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 <[email protected]> 22 * Paul E. McKenney <[email protected]> 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 34 /* 35 * Define shape of hierarchy based on NR_CPUS, CONFIG_RCU_FANOUT, and 36 * CONFIG_RCU_FANOUT_LEAF. 37 * In theory, it should be possible to add more levels straightforwardly. 38 * In practice, this did work well going from three levels to four. 39 * Of course, your mileage may vary. 40 */ 41 42 #ifdef CONFIG_RCU_FANOUT 43 #define RCU_FANOUT CONFIG_RCU_FANOUT 44 #else /* #ifdef CONFIG_RCU_FANOUT */ 45 # ifdef CONFIG_64BIT 46 # define RCU_FANOUT 64 47 # else 48 # define RCU_FANOUT 32 49 # endif 50 #endif /* #else #ifdef CONFIG_RCU_FANOUT */ 51 52 #ifdef CONFIG_RCU_FANOUT_LEAF 53 #define RCU_FANOUT_LEAF CONFIG_RCU_FANOUT_LEAF 54 #else /* #ifdef CONFIG_RCU_FANOUT_LEAF */ 55 # ifdef CONFIG_64BIT 56 # define RCU_FANOUT_LEAF 64 57 # else 58 # define RCU_FANOUT_LEAF 32 59 # endif 60 #endif /* #else #ifdef CONFIG_RCU_FANOUT_LEAF */ 61 62 #define RCU_FANOUT_1 (RCU_FANOUT_LEAF) 63 #define RCU_FANOUT_2 (RCU_FANOUT_1 * RCU_FANOUT) 64 #define RCU_FANOUT_3 (RCU_FANOUT_2 * RCU_FANOUT) 65 #define RCU_FANOUT_4 (RCU_FANOUT_3 * RCU_FANOUT) 66 67 #if NR_CPUS <= RCU_FANOUT_1 68 # define RCU_NUM_LVLS 1 69 # define NUM_RCU_LVL_0 1 70 # define NUM_RCU_NODES NUM_RCU_LVL_0 71 # define NUM_RCU_LVL_INIT { NUM_RCU_LVL_0 } 72 # define RCU_NODE_NAME_INIT { "rcu_node_0" } 73 # define RCU_FQS_NAME_INIT { "rcu_node_fqs_0" } 74 #elif NR_CPUS <= RCU_FANOUT_2 75 # define RCU_NUM_LVLS 2 76 # define NUM_RCU_LVL_0 1 77 # define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1) 78 # define NUM_RCU_NODES (NUM_RCU_LVL_0 + NUM_RCU_LVL_1) 79 # define NUM_RCU_LVL_INIT { NUM_RCU_LVL_0, NUM_RCU_LVL_1 } 80 # define RCU_NODE_NAME_INIT { "rcu_node_0", "rcu_node_1" } 81 # define RCU_FQS_NAME_INIT { "rcu_node_fqs_0", "rcu_node_fqs_1" } 82 #elif NR_CPUS <= RCU_FANOUT_3 83 # define RCU_NUM_LVLS 3 84 # define NUM_RCU_LVL_0 1 85 # define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2) 86 # define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1) 87 # define NUM_RCU_NODES (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2) 88 # define NUM_RCU_LVL_INIT { NUM_RCU_LVL_0, NUM_RCU_LVL_1, NUM_RCU_LVL_2 } 89 # define RCU_NODE_NAME_INIT { "rcu_node_0", "rcu_node_1", "rcu_node_2" } 90 # define RCU_FQS_NAME_INIT { "rcu_node_fqs_0", "rcu_node_fqs_1", "rcu_node_fqs_2" } 91 #elif NR_CPUS <= RCU_FANOUT_4 92 # define RCU_NUM_LVLS 4 93 # define NUM_RCU_LVL_0 1 94 # define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3) 95 # define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2) 96 # define NUM_RCU_LVL_3 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1) 97 # define NUM_RCU_NODES (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3) 98 # define NUM_RCU_LVL_INIT { NUM_RCU_LVL_0, NUM_RCU_LVL_1, NUM_RCU_LVL_2, NUM_RCU_LVL_3 } 99 # define RCU_NODE_NAME_INIT { "rcu_node_0", "rcu_node_1", "rcu_node_2", "rcu_node_3" } 100 # define RCU_FQS_NAME_INIT { "rcu_node_fqs_0", "rcu_node_fqs_1", "rcu_node_fqs_2", "rcu_node_fqs_3" } 101 #else 102 # error "CONFIG_RCU_FANOUT insufficient for NR_CPUS" 103 #endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */ 104 105 extern int rcu_num_lvls; 106 extern int rcu_num_nodes; 107 108 /* 109 * Dynticks per-CPU state. 110 */ 111 struct rcu_dynticks { 112 long long dynticks_nesting; /* Track irq/process nesting level. */ 113 /* Process level is worth LLONG_MAX/2. */ 114 int dynticks_nmi_nesting; /* Track NMI nesting level. */ 115 atomic_t dynticks; /* Even value for idle, else odd. */ 116 bool rcu_need_heavy_qs; /* GP old, need heavy quiescent state. */ 117 unsigned long rcu_qs_ctr; /* Light universal quiescent state ctr. */ 118 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE 119 long long dynticks_idle_nesting; 120 /* irq/process nesting level from idle. */ 121 atomic_t dynticks_idle; /* Even value for idle, else odd. */ 122 /* "Idle" excludes userspace execution. */ 123 unsigned long dynticks_idle_jiffies; 124 /* End of last non-NMI non-idle period. */ 125 #endif /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ 126 #ifdef CONFIG_RCU_FAST_NO_HZ 127 bool all_lazy; /* Are all CPU's CBs lazy? */ 128 unsigned long nonlazy_posted; 129 /* # times non-lazy CBs posted to CPU. */ 130 unsigned long nonlazy_posted_snap; 131 /* idle-period nonlazy_posted snapshot. */ 132 unsigned long last_accelerate; 133 /* Last jiffy CBs were accelerated. */ 134 unsigned long last_advance_all; 135 /* Last jiffy CBs were all advanced. */ 136 int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */ 137 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */ 138 }; 139 140 /* RCU's kthread states for tracing. */ 141 #define RCU_KTHREAD_STOPPED 0 142 #define RCU_KTHREAD_RUNNING 1 143 #define RCU_KTHREAD_WAITING 2 144 #define RCU_KTHREAD_OFFCPU 3 145 #define RCU_KTHREAD_YIELDING 4 146 #define RCU_KTHREAD_MAX 4 147 148 /* 149 * Definition for node within the RCU grace-period-detection hierarchy. 150 */ 151 struct rcu_node { 152 raw_spinlock_t __private lock; /* Root rcu_node's lock protects */ 153 /* some rcu_state fields as well as */ 154 /* following. */ 155 unsigned long gpnum; /* Current grace period for this node. */ 156 /* This will either be equal to or one */ 157 /* behind the root rcu_node's gpnum. */ 158 unsigned long completed; /* Last GP completed for this node. */ 159 /* This will either be equal to or one */ 160 /* behind the root rcu_node's gpnum. */ 161 unsigned long qsmask; /* CPUs or groups that need to switch in */ 162 /* order for current grace period to proceed.*/ 163 /* In leaf rcu_node, each bit corresponds to */ 164 /* an rcu_data structure, otherwise, each */ 165 /* bit corresponds to a child rcu_node */ 166 /* structure. */ 167 unsigned long qsmaskinit; 168 /* Per-GP initial value for qsmask. */ 169 /* Initialized from ->qsmaskinitnext at the */ 170 /* beginning of each grace period. */ 171 unsigned long qsmaskinitnext; 172 /* Online CPUs for next grace period. */ 173 unsigned long expmask; /* CPUs or groups that need to check in */ 174 /* to allow the current expedited GP */ 175 /* to complete. */ 176 unsigned long expmaskinit; 177 /* Per-GP initial values for expmask. */ 178 /* Initialized from ->expmaskinitnext at the */ 179 /* beginning of each expedited GP. */ 180 unsigned long expmaskinitnext; 181 /* Online CPUs for next expedited GP. */ 182 /* Any CPU that has ever been online will */ 183 /* have its bit set. */ 184 unsigned long grpmask; /* Mask to apply to parent qsmask. */ 185 /* Only one bit will be set in this mask. */ 186 int grplo; /* lowest-numbered CPU or group here. */ 187 int grphi; /* highest-numbered CPU or group here. */ 188 u8 grpnum; /* CPU/group number for next level up. */ 189 u8 level; /* root is at level 0. */ 190 bool wait_blkd_tasks;/* Necessary to wait for blocked tasks to */ 191 /* exit RCU read-side critical sections */ 192 /* before propagating offline up the */ 193 /* rcu_node tree? */ 194 struct rcu_node *parent; 195 struct list_head blkd_tasks; 196 /* Tasks blocked in RCU read-side critical */ 197 /* section. Tasks are placed at the head */ 198 /* of this list and age towards the tail. */ 199 struct list_head *gp_tasks; 200 /* Pointer to the first task blocking the */ 201 /* current grace period, or NULL if there */ 202 /* is no such task. */ 203 struct list_head *exp_tasks; 204 /* Pointer to the first task blocking the */ 205 /* current expedited grace period, or NULL */ 206 /* if there is no such task. If there */ 207 /* is no current expedited grace period, */ 208 /* then there can cannot be any such task. */ 209 struct list_head *boost_tasks; 210 /* Pointer to first task that needs to be */ 211 /* priority boosted, or NULL if no priority */ 212 /* boosting is needed for this rcu_node */ 213 /* structure. If there are no tasks */ 214 /* queued on this rcu_node structure that */ 215 /* are blocking the current grace period, */ 216 /* there can be no such task. */ 217 struct rt_mutex boost_mtx; 218 /* Used only for the priority-boosting */ 219 /* side effect, not as a lock. */ 220 unsigned long boost_time; 221 /* When to start boosting (jiffies). */ 222 struct task_struct *boost_kthread_task; 223 /* kthread that takes care of priority */ 224 /* boosting for this rcu_node structure. */ 225 unsigned int boost_kthread_status; 226 /* State of boost_kthread_task for tracing. */ 227 unsigned long n_tasks_boosted; 228 /* Total number of tasks boosted. */ 229 unsigned long n_exp_boosts; 230 /* Number of tasks boosted for expedited GP. */ 231 unsigned long n_normal_boosts; 232 /* Number of tasks boosted for normal GP. */ 233 unsigned long n_balk_blkd_tasks; 234 /* Refused to boost: no blocked tasks. */ 235 unsigned long n_balk_exp_gp_tasks; 236 /* Refused to boost: nothing blocking GP. */ 237 unsigned long n_balk_boost_tasks; 238 /* Refused to boost: already boosting. */ 239 unsigned long n_balk_notblocked; 240 /* Refused to boost: RCU RS CS still running. */ 241 unsigned long n_balk_notyet; 242 /* Refused to boost: not yet time. */ 243 unsigned long n_balk_nos; 244 /* Refused to boost: not sure why, though. */ 245 /* This can happen due to race conditions. */ 246 #ifdef CONFIG_RCU_NOCB_CPU 247 struct swait_queue_head nocb_gp_wq[2]; 248 /* Place for rcu_nocb_kthread() to wait GP. */ 249 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ 250 int need_future_gp[2]; 251 /* Counts of upcoming no-CB GP requests. */ 252 raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp; 253 254 spinlock_t exp_lock ____cacheline_internodealigned_in_smp; 255 unsigned long exp_seq_rq; 256 wait_queue_head_t exp_wq[4]; 257 } ____cacheline_internodealigned_in_smp; 258 259 /* 260 * Bitmasks in an rcu_node cover the interval [grplo, grphi] of CPU IDs, and 261 * are indexed relative to this interval rather than the global CPU ID space. 262 * This generates the bit for a CPU in node-local masks. 263 */ 264 #define leaf_node_cpu_bit(rnp, cpu) (1UL << ((cpu) - (rnp)->grplo)) 265 266 /* 267 * Do a full breadth-first scan of the rcu_node structures for the 268 * specified rcu_state structure. 269 */ 270 #define rcu_for_each_node_breadth_first(rsp, rnp) \ 271 for ((rnp) = &(rsp)->node[0]; \ 272 (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++) 273 274 /* 275 * Do a breadth-first scan of the non-leaf rcu_node structures for the 276 * specified rcu_state structure. Note that if there is a singleton 277 * rcu_node tree with but one rcu_node structure, this loop is a no-op. 278 */ 279 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \ 280 for ((rnp) = &(rsp)->node[0]; \ 281 (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++) 282 283 /* 284 * Scan the leaves of the rcu_node hierarchy for the specified rcu_state 285 * structure. Note that if there is a singleton rcu_node tree with but 286 * one rcu_node structure, this loop -will- visit the rcu_node structure. 287 * It is still a leaf node, even if it is also the root node. 288 */ 289 #define rcu_for_each_leaf_node(rsp, rnp) \ 290 for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \ 291 (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++) 292 293 /* 294 * Iterate over all possible CPUs in a leaf RCU node. 295 */ 296 #define for_each_leaf_node_possible_cpu(rnp, cpu) \ 297 for ((cpu) = cpumask_next(rnp->grplo - 1, cpu_possible_mask); \ 298 cpu <= rnp->grphi; \ 299 cpu = cpumask_next((cpu), cpu_possible_mask)) 300 301 /* 302 * Union to allow "aggregate OR" operation on the need for a quiescent 303 * state by the normal and expedited grace periods. 304 */ 305 union rcu_noqs { 306 struct { 307 u8 norm; 308 u8 exp; 309 } b; /* Bits. */ 310 u16 s; /* Set of bits, aggregate OR here. */ 311 }; 312 313 /* Index values for nxttail array in struct rcu_data. */ 314 #define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */ 315 #define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */ 316 #define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */ 317 #define RCU_NEXT_TAIL 3 318 #define RCU_NEXT_SIZE 4 319 320 /* Per-CPU data for read-copy update. */ 321 struct rcu_data { 322 /* 1) quiescent-state and grace-period handling : */ 323 unsigned long completed; /* Track rsp->completed gp number */ 324 /* in order to detect GP end. */ 325 unsigned long gpnum; /* Highest gp number that this CPU */ 326 /* is aware of having started. */ 327 unsigned long rcu_qs_ctr_snap;/* Snapshot of rcu_qs_ctr to check */ 328 /* for rcu_all_qs() invocations. */ 329 union rcu_noqs cpu_no_qs; /* No QSes yet for this CPU. */ 330 bool core_needs_qs; /* Core waits for quiesc state. */ 331 bool beenonline; /* CPU online at least once. */ 332 bool gpwrap; /* Possible gpnum/completed wrap. */ 333 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */ 334 unsigned long grpmask; /* Mask to apply to leaf qsmask. */ 335 unsigned long ticks_this_gp; /* The number of scheduling-clock */ 336 /* ticks this CPU has handled */ 337 /* during and after the last grace */ 338 /* period it is aware of. */ 339 340 /* 2) batch handling */ 341 /* 342 * If nxtlist is not NULL, it is partitioned as follows. 343 * Any of the partitions might be empty, in which case the 344 * pointer to that partition will be equal to the pointer for 345 * the following partition. When the list is empty, all of 346 * the nxttail elements point to the ->nxtlist pointer itself, 347 * which in that case is NULL. 348 * 349 * [nxtlist, *nxttail[RCU_DONE_TAIL]): 350 * Entries that batch # <= ->completed 351 * The grace period for these entries has completed, and 352 * the other grace-period-completed entries may be moved 353 * here temporarily in rcu_process_callbacks(). 354 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]): 355 * Entries that batch # <= ->completed - 1: waiting for current GP 356 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]): 357 * Entries known to have arrived before current GP ended 358 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]): 359 * Entries that might have arrived after current GP ended 360 * Note that the value of *nxttail[RCU_NEXT_TAIL] will 361 * always be NULL, as this is the end of the list. 362 */ 363 struct rcu_head *nxtlist; 364 struct rcu_head **nxttail[RCU_NEXT_SIZE]; 365 unsigned long nxtcompleted[RCU_NEXT_SIZE]; 366 /* grace periods for sublists. */ 367 long qlen_lazy; /* # of lazy queued callbacks */ 368 long qlen; /* # of queued callbacks, incl lazy */ 369 long qlen_last_fqs_check; 370 /* qlen at last check for QS forcing */ 371 unsigned long n_cbs_invoked; /* count of RCU cbs invoked. */ 372 unsigned long n_nocbs_invoked; /* count of no-CBs RCU cbs invoked. */ 373 unsigned long n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */ 374 unsigned long n_cbs_adopted; /* RCU cbs adopted from dying CPU */ 375 unsigned long n_force_qs_snap; 376 /* did other CPU force QS recently? */ 377 long blimit; /* Upper limit on a processed batch */ 378 379 /* 3) dynticks interface. */ 380 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */ 381 int dynticks_snap; /* Per-GP tracking for dynticks. */ 382 383 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */ 384 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */ 385 unsigned long offline_fqs; /* Kicked due to being offline. */ 386 unsigned long cond_resched_completed; 387 /* Grace period that needs help */ 388 /* from cond_resched(). */ 389 390 /* 5) __rcu_pending() statistics. */ 391 unsigned long n_rcu_pending; /* rcu_pending() calls since boot. */ 392 unsigned long n_rp_core_needs_qs; 393 unsigned long n_rp_report_qs; 394 unsigned long n_rp_cb_ready; 395 unsigned long n_rp_cpu_needs_gp; 396 unsigned long n_rp_gp_completed; 397 unsigned long n_rp_gp_started; 398 unsigned long n_rp_nocb_defer_wakeup; 399 unsigned long n_rp_need_nothing; 400 401 /* 6) _rcu_barrier(), OOM callbacks, and expediting. */ 402 struct rcu_head barrier_head; 403 #ifdef CONFIG_RCU_FAST_NO_HZ 404 struct rcu_head oom_head; 405 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */ 406 atomic_long_t exp_workdone0; /* # done by workqueue. */ 407 atomic_long_t exp_workdone1; /* # done by others #1. */ 408 atomic_long_t exp_workdone2; /* # done by others #2. */ 409 atomic_long_t exp_workdone3; /* # done by others #3. */ 410 int exp_dynticks_snap; /* Double-check need for IPI. */ 411 412 /* 7) Callback offloading. */ 413 #ifdef CONFIG_RCU_NOCB_CPU 414 struct rcu_head *nocb_head; /* CBs waiting for kthread. */ 415 struct rcu_head **nocb_tail; 416 atomic_long_t nocb_q_count; /* # CBs waiting for nocb */ 417 atomic_long_t nocb_q_count_lazy; /* invocation (all stages). */ 418 struct rcu_head *nocb_follower_head; /* CBs ready to invoke. */ 419 struct rcu_head **nocb_follower_tail; 420 struct swait_queue_head nocb_wq; /* For nocb kthreads to sleep on. */ 421 struct task_struct *nocb_kthread; 422 int nocb_defer_wakeup; /* Defer wakeup of nocb_kthread. */ 423 424 /* The following fields are used by the leader, hence own cacheline. */ 425 struct rcu_head *nocb_gp_head ____cacheline_internodealigned_in_smp; 426 /* CBs waiting for GP. */ 427 struct rcu_head **nocb_gp_tail; 428 bool nocb_leader_sleep; /* Is the nocb leader thread asleep? */ 429 struct rcu_data *nocb_next_follower; 430 /* Next follower in wakeup chain. */ 431 432 /* The following fields are used by the follower, hence new cachline. */ 433 struct rcu_data *nocb_leader ____cacheline_internodealigned_in_smp; 434 /* Leader CPU takes GP-end wakeups. */ 435 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ 436 437 /* 8) RCU CPU stall data. */ 438 unsigned int softirq_snap; /* Snapshot of softirq activity. */ 439 440 int cpu; 441 struct rcu_state *rsp; 442 }; 443 444 /* Values for nocb_defer_wakeup field in struct rcu_data. */ 445 #define RCU_NOGP_WAKE_NOT 0 446 #define RCU_NOGP_WAKE 1 447 #define RCU_NOGP_WAKE_FORCE 2 448 449 #define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500)) 450 /* For jiffies_till_first_fqs and */ 451 /* and jiffies_till_next_fqs. */ 452 453 #define RCU_JIFFIES_FQS_DIV 256 /* Very large systems need more */ 454 /* delay between bouts of */ 455 /* quiescent-state forcing. */ 456 457 #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time to take */ 458 /* at least one scheduling clock */ 459 /* irq before ratting on them. */ 460 461 #define rcu_wait(cond) \ 462 do { \ 463 for (;;) { \ 464 set_current_state(TASK_INTERRUPTIBLE); \ 465 if (cond) \ 466 break; \ 467 schedule(); \ 468 } \ 469 __set_current_state(TASK_RUNNING); \ 470 } while (0) 471 472 /* 473 * RCU global state, including node hierarchy. This hierarchy is 474 * represented in "heap" form in a dense array. The root (first level) 475 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second 476 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]), 477 * and the third level in ->node[m+1] and following (->node[m+1] referenced 478 * by ->level[2]). The number of levels is determined by the number of 479 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy" 480 * consisting of a single rcu_node. 481 */ 482 struct rcu_state { 483 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */ 484 struct rcu_node *level[RCU_NUM_LVLS + 1]; 485 /* Hierarchy levels (+1 to */ 486 /* shut bogus gcc warning) */ 487 struct rcu_data __percpu *rda; /* pointer of percu rcu_data. */ 488 call_rcu_func_t call; /* call_rcu() flavor. */ 489 int ncpus; /* # CPUs seen so far. */ 490 491 /* The following fields are guarded by the root rcu_node's lock. */ 492 493 u8 boost ____cacheline_internodealigned_in_smp; 494 /* Subject to priority boost. */ 495 unsigned long gpnum; /* Current gp number. */ 496 unsigned long completed; /* # of last completed gp. */ 497 struct task_struct *gp_kthread; /* Task for grace periods. */ 498 struct swait_queue_head gp_wq; /* Where GP task waits. */ 499 short gp_flags; /* Commands for GP task. */ 500 short gp_state; /* GP kthread sleep state. */ 501 502 /* End of fields guarded by root rcu_node's lock. */ 503 504 raw_spinlock_t orphan_lock ____cacheline_internodealigned_in_smp; 505 /* Protect following fields. */ 506 struct rcu_head *orphan_nxtlist; /* Orphaned callbacks that */ 507 /* need a grace period. */ 508 struct rcu_head **orphan_nxttail; /* Tail of above. */ 509 struct rcu_head *orphan_donelist; /* Orphaned callbacks that */ 510 /* are ready to invoke. */ 511 struct rcu_head **orphan_donetail; /* Tail of above. */ 512 long qlen_lazy; /* Number of lazy callbacks. */ 513 long qlen; /* Total number of callbacks. */ 514 /* End of fields guarded by orphan_lock. */ 515 516 struct mutex barrier_mutex; /* Guards barrier fields. */ 517 atomic_t barrier_cpu_count; /* # CPUs waiting on. */ 518 struct completion barrier_completion; /* Wake at barrier end. */ 519 unsigned long barrier_sequence; /* ++ at start and end of */ 520 /* _rcu_barrier(). */ 521 /* End of fields guarded by barrier_mutex. */ 522 523 struct mutex exp_mutex; /* Serialize expedited GP. */ 524 struct mutex exp_wake_mutex; /* Serialize wakeup. */ 525 unsigned long expedited_sequence; /* Take a ticket. */ 526 atomic_t expedited_need_qs; /* # CPUs left to check in. */ 527 struct swait_queue_head expedited_wq; /* Wait for check-ins. */ 528 int ncpus_snap; /* # CPUs seen last time. */ 529 530 unsigned long jiffies_force_qs; /* Time at which to invoke */ 531 /* force_quiescent_state(). */ 532 unsigned long jiffies_kick_kthreads; /* Time at which to kick */ 533 /* kthreads, if configured. */ 534 unsigned long n_force_qs; /* Number of calls to */ 535 /* force_quiescent_state(). */ 536 unsigned long n_force_qs_lh; /* ~Number of calls leaving */ 537 /* due to lock unavailable. */ 538 unsigned long n_force_qs_ngp; /* Number of calls leaving */ 539 /* due to no GP active. */ 540 unsigned long gp_start; /* Time at which GP started, */ 541 /* but in jiffies. */ 542 unsigned long gp_activity; /* Time of last GP kthread */ 543 /* activity in jiffies. */ 544 unsigned long jiffies_stall; /* Time at which to check */ 545 /* for CPU stalls. */ 546 unsigned long jiffies_resched; /* Time at which to resched */ 547 /* a reluctant CPU. */ 548 unsigned long n_force_qs_gpstart; /* Snapshot of n_force_qs at */ 549 /* GP start. */ 550 unsigned long gp_max; /* Maximum GP duration in */ 551 /* jiffies. */ 552 const char *name; /* Name of structure. */ 553 char abbr; /* Abbreviated name. */ 554 struct list_head flavors; /* List of RCU flavors. */ 555 }; 556 557 /* Values for rcu_state structure's gp_flags field. */ 558 #define RCU_GP_FLAG_INIT 0x1 /* Need grace-period initialization. */ 559 #define RCU_GP_FLAG_FQS 0x2 /* Need grace-period quiescent-state forcing. */ 560 561 /* Values for rcu_state structure's gp_state field. */ 562 #define RCU_GP_IDLE 0 /* Initial state and no GP in progress. */ 563 #define RCU_GP_WAIT_GPS 1 /* Wait for grace-period start. */ 564 #define RCU_GP_DONE_GPS 2 /* Wait done for grace-period start. */ 565 #define RCU_GP_WAIT_FQS 3 /* Wait for force-quiescent-state time. */ 566 #define RCU_GP_DOING_FQS 4 /* Wait done for force-quiescent-state time. */ 567 #define RCU_GP_CLEANUP 5 /* Grace-period cleanup started. */ 568 #define RCU_GP_CLEANED 6 /* Grace-period cleanup complete. */ 569 570 #ifndef RCU_TREE_NONCORE 571 static const char * const gp_state_names[] = { 572 "RCU_GP_IDLE", 573 "RCU_GP_WAIT_GPS", 574 "RCU_GP_DONE_GPS", 575 "RCU_GP_WAIT_FQS", 576 "RCU_GP_DOING_FQS", 577 "RCU_GP_CLEANUP", 578 "RCU_GP_CLEANED", 579 }; 580 #endif /* #ifndef RCU_TREE_NONCORE */ 581 582 extern struct list_head rcu_struct_flavors; 583 584 /* Sequence through rcu_state structures for each RCU flavor. */ 585 #define for_each_rcu_flavor(rsp) \ 586 list_for_each_entry((rsp), &rcu_struct_flavors, flavors) 587 588 /* 589 * RCU implementation internal declarations: 590 */ 591 extern struct rcu_state rcu_sched_state; 592 593 extern struct rcu_state rcu_bh_state; 594 595 #ifdef CONFIG_PREEMPT_RCU 596 extern struct rcu_state rcu_preempt_state; 597 #endif /* #ifdef CONFIG_PREEMPT_RCU */ 598 599 int rcu_dynticks_snap(struct rcu_dynticks *rdtp); 600 bool rcu_eqs_special_set(int cpu); 601 602 #ifdef CONFIG_RCU_BOOST 603 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status); 604 DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu); 605 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops); 606 DECLARE_PER_CPU(char, rcu_cpu_has_work); 607 #endif /* #ifdef CONFIG_RCU_BOOST */ 608 609 #ifndef RCU_TREE_NONCORE 610 611 /* Forward declarations for rcutree_plugin.h */ 612 static void rcu_bootup_announce(void); 613 static void rcu_preempt_note_context_switch(void); 614 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp); 615 #ifdef CONFIG_HOTPLUG_CPU 616 static bool rcu_preempt_has_tasks(struct rcu_node *rnp); 617 #endif /* #ifdef CONFIG_HOTPLUG_CPU */ 618 static void rcu_print_detail_task_stall(struct rcu_state *rsp); 619 static int rcu_print_task_stall(struct rcu_node *rnp); 620 static int rcu_print_task_exp_stall(struct rcu_node *rnp); 621 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp); 622 static void rcu_preempt_check_callbacks(void); 623 void call_rcu(struct rcu_head *head, rcu_callback_t func); 624 static void __init __rcu_init_preempt(void); 625 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags); 626 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp); 627 static void invoke_rcu_callbacks_kthread(void); 628 static bool rcu_is_callbacks_kthread(void); 629 #ifdef CONFIG_RCU_BOOST 630 static void rcu_preempt_do_callbacks(void); 631 static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp, 632 struct rcu_node *rnp); 633 #endif /* #ifdef CONFIG_RCU_BOOST */ 634 static void __init rcu_spawn_boost_kthreads(void); 635 static void rcu_prepare_kthreads(int cpu); 636 static void rcu_cleanup_after_idle(void); 637 static void rcu_prepare_for_idle(void); 638 static void rcu_idle_count_callbacks_posted(void); 639 static bool rcu_preempt_has_tasks(struct rcu_node *rnp); 640 static void print_cpu_stall_info_begin(void); 641 static void print_cpu_stall_info(struct rcu_state *rsp, int cpu); 642 static void print_cpu_stall_info_end(void); 643 static void zero_cpu_stall_ticks(struct rcu_data *rdp); 644 static void increment_cpu_stall_ticks(void); 645 static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu); 646 static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq); 647 static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp); 648 static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq); 649 static void rcu_init_one_nocb(struct rcu_node *rnp); 650 static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, 651 bool lazy, unsigned long flags); 652 static bool rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp, 653 struct rcu_data *rdp, 654 unsigned long flags); 655 static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp); 656 static void do_nocb_deferred_wakeup(struct rcu_data *rdp); 657 static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp); 658 static void rcu_spawn_all_nocb_kthreads(int cpu); 659 static void __init rcu_spawn_nocb_kthreads(void); 660 #ifdef CONFIG_RCU_NOCB_CPU 661 static void __init rcu_organize_nocb_kthreads(struct rcu_state *rsp); 662 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ 663 static void __maybe_unused rcu_kick_nohz_cpu(int cpu); 664 static bool init_nocb_callback_list(struct rcu_data *rdp); 665 static void rcu_sysidle_enter(int irq); 666 static void rcu_sysidle_exit(int irq); 667 static void rcu_sysidle_check_cpu(struct rcu_data *rdp, bool *isidle, 668 unsigned long *maxj); 669 static bool is_sysidle_rcu_state(struct rcu_state *rsp); 670 static void rcu_sysidle_report_gp(struct rcu_state *rsp, int isidle, 671 unsigned long maxj); 672 static void rcu_bind_gp_kthread(void); 673 static void rcu_sysidle_init_percpu_data(struct rcu_dynticks *rdtp); 674 static bool rcu_nohz_full_cpu(struct rcu_state *rsp); 675 static void rcu_dynticks_task_enter(void); 676 static void rcu_dynticks_task_exit(void); 677 678 #endif /* #ifndef RCU_TREE_NONCORE */ 679 680 #ifdef CONFIG_RCU_TRACE 681 /* Read out queue lengths for tracing. */ 682 static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll) 683 { 684 #ifdef CONFIG_RCU_NOCB_CPU 685 *ql = atomic_long_read(&rdp->nocb_q_count); 686 *qll = atomic_long_read(&rdp->nocb_q_count_lazy); 687 #else /* #ifdef CONFIG_RCU_NOCB_CPU */ 688 *ql = 0; 689 *qll = 0; 690 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ 691 } 692 #endif /* #ifdef CONFIG_RCU_TRACE */ 693 694 /* 695 * Wrappers for the rcu_node::lock acquire and release. 696 * 697 * Because the rcu_nodes form a tree, the tree traversal locking will observe 698 * different lock values, this in turn means that an UNLOCK of one level 699 * followed by a LOCK of another level does not imply a full memory barrier; 700 * and most importantly transitivity is lost. 701 * 702 * In order to restore full ordering between tree levels, augment the regular 703 * lock acquire functions with smp_mb__after_unlock_lock(). 704 * 705 * As ->lock of struct rcu_node is a __private field, therefore one should use 706 * these wrappers rather than directly call raw_spin_{lock,unlock}* on ->lock. 707 */ 708 static inline void raw_spin_lock_rcu_node(struct rcu_node *rnp) 709 { 710 raw_spin_lock(&ACCESS_PRIVATE(rnp, lock)); 711 smp_mb__after_unlock_lock(); 712 } 713 714 static inline void raw_spin_unlock_rcu_node(struct rcu_node *rnp) 715 { 716 raw_spin_unlock(&ACCESS_PRIVATE(rnp, lock)); 717 } 718 719 static inline void raw_spin_lock_irq_rcu_node(struct rcu_node *rnp) 720 { 721 raw_spin_lock_irq(&ACCESS_PRIVATE(rnp, lock)); 722 smp_mb__after_unlock_lock(); 723 } 724 725 static inline void raw_spin_unlock_irq_rcu_node(struct rcu_node *rnp) 726 { 727 raw_spin_unlock_irq(&ACCESS_PRIVATE(rnp, lock)); 728 } 729 730 #define raw_spin_lock_irqsave_rcu_node(rnp, flags) \ 731 do { \ 732 typecheck(unsigned long, flags); \ 733 raw_spin_lock_irqsave(&ACCESS_PRIVATE(rnp, lock), flags); \ 734 smp_mb__after_unlock_lock(); \ 735 } while (0) 736 737 #define raw_spin_unlock_irqrestore_rcu_node(rnp, flags) \ 738 do { \ 739 typecheck(unsigned long, flags); \ 740 raw_spin_unlock_irqrestore(&ACCESS_PRIVATE(rnp, lock), flags); \ 741 } while (0) 742 743 static inline bool raw_spin_trylock_rcu_node(struct rcu_node *rnp) 744 { 745 bool locked = raw_spin_trylock(&ACCESS_PRIVATE(rnp, lock)); 746 747 if (locked) 748 smp_mb__after_unlock_lock(); 749 return locked; 750 } 751