2 * Read-Copy Update mechanism for mutual exclusion (tree-based version)
3 * Internal non-public definitions.
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.
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.
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.
19 * Copyright IBM Corporation, 2008
21 * Author: Ingo Molnar <mingo@elte.hu>
22 * Paul E. McKenney <paulmck@linux.vnet.ibm.com>
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>
35 #include "rcu_segcblist.h"
38 * Dynticks per-CPU state.
41 long long dynticks_nesting
; /* Track irq/process nesting level. */
42 /* Process level is worth LLONG_MAX/2. */
43 int dynticks_nmi_nesting
; /* Track NMI nesting level. */
44 atomic_t dynticks
; /* Even value for idle, else odd. */
45 bool rcu_need_heavy_qs
; /* GP old, need heavy quiescent state. */
46 unsigned long rcu_qs_ctr
; /* Light universal quiescent state ctr. */
47 bool rcu_urgent_qs
; /* GP old need light quiescent state. */
48 #ifdef CONFIG_RCU_FAST_NO_HZ
49 bool all_lazy
; /* Are all CPU's CBs lazy? */
50 unsigned long nonlazy_posted
;
51 /* # times non-lazy CBs posted to CPU. */
52 unsigned long nonlazy_posted_snap
;
53 /* idle-period nonlazy_posted snapshot. */
54 unsigned long last_accelerate
;
55 /* Last jiffy CBs were accelerated. */
56 unsigned long last_advance_all
;
57 /* Last jiffy CBs were all advanced. */
58 int tick_nohz_enabled_snap
; /* Previously seen value from sysfs. */
59 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
62 /* RCU's kthread states for tracing. */
63 #define RCU_KTHREAD_STOPPED 0
64 #define RCU_KTHREAD_RUNNING 1
65 #define RCU_KTHREAD_WAITING 2
66 #define RCU_KTHREAD_OFFCPU 3
67 #define RCU_KTHREAD_YIELDING 4
68 #define RCU_KTHREAD_MAX 4
71 * Definition for node within the RCU grace-period-detection hierarchy.
74 raw_spinlock_t __private lock
; /* Root rcu_node's lock protects */
75 /* some rcu_state fields as well as */
77 unsigned long gpnum
; /* Current grace period for this node. */
78 /* This will either be equal to or one */
79 /* behind the root rcu_node's gpnum. */
80 unsigned long completed
; /* Last GP completed for this node. */
81 /* This will either be equal to or one */
82 /* behind the root rcu_node's gpnum. */
83 unsigned long qsmask
; /* CPUs or groups that need to switch in */
84 /* order for current grace period to proceed.*/
85 /* In leaf rcu_node, each bit corresponds to */
86 /* an rcu_data structure, otherwise, each */
87 /* bit corresponds to a child rcu_node */
89 unsigned long qsmaskinit
;
90 /* Per-GP initial value for qsmask. */
91 /* Initialized from ->qsmaskinitnext at the */
92 /* beginning of each grace period. */
93 unsigned long qsmaskinitnext
;
94 /* Online CPUs for next grace period. */
95 unsigned long expmask
; /* CPUs or groups that need to check in */
96 /* to allow the current expedited GP */
98 unsigned long expmaskinit
;
99 /* Per-GP initial values for expmask. */
100 /* Initialized from ->expmaskinitnext at the */
101 /* beginning of each expedited GP. */
102 unsigned long expmaskinitnext
;
103 /* Online CPUs for next expedited GP. */
104 /* Any CPU that has ever been online will */
105 /* have its bit set. */
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 */
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 unsigned long n_tasks_boosted
;
150 /* Total number of tasks boosted. */
151 unsigned long n_exp_boosts
;
152 /* Number of tasks boosted for expedited GP. */
153 unsigned long n_normal_boosts
;
154 /* Number of tasks boosted for normal GP. */
155 #ifdef CONFIG_RCU_NOCB_CPU
156 struct swait_queue_head nocb_gp_wq
[2];
157 /* Place for rcu_nocb_kthread() to wait GP. */
158 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
159 int need_future_gp
[2];
160 /* Counts of upcoming no-CB GP requests. */
161 raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp
;
163 spinlock_t exp_lock ____cacheline_internodealigned_in_smp
;
164 unsigned long exp_seq_rq
;
165 wait_queue_head_t exp_wq
[4];
166 } ____cacheline_internodealigned_in_smp
;
169 * Bitmasks in an rcu_node cover the interval [grplo, grphi] of CPU IDs, and
170 * are indexed relative to this interval rather than the global CPU ID space.
171 * This generates the bit for a CPU in node-local masks.
173 #define leaf_node_cpu_bit(rnp, cpu) (1UL << ((cpu) - (rnp)->grplo))
176 * Union to allow "aggregate OR" operation on the need for a quiescent
177 * state by the normal and expedited grace periods.
184 u16 s
; /* Set of bits, aggregate OR here. */
187 /* Index values for nxttail array in struct rcu_data. */
188 #define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
189 #define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
190 #define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
191 #define RCU_NEXT_TAIL 3
192 #define RCU_NEXT_SIZE 4
194 /* Per-CPU data for read-copy update. */
196 /* 1) quiescent-state and grace-period handling : */
197 unsigned long completed
; /* Track rsp->completed gp number */
198 /* in order to detect GP end. */
199 unsigned long gpnum
; /* Highest gp number that this CPU */
200 /* is aware of having started. */
201 unsigned long rcu_qs_ctr_snap
;/* Snapshot of rcu_qs_ctr to check */
202 /* for rcu_all_qs() invocations. */
203 union rcu_noqs cpu_no_qs
; /* No QSes yet for this CPU. */
204 bool core_needs_qs
; /* Core waits for quiesc state. */
205 bool beenonline
; /* CPU online at least once. */
206 bool gpwrap
; /* Possible gpnum/completed wrap. */
207 struct rcu_node
*mynode
; /* This CPU's leaf of hierarchy */
208 unsigned long grpmask
; /* Mask to apply to leaf qsmask. */
209 unsigned long ticks_this_gp
; /* The number of scheduling-clock */
210 /* ticks this CPU has handled */
211 /* during and after the last grace */
212 /* period it is aware of. */
214 /* 2) batch handling */
215 struct rcu_segcblist cblist
; /* Segmented callback list, with */
216 /* different callbacks waiting for */
217 /* different grace periods. */
218 long qlen_last_fqs_check
;
219 /* qlen at last check for QS forcing */
220 unsigned long n_cbs_invoked
; /* count of RCU cbs invoked. */
221 unsigned long n_nocbs_invoked
; /* count of no-CBs RCU cbs invoked. */
222 unsigned long n_force_qs_snap
;
223 /* did other CPU force QS recently? */
224 long blimit
; /* Upper limit on a processed batch */
226 /* 3) dynticks interface. */
227 struct rcu_dynticks
*dynticks
; /* Shared per-CPU dynticks state. */
228 int dynticks_snap
; /* Per-GP tracking for dynticks. */
230 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
231 unsigned long dynticks_fqs
; /* Kicked due to dynticks idle. */
232 unsigned long offline_fqs
; /* Kicked due to being offline. */
233 unsigned long cond_resched_completed
;
234 /* Grace period that needs help */
235 /* from cond_resched(). */
237 /* 5) __rcu_pending() statistics. */
238 unsigned long n_rcu_pending
; /* rcu_pending() calls since boot. */
239 unsigned long n_rp_core_needs_qs
;
240 unsigned long n_rp_report_qs
;
241 unsigned long n_rp_cb_ready
;
242 unsigned long n_rp_cpu_needs_gp
;
243 unsigned long n_rp_gp_completed
;
244 unsigned long n_rp_gp_started
;
245 unsigned long n_rp_nocb_defer_wakeup
;
246 unsigned long n_rp_need_nothing
;
248 /* 6) _rcu_barrier(), OOM callbacks, and expediting. */
249 struct rcu_head barrier_head
;
250 #ifdef CONFIG_RCU_FAST_NO_HZ
251 struct rcu_head oom_head
;
252 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
253 atomic_long_t exp_workdone0
; /* # done by workqueue. */
254 atomic_long_t exp_workdone1
; /* # done by others #1. */
255 atomic_long_t exp_workdone2
; /* # done by others #2. */
256 atomic_long_t exp_workdone3
; /* # done by others #3. */
257 int exp_dynticks_snap
; /* Double-check need for IPI. */
259 /* 7) Callback offloading. */
260 #ifdef CONFIG_RCU_NOCB_CPU
261 struct rcu_head
*nocb_head
; /* CBs waiting for kthread. */
262 struct rcu_head
**nocb_tail
;
263 atomic_long_t nocb_q_count
; /* # CBs waiting for nocb */
264 atomic_long_t nocb_q_count_lazy
; /* invocation (all stages). */
265 struct rcu_head
*nocb_follower_head
; /* CBs ready to invoke. */
266 struct rcu_head
**nocb_follower_tail
;
267 struct swait_queue_head nocb_wq
; /* For nocb kthreads to sleep on. */
268 struct task_struct
*nocb_kthread
;
269 raw_spinlock_t nocb_lock
; /* Guard following pair of fields. */
270 int nocb_defer_wakeup
; /* Defer wakeup of nocb_kthread. */
271 struct timer_list nocb_timer
; /* Enforce finite deferral. */
273 /* The following fields are used by the leader, hence own cacheline. */
274 struct rcu_head
*nocb_gp_head ____cacheline_internodealigned_in_smp
;
275 /* CBs waiting for GP. */
276 struct rcu_head
**nocb_gp_tail
;
277 bool nocb_leader_sleep
; /* Is the nocb leader thread asleep? */
278 struct rcu_data
*nocb_next_follower
;
279 /* Next follower in wakeup chain. */
281 /* The following fields are used by the follower, hence new cachline. */
282 struct rcu_data
*nocb_leader ____cacheline_internodealigned_in_smp
;
283 /* Leader CPU takes GP-end wakeups. */
284 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
286 /* 8) RCU CPU stall data. */
287 unsigned int softirq_snap
; /* Snapshot of softirq activity. */
290 struct rcu_state
*rsp
;
293 /* Values for nocb_defer_wakeup field in struct rcu_data. */
294 #define RCU_NOCB_WAKE_NOT 0
295 #define RCU_NOCB_WAKE 1
296 #define RCU_NOCB_WAKE_FORCE 2
298 #define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500))
299 /* For jiffies_till_first_fqs and */
300 /* and jiffies_till_next_fqs. */
302 #define RCU_JIFFIES_FQS_DIV 256 /* Very large systems need more */
303 /* delay between bouts of */
304 /* quiescent-state forcing. */
306 #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time to take */
307 /* at least one scheduling clock */
308 /* irq before ratting on them. */
310 #define rcu_wait(cond) \
313 set_current_state(TASK_INTERRUPTIBLE); \
318 __set_current_state(TASK_RUNNING); \
322 * RCU global state, including node hierarchy. This hierarchy is
323 * represented in "heap" form in a dense array. The root (first level)
324 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
325 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
326 * and the third level in ->node[m+1] and following (->node[m+1] referenced
327 * by ->level[2]). The number of levels is determined by the number of
328 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
329 * consisting of a single rcu_node.
332 struct rcu_node node
[NUM_RCU_NODES
]; /* Hierarchy. */
333 struct rcu_node
*level
[RCU_NUM_LVLS
+ 1];
334 /* Hierarchy levels (+1 to */
335 /* shut bogus gcc warning) */
336 struct rcu_data __percpu
*rda
; /* pointer of percu rcu_data. */
337 call_rcu_func_t call
; /* call_rcu() flavor. */
338 int ncpus
; /* # CPUs seen so far. */
340 /* The following fields are guarded by the root rcu_node's lock. */
342 u8 boost ____cacheline_internodealigned_in_smp
;
343 /* Subject to priority boost. */
344 unsigned long gpnum
; /* Current gp number. */
345 unsigned long completed
; /* # of last completed gp. */
346 struct task_struct
*gp_kthread
; /* Task for grace periods. */
347 struct swait_queue_head gp_wq
; /* Where GP task waits. */
348 short gp_flags
; /* Commands for GP task. */
349 short gp_state
; /* GP kthread sleep state. */
351 /* End of fields guarded by root rcu_node's lock. */
353 struct mutex barrier_mutex
; /* Guards barrier fields. */
354 atomic_t barrier_cpu_count
; /* # CPUs waiting on. */
355 struct completion barrier_completion
; /* Wake at barrier end. */
356 unsigned long barrier_sequence
; /* ++ at start and end of */
357 /* _rcu_barrier(). */
358 /* End of fields guarded by barrier_mutex. */
360 struct mutex exp_mutex
; /* Serialize expedited GP. */
361 struct mutex exp_wake_mutex
; /* Serialize wakeup. */
362 unsigned long expedited_sequence
; /* Take a ticket. */
363 atomic_t expedited_need_qs
; /* # CPUs left to check in. */
364 struct swait_queue_head expedited_wq
; /* Wait for check-ins. */
365 int ncpus_snap
; /* # CPUs seen last time. */
367 unsigned long jiffies_force_qs
; /* Time at which to invoke */
368 /* force_quiescent_state(). */
369 unsigned long jiffies_kick_kthreads
; /* Time at which to kick */
370 /* kthreads, if configured. */
371 unsigned long n_force_qs
; /* Number of calls to */
372 /* force_quiescent_state(). */
373 unsigned long n_force_qs_lh
; /* ~Number of calls leaving */
374 /* due to lock unavailable. */
375 unsigned long n_force_qs_ngp
; /* Number of calls leaving */
376 /* due to no GP active. */
377 unsigned long gp_start
; /* Time at which GP started, */
378 /* but in jiffies. */
379 unsigned long gp_activity
; /* Time of last GP kthread */
380 /* activity in jiffies. */
381 unsigned long jiffies_stall
; /* Time at which to check */
382 /* for CPU stalls. */
383 unsigned long jiffies_resched
; /* Time at which to resched */
384 /* a reluctant CPU. */
385 unsigned long n_force_qs_gpstart
; /* Snapshot of n_force_qs at */
387 unsigned long gp_max
; /* Maximum GP duration in */
389 const char *name
; /* Name of structure. */
390 char abbr
; /* Abbreviated name. */
391 struct list_head flavors
; /* List of RCU flavors. */
394 /* Values for rcu_state structure's gp_flags field. */
395 #define RCU_GP_FLAG_INIT 0x1 /* Need grace-period initialization. */
396 #define RCU_GP_FLAG_FQS 0x2 /* Need grace-period quiescent-state forcing. */
398 /* Values for rcu_state structure's gp_state field. */
399 #define RCU_GP_IDLE 0 /* Initial state and no GP in progress. */
400 #define RCU_GP_WAIT_GPS 1 /* Wait for grace-period start. */
401 #define RCU_GP_DONE_GPS 2 /* Wait done for grace-period start. */
402 #define RCU_GP_WAIT_FQS 3 /* Wait for force-quiescent-state time. */
403 #define RCU_GP_DOING_FQS 4 /* Wait done for force-quiescent-state time. */
404 #define RCU_GP_CLEANUP 5 /* Grace-period cleanup started. */
405 #define RCU_GP_CLEANED 6 /* Grace-period cleanup complete. */
407 #ifndef RCU_TREE_NONCORE
408 static const char * const gp_state_names
[] = {
417 #endif /* #ifndef RCU_TREE_NONCORE */
419 extern struct list_head rcu_struct_flavors
;
421 /* Sequence through rcu_state structures for each RCU flavor. */
422 #define for_each_rcu_flavor(rsp) \
423 list_for_each_entry((rsp), &rcu_struct_flavors, flavors)
426 * RCU implementation internal declarations:
428 extern struct rcu_state rcu_sched_state
;
430 extern struct rcu_state rcu_bh_state
;
432 #ifdef CONFIG_PREEMPT_RCU
433 extern struct rcu_state rcu_preempt_state
;
434 #endif /* #ifdef CONFIG_PREEMPT_RCU */
436 int rcu_dynticks_snap(struct rcu_dynticks
*rdtp
);
437 bool rcu_eqs_special_set(int cpu
);
439 #ifdef CONFIG_RCU_BOOST
440 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status
);
441 DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu
);
442 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops
);
443 DECLARE_PER_CPU(char, rcu_cpu_has_work
);
444 #endif /* #ifdef CONFIG_RCU_BOOST */
446 #ifndef RCU_TREE_NONCORE
448 /* Forward declarations for rcutree_plugin.h */
449 static void rcu_bootup_announce(void);
450 static void rcu_preempt_note_context_switch(bool preempt
);
451 static int rcu_preempt_blocked_readers_cgp(struct rcu_node
*rnp
);
452 #ifdef CONFIG_HOTPLUG_CPU
453 static bool rcu_preempt_has_tasks(struct rcu_node
*rnp
);
454 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
455 static void rcu_print_detail_task_stall(struct rcu_state
*rsp
);
456 static int rcu_print_task_stall(struct rcu_node
*rnp
);
457 static int rcu_print_task_exp_stall(struct rcu_node
*rnp
);
458 static void rcu_preempt_check_blocked_tasks(struct rcu_node
*rnp
);
459 static void rcu_preempt_check_callbacks(void);
460 void call_rcu(struct rcu_head
*head
, rcu_callback_t func
);
461 static void __init
__rcu_init_preempt(void);
462 static void rcu_initiate_boost(struct rcu_node
*rnp
, unsigned long flags
);
463 static void rcu_preempt_boost_start_gp(struct rcu_node
*rnp
);
464 static void invoke_rcu_callbacks_kthread(void);
465 static bool rcu_is_callbacks_kthread(void);
466 #ifdef CONFIG_RCU_BOOST
467 static void rcu_preempt_do_callbacks(void);
468 static int rcu_spawn_one_boost_kthread(struct rcu_state
*rsp
,
469 struct rcu_node
*rnp
);
470 #endif /* #ifdef CONFIG_RCU_BOOST */
471 static void __init
rcu_spawn_boost_kthreads(void);
472 static void rcu_prepare_kthreads(int cpu
);
473 static void rcu_cleanup_after_idle(void);
474 static void rcu_prepare_for_idle(void);
475 static void rcu_idle_count_callbacks_posted(void);
476 static bool rcu_preempt_has_tasks(struct rcu_node
*rnp
);
477 static void print_cpu_stall_info_begin(void);
478 static void print_cpu_stall_info(struct rcu_state
*rsp
, int cpu
);
479 static void print_cpu_stall_info_end(void);
480 static void zero_cpu_stall_ticks(struct rcu_data
*rdp
);
481 static void increment_cpu_stall_ticks(void);
482 static bool rcu_nocb_cpu_needs_barrier(struct rcu_state
*rsp
, int cpu
);
483 static void rcu_nocb_gp_set(struct rcu_node
*rnp
, int nrq
);
484 static struct swait_queue_head
*rcu_nocb_gp_get(struct rcu_node
*rnp
);
485 static void rcu_nocb_gp_cleanup(struct swait_queue_head
*sq
);
486 static void rcu_init_one_nocb(struct rcu_node
*rnp
);
487 static bool __call_rcu_nocb(struct rcu_data
*rdp
, struct rcu_head
*rhp
,
488 bool lazy
, unsigned long flags
);
489 static bool rcu_nocb_adopt_orphan_cbs(struct rcu_data
*my_rdp
,
490 struct rcu_data
*rdp
,
491 unsigned long flags
);
492 static int rcu_nocb_need_deferred_wakeup(struct rcu_data
*rdp
);
493 static void do_nocb_deferred_wakeup(struct rcu_data
*rdp
);
494 static void rcu_boot_init_nocb_percpu_data(struct rcu_data
*rdp
);
495 static void rcu_spawn_all_nocb_kthreads(int cpu
);
496 static void __init
rcu_spawn_nocb_kthreads(void);
497 #ifdef CONFIG_RCU_NOCB_CPU
498 static void __init
rcu_organize_nocb_kthreads(struct rcu_state
*rsp
);
499 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
500 static void __maybe_unused
rcu_kick_nohz_cpu(int cpu
);
501 static bool init_nocb_callback_list(struct rcu_data
*rdp
);
502 static void rcu_bind_gp_kthread(void);
503 static bool rcu_nohz_full_cpu(struct rcu_state
*rsp
);
504 static void rcu_dynticks_task_enter(void);
505 static void rcu_dynticks_task_exit(void);
508 void srcu_online_cpu(unsigned int cpu
);
509 void srcu_offline_cpu(unsigned int cpu
);
510 #else /* #ifdef CONFIG_SRCU */
511 void srcu_online_cpu(unsigned int cpu
) { }
512 void srcu_offline_cpu(unsigned int cpu
) { }
513 #endif /* #else #ifdef CONFIG_SRCU */
515 #endif /* #ifndef RCU_TREE_NONCORE */