xref: /openbmc/linux/kernel/rcu/tree.h (revision eb3fcf00)
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/stop_machine.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 
40 #ifdef CONFIG_RCU_FANOUT
41 #define RCU_FANOUT CONFIG_RCU_FANOUT
42 #else /* #ifdef CONFIG_RCU_FANOUT */
43 # ifdef CONFIG_64BIT
44 # define RCU_FANOUT 64
45 # else
46 # define RCU_FANOUT 32
47 # endif
48 #endif /* #else #ifdef CONFIG_RCU_FANOUT */
49 
50 #ifdef CONFIG_RCU_FANOUT_LEAF
51 #define RCU_FANOUT_LEAF CONFIG_RCU_FANOUT_LEAF
52 #else /* #ifdef CONFIG_RCU_FANOUT_LEAF */
53 # ifdef CONFIG_64BIT
54 # define RCU_FANOUT_LEAF 64
55 # else
56 # define RCU_FANOUT_LEAF 32
57 # endif
58 #endif /* #else #ifdef CONFIG_RCU_FANOUT_LEAF */
59 
60 #define RCU_FANOUT_1	      (RCU_FANOUT_LEAF)
61 #define RCU_FANOUT_2	      (RCU_FANOUT_1 * RCU_FANOUT)
62 #define RCU_FANOUT_3	      (RCU_FANOUT_2 * RCU_FANOUT)
63 #define RCU_FANOUT_4	      (RCU_FANOUT_3 * RCU_FANOUT)
64 
65 #if NR_CPUS <= RCU_FANOUT_1
66 #  define RCU_NUM_LVLS	      1
67 #  define NUM_RCU_LVL_0	      1
68 #  define NUM_RCU_NODES	      NUM_RCU_LVL_0
69 #  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0 }
70 #  define RCU_NODE_NAME_INIT  { "rcu_node_0" }
71 #  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0" }
72 #  define RCU_EXP_NAME_INIT   { "rcu_node_exp_0" }
73 #  define RCU_EXP_SCHED_NAME_INIT \
74 			      { "rcu_node_exp_sched_0" }
75 #elif NR_CPUS <= RCU_FANOUT_2
76 #  define RCU_NUM_LVLS	      2
77 #  define NUM_RCU_LVL_0	      1
78 #  define NUM_RCU_LVL_1	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
79 #  define NUM_RCU_NODES	      (NUM_RCU_LVL_0 + NUM_RCU_LVL_1)
80 #  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0, NUM_RCU_LVL_1 }
81 #  define RCU_NODE_NAME_INIT  { "rcu_node_0", "rcu_node_1" }
82 #  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0", "rcu_node_fqs_1" }
83 #  define RCU_EXP_NAME_INIT   { "rcu_node_exp_0", "rcu_node_exp_1" }
84 #  define RCU_EXP_SCHED_NAME_INIT \
85 			      { "rcu_node_exp_sched_0", "rcu_node_exp_sched_1" }
86 #elif NR_CPUS <= RCU_FANOUT_3
87 #  define RCU_NUM_LVLS	      3
88 #  define NUM_RCU_LVL_0	      1
89 #  define NUM_RCU_LVL_1	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
90 #  define NUM_RCU_LVL_2	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
91 #  define NUM_RCU_NODES	      (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2)
92 #  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0, NUM_RCU_LVL_1, NUM_RCU_LVL_2 }
93 #  define RCU_NODE_NAME_INIT  { "rcu_node_0", "rcu_node_1", "rcu_node_2" }
94 #  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0", "rcu_node_fqs_1", "rcu_node_fqs_2" }
95 #  define RCU_EXP_NAME_INIT   { "rcu_node_exp_0", "rcu_node_exp_1", "rcu_node_exp_2" }
96 #  define RCU_EXP_SCHED_NAME_INIT \
97 			      { "rcu_node_exp_sched_0", "rcu_node_exp_sched_1", "rcu_node_exp_sched_2" }
98 #elif NR_CPUS <= RCU_FANOUT_4
99 #  define RCU_NUM_LVLS	      4
100 #  define NUM_RCU_LVL_0	      1
101 #  define NUM_RCU_LVL_1	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3)
102 #  define NUM_RCU_LVL_2	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
103 #  define NUM_RCU_LVL_3	      DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
104 #  define NUM_RCU_NODES	      (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3)
105 #  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0, NUM_RCU_LVL_1, NUM_RCU_LVL_2, NUM_RCU_LVL_3 }
106 #  define RCU_NODE_NAME_INIT  { "rcu_node_0", "rcu_node_1", "rcu_node_2", "rcu_node_3" }
107 #  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0", "rcu_node_fqs_1", "rcu_node_fqs_2", "rcu_node_fqs_3" }
108 #  define RCU_EXP_NAME_INIT   { "rcu_node_exp_0", "rcu_node_exp_1", "rcu_node_exp_2", "rcu_node_exp_3" }
109 #  define RCU_EXP_SCHED_NAME_INIT \
110 			      { "rcu_node_exp_sched_0", "rcu_node_exp_sched_1", "rcu_node_exp_sched_2", "rcu_node_exp_sched_3" }
111 #else
112 # error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
113 #endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
114 
115 extern int rcu_num_lvls;
116 extern int rcu_num_nodes;
117 
118 /*
119  * Dynticks per-CPU state.
120  */
121 struct rcu_dynticks {
122 	long long dynticks_nesting; /* Track irq/process nesting level. */
123 				    /* Process level is worth LLONG_MAX/2. */
124 	int dynticks_nmi_nesting;   /* Track NMI nesting level. */
125 	atomic_t dynticks;	    /* Even value for idle, else odd. */
126 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
127 	long long dynticks_idle_nesting;
128 				    /* irq/process nesting level from idle. */
129 	atomic_t dynticks_idle;	    /* Even value for idle, else odd. */
130 				    /*  "Idle" excludes userspace execution. */
131 	unsigned long dynticks_idle_jiffies;
132 				    /* End of last non-NMI non-idle period. */
133 #endif /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
134 #ifdef CONFIG_RCU_FAST_NO_HZ
135 	bool all_lazy;		    /* Are all CPU's CBs lazy? */
136 	unsigned long nonlazy_posted;
137 				    /* # times non-lazy CBs posted to CPU. */
138 	unsigned long nonlazy_posted_snap;
139 				    /* idle-period nonlazy_posted snapshot. */
140 	unsigned long last_accelerate;
141 				    /* Last jiffy CBs were accelerated. */
142 	unsigned long last_advance_all;
143 				    /* Last jiffy CBs were all advanced. */
144 	int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */
145 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
146 };
147 
148 /* RCU's kthread states for tracing. */
149 #define RCU_KTHREAD_STOPPED  0
150 #define RCU_KTHREAD_RUNNING  1
151 #define RCU_KTHREAD_WAITING  2
152 #define RCU_KTHREAD_OFFCPU   3
153 #define RCU_KTHREAD_YIELDING 4
154 #define RCU_KTHREAD_MAX      4
155 
156 /*
157  * Definition for node within the RCU grace-period-detection hierarchy.
158  */
159 struct rcu_node {
160 	raw_spinlock_t lock;	/* Root rcu_node's lock protects some */
161 				/*  rcu_state fields as well as following. */
162 	unsigned long gpnum;	/* Current grace period for this node. */
163 				/*  This will either be equal to or one */
164 				/*  behind the root rcu_node's gpnum. */
165 	unsigned long completed; /* Last GP completed for this node. */
166 				/*  This will either be equal to or one */
167 				/*  behind the root rcu_node's gpnum. */
168 	unsigned long qsmask;	/* CPUs or groups that need to switch in */
169 				/*  order for current grace period to proceed.*/
170 				/*  In leaf rcu_node, each bit corresponds to */
171 				/*  an rcu_data structure, otherwise, each */
172 				/*  bit corresponds to a child rcu_node */
173 				/*  structure. */
174 	unsigned long expmask;	/* Groups that have ->blkd_tasks */
175 				/*  elements that need to drain to allow the */
176 				/*  current expedited grace period to */
177 				/*  complete (only for PREEMPT_RCU). */
178 	unsigned long qsmaskinit;
179 				/* Per-GP initial value for qsmask & expmask. */
180 				/*  Initialized from ->qsmaskinitnext at the */
181 				/*  beginning of each grace period. */
182 	unsigned long qsmaskinitnext;
183 				/* Online CPUs for next grace period. */
184 	unsigned long grpmask;	/* Mask to apply to parent qsmask. */
185 				/*  Only one bit will be set in this mask. */
186 	int	grplo;		/* lowest-numbered CPU or group here. */
187 	int	grphi;		/* highest-numbered CPU or group here. */
188 	u8	grpnum;		/* CPU/group number for next level up. */
189 	u8	level;		/* root is at level 0. */
190 	bool	wait_blkd_tasks;/* Necessary to wait for blocked tasks to */
191 				/*  exit RCU read-side critical sections */
192 				/*  before propagating offline up the */
193 				/*  rcu_node tree? */
194 	struct rcu_node *parent;
195 	struct list_head blkd_tasks;
196 				/* Tasks blocked in RCU read-side critical */
197 				/*  section.  Tasks are placed at the head */
198 				/*  of this list and age towards the tail. */
199 	struct list_head *gp_tasks;
200 				/* Pointer to the first task blocking the */
201 				/*  current grace period, or NULL if there */
202 				/*  is no such task. */
203 	struct list_head *exp_tasks;
204 				/* Pointer to the first task blocking the */
205 				/*  current expedited grace period, or NULL */
206 				/*  if there is no such task.  If there */
207 				/*  is no current expedited grace period, */
208 				/*  then there can cannot be any such task. */
209 	struct list_head *boost_tasks;
210 				/* Pointer to first task that needs to be */
211 				/*  priority boosted, or NULL if no priority */
212 				/*  boosting is needed for this rcu_node */
213 				/*  structure.  If there are no tasks */
214 				/*  queued on this rcu_node structure that */
215 				/*  are blocking the current grace period, */
216 				/*  there can be no such task. */
217 	struct rt_mutex boost_mtx;
218 				/* Used only for the priority-boosting */
219 				/*  side effect, not as a lock. */
220 	unsigned long boost_time;
221 				/* When to start boosting (jiffies). */
222 	struct task_struct *boost_kthread_task;
223 				/* kthread that takes care of priority */
224 				/*  boosting for this rcu_node structure. */
225 	unsigned int boost_kthread_status;
226 				/* State of boost_kthread_task for tracing. */
227 	unsigned long n_tasks_boosted;
228 				/* Total number of tasks boosted. */
229 	unsigned long n_exp_boosts;
230 				/* Number of tasks boosted for expedited GP. */
231 	unsigned long n_normal_boosts;
232 				/* Number of tasks boosted for normal GP. */
233 	unsigned long n_balk_blkd_tasks;
234 				/* Refused to boost: no blocked tasks. */
235 	unsigned long n_balk_exp_gp_tasks;
236 				/* Refused to boost: nothing blocking GP. */
237 	unsigned long n_balk_boost_tasks;
238 				/* Refused to boost: already boosting. */
239 	unsigned long n_balk_notblocked;
240 				/* Refused to boost: RCU RS CS still running. */
241 	unsigned long n_balk_notyet;
242 				/* Refused to boost: not yet time. */
243 	unsigned long n_balk_nos;
244 				/* Refused to boost: not sure why, though. */
245 				/*  This can happen due to race conditions. */
246 #ifdef CONFIG_RCU_NOCB_CPU
247 	wait_queue_head_t nocb_gp_wq[2];
248 				/* Place for rcu_nocb_kthread() to wait GP. */
249 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
250 	int need_future_gp[2];
251 				/* Counts of upcoming no-CB GP requests. */
252 	raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp;
253 
254 	struct mutex exp_funnel_mutex ____cacheline_internodealigned_in_smp;
255 } ____cacheline_internodealigned_in_smp;
256 
257 /*
258  * Do a full breadth-first scan of the rcu_node structures for the
259  * specified rcu_state structure.
260  */
261 #define rcu_for_each_node_breadth_first(rsp, rnp) \
262 	for ((rnp) = &(rsp)->node[0]; \
263 	     (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
264 
265 /*
266  * Do a breadth-first scan of the non-leaf rcu_node structures for the
267  * specified rcu_state structure.  Note that if there is a singleton
268  * rcu_node tree with but one rcu_node structure, this loop is a no-op.
269  */
270 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
271 	for ((rnp) = &(rsp)->node[0]; \
272 	     (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
273 
274 /*
275  * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
276  * structure.  Note that if there is a singleton rcu_node tree with but
277  * one rcu_node structure, this loop -will- visit the rcu_node structure.
278  * It is still a leaf node, even if it is also the root node.
279  */
280 #define rcu_for_each_leaf_node(rsp, rnp) \
281 	for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
282 	     (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
283 
284 /* Index values for nxttail array in struct rcu_data. */
285 #define RCU_DONE_TAIL		0	/* Also RCU_WAIT head. */
286 #define RCU_WAIT_TAIL		1	/* Also RCU_NEXT_READY head. */
287 #define RCU_NEXT_READY_TAIL	2	/* Also RCU_NEXT head. */
288 #define RCU_NEXT_TAIL		3
289 #define RCU_NEXT_SIZE		4
290 
291 /* Per-CPU data for read-copy update. */
292 struct rcu_data {
293 	/* 1) quiescent-state and grace-period handling : */
294 	unsigned long	completed;	/* Track rsp->completed gp number */
295 					/*  in order to detect GP end. */
296 	unsigned long	gpnum;		/* Highest gp number that this CPU */
297 					/*  is aware of having started. */
298 	unsigned long	rcu_qs_ctr_snap;/* Snapshot of rcu_qs_ctr to check */
299 					/*  for rcu_all_qs() invocations. */
300 	bool		passed_quiesce;	/* User-mode/idle loop etc. */
301 	bool		qs_pending;	/* Core waits for quiesc state. */
302 	bool		beenonline;	/* CPU online at least once. */
303 	bool		gpwrap;		/* Possible gpnum/completed wrap. */
304 	struct rcu_node *mynode;	/* This CPU's leaf of hierarchy */
305 	unsigned long grpmask;		/* Mask to apply to leaf qsmask. */
306 	unsigned long	ticks_this_gp;	/* The number of scheduling-clock */
307 					/*  ticks this CPU has handled */
308 					/*  during and after the last grace */
309 					/* period it is aware of. */
310 	struct cpu_stop_work exp_stop_work;
311 					/* Expedited grace-period control */
312 					/*  for CPU stopping. */
313 
314 	/* 2) batch handling */
315 	/*
316 	 * If nxtlist is not NULL, it is partitioned as follows.
317 	 * Any of the partitions might be empty, in which case the
318 	 * pointer to that partition will be equal to the pointer for
319 	 * the following partition.  When the list is empty, all of
320 	 * the nxttail elements point to the ->nxtlist pointer itself,
321 	 * which in that case is NULL.
322 	 *
323 	 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
324 	 *	Entries that batch # <= ->completed
325 	 *	The grace period for these entries has completed, and
326 	 *	the other grace-period-completed entries may be moved
327 	 *	here temporarily in rcu_process_callbacks().
328 	 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
329 	 *	Entries that batch # <= ->completed - 1: waiting for current GP
330 	 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
331 	 *	Entries known to have arrived before current GP ended
332 	 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
333 	 *	Entries that might have arrived after current GP ended
334 	 *	Note that the value of *nxttail[RCU_NEXT_TAIL] will
335 	 *	always be NULL, as this is the end of the list.
336 	 */
337 	struct rcu_head *nxtlist;
338 	struct rcu_head **nxttail[RCU_NEXT_SIZE];
339 	unsigned long	nxtcompleted[RCU_NEXT_SIZE];
340 					/* grace periods for sublists. */
341 	long		qlen_lazy;	/* # of lazy queued callbacks */
342 	long		qlen;		/* # of queued callbacks, incl lazy */
343 	long		qlen_last_fqs_check;
344 					/* qlen at last check for QS forcing */
345 	unsigned long	n_cbs_invoked;	/* count of RCU cbs invoked. */
346 	unsigned long	n_nocbs_invoked; /* count of no-CBs RCU cbs invoked. */
347 	unsigned long   n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
348 	unsigned long   n_cbs_adopted;  /* RCU cbs adopted from dying CPU */
349 	unsigned long	n_force_qs_snap;
350 					/* did other CPU force QS recently? */
351 	long		blimit;		/* Upper limit on a processed batch */
352 
353 	/* 3) dynticks interface. */
354 	struct rcu_dynticks *dynticks;	/* Shared per-CPU dynticks state. */
355 	int dynticks_snap;		/* Per-GP tracking for dynticks. */
356 
357 	/* 4) reasons this CPU needed to be kicked by force_quiescent_state */
358 	unsigned long dynticks_fqs;	/* Kicked due to dynticks idle. */
359 	unsigned long offline_fqs;	/* Kicked due to being offline. */
360 	unsigned long cond_resched_completed;
361 					/* Grace period that needs help */
362 					/*  from cond_resched(). */
363 
364 	/* 5) __rcu_pending() statistics. */
365 	unsigned long n_rcu_pending;	/* rcu_pending() calls since boot. */
366 	unsigned long n_rp_qs_pending;
367 	unsigned long n_rp_report_qs;
368 	unsigned long n_rp_cb_ready;
369 	unsigned long n_rp_cpu_needs_gp;
370 	unsigned long n_rp_gp_completed;
371 	unsigned long n_rp_gp_started;
372 	unsigned long n_rp_nocb_defer_wakeup;
373 	unsigned long n_rp_need_nothing;
374 
375 	/* 6) _rcu_barrier(), OOM callbacks, and expediting. */
376 	struct rcu_head barrier_head;
377 #ifdef CONFIG_RCU_FAST_NO_HZ
378 	struct rcu_head oom_head;
379 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
380 	struct mutex exp_funnel_mutex;
381 	bool exp_done;			/* Expedited QS for this CPU? */
382 
383 	/* 7) Callback offloading. */
384 #ifdef CONFIG_RCU_NOCB_CPU
385 	struct rcu_head *nocb_head;	/* CBs waiting for kthread. */
386 	struct rcu_head **nocb_tail;
387 	atomic_long_t nocb_q_count;	/* # CBs waiting for nocb */
388 	atomic_long_t nocb_q_count_lazy; /*  invocation (all stages). */
389 	struct rcu_head *nocb_follower_head; /* CBs ready to invoke. */
390 	struct rcu_head **nocb_follower_tail;
391 	wait_queue_head_t nocb_wq;	/* For nocb kthreads to sleep on. */
392 	struct task_struct *nocb_kthread;
393 	int nocb_defer_wakeup;		/* Defer wakeup of nocb_kthread. */
394 
395 	/* The following fields are used by the leader, hence own cacheline. */
396 	struct rcu_head *nocb_gp_head ____cacheline_internodealigned_in_smp;
397 					/* CBs waiting for GP. */
398 	struct rcu_head **nocb_gp_tail;
399 	bool nocb_leader_sleep;		/* Is the nocb leader thread asleep? */
400 	struct rcu_data *nocb_next_follower;
401 					/* Next follower in wakeup chain. */
402 
403 	/* The following fields are used by the follower, hence new cachline. */
404 	struct rcu_data *nocb_leader ____cacheline_internodealigned_in_smp;
405 					/* Leader CPU takes GP-end wakeups. */
406 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
407 
408 	/* 8) RCU CPU stall data. */
409 	unsigned int softirq_snap;	/* Snapshot of softirq activity. */
410 
411 	int cpu;
412 	struct rcu_state *rsp;
413 };
414 
415 /* Values for fqs_state field in struct rcu_state. */
416 #define RCU_GP_IDLE		0	/* No grace period in progress. */
417 #define RCU_GP_INIT		1	/* Grace period being initialized. */
418 #define RCU_SAVE_DYNTICK	2	/* Need to scan dyntick state. */
419 #define RCU_FORCE_QS		3	/* Need to force quiescent state. */
420 #define RCU_SIGNAL_INIT		RCU_SAVE_DYNTICK
421 
422 /* Values for nocb_defer_wakeup field in struct rcu_data. */
423 #define RCU_NOGP_WAKE_NOT	0
424 #define RCU_NOGP_WAKE		1
425 #define RCU_NOGP_WAKE_FORCE	2
426 
427 #define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500))
428 					/* For jiffies_till_first_fqs and */
429 					/*  and jiffies_till_next_fqs. */
430 
431 #define RCU_JIFFIES_FQS_DIV	256	/* Very large systems need more */
432 					/*  delay between bouts of */
433 					/*  quiescent-state forcing. */
434 
435 #define RCU_STALL_RAT_DELAY	2	/* Allow other CPUs time to take */
436 					/*  at least one scheduling clock */
437 					/*  irq before ratting on them. */
438 
439 #define rcu_wait(cond)							\
440 do {									\
441 	for (;;) {							\
442 		set_current_state(TASK_INTERRUPTIBLE);			\
443 		if (cond)						\
444 			break;						\
445 		schedule();						\
446 	}								\
447 	__set_current_state(TASK_RUNNING);				\
448 } while (0)
449 
450 /*
451  * RCU global state, including node hierarchy.  This hierarchy is
452  * represented in "heap" form in a dense array.  The root (first level)
453  * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
454  * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
455  * and the third level in ->node[m+1] and following (->node[m+1] referenced
456  * by ->level[2]).  The number of levels is determined by the number of
457  * CPUs and by CONFIG_RCU_FANOUT.  Small systems will have a "hierarchy"
458  * consisting of a single rcu_node.
459  */
460 struct rcu_state {
461 	struct rcu_node node[NUM_RCU_NODES];	/* Hierarchy. */
462 	struct rcu_node *level[RCU_NUM_LVLS + 1];
463 						/* Hierarchy levels (+1 to */
464 						/*  shut bogus gcc warning) */
465 	u8 flavor_mask;				/* bit in flavor mask. */
466 	struct rcu_data __percpu *rda;		/* pointer of percu rcu_data. */
467 	void (*call)(struct rcu_head *head,	/* call_rcu() flavor. */
468 		     void (*func)(struct rcu_head *head));
469 
470 	/* The following fields are guarded by the root rcu_node's lock. */
471 
472 	u8	fqs_state ____cacheline_internodealigned_in_smp;
473 						/* Force QS state. */
474 	u8	boost;				/* Subject to priority boost. */
475 	unsigned long gpnum;			/* Current gp number. */
476 	unsigned long completed;		/* # of last completed gp. */
477 	struct task_struct *gp_kthread;		/* Task for grace periods. */
478 	wait_queue_head_t gp_wq;		/* Where GP task waits. */
479 	short gp_flags;				/* Commands for GP task. */
480 	short gp_state;				/* GP kthread sleep state. */
481 
482 	/* End of fields guarded by root rcu_node's lock. */
483 
484 	raw_spinlock_t orphan_lock ____cacheline_internodealigned_in_smp;
485 						/* Protect following fields. */
486 	struct rcu_head *orphan_nxtlist;	/* Orphaned callbacks that */
487 						/*  need a grace period. */
488 	struct rcu_head **orphan_nxttail;	/* Tail of above. */
489 	struct rcu_head *orphan_donelist;	/* Orphaned callbacks that */
490 						/*  are ready to invoke. */
491 	struct rcu_head **orphan_donetail;	/* Tail of above. */
492 	long qlen_lazy;				/* Number of lazy callbacks. */
493 	long qlen;				/* Total number of callbacks. */
494 	/* End of fields guarded by orphan_lock. */
495 
496 	struct mutex barrier_mutex;		/* Guards barrier fields. */
497 	atomic_t barrier_cpu_count;		/* # CPUs waiting on. */
498 	struct completion barrier_completion;	/* Wake at barrier end. */
499 	unsigned long barrier_sequence;		/* ++ at start and end of */
500 						/*  _rcu_barrier(). */
501 	/* End of fields guarded by barrier_mutex. */
502 
503 	unsigned long expedited_sequence;	/* Take a ticket. */
504 	atomic_long_t expedited_workdone0;	/* # done by others #0. */
505 	atomic_long_t expedited_workdone1;	/* # done by others #1. */
506 	atomic_long_t expedited_workdone2;	/* # done by others #2. */
507 	atomic_long_t expedited_workdone3;	/* # done by others #3. */
508 	atomic_long_t expedited_normal;		/* # fallbacks to normal. */
509 	atomic_t expedited_need_qs;		/* # CPUs left to check in. */
510 	wait_queue_head_t expedited_wq;		/* Wait for check-ins. */
511 
512 	unsigned long jiffies_force_qs;		/* Time at which to invoke */
513 						/*  force_quiescent_state(). */
514 	unsigned long n_force_qs;		/* Number of calls to */
515 						/*  force_quiescent_state(). */
516 	unsigned long n_force_qs_lh;		/* ~Number of calls leaving */
517 						/*  due to lock unavailable. */
518 	unsigned long n_force_qs_ngp;		/* Number of calls leaving */
519 						/*  due to no GP active. */
520 	unsigned long gp_start;			/* Time at which GP started, */
521 						/*  but in jiffies. */
522 	unsigned long gp_activity;		/* Time of last GP kthread */
523 						/*  activity in jiffies. */
524 	unsigned long jiffies_stall;		/* Time at which to check */
525 						/*  for CPU stalls. */
526 	unsigned long jiffies_resched;		/* Time at which to resched */
527 						/*  a reluctant CPU. */
528 	unsigned long n_force_qs_gpstart;	/* Snapshot of n_force_qs at */
529 						/*  GP start. */
530 	unsigned long gp_max;			/* Maximum GP duration in */
531 						/*  jiffies. */
532 	const char *name;			/* Name of structure. */
533 	char abbr;				/* Abbreviated name. */
534 	struct list_head flavors;		/* List of RCU flavors. */
535 };
536 
537 /* Values for rcu_state structure's gp_flags field. */
538 #define RCU_GP_FLAG_INIT 0x1	/* Need grace-period initialization. */
539 #define RCU_GP_FLAG_FQS  0x2	/* Need grace-period quiescent-state forcing. */
540 
541 /* Values for rcu_state structure's gp_flags field. */
542 #define RCU_GP_WAIT_INIT 0	/* Initial state. */
543 #define RCU_GP_WAIT_GPS  1	/* Wait for grace-period start. */
544 #define RCU_GP_DONE_GPS  2	/* Wait done for grace-period start. */
545 #define RCU_GP_WAIT_FQS  3	/* Wait for force-quiescent-state time. */
546 #define RCU_GP_DOING_FQS 4	/* Wait done for force-quiescent-state time. */
547 #define RCU_GP_CLEANUP   5	/* Grace-period cleanup started. */
548 #define RCU_GP_CLEANED   6	/* Grace-period cleanup complete. */
549 
550 extern struct list_head rcu_struct_flavors;
551 
552 /* Sequence through rcu_state structures for each RCU flavor. */
553 #define for_each_rcu_flavor(rsp) \
554 	list_for_each_entry((rsp), &rcu_struct_flavors, flavors)
555 
556 /*
557  * RCU implementation internal declarations:
558  */
559 extern struct rcu_state rcu_sched_state;
560 
561 extern struct rcu_state rcu_bh_state;
562 
563 #ifdef CONFIG_PREEMPT_RCU
564 extern struct rcu_state rcu_preempt_state;
565 #endif /* #ifdef CONFIG_PREEMPT_RCU */
566 
567 #ifdef CONFIG_RCU_BOOST
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);
572 #endif /* #ifdef CONFIG_RCU_BOOST */
573 
574 #ifndef RCU_TREE_NONCORE
575 
576 /* Forward declarations for rcutree_plugin.h */
577 static void rcu_bootup_announce(void);
578 static void rcu_preempt_note_context_switch(void);
579 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
580 #ifdef CONFIG_HOTPLUG_CPU
581 static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
582 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
583 static void rcu_print_detail_task_stall(struct rcu_state *rsp);
584 static int rcu_print_task_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, void (*func)(struct rcu_head *rcu));
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 void invoke_rcu_callbacks_kthread(void);
592 static bool rcu_is_callbacks_kthread(void);
593 #ifdef CONFIG_RCU_BOOST
594 static void rcu_preempt_do_callbacks(void);
595 static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
596 						 struct rcu_node *rnp);
597 #endif /* #ifdef CONFIG_RCU_BOOST */
598 static void __init rcu_spawn_boost_kthreads(void);
599 static void rcu_prepare_kthreads(int cpu);
600 static void rcu_cleanup_after_idle(void);
601 static void rcu_prepare_for_idle(void);
602 static void rcu_idle_count_callbacks_posted(void);
603 static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
604 static void print_cpu_stall_info_begin(void);
605 static void print_cpu_stall_info(struct rcu_state *rsp, int cpu);
606 static void print_cpu_stall_info_end(void);
607 static void zero_cpu_stall_ticks(struct rcu_data *rdp);
608 static void increment_cpu_stall_ticks(void);
609 static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu);
610 static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq);
611 static void rcu_nocb_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp);
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,
634 				  unsigned long maxj);
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);
640 
641 #endif /* #ifndef RCU_TREE_NONCORE */
642 
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)
646 {
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 */
651 	*ql = 0;
652 	*qll = 0;
653 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
654 }
655 #endif /* #ifdef CONFIG_RCU_TRACE */
656 
657 /*
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.
662  */
663 #ifdef CONFIG_PPC
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 */
668