2 * Read-Copy Update mechanism for mutual exclusion (tree-based version)
3 * Internal non-public definitions.
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, you can access it online at
17 * http://www.gnu.org/licenses/gpl-2.0.html.
19 * Copyright IBM Corporation, 2008
21 * Author: Ingo Molnar <mingo@elte.hu>
22 * Paul E. McKenney <paulmck@linux.vnet.ibm.com>
25 #include <linux/cache.h>
26 #include <linux/spinlock.h>
27 #include <linux/threads.h>
28 #include <linux/cpumask.h>
29 #include <linux/seqlock.h>
30 #include <linux/swait.h>
31 #include <linux/stop_machine.h>
34 * Define shape of hierarchy based on NR_CPUS, CONFIG_RCU_FANOUT, and
35 * CONFIG_RCU_FANOUT_LEAF.
36 * In theory, it should be possible to add more levels straightforwardly.
37 * In practice, this did work well going from three levels to four.
38 * Of course, your mileage may vary.
41 #ifdef CONFIG_RCU_FANOUT
42 #define RCU_FANOUT CONFIG_RCU_FANOUT
43 #else /* #ifdef CONFIG_RCU_FANOUT */
45 # define RCU_FANOUT 64
47 # define RCU_FANOUT 32
49 #endif /* #else #ifdef CONFIG_RCU_FANOUT */
51 #ifdef CONFIG_RCU_FANOUT_LEAF
52 #define RCU_FANOUT_LEAF CONFIG_RCU_FANOUT_LEAF
53 #else /* #ifdef CONFIG_RCU_FANOUT_LEAF */
55 # define RCU_FANOUT_LEAF 64
57 # define RCU_FANOUT_LEAF 32
59 #endif /* #else #ifdef CONFIG_RCU_FANOUT_LEAF */
61 #define RCU_FANOUT_1 (RCU_FANOUT_LEAF)
62 #define RCU_FANOUT_2 (RCU_FANOUT_1 * RCU_FANOUT)
63 #define RCU_FANOUT_3 (RCU_FANOUT_2 * RCU_FANOUT)
64 #define RCU_FANOUT_4 (RCU_FANOUT_3 * RCU_FANOUT)
66 #if NR_CPUS <= RCU_FANOUT_1
67 # define RCU_NUM_LVLS 1
68 # define NUM_RCU_LVL_0 1
69 # define NUM_RCU_NODES NUM_RCU_LVL_0
70 # define NUM_RCU_LVL_INIT { NUM_RCU_LVL_0 }
71 # define RCU_NODE_NAME_INIT { "rcu_node_0" }
72 # define RCU_FQS_NAME_INIT { "rcu_node_fqs_0" }
73 # define RCU_EXP_NAME_INIT { "rcu_node_exp_0" }
74 #elif NR_CPUS <= RCU_FANOUT_2
75 # define RCU_NUM_LVLS 2
76 # define NUM_RCU_LVL_0 1
77 # define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
78 # define NUM_RCU_NODES (NUM_RCU_LVL_0 + NUM_RCU_LVL_1)
79 # define NUM_RCU_LVL_INIT { NUM_RCU_LVL_0, NUM_RCU_LVL_1 }
80 # define RCU_NODE_NAME_INIT { "rcu_node_0", "rcu_node_1" }
81 # define RCU_FQS_NAME_INIT { "rcu_node_fqs_0", "rcu_node_fqs_1" }
82 # define RCU_EXP_NAME_INIT { "rcu_node_exp_0", "rcu_node_exp_1" }
83 #elif NR_CPUS <= RCU_FANOUT_3
84 # define RCU_NUM_LVLS 3
85 # define NUM_RCU_LVL_0 1
86 # define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
87 # define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
88 # define NUM_RCU_NODES (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2)
89 # define NUM_RCU_LVL_INIT { NUM_RCU_LVL_0, NUM_RCU_LVL_1, NUM_RCU_LVL_2 }
90 # define RCU_NODE_NAME_INIT { "rcu_node_0", "rcu_node_1", "rcu_node_2" }
91 # define RCU_FQS_NAME_INIT { "rcu_node_fqs_0", "rcu_node_fqs_1", "rcu_node_fqs_2" }
92 # define RCU_EXP_NAME_INIT { "rcu_node_exp_0", "rcu_node_exp_1", "rcu_node_exp_2" }
93 #elif NR_CPUS <= RCU_FANOUT_4
94 # define RCU_NUM_LVLS 4
95 # define NUM_RCU_LVL_0 1
96 # define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3)
97 # define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
98 # define NUM_RCU_LVL_3 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
99 # define NUM_RCU_NODES (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3)
100 # define NUM_RCU_LVL_INIT { NUM_RCU_LVL_0, NUM_RCU_LVL_1, NUM_RCU_LVL_2, NUM_RCU_LVL_3 }
101 # define RCU_NODE_NAME_INIT { "rcu_node_0", "rcu_node_1", "rcu_node_2", "rcu_node_3" }
102 # define RCU_FQS_NAME_INIT { "rcu_node_fqs_0", "rcu_node_fqs_1", "rcu_node_fqs_2", "rcu_node_fqs_3" }
103 # define RCU_EXP_NAME_INIT { "rcu_node_exp_0", "rcu_node_exp_1", "rcu_node_exp_2", "rcu_node_exp_3" }
105 # error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
106 #endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
108 extern int rcu_num_lvls;
109 extern int rcu_num_nodes;
112 * Dynticks per-CPU state.
114 struct rcu_dynticks {
115 long long dynticks_nesting; /* Track irq/process nesting level. */
116 /* Process level is worth LLONG_MAX/2. */
117 int dynticks_nmi_nesting; /* Track NMI nesting level. */
118 atomic_t dynticks; /* Even value for idle, else odd. */
119 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
120 long long dynticks_idle_nesting;
121 /* irq/process nesting level from idle. */
122 atomic_t dynticks_idle; /* Even value for idle, else odd. */
123 /* "Idle" excludes userspace execution. */
124 unsigned long dynticks_idle_jiffies;
125 /* End of last non-NMI non-idle period. */
126 #endif /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
127 #ifdef CONFIG_RCU_FAST_NO_HZ
128 bool all_lazy; /* Are all CPU's CBs lazy? */
129 unsigned long nonlazy_posted;
130 /* # times non-lazy CBs posted to CPU. */
131 unsigned long nonlazy_posted_snap;
132 /* idle-period nonlazy_posted snapshot. */
133 unsigned long last_accelerate;
134 /* Last jiffy CBs were accelerated. */
135 unsigned long last_advance_all;
136 /* Last jiffy CBs were all advanced. */
137 int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */
138 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
141 /* RCU's kthread states for tracing. */
142 #define RCU_KTHREAD_STOPPED 0
143 #define RCU_KTHREAD_RUNNING 1
144 #define RCU_KTHREAD_WAITING 2
145 #define RCU_KTHREAD_OFFCPU 3
146 #define RCU_KTHREAD_YIELDING 4
147 #define RCU_KTHREAD_MAX 4
150 * Definition for node within the RCU grace-period-detection hierarchy.
153 raw_spinlock_t lock; /* Root rcu_node's lock protects some */
154 /* rcu_state fields as well as following. */
155 unsigned long gpnum; /* Current grace period for this node. */
156 /* This will either be equal to or one */
157 /* behind the root rcu_node's gpnum. */
158 unsigned long completed; /* Last GP completed for this node. */
159 /* This will either be equal to or one */
160 /* behind the root rcu_node's gpnum. */
161 unsigned long qsmask; /* CPUs or groups that need to switch in */
162 /* order for current grace period to proceed.*/
163 /* In leaf rcu_node, each bit corresponds to */
164 /* an rcu_data structure, otherwise, each */
165 /* bit corresponds to a child rcu_node */
167 unsigned long qsmaskinit;
168 /* Per-GP initial value for qsmask. */
169 /* Initialized from ->qsmaskinitnext at the */
170 /* beginning of each grace period. */
171 unsigned long qsmaskinitnext;
172 /* Online CPUs for next grace period. */
173 unsigned long expmask; /* CPUs or groups that need to check in */
174 /* to allow the current expedited GP */
176 unsigned long expmaskinit;
177 /* Per-GP initial values for expmask. */
178 /* Initialized from ->expmaskinitnext at the */
179 /* beginning of each expedited GP. */
180 unsigned long expmaskinitnext;
181 /* Online CPUs for next expedited GP. */
182 unsigned long grpmask; /* Mask to apply to parent qsmask. */
183 /* Only one bit will be set in this mask. */
184 int grplo; /* lowest-numbered CPU or group here. */
185 int grphi; /* highest-numbered CPU or group here. */
186 u8 grpnum; /* CPU/group number for next level up. */
187 u8 level; /* root is at level 0. */
188 bool wait_blkd_tasks;/* Necessary to wait for blocked tasks to */
189 /* exit RCU read-side critical sections */
190 /* before propagating offline up the */
192 struct rcu_node *parent;
193 struct list_head blkd_tasks;
194 /* Tasks blocked in RCU read-side critical */
195 /* section. Tasks are placed at the head */
196 /* of this list and age towards the tail. */
197 struct list_head *gp_tasks;
198 /* Pointer to the first task blocking the */
199 /* current grace period, or NULL if there */
200 /* is no such task. */
201 struct list_head *exp_tasks;
202 /* Pointer to the first task blocking the */
203 /* current expedited grace period, or NULL */
204 /* if there is no such task. If there */
205 /* is no current expedited grace period, */
206 /* then there can cannot be any such task. */
207 struct list_head *boost_tasks;
208 /* Pointer to first task that needs to be */
209 /* priority boosted, or NULL if no priority */
210 /* boosting is needed for this rcu_node */
211 /* structure. If there are no tasks */
212 /* queued on this rcu_node structure that */
213 /* are blocking the current grace period, */
214 /* there can be no such task. */
215 struct rt_mutex boost_mtx;
216 /* Used only for the priority-boosting */
217 /* side effect, not as a lock. */
218 unsigned long boost_time;
219 /* When to start boosting (jiffies). */
220 struct task_struct *boost_kthread_task;
221 /* kthread that takes care of priority */
222 /* boosting for this rcu_node structure. */
223 unsigned int boost_kthread_status;
224 /* State of boost_kthread_task for tracing. */
225 unsigned long n_tasks_boosted;
226 /* Total number of tasks boosted. */
227 unsigned long n_exp_boosts;
228 /* Number of tasks boosted for expedited GP. */
229 unsigned long n_normal_boosts;
230 /* Number of tasks boosted for normal GP. */
231 unsigned long n_balk_blkd_tasks;
232 /* Refused to boost: no blocked tasks. */
233 unsigned long n_balk_exp_gp_tasks;
234 /* Refused to boost: nothing blocking GP. */
235 unsigned long n_balk_boost_tasks;
236 /* Refused to boost: already boosting. */
237 unsigned long n_balk_notblocked;
238 /* Refused to boost: RCU RS CS still running. */
239 unsigned long n_balk_notyet;
240 /* Refused to boost: not yet time. */
241 unsigned long n_balk_nos;
242 /* Refused to boost: not sure why, though. */
243 /* This can happen due to race conditions. */
244 #ifdef CONFIG_RCU_NOCB_CPU
245 struct swait_queue_head nocb_gp_wq[2];
246 /* Place for rcu_nocb_kthread() to wait GP. */
247 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
248 int need_future_gp[2];
249 /* Counts of upcoming no-CB GP requests. */
250 raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp;
252 struct mutex exp_funnel_mutex ____cacheline_internodealigned_in_smp;
253 } ____cacheline_internodealigned_in_smp;
256 * Do a full breadth-first scan of the rcu_node structures for the
257 * specified rcu_state structure.
259 #define rcu_for_each_node_breadth_first(rsp, rnp) \
260 for ((rnp) = &(rsp)->node[0]; \
261 (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
264 * Do a breadth-first scan of the non-leaf rcu_node structures for the
265 * specified rcu_state structure. Note that if there is a singleton
266 * rcu_node tree with but one rcu_node structure, this loop is a no-op.
268 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
269 for ((rnp) = &(rsp)->node[0]; \
270 (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
273 * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
274 * structure. Note that if there is a singleton rcu_node tree with but
275 * one rcu_node structure, this loop -will- visit the rcu_node structure.
276 * It is still a leaf node, even if it is also the root node.
278 #define rcu_for_each_leaf_node(rsp, rnp) \
279 for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
280 (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
283 * Union to allow "aggregate OR" operation on the need for a quiescent
284 * state by the normal and expedited grace periods.
291 u16 s; /* Set of bits, aggregate OR here. */
294 /* Index values for nxttail array in struct rcu_data. */
295 #define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
296 #define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
297 #define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
298 #define RCU_NEXT_TAIL 3
299 #define RCU_NEXT_SIZE 4
301 /* Per-CPU data for read-copy update. */
303 /* 1) quiescent-state and grace-period handling : */
304 unsigned long completed; /* Track rsp->completed gp number */
305 /* in order to detect GP end. */
306 unsigned long gpnum; /* Highest gp number that this CPU */
307 /* is aware of having started. */
308 unsigned long rcu_qs_ctr_snap;/* Snapshot of rcu_qs_ctr to check */
309 /* for rcu_all_qs() invocations. */
310 union rcu_noqs cpu_no_qs; /* No QSes yet for this CPU. */
311 bool core_needs_qs; /* Core waits for quiesc state. */
312 bool beenonline; /* CPU online at least once. */
313 bool gpwrap; /* Possible gpnum/completed wrap. */
314 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
315 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
316 unsigned long ticks_this_gp; /* The number of scheduling-clock */
317 /* ticks this CPU has handled */
318 /* during and after the last grace */
319 /* period it is aware of. */
321 /* 2) batch handling */
323 * If nxtlist is not NULL, it is partitioned as follows.
324 * Any of the partitions might be empty, in which case the
325 * pointer to that partition will be equal to the pointer for
326 * the following partition. When the list is empty, all of
327 * the nxttail elements point to the ->nxtlist pointer itself,
328 * which in that case is NULL.
330 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
331 * Entries that batch # <= ->completed
332 * The grace period for these entries has completed, and
333 * the other grace-period-completed entries may be moved
334 * here temporarily in rcu_process_callbacks().
335 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
336 * Entries that batch # <= ->completed - 1: waiting for current GP
337 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
338 * Entries known to have arrived before current GP ended
339 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
340 * Entries that might have arrived after current GP ended
341 * Note that the value of *nxttail[RCU_NEXT_TAIL] will
342 * always be NULL, as this is the end of the list.
344 struct rcu_head *nxtlist;
345 struct rcu_head **nxttail[RCU_NEXT_SIZE];
346 unsigned long nxtcompleted[RCU_NEXT_SIZE];
347 /* grace periods for sublists. */
348 long qlen_lazy; /* # of lazy queued callbacks */
349 long qlen; /* # of queued callbacks, incl lazy */
350 long qlen_last_fqs_check;
351 /* qlen at last check for QS forcing */
352 unsigned long n_cbs_invoked; /* count of RCU cbs invoked. */
353 unsigned long n_nocbs_invoked; /* count of no-CBs RCU cbs invoked. */
354 unsigned long n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
355 unsigned long n_cbs_adopted; /* RCU cbs adopted from dying CPU */
356 unsigned long n_force_qs_snap;
357 /* did other CPU force QS recently? */
358 long blimit; /* Upper limit on a processed batch */
360 /* 3) dynticks interface. */
361 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
362 int dynticks_snap; /* Per-GP tracking for dynticks. */
364 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
365 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
366 unsigned long offline_fqs; /* Kicked due to being offline. */
367 unsigned long cond_resched_completed;
368 /* Grace period that needs help */
369 /* from cond_resched(). */
371 /* 5) __rcu_pending() statistics. */
372 unsigned long n_rcu_pending; /* rcu_pending() calls since boot. */
373 unsigned long n_rp_core_needs_qs;
374 unsigned long n_rp_report_qs;
375 unsigned long n_rp_cb_ready;
376 unsigned long n_rp_cpu_needs_gp;
377 unsigned long n_rp_gp_completed;
378 unsigned long n_rp_gp_started;
379 unsigned long n_rp_nocb_defer_wakeup;
380 unsigned long n_rp_need_nothing;
382 /* 6) _rcu_barrier(), OOM callbacks, and expediting. */
383 struct rcu_head barrier_head;
384 #ifdef CONFIG_RCU_FAST_NO_HZ
385 struct rcu_head oom_head;
386 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
387 struct mutex exp_funnel_mutex;
389 /* 7) Callback offloading. */
390 #ifdef CONFIG_RCU_NOCB_CPU
391 struct rcu_head *nocb_head; /* CBs waiting for kthread. */
392 struct rcu_head **nocb_tail;
393 atomic_long_t nocb_q_count; /* # CBs waiting for nocb */
394 atomic_long_t nocb_q_count_lazy; /* invocation (all stages). */
395 struct rcu_head *nocb_follower_head; /* CBs ready to invoke. */
396 struct rcu_head **nocb_follower_tail;
397 struct swait_queue_head nocb_wq; /* For nocb kthreads to sleep on. */
398 struct task_struct *nocb_kthread;
399 int nocb_defer_wakeup; /* Defer wakeup of nocb_kthread. */
401 /* The following fields are used by the leader, hence own cacheline. */
402 struct rcu_head *nocb_gp_head ____cacheline_internodealigned_in_smp;
403 /* CBs waiting for GP. */
404 struct rcu_head **nocb_gp_tail;
405 bool nocb_leader_sleep; /* Is the nocb leader thread asleep? */
406 struct rcu_data *nocb_next_follower;
407 /* Next follower in wakeup chain. */
409 /* The following fields are used by the follower, hence new cachline. */
410 struct rcu_data *nocb_leader ____cacheline_internodealigned_in_smp;
411 /* Leader CPU takes GP-end wakeups. */
412 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
414 /* 8) RCU CPU stall data. */
415 unsigned int softirq_snap; /* Snapshot of softirq activity. */
418 struct rcu_state *rsp;
421 /* Values for nocb_defer_wakeup field in struct rcu_data. */
422 #define RCU_NOGP_WAKE_NOT 0
423 #define RCU_NOGP_WAKE 1
424 #define RCU_NOGP_WAKE_FORCE 2
426 #define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500))
427 /* For jiffies_till_first_fqs and */
428 /* and jiffies_till_next_fqs. */
430 #define RCU_JIFFIES_FQS_DIV 256 /* Very large systems need more */
431 /* delay between bouts of */
432 /* quiescent-state forcing. */
434 #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time to take */
435 /* at least one scheduling clock */
436 /* irq before ratting on them. */
438 #define rcu_wait(cond) \
441 set_current_state(TASK_INTERRUPTIBLE); \
446 __set_current_state(TASK_RUNNING); \
450 * RCU global state, including node hierarchy. This hierarchy is
451 * represented in "heap" form in a dense array. The root (first level)
452 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
453 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
454 * and the third level in ->node[m+1] and following (->node[m+1] referenced
455 * by ->level[2]). The number of levels is determined by the number of
456 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
457 * consisting of a single rcu_node.
460 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
461 struct rcu_node *level[RCU_NUM_LVLS + 1];
462 /* Hierarchy levels (+1 to */
463 /* shut bogus gcc warning) */
464 u8 flavor_mask; /* bit in flavor mask. */
465 struct rcu_data __percpu *rda; /* pointer of percu rcu_data. */
466 call_rcu_func_t call; /* call_rcu() flavor. */
467 int ncpus; /* # CPUs seen so far. */
469 /* The following fields are guarded by the root rcu_node's lock. */
471 u8 boost ____cacheline_internodealigned_in_smp;
472 /* Subject to priority boost. */
473 unsigned long gpnum; /* Current gp number. */
474 unsigned long completed; /* # of last completed gp. */
475 struct task_struct *gp_kthread; /* Task for grace periods. */
476 struct swait_queue_head gp_wq; /* Where GP task waits. */
477 short gp_flags; /* Commands for GP task. */
478 short gp_state; /* GP kthread sleep state. */
480 /* End of fields guarded by root rcu_node's lock. */
482 raw_spinlock_t orphan_lock ____cacheline_internodealigned_in_smp;
483 /* Protect following fields. */
484 struct rcu_head *orphan_nxtlist; /* Orphaned callbacks that */
485 /* need a grace period. */
486 struct rcu_head **orphan_nxttail; /* Tail of above. */
487 struct rcu_head *orphan_donelist; /* Orphaned callbacks that */
488 /* are ready to invoke. */
489 struct rcu_head **orphan_donetail; /* Tail of above. */
490 long qlen_lazy; /* Number of lazy callbacks. */
491 long qlen; /* Total number of callbacks. */
492 /* End of fields guarded by orphan_lock. */
494 struct mutex barrier_mutex; /* Guards barrier fields. */
495 atomic_t barrier_cpu_count; /* # CPUs waiting on. */
496 struct completion barrier_completion; /* Wake at barrier end. */
497 unsigned long barrier_sequence; /* ++ at start and end of */
498 /* _rcu_barrier(). */
499 /* End of fields guarded by barrier_mutex. */
501 unsigned long expedited_sequence; /* Take a ticket. */
502 atomic_long_t expedited_workdone0; /* # done by others #0. */
503 atomic_long_t expedited_workdone1; /* # done by others #1. */
504 atomic_long_t expedited_workdone2; /* # done by others #2. */
505 atomic_long_t expedited_workdone3; /* # done by others #3. */
506 atomic_long_t expedited_normal; /* # fallbacks to normal. */
507 atomic_t expedited_need_qs; /* # CPUs left to check in. */
508 struct swait_queue_head expedited_wq; /* Wait for check-ins. */
509 int ncpus_snap; /* # CPUs seen last time. */
511 unsigned long jiffies_force_qs; /* Time at which to invoke */
512 /* force_quiescent_state(). */
513 unsigned long n_force_qs; /* Number of calls to */
514 /* force_quiescent_state(). */
515 unsigned long n_force_qs_lh; /* ~Number of calls leaving */
516 /* due to lock unavailable. */
517 unsigned long n_force_qs_ngp; /* Number of calls leaving */
518 /* due to no GP active. */
519 unsigned long gp_start; /* Time at which GP started, */
520 /* but in jiffies. */
521 unsigned long gp_activity; /* Time of last GP kthread */
522 /* activity in jiffies. */
523 unsigned long jiffies_stall; /* Time at which to check */
524 /* for CPU stalls. */
525 unsigned long jiffies_resched; /* Time at which to resched */
526 /* a reluctant CPU. */
527 unsigned long n_force_qs_gpstart; /* Snapshot of n_force_qs at */
529 unsigned long gp_max; /* Maximum GP duration in */
531 const char *name; /* Name of structure. */
532 char abbr; /* Abbreviated name. */
533 struct list_head flavors; /* List of RCU flavors. */
536 /* Values for rcu_state structure's gp_flags field. */
537 #define RCU_GP_FLAG_INIT 0x1 /* Need grace-period initialization. */
538 #define RCU_GP_FLAG_FQS 0x2 /* Need grace-period quiescent-state forcing. */
540 /* Values for rcu_state structure's gp_state field. */
541 #define RCU_GP_IDLE 0 /* Initial state and no GP in progress. */
542 #define RCU_GP_WAIT_GPS 1 /* Wait for grace-period start. */
543 #define RCU_GP_DONE_GPS 2 /* Wait done for grace-period start. */
544 #define RCU_GP_WAIT_FQS 3 /* Wait for force-quiescent-state time. */
545 #define RCU_GP_DOING_FQS 4 /* Wait done for force-quiescent-state time. */
546 #define RCU_GP_CLEANUP 5 /* Grace-period cleanup started. */
547 #define RCU_GP_CLEANED 6 /* Grace-period cleanup complete. */
549 extern struct list_head rcu_struct_flavors;
551 /* Sequence through rcu_state structures for each RCU flavor. */
552 #define for_each_rcu_flavor(rsp) \
553 list_for_each_entry((rsp), &rcu_struct_flavors, flavors)
556 * RCU implementation internal declarations:
558 extern struct rcu_state rcu_sched_state;
560 #ifndef CONFIG_PREEMPT_RT_FULL
561 extern struct rcu_state rcu_bh_state;
564 #ifdef CONFIG_PREEMPT_RCU
565 extern struct rcu_state rcu_preempt_state;
566 #endif /* #ifdef CONFIG_PREEMPT_RCU */
568 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
569 DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu);
570 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
571 DECLARE_PER_CPU(char, rcu_cpu_has_work);
573 #ifndef RCU_TREE_NONCORE
575 /* Forward declarations for rcutree_plugin.h */
576 static void rcu_bootup_announce(void);
577 static void rcu_preempt_note_context_switch(void);
578 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
579 #ifdef CONFIG_HOTPLUG_CPU
580 static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
581 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
582 static void rcu_print_detail_task_stall(struct rcu_state *rsp);
583 static int rcu_print_task_stall(struct rcu_node *rnp);
584 static int rcu_print_task_exp_stall(struct rcu_node *rnp);
585 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
586 static void rcu_preempt_check_callbacks(void);
587 void call_rcu(struct rcu_head *head, rcu_callback_t func);
588 static void __init __rcu_init_preempt(void);
589 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
590 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
591 static bool rcu_is_callbacks_kthread(void);
592 static void rcu_cpu_kthread_setup(unsigned int cpu);
593 #ifdef CONFIG_RCU_BOOST
594 static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
595 struct rcu_node *rnp);
596 #endif /* #ifdef CONFIG_RCU_BOOST */
597 static void __init rcu_spawn_boost_kthreads(void);
598 static void rcu_prepare_kthreads(int cpu);
599 static void rcu_cleanup_after_idle(void);
600 static void rcu_prepare_for_idle(void);
601 static void rcu_idle_count_callbacks_posted(void);
602 static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
603 static void print_cpu_stall_info_begin(void);
604 static void print_cpu_stall_info(struct rcu_state *rsp, int cpu);
605 static void print_cpu_stall_info_end(void);
606 static void zero_cpu_stall_ticks(struct rcu_data *rdp);
607 static void increment_cpu_stall_ticks(void);
608 static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu);
609 static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq);
610 static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp);
611 static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq);
612 static void rcu_init_one_nocb(struct rcu_node *rnp);
613 static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
614 bool lazy, unsigned long flags);
615 static bool rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp,
616 struct rcu_data *rdp,
617 unsigned long flags);
618 static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp);
619 static void do_nocb_deferred_wakeup(struct rcu_data *rdp);
620 static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp);
621 static void rcu_spawn_all_nocb_kthreads(int cpu);
622 static void __init rcu_spawn_nocb_kthreads(void);
623 #ifdef CONFIG_RCU_NOCB_CPU
624 static void __init rcu_organize_nocb_kthreads(struct rcu_state *rsp);
625 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
626 static void __maybe_unused rcu_kick_nohz_cpu(int cpu);
627 static bool init_nocb_callback_list(struct rcu_data *rdp);
628 static void rcu_sysidle_enter(int irq);
629 static void rcu_sysidle_exit(int irq);
630 static void rcu_sysidle_check_cpu(struct rcu_data *rdp, bool *isidle,
631 unsigned long *maxj);
632 static bool is_sysidle_rcu_state(struct rcu_state *rsp);
633 static void rcu_sysidle_report_gp(struct rcu_state *rsp, int isidle,
635 static void rcu_bind_gp_kthread(void);
636 static void rcu_sysidle_init_percpu_data(struct rcu_dynticks *rdtp);
637 static bool rcu_nohz_full_cpu(struct rcu_state *rsp);
638 static void rcu_dynticks_task_enter(void);
639 static void rcu_dynticks_task_exit(void);
641 #endif /* #ifndef RCU_TREE_NONCORE */
643 #ifdef CONFIG_RCU_TRACE
644 /* Read out queue lengths for tracing. */
645 static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll)
647 #ifdef CONFIG_RCU_NOCB_CPU
648 *ql = atomic_long_read(&rdp->nocb_q_count);
649 *qll = atomic_long_read(&rdp->nocb_q_count_lazy);
650 #else /* #ifdef CONFIG_RCU_NOCB_CPU */
653 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
655 #endif /* #ifdef CONFIG_RCU_TRACE */
658 * Place this after a lock-acquisition primitive to guarantee that
659 * an UNLOCK+LOCK pair act as a full barrier. This guarantee applies
660 * if the UNLOCK and LOCK are executed by the same CPU or if the
661 * UNLOCK and LOCK operate on the same lock variable.
664 #define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */
665 #else /* #ifdef CONFIG_PPC */
666 #define smp_mb__after_unlock_lock() do { } while (0)
667 #endif /* #else #ifdef CONFIG_PPC */