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 | 77 | .onofflock = __RAW_SPIN_LOCK_UNLOCKED(&structname##_state.onofflock), \ |
b1420f1c PM |
78 | .orphan_nxttail = &structname##_state.orphan_nxtlist, \ |
79 | .orphan_donetail = &structname##_state.orphan_donelist, \ | |
e99033c5 | 80 | .fqslock = __RAW_SPIN_LOCK_UNLOCKED(&structname##_state.fqslock), \ |
64db4cff PM |
81 | .n_force_qs = 0, \ |
82 | .n_force_qs_ngp = 0, \ | |
4300aa64 | 83 | .name = #structname, \ |
64db4cff PM |
84 | } |
85 | ||
e99033c5 | 86 | struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched); |
d6714c22 | 87 | DEFINE_PER_CPU(struct rcu_data, rcu_sched_data); |
64db4cff | 88 | |
e99033c5 | 89 | struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh); |
6258c4fb | 90 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); |
b1f77b05 | 91 | |
27f4d280 PM |
92 | static struct rcu_state *rcu_state; |
93 | ||
b0d30417 PM |
94 | /* |
95 | * The rcu_scheduler_active variable transitions from zero to one just | |
96 | * before the first task is spawned. So when this variable is zero, RCU | |
97 | * can assume that there is but one task, allowing RCU to (for example) | |
98 | * optimized synchronize_sched() to a simple barrier(). When this variable | |
99 | * is one, RCU must actually do all the hard work required to detect real | |
100 | * grace periods. This variable is also used to suppress boot-time false | |
101 | * positives from lockdep-RCU error checking. | |
102 | */ | |
bbad9379 PM |
103 | int rcu_scheduler_active __read_mostly; |
104 | EXPORT_SYMBOL_GPL(rcu_scheduler_active); | |
105 | ||
b0d30417 PM |
106 | /* |
107 | * The rcu_scheduler_fully_active variable transitions from zero to one | |
108 | * during the early_initcall() processing, which is after the scheduler | |
109 | * is capable of creating new tasks. So RCU processing (for example, | |
110 | * creating tasks for RCU priority boosting) must be delayed until after | |
111 | * rcu_scheduler_fully_active transitions from zero to one. We also | |
112 | * currently delay invocation of any RCU callbacks until after this point. | |
113 | * | |
114 | * It might later prove better for people registering RCU callbacks during | |
115 | * early boot to take responsibility for these callbacks, but one step at | |
116 | * a time. | |
117 | */ | |
118 | static int rcu_scheduler_fully_active __read_mostly; | |
119 | ||
a46e0899 PM |
120 | #ifdef CONFIG_RCU_BOOST |
121 | ||
a26ac245 PM |
122 | /* |
123 | * Control variables for per-CPU and per-rcu_node kthreads. These | |
124 | * handle all flavors of RCU. | |
125 | */ | |
126 | static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task); | |
d71df90e | 127 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status); |
15ba0ba8 | 128 | DEFINE_PER_CPU(int, rcu_cpu_kthread_cpu); |
5ece5bab | 129 | DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops); |
d71df90e | 130 | DEFINE_PER_CPU(char, rcu_cpu_has_work); |
a26ac245 | 131 | |
a46e0899 PM |
132 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
133 | ||
0f962a5e | 134 | static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); |
a46e0899 PM |
135 | static void invoke_rcu_core(void); |
136 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp); | |
a26ac245 | 137 | |
4a298656 PM |
138 | /* |
139 | * Track the rcutorture test sequence number and the update version | |
140 | * number within a given test. The rcutorture_testseq is incremented | |
141 | * on every rcutorture module load and unload, so has an odd value | |
142 | * when a test is running. The rcutorture_vernum is set to zero | |
143 | * when rcutorture starts and is incremented on each rcutorture update. | |
144 | * These variables enable correlating rcutorture output with the | |
145 | * RCU tracing information. | |
146 | */ | |
147 | unsigned long rcutorture_testseq; | |
148 | unsigned long rcutorture_vernum; | |
149 | ||
b1420f1c PM |
150 | /* State information for rcu_barrier() and friends. */ |
151 | ||
152 | static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL}; | |
153 | static atomic_t rcu_barrier_cpu_count; | |
154 | static DEFINE_MUTEX(rcu_barrier_mutex); | |
155 | static struct completion rcu_barrier_completion; | |
156 | ||
fc2219d4 PM |
157 | /* |
158 | * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s | |
159 | * permit this function to be invoked without holding the root rcu_node | |
160 | * structure's ->lock, but of course results can be subject to change. | |
161 | */ | |
162 | static int rcu_gp_in_progress(struct rcu_state *rsp) | |
163 | { | |
164 | return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum); | |
165 | } | |
166 | ||
b1f77b05 | 167 | /* |
d6714c22 | 168 | * Note a quiescent state. Because we do not need to know |
b1f77b05 | 169 | * how many quiescent states passed, just if there was at least |
d6714c22 | 170 | * one since the start of the grace period, this just sets a flag. |
e4cc1f22 | 171 | * The caller must have disabled preemption. |
b1f77b05 | 172 | */ |
d6714c22 | 173 | void rcu_sched_qs(int cpu) |
b1f77b05 | 174 | { |
25502a6c | 175 | struct rcu_data *rdp = &per_cpu(rcu_sched_data, cpu); |
f41d911f | 176 | |
e4cc1f22 | 177 | rdp->passed_quiesce_gpnum = rdp->gpnum; |
c3422bea | 178 | barrier(); |
e4cc1f22 | 179 | if (rdp->passed_quiesce == 0) |
d4c08f2a | 180 | trace_rcu_grace_period("rcu_sched", rdp->gpnum, "cpuqs"); |
e4cc1f22 | 181 | rdp->passed_quiesce = 1; |
b1f77b05 IM |
182 | } |
183 | ||
d6714c22 | 184 | void rcu_bh_qs(int cpu) |
b1f77b05 | 185 | { |
25502a6c | 186 | struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu); |
f41d911f | 187 | |
e4cc1f22 | 188 | rdp->passed_quiesce_gpnum = rdp->gpnum; |
c3422bea | 189 | barrier(); |
e4cc1f22 | 190 | if (rdp->passed_quiesce == 0) |
d4c08f2a | 191 | trace_rcu_grace_period("rcu_bh", rdp->gpnum, "cpuqs"); |
e4cc1f22 | 192 | rdp->passed_quiesce = 1; |
b1f77b05 | 193 | } |
64db4cff | 194 | |
25502a6c PM |
195 | /* |
196 | * Note a context switch. This is a quiescent state for RCU-sched, | |
197 | * and requires special handling for preemptible RCU. | |
e4cc1f22 | 198 | * The caller must have disabled preemption. |
25502a6c PM |
199 | */ |
200 | void rcu_note_context_switch(int cpu) | |
201 | { | |
300df91c | 202 | trace_rcu_utilization("Start context switch"); |
25502a6c | 203 | rcu_sched_qs(cpu); |
cba6d0d6 | 204 | rcu_preempt_note_context_switch(cpu); |
300df91c | 205 | trace_rcu_utilization("End context switch"); |
25502a6c | 206 | } |
29ce8310 | 207 | EXPORT_SYMBOL_GPL(rcu_note_context_switch); |
25502a6c | 208 | |
90a4d2c0 | 209 | DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { |
29e37d81 | 210 | .dynticks_nesting = DYNTICK_TASK_EXIT_IDLE, |
23b5c8fa | 211 | .dynticks = ATOMIC_INIT(1), |
90a4d2c0 | 212 | }; |
64db4cff | 213 | |
e0f23060 | 214 | static int blimit = 10; /* Maximum callbacks per rcu_do_batch. */ |
64db4cff PM |
215 | static int qhimark = 10000; /* If this many pending, ignore blimit. */ |
216 | static int qlowmark = 100; /* Once only this many pending, use blimit. */ | |
217 | ||
3d76c082 PM |
218 | module_param(blimit, int, 0); |
219 | module_param(qhimark, int, 0); | |
220 | module_param(qlowmark, int, 0); | |
221 | ||
13cfcca0 PM |
222 | int rcu_cpu_stall_suppress __read_mostly; /* 1 = suppress stall warnings. */ |
223 | int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT; | |
224 | ||
f2e0dd70 | 225 | module_param(rcu_cpu_stall_suppress, int, 0644); |
13cfcca0 | 226 | module_param(rcu_cpu_stall_timeout, int, 0644); |
742734ee | 227 | |
64db4cff | 228 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed); |
a157229c | 229 | static int rcu_pending(int cpu); |
64db4cff PM |
230 | |
231 | /* | |
d6714c22 | 232 | * Return the number of RCU-sched batches processed thus far for debug & stats. |
64db4cff | 233 | */ |
d6714c22 | 234 | long rcu_batches_completed_sched(void) |
64db4cff | 235 | { |
d6714c22 | 236 | return rcu_sched_state.completed; |
64db4cff | 237 | } |
d6714c22 | 238 | EXPORT_SYMBOL_GPL(rcu_batches_completed_sched); |
64db4cff PM |
239 | |
240 | /* | |
241 | * Return the number of RCU BH batches processed thus far for debug & stats. | |
242 | */ | |
243 | long rcu_batches_completed_bh(void) | |
244 | { | |
245 | return rcu_bh_state.completed; | |
246 | } | |
247 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); | |
248 | ||
bf66f18e PM |
249 | /* |
250 | * Force a quiescent state for RCU BH. | |
251 | */ | |
252 | void rcu_bh_force_quiescent_state(void) | |
253 | { | |
254 | force_quiescent_state(&rcu_bh_state, 0); | |
255 | } | |
256 | EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state); | |
257 | ||
4a298656 PM |
258 | /* |
259 | * Record the number of times rcutorture tests have been initiated and | |
260 | * terminated. This information allows the debugfs tracing stats to be | |
261 | * correlated to the rcutorture messages, even when the rcutorture module | |
262 | * is being repeatedly loaded and unloaded. In other words, we cannot | |
263 | * store this state in rcutorture itself. | |
264 | */ | |
265 | void rcutorture_record_test_transition(void) | |
266 | { | |
267 | rcutorture_testseq++; | |
268 | rcutorture_vernum = 0; | |
269 | } | |
270 | EXPORT_SYMBOL_GPL(rcutorture_record_test_transition); | |
271 | ||
272 | /* | |
273 | * Record the number of writer passes through the current rcutorture test. | |
274 | * This is also used to correlate debugfs tracing stats with the rcutorture | |
275 | * messages. | |
276 | */ | |
277 | void rcutorture_record_progress(unsigned long vernum) | |
278 | { | |
279 | rcutorture_vernum++; | |
280 | } | |
281 | EXPORT_SYMBOL_GPL(rcutorture_record_progress); | |
282 | ||
bf66f18e PM |
283 | /* |
284 | * Force a quiescent state for RCU-sched. | |
285 | */ | |
286 | void rcu_sched_force_quiescent_state(void) | |
287 | { | |
288 | force_quiescent_state(&rcu_sched_state, 0); | |
289 | } | |
290 | EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state); | |
291 | ||
64db4cff PM |
292 | /* |
293 | * Does the CPU have callbacks ready to be invoked? | |
294 | */ | |
295 | static int | |
296 | cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) | |
297 | { | |
298 | return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]; | |
299 | } | |
300 | ||
301 | /* | |
302 | * Does the current CPU require a yet-as-unscheduled grace period? | |
303 | */ | |
304 | static int | |
305 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) | |
306 | { | |
fc2219d4 | 307 | return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp); |
64db4cff PM |
308 | } |
309 | ||
310 | /* | |
311 | * Return the root node of the specified rcu_state structure. | |
312 | */ | |
313 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) | |
314 | { | |
315 | return &rsp->node[0]; | |
316 | } | |
317 | ||
64db4cff PM |
318 | /* |
319 | * If the specified CPU is offline, tell the caller that it is in | |
320 | * a quiescent state. Otherwise, whack it with a reschedule IPI. | |
321 | * Grace periods can end up waiting on an offline CPU when that | |
322 | * CPU is in the process of coming online -- it will be added to the | |
323 | * rcu_node bitmasks before it actually makes it online. The same thing | |
324 | * can happen while a CPU is in the process of coming online. Because this | |
325 | * race is quite rare, we check for it after detecting that the grace | |
326 | * period has been delayed rather than checking each and every CPU | |
327 | * each and every time we start a new grace period. | |
328 | */ | |
329 | static int rcu_implicit_offline_qs(struct rcu_data *rdp) | |
330 | { | |
331 | /* | |
2036d94a PM |
332 | * If the CPU is offline for more than a jiffy, it is in a quiescent |
333 | * state. We can trust its state not to change because interrupts | |
334 | * are disabled. The reason for the jiffy's worth of slack is to | |
335 | * handle CPUs initializing on the way up and finding their way | |
336 | * to the idle loop on the way down. | |
64db4cff | 337 | */ |
2036d94a PM |
338 | if (cpu_is_offline(rdp->cpu) && |
339 | ULONG_CMP_LT(rdp->rsp->gp_start + 2, jiffies)) { | |
d4c08f2a | 340 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, "ofl"); |
64db4cff PM |
341 | rdp->offline_fqs++; |
342 | return 1; | |
343 | } | |
64db4cff PM |
344 | return 0; |
345 | } | |
346 | ||
9b2e4f18 PM |
347 | /* |
348 | * rcu_idle_enter_common - inform RCU that current CPU is moving towards idle | |
349 | * | |
350 | * If the new value of the ->dynticks_nesting counter now is zero, | |
351 | * we really have entered idle, and must do the appropriate accounting. | |
352 | * The caller must have disabled interrupts. | |
353 | */ | |
4145fa7f | 354 | static void rcu_idle_enter_common(struct rcu_dynticks *rdtp, long long oldval) |
9b2e4f18 | 355 | { |
facc4e15 | 356 | trace_rcu_dyntick("Start", oldval, 0); |
99745b6a | 357 | if (!is_idle_task(current)) { |
0989cb46 PM |
358 | struct task_struct *idle = idle_task(smp_processor_id()); |
359 | ||
facc4e15 | 360 | trace_rcu_dyntick("Error on entry: not idle task", oldval, 0); |
9b2e4f18 | 361 | ftrace_dump(DUMP_ALL); |
0989cb46 PM |
362 | WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s", |
363 | current->pid, current->comm, | |
364 | idle->pid, idle->comm); /* must be idle task! */ | |
9b2e4f18 | 365 | } |
aea1b35e | 366 | rcu_prepare_for_idle(smp_processor_id()); |
9b2e4f18 PM |
367 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ |
368 | smp_mb__before_atomic_inc(); /* See above. */ | |
369 | atomic_inc(&rdtp->dynticks); | |
370 | smp_mb__after_atomic_inc(); /* Force ordering with next sojourn. */ | |
371 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); | |
c44e2cdd PM |
372 | |
373 | /* | |
374 | * The idle task is not permitted to enter the idle loop while | |
375 | * in an RCU read-side critical section. | |
376 | */ | |
377 | rcu_lockdep_assert(!lock_is_held(&rcu_lock_map), | |
378 | "Illegal idle entry in RCU read-side critical section."); | |
379 | rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map), | |
380 | "Illegal idle entry in RCU-bh read-side critical section."); | |
381 | rcu_lockdep_assert(!lock_is_held(&rcu_sched_lock_map), | |
382 | "Illegal idle entry in RCU-sched read-side critical section."); | |
9b2e4f18 | 383 | } |
64db4cff PM |
384 | |
385 | /** | |
9b2e4f18 | 386 | * rcu_idle_enter - inform RCU that current CPU is entering idle |
64db4cff | 387 | * |
9b2e4f18 | 388 | * Enter idle mode, in other words, -leave- the mode in which RCU |
64db4cff | 389 | * read-side critical sections can occur. (Though RCU read-side |
9b2e4f18 PM |
390 | * critical sections can occur in irq handlers in idle, a possibility |
391 | * handled by irq_enter() and irq_exit().) | |
392 | * | |
393 | * We crowbar the ->dynticks_nesting field to zero to allow for | |
394 | * the possibility of usermode upcalls having messed up our count | |
395 | * of interrupt nesting level during the prior busy period. | |
64db4cff | 396 | */ |
9b2e4f18 | 397 | void rcu_idle_enter(void) |
64db4cff PM |
398 | { |
399 | unsigned long flags; | |
4145fa7f | 400 | long long oldval; |
64db4cff PM |
401 | struct rcu_dynticks *rdtp; |
402 | ||
64db4cff PM |
403 | local_irq_save(flags); |
404 | rdtp = &__get_cpu_var(rcu_dynticks); | |
4145fa7f | 405 | oldval = rdtp->dynticks_nesting; |
29e37d81 PM |
406 | WARN_ON_ONCE((oldval & DYNTICK_TASK_NEST_MASK) == 0); |
407 | if ((oldval & DYNTICK_TASK_NEST_MASK) == DYNTICK_TASK_NEST_VALUE) | |
408 | rdtp->dynticks_nesting = 0; | |
409 | else | |
410 | rdtp->dynticks_nesting -= DYNTICK_TASK_NEST_VALUE; | |
4145fa7f | 411 | rcu_idle_enter_common(rdtp, oldval); |
64db4cff PM |
412 | local_irq_restore(flags); |
413 | } | |
8a2ecf47 | 414 | EXPORT_SYMBOL_GPL(rcu_idle_enter); |
64db4cff | 415 | |
9b2e4f18 PM |
416 | /** |
417 | * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle | |
418 | * | |
419 | * Exit from an interrupt handler, which might possibly result in entering | |
420 | * idle mode, in other words, leaving the mode in which read-side critical | |
421 | * sections can occur. | |
64db4cff | 422 | * |
9b2e4f18 PM |
423 | * This code assumes that the idle loop never does anything that might |
424 | * result in unbalanced calls to irq_enter() and irq_exit(). If your | |
425 | * architecture violates this assumption, RCU will give you what you | |
426 | * deserve, good and hard. But very infrequently and irreproducibly. | |
427 | * | |
428 | * Use things like work queues to work around this limitation. | |
429 | * | |
430 | * You have been warned. | |
64db4cff | 431 | */ |
9b2e4f18 | 432 | void rcu_irq_exit(void) |
64db4cff PM |
433 | { |
434 | unsigned long flags; | |
4145fa7f | 435 | long long oldval; |
64db4cff PM |
436 | struct rcu_dynticks *rdtp; |
437 | ||
438 | local_irq_save(flags); | |
439 | rdtp = &__get_cpu_var(rcu_dynticks); | |
4145fa7f | 440 | oldval = rdtp->dynticks_nesting; |
9b2e4f18 PM |
441 | rdtp->dynticks_nesting--; |
442 | WARN_ON_ONCE(rdtp->dynticks_nesting < 0); | |
b6fc6020 FW |
443 | if (rdtp->dynticks_nesting) |
444 | trace_rcu_dyntick("--=", oldval, rdtp->dynticks_nesting); | |
445 | else | |
446 | rcu_idle_enter_common(rdtp, oldval); | |
9b2e4f18 PM |
447 | local_irq_restore(flags); |
448 | } | |
449 | ||
450 | /* | |
451 | * rcu_idle_exit_common - inform RCU that current CPU is moving away from idle | |
452 | * | |
453 | * If the new value of the ->dynticks_nesting counter was previously zero, | |
454 | * we really have exited idle, and must do the appropriate accounting. | |
455 | * The caller must have disabled interrupts. | |
456 | */ | |
457 | static void rcu_idle_exit_common(struct rcu_dynticks *rdtp, long long oldval) | |
458 | { | |
23b5c8fa PM |
459 | smp_mb__before_atomic_inc(); /* Force ordering w/previous sojourn. */ |
460 | atomic_inc(&rdtp->dynticks); | |
461 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ | |
462 | smp_mb__after_atomic_inc(); /* See above. */ | |
463 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); | |
7cb92499 | 464 | rcu_cleanup_after_idle(smp_processor_id()); |
4145fa7f | 465 | trace_rcu_dyntick("End", oldval, rdtp->dynticks_nesting); |
99745b6a | 466 | if (!is_idle_task(current)) { |
0989cb46 PM |
467 | struct task_struct *idle = idle_task(smp_processor_id()); |
468 | ||
4145fa7f PM |
469 | trace_rcu_dyntick("Error on exit: not idle task", |
470 | oldval, rdtp->dynticks_nesting); | |
9b2e4f18 | 471 | ftrace_dump(DUMP_ALL); |
0989cb46 PM |
472 | WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s", |
473 | current->pid, current->comm, | |
474 | idle->pid, idle->comm); /* must be idle task! */ | |
9b2e4f18 PM |
475 | } |
476 | } | |
477 | ||
478 | /** | |
479 | * rcu_idle_exit - inform RCU that current CPU is leaving idle | |
480 | * | |
481 | * Exit idle mode, in other words, -enter- the mode in which RCU | |
482 | * read-side critical sections can occur. | |
483 | * | |
29e37d81 | 484 | * We crowbar the ->dynticks_nesting field to DYNTICK_TASK_NEST to |
4145fa7f | 485 | * allow for the possibility of usermode upcalls messing up our count |
9b2e4f18 PM |
486 | * of interrupt nesting level during the busy period that is just |
487 | * now starting. | |
488 | */ | |
489 | void rcu_idle_exit(void) | |
490 | { | |
491 | unsigned long flags; | |
492 | struct rcu_dynticks *rdtp; | |
493 | long long oldval; | |
494 | ||
495 | local_irq_save(flags); | |
496 | rdtp = &__get_cpu_var(rcu_dynticks); | |
497 | oldval = rdtp->dynticks_nesting; | |
29e37d81 PM |
498 | WARN_ON_ONCE(oldval < 0); |
499 | if (oldval & DYNTICK_TASK_NEST_MASK) | |
500 | rdtp->dynticks_nesting += DYNTICK_TASK_NEST_VALUE; | |
501 | else | |
502 | rdtp->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE; | |
9b2e4f18 PM |
503 | rcu_idle_exit_common(rdtp, oldval); |
504 | local_irq_restore(flags); | |
505 | } | |
8a2ecf47 | 506 | EXPORT_SYMBOL_GPL(rcu_idle_exit); |
9b2e4f18 PM |
507 | |
508 | /** | |
509 | * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle | |
510 | * | |
511 | * Enter an interrupt handler, which might possibly result in exiting | |
512 | * idle mode, in other words, entering the mode in which read-side critical | |
513 | * sections can occur. | |
514 | * | |
515 | * Note that the Linux kernel is fully capable of entering an interrupt | |
516 | * handler that it never exits, for example when doing upcalls to | |
517 | * user mode! This code assumes that the idle loop never does upcalls to | |
518 | * user mode. If your architecture does do upcalls from the idle loop (or | |
519 | * does anything else that results in unbalanced calls to the irq_enter() | |
520 | * and irq_exit() functions), RCU will give you what you deserve, good | |
521 | * and hard. But very infrequently and irreproducibly. | |
522 | * | |
523 | * Use things like work queues to work around this limitation. | |
524 | * | |
525 | * You have been warned. | |
526 | */ | |
527 | void rcu_irq_enter(void) | |
528 | { | |
529 | unsigned long flags; | |
530 | struct rcu_dynticks *rdtp; | |
531 | long long oldval; | |
532 | ||
533 | local_irq_save(flags); | |
534 | rdtp = &__get_cpu_var(rcu_dynticks); | |
535 | oldval = rdtp->dynticks_nesting; | |
536 | rdtp->dynticks_nesting++; | |
537 | WARN_ON_ONCE(rdtp->dynticks_nesting == 0); | |
b6fc6020 FW |
538 | if (oldval) |
539 | trace_rcu_dyntick("++=", oldval, rdtp->dynticks_nesting); | |
540 | else | |
541 | rcu_idle_exit_common(rdtp, oldval); | |
64db4cff | 542 | local_irq_restore(flags); |
64db4cff PM |
543 | } |
544 | ||
545 | /** | |
546 | * rcu_nmi_enter - inform RCU of entry to NMI context | |
547 | * | |
548 | * If the CPU was idle with dynamic ticks active, and there is no | |
549 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
550 | * RCU grace-period handling know that the CPU is active. | |
551 | */ | |
552 | void rcu_nmi_enter(void) | |
553 | { | |
554 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
555 | ||
23b5c8fa PM |
556 | if (rdtp->dynticks_nmi_nesting == 0 && |
557 | (atomic_read(&rdtp->dynticks) & 0x1)) | |
64db4cff | 558 | return; |
23b5c8fa PM |
559 | rdtp->dynticks_nmi_nesting++; |
560 | smp_mb__before_atomic_inc(); /* Force delay from prior write. */ | |
561 | atomic_inc(&rdtp->dynticks); | |
562 | /* CPUs seeing atomic_inc() must see later RCU read-side crit sects */ | |
563 | smp_mb__after_atomic_inc(); /* See above. */ | |
564 | WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); | |
64db4cff PM |
565 | } |
566 | ||
567 | /** | |
568 | * rcu_nmi_exit - inform RCU of exit from NMI context | |
569 | * | |
570 | * If the CPU was idle with dynamic ticks active, and there is no | |
571 | * irq handler running, this updates rdtp->dynticks_nmi to let the | |
572 | * RCU grace-period handling know that the CPU is no longer active. | |
573 | */ | |
574 | void rcu_nmi_exit(void) | |
575 | { | |
576 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); | |
577 | ||
23b5c8fa PM |
578 | if (rdtp->dynticks_nmi_nesting == 0 || |
579 | --rdtp->dynticks_nmi_nesting != 0) | |
64db4cff | 580 | return; |
23b5c8fa PM |
581 | /* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */ |
582 | smp_mb__before_atomic_inc(); /* See above. */ | |
583 | atomic_inc(&rdtp->dynticks); | |
584 | smp_mb__after_atomic_inc(); /* Force delay to next write. */ | |
585 | WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); | |
64db4cff PM |
586 | } |
587 | ||
9b2e4f18 PM |
588 | #ifdef CONFIG_PROVE_RCU |
589 | ||
64db4cff | 590 | /** |
9b2e4f18 | 591 | * rcu_is_cpu_idle - see if RCU thinks that the current CPU is idle |
64db4cff | 592 | * |
9b2e4f18 | 593 | * If the current CPU is in its idle loop and is neither in an interrupt |
34240697 | 594 | * or NMI handler, return true. |
64db4cff | 595 | */ |
9b2e4f18 | 596 | int rcu_is_cpu_idle(void) |
64db4cff | 597 | { |
34240697 PM |
598 | int ret; |
599 | ||
600 | preempt_disable(); | |
601 | ret = (atomic_read(&__get_cpu_var(rcu_dynticks).dynticks) & 0x1) == 0; | |
602 | preempt_enable(); | |
603 | return ret; | |
64db4cff | 604 | } |
e6b80a3b | 605 | EXPORT_SYMBOL(rcu_is_cpu_idle); |
64db4cff | 606 | |
c0d6d01b PM |
607 | #ifdef CONFIG_HOTPLUG_CPU |
608 | ||
609 | /* | |
610 | * Is the current CPU online? Disable preemption to avoid false positives | |
611 | * that could otherwise happen due to the current CPU number being sampled, | |
612 | * this task being preempted, its old CPU being taken offline, resuming | |
613 | * on some other CPU, then determining that its old CPU is now offline. | |
614 | * It is OK to use RCU on an offline processor during initial boot, hence | |
2036d94a PM |
615 | * the check for rcu_scheduler_fully_active. Note also that it is OK |
616 | * for a CPU coming online to use RCU for one jiffy prior to marking itself | |
617 | * online in the cpu_online_mask. Similarly, it is OK for a CPU going | |
618 | * offline to continue to use RCU for one jiffy after marking itself | |
619 | * offline in the cpu_online_mask. This leniency is necessary given the | |
620 | * non-atomic nature of the online and offline processing, for example, | |
621 | * the fact that a CPU enters the scheduler after completing the CPU_DYING | |
622 | * notifiers. | |
623 | * | |
624 | * This is also why RCU internally marks CPUs online during the | |
625 | * CPU_UP_PREPARE phase and offline during the CPU_DEAD phase. | |
c0d6d01b PM |
626 | * |
627 | * Disable checking if in an NMI handler because we cannot safely report | |
628 | * errors from NMI handlers anyway. | |
629 | */ | |
630 | bool rcu_lockdep_current_cpu_online(void) | |
631 | { | |
2036d94a PM |
632 | struct rcu_data *rdp; |
633 | struct rcu_node *rnp; | |
c0d6d01b PM |
634 | bool ret; |
635 | ||
636 | if (in_nmi()) | |
637 | return 1; | |
638 | preempt_disable(); | |
2036d94a PM |
639 | rdp = &__get_cpu_var(rcu_sched_data); |
640 | rnp = rdp->mynode; | |
641 | ret = (rdp->grpmask & rnp->qsmaskinit) || | |
c0d6d01b PM |
642 | !rcu_scheduler_fully_active; |
643 | preempt_enable(); | |
644 | return ret; | |
645 | } | |
646 | EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online); | |
647 | ||
648 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | |
649 | ||
9b2e4f18 PM |
650 | #endif /* #ifdef CONFIG_PROVE_RCU */ |
651 | ||
64db4cff | 652 | /** |
9b2e4f18 | 653 | * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle |
64db4cff | 654 | * |
9b2e4f18 PM |
655 | * If the current CPU is idle or running at a first-level (not nested) |
656 | * interrupt from idle, return true. The caller must have at least | |
657 | * disabled preemption. | |
64db4cff | 658 | */ |
9b2e4f18 | 659 | int rcu_is_cpu_rrupt_from_idle(void) |
64db4cff | 660 | { |
9b2e4f18 | 661 | return __get_cpu_var(rcu_dynticks).dynticks_nesting <= 1; |
64db4cff PM |
662 | } |
663 | ||
64db4cff PM |
664 | /* |
665 | * Snapshot the specified CPU's dynticks counter so that we can later | |
666 | * credit them with an implicit quiescent state. Return 1 if this CPU | |
1eba8f84 | 667 | * is in dynticks idle mode, which is an extended quiescent state. |
64db4cff PM |
668 | */ |
669 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | |
670 | { | |
23b5c8fa | 671 | rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks); |
f0e7c19d | 672 | return (rdp->dynticks_snap & 0x1) == 0; |
64db4cff PM |
673 | } |
674 | ||
675 | /* | |
676 | * Return true if the specified CPU has passed through a quiescent | |
677 | * state by virtue of being in or having passed through an dynticks | |
678 | * idle state since the last call to dyntick_save_progress_counter() | |
679 | * for this same CPU. | |
680 | */ | |
681 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) | |
682 | { | |
7eb4f455 PM |
683 | unsigned int curr; |
684 | unsigned int snap; | |
64db4cff | 685 | |
7eb4f455 PM |
686 | curr = (unsigned int)atomic_add_return(0, &rdp->dynticks->dynticks); |
687 | snap = (unsigned int)rdp->dynticks_snap; | |
64db4cff PM |
688 | |
689 | /* | |
690 | * If the CPU passed through or entered a dynticks idle phase with | |
691 | * no active irq/NMI handlers, then we can safely pretend that the CPU | |
692 | * already acknowledged the request to pass through a quiescent | |
693 | * state. Either way, that CPU cannot possibly be in an RCU | |
694 | * read-side critical section that started before the beginning | |
695 | * of the current RCU grace period. | |
696 | */ | |
7eb4f455 | 697 | if ((curr & 0x1) == 0 || UINT_CMP_GE(curr, snap + 2)) { |
d4c08f2a | 698 | trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, "dti"); |
64db4cff PM |
699 | rdp->dynticks_fqs++; |
700 | return 1; | |
701 | } | |
702 | ||
703 | /* Go check for the CPU being offline. */ | |
704 | return rcu_implicit_offline_qs(rdp); | |
705 | } | |
706 | ||
13cfcca0 PM |
707 | static int jiffies_till_stall_check(void) |
708 | { | |
709 | int till_stall_check = ACCESS_ONCE(rcu_cpu_stall_timeout); | |
710 | ||
711 | /* | |
712 | * Limit check must be consistent with the Kconfig limits | |
713 | * for CONFIG_RCU_CPU_STALL_TIMEOUT. | |
714 | */ | |
715 | if (till_stall_check < 3) { | |
716 | ACCESS_ONCE(rcu_cpu_stall_timeout) = 3; | |
717 | till_stall_check = 3; | |
718 | } else if (till_stall_check > 300) { | |
719 | ACCESS_ONCE(rcu_cpu_stall_timeout) = 300; | |
720 | till_stall_check = 300; | |
721 | } | |
722 | return till_stall_check * HZ + RCU_STALL_DELAY_DELTA; | |
723 | } | |
724 | ||
64db4cff PM |
725 | static void record_gp_stall_check_time(struct rcu_state *rsp) |
726 | { | |
727 | rsp->gp_start = jiffies; | |
13cfcca0 | 728 | rsp->jiffies_stall = jiffies + jiffies_till_stall_check(); |
64db4cff PM |
729 | } |
730 | ||
731 | static void print_other_cpu_stall(struct rcu_state *rsp) | |
732 | { | |
733 | int cpu; | |
734 | long delta; | |
735 | unsigned long flags; | |
285fe294 | 736 | int ndetected = 0; |
64db4cff | 737 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff PM |
738 | |
739 | /* Only let one CPU complain about others per time interval. */ | |
740 | ||
1304afb2 | 741 | raw_spin_lock_irqsave(&rnp->lock, flags); |
64db4cff | 742 | delta = jiffies - rsp->jiffies_stall; |
fc2219d4 | 743 | if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) { |
1304afb2 | 744 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
745 | return; |
746 | } | |
13cfcca0 | 747 | rsp->jiffies_stall = jiffies + 3 * jiffies_till_stall_check() + 3; |
1304afb2 | 748 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 749 | |
8cdd32a9 PM |
750 | /* |
751 | * OK, time to rat on our buddy... | |
752 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
753 | * RCU CPU stall warnings. | |
754 | */ | |
a858af28 | 755 | printk(KERN_ERR "INFO: %s detected stalls on CPUs/tasks:", |
4300aa64 | 756 | rsp->name); |
a858af28 | 757 | print_cpu_stall_info_begin(); |
a0b6c9a7 | 758 | rcu_for_each_leaf_node(rsp, rnp) { |
3acd9eb3 | 759 | raw_spin_lock_irqsave(&rnp->lock, flags); |
9bc8b558 | 760 | ndetected += rcu_print_task_stall(rnp); |
3acd9eb3 | 761 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
a0b6c9a7 | 762 | if (rnp->qsmask == 0) |
64db4cff | 763 | continue; |
a0b6c9a7 | 764 | for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++) |
9bc8b558 | 765 | if (rnp->qsmask & (1UL << cpu)) { |
a858af28 | 766 | print_cpu_stall_info(rsp, rnp->grplo + cpu); |
9bc8b558 PM |
767 | ndetected++; |
768 | } | |
64db4cff | 769 | } |
a858af28 PM |
770 | |
771 | /* | |
772 | * Now rat on any tasks that got kicked up to the root rcu_node | |
773 | * due to CPU offlining. | |
774 | */ | |
775 | rnp = rcu_get_root(rsp); | |
776 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
285fe294 | 777 | ndetected += rcu_print_task_stall(rnp); |
a858af28 PM |
778 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
779 | ||
780 | print_cpu_stall_info_end(); | |
781 | printk(KERN_CONT "(detected by %d, t=%ld jiffies)\n", | |
64db4cff | 782 | smp_processor_id(), (long)(jiffies - rsp->gp_start)); |
9bc8b558 PM |
783 | if (ndetected == 0) |
784 | printk(KERN_ERR "INFO: Stall ended before state dump start\n"); | |
785 | else if (!trigger_all_cpu_backtrace()) | |
4627e240 | 786 | dump_stack(); |
c1dc0b9c | 787 | |
1ed509a2 PM |
788 | /* If so configured, complain about tasks blocking the grace period. */ |
789 | ||
790 | rcu_print_detail_task_stall(rsp); | |
791 | ||
64db4cff PM |
792 | force_quiescent_state(rsp, 0); /* Kick them all. */ |
793 | } | |
794 | ||
795 | static void print_cpu_stall(struct rcu_state *rsp) | |
796 | { | |
797 | unsigned long flags; | |
798 | struct rcu_node *rnp = rcu_get_root(rsp); | |
799 | ||
8cdd32a9 PM |
800 | /* |
801 | * OK, time to rat on ourselves... | |
802 | * See Documentation/RCU/stallwarn.txt for info on how to debug | |
803 | * RCU CPU stall warnings. | |
804 | */ | |
a858af28 PM |
805 | printk(KERN_ERR "INFO: %s self-detected stall on CPU", rsp->name); |
806 | print_cpu_stall_info_begin(); | |
807 | print_cpu_stall_info(rsp, smp_processor_id()); | |
808 | print_cpu_stall_info_end(); | |
809 | printk(KERN_CONT " (t=%lu jiffies)\n", jiffies - rsp->gp_start); | |
4627e240 PM |
810 | if (!trigger_all_cpu_backtrace()) |
811 | dump_stack(); | |
c1dc0b9c | 812 | |
1304afb2 | 813 | raw_spin_lock_irqsave(&rnp->lock, flags); |
20133cfc | 814 | if (ULONG_CMP_GE(jiffies, rsp->jiffies_stall)) |
13cfcca0 PM |
815 | rsp->jiffies_stall = jiffies + |
816 | 3 * jiffies_till_stall_check() + 3; | |
1304afb2 | 817 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
c1dc0b9c | 818 | |
64db4cff PM |
819 | set_need_resched(); /* kick ourselves to get things going. */ |
820 | } | |
821 | ||
822 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) | |
823 | { | |
bad6e139 PM |
824 | unsigned long j; |
825 | unsigned long js; | |
64db4cff PM |
826 | struct rcu_node *rnp; |
827 | ||
742734ee | 828 | if (rcu_cpu_stall_suppress) |
c68de209 | 829 | return; |
bad6e139 PM |
830 | j = ACCESS_ONCE(jiffies); |
831 | js = ACCESS_ONCE(rsp->jiffies_stall); | |
64db4cff | 832 | rnp = rdp->mynode; |
bad6e139 | 833 | if ((ACCESS_ONCE(rnp->qsmask) & rdp->grpmask) && ULONG_CMP_GE(j, js)) { |
64db4cff PM |
834 | |
835 | /* We haven't checked in, so go dump stack. */ | |
836 | print_cpu_stall(rsp); | |
837 | ||
bad6e139 PM |
838 | } else if (rcu_gp_in_progress(rsp) && |
839 | ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) { | |
64db4cff | 840 | |
bad6e139 | 841 | /* They had a few time units to dump stack, so complain. */ |
64db4cff PM |
842 | print_other_cpu_stall(rsp); |
843 | } | |
844 | } | |
845 | ||
c68de209 PM |
846 | static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr) |
847 | { | |
742734ee | 848 | rcu_cpu_stall_suppress = 1; |
c68de209 PM |
849 | return NOTIFY_DONE; |
850 | } | |
851 | ||
53d84e00 PM |
852 | /** |
853 | * rcu_cpu_stall_reset - prevent further stall warnings in current grace period | |
854 | * | |
855 | * Set the stall-warning timeout way off into the future, thus preventing | |
856 | * any RCU CPU stall-warning messages from appearing in the current set of | |
857 | * RCU grace periods. | |
858 | * | |
859 | * The caller must disable hard irqs. | |
860 | */ | |
861 | void rcu_cpu_stall_reset(void) | |
862 | { | |
863 | rcu_sched_state.jiffies_stall = jiffies + ULONG_MAX / 2; | |
864 | rcu_bh_state.jiffies_stall = jiffies + ULONG_MAX / 2; | |
865 | rcu_preempt_stall_reset(); | |
866 | } | |
867 | ||
c68de209 PM |
868 | static struct notifier_block rcu_panic_block = { |
869 | .notifier_call = rcu_panic, | |
870 | }; | |
871 | ||
872 | static void __init check_cpu_stall_init(void) | |
873 | { | |
874 | atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block); | |
875 | } | |
876 | ||
64db4cff PM |
877 | /* |
878 | * Update CPU-local rcu_data state to record the newly noticed grace period. | |
879 | * This is used both when we started the grace period and when we notice | |
9160306e PM |
880 | * that someone else started the grace period. The caller must hold the |
881 | * ->lock of the leaf rcu_node structure corresponding to the current CPU, | |
882 | * and must have irqs disabled. | |
64db4cff | 883 | */ |
9160306e PM |
884 | static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) |
885 | { | |
886 | if (rdp->gpnum != rnp->gpnum) { | |
121dfc4b PM |
887 | /* |
888 | * If the current grace period is waiting for this CPU, | |
889 | * set up to detect a quiescent state, otherwise don't | |
890 | * go looking for one. | |
891 | */ | |
9160306e | 892 | rdp->gpnum = rnp->gpnum; |
d4c08f2a | 893 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpustart"); |
121dfc4b PM |
894 | if (rnp->qsmask & rdp->grpmask) { |
895 | rdp->qs_pending = 1; | |
e4cc1f22 | 896 | rdp->passed_quiesce = 0; |
121dfc4b PM |
897 | } else |
898 | rdp->qs_pending = 0; | |
a858af28 | 899 | zero_cpu_stall_ticks(rdp); |
9160306e PM |
900 | } |
901 | } | |
902 | ||
64db4cff PM |
903 | static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp) |
904 | { | |
9160306e PM |
905 | unsigned long flags; |
906 | struct rcu_node *rnp; | |
907 | ||
908 | local_irq_save(flags); | |
909 | rnp = rdp->mynode; | |
910 | if (rdp->gpnum == ACCESS_ONCE(rnp->gpnum) || /* outside lock. */ | |
1304afb2 | 911 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
9160306e PM |
912 | local_irq_restore(flags); |
913 | return; | |
914 | } | |
915 | __note_new_gpnum(rsp, rnp, rdp); | |
1304afb2 | 916 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
917 | } |
918 | ||
919 | /* | |
920 | * Did someone else start a new RCU grace period start since we last | |
921 | * checked? Update local state appropriately if so. Must be called | |
922 | * on the CPU corresponding to rdp. | |
923 | */ | |
924 | static int | |
925 | check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp) | |
926 | { | |
927 | unsigned long flags; | |
928 | int ret = 0; | |
929 | ||
930 | local_irq_save(flags); | |
931 | if (rdp->gpnum != rsp->gpnum) { | |
932 | note_new_gpnum(rsp, rdp); | |
933 | ret = 1; | |
934 | } | |
935 | local_irq_restore(flags); | |
936 | return ret; | |
937 | } | |
938 | ||
3f5d3ea6 PM |
939 | /* |
940 | * Initialize the specified rcu_data structure's callback list to empty. | |
941 | */ | |
942 | static void init_callback_list(struct rcu_data *rdp) | |
943 | { | |
944 | int i; | |
945 | ||
946 | rdp->nxtlist = NULL; | |
947 | for (i = 0; i < RCU_NEXT_SIZE; i++) | |
948 | rdp->nxttail[i] = &rdp->nxtlist; | |
949 | } | |
950 | ||
d09b62df PM |
951 | /* |
952 | * Advance this CPU's callbacks, but only if the current grace period | |
953 | * has ended. This may be called only from the CPU to whom the rdp | |
954 | * belongs. In addition, the corresponding leaf rcu_node structure's | |
955 | * ->lock must be held by the caller, with irqs disabled. | |
956 | */ | |
957 | static void | |
958 | __rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
959 | { | |
960 | /* Did another grace period end? */ | |
961 | if (rdp->completed != rnp->completed) { | |
962 | ||
963 | /* Advance callbacks. No harm if list empty. */ | |
964 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL]; | |
965 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL]; | |
966 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
967 | ||
968 | /* Remember that we saw this grace-period completion. */ | |
969 | rdp->completed = rnp->completed; | |
d4c08f2a | 970 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuend"); |
20377f32 | 971 | |
5ff8e6f0 FW |
972 | /* |
973 | * If we were in an extended quiescent state, we may have | |
121dfc4b | 974 | * missed some grace periods that others CPUs handled on |
5ff8e6f0 | 975 | * our behalf. Catch up with this state to avoid noting |
121dfc4b PM |
976 | * spurious new grace periods. If another grace period |
977 | * has started, then rnp->gpnum will have advanced, so | |
978 | * we will detect this later on. | |
5ff8e6f0 | 979 | */ |
121dfc4b | 980 | if (ULONG_CMP_LT(rdp->gpnum, rdp->completed)) |
5ff8e6f0 FW |
981 | rdp->gpnum = rdp->completed; |
982 | ||
20377f32 | 983 | /* |
121dfc4b PM |
984 | * If RCU does not need a quiescent state from this CPU, |
985 | * then make sure that this CPU doesn't go looking for one. | |
20377f32 | 986 | */ |
121dfc4b | 987 | if ((rnp->qsmask & rdp->grpmask) == 0) |
20377f32 | 988 | rdp->qs_pending = 0; |
d09b62df PM |
989 | } |
990 | } | |
991 | ||
992 | /* | |
993 | * Advance this CPU's callbacks, but only if the current grace period | |
994 | * has ended. This may be called only from the CPU to whom the rdp | |
995 | * belongs. | |
996 | */ | |
997 | static void | |
998 | rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp) | |
999 | { | |
1000 | unsigned long flags; | |
1001 | struct rcu_node *rnp; | |
1002 | ||
1003 | local_irq_save(flags); | |
1004 | rnp = rdp->mynode; | |
1005 | if (rdp->completed == ACCESS_ONCE(rnp->completed) || /* outside lock. */ | |
1304afb2 | 1006 | !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */ |
d09b62df PM |
1007 | local_irq_restore(flags); |
1008 | return; | |
1009 | } | |
1010 | __rcu_process_gp_end(rsp, rnp, rdp); | |
1304afb2 | 1011 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
d09b62df PM |
1012 | } |
1013 | ||
1014 | /* | |
1015 | * Do per-CPU grace-period initialization for running CPU. The caller | |
1016 | * must hold the lock of the leaf rcu_node structure corresponding to | |
1017 | * this CPU. | |
1018 | */ | |
1019 | static void | |
1020 | rcu_start_gp_per_cpu(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp) | |
1021 | { | |
1022 | /* Prior grace period ended, so advance callbacks for current CPU. */ | |
1023 | __rcu_process_gp_end(rsp, rnp, rdp); | |
1024 | ||
1025 | /* | |
1026 | * Because this CPU just now started the new grace period, we know | |
1027 | * that all of its callbacks will be covered by this upcoming grace | |
1028 | * period, even the ones that were registered arbitrarily recently. | |
1029 | * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL. | |
1030 | * | |
1031 | * Other CPUs cannot be sure exactly when the grace period started. | |
1032 | * Therefore, their recently registered callbacks must pass through | |
1033 | * an additional RCU_NEXT_READY stage, so that they will be handled | |
1034 | * by the next RCU grace period. | |
1035 | */ | |
1036 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
1037 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; | |
9160306e PM |
1038 | |
1039 | /* Set state so that this CPU will detect the next quiescent state. */ | |
1040 | __note_new_gpnum(rsp, rnp, rdp); | |
d09b62df PM |
1041 | } |
1042 | ||
64db4cff PM |
1043 | /* |
1044 | * Start a new RCU grace period if warranted, re-initializing the hierarchy | |
1045 | * in preparation for detecting the next grace period. The caller must hold | |
1046 | * the root node's ->lock, which is released before return. Hard irqs must | |
1047 | * be disabled. | |
e5601400 PM |
1048 | * |
1049 | * Note that it is legal for a dying CPU (which is marked as offline) to | |
1050 | * invoke this function. This can happen when the dying CPU reports its | |
1051 | * quiescent state. | |
64db4cff PM |
1052 | */ |
1053 | static void | |
1054 | rcu_start_gp(struct rcu_state *rsp, unsigned long flags) | |
1055 | __releases(rcu_get_root(rsp)->lock) | |
1056 | { | |
394f99a9 | 1057 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); |
64db4cff | 1058 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff | 1059 | |
037067a1 | 1060 | if (!rcu_scheduler_fully_active || |
afe24b12 PM |
1061 | !cpu_needs_another_gp(rsp, rdp)) { |
1062 | /* | |
1063 | * Either the scheduler hasn't yet spawned the first | |
1064 | * non-idle task or this CPU does not need another | |
1065 | * grace period. Either way, don't start a new grace | |
1066 | * period. | |
1067 | */ | |
1068 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
1069 | return; | |
1070 | } | |
b32e9eb6 | 1071 | |
afe24b12 | 1072 | if (rsp->fqs_active) { |
b32e9eb6 | 1073 | /* |
afe24b12 PM |
1074 | * This CPU needs a grace period, but force_quiescent_state() |
1075 | * is running. Tell it to start one on this CPU's behalf. | |
b32e9eb6 | 1076 | */ |
afe24b12 PM |
1077 | rsp->fqs_need_gp = 1; |
1078 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
64db4cff PM |
1079 | return; |
1080 | } | |
1081 | ||
1082 | /* Advance to a new grace period and initialize state. */ | |
1083 | rsp->gpnum++; | |
d4c08f2a | 1084 | trace_rcu_grace_period(rsp->name, rsp->gpnum, "start"); |
af446b70 PM |
1085 | WARN_ON_ONCE(rsp->fqs_state == RCU_GP_INIT); |
1086 | rsp->fqs_state = RCU_GP_INIT; /* Hold off force_quiescent_state. */ | |
64db4cff | 1087 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
64db4cff | 1088 | record_gp_stall_check_time(rsp); |
1304afb2 | 1089 | raw_spin_unlock(&rnp->lock); /* leave irqs disabled. */ |
64db4cff | 1090 | |
64db4cff | 1091 | /* Exclude any concurrent CPU-hotplug operations. */ |
1304afb2 | 1092 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
64db4cff PM |
1093 | |
1094 | /* | |
b835db1f PM |
1095 | * Set the quiescent-state-needed bits in all the rcu_node |
1096 | * structures for all currently online CPUs in breadth-first | |
1097 | * order, starting from the root rcu_node structure. This | |
1098 | * operation relies on the layout of the hierarchy within the | |
1099 | * rsp->node[] array. Note that other CPUs will access only | |
1100 | * the leaves of the hierarchy, which still indicate that no | |
1101 | * grace period is in progress, at least until the corresponding | |
1102 | * leaf node has been initialized. In addition, we have excluded | |
1103 | * CPU-hotplug operations. | |
64db4cff PM |
1104 | * |
1105 | * Note that the grace period cannot complete until we finish | |
1106 | * the initialization process, as there will be at least one | |
1107 | * qsmask bit set in the root node until that time, namely the | |
b835db1f PM |
1108 | * one corresponding to this CPU, due to the fact that we have |
1109 | * irqs disabled. | |
64db4cff | 1110 | */ |
a0b6c9a7 | 1111 | rcu_for_each_node_breadth_first(rsp, rnp) { |
1304afb2 | 1112 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
b0e165c0 | 1113 | rcu_preempt_check_blocked_tasks(rnp); |
49e29126 | 1114 | rnp->qsmask = rnp->qsmaskinit; |
de078d87 | 1115 | rnp->gpnum = rsp->gpnum; |
d09b62df PM |
1116 | rnp->completed = rsp->completed; |
1117 | if (rnp == rdp->mynode) | |
1118 | rcu_start_gp_per_cpu(rsp, rnp, rdp); | |
27f4d280 | 1119 | rcu_preempt_boost_start_gp(rnp); |
d4c08f2a PM |
1120 | trace_rcu_grace_period_init(rsp->name, rnp->gpnum, |
1121 | rnp->level, rnp->grplo, | |
1122 | rnp->grphi, rnp->qsmask); | |
1304afb2 | 1123 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
1124 | } |
1125 | ||
83f5b01f | 1126 | rnp = rcu_get_root(rsp); |
1304afb2 | 1127 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
af446b70 | 1128 | rsp->fqs_state = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */ |
1304afb2 PM |
1129 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
1130 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); | |
64db4cff PM |
1131 | } |
1132 | ||
f41d911f | 1133 | /* |
d3f6bad3 PM |
1134 | * Report a full set of quiescent states to the specified rcu_state |
1135 | * data structure. This involves cleaning up after the prior grace | |
1136 | * period and letting rcu_start_gp() start up the next grace period | |
1137 | * if one is needed. Note that the caller must hold rnp->lock, as | |
1138 | * required by rcu_start_gp(), which will release it. | |
f41d911f | 1139 | */ |
d3f6bad3 | 1140 | static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags) |
fc2219d4 | 1141 | __releases(rcu_get_root(rsp)->lock) |
f41d911f | 1142 | { |
15ba0ba8 | 1143 | unsigned long gp_duration; |
afe24b12 PM |
1144 | struct rcu_node *rnp = rcu_get_root(rsp); |
1145 | struct rcu_data *rdp = this_cpu_ptr(rsp->rda); | |
15ba0ba8 | 1146 | |
fc2219d4 | 1147 | WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); |
0bbcc529 PM |
1148 | |
1149 | /* | |
1150 | * Ensure that all grace-period and pre-grace-period activity | |
1151 | * is seen before the assignment to rsp->completed. | |
1152 | */ | |
1153 | smp_mb(); /* See above block comment. */ | |
15ba0ba8 PM |
1154 | gp_duration = jiffies - rsp->gp_start; |
1155 | if (gp_duration > rsp->gp_max) | |
1156 | rsp->gp_max = gp_duration; | |
afe24b12 PM |
1157 | |
1158 | /* | |
1159 | * We know the grace period is complete, but to everyone else | |
1160 | * it appears to still be ongoing. But it is also the case | |
1161 | * that to everyone else it looks like there is nothing that | |
1162 | * they can do to advance the grace period. It is therefore | |
1163 | * safe for us to drop the lock in order to mark the grace | |
1164 | * period as completed in all of the rcu_node structures. | |
1165 | * | |
1166 | * But if this CPU needs another grace period, it will take | |
1167 | * care of this while initializing the next grace period. | |
1168 | * We use RCU_WAIT_TAIL instead of the usual RCU_DONE_TAIL | |
1169 | * because the callbacks have not yet been advanced: Those | |
1170 | * callbacks are waiting on the grace period that just now | |
1171 | * completed. | |
1172 | */ | |
1173 | if (*rdp->nxttail[RCU_WAIT_TAIL] == NULL) { | |
1174 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ | |
1175 | ||
1176 | /* | |
1177 | * Propagate new ->completed value to rcu_node structures | |
1178 | * so that other CPUs don't have to wait until the start | |
1179 | * of the next grace period to process their callbacks. | |
1180 | */ | |
1181 | rcu_for_each_node_breadth_first(rsp, rnp) { | |
1182 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ | |
1183 | rnp->completed = rsp->gpnum; | |
1184 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ | |
1185 | } | |
1186 | rnp = rcu_get_root(rsp); | |
1187 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ | |
1188 | } | |
1189 | ||
1190 | rsp->completed = rsp->gpnum; /* Declare the grace period complete. */ | |
d4c08f2a | 1191 | trace_rcu_grace_period(rsp->name, rsp->completed, "end"); |
af446b70 | 1192 | rsp->fqs_state = RCU_GP_IDLE; |
f41d911f PM |
1193 | rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */ |
1194 | } | |
1195 | ||
64db4cff | 1196 | /* |
d3f6bad3 PM |
1197 | * Similar to rcu_report_qs_rdp(), for which it is a helper function. |
1198 | * Allows quiescent states for a group of CPUs to be reported at one go | |
1199 | * to the specified rcu_node structure, though all the CPUs in the group | |
1200 | * must be represented by the same rcu_node structure (which need not be | |
1201 | * a leaf rcu_node structure, though it often will be). That structure's | |
1202 | * lock must be held upon entry, and it is released before return. | |
64db4cff PM |
1203 | */ |
1204 | static void | |
d3f6bad3 PM |
1205 | rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp, |
1206 | struct rcu_node *rnp, unsigned long flags) | |
64db4cff PM |
1207 | __releases(rnp->lock) |
1208 | { | |
28ecd580 PM |
1209 | struct rcu_node *rnp_c; |
1210 | ||
64db4cff PM |
1211 | /* Walk up the rcu_node hierarchy. */ |
1212 | for (;;) { | |
1213 | if (!(rnp->qsmask & mask)) { | |
1214 | ||
1215 | /* Our bit has already been cleared, so done. */ | |
1304afb2 | 1216 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1217 | return; |
1218 | } | |
1219 | rnp->qsmask &= ~mask; | |
d4c08f2a PM |
1220 | trace_rcu_quiescent_state_report(rsp->name, rnp->gpnum, |
1221 | mask, rnp->qsmask, rnp->level, | |
1222 | rnp->grplo, rnp->grphi, | |
1223 | !!rnp->gp_tasks); | |
27f4d280 | 1224 | if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) { |
64db4cff PM |
1225 | |
1226 | /* Other bits still set at this level, so done. */ | |
1304afb2 | 1227 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1228 | return; |
1229 | } | |
1230 | mask = rnp->grpmask; | |
1231 | if (rnp->parent == NULL) { | |
1232 | ||
1233 | /* No more levels. Exit loop holding root lock. */ | |
1234 | ||
1235 | break; | |
1236 | } | |
1304afb2 | 1237 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
28ecd580 | 1238 | rnp_c = rnp; |
64db4cff | 1239 | rnp = rnp->parent; |
1304afb2 | 1240 | raw_spin_lock_irqsave(&rnp->lock, flags); |
28ecd580 | 1241 | WARN_ON_ONCE(rnp_c->qsmask); |
64db4cff PM |
1242 | } |
1243 | ||
1244 | /* | |
1245 | * Get here if we are the last CPU to pass through a quiescent | |
d3f6bad3 | 1246 | * state for this grace period. Invoke rcu_report_qs_rsp() |
f41d911f | 1247 | * to clean up and start the next grace period if one is needed. |
64db4cff | 1248 | */ |
d3f6bad3 | 1249 | rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */ |
64db4cff PM |
1250 | } |
1251 | ||
1252 | /* | |
d3f6bad3 PM |
1253 | * Record a quiescent state for the specified CPU to that CPU's rcu_data |
1254 | * structure. This must be either called from the specified CPU, or | |
1255 | * called when the specified CPU is known to be offline (and when it is | |
1256 | * also known that no other CPU is concurrently trying to help the offline | |
1257 | * CPU). The lastcomp argument is used to make sure we are still in the | |
1258 | * grace period of interest. We don't want to end the current grace period | |
1259 | * based on quiescent states detected in an earlier grace period! | |
64db4cff PM |
1260 | */ |
1261 | static void | |
e4cc1f22 | 1262 | rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastgp) |
64db4cff PM |
1263 | { |
1264 | unsigned long flags; | |
1265 | unsigned long mask; | |
1266 | struct rcu_node *rnp; | |
1267 | ||
1268 | rnp = rdp->mynode; | |
1304afb2 | 1269 | raw_spin_lock_irqsave(&rnp->lock, flags); |
e4cc1f22 | 1270 | if (lastgp != rnp->gpnum || rnp->completed == rnp->gpnum) { |
64db4cff PM |
1271 | |
1272 | /* | |
e4cc1f22 PM |
1273 | * The grace period in which this quiescent state was |
1274 | * recorded has ended, so don't report it upwards. | |
1275 | * We will instead need a new quiescent state that lies | |
1276 | * within the current grace period. | |
64db4cff | 1277 | */ |
e4cc1f22 | 1278 | rdp->passed_quiesce = 0; /* need qs for new gp. */ |
1304afb2 | 1279 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1280 | return; |
1281 | } | |
1282 | mask = rdp->grpmask; | |
1283 | if ((rnp->qsmask & mask) == 0) { | |
1304afb2 | 1284 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff PM |
1285 | } else { |
1286 | rdp->qs_pending = 0; | |
1287 | ||
1288 | /* | |
1289 | * This GP can't end until cpu checks in, so all of our | |
1290 | * callbacks can be processed during the next GP. | |
1291 | */ | |
64db4cff PM |
1292 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
1293 | ||
d3f6bad3 | 1294 | rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */ |
64db4cff PM |
1295 | } |
1296 | } | |
1297 | ||
1298 | /* | |
1299 | * Check to see if there is a new grace period of which this CPU | |
1300 | * is not yet aware, and if so, set up local rcu_data state for it. | |
1301 | * Otherwise, see if this CPU has just passed through its first | |
1302 | * quiescent state for this grace period, and record that fact if so. | |
1303 | */ | |
1304 | static void | |
1305 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) | |
1306 | { | |
1307 | /* If there is now a new grace period, record and return. */ | |
1308 | if (check_for_new_grace_period(rsp, rdp)) | |
1309 | return; | |
1310 | ||
1311 | /* | |
1312 | * Does this CPU still need to do its part for current grace period? | |
1313 | * If no, return and let the other CPUs do their part as well. | |
1314 | */ | |
1315 | if (!rdp->qs_pending) | |
1316 | return; | |
1317 | ||
1318 | /* | |
1319 | * Was there a quiescent state since the beginning of the grace | |
1320 | * period? If no, then exit and wait for the next call. | |
1321 | */ | |
e4cc1f22 | 1322 | if (!rdp->passed_quiesce) |
64db4cff PM |
1323 | return; |
1324 | ||
d3f6bad3 PM |
1325 | /* |
1326 | * Tell RCU we are done (but rcu_report_qs_rdp() will be the | |
1327 | * judge of that). | |
1328 | */ | |
e4cc1f22 | 1329 | rcu_report_qs_rdp(rdp->cpu, rsp, rdp, rdp->passed_quiesce_gpnum); |
64db4cff PM |
1330 | } |
1331 | ||
1332 | #ifdef CONFIG_HOTPLUG_CPU | |
1333 | ||
e74f4c45 | 1334 | /* |
b1420f1c PM |
1335 | * Send the specified CPU's RCU callbacks to the orphanage. The |
1336 | * specified CPU must be offline, and the caller must hold the | |
1337 | * ->onofflock. | |
e74f4c45 | 1338 | */ |
b1420f1c PM |
1339 | static void |
1340 | rcu_send_cbs_to_orphanage(int cpu, struct rcu_state *rsp, | |
1341 | struct rcu_node *rnp, struct rcu_data *rdp) | |
e74f4c45 | 1342 | { |
b1420f1c PM |
1343 | /* |
1344 | * Orphan the callbacks. First adjust the counts. This is safe | |
1345 | * because ->onofflock excludes _rcu_barrier()'s adoption of | |
1346 | * the callbacks, thus no memory barrier is required. | |
1347 | */ | |
a50c3af9 | 1348 | if (rdp->nxtlist != NULL) { |
b1420f1c PM |
1349 | rsp->qlen_lazy += rdp->qlen_lazy; |
1350 | rsp->qlen += rdp->qlen; | |
1351 | rdp->n_cbs_orphaned += rdp->qlen; | |
a50c3af9 | 1352 | rdp->qlen_lazy = 0; |
1d1fb395 | 1353 | ACCESS_ONCE(rdp->qlen) = 0; |
a50c3af9 PM |
1354 | } |
1355 | ||
1356 | /* | |
b1420f1c PM |
1357 | * Next, move those callbacks still needing a grace period to |
1358 | * the orphanage, where some other CPU will pick them up. | |
1359 | * Some of the callbacks might have gone partway through a grace | |
1360 | * period, but that is too bad. They get to start over because we | |
1361 | * cannot assume that grace periods are synchronized across CPUs. | |
1362 | * We don't bother updating the ->nxttail[] array yet, instead | |
1363 | * we just reset the whole thing later on. | |
a50c3af9 | 1364 | */ |
b1420f1c PM |
1365 | if (*rdp->nxttail[RCU_DONE_TAIL] != NULL) { |
1366 | *rsp->orphan_nxttail = *rdp->nxttail[RCU_DONE_TAIL]; | |
1367 | rsp->orphan_nxttail = rdp->nxttail[RCU_NEXT_TAIL]; | |
1368 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | |
a50c3af9 PM |
1369 | } |
1370 | ||
1371 | /* | |
b1420f1c PM |
1372 | * Then move the ready-to-invoke callbacks to the orphanage, |
1373 | * where some other CPU will pick them up. These will not be | |
1374 | * required to pass though another grace period: They are done. | |
a50c3af9 | 1375 | */ |
e5601400 | 1376 | if (rdp->nxtlist != NULL) { |
b1420f1c PM |
1377 | *rsp->orphan_donetail = rdp->nxtlist; |
1378 | rsp->orphan_donetail = rdp->nxttail[RCU_DONE_TAIL]; | |
e5601400 | 1379 | } |
e74f4c45 | 1380 | |
b1420f1c | 1381 | /* Finally, initialize the rcu_data structure's list to empty. */ |
3f5d3ea6 | 1382 | init_callback_list(rdp); |
b1420f1c PM |
1383 | } |
1384 | ||
1385 | /* | |
1386 | * Adopt the RCU callbacks from the specified rcu_state structure's | |
1387 | * orphanage. The caller must hold the ->onofflock. | |
1388 | */ | |
1389 | static void rcu_adopt_orphan_cbs(struct rcu_state *rsp) | |
1390 | { | |
1391 | int i; | |
1392 | struct rcu_data *rdp = __this_cpu_ptr(rsp->rda); | |
1393 | ||
a50c3af9 | 1394 | /* |
b1420f1c PM |
1395 | * If there is an rcu_barrier() operation in progress, then |
1396 | * only the task doing that operation is permitted to adopt | |
1397 | * callbacks. To do otherwise breaks rcu_barrier() and friends | |
1398 | * by causing them to fail to wait for the callbacks in the | |
1399 | * orphanage. | |
a50c3af9 | 1400 | */ |
b1420f1c PM |
1401 | if (rsp->rcu_barrier_in_progress && |
1402 | rsp->rcu_barrier_in_progress != current) | |
1403 | return; | |
1404 | ||
1405 | /* Do the accounting first. */ | |
1406 | rdp->qlen_lazy += rsp->qlen_lazy; | |
1407 | rdp->qlen += rsp->qlen; | |
1408 | rdp->n_cbs_adopted += rsp->qlen; | |
8f5af6f1 PM |
1409 | if (rsp->qlen_lazy != rsp->qlen) |
1410 | rcu_idle_count_callbacks_posted(); | |
b1420f1c PM |
1411 | rsp->qlen_lazy = 0; |
1412 | rsp->qlen = 0; | |
1413 | ||
1414 | /* | |
1415 | * We do not need a memory barrier here because the only way we | |
1416 | * can get here if there is an rcu_barrier() in flight is if | |
1417 | * we are the task doing the rcu_barrier(). | |
1418 | */ | |
1419 | ||
1420 | /* First adopt the ready-to-invoke callbacks. */ | |
1421 | if (rsp->orphan_donelist != NULL) { | |
1422 | *rsp->orphan_donetail = *rdp->nxttail[RCU_DONE_TAIL]; | |
1423 | *rdp->nxttail[RCU_DONE_TAIL] = rsp->orphan_donelist; | |
1424 | for (i = RCU_NEXT_SIZE - 1; i >= RCU_DONE_TAIL; i--) | |
1425 | if (rdp->nxttail[i] == rdp->nxttail[RCU_DONE_TAIL]) | |
1426 | rdp->nxttail[i] = rsp->orphan_donetail; | |
1427 | rsp->orphan_donelist = NULL; | |
1428 | rsp->orphan_donetail = &rsp->orphan_donelist; | |
1429 | } | |
1430 | ||
1431 | /* And then adopt the callbacks that still need a grace period. */ | |
1432 | if (rsp->orphan_nxtlist != NULL) { | |
1433 | *rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_nxtlist; | |
1434 | rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_nxttail; | |
1435 | rsp->orphan_nxtlist = NULL; | |
1436 | rsp->orphan_nxttail = &rsp->orphan_nxtlist; | |
1437 | } | |
1438 | } | |
1439 | ||
1440 | /* | |
1441 | * Trace the fact that this CPU is going offline. | |
1442 | */ | |
1443 | static void rcu_cleanup_dying_cpu(struct rcu_state *rsp) | |
1444 | { | |
1445 | RCU_TRACE(unsigned long mask); | |
1446 | RCU_TRACE(struct rcu_data *rdp = this_cpu_ptr(rsp->rda)); | |
1447 | RCU_TRACE(struct rcu_node *rnp = rdp->mynode); | |
1448 | ||
1449 | RCU_TRACE(mask = rdp->grpmask); | |
e5601400 PM |
1450 | trace_rcu_grace_period(rsp->name, |
1451 | rnp->gpnum + 1 - !!(rnp->qsmask & mask), | |
1452 | "cpuofl"); | |
64db4cff PM |
1453 | } |
1454 | ||
1455 | /* | |
e5601400 | 1456 | * The CPU has been completely removed, and some other CPU is reporting |
b1420f1c PM |
1457 | * this fact from process context. Do the remainder of the cleanup, |
1458 | * including orphaning the outgoing CPU's RCU callbacks, and also | |
1459 | * adopting them, if there is no _rcu_barrier() instance running. | |
e5601400 PM |
1460 | * There can only be one CPU hotplug operation at a time, so no other |
1461 | * CPU can be attempting to update rcu_cpu_kthread_task. | |
64db4cff | 1462 | */ |
e5601400 | 1463 | static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp) |
64db4cff | 1464 | { |
2036d94a PM |
1465 | unsigned long flags; |
1466 | unsigned long mask; | |
1467 | int need_report = 0; | |
e5601400 | 1468 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
b1420f1c | 1469 | struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ |
e5601400 | 1470 | |
2036d94a | 1471 | /* Adjust any no-longer-needed kthreads. */ |
e5601400 PM |
1472 | rcu_stop_cpu_kthread(cpu); |
1473 | rcu_node_kthread_setaffinity(rnp, -1); | |
2036d94a | 1474 | |
b1420f1c | 1475 | /* Remove the dead CPU from the bitmasks in the rcu_node hierarchy. */ |
2036d94a PM |
1476 | |
1477 | /* Exclude any attempts to start a new grace period. */ | |
1478 | raw_spin_lock_irqsave(&rsp->onofflock, flags); | |
1479 | ||
b1420f1c PM |
1480 | /* Orphan the dead CPU's callbacks, and adopt them if appropriate. */ |
1481 | rcu_send_cbs_to_orphanage(cpu, rsp, rnp, rdp); | |
1482 | rcu_adopt_orphan_cbs(rsp); | |
1483 | ||
2036d94a PM |
1484 | /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */ |
1485 | mask = rdp->grpmask; /* rnp->grplo is constant. */ | |
1486 | do { | |
1487 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ | |
1488 | rnp->qsmaskinit &= ~mask; | |
1489 | if (rnp->qsmaskinit != 0) { | |
1490 | if (rnp != rdp->mynode) | |
1491 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ | |
1492 | break; | |
1493 | } | |
1494 | if (rnp == rdp->mynode) | |
1495 | need_report = rcu_preempt_offline_tasks(rsp, rnp, rdp); | |
1496 | else | |
1497 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ | |
1498 | mask = rnp->grpmask; | |
1499 | rnp = rnp->parent; | |
1500 | } while (rnp != NULL); | |
1501 | ||
1502 | /* | |
1503 | * We still hold the leaf rcu_node structure lock here, and | |
1504 | * irqs are still disabled. The reason for this subterfuge is | |
1505 | * because invoking rcu_report_unblock_qs_rnp() with ->onofflock | |
1506 | * held leads to deadlock. | |
1507 | */ | |
1508 | raw_spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ | |
1509 | rnp = rdp->mynode; | |
1510 | if (need_report & RCU_OFL_TASKS_NORM_GP) | |
1511 | rcu_report_unblock_qs_rnp(rnp, flags); | |
1512 | else | |
1513 | raw_spin_unlock_irqrestore(&rnp->lock, flags); | |
1514 | if (need_report & RCU_OFL_TASKS_EXP_GP) | |
1515 | rcu_report_exp_rnp(rsp, rnp, true); | |
64db4cff PM |
1516 | } |
1517 | ||
1518 | #else /* #ifdef CONFIG_HOTPLUG_CPU */ | |
1519 | ||
b1420f1c PM |
1520 | static void rcu_adopt_orphan_cbs(struct rcu_state *rsp) |
1521 | { | |
1522 | } | |
1523 | ||
e5601400 | 1524 | static void rcu_cleanup_dying_cpu(struct rcu_state *rsp) |
e74f4c45 PM |
1525 | { |
1526 | } | |
1527 | ||
e5601400 | 1528 | static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp) |
64db4cff PM |
1529 | { |
1530 | } | |
1531 | ||
1532 | #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */ | |
1533 | ||
1534 | /* | |
1535 | * Invoke any RCU callbacks that have made it to the end of their grace | |
1536 | * period. Thottle as specified by rdp->blimit. | |
1537 | */ | |
37c72e56 | 1538 | static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp) |
64db4cff PM |
1539 | { |
1540 | unsigned long flags; | |
1541 | struct rcu_head *next, *list, **tail; | |
b41772ab | 1542 | int bl, count, count_lazy, i; |
64db4cff PM |
1543 | |
1544 | /* If no callbacks are ready, just return.*/ | |
29c00b4a | 1545 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) { |
486e2593 | 1546 | trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, 0); |
4968c300 PM |
1547 | trace_rcu_batch_end(rsp->name, 0, !!ACCESS_ONCE(rdp->nxtlist), |
1548 | need_resched(), is_idle_task(current), | |
1549 | rcu_is_callbacks_kthread()); | |
64db4cff | 1550 | return; |
29c00b4a | 1551 | } |
64db4cff PM |
1552 | |
1553 | /* | |
1554 | * Extract the list of ready callbacks, disabling to prevent | |
1555 | * races with call_rcu() from interrupt handlers. | |
1556 | */ | |
1557 | local_irq_save(flags); | |
8146c4e2 | 1558 | WARN_ON_ONCE(cpu_is_offline(smp_processor_id())); |
29c00b4a | 1559 | bl = rdp->blimit; |
486e2593 | 1560 | trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, bl); |
64db4cff PM |
1561 | list = rdp->nxtlist; |
1562 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; | |
1563 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; | |
1564 | tail = rdp->nxttail[RCU_DONE_TAIL]; | |
b41772ab PM |
1565 | for (i = RCU_NEXT_SIZE - 1; i >= 0; i--) |
1566 | if (rdp->nxttail[i] == rdp->nxttail[RCU_DONE_TAIL]) | |
1567 | rdp->nxttail[i] = &rdp->nxtlist; | |
64db4cff PM |
1568 | local_irq_restore(flags); |
1569 | ||
1570 | /* Invoke callbacks. */ | |
486e2593 | 1571 | count = count_lazy = 0; |
64db4cff PM |
1572 | while (list) { |
1573 | next = list->next; | |
1574 | prefetch(next); | |
551d55a9 | 1575 | debug_rcu_head_unqueue(list); |
486e2593 PM |
1576 | if (__rcu_reclaim(rsp->name, list)) |
1577 | count_lazy++; | |
64db4cff | 1578 | list = next; |
dff1672d PM |
1579 | /* Stop only if limit reached and CPU has something to do. */ |
1580 | if (++count >= bl && | |
1581 | (need_resched() || | |
1582 | (!is_idle_task(current) && !rcu_is_callbacks_kthread()))) | |
64db4cff PM |
1583 | break; |
1584 | } | |
1585 | ||
1586 | local_irq_save(flags); | |
4968c300 PM |
1587 | trace_rcu_batch_end(rsp->name, count, !!list, need_resched(), |
1588 | is_idle_task(current), | |
1589 | rcu_is_callbacks_kthread()); | |
64db4cff PM |
1590 | |
1591 | /* Update count, and requeue any remaining callbacks. */ | |
64db4cff PM |
1592 | if (list != NULL) { |
1593 | *tail = rdp->nxtlist; | |
1594 | rdp->nxtlist = list; | |
b41772ab PM |
1595 | for (i = 0; i < RCU_NEXT_SIZE; i++) |
1596 | if (&rdp->nxtlist == rdp->nxttail[i]) | |
1597 | rdp->nxttail[i] = tail; | |
64db4cff PM |
1598 | else |
1599 | break; | |
1600 | } | |
b1420f1c PM |
1601 | smp_mb(); /* List handling before counting for rcu_barrier(). */ |
1602 | rdp->qlen_lazy -= count_lazy; | |
1d1fb395 | 1603 | ACCESS_ONCE(rdp->qlen) -= count; |
b1420f1c | 1604 | rdp->n_cbs_invoked += count; |
64db4cff PM |
1605 | |
1606 | /* Reinstate batch limit if we have worked down the excess. */ | |
1607 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) | |
1608 | rdp->blimit = blimit; | |
1609 | ||
37c72e56 PM |
1610 | /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */ |
1611 | if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) { | |
1612 | rdp->qlen_last_fqs_check = 0; | |
1613 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1614 | } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark) | |
1615 | rdp->qlen_last_fqs_check = rdp->qlen; | |
1616 | ||
64db4cff PM |
1617 | local_irq_restore(flags); |
1618 | ||
e0f23060 | 1619 | /* Re-invoke RCU core processing if there are callbacks remaining. */ |
64db4cff | 1620 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
a46e0899 | 1621 | invoke_rcu_core(); |
64db4cff PM |
1622 | } |
1623 | ||
1624 | /* | |
1625 | * Check to see if this CPU is in a non-context-switch quiescent state | |
1626 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). | |
e0f23060 | 1627 | * Also schedule RCU core processing. |
64db4cff | 1628 | * |
9b2e4f18 | 1629 | * This function must be called from hardirq context. It is normally |
64db4cff PM |
1630 | * invoked from the scheduling-clock interrupt. If rcu_pending returns |
1631 | * false, there is no point in invoking rcu_check_callbacks(). | |
1632 | */ | |
1633 | void rcu_check_callbacks(int cpu, int user) | |
1634 | { | |
300df91c | 1635 | trace_rcu_utilization("Start scheduler-tick"); |
a858af28 | 1636 | increment_cpu_stall_ticks(); |
9b2e4f18 | 1637 | if (user || rcu_is_cpu_rrupt_from_idle()) { |
64db4cff PM |
1638 | |
1639 | /* | |
1640 | * Get here if this CPU took its interrupt from user | |
1641 | * mode or from the idle loop, and if this is not a | |
1642 | * nested interrupt. In this case, the CPU is in | |
d6714c22 | 1643 | * a quiescent state, so note it. |
64db4cff PM |
1644 | * |
1645 | * No memory barrier is required here because both | |
d6714c22 PM |
1646 | * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local |
1647 | * variables that other CPUs neither access nor modify, | |
1648 | * at least not while the corresponding CPU is online. | |
64db4cff PM |
1649 | */ |
1650 | ||
d6714c22 PM |
1651 | rcu_sched_qs(cpu); |
1652 | rcu_bh_qs(cpu); | |
64db4cff PM |
1653 | |
1654 | } else if (!in_softirq()) { | |
1655 | ||
1656 | /* | |
1657 | * Get here if this CPU did not take its interrupt from | |
1658 | * softirq, in other words, if it is not interrupting | |
1659 | * a rcu_bh read-side critical section. This is an _bh | |
d6714c22 | 1660 | * critical section, so note it. |
64db4cff PM |
1661 | */ |
1662 | ||
d6714c22 | 1663 | rcu_bh_qs(cpu); |
64db4cff | 1664 | } |
f41d911f | 1665 | rcu_preempt_check_callbacks(cpu); |
d21670ac | 1666 | if (rcu_pending(cpu)) |
a46e0899 | 1667 | invoke_rcu_core(); |
300df91c | 1668 | trace_rcu_utilization("End scheduler-tick"); |
64db4cff PM |
1669 | } |
1670 | ||
64db4cff PM |
1671 | /* |
1672 | * Scan the leaf rcu_node structures, processing dyntick state for any that | |
1673 | * have not yet encountered a quiescent state, using the function specified. | |
27f4d280 PM |
1674 | * Also initiate boosting for any threads blocked on the root rcu_node. |
1675 | * | |
ee47eb9f | 1676 | * The caller must have suppressed start of new grace periods. |
64db4cff | 1677 | */ |
45f014c5 | 1678 | static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *)) |
64db4cff PM |
1679 | { |
1680 | unsigned long bit; | |
1681 | int cpu; | |
1682 | unsigned long flags; | |
1683 | unsigned long mask; | |
a0b6c9a7 | 1684 | struct rcu_node *rnp; |
64db4cff | 1685 | |
a0b6c9a7 | 1686 | rcu_for_each_leaf_node(rsp, rnp) { |
64db4cff | 1687 | mask = 0; |
1304afb2 | 1688 | raw_spin_lock_irqsave(&rnp->lock, flags); |
ee47eb9f | 1689 | if (!rcu_gp_in_progress(rsp)) { |
1304afb2 | 1690 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
0f10dc82 | 1691 | return; |
64db4cff | 1692 | } |
a0b6c9a7 | 1693 | if (rnp->qsmask == 0) { |
1217ed1b | 1694 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock */ |
64db4cff PM |
1695 | continue; |
1696 | } | |
a0b6c9a7 | 1697 | cpu = rnp->grplo; |
64db4cff | 1698 | bit = 1; |
a0b6c9a7 | 1699 | for (; cpu <= rnp->grphi; cpu++, bit <<= 1) { |
394f99a9 LJ |
1700 | if ((rnp->qsmask & bit) != 0 && |
1701 | f(per_cpu_ptr(rsp->rda, cpu))) | |
64db4cff PM |
1702 | mask |= bit; |
1703 | } | |
45f014c5 | 1704 | if (mask != 0) { |
64db4cff | 1705 | |
d3f6bad3 PM |
1706 | /* rcu_report_qs_rnp() releases rnp->lock. */ |
1707 | rcu_report_qs_rnp(mask, rsp, rnp, flags); | |
64db4cff PM |
1708 | continue; |
1709 | } | |
1304afb2 | 1710 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
64db4cff | 1711 | } |
27f4d280 | 1712 | rnp = rcu_get_root(rsp); |
1217ed1b PM |
1713 | if (rnp->qsmask == 0) { |
1714 | raw_spin_lock_irqsave(&rnp->lock, flags); | |
1715 | rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */ | |
1716 | } | |
64db4cff PM |
1717 | } |
1718 | ||
1719 | /* | |
1720 | * Force quiescent states on reluctant CPUs, and also detect which | |
1721 | * CPUs are in dyntick-idle mode. | |
1722 | */ | |
1723 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) | |
1724 | { | |
1725 | unsigned long flags; | |
64db4cff | 1726 | struct rcu_node *rnp = rcu_get_root(rsp); |
64db4cff | 1727 | |
300df91c PM |
1728 | trace_rcu_utilization("Start fqs"); |
1729 | if (!rcu_gp_in_progress(rsp)) { | |
1730 | trace_rcu_utilization("End fqs"); | |
64db4cff | 1731 | return; /* No grace period in progress, nothing to force. */ |
300df91c | 1732 | } |
1304afb2 | 1733 | if (!raw_spin_trylock_irqsave(&rsp->fqslock, flags)) { |
64db4cff | 1734 | rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ |
300df91c | 1735 | trace_rcu_utilization("End fqs"); |
64db4cff PM |
1736 | return; /* Someone else is already on the job. */ |
1737 | } | |
20133cfc | 1738 | if (relaxed && ULONG_CMP_GE(rsp->jiffies_force_qs, jiffies)) |
f96e9232 | 1739 | goto unlock_fqs_ret; /* no emergency and done recently. */ |
64db4cff | 1740 | rsp->n_force_qs++; |
1304afb2 | 1741 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
64db4cff | 1742 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
560d4bc0 | 1743 | if(!rcu_gp_in_progress(rsp)) { |
64db4cff | 1744 | rsp->n_force_qs_ngp++; |
1304afb2 | 1745 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1746 | goto unlock_fqs_ret; /* no GP in progress, time updated. */ |
64db4cff | 1747 | } |
07079d53 | 1748 | rsp->fqs_active = 1; |
af446b70 | 1749 | switch (rsp->fqs_state) { |
83f5b01f | 1750 | case RCU_GP_IDLE: |
64db4cff PM |
1751 | case RCU_GP_INIT: |
1752 | ||
83f5b01f | 1753 | break; /* grace period idle or initializing, ignore. */ |
64db4cff PM |
1754 | |
1755 | case RCU_SAVE_DYNTICK: | |
64db4cff PM |
1756 | if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK) |
1757 | break; /* So gcc recognizes the dead code. */ | |
1758 | ||
f261414f LJ |
1759 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
1760 | ||
64db4cff | 1761 | /* Record dyntick-idle state. */ |
45f014c5 | 1762 | force_qs_rnp(rsp, dyntick_save_progress_counter); |
1304afb2 | 1763 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
ee47eb9f | 1764 | if (rcu_gp_in_progress(rsp)) |
af446b70 | 1765 | rsp->fqs_state = RCU_FORCE_QS; |
ee47eb9f | 1766 | break; |
64db4cff PM |
1767 | |
1768 | case RCU_FORCE_QS: | |
1769 | ||
1770 | /* Check dyntick-idle state, send IPI to laggarts. */ | |
1304afb2 | 1771 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
45f014c5 | 1772 | force_qs_rnp(rsp, rcu_implicit_dynticks_qs); |
64db4cff PM |
1773 | |
1774 | /* Leave state in case more forcing is required. */ | |
1775 | ||
1304afb2 | 1776 | raw_spin_lock(&rnp->lock); /* irqs already disabled */ |
f96e9232 | 1777 | break; |
64db4cff | 1778 | } |
07079d53 | 1779 | rsp->fqs_active = 0; |
46a1e34e | 1780 | if (rsp->fqs_need_gp) { |
1304afb2 | 1781 | raw_spin_unlock(&rsp->fqslock); /* irqs remain disabled */ |
46a1e34e PM |
1782 | rsp->fqs_need_gp = 0; |
1783 | rcu_start_gp(rsp, flags); /* releases rnp->lock */ | |
300df91c | 1784 | trace_rcu_utilization("End fqs"); |
46a1e34e PM |
1785 | return; |
1786 | } | |
1304afb2 | 1787 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ |
f96e9232 | 1788 | unlock_fqs_ret: |
1304afb2 | 1789 | raw_spin_unlock_irqrestore(&rsp->fqslock, flags); |
300df91c | 1790 | trace_rcu_utilization("End fqs"); |
64db4cff PM |
1791 | } |
1792 | ||
64db4cff | 1793 | /* |
e0f23060 PM |
1794 | * This does the RCU core processing work for the specified rcu_state |
1795 | * and rcu_data structures. This may be called only from the CPU to | |
1796 | * whom the rdp belongs. | |
64db4cff PM |
1797 | */ |
1798 | static void | |
1799 | __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) | |
1800 | { | |
1801 | unsigned long flags; | |
1802 | ||
2e597558 PM |
1803 | WARN_ON_ONCE(rdp->beenonline == 0); |
1804 | ||
64db4cff PM |
1805 | /* |
1806 | * If an RCU GP has gone long enough, go check for dyntick | |
1807 | * idle CPUs and, if needed, send resched IPIs. | |
1808 | */ | |
20133cfc | 1809 | if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
64db4cff PM |
1810 | force_quiescent_state(rsp, 1); |
1811 | ||
1812 | /* | |
1813 | * Advance callbacks in response to end of earlier grace | |
1814 | * period that some other CPU ended. | |
1815 | */ | |
1816 | rcu_process_gp_end(rsp, rdp); | |
1817 | ||
1818 | /* Update RCU state based on any recent quiescent states. */ | |
1819 | rcu_check_quiescent_state(rsp, rdp); | |
1820 | ||
1821 | /* Does this CPU require a not-yet-started grace period? */ | |
1822 | if (cpu_needs_another_gp(rsp, rdp)) { | |
1304afb2 | 1823 | raw_spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags); |
64db4cff PM |
1824 | rcu_start_gp(rsp, flags); /* releases above lock */ |
1825 | } | |
1826 | ||
1827 | /* If there are callbacks ready, invoke them. */ | |
09223371 | 1828 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
a46e0899 | 1829 | invoke_rcu_callbacks(rsp, rdp); |
09223371 SL |
1830 | } |
1831 | ||
64db4cff | 1832 | /* |
e0f23060 | 1833 | * Do RCU core processing for the current CPU. |
64db4cff | 1834 | */ |
09223371 | 1835 | static void rcu_process_callbacks(struct softirq_action *unused) |
64db4cff | 1836 | { |
300df91c | 1837 | trace_rcu_utilization("Start RCU core"); |
d6714c22 PM |
1838 | __rcu_process_callbacks(&rcu_sched_state, |
1839 | &__get_cpu_var(rcu_sched_data)); | |
64db4cff | 1840 | __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); |
f41d911f | 1841 | rcu_preempt_process_callbacks(); |
300df91c | 1842 | trace_rcu_utilization("End RCU core"); |
64db4cff PM |
1843 | } |
1844 | ||
a26ac245 | 1845 | /* |
e0f23060 PM |
1846 | * Schedule RCU callback invocation. If the specified type of RCU |
1847 | * does not support RCU priority boosting, just do a direct call, | |
1848 | * otherwise wake up the per-CPU kernel kthread. Note that because we | |
1849 | * are running on the current CPU with interrupts disabled, the | |
1850 | * rcu_cpu_kthread_task cannot disappear out from under us. | |
a26ac245 | 1851 | */ |
a46e0899 | 1852 | static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) |
a26ac245 | 1853 | { |
b0d30417 PM |
1854 | if (unlikely(!ACCESS_ONCE(rcu_scheduler_fully_active))) |
1855 | return; | |
a46e0899 PM |
1856 | if (likely(!rsp->boost)) { |
1857 | rcu_do_batch(rsp, rdp); | |
a26ac245 PM |
1858 | return; |
1859 | } | |
a46e0899 | 1860 | invoke_rcu_callbacks_kthread(); |
a26ac245 PM |
1861 | } |
1862 | ||
a46e0899 | 1863 | static void invoke_rcu_core(void) |
09223371 SL |
1864 | { |
1865 | raise_softirq(RCU_SOFTIRQ); | |
1866 | } | |
1867 | ||
64db4cff PM |
1868 | static void |
1869 | __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu), | |
486e2593 | 1870 | struct rcu_state *rsp, bool lazy) |
64db4cff PM |
1871 | { |
1872 | unsigned long flags; | |
1873 | struct rcu_data *rdp; | |
1874 | ||
0bb7b59d | 1875 | WARN_ON_ONCE((unsigned long)head & 0x3); /* Misaligned rcu_head! */ |
551d55a9 | 1876 | debug_rcu_head_queue(head); |
64db4cff PM |
1877 | head->func = func; |
1878 | head->next = NULL; | |
1879 | ||
1880 | smp_mb(); /* Ensure RCU update seen before callback registry. */ | |
1881 | ||
1882 | /* | |
1883 | * Opportunistically note grace-period endings and beginnings. | |
1884 | * Note that we might see a beginning right after we see an | |
1885 | * end, but never vice versa, since this CPU has to pass through | |
1886 | * a quiescent state betweentimes. | |
1887 | */ | |
1888 | local_irq_save(flags); | |
394f99a9 | 1889 | rdp = this_cpu_ptr(rsp->rda); |
64db4cff PM |
1890 | |
1891 | /* Add the callback to our list. */ | |
1d1fb395 | 1892 | ACCESS_ONCE(rdp->qlen)++; |
486e2593 PM |
1893 | if (lazy) |
1894 | rdp->qlen_lazy++; | |
c57afe80 PM |
1895 | else |
1896 | rcu_idle_count_callbacks_posted(); | |
b1420f1c PM |
1897 | smp_mb(); /* Count before adding callback for rcu_barrier(). */ |
1898 | *rdp->nxttail[RCU_NEXT_TAIL] = head; | |
1899 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; | |
2655d57e | 1900 | |
d4c08f2a PM |
1901 | if (__is_kfree_rcu_offset((unsigned long)func)) |
1902 | trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func, | |
486e2593 | 1903 | rdp->qlen_lazy, rdp->qlen); |
d4c08f2a | 1904 | else |
486e2593 | 1905 | trace_rcu_callback(rsp->name, head, rdp->qlen_lazy, rdp->qlen); |
d4c08f2a | 1906 | |
2655d57e PM |
1907 | /* If interrupts were disabled, don't dive into RCU core. */ |
1908 | if (irqs_disabled_flags(flags)) { | |
1909 | local_irq_restore(flags); | |
1910 | return; | |
1911 | } | |
64db4cff | 1912 | |
37c72e56 PM |
1913 | /* |
1914 | * Force the grace period if too many callbacks or too long waiting. | |
1915 | * Enforce hysteresis, and don't invoke force_quiescent_state() | |
1916 | * if some other CPU has recently done so. Also, don't bother | |
1917 | * invoking force_quiescent_state() if the newly enqueued callback | |
1918 | * is the only one waiting for a grace period to complete. | |
1919 | */ | |
2655d57e | 1920 | if (unlikely(rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) { |
b52573d2 PM |
1921 | |
1922 | /* Are we ignoring a completed grace period? */ | |
1923 | rcu_process_gp_end(rsp, rdp); | |
1924 | check_for_new_grace_period(rsp, rdp); | |
1925 | ||
1926 | /* Start a new grace period if one not already started. */ | |
1927 | if (!rcu_gp_in_progress(rsp)) { | |
1928 | unsigned long nestflag; | |
1929 | struct rcu_node *rnp_root = rcu_get_root(rsp); | |
1930 | ||
1931 | raw_spin_lock_irqsave(&rnp_root->lock, nestflag); | |
1932 | rcu_start_gp(rsp, nestflag); /* rlses rnp_root->lock */ | |
1933 | } else { | |
1934 | /* Give the grace period a kick. */ | |
1935 | rdp->blimit = LONG_MAX; | |
1936 | if (rsp->n_force_qs == rdp->n_force_qs_snap && | |
1937 | *rdp->nxttail[RCU_DONE_TAIL] != head) | |
1938 | force_quiescent_state(rsp, 0); | |
1939 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
1940 | rdp->qlen_last_fqs_check = rdp->qlen; | |
1941 | } | |
20133cfc | 1942 | } else if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) |
64db4cff PM |
1943 | force_quiescent_state(rsp, 1); |
1944 | local_irq_restore(flags); | |
1945 | } | |
1946 | ||
1947 | /* | |
d6714c22 | 1948 | * Queue an RCU-sched callback for invocation after a grace period. |
64db4cff | 1949 | */ |
d6714c22 | 1950 | void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) |
64db4cff | 1951 | { |
486e2593 | 1952 | __call_rcu(head, func, &rcu_sched_state, 0); |
64db4cff | 1953 | } |
d6714c22 | 1954 | EXPORT_SYMBOL_GPL(call_rcu_sched); |
64db4cff PM |
1955 | |
1956 | /* | |
486e2593 | 1957 | * Queue an RCU callback for invocation after a quicker grace period. |
64db4cff PM |
1958 | */ |
1959 | void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) | |
1960 | { | |
486e2593 | 1961 | __call_rcu(head, func, &rcu_bh_state, 0); |
64db4cff PM |
1962 | } |
1963 | EXPORT_SYMBOL_GPL(call_rcu_bh); | |
1964 | ||
6d813391 PM |
1965 | /* |
1966 | * Because a context switch is a grace period for RCU-sched and RCU-bh, | |
1967 | * any blocking grace-period wait automatically implies a grace period | |
1968 | * if there is only one CPU online at any point time during execution | |
1969 | * of either synchronize_sched() or synchronize_rcu_bh(). It is OK to | |
1970 | * occasionally incorrectly indicate that there are multiple CPUs online | |
1971 | * when there was in fact only one the whole time, as this just adds | |
1972 | * some overhead: RCU still operates correctly. | |
1973 | * | |
1974 | * Of course, sampling num_online_cpus() with preemption enabled can | |
1975 | * give erroneous results if there are concurrent CPU-hotplug operations. | |
1976 | * For example, given a demonic sequence of preemptions in num_online_cpus() | |
1977 | * and CPU-hotplug operations, there could be two or more CPUs online at | |
1978 | * all times, but num_online_cpus() might well return one (or even zero). | |
1979 | * | |
1980 | * However, all such demonic sequences require at least one CPU-offline | |
1981 | * operation. Furthermore, rcu_blocking_is_gp() giving the wrong answer | |
1982 | * is only a problem if there is an RCU read-side critical section executing | |
1983 | * throughout. But RCU-sched and RCU-bh read-side critical sections | |
1984 | * disable either preemption or bh, which prevents a CPU from going offline. | |
1985 | * Therefore, the only way that rcu_blocking_is_gp() can incorrectly return | |
1986 | * that there is only one CPU when in fact there was more than one throughout | |
1987 | * is when there were no RCU readers in the system. If there are no | |
1988 | * RCU readers, the grace period by definition can be of zero length, | |
1989 | * regardless of the number of online CPUs. | |
1990 | */ | |
1991 | static inline int rcu_blocking_is_gp(void) | |
1992 | { | |
1993 | might_sleep(); /* Check for RCU read-side critical section. */ | |
1994 | return num_online_cpus() <= 1; | |
1995 | } | |
1996 | ||
6ebb237b PM |
1997 | /** |
1998 | * synchronize_sched - wait until an rcu-sched grace period has elapsed. | |
1999 | * | |
2000 | * Control will return to the caller some time after a full rcu-sched | |
2001 | * grace period has elapsed, in other words after all currently executing | |
2002 | * rcu-sched read-side critical sections have completed. These read-side | |
2003 | * critical sections are delimited by rcu_read_lock_sched() and | |
2004 | * rcu_read_unlock_sched(), and may be nested. Note that preempt_disable(), | |
2005 | * local_irq_disable(), and so on may be used in place of | |
2006 | * rcu_read_lock_sched(). | |
2007 | * | |
2008 | * This means that all preempt_disable code sequences, including NMI and | |
2009 | * hardware-interrupt handlers, in progress on entry will have completed | |
2010 | * before this primitive returns. However, this does not guarantee that | |
2011 | * softirq handlers will have completed, since in some kernels, these | |
2012 | * handlers can run in process context, and can block. | |
2013 | * | |
2014 | * This primitive provides the guarantees made by the (now removed) | |
2015 | * synchronize_kernel() API. In contrast, synchronize_rcu() only | |
2016 | * guarantees that rcu_read_lock() sections will have completed. | |
2017 | * In "classic RCU", these two guarantees happen to be one and | |
2018 | * the same, but can differ in realtime RCU implementations. | |
2019 | */ | |
2020 | void synchronize_sched(void) | |
2021 | { | |
fe15d706 PM |
2022 | rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map) && |
2023 | !lock_is_held(&rcu_lock_map) && | |
2024 | !lock_is_held(&rcu_sched_lock_map), | |
2025 | "Illegal synchronize_sched() in RCU-sched read-side critical section"); | |
6ebb237b PM |
2026 | if (rcu_blocking_is_gp()) |
2027 | return; | |
2c42818e | 2028 | wait_rcu_gp(call_rcu_sched); |
6ebb237b PM |
2029 | } |
2030 | EXPORT_SYMBOL_GPL(synchronize_sched); | |
2031 | ||
2032 | /** | |
2033 | * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed. | |
2034 | * | |
2035 | * Control will return to the caller some time after a full rcu_bh grace | |
2036 | * period has elapsed, in other words after all currently executing rcu_bh | |
2037 | * read-side critical sections have completed. RCU read-side critical | |
2038 | * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(), | |
2039 | * and may be nested. | |
2040 | */ | |
2041 | void synchronize_rcu_bh(void) | |
2042 | { | |
fe15d706 PM |
2043 | rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map) && |
2044 | !lock_is_held(&rcu_lock_map) && | |
2045 | !lock_is_held(&rcu_sched_lock_map), | |
2046 | "Illegal synchronize_rcu_bh() in RCU-bh read-side critical section"); | |
6ebb237b PM |
2047 | if (rcu_blocking_is_gp()) |
2048 | return; | |
2c42818e | 2049 | wait_rcu_gp(call_rcu_bh); |
6ebb237b PM |
2050 | } |
2051 | EXPORT_SYMBOL_GPL(synchronize_rcu_bh); | |
2052 | ||
3d3b7db0 PM |
2053 | static atomic_t sync_sched_expedited_started = ATOMIC_INIT(0); |
2054 | static atomic_t sync_sched_expedited_done = ATOMIC_INIT(0); | |
2055 | ||
2056 | static int synchronize_sched_expedited_cpu_stop(void *data) | |
2057 | { | |
2058 | /* | |
2059 | * There must be a full memory barrier on each affected CPU | |
2060 | * between the time that try_stop_cpus() is called and the | |
2061 | * time that it returns. | |
2062 | * | |
2063 | * In the current initial implementation of cpu_stop, the | |
2064 | * above condition is already met when the control reaches | |
2065 | * this point and the following smp_mb() is not strictly | |
2066 | * necessary. Do smp_mb() anyway for documentation and | |
2067 | * robustness against future implementation changes. | |
2068 | */ | |
2069 | smp_mb(); /* See above comment block. */ | |
2070 | return 0; | |
2071 | } | |
2072 | ||
236fefaf PM |
2073 | /** |
2074 | * synchronize_sched_expedited - Brute-force RCU-sched grace period | |
2075 | * | |
2076 | * Wait for an RCU-sched grace period to elapse, but use a "big hammer" | |
2077 | * approach to force the grace period to end quickly. This consumes | |
2078 | * significant time on all CPUs and is unfriendly to real-time workloads, | |
2079 | * so is thus not recommended for any sort of common-case code. In fact, | |
2080 | * if you are using synchronize_sched_expedited() in a loop, please | |
2081 | * restructure your code to batch your updates, and then use a single | |
2082 | * synchronize_sched() instead. | |
3d3b7db0 | 2083 | * |
236fefaf PM |
2084 | * Note that it is illegal to call this function while holding any lock |
2085 | * that is acquired by a CPU-hotplug notifier. And yes, it is also illegal | |
2086 | * to call this function from a CPU-hotplug notifier. Failing to observe | |
2087 | * these restriction will result in deadlock. | |
3d3b7db0 PM |
2088 | * |
2089 | * This implementation can be thought of as an application of ticket | |
2090 | * locking to RCU, with sync_sched_expedited_started and | |
2091 | * sync_sched_expedited_done taking on the roles of the halves | |
2092 | * of the ticket-lock word. Each task atomically increments | |
2093 | * sync_sched_expedited_started upon entry, snapshotting the old value, | |
2094 | * then attempts to stop all the CPUs. If this succeeds, then each | |
2095 | * CPU will have executed a context switch, resulting in an RCU-sched | |
2096 | * grace period. We are then done, so we use atomic_cmpxchg() to | |
2097 | * update sync_sched_expedited_done to match our snapshot -- but | |
2098 | * only if someone else has not already advanced past our snapshot. | |
2099 | * | |
2100 | * On the other hand, if try_stop_cpus() fails, we check the value | |
2101 | * of sync_sched_expedited_done. If it has advanced past our | |
2102 | * initial snapshot, then someone else must have forced a grace period | |
2103 | * some time after we took our snapshot. In this case, our work is | |
2104 | * done for us, and we can simply return. Otherwise, we try again, | |
2105 | * but keep our initial snapshot for purposes of checking for someone | |
2106 | * doing our work for us. | |
2107 | * | |
2108 | * If we fail too many times in a row, we fall back to synchronize_sched(). | |
2109 | */ | |
2110 | void synchronize_sched_expedited(void) | |
2111 | { | |
2112 | int firstsnap, s, snap, trycount = 0; | |
2113 | ||
2114 | /* Note that atomic_inc_return() implies full memory barrier. */ | |
2115 | firstsnap = snap = atomic_inc_return(&sync_sched_expedited_started); | |
2116 | get_online_cpus(); | |
1cc85961 | 2117 | WARN_ON_ONCE(cpu_is_offline(raw_smp_processor_id())); |
3d3b7db0 PM |
2118 | |
2119 | /* | |
2120 | * Each pass through the following loop attempts to force a | |
2121 | * context switch on each CPU. | |
2122 | */ | |
2123 | while (try_stop_cpus(cpu_online_mask, | |
2124 | synchronize_sched_expedited_cpu_stop, | |
2125 | NULL) == -EAGAIN) { | |
2126 | put_online_cpus(); | |
2127 | ||
2128 | /* No joy, try again later. Or just synchronize_sched(). */ | |
2129 | if (trycount++ < 10) | |
2130 | udelay(trycount * num_online_cpus()); | |
2131 | else { | |
2132 | synchronize_sched(); | |
2133 | return; | |
2134 | } | |
2135 | ||
2136 | /* Check to see if someone else did our work for us. */ | |
2137 | s = atomic_read(&sync_sched_expedited_done); | |
2138 | if (UINT_CMP_GE((unsigned)s, (unsigned)firstsnap)) { | |
2139 | smp_mb(); /* ensure test happens before caller kfree */ | |
2140 | return; | |
2141 | } | |
2142 | ||
2143 | /* | |
2144 | * Refetching sync_sched_expedited_started allows later | |
2145 | * callers to piggyback on our grace period. We subtract | |
2146 | * 1 to get the same token that the last incrementer got. | |
2147 | * We retry after they started, so our grace period works | |
2148 | * for them, and they started after our first try, so their | |
2149 | * grace period works for us. | |
2150 | */ | |
2151 | get_online_cpus(); | |
2152 | snap = atomic_read(&sync_sched_expedited_started); | |
2153 | smp_mb(); /* ensure read is before try_stop_cpus(). */ | |
2154 | } | |
2155 | ||
2156 | /* | |
2157 | * Everyone up to our most recent fetch is covered by our grace | |
2158 | * period. Update the counter, but only if our work is still | |
2159 | * relevant -- which it won't be if someone who started later | |
2160 | * than we did beat us to the punch. | |
2161 | */ | |
2162 | do { | |
2163 | s = atomic_read(&sync_sched_expedited_done); | |
2164 | if (UINT_CMP_GE((unsigned)s, (unsigned)snap)) { | |
2165 | smp_mb(); /* ensure test happens before caller kfree */ | |
2166 | break; | |
2167 | } | |
2168 | } while (atomic_cmpxchg(&sync_sched_expedited_done, s, snap) != s); | |
2169 | ||
2170 | put_online_cpus(); | |
2171 | } | |
2172 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
2173 | ||
64db4cff PM |
2174 | /* |
2175 | * Check to see if there is any immediate RCU-related work to be done | |
2176 | * by the current CPU, for the specified type of RCU, returning 1 if so. | |
2177 | * The checks are in order of increasing expense: checks that can be | |
2178 | * carried out against CPU-local state are performed first. However, | |
2179 | * we must check for CPU stalls first, else we might not get a chance. | |
2180 | */ | |
2181 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) | |
2182 | { | |
2f51f988 PM |
2183 | struct rcu_node *rnp = rdp->mynode; |
2184 | ||
64db4cff PM |
2185 | rdp->n_rcu_pending++; |
2186 | ||
2187 | /* Check for CPU stalls, if enabled. */ | |
2188 | check_cpu_stall(rsp, rdp); | |
2189 | ||
2190 | /* Is the RCU core waiting for a quiescent state from this CPU? */ | |
5c51dd73 PM |
2191 | if (rcu_scheduler_fully_active && |
2192 | rdp->qs_pending && !rdp->passed_quiesce) { | |
d25eb944 PM |
2193 | |
2194 | /* | |
2195 | * If force_quiescent_state() coming soon and this CPU | |
2196 | * needs a quiescent state, and this is either RCU-sched | |
2197 | * or RCU-bh, force a local reschedule. | |
2198 | */ | |
d21670ac | 2199 | rdp->n_rp_qs_pending++; |
6cc68793 | 2200 | if (!rdp->preemptible && |
d25eb944 PM |
2201 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs) - 1, |
2202 | jiffies)) | |
2203 | set_need_resched(); | |
e4cc1f22 | 2204 | } else if (rdp->qs_pending && rdp->passed_quiesce) { |
d21670ac | 2205 | rdp->n_rp_report_qs++; |
64db4cff | 2206 | return 1; |
7ba5c840 | 2207 | } |
64db4cff PM |
2208 | |
2209 | /* Does this CPU have callbacks ready to invoke? */ | |
7ba5c840 PM |
2210 | if (cpu_has_callbacks_ready_to_invoke(rdp)) { |
2211 | rdp->n_rp_cb_ready++; | |
64db4cff | 2212 | return 1; |
7ba5c840 | 2213 | } |
64db4cff PM |
2214 | |
2215 | /* Has RCU gone idle with this CPU needing another grace period? */ | |
7ba5c840 PM |
2216 | if (cpu_needs_another_gp(rsp, rdp)) { |
2217 | rdp->n_rp_cpu_needs_gp++; | |
64db4cff | 2218 | return 1; |
7ba5c840 | 2219 | } |
64db4cff PM |
2220 | |
2221 | /* Has another RCU grace period completed? */ | |
2f51f988 | 2222 | if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */ |
7ba5c840 | 2223 | rdp->n_rp_gp_completed++; |
64db4cff | 2224 | return 1; |
7ba5c840 | 2225 | } |
64db4cff PM |
2226 | |
2227 | /* Has a new RCU grace period started? */ | |
2f51f988 | 2228 | if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */ |
7ba5c840 | 2229 | rdp->n_rp_gp_started++; |
64db4cff | 2230 | return 1; |
7ba5c840 | 2231 | } |
64db4cff PM |
2232 | |
2233 | /* Has an RCU GP gone long enough to send resched IPIs &c? */ | |
fc2219d4 | 2234 | if (rcu_gp_in_progress(rsp) && |
20133cfc | 2235 | ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) { |
7ba5c840 | 2236 | rdp->n_rp_need_fqs++; |
64db4cff | 2237 | return 1; |
7ba5c840 | 2238 | } |
64db4cff PM |
2239 | |
2240 | /* nothing to do */ | |
7ba5c840 | 2241 | rdp->n_rp_need_nothing++; |
64db4cff PM |
2242 | return 0; |
2243 | } | |
2244 | ||
2245 | /* | |
2246 | * Check to see if there is any immediate RCU-related work to be done | |
2247 | * by the current CPU, returning 1 if so. This function is part of the | |
2248 | * RCU implementation; it is -not- an exported member of the RCU API. | |
2249 | */ | |
a157229c | 2250 | static int rcu_pending(int cpu) |
64db4cff | 2251 | { |
d6714c22 | 2252 | return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) || |
f41d911f PM |
2253 | __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) || |
2254 | rcu_preempt_pending(cpu); | |
64db4cff PM |
2255 | } |
2256 | ||
2257 | /* | |
2258 | * Check to see if any future RCU-related work will need to be done | |
2259 | * by the current CPU, even if none need be done immediately, returning | |
8bd93a2c | 2260 | * 1 if so. |
64db4cff | 2261 | */ |
aea1b35e | 2262 | static int rcu_cpu_has_callbacks(int cpu) |
64db4cff PM |
2263 | { |
2264 | /* RCU callbacks either ready or pending? */ | |
d6714c22 | 2265 | return per_cpu(rcu_sched_data, cpu).nxtlist || |
f41d911f | 2266 | per_cpu(rcu_bh_data, cpu).nxtlist || |
30fbcc90 | 2267 | rcu_preempt_cpu_has_callbacks(cpu); |
64db4cff PM |
2268 | } |
2269 | ||
b1420f1c PM |
2270 | /* |
2271 | * RCU callback function for _rcu_barrier(). If we are last, wake | |
2272 | * up the task executing _rcu_barrier(). | |
2273 | */ | |
d0ec774c PM |
2274 | static void rcu_barrier_callback(struct rcu_head *notused) |
2275 | { | |
2276 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) | |
2277 | complete(&rcu_barrier_completion); | |
2278 | } | |
2279 | ||
2280 | /* | |
2281 | * Called with preemption disabled, and from cross-cpu IRQ context. | |
2282 | */ | |
2283 | static void rcu_barrier_func(void *type) | |
2284 | { | |
2285 | int cpu = smp_processor_id(); | |
2286 | struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu); | |
2287 | void (*call_rcu_func)(struct rcu_head *head, | |
2288 | void (*func)(struct rcu_head *head)); | |
2289 | ||
2290 | atomic_inc(&rcu_barrier_cpu_count); | |
2291 | call_rcu_func = type; | |
2292 | call_rcu_func(head, rcu_barrier_callback); | |
2293 | } | |
2294 | ||
d0ec774c PM |
2295 | /* |
2296 | * Orchestrate the specified type of RCU barrier, waiting for all | |
2297 | * RCU callbacks of the specified type to complete. | |
2298 | */ | |
e74f4c45 PM |
2299 | static void _rcu_barrier(struct rcu_state *rsp, |
2300 | void (*call_rcu_func)(struct rcu_head *head, | |
d0ec774c PM |
2301 | void (*func)(struct rcu_head *head))) |
2302 | { | |
b1420f1c PM |
2303 | int cpu; |
2304 | unsigned long flags; | |
2305 | struct rcu_data *rdp; | |
2306 | struct rcu_head rh; | |
2307 | ||
2308 | init_rcu_head_on_stack(&rh); | |
2309 | ||
e74f4c45 | 2310 | /* Take mutex to serialize concurrent rcu_barrier() requests. */ |
d0ec774c | 2311 | mutex_lock(&rcu_barrier_mutex); |
b1420f1c PM |
2312 | |
2313 | smp_mb(); /* Prevent any prior operations from leaking in. */ | |
2314 | ||
d0ec774c | 2315 | /* |
b1420f1c PM |
2316 | * Initialize the count to one rather than to zero in order to |
2317 | * avoid a too-soon return to zero in case of a short grace period | |
2318 | * (or preemption of this task). Also flag this task as doing | |
2319 | * an rcu_barrier(). This will prevent anyone else from adopting | |
2320 | * orphaned callbacks, which could cause otherwise failure if a | |
2321 | * CPU went offline and quickly came back online. To see this, | |
2322 | * consider the following sequence of events: | |
2323 | * | |
2324 | * 1. We cause CPU 0 to post an rcu_barrier_callback() callback. | |
2325 | * 2. CPU 1 goes offline, orphaning its callbacks. | |
2326 | * 3. CPU 0 adopts CPU 1's orphaned callbacks. | |
2327 | * 4. CPU 1 comes back online. | |
2328 | * 5. We cause CPU 1 to post an rcu_barrier_callback() callback. | |
2329 | * 6. Both rcu_barrier_callback() callbacks are invoked, awakening | |
2330 | * us -- but before CPU 1's orphaned callbacks are invoked!!! | |
d0ec774c | 2331 | */ |
b1420f1c | 2332 | init_completion(&rcu_barrier_completion); |
d0ec774c | 2333 | atomic_set(&rcu_barrier_cpu_count, 1); |
b1420f1c PM |
2334 | raw_spin_lock_irqsave(&rsp->onofflock, flags); |
2335 | rsp->rcu_barrier_in_progress = current; | |
2336 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); | |
2337 | ||
2338 | /* | |
2339 | * Force every CPU with callbacks to register a new callback | |
2340 | * that will tell us when all the preceding callbacks have | |
2341 | * been invoked. If an offline CPU has callbacks, wait for | |
2342 | * it to either come back online or to finish orphaning those | |
2343 | * callbacks. | |
2344 | */ | |
2345 | for_each_possible_cpu(cpu) { | |
2346 | preempt_disable(); | |
2347 | rdp = per_cpu_ptr(rsp->rda, cpu); | |
2348 | if (cpu_is_offline(cpu)) { | |
2349 | preempt_enable(); | |
2350 | while (cpu_is_offline(cpu) && ACCESS_ONCE(rdp->qlen)) | |
2351 | schedule_timeout_interruptible(1); | |
2352 | } else if (ACCESS_ONCE(rdp->qlen)) { | |
2353 | smp_call_function_single(cpu, rcu_barrier_func, | |
2354 | (void *)call_rcu_func, 1); | |
2355 | preempt_enable(); | |
2356 | } else { | |
2357 | preempt_enable(); | |
2358 | } | |
2359 | } | |
2360 | ||
2361 | /* | |
2362 | * Now that all online CPUs have rcu_barrier_callback() callbacks | |
2363 | * posted, we can adopt all of the orphaned callbacks and place | |
2364 | * an rcu_barrier_callback() callback after them. When that is done, | |
2365 | * we are guaranteed to have an rcu_barrier_callback() callback | |
2366 | * following every callback that could possibly have been | |
2367 | * registered before _rcu_barrier() was called. | |
2368 | */ | |
2369 | raw_spin_lock_irqsave(&rsp->onofflock, flags); | |
2370 | rcu_adopt_orphan_cbs(rsp); | |
2371 | rsp->rcu_barrier_in_progress = NULL; | |
2372 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); | |
2373 | atomic_inc(&rcu_barrier_cpu_count); | |
2374 | smp_mb__after_atomic_inc(); /* Ensure atomic_inc() before callback. */ | |
2375 | call_rcu_func(&rh, rcu_barrier_callback); | |
2376 | ||
2377 | /* | |
2378 | * Now that we have an rcu_barrier_callback() callback on each | |
2379 | * CPU, and thus each counted, remove the initial count. | |
2380 | */ | |
d0ec774c PM |
2381 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) |
2382 | complete(&rcu_barrier_completion); | |
b1420f1c PM |
2383 | |
2384 | /* Wait for all rcu_barrier_callback() callbacks to be invoked. */ | |
d0ec774c | 2385 | wait_for_completion(&rcu_barrier_completion); |
b1420f1c PM |
2386 | |
2387 | /* Other rcu_barrier() invocations can now safely proceed. */ | |
d0ec774c | 2388 | mutex_unlock(&rcu_barrier_mutex); |
b1420f1c PM |
2389 | |
2390 | destroy_rcu_head_on_stack(&rh); | |
d0ec774c | 2391 | } |
d0ec774c PM |
2392 | |
2393 | /** | |
2394 | * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete. | |
2395 | */ | |
2396 | void rcu_barrier_bh(void) | |
2397 | { | |
e74f4c45 | 2398 | _rcu_barrier(&rcu_bh_state, call_rcu_bh); |
d0ec774c PM |
2399 | } |
2400 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); | |
2401 | ||
2402 | /** | |
2403 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. | |
2404 | */ | |
2405 | void rcu_barrier_sched(void) | |
2406 | { | |
e74f4c45 | 2407 | _rcu_barrier(&rcu_sched_state, call_rcu_sched); |
d0ec774c PM |
2408 | } |
2409 | EXPORT_SYMBOL_GPL(rcu_barrier_sched); | |
2410 | ||
64db4cff | 2411 | /* |
27569620 | 2412 | * Do boot-time initialization of a CPU's per-CPU RCU data. |
64db4cff | 2413 | */ |
27569620 PM |
2414 | static void __init |
2415 | rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp) | |
64db4cff PM |
2416 | { |
2417 | unsigned long flags; | |
394f99a9 | 2418 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
27569620 PM |
2419 | struct rcu_node *rnp = rcu_get_root(rsp); |
2420 | ||
2421 | /* Set up local state, ensuring consistent view of global state. */ | |
1304afb2 | 2422 | raw_spin_lock_irqsave(&rnp->lock, flags); |
27569620 | 2423 | rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo); |
3f5d3ea6 | 2424 | init_callback_list(rdp); |
486e2593 | 2425 | rdp->qlen_lazy = 0; |
1d1fb395 | 2426 | ACCESS_ONCE(rdp->qlen) = 0; |
27569620 | 2427 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); |
29e37d81 | 2428 | WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != DYNTICK_TASK_EXIT_IDLE); |
9b2e4f18 | 2429 | WARN_ON_ONCE(atomic_read(&rdp->dynticks->dynticks) != 1); |
27569620 | 2430 | rdp->cpu = cpu; |
d4c08f2a | 2431 | rdp->rsp = rsp; |
1304afb2 | 2432 | raw_spin_unlock_irqrestore(&rnp->lock, flags); |
27569620 PM |
2433 | } |
2434 | ||
2435 | /* | |
2436 | * Initialize a CPU's per-CPU RCU data. Note that only one online or | |
2437 | * offline event can be happening at a given time. Note also that we | |
2438 | * can accept some slop in the rsp->completed access due to the fact | |
2439 | * that this CPU cannot possibly have any RCU callbacks in flight yet. | |
64db4cff | 2440 | */ |
e4fa4c97 | 2441 | static void __cpuinit |
6cc68793 | 2442 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptible) |
64db4cff PM |
2443 | { |
2444 | unsigned long flags; | |
64db4cff | 2445 | unsigned long mask; |
394f99a9 | 2446 | struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); |
64db4cff PM |
2447 | struct rcu_node *rnp = rcu_get_root(rsp); |
2448 | ||
2449 | /* Set up local state, ensuring consistent view of global state. */ | |
1304afb2 | 2450 | raw_spin_lock_irqsave(&rnp->lock, flags); |
64db4cff | 2451 | rdp->beenonline = 1; /* We have now been online. */ |
6cc68793 | 2452 | rdp->preemptible = preemptible; |
37c72e56 PM |
2453 | rdp->qlen_last_fqs_check = 0; |
2454 | rdp->n_force_qs_snap = rsp->n_force_qs; | |
64db4cff | 2455 | rdp->blimit = blimit; |
29e37d81 | 2456 | rdp->dynticks->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE; |
c92b131b PM |
2457 | atomic_set(&rdp->dynticks->dynticks, |
2458 | (atomic_read(&rdp->dynticks->dynticks) & ~0x1) + 1); | |
7cb92499 | 2459 | rcu_prepare_for_idle_init(cpu); |
1304afb2 | 2460 | raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
64db4cff PM |
2461 | |
2462 | /* | |
2463 | * A new grace period might start here. If so, we won't be part | |
2464 | * of it, but that is OK, as we are currently in a quiescent state. | |
2465 | */ | |
2466 | ||
2467 | /* Exclude any attempts to start a new GP on large systems. */ | |
1304afb2 | 2468 | raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
64db4cff PM |
2469 | |
2470 | /* Add CPU to rcu_node bitmasks. */ | |
2471 | rnp = rdp->mynode; | |
2472 | mask = rdp->grpmask; | |
2473 | do { | |
2474 | /* Exclude any attempts to start a new GP on small systems. */ | |
1304afb2 | 2475 | raw_spin_lock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
2476 | rnp->qsmaskinit |= mask; |
2477 | mask = rnp->grpmask; | |
d09b62df | 2478 | if (rnp == rdp->mynode) { |
06ae115a PM |
2479 | /* |
2480 | * If there is a grace period in progress, we will | |
2481 | * set up to wait for it next time we run the | |
2482 | * RCU core code. | |
2483 | */ | |
2484 | rdp->gpnum = rnp->completed; | |
d09b62df | 2485 | rdp->completed = rnp->completed; |
06ae115a PM |
2486 | rdp->passed_quiesce = 0; |
2487 | rdp->qs_pending = 0; | |
e4cc1f22 | 2488 | rdp->passed_quiesce_gpnum = rnp->gpnum - 1; |
d4c08f2a | 2489 | trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuonl"); |
d09b62df | 2490 | } |
1304afb2 | 2491 | raw_spin_unlock(&rnp->lock); /* irqs already disabled. */ |
64db4cff PM |
2492 | rnp = rnp->parent; |
2493 | } while (rnp != NULL && !(rnp->qsmaskinit & mask)); | |
2494 | ||
1304afb2 | 2495 | raw_spin_unlock_irqrestore(&rsp->onofflock, flags); |
64db4cff PM |
2496 | } |
2497 | ||
d72bce0e | 2498 | static void __cpuinit rcu_prepare_cpu(int cpu) |
64db4cff | 2499 | { |
f41d911f PM |
2500 | rcu_init_percpu_data(cpu, &rcu_sched_state, 0); |
2501 | rcu_init_percpu_data(cpu, &rcu_bh_state, 0); | |
2502 | rcu_preempt_init_percpu_data(cpu); | |
64db4cff PM |
2503 | } |
2504 | ||
2505 | /* | |
f41d911f | 2506 | * Handle CPU online/offline notification events. |
64db4cff | 2507 | */ |
9f680ab4 PM |
2508 | static int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
2509 | unsigned long action, void *hcpu) | |
64db4cff PM |
2510 | { |
2511 | long cpu = (long)hcpu; | |
27f4d280 | 2512 | struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, cpu); |
a26ac245 | 2513 | struct rcu_node *rnp = rdp->mynode; |
64db4cff | 2514 | |
300df91c | 2515 | trace_rcu_utilization("Start CPU hotplug"); |
64db4cff PM |
2516 | switch (action) { |
2517 | case CPU_UP_PREPARE: | |
2518 | case CPU_UP_PREPARE_FROZEN: | |
d72bce0e PZ |
2519 | rcu_prepare_cpu(cpu); |
2520 | rcu_prepare_kthreads(cpu); | |
a26ac245 PM |
2521 | break; |
2522 | case CPU_ONLINE: | |
0f962a5e PM |
2523 | case CPU_DOWN_FAILED: |
2524 | rcu_node_kthread_setaffinity(rnp, -1); | |
e3995a25 | 2525 | rcu_cpu_kthread_setrt(cpu, 1); |
0f962a5e PM |
2526 | break; |
2527 | case CPU_DOWN_PREPARE: | |
2528 | rcu_node_kthread_setaffinity(rnp, cpu); | |
e3995a25 | 2529 | rcu_cpu_kthread_setrt(cpu, 0); |
64db4cff | 2530 | break; |
d0ec774c PM |
2531 | case CPU_DYING: |
2532 | case CPU_DYING_FROZEN: | |
2533 | /* | |
2d999e03 PM |
2534 | * The whole machine is "stopped" except this CPU, so we can |
2535 | * touch any data without introducing corruption. We send the | |
2536 | * dying CPU's callbacks to an arbitrarily chosen online CPU. | |
d0ec774c | 2537 | */ |
e5601400 PM |
2538 | rcu_cleanup_dying_cpu(&rcu_bh_state); |
2539 | rcu_cleanup_dying_cpu(&rcu_sched_state); | |
2540 | rcu_preempt_cleanup_dying_cpu(); | |
7cb92499 | 2541 | rcu_cleanup_after_idle(cpu); |
d0ec774c | 2542 | break; |
64db4cff PM |
2543 | case CPU_DEAD: |
2544 | case CPU_DEAD_FROZEN: | |
2545 | case CPU_UP_CANCELED: | |
2546 | case CPU_UP_CANCELED_FROZEN: | |
e5601400 PM |
2547 | rcu_cleanup_dead_cpu(cpu, &rcu_bh_state); |
2548 | rcu_cleanup_dead_cpu(cpu, &rcu_sched_state); | |
2549 | rcu_preempt_cleanup_dead_cpu(cpu); | |
64db4cff PM |
2550 | break; |
2551 | default: | |
2552 | break; | |
2553 | } | |
300df91c | 2554 | trace_rcu_utilization("End CPU hotplug"); |
64db4cff PM |
2555 | return NOTIFY_OK; |
2556 | } | |
2557 | ||
bbad9379 PM |
2558 | /* |
2559 | * This function is invoked towards the end of the scheduler's initialization | |
2560 | * process. Before this is called, the idle task might contain | |
2561 | * RCU read-side critical sections (during which time, this idle | |
2562 | * task is booting the system). After this function is called, the | |
2563 | * idle tasks are prohibited from containing RCU read-side critical | |
2564 | * sections. This function also enables RCU lockdep checking. | |
2565 | */ | |
2566 | void rcu_scheduler_starting(void) | |
2567 | { | |
2568 | WARN_ON(num_online_cpus() != 1); | |
2569 | WARN_ON(nr_context_switches() > 0); | |
2570 | rcu_scheduler_active = 1; | |
2571 | } | |
2572 | ||
64db4cff PM |
2573 | /* |
2574 | * Compute the per-level fanout, either using the exact fanout specified | |
2575 | * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT. | |
2576 | */ | |
2577 | #ifdef CONFIG_RCU_FANOUT_EXACT | |
2578 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
2579 | { | |
2580 | int i; | |
2581 | ||
0209f649 | 2582 | for (i = NUM_RCU_LVLS - 1; i > 0; i--) |
64db4cff | 2583 | rsp->levelspread[i] = CONFIG_RCU_FANOUT; |
8932a63d | 2584 | rsp->levelspread[0] = CONFIG_RCU_FANOUT_LEAF; |
64db4cff PM |
2585 | } |
2586 | #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
2587 | static void __init rcu_init_levelspread(struct rcu_state *rsp) | |
2588 | { | |
2589 | int ccur; | |
2590 | int cprv; | |
2591 | int i; | |
2592 | ||
2593 | cprv = NR_CPUS; | |
2594 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
2595 | ccur = rsp->levelcnt[i]; | |
2596 | rsp->levelspread[i] = (cprv + ccur - 1) / ccur; | |
2597 | cprv = ccur; | |
2598 | } | |
2599 | } | |
2600 | #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */ | |
2601 | ||
2602 | /* | |
2603 | * Helper function for rcu_init() that initializes one rcu_state structure. | |
2604 | */ | |
394f99a9 LJ |
2605 | static void __init rcu_init_one(struct rcu_state *rsp, |
2606 | struct rcu_data __percpu *rda) | |
64db4cff | 2607 | { |
b6407e86 PM |
2608 | static char *buf[] = { "rcu_node_level_0", |
2609 | "rcu_node_level_1", | |
2610 | "rcu_node_level_2", | |
2611 | "rcu_node_level_3" }; /* Match MAX_RCU_LVLS */ | |
64db4cff PM |
2612 | int cpustride = 1; |
2613 | int i; | |
2614 | int j; | |
2615 | struct rcu_node *rnp; | |
2616 | ||
b6407e86 PM |
2617 | BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf)); /* Fix buf[] init! */ |
2618 | ||
64db4cff PM |
2619 | /* Initialize the level-tracking arrays. */ |
2620 | ||
2621 | for (i = 1; i < NUM_RCU_LVLS; i++) | |
2622 | rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1]; | |
2623 | rcu_init_levelspread(rsp); | |
2624 | ||
2625 | /* Initialize the elements themselves, starting from the leaves. */ | |
2626 | ||
2627 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { | |
2628 | cpustride *= rsp->levelspread[i]; | |
2629 | rnp = rsp->level[i]; | |
2630 | for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { | |
1304afb2 | 2631 | raw_spin_lock_init(&rnp->lock); |
b6407e86 PM |
2632 | lockdep_set_class_and_name(&rnp->lock, |
2633 | &rcu_node_class[i], buf[i]); | |
f41d911f | 2634 | rnp->gpnum = 0; |
64db4cff PM |
2635 | rnp->qsmask = 0; |
2636 | rnp->qsmaskinit = 0; | |
2637 | rnp->grplo = j * cpustride; | |
2638 | rnp->grphi = (j + 1) * cpustride - 1; | |
2639 | if (rnp->grphi >= NR_CPUS) | |
2640 | rnp->grphi = NR_CPUS - 1; | |
2641 | if (i == 0) { | |
2642 | rnp->grpnum = 0; | |
2643 | rnp->grpmask = 0; | |
2644 | rnp->parent = NULL; | |
2645 | } else { | |
2646 | rnp->grpnum = j % rsp->levelspread[i - 1]; | |
2647 | rnp->grpmask = 1UL << rnp->grpnum; | |
2648 | rnp->parent = rsp->level[i - 1] + | |
2649 | j / rsp->levelspread[i - 1]; | |
2650 | } | |
2651 | rnp->level = i; | |
12f5f524 | 2652 | INIT_LIST_HEAD(&rnp->blkd_tasks); |
64db4cff PM |
2653 | } |
2654 | } | |
0c34029a | 2655 | |
394f99a9 | 2656 | rsp->rda = rda; |
0c34029a LJ |
2657 | rnp = rsp->level[NUM_RCU_LVLS - 1]; |
2658 | for_each_possible_cpu(i) { | |
4a90a068 | 2659 | while (i > rnp->grphi) |
0c34029a | 2660 | rnp++; |
394f99a9 | 2661 | per_cpu_ptr(rsp->rda, i)->mynode = rnp; |
0c34029a LJ |
2662 | rcu_boot_init_percpu_data(i, rsp); |
2663 | } | |
64db4cff PM |
2664 | } |
2665 | ||
9f680ab4 | 2666 | void __init rcu_init(void) |
64db4cff | 2667 | { |
017c4261 | 2668 | int cpu; |
9f680ab4 | 2669 | |
f41d911f | 2670 | rcu_bootup_announce(); |
394f99a9 LJ |
2671 | rcu_init_one(&rcu_sched_state, &rcu_sched_data); |
2672 | rcu_init_one(&rcu_bh_state, &rcu_bh_data); | |
f41d911f | 2673 | __rcu_init_preempt(); |
09223371 | 2674 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
9f680ab4 PM |
2675 | |
2676 | /* | |
2677 | * We don't need protection against CPU-hotplug here because | |
2678 | * this is called early in boot, before either interrupts | |
2679 | * or the scheduler are operational. | |
2680 | */ | |
2681 | cpu_notifier(rcu_cpu_notify, 0); | |
017c4261 PM |
2682 | for_each_online_cpu(cpu) |
2683 | rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu); | |
c68de209 | 2684 | check_cpu_stall_init(); |
64db4cff PM |
2685 | } |
2686 | ||
1eba8f84 | 2687 | #include "rcutree_plugin.h" |