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