Loading...
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);
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/rtmutex.h>
28#include <linux/threads.h>
29#include <linux/cpumask.h>
30#include <linux/seqlock.h>
31#include <linux/swait.h>
32#include <linux/stop_machine.h>
33#include <linux/rcu_node_tree.h>
34
35#include "rcu_segcblist.h"
36
37/*
38 * Dynticks per-CPU state.
39 */
40struct rcu_dynticks {
41 long dynticks_nesting; /* Track process nesting level. */
42 long dynticks_nmi_nesting; /* Track irq/NMI nesting level. */
43 atomic_t dynticks; /* Even value for idle, else odd. */
44 bool rcu_need_heavy_qs; /* GP old, need heavy quiescent state. */
45 unsigned long rcu_qs_ctr; /* Light universal quiescent state ctr. */
46 bool rcu_urgent_qs; /* GP old need light quiescent state. */
47#ifdef CONFIG_RCU_FAST_NO_HZ
48 bool all_lazy; /* Are all CPU's CBs lazy? */
49 unsigned long nonlazy_posted;
50 /* # times non-lazy CBs posted to CPU. */
51 unsigned long nonlazy_posted_snap;
52 /* idle-period nonlazy_posted snapshot. */
53 unsigned long last_accelerate;
54 /* Last jiffy CBs were accelerated. */
55 unsigned long last_advance_all;
56 /* Last jiffy CBs were all advanced. */
57 int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */
58#endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
59};
60
61/* RCU's kthread states for tracing. */
62#define RCU_KTHREAD_STOPPED 0
63#define RCU_KTHREAD_RUNNING 1
64#define RCU_KTHREAD_WAITING 2
65#define RCU_KTHREAD_OFFCPU 3
66#define RCU_KTHREAD_YIELDING 4
67#define RCU_KTHREAD_MAX 4
68
69/*
70 * Definition for node within the RCU grace-period-detection hierarchy.
71 */
72struct rcu_node {
73 raw_spinlock_t __private lock; /* Root rcu_node's lock protects */
74 /* some rcu_state fields as well as */
75 /* following. */
76 unsigned long gpnum; /* Current grace period for this node. */
77 /* This will either be equal to or one */
78 /* behind the root rcu_node's gpnum. */
79 unsigned long completed; /* Last GP completed for this node. */
80 /* This will either be equal to or one */
81 /* behind the root rcu_node's gpnum. */
82 unsigned long qsmask; /* CPUs or groups that need to switch in */
83 /* order for current grace period to proceed.*/
84 /* In leaf rcu_node, each bit corresponds to */
85 /* an rcu_data structure, otherwise, each */
86 /* bit corresponds to a child rcu_node */
87 /* structure. */
88 unsigned long qsmaskinit;
89 /* Per-GP initial value for qsmask. */
90 /* Initialized from ->qsmaskinitnext at the */
91 /* beginning of each grace period. */
92 unsigned long qsmaskinitnext;
93 /* Online CPUs for next grace period. */
94 unsigned long expmask; /* CPUs or groups that need to check in */
95 /* to allow the current expedited GP */
96 /* to complete. */
97 unsigned long expmaskinit;
98 /* Per-GP initial values for expmask. */
99 /* Initialized from ->expmaskinitnext at the */
100 /* beginning of each expedited GP. */
101 unsigned long expmaskinitnext;
102 /* Online CPUs for next expedited GP. */
103 /* Any CPU that has ever been online will */
104 /* have its bit set. */
105 unsigned long ffmask; /* Fully functional CPUs. */
106 unsigned long grpmask; /* Mask to apply to parent qsmask. */
107 /* Only one bit will be set in this mask. */
108 int grplo; /* lowest-numbered CPU or group here. */
109 int grphi; /* highest-numbered CPU or group here. */
110 u8 grpnum; /* CPU/group number for next level up. */
111 u8 level; /* root is at level 0. */
112 bool wait_blkd_tasks;/* Necessary to wait for blocked tasks to */
113 /* exit RCU read-side critical sections */
114 /* before propagating offline up the */
115 /* rcu_node tree? */
116 struct rcu_node *parent;
117 struct list_head blkd_tasks;
118 /* Tasks blocked in RCU read-side critical */
119 /* section. Tasks are placed at the head */
120 /* of this list and age towards the tail. */
121 struct list_head *gp_tasks;
122 /* Pointer to the first task blocking the */
123 /* current grace period, or NULL if there */
124 /* is no such task. */
125 struct list_head *exp_tasks;
126 /* Pointer to the first task blocking the */
127 /* current expedited grace period, or NULL */
128 /* if there is no such task. If there */
129 /* is no current expedited grace period, */
130 /* then there can cannot be any such task. */
131 struct list_head *boost_tasks;
132 /* Pointer to first task that needs to be */
133 /* priority boosted, or NULL if no priority */
134 /* boosting is needed for this rcu_node */
135 /* structure. If there are no tasks */
136 /* queued on this rcu_node structure that */
137 /* are blocking the current grace period, */
138 /* there can be no such task. */
139 struct rt_mutex boost_mtx;
140 /* Used only for the priority-boosting */
141 /* side effect, not as a lock. */
142 unsigned long boost_time;
143 /* When to start boosting (jiffies). */
144 struct task_struct *boost_kthread_task;
145 /* kthread that takes care of priority */
146 /* boosting for this rcu_node structure. */
147 unsigned int boost_kthread_status;
148 /* State of boost_kthread_task for tracing. */
149#ifdef CONFIG_RCU_NOCB_CPU
150 struct swait_queue_head nocb_gp_wq[2];
151 /* Place for rcu_nocb_kthread() to wait GP. */
152#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
153 int need_future_gp[2];
154 /* Counts of upcoming no-CB GP requests. */
155 raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp;
156
157 spinlock_t exp_lock ____cacheline_internodealigned_in_smp;
158 unsigned long exp_seq_rq;
159 wait_queue_head_t exp_wq[4];
160} ____cacheline_internodealigned_in_smp;
161
162/*
163 * Bitmasks in an rcu_node cover the interval [grplo, grphi] of CPU IDs, and
164 * are indexed relative to this interval rather than the global CPU ID space.
165 * This generates the bit for a CPU in node-local masks.
166 */
167#define leaf_node_cpu_bit(rnp, cpu) (1UL << ((cpu) - (rnp)->grplo))
168
169/*
170 * Union to allow "aggregate OR" operation on the need for a quiescent
171 * state by the normal and expedited grace periods.
172 */
173union rcu_noqs {
174 struct {
175 u8 norm;
176 u8 exp;
177 } b; /* Bits. */
178 u16 s; /* Set of bits, aggregate OR here. */
179};
180
181/* Per-CPU data for read-copy update. */
182struct rcu_data {
183 /* 1) quiescent-state and grace-period handling : */
184 unsigned long completed; /* Track rsp->completed gp number */
185 /* in order to detect GP end. */
186 unsigned long gpnum; /* Highest gp number that this CPU */
187 /* is aware of having started. */
188 unsigned long rcu_qs_ctr_snap;/* Snapshot of rcu_qs_ctr to check */
189 /* for rcu_all_qs() invocations. */
190 union rcu_noqs cpu_no_qs; /* No QSes yet for this CPU. */
191 bool core_needs_qs; /* Core waits for quiesc state. */
192 bool beenonline; /* CPU online at least once. */
193 bool gpwrap; /* Possible gpnum/completed wrap. */
194 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
195 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
196 unsigned long ticks_this_gp; /* The number of scheduling-clock */
197 /* ticks this CPU has handled */
198 /* during and after the last grace */
199 /* period it is aware of. */
200
201 /* 2) batch handling */
202 struct rcu_segcblist cblist; /* Segmented callback list, with */
203 /* different callbacks waiting for */
204 /* different grace periods. */
205 long qlen_last_fqs_check;
206 /* qlen at last check for QS forcing */
207 unsigned long n_force_qs_snap;
208 /* did other CPU force QS recently? */
209 long blimit; /* Upper limit on a processed batch */
210
211 /* 3) dynticks interface. */
212 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
213 int dynticks_snap; /* Per-GP tracking for dynticks. */
214
215 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
216 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
217 unsigned long offline_fqs; /* Kicked due to being offline. */
218 unsigned long cond_resched_completed;
219 /* Grace period that needs help */
220 /* from cond_resched(). */
221
222 /* 5) _rcu_barrier(), OOM callbacks, and expediting. */
223 struct rcu_head barrier_head;
224#ifdef CONFIG_RCU_FAST_NO_HZ
225 struct rcu_head oom_head;
226#endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
227 atomic_long_t exp_workdone0; /* # done by workqueue. */
228 atomic_long_t exp_workdone1; /* # done by others #1. */
229 atomic_long_t exp_workdone2; /* # done by others #2. */
230 atomic_long_t exp_workdone3; /* # done by others #3. */
231 int exp_dynticks_snap; /* Double-check need for IPI. */
232
233 /* 6) Callback offloading. */
234#ifdef CONFIG_RCU_NOCB_CPU
235 struct rcu_head *nocb_head; /* CBs waiting for kthread. */
236 struct rcu_head **nocb_tail;
237 atomic_long_t nocb_q_count; /* # CBs waiting for nocb */
238 atomic_long_t nocb_q_count_lazy; /* invocation (all stages). */
239 struct rcu_head *nocb_follower_head; /* CBs ready to invoke. */
240 struct rcu_head **nocb_follower_tail;
241 struct swait_queue_head nocb_wq; /* For nocb kthreads to sleep on. */
242 struct task_struct *nocb_kthread;
243 raw_spinlock_t nocb_lock; /* Guard following pair of fields. */
244 int nocb_defer_wakeup; /* Defer wakeup of nocb_kthread. */
245 struct timer_list nocb_timer; /* Enforce finite deferral. */
246
247 /* The following fields are used by the leader, hence own cacheline. */
248 struct rcu_head *nocb_gp_head ____cacheline_internodealigned_in_smp;
249 /* CBs waiting for GP. */
250 struct rcu_head **nocb_gp_tail;
251 bool nocb_leader_sleep; /* Is the nocb leader thread asleep? */
252 struct rcu_data *nocb_next_follower;
253 /* Next follower in wakeup chain. */
254
255 /* The following fields are used by the follower, hence new cachline. */
256 struct rcu_data *nocb_leader ____cacheline_internodealigned_in_smp;
257 /* Leader CPU takes GP-end wakeups. */
258#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
259
260 /* 7) RCU CPU stall data. */
261 unsigned int softirq_snap; /* Snapshot of softirq activity. */
262 /* ->rcu_iw* fields protected by leaf rcu_node ->lock. */
263 struct irq_work rcu_iw; /* Check for non-irq activity. */
264 bool rcu_iw_pending; /* Is ->rcu_iw pending? */
265 unsigned long rcu_iw_gpnum; /* ->gpnum associated with ->rcu_iw. */
266
267 int cpu;
268 struct rcu_state *rsp;
269};
270
271/* Values for nocb_defer_wakeup field in struct rcu_data. */
272#define RCU_NOCB_WAKE_NOT 0
273#define RCU_NOCB_WAKE 1
274#define RCU_NOCB_WAKE_FORCE 2
275
276#define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500))
277 /* For jiffies_till_first_fqs and */
278 /* and jiffies_till_next_fqs. */
279
280#define RCU_JIFFIES_FQS_DIV 256 /* Very large systems need more */
281 /* delay between bouts of */
282 /* quiescent-state forcing. */
283
284#define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time to take */
285 /* at least one scheduling clock */
286 /* irq before ratting on them. */
287
288#define rcu_wait(cond) \
289do { \
290 for (;;) { \
291 set_current_state(TASK_INTERRUPTIBLE); \
292 if (cond) \
293 break; \
294 schedule(); \
295 } \
296 __set_current_state(TASK_RUNNING); \
297} while (0)
298
299/*
300 * RCU global state, including node hierarchy. This hierarchy is
301 * represented in "heap" form in a dense array. The root (first level)
302 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
303 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
304 * and the third level in ->node[m+1] and following (->node[m+1] referenced
305 * by ->level[2]). The number of levels is determined by the number of
306 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
307 * consisting of a single rcu_node.
308 */
309struct rcu_state {
310 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
311 struct rcu_node *level[RCU_NUM_LVLS + 1];
312 /* Hierarchy levels (+1 to */
313 /* shut bogus gcc warning) */
314 struct rcu_data __percpu *rda; /* pointer of percu rcu_data. */
315 call_rcu_func_t call; /* call_rcu() flavor. */
316 int ncpus; /* # CPUs seen so far. */
317
318 /* The following fields are guarded by the root rcu_node's lock. */
319
320 u8 boost ____cacheline_internodealigned_in_smp;
321 /* Subject to priority boost. */
322 unsigned long gpnum; /* Current gp number. */
323 unsigned long completed; /* # of last completed gp. */
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
329 /* End of fields guarded by root rcu_node's lock. */
330
331 struct mutex barrier_mutex; /* Guards barrier fields. */
332 atomic_t barrier_cpu_count; /* # CPUs waiting on. */
333 struct completion barrier_completion; /* Wake at barrier end. */
334 unsigned long barrier_sequence; /* ++ at start and end of */
335 /* _rcu_barrier(). */
336 /* End of fields guarded by barrier_mutex. */
337
338 struct mutex exp_mutex; /* Serialize expedited GP. */
339 struct mutex exp_wake_mutex; /* Serialize wakeup. */
340 unsigned long expedited_sequence; /* Take a ticket. */
341 atomic_t expedited_need_qs; /* # CPUs left to check in. */
342 struct swait_queue_head expedited_wq; /* Wait for check-ins. */
343 int ncpus_snap; /* # CPUs seen last time. */
344
345 unsigned long jiffies_force_qs; /* Time at which to invoke */
346 /* force_quiescent_state(). */
347 unsigned long jiffies_kick_kthreads; /* Time at which to kick */
348 /* kthreads, if configured. */
349 unsigned long n_force_qs; /* Number of calls to */
350 /* force_quiescent_state(). */
351 unsigned long gp_start; /* Time at which GP started, */
352 /* but in jiffies. */
353 unsigned long gp_activity; /* Time of last GP kthread */
354 /* activity in jiffies. */
355 unsigned long jiffies_stall; /* Time at which to check */
356 /* for CPU stalls. */
357 unsigned long jiffies_resched; /* Time at which to resched */
358 /* a reluctant CPU. */
359 unsigned long n_force_qs_gpstart; /* Snapshot of n_force_qs at */
360 /* GP start. */
361 unsigned long gp_max; /* Maximum GP duration in */
362 /* jiffies. */
363 const char *name; /* Name of structure. */
364 char abbr; /* Abbreviated name. */
365 struct list_head flavors; /* List of RCU flavors. */
366};
367
368/* Values for rcu_state structure's gp_flags field. */
369#define RCU_GP_FLAG_INIT 0x1 /* Need grace-period initialization. */
370#define RCU_GP_FLAG_FQS 0x2 /* Need grace-period quiescent-state forcing. */
371
372/* Values for rcu_state structure's gp_state field. */
373#define RCU_GP_IDLE 0 /* Initial state and no GP in progress. */
374#define RCU_GP_WAIT_GPS 1 /* Wait for grace-period start. */
375#define RCU_GP_DONE_GPS 2 /* Wait done for grace-period start. */
376#define RCU_GP_WAIT_FQS 3 /* Wait for force-quiescent-state time. */
377#define RCU_GP_DOING_FQS 4 /* Wait done for force-quiescent-state time. */
378#define RCU_GP_CLEANUP 5 /* Grace-period cleanup started. */
379#define RCU_GP_CLEANED 6 /* Grace-period cleanup complete. */
380
381#ifndef RCU_TREE_NONCORE
382static const char * const gp_state_names[] = {
383 "RCU_GP_IDLE",
384 "RCU_GP_WAIT_GPS",
385 "RCU_GP_DONE_GPS",
386 "RCU_GP_WAIT_FQS",
387 "RCU_GP_DOING_FQS",
388 "RCU_GP_CLEANUP",
389 "RCU_GP_CLEANED",
390};
391#endif /* #ifndef RCU_TREE_NONCORE */
392
393extern struct list_head rcu_struct_flavors;
394
395/* Sequence through rcu_state structures for each RCU flavor. */
396#define for_each_rcu_flavor(rsp) \
397 list_for_each_entry((rsp), &rcu_struct_flavors, flavors)
398
399/*
400 * RCU implementation internal declarations:
401 */
402extern struct rcu_state rcu_sched_state;
403
404extern struct rcu_state rcu_bh_state;
405
406#ifdef CONFIG_PREEMPT_RCU
407extern struct rcu_state rcu_preempt_state;
408#endif /* #ifdef CONFIG_PREEMPT_RCU */
409
410int rcu_dynticks_snap(struct rcu_dynticks *rdtp);
411bool rcu_eqs_special_set(int cpu);
412
413#ifdef CONFIG_RCU_BOOST
414DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
415DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu);
416DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
417DECLARE_PER_CPU(char, rcu_cpu_has_work);
418#endif /* #ifdef CONFIG_RCU_BOOST */
419
420#ifndef RCU_TREE_NONCORE
421
422/* Forward declarations for rcutree_plugin.h */
423static void rcu_bootup_announce(void);
424static void rcu_preempt_note_context_switch(bool preempt);
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 void rcu_print_detail_task_stall(struct rcu_state *rsp);
430static int rcu_print_task_stall(struct rcu_node *rnp);
431static int rcu_print_task_exp_stall(struct rcu_node *rnp);
432static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
433static void rcu_preempt_check_callbacks(void);
434void call_rcu(struct rcu_head *head, rcu_callback_t func);
435static void __init __rcu_init_preempt(void);
436static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
437static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
438static void invoke_rcu_callbacks_kthread(void);
439static bool rcu_is_callbacks_kthread(void);
440#ifdef CONFIG_RCU_BOOST
441static void rcu_preempt_do_callbacks(void);
442static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
443 struct rcu_node *rnp);
444#endif /* #ifdef CONFIG_RCU_BOOST */
445static void __init rcu_spawn_boost_kthreads(void);
446static void rcu_prepare_kthreads(int cpu);
447static void rcu_cleanup_after_idle(void);
448static void rcu_prepare_for_idle(void);
449static void rcu_idle_count_callbacks_posted(void);
450static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
451static void print_cpu_stall_info_begin(void);
452static void print_cpu_stall_info(struct rcu_state *rsp, int cpu);
453static void print_cpu_stall_info_end(void);
454static void zero_cpu_stall_ticks(struct rcu_data *rdp);
455static void increment_cpu_stall_ticks(void);
456static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu);
457static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq);
458static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp);
459static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq);
460static void rcu_init_one_nocb(struct rcu_node *rnp);
461static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
462 bool lazy, unsigned long flags);
463static bool rcu_nocb_adopt_orphan_cbs(struct rcu_data *my_rdp,
464 struct rcu_data *rdp,
465 unsigned long flags);
466static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp);
467static void do_nocb_deferred_wakeup(struct rcu_data *rdp);
468static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp);
469static void rcu_spawn_all_nocb_kthreads(int cpu);
470static void __init rcu_spawn_nocb_kthreads(void);
471#ifdef CONFIG_RCU_NOCB_CPU
472static void __init rcu_organize_nocb_kthreads(struct rcu_state *rsp);
473#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
474static void __maybe_unused rcu_kick_nohz_cpu(int cpu);
475static bool init_nocb_callback_list(struct rcu_data *rdp);
476static void rcu_bind_gp_kthread(void);
477static bool rcu_nohz_full_cpu(struct rcu_state *rsp);
478static void rcu_dynticks_task_enter(void);
479static void rcu_dynticks_task_exit(void);
480
481#ifdef CONFIG_SRCU
482void srcu_online_cpu(unsigned int cpu);
483void srcu_offline_cpu(unsigned int cpu);
484#else /* #ifdef CONFIG_SRCU */
485void srcu_online_cpu(unsigned int cpu) { }
486void srcu_offline_cpu(unsigned int cpu) { }
487#endif /* #else #ifdef CONFIG_SRCU */
488
489#endif /* #ifndef RCU_TREE_NONCORE */