xref: /linux/kernel/rcu/tree.h (revision 3932b9ca55b0be314a36d3e84faff3e823c081f5)
1 /*
2  * Read-Copy Update mechanism for mutual exclusion (tree-based version)
3  * Internal non-public definitions.
4  *
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.
9  *
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.
14  *
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.
18  *
19  * Copyright IBM Corporation, 2008
20  *
21  * Author: Ingo Molnar <mingo@elte.hu>
22  *	   Paul E. McKenney <paulmck@linux.vnet.ibm.com>
23  */
24 
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/irq_work.h>
31 
32 /*
33  * Define shape of hierarchy based on NR_CPUS, CONFIG_RCU_FANOUT, and
34  * CONFIG_RCU_FANOUT_LEAF.
35  * In theory, it should be possible to add more levels straightforwardly.
36  * In practice, this did work well going from three levels to four.
37  * Of course, your mileage may vary.
38  */
39 #define MAX_RCU_LVLS 4
40 #define RCU_FANOUT_1	      (CONFIG_RCU_FANOUT_LEAF)
41 #define RCU_FANOUT_2	      (RCU_FANOUT_1 * CONFIG_RCU_FANOUT)
42 #define RCU_FANOUT_3	      (RCU_FANOUT_2 * CONFIG_RCU_FANOUT)
43 #define RCU_FANOUT_4	      (RCU_FANOUT_3 * CONFIG_RCU_FANOUT)
44 
45 #if NR_CPUS <= RCU_FANOUT_1
46 #  define RCU_NUM_LVLS	      1
47 #  define NUM_RCU_LVL_0	      1
48 #  define NUM_RCU_LVL_1	      (NR_CPUS)
49 #  define NUM_RCU_LVL_2	      0
50 #  define NUM_RCU_LVL_3	      0
51 #  define NUM_RCU_LVL_4	      0
52 #elif NR_CPUS <= RCU_FANOUT_2
53 #  define RCU_NUM_LVLS	      2
54 #  define NUM_RCU_LVL_0	      1
55 #  define NUM_RCU_LVL_1	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
56 #  define NUM_RCU_LVL_2	      (NR_CPUS)
57 #  define NUM_RCU_LVL_3	      0
58 #  define NUM_RCU_LVL_4	      0
59 #elif NR_CPUS <= RCU_FANOUT_3
60 #  define RCU_NUM_LVLS	      3
61 #  define NUM_RCU_LVL_0	      1
62 #  define NUM_RCU_LVL_1	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
63 #  define NUM_RCU_LVL_2	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
64 #  define NUM_RCU_LVL_3	      (NR_CPUS)
65 #  define NUM_RCU_LVL_4	      0
66 #elif NR_CPUS <= RCU_FANOUT_4
67 #  define RCU_NUM_LVLS	      4
68 #  define NUM_RCU_LVL_0	      1
69 #  define NUM_RCU_LVL_1	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3)
70 #  define NUM_RCU_LVL_2	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
71 #  define NUM_RCU_LVL_3	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
72 #  define NUM_RCU_LVL_4	      (NR_CPUS)
73 #else
74 # error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
75 #endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
76 
77 #define RCU_SUM (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3 + NUM_RCU_LVL_4)
78 #define NUM_RCU_NODES (RCU_SUM - NR_CPUS)
79 
80 extern int rcu_num_lvls;
81 extern int rcu_num_nodes;
82 
83 /*
84  * Dynticks per-CPU state.
85  */
86 struct rcu_dynticks {
87 	long long dynticks_nesting; /* Track irq/process nesting level. */
88 				    /* Process level is worth LLONG_MAX/2. */
89 	int dynticks_nmi_nesting;   /* Track NMI nesting level. */
90 	atomic_t dynticks;	    /* Even value for idle, else odd. */
91 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
92 	long long dynticks_idle_nesting;
93 				    /* irq/process nesting level from idle. */
94 	atomic_t dynticks_idle;	    /* Even value for idle, else odd. */
95 				    /*  "Idle" excludes userspace execution. */
96 	unsigned long dynticks_idle_jiffies;
97 				    /* End of last non-NMI non-idle period. */
98 #endif /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
99 #ifdef CONFIG_RCU_FAST_NO_HZ
100 	bool all_lazy;		    /* Are all CPU's CBs lazy? */
101 	unsigned long nonlazy_posted;
102 				    /* # times non-lazy CBs posted to CPU. */
103 	unsigned long nonlazy_posted_snap;
104 				    /* idle-period nonlazy_posted snapshot. */
105 	unsigned long last_accelerate;
106 				    /* Last jiffy CBs were accelerated. */
107 	unsigned long last_advance_all;
108 				    /* Last jiffy CBs were all advanced. */
109 	int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */
110 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
111 };
112 
113 /* RCU's kthread states for tracing. */
114 #define RCU_KTHREAD_STOPPED  0
115 #define RCU_KTHREAD_RUNNING  1
116 #define RCU_KTHREAD_WAITING  2
117 #define RCU_KTHREAD_OFFCPU   3
118 #define RCU_KTHREAD_YIELDING 4
119 #define RCU_KTHREAD_MAX      4
120 
121 /*
122  * Definition for node within the RCU grace-period-detection hierarchy.
123  */
124 struct rcu_node {
125 	raw_spinlock_t lock;	/* Root rcu_node's lock protects some */
126 				/*  rcu_state fields as well as following. */
127 	unsigned long gpnum;	/* Current grace period for this node. */
128 				/*  This will either be equal to or one */
129 				/*  behind the root rcu_node's gpnum. */
130 	unsigned long completed; /* Last GP completed for this node. */
131 				/*  This will either be equal to or one */
132 				/*  behind the root rcu_node's gpnum. */
133 	unsigned long qsmask;	/* CPUs or groups that need to switch in */
134 				/*  order for current grace period to proceed.*/
135 				/*  In leaf rcu_node, each bit corresponds to */
136 				/*  an rcu_data structure, otherwise, each */
137 				/*  bit corresponds to a child rcu_node */
138 				/*  structure. */
139 	unsigned long expmask;	/* Groups that have ->blkd_tasks */
140 				/*  elements that need to drain to allow the */
141 				/*  current expedited grace period to */
142 				/*  complete (only for TREE_PREEMPT_RCU). */
143 	unsigned long qsmaskinit;
144 				/* Per-GP initial value for qsmask & expmask. */
145 	unsigned long grpmask;	/* Mask to apply to parent qsmask. */
146 				/*  Only one bit will be set in this mask. */
147 	int	grplo;		/* lowest-numbered CPU or group here. */
148 	int	grphi;		/* highest-numbered CPU or group here. */
149 	u8	grpnum;		/* CPU/group number for next level up. */
150 	u8	level;		/* root is at level 0. */
151 	struct rcu_node *parent;
152 	struct list_head blkd_tasks;
153 				/* Tasks blocked in RCU read-side critical */
154 				/*  section.  Tasks are placed at the head */
155 				/*  of this list and age towards the tail. */
156 	struct list_head *gp_tasks;
157 				/* Pointer to the first task blocking the */
158 				/*  current grace period, or NULL if there */
159 				/*  is no such task. */
160 	struct list_head *exp_tasks;
161 				/* Pointer to the first task blocking the */
162 				/*  current expedited grace period, or NULL */
163 				/*  if there is no such task.  If there */
164 				/*  is no current expedited grace period, */
165 				/*  then there can cannot be any such task. */
166 #ifdef CONFIG_RCU_BOOST
167 	struct list_head *boost_tasks;
168 				/* Pointer to first task that needs to be */
169 				/*  priority boosted, or NULL if no priority */
170 				/*  boosting is needed for this rcu_node */
171 				/*  structure.  If there are no tasks */
172 				/*  queued on this rcu_node structure that */
173 				/*  are blocking the current grace period, */
174 				/*  there can be no such task. */
175 	struct completion boost_completion;
176 				/* Used to ensure that the rt_mutex used */
177 				/*  to carry out the boosting is fully */
178 				/*  released with no future boostee accesses */
179 				/*  before that rt_mutex is re-initialized. */
180 	struct rt_mutex boost_mtx;
181 				/* Used only for the priority-boosting */
182 				/*  side effect, not as a lock. */
183 	unsigned long boost_time;
184 				/* When to start boosting (jiffies). */
185 	struct task_struct *boost_kthread_task;
186 				/* kthread that takes care of priority */
187 				/*  boosting for this rcu_node structure. */
188 	unsigned int boost_kthread_status;
189 				/* State of boost_kthread_task for tracing. */
190 	unsigned long n_tasks_boosted;
191 				/* Total number of tasks boosted. */
192 	unsigned long n_exp_boosts;
193 				/* Number of tasks boosted for expedited GP. */
194 	unsigned long n_normal_boosts;
195 				/* Number of tasks boosted for normal GP. */
196 	unsigned long n_balk_blkd_tasks;
197 				/* Refused to boost: no blocked tasks. */
198 	unsigned long n_balk_exp_gp_tasks;
199 				/* Refused to boost: nothing blocking GP. */
200 	unsigned long n_balk_boost_tasks;
201 				/* Refused to boost: already boosting. */
202 	unsigned long n_balk_notblocked;
203 				/* Refused to boost: RCU RS CS still running. */
204 	unsigned long n_balk_notyet;
205 				/* Refused to boost: not yet time. */
206 	unsigned long n_balk_nos;
207 				/* Refused to boost: not sure why, though. */
208 				/*  This can happen due to race conditions. */
209 #endif /* #ifdef CONFIG_RCU_BOOST */
210 #ifdef CONFIG_RCU_NOCB_CPU
211 	wait_queue_head_t nocb_gp_wq[2];
212 				/* Place for rcu_nocb_kthread() to wait GP. */
213 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
214 	int need_future_gp[2];
215 				/* Counts of upcoming no-CB GP requests. */
216 	raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp;
217 } ____cacheline_internodealigned_in_smp;
218 
219 /*
220  * Do a full breadth-first scan of the rcu_node structures for the
221  * specified rcu_state structure.
222  */
223 #define rcu_for_each_node_breadth_first(rsp, rnp) \
224 	for ((rnp) = &(rsp)->node[0]; \
225 	     (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
226 
227 /*
228  * Do a breadth-first scan of the non-leaf rcu_node structures for the
229  * specified rcu_state structure.  Note that if there is a singleton
230  * rcu_node tree with but one rcu_node structure, this loop is a no-op.
231  */
232 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
233 	for ((rnp) = &(rsp)->node[0]; \
234 	     (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
235 
236 /*
237  * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
238  * structure.  Note that if there is a singleton rcu_node tree with but
239  * one rcu_node structure, this loop -will- visit the rcu_node structure.
240  * It is still a leaf node, even if it is also the root node.
241  */
242 #define rcu_for_each_leaf_node(rsp, rnp) \
243 	for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
244 	     (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
245 
246 /* Index values for nxttail array in struct rcu_data. */
247 #define RCU_DONE_TAIL		0	/* Also RCU_WAIT head. */
248 #define RCU_WAIT_TAIL		1	/* Also RCU_NEXT_READY head. */
249 #define RCU_NEXT_READY_TAIL	2	/* Also RCU_NEXT head. */
250 #define RCU_NEXT_TAIL		3
251 #define RCU_NEXT_SIZE		4
252 
253 /* Per-CPU data for read-copy update. */
254 struct rcu_data {
255 	/* 1) quiescent-state and grace-period handling : */
256 	unsigned long	completed;	/* Track rsp->completed gp number */
257 					/*  in order to detect GP end. */
258 	unsigned long	gpnum;		/* Highest gp number that this CPU */
259 					/*  is aware of having started. */
260 	bool		passed_quiesce;	/* User-mode/idle loop etc. */
261 	bool		qs_pending;	/* Core waits for quiesc state. */
262 	bool		beenonline;	/* CPU online at least once. */
263 	struct rcu_node *mynode;	/* This CPU's leaf of hierarchy */
264 	unsigned long grpmask;		/* Mask to apply to leaf qsmask. */
265 #ifdef CONFIG_RCU_CPU_STALL_INFO
266 	unsigned long	ticks_this_gp;	/* The number of scheduling-clock */
267 					/*  ticks this CPU has handled */
268 					/*  during and after the last grace */
269 					/* period it is aware of. */
270 #endif /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
271 
272 	/* 2) batch handling */
273 	/*
274 	 * If nxtlist is not NULL, it is partitioned as follows.
275 	 * Any of the partitions might be empty, in which case the
276 	 * pointer to that partition will be equal to the pointer for
277 	 * the following partition.  When the list is empty, all of
278 	 * the nxttail elements point to the ->nxtlist pointer itself,
279 	 * which in that case is NULL.
280 	 *
281 	 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
282 	 *	Entries that batch # <= ->completed
283 	 *	The grace period for these entries has completed, and
284 	 *	the other grace-period-completed entries may be moved
285 	 *	here temporarily in rcu_process_callbacks().
286 	 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
287 	 *	Entries that batch # <= ->completed - 1: waiting for current GP
288 	 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
289 	 *	Entries known to have arrived before current GP ended
290 	 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
291 	 *	Entries that might have arrived after current GP ended
292 	 *	Note that the value of *nxttail[RCU_NEXT_TAIL] will
293 	 *	always be NULL, as this is the end of the list.
294 	 */
295 	struct rcu_head *nxtlist;
296 	struct rcu_head **nxttail[RCU_NEXT_SIZE];
297 	unsigned long	nxtcompleted[RCU_NEXT_SIZE];
298 					/* grace periods for sublists. */
299 	long		qlen_lazy;	/* # of lazy queued callbacks */
300 	long		qlen;		/* # of queued callbacks, incl lazy */
301 	long		qlen_last_fqs_check;
302 					/* qlen at last check for QS forcing */
303 	unsigned long	n_cbs_invoked;	/* count of RCU cbs invoked. */
304 	unsigned long	n_nocbs_invoked; /* count of no-CBs RCU cbs invoked. */
305 	unsigned long   n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
306 	unsigned long   n_cbs_adopted;  /* RCU cbs adopted from dying CPU */
307 	unsigned long	n_force_qs_snap;
308 					/* did other CPU force QS recently? */
309 	long		blimit;		/* Upper limit on a processed batch */
310 
311 	/* 3) dynticks interface. */
312 	struct rcu_dynticks *dynticks;	/* Shared per-CPU dynticks state. */
313 	int dynticks_snap;		/* Per-GP tracking for dynticks. */
314 
315 	/* 4) reasons this CPU needed to be kicked by force_quiescent_state */
316 	unsigned long dynticks_fqs;	/* Kicked due to dynticks idle. */
317 	unsigned long offline_fqs;	/* Kicked due to being offline. */
318 	unsigned long cond_resched_completed;
319 					/* Grace period that needs help */
320 					/*  from cond_resched(). */
321 
322 	/* 5) __rcu_pending() statistics. */
323 	unsigned long n_rcu_pending;	/* rcu_pending() calls since boot. */
324 	unsigned long n_rp_qs_pending;
325 	unsigned long n_rp_report_qs;
326 	unsigned long n_rp_cb_ready;
327 	unsigned long n_rp_cpu_needs_gp;
328 	unsigned long n_rp_gp_completed;
329 	unsigned long n_rp_gp_started;
330 	unsigned long n_rp_nocb_defer_wakeup;
331 	unsigned long n_rp_need_nothing;
332 
333 	/* 6) _rcu_barrier() and OOM callbacks. */
334 	struct rcu_head barrier_head;
335 #ifdef CONFIG_RCU_FAST_NO_HZ
336 	struct rcu_head oom_head;
337 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
338 
339 	/* 7) Callback offloading. */
340 #ifdef CONFIG_RCU_NOCB_CPU
341 	struct rcu_head *nocb_head;	/* CBs waiting for kthread. */
342 	struct rcu_head **nocb_tail;
343 	atomic_long_t nocb_q_count;	/* # CBs waiting for kthread */
344 	atomic_long_t nocb_q_count_lazy; /*  (approximate). */
345 	struct rcu_head *nocb_follower_head; /* CBs ready to invoke. */
346 	struct rcu_head **nocb_follower_tail;
347 	atomic_long_t nocb_follower_count; /* # CBs ready to invoke. */
348 	atomic_long_t nocb_follower_count_lazy; /*  (approximate). */
349 	int nocb_p_count;		/* # CBs being invoked by kthread */
350 	int nocb_p_count_lazy;		/*  (approximate). */
351 	wait_queue_head_t nocb_wq;	/* For nocb kthreads to sleep on. */
352 	struct task_struct *nocb_kthread;
353 	bool nocb_defer_wakeup;		/* Defer wakeup of nocb_kthread. */
354 
355 	/* The following fields are used by the leader, hence own cacheline. */
356 	struct rcu_head *nocb_gp_head ____cacheline_internodealigned_in_smp;
357 					/* CBs waiting for GP. */
358 	struct rcu_head **nocb_gp_tail;
359 	long nocb_gp_count;
360 	long nocb_gp_count_lazy;
361 	bool nocb_leader_sleep;		/* Is the nocb leader thread asleep? */
362 	struct rcu_data *nocb_next_follower;
363 					/* Next follower in wakeup chain. */
364 
365 	/* The following fields are used by the follower, hence new cachline. */
366 	struct rcu_data *nocb_leader ____cacheline_internodealigned_in_smp;
367 					/* Leader CPU takes GP-end wakeups. */
368 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
369 
370 	/* 8) RCU CPU stall data. */
371 #ifdef CONFIG_RCU_CPU_STALL_INFO
372 	unsigned int softirq_snap;	/* Snapshot of softirq activity. */
373 #endif /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
374 
375 	int cpu;
376 	struct rcu_state *rsp;
377 };
378 
379 /* Values for fqs_state field in struct rcu_state. */
380 #define RCU_GP_IDLE		0	/* No grace period in progress. */
381 #define RCU_GP_INIT		1	/* Grace period being initialized. */
382 #define RCU_SAVE_DYNTICK	2	/* Need to scan dyntick state. */
383 #define RCU_FORCE_QS		3	/* Need to force quiescent state. */
384 #define RCU_SIGNAL_INIT		RCU_SAVE_DYNTICK
385 
386 #define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500))
387 					/* For jiffies_till_first_fqs and */
388 					/*  and jiffies_till_next_fqs. */
389 
390 #define RCU_JIFFIES_FQS_DIV	256	/* Very large systems need more */
391 					/*  delay between bouts of */
392 					/*  quiescent-state forcing. */
393 
394 #define RCU_STALL_RAT_DELAY	2	/* Allow other CPUs time to take */
395 					/*  at least one scheduling clock */
396 					/*  irq before ratting on them. */
397 
398 #define rcu_wait(cond)							\
399 do {									\
400 	for (;;) {							\
401 		set_current_state(TASK_INTERRUPTIBLE);			\
402 		if (cond)						\
403 			break;						\
404 		schedule();						\
405 	}								\
406 	__set_current_state(TASK_RUNNING);				\
407 } while (0)
408 
409 /*
410  * RCU global state, including node hierarchy.  This hierarchy is
411  * represented in "heap" form in a dense array.  The root (first level)
412  * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
413  * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
414  * and the third level in ->node[m+1] and following (->node[m+1] referenced
415  * by ->level[2]).  The number of levels is determined by the number of
416  * CPUs and by CONFIG_RCU_FANOUT.  Small systems will have a "hierarchy"
417  * consisting of a single rcu_node.
418  */
419 struct rcu_state {
420 	struct rcu_node node[NUM_RCU_NODES];	/* Hierarchy. */
421 	struct rcu_node *level[RCU_NUM_LVLS];	/* Hierarchy levels. */
422 	u32 levelcnt[MAX_RCU_LVLS + 1];		/* # nodes in each level. */
423 	u8 levelspread[RCU_NUM_LVLS];		/* kids/node in each level. */
424 	u8 flavor_mask;				/* bit in flavor mask. */
425 	struct rcu_data __percpu *rda;		/* pointer of percu rcu_data. */
426 	void (*call)(struct rcu_head *head,	/* call_rcu() flavor. */
427 		     void (*func)(struct rcu_head *head));
428 
429 	/* The following fields are guarded by the root rcu_node's lock. */
430 
431 	u8	fqs_state ____cacheline_internodealigned_in_smp;
432 						/* Force QS state. */
433 	u8	boost;				/* Subject to priority boost. */
434 	unsigned long gpnum;			/* Current gp number. */
435 	unsigned long completed;		/* # of last completed gp. */
436 	struct task_struct *gp_kthread;		/* Task for grace periods. */
437 	wait_queue_head_t gp_wq;		/* Where GP task waits. */
438 	short gp_flags;				/* Commands for GP task. */
439 	short gp_state;				/* GP kthread sleep state. */
440 
441 	/* End of fields guarded by root rcu_node's lock. */
442 
443 	raw_spinlock_t orphan_lock ____cacheline_internodealigned_in_smp;
444 						/* Protect following fields. */
445 	struct rcu_head *orphan_nxtlist;	/* Orphaned callbacks that */
446 						/*  need a grace period. */
447 	struct rcu_head **orphan_nxttail;	/* Tail of above. */
448 	struct rcu_head *orphan_donelist;	/* Orphaned callbacks that */
449 						/*  are ready to invoke. */
450 	struct rcu_head **orphan_donetail;	/* Tail of above. */
451 	long qlen_lazy;				/* Number of lazy callbacks. */
452 	long qlen;				/* Total number of callbacks. */
453 	/* End of fields guarded by orphan_lock. */
454 
455 	struct mutex onoff_mutex;		/* Coordinate hotplug & GPs. */
456 
457 	struct mutex barrier_mutex;		/* Guards barrier fields. */
458 	atomic_t barrier_cpu_count;		/* # CPUs waiting on. */
459 	struct completion barrier_completion;	/* Wake at barrier end. */
460 	unsigned long n_barrier_done;		/* ++ at start and end of */
461 						/*  _rcu_barrier(). */
462 	/* End of fields guarded by barrier_mutex. */
463 
464 	atomic_long_t expedited_start;		/* Starting ticket. */
465 	atomic_long_t expedited_done;		/* Done ticket. */
466 	atomic_long_t expedited_wrap;		/* # near-wrap incidents. */
467 	atomic_long_t expedited_tryfail;	/* # acquisition failures. */
468 	atomic_long_t expedited_workdone1;	/* # done by others #1. */
469 	atomic_long_t expedited_workdone2;	/* # done by others #2. */
470 	atomic_long_t expedited_normal;		/* # fallbacks to normal. */
471 	atomic_long_t expedited_stoppedcpus;	/* # successful stop_cpus. */
472 	atomic_long_t expedited_done_tries;	/* # tries to update _done. */
473 	atomic_long_t expedited_done_lost;	/* # times beaten to _done. */
474 	atomic_long_t expedited_done_exit;	/* # times exited _done loop. */
475 
476 	unsigned long jiffies_force_qs;		/* Time at which to invoke */
477 						/*  force_quiescent_state(). */
478 	unsigned long n_force_qs;		/* Number of calls to */
479 						/*  force_quiescent_state(). */
480 	unsigned long n_force_qs_lh;		/* ~Number of calls leaving */
481 						/*  due to lock unavailable. */
482 	unsigned long n_force_qs_ngp;		/* Number of calls leaving */
483 						/*  due to no GP active. */
484 	unsigned long gp_start;			/* Time at which GP started, */
485 						/*  but in jiffies. */
486 	unsigned long jiffies_stall;		/* Time at which to check */
487 						/*  for CPU stalls. */
488 	unsigned long jiffies_resched;		/* Time at which to resched */
489 						/*  a reluctant CPU. */
490 	unsigned long gp_max;			/* Maximum GP duration in */
491 						/*  jiffies. */
492 	const char *name;			/* Name of structure. */
493 	char abbr;				/* Abbreviated name. */
494 	struct list_head flavors;		/* List of RCU flavors. */
495 };
496 
497 /* Values for rcu_state structure's gp_flags field. */
498 #define RCU_GP_FLAG_INIT 0x1	/* Need grace-period initialization. */
499 #define RCU_GP_FLAG_FQS  0x2	/* Need grace-period quiescent-state forcing. */
500 
501 /* Values for rcu_state structure's gp_flags field. */
502 #define RCU_GP_WAIT_INIT 0	/* Initial state. */
503 #define RCU_GP_WAIT_GPS  1	/* Wait for grace-period start. */
504 #define RCU_GP_WAIT_FQS  2	/* Wait for force-quiescent-state time. */
505 
506 extern struct list_head rcu_struct_flavors;
507 
508 /* Sequence through rcu_state structures for each RCU flavor. */
509 #define for_each_rcu_flavor(rsp) \
510 	list_for_each_entry((rsp), &rcu_struct_flavors, flavors)
511 
512 /* Return values for rcu_preempt_offline_tasks(). */
513 
514 #define RCU_OFL_TASKS_NORM_GP	0x1		/* Tasks blocking normal */
515 						/*  GP were moved to root. */
516 #define RCU_OFL_TASKS_EXP_GP	0x2		/* Tasks blocking expedited */
517 						/*  GP were moved to root. */
518 
519 /*
520  * RCU implementation internal declarations:
521  */
522 extern struct rcu_state rcu_sched_state;
523 DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
524 
525 extern struct rcu_state rcu_bh_state;
526 DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
527 
528 #ifdef CONFIG_TREE_PREEMPT_RCU
529 extern struct rcu_state rcu_preempt_state;
530 DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
531 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
532 
533 #ifdef CONFIG_RCU_BOOST
534 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
535 DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu);
536 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
537 DECLARE_PER_CPU(char, rcu_cpu_has_work);
538 #endif /* #ifdef CONFIG_RCU_BOOST */
539 
540 #ifndef RCU_TREE_NONCORE
541 
542 /* Forward declarations for rcutree_plugin.h */
543 static void rcu_bootup_announce(void);
544 long rcu_batches_completed(void);
545 static void rcu_preempt_note_context_switch(int cpu);
546 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
547 #ifdef CONFIG_HOTPLUG_CPU
548 static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp,
549 				      unsigned long flags);
550 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
551 static void rcu_print_detail_task_stall(struct rcu_state *rsp);
552 static int rcu_print_task_stall(struct rcu_node *rnp);
553 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
554 #ifdef CONFIG_HOTPLUG_CPU
555 static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
556 				     struct rcu_node *rnp,
557 				     struct rcu_data *rdp);
558 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
559 static void rcu_preempt_check_callbacks(int cpu);
560 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
561 #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU)
562 static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
563 			       bool wake);
564 #endif /* #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) */
565 static void __init __rcu_init_preempt(void);
566 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
567 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
568 static void invoke_rcu_callbacks_kthread(void);
569 static bool rcu_is_callbacks_kthread(void);
570 #ifdef CONFIG_RCU_BOOST
571 static void rcu_preempt_do_callbacks(void);
572 static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
573 						 struct rcu_node *rnp);
574 #endif /* #ifdef CONFIG_RCU_BOOST */
575 static void rcu_prepare_kthreads(int cpu);
576 static void rcu_cleanup_after_idle(int cpu);
577 static void rcu_prepare_for_idle(int cpu);
578 static void rcu_idle_count_callbacks_posted(void);
579 static void print_cpu_stall_info_begin(void);
580 static void print_cpu_stall_info(struct rcu_state *rsp, int cpu);
581 static void print_cpu_stall_info_end(void);
582 static void zero_cpu_stall_ticks(struct rcu_data *rdp);
583 static void increment_cpu_stall_ticks(void);
584 static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq);
585 static void rcu_nocb_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp);
586 static void rcu_init_one_nocb(struct rcu_node *rnp);
587 static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
588 			    bool lazy, unsigned long flags);
589 static bool rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp,
590 				      struct rcu_data *rdp,
591 				      unsigned long flags);
592 static bool rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp);
593 static void do_nocb_deferred_wakeup(struct rcu_data *rdp);
594 static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp);
595 static void rcu_spawn_nocb_kthreads(struct rcu_state *rsp);
596 static void __maybe_unused rcu_kick_nohz_cpu(int cpu);
597 static bool init_nocb_callback_list(struct rcu_data *rdp);
598 static void rcu_sysidle_enter(struct rcu_dynticks *rdtp, int irq);
599 static void rcu_sysidle_exit(struct rcu_dynticks *rdtp, int irq);
600 static void rcu_sysidle_check_cpu(struct rcu_data *rdp, bool *isidle,
601 				  unsigned long *maxj);
602 static bool is_sysidle_rcu_state(struct rcu_state *rsp);
603 static void rcu_sysidle_report_gp(struct rcu_state *rsp, int isidle,
604 				  unsigned long maxj);
605 static void rcu_bind_gp_kthread(void);
606 static void rcu_sysidle_init_percpu_data(struct rcu_dynticks *rdtp);
607 static bool rcu_nohz_full_cpu(struct rcu_state *rsp);
608 
609 #endif /* #ifndef RCU_TREE_NONCORE */
610 
611 #ifdef CONFIG_RCU_TRACE
612 #ifdef CONFIG_RCU_NOCB_CPU
613 /* Sum up queue lengths for tracing. */
614 static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll)
615 {
616 	*ql = atomic_long_read(&rdp->nocb_q_count) +
617 	      rdp->nocb_p_count +
618 	      atomic_long_read(&rdp->nocb_follower_count) +
619 	      rdp->nocb_p_count + rdp->nocb_gp_count;
620 	*qll = atomic_long_read(&rdp->nocb_q_count_lazy) +
621 	       rdp->nocb_p_count_lazy +
622 	       atomic_long_read(&rdp->nocb_follower_count_lazy) +
623 	       rdp->nocb_p_count_lazy + rdp->nocb_gp_count_lazy;
624 }
625 #else /* #ifdef CONFIG_RCU_NOCB_CPU */
626 static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll)
627 {
628 	*ql = 0;
629 	*qll = 0;
630 }
631 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
632 #endif /* #ifdef CONFIG_RCU_TRACE */
633