Loading...
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
80extern int rcu_num_lvls;
81extern int rcu_num_nodes;
82
83/*
84 * Dynticks per-CPU state.
85 */
86struct 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 */
124struct 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 unsigned long boost_time;
176 /* When to start boosting (jiffies). */
177 struct task_struct *boost_kthread_task;
178 /* kthread that takes care of priority */
179 /* boosting for this rcu_node structure. */
180 unsigned int boost_kthread_status;
181 /* State of boost_kthread_task for tracing. */
182 unsigned long n_tasks_boosted;
183 /* Total number of tasks boosted. */
184 unsigned long n_exp_boosts;
185 /* Number of tasks boosted for expedited GP. */
186 unsigned long n_normal_boosts;
187 /* Number of tasks boosted for normal GP. */
188 unsigned long n_balk_blkd_tasks;
189 /* Refused to boost: no blocked tasks. */
190 unsigned long n_balk_exp_gp_tasks;
191 /* Refused to boost: nothing blocking GP. */
192 unsigned long n_balk_boost_tasks;
193 /* Refused to boost: already boosting. */
194 unsigned long n_balk_notblocked;
195 /* Refused to boost: RCU RS CS still running. */
196 unsigned long n_balk_notyet;
197 /* Refused to boost: not yet time. */
198 unsigned long n_balk_nos;
199 /* Refused to boost: not sure why, though. */
200 /* This can happen due to race conditions. */
201#endif /* #ifdef CONFIG_RCU_BOOST */
202#ifdef CONFIG_RCU_NOCB_CPU
203 wait_queue_head_t nocb_gp_wq[2];
204 /* Place for rcu_nocb_kthread() to wait GP. */
205#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
206 int need_future_gp[2];
207 /* Counts of upcoming no-CB GP requests. */
208 raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp;
209} ____cacheline_internodealigned_in_smp;
210
211/*
212 * Do a full breadth-first scan of the rcu_node structures for the
213 * specified rcu_state structure.
214 */
215#define rcu_for_each_node_breadth_first(rsp, rnp) \
216 for ((rnp) = &(rsp)->node[0]; \
217 (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
218
219/*
220 * Do a breadth-first scan of the non-leaf rcu_node structures for the
221 * specified rcu_state structure. Note that if there is a singleton
222 * rcu_node tree with but one rcu_node structure, this loop is a no-op.
223 */
224#define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
225 for ((rnp) = &(rsp)->node[0]; \
226 (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
227
228/*
229 * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
230 * structure. Note that if there is a singleton rcu_node tree with but
231 * one rcu_node structure, this loop -will- visit the rcu_node structure.
232 * It is still a leaf node, even if it is also the root node.
233 */
234#define rcu_for_each_leaf_node(rsp, rnp) \
235 for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
236 (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
237
238/* Index values for nxttail array in struct rcu_data. */
239#define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
240#define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
241#define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
242#define RCU_NEXT_TAIL 3
243#define RCU_NEXT_SIZE 4
244
245/* Per-CPU data for read-copy update. */
246struct rcu_data {
247 /* 1) quiescent-state and grace-period handling : */
248 unsigned long completed; /* Track rsp->completed gp number */
249 /* in order to detect GP end. */
250 unsigned long gpnum; /* Highest gp number that this CPU */
251 /* is aware of having started. */
252 bool passed_quiesce; /* User-mode/idle loop etc. */
253 bool qs_pending; /* Core waits for quiesc state. */
254 bool beenonline; /* CPU online at least once. */
255 bool preemptible; /* Preemptible RCU? */
256 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
257 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
258#ifdef CONFIG_RCU_CPU_STALL_INFO
259 unsigned long ticks_this_gp; /* The number of scheduling-clock */
260 /* ticks this CPU has handled */
261 /* during and after the last grace */
262 /* period it is aware of. */
263#endif /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
264
265 /* 2) batch handling */
266 /*
267 * If nxtlist is not NULL, it is partitioned as follows.
268 * Any of the partitions might be empty, in which case the
269 * pointer to that partition will be equal to the pointer for
270 * the following partition. When the list is empty, all of
271 * the nxttail elements point to the ->nxtlist pointer itself,
272 * which in that case is NULL.
273 *
274 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
275 * Entries that batch # <= ->completed
276 * The grace period for these entries has completed, and
277 * the other grace-period-completed entries may be moved
278 * here temporarily in rcu_process_callbacks().
279 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
280 * Entries that batch # <= ->completed - 1: waiting for current GP
281 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
282 * Entries known to have arrived before current GP ended
283 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
284 * Entries that might have arrived after current GP ended
285 * Note that the value of *nxttail[RCU_NEXT_TAIL] will
286 * always be NULL, as this is the end of the list.
287 */
288 struct rcu_head *nxtlist;
289 struct rcu_head **nxttail[RCU_NEXT_SIZE];
290 unsigned long nxtcompleted[RCU_NEXT_SIZE];
291 /* grace periods for sublists. */
292 long qlen_lazy; /* # of lazy queued callbacks */
293 long qlen; /* # of queued callbacks, incl lazy */
294 long qlen_last_fqs_check;
295 /* qlen at last check for QS forcing */
296 unsigned long n_cbs_invoked; /* count of RCU cbs invoked. */
297 unsigned long n_nocbs_invoked; /* count of no-CBs RCU cbs invoked. */
298 unsigned long n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
299 unsigned long n_cbs_adopted; /* RCU cbs adopted from dying CPU */
300 unsigned long n_force_qs_snap;
301 /* did other CPU force QS recently? */
302 long blimit; /* Upper limit on a processed batch */
303
304 /* 3) dynticks interface. */
305 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
306 int dynticks_snap; /* Per-GP tracking for dynticks. */
307
308 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
309 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
310 unsigned long offline_fqs; /* Kicked due to being offline. */
311
312 /* 5) __rcu_pending() statistics. */
313 unsigned long n_rcu_pending; /* rcu_pending() calls since boot. */
314 unsigned long n_rp_qs_pending;
315 unsigned long n_rp_report_qs;
316 unsigned long n_rp_cb_ready;
317 unsigned long n_rp_cpu_needs_gp;
318 unsigned long n_rp_gp_completed;
319 unsigned long n_rp_gp_started;
320 unsigned long n_rp_nocb_defer_wakeup;
321 unsigned long n_rp_need_nothing;
322
323 /* 6) _rcu_barrier() and OOM callbacks. */
324 struct rcu_head barrier_head;
325#ifdef CONFIG_RCU_FAST_NO_HZ
326 struct rcu_head oom_head;
327#endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
328
329 /* 7) Callback offloading. */
330#ifdef CONFIG_RCU_NOCB_CPU
331 struct rcu_head *nocb_head; /* CBs waiting for kthread. */
332 struct rcu_head **nocb_tail;
333 atomic_long_t nocb_q_count; /* # CBs waiting for kthread */
334 atomic_long_t nocb_q_count_lazy; /* (approximate). */
335 int nocb_p_count; /* # CBs being invoked by kthread */
336 int nocb_p_count_lazy; /* (approximate). */
337 wait_queue_head_t nocb_wq; /* For nocb kthreads to sleep on. */
338 struct task_struct *nocb_kthread;
339 bool nocb_defer_wakeup; /* Defer wakeup of nocb_kthread. */
340#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
341
342 /* 8) RCU CPU stall data. */
343#ifdef CONFIG_RCU_CPU_STALL_INFO
344 unsigned int softirq_snap; /* Snapshot of softirq activity. */
345#endif /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
346
347 int cpu;
348 struct rcu_state *rsp;
349};
350
351/* Values for fqs_state field in struct rcu_state. */
352#define RCU_GP_IDLE 0 /* No grace period in progress. */
353#define RCU_GP_INIT 1 /* Grace period being initialized. */
354#define RCU_SAVE_DYNTICK 2 /* Need to scan dyntick state. */
355#define RCU_FORCE_QS 3 /* Need to force quiescent state. */
356#define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
357
358#define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500))
359 /* For jiffies_till_first_fqs and */
360 /* and jiffies_till_next_fqs. */
361
362#define RCU_JIFFIES_FQS_DIV 256 /* Very large systems need more */
363 /* delay between bouts of */
364 /* quiescent-state forcing. */
365
366#define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time to take */
367 /* at least one scheduling clock */
368 /* irq before ratting on them. */
369
370#define rcu_wait(cond) \
371do { \
372 for (;;) { \
373 set_current_state(TASK_INTERRUPTIBLE); \
374 if (cond) \
375 break; \
376 schedule(); \
377 } \
378 __set_current_state(TASK_RUNNING); \
379} while (0)
380
381/*
382 * RCU global state, including node hierarchy. This hierarchy is
383 * represented in "heap" form in a dense array. The root (first level)
384 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
385 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
386 * and the third level in ->node[m+1] and following (->node[m+1] referenced
387 * by ->level[2]). The number of levels is determined by the number of
388 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
389 * consisting of a single rcu_node.
390 */
391struct rcu_state {
392 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
393 struct rcu_node *level[RCU_NUM_LVLS]; /* Hierarchy levels. */
394 u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */
395 u8 levelspread[RCU_NUM_LVLS]; /* kids/node in each level. */
396 struct rcu_data __percpu *rda; /* pointer of percu rcu_data. */
397 void (*call)(struct rcu_head *head, /* call_rcu() flavor. */
398 void (*func)(struct rcu_head *head));
399
400 /* The following fields are guarded by the root rcu_node's lock. */
401
402 u8 fqs_state ____cacheline_internodealigned_in_smp;
403 /* Force QS state. */
404 u8 boost; /* Subject to priority boost. */
405 unsigned long gpnum; /* Current gp number. */
406 unsigned long completed; /* # of last completed gp. */
407 struct task_struct *gp_kthread; /* Task for grace periods. */
408 wait_queue_head_t gp_wq; /* Where GP task waits. */
409 int gp_flags; /* Commands for GP task. */
410
411 /* End of fields guarded by root rcu_node's lock. */
412
413 raw_spinlock_t orphan_lock ____cacheline_internodealigned_in_smp;
414 /* Protect following fields. */
415 struct rcu_head *orphan_nxtlist; /* Orphaned callbacks that */
416 /* need a grace period. */
417 struct rcu_head **orphan_nxttail; /* Tail of above. */
418 struct rcu_head *orphan_donelist; /* Orphaned callbacks that */
419 /* are ready to invoke. */
420 struct rcu_head **orphan_donetail; /* Tail of above. */
421 long qlen_lazy; /* Number of lazy callbacks. */
422 long qlen; /* Total number of callbacks. */
423 /* End of fields guarded by orphan_lock. */
424
425 struct mutex onoff_mutex; /* Coordinate hotplug & GPs. */
426
427 struct mutex barrier_mutex; /* Guards barrier fields. */
428 atomic_t barrier_cpu_count; /* # CPUs waiting on. */
429 struct completion barrier_completion; /* Wake at barrier end. */
430 unsigned long n_barrier_done; /* ++ at start and end of */
431 /* _rcu_barrier(). */
432 /* End of fields guarded by barrier_mutex. */
433
434 atomic_long_t expedited_start; /* Starting ticket. */
435 atomic_long_t expedited_done; /* Done ticket. */
436 atomic_long_t expedited_wrap; /* # near-wrap incidents. */
437 atomic_long_t expedited_tryfail; /* # acquisition failures. */
438 atomic_long_t expedited_workdone1; /* # done by others #1. */
439 atomic_long_t expedited_workdone2; /* # done by others #2. */
440 atomic_long_t expedited_normal; /* # fallbacks to normal. */
441 atomic_long_t expedited_stoppedcpus; /* # successful stop_cpus. */
442 atomic_long_t expedited_done_tries; /* # tries to update _done. */
443 atomic_long_t expedited_done_lost; /* # times beaten to _done. */
444 atomic_long_t expedited_done_exit; /* # times exited _done loop. */
445
446 unsigned long jiffies_force_qs; /* Time at which to invoke */
447 /* force_quiescent_state(). */
448 unsigned long n_force_qs; /* Number of calls to */
449 /* force_quiescent_state(). */
450 unsigned long n_force_qs_lh; /* ~Number of calls leaving */
451 /* due to lock unavailable. */
452 unsigned long n_force_qs_ngp; /* Number of calls leaving */
453 /* due to no GP active. */
454 unsigned long gp_start; /* Time at which GP started, */
455 /* but in jiffies. */
456 unsigned long jiffies_stall; /* Time at which to check */
457 /* for CPU stalls. */
458 unsigned long jiffies_resched; /* Time at which to resched */
459 /* a reluctant CPU. */
460 unsigned long gp_max; /* Maximum GP duration in */
461 /* jiffies. */
462 const char *name; /* Name of structure. */
463 char abbr; /* Abbreviated name. */
464 struct list_head flavors; /* List of RCU flavors. */
465 struct irq_work wakeup_work; /* Postponed wakeups */
466};
467
468/* Values for rcu_state structure's gp_flags field. */
469#define RCU_GP_FLAG_INIT 0x1 /* Need grace-period initialization. */
470#define RCU_GP_FLAG_FQS 0x2 /* Need grace-period quiescent-state forcing. */
471
472extern struct list_head rcu_struct_flavors;
473
474/* Sequence through rcu_state structures for each RCU flavor. */
475#define for_each_rcu_flavor(rsp) \
476 list_for_each_entry((rsp), &rcu_struct_flavors, flavors)
477
478/* Return values for rcu_preempt_offline_tasks(). */
479
480#define RCU_OFL_TASKS_NORM_GP 0x1 /* Tasks blocking normal */
481 /* GP were moved to root. */
482#define RCU_OFL_TASKS_EXP_GP 0x2 /* Tasks blocking expedited */
483 /* GP were moved to root. */
484
485/*
486 * RCU implementation internal declarations:
487 */
488extern struct rcu_state rcu_sched_state;
489DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
490
491extern struct rcu_state rcu_bh_state;
492DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
493
494#ifdef CONFIG_TREE_PREEMPT_RCU
495extern struct rcu_state rcu_preempt_state;
496DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
497#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
498
499#ifdef CONFIG_RCU_BOOST
500DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
501DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu);
502DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
503DECLARE_PER_CPU(char, rcu_cpu_has_work);
504#endif /* #ifdef CONFIG_RCU_BOOST */
505
506#ifndef RCU_TREE_NONCORE
507
508/* Forward declarations for rcutree_plugin.h */
509static void rcu_bootup_announce(void);
510long rcu_batches_completed(void);
511static void rcu_preempt_note_context_switch(int cpu);
512static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
513#ifdef CONFIG_HOTPLUG_CPU
514static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp,
515 unsigned long flags);
516#endif /* #ifdef CONFIG_HOTPLUG_CPU */
517static void rcu_print_detail_task_stall(struct rcu_state *rsp);
518static int rcu_print_task_stall(struct rcu_node *rnp);
519static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
520#ifdef CONFIG_HOTPLUG_CPU
521static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
522 struct rcu_node *rnp,
523 struct rcu_data *rdp);
524#endif /* #ifdef CONFIG_HOTPLUG_CPU */
525static void rcu_preempt_check_callbacks(int cpu);
526void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
527#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU)
528static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
529 bool wake);
530#endif /* #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) */
531static void __init __rcu_init_preempt(void);
532static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
533static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
534static void invoke_rcu_callbacks_kthread(void);
535static bool rcu_is_callbacks_kthread(void);
536#ifdef CONFIG_RCU_BOOST
537static void rcu_preempt_do_callbacks(void);
538static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
539 struct rcu_node *rnp);
540#endif /* #ifdef CONFIG_RCU_BOOST */
541static void rcu_prepare_kthreads(int cpu);
542static void rcu_cleanup_after_idle(int cpu);
543static void rcu_prepare_for_idle(int cpu);
544static void rcu_idle_count_callbacks_posted(void);
545static void print_cpu_stall_info_begin(void);
546static void print_cpu_stall_info(struct rcu_state *rsp, int cpu);
547static void print_cpu_stall_info_end(void);
548static void zero_cpu_stall_ticks(struct rcu_data *rdp);
549static void increment_cpu_stall_ticks(void);
550static int rcu_nocb_needs_gp(struct rcu_state *rsp);
551static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq);
552static void rcu_nocb_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp);
553static void rcu_init_one_nocb(struct rcu_node *rnp);
554static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
555 bool lazy, unsigned long flags);
556static bool rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp,
557 struct rcu_data *rdp,
558 unsigned long flags);
559static bool rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp);
560static void do_nocb_deferred_wakeup(struct rcu_data *rdp);
561static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp);
562static void rcu_spawn_nocb_kthreads(struct rcu_state *rsp);
563static void rcu_kick_nohz_cpu(int cpu);
564static bool init_nocb_callback_list(struct rcu_data *rdp);
565static void rcu_sysidle_enter(struct rcu_dynticks *rdtp, int irq);
566static void rcu_sysidle_exit(struct rcu_dynticks *rdtp, int irq);
567static void rcu_sysidle_check_cpu(struct rcu_data *rdp, bool *isidle,
568 unsigned long *maxj);
569static bool is_sysidle_rcu_state(struct rcu_state *rsp);
570static void rcu_sysidle_report_gp(struct rcu_state *rsp, int isidle,
571 unsigned long maxj);
572static void rcu_bind_gp_kthread(void);
573static void rcu_sysidle_init_percpu_data(struct rcu_dynticks *rdtp);
574static bool rcu_nohz_full_cpu(struct rcu_state *rsp);
575
576#endif /* #ifndef RCU_TREE_NONCORE */
577
578#ifdef CONFIG_RCU_TRACE
579#ifdef CONFIG_RCU_NOCB_CPU
580/* Sum up queue lengths for tracing. */
581static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll)
582{
583 *ql = atomic_long_read(&rdp->nocb_q_count) + rdp->nocb_p_count;
584 *qll = atomic_long_read(&rdp->nocb_q_count_lazy) + rdp->nocb_p_count_lazy;
585}
586#else /* #ifdef CONFIG_RCU_NOCB_CPU */
587static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll)
588{
589 *ql = 0;
590 *qll = 0;
591}
592#endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
593#endif /* #ifdef CONFIG_RCU_TRACE */
1/* SPDX-License-Identifier: GPL-2.0+ */
2/*
3 * Read-Copy Update mechanism for mutual exclusion (tree-based version)
4 * Internal non-public definitions.
5 *
6 * Copyright IBM Corporation, 2008
7 *
8 * Author: Ingo Molnar <mingo@elte.hu>
9 * Paul E. McKenney <paulmck@linux.ibm.com>
10 */
11
12#include <linux/cache.h>
13#include <linux/kthread.h>
14#include <linux/spinlock.h>
15#include <linux/rtmutex.h>
16#include <linux/threads.h>
17#include <linux/cpumask.h>
18#include <linux/seqlock.h>
19#include <linux/swait.h>
20#include <linux/rcu_node_tree.h>
21
22#include "rcu_segcblist.h"
23
24/* Communicate arguments to a kthread worker handler. */
25struct rcu_exp_work {
26 unsigned long rew_s;
27 struct kthread_work rew_work;
28};
29
30/* RCU's kthread states for tracing. */
31#define RCU_KTHREAD_STOPPED 0
32#define RCU_KTHREAD_RUNNING 1
33#define RCU_KTHREAD_WAITING 2
34#define RCU_KTHREAD_OFFCPU 3
35#define RCU_KTHREAD_YIELDING 4
36#define RCU_KTHREAD_MAX 4
37
38/*
39 * Definition for node within the RCU grace-period-detection hierarchy.
40 */
41struct rcu_node {
42 raw_spinlock_t __private lock; /* Root rcu_node's lock protects */
43 /* some rcu_state fields as well as */
44 /* following. */
45 unsigned long gp_seq; /* Track rsp->gp_seq. */
46 unsigned long gp_seq_needed; /* Track furthest future GP request. */
47 unsigned long completedqs; /* All QSes done for this node. */
48 unsigned long qsmask; /* CPUs or groups that need to switch in */
49 /* order for current grace period to proceed.*/
50 /* In leaf rcu_node, each bit corresponds to */
51 /* an rcu_data structure, otherwise, each */
52 /* bit corresponds to a child rcu_node */
53 /* structure. */
54 unsigned long rcu_gp_init_mask; /* Mask of offline CPUs at GP init. */
55 unsigned long qsmaskinit;
56 /* Per-GP initial value for qsmask. */
57 /* Initialized from ->qsmaskinitnext at the */
58 /* beginning of each grace period. */
59 unsigned long qsmaskinitnext;
60 unsigned long expmask; /* CPUs or groups that need to check in */
61 /* to allow the current expedited GP */
62 /* to complete. */
63 unsigned long expmaskinit;
64 /* Per-GP initial values for expmask. */
65 /* Initialized from ->expmaskinitnext at the */
66 /* beginning of each expedited GP. */
67 unsigned long expmaskinitnext;
68 /* Online CPUs for next expedited GP. */
69 /* Any CPU that has ever been online will */
70 /* have its bit set. */
71 struct kthread_worker *exp_kworker;
72 /* Workers performing per node expedited GP */
73 /* initialization. */
74 unsigned long cbovldmask;
75 /* CPUs experiencing callback overload. */
76 unsigned long ffmask; /* Fully functional CPUs. */
77 unsigned long grpmask; /* Mask to apply to parent qsmask. */
78 /* Only one bit will be set in this mask. */
79 int grplo; /* lowest-numbered CPU here. */
80 int grphi; /* highest-numbered CPU here. */
81 u8 grpnum; /* group number for next level up. */
82 u8 level; /* root is at level 0. */
83 bool wait_blkd_tasks;/* Necessary to wait for blocked tasks to */
84 /* exit RCU read-side critical sections */
85 /* before propagating offline up the */
86 /* rcu_node tree? */
87 struct rcu_node *parent;
88 struct list_head blkd_tasks;
89 /* Tasks blocked in RCU read-side critical */
90 /* section. Tasks are placed at the head */
91 /* of this list and age towards the tail. */
92 struct list_head *gp_tasks;
93 /* Pointer to the first task blocking the */
94 /* current grace period, or NULL if there */
95 /* is no such task. */
96 struct list_head *exp_tasks;
97 /* Pointer to the first task blocking the */
98 /* current expedited grace period, or NULL */
99 /* if there is no such task. If there */
100 /* is no current expedited grace period, */
101 /* then there can cannot be any such task. */
102 struct list_head *boost_tasks;
103 /* Pointer to first task that needs to be */
104 /* priority boosted, or NULL if no priority */
105 /* boosting is needed for this rcu_node */
106 /* structure. If there are no tasks */
107 /* queued on this rcu_node structure that */
108 /* are blocking the current grace period, */
109 /* there can be no such task. */
110 struct rt_mutex boost_mtx;
111 /* Used only for the priority-boosting */
112 /* side effect, not as a lock. */
113 unsigned long boost_time;
114 /* When to start boosting (jiffies). */
115 struct mutex kthread_mutex;
116 /* Exclusion for thread spawning and affinity */
117 /* manipulation. */
118 struct task_struct *boost_kthread_task;
119 /* kthread that takes care of priority */
120 /* boosting for this rcu_node structure. */
121 unsigned int boost_kthread_status;
122 /* State of boost_kthread_task for tracing. */
123 unsigned long n_boosts; /* Number of boosts for this rcu_node structure. */
124#ifdef CONFIG_RCU_NOCB_CPU
125 struct swait_queue_head nocb_gp_wq[2];
126 /* Place for rcu_nocb_kthread() to wait GP. */
127#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
128 raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp;
129
130 spinlock_t exp_lock ____cacheline_internodealigned_in_smp;
131 unsigned long exp_seq_rq;
132 wait_queue_head_t exp_wq[4];
133 struct rcu_exp_work rew;
134 bool exp_need_flush; /* Need to flush workitem? */
135 raw_spinlock_t exp_poll_lock;
136 /* Lock and data for polled expedited grace periods. */
137 unsigned long exp_seq_poll_rq;
138 struct work_struct exp_poll_wq;
139} ____cacheline_internodealigned_in_smp;
140
141/*
142 * Bitmasks in an rcu_node cover the interval [grplo, grphi] of CPU IDs, and
143 * are indexed relative to this interval rather than the global CPU ID space.
144 * This generates the bit for a CPU in node-local masks.
145 */
146#define leaf_node_cpu_bit(rnp, cpu) (BIT((cpu) - (rnp)->grplo))
147
148/*
149 * Union to allow "aggregate OR" operation on the need for a quiescent
150 * state by the normal and expedited grace periods.
151 */
152union rcu_noqs {
153 struct {
154 u8 norm;
155 u8 exp;
156 } b; /* Bits. */
157 u16 s; /* Set of bits, aggregate OR here. */
158};
159
160/*
161 * Record the snapshot of the core stats at half of the first RCU stall timeout.
162 * The member gp_seq is used to ensure that all members are updated only once
163 * during the sampling period. The snapshot is taken only if this gp_seq is not
164 * equal to rdp->gp_seq.
165 */
166struct rcu_snap_record {
167 unsigned long gp_seq; /* Track rdp->gp_seq counter */
168 u64 cputime_irq; /* Accumulated cputime of hard irqs */
169 u64 cputime_softirq;/* Accumulated cputime of soft irqs */
170 u64 cputime_system; /* Accumulated cputime of kernel tasks */
171 unsigned long nr_hardirqs; /* Accumulated number of hard irqs */
172 unsigned int nr_softirqs; /* Accumulated number of soft irqs */
173 unsigned long long nr_csw; /* Accumulated number of task switches */
174 unsigned long jiffies; /* Track jiffies value */
175};
176
177/* Per-CPU data for read-copy update. */
178struct rcu_data {
179 /* 1) quiescent-state and grace-period handling : */
180 unsigned long gp_seq; /* Track rsp->gp_seq counter. */
181 unsigned long gp_seq_needed; /* Track furthest future GP request. */
182 union rcu_noqs cpu_no_qs; /* No QSes yet for this CPU. */
183 bool core_needs_qs; /* Core waits for quiescent state. */
184 bool beenonline; /* CPU online at least once. */
185 bool gpwrap; /* Possible ->gp_seq wrap. */
186 bool cpu_started; /* RCU watching this onlining CPU. */
187 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
188 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
189 unsigned long ticks_this_gp; /* The number of scheduling-clock */
190 /* ticks this CPU has handled */
191 /* during and after the last grace */
192 /* period it is aware of. */
193 struct irq_work defer_qs_iw; /* Obtain later scheduler attention. */
194 bool defer_qs_iw_pending; /* Scheduler attention pending? */
195 struct work_struct strict_work; /* Schedule readers for strict GPs. */
196
197 /* 2) batch handling */
198 struct rcu_segcblist cblist; /* Segmented callback list, with */
199 /* different callbacks waiting for */
200 /* different grace periods. */
201 long qlen_last_fqs_check;
202 /* qlen at last check for QS forcing */
203 unsigned long n_cbs_invoked; /* # callbacks invoked since boot. */
204 unsigned long n_force_qs_snap;
205 /* did other CPU force QS recently? */
206 long blimit; /* Upper limit on a processed batch */
207
208 /* 3) dynticks interface. */
209 int watching_snap; /* Per-GP tracking for dynticks. */
210 bool rcu_need_heavy_qs; /* GP old, so heavy quiescent state! */
211 bool rcu_urgent_qs; /* GP old need light quiescent state. */
212 bool rcu_forced_tick; /* Forced tick to provide QS. */
213 bool rcu_forced_tick_exp; /* ... provide QS to expedited GP. */
214
215 /* 4) rcu_barrier(), OOM callbacks, and expediting. */
216 unsigned long barrier_seq_snap; /* Snap of rcu_state.barrier_sequence. */
217 struct rcu_head barrier_head;
218 int exp_watching_snap; /* Double-check need for IPI. */
219
220 /* 5) Callback offloading. */
221#ifdef CONFIG_RCU_NOCB_CPU
222 struct swait_queue_head nocb_cb_wq; /* For nocb kthreads to sleep on. */
223 struct swait_queue_head nocb_state_wq; /* For offloading state changes */
224 struct task_struct *nocb_gp_kthread;
225 raw_spinlock_t nocb_lock; /* Guard following pair of fields. */
226 int nocb_defer_wakeup; /* Defer wakeup of nocb_kthread. */
227 struct timer_list nocb_timer; /* Enforce finite deferral. */
228 unsigned long nocb_gp_adv_time; /* Last call_rcu() CB adv (jiffies). */
229 struct mutex nocb_gp_kthread_mutex; /* Exclusion for nocb gp kthread */
230 /* spawning */
231
232 /* The following fields are used by call_rcu, hence own cacheline. */
233 raw_spinlock_t nocb_bypass_lock ____cacheline_internodealigned_in_smp;
234 struct rcu_cblist nocb_bypass; /* Lock-contention-bypass CB list. */
235 unsigned long nocb_bypass_first; /* Time (jiffies) of first enqueue. */
236 unsigned long nocb_nobypass_last; /* Last ->cblist enqueue (jiffies). */
237 int nocb_nobypass_count; /* # ->cblist enqueues at ^^^ time. */
238
239 /* The following fields are used by GP kthread, hence own cacheline. */
240 raw_spinlock_t nocb_gp_lock ____cacheline_internodealigned_in_smp;
241 u8 nocb_gp_sleep; /* Is the nocb GP thread asleep? */
242 u8 nocb_gp_bypass; /* Found a bypass on last scan? */
243 u8 nocb_gp_gp; /* GP to wait for on last scan? */
244 unsigned long nocb_gp_seq; /* If so, ->gp_seq to wait for. */
245 unsigned long nocb_gp_loops; /* # passes through wait code. */
246 struct swait_queue_head nocb_gp_wq; /* For nocb kthreads to sleep on. */
247 bool nocb_cb_sleep; /* Is the nocb CB thread asleep? */
248 struct task_struct *nocb_cb_kthread;
249 struct list_head nocb_head_rdp; /*
250 * Head of rcu_data list in wakeup chain,
251 * if rdp_gp.
252 */
253 struct list_head nocb_entry_rdp; /* rcu_data node in wakeup chain. */
254 struct rcu_data *nocb_toggling_rdp; /* rdp queued for (de-)offloading */
255
256 /* The following fields are used by CB kthread, hence new cacheline. */
257 struct rcu_data *nocb_gp_rdp ____cacheline_internodealigned_in_smp;
258 /* GP rdp takes GP-end wakeups. */
259#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
260
261 /* 6) RCU priority boosting. */
262 struct task_struct *rcu_cpu_kthread_task;
263 /* rcuc per-CPU kthread or NULL. */
264 unsigned int rcu_cpu_kthread_status;
265 char rcu_cpu_has_work;
266 unsigned long rcuc_activity;
267
268 /* 7) Diagnostic data, including RCU CPU stall warnings. */
269 unsigned int softirq_snap; /* Snapshot of softirq activity. */
270 /* ->rcu_iw* fields protected by leaf rcu_node ->lock. */
271 struct irq_work rcu_iw; /* Check for non-irq activity. */
272 bool rcu_iw_pending; /* Is ->rcu_iw pending? */
273 unsigned long rcu_iw_gp_seq; /* ->gp_seq associated with ->rcu_iw. */
274 unsigned long rcu_ofl_gp_seq; /* ->gp_seq at last offline. */
275 short rcu_ofl_gp_state; /* ->gp_state at last offline. */
276 unsigned long rcu_onl_gp_seq; /* ->gp_seq at last online. */
277 short rcu_onl_gp_state; /* ->gp_state at last online. */
278 unsigned long last_fqs_resched; /* Time of last rcu_resched(). */
279 unsigned long last_sched_clock; /* Jiffies of last rcu_sched_clock_irq(). */
280 struct rcu_snap_record snap_record; /* Snapshot of core stats at half of */
281 /* the first RCU stall timeout */
282
283 long lazy_len; /* Length of buffered lazy callbacks. */
284 int cpu;
285};
286
287/* Values for nocb_defer_wakeup field in struct rcu_data. */
288#define RCU_NOCB_WAKE_NOT 0
289#define RCU_NOCB_WAKE_BYPASS 1
290#define RCU_NOCB_WAKE_LAZY 2
291#define RCU_NOCB_WAKE 3
292#define RCU_NOCB_WAKE_FORCE 4
293
294#define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500))
295 /* For jiffies_till_first_fqs and */
296 /* and jiffies_till_next_fqs. */
297
298#define RCU_JIFFIES_FQS_DIV 256 /* Very large systems need more */
299 /* delay between bouts of */
300 /* quiescent-state forcing. */
301
302#define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time to take */
303 /* at least one scheduling clock */
304 /* irq before ratting on them. */
305
306#define rcu_wait(cond) \
307do { \
308 for (;;) { \
309 set_current_state(TASK_INTERRUPTIBLE); \
310 if (cond) \
311 break; \
312 schedule(); \
313 } \
314 __set_current_state(TASK_RUNNING); \
315} while (0)
316
317/*
318 * A max threshold for synchronize_rcu() users which are
319 * awaken directly by the rcu_gp_kthread(). Left part is
320 * deferred to the main worker.
321 */
322#define SR_MAX_USERS_WAKE_FROM_GP 5
323#define SR_NORMAL_GP_WAIT_HEAD_MAX 5
324
325struct sr_wait_node {
326 atomic_t inuse;
327 struct llist_node node;
328};
329
330/*
331 * RCU global state, including node hierarchy. This hierarchy is
332 * represented in "heap" form in a dense array. The root (first level)
333 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
334 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
335 * and the third level in ->node[m+1] and following (->node[m+1] referenced
336 * by ->level[2]). The number of levels is determined by the number of
337 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
338 * consisting of a single rcu_node.
339 */
340struct rcu_state {
341 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
342 struct rcu_node *level[RCU_NUM_LVLS + 1];
343 /* Hierarchy levels (+1 to */
344 /* shut bogus gcc warning) */
345 int ncpus; /* # CPUs seen so far. */
346 int n_online_cpus; /* # CPUs online for RCU. */
347
348 /* The following fields are guarded by the root rcu_node's lock. */
349
350 unsigned long gp_seq ____cacheline_internodealigned_in_smp;
351 /* Grace-period sequence #. */
352 unsigned long gp_max; /* Maximum GP duration in */
353 /* jiffies. */
354 struct task_struct *gp_kthread; /* Task for grace periods. */
355 struct swait_queue_head gp_wq; /* Where GP task waits. */
356 short gp_flags; /* Commands for GP task. */
357 short gp_state; /* GP kthread sleep state. */
358 unsigned long gp_wake_time; /* Last GP kthread wake. */
359 unsigned long gp_wake_seq; /* ->gp_seq at ^^^. */
360 unsigned long gp_seq_polled; /* GP seq for polled API. */
361 unsigned long gp_seq_polled_snap; /* ->gp_seq_polled at normal GP start. */
362 unsigned long gp_seq_polled_exp_snap; /* ->gp_seq_polled at expedited GP start. */
363
364 /* End of fields guarded by root rcu_node's lock. */
365
366 struct mutex barrier_mutex; /* Guards barrier fields. */
367 atomic_t barrier_cpu_count; /* # CPUs waiting on. */
368 struct completion barrier_completion; /* Wake at barrier end. */
369 unsigned long barrier_sequence; /* ++ at start and end of */
370 /* rcu_barrier(). */
371 /* End of fields guarded by barrier_mutex. */
372
373 raw_spinlock_t barrier_lock; /* Protects ->barrier_seq_snap. */
374
375 struct mutex exp_mutex; /* Serialize expedited GP. */
376 struct mutex exp_wake_mutex; /* Serialize wakeup. */
377 unsigned long expedited_sequence; /* Take a ticket. */
378 atomic_t expedited_need_qs; /* # CPUs left to check in. */
379 struct swait_queue_head expedited_wq; /* Wait for check-ins. */
380 int ncpus_snap; /* # CPUs seen last time. */
381 u8 cbovld; /* Callback overload now? */
382 u8 cbovldnext; /* ^ ^ next time? */
383
384 unsigned long jiffies_force_qs; /* Time at which to invoke */
385 /* force_quiescent_state(). */
386 unsigned long jiffies_kick_kthreads; /* Time at which to kick */
387 /* kthreads, if configured. */
388 unsigned long n_force_qs; /* Number of calls to */
389 /* force_quiescent_state(). */
390 unsigned long gp_start; /* Time at which GP started, */
391 /* but in jiffies. */
392 unsigned long gp_end; /* Time last GP ended, again */
393 /* in jiffies. */
394 unsigned long gp_activity; /* Time of last GP kthread */
395 /* activity in jiffies. */
396 unsigned long gp_req_activity; /* Time of last GP request */
397 /* in jiffies. */
398 unsigned long jiffies_stall; /* Time at which to check */
399 /* for CPU stalls. */
400 int nr_fqs_jiffies_stall; /* Number of fqs loops after
401 * which read jiffies and set
402 * jiffies_stall. Stall
403 * warnings disabled if !0. */
404 unsigned long jiffies_resched; /* Time at which to resched */
405 /* a reluctant CPU. */
406 unsigned long n_force_qs_gpstart; /* Snapshot of n_force_qs at */
407 /* GP start. */
408 const char *name; /* Name of structure. */
409 char abbr; /* Abbreviated name. */
410
411 arch_spinlock_t ofl_lock ____cacheline_internodealigned_in_smp;
412 /* Synchronize offline with */
413 /* GP pre-initialization. */
414
415 /* synchronize_rcu() part. */
416 struct llist_head srs_next; /* request a GP users. */
417 struct llist_node *srs_wait_tail; /* wait for GP users. */
418 struct llist_node *srs_done_tail; /* ready for GP users. */
419 struct sr_wait_node srs_wait_nodes[SR_NORMAL_GP_WAIT_HEAD_MAX];
420 struct work_struct srs_cleanup_work;
421 atomic_t srs_cleanups_pending; /* srs inflight worker cleanups. */
422
423#ifdef CONFIG_RCU_NOCB_CPU
424 struct mutex nocb_mutex; /* Guards (de-)offloading */
425 int nocb_is_setup; /* nocb is setup from boot */
426#endif
427};
428
429/* Values for rcu_state structure's gp_flags field. */
430#define RCU_GP_FLAG_INIT 0x1 /* Need grace-period initialization. */
431#define RCU_GP_FLAG_FQS 0x2 /* Need grace-period quiescent-state forcing. */
432#define RCU_GP_FLAG_OVLD 0x4 /* Experiencing callback overload. */
433
434/* Values for rcu_state structure's gp_state field. */
435#define RCU_GP_IDLE 0 /* Initial state and no GP in progress. */
436#define RCU_GP_WAIT_GPS 1 /* Wait for grace-period start. */
437#define RCU_GP_DONE_GPS 2 /* Wait done for grace-period start. */
438#define RCU_GP_ONOFF 3 /* Grace-period initialization hotplug. */
439#define RCU_GP_INIT 4 /* Grace-period initialization. */
440#define RCU_GP_WAIT_FQS 5 /* Wait for force-quiescent-state time. */
441#define RCU_GP_DOING_FQS 6 /* Wait done for force-quiescent-state time. */
442#define RCU_GP_CLEANUP 7 /* Grace-period cleanup started. */
443#define RCU_GP_CLEANED 8 /* Grace-period cleanup complete. */
444
445/*
446 * In order to export the rcu_state name to the tracing tools, it
447 * needs to be added in the __tracepoint_string section.
448 * This requires defining a separate variable tp_<sname>_varname
449 * that points to the string being used, and this will allow
450 * the tracing userspace tools to be able to decipher the string
451 * address to the matching string.
452 */
453#ifdef CONFIG_PREEMPT_RCU
454#define RCU_ABBR 'p'
455#define RCU_NAME_RAW "rcu_preempt"
456#else /* #ifdef CONFIG_PREEMPT_RCU */
457#define RCU_ABBR 's'
458#define RCU_NAME_RAW "rcu_sched"
459#endif /* #else #ifdef CONFIG_PREEMPT_RCU */
460#ifndef CONFIG_TRACING
461#define RCU_NAME RCU_NAME_RAW
462#else /* #ifdef CONFIG_TRACING */
463static char rcu_name[] = RCU_NAME_RAW;
464static const char *tp_rcu_varname __used __tracepoint_string = rcu_name;
465#define RCU_NAME rcu_name
466#endif /* #else #ifdef CONFIG_TRACING */
467
468/* Forward declarations for tree_plugin.h */
469static void rcu_bootup_announce(void);
470static void rcu_qs(void);
471static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
472#ifdef CONFIG_HOTPLUG_CPU
473static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
474#endif /* #ifdef CONFIG_HOTPLUG_CPU */
475static int rcu_print_task_exp_stall(struct rcu_node *rnp);
476static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
477static void rcu_flavor_sched_clock_irq(int user);
478static void dump_blkd_tasks(struct rcu_node *rnp, int ncheck);
479static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
480static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
481static bool rcu_is_callbacks_kthread(struct rcu_data *rdp);
482static void rcu_cpu_kthread_setup(unsigned int cpu);
483static void rcu_spawn_one_boost_kthread(struct rcu_node *rnp);
484static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
485static bool rcu_preempt_need_deferred_qs(struct task_struct *t);
486static void zero_cpu_stall_ticks(struct rcu_data *rdp);
487static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp);
488static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq);
489static void rcu_init_one_nocb(struct rcu_node *rnp);
490static bool wake_nocb_gp(struct rcu_data *rdp, bool force);
491static bool rcu_nocb_flush_bypass(struct rcu_data *rdp, struct rcu_head *rhp,
492 unsigned long j, bool lazy);
493static void call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *head,
494 rcu_callback_t func, unsigned long flags, bool lazy);
495static void __maybe_unused __call_rcu_nocb_wake(struct rcu_data *rdp, bool was_empty,
496 unsigned long flags);
497static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp, int level);
498static bool do_nocb_deferred_wakeup(struct rcu_data *rdp);
499static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp);
500static void rcu_spawn_cpu_nocb_kthread(int cpu);
501static void show_rcu_nocb_state(struct rcu_data *rdp);
502static void rcu_nocb_lock(struct rcu_data *rdp);
503static void rcu_nocb_unlock(struct rcu_data *rdp);
504static void rcu_nocb_unlock_irqrestore(struct rcu_data *rdp,
505 unsigned long flags);
506static void rcu_lockdep_assert_cblist_protected(struct rcu_data *rdp);
507#ifdef CONFIG_RCU_NOCB_CPU
508static void __init rcu_organize_nocb_kthreads(void);
509
510/*
511 * Disable IRQs before checking offloaded state so that local
512 * locking is safe against concurrent de-offloading.
513 */
514#define rcu_nocb_lock_irqsave(rdp, flags) \
515do { \
516 local_irq_save(flags); \
517 if (rcu_segcblist_is_offloaded(&(rdp)->cblist)) \
518 raw_spin_lock(&(rdp)->nocb_lock); \
519} while (0)
520#else /* #ifdef CONFIG_RCU_NOCB_CPU */
521#define rcu_nocb_lock_irqsave(rdp, flags) local_irq_save(flags)
522#endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
523
524static void rcu_bind_gp_kthread(void);
525static bool rcu_nohz_full_cpu(void);
526
527/* Forward declarations for tree_stall.h */
528static void record_gp_stall_check_time(void);
529static void rcu_iw_handler(struct irq_work *iwp);
530static void check_cpu_stall(struct rcu_data *rdp);
531static void rcu_check_gp_start_stall(struct rcu_node *rnp, struct rcu_data *rdp,
532 const unsigned long gpssdelay);
533
534/* Forward declarations for tree_exp.h. */
535static void sync_rcu_do_polled_gp(struct work_struct *wp);