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/swait.h>
31#include <linux/stop_machine.h>
32
33/*
34 * Define shape of hierarchy based on NR_CPUS, CONFIG_RCU_FANOUT, and
35 * CONFIG_RCU_FANOUT_LEAF.
36 * In theory, it should be possible to add more levels straightforwardly.
37 * In practice, this did work well going from three levels to four.
38 * Of course, your mileage may vary.
39 */
40
41#ifdef CONFIG_RCU_FANOUT
42#define RCU_FANOUT CONFIG_RCU_FANOUT
43#else /* #ifdef CONFIG_RCU_FANOUT */
44# ifdef CONFIG_64BIT
45# define RCU_FANOUT 64
46# else
47# define RCU_FANOUT 32
48# endif
49#endif /* #else #ifdef CONFIG_RCU_FANOUT */
50
51#ifdef CONFIG_RCU_FANOUT_LEAF
52#define RCU_FANOUT_LEAF CONFIG_RCU_FANOUT_LEAF
53#else /* #ifdef CONFIG_RCU_FANOUT_LEAF */
54# ifdef CONFIG_64BIT
55# define RCU_FANOUT_LEAF 64
56# else
57# define RCU_FANOUT_LEAF 32
58# endif
59#endif /* #else #ifdef CONFIG_RCU_FANOUT_LEAF */
60
61#define RCU_FANOUT_1 (RCU_FANOUT_LEAF)
62#define RCU_FANOUT_2 (RCU_FANOUT_1 * RCU_FANOUT)
63#define RCU_FANOUT_3 (RCU_FANOUT_2 * RCU_FANOUT)
64#define RCU_FANOUT_4 (RCU_FANOUT_3 * RCU_FANOUT)
65
66#if NR_CPUS <= RCU_FANOUT_1
67# define RCU_NUM_LVLS 1
68# define NUM_RCU_LVL_0 1
69# define NUM_RCU_NODES NUM_RCU_LVL_0
70# define NUM_RCU_LVL_INIT { NUM_RCU_LVL_0 }
71# define RCU_NODE_NAME_INIT { "rcu_node_0" }
72# define RCU_FQS_NAME_INIT { "rcu_node_fqs_0" }
73#elif NR_CPUS <= RCU_FANOUT_2
74# define RCU_NUM_LVLS 2
75# define NUM_RCU_LVL_0 1
76# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
77# define NUM_RCU_NODES (NUM_RCU_LVL_0 + NUM_RCU_LVL_1)
78# define NUM_RCU_LVL_INIT { NUM_RCU_LVL_0, NUM_RCU_LVL_1 }
79# define RCU_NODE_NAME_INIT { "rcu_node_0", "rcu_node_1" }
80# define RCU_FQS_NAME_INIT { "rcu_node_fqs_0", "rcu_node_fqs_1" }
81#elif NR_CPUS <= RCU_FANOUT_3
82# define RCU_NUM_LVLS 3
83# define NUM_RCU_LVL_0 1
84# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
85# define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
86# define NUM_RCU_NODES (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2)
87# define NUM_RCU_LVL_INIT { NUM_RCU_LVL_0, NUM_RCU_LVL_1, NUM_RCU_LVL_2 }
88# define RCU_NODE_NAME_INIT { "rcu_node_0", "rcu_node_1", "rcu_node_2" }
89# define RCU_FQS_NAME_INIT { "rcu_node_fqs_0", "rcu_node_fqs_1", "rcu_node_fqs_2" }
90#elif NR_CPUS <= RCU_FANOUT_4
91# define RCU_NUM_LVLS 4
92# define NUM_RCU_LVL_0 1
93# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3)
94# define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
95# define NUM_RCU_LVL_3 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
96# define NUM_RCU_NODES (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3)
97# define NUM_RCU_LVL_INIT { NUM_RCU_LVL_0, NUM_RCU_LVL_1, NUM_RCU_LVL_2, NUM_RCU_LVL_3 }
98# define RCU_NODE_NAME_INIT { "rcu_node_0", "rcu_node_1", "rcu_node_2", "rcu_node_3" }
99# define RCU_FQS_NAME_INIT { "rcu_node_fqs_0", "rcu_node_fqs_1", "rcu_node_fqs_2", "rcu_node_fqs_3" }
100#else
101# error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
102#endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
103
104extern int rcu_num_lvls;
105extern int rcu_num_nodes;
106
107/*
108 * Dynticks per-CPU state.
109 */
110struct rcu_dynticks {
111 long long dynticks_nesting; /* Track irq/process nesting level. */
112 /* Process level is worth LLONG_MAX/2. */
113 int dynticks_nmi_nesting; /* Track NMI nesting level. */
114 atomic_t dynticks; /* Even value for idle, else odd. */
115#ifdef CONFIG_NO_HZ_FULL_SYSIDLE
116 long long dynticks_idle_nesting;
117 /* irq/process nesting level from idle. */
118 atomic_t dynticks_idle; /* Even value for idle, else odd. */
119 /* "Idle" excludes userspace execution. */
120 unsigned long dynticks_idle_jiffies;
121 /* End of last non-NMI non-idle period. */
122#endif /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
123#ifdef CONFIG_RCU_FAST_NO_HZ
124 bool all_lazy; /* Are all CPU's CBs lazy? */
125 unsigned long nonlazy_posted;
126 /* # times non-lazy CBs posted to CPU. */
127 unsigned long nonlazy_posted_snap;
128 /* idle-period nonlazy_posted snapshot. */
129 unsigned long last_accelerate;
130 /* Last jiffy CBs were accelerated. */
131 unsigned long last_advance_all;
132 /* Last jiffy CBs were all advanced. */
133 int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */
134#endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
135};
136
137/* RCU's kthread states for tracing. */
138#define RCU_KTHREAD_STOPPED 0
139#define RCU_KTHREAD_RUNNING 1
140#define RCU_KTHREAD_WAITING 2
141#define RCU_KTHREAD_OFFCPU 3
142#define RCU_KTHREAD_YIELDING 4
143#define RCU_KTHREAD_MAX 4
144
145/*
146 * Definition for node within the RCU grace-period-detection hierarchy.
147 */
148struct rcu_node {
149 raw_spinlock_t __private lock; /* Root rcu_node's lock protects */
150 /* some rcu_state fields as well as */
151 /* following. */
152 unsigned long gpnum; /* Current grace period for this node. */
153 /* This will either be equal to or one */
154 /* behind the root rcu_node's gpnum. */
155 unsigned long completed; /* Last GP completed for this node. */
156 /* This will either be equal to or one */
157 /* behind the root rcu_node's gpnum. */
158 unsigned long qsmask; /* CPUs or groups that need to switch in */
159 /* order for current grace period to proceed.*/
160 /* In leaf rcu_node, each bit corresponds to */
161 /* an rcu_data structure, otherwise, each */
162 /* bit corresponds to a child rcu_node */
163 /* structure. */
164 unsigned long qsmaskinit;
165 /* Per-GP initial value for qsmask. */
166 /* Initialized from ->qsmaskinitnext at the */
167 /* beginning of each grace period. */
168 unsigned long qsmaskinitnext;
169 /* Online CPUs for next grace period. */
170 unsigned long expmask; /* CPUs or groups that need to check in */
171 /* to allow the current expedited GP */
172 /* to complete. */
173 unsigned long expmaskinit;
174 /* Per-GP initial values for expmask. */
175 /* Initialized from ->expmaskinitnext at the */
176 /* beginning of each expedited GP. */
177 unsigned long expmaskinitnext;
178 /* Online CPUs for next expedited GP. */
179 /* Any CPU that has ever been online will */
180 /* have its bit set. */
181 unsigned long grpmask; /* Mask to apply to parent qsmask. */
182 /* Only one bit will be set in this mask. */
183 int grplo; /* lowest-numbered CPU or group here. */
184 int grphi; /* highest-numbered CPU or group here. */
185 u8 grpnum; /* CPU/group number for next level up. */
186 u8 level; /* root is at level 0. */
187 bool wait_blkd_tasks;/* Necessary to wait for blocked tasks to */
188 /* exit RCU read-side critical sections */
189 /* before propagating offline up the */
190 /* rcu_node tree? */
191 struct rcu_node *parent;
192 struct list_head blkd_tasks;
193 /* Tasks blocked in RCU read-side critical */
194 /* section. Tasks are placed at the head */
195 /* of this list and age towards the tail. */
196 struct list_head *gp_tasks;
197 /* Pointer to the first task blocking the */
198 /* current grace period, or NULL if there */
199 /* is no such task. */
200 struct list_head *exp_tasks;
201 /* Pointer to the first task blocking the */
202 /* current expedited grace period, or NULL */
203 /* if there is no such task. If there */
204 /* is no current expedited grace period, */
205 /* then there can cannot be any such task. */
206 struct list_head *boost_tasks;
207 /* Pointer to first task that needs to be */
208 /* priority boosted, or NULL if no priority */
209 /* boosting is needed for this rcu_node */
210 /* structure. If there are no tasks */
211 /* queued on this rcu_node structure that */
212 /* are blocking the current grace period, */
213 /* there can be no such task. */
214 struct rt_mutex boost_mtx;
215 /* Used only for the priority-boosting */
216 /* side effect, not as a lock. */
217 unsigned long boost_time;
218 /* When to start boosting (jiffies). */
219 struct task_struct *boost_kthread_task;
220 /* kthread that takes care of priority */
221 /* boosting for this rcu_node structure. */
222 unsigned int boost_kthread_status;
223 /* State of boost_kthread_task for tracing. */
224 unsigned long n_tasks_boosted;
225 /* Total number of tasks boosted. */
226 unsigned long n_exp_boosts;
227 /* Number of tasks boosted for expedited GP. */
228 unsigned long n_normal_boosts;
229 /* Number of tasks boosted for normal GP. */
230 unsigned long n_balk_blkd_tasks;
231 /* Refused to boost: no blocked tasks. */
232 unsigned long n_balk_exp_gp_tasks;
233 /* Refused to boost: nothing blocking GP. */
234 unsigned long n_balk_boost_tasks;
235 /* Refused to boost: already boosting. */
236 unsigned long n_balk_notblocked;
237 /* Refused to boost: RCU RS CS still running. */
238 unsigned long n_balk_notyet;
239 /* Refused to boost: not yet time. */
240 unsigned long n_balk_nos;
241 /* Refused to boost: not sure why, though. */
242 /* This can happen due to race conditions. */
243#ifdef CONFIG_RCU_NOCB_CPU
244 struct swait_queue_head nocb_gp_wq[2];
245 /* Place for rcu_nocb_kthread() to wait GP. */
246#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
247 int need_future_gp[2];
248 /* Counts of upcoming no-CB GP requests. */
249 raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp;
250
251 spinlock_t exp_lock ____cacheline_internodealigned_in_smp;
252 unsigned long exp_seq_rq;
253 wait_queue_head_t exp_wq[4];
254} ____cacheline_internodealigned_in_smp;
255
256/*
257 * Bitmasks in an rcu_node cover the interval [grplo, grphi] of CPU IDs, and
258 * are indexed relative to this interval rather than the global CPU ID space.
259 * This generates the bit for a CPU in node-local masks.
260 */
261#define leaf_node_cpu_bit(rnp, cpu) (1UL << ((cpu) - (rnp)->grplo))
262
263/*
264 * Do a full breadth-first scan of the rcu_node structures for the
265 * specified rcu_state structure.
266 */
267#define rcu_for_each_node_breadth_first(rsp, rnp) \
268 for ((rnp) = &(rsp)->node[0]; \
269 (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
270
271/*
272 * Do a breadth-first scan of the non-leaf rcu_node structures for the
273 * specified rcu_state structure. Note that if there is a singleton
274 * rcu_node tree with but one rcu_node structure, this loop is a no-op.
275 */
276#define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
277 for ((rnp) = &(rsp)->node[0]; \
278 (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
279
280/*
281 * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
282 * structure. Note that if there is a singleton rcu_node tree with but
283 * one rcu_node structure, this loop -will- visit the rcu_node structure.
284 * It is still a leaf node, even if it is also the root node.
285 */
286#define rcu_for_each_leaf_node(rsp, rnp) \
287 for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
288 (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
289
290/*
291 * Iterate over all possible CPUs in a leaf RCU node.
292 */
293#define for_each_leaf_node_possible_cpu(rnp, cpu) \
294 for ((cpu) = cpumask_next(rnp->grplo - 1, cpu_possible_mask); \
295 cpu <= rnp->grphi; \
296 cpu = cpumask_next((cpu), cpu_possible_mask))
297
298/*
299 * Union to allow "aggregate OR" operation on the need for a quiescent
300 * state by the normal and expedited grace periods.
301 */
302union rcu_noqs {
303 struct {
304 u8 norm;
305 u8 exp;
306 } b; /* Bits. */
307 u16 s; /* Set of bits, aggregate OR here. */
308};
309
310/* Index values for nxttail array in struct rcu_data. */
311#define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
312#define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
313#define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
314#define RCU_NEXT_TAIL 3
315#define RCU_NEXT_SIZE 4
316
317/* Per-CPU data for read-copy update. */
318struct rcu_data {
319 /* 1) quiescent-state and grace-period handling : */
320 unsigned long completed; /* Track rsp->completed gp number */
321 /* in order to detect GP end. */
322 unsigned long gpnum; /* Highest gp number that this CPU */
323 /* is aware of having started. */
324 unsigned long rcu_qs_ctr_snap;/* Snapshot of rcu_qs_ctr to check */
325 /* for rcu_all_qs() invocations. */
326 union rcu_noqs cpu_no_qs; /* No QSes yet for this CPU. */
327 bool core_needs_qs; /* Core waits for quiesc state. */
328 bool beenonline; /* CPU online at least once. */
329 bool gpwrap; /* Possible gpnum/completed wrap. */
330 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
331 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
332 unsigned long ticks_this_gp; /* The number of scheduling-clock */
333 /* ticks this CPU has handled */
334 /* during and after the last grace */
335 /* period it is aware of. */
336
337 /* 2) batch handling */
338 /*
339 * If nxtlist is not NULL, it is partitioned as follows.
340 * Any of the partitions might be empty, in which case the
341 * pointer to that partition will be equal to the pointer for
342 * the following partition. When the list is empty, all of
343 * the nxttail elements point to the ->nxtlist pointer itself,
344 * which in that case is NULL.
345 *
346 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
347 * Entries that batch # <= ->completed
348 * The grace period for these entries has completed, and
349 * the other grace-period-completed entries may be moved
350 * here temporarily in rcu_process_callbacks().
351 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
352 * Entries that batch # <= ->completed - 1: waiting for current GP
353 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
354 * Entries known to have arrived before current GP ended
355 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
356 * Entries that might have arrived after current GP ended
357 * Note that the value of *nxttail[RCU_NEXT_TAIL] will
358 * always be NULL, as this is the end of the list.
359 */
360 struct rcu_head *nxtlist;
361 struct rcu_head **nxttail[RCU_NEXT_SIZE];
362 unsigned long nxtcompleted[RCU_NEXT_SIZE];
363 /* grace periods for sublists. */
364 long qlen_lazy; /* # of lazy queued callbacks */
365 long qlen; /* # of queued callbacks, incl lazy */
366 long qlen_last_fqs_check;
367 /* qlen at last check for QS forcing */
368 unsigned long n_cbs_invoked; /* count of RCU cbs invoked. */
369 unsigned long n_nocbs_invoked; /* count of no-CBs RCU cbs invoked. */
370 unsigned long n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
371 unsigned long n_cbs_adopted; /* RCU cbs adopted from dying CPU */
372 unsigned long n_force_qs_snap;
373 /* did other CPU force QS recently? */
374 long blimit; /* Upper limit on a processed batch */
375
376 /* 3) dynticks interface. */
377 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
378 int dynticks_snap; /* Per-GP tracking for dynticks. */
379
380 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
381 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
382 unsigned long offline_fqs; /* Kicked due to being offline. */
383 unsigned long cond_resched_completed;
384 /* Grace period that needs help */
385 /* from cond_resched(). */
386
387 /* 5) __rcu_pending() statistics. */
388 unsigned long n_rcu_pending; /* rcu_pending() calls since boot. */
389 unsigned long n_rp_core_needs_qs;
390 unsigned long n_rp_report_qs;
391 unsigned long n_rp_cb_ready;
392 unsigned long n_rp_cpu_needs_gp;
393 unsigned long n_rp_gp_completed;
394 unsigned long n_rp_gp_started;
395 unsigned long n_rp_nocb_defer_wakeup;
396 unsigned long n_rp_need_nothing;
397
398 /* 6) _rcu_barrier(), OOM callbacks, and expediting. */
399 struct rcu_head barrier_head;
400#ifdef CONFIG_RCU_FAST_NO_HZ
401 struct rcu_head oom_head;
402#endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
403 atomic_long_t exp_workdone0; /* # done by workqueue. */
404 atomic_long_t exp_workdone1; /* # done by others #1. */
405 atomic_long_t exp_workdone2; /* # done by others #2. */
406 atomic_long_t exp_workdone3; /* # done by others #3. */
407 int exp_dynticks_snap; /* Double-check need for IPI. */
408
409 /* 7) Callback offloading. */
410#ifdef CONFIG_RCU_NOCB_CPU
411 struct rcu_head *nocb_head; /* CBs waiting for kthread. */
412 struct rcu_head **nocb_tail;
413 atomic_long_t nocb_q_count; /* # CBs waiting for nocb */
414 atomic_long_t nocb_q_count_lazy; /* invocation (all stages). */
415 struct rcu_head *nocb_follower_head; /* CBs ready to invoke. */
416 struct rcu_head **nocb_follower_tail;
417 struct swait_queue_head nocb_wq; /* For nocb kthreads to sleep on. */
418 struct task_struct *nocb_kthread;
419 int nocb_defer_wakeup; /* Defer wakeup of nocb_kthread. */
420
421 /* The following fields are used by the leader, hence own cacheline. */
422 struct rcu_head *nocb_gp_head ____cacheline_internodealigned_in_smp;
423 /* CBs waiting for GP. */
424 struct rcu_head **nocb_gp_tail;
425 bool nocb_leader_sleep; /* Is the nocb leader thread asleep? */
426 struct rcu_data *nocb_next_follower;
427 /* Next follower in wakeup chain. */
428
429 /* The following fields are used by the follower, hence new cachline. */
430 struct rcu_data *nocb_leader ____cacheline_internodealigned_in_smp;
431 /* Leader CPU takes GP-end wakeups. */
432#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
433
434 /* 8) RCU CPU stall data. */
435 unsigned int softirq_snap; /* Snapshot of softirq activity. */
436
437 int cpu;
438 struct rcu_state *rsp;
439};
440
441/* Values for nocb_defer_wakeup field in struct rcu_data. */
442#define RCU_NOGP_WAKE_NOT 0
443#define RCU_NOGP_WAKE 1
444#define RCU_NOGP_WAKE_FORCE 2
445
446#define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500))
447 /* For jiffies_till_first_fqs and */
448 /* and jiffies_till_next_fqs. */
449
450#define RCU_JIFFIES_FQS_DIV 256 /* Very large systems need more */
451 /* delay between bouts of */
452 /* quiescent-state forcing. */
453
454#define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time to take */
455 /* at least one scheduling clock */
456 /* irq before ratting on them. */
457
458#define rcu_wait(cond) \
459do { \
460 for (;;) { \
461 set_current_state(TASK_INTERRUPTIBLE); \
462 if (cond) \
463 break; \
464 schedule(); \
465 } \
466 __set_current_state(TASK_RUNNING); \
467} while (0)
468
469/*
470 * RCU global state, including node hierarchy. This hierarchy is
471 * represented in "heap" form in a dense array. The root (first level)
472 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
473 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
474 * and the third level in ->node[m+1] and following (->node[m+1] referenced
475 * by ->level[2]). The number of levels is determined by the number of
476 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
477 * consisting of a single rcu_node.
478 */
479struct rcu_state {
480 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
481 struct rcu_node *level[RCU_NUM_LVLS + 1];
482 /* Hierarchy levels (+1 to */
483 /* shut bogus gcc warning) */
484 u8 flavor_mask; /* bit in flavor mask. */
485 struct rcu_data __percpu *rda; /* pointer of percu rcu_data. */
486 call_rcu_func_t call; /* call_rcu() flavor. */
487 int ncpus; /* # CPUs seen so far. */
488
489 /* The following fields are guarded by the root rcu_node's lock. */
490
491 u8 boost ____cacheline_internodealigned_in_smp;
492 /* Subject to priority boost. */
493 unsigned long gpnum; /* Current gp number. */
494 unsigned long completed; /* # of last completed gp. */
495 struct task_struct *gp_kthread; /* Task for grace periods. */
496 struct swait_queue_head gp_wq; /* Where GP task waits. */
497 short gp_flags; /* Commands for GP task. */
498 short gp_state; /* GP kthread sleep state. */
499
500 /* End of fields guarded by root rcu_node's lock. */
501
502 raw_spinlock_t orphan_lock ____cacheline_internodealigned_in_smp;
503 /* Protect following fields. */
504 struct rcu_head *orphan_nxtlist; /* Orphaned callbacks that */
505 /* need a grace period. */
506 struct rcu_head **orphan_nxttail; /* Tail of above. */
507 struct rcu_head *orphan_donelist; /* Orphaned callbacks that */
508 /* are ready to invoke. */
509 struct rcu_head **orphan_donetail; /* Tail of above. */
510 long qlen_lazy; /* Number of lazy callbacks. */
511 long qlen; /* Total number of callbacks. */
512 /* End of fields guarded by orphan_lock. */
513
514 struct mutex barrier_mutex; /* Guards barrier fields. */
515 atomic_t barrier_cpu_count; /* # CPUs waiting on. */
516 struct completion barrier_completion; /* Wake at barrier end. */
517 unsigned long barrier_sequence; /* ++ at start and end of */
518 /* _rcu_barrier(). */
519 /* End of fields guarded by barrier_mutex. */
520
521 struct mutex exp_mutex; /* Serialize expedited GP. */
522 struct mutex exp_wake_mutex; /* Serialize wakeup. */
523 unsigned long expedited_sequence; /* Take a ticket. */
524 atomic_long_t expedited_normal; /* # fallbacks to normal. */
525 atomic_t expedited_need_qs; /* # CPUs left to check in. */
526 struct swait_queue_head expedited_wq; /* Wait for check-ins. */
527 int ncpus_snap; /* # CPUs seen last time. */
528
529 unsigned long jiffies_force_qs; /* Time at which to invoke */
530 /* force_quiescent_state(). */
531 unsigned long jiffies_kick_kthreads; /* Time at which to kick */
532 /* kthreads, if configured. */
533 unsigned long n_force_qs; /* Number of calls to */
534 /* force_quiescent_state(). */
535 unsigned long n_force_qs_lh; /* ~Number of calls leaving */
536 /* due to lock unavailable. */
537 unsigned long n_force_qs_ngp; /* Number of calls leaving */
538 /* due to no GP active. */
539 unsigned long gp_start; /* Time at which GP started, */
540 /* but in jiffies. */
541 unsigned long gp_activity; /* Time of last GP kthread */
542 /* activity in jiffies. */
543 unsigned long jiffies_stall; /* Time at which to check */
544 /* for CPU stalls. */
545 unsigned long jiffies_resched; /* Time at which to resched */
546 /* a reluctant CPU. */
547 unsigned long n_force_qs_gpstart; /* Snapshot of n_force_qs at */
548 /* GP start. */
549 unsigned long gp_max; /* Maximum GP duration in */
550 /* jiffies. */
551 const char *name; /* Name of structure. */
552 char abbr; /* Abbreviated name. */
553 struct list_head flavors; /* List of RCU flavors. */
554};
555
556/* Values for rcu_state structure's gp_flags field. */
557#define RCU_GP_FLAG_INIT 0x1 /* Need grace-period initialization. */
558#define RCU_GP_FLAG_FQS 0x2 /* Need grace-period quiescent-state forcing. */
559
560/* Values for rcu_state structure's gp_state field. */
561#define RCU_GP_IDLE 0 /* Initial state and no GP in progress. */
562#define RCU_GP_WAIT_GPS 1 /* Wait for grace-period start. */
563#define RCU_GP_DONE_GPS 2 /* Wait done for grace-period start. */
564#define RCU_GP_WAIT_FQS 3 /* Wait for force-quiescent-state time. */
565#define RCU_GP_DOING_FQS 4 /* Wait done for force-quiescent-state time. */
566#define RCU_GP_CLEANUP 5 /* Grace-period cleanup started. */
567#define RCU_GP_CLEANED 6 /* Grace-period cleanup complete. */
568
569#ifndef RCU_TREE_NONCORE
570static const char * const gp_state_names[] = {
571 "RCU_GP_IDLE",
572 "RCU_GP_WAIT_GPS",
573 "RCU_GP_DONE_GPS",
574 "RCU_GP_WAIT_FQS",
575 "RCU_GP_DOING_FQS",
576 "RCU_GP_CLEANUP",
577 "RCU_GP_CLEANED",
578};
579#endif /* #ifndef RCU_TREE_NONCORE */
580
581extern struct list_head rcu_struct_flavors;
582
583/* Sequence through rcu_state structures for each RCU flavor. */
584#define for_each_rcu_flavor(rsp) \
585 list_for_each_entry((rsp), &rcu_struct_flavors, flavors)
586
587/*
588 * RCU implementation internal declarations:
589 */
590extern struct rcu_state rcu_sched_state;
591
592extern struct rcu_state rcu_bh_state;
593
594#ifdef CONFIG_PREEMPT_RCU
595extern struct rcu_state rcu_preempt_state;
596#endif /* #ifdef CONFIG_PREEMPT_RCU */
597
598#ifdef CONFIG_RCU_BOOST
599DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
600DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu);
601DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
602DECLARE_PER_CPU(char, rcu_cpu_has_work);
603#endif /* #ifdef CONFIG_RCU_BOOST */
604
605#ifndef RCU_TREE_NONCORE
606
607/* Forward declarations for rcutree_plugin.h */
608static void rcu_bootup_announce(void);
609static void rcu_preempt_note_context_switch(void);
610static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
611#ifdef CONFIG_HOTPLUG_CPU
612static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
613#endif /* #ifdef CONFIG_HOTPLUG_CPU */
614static void rcu_print_detail_task_stall(struct rcu_state *rsp);
615static int rcu_print_task_stall(struct rcu_node *rnp);
616static int rcu_print_task_exp_stall(struct rcu_node *rnp);
617static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
618static void rcu_preempt_check_callbacks(void);
619void call_rcu(struct rcu_head *head, rcu_callback_t func);
620static void __init __rcu_init_preempt(void);
621static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
622static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
623static void invoke_rcu_callbacks_kthread(void);
624static bool rcu_is_callbacks_kthread(void);
625#ifdef CONFIG_RCU_BOOST
626static void rcu_preempt_do_callbacks(void);
627static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
628 struct rcu_node *rnp);
629#endif /* #ifdef CONFIG_RCU_BOOST */
630static void __init rcu_spawn_boost_kthreads(void);
631static void rcu_prepare_kthreads(int cpu);
632static void rcu_cleanup_after_idle(void);
633static void rcu_prepare_for_idle(void);
634static void rcu_idle_count_callbacks_posted(void);
635static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
636static void print_cpu_stall_info_begin(void);
637static void print_cpu_stall_info(struct rcu_state *rsp, int cpu);
638static void print_cpu_stall_info_end(void);
639static void zero_cpu_stall_ticks(struct rcu_data *rdp);
640static void increment_cpu_stall_ticks(void);
641static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu);
642static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq);
643static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp);
644static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq);
645static void rcu_init_one_nocb(struct rcu_node *rnp);
646static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
647 bool lazy, unsigned long flags);
648static bool rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp,
649 struct rcu_data *rdp,
650 unsigned long flags);
651static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp);
652static void do_nocb_deferred_wakeup(struct rcu_data *rdp);
653static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp);
654static void rcu_spawn_all_nocb_kthreads(int cpu);
655static void __init rcu_spawn_nocb_kthreads(void);
656#ifdef CONFIG_RCU_NOCB_CPU
657static void __init rcu_organize_nocb_kthreads(struct rcu_state *rsp);
658#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
659static void __maybe_unused rcu_kick_nohz_cpu(int cpu);
660static bool init_nocb_callback_list(struct rcu_data *rdp);
661static void rcu_sysidle_enter(int irq);
662static void rcu_sysidle_exit(int irq);
663static void rcu_sysidle_check_cpu(struct rcu_data *rdp, bool *isidle,
664 unsigned long *maxj);
665static bool is_sysidle_rcu_state(struct rcu_state *rsp);
666static void rcu_sysidle_report_gp(struct rcu_state *rsp, int isidle,
667 unsigned long maxj);
668static void rcu_bind_gp_kthread(void);
669static void rcu_sysidle_init_percpu_data(struct rcu_dynticks *rdtp);
670static bool rcu_nohz_full_cpu(struct rcu_state *rsp);
671static void rcu_dynticks_task_enter(void);
672static void rcu_dynticks_task_exit(void);
673
674#endif /* #ifndef RCU_TREE_NONCORE */
675
676#ifdef CONFIG_RCU_TRACE
677/* Read out queue lengths for tracing. */
678static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll)
679{
680#ifdef CONFIG_RCU_NOCB_CPU
681 *ql = atomic_long_read(&rdp->nocb_q_count);
682 *qll = atomic_long_read(&rdp->nocb_q_count_lazy);
683#else /* #ifdef CONFIG_RCU_NOCB_CPU */
684 *ql = 0;
685 *qll = 0;
686#endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
687}
688#endif /* #ifdef CONFIG_RCU_TRACE */
689
690/*
691 * Place this after a lock-acquisition primitive to guarantee that
692 * an UNLOCK+LOCK pair act as a full barrier. This guarantee applies
693 * if the UNLOCK and LOCK are executed by the same CPU or if the
694 * UNLOCK and LOCK operate on the same lock variable.
695 */
696#ifdef CONFIG_PPC
697#define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */
698#else /* #ifdef CONFIG_PPC */
699#define smp_mb__after_unlock_lock() do { } while (0)
700#endif /* #else #ifdef CONFIG_PPC */
701
702/*
703 * Wrappers for the rcu_node::lock acquire and release.
704 *
705 * Because the rcu_nodes form a tree, the tree traversal locking will observe
706 * different lock values, this in turn means that an UNLOCK of one level
707 * followed by a LOCK of another level does not imply a full memory barrier;
708 * and most importantly transitivity is lost.
709 *
710 * In order to restore full ordering between tree levels, augment the regular
711 * lock acquire functions with smp_mb__after_unlock_lock().
712 *
713 * As ->lock of struct rcu_node is a __private field, therefore one should use
714 * these wrappers rather than directly call raw_spin_{lock,unlock}* on ->lock.
715 */
716static inline void raw_spin_lock_rcu_node(struct rcu_node *rnp)
717{
718 raw_spin_lock(&ACCESS_PRIVATE(rnp, lock));
719 smp_mb__after_unlock_lock();
720}
721
722static inline void raw_spin_unlock_rcu_node(struct rcu_node *rnp)
723{
724 raw_spin_unlock(&ACCESS_PRIVATE(rnp, lock));
725}
726
727static inline void raw_spin_lock_irq_rcu_node(struct rcu_node *rnp)
728{
729 raw_spin_lock_irq(&ACCESS_PRIVATE(rnp, lock));
730 smp_mb__after_unlock_lock();
731}
732
733static inline void raw_spin_unlock_irq_rcu_node(struct rcu_node *rnp)
734{
735 raw_spin_unlock_irq(&ACCESS_PRIVATE(rnp, lock));
736}
737
738#define raw_spin_lock_irqsave_rcu_node(rnp, flags) \
739do { \
740 typecheck(unsigned long, flags); \
741 raw_spin_lock_irqsave(&ACCESS_PRIVATE(rnp, lock), flags); \
742 smp_mb__after_unlock_lock(); \
743} while (0)
744
745#define raw_spin_unlock_irqrestore_rcu_node(rnp, flags) \
746do { \
747 typecheck(unsigned long, flags); \
748 raw_spin_unlock_irqrestore(&ACCESS_PRIVATE(rnp, lock), flags); \
749} while (0)
750
751static inline bool raw_spin_trylock_rcu_node(struct rcu_node *rnp)
752{
753 bool locked = raw_spin_trylock(&ACCESS_PRIVATE(rnp, lock));
754
755 if (locked)
756 smp_mb__after_unlock_lock();
757 return locked;
758}
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 workqueue handler. */
25struct rcu_exp_work {
26 unsigned long rew_s;
27#ifdef CONFIG_RCU_EXP_KTHREAD
28 struct kthread_work rew_work;
29#else
30 struct work_struct rew_work;
31#endif /* CONFIG_RCU_EXP_KTHREAD */
32};
33
34/* RCU's kthread states for tracing. */
35#define RCU_KTHREAD_STOPPED 0
36#define RCU_KTHREAD_RUNNING 1
37#define RCU_KTHREAD_WAITING 2
38#define RCU_KTHREAD_OFFCPU 3
39#define RCU_KTHREAD_YIELDING 4
40#define RCU_KTHREAD_MAX 4
41
42/*
43 * Definition for node within the RCU grace-period-detection hierarchy.
44 */
45struct rcu_node {
46 raw_spinlock_t __private lock; /* Root rcu_node's lock protects */
47 /* some rcu_state fields as well as */
48 /* following. */
49 unsigned long gp_seq; /* Track rsp->gp_seq. */
50 unsigned long gp_seq_needed; /* Track furthest future GP request. */
51 unsigned long completedqs; /* All QSes done for this node. */
52 unsigned long qsmask; /* CPUs or groups that need to switch in */
53 /* order for current grace period to proceed.*/
54 /* In leaf rcu_node, each bit corresponds to */
55 /* an rcu_data structure, otherwise, each */
56 /* bit corresponds to a child rcu_node */
57 /* structure. */
58 unsigned long rcu_gp_init_mask; /* Mask of offline CPUs at GP init. */
59 unsigned long qsmaskinit;
60 /* Per-GP initial value for qsmask. */
61 /* Initialized from ->qsmaskinitnext at the */
62 /* beginning of each grace period. */
63 unsigned long qsmaskinitnext;
64 unsigned long expmask; /* CPUs or groups that need to check in */
65 /* to allow the current expedited GP */
66 /* to complete. */
67 unsigned long expmaskinit;
68 /* Per-GP initial values for expmask. */
69 /* Initialized from ->expmaskinitnext at the */
70 /* beginning of each expedited GP. */
71 unsigned long expmaskinitnext;
72 /* Online CPUs for next expedited GP. */
73 /* Any CPU that has ever been online will */
74 /* have its bit set. */
75 unsigned long cbovldmask;
76 /* CPUs experiencing callback overload. */
77 unsigned long ffmask; /* Fully functional CPUs. */
78 unsigned long grpmask; /* Mask to apply to parent qsmask. */
79 /* Only one bit will be set in this mask. */
80 int grplo; /* lowest-numbered CPU here. */
81 int grphi; /* highest-numbered CPU here. */
82 u8 grpnum; /* group number for next level up. */
83 u8 level; /* root is at level 0. */
84 bool wait_blkd_tasks;/* Necessary to wait for blocked tasks to */
85 /* exit RCU read-side critical sections */
86 /* before propagating offline up the */
87 /* rcu_node tree? */
88 struct rcu_node *parent;
89 struct list_head blkd_tasks;
90 /* Tasks blocked in RCU read-side critical */
91 /* section. Tasks are placed at the head */
92 /* of this list and age towards the tail. */
93 struct list_head *gp_tasks;
94 /* Pointer to the first task blocking the */
95 /* current grace period, or NULL if there */
96 /* is no such task. */
97 struct list_head *exp_tasks;
98 /* Pointer to the first task blocking the */
99 /* current expedited grace period, or NULL */
100 /* if there is no such task. If there */
101 /* is no current expedited grace period, */
102 /* then there can cannot be any such task. */
103 struct list_head *boost_tasks;
104 /* Pointer to first task that needs to be */
105 /* priority boosted, or NULL if no priority */
106 /* boosting is needed for this rcu_node */
107 /* structure. If there are no tasks */
108 /* queued on this rcu_node structure that */
109 /* are blocking the current grace period, */
110 /* there can be no such task. */
111 struct rt_mutex boost_mtx;
112 /* Used only for the priority-boosting */
113 /* side effect, not as a lock. */
114 unsigned long boost_time;
115 /* When to start boosting (jiffies). */
116 struct mutex boost_kthread_mutex;
117 /* Exclusion for thread spawning and affinity */
118 /* manipulation. */
119 struct task_struct *boost_kthread_task;
120 /* kthread that takes care of priority */
121 /* boosting for this rcu_node structure. */
122 unsigned int boost_kthread_status;
123 /* State of boost_kthread_task for tracing. */
124 unsigned long n_boosts; /* Number of boosts for this rcu_node structure. */
125#ifdef CONFIG_RCU_NOCB_CPU
126 struct swait_queue_head nocb_gp_wq[2];
127 /* Place for rcu_nocb_kthread() to wait GP. */
128#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
129 raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp;
130
131 spinlock_t exp_lock ____cacheline_internodealigned_in_smp;
132 unsigned long exp_seq_rq;
133 wait_queue_head_t exp_wq[4];
134 struct rcu_exp_work rew;
135 bool exp_need_flush; /* Need to flush workitem? */
136 raw_spinlock_t exp_poll_lock;
137 /* Lock and data for polled expedited grace periods. */
138 unsigned long exp_seq_poll_rq;
139 struct work_struct exp_poll_wq;
140} ____cacheline_internodealigned_in_smp;
141
142/*
143 * Bitmasks in an rcu_node cover the interval [grplo, grphi] of CPU IDs, and
144 * are indexed relative to this interval rather than the global CPU ID space.
145 * This generates the bit for a CPU in node-local masks.
146 */
147#define leaf_node_cpu_bit(rnp, cpu) (BIT((cpu) - (rnp)->grplo))
148
149/*
150 * Union to allow "aggregate OR" operation on the need for a quiescent
151 * state by the normal and expedited grace periods.
152 */
153union rcu_noqs {
154 struct {
155 u8 norm;
156 u8 exp;
157 } b; /* Bits. */
158 u16 s; /* Set of bits, aggregate OR here. */
159};
160
161/* Per-CPU data for read-copy update. */
162struct rcu_data {
163 /* 1) quiescent-state and grace-period handling : */
164 unsigned long gp_seq; /* Track rsp->gp_seq counter. */
165 unsigned long gp_seq_needed; /* Track furthest future GP request. */
166 union rcu_noqs cpu_no_qs; /* No QSes yet for this CPU. */
167 bool core_needs_qs; /* Core waits for quiescent state. */
168 bool beenonline; /* CPU online at least once. */
169 bool gpwrap; /* Possible ->gp_seq wrap. */
170 bool cpu_started; /* RCU watching this onlining CPU. */
171 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
172 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
173 unsigned long ticks_this_gp; /* The number of scheduling-clock */
174 /* ticks this CPU has handled */
175 /* during and after the last grace */
176 /* period it is aware of. */
177 struct irq_work defer_qs_iw; /* Obtain later scheduler attention. */
178 bool defer_qs_iw_pending; /* Scheduler attention pending? */
179 struct work_struct strict_work; /* Schedule readers for strict GPs. */
180
181 /* 2) batch handling */
182 struct rcu_segcblist cblist; /* Segmented callback list, with */
183 /* different callbacks waiting for */
184 /* different grace periods. */
185 long qlen_last_fqs_check;
186 /* qlen at last check for QS forcing */
187 unsigned long n_cbs_invoked; /* # callbacks invoked since boot. */
188 unsigned long n_force_qs_snap;
189 /* did other CPU force QS recently? */
190 long blimit; /* Upper limit on a processed batch */
191
192 /* 3) dynticks interface. */
193 int dynticks_snap; /* Per-GP tracking for dynticks. */
194 bool rcu_need_heavy_qs; /* GP old, so heavy quiescent state! */
195 bool rcu_urgent_qs; /* GP old need light quiescent state. */
196 bool rcu_forced_tick; /* Forced tick to provide QS. */
197 bool rcu_forced_tick_exp; /* ... provide QS to expedited GP. */
198
199 /* 4) rcu_barrier(), OOM callbacks, and expediting. */
200 unsigned long barrier_seq_snap; /* Snap of rcu_state.barrier_sequence. */
201 struct rcu_head barrier_head;
202 int exp_dynticks_snap; /* Double-check need for IPI. */
203
204 /* 5) Callback offloading. */
205#ifdef CONFIG_RCU_NOCB_CPU
206 struct swait_queue_head nocb_cb_wq; /* For nocb kthreads to sleep on. */
207 struct swait_queue_head nocb_state_wq; /* For offloading state changes */
208 struct task_struct *nocb_gp_kthread;
209 raw_spinlock_t nocb_lock; /* Guard following pair of fields. */
210 atomic_t nocb_lock_contended; /* Contention experienced. */
211 int nocb_defer_wakeup; /* Defer wakeup of nocb_kthread. */
212 struct timer_list nocb_timer; /* Enforce finite deferral. */
213 unsigned long nocb_gp_adv_time; /* Last call_rcu() CB adv (jiffies). */
214 struct mutex nocb_gp_kthread_mutex; /* Exclusion for nocb gp kthread */
215 /* spawning */
216
217 /* The following fields are used by call_rcu, hence own cacheline. */
218 raw_spinlock_t nocb_bypass_lock ____cacheline_internodealigned_in_smp;
219 struct rcu_cblist nocb_bypass; /* Lock-contention-bypass CB list. */
220 unsigned long nocb_bypass_first; /* Time (jiffies) of first enqueue. */
221 unsigned long nocb_nobypass_last; /* Last ->cblist enqueue (jiffies). */
222 int nocb_nobypass_count; /* # ->cblist enqueues at ^^^ time. */
223
224 /* The following fields are used by GP kthread, hence own cacheline. */
225 raw_spinlock_t nocb_gp_lock ____cacheline_internodealigned_in_smp;
226 u8 nocb_gp_sleep; /* Is the nocb GP thread asleep? */
227 u8 nocb_gp_bypass; /* Found a bypass on last scan? */
228 u8 nocb_gp_gp; /* GP to wait for on last scan? */
229 unsigned long nocb_gp_seq; /* If so, ->gp_seq to wait for. */
230 unsigned long nocb_gp_loops; /* # passes through wait code. */
231 struct swait_queue_head nocb_gp_wq; /* For nocb kthreads to sleep on. */
232 bool nocb_cb_sleep; /* Is the nocb CB thread asleep? */
233 struct task_struct *nocb_cb_kthread;
234 struct list_head nocb_head_rdp; /*
235 * Head of rcu_data list in wakeup chain,
236 * if rdp_gp.
237 */
238 struct list_head nocb_entry_rdp; /* rcu_data node in wakeup chain. */
239 struct rcu_data *nocb_toggling_rdp; /* rdp queued for (de-)offloading */
240
241 /* The following fields are used by CB kthread, hence new cacheline. */
242 struct rcu_data *nocb_gp_rdp ____cacheline_internodealigned_in_smp;
243 /* GP rdp takes GP-end wakeups. */
244#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
245
246 /* 6) RCU priority boosting. */
247 struct task_struct *rcu_cpu_kthread_task;
248 /* rcuc per-CPU kthread or NULL. */
249 unsigned int rcu_cpu_kthread_status;
250 char rcu_cpu_has_work;
251 unsigned long rcuc_activity;
252
253 /* 7) Diagnostic data, including RCU CPU stall warnings. */
254 unsigned int softirq_snap; /* Snapshot of softirq activity. */
255 /* ->rcu_iw* fields protected by leaf rcu_node ->lock. */
256 struct irq_work rcu_iw; /* Check for non-irq activity. */
257 bool rcu_iw_pending; /* Is ->rcu_iw pending? */
258 unsigned long rcu_iw_gp_seq; /* ->gp_seq associated with ->rcu_iw. */
259 unsigned long rcu_ofl_gp_seq; /* ->gp_seq at last offline. */
260 short rcu_ofl_gp_flags; /* ->gp_flags at last offline. */
261 unsigned long rcu_onl_gp_seq; /* ->gp_seq at last online. */
262 short rcu_onl_gp_flags; /* ->gp_flags at last online. */
263 unsigned long last_fqs_resched; /* Time of last rcu_resched(). */
264 unsigned long last_sched_clock; /* Jiffies of last rcu_sched_clock_irq(). */
265
266 long lazy_len; /* Length of buffered lazy callbacks. */
267 int cpu;
268};
269
270/* Values for nocb_defer_wakeup field in struct rcu_data. */
271#define RCU_NOCB_WAKE_NOT 0
272#define RCU_NOCB_WAKE_BYPASS 1
273#define RCU_NOCB_WAKE_LAZY 2
274#define RCU_NOCB_WAKE 3
275#define RCU_NOCB_WAKE_FORCE 4
276
277#define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500))
278 /* For jiffies_till_first_fqs and */
279 /* and jiffies_till_next_fqs. */
280
281#define RCU_JIFFIES_FQS_DIV 256 /* Very large systems need more */
282 /* delay between bouts of */
283 /* quiescent-state forcing. */
284
285#define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time to take */
286 /* at least one scheduling clock */
287 /* irq before ratting on them. */
288
289#define rcu_wait(cond) \
290do { \
291 for (;;) { \
292 set_current_state(TASK_INTERRUPTIBLE); \
293 if (cond) \
294 break; \
295 schedule(); \
296 } \
297 __set_current_state(TASK_RUNNING); \
298} while (0)
299
300/*
301 * RCU global state, including node hierarchy. This hierarchy is
302 * represented in "heap" form in a dense array. The root (first level)
303 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
304 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
305 * and the third level in ->node[m+1] and following (->node[m+1] referenced
306 * by ->level[2]). The number of levels is determined by the number of
307 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
308 * consisting of a single rcu_node.
309 */
310struct rcu_state {
311 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
312 struct rcu_node *level[RCU_NUM_LVLS + 1];
313 /* Hierarchy levels (+1 to */
314 /* shut bogus gcc warning) */
315 int ncpus; /* # CPUs seen so far. */
316 int n_online_cpus; /* # CPUs online for RCU. */
317
318 /* The following fields are guarded by the root rcu_node's lock. */
319
320 unsigned long gp_seq ____cacheline_internodealigned_in_smp;
321 /* Grace-period sequence #. */
322 unsigned long gp_max; /* Maximum GP duration in */
323 /* jiffies. */
324 struct task_struct *gp_kthread; /* Task for grace periods. */
325 struct swait_queue_head gp_wq; /* Where GP task waits. */
326 short gp_flags; /* Commands for GP task. */
327 short gp_state; /* GP kthread sleep state. */
328 unsigned long gp_wake_time; /* Last GP kthread wake. */
329 unsigned long gp_wake_seq; /* ->gp_seq at ^^^. */
330 unsigned long gp_seq_polled; /* GP seq for polled API. */
331 unsigned long gp_seq_polled_snap; /* ->gp_seq_polled at normal GP start. */
332 unsigned long gp_seq_polled_exp_snap; /* ->gp_seq_polled at expedited GP start. */
333
334 /* End of fields guarded by root rcu_node's lock. */
335
336 struct mutex barrier_mutex; /* Guards barrier fields. */
337 atomic_t barrier_cpu_count; /* # CPUs waiting on. */
338 struct completion barrier_completion; /* Wake at barrier end. */
339 unsigned long barrier_sequence; /* ++ at start and end of */
340 /* rcu_barrier(). */
341 /* End of fields guarded by barrier_mutex. */
342
343 raw_spinlock_t barrier_lock; /* Protects ->barrier_seq_snap. */
344
345 struct mutex exp_mutex; /* Serialize expedited GP. */
346 struct mutex exp_wake_mutex; /* Serialize wakeup. */
347 unsigned long expedited_sequence; /* Take a ticket. */
348 atomic_t expedited_need_qs; /* # CPUs left to check in. */
349 struct swait_queue_head expedited_wq; /* Wait for check-ins. */
350 int ncpus_snap; /* # CPUs seen last time. */
351 u8 cbovld; /* Callback overload now? */
352 u8 cbovldnext; /* ^ ^ next time? */
353
354 unsigned long jiffies_force_qs; /* Time at which to invoke */
355 /* force_quiescent_state(). */
356 unsigned long jiffies_kick_kthreads; /* Time at which to kick */
357 /* kthreads, if configured. */
358 unsigned long n_force_qs; /* Number of calls to */
359 /* force_quiescent_state(). */
360 unsigned long gp_start; /* Time at which GP started, */
361 /* but in jiffies. */
362 unsigned long gp_end; /* Time last GP ended, again */
363 /* in jiffies. */
364 unsigned long gp_activity; /* Time of last GP kthread */
365 /* activity in jiffies. */
366 unsigned long gp_req_activity; /* Time of last GP request */
367 /* in jiffies. */
368 unsigned long jiffies_stall; /* Time at which to check */
369 /* for CPU stalls. */
370 unsigned long jiffies_resched; /* Time at which to resched */
371 /* a reluctant CPU. */
372 unsigned long n_force_qs_gpstart; /* Snapshot of n_force_qs at */
373 /* GP start. */
374 const char *name; /* Name of structure. */
375 char abbr; /* Abbreviated name. */
376
377 arch_spinlock_t ofl_lock ____cacheline_internodealigned_in_smp;
378 /* Synchronize offline with */
379 /* GP pre-initialization. */
380 int nocb_is_setup; /* nocb is setup from boot */
381};
382
383/* Values for rcu_state structure's gp_flags field. */
384#define RCU_GP_FLAG_INIT 0x1 /* Need grace-period initialization. */
385#define RCU_GP_FLAG_FQS 0x2 /* Need grace-period quiescent-state forcing. */
386#define RCU_GP_FLAG_OVLD 0x4 /* Experiencing callback overload. */
387
388/* Values for rcu_state structure's gp_state field. */
389#define RCU_GP_IDLE 0 /* Initial state and no GP in progress. */
390#define RCU_GP_WAIT_GPS 1 /* Wait for grace-period start. */
391#define RCU_GP_DONE_GPS 2 /* Wait done for grace-period start. */
392#define RCU_GP_ONOFF 3 /* Grace-period initialization hotplug. */
393#define RCU_GP_INIT 4 /* Grace-period initialization. */
394#define RCU_GP_WAIT_FQS 5 /* Wait for force-quiescent-state time. */
395#define RCU_GP_DOING_FQS 6 /* Wait done for force-quiescent-state time. */
396#define RCU_GP_CLEANUP 7 /* Grace-period cleanup started. */
397#define RCU_GP_CLEANED 8 /* Grace-period cleanup complete. */
398
399/*
400 * In order to export the rcu_state name to the tracing tools, it
401 * needs to be added in the __tracepoint_string section.
402 * This requires defining a separate variable tp_<sname>_varname
403 * that points to the string being used, and this will allow
404 * the tracing userspace tools to be able to decipher the string
405 * address to the matching string.
406 */
407#ifdef CONFIG_PREEMPT_RCU
408#define RCU_ABBR 'p'
409#define RCU_NAME_RAW "rcu_preempt"
410#else /* #ifdef CONFIG_PREEMPT_RCU */
411#define RCU_ABBR 's'
412#define RCU_NAME_RAW "rcu_sched"
413#endif /* #else #ifdef CONFIG_PREEMPT_RCU */
414#ifndef CONFIG_TRACING
415#define RCU_NAME RCU_NAME_RAW
416#else /* #ifdef CONFIG_TRACING */
417static char rcu_name[] = RCU_NAME_RAW;
418static const char *tp_rcu_varname __used __tracepoint_string = rcu_name;
419#define RCU_NAME rcu_name
420#endif /* #else #ifdef CONFIG_TRACING */
421
422/* Forward declarations for tree_plugin.h */
423static void rcu_bootup_announce(void);
424static void rcu_qs(void);
425static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
426#ifdef CONFIG_HOTPLUG_CPU
427static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
428#endif /* #ifdef CONFIG_HOTPLUG_CPU */
429static int rcu_print_task_exp_stall(struct rcu_node *rnp);
430static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
431static void rcu_flavor_sched_clock_irq(int user);
432static void dump_blkd_tasks(struct rcu_node *rnp, int ncheck);
433static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
434static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
435static bool rcu_is_callbacks_kthread(struct rcu_data *rdp);
436static void rcu_cpu_kthread_setup(unsigned int cpu);
437static void rcu_spawn_one_boost_kthread(struct rcu_node *rnp);
438static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
439static bool rcu_preempt_need_deferred_qs(struct task_struct *t);
440static void zero_cpu_stall_ticks(struct rcu_data *rdp);
441static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp);
442static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq);
443static void rcu_init_one_nocb(struct rcu_node *rnp);
444static bool wake_nocb_gp(struct rcu_data *rdp, bool force);
445static bool rcu_nocb_flush_bypass(struct rcu_data *rdp, struct rcu_head *rhp,
446 unsigned long j, bool lazy);
447static bool rcu_nocb_try_bypass(struct rcu_data *rdp, struct rcu_head *rhp,
448 bool *was_alldone, unsigned long flags,
449 bool lazy);
450static void __call_rcu_nocb_wake(struct rcu_data *rdp, bool was_empty,
451 unsigned long flags);
452static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp, int level);
453static bool do_nocb_deferred_wakeup(struct rcu_data *rdp);
454static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp);
455static void rcu_spawn_cpu_nocb_kthread(int cpu);
456static void show_rcu_nocb_state(struct rcu_data *rdp);
457static void rcu_nocb_lock(struct rcu_data *rdp);
458static void rcu_nocb_unlock(struct rcu_data *rdp);
459static void rcu_nocb_unlock_irqrestore(struct rcu_data *rdp,
460 unsigned long flags);
461static void rcu_lockdep_assert_cblist_protected(struct rcu_data *rdp);
462#ifdef CONFIG_RCU_NOCB_CPU
463static void __init rcu_organize_nocb_kthreads(void);
464
465/*
466 * Disable IRQs before checking offloaded state so that local
467 * locking is safe against concurrent de-offloading.
468 */
469#define rcu_nocb_lock_irqsave(rdp, flags) \
470do { \
471 local_irq_save(flags); \
472 if (rcu_segcblist_is_offloaded(&(rdp)->cblist)) \
473 raw_spin_lock(&(rdp)->nocb_lock); \
474} while (0)
475#else /* #ifdef CONFIG_RCU_NOCB_CPU */
476#define rcu_nocb_lock_irqsave(rdp, flags) local_irq_save(flags)
477#endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
478
479static void rcu_bind_gp_kthread(void);
480static bool rcu_nohz_full_cpu(void);
481
482/* Forward declarations for tree_stall.h */
483static void record_gp_stall_check_time(void);
484static void rcu_iw_handler(struct irq_work *iwp);
485static void check_cpu_stall(struct rcu_data *rdp);
486static void rcu_check_gp_start_stall(struct rcu_node *rnp, struct rcu_data *rdp,
487 const unsigned long gpssdelay);
488
489/* Forward declarations for tree_exp.h. */
490static void sync_rcu_do_polled_gp(struct work_struct *wp);