]>
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 | 40 | #include <linux/bitops.h> |
9984de1a | 41 | #include <linux/export.h> |
64db4cff PM |
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> |
3d3b7db0 PM |
53 | #include <linux/delay.h> |
54 | #include <linux/stop_machine.h> | |
64db4cff | 55 | |
9f77da9f | 56 | #include "rcutree.h" |
29c00b4a PM |
57 | #include <trace/events/rcu.h> |
58 | ||
59 | #include "rcu.h" | |
9f77da9f | 60 | |
64db4cff PM |
61 | /* Data structures. */ |
62 | ||
b668c9cf | 63 | static struct lock_class_key rcu_node_class[NUM_RCU_LVLS]; |
88b91c7c | 64 | |
4300aa64 | 65 | #define RCU_STATE_INITIALIZER(structname) { \ |
e99033c5 | 66 | .level = { &structname##_state.node[0] }, \ |
64db4cff PM |
67 | .levelcnt = { \ |
68 | NUM_RCU_LVL_0, /* root of hierarchy. */ \ | |
69 | NUM_RCU_LVL_1, \ | |
70 | NUM_RCU_LVL_2, \ | |
cf244dc0 PM |
71 | NUM_RCU_LVL_3, \ |
72 | NUM_RCU_LVL_4, /* == MAX_RCU_LVLS */ \ | |
64db4cff | 73 | }, \ |
af446b70 | 74 | .fqs_state = RCU_GP_IDLE, \ |
64db4cff PM |
75 | .gpnum = -300, \ |
76 | .completed = -300, \ | |
e99033c5 PM |
77 | .onofflock = __RAW_SPIN_LOCK_UNLOCKED(&structname##_state.onofflock), \ |
78 | .fqslock = __RAW_SPIN_LOCK_UNLOCKED(&structname##_state.fqslock), \ | |
64db4cff PM |
79 | .n_force_qs = 0, \ |
80 | .n_force_qs_ngp = 0, \ | |
4300aa64 | 81 | .name = #structname, \ |
64db4cff PM |
82 | } |
83 | ||
e99033c5 | 84 | struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched); |
d6714c22 | 85 | DEFINE_PER_CPU(struct rcu_data, rcu_sched_data); |
64db4cff | 86 | |
e99033c5 | 87 | struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh); |
6258c4fb | 88 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); |
b1f77b05 | 89 | |
27f4d280 PM |
90 | static struct rcu_state *rcu_state; |
91 | ||
b0d30417 PM |
92 | /* |
93 | * The rcu_scheduler_active variable transitions from zero to one just | |
94 | * before the first task is spawned. So when this variable is zero, RCU | |
95 | * can assume that there is but one task, allowing RCU to (for example) | |
96 | * optimized synchronize_sched() to a simple barrier(). When this variable | |
97 | * is one, RCU must actually do all the hard work required to detect real | |
98 | * grace periods. This variable is also used to suppress boot-time false | |
99 | * positives from lockdep-RCU error checking. | |
100 | */ | |
bbad9379 PM |
101 | int rcu_scheduler_active __read_mostly; |
102 | EXPORT_SYMBOL_GPL(rcu_scheduler_active); | |
103 | ||
b0d30417 PM |
104 | /* |
105 | * The rcu_scheduler_fully_active variable transitions from zero to one | |
106 | * during the early_initcall() processing, which is after the scheduler | |
107 | * is capable of creating new tasks. So RCU processing (for example, | |
108 | * creating tasks for RCU priority boosting) must be delayed until after | |
109 | * rcu_scheduler_fully_active transitions from zero to one. We also | |
110 | * currently delay invocation of any RCU callbacks until after this point. | |
111 | * | |
112 | * It might later prove better for people registering RCU callbacks during | |
113 | * early boot to take responsibility for these callbacks, but one step at | |
114 | * a time. | |
115 | */ | |
116 | static int rcu_scheduler_fully_active __read_mostly; | |
117 | ||
a46e0899 PM |
118 | #ifdef CONFIG_RCU_BOOST |
119 | ||
a26ac245 PM |
120 | /* |
121 | * Control variables for per-CPU and per-rcu_node kthreads. These | |
122 | * handle all flavors of RCU. | |
123 | */ | |
124 | static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task); | |
d71df90e | 125 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status); |
15ba0ba8 | 126 | DEFINE_PER_CPU(int, rcu_cpu_kthread_cpu); |
5ece5bab | 127 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops); |
d71df90e | 128 | DEFINE_PER_CPU(char, rcu_cpu_has_work); |
a26ac245 | 129 | |
a46e0899 PM |
130 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
131 | ||
0f962a5e | 132 | static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); |
a46e0899 PM |
133 | static void invoke_rcu_core(void); |
134 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp); | |
a26ac245 | 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 | |
90a4d2c0 | 200 | DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { |
4145fa7f | 201 | .dynticks_nesting = DYNTICK_TASK_NESTING, |
23b5c8fa | 202 | .dynticks = ATOMIC_INIT(1), |
90a4d2c0 | 203 | }; |
64db4cff | 204 | |
e0f23060 | 205 | static int blimit = 10; /* Maximum callbacks per rcu_do_batch. */ |
64db4cff PM |
206 | static int qhimark = 10000; /* If this many pending, ignore blimit. */ |
207 | static int qlowmark = 100; /* Once only this many pending, use blimit. */ | |
208 | ||
3d76c082 PM |
209 | module_param(blimit, int, 0); |
210 | module_param(qhimark, int, 0); | |
211 | module_param(qlowmark, int, 0); | |
212 | ||
13cfcca0 PM |
213 | int rcu_cpu_stall_suppress __read_mostly; /* 1 = suppress stall warnings. */ |
214 | int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT; | |
215 | ||
f2e0dd70 | 216 | module_param(rcu_cpu_stall_suppress, int, 0644); |
13cfcca0 | 217 | module_param(rcu_cpu_stall_timeout, int, 0644); |
742734ee | 218 | |
64db4cff | 219 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed); |
a157229c | 220 | static int rcu_pending(int cpu); |
64db4cff PM |
221 | |
222 | /* | |
d6714c22 | 223 | * Return the number of RCU-sched batches processed thus far for debug & stats. |
64db4cff | 224 | */ |
d6714c22 | 225 | long rcu_batches_completed_sched(void) |
64db4cff | 226 | { |
d6714c22 | 227 | return rcu_sched_state.completed; |
64db4cff | 228 | } |
d6714c22 | 229 | EXPORT_SYMBOL_GPL(rcu_batches_completed_sched); |
64db4cff PM |
230 | |
231 | /* | |
232 | * Return the number of RCU BH batches processed thus far for debug & stats. | |
233 | */ | |
234 | long rcu_batches_completed_bh(void) | |
235 | { | |
236 | return rcu_bh_state.completed; | |
237 | } | |
238 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); | |
239 | ||
bf66f18e PM |
240 | /* |
241 | * Force a quiescent state for RCU BH. | |
242 | */ | |
243 | void rcu_bh_force_quiescent_state(void) | |
244 | { | |
245 | force_quiescent_state(&rcu_bh_state, 0); | |
246 | } | |
247 | EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state); | |
248 | ||
4a298656 PM |
249 | /* |
250 | * Record the number of times rcutorture tests have been initiated and | |
251 | * terminated. This information allows the debugfs tracing stats to be | |
252 | * correlated to the rcutorture messages, even when the rcutorture module | |
253 | * is being repeatedly loaded and unloaded. In other words, we cannot | |
254 | * store this state in rcutorture itself. | |
255 | */ | |
256 | void rcutorture_record_test_transition(void) | |
257 | { | |
258 | rcutorture_testseq++; | |
259 | rcutorture_vernum = 0; | |
260 | } | |
261 | EXPORT_SYMBOL_GPL(rcutorture_record_test_transition); | |
262 | ||
263 | /* | |
264 | * Record the number of writer passes through the current rcutorture test. | |
265 | * This is also used to correlate debugfs tracing stats with the rcutorture | |
266 | * messages. | |
267 | */ | |
268 | void rcutorture_record_progress(unsigned long vernum) | |
269 | { | |
270 | rcutorture_vernum++; | |
271 | } | |
272 | EXPORT_SYMBOL_GPL(rcutorture_record_progress); | |
273 | ||
bf66f18e PM |
274 | /* |
275 | * Force a quiescent state for RCU-sched. | |
276 | */ | |
277 | void rcu_sched_force_quiescent_state(void) | |
278 | { | |
279 | force_quiescent_state(&rcu_sched_state, 0); | |
280 | } | |
281 | EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state); | |
282 | ||
64db4cff PM |
283 | /* |
284 | * Does the CPU have callbacks ready to be invoked? | |
285 | */ | |
286 | static int | |
287 | cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) | |
288 | { | |
289 | return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]; | |
290 | } | |
291 | ||
292 | /* | |
293 | * Does the current CPU require a yet-as-unscheduled grace period? | |
294 | */ | |
295 | static int | |
296 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) | |
297 | { | |
fc2219d4 | 298 | return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp); |
64db4cff PM |
299 | } |
300 | ||
301 | /* | |
302 | * Return the root node of the specified rcu_state structure. | |
303 | */ | |
304 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) | |
305 | { | |
306 | return &rsp->node[0]; | |
307 | } | |
308 | ||
64db4cff PM |
309 | /* |
310 | * If the specified CPU is offline, tell the caller that it is in | |
311 | * a quiescent state. Otherwise, whack it with a reschedule IPI. | |
312 | * Grace periods can end up waiting on an offline CPU when that | |
313 | * CPU is in the process of coming online -- it will be added to the | |
314 | * rcu_node bitmasks before it actually makes it online. The same thing | |
315 | * can happen while a CPU is in the process of coming online. Because this | |
316 | * race is quite rare, we check for it after detecting that the grace | |
317 | * period has been delayed rather than checking each and every CPU | |
318 | * each and every time we start a new grace period. | |
319 | */ | |
320 | static int rcu_implicit_offline_qs(struct rcu_data *rdp) | |
321 | { | |
322 | /* | |
323 | * If the CPU is offline, it is in a quiescent state. We can | |
324 | * trust its state not to change because interrupts are disabled. | |
325 | */ | |
326 | if (cpu_is_offline(rdp->cpu)) { | |
d4c08f2a | 327 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, "ofl"); |
64db4cff PM |
328 | rdp->offline_fqs++; |
329 | return 1; | |
330 | } | |
331 | ||
9b2e4f18 PM |
332 | /* |
333 | * The CPU is online, so send it a reschedule IPI. This forces | |
334 | * it through the scheduler, and (inefficiently) also handles cases | |
335 | * where idle loops fail to inform RCU about the CPU being idle. | |
336 | */ | |
64db4cff PM |
337 | if (rdp->cpu != smp_processor_id()) |
338 | smp_send_reschedule(rdp->cpu); | |
339 | else | |
340 | set_need_resched(); | |
341 | rdp->resched_ipi++; | |
342 | return 0; | |
343 | } | |
344 | ||
9b2e4f18 PM |
345 | /* |
346 | * rcu_idle_enter_common - inform RCU that current CPU is moving towards idle | |
347 | * | |
348 | * If the new value of the ->dynticks_nesting counter now is zero, | |
349 | * we really have entered idle, and must do the appropriate accounting. | |
350 | * The caller must have disabled interrupts. | |
351 | */ | |
4145fa7f | 352 | static void rcu_idle_enter_common(struct rcu_dynticks *rdtp, long long oldval) |
9b2e4f18 | 353 | { |
facc4e15 | 354 | trace_rcu_dyntick("Start", oldval, 0); |
99745b6a | 355 | if (!is_idle_task(current)) { |
0989cb46 PM |
356 | struct task_struct *idle = idle_task(smp_processor_id()); |
357 | ||
facc4e15 | 358 | trace_rcu_dyntick("Error on entry: not idle task", oldval, 0); |
9b2e4f18 | 359 | ftrace_dump(DUMP_ALL); |
0989cb46 PM |
360 | WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s", |
361 | current->pid, current->comm, | |
362 | idle->pid, idle->comm); /* must be idle task! */ | |
9b2e4f18 | 363 | } |
aea1b35e | 364 | rcu_prepare_for_idle(smp_processor_id()); |
9b2e4f18 PM |
365 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ |
366 | smp_mb__before_atomic_inc(); /* See above. */ | |
367 | atomic_inc(&rdtp->dynticks); | |
368 | smp_mb__after_atomic_inc(); /* Force ordering with next sojourn. */ | |
369 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); | |
c44e2cdd PM |
370 | |
371 | /* | |
372 | * The idle task is not permitted to enter the idle loop while | |
373 | * in an RCU read-side critical section. | |
374 | */ | |
375 | rcu_lockdep_assert(!lock_is_held(&rcu_lock_map), | |
376 | "Illegal idle entry in RCU read-side critical section."); | |
377 | rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map), | |
378 | "Illegal idle entry in RCU-bh read-side critical section."); | |
379 | rcu_lockdep_assert(!lock_is_held(&rcu_sched_lock_map), | |
380 | "Illegal idle entry in RCU-sched read-side critical section."); | |
9b2e4f18 | 381 | } |
64db4cff PM |
382 | |
383 | /** | |
9b2e4f18 | 384 | * rcu_idle_enter - inform RCU that current CPU is entering idle |
64db4cff | 385 | * |
9b2e4f18 | 386 | * Enter idle mode, in other words, -leave- the mode in which RCU |
64db4cff | 387 | * read-side critical sections can occur. (Though RCU read-side |
9b2e4f18 PM |
388 | * critical sections can occur in irq handlers in idle, a possibility |
389 | * handled by irq_enter() and irq_exit().) | |
390 | * | |
391 | * We crowbar the ->dynticks_nesting field to zero to allow for | |
392 | * the possibility of usermode upcalls having messed up our count | |
393 | * of interrupt nesting level during the prior busy period. | |
64db4cff | 394 | */ |
9b2e4f18 | 395 | void rcu_idle_enter(void) |
64db4cff PM |
396 | { |
397 | unsigned long flags; | |
4145fa7f | 398 | long long oldval; |
64db4cff PM |
399 | struct rcu_dynticks *rdtp; |
400 | ||
64db4cff PM |
401 | local_irq_save(flags); |
402 | rdtp = &__get_cpu_var(rcu_dynticks); | |
4145fa7f | 403 | oldval = rdtp->dynticks_nesting; |
9b2e4f18 | 404 | rdtp->dynticks_nesting = 0; |
4145fa7f | 405 | rcu_idle_enter_common(rdtp, oldval); |
64db4cff PM |
406 | local_irq_restore(flags); |
407 | } | |
408 | ||
9b2e4f18 PM |
409 | /** |
410 | * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle | |
411 | * | |
412 | * Exit from an interrupt handler, which might possibly result in entering | |
413 | * idle mode, in other words, leaving the mode in which read-side critical | |
414 | * sections can occur. | |
64db4cff | 415 | * |
9b2e4f18 PM |
416 | * This code assumes that the idle loop never does anything that might |
417 | * result in unbalanced calls to irq_enter() and irq_exit(). If your | |
418 | * architecture violates this assumption, RCU will give you what you | |
419 | * deserve, good and hard. But very infrequently and irreproducibly. | |
420 | * | |
421 | * Use things like work queues to work around this limitation. | |
422 | * | |
423 | * You have been warned. | |
64db4cff | 424 | */ |
9b2e4f18 | 425 | void rcu_irq_exit(void) |
64db4cff PM |
426 | { |
427 | unsigned long flags; | |
4145fa7f | 428 | long long oldval; |
64db4cff PM |
429 | struct rcu_dynticks *rdtp; |
430 | ||
431 | local_irq_save(flags); | |
432 | rdtp = &__get_cpu_var(rcu_dynticks); | |
4145fa7f | 433 | oldval = rdtp->dynticks_nesting; |
9b2e4f18 PM |
434 | rdtp->dynticks_nesting--; |
435 | WARN_ON_ONCE(rdtp->dynticks_nesting < 0); | |
b6fc6020 FW |
436 | if (rdtp->dynticks_nesting) |
437 | trace_rcu_dyntick("--=", oldval, rdtp->dynticks_nesting); | |
438 | else | |
439 | rcu_idle_enter_common(rdtp, oldval); | |
9b2e4f18 PM |
440 | local_irq_restore(flags); |
441 | } | |
442 | ||
443 | /* | |
444 | * rcu_idle_exit_common - inform RCU that current CPU is moving away from idle | |
445 | * | |
446 | * If the new value of the ->dynticks_nesting counter was previously zero, | |
447 | * we really have exited idle, and must do the appropriate accounting. | |
448 | * The caller must have disabled interrupts. | |
449 | */ | |
450 | static void rcu_idle_exit_common(struct rcu_dynticks *rdtp, long long oldval) | |
451 | { | |
23b5c8fa PM |
452 | smp_mb__before_atomic_inc(); /* Force ordering w/previous sojourn. */ |
453 | atomic_inc(&rdtp->dynticks); | |
454 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ | |
455 | smp_mb__after_atomic_inc(); /* See above. */ | |
456 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); | |
7cb92499 | 457 | rcu_cleanup_after_idle(smp_processor_id()); |
4145fa7f | 458 | trace_rcu_dyntick("End", oldval, rdtp->dynticks_nesting); |
99745b6a | 459 | if (!is_idle_task(current)) { |
0989cb46 PM |
460 | struct task_struct *idle = idle_task(smp_processor_id()); |
461 | ||
4145fa7f PM |
462 | trace_rcu_dyntick("Error on exit: not idle task", |
463 | oldval, rdtp->dynticks_nesting); | |
9b2e4f18 | 464 | ftrace_dump(DUMP_ALL); |
0989cb46 PM |
465 | WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s", |
466 | current->pid, current->comm, | |
467 | idle->pid, idle->comm); /* must be idle task! */ | |
9b2e4f18 PM |
468 | } |
469 | } | |
470 | ||
471 | /** | |
472 | * rcu_idle_exit - inform RCU that current CPU is leaving idle | |
473 | * | |
474 | * Exit idle mode, in other words, -enter- the mode in which RCU | |
475 | * read-side critical sections can occur. | |
476 | * | |
4145fa7f PM |
477 | * We crowbar the ->dynticks_nesting field to DYNTICK_TASK_NESTING to |
478 | * allow for the possibility of usermode upcalls messing up our count | |
9b2e4f18 PM |
479 | * of interrupt nesting level during the busy period that is just |
480 | * now starting. | |
481 | */ | |
482 | void rcu_idle_exit(void) | |
483 | { | |
484 | unsigned long flags; | |
485 | struct rcu_dynticks *rdtp; | |
486 | long long oldval; | |
487 | ||
488 | local_irq_save(flags); | |
489 | rdtp = &__get_cpu_var(rcu_dynticks); | |
490 | oldval = rdtp->dynticks_nesting; | |
491 | WARN_ON_ONCE(oldval != 0); | |
4145fa7f | 492 | rdtp->dynticks_nesting = DYNTICK_TASK_NESTING; |
9b2e4f18 PM |
493 | rcu_idle_exit_common(rdtp, oldval); |
494 | local_irq_restore(flags); | |
495 | } | |
496 | ||
497 | /** | |
498 | * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle | |
499 | * | |
500 | * Enter an interrupt handler, which might possibly result in exiting | |
501 | * idle mode, in other words, entering the mode in which read-side critical | |
502 | * sections can occur. | |
503 | * | |
504 | * Note that the Linux kernel is fully capable of entering an interrupt | |
505 | * handler that it never exits, for example when doing upcalls to | |
506 | * user mode! This code assumes that the idle loop never does upcalls to | |
507 | * user mode. If your architecture does do upcalls from the idle loop (or | |
508 | * does anything else that results in unbalanced calls to the irq_enter() | |
509 | * and irq_exit() functions), RCU will give you what you deserve, good | |
510 | * and hard. But very infrequently and irreproducibly. | |
511 | * | |
512 | * Use things like work queues to work around this limitation. | |
513 | * | |
514 | * You have been warned. | |
515 | */ | |
516 | void rcu_irq_enter(void) | |
517 | { | |
518 | unsigned long flags; | |
519 | struct rcu_dynticks *rdtp; | |
520 | long long oldval; | |
521 | ||
522 | local_irq_save(flags); | |
523 | rdtp = &__get_cpu_var(rcu_dynticks); | |
524 | oldval = rdtp->dynticks_nesting; | |
525 | rdtp->dynticks_nesting++; | |
526 | WARN_ON_ONCE(rdtp->dynticks_nesting == 0); | |
b6fc6020 FW |
527 | if (oldval) |
528 | trace_rcu_dyntick("++=", oldval, rdtp->dynticks_nesting); | |
529 | else | |
530 | rcu_idle_exit_common(rdtp, oldval); | |
64db4cff | 531 | local_irq_restore(flags); |
64db4cff PM |
532 | } |
533 | ||
534 | /** | |
535 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
536 | * | |
537 | * If the CPU was idle with dynamic ticks active, and there is no | |
538 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
539 | * RCU grace-period handling know that the CPU is active. | |
540 | */ | |
541 | void rcu_nmi_enter(void) | |
542 | { | |
543 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
544 | ||
23b5c8fa PM |
545 | if (rdtp->dynticks_nmi_nesting == 0 && |
546 | (atomic_read(&rdtp->dynticks) & 0x1)) | |
64db4cff | 547 | return; |
23b5c8fa PM |
548 | rdtp->dynticks_nmi_nesting++; |
549 | smp_mb__before_atomic_inc(); /* Force delay from prior write. */ | |
550 | atomic_inc(&rdtp->dynticks); | |
551 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ | |
552 | smp_mb__after_atomic_inc(); /* See above. */ | |
553 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); | |
64db4cff PM |
554 | } |
555 | ||
556 | /** | |
557 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
558 | * | |
559 | * If the CPU was idle with dynamic ticks active, and there is no | |
560 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
561 | * RCU grace-period handling know that the CPU is no longer active. | |
562 | */ | |
563 | void rcu_nmi_exit(void) | |
564 | { | |
565 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
566 | ||
23b5c8fa PM |
567 | if (rdtp->dynticks_nmi_nesting == 0 || |
568 | --rdtp->dynticks_nmi_nesting != 0) | |
64db4cff | 569 | return; |
23b5c8fa PM |
570 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ |
571 | smp_mb__before_atomic_inc(); /* See above. */ | |
572 | atomic_inc(&rdtp->dynticks); | |
573 | smp_mb__after_atomic_inc(); /* Force delay to next write. */ | |
574 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); | |
64db4cff PM |
575 | } |
576 | ||
9b2e4f18 PM |
577 | #ifdef CONFIG_PROVE_RCU |
578 | ||
64db4cff | 579 | /** |
9b2e4f18 | 580 | * rcu_is_cpu_idle - see if RCU thinks that the current CPU is idle |
64db4cff | 581 | * |
9b2e4f18 | 582 | * If the current CPU is in its idle loop and is neither in an interrupt |
34240697 | 583 | * or NMI handler, return true. |
64db4cff | 584 | */ |
9b2e4f18 | 585 | int rcu_is_cpu_idle(void) |
64db4cff | 586 | { |
34240697 PM |
587 | int ret; |
588 | ||
589 | preempt_disable(); | |
590 | ret = (atomic_read(&__get_cpu_var(rcu_dynticks).dynticks) & 0x1) == 0; | |
591 | preempt_enable(); | |
592 | return ret; | |
64db4cff | 593 | } |
e6b80a3b | 594 | EXPORT_SYMBOL(rcu_is_cpu_idle); |
64db4cff | 595 | |
c0d6d01b PM |
596 | #ifdef CONFIG_HOTPLUG_CPU |
597 | ||
598 | /* | |
599 | * Is the current CPU online? Disable preemption to avoid false positives | |
600 | * that could otherwise happen due to the current CPU number being sampled, | |
601 | * this task being preempted, its old CPU being taken offline, resuming | |
602 | * on some other CPU, then determining that its old CPU is now offline. | |
603 | * It is OK to use RCU on an offline processor during initial boot, hence | |
604 | * the check for rcu_scheduler_fully_active. | |
605 | * | |
606 | * Disable checking if in an NMI handler because we cannot safely report | |
607 | * errors from NMI handlers anyway. | |
608 | */ | |
609 | bool rcu_lockdep_current_cpu_online(void) | |
610 | { | |
611 | bool ret; | |
612 | ||
613 | if (in_nmi()) | |
614 | return 1; | |
615 | preempt_disable(); | |
616 | ret = cpu_online(smp_processor_id()) || | |
617 | !rcu_scheduler_fully_active; | |
618 | preempt_enable(); | |
619 | return ret; | |
620 | } | |
621 | EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online); | |
622 | ||
623 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | |
624 | ||
9b2e4f18 PM |
625 | #endif /* #ifdef CONFIG_PROVE_RCU */ |
626 | ||
64db4cff | 627 | /** |
9b2e4f18 | 628 | * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle |
64db4cff | 629 | * |
9b2e4f18 PM |
630 | * If the current CPU is idle or running at a first-level (not nested) |
631 | * interrupt from idle, return true. The caller must have at least | |
632 | * disabled preemption. | |
64db4cff | 633 | */ |
9b2e4f18 | 634 | int rcu_is_cpu_rrupt_from_idle(void) |
64db4cff | 635 | { |
9b2e4f18 | 636 | return __get_cpu_var(rcu_dynticks).dynticks_nesting <= 1; |
64db4cff PM |
637 | } |
638 | ||
64db4cff PM |
639 | /* |
640 | * Snapshot the specified CPU's dynticks counter so that we can later | |
641 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 642 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff PM |
643 | */ |
644 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | |
645 | { | |
23b5c8fa | 646 | rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks); |
f0e7c19d | 647 | return (rdp->dynticks_snap & 0x1) == 0; |
64db4cff PM |
648 | } |
649 | ||
650 | /* | |
651 | * Return true if the specified CPU has passed through a quiescent | |
652 | * state by virtue of being in or having passed through an dynticks | |
653 | * idle state since the last call to dyntick_save_progress_counter() | |
654 | * for this same CPU. | |
655 | */ | |
656 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
657 | { | |
7eb4f455 PM |
658 | unsigned int curr; |
659 | unsigned int snap; | |
64db4cff | 660 | |
7eb4f455 PM |
661 | curr = (unsigned int)atomic_add_return(0, &rdp->dynticks->dynticks); |
662 | snap = (unsigned int)rdp->dynticks_snap; | |
64db4cff PM |
663 | |
664 | /* | |
665 | * If the CPU passed through or entered a dynticks idle phase with | |
666 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
667 | * already acknowledged the request to pass through a quiescent | |
668 | * state. Either way, that CPU cannot possibly be in an RCU | |
669 | * read-side critical section that started before the beginning | |
670 | * of the current RCU grace period. | |
671 | */ | |
7eb4f455 | 672 | if ((curr & 0x1) == 0 || UINT_CMP_GE(curr, snap + 2)) { |
d4c08f2a | 673 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, "dti"); |
64db4cff PM |
674 | rdp->dynticks_fqs++; |
675 | return 1; | |
676 | } | |
677 | ||
678 | /* Go check for the CPU being offline. */ | |
679 | return rcu_implicit_offline_qs(rdp); | |
680 | } | |
681 | ||
13cfcca0 PM |
682 | static int jiffies_till_stall_check(void) |
683 | { | |
684 | int till_stall_check = ACCESS_ONCE(rcu_cpu_stall_timeout); | |
685 | ||
686 | /* | |
687 | * Limit check must be consistent with the Kconfig limits | |
688 | * for CONFIG_RCU_CPU_STALL_TIMEOUT. | |
689 | */ | |
690 | if (till_stall_check < 3) { | |
691 | ACCESS_ONCE(rcu_cpu_stall_timeout) = 3; | |
692 | till_stall_check = 3; | |
693 | } else if (till_stall_check > 300) { | |
694 | ACCESS_ONCE(rcu_cpu_stall_timeout) = 300; | |
695 | till_stall_check = 300; | |
696 | } | |
697 | return till_stall_check * HZ + RCU_STALL_DELAY_DELTA; | |
698 | } | |
699 | ||
64db4cff PM |
700 | static void record_gp_stall_check_time(struct rcu_state *rsp) |
701 | { | |
702 | rsp->gp_start = jiffies; | |
13cfcca0 | 703 | rsp->jiffies_stall = jiffies + jiffies_till_stall_check(); |
64db4cff PM |
704 | } |
705 | ||
706 | static void print_other_cpu_stall(struct rcu_state *rsp) | |
707 | { | |
708 | int cpu; | |
709 | long delta; | |
710 | unsigned long flags; | |
9bc8b558 | 711 | int ndetected; |
64db4cff | 712 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff PM |
713 | |
714 | /* Only let one CPU complain about others per time interval. */ | |
715 | ||
1304afb2 | 716 | raw_spin_lock_irqsave(&rnp->lock, flags); |
64db4cff | 717 | delta = jiffies - rsp->jiffies_stall; |
fc2219d4 | 718 | if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) { |
1304afb2 | 719 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
720 | return; |
721 | } | |
13cfcca0 | 722 | rsp->jiffies_stall = jiffies + 3 * jiffies_till_stall_check() + 3; |
1304afb2 | 723 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 724 | |
8cdd32a9 PM |
725 | /* |
726 | * OK, time to rat on our buddy... | |
727 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
728 | * RCU CPU stall warnings. | |
729 | */ | |
a858af28 | 730 | printk(KERN_ERR "INFO: %s detected stalls on CPUs/tasks:", |
4300aa64 | 731 | rsp->name); |
a858af28 | 732 | print_cpu_stall_info_begin(); |
a0b6c9a7 | 733 | rcu_for_each_leaf_node(rsp, rnp) { |
3acd9eb3 | 734 | raw_spin_lock_irqsave(&rnp->lock, flags); |
9bc8b558 | 735 | ndetected += rcu_print_task_stall(rnp); |
3acd9eb3 | 736 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
a0b6c9a7 | 737 | if (rnp->qsmask == 0) |
64db4cff | 738 | continue; |
a0b6c9a7 | 739 | for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++) |
9bc8b558 | 740 | if (rnp->qsmask & (1UL << cpu)) { |
a858af28 | 741 | print_cpu_stall_info(rsp, rnp->grplo + cpu); |
9bc8b558 PM |
742 | ndetected++; |
743 | } | |
64db4cff | 744 | } |
a858af28 PM |
745 | |
746 | /* | |
747 | * Now rat on any tasks that got kicked up to the root rcu_node | |
748 | * due to CPU offlining. | |
749 | */ | |
750 | rnp = rcu_get_root(rsp); | |
751 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
752 | ndetected = rcu_print_task_stall(rnp); | |
753 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
754 | ||
755 | print_cpu_stall_info_end(); | |
756 | printk(KERN_CONT "(detected by %d, t=%ld jiffies)\n", | |
64db4cff | 757 | smp_processor_id(), (long)(jiffies - rsp->gp_start)); |
9bc8b558 PM |
758 | if (ndetected == 0) |
759 | printk(KERN_ERR "INFO: Stall ended before state dump start\n"); | |
760 | else if (!trigger_all_cpu_backtrace()) | |
4627e240 | 761 | dump_stack(); |
c1dc0b9c | 762 | |
1ed509a2 PM |
763 | /* If so configured, complain about tasks blocking the grace period. */ |
764 | ||
765 | rcu_print_detail_task_stall(rsp); | |
766 | ||
64db4cff PM |
767 | force_quiescent_state(rsp, 0); /* Kick them all. */ |
768 | } | |
769 | ||
770 | static void print_cpu_stall(struct rcu_state *rsp) | |
771 | { | |
772 | unsigned long flags; | |
773 | struct rcu_node *rnp = rcu_get_root(rsp); | |
774 | ||
8cdd32a9 PM |
775 | /* |
776 | * OK, time to rat on ourselves... | |
777 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
778 | * RCU CPU stall warnings. | |
779 | */ | |
a858af28 PM |
780 | printk(KERN_ERR "INFO: %s self-detected stall on CPU", rsp->name); |
781 | print_cpu_stall_info_begin(); | |
782 | print_cpu_stall_info(rsp, smp_processor_id()); | |
783 | print_cpu_stall_info_end(); | |
784 | printk(KERN_CONT " (t=%lu jiffies)\n", jiffies - rsp->gp_start); | |
4627e240 PM |
785 | if (!trigger_all_cpu_backtrace()) |
786 | dump_stack(); | |
c1dc0b9c | 787 | |
1304afb2 | 788 | raw_spin_lock_irqsave(&rnp->lock, flags); |
20133cfc | 789 | if (ULONG_CMP_GE(jiffies, rsp->jiffies_stall)) |
13cfcca0 PM |
790 | rsp->jiffies_stall = jiffies + |
791 | 3 * jiffies_till_stall_check() + 3; | |
1304afb2 | 792 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
c1dc0b9c | 793 | |
64db4cff PM |
794 | set_need_resched(); /* kick ourselves to get things going. */ |
795 | } | |
796 | ||
797 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
798 | { | |
bad6e139 PM |
799 | unsigned long j; |
800 | unsigned long js; | |
64db4cff PM |
801 | struct rcu_node *rnp; |
802 | ||
742734ee | 803 | if (rcu_cpu_stall_suppress) |
c68de209 | 804 | return; |
bad6e139 PM |
805 | j = ACCESS_ONCE(jiffies); |
806 | js = ACCESS_ONCE(rsp->jiffies_stall); | |
64db4cff | 807 | rnp = rdp->mynode; |
bad6e139 | 808 | if ((ACCESS_ONCE(rnp->qsmask) & rdp->grpmask) && ULONG_CMP_GE(j, js)) { |
64db4cff PM |
809 | |
810 | /* We haven't checked in, so go dump stack. */ | |
811 | print_cpu_stall(rsp); | |
812 | ||
bad6e139 PM |
813 | } else if (rcu_gp_in_progress(rsp) && |
814 | ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) { | |
64db4cff | 815 | |
bad6e139 | 816 | /* They had a few time units to dump stack, so complain. */ |
64db4cff PM |
817 | print_other_cpu_stall(rsp); |
818 | } | |
819 | } | |
820 | ||
c68de209 PM |
821 | static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr) |
822 | { | |
742734ee | 823 | rcu_cpu_stall_suppress = 1; |
c68de209 PM |
824 | return NOTIFY_DONE; |
825 | } | |
826 | ||
53d84e00 PM |
827 | /** |
828 | * rcu_cpu_stall_reset - prevent further stall warnings in current grace period | |
829 | * | |
830 | * Set the stall-warning timeout way off into the future, thus preventing | |
831 | * any RCU CPU stall-warning messages from appearing in the current set of | |
832 | * RCU grace periods. | |
833 | * | |
834 | * The caller must disable hard irqs. | |
835 | */ | |
836 | void rcu_cpu_stall_reset(void) | |
837 | { | |
838 | rcu_sched_state.jiffies_stall = jiffies + ULONG_MAX / 2; | |
839 | rcu_bh_state.jiffies_stall = jiffies + ULONG_MAX / 2; | |
840 | rcu_preempt_stall_reset(); | |
841 | } | |
842 | ||
c68de209 PM |
843 | static struct notifier_block rcu_panic_block = { |
844 | .notifier_call = rcu_panic, | |
845 | }; | |
846 | ||
847 | static void __init check_cpu_stall_init(void) | |
848 | { | |
849 | atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block); | |
850 | } | |
851 | ||
64db4cff PM |
852 | /* |
853 | * Update CPU-local rcu_data state to record the newly noticed grace period. | |
854 | * This is used both when we started the grace period and when we notice | |
9160306e PM |
855 | * that someone else started the grace period. The caller must hold the |
856 | * ->lock of the leaf rcu_node structure corresponding to the current CPU, | |
857 | * and must have irqs disabled. | |
64db4cff | 858 | */ |
9160306e PM |
859 | static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) |
860 | { | |
861 | if (rdp->gpnum != rnp->gpnum) { | |
121dfc4b PM |
862 | /* |
863 | * If the current grace period is waiting for this CPU, | |
864 | * set up to detect a quiescent state, otherwise don't | |
865 | * go looking for one. | |
866 | */ | |
9160306e | 867 | rdp->gpnum = rnp->gpnum; |
d4c08f2a | 868 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpustart"); |
121dfc4b PM |
869 | if (rnp->qsmask & rdp->grpmask) { |
870 | rdp->qs_pending = 1; | |
e4cc1f22 | 871 | rdp->passed_quiesce = 0; |
121dfc4b PM |
872 | } else |
873 | rdp->qs_pending = 0; | |
a858af28 | 874 | zero_cpu_stall_ticks(rdp); |
9160306e PM |
875 | } |
876 | } | |
877 | ||
64db4cff PM |
878 | static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp) |
879 | { | |
9160306e PM |
880 | unsigned long flags; |
881 | struct rcu_node *rnp; | |
882 | ||
883 | local_irq_save(flags); | |
884 | rnp = rdp->mynode; | |
885 | if (rdp->gpnum == ACCESS_ONCE(rnp->gpnum) || /* outside lock. */ | |
1304afb2 | 886 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
9160306e PM |
887 | local_irq_restore(flags); |
888 | return; | |
889 | } | |
890 | __note_new_gpnum(rsp, rnp, rdp); | |
1304afb2 | 891 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
892 | } |
893 | ||
894 | /* | |
895 | * Did someone else start a new RCU grace period start since we last | |
896 | * checked? Update local state appropriately if so. Must be called | |
897 | * on the CPU corresponding to rdp. | |
898 | */ | |
899 | static int | |
900 | check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp) | |
901 | { | |
902 | unsigned long flags; | |
903 | int ret = 0; | |
904 | ||
905 | local_irq_save(flags); | |
906 | if (rdp->gpnum != rsp->gpnum) { | |
907 | note_new_gpnum(rsp, rdp); | |
908 | ret = 1; | |
909 | } | |
910 | local_irq_restore(flags); | |
911 | return ret; | |
912 | } | |
913 | ||
d09b62df PM |
914 | /* |
915 | * Advance this CPU's callbacks, but only if the current grace period | |
916 | * has ended. This may be called only from the CPU to whom the rdp | |
917 | * belongs. In addition, the corresponding leaf rcu_node structure's | |
918 | * ->lock must be held by the caller, with irqs disabled. | |
919 | */ | |
920 | static void | |
921 | __rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
922 | { | |
923 | /* Did another grace period end? */ | |
924 | if (rdp->completed != rnp->completed) { | |
925 | ||
926 | /* Advance callbacks. No harm if list empty. */ | |
927 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL]; | |
928 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL]; | |
929 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
930 | ||
931 | /* Remember that we saw this grace-period completion. */ | |
932 | rdp->completed = rnp->completed; | |
d4c08f2a | 933 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuend"); |
20377f32 | 934 | |
5ff8e6f0 FW |
935 | /* |
936 | * If we were in an extended quiescent state, we may have | |
121dfc4b | 937 | * missed some grace periods that others CPUs handled on |
5ff8e6f0 | 938 | * our behalf. Catch up with this state to avoid noting |
121dfc4b PM |
939 | * spurious new grace periods. If another grace period |
940 | * has started, then rnp->gpnum will have advanced, so | |
941 | * we will detect this later on. | |
5ff8e6f0 | 942 | */ |
121dfc4b | 943 | if (ULONG_CMP_LT(rdp->gpnum, rdp->completed)) |
5ff8e6f0 FW |
944 | rdp->gpnum = rdp->completed; |
945 | ||
20377f32 | 946 | /* |
121dfc4b PM |
947 | * If RCU does not need a quiescent state from this CPU, |
948 | * then make sure that this CPU doesn't go looking for one. | |
20377f32 | 949 | */ |
121dfc4b | 950 | if ((rnp->qsmask & rdp->grpmask) == 0) |
20377f32 | 951 | rdp->qs_pending = 0; |
d09b62df PM |
952 | } |
953 | } | |
954 | ||
955 | /* | |
956 | * Advance this CPU's callbacks, but only if the current grace period | |
957 | * has ended. This may be called only from the CPU to whom the rdp | |
958 | * belongs. | |
959 | */ | |
960 | static void | |
961 | rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp) | |
962 | { | |
963 | unsigned long flags; | |
964 | struct rcu_node *rnp; | |
965 | ||
966 | local_irq_save(flags); | |
967 | rnp = rdp->mynode; | |
968 | if (rdp->completed == ACCESS_ONCE(rnp->completed) || /* outside lock. */ | |
1304afb2 | 969 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
d09b62df PM |
970 | local_irq_restore(flags); |
971 | return; | |
972 | } | |
973 | __rcu_process_gp_end(rsp, rnp, rdp); | |
1304afb2 | 974 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
d09b62df PM |
975 | } |
976 | ||
977 | /* | |
978 | * Do per-CPU grace-period initialization for running CPU. The caller | |
979 | * must hold the lock of the leaf rcu_node structure corresponding to | |
980 | * this CPU. | |
981 | */ | |
982 | static void | |
983 | rcu_start_gp_per_cpu(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
984 | { | |
985 | /* Prior grace period ended, so advance callbacks for current CPU. */ | |
986 | __rcu_process_gp_end(rsp, rnp, rdp); | |
987 | ||
988 | /* | |
989 | * Because this CPU just now started the new grace period, we know | |
990 | * that all of its callbacks will be covered by this upcoming grace | |
991 | * period, even the ones that were registered arbitrarily recently. | |
992 | * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL. | |
993 | * | |
994 | * Other CPUs cannot be sure exactly when the grace period started. | |
995 | * Therefore, their recently registered callbacks must pass through | |
996 | * an additional RCU_NEXT_READY stage, so that they will be handled | |
997 | * by the next RCU grace period. | |
998 | */ | |
999 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
1000 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
9160306e PM |
1001 | |
1002 | /* Set state so that this CPU will detect the next quiescent state. */ | |
1003 | __note_new_gpnum(rsp, rnp, rdp); | |
d09b62df PM |
1004 | } |
1005 | ||
64db4cff PM |
1006 | /* |
1007 | * Start a new RCU grace period if warranted, re-initializing the hierarchy | |
1008 | * in preparation for detecting the next grace period. The caller must hold | |
1009 | * the root node's ->lock, which is released before return. Hard irqs must | |
1010 | * be disabled. | |
e5601400 PM |
1011 | * |
1012 | * Note that it is legal for a dying CPU (which is marked as offline) to | |
1013 | * invoke this function. This can happen when the dying CPU reports its | |
1014 | * quiescent state. | |
64db4cff PM |
1015 | */ |
1016 | static void | |
1017 | rcu_start_gp(struct rcu_state *rsp, unsigned long flags) | |
1018 | __releases(rcu_get_root(rsp)->lock) | |
1019 | { | |
394f99a9 | 1020 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
64db4cff | 1021 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff | 1022 | |
037067a1 | 1023 | if (!rcu_scheduler_fully_active || |
afe24b12 PM |
1024 | !cpu_needs_another_gp(rsp, rdp)) { |
1025 | /* | |
1026 | * Either the scheduler hasn't yet spawned the first | |
1027 | * non-idle task or this CPU does not need another | |
1028 | * grace period. Either way, don't start a new grace | |
1029 | * period. | |
1030 | */ | |
1031 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
1032 | return; | |
1033 | } | |
b32e9eb6 | 1034 | |
afe24b12 | 1035 | if (rsp->fqs_active) { |
b32e9eb6 | 1036 | /* |
afe24b12 PM |
1037 | * This CPU needs a grace period, but force_quiescent_state() |
1038 | * is running. Tell it to start one on this CPU's behalf. | |
b32e9eb6 | 1039 | */ |
afe24b12 PM |
1040 | rsp->fqs_need_gp = 1; |
1041 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
64db4cff PM |
1042 | return; |
1043 | } | |
1044 | ||
1045 | /* Advance to a new grace period and initialize state. */ | |
1046 | rsp->gpnum++; | |
d4c08f2a | 1047 | trace_rcu_grace_period(rsp->name, rsp->gpnum, "start"); |
af446b70 PM |
1048 | WARN_ON_ONCE(rsp->fqs_state == RCU_GP_INIT); |
1049 | rsp->fqs_state = RCU_GP_INIT; /* Hold off force_quiescent_state. */ | |
64db4cff | 1050 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
64db4cff | 1051 | record_gp_stall_check_time(rsp); |
1304afb2 | 1052 | raw_spin_unlock(&rnp->lock); /* leave irqs disabled. */ |
64db4cff | 1053 | |
64db4cff | 1054 | /* Exclude any concurrent CPU-hotplug operations. */ |
1304afb2 | 1055 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
64db4cff PM |
1056 | |
1057 | /* | |
b835db1f PM |
1058 | * Set the quiescent-state-needed bits in all the rcu_node |
1059 | * structures for all currently online CPUs in breadth-first | |
1060 | * order, starting from the root rcu_node structure. This | |
1061 | * operation relies on the layout of the hierarchy within the | |
1062 | * rsp->node[] array. Note that other CPUs will access only | |
1063 | * the leaves of the hierarchy, which still indicate that no | |
1064 | * grace period is in progress, at least until the corresponding | |
1065 | * leaf node has been initialized. In addition, we have excluded | |
1066 | * CPU-hotplug operations. | |
64db4cff PM |
1067 | * |
1068 | * Note that the grace period cannot complete until we finish | |
1069 | * the initialization process, as there will be at least one | |
1070 | * qsmask bit set in the root node until that time, namely the | |
b835db1f PM |
1071 | * one corresponding to this CPU, due to the fact that we have |
1072 | * irqs disabled. | |
64db4cff | 1073 | */ |
a0b6c9a7 | 1074 | rcu_for_each_node_breadth_first(rsp, rnp) { |
1304afb2 | 1075 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
b0e165c0 | 1076 | rcu_preempt_check_blocked_tasks(rnp); |
49e29126 | 1077 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 1078 | rnp->gpnum = rsp->gpnum; |
d09b62df PM |
1079 | rnp->completed = rsp->completed; |
1080 | if (rnp == rdp->mynode) | |
1081 | rcu_start_gp_per_cpu(rsp, rnp, rdp); | |
27f4d280 | 1082 | rcu_preempt_boost_start_gp(rnp); |
d4c08f2a PM |
1083 | trace_rcu_grace_period_init(rsp->name, rnp->gpnum, |
1084 | rnp->level, rnp->grplo, | |
1085 | rnp->grphi, rnp->qsmask); | |
1304afb2 | 1086 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
1087 | } |
1088 | ||
83f5b01f | 1089 | rnp = rcu_get_root(rsp); |
1304afb2 | 1090 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
af446b70 | 1091 | rsp->fqs_state = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */ |
1304afb2 PM |
1092 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
1093 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); | |
64db4cff PM |
1094 | } |
1095 | ||
f41d911f | 1096 | /* |
d3f6bad3 PM |
1097 | * Report a full set of quiescent states to the specified rcu_state |
1098 | * data structure. This involves cleaning up after the prior grace | |
1099 | * period and letting rcu_start_gp() start up the next grace period | |
1100 | * if one is needed. Note that the caller must hold rnp->lock, as | |
1101 | * required by rcu_start_gp(), which will release it. | |
f41d911f | 1102 | */ |
d3f6bad3 | 1103 | static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags) |
fc2219d4 | 1104 | __releases(rcu_get_root(rsp)->lock) |
f41d911f | 1105 | { |
15ba0ba8 | 1106 | unsigned long gp_duration; |
afe24b12 PM |
1107 | struct rcu_node *rnp = rcu_get_root(rsp); |
1108 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); | |
15ba0ba8 | 1109 | |
fc2219d4 | 1110 | WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); |
0bbcc529 PM |
1111 | |
1112 | /* | |
1113 | * Ensure that all grace-period and pre-grace-period activity | |
1114 | * is seen before the assignment to rsp->completed. | |
1115 | */ | |
1116 | smp_mb(); /* See above block comment. */ | |
15ba0ba8 PM |
1117 | gp_duration = jiffies - rsp->gp_start; |
1118 | if (gp_duration > rsp->gp_max) | |
1119 | rsp->gp_max = gp_duration; | |
afe24b12 PM |
1120 | |
1121 | /* | |
1122 | * We know the grace period is complete, but to everyone else | |
1123 | * it appears to still be ongoing. But it is also the case | |
1124 | * that to everyone else it looks like there is nothing that | |
1125 | * they can do to advance the grace period. It is therefore | |
1126 | * safe for us to drop the lock in order to mark the grace | |
1127 | * period as completed in all of the rcu_node structures. | |
1128 | * | |
1129 | * But if this CPU needs another grace period, it will take | |
1130 | * care of this while initializing the next grace period. | |
1131 | * We use RCU_WAIT_TAIL instead of the usual RCU_DONE_TAIL | |
1132 | * because the callbacks have not yet been advanced: Those | |
1133 | * callbacks are waiting on the grace period that just now | |
1134 | * completed. | |
1135 | */ | |
1136 | if (*rdp->nxttail[RCU_WAIT_TAIL] == NULL) { | |
1137 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ | |
1138 | ||
1139 | /* | |
1140 | * Propagate new ->completed value to rcu_node structures | |
1141 | * so that other CPUs don't have to wait until the start | |
1142 | * of the next grace period to process their callbacks. | |
1143 | */ | |
1144 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
1145 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ | |
1146 | rnp->completed = rsp->gpnum; | |
1147 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ | |
1148 | } | |
1149 | rnp = rcu_get_root(rsp); | |
1150 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ | |
1151 | } | |
1152 | ||
1153 | rsp->completed = rsp->gpnum; /* Declare the grace period complete. */ | |
d4c08f2a | 1154 | trace_rcu_grace_period(rsp->name, rsp->completed, "end"); |
af446b70 | 1155 | rsp->fqs_state = RCU_GP_IDLE; |
f41d911f PM |
1156 | rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */ |
1157 | } | |
1158 | ||
64db4cff | 1159 | /* |
d3f6bad3 PM |
1160 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
1161 | * Allows quiescent states for a group of CPUs to be reported at one go | |
1162 | * to the specified rcu_node structure, though all the CPUs in the group | |
1163 | * must be represented by the same rcu_node structure (which need not be | |
1164 | * a leaf rcu_node structure, though it often will be). That structure's | |
1165 | * lock must be held upon entry, and it is released before return. | |
64db4cff PM |
1166 | */ |
1167 | static void | |
d3f6bad3 PM |
1168 | rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp, |
1169 | struct rcu_node *rnp, unsigned long flags) | |
64db4cff PM |
1170 | __releases(rnp->lock) |
1171 | { | |
28ecd580 PM |
1172 | struct rcu_node *rnp_c; |
1173 | ||
64db4cff PM |
1174 | /* Walk up the rcu_node hierarchy. */ |
1175 | for (;;) { | |
1176 | if (!(rnp->qsmask & mask)) { | |
1177 | ||
1178 | /* Our bit has already been cleared, so done. */ | |
1304afb2 | 1179 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1180 | return; |
1181 | } | |
1182 | rnp->qsmask &= ~mask; | |
d4c08f2a PM |
1183 | trace_rcu_quiescent_state_report(rsp->name, rnp->gpnum, |
1184 | mask, rnp->qsmask, rnp->level, | |
1185 | rnp->grplo, rnp->grphi, | |
1186 | !!rnp->gp_tasks); | |
27f4d280 | 1187 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
64db4cff PM |
1188 | |
1189 | /* Other bits still set at this level, so done. */ | |
1304afb2 | 1190 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1191 | return; |
1192 | } | |
1193 | mask = rnp->grpmask; | |
1194 | if (rnp->parent == NULL) { | |
1195 | ||
1196 | /* No more levels. Exit loop holding root lock. */ | |
1197 | ||
1198 | break; | |
1199 | } | |
1304afb2 | 1200 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
28ecd580 | 1201 | rnp_c = rnp; |
64db4cff | 1202 | rnp = rnp->parent; |
1304afb2 | 1203 | raw_spin_lock_irqsave(&rnp->lock, flags); |
28ecd580 | 1204 | WARN_ON_ONCE(rnp_c->qsmask); |
64db4cff PM |
1205 | } |
1206 | ||
1207 | /* | |
1208 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 1209 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 1210 | * to clean up and start the next grace period if one is needed. |
64db4cff | 1211 | */ |
d3f6bad3 | 1212 | rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */ |
64db4cff PM |
1213 | } |
1214 | ||
1215 | /* | |
d3f6bad3 PM |
1216 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
1217 | * structure. This must be either called from the specified CPU, or | |
1218 | * called when the specified CPU is known to be offline (and when it is | |
1219 | * also known that no other CPU is concurrently trying to help the offline | |
1220 | * CPU). The lastcomp argument is used to make sure we are still in the | |
1221 | * grace period of interest. We don't want to end the current grace period | |
1222 | * based on quiescent states detected in an earlier grace period! | |
64db4cff PM |
1223 | */ |
1224 | static void | |
e4cc1f22 | 1225 | rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastgp) |
64db4cff PM |
1226 | { |
1227 | unsigned long flags; | |
1228 | unsigned long mask; | |
1229 | struct rcu_node *rnp; | |
1230 | ||
1231 | rnp = rdp->mynode; | |
1304afb2 | 1232 | raw_spin_lock_irqsave(&rnp->lock, flags); |
e4cc1f22 | 1233 | if (lastgp != rnp->gpnum || rnp->completed == rnp->gpnum) { |
64db4cff PM |
1234 | |
1235 | /* | |
e4cc1f22 PM |
1236 | * The grace period in which this quiescent state was |
1237 | * recorded has ended, so don't report it upwards. | |
1238 | * We will instead need a new quiescent state that lies | |
1239 | * within the current grace period. | |
64db4cff | 1240 | */ |
e4cc1f22 | 1241 | rdp->passed_quiesce = 0; /* need qs for new gp. */ |
1304afb2 | 1242 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1243 | return; |
1244 | } | |
1245 | mask = rdp->grpmask; | |
1246 | if ((rnp->qsmask & mask) == 0) { | |
1304afb2 | 1247 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1248 | } else { |
1249 | rdp->qs_pending = 0; | |
1250 | ||
1251 | /* | |
1252 | * This GP can't end until cpu checks in, so all of our | |
1253 | * callbacks can be processed during the next GP. | |
1254 | */ | |
64db4cff PM |
1255 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
1256 | ||
d3f6bad3 | 1257 | rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */ |
64db4cff PM |
1258 | } |
1259 | } | |
1260 | ||
1261 | /* | |
1262 | * Check to see if there is a new grace period of which this CPU | |
1263 | * is not yet aware, and if so, set up local rcu_data state for it. | |
1264 | * Otherwise, see if this CPU has just passed through its first | |
1265 | * quiescent state for this grace period, and record that fact if so. | |
1266 | */ | |
1267 | static void | |
1268 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | |
1269 | { | |
1270 | /* If there is now a new grace period, record and return. */ | |
1271 | if (check_for_new_grace_period(rsp, rdp)) | |
1272 | return; | |
1273 | ||
1274 | /* | |
1275 | * Does this CPU still need to do its part for current grace period? | |
1276 | * If no, return and let the other CPUs do their part as well. | |
1277 | */ | |
1278 | if (!rdp->qs_pending) | |
1279 | return; | |
1280 | ||
1281 | /* | |
1282 | * Was there a quiescent state since the beginning of the grace | |
1283 | * period? If no, then exit and wait for the next call. | |
1284 | */ | |
e4cc1f22 | 1285 | if (!rdp->passed_quiesce) |
64db4cff PM |
1286 | return; |
1287 | ||
d3f6bad3 PM |
1288 | /* |
1289 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
1290 | * judge of that). | |
1291 | */ | |
e4cc1f22 | 1292 | rcu_report_qs_rdp(rdp->cpu, rsp, rdp, rdp->passed_quiesce_gpnum); |
64db4cff PM |
1293 | } |
1294 | ||
1295 | #ifdef CONFIG_HOTPLUG_CPU | |
1296 | ||
e74f4c45 | 1297 | /* |
29494be7 | 1298 | * Move a dying CPU's RCU callbacks to online CPU's callback list. |
e5601400 PM |
1299 | * Also record a quiescent state for this CPU for the current grace period. |
1300 | * Synchronization and interrupt disabling are not required because | |
1301 | * this function executes in stop_machine() context. Therefore, cleanup | |
1302 | * operations that might block must be done later from the CPU_DEAD | |
1303 | * notifier. | |
1304 | * | |
1305 | * Note that the outgoing CPU's bit has already been cleared in the | |
1306 | * cpu_online_mask. This allows us to randomly pick a callback | |
1307 | * destination from the bits set in that mask. | |
e74f4c45 | 1308 | */ |
e5601400 | 1309 | static void rcu_cleanup_dying_cpu(struct rcu_state *rsp) |
e74f4c45 | 1310 | { |
e5601400 | 1311 | unsigned long flags; |
e74f4c45 | 1312 | int i; |
e5601400 PM |
1313 | unsigned long mask; |
1314 | int need_report; | |
29494be7 | 1315 | int receive_cpu = cpumask_any(cpu_online_mask); |
394f99a9 | 1316 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
29494be7 | 1317 | struct rcu_data *receive_rdp = per_cpu_ptr(rsp->rda, receive_cpu); |
e5601400 PM |
1318 | struct rcu_node *rnp = rdp->mynode; /* For dying CPU. */ |
1319 | ||
a50c3af9 PM |
1320 | /* First, adjust the counts. */ |
1321 | if (rdp->nxtlist != NULL) { | |
1322 | receive_rdp->qlen_lazy += rdp->qlen_lazy; | |
1323 | receive_rdp->qlen += rdp->qlen; | |
1324 | rdp->qlen_lazy = 0; | |
1325 | rdp->qlen = 0; | |
1326 | } | |
1327 | ||
1328 | /* | |
1329 | * Next, move ready-to-invoke callbacks to be invoked on some | |
1330 | * other CPU. These will not be required to pass through another | |
1331 | * grace period: They are done, regardless of CPU. | |
1332 | */ | |
1333 | if (rdp->nxtlist != NULL && | |
1334 | rdp->nxttail[RCU_DONE_TAIL] != &rdp->nxtlist) { | |
1335 | struct rcu_head *oldhead; | |
1336 | struct rcu_head **oldtail; | |
1337 | struct rcu_head **newtail; | |
1338 | ||
1339 | oldhead = rdp->nxtlist; | |
1340 | oldtail = receive_rdp->nxttail[RCU_DONE_TAIL]; | |
1341 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; | |
1342 | *rdp->nxttail[RCU_DONE_TAIL] = *oldtail; | |
1343 | *receive_rdp->nxttail[RCU_DONE_TAIL] = oldhead; | |
1344 | newtail = rdp->nxttail[RCU_DONE_TAIL]; | |
1345 | for (i = RCU_DONE_TAIL; i < RCU_NEXT_SIZE; i++) { | |
1346 | if (receive_rdp->nxttail[i] == oldtail) | |
1347 | receive_rdp->nxttail[i] = newtail; | |
1348 | if (rdp->nxttail[i] == newtail) | |
1349 | rdp->nxttail[i] = &rdp->nxtlist; | |
1350 | } | |
1351 | } | |
1352 | ||
1353 | /* | |
1354 | * Finally, put the rest of the callbacks at the end of the list. | |
1355 | * The ones that made it partway through get to start over: We | |
1356 | * cannot assume that grace periods are synchronized across CPUs. | |
1357 | * (We could splice RCU_WAIT_TAIL into RCU_NEXT_READY_TAIL, but | |
1358 | * this does not seem compelling. Not yet, anyway.) | |
1359 | */ | |
e5601400 PM |
1360 | if (rdp->nxtlist != NULL) { |
1361 | *receive_rdp->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist; | |
1362 | receive_rdp->nxttail[RCU_NEXT_TAIL] = | |
1363 | rdp->nxttail[RCU_NEXT_TAIL]; | |
e5601400 PM |
1364 | receive_rdp->n_cbs_adopted += rdp->qlen; |
1365 | rdp->n_cbs_orphaned += rdp->qlen; | |
1366 | ||
1367 | rdp->nxtlist = NULL; | |
1368 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
1369 | rdp->nxttail[i] = &rdp->nxtlist; | |
e5601400 | 1370 | } |
e74f4c45 | 1371 | |
a50c3af9 PM |
1372 | /* |
1373 | * Record a quiescent state for the dying CPU. This is safe | |
1374 | * only because we have already cleared out the callbacks. | |
1375 | * (Otherwise, the RCU core might try to schedule the invocation | |
1376 | * of callbacks on this now-offline CPU, which would be bad.) | |
1377 | */ | |
64db4cff | 1378 | mask = rdp->grpmask; /* rnp->grplo is constant. */ |
e5601400 PM |
1379 | trace_rcu_grace_period(rsp->name, |
1380 | rnp->gpnum + 1 - !!(rnp->qsmask & mask), | |
1381 | "cpuofl"); | |
1382 | rcu_report_qs_rdp(smp_processor_id(), rsp, rdp, rsp->gpnum); | |
1383 | /* Note that rcu_report_qs_rdp() might call trace_rcu_grace_period(). */ | |
1384 | ||
1385 | /* | |
1386 | * Remove the dying CPU from the bitmasks in the rcu_node | |
1387 | * hierarchy. Because we are in stop_machine() context, we | |
1388 | * automatically exclude ->onofflock critical sections. | |
1389 | */ | |
64db4cff | 1390 | do { |
e5601400 | 1391 | raw_spin_lock_irqsave(&rnp->lock, flags); |
64db4cff PM |
1392 | rnp->qsmaskinit &= ~mask; |
1393 | if (rnp->qsmaskinit != 0) { | |
e5601400 | 1394 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1395 | break; |
1396 | } | |
d4c08f2a | 1397 | if (rnp == rdp->mynode) { |
d9a3da06 | 1398 | need_report = rcu_preempt_offline_tasks(rsp, rnp, rdp); |
e5601400 PM |
1399 | if (need_report & RCU_OFL_TASKS_NORM_GP) |
1400 | rcu_report_unblock_qs_rnp(rnp, flags); | |
1401 | else | |
1402 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
1403 | if (need_report & RCU_OFL_TASKS_EXP_GP) | |
1404 | rcu_report_exp_rnp(rsp, rnp, true); | |
d4c08f2a | 1405 | } else |
e5601400 | 1406 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 1407 | mask = rnp->grpmask; |
64db4cff PM |
1408 | rnp = rnp->parent; |
1409 | } while (rnp != NULL); | |
64db4cff PM |
1410 | } |
1411 | ||
1412 | /* | |
e5601400 PM |
1413 | * The CPU has been completely removed, and some other CPU is reporting |
1414 | * this fact from process context. Do the remainder of the cleanup. | |
1415 | * There can only be one CPU hotplug operation at a time, so no other | |
1416 | * CPU can be attempting to update rcu_cpu_kthread_task. | |
64db4cff | 1417 | */ |
e5601400 | 1418 | static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp) |
64db4cff | 1419 | { |
e5601400 PM |
1420 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
1421 | struct rcu_node *rnp = rdp->mynode; | |
1422 | ||
1423 | rcu_stop_cpu_kthread(cpu); | |
1424 | rcu_node_kthread_setaffinity(rnp, -1); | |
64db4cff PM |
1425 | } |
1426 | ||
1427 | #else /* #ifdef CONFIG_HOTPLUG_CPU */ | |
1428 | ||
e5601400 | 1429 | static void rcu_cleanup_dying_cpu(struct rcu_state *rsp) |
e74f4c45 PM |
1430 | { |
1431 | } | |
1432 | ||
e5601400 | 1433 | static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp) |
64db4cff PM |
1434 | { |
1435 | } | |
1436 | ||
1437 | #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */ | |
1438 | ||
1439 | /* | |
1440 | * Invoke any RCU callbacks that have made it to the end of their grace | |
1441 | * period. Thottle as specified by rdp->blimit. | |
1442 | */ | |
37c72e56 | 1443 | static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
1444 | { |
1445 | unsigned long flags; | |
1446 | struct rcu_head *next, *list, **tail; | |
486e2593 | 1447 | int bl, count, count_lazy; |
64db4cff PM |
1448 | |
1449 | /* If no callbacks are ready, just return.*/ | |
29c00b4a | 1450 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) { |
486e2593 | 1451 | trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, 0); |
4968c300 PM |
1452 | trace_rcu_batch_end(rsp->name, 0, !!ACCESS_ONCE(rdp->nxtlist), |
1453 | need_resched(), is_idle_task(current), | |
1454 | rcu_is_callbacks_kthread()); | |
64db4cff | 1455 | return; |
29c00b4a | 1456 | } |
64db4cff PM |
1457 | |
1458 | /* | |
1459 | * Extract the list of ready callbacks, disabling to prevent | |
1460 | * races with call_rcu() from interrupt handlers. | |
1461 | */ | |
1462 | local_irq_save(flags); | |
8146c4e2 | 1463 | WARN_ON_ONCE(cpu_is_offline(smp_processor_id())); |
29c00b4a | 1464 | bl = rdp->blimit; |
486e2593 | 1465 | trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, bl); |
64db4cff PM |
1466 | list = rdp->nxtlist; |
1467 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; | |
1468 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | |
1469 | tail = rdp->nxttail[RCU_DONE_TAIL]; | |
1470 | for (count = RCU_NEXT_SIZE - 1; count >= 0; count--) | |
1471 | if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL]) | |
1472 | rdp->nxttail[count] = &rdp->nxtlist; | |
1473 | local_irq_restore(flags); | |
1474 | ||
1475 | /* Invoke callbacks. */ | |
486e2593 | 1476 | count = count_lazy = 0; |
64db4cff PM |
1477 | while (list) { |
1478 | next = list->next; | |
1479 | prefetch(next); | |
551d55a9 | 1480 | debug_rcu_head_unqueue(list); |
486e2593 PM |
1481 | if (__rcu_reclaim(rsp->name, list)) |
1482 | count_lazy++; | |
64db4cff | 1483 | list = next; |
dff1672d PM |
1484 | /* Stop only if limit reached and CPU has something to do. */ |
1485 | if (++count >= bl && | |
1486 | (need_resched() || | |
1487 | (!is_idle_task(current) && !rcu_is_callbacks_kthread()))) | |
64db4cff PM |
1488 | break; |
1489 | } | |
1490 | ||
1491 | local_irq_save(flags); | |
4968c300 PM |
1492 | trace_rcu_batch_end(rsp->name, count, !!list, need_resched(), |
1493 | is_idle_task(current), | |
1494 | rcu_is_callbacks_kthread()); | |
64db4cff PM |
1495 | |
1496 | /* Update count, and requeue any remaining callbacks. */ | |
486e2593 | 1497 | rdp->qlen_lazy -= count_lazy; |
64db4cff | 1498 | rdp->qlen -= count; |
269dcc1c | 1499 | rdp->n_cbs_invoked += count; |
64db4cff PM |
1500 | if (list != NULL) { |
1501 | *tail = rdp->nxtlist; | |
1502 | rdp->nxtlist = list; | |
1503 | for (count = 0; count < RCU_NEXT_SIZE; count++) | |
1504 | if (&rdp->nxtlist == rdp->nxttail[count]) | |
1505 | rdp->nxttail[count] = tail; | |
1506 | else | |
1507 | break; | |
1508 | } | |
1509 | ||
1510 | /* Reinstate batch limit if we have worked down the excess. */ | |
1511 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) | |
1512 | rdp->blimit = blimit; | |
1513 | ||
37c72e56 PM |
1514 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
1515 | if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) { | |
1516 | rdp->qlen_last_fqs_check = 0; | |
1517 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1518 | } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark) | |
1519 | rdp->qlen_last_fqs_check = rdp->qlen; | |
1520 | ||
64db4cff PM |
1521 | local_irq_restore(flags); |
1522 | ||
e0f23060 | 1523 | /* Re-invoke RCU core processing if there are callbacks remaining. */ |
64db4cff | 1524 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
a46e0899 | 1525 | invoke_rcu_core(); |
64db4cff PM |
1526 | } |
1527 | ||
1528 | /* | |
1529 | * Check to see if this CPU is in a non-context-switch quiescent state | |
1530 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
e0f23060 | 1531 | * Also schedule RCU core processing. |
64db4cff | 1532 | * |
9b2e4f18 | 1533 | * This function must be called from hardirq context. It is normally |
64db4cff PM |
1534 | * invoked from the scheduling-clock interrupt. If rcu_pending returns |
1535 | * false, there is no point in invoking rcu_check_callbacks(). | |
1536 | */ | |
1537 | void rcu_check_callbacks(int cpu, int user) | |
1538 | { | |
300df91c | 1539 | trace_rcu_utilization("Start scheduler-tick"); |
a858af28 | 1540 | increment_cpu_stall_ticks(); |
9b2e4f18 | 1541 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
64db4cff PM |
1542 | |
1543 | /* | |
1544 | * Get here if this CPU took its interrupt from user | |
1545 | * mode or from the idle loop, and if this is not a | |
1546 | * nested interrupt. In this case, the CPU is in | |
d6714c22 | 1547 | * a quiescent state, so note it. |
64db4cff PM |
1548 | * |
1549 | * No memory barrier is required here because both | |
d6714c22 PM |
1550 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
1551 | * variables that other CPUs neither access nor modify, | |
1552 | * at least not while the corresponding CPU is online. | |
64db4cff PM |
1553 | */ |
1554 | ||
d6714c22 PM |
1555 | rcu_sched_qs(cpu); |
1556 | rcu_bh_qs(cpu); | |
64db4cff PM |
1557 | |
1558 | } else if (!in_softirq()) { | |
1559 | ||
1560 | /* | |
1561 | * Get here if this CPU did not take its interrupt from | |
1562 | * softirq, in other words, if it is not interrupting | |
1563 | * a rcu_bh read-side critical section. This is an _bh | |
d6714c22 | 1564 | * critical section, so note it. |
64db4cff PM |
1565 | */ |
1566 | ||
d6714c22 | 1567 | rcu_bh_qs(cpu); |
64db4cff | 1568 | } |
f41d911f | 1569 | rcu_preempt_check_callbacks(cpu); |
d21670ac | 1570 | if (rcu_pending(cpu)) |
a46e0899 | 1571 | invoke_rcu_core(); |
300df91c | 1572 | trace_rcu_utilization("End scheduler-tick"); |
64db4cff PM |
1573 | } |
1574 | ||
64db4cff PM |
1575 | /* |
1576 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
1577 | * have not yet encountered a quiescent state, using the function specified. | |
27f4d280 PM |
1578 | * Also initiate boosting for any threads blocked on the root rcu_node. |
1579 | * | |
ee47eb9f | 1580 | * The caller must have suppressed start of new grace periods. |
64db4cff | 1581 | */ |
45f014c5 | 1582 | static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *)) |
64db4cff PM |
1583 | { |
1584 | unsigned long bit; | |
1585 | int cpu; | |
1586 | unsigned long flags; | |
1587 | unsigned long mask; | |
a0b6c9a7 | 1588 | struct rcu_node *rnp; |
64db4cff | 1589 | |
a0b6c9a7 | 1590 | rcu_for_each_leaf_node(rsp, rnp) { |
64db4cff | 1591 | mask = 0; |
1304afb2 | 1592 | raw_spin_lock_irqsave(&rnp->lock, flags); |
ee47eb9f | 1593 | if (!rcu_gp_in_progress(rsp)) { |
1304afb2 | 1594 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
0f10dc82 | 1595 | return; |
64db4cff | 1596 | } |
a0b6c9a7 | 1597 | if (rnp->qsmask == 0) { |
1217ed1b | 1598 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock */ |
64db4cff PM |
1599 | continue; |
1600 | } | |
a0b6c9a7 | 1601 | cpu = rnp->grplo; |
64db4cff | 1602 | bit = 1; |
a0b6c9a7 | 1603 | for (; cpu <= rnp->grphi; cpu++, bit <<= 1) { |
394f99a9 LJ |
1604 | if ((rnp->qsmask & bit) != 0 && |
1605 | f(per_cpu_ptr(rsp->rda, cpu))) | |
64db4cff PM |
1606 | mask |= bit; |
1607 | } | |
45f014c5 | 1608 | if (mask != 0) { |
64db4cff | 1609 | |
d3f6bad3 PM |
1610 | /* rcu_report_qs_rnp() releases rnp->lock. */ |
1611 | rcu_report_qs_rnp(mask, rsp, rnp, flags); | |
64db4cff PM |
1612 | continue; |
1613 | } | |
1304afb2 | 1614 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 1615 | } |
27f4d280 | 1616 | rnp = rcu_get_root(rsp); |
1217ed1b PM |
1617 | if (rnp->qsmask == 0) { |
1618 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
1619 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */ | |
1620 | } | |
64db4cff PM |
1621 | } |
1622 | ||
1623 | /* | |
1624 | * Force quiescent states on reluctant CPUs, and also detect which | |
1625 | * CPUs are in dyntick-idle mode. | |
1626 | */ | |
1627 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1628 | { | |
1629 | unsigned long flags; | |
64db4cff | 1630 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff | 1631 | |
300df91c PM |
1632 | trace_rcu_utilization("Start fqs"); |
1633 | if (!rcu_gp_in_progress(rsp)) { | |
1634 | trace_rcu_utilization("End fqs"); | |
64db4cff | 1635 | return; /* No grace period in progress, nothing to force. */ |
300df91c | 1636 | } |
1304afb2 | 1637 | if (!raw_spin_trylock_irqsave(&rsp->fqslock, flags)) { |
64db4cff | 1638 | rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ |
300df91c | 1639 | trace_rcu_utilization("End fqs"); |
64db4cff PM |
1640 | return; /* Someone else is already on the job. */ |
1641 | } | |
20133cfc | 1642 | if (relaxed && ULONG_CMP_GE(rsp->jiffies_force_qs, jiffies)) |
f96e9232 | 1643 | goto unlock_fqs_ret; /* no emergency and done recently. */ |
64db4cff | 1644 | rsp->n_force_qs++; |
1304afb2 | 1645 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
64db4cff | 1646 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
560d4bc0 | 1647 | if(!rcu_gp_in_progress(rsp)) { |
64db4cff | 1648 | rsp->n_force_qs_ngp++; |
1304afb2 | 1649 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1650 | goto unlock_fqs_ret; /* no GP in progress, time updated. */ |
64db4cff | 1651 | } |
07079d53 | 1652 | rsp->fqs_active = 1; |
af446b70 | 1653 | switch (rsp->fqs_state) { |
83f5b01f | 1654 | case RCU_GP_IDLE: |
64db4cff PM |
1655 | case RCU_GP_INIT: |
1656 | ||
83f5b01f | 1657 | break; /* grace period idle or initializing, ignore. */ |
64db4cff PM |
1658 | |
1659 | case RCU_SAVE_DYNTICK: | |
64db4cff PM |
1660 | if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK) |
1661 | break; /* So gcc recognizes the dead code. */ | |
1662 | ||
f261414f LJ |
1663 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
1664 | ||
64db4cff | 1665 | /* Record dyntick-idle state. */ |
45f014c5 | 1666 | force_qs_rnp(rsp, dyntick_save_progress_counter); |
1304afb2 | 1667 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
ee47eb9f | 1668 | if (rcu_gp_in_progress(rsp)) |
af446b70 | 1669 | rsp->fqs_state = RCU_FORCE_QS; |
ee47eb9f | 1670 | break; |
64db4cff PM |
1671 | |
1672 | case RCU_FORCE_QS: | |
1673 | ||
1674 | /* Check dyntick-idle state, send IPI to laggarts. */ | |
1304afb2 | 1675 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
45f014c5 | 1676 | force_qs_rnp(rsp, rcu_implicit_dynticks_qs); |
64db4cff PM |
1677 | |
1678 | /* Leave state in case more forcing is required. */ | |
1679 | ||
1304afb2 | 1680 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
f96e9232 | 1681 | break; |
64db4cff | 1682 | } |
07079d53 | 1683 | rsp->fqs_active = 0; |
46a1e34e | 1684 | if (rsp->fqs_need_gp) { |
1304afb2 | 1685 | raw_spin_unlock(&rsp->fqslock); /* irqs remain disabled */ |
46a1e34e PM |
1686 | rsp->fqs_need_gp = 0; |
1687 | rcu_start_gp(rsp, flags); /* releases rnp->lock */ | |
300df91c | 1688 | trace_rcu_utilization("End fqs"); |
46a1e34e PM |
1689 | return; |
1690 | } | |
1304afb2 | 1691 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1692 | unlock_fqs_ret: |
1304afb2 | 1693 | raw_spin_unlock_irqrestore(&rsp->fqslock, flags); |
300df91c | 1694 | trace_rcu_utilization("End fqs"); |
64db4cff PM |
1695 | } |
1696 | ||
64db4cff | 1697 | /* |
e0f23060 PM |
1698 | * This does the RCU core processing work for the specified rcu_state |
1699 | * and rcu_data structures. This may be called only from the CPU to | |
1700 | * whom the rdp belongs. | |
64db4cff PM |
1701 | */ |
1702 | static void | |
1703 | __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) | |
1704 | { | |
1705 | unsigned long flags; | |
1706 | ||
2e597558 PM |
1707 | WARN_ON_ONCE(rdp->beenonline == 0); |
1708 | ||
64db4cff PM |
1709 | /* |
1710 | * If an RCU GP has gone long enough, go check for dyntick | |
1711 | * idle CPUs and, if needed, send resched IPIs. | |
1712 | */ | |
20133cfc | 1713 | if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
64db4cff PM |
1714 | force_quiescent_state(rsp, 1); |
1715 | ||
1716 | /* | |
1717 | * Advance callbacks in response to end of earlier grace | |
1718 | * period that some other CPU ended. | |
1719 | */ | |
1720 | rcu_process_gp_end(rsp, rdp); | |
1721 | ||
1722 | /* Update RCU state based on any recent quiescent states. */ | |
1723 | rcu_check_quiescent_state(rsp, rdp); | |
1724 | ||
1725 | /* Does this CPU require a not-yet-started grace period? */ | |
1726 | if (cpu_needs_another_gp(rsp, rdp)) { | |
1304afb2 | 1727 | raw_spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags); |
64db4cff PM |
1728 | rcu_start_gp(rsp, flags); /* releases above lock */ |
1729 | } | |
1730 | ||
1731 | /* If there are callbacks ready, invoke them. */ | |
09223371 | 1732 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
a46e0899 | 1733 | invoke_rcu_callbacks(rsp, rdp); |
09223371 SL |
1734 | } |
1735 | ||
64db4cff | 1736 | /* |
e0f23060 | 1737 | * Do RCU core processing for the current CPU. |
64db4cff | 1738 | */ |
09223371 | 1739 | static void rcu_process_callbacks(struct softirq_action *unused) |
64db4cff | 1740 | { |
300df91c | 1741 | trace_rcu_utilization("Start RCU core"); |
d6714c22 PM |
1742 | __rcu_process_callbacks(&rcu_sched_state, |
1743 | &__get_cpu_var(rcu_sched_data)); | |
64db4cff | 1744 | __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); |
f41d911f | 1745 | rcu_preempt_process_callbacks(); |
300df91c | 1746 | trace_rcu_utilization("End RCU core"); |
64db4cff PM |
1747 | } |
1748 | ||
a26ac245 | 1749 | /* |
e0f23060 PM |
1750 | * Schedule RCU callback invocation. If the specified type of RCU |
1751 | * does not support RCU priority boosting, just do a direct call, | |
1752 | * otherwise wake up the per-CPU kernel kthread. Note that because we | |
1753 | * are running on the current CPU with interrupts disabled, the | |
1754 | * rcu_cpu_kthread_task cannot disappear out from under us. | |
a26ac245 | 1755 | */ |
a46e0899 | 1756 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) |
a26ac245 | 1757 | { |
b0d30417 PM |
1758 | if (unlikely(!ACCESS_ONCE(rcu_scheduler_fully_active))) |
1759 | return; | |
a46e0899 PM |
1760 | if (likely(!rsp->boost)) { |
1761 | rcu_do_batch(rsp, rdp); | |
a26ac245 PM |
1762 | return; |
1763 | } | |
a46e0899 | 1764 | invoke_rcu_callbacks_kthread(); |
a26ac245 PM |
1765 | } |
1766 | ||
a46e0899 | 1767 | static void invoke_rcu_core(void) |
09223371 SL |
1768 | { |
1769 | raise_softirq(RCU_SOFTIRQ); | |
1770 | } | |
1771 | ||
64db4cff PM |
1772 | static void |
1773 | __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu), | |
486e2593 | 1774 | struct rcu_state *rsp, bool lazy) |
64db4cff PM |
1775 | { |
1776 | unsigned long flags; | |
1777 | struct rcu_data *rdp; | |
1778 | ||
0bb7b59d | 1779 | WARN_ON_ONCE((unsigned long)head & 0x3); /* Misaligned rcu_head! */ |
551d55a9 | 1780 | debug_rcu_head_queue(head); |
64db4cff PM |
1781 | head->func = func; |
1782 | head->next = NULL; | |
1783 | ||
1784 | smp_mb(); /* Ensure RCU update seen before callback registry. */ | |
1785 | ||
1786 | /* | |
1787 | * Opportunistically note grace-period endings and beginnings. | |
1788 | * Note that we might see a beginning right after we see an | |
1789 | * end, but never vice versa, since this CPU has to pass through | |
1790 | * a quiescent state betweentimes. | |
1791 | */ | |
1792 | local_irq_save(flags); | |
e5601400 | 1793 | WARN_ON_ONCE(cpu_is_offline(smp_processor_id())); |
394f99a9 | 1794 | rdp = this_cpu_ptr(rsp->rda); |
64db4cff PM |
1795 | |
1796 | /* Add the callback to our list. */ | |
1797 | *rdp->nxttail[RCU_NEXT_TAIL] = head; | |
1798 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; | |
2655d57e | 1799 | rdp->qlen++; |
486e2593 PM |
1800 | if (lazy) |
1801 | rdp->qlen_lazy++; | |
2655d57e | 1802 | |
d4c08f2a PM |
1803 | if (__is_kfree_rcu_offset((unsigned long)func)) |
1804 | trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func, | |
486e2593 | 1805 | rdp->qlen_lazy, rdp->qlen); |
d4c08f2a | 1806 | else |
486e2593 | 1807 | trace_rcu_callback(rsp->name, head, rdp->qlen_lazy, rdp->qlen); |
d4c08f2a | 1808 | |
2655d57e PM |
1809 | /* If interrupts were disabled, don't dive into RCU core. */ |
1810 | if (irqs_disabled_flags(flags)) { | |
1811 | local_irq_restore(flags); | |
1812 | return; | |
1813 | } | |
64db4cff | 1814 | |
37c72e56 PM |
1815 | /* |
1816 | * Force the grace period if too many callbacks or too long waiting. | |
1817 | * Enforce hysteresis, and don't invoke force_quiescent_state() | |
1818 | * if some other CPU has recently done so. Also, don't bother | |
1819 | * invoking force_quiescent_state() if the newly enqueued callback | |
1820 | * is the only one waiting for a grace period to complete. | |
1821 | */ | |
2655d57e | 1822 | if (unlikely(rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) { |
b52573d2 PM |
1823 | |
1824 | /* Are we ignoring a completed grace period? */ | |
1825 | rcu_process_gp_end(rsp, rdp); | |
1826 | check_for_new_grace_period(rsp, rdp); | |
1827 | ||
1828 | /* Start a new grace period if one not already started. */ | |
1829 | if (!rcu_gp_in_progress(rsp)) { | |
1830 | unsigned long nestflag; | |
1831 | struct rcu_node *rnp_root = rcu_get_root(rsp); | |
1832 | ||
1833 | raw_spin_lock_irqsave(&rnp_root->lock, nestflag); | |
1834 | rcu_start_gp(rsp, nestflag); /* rlses rnp_root->lock */ | |
1835 | } else { | |
1836 | /* Give the grace period a kick. */ | |
1837 | rdp->blimit = LONG_MAX; | |
1838 | if (rsp->n_force_qs == rdp->n_force_qs_snap && | |
1839 | *rdp->nxttail[RCU_DONE_TAIL] != head) | |
1840 | force_quiescent_state(rsp, 0); | |
1841 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1842 | rdp->qlen_last_fqs_check = rdp->qlen; | |
1843 | } | |
20133cfc | 1844 | } else if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
64db4cff PM |
1845 | force_quiescent_state(rsp, 1); |
1846 | local_irq_restore(flags); | |
1847 | } | |
1848 | ||
1849 | /* | |
d6714c22 | 1850 | * Queue an RCU-sched callback for invocation after a grace period. |
64db4cff | 1851 | */ |
d6714c22 | 1852 | void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) |
64db4cff | 1853 | { |
486e2593 | 1854 | __call_rcu(head, func, &rcu_sched_state, 0); |
64db4cff | 1855 | } |
d6714c22 | 1856 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
64db4cff PM |
1857 | |
1858 | /* | |
486e2593 | 1859 | * Queue an RCU callback for invocation after a quicker grace period. |
64db4cff PM |
1860 | */ |
1861 | void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
1862 | { | |
486e2593 | 1863 | __call_rcu(head, func, &rcu_bh_state, 0); |
64db4cff PM |
1864 | } |
1865 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
1866 | ||
6ebb237b PM |
1867 | /** |
1868 | * synchronize_sched - wait until an rcu-sched grace period has elapsed. | |
1869 | * | |
1870 | * Control will return to the caller some time after a full rcu-sched | |
1871 | * grace period has elapsed, in other words after all currently executing | |
1872 | * rcu-sched read-side critical sections have completed. These read-side | |
1873 | * critical sections are delimited by rcu_read_lock_sched() and | |
1874 | * rcu_read_unlock_sched(), and may be nested. Note that preempt_disable(), | |
1875 | * local_irq_disable(), and so on may be used in place of | |
1876 | * rcu_read_lock_sched(). | |
1877 | * | |
1878 | * This means that all preempt_disable code sequences, including NMI and | |
1879 | * hardware-interrupt handlers, in progress on entry will have completed | |
1880 | * before this primitive returns. However, this does not guarantee that | |
1881 | * softirq handlers will have completed, since in some kernels, these | |
1882 | * handlers can run in process context, and can block. | |
1883 | * | |
1884 | * This primitive provides the guarantees made by the (now removed) | |
1885 | * synchronize_kernel() API. In contrast, synchronize_rcu() only | |
1886 | * guarantees that rcu_read_lock() sections will have completed. | |
1887 | * In "classic RCU", these two guarantees happen to be one and | |
1888 | * the same, but can differ in realtime RCU implementations. | |
1889 | */ | |
1890 | void synchronize_sched(void) | |
1891 | { | |
fe15d706 PM |
1892 | rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map) && |
1893 | !lock_is_held(&rcu_lock_map) && | |
1894 | !lock_is_held(&rcu_sched_lock_map), | |
1895 | "Illegal synchronize_sched() in RCU-sched read-side critical section"); | |
6ebb237b PM |
1896 | if (rcu_blocking_is_gp()) |
1897 | return; | |
2c42818e | 1898 | wait_rcu_gp(call_rcu_sched); |
6ebb237b PM |
1899 | } |
1900 | EXPORT_SYMBOL_GPL(synchronize_sched); | |
1901 | ||
1902 | /** | |
1903 | * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed. | |
1904 | * | |
1905 | * Control will return to the caller some time after a full rcu_bh grace | |
1906 | * period has elapsed, in other words after all currently executing rcu_bh | |
1907 | * read-side critical sections have completed. RCU read-side critical | |
1908 | * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(), | |
1909 | * and may be nested. | |
1910 | */ | |
1911 | void synchronize_rcu_bh(void) | |
1912 | { | |
fe15d706 PM |
1913 | rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map) && |
1914 | !lock_is_held(&rcu_lock_map) && | |
1915 | !lock_is_held(&rcu_sched_lock_map), | |
1916 | "Illegal synchronize_rcu_bh() in RCU-bh read-side critical section"); | |
6ebb237b PM |
1917 | if (rcu_blocking_is_gp()) |
1918 | return; | |
2c42818e | 1919 | wait_rcu_gp(call_rcu_bh); |
6ebb237b PM |
1920 | } |
1921 | EXPORT_SYMBOL_GPL(synchronize_rcu_bh); | |
1922 | ||
3d3b7db0 PM |
1923 | static atomic_t sync_sched_expedited_started = ATOMIC_INIT(0); |
1924 | static atomic_t sync_sched_expedited_done = ATOMIC_INIT(0); | |
1925 | ||
1926 | static int synchronize_sched_expedited_cpu_stop(void *data) | |
1927 | { | |
1928 | /* | |
1929 | * There must be a full memory barrier on each affected CPU | |
1930 | * between the time that try_stop_cpus() is called and the | |
1931 | * time that it returns. | |
1932 | * | |
1933 | * In the current initial implementation of cpu_stop, the | |
1934 | * above condition is already met when the control reaches | |
1935 | * this point and the following smp_mb() is not strictly | |
1936 | * necessary. Do smp_mb() anyway for documentation and | |
1937 | * robustness against future implementation changes. | |
1938 | */ | |
1939 | smp_mb(); /* See above comment block. */ | |
1940 | return 0; | |
1941 | } | |
1942 | ||
1943 | /* | |
1944 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
1945 | * approach to force grace period to end quickly. This consumes | |
1946 | * significant time on all CPUs, and is thus not recommended for | |
1947 | * any sort of common-case code. | |
1948 | * | |
1949 | * Note that it is illegal to call this function while holding any | |
1950 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
1951 | * observe this restriction will result in deadlock. | |
1952 | * | |
1953 | * This implementation can be thought of as an application of ticket | |
1954 | * locking to RCU, with sync_sched_expedited_started and | |
1955 | * sync_sched_expedited_done taking on the roles of the halves | |
1956 | * of the ticket-lock word. Each task atomically increments | |
1957 | * sync_sched_expedited_started upon entry, snapshotting the old value, | |
1958 | * then attempts to stop all the CPUs. If this succeeds, then each | |
1959 | * CPU will have executed a context switch, resulting in an RCU-sched | |
1960 | * grace period. We are then done, so we use atomic_cmpxchg() to | |
1961 | * update sync_sched_expedited_done to match our snapshot -- but | |
1962 | * only if someone else has not already advanced past our snapshot. | |
1963 | * | |
1964 | * On the other hand, if try_stop_cpus() fails, we check the value | |
1965 | * of sync_sched_expedited_done. If it has advanced past our | |
1966 | * initial snapshot, then someone else must have forced a grace period | |
1967 | * some time after we took our snapshot. In this case, our work is | |
1968 | * done for us, and we can simply return. Otherwise, we try again, | |
1969 | * but keep our initial snapshot for purposes of checking for someone | |
1970 | * doing our work for us. | |
1971 | * | |
1972 | * If we fail too many times in a row, we fall back to synchronize_sched(). | |
1973 | */ | |
1974 | void synchronize_sched_expedited(void) | |
1975 | { | |
1976 | int firstsnap, s, snap, trycount = 0; | |
1977 | ||
1978 | /* Note that atomic_inc_return() implies full memory barrier. */ | |
1979 | firstsnap = snap = atomic_inc_return(&sync_sched_expedited_started); | |
1980 | get_online_cpus(); | |
1981 | WARN_ON_ONCE(cpu_is_offline(smp_processor_id())); | |
1982 | ||
1983 | /* | |
1984 | * Each pass through the following loop attempts to force a | |
1985 | * context switch on each CPU. | |
1986 | */ | |
1987 | while (try_stop_cpus(cpu_online_mask, | |
1988 | synchronize_sched_expedited_cpu_stop, | |
1989 | NULL) == -EAGAIN) { | |
1990 | put_online_cpus(); | |
1991 | ||
1992 | /* No joy, try again later. Or just synchronize_sched(). */ | |
1993 | if (trycount++ < 10) | |
1994 | udelay(trycount * num_online_cpus()); | |
1995 | else { | |
1996 | synchronize_sched(); | |
1997 | return; | |
1998 | } | |
1999 | ||
2000 | /* Check to see if someone else did our work for us. */ | |
2001 | s = atomic_read(&sync_sched_expedited_done); | |
2002 | if (UINT_CMP_GE((unsigned)s, (unsigned)firstsnap)) { | |
2003 | smp_mb(); /* ensure test happens before caller kfree */ | |
2004 | return; | |
2005 | } | |
2006 | ||
2007 | /* | |
2008 | * Refetching sync_sched_expedited_started allows later | |
2009 | * callers to piggyback on our grace period. We subtract | |
2010 | * 1 to get the same token that the last incrementer got. | |
2011 | * We retry after they started, so our grace period works | |
2012 | * for them, and they started after our first try, so their | |
2013 | * grace period works for us. | |
2014 | */ | |
2015 | get_online_cpus(); | |
2016 | snap = atomic_read(&sync_sched_expedited_started); | |
2017 | smp_mb(); /* ensure read is before try_stop_cpus(). */ | |
2018 | } | |
2019 | ||
2020 | /* | |
2021 | * Everyone up to our most recent fetch is covered by our grace | |
2022 | * period. Update the counter, but only if our work is still | |
2023 | * relevant -- which it won't be if someone who started later | |
2024 | * than we did beat us to the punch. | |
2025 | */ | |
2026 | do { | |
2027 | s = atomic_read(&sync_sched_expedited_done); | |
2028 | if (UINT_CMP_GE((unsigned)s, (unsigned)snap)) { | |
2029 | smp_mb(); /* ensure test happens before caller kfree */ | |
2030 | break; | |
2031 | } | |
2032 | } while (atomic_cmpxchg(&sync_sched_expedited_done, s, snap) != s); | |
2033 | ||
2034 | put_online_cpus(); | |
2035 | } | |
2036 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
2037 | ||
64db4cff PM |
2038 | /* |
2039 | * Check to see if there is any immediate RCU-related work to be done | |
2040 | * by the current CPU, for the specified type of RCU, returning 1 if so. | |
2041 | * The checks are in order of increasing expense: checks that can be | |
2042 | * carried out against CPU-local state are performed first. However, | |
2043 | * we must check for CPU stalls first, else we might not get a chance. | |
2044 | */ | |
2045 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | |
2046 | { | |
2f51f988 PM |
2047 | struct rcu_node *rnp = rdp->mynode; |
2048 | ||
64db4cff PM |
2049 | rdp->n_rcu_pending++; |
2050 | ||
2051 | /* Check for CPU stalls, if enabled. */ | |
2052 | check_cpu_stall(rsp, rdp); | |
2053 | ||
2054 | /* Is the RCU core waiting for a quiescent state from this CPU? */ | |
5c51dd73 PM |
2055 | if (rcu_scheduler_fully_active && |
2056 | rdp->qs_pending && !rdp->passed_quiesce) { | |
d25eb944 PM |
2057 | |
2058 | /* | |
2059 | * If force_quiescent_state() coming soon and this CPU | |
2060 | * needs a quiescent state, and this is either RCU-sched | |
2061 | * or RCU-bh, force a local reschedule. | |
2062 | */ | |
d21670ac | 2063 | rdp->n_rp_qs_pending++; |
6cc68793 | 2064 | if (!rdp->preemptible && |
d25eb944 PM |
2065 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs) - 1, |
2066 | jiffies)) | |
2067 | set_need_resched(); | |
e4cc1f22 | 2068 | } else if (rdp->qs_pending && rdp->passed_quiesce) { |
d21670ac | 2069 | rdp->n_rp_report_qs++; |
64db4cff | 2070 | return 1; |
7ba5c840 | 2071 | } |
64db4cff PM |
2072 | |
2073 | /* Does this CPU have callbacks ready to invoke? */ | |
7ba5c840 PM |
2074 | if (cpu_has_callbacks_ready_to_invoke(rdp)) { |
2075 | rdp->n_rp_cb_ready++; | |
64db4cff | 2076 | return 1; |
7ba5c840 | 2077 | } |
64db4cff PM |
2078 | |
2079 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
7ba5c840 PM |
2080 | if (cpu_needs_another_gp(rsp, rdp)) { |
2081 | rdp->n_rp_cpu_needs_gp++; | |
64db4cff | 2082 | return 1; |
7ba5c840 | 2083 | } |
64db4cff PM |
2084 | |
2085 | /* Has another RCU grace period completed? */ | |
2f51f988 | 2086 | if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */ |
7ba5c840 | 2087 | rdp->n_rp_gp_completed++; |
64db4cff | 2088 | return 1; |
7ba5c840 | 2089 | } |
64db4cff PM |
2090 | |
2091 | /* Has a new RCU grace period started? */ | |
2f51f988 | 2092 | if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */ |
7ba5c840 | 2093 | rdp->n_rp_gp_started++; |
64db4cff | 2094 | return 1; |
7ba5c840 | 2095 | } |
64db4cff PM |
2096 | |
2097 | /* Has an RCU GP gone long enough to send resched IPIs &c? */ | |
fc2219d4 | 2098 | if (rcu_gp_in_progress(rsp) && |
20133cfc | 2099 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) { |
7ba5c840 | 2100 | rdp->n_rp_need_fqs++; |
64db4cff | 2101 | return 1; |
7ba5c840 | 2102 | } |
64db4cff PM |
2103 | |
2104 | /* nothing to do */ | |
7ba5c840 | 2105 | rdp->n_rp_need_nothing++; |
64db4cff PM |
2106 | return 0; |
2107 | } | |
2108 | ||
2109 | /* | |
2110 | * Check to see if there is any immediate RCU-related work to be done | |
2111 | * by the current CPU, returning 1 if so. This function is part of the | |
2112 | * RCU implementation; it is -not- an exported member of the RCU API. | |
2113 | */ | |
a157229c | 2114 | static int rcu_pending(int cpu) |
64db4cff | 2115 | { |
d6714c22 | 2116 | return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) || |
f41d911f PM |
2117 | __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) || |
2118 | rcu_preempt_pending(cpu); | |
64db4cff PM |
2119 | } |
2120 | ||
2121 | /* | |
2122 | * Check to see if any future RCU-related work will need to be done | |
2123 | * by the current CPU, even if none need be done immediately, returning | |
8bd93a2c | 2124 | * 1 if so. |
64db4cff | 2125 | */ |
aea1b35e | 2126 | static int rcu_cpu_has_callbacks(int cpu) |
64db4cff PM |
2127 | { |
2128 | /* RCU callbacks either ready or pending? */ | |
d6714c22 | 2129 | return per_cpu(rcu_sched_data, cpu).nxtlist || |
f41d911f | 2130 | per_cpu(rcu_bh_data, cpu).nxtlist || |
30fbcc90 | 2131 | rcu_preempt_cpu_has_callbacks(cpu); |
64db4cff PM |
2132 | } |
2133 | ||
d0ec774c PM |
2134 | static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL}; |
2135 | static atomic_t rcu_barrier_cpu_count; | |
2136 | static DEFINE_MUTEX(rcu_barrier_mutex); | |
2137 | static struct completion rcu_barrier_completion; | |
d0ec774c PM |
2138 | |
2139 | static void rcu_barrier_callback(struct rcu_head *notused) | |
2140 | { | |
2141 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
2142 | complete(&rcu_barrier_completion); | |
2143 | } | |
2144 | ||
2145 | /* | |
2146 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
2147 | */ | |
2148 | static void rcu_barrier_func(void *type) | |
2149 | { | |
2150 | int cpu = smp_processor_id(); | |
2151 | struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu); | |
2152 | void (*call_rcu_func)(struct rcu_head *head, | |
2153 | void (*func)(struct rcu_head *head)); | |
2154 | ||
2155 | atomic_inc(&rcu_barrier_cpu_count); | |
2156 | call_rcu_func = type; | |
2157 | call_rcu_func(head, rcu_barrier_callback); | |
2158 | } | |
2159 | ||
d0ec774c PM |
2160 | /* |
2161 | * Orchestrate the specified type of RCU barrier, waiting for all | |
2162 | * RCU callbacks of the specified type to complete. | |
2163 | */ | |
e74f4c45 PM |
2164 | static void _rcu_barrier(struct rcu_state *rsp, |
2165 | void (*call_rcu_func)(struct rcu_head *head, | |
d0ec774c PM |
2166 | void (*func)(struct rcu_head *head))) |
2167 | { | |
2168 | BUG_ON(in_interrupt()); | |
e74f4c45 | 2169 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
d0ec774c PM |
2170 | mutex_lock(&rcu_barrier_mutex); |
2171 | init_completion(&rcu_barrier_completion); | |
2172 | /* | |
2173 | * Initialize rcu_barrier_cpu_count to 1, then invoke | |
2174 | * rcu_barrier_func() on each CPU, so that each CPU also has | |
2175 | * incremented rcu_barrier_cpu_count. Only then is it safe to | |
2176 | * decrement rcu_barrier_cpu_count -- otherwise the first CPU | |
2177 | * might complete its grace period before all of the other CPUs | |
2178 | * did their increment, causing this function to return too | |
2d999e03 PM |
2179 | * early. Note that on_each_cpu() disables irqs, which prevents |
2180 | * any CPUs from coming online or going offline until each online | |
2181 | * CPU has queued its RCU-barrier callback. | |
d0ec774c PM |
2182 | */ |
2183 | atomic_set(&rcu_barrier_cpu_count, 1); | |
2184 | on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1); | |
2185 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
2186 | complete(&rcu_barrier_completion); | |
2187 | wait_for_completion(&rcu_barrier_completion); | |
2188 | mutex_unlock(&rcu_barrier_mutex); | |
d0ec774c | 2189 | } |
d0ec774c PM |
2190 | |
2191 | /** | |
2192 | * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete. | |
2193 | */ | |
2194 | void rcu_barrier_bh(void) | |
2195 | { | |
e74f4c45 | 2196 | _rcu_barrier(&rcu_bh_state, call_rcu_bh); |
d0ec774c PM |
2197 | } |
2198 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); | |
2199 | ||
2200 | /** | |
2201 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. | |
2202 | */ | |
2203 | void rcu_barrier_sched(void) | |
2204 | { | |
e74f4c45 | 2205 | _rcu_barrier(&rcu_sched_state, call_rcu_sched); |
d0ec774c PM |
2206 | } |
2207 | EXPORT_SYMBOL_GPL(rcu_barrier_sched); | |
2208 | ||
64db4cff | 2209 | /* |
27569620 | 2210 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 2211 | */ |
27569620 PM |
2212 | static void __init |
2213 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) | |
64db4cff PM |
2214 | { |
2215 | unsigned long flags; | |
2216 | int i; | |
394f99a9 | 2217 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
27569620 PM |
2218 | struct rcu_node *rnp = rcu_get_root(rsp); |
2219 | ||
2220 | /* Set up local state, ensuring consistent view of global state. */ | |
1304afb2 | 2221 | raw_spin_lock_irqsave(&rnp->lock, flags); |
27569620 PM |
2222 | rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo); |
2223 | rdp->nxtlist = NULL; | |
2224 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
2225 | rdp->nxttail[i] = &rdp->nxtlist; | |
486e2593 | 2226 | rdp->qlen_lazy = 0; |
27569620 | 2227 | rdp->qlen = 0; |
27569620 | 2228 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); |
4145fa7f | 2229 | WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != DYNTICK_TASK_NESTING); |
9b2e4f18 | 2230 | WARN_ON_ONCE(atomic_read(&rdp->dynticks->dynticks) != 1); |
27569620 | 2231 | rdp->cpu = cpu; |
d4c08f2a | 2232 | rdp->rsp = rsp; |
1304afb2 | 2233 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
27569620 PM |
2234 | } |
2235 | ||
2236 | /* | |
2237 | * Initialize a CPU's per-CPU RCU data. Note that only one online or | |
2238 | * offline event can be happening at a given time. Note also that we | |
2239 | * can accept some slop in the rsp->completed access due to the fact | |
2240 | * that this CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 2241 | */ |
e4fa4c97 | 2242 | static void __cpuinit |
6cc68793 | 2243 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptible) |
64db4cff PM |
2244 | { |
2245 | unsigned long flags; | |
64db4cff | 2246 | unsigned long mask; |
394f99a9 | 2247 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff PM |
2248 | struct rcu_node *rnp = rcu_get_root(rsp); |
2249 | ||
2250 | /* Set up local state, ensuring consistent view of global state. */ | |
1304afb2 | 2251 | raw_spin_lock_irqsave(&rnp->lock, flags); |
64db4cff | 2252 | rdp->beenonline = 1; /* We have now been online. */ |
6cc68793 | 2253 | rdp->preemptible = preemptible; |
37c72e56 PM |
2254 | rdp->qlen_last_fqs_check = 0; |
2255 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
64db4cff | 2256 | rdp->blimit = blimit; |
c92b131b PM |
2257 | rdp->dynticks->dynticks_nesting = DYNTICK_TASK_NESTING; |
2258 | atomic_set(&rdp->dynticks->dynticks, | |
2259 | (atomic_read(&rdp->dynticks->dynticks) & ~0x1) + 1); | |
7cb92499 | 2260 | rcu_prepare_for_idle_init(cpu); |
1304afb2 | 2261 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
2262 | |
2263 | /* | |
2264 | * A new grace period might start here. If so, we won't be part | |
2265 | * of it, but that is OK, as we are currently in a quiescent state. | |
2266 | */ | |
2267 | ||
2268 | /* Exclude any attempts to start a new GP on large systems. */ | |
1304afb2 | 2269 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
64db4cff PM |
2270 | |
2271 | /* Add CPU to rcu_node bitmasks. */ | |
2272 | rnp = rdp->mynode; | |
2273 | mask = rdp->grpmask; | |
2274 | do { | |
2275 | /* Exclude any attempts to start a new GP on small systems. */ | |
1304afb2 | 2276 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
2277 | rnp->qsmaskinit |= mask; |
2278 | mask = rnp->grpmask; | |
d09b62df | 2279 | if (rnp == rdp->mynode) { |
06ae115a PM |
2280 | /* |
2281 | * If there is a grace period in progress, we will | |
2282 | * set up to wait for it next time we run the | |
2283 | * RCU core code. | |
2284 | */ | |
2285 | rdp->gpnum = rnp->completed; | |
d09b62df | 2286 | rdp->completed = rnp->completed; |
06ae115a PM |
2287 | rdp->passed_quiesce = 0; |
2288 | rdp->qs_pending = 0; | |
e4cc1f22 | 2289 | rdp->passed_quiesce_gpnum = rnp->gpnum - 1; |
d4c08f2a | 2290 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuonl"); |
d09b62df | 2291 | } |
1304afb2 | 2292 | raw_spin_unlock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
2293 | rnp = rnp->parent; |
2294 | } while (rnp != NULL && !(rnp->qsmaskinit & mask)); | |
2295 | ||
1304afb2 | 2296 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); |
64db4cff PM |
2297 | } |
2298 | ||
d72bce0e | 2299 | static void __cpuinit rcu_prepare_cpu(int cpu) |
64db4cff | 2300 | { |
f41d911f PM |
2301 | rcu_init_percpu_data(cpu, &rcu_sched_state, 0); |
2302 | rcu_init_percpu_data(cpu, &rcu_bh_state, 0); | |
2303 | rcu_preempt_init_percpu_data(cpu); | |
64db4cff PM |
2304 | } |
2305 | ||
2306 | /* | |
f41d911f | 2307 | * Handle CPU online/offline notification events. |
64db4cff | 2308 | */ |
9f680ab4 PM |
2309 | static int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
2310 | unsigned long action, void *hcpu) | |
64db4cff PM |
2311 | { |
2312 | long cpu = (long)hcpu; | |
27f4d280 | 2313 | struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu); |
a26ac245 | 2314 | struct rcu_node *rnp = rdp->mynode; |
64db4cff | 2315 | |
300df91c | 2316 | trace_rcu_utilization("Start CPU hotplug"); |
64db4cff PM |
2317 | switch (action) { |
2318 | case CPU_UP_PREPARE: | |
2319 | case CPU_UP_PREPARE_FROZEN: | |
d72bce0e PZ |
2320 | rcu_prepare_cpu(cpu); |
2321 | rcu_prepare_kthreads(cpu); | |
a26ac245 PM |
2322 | break; |
2323 | case CPU_ONLINE: | |
0f962a5e PM |
2324 | case CPU_DOWN_FAILED: |
2325 | rcu_node_kthread_setaffinity(rnp, -1); | |
e3995a25 | 2326 | rcu_cpu_kthread_setrt(cpu, 1); |
0f962a5e PM |
2327 | break; |
2328 | case CPU_DOWN_PREPARE: | |
2329 | rcu_node_kthread_setaffinity(rnp, cpu); | |
e3995a25 | 2330 | rcu_cpu_kthread_setrt(cpu, 0); |
64db4cff | 2331 | break; |
d0ec774c PM |
2332 | case CPU_DYING: |
2333 | case CPU_DYING_FROZEN: | |
2334 | /* | |
2d999e03 PM |
2335 | * The whole machine is "stopped" except this CPU, so we can |
2336 | * touch any data without introducing corruption. We send the | |
2337 | * dying CPU's callbacks to an arbitrarily chosen online CPU. | |
d0ec774c | 2338 | */ |
e5601400 PM |
2339 | rcu_cleanup_dying_cpu(&rcu_bh_state); |
2340 | rcu_cleanup_dying_cpu(&rcu_sched_state); | |
2341 | rcu_preempt_cleanup_dying_cpu(); | |
7cb92499 | 2342 | rcu_cleanup_after_idle(cpu); |
d0ec774c | 2343 | break; |
64db4cff PM |
2344 | case CPU_DEAD: |
2345 | case CPU_DEAD_FROZEN: | |
2346 | case CPU_UP_CANCELED: | |
2347 | case CPU_UP_CANCELED_FROZEN: | |
e5601400 PM |
2348 | rcu_cleanup_dead_cpu(cpu, &rcu_bh_state); |
2349 | rcu_cleanup_dead_cpu(cpu, &rcu_sched_state); | |
2350 | rcu_preempt_cleanup_dead_cpu(cpu); | |
64db4cff PM |
2351 | break; |
2352 | default: | |
2353 | break; | |
2354 | } | |
300df91c | 2355 | trace_rcu_utilization("End CPU hotplug"); |
64db4cff PM |
2356 | return NOTIFY_OK; |
2357 | } | |
2358 | ||
bbad9379 PM |
2359 | /* |
2360 | * This function is invoked towards the end of the scheduler's initialization | |
2361 | * process. Before this is called, the idle task might contain | |
2362 | * RCU read-side critical sections (during which time, this idle | |
2363 | * task is booting the system). After this function is called, the | |
2364 | * idle tasks are prohibited from containing RCU read-side critical | |
2365 | * sections. This function also enables RCU lockdep checking. | |
2366 | */ | |
2367 | void rcu_scheduler_starting(void) | |
2368 | { | |
2369 | WARN_ON(num_online_cpus() != 1); | |
2370 | WARN_ON(nr_context_switches() > 0); | |
2371 | rcu_scheduler_active = 1; | |
2372 | } | |
2373 | ||
64db4cff PM |
2374 | /* |
2375 | * Compute the per-level fanout, either using the exact fanout specified | |
2376 | * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT. | |
2377 | */ | |
2378 | #ifdef CONFIG_RCU_FANOUT_EXACT | |
2379 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
2380 | { | |
2381 | int i; | |
2382 | ||
0209f649 | 2383 | for (i = NUM_RCU_LVLS - 1; i > 0; i--) |
64db4cff | 2384 | rsp->levelspread[i] = CONFIG_RCU_FANOUT; |
0209f649 | 2385 | rsp->levelspread[0] = RCU_FANOUT_LEAF; |
64db4cff PM |
2386 | } |
2387 | #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
2388 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
2389 | { | |
2390 | int ccur; | |
2391 | int cprv; | |
2392 | int i; | |
2393 | ||
2394 | cprv = NR_CPUS; | |
2395 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
2396 | ccur = rsp->levelcnt[i]; | |
2397 | rsp->levelspread[i] = (cprv + ccur - 1) / ccur; | |
2398 | cprv = ccur; | |
2399 | } | |
2400 | } | |
2401 | #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
2402 | ||
2403 | /* | |
2404 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
2405 | */ | |
394f99a9 LJ |
2406 | static void __init rcu_init_one(struct rcu_state *rsp, |
2407 | struct rcu_data __percpu *rda) | |
64db4cff | 2408 | { |
b6407e86 PM |
2409 | static char *buf[] = { "rcu_node_level_0", |
2410 | "rcu_node_level_1", | |
2411 | "rcu_node_level_2", | |
2412 | "rcu_node_level_3" }; /* Match MAX_RCU_LVLS */ | |
64db4cff PM |
2413 | int cpustride = 1; |
2414 | int i; | |
2415 | int j; | |
2416 | struct rcu_node *rnp; | |
2417 | ||
b6407e86 PM |
2418 | BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
2419 | ||
64db4cff PM |
2420 | /* Initialize the level-tracking arrays. */ |
2421 | ||
2422 | for (i = 1; i < NUM_RCU_LVLS; i++) | |
2423 | rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1]; | |
2424 | rcu_init_levelspread(rsp); | |
2425 | ||
2426 | /* Initialize the elements themselves, starting from the leaves. */ | |
2427 | ||
2428 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
2429 | cpustride *= rsp->levelspread[i]; | |
2430 | rnp = rsp->level[i]; | |
2431 | for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { | |
1304afb2 | 2432 | raw_spin_lock_init(&rnp->lock); |
b6407e86 PM |
2433 | lockdep_set_class_and_name(&rnp->lock, |
2434 | &rcu_node_class[i], buf[i]); | |
f41d911f | 2435 | rnp->gpnum = 0; |
64db4cff PM |
2436 | rnp->qsmask = 0; |
2437 | rnp->qsmaskinit = 0; | |
2438 | rnp->grplo = j * cpustride; | |
2439 | rnp->grphi = (j + 1) * cpustride - 1; | |
2440 | if (rnp->grphi >= NR_CPUS) | |
2441 | rnp->grphi = NR_CPUS - 1; | |
2442 | if (i == 0) { | |
2443 | rnp->grpnum = 0; | |
2444 | rnp->grpmask = 0; | |
2445 | rnp->parent = NULL; | |
2446 | } else { | |
2447 | rnp->grpnum = j % rsp->levelspread[i - 1]; | |
2448 | rnp->grpmask = 1UL << rnp->grpnum; | |
2449 | rnp->parent = rsp->level[i - 1] + | |
2450 | j / rsp->levelspread[i - 1]; | |
2451 | } | |
2452 | rnp->level = i; | |
12f5f524 | 2453 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
64db4cff PM |
2454 | } |
2455 | } | |
0c34029a | 2456 | |
394f99a9 | 2457 | rsp->rda = rda; |
0c34029a LJ |
2458 | rnp = rsp->level[NUM_RCU_LVLS - 1]; |
2459 | for_each_possible_cpu(i) { | |
4a90a068 | 2460 | while (i > rnp->grphi) |
0c34029a | 2461 | rnp++; |
394f99a9 | 2462 | per_cpu_ptr(rsp->rda, i)->mynode = rnp; |
0c34029a LJ |
2463 | rcu_boot_init_percpu_data(i, rsp); |
2464 | } | |
64db4cff PM |
2465 | } |
2466 | ||
9f680ab4 | 2467 | void __init rcu_init(void) |
64db4cff | 2468 | { |
017c4261 | 2469 | int cpu; |
9f680ab4 | 2470 | |
f41d911f | 2471 | rcu_bootup_announce(); |
394f99a9 LJ |
2472 | rcu_init_one(&rcu_sched_state, &rcu_sched_data); |
2473 | rcu_init_one(&rcu_bh_state, &rcu_bh_data); | |
f41d911f | 2474 | __rcu_init_preempt(); |
09223371 | 2475 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
9f680ab4 PM |
2476 | |
2477 | /* | |
2478 | * We don't need protection against CPU-hotplug here because | |
2479 | * this is called early in boot, before either interrupts | |
2480 | * or the scheduler are operational. | |
2481 | */ | |
2482 | cpu_notifier(rcu_cpu_notify, 0); | |
017c4261 PM |
2483 | for_each_online_cpu(cpu) |
2484 | rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu); | |
c68de209 | 2485 | check_cpu_stall_init(); |
64db4cff PM |
2486 | } |
2487 | ||
1eba8f84 | 2488 | #include "rcutree_plugin.h" |