacks(struct softirq_action *unused) |
64db4cff | 1517 | { |
300df91c | 1518 | trace_rcu_utilization("Start RCU core"); |
d6714c22 PM |
1519 | __rcu_process_callbacks(&rcu_sched_state, |
1520 | &__get_cpu_var(rcu_sched_data)); | |
64db4cff | 1521 | __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); |
f41d911f | 1522 | rcu_preempt_process_callbacks(); |
a47cd880 PM |
1523 | |
1524 | /* If we are last CPU on way to dyntick-idle mode, accelerate it. */ | |
1525 | rcu_needs_cpu_flush(); | |
300df91c | 1526 | trace_rcu_utilization("End RCU core"); |
64db4cff PM |
1527 | } |
1528 | ||
a26ac245 | 1529 | /* |
e0f23060 PM |
1530 | * Schedule RCU callback invocation. If the specified type of RCU |
1531 | * does not support RCU priority boosting, just do a direct call, | |
1532 | * otherwise wake up the per-CPU kernel kthread. Note that because we | |
1533 | * are running on the current CPU with interrupts disabled, the | |
1534 | * rcu_cpu_kthread_task cannot disappear out from under us. | |
a26ac245 | 1535 | */ |
a46e0899 | 1536 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) |
a26ac245 | 1537 | { |
b0d30417 PM |
1538 | if (unlikely(!ACCESS_ONCE(rcu_scheduler_fully_active))) |
1539 | return; | |
a46e0899 PM |
1540 | if (likely(!rsp->boost)) { |
1541 | rcu_do_batch(rsp, rdp); | |
a26ac245 PM |
1542 | return; |
1543 | } | |
a46e0899 | 1544 | invoke_rcu_callbacks_kthread(); |
a26ac245 PM |
1545 | } |
1546 | ||
a46e0899 | 1547 | static void invoke_rcu_core(void) |
09223371 SL |
1548 | { |
1549 | raise_softirq(RCU_SOFTIRQ); | |
1550 | } | |
1551 | ||
64db4cff PM |
1552 | static void |
1553 | __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu), | |
1554 | struct rcu_state *rsp) | |
1555 | { | |
1556 | unsigned long flags; | |
1557 | struct rcu_data *rdp; | |
1558 | ||
551d55a9 | 1559 | debug_rcu_head_queue(head); |
64db4cff PM |
1560 | head->func = func; |
1561 | head->next = NULL; | |
1562 | ||
1563 | smp_mb(); /* Ensure RCU update seen before callback registry. */ | |
1564 | ||
1565 | /* | |
1566 | * Opportunistically note grace-period endings and beginnings. | |
1567 | * Note that we might see a beginning right after we see an | |
1568 | * end, but never vice versa, since this CPU has to pass through | |
1569 | * a quiescent state betweentimes. | |
1570 | */ | |
1571 | local_irq_save(flags); | |
394f99a9 | 1572 | rdp = this_cpu_ptr(rsp->rda); |
64db4cff PM |
1573 | |
1574 | /* Add the callback to our list. */ | |
1575 | *rdp->nxttail[RCU_NEXT_TAIL] = head; | |
1576 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; | |
2655d57e PM |
1577 | rdp->qlen++; |
1578 | ||
d4c08f2a PM |
1579 | if (__is_kfree_rcu_offset((unsigned long)func)) |
1580 | trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func, | |
1581 | rdp->qlen); | |
1582 | else | |
1583 | trace_rcu_callback(rsp->name, head, rdp->qlen); | |
1584 | ||
2655d57e PM |
1585 | /* If interrupts were disabled, don't dive into RCU core. */ |
1586 | if (irqs_disabled_flags(flags)) { | |
1587 | local_irq_restore(flags); | |
1588 | return; | |
1589 | } | |
64db4cff | 1590 | |
37c72e56 PM |
1591 | /* |
1592 | * Force the grace period if too many callbacks or too long waiting. | |
1593 | * Enforce hysteresis, and don't invoke force_quiescent_state() | |
1594 | * if some other CPU has recently done so. Also, don't bother | |
1595 | * invoking force_quiescent_state() if the newly enqueued callback | |
1596 | * is the only one waiting for a grace period to complete. | |
1597 | */ | |
2655d57e | 1598 | if (unlikely(rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) { |
b52573d2 PM |
1599 | |
1600 | /* Are we ignoring a completed grace period? */ | |
1601 | rcu_process_gp_end(rsp, rdp); | |
1602 | check_for_new_grace_period(rsp, rdp); | |
1603 | ||
1604 | /* Start a new grace period if one not already started. */ | |
1605 | if (!rcu_gp_in_progress(rsp)) { | |
1606 | unsigned long nestflag; | |
1607 | struct rcu_node *rnp_root = rcu_get_root(rsp); | |
1608 | ||
1609 | raw_spin_lock_irqsave(&rnp_root->lock, nestflag); | |
1610 | rcu_start_gp(rsp, nestflag); /* rlses rnp_root->lock */ | |
1611 | } else { | |
1612 | /* Give the grace period a kick. */ | |
1613 | rdp->blimit = LONG_MAX; | |
1614 | if (rsp->n_force_qs == rdp->n_force_qs_snap && | |
1615 | *rdp->nxttail[RCU_DONE_TAIL] != head) | |
1616 | force_quiescent_state(rsp, 0); | |
1617 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1618 | rdp->qlen_last_fqs_check = rdp->qlen; | |
1619 | } | |
20133cfc | 1620 | } else if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
64db4cff PM |
1621 | force_quiescent_state(rsp, 1); |
1622 | local_irq_restore(flags); | |
1623 | } | |
1624 | ||
1625 | /* | |
d6714c22 | 1626 | * Queue an RCU-sched callback for invocation after a grace period. |
64db4cff | 1627 | */ |
d6714c22 | 1628 | void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) |
64db4cff | 1629 | { |
d6714c22 | 1630 | __call_rcu(head, func, &rcu_sched_state); |
64db4cff | 1631 | } |
d6714c22 | 1632 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
64db4cff PM |
1633 | |
1634 | /* | |
1635 | * Queue an RCU for invocation after a quicker grace period. | |
1636 | */ | |
1637 | void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
1638 | { | |
1639 | __call_rcu(head, func, &rcu_bh_state); | |
1640 | } | |
1641 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
1642 | ||
6ebb237b PM |
1643 | /** |
1644 | * synchronize_sched - wait until an rcu-sched grace period has elapsed. | |
1645 | * | |
1646 | * Control will return to the caller some time after a full rcu-sched | |
1647 | * grace period has elapsed, in other words after all currently executing | |
1648 | * rcu-sched read-side critical sections have completed. These read-side | |
1649 | * critical sections are delimited by rcu_read_lock_sched() and | |
1650 | * rcu_read_unlock_sched(), and may be nested. Note that preempt_disable(), | |
1651 | * local_irq_disable(), and so on may be used in place of | |
1652 | * rcu_read_lock_sched(). | |
1653 | * | |
1654 | * This means that all preempt_disable code sequences, including NMI and | |
1655 | * hardware-interrupt handlers, in progress on entry will have completed | |
1656 | * before this primitive returns. However, this does not guarantee that | |
1657 | * softirq handlers will have completed, since in some kernels, these | |
1658 | * handlers can run in process context, and can block. | |
1659 | * | |
1660 | * This primitive provides the guarantees made by the (now removed) | |
1661 | * synchronize_kernel() API. In contrast, synchronize_rcu() only | |
1662 | * guarantees that rcu_read_lock() sections will have completed. | |
1663 | * In "classic RCU", these two guarantees happen to be one and | |
1664 | * the same, but can differ in realtime RCU implementations. | |
1665 | */ | |
1666 | void synchronize_sched(void) | |
1667 | { | |
6ebb237b PM |
1668 | if (rcu_blocking_is_gp()) |
1669 | return; | |
2c42818e | 1670 | wait_rcu_gp(call_rcu_sched); |
6ebb237b PM |
1671 | } |
1672 | EXPORT_SYMBOL_GPL(synchronize_sched); | |
1673 | ||
1674 | /** | |
1675 | * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed. | |
1676 | * | |
1677 | * Control will return to the caller some time after a full rcu_bh grace | |
1678 | * period has elapsed, in other words after all currently executing rcu_bh | |
1679 | * read-side critical sections have completed. RCU read-side critical | |
1680 | * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(), | |
1681 | * and may be nested. | |
1682 | */ | |
1683 | void synchronize_rcu_bh(void) | |
1684 | { | |
6ebb237b PM |
1685 | if (rcu_blocking_is_gp()) |
1686 | return; | |
2c42818e | 1687 | wait_rcu_gp(call_rcu_bh); |
6ebb237b PM |
1688 | } |
1689 | EXPORT_SYMBOL_GPL(synchronize_rcu_bh); | |
1690 | ||
64db4cff PM |
1691 | /* |
1692 | * Check to see if there is any immediate RCU-related work to be done | |
1693 | * by the current CPU, for the specified type of RCU, returning 1 if so. | |
1694 | * The checks are in order of increasing expense: checks that can be | |
1695 | * carried out against CPU-local state are performed first. However, | |
1696 | * we must check for CPU stalls first, else we might not get a chance. | |
1697 | */ | |
1698 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | |
1699 | { | |
2f51f988 PM |
1700 | struct rcu_node *rnp = rdp->mynode; |
1701 | ||
64db4cff PM |
1702 | rdp->n_rcu_pending++; |
1703 | ||
1704 | /* Check for CPU stalls, if enabled. */ | |
1705 | check_cpu_stall(rsp, rdp); | |
1706 | ||
1707 | /* Is the RCU core waiting for a quiescent state from this CPU? */ | |
e4cc1f22 | 1708 | if (rdp->qs_pending && !rdp->passed_quiesce) { |
d25eb944 PM |
1709 | |
1710 | /* | |
1711 | * If force_quiescent_state() coming soon and this CPU | |
1712 | * needs a quiescent state, and this is either RCU-sched | |
1713 | * or RCU-bh, force a local reschedule. | |
1714 | */ | |
d21670ac | 1715 | rdp->n_rp_qs_pending++; |
6cc68793 | 1716 | if (!rdp->preemptible && |
d25eb944 PM |
1717 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs) - 1, |
1718 | jiffies)) | |
1719 | set_need_resched(); | |
e4cc1f22 | 1720 | } else if (rdp->qs_pending && rdp->passed_quiesce) { |
d21670ac | 1721 | rdp->n_rp_report_qs++; |
64db4cff | 1722 | return 1; |
7ba5c840 | 1723 | } |
64db4cff PM |
1724 | |
1725 | /* Does this CPU have callbacks ready to invoke? */ | |
7ba5c840 PM |
1726 | if (cpu_has_callbacks_ready_to_invoke(rdp)) { |
1727 | rdp->n_rp_cb_ready++; | |
64db4cff | 1728 | return 1; |
7ba5c840 | 1729 | } |
64db4cff PM |
1730 | |
1731 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
7ba5c840 PM |
1732 | if (cpu_needs_another_gp(rsp, rdp)) { |
1733 | rdp->n_rp_cpu_needs_gp++; | |
64db4cff | 1734 | return 1; |
7ba5c840 | 1735 | } |
64db4cff PM |
1736 | |
1737 | /* Has another RCU grace period completed? */ | |
2f51f988 | 1738 | if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */ |
7ba5c840 | 1739 | rdp->n_rp_gp_completed++; |
64db4cff | 1740 | return 1; |
7ba5c840 | 1741 | } |
64db4cff PM |
1742 | |
1743 | /* Has a new RCU grace period started? */ | |
2f51f988 | 1744 | if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */ |
7ba5c840 | 1745 | rdp->n_rp_gp_started++; |
64db4cff | 1746 | return 1; |
7ba5c840 | 1747 | } |
64db4cff PM |
1748 | |
1749 | /* Has an RCU GP gone long enough to send resched IPIs &c? */ | |
fc2219d4 | 1750 | if (rcu_gp_in_progress(rsp) && |
20133cfc | 1751 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) { |
7ba5c840 | 1752 | rdp->n_rp_need_fqs++; |
64db4cff | 1753 | return 1; |
7ba5c840 | 1754 | } |
64db4cff PM |
1755 | |
1756 | /* nothing to do */ | |
7ba5c840 | 1757 | rdp->n_rp_need_nothing++; |
64db4cff PM |
1758 | return 0; |
1759 | } | |
1760 | ||
1761 | /* | |
1762 | * Check to see if there is any immediate RCU-related work to be done | |
1763 | * by the current CPU, returning 1 if so. This function is part of the | |
1764 | * RCU implementation; it is -not- an exported member of the RCU API. | |
1765 | */ | |
a157229c | 1766 | static int rcu_pending(int cpu) |
64db4cff | 1767 | { |
d6714c22 | 1768 | return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) || |
f41d911f PM |
1769 | __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) || |
1770 | rcu_preempt_pending(cpu); | |
64db4cff PM |
1771 | } |
1772 | ||
1773 | /* | |
1774 | * Check to see if any future RCU-related work will need to be done | |
1775 | * by the current CPU, even if none need be done immediately, returning | |
8bd93a2c | 1776 | * 1 if so. |
64db4cff | 1777 | */ |
8bd93a2c | 1778 | static int rcu_needs_cpu_quick_check(int cpu) |
64db4cff PM |
1779 | { |
1780 | /* RCU callbacks either ready or pending? */ | |
d6714c22 | 1781 | return per_cpu(rcu_sched_data, cpu).nxtlist || |
f41d911f PM |
1782 | per_cpu(rcu_bh_data, cpu).nxtlist || |
1783 | rcu_preempt_needs_cpu(cpu); | |
64db4cff PM |
1784 | } |
1785 | ||
d0ec774c PM |
1786 | static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL}; |
1787 | static atomic_t rcu_barrier_cpu_count; | |
1788 | static DEFINE_MUTEX(rcu_barrier_mutex); | |
1789 | static struct completion rcu_barrier_completion; | |
d0ec774c PM |
1790 | |
1791 | static void rcu_barrier_callback(struct rcu_head *notused) | |
1792 | { | |
1793 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
1794 | complete(&rcu_barrier_completion); | |
1795 | } | |
1796 | ||
1797 | /* | |
1798 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
1799 | */ | |
1800 | static void rcu_barrier_func(void *type) | |
1801 | { | |
1802 | int cpu = smp_processor_id(); | |
1803 | struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu); | |
1804 | void (*call_rcu_func)(struct rcu_head *head, | |
1805 | void (*func)(struct rcu_head *head)); | |
1806 | ||
1807 | atomic_inc(&rcu_barrier_cpu_count); | |
1808 | call_rcu_func = type; | |
1809 | call_rcu_func(head, rcu_barrier_callback); | |
1810 | } | |
1811 | ||
d0ec774c PM |
1812 | /* |
1813 | * Orchestrate the specified type of RCU barrier, waiting for all | |
1814 | * RCU callbacks of the specified type to complete. | |
1815 | */ | |
e74f4c45 PM |
1816 | static void _rcu_barrier(struct rcu_state *rsp, |
1817 | void (*call_rcu_func)(struct rcu_head *head, | |
d0ec774c PM |
1818 | void (*func)(struct rcu_head *head))) |
1819 | { | |
1820 | BUG_ON(in_interrupt()); | |
e74f4c45 | 1821 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
d0ec774c PM |
1822 | mutex_lock(&rcu_barrier_mutex); |
1823 | init_completion(&rcu_barrier_completion); | |
1824 | /* | |
1825 | * Initialize rcu_barrier_cpu_count to 1, then invoke | |
1826 | * rcu_barrier_func() on each CPU, so that each CPU also has | |
1827 | * incremented rcu_barrier_cpu_count. Only then is it safe to | |
1828 | * decrement rcu_barrier_cpu_count -- otherwise the first CPU | |
1829 | * might complete its grace period before all of the other CPUs | |
1830 | * did their increment, causing this function to return too | |
2d999e03 PM |
1831 | * early. Note that on_each_cpu() disables irqs, which prevents |
1832 | * any CPUs from coming online or going offline until each online | |
1833 | * CPU has queued its RCU-barrier callback. | |
d0ec774c PM |
1834 | */ |
1835 | atomic_set(&rcu_barrier_cpu_count, 1); | |
1836 | on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1); | |
1837 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
1838 | complete(&rcu_barrier_completion); | |
1839 | wait_for_completion(&rcu_barrier_completion); | |
1840 | mutex_unlock(&rcu_barrier_mutex); | |
d0ec774c | 1841 | } |
d0ec774c PM |
1842 | |
1843 | /** | |
1844 | * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete. | |
1845 | */ | |
1846 | void rcu_barrier_bh(void) | |
1847 | { | |
e74f4c45 | 1848 | _rcu_barrier(&rcu_bh_state, call_rcu_bh); |
d0ec774c PM |
1849 | } |
1850 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); | |
1851 | ||
1852 | /** | |
1853 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. | |
1854 | */ | |
1855 | void rcu_barrier_sched(void) | |
1856 | { | |
e74f4c45 | 1857 | _rcu_barrier(&rcu_sched_state, call_rcu_sched); |
d0ec774c PM |
1858 | } |
1859 | EXPORT_SYMBOL_GPL(rcu_barrier_sched); | |
1860 | ||
64db4cff | 1861 | /* |
27569620 | 1862 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 1863 | */ |
27569620 PM |
1864 | static void __init |
1865 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) | |
64db4cff PM |
1866 | { |
1867 | unsigned long flags; | |
1868 | int i; | |
394f99a9 | 1869 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
27569620 PM |
1870 | struct rcu_node *rnp = rcu_get_root(rsp); |
1871 | ||
1872 | /* Set up local state, ensuring consistent view of global state. */ | |
1304afb2 | 1873 | raw_spin_lock_irqsave(&rnp->lock, flags); |
27569620 PM |
1874 | rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo); |
1875 | rdp->nxtlist = NULL; | |
1876 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
1877 | rdp->nxttail[i] = &rdp->nxtlist; | |
1878 | rdp->qlen = 0; | |
1879 | #ifdef CONFIG_NO_HZ | |
1880 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); | |
1881 | #endif /* #ifdef CONFIG_NO_HZ */ | |
1882 | rdp->cpu = cpu; | |
d4c08f2a | 1883 | rdp->rsp = rsp; |
1304afb2 | 1884 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
27569620 PM |
1885 | } |
1886 | ||
1887 | /* | |
1888 | * Initialize a CPU's per-CPU RCU data. Note that only one online or | |
1889 | * offline event can be happening at a given time. Note also that we | |
1890 | * can accept some slop in the rsp->completed access due to the fact | |
1891 | * that this CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 1892 | */ |
e4fa4c97 | 1893 | static void __cpuinit |
6cc68793 | 1894 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptible) |
64db4cff PM |
1895 | { |
1896 | unsigned long flags; | |
64db4cff | 1897 | unsigned long mask; |
394f99a9 | 1898 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff PM |
1899 | struct rcu_node *rnp = rcu_get_root(rsp); |
1900 | ||
1901 | /* Set up local state, ensuring consistent view of global state. */ | |
1304afb2 | 1902 | raw_spin_lock_irqsave(&rnp->lock, flags); |
e4cc1f22 | 1903 | rdp->passed_quiesce = 0; /* We could be racing with new GP, */ |
64db4cff PM |
1904 | rdp->qs_pending = 1; /* so set up to respond to current GP. */ |
1905 | rdp->beenonline = 1; /* We have now been online. */ | |
6cc68793 | 1906 | rdp->preemptible = preemptible; |
37c72e56 PM |
1907 | rdp->qlen_last_fqs_check = 0; |
1908 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
64db4cff | 1909 | rdp->blimit = blimit; |
1304afb2 | 1910 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
1911 | |
1912 | /* | |
1913 | * A new grace period might start here. If so, we won't be part | |
1914 | * of it, but that is OK, as we are currently in a quiescent state. | |
1915 | */ | |
1916 | ||
1917 | /* Exclude any attempts to start a new GP on large systems. */ | |
1304afb2 | 1918 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
64db4cff PM |
1919 | |
1920 | /* Add CPU to rcu_node bitmasks. */ | |
1921 | rnp = rdp->mynode; | |
1922 | mask = rdp->grpmask; | |
1923 | do { | |
1924 | /* Exclude any attempts to start a new GP on small systems. */ | |
1304afb2 | 1925 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
1926 | rnp->qsmaskinit |= mask; |
1927 | mask = rnp->grpmask; | |
d09b62df PM |
1928 | if (rnp == rdp->mynode) { |
1929 | rdp->gpnum = rnp->completed; /* if GP in progress... */ | |
1930 | rdp->completed = rnp->completed; | |
e4cc1f22 | 1931 | rdp->passed_quiesce_gpnum = rnp->gpnum - 1; |
d4c08f2a | 1932 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuonl"); |
d09b62df | 1933 | } |
1304afb2 | 1934 | raw_spin_unlock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
1935 | rnp = rnp->parent; |
1936 | } while (rnp != NULL && !(rnp->qsmaskinit & mask)); | |
1937 | ||
1304afb2 | 1938 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); |
64db4cff PM |
1939 | } |
1940 | ||
d72bce0e | 1941 | static void __cpuinit rcu_prepare_cpu(int cpu) |
64db4cff | 1942 | { |
f41d911f PM |
1943 | rcu_init_percpu_data(cpu, &rcu_sched_state, 0); |
1944 | rcu_init_percpu_data(cpu, &rcu_bh_state, 0); | |
1945 | rcu_preempt_init_percpu_data(cpu); | |
64db4cff PM |
1946 | } |
1947 | ||
1948 | /* | |
f41d911f | 1949 | * Handle CPU online/offline notification events. |
64db4cff | 1950 | */ |
9f680ab4 PM |
1951 | static int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
1952 | unsigned long action, void *hcpu) | |
64db4cff PM |
1953 | { |
1954 | long cpu = (long)hcpu; | |
27f4d280 | 1955 | struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu); |
a26ac245 | 1956 | struct rcu_node *rnp = rdp->mynode; |
64db4cff | 1957 | |
300df91c | 1958 | trace_rcu_utilization("Start CPU hotplug"); |
64db4cff PM |
1959 | switch (action) { |
1960 | case CPU_UP_PREPARE: | |
1961 | case CPU_UP_PREPARE_FROZEN: | |
d72bce0e PZ |
1962 | rcu_prepare_cpu(cpu); |
1963 | rcu_prepare_kthreads(cpu); | |
a26ac245 PM |
1964 | break; |
1965 | case CPU_ONLINE: | |
0f962a5e PM |
1966 | case CPU_DOWN_FAILED: |
1967 | rcu_node_kthread_setaffinity(rnp, -1); | |
e3995a25 | 1968 | rcu_cpu_kthread_setrt(cpu, 1); |
0f962a5e PM |
1969 | break; |
1970 | case CPU_DOWN_PREPARE: | |
1971 | rcu_node_kthread_setaffinity(rnp, cpu); | |
e3995a25 | 1972 | rcu_cpu_kthread_setrt(cpu, 0); |
64db4cff | 1973 | break; |
d0ec774c PM |
1974 | case CPU_DYING: |
1975 | case CPU_DYING_FROZEN: | |
1976 | /* | |
2d999e03 PM |
1977 | * The whole machine is "stopped" except this CPU, so we can |
1978 | * touch any data without introducing corruption. We send the | |
1979 | * dying CPU's callbacks to an arbitrarily chosen online CPU. | |
d0ec774c | 1980 | */ |
29494be7 LJ |
1981 | rcu_send_cbs_to_online(&rcu_bh_state); |
1982 | rcu_send_cbs_to_online(&rcu_sched_state); | |
1983 | rcu_preempt_send_cbs_to_online(); | |
d0ec774c | 1984 | break; |
64db4cff PM |
1985 | case CPU_DEAD: |
1986 | case CPU_DEAD_FROZEN: | |
1987 | case CPU_UP_CANCELED: | |
1988 | case CPU_UP_CANCELED_FROZEN: | |
1989 | rcu_offline_cpu(cpu); | |
1990 | break; | |
1991 | default: | |
1992 | break; | |
1993 | } | |
300df91c | 1994 | trace_rcu_utilization("End CPU hotplug"); |
64db4cff PM |
1995 | return NOTIFY_OK; |
1996 | } | |
1997 | ||
bbad9379 PM |
1998 | /* |
1999 | * This function is invoked towards the end of the scheduler's initialization | |
2000 | * process. Before this is called, the idle task might contain | |
2001 | * RCU read-side critical sections (during which time, this idle | |
2002 | * task is booting the system). After this function is called, the | |
2003 | * idle tasks are prohibited from containing RCU read-side critical | |
2004 | * sections. This function also enables RCU lockdep checking. | |
2005 | */ | |
2006 | void rcu_scheduler_starting(void) | |
2007 | { | |
2008 | WARN_ON(num_online_cpus() != 1); | |
2009 | WARN_ON(nr_context_switches() > 0); | |
2010 | rcu_scheduler_active = 1; | |
2011 | } | |
2012 | ||
64db4cff PM |
2013 | /* |
2014 | * Compute the per-level fanout, either using the exact fanout specified | |
2015 | * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT. | |
2016 | */ | |
2017 | #ifdef CONFIG_RCU_FANOUT_EXACT | |
2018 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
2019 | { | |
2020 | int i; | |
2021 | ||
0209f649 | 2022 | for (i = NUM_RCU_LVLS - 1; i > 0; i--) |
64db4cff | 2023 | rsp->levelspread[i] = CONFIG_RCU_FANOUT; |
0209f649 | 2024 | rsp->levelspread[0] = RCU_FANOUT_LEAF; |
64db4cff PM |
2025 | } |
2026 | #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
2027 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
2028 | { | |
2029 | int ccur; | |
2030 | int cprv; | |
2031 | int i; | |
2032 | ||
2033 | cprv = NR_CPUS; | |
2034 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
2035 | ccur = rsp->levelcnt[i]; | |
2036 | rsp->levelspread[i] = (cprv + ccur - 1) / ccur; | |
2037 | cprv = ccur; | |
2038 | } | |
2039 | } | |
2040 | #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
2041 | ||
2042 | /* | |
2043 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
2044 | */ | |
394f99a9 LJ |
2045 | static void __init rcu_init_one(struct rcu_state *rsp, |
2046 | struct rcu_data __percpu *rda) | |
64db4cff | 2047 | { |
b6407e86 PM |
2048 | static char *buf[] = { "rcu_node_level_0", |
2049 | "rcu_node_level_1", | |
2050 | "rcu_node_level_2", | |
2051 | "rcu_node_level_3" }; /* Match MAX_RCU_LVLS */ | |
64db4cff PM |
2052 | int cpustride = 1; |
2053 | int i; | |
2054 | int j; | |
2055 | struct rcu_node *rnp; | |
2056 | ||
b6407e86 PM |
2057 | BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
2058 | ||
64db4cff PM |
2059 | /* Initialize the level-tracking arrays. */ |
2060 | ||
2061 | for (i = 1; i < NUM_RCU_LVLS; i++) | |
2062 | rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1]; | |
2063 | rcu_init_levelspread(rsp); | |
2064 | ||
2065 | /* Initialize the elements themselves, starting from the leaves. */ | |
2066 | ||
2067 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
2068 | cpustride *= rsp->levelspread[i]; | |
2069 | rnp = rsp->level[i]; | |
2070 | for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { | |
1304afb2 | 2071 | raw_spin_lock_init(&rnp->lock); |
b6407e86 PM |
2072 | lockdep_set_class_and_name(&rnp->lock, |
2073 | &rcu_node_class[i], buf[i]); | |
f41d911f | 2074 | rnp->gpnum = 0; |
64db4cff PM |
2075 | rnp->qsmask = 0; |
2076 | rnp->qsmaskinit = 0; | |
2077 | rnp->grplo = j * cpustride; | |
2078 | rnp->grphi = (j + 1) * cpustride - 1; | |
2079 | if (rnp->grphi >= NR_CPUS) | |
2080 | rnp->grphi = NR_CPUS - 1; | |
2081 | if (i == 0) { | |
2082 | rnp->grpnum = 0; | |
2083 | rnp->grpmask = 0; | |
2084 | rnp->parent = NULL; | |
2085 | } else { | |
2086 | rnp->grpnum = j % rsp->levelspread[i - 1]; | |
2087 | rnp->grpmask = 1UL << rnp->grpnum; | |
2088 | rnp->parent = rsp->level[i - 1] + | |
2089 | j / rsp->levelspread[i - 1]; | |
2090 | } | |
2091 | rnp->level = i; | |
12f5f524 | 2092 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
64db4cff PM |
2093 | } |
2094 | } | |
0c34029a | 2095 | |
394f99a9 | 2096 | rsp->rda = rda; |
0c34029a LJ |
2097 | rnp = rsp->level[NUM_RCU_LVLS - 1]; |
2098 | for_each_possible_cpu(i) { | |
4a90a068 | 2099 | while (i > rnp->grphi) |
0c34029a | 2100 | rnp++; |
394f99a9 | 2101 | per_cpu_ptr(rsp->rda, i)->mynode = rnp; |
0c34029a LJ |
2102 | rcu_boot_init_percpu_data(i, rsp); |
2103 | } | |
64db4cff PM |
2104 | } |
2105 | ||
9f680ab4 | 2106 | void __init rcu_init(void) |
64db4cff | 2107 | { |
017c4261 | 2108 | int cpu; |
9f680ab4 | 2109 | |
f41d911f | 2110 | rcu_bootup_announce(); |
394f99a9 LJ |
2111 | rcu_init_one(&rcu_sched_state, &rcu_sched_data); |
2112 | rcu_init_one(&rcu_bh_state, &rcu_bh_data); | |
f41d911f | 2113 | __rcu_init_preempt(); |
09223371 | 2114 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
9f680ab4 PM |
2115 | |
2116 | /* | |
2117 | * We don't need protection against CPU-hotplug here because | |
2118 | * this is called early in boot, before either interrupts | |
2119 | * or the scheduler are operational. | |
2120 | */ | |
2121 | cpu_notifier(rcu_cpu_notify, 0); | |
017c4261 PM |
2122 | for_each_online_cpu(cpu) |
2123 | rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu); | |
c68de209 | 2124 | check_cpu_stall_init(); |
64db4cff PM |
2125 | } |
2126 | ||
1eba8f84 | 2127 | #include "rcutree_plugin.h" |