1 /* 2 * Read-Copy Update mechanism for mutual exclusion (tree-based version) 3 * Internal non-public definitions. 4 * 5 * This program is free software; you can redistribute it and/or modify 6 * it under the terms of the GNU General Public License as published by 7 * the Free Software Foundation; either version 2 of the License, or 8 * (at your option) any later version. 9 * 10 * This program is distributed in the hope that it will be useful, 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 13 * GNU General Public License for more details. 14 * 15 * You should have received a copy of the GNU General Public License 16 * along with this program; if not, you can access it online at 17 * http://www.gnu.org/licenses/gpl-2.0.html. 18 * 19 * Copyright IBM Corporation, 2008 20 * 21 * Author: Ingo Molnar <mingo@elte.hu> 22 * Paul E. McKenney <paulmck@linux.vnet.ibm.com> 23 */ 24 25 #include <linux/cache.h> 26 #include <linux/spinlock.h> 27 #include <linux/threads.h> 28 #include <linux/cpumask.h> 29 #include <linux/seqlock.h> 30 31 /* 32 * Define shape of hierarchy based on NR_CPUS, CONFIG_RCU_FANOUT, and 33 * CONFIG_RCU_FANOUT_LEAF. 34 * In theory, it should be possible to add more levels straightforwardly. 35 * In practice, this did work well going from three levels to four. 36 * Of course, your mileage may vary. 37 */ 38 #define MAX_RCU_LVLS 4 39 #define RCU_FANOUT_1 (CONFIG_RCU_FANOUT_LEAF) 40 #define RCU_FANOUT_2 (RCU_FANOUT_1 * CONFIG_RCU_FANOUT) 41 #define RCU_FANOUT_3 (RCU_FANOUT_2 * CONFIG_RCU_FANOUT) 42 #define RCU_FANOUT_4 (RCU_FANOUT_3 * CONFIG_RCU_FANOUT) 43 44 #if NR_CPUS <= RCU_FANOUT_1 45 # define RCU_NUM_LVLS 1 46 # define NUM_RCU_LVL_0 1 47 # define NUM_RCU_LVL_1 (NR_CPUS) 48 # define NUM_RCU_LVL_2 0 49 # define NUM_RCU_LVL_3 0 50 # define NUM_RCU_LVL_4 0 51 #elif NR_CPUS <= RCU_FANOUT_2 52 # define RCU_NUM_LVLS 2 53 # define NUM_RCU_LVL_0 1 54 # define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1) 55 # define NUM_RCU_LVL_2 (NR_CPUS) 56 # define NUM_RCU_LVL_3 0 57 # define NUM_RCU_LVL_4 0 58 #elif NR_CPUS <= RCU_FANOUT_3 59 # define RCU_NUM_LVLS 3 60 # define NUM_RCU_LVL_0 1 61 # define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2) 62 # define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1) 63 # define NUM_RCU_LVL_3 (NR_CPUS) 64 # define NUM_RCU_LVL_4 0 65 #elif NR_CPUS <= RCU_FANOUT_4 66 # define RCU_NUM_LVLS 4 67 # define NUM_RCU_LVL_0 1 68 # define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3) 69 # define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2) 70 # define NUM_RCU_LVL_3 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1) 71 # define NUM_RCU_LVL_4 (NR_CPUS) 72 #else 73 # error "CONFIG_RCU_FANOUT insufficient for NR_CPUS" 74 #endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */ 75 76 #define RCU_SUM (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3 + NUM_RCU_LVL_4) 77 #define NUM_RCU_NODES (RCU_SUM - NR_CPUS) 78 79 extern int rcu_num_lvls; 80 extern int rcu_num_nodes; 81 82 /* 83 * Dynticks per-CPU state. 84 */ 85 struct rcu_dynticks { 86 long long dynticks_nesting; /* Track irq/process nesting level. */ 87 /* Process level is worth LLONG_MAX/2. */ 88 int dynticks_nmi_nesting; /* Track NMI nesting level. */ 89 atomic_t dynticks; /* Even value for idle, else odd. */ 90 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE 91 long long dynticks_idle_nesting; 92 /* irq/process nesting level from idle. */ 93 atomic_t dynticks_idle; /* Even value for idle, else odd. */ 94 /* "Idle" excludes userspace execution. */ 95 unsigned long dynticks_idle_jiffies; 96 /* End of last non-NMI non-idle period. */ 97 #endif /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ 98 #ifdef CONFIG_RCU_FAST_NO_HZ 99 bool all_lazy; /* Are all CPU's CBs lazy? */ 100 unsigned long nonlazy_posted; 101 /* # times non-lazy CBs posted to CPU. */ 102 unsigned long nonlazy_posted_snap; 103 /* idle-period nonlazy_posted snapshot. */ 104 unsigned long last_accelerate; 105 /* Last jiffy CBs were accelerated. */ 106 unsigned long last_advance_all; 107 /* Last jiffy CBs were all advanced. */ 108 int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */ 109 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */ 110 }; 111 112 /* RCU's kthread states for tracing. */ 113 #define RCU_KTHREAD_STOPPED 0 114 #define RCU_KTHREAD_RUNNING 1 115 #define RCU_KTHREAD_WAITING 2 116 #define RCU_KTHREAD_OFFCPU 3 117 #define RCU_KTHREAD_YIELDING 4 118 #define RCU_KTHREAD_MAX 4 119 120 /* 121 * Definition for node within the RCU grace-period-detection hierarchy. 122 */ 123 struct rcu_node { 124 raw_spinlock_t lock; /* Root rcu_node's lock protects some */ 125 /* rcu_state fields as well as following. */ 126 unsigned long gpnum; /* Current grace period for this node. */ 127 /* This will either be equal to or one */ 128 /* behind the root rcu_node's gpnum. */ 129 unsigned long completed; /* Last GP completed for this node. */ 130 /* This will either be equal to or one */ 131 /* behind the root rcu_node's gpnum. */ 132 unsigned long qsmask; /* CPUs or groups that need to switch in */ 133 /* order for current grace period to proceed.*/ 134 /* In leaf rcu_node, each bit corresponds to */ 135 /* an rcu_data structure, otherwise, each */ 136 /* bit corresponds to a child rcu_node */ 137 /* structure. */ 138 unsigned long expmask; /* Groups that have ->blkd_tasks */ 139 /* elements that need to drain to allow the */ 140 /* current expedited grace period to */ 141 /* complete (only for PREEMPT_RCU). */ 142 unsigned long qsmaskinit; 143 /* Per-GP initial value for qsmask & expmask. */ 144 unsigned long grpmask; /* Mask to apply to parent qsmask. */ 145 /* Only one bit will be set in this mask. */ 146 int grplo; /* lowest-numbered CPU or group here. */ 147 int grphi; /* highest-numbered CPU or group here. */ 148 u8 grpnum; /* CPU/group number for next level up. */ 149 u8 level; /* root is at level 0. */ 150 struct rcu_node *parent; 151 struct list_head blkd_tasks; 152 /* Tasks blocked in RCU read-side critical */ 153 /* section. Tasks are placed at the head */ 154 /* of this list and age towards the tail. */ 155 struct list_head *gp_tasks; 156 /* Pointer to the first task blocking the */ 157 /* current grace period, or NULL if there */ 158 /* is no such task. */ 159 struct list_head *exp_tasks; 160 /* Pointer to the first task blocking the */ 161 /* current expedited grace period, or NULL */ 162 /* if there is no such task. If there */ 163 /* is no current expedited grace period, */ 164 /* then there can cannot be any such task. */ 165 #ifdef CONFIG_RCU_BOOST 166 struct list_head *boost_tasks; 167 /* Pointer to first task that needs to be */ 168 /* priority boosted, or NULL if no priority */ 169 /* boosting is needed for this rcu_node */ 170 /* structure. If there are no tasks */ 171 /* queued on this rcu_node structure that */ 172 /* are blocking the current grace period, */ 173 /* there can be no such task. */ 174 struct rt_mutex boost_mtx; 175 /* Used only for the priority-boosting */ 176 /* side effect, not as a lock. */ 177 unsigned long boost_time; 178 /* When to start boosting (jiffies). */ 179 struct task_struct *boost_kthread_task; 180 /* kthread that takes care of priority */ 181 /* boosting for this rcu_node structure. */ 182 unsigned int boost_kthread_status; 183 /* State of boost_kthread_task for tracing. */ 184 unsigned long n_tasks_boosted; 185 /* Total number of tasks boosted. */ 186 unsigned long n_exp_boosts; 187 /* Number of tasks boosted for expedited GP. */ 188 unsigned long n_normal_boosts; 189 /* Number of tasks boosted for normal GP. */ 190 unsigned long n_balk_blkd_tasks; 191 /* Refused to boost: no blocked tasks. */ 192 unsigned long n_balk_exp_gp_tasks; 193 /* Refused to boost: nothing blocking GP. */ 194 unsigned long n_balk_boost_tasks; 195 /* Refused to boost: already boosting. */ 196 unsigned long n_balk_notblocked; 197 /* Refused to boost: RCU RS CS still running. */ 198 unsigned long n_balk_notyet; 199 /* Refused to boost: not yet time. */ 200 unsigned long n_balk_nos; 201 /* Refused to boost: not sure why, though. */ 202 /* This can happen due to race conditions. */ 203 #endif /* #ifdef CONFIG_RCU_BOOST */ 204 #ifdef CONFIG_RCU_NOCB_CPU 205 wait_queue_head_t nocb_gp_wq[2]; 206 /* Place for rcu_nocb_kthread() to wait GP. */ 207 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ 208 int need_future_gp[2]; 209 /* Counts of upcoming no-CB GP requests. */ 210 raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp; 211 } ____cacheline_internodealigned_in_smp; 212 213 /* 214 * Do a full breadth-first scan of the rcu_node structures for the 215 * specified rcu_state structure. 216 */ 217 #define rcu_for_each_node_breadth_first(rsp, rnp) \ 218 for ((rnp) = &(rsp)->node[0]; \ 219 (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++) 220 221 /* 222 * Do a breadth-first scan of the non-leaf rcu_node structures for the 223 * specified rcu_state structure. Note that if there is a singleton 224 * rcu_node tree with but one rcu_node structure, this loop is a no-op. 225 */ 226 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \ 227 for ((rnp) = &(rsp)->node[0]; \ 228 (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++) 229 230 /* 231 * Scan the leaves of the rcu_node hierarchy for the specified rcu_state 232 * structure. Note that if there is a singleton rcu_node tree with but 233 * one rcu_node structure, this loop -will- visit the rcu_node structure. 234 * It is still a leaf node, even if it is also the root node. 235 */ 236 #define rcu_for_each_leaf_node(rsp, rnp) \ 237 for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \ 238 (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++) 239 240 /* Index values for nxttail array in struct rcu_data. */ 241 #define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */ 242 #define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */ 243 #define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */ 244 #define RCU_NEXT_TAIL 3 245 #define RCU_NEXT_SIZE 4 246 247 /* Per-CPU data for read-copy update. */ 248 struct rcu_data { 249 /* 1) quiescent-state and grace-period handling : */ 250 unsigned long completed; /* Track rsp->completed gp number */ 251 /* in order to detect GP end. */ 252 unsigned long gpnum; /* Highest gp number that this CPU */ 253 /* is aware of having started. */ 254 unsigned long rcu_qs_ctr_snap;/* Snapshot of rcu_qs_ctr to check */ 255 /* for rcu_all_qs() invocations. */ 256 bool passed_quiesce; /* User-mode/idle loop etc. */ 257 bool qs_pending; /* Core waits for quiesc state. */ 258 bool beenonline; /* CPU online at least once. */ 259 bool gpwrap; /* Possible gpnum/completed wrap. */ 260 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */ 261 unsigned long grpmask; /* Mask to apply to leaf qsmask. */ 262 #ifdef CONFIG_RCU_CPU_STALL_INFO 263 unsigned long ticks_this_gp; /* The number of scheduling-clock */ 264 /* ticks this CPU has handled */ 265 /* during and after the last grace */ 266 /* period it is aware of. */ 267 #endif /* #ifdef CONFIG_RCU_CPU_STALL_INFO */ 268 269 /* 2) batch handling */ 270 /* 271 * If nxtlist is not NULL, it is partitioned as follows. 272 * Any of the partitions might be empty, in which case the 273 * pointer to that partition will be equal to the pointer for 274 * the following partition. When the list is empty, all of 275 * the nxttail elements point to the ->nxtlist pointer itself, 276 * which in that case is NULL. 277 * 278 * [nxtlist, *nxttail[RCU_DONE_TAIL]): 279 * Entries that batch # <= ->completed 280 * The grace period for these entries has completed, and 281 * the other grace-period-completed entries may be moved 282 * here temporarily in rcu_process_callbacks(). 283 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]): 284 * Entries that batch # <= ->completed - 1: waiting for current GP 285 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]): 286 * Entries known to have arrived before current GP ended 287 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]): 288 * Entries that might have arrived after current GP ended 289 * Note that the value of *nxttail[RCU_NEXT_TAIL] will 290 * always be NULL, as this is the end of the list. 291 */ 292 struct rcu_head *nxtlist; 293 struct rcu_head **nxttail[RCU_NEXT_SIZE]; 294 unsigned long nxtcompleted[RCU_NEXT_SIZE]; 295 /* grace periods for sublists. */ 296 long qlen_lazy; /* # of lazy queued callbacks */ 297 long qlen; /* # of queued callbacks, incl lazy */ 298 long qlen_last_fqs_check; 299 /* qlen at last check for QS forcing */ 300 unsigned long n_cbs_invoked; /* count of RCU cbs invoked. */ 301 unsigned long n_nocbs_invoked; /* count of no-CBs RCU cbs invoked. */ 302 unsigned long n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */ 303 unsigned long n_cbs_adopted; /* RCU cbs adopted from dying CPU */ 304 unsigned long n_force_qs_snap; 305 /* did other CPU force QS recently? */ 306 long blimit; /* Upper limit on a processed batch */ 307 308 /* 3) dynticks interface. */ 309 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */ 310 int dynticks_snap; /* Per-GP tracking for dynticks. */ 311 312 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */ 313 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */ 314 unsigned long offline_fqs; /* Kicked due to being offline. */ 315 unsigned long cond_resched_completed; 316 /* Grace period that needs help */ 317 /* from cond_resched(). */ 318 319 /* 5) __rcu_pending() statistics. */ 320 unsigned long n_rcu_pending; /* rcu_pending() calls since boot. */ 321 unsigned long n_rp_qs_pending; 322 unsigned long n_rp_report_qs; 323 unsigned long n_rp_cb_ready; 324 unsigned long n_rp_cpu_needs_gp; 325 unsigned long n_rp_gp_completed; 326 unsigned long n_rp_gp_started; 327 unsigned long n_rp_nocb_defer_wakeup; 328 unsigned long n_rp_need_nothing; 329 330 /* 6) _rcu_barrier() and OOM callbacks. */ 331 struct rcu_head barrier_head; 332 #ifdef CONFIG_RCU_FAST_NO_HZ 333 struct rcu_head oom_head; 334 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */ 335 336 /* 7) Callback offloading. */ 337 #ifdef CONFIG_RCU_NOCB_CPU 338 struct rcu_head *nocb_head; /* CBs waiting for kthread. */ 339 struct rcu_head **nocb_tail; 340 atomic_long_t nocb_q_count; /* # CBs waiting for nocb */ 341 atomic_long_t nocb_q_count_lazy; /* invocation (all stages). */ 342 struct rcu_head *nocb_follower_head; /* CBs ready to invoke. */ 343 struct rcu_head **nocb_follower_tail; 344 wait_queue_head_t nocb_wq; /* For nocb kthreads to sleep on. */ 345 struct task_struct *nocb_kthread; 346 int nocb_defer_wakeup; /* Defer wakeup of nocb_kthread. */ 347 348 /* The following fields are used by the leader, hence own cacheline. */ 349 struct rcu_head *nocb_gp_head ____cacheline_internodealigned_in_smp; 350 /* CBs waiting for GP. */ 351 struct rcu_head **nocb_gp_tail; 352 bool nocb_leader_sleep; /* Is the nocb leader thread asleep? */ 353 struct rcu_data *nocb_next_follower; 354 /* Next follower in wakeup chain. */ 355 356 /* The following fields are used by the follower, hence new cachline. */ 357 struct rcu_data *nocb_leader ____cacheline_internodealigned_in_smp; 358 /* Leader CPU takes GP-end wakeups. */ 359 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ 360 361 /* 8) RCU CPU stall data. */ 362 #ifdef CONFIG_RCU_CPU_STALL_INFO 363 unsigned int softirq_snap; /* Snapshot of softirq activity. */ 364 #endif /* #ifdef CONFIG_RCU_CPU_STALL_INFO */ 365 366 int cpu; 367 struct rcu_state *rsp; 368 }; 369 370 /* Values for fqs_state field in struct rcu_state. */ 371 #define RCU_GP_IDLE 0 /* No grace period in progress. */ 372 #define RCU_GP_INIT 1 /* Grace period being initialized. */ 373 #define RCU_SAVE_DYNTICK 2 /* Need to scan dyntick state. */ 374 #define RCU_FORCE_QS 3 /* Need to force quiescent state. */ 375 #define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK 376 377 /* Values for nocb_defer_wakeup field in struct rcu_data. */ 378 #define RCU_NOGP_WAKE_NOT 0 379 #define RCU_NOGP_WAKE 1 380 #define RCU_NOGP_WAKE_FORCE 2 381 382 #define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500)) 383 /* For jiffies_till_first_fqs and */ 384 /* and jiffies_till_next_fqs. */ 385 386 #define RCU_JIFFIES_FQS_DIV 256 /* Very large systems need more */ 387 /* delay between bouts of */ 388 /* quiescent-state forcing. */ 389 390 #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time to take */ 391 /* at least one scheduling clock */ 392 /* irq before ratting on them. */ 393 394 #define rcu_wait(cond) \ 395 do { \ 396 for (;;) { \ 397 set_current_state(TASK_INTERRUPTIBLE); \ 398 if (cond) \ 399 break; \ 400 schedule(); \ 401 } \ 402 __set_current_state(TASK_RUNNING); \ 403 } while (0) 404 405 /* 406 * RCU global state, including node hierarchy. This hierarchy is 407 * represented in "heap" form in a dense array. The root (first level) 408 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second 409 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]), 410 * and the third level in ->node[m+1] and following (->node[m+1] referenced 411 * by ->level[2]). The number of levels is determined by the number of 412 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy" 413 * consisting of a single rcu_node. 414 */ 415 struct rcu_state { 416 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */ 417 struct rcu_node *level[RCU_NUM_LVLS]; /* Hierarchy levels. */ 418 u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */ 419 u8 levelspread[RCU_NUM_LVLS]; /* kids/node in each level. */ 420 u8 flavor_mask; /* bit in flavor mask. */ 421 struct rcu_data __percpu *rda; /* pointer of percu rcu_data. */ 422 void (*call)(struct rcu_head *head, /* call_rcu() flavor. */ 423 void (*func)(struct rcu_head *head)); 424 425 /* The following fields are guarded by the root rcu_node's lock. */ 426 427 u8 fqs_state ____cacheline_internodealigned_in_smp; 428 /* Force QS state. */ 429 u8 boost; /* Subject to priority boost. */ 430 unsigned long gpnum; /* Current gp number. */ 431 unsigned long completed; /* # of last completed gp. */ 432 struct task_struct *gp_kthread; /* Task for grace periods. */ 433 wait_queue_head_t gp_wq; /* Where GP task waits. */ 434 short gp_flags; /* Commands for GP task. */ 435 short gp_state; /* GP kthread sleep state. */ 436 437 /* End of fields guarded by root rcu_node's lock. */ 438 439 raw_spinlock_t orphan_lock ____cacheline_internodealigned_in_smp; 440 /* Protect following fields. */ 441 struct rcu_head *orphan_nxtlist; /* Orphaned callbacks that */ 442 /* need a grace period. */ 443 struct rcu_head **orphan_nxttail; /* Tail of above. */ 444 struct rcu_head *orphan_donelist; /* Orphaned callbacks that */ 445 /* are ready to invoke. */ 446 struct rcu_head **orphan_donetail; /* Tail of above. */ 447 long qlen_lazy; /* Number of lazy callbacks. */ 448 long qlen; /* Total number of callbacks. */ 449 /* End of fields guarded by orphan_lock. */ 450 451 struct mutex onoff_mutex; /* Coordinate hotplug & GPs. */ 452 453 struct mutex barrier_mutex; /* Guards barrier fields. */ 454 atomic_t barrier_cpu_count; /* # CPUs waiting on. */ 455 struct completion barrier_completion; /* Wake at barrier end. */ 456 unsigned long n_barrier_done; /* ++ at start and end of */ 457 /* _rcu_barrier(). */ 458 /* End of fields guarded by barrier_mutex. */ 459 460 atomic_long_t expedited_start; /* Starting ticket. */ 461 atomic_long_t expedited_done; /* Done ticket. */ 462 atomic_long_t expedited_wrap; /* # near-wrap incidents. */ 463 atomic_long_t expedited_tryfail; /* # acquisition failures. */ 464 atomic_long_t expedited_workdone1; /* # done by others #1. */ 465 atomic_long_t expedited_workdone2; /* # done by others #2. */ 466 atomic_long_t expedited_normal; /* # fallbacks to normal. */ 467 atomic_long_t expedited_stoppedcpus; /* # successful stop_cpus. */ 468 atomic_long_t expedited_done_tries; /* # tries to update _done. */ 469 atomic_long_t expedited_done_lost; /* # times beaten to _done. */ 470 atomic_long_t expedited_done_exit; /* # times exited _done loop. */ 471 472 unsigned long jiffies_force_qs; /* Time at which to invoke */ 473 /* force_quiescent_state(). */ 474 unsigned long n_force_qs; /* Number of calls to */ 475 /* force_quiescent_state(). */ 476 unsigned long n_force_qs_lh; /* ~Number of calls leaving */ 477 /* due to lock unavailable. */ 478 unsigned long n_force_qs_ngp; /* Number of calls leaving */ 479 /* due to no GP active. */ 480 unsigned long gp_start; /* Time at which GP started, */ 481 /* but in jiffies. */ 482 unsigned long gp_activity; /* Time of last GP kthread */ 483 /* activity in jiffies. */ 484 unsigned long jiffies_stall; /* Time at which to check */ 485 /* for CPU stalls. */ 486 unsigned long jiffies_resched; /* Time at which to resched */ 487 /* a reluctant CPU. */ 488 unsigned long n_force_qs_gpstart; /* Snapshot of n_force_qs at */ 489 /* GP start. */ 490 unsigned long gp_max; /* Maximum GP duration in */ 491 /* jiffies. */ 492 const char *name; /* Name of structure. */ 493 char abbr; /* Abbreviated name. */ 494 struct list_head flavors; /* List of RCU flavors. */ 495 }; 496 497 /* Values for rcu_state structure's gp_flags field. */ 498 #define RCU_GP_FLAG_INIT 0x1 /* Need grace-period initialization. */ 499 #define RCU_GP_FLAG_FQS 0x2 /* Need grace-period quiescent-state forcing. */ 500 501 /* Values for rcu_state structure's gp_flags field. */ 502 #define RCU_GP_WAIT_INIT 0 /* Initial state. */ 503 #define RCU_GP_WAIT_GPS 1 /* Wait for grace-period start. */ 504 #define RCU_GP_WAIT_FQS 2 /* Wait for force-quiescent-state time. */ 505 506 extern struct list_head rcu_struct_flavors; 507 508 /* Sequence through rcu_state structures for each RCU flavor. */ 509 #define for_each_rcu_flavor(rsp) \ 510 list_for_each_entry((rsp), &rcu_struct_flavors, flavors) 511 512 /* 513 * RCU implementation internal declarations: 514 */ 515 extern struct rcu_state rcu_sched_state; 516 DECLARE_PER_CPU(struct rcu_data, rcu_sched_data); 517 518 extern struct rcu_state rcu_bh_state; 519 DECLARE_PER_CPU(struct rcu_data, rcu_bh_data); 520 521 #ifdef CONFIG_PREEMPT_RCU 522 extern struct rcu_state rcu_preempt_state; 523 DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data); 524 #endif /* #ifdef CONFIG_PREEMPT_RCU */ 525 526 #ifdef CONFIG_RCU_BOOST 527 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status); 528 DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu); 529 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops); 530 DECLARE_PER_CPU(char, rcu_cpu_has_work); 531 #endif /* #ifdef CONFIG_RCU_BOOST */ 532 533 #ifndef RCU_TREE_NONCORE 534 535 /* Forward declarations for rcutree_plugin.h */ 536 static void rcu_bootup_announce(void); 537 static void rcu_preempt_note_context_switch(void); 538 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp); 539 #ifdef CONFIG_HOTPLUG_CPU 540 static bool rcu_preempt_has_tasks(struct rcu_node *rnp); 541 #endif /* #ifdef CONFIG_HOTPLUG_CPU */ 542 static void rcu_print_detail_task_stall(struct rcu_state *rsp); 543 static int rcu_print_task_stall(struct rcu_node *rnp); 544 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp); 545 static void rcu_preempt_check_callbacks(void); 546 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu)); 547 static void __init __rcu_init_preempt(void); 548 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags); 549 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp); 550 static void invoke_rcu_callbacks_kthread(void); 551 static bool rcu_is_callbacks_kthread(void); 552 #ifdef CONFIG_RCU_BOOST 553 static void rcu_preempt_do_callbacks(void); 554 static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp, 555 struct rcu_node *rnp); 556 #endif /* #ifdef CONFIG_RCU_BOOST */ 557 static void __init rcu_spawn_boost_kthreads(void); 558 static void rcu_prepare_kthreads(int cpu); 559 static void rcu_cleanup_after_idle(void); 560 static void rcu_prepare_for_idle(void); 561 static void rcu_idle_count_callbacks_posted(void); 562 static void print_cpu_stall_info_begin(void); 563 static void print_cpu_stall_info(struct rcu_state *rsp, int cpu); 564 static void print_cpu_stall_info_end(void); 565 static void zero_cpu_stall_ticks(struct rcu_data *rdp); 566 static void increment_cpu_stall_ticks(void); 567 static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu); 568 static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq); 569 static void rcu_nocb_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp); 570 static void rcu_init_one_nocb(struct rcu_node *rnp); 571 static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, 572 bool lazy, unsigned long flags); 573 static bool rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp, 574 struct rcu_data *rdp, 575 unsigned long flags); 576 static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp); 577 static void do_nocb_deferred_wakeup(struct rcu_data *rdp); 578 static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp); 579 static void rcu_spawn_all_nocb_kthreads(int cpu); 580 static void __init rcu_spawn_nocb_kthreads(void); 581 #ifdef CONFIG_RCU_NOCB_CPU 582 static void __init rcu_organize_nocb_kthreads(struct rcu_state *rsp); 583 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ 584 static void __maybe_unused rcu_kick_nohz_cpu(int cpu); 585 static bool init_nocb_callback_list(struct rcu_data *rdp); 586 static void rcu_sysidle_enter(int irq); 587 static void rcu_sysidle_exit(int irq); 588 static void rcu_sysidle_check_cpu(struct rcu_data *rdp, bool *isidle, 589 unsigned long *maxj); 590 static bool is_sysidle_rcu_state(struct rcu_state *rsp); 591 static void rcu_sysidle_report_gp(struct rcu_state *rsp, int isidle, 592 unsigned long maxj); 593 static void rcu_bind_gp_kthread(void); 594 static void rcu_sysidle_init_percpu_data(struct rcu_dynticks *rdtp); 595 static bool rcu_nohz_full_cpu(struct rcu_state *rsp); 596 static void rcu_dynticks_task_enter(void); 597 static void rcu_dynticks_task_exit(void); 598 599 #endif /* #ifndef RCU_TREE_NONCORE */ 600 601 #ifdef CONFIG_RCU_TRACE 602 /* Read out queue lengths for tracing. */ 603 static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll) 604 { 605 #ifdef CONFIG_RCU_NOCB_CPU 606 *ql = atomic_long_read(&rdp->nocb_q_count); 607 *qll = atomic_long_read(&rdp->nocb_q_count_lazy); 608 #else /* #ifdef CONFIG_RCU_NOCB_CPU */ 609 *ql = 0; 610 *qll = 0; 611 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ 612 } 613 #endif /* #ifdef CONFIG_RCU_TRACE */ 614