xref: /openbmc/linux/kernel/rcu/tree.h (revision d7a3d85e)
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 
31 /*
32  * Define shape of hierarchy based on NR_CPUS, CONFIG_RCU_FANOUT, and
33  * CONFIG_RCU_FANOUT_LEAF.
34  * In theory, it should be possible to add more levels straightforwardly.
35  * In practice, this did work well going from three levels to four.
36  * Of course, your mileage may vary.
37  */
38 #define MAX_RCU_LVLS 4
39 #define RCU_FANOUT_1	      (CONFIG_RCU_FANOUT_LEAF)
40 #define RCU_FANOUT_2	      (RCU_FANOUT_1 * CONFIG_RCU_FANOUT)
41 #define RCU_FANOUT_3	      (RCU_FANOUT_2 * CONFIG_RCU_FANOUT)
42 #define RCU_FANOUT_4	      (RCU_FANOUT_3 * CONFIG_RCU_FANOUT)
43 
44 #if NR_CPUS <= RCU_FANOUT_1
45 #  define RCU_NUM_LVLS	      1
46 #  define NUM_RCU_LVL_0	      1
47 #  define NUM_RCU_LVL_1	      (NR_CPUS)
48 #  define NUM_RCU_LVL_2	      0
49 #  define NUM_RCU_LVL_3	      0
50 #  define NUM_RCU_LVL_4	      0
51 #elif NR_CPUS <= RCU_FANOUT_2
52 #  define RCU_NUM_LVLS	      2
53 #  define NUM_RCU_LVL_0	      1
54 #  define NUM_RCU_LVL_1	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
55 #  define NUM_RCU_LVL_2	      (NR_CPUS)
56 #  define NUM_RCU_LVL_3	      0
57 #  define NUM_RCU_LVL_4	      0
58 #elif NR_CPUS <= RCU_FANOUT_3
59 #  define RCU_NUM_LVLS	      3
60 #  define NUM_RCU_LVL_0	      1
61 #  define NUM_RCU_LVL_1	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
62 #  define NUM_RCU_LVL_2	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
63 #  define NUM_RCU_LVL_3	      (NR_CPUS)
64 #  define NUM_RCU_LVL_4	      0
65 #elif NR_CPUS <= RCU_FANOUT_4
66 #  define RCU_NUM_LVLS	      4
67 #  define NUM_RCU_LVL_0	      1
68 #  define NUM_RCU_LVL_1	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3)
69 #  define NUM_RCU_LVL_2	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
70 #  define NUM_RCU_LVL_3	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
71 #  define NUM_RCU_LVL_4	      (NR_CPUS)
72 #else
73 # error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
74 #endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
75 
76 #define RCU_SUM (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3 + NUM_RCU_LVL_4)
77 #define NUM_RCU_NODES (RCU_SUM - NR_CPUS)
78 
79 extern int rcu_num_lvls;
80 extern int rcu_num_nodes;
81 
82 /*
83  * Dynticks per-CPU state.
84  */
85 struct rcu_dynticks {
86 	long long dynticks_nesting; /* Track irq/process nesting level. */
87 				    /* Process level is worth LLONG_MAX/2. */
88 	int dynticks_nmi_nesting;   /* Track NMI nesting level. */
89 	atomic_t dynticks;	    /* Even value for idle, else odd. */
90 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
91 	long long dynticks_idle_nesting;
92 				    /* irq/process nesting level from idle. */
93 	atomic_t dynticks_idle;	    /* Even value for idle, else odd. */
94 				    /*  "Idle" excludes userspace execution. */
95 	unsigned long dynticks_idle_jiffies;
96 				    /* End of last non-NMI non-idle period. */
97 #endif /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
98 #ifdef CONFIG_RCU_FAST_NO_HZ
99 	bool all_lazy;		    /* Are all CPU's CBs lazy? */
100 	unsigned long nonlazy_posted;
101 				    /* # times non-lazy CBs posted to CPU. */
102 	unsigned long nonlazy_posted_snap;
103 				    /* idle-period nonlazy_posted snapshot. */
104 	unsigned long last_accelerate;
105 				    /* Last jiffy CBs were accelerated. */
106 	unsigned long last_advance_all;
107 				    /* Last jiffy CBs were all advanced. */
108 	int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */
109 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
110 };
111 
112 /* RCU's kthread states for tracing. */
113 #define RCU_KTHREAD_STOPPED  0
114 #define RCU_KTHREAD_RUNNING  1
115 #define RCU_KTHREAD_WAITING  2
116 #define RCU_KTHREAD_OFFCPU   3
117 #define RCU_KTHREAD_YIELDING 4
118 #define RCU_KTHREAD_MAX      4
119 
120 /*
121  * Definition for node within the RCU grace-period-detection hierarchy.
122  */
123 struct rcu_node {
124 	raw_spinlock_t lock;	/* Root rcu_node's lock protects some */
125 				/*  rcu_state fields as well as following. */
126 	unsigned long gpnum;	/* Current grace period for this node. */
127 				/*  This will either be equal to or one */
128 				/*  behind the root rcu_node's gpnum. */
129 	unsigned long completed; /* Last GP completed for this node. */
130 				/*  This will either be equal to or one */
131 				/*  behind the root rcu_node's gpnum. */
132 	unsigned long qsmask;	/* CPUs or groups that need to switch in */
133 				/*  order for current grace period to proceed.*/
134 				/*  In leaf rcu_node, each bit corresponds to */
135 				/*  an rcu_data structure, otherwise, each */
136 				/*  bit corresponds to a child rcu_node */
137 				/*  structure. */
138 	unsigned long expmask;	/* Groups that have ->blkd_tasks */
139 				/*  elements that need to drain to allow the */
140 				/*  current expedited grace period to */
141 				/*  complete (only for PREEMPT_RCU). */
142 	unsigned long qsmaskinit;
143 				/* Per-GP initial value for qsmask & expmask. */
144 				/*  Initialized from ->qsmaskinitnext at the */
145 				/*  beginning of each grace period. */
146 	unsigned long qsmaskinitnext;
147 				/* Online CPUs for next grace period. */
148 	unsigned long grpmask;	/* Mask to apply to parent qsmask. */
149 				/*  Only one bit will be set in this mask. */
150 	int	grplo;		/* lowest-numbered CPU or group here. */
151 	int	grphi;		/* highest-numbered CPU or group here. */
152 	u8	grpnum;		/* CPU/group number for next level up. */
153 	u8	level;		/* root is at level 0. */
154 	bool	wait_blkd_tasks;/* Necessary to wait for blocked tasks to */
155 				/*  exit RCU read-side critical sections */
156 				/*  before propagating offline up the */
157 				/*  rcu_node tree? */
158 	struct rcu_node *parent;
159 	struct list_head blkd_tasks;
160 				/* Tasks blocked in RCU read-side critical */
161 				/*  section.  Tasks are placed at the head */
162 				/*  of this list and age towards the tail. */
163 	struct list_head *gp_tasks;
164 				/* Pointer to the first task blocking the */
165 				/*  current grace period, or NULL if there */
166 				/*  is no such task. */
167 	struct list_head *exp_tasks;
168 				/* Pointer to the first task blocking the */
169 				/*  current expedited grace period, or NULL */
170 				/*  if there is no such task.  If there */
171 				/*  is no current expedited grace period, */
172 				/*  then there can cannot be any such task. */
173 #ifdef CONFIG_RCU_BOOST
174 	struct list_head *boost_tasks;
175 				/* Pointer to first task that needs to be */
176 				/*  priority boosted, or NULL if no priority */
177 				/*  boosting is needed for this rcu_node */
178 				/*  structure.  If there are no tasks */
179 				/*  queued on this rcu_node structure that */
180 				/*  are blocking the current grace period, */
181 				/*  there can be no such task. */
182 	struct rt_mutex boost_mtx;
183 				/* Used only for the priority-boosting */
184 				/*  side effect, not as a lock. */
185 	unsigned long boost_time;
186 				/* When to start boosting (jiffies). */
187 	struct task_struct *boost_kthread_task;
188 				/* kthread that takes care of priority */
189 				/*  boosting for this rcu_node structure. */
190 	unsigned int boost_kthread_status;
191 				/* State of boost_kthread_task for tracing. */
192 	unsigned long n_tasks_boosted;
193 				/* Total number of tasks boosted. */
194 	unsigned long n_exp_boosts;
195 				/* Number of tasks boosted for expedited GP. */
196 	unsigned long n_normal_boosts;
197 				/* Number of tasks boosted for normal GP. */
198 	unsigned long n_balk_blkd_tasks;
199 				/* Refused to boost: no blocked tasks. */
200 	unsigned long n_balk_exp_gp_tasks;
201 				/* Refused to boost: nothing blocking GP. */
202 	unsigned long n_balk_boost_tasks;
203 				/* Refused to boost: already boosting. */
204 	unsigned long n_balk_notblocked;
205 				/* Refused to boost: RCU RS CS still running. */
206 	unsigned long n_balk_notyet;
207 				/* Refused to boost: not yet time. */
208 	unsigned long n_balk_nos;
209 				/* Refused to boost: not sure why, though. */
210 				/*  This can happen due to race conditions. */
211 #endif /* #ifdef CONFIG_RCU_BOOST */
212 #ifdef CONFIG_RCU_NOCB_CPU
213 	wait_queue_head_t nocb_gp_wq[2];
214 				/* Place for rcu_nocb_kthread() to wait GP. */
215 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
216 	int need_future_gp[2];
217 				/* Counts of upcoming no-CB GP requests. */
218 	raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp;
219 } ____cacheline_internodealigned_in_smp;
220 
221 /*
222  * Do a full breadth-first scan of the rcu_node structures for the
223  * specified rcu_state structure.
224  */
225 #define rcu_for_each_node_breadth_first(rsp, rnp) \
226 	for ((rnp) = &(rsp)->node[0]; \
227 	     (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
228 
229 /*
230  * Do a breadth-first scan of the non-leaf rcu_node structures for the
231  * specified rcu_state structure.  Note that if there is a singleton
232  * rcu_node tree with but one rcu_node structure, this loop is a no-op.
233  */
234 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
235 	for ((rnp) = &(rsp)->node[0]; \
236 	     (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
237 
238 /*
239  * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
240  * structure.  Note that if there is a singleton rcu_node tree with but
241  * one rcu_node structure, this loop -will- visit the rcu_node structure.
242  * It is still a leaf node, even if it is also the root node.
243  */
244 #define rcu_for_each_leaf_node(rsp, rnp) \
245 	for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
246 	     (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
247 
248 /* Index values for nxttail array in struct rcu_data. */
249 #define RCU_DONE_TAIL		0	/* Also RCU_WAIT head. */
250 #define RCU_WAIT_TAIL		1	/* Also RCU_NEXT_READY head. */
251 #define RCU_NEXT_READY_TAIL	2	/* Also RCU_NEXT head. */
252 #define RCU_NEXT_TAIL		3
253 #define RCU_NEXT_SIZE		4
254 
255 /* Per-CPU data for read-copy update. */
256 struct rcu_data {
257 	/* 1) quiescent-state and grace-period handling : */
258 	unsigned long	completed;	/* Track rsp->completed gp number */
259 					/*  in order to detect GP end. */
260 	unsigned long	gpnum;		/* Highest gp number that this CPU */
261 					/*  is aware of having started. */
262 	unsigned long	rcu_qs_ctr_snap;/* Snapshot of rcu_qs_ctr to check */
263 					/*  for rcu_all_qs() invocations. */
264 	bool		passed_quiesce;	/* User-mode/idle loop etc. */
265 	bool		qs_pending;	/* Core waits for quiesc state. */
266 	bool		beenonline;	/* CPU online at least once. */
267 	bool		gpwrap;		/* Possible gpnum/completed wrap. */
268 	struct rcu_node *mynode;	/* This CPU's leaf of hierarchy */
269 	unsigned long grpmask;		/* Mask to apply to leaf qsmask. */
270 #ifdef CONFIG_RCU_CPU_STALL_INFO
271 	unsigned long	ticks_this_gp;	/* The number of scheduling-clock */
272 					/*  ticks this CPU has handled */
273 					/*  during and after the last grace */
274 					/* period it is aware of. */
275 #endif /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
276 
277 	/* 2) batch handling */
278 	/*
279 	 * If nxtlist is not NULL, it is partitioned as follows.
280 	 * Any of the partitions might be empty, in which case the
281 	 * pointer to that partition will be equal to the pointer for
282 	 * the following partition.  When the list is empty, all of
283 	 * the nxttail elements point to the ->nxtlist pointer itself,
284 	 * which in that case is NULL.
285 	 *
286 	 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
287 	 *	Entries that batch # <= ->completed
288 	 *	The grace period for these entries has completed, and
289 	 *	the other grace-period-completed entries may be moved
290 	 *	here temporarily in rcu_process_callbacks().
291 	 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
292 	 *	Entries that batch # <= ->completed - 1: waiting for current GP
293 	 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
294 	 *	Entries known to have arrived before current GP ended
295 	 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
296 	 *	Entries that might have arrived after current GP ended
297 	 *	Note that the value of *nxttail[RCU_NEXT_TAIL] will
298 	 *	always be NULL, as this is the end of the list.
299 	 */
300 	struct rcu_head *nxtlist;
301 	struct rcu_head **nxttail[RCU_NEXT_SIZE];
302 	unsigned long	nxtcompleted[RCU_NEXT_SIZE];
303 					/* grace periods for sublists. */
304 	long		qlen_lazy;	/* # of lazy queued callbacks */
305 	long		qlen;		/* # of queued callbacks, incl lazy */
306 	long		qlen_last_fqs_check;
307 					/* qlen at last check for QS forcing */
308 	unsigned long	n_cbs_invoked;	/* count of RCU cbs invoked. */
309 	unsigned long	n_nocbs_invoked; /* count of no-CBs RCU cbs invoked. */
310 	unsigned long   n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
311 	unsigned long   n_cbs_adopted;  /* RCU cbs adopted from dying CPU */
312 	unsigned long	n_force_qs_snap;
313 					/* did other CPU force QS recently? */
314 	long		blimit;		/* Upper limit on a processed batch */
315 
316 	/* 3) dynticks interface. */
317 	struct rcu_dynticks *dynticks;	/* Shared per-CPU dynticks state. */
318 	int dynticks_snap;		/* Per-GP tracking for dynticks. */
319 
320 	/* 4) reasons this CPU needed to be kicked by force_quiescent_state */
321 	unsigned long dynticks_fqs;	/* Kicked due to dynticks idle. */
322 	unsigned long offline_fqs;	/* Kicked due to being offline. */
323 	unsigned long cond_resched_completed;
324 					/* Grace period that needs help */
325 					/*  from cond_resched(). */
326 
327 	/* 5) __rcu_pending() statistics. */
328 	unsigned long n_rcu_pending;	/* rcu_pending() calls since boot. */
329 	unsigned long n_rp_qs_pending;
330 	unsigned long n_rp_report_qs;
331 	unsigned long n_rp_cb_ready;
332 	unsigned long n_rp_cpu_needs_gp;
333 	unsigned long n_rp_gp_completed;
334 	unsigned long n_rp_gp_started;
335 	unsigned long n_rp_nocb_defer_wakeup;
336 	unsigned long n_rp_need_nothing;
337 
338 	/* 6) _rcu_barrier() and OOM callbacks. */
339 	struct rcu_head barrier_head;
340 #ifdef CONFIG_RCU_FAST_NO_HZ
341 	struct rcu_head oom_head;
342 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
343 
344 	/* 7) Callback offloading. */
345 #ifdef CONFIG_RCU_NOCB_CPU
346 	struct rcu_head *nocb_head;	/* CBs waiting for kthread. */
347 	struct rcu_head **nocb_tail;
348 	atomic_long_t nocb_q_count;	/* # CBs waiting for nocb */
349 	atomic_long_t nocb_q_count_lazy; /*  invocation (all stages). */
350 	struct rcu_head *nocb_follower_head; /* CBs ready to invoke. */
351 	struct rcu_head **nocb_follower_tail;
352 	wait_queue_head_t nocb_wq;	/* For nocb kthreads to sleep on. */
353 	struct task_struct *nocb_kthread;
354 	int nocb_defer_wakeup;		/* Defer wakeup of nocb_kthread. */
355 
356 	/* The following fields are used by the leader, hence own cacheline. */
357 	struct rcu_head *nocb_gp_head ____cacheline_internodealigned_in_smp;
358 					/* CBs waiting for GP. */
359 	struct rcu_head **nocb_gp_tail;
360 	bool nocb_leader_sleep;		/* Is the nocb leader thread asleep? */
361 	struct rcu_data *nocb_next_follower;
362 					/* Next follower in wakeup chain. */
363 
364 	/* The following fields are used by the follower, hence new cachline. */
365 	struct rcu_data *nocb_leader ____cacheline_internodealigned_in_smp;
366 					/* Leader CPU takes GP-end wakeups. */
367 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
368 
369 	/* 8) RCU CPU stall data. */
370 #ifdef CONFIG_RCU_CPU_STALL_INFO
371 	unsigned int softirq_snap;	/* Snapshot of softirq activity. */
372 #endif /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
373 
374 	int cpu;
375 	struct rcu_state *rsp;
376 };
377 
378 /* Values for fqs_state field in struct rcu_state. */
379 #define RCU_GP_IDLE		0	/* No grace period in progress. */
380 #define RCU_GP_INIT		1	/* Grace period being initialized. */
381 #define RCU_SAVE_DYNTICK	2	/* Need to scan dyntick state. */
382 #define RCU_FORCE_QS		3	/* Need to force quiescent state. */
383 #define RCU_SIGNAL_INIT		RCU_SAVE_DYNTICK
384 
385 /* Values for nocb_defer_wakeup field in struct rcu_data. */
386 #define RCU_NOGP_WAKE_NOT	0
387 #define RCU_NOGP_WAKE		1
388 #define RCU_NOGP_WAKE_FORCE	2
389 
390 #define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500))
391 					/* For jiffies_till_first_fqs and */
392 					/*  and jiffies_till_next_fqs. */
393 
394 #define RCU_JIFFIES_FQS_DIV	256	/* Very large systems need more */
395 					/*  delay between bouts of */
396 					/*  quiescent-state forcing. */
397 
398 #define RCU_STALL_RAT_DELAY	2	/* Allow other CPUs time to take */
399 					/*  at least one scheduling clock */
400 					/*  irq before ratting on them. */
401 
402 #define rcu_wait(cond)							\
403 do {									\
404 	for (;;) {							\
405 		set_current_state(TASK_INTERRUPTIBLE);			\
406 		if (cond)						\
407 			break;						\
408 		schedule();						\
409 	}								\
410 	__set_current_state(TASK_RUNNING);				\
411 } while (0)
412 
413 /*
414  * RCU global state, including node hierarchy.  This hierarchy is
415  * represented in "heap" form in a dense array.  The root (first level)
416  * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
417  * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
418  * and the third level in ->node[m+1] and following (->node[m+1] referenced
419  * by ->level[2]).  The number of levels is determined by the number of
420  * CPUs and by CONFIG_RCU_FANOUT.  Small systems will have a "hierarchy"
421  * consisting of a single rcu_node.
422  */
423 struct rcu_state {
424 	struct rcu_node node[NUM_RCU_NODES];	/* Hierarchy. */
425 	struct rcu_node *level[RCU_NUM_LVLS];	/* Hierarchy levels. */
426 	u32 levelcnt[MAX_RCU_LVLS + 1];		/* # nodes in each level. */
427 	u8 levelspread[RCU_NUM_LVLS];		/* kids/node in each level. */
428 	u8 flavor_mask;				/* bit in flavor mask. */
429 	struct rcu_data __percpu *rda;		/* pointer of percu rcu_data. */
430 	void (*call)(struct rcu_head *head,	/* call_rcu() flavor. */
431 		     void (*func)(struct rcu_head *head));
432 
433 	/* The following fields are guarded by the root rcu_node's lock. */
434 
435 	u8	fqs_state ____cacheline_internodealigned_in_smp;
436 						/* Force QS state. */
437 	u8	boost;				/* Subject to priority boost. */
438 	unsigned long gpnum;			/* Current gp number. */
439 	unsigned long completed;		/* # of last completed gp. */
440 	struct task_struct *gp_kthread;		/* Task for grace periods. */
441 	wait_queue_head_t gp_wq;		/* Where GP task waits. */
442 	short gp_flags;				/* Commands for GP task. */
443 	short gp_state;				/* GP kthread sleep state. */
444 
445 	/* End of fields guarded by root rcu_node's lock. */
446 
447 	raw_spinlock_t orphan_lock ____cacheline_internodealigned_in_smp;
448 						/* Protect following fields. */
449 	struct rcu_head *orphan_nxtlist;	/* Orphaned callbacks that */
450 						/*  need a grace period. */
451 	struct rcu_head **orphan_nxttail;	/* Tail of above. */
452 	struct rcu_head *orphan_donelist;	/* Orphaned callbacks that */
453 						/*  are ready to invoke. */
454 	struct rcu_head **orphan_donetail;	/* Tail of above. */
455 	long qlen_lazy;				/* Number of lazy callbacks. */
456 	long qlen;				/* Total number of callbacks. */
457 	/* End of fields guarded by orphan_lock. */
458 
459 	struct mutex barrier_mutex;		/* Guards barrier fields. */
460 	atomic_t barrier_cpu_count;		/* # CPUs waiting on. */
461 	struct completion barrier_completion;	/* Wake at barrier end. */
462 	unsigned long n_barrier_done;		/* ++ at start and end of */
463 						/*  _rcu_barrier(). */
464 	/* End of fields guarded by barrier_mutex. */
465 
466 	atomic_long_t expedited_start;		/* Starting ticket. */
467 	atomic_long_t expedited_done;		/* Done ticket. */
468 	atomic_long_t expedited_wrap;		/* # near-wrap incidents. */
469 	atomic_long_t expedited_tryfail;	/* # acquisition failures. */
470 	atomic_long_t expedited_workdone1;	/* # done by others #1. */
471 	atomic_long_t expedited_workdone2;	/* # done by others #2. */
472 	atomic_long_t expedited_normal;		/* # fallbacks to normal. */
473 	atomic_long_t expedited_stoppedcpus;	/* # successful stop_cpus. */
474 	atomic_long_t expedited_done_tries;	/* # tries to update _done. */
475 	atomic_long_t expedited_done_lost;	/* # times beaten to _done. */
476 	atomic_long_t expedited_done_exit;	/* # times exited _done loop. */
477 
478 	unsigned long jiffies_force_qs;		/* Time at which to invoke */
479 						/*  force_quiescent_state(). */
480 	unsigned long n_force_qs;		/* Number of calls to */
481 						/*  force_quiescent_state(). */
482 	unsigned long n_force_qs_lh;		/* ~Number of calls leaving */
483 						/*  due to lock unavailable. */
484 	unsigned long n_force_qs_ngp;		/* Number of calls leaving */
485 						/*  due to no GP active. */
486 	unsigned long gp_start;			/* Time at which GP started, */
487 						/*  but in jiffies. */
488 	unsigned long gp_activity;		/* Time of last GP kthread */
489 						/*  activity in jiffies. */
490 	unsigned long jiffies_stall;		/* Time at which to check */
491 						/*  for CPU stalls. */
492 	unsigned long jiffies_resched;		/* Time at which to resched */
493 						/*  a reluctant CPU. */
494 	unsigned long n_force_qs_gpstart;	/* Snapshot of n_force_qs at */
495 						/*  GP start. */
496 	unsigned long gp_max;			/* Maximum GP duration in */
497 						/*  jiffies. */
498 	const char *name;			/* Name of structure. */
499 	char abbr;				/* Abbreviated name. */
500 	struct list_head flavors;		/* List of RCU flavors. */
501 };
502 
503 /* Values for rcu_state structure's gp_flags field. */
504 #define RCU_GP_FLAG_INIT 0x1	/* Need grace-period initialization. */
505 #define RCU_GP_FLAG_FQS  0x2	/* Need grace-period quiescent-state forcing. */
506 
507 /* Values for rcu_state structure's gp_flags field. */
508 #define RCU_GP_WAIT_INIT 0	/* Initial state. */
509 #define RCU_GP_WAIT_GPS  1	/* Wait for grace-period start. */
510 #define RCU_GP_WAIT_FQS  2	/* Wait for force-quiescent-state time. */
511 
512 extern struct list_head rcu_struct_flavors;
513 
514 /* Sequence through rcu_state structures for each RCU flavor. */
515 #define for_each_rcu_flavor(rsp) \
516 	list_for_each_entry((rsp), &rcu_struct_flavors, flavors)
517 
518 /*
519  * RCU implementation internal declarations:
520  */
521 extern struct rcu_state rcu_sched_state;
522 DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
523 
524 extern struct rcu_state rcu_bh_state;
525 DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
526 
527 #ifdef CONFIG_PREEMPT_RCU
528 extern struct rcu_state rcu_preempt_state;
529 DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
530 #endif /* #ifdef CONFIG_PREEMPT_RCU */
531 
532 #ifdef CONFIG_RCU_BOOST
533 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
534 DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu);
535 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
536 DECLARE_PER_CPU(char, rcu_cpu_has_work);
537 #endif /* #ifdef CONFIG_RCU_BOOST */
538 
539 #ifndef RCU_TREE_NONCORE
540 
541 /* Forward declarations for rcutree_plugin.h */
542 static void rcu_bootup_announce(void);
543 static void rcu_preempt_note_context_switch(void);
544 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
545 #ifdef CONFIG_HOTPLUG_CPU
546 static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
547 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
548 static void rcu_print_detail_task_stall(struct rcu_state *rsp);
549 static int rcu_print_task_stall(struct rcu_node *rnp);
550 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
551 static void rcu_preempt_check_callbacks(void);
552 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
553 static void __init __rcu_init_preempt(void);
554 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
555 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
556 static void invoke_rcu_callbacks_kthread(void);
557 static bool rcu_is_callbacks_kthread(void);
558 #ifdef CONFIG_RCU_BOOST
559 static void rcu_preempt_do_callbacks(void);
560 static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
561 						 struct rcu_node *rnp);
562 #endif /* #ifdef CONFIG_RCU_BOOST */
563 static void __init rcu_spawn_boost_kthreads(void);
564 static void rcu_prepare_kthreads(int cpu);
565 static void rcu_cleanup_after_idle(void);
566 static void rcu_prepare_for_idle(void);
567 static void rcu_idle_count_callbacks_posted(void);
568 static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
569 static void print_cpu_stall_info_begin(void);
570 static void print_cpu_stall_info(struct rcu_state *rsp, int cpu);
571 static void print_cpu_stall_info_end(void);
572 static void zero_cpu_stall_ticks(struct rcu_data *rdp);
573 static void increment_cpu_stall_ticks(void);
574 static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu);
575 static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq);
576 static void rcu_nocb_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp);
577 static void rcu_init_one_nocb(struct rcu_node *rnp);
578 static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
579 			    bool lazy, unsigned long flags);
580 static bool rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp,
581 				      struct rcu_data *rdp,
582 				      unsigned long flags);
583 static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp);
584 static void do_nocb_deferred_wakeup(struct rcu_data *rdp);
585 static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp);
586 static void rcu_spawn_all_nocb_kthreads(int cpu);
587 static void __init rcu_spawn_nocb_kthreads(void);
588 #ifdef CONFIG_RCU_NOCB_CPU
589 static void __init rcu_organize_nocb_kthreads(struct rcu_state *rsp);
590 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
591 static void __maybe_unused rcu_kick_nohz_cpu(int cpu);
592 static bool init_nocb_callback_list(struct rcu_data *rdp);
593 static void rcu_sysidle_enter(int irq);
594 static void rcu_sysidle_exit(int irq);
595 static void rcu_sysidle_check_cpu(struct rcu_data *rdp, bool *isidle,
596 				  unsigned long *maxj);
597 static bool is_sysidle_rcu_state(struct rcu_state *rsp);
598 static void rcu_sysidle_report_gp(struct rcu_state *rsp, int isidle,
599 				  unsigned long maxj);
600 static void rcu_bind_gp_kthread(void);
601 static void rcu_sysidle_init_percpu_data(struct rcu_dynticks *rdtp);
602 static bool rcu_nohz_full_cpu(struct rcu_state *rsp);
603 static void rcu_dynticks_task_enter(void);
604 static void rcu_dynticks_task_exit(void);
605 
606 #endif /* #ifndef RCU_TREE_NONCORE */
607 
608 #ifdef CONFIG_RCU_TRACE
609 /* Read out queue lengths for tracing. */
610 static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll)
611 {
612 #ifdef CONFIG_RCU_NOCB_CPU
613 	*ql = atomic_long_read(&rdp->nocb_q_count);
614 	*qll = atomic_long_read(&rdp->nocb_q_count_lazy);
615 #else /* #ifdef CONFIG_RCU_NOCB_CPU */
616 	*ql = 0;
617 	*qll = 0;
618 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
619 }
620 #endif /* #ifdef CONFIG_RCU_TRACE */
621