1 /* SPDX-License-Identifier: GPL-2.0+ */
3 * Read-Copy Update mechanism for mutual exclusion (tree-based version)
4 * Internal non-public definitions.
6 * Copyright IBM Corporation, 2008
8 * Author: Ingo Molnar <mingo@elte.hu>
9 * Paul E. McKenney <paulmck@linux.ibm.com>
12 #include <linux/cache.h>
13 #include <linux/spinlock.h>
14 #include <linux/rtmutex.h>
15 #include <linux/threads.h>
16 #include <linux/cpumask.h>
17 #include <linux/seqlock.h>
18 #include <linux/swait.h>
19 #include <linux/rcu_node_tree.h>
21 #include "rcu_segcblist.h"
23 /* Communicate arguments to a workqueue handler. */
26 struct work_struct rew_work
;
29 /* RCU's kthread states for tracing. */
30 #define RCU_KTHREAD_STOPPED 0
31 #define RCU_KTHREAD_RUNNING 1
32 #define RCU_KTHREAD_WAITING 2
33 #define RCU_KTHREAD_OFFCPU 3
34 #define RCU_KTHREAD_YIELDING 4
35 #define RCU_KTHREAD_MAX 4
38 * Definition for node within the RCU grace-period-detection hierarchy.
41 raw_spinlock_t __private lock
; /* Root rcu_node's lock protects */
42 /* some rcu_state fields as well as */
44 unsigned long gp_seq
; /* Track rsp->gp_seq. */
45 unsigned long gp_seq_needed
; /* Track furthest future GP request. */
46 unsigned long completedqs
; /* All QSes done for this node. */
47 unsigned long qsmask
; /* CPUs or groups that need to switch in */
48 /* order for current grace period to proceed.*/
49 /* In leaf rcu_node, each bit corresponds to */
50 /* an rcu_data structure, otherwise, each */
51 /* bit corresponds to a child rcu_node */
53 unsigned long rcu_gp_init_mask
; /* Mask of offline CPUs at GP init. */
54 unsigned long qsmaskinit
;
55 /* Per-GP initial value for qsmask. */
56 /* Initialized from ->qsmaskinitnext at the */
57 /* beginning of each grace period. */
58 unsigned long qsmaskinitnext
;
59 unsigned long ofl_seq
; /* CPU-hotplug operation sequence count. */
60 /* Online CPUs for next grace period. */
61 unsigned long expmask
; /* CPUs or groups that need to check in */
62 /* to allow the current expedited GP */
64 unsigned long expmaskinit
;
65 /* Per-GP initial values for expmask. */
66 /* Initialized from ->expmaskinitnext at the */
67 /* beginning of each expedited GP. */
68 unsigned long expmaskinitnext
;
69 /* Online CPUs for next expedited GP. */
70 /* Any CPU that has ever been online will */
71 /* have its bit set. */
72 unsigned long cbovldmask
;
73 /* CPUs experiencing callback overload. */
74 unsigned long ffmask
; /* Fully functional CPUs. */
75 unsigned long grpmask
; /* Mask to apply to parent qsmask. */
76 /* Only one bit will be set in this mask. */
77 int grplo
; /* lowest-numbered CPU here. */
78 int grphi
; /* highest-numbered CPU here. */
79 u8 grpnum
; /* group number for next level up. */
80 u8 level
; /* root is at level 0. */
81 bool wait_blkd_tasks
;/* Necessary to wait for blocked tasks to */
82 /* exit RCU read-side critical sections */
83 /* before propagating offline up the */
85 struct rcu_node
*parent
;
86 struct list_head blkd_tasks
;
87 /* Tasks blocked in RCU read-side critical */
88 /* section. Tasks are placed at the head */
89 /* of this list and age towards the tail. */
90 struct list_head
*gp_tasks
;
91 /* Pointer to the first task blocking the */
92 /* current grace period, or NULL if there */
93 /* is no such task. */
94 struct list_head
*exp_tasks
;
95 /* Pointer to the first task blocking the */
96 /* current expedited grace period, or NULL */
97 /* if there is no such task. If there */
98 /* is no current expedited grace period, */
99 /* then there can cannot be any such task. */
100 struct list_head
*boost_tasks
;
101 /* Pointer to first task that needs to be */
102 /* priority boosted, or NULL if no priority */
103 /* boosting is needed for this rcu_node */
104 /* structure. If there are no tasks */
105 /* queued on this rcu_node structure that */
106 /* are blocking the current grace period, */
107 /* there can be no such task. */
108 struct rt_mutex boost_mtx
;
109 /* Used only for the priority-boosting */
110 /* side effect, not as a lock. */
111 unsigned long boost_time
;
112 /* When to start boosting (jiffies). */
113 struct task_struct
*boost_kthread_task
;
114 /* kthread that takes care of priority */
115 /* boosting for this rcu_node structure. */
116 unsigned int boost_kthread_status
;
117 /* State of boost_kthread_task for tracing. */
118 #ifdef CONFIG_RCU_NOCB_CPU
119 struct swait_queue_head nocb_gp_wq
[2];
120 /* Place for rcu_nocb_kthread() to wait GP. */
121 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
122 raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp
;
124 spinlock_t exp_lock ____cacheline_internodealigned_in_smp
;
125 unsigned long exp_seq_rq
;
126 wait_queue_head_t exp_wq
[4];
127 struct rcu_exp_work rew
;
128 bool exp_need_flush
; /* Need to flush workitem? */
129 } ____cacheline_internodealigned_in_smp
;
132 * Bitmasks in an rcu_node cover the interval [grplo, grphi] of CPU IDs, and
133 * are indexed relative to this interval rather than the global CPU ID space.
134 * This generates the bit for a CPU in node-local masks.
136 #define leaf_node_cpu_bit(rnp, cpu) (BIT((cpu) - (rnp)->grplo))
139 * Union to allow "aggregate OR" operation on the need for a quiescent
140 * state by the normal and expedited grace periods.
147 u16 s
; /* Set of bits, aggregate OR here. */
150 /* Per-CPU data for read-copy update. */
152 /* 1) quiescent-state and grace-period handling : */
153 unsigned long gp_seq
; /* Track rsp->gp_seq counter. */
154 unsigned long gp_seq_needed
; /* Track furthest future GP request. */
155 union rcu_noqs cpu_no_qs
; /* No QSes yet for this CPU. */
156 bool core_needs_qs
; /* Core waits for quiesc state. */
157 bool beenonline
; /* CPU online at least once. */
158 bool gpwrap
; /* Possible ->gp_seq wrap. */
159 bool exp_deferred_qs
; /* This CPU awaiting a deferred QS? */
160 bool cpu_started
; /* RCU watching this onlining CPU. */
161 struct rcu_node
*mynode
; /* This CPU's leaf of hierarchy */
162 unsigned long grpmask
; /* Mask to apply to leaf qsmask. */
163 unsigned long ticks_this_gp
; /* The number of scheduling-clock */
164 /* ticks this CPU has handled */
165 /* during and after the last grace */
166 /* period it is aware of. */
167 struct irq_work defer_qs_iw
; /* Obtain later scheduler attention. */
168 bool defer_qs_iw_pending
; /* Scheduler attention pending? */
169 struct work_struct strict_work
; /* Schedule readers for strict GPs. */
171 /* 2) batch handling */
172 struct rcu_segcblist cblist
; /* Segmented callback list, with */
173 /* different callbacks waiting for */
174 /* different grace periods. */
175 long qlen_last_fqs_check
;
176 /* qlen at last check for QS forcing */
177 unsigned long n_cbs_invoked
; /* # callbacks invoked since boot. */
178 unsigned long n_force_qs_snap
;
179 /* did other CPU force QS recently? */
180 long blimit
; /* Upper limit on a processed batch */
182 /* 3) dynticks interface. */
183 int dynticks_snap
; /* Per-GP tracking for dynticks. */
184 long dynticks_nesting
; /* Track process nesting level. */
185 long dynticks_nmi_nesting
; /* Track irq/NMI nesting level. */
186 atomic_t dynticks
; /* Even value for idle, else odd. */
187 bool rcu_need_heavy_qs
; /* GP old, so heavy quiescent state! */
188 bool rcu_urgent_qs
; /* GP old need light quiescent state. */
189 bool rcu_forced_tick
; /* Forced tick to provide QS. */
190 bool rcu_forced_tick_exp
; /* ... provide QS to expedited GP. */
191 #ifdef CONFIG_RCU_FAST_NO_HZ
192 unsigned long last_accelerate
; /* Last jiffy CBs were accelerated. */
193 unsigned long last_advance_all
; /* Last jiffy CBs were all advanced. */
194 int tick_nohz_enabled_snap
; /* Previously seen value from sysfs. */
195 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
197 /* 4) rcu_barrier(), OOM callbacks, and expediting. */
198 struct rcu_head barrier_head
;
199 int exp_dynticks_snap
; /* Double-check need for IPI. */
201 /* 5) Callback offloading. */
202 #ifdef CONFIG_RCU_NOCB_CPU
203 struct swait_queue_head nocb_cb_wq
; /* For nocb kthreads to sleep on. */
204 struct task_struct
*nocb_gp_kthread
;
205 raw_spinlock_t nocb_lock
; /* Guard following pair of fields. */
206 atomic_t nocb_lock_contended
; /* Contention experienced. */
207 int nocb_defer_wakeup
; /* Defer wakeup of nocb_kthread. */
208 struct timer_list nocb_timer
; /* Enforce finite deferral. */
209 unsigned long nocb_gp_adv_time
; /* Last call_rcu() CB adv (jiffies). */
211 /* The following fields are used by call_rcu, hence own cacheline. */
212 raw_spinlock_t nocb_bypass_lock ____cacheline_internodealigned_in_smp
;
213 struct rcu_cblist nocb_bypass
; /* Lock-contention-bypass CB list. */
214 unsigned long nocb_bypass_first
; /* Time (jiffies) of first enqueue. */
215 unsigned long nocb_nobypass_last
; /* Last ->cblist enqueue (jiffies). */
216 int nocb_nobypass_count
; /* # ->cblist enqueues at ^^^ time. */
218 /* The following fields are used by GP kthread, hence own cacheline. */
219 raw_spinlock_t nocb_gp_lock ____cacheline_internodealigned_in_smp
;
220 struct timer_list nocb_bypass_timer
; /* Force nocb_bypass flush. */
221 u8 nocb_gp_sleep
; /* Is the nocb GP thread asleep? */
222 u8 nocb_gp_bypass
; /* Found a bypass on last scan? */
223 u8 nocb_gp_gp
; /* GP to wait for on last scan? */
224 unsigned long nocb_gp_seq
; /* If so, ->gp_seq to wait for. */
225 unsigned long nocb_gp_loops
; /* # passes through wait code. */
226 struct swait_queue_head nocb_gp_wq
; /* For nocb kthreads to sleep on. */
227 bool nocb_cb_sleep
; /* Is the nocb CB thread asleep? */
228 struct task_struct
*nocb_cb_kthread
;
229 struct rcu_data
*nocb_next_cb_rdp
;
230 /* Next rcu_data in wakeup chain. */
232 /* The following fields are used by CB kthread, hence new cacheline. */
233 struct rcu_data
*nocb_gp_rdp ____cacheline_internodealigned_in_smp
;
234 /* GP rdp takes GP-end wakeups. */
235 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
237 /* 6) RCU priority boosting. */
238 struct task_struct
*rcu_cpu_kthread_task
;
239 /* rcuc per-CPU kthread or NULL. */
240 unsigned int rcu_cpu_kthread_status
;
241 char rcu_cpu_has_work
;
243 /* 7) Diagnostic data, including RCU CPU stall warnings. */
244 unsigned int softirq_snap
; /* Snapshot of softirq activity. */
245 /* ->rcu_iw* fields protected by leaf rcu_node ->lock. */
246 struct irq_work rcu_iw
; /* Check for non-irq activity. */
247 bool rcu_iw_pending
; /* Is ->rcu_iw pending? */
248 unsigned long rcu_iw_gp_seq
; /* ->gp_seq associated with ->rcu_iw. */
249 unsigned long rcu_ofl_gp_seq
; /* ->gp_seq at last offline. */
250 short rcu_ofl_gp_flags
; /* ->gp_flags at last offline. */
251 unsigned long rcu_onl_gp_seq
; /* ->gp_seq at last online. */
252 short rcu_onl_gp_flags
; /* ->gp_flags at last online. */
253 unsigned long last_fqs_resched
; /* Time of last rcu_resched(). */
258 /* Values for nocb_defer_wakeup field in struct rcu_data. */
259 #define RCU_NOCB_WAKE_NOT 0
260 #define RCU_NOCB_WAKE 1
261 #define RCU_NOCB_WAKE_FORCE 2
263 #define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500))
264 /* For jiffies_till_first_fqs and */
265 /* and jiffies_till_next_fqs. */
267 #define RCU_JIFFIES_FQS_DIV 256 /* Very large systems need more */
268 /* delay between bouts of */
269 /* quiescent-state forcing. */
271 #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time to take */
272 /* at least one scheduling clock */
273 /* irq before ratting on them. */
275 #define rcu_wait(cond) \
278 set_current_state(TASK_INTERRUPTIBLE); \
283 __set_current_state(TASK_RUNNING); \
287 * RCU global state, including node hierarchy. This hierarchy is
288 * represented in "heap" form in a dense array. The root (first level)
289 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
290 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
291 * and the third level in ->node[m+1] and following (->node[m+1] referenced
292 * by ->level[2]). The number of levels is determined by the number of
293 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
294 * consisting of a single rcu_node.
297 struct rcu_node node
[NUM_RCU_NODES
]; /* Hierarchy. */
298 struct rcu_node
*level
[RCU_NUM_LVLS
+ 1];
299 /* Hierarchy levels (+1 to */
300 /* shut bogus gcc warning) */
301 int ncpus
; /* # CPUs seen so far. */
302 int n_online_cpus
; /* # CPUs online for RCU. */
304 /* The following fields are guarded by the root rcu_node's lock. */
306 u8 boost ____cacheline_internodealigned_in_smp
;
307 /* Subject to priority boost. */
308 unsigned long gp_seq
; /* Grace-period sequence #. */
309 unsigned long gp_max
; /* Maximum GP duration in */
311 struct task_struct
*gp_kthread
; /* Task for grace periods. */
312 struct swait_queue_head gp_wq
; /* Where GP task waits. */
313 short gp_flags
; /* Commands for GP task. */
314 short gp_state
; /* GP kthread sleep state. */
315 unsigned long gp_wake_time
; /* Last GP kthread wake. */
316 unsigned long gp_wake_seq
; /* ->gp_seq at ^^^. */
318 /* End of fields guarded by root rcu_node's lock. */
320 struct mutex barrier_mutex
; /* Guards barrier fields. */
321 atomic_t barrier_cpu_count
; /* # CPUs waiting on. */
322 struct completion barrier_completion
; /* Wake at barrier end. */
323 unsigned long barrier_sequence
; /* ++ at start and end of */
325 /* End of fields guarded by barrier_mutex. */
327 struct mutex exp_mutex
; /* Serialize expedited GP. */
328 struct mutex exp_wake_mutex
; /* Serialize wakeup. */
329 unsigned long expedited_sequence
; /* Take a ticket. */
330 atomic_t expedited_need_qs
; /* # CPUs left to check in. */
331 struct swait_queue_head expedited_wq
; /* Wait for check-ins. */
332 int ncpus_snap
; /* # CPUs seen last time. */
333 u8 cbovld
; /* Callback overload now? */
334 u8 cbovldnext
; /* ^ ^ next time? */
336 unsigned long jiffies_force_qs
; /* Time at which to invoke */
337 /* force_quiescent_state(). */
338 unsigned long jiffies_kick_kthreads
; /* Time at which to kick */
339 /* kthreads, if configured. */
340 unsigned long n_force_qs
; /* Number of calls to */
341 /* force_quiescent_state(). */
342 unsigned long gp_start
; /* Time at which GP started, */
343 /* but in jiffies. */
344 unsigned long gp_end
; /* Time last GP ended, again */
346 unsigned long gp_activity
; /* Time of last GP kthread */
347 /* activity in jiffies. */
348 unsigned long gp_req_activity
; /* Time of last GP request */
350 unsigned long jiffies_stall
; /* Time at which to check */
351 /* for CPU stalls. */
352 unsigned long jiffies_resched
; /* Time at which to resched */
353 /* a reluctant CPU. */
354 unsigned long n_force_qs_gpstart
; /* Snapshot of n_force_qs at */
356 const char *name
; /* Name of structure. */
357 char abbr
; /* Abbreviated name. */
359 raw_spinlock_t ofl_lock ____cacheline_internodealigned_in_smp
;
360 /* Synchronize offline with */
361 /* GP pre-initialization. */
364 /* Values for rcu_state structure's gp_flags field. */
365 #define RCU_GP_FLAG_INIT 0x1 /* Need grace-period initialization. */
366 #define RCU_GP_FLAG_FQS 0x2 /* Need grace-period quiescent-state forcing. */
367 #define RCU_GP_FLAG_OVLD 0x4 /* Experiencing callback overload. */
369 /* Values for rcu_state structure's gp_state field. */
370 #define RCU_GP_IDLE 0 /* Initial state and no GP in progress. */
371 #define RCU_GP_WAIT_GPS 1 /* Wait for grace-period start. */
372 #define RCU_GP_DONE_GPS 2 /* Wait done for grace-period start. */
373 #define RCU_GP_ONOFF 3 /* Grace-period initialization hotplug. */
374 #define RCU_GP_INIT 4 /* Grace-period initialization. */
375 #define RCU_GP_WAIT_FQS 5 /* Wait for force-quiescent-state time. */
376 #define RCU_GP_DOING_FQS 6 /* Wait done for force-quiescent-state time. */
377 #define RCU_GP_CLEANUP 7 /* Grace-period cleanup started. */
378 #define RCU_GP_CLEANED 8 /* Grace-period cleanup complete. */
381 * In order to export the rcu_state name to the tracing tools, it
382 * needs to be added in the __tracepoint_string section.
383 * This requires defining a separate variable tp_<sname>_varname
384 * that points to the string being used, and this will allow
385 * the tracing userspace tools to be able to decipher the string
386 * address to the matching string.
388 #ifdef CONFIG_PREEMPT_RCU
390 #define RCU_NAME_RAW "rcu_preempt"
391 #else /* #ifdef CONFIG_PREEMPT_RCU */
393 #define RCU_NAME_RAW "rcu_sched"
394 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
395 #ifndef CONFIG_TRACING
396 #define RCU_NAME RCU_NAME_RAW
397 #else /* #ifdef CONFIG_TRACING */
398 static char rcu_name
[] = RCU_NAME_RAW
;
399 static const char *tp_rcu_varname __used __tracepoint_string
= rcu_name
;
400 #define RCU_NAME rcu_name
401 #endif /* #else #ifdef CONFIG_TRACING */
403 /* Forward declarations for tree_plugin.h */
404 static void rcu_bootup_announce(void);
405 static void rcu_qs(void);
406 static int rcu_preempt_blocked_readers_cgp(struct rcu_node
*rnp
);
407 #ifdef CONFIG_HOTPLUG_CPU
408 static bool rcu_preempt_has_tasks(struct rcu_node
*rnp
);
409 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
410 static int rcu_print_task_exp_stall(struct rcu_node
*rnp
);
411 static void rcu_preempt_check_blocked_tasks(struct rcu_node
*rnp
);
412 static void rcu_flavor_sched_clock_irq(int user
);
413 static void dump_blkd_tasks(struct rcu_node
*rnp
, int ncheck
);
414 static void rcu_initiate_boost(struct rcu_node
*rnp
, unsigned long flags
);
415 static void rcu_preempt_boost_start_gp(struct rcu_node
*rnp
);
416 static bool rcu_is_callbacks_kthread(void);
417 static void rcu_cpu_kthread_setup(unsigned int cpu
);
418 static void __init
rcu_spawn_boost_kthreads(void);
419 static void rcu_prepare_kthreads(int cpu
);
420 static void rcu_cleanup_after_idle(void);
421 static void rcu_prepare_for_idle(void);
422 static bool rcu_preempt_has_tasks(struct rcu_node
*rnp
);
423 static bool rcu_preempt_need_deferred_qs(struct task_struct
*t
);
424 static void rcu_preempt_deferred_qs(struct task_struct
*t
);
425 static void zero_cpu_stall_ticks(struct rcu_data
*rdp
);
426 static struct swait_queue_head
*rcu_nocb_gp_get(struct rcu_node
*rnp
);
427 static void rcu_nocb_gp_cleanup(struct swait_queue_head
*sq
);
428 static void rcu_init_one_nocb(struct rcu_node
*rnp
);
429 static bool rcu_nocb_flush_bypass(struct rcu_data
*rdp
, struct rcu_head
*rhp
,
431 static bool rcu_nocb_try_bypass(struct rcu_data
*rdp
, struct rcu_head
*rhp
,
432 bool *was_alldone
, unsigned long flags
);
433 static void __call_rcu_nocb_wake(struct rcu_data
*rdp
, bool was_empty
,
434 unsigned long flags
);
435 static int rcu_nocb_need_deferred_wakeup(struct rcu_data
*rdp
);
436 static void do_nocb_deferred_wakeup(struct rcu_data
*rdp
);
437 static void rcu_boot_init_nocb_percpu_data(struct rcu_data
*rdp
);
438 static void rcu_spawn_cpu_nocb_kthread(int cpu
);
439 static void __init
rcu_spawn_nocb_kthreads(void);
440 static void show_rcu_nocb_state(struct rcu_data
*rdp
);
441 static void rcu_nocb_lock(struct rcu_data
*rdp
);
442 static void rcu_nocb_unlock(struct rcu_data
*rdp
);
443 static void rcu_nocb_unlock_irqrestore(struct rcu_data
*rdp
,
444 unsigned long flags
);
445 static void rcu_lockdep_assert_cblist_protected(struct rcu_data
*rdp
);
446 #ifdef CONFIG_RCU_NOCB_CPU
447 static void __init
rcu_organize_nocb_kthreads(void);
448 #define rcu_nocb_lock_irqsave(rdp, flags) \
450 if (!rcu_segcblist_is_offloaded(&(rdp)->cblist)) \
451 local_irq_save(flags); \
453 raw_spin_lock_irqsave(&(rdp)->nocb_lock, (flags)); \
455 #else /* #ifdef CONFIG_RCU_NOCB_CPU */
456 #define rcu_nocb_lock_irqsave(rdp, flags) local_irq_save(flags)
457 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
459 static void rcu_bind_gp_kthread(void);
460 static bool rcu_nohz_full_cpu(void);
461 static void rcu_dynticks_task_enter(void);
462 static void rcu_dynticks_task_exit(void);
463 static void rcu_dynticks_task_trace_enter(void);
464 static void rcu_dynticks_task_trace_exit(void);
466 /* Forward declarations for tree_stall.h */
467 static void record_gp_stall_check_time(void);
468 static void rcu_iw_handler(struct irq_work
*iwp
);
469 static void check_cpu_stall(struct rcu_data
*rdp
);
470 static void rcu_check_gp_start_stall(struct rcu_node
*rnp
, struct rcu_data
*rdp
,
471 const unsigned long gpssdelay
);