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 #include <linux/stop_machine.h> 31 32 /* 33 * Define shape of hierarchy based on NR_CPUS, CONFIG_RCU_FANOUT, and 34 * CONFIG_RCU_FANOUT_LEAF. 35 * In theory, it should be possible to add more levels straightforwardly. 36 * In practice, this did work well going from three levels to four. 37 * Of course, your mileage may vary. 38 */ 39 40 #ifdef CONFIG_RCU_FANOUT 41 #define RCU_FANOUT CONFIG_RCU_FANOUT 42 #else /* #ifdef CONFIG_RCU_FANOUT */ 43 # ifdef CONFIG_64BIT 44 # define RCU_FANOUT 64 45 # else 46 # define RCU_FANOUT 32 47 # endif 48 #endif /* #else #ifdef CONFIG_RCU_FANOUT */ 49 50 #ifdef CONFIG_RCU_FANOUT_LEAF 51 #define RCU_FANOUT_LEAF CONFIG_RCU_FANOUT_LEAF 52 #else /* #ifdef CONFIG_RCU_FANOUT_LEAF */ 53 # ifdef CONFIG_64BIT 54 # define RCU_FANOUT_LEAF 64 55 # else 56 # define RCU_FANOUT_LEAF 32 57 # endif 58 #endif /* #else #ifdef CONFIG_RCU_FANOUT_LEAF */ 59 60 #define RCU_FANOUT_1 (RCU_FANOUT_LEAF) 61 #define RCU_FANOUT_2 (RCU_FANOUT_1 * RCU_FANOUT) 62 #define RCU_FANOUT_3 (RCU_FANOUT_2 * RCU_FANOUT) 63 #define RCU_FANOUT_4 (RCU_FANOUT_3 * RCU_FANOUT) 64 65 #if NR_CPUS <= RCU_FANOUT_1 66 # define RCU_NUM_LVLS 1 67 # define NUM_RCU_LVL_0 1 68 # define NUM_RCU_NODES NUM_RCU_LVL_0 69 # define NUM_RCU_LVL_INIT { NUM_RCU_LVL_0 } 70 # define RCU_NODE_NAME_INIT { "rcu_node_0" } 71 # define RCU_FQS_NAME_INIT { "rcu_node_fqs_0" } 72 # define RCU_EXP_NAME_INIT { "rcu_node_exp_0" } 73 #elif NR_CPUS <= RCU_FANOUT_2 74 # define RCU_NUM_LVLS 2 75 # define NUM_RCU_LVL_0 1 76 # define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1) 77 # define NUM_RCU_NODES (NUM_RCU_LVL_0 + NUM_RCU_LVL_1) 78 # define NUM_RCU_LVL_INIT { NUM_RCU_LVL_0, NUM_RCU_LVL_1 } 79 # define RCU_NODE_NAME_INIT { "rcu_node_0", "rcu_node_1" } 80 # define RCU_FQS_NAME_INIT { "rcu_node_fqs_0", "rcu_node_fqs_1" } 81 # define RCU_EXP_NAME_INIT { "rcu_node_exp_0", "rcu_node_exp_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 # define RCU_EXP_NAME_INIT { "rcu_node_exp_0", "rcu_node_exp_1", "rcu_node_exp_2" } 92 #elif NR_CPUS <= RCU_FANOUT_4 93 # define RCU_NUM_LVLS 4 94 # define NUM_RCU_LVL_0 1 95 # define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3) 96 # define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2) 97 # define NUM_RCU_LVL_3 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1) 98 # define NUM_RCU_NODES (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3) 99 # define NUM_RCU_LVL_INIT { NUM_RCU_LVL_0, NUM_RCU_LVL_1, NUM_RCU_LVL_2, NUM_RCU_LVL_3 } 100 # define RCU_NODE_NAME_INIT { "rcu_node_0", "rcu_node_1", "rcu_node_2", "rcu_node_3" } 101 # define RCU_FQS_NAME_INIT { "rcu_node_fqs_0", "rcu_node_fqs_1", "rcu_node_fqs_2", "rcu_node_fqs_3" } 102 # define RCU_EXP_NAME_INIT { "rcu_node_exp_0", "rcu_node_exp_1", "rcu_node_exp_2", "rcu_node_exp_3" } 103 #else 104 # error "CONFIG_RCU_FANOUT insufficient for NR_CPUS" 105 #endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */ 106 107 extern int rcu_num_lvls; 108 extern int rcu_num_nodes; 109 110 /* 111 * Dynticks per-CPU state. 112 */ 113 struct rcu_dynticks { 114 long long dynticks_nesting; /* Track irq/process nesting level. */ 115 /* Process level is worth LLONG_MAX/2. */ 116 int dynticks_nmi_nesting; /* Track NMI nesting level. */ 117 atomic_t dynticks; /* Even value for idle, else odd. */ 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 lock; /* Root rcu_node's lock protects some */ 153 /* rcu_state fields as well as following. */ 154 unsigned long gpnum; /* Current grace period for this node. */ 155 /* This will either be equal to or one */ 156 /* behind the root rcu_node's gpnum. */ 157 unsigned long completed; /* Last GP completed for this node. */ 158 /* This will either be equal to or one */ 159 /* behind the root rcu_node's gpnum. */ 160 unsigned long qsmask; /* CPUs or groups that need to switch in */ 161 /* order for current grace period to proceed.*/ 162 /* In leaf rcu_node, each bit corresponds to */ 163 /* an rcu_data structure, otherwise, each */ 164 /* bit corresponds to a child rcu_node */ 165 /* structure. */ 166 unsigned long qsmaskinit; 167 /* Per-GP initial value for qsmask. */ 168 /* Initialized from ->qsmaskinitnext at the */ 169 /* beginning of each grace period. */ 170 unsigned long qsmaskinitnext; 171 /* Online CPUs for next grace period. */ 172 unsigned long expmask; /* CPUs or groups that need to check in */ 173 /* to allow the current expedited GP */ 174 /* to complete. */ 175 unsigned long expmaskinit; 176 /* Per-GP initial values for expmask. */ 177 /* Initialized from ->expmaskinitnext at the */ 178 /* beginning of each expedited GP. */ 179 unsigned long expmaskinitnext; 180 /* Online CPUs for next expedited GP. */ 181 unsigned long grpmask; /* Mask to apply to parent qsmask. */ 182 /* Only one bit will be set in this mask. */ 183 int grplo; /* lowest-numbered CPU or group here. */ 184 int grphi; /* highest-numbered CPU or group here. */ 185 u8 grpnum; /* CPU/group number for next level up. */ 186 u8 level; /* root is at level 0. */ 187 bool wait_blkd_tasks;/* Necessary to wait for blocked tasks to */ 188 /* exit RCU read-side critical sections */ 189 /* before propagating offline up the */ 190 /* rcu_node tree? */ 191 struct rcu_node *parent; 192 struct list_head blkd_tasks; 193 /* Tasks blocked in RCU read-side critical */ 194 /* section. Tasks are placed at the head */ 195 /* of this list and age towards the tail. */ 196 struct list_head *gp_tasks; 197 /* Pointer to the first task blocking the */ 198 /* current grace period, or NULL if there */ 199 /* is no such task. */ 200 struct list_head *exp_tasks; 201 /* Pointer to the first task blocking the */ 202 /* current expedited grace period, or NULL */ 203 /* if there is no such task. If there */ 204 /* is no current expedited grace period, */ 205 /* then there can cannot be any such task. */ 206 struct list_head *boost_tasks; 207 /* Pointer to first task that needs to be */ 208 /* priority boosted, or NULL if no priority */ 209 /* boosting is needed for this rcu_node */ 210 /* structure. If there are no tasks */ 211 /* queued on this rcu_node structure that */ 212 /* are blocking the current grace period, */ 213 /* there can be no such task. */ 214 struct rt_mutex boost_mtx; 215 /* Used only for the priority-boosting */ 216 /* side effect, not as a lock. */ 217 unsigned long boost_time; 218 /* When to start boosting (jiffies). */ 219 struct task_struct *boost_kthread_task; 220 /* kthread that takes care of priority */ 221 /* boosting for this rcu_node structure. */ 222 unsigned int boost_kthread_status; 223 /* State of boost_kthread_task for tracing. */ 224 unsigned long n_tasks_boosted; 225 /* Total number of tasks boosted. */ 226 unsigned long n_exp_boosts; 227 /* Number of tasks boosted for expedited GP. */ 228 unsigned long n_normal_boosts; 229 /* Number of tasks boosted for normal GP. */ 230 unsigned long n_balk_blkd_tasks; 231 /* Refused to boost: no blocked tasks. */ 232 unsigned long n_balk_exp_gp_tasks; 233 /* Refused to boost: nothing blocking GP. */ 234 unsigned long n_balk_boost_tasks; 235 /* Refused to boost: already boosting. */ 236 unsigned long n_balk_notblocked; 237 /* Refused to boost: RCU RS CS still running. */ 238 unsigned long n_balk_notyet; 239 /* Refused to boost: not yet time. */ 240 unsigned long n_balk_nos; 241 /* Refused to boost: not sure why, though. */ 242 /* This can happen due to race conditions. */ 243 #ifdef CONFIG_RCU_NOCB_CPU 244 wait_queue_head_t nocb_gp_wq[2]; 245 /* Place for rcu_nocb_kthread() to wait GP. */ 246 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ 247 int need_future_gp[2]; 248 /* Counts of upcoming no-CB GP requests. */ 249 raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp; 250 251 struct mutex exp_funnel_mutex ____cacheline_internodealigned_in_smp; 252 } ____cacheline_internodealigned_in_smp; 253 254 /* 255 * Do a full breadth-first scan of the rcu_node structures for the 256 * specified rcu_state structure. 257 */ 258 #define rcu_for_each_node_breadth_first(rsp, rnp) \ 259 for ((rnp) = &(rsp)->node[0]; \ 260 (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++) 261 262 /* 263 * Do a breadth-first scan of the non-leaf rcu_node structures for the 264 * specified rcu_state structure. Note that if there is a singleton 265 * rcu_node tree with but one rcu_node structure, this loop is a no-op. 266 */ 267 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \ 268 for ((rnp) = &(rsp)->node[0]; \ 269 (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++) 270 271 /* 272 * Scan the leaves of the rcu_node hierarchy for the specified rcu_state 273 * structure. Note that if there is a singleton rcu_node tree with but 274 * one rcu_node structure, this loop -will- visit the rcu_node structure. 275 * It is still a leaf node, even if it is also the root node. 276 */ 277 #define rcu_for_each_leaf_node(rsp, rnp) \ 278 for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \ 279 (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++) 280 281 /* 282 * Union to allow "aggregate OR" operation on the need for a quiescent 283 * state by the normal and expedited grace periods. 284 */ 285 union rcu_noqs { 286 struct { 287 u8 norm; 288 u8 exp; 289 } b; /* Bits. */ 290 u16 s; /* Set of bits, aggregate OR here. */ 291 }; 292 293 /* Index values for nxttail array in struct rcu_data. */ 294 #define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */ 295 #define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */ 296 #define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */ 297 #define RCU_NEXT_TAIL 3 298 #define RCU_NEXT_SIZE 4 299 300 /* Per-CPU data for read-copy update. */ 301 struct rcu_data { 302 /* 1) quiescent-state and grace-period handling : */ 303 unsigned long completed; /* Track rsp->completed gp number */ 304 /* in order to detect GP end. */ 305 unsigned long gpnum; /* Highest gp number that this CPU */ 306 /* is aware of having started. */ 307 unsigned long rcu_qs_ctr_snap;/* Snapshot of rcu_qs_ctr to check */ 308 /* for rcu_all_qs() invocations. */ 309 union rcu_noqs cpu_no_qs; /* No QSes yet for this CPU. */ 310 bool core_needs_qs; /* Core waits for quiesc state. */ 311 bool beenonline; /* CPU online at least once. */ 312 bool gpwrap; /* Possible gpnum/completed wrap. */ 313 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */ 314 unsigned long grpmask; /* Mask to apply to leaf qsmask. */ 315 unsigned long ticks_this_gp; /* The number of scheduling-clock */ 316 /* ticks this CPU has handled */ 317 /* during and after the last grace */ 318 /* period it is aware of. */ 319 320 /* 2) batch handling */ 321 /* 322 * If nxtlist is not NULL, it is partitioned as follows. 323 * Any of the partitions might be empty, in which case the 324 * pointer to that partition will be equal to the pointer for 325 * the following partition. When the list is empty, all of 326 * the nxttail elements point to the ->nxtlist pointer itself, 327 * which in that case is NULL. 328 * 329 * [nxtlist, *nxttail[RCU_DONE_TAIL]): 330 * Entries that batch # <= ->completed 331 * The grace period for these entries has completed, and 332 * the other grace-period-completed entries may be moved 333 * here temporarily in rcu_process_callbacks(). 334 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]): 335 * Entries that batch # <= ->completed - 1: waiting for current GP 336 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]): 337 * Entries known to have arrived before current GP ended 338 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]): 339 * Entries that might have arrived after current GP ended 340 * Note that the value of *nxttail[RCU_NEXT_TAIL] will 341 * always be NULL, as this is the end of the list. 342 */ 343 struct rcu_head *nxtlist; 344 struct rcu_head **nxttail[RCU_NEXT_SIZE]; 345 unsigned long nxtcompleted[RCU_NEXT_SIZE]; 346 /* grace periods for sublists. */ 347 long qlen_lazy; /* # of lazy queued callbacks */ 348 long qlen; /* # of queued callbacks, incl lazy */ 349 long qlen_last_fqs_check; 350 /* qlen at last check for QS forcing */ 351 unsigned long n_cbs_invoked; /* count of RCU cbs invoked. */ 352 unsigned long n_nocbs_invoked; /* count of no-CBs RCU cbs invoked. */ 353 unsigned long n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */ 354 unsigned long n_cbs_adopted; /* RCU cbs adopted from dying CPU */ 355 unsigned long n_force_qs_snap; 356 /* did other CPU force QS recently? */ 357 long blimit; /* Upper limit on a processed batch */ 358 359 /* 3) dynticks interface. */ 360 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */ 361 int dynticks_snap; /* Per-GP tracking for dynticks. */ 362 363 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */ 364 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */ 365 unsigned long offline_fqs; /* Kicked due to being offline. */ 366 unsigned long cond_resched_completed; 367 /* Grace period that needs help */ 368 /* from cond_resched(). */ 369 370 /* 5) __rcu_pending() statistics. */ 371 unsigned long n_rcu_pending; /* rcu_pending() calls since boot. */ 372 unsigned long n_rp_core_needs_qs; 373 unsigned long n_rp_report_qs; 374 unsigned long n_rp_cb_ready; 375 unsigned long n_rp_cpu_needs_gp; 376 unsigned long n_rp_gp_completed; 377 unsigned long n_rp_gp_started; 378 unsigned long n_rp_nocb_defer_wakeup; 379 unsigned long n_rp_need_nothing; 380 381 /* 6) _rcu_barrier(), OOM callbacks, and expediting. */ 382 struct rcu_head barrier_head; 383 #ifdef CONFIG_RCU_FAST_NO_HZ 384 struct rcu_head oom_head; 385 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */ 386 struct mutex exp_funnel_mutex; 387 388 /* 7) Callback offloading. */ 389 #ifdef CONFIG_RCU_NOCB_CPU 390 struct rcu_head *nocb_head; /* CBs waiting for kthread. */ 391 struct rcu_head **nocb_tail; 392 atomic_long_t nocb_q_count; /* # CBs waiting for nocb */ 393 atomic_long_t nocb_q_count_lazy; /* invocation (all stages). */ 394 struct rcu_head *nocb_follower_head; /* CBs ready to invoke. */ 395 struct rcu_head **nocb_follower_tail; 396 wait_queue_head_t nocb_wq; /* For nocb kthreads to sleep on. */ 397 struct task_struct *nocb_kthread; 398 int nocb_defer_wakeup; /* Defer wakeup of nocb_kthread. */ 399 400 /* The following fields are used by the leader, hence own cacheline. */ 401 struct rcu_head *nocb_gp_head ____cacheline_internodealigned_in_smp; 402 /* CBs waiting for GP. */ 403 struct rcu_head **nocb_gp_tail; 404 bool nocb_leader_sleep; /* Is the nocb leader thread asleep? */ 405 struct rcu_data *nocb_next_follower; 406 /* Next follower in wakeup chain. */ 407 408 /* The following fields are used by the follower, hence new cachline. */ 409 struct rcu_data *nocb_leader ____cacheline_internodealigned_in_smp; 410 /* Leader CPU takes GP-end wakeups. */ 411 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ 412 413 /* 8) RCU CPU stall data. */ 414 unsigned int softirq_snap; /* Snapshot of softirq activity. */ 415 416 int cpu; 417 struct rcu_state *rsp; 418 }; 419 420 /* Values for nocb_defer_wakeup field in struct rcu_data. */ 421 #define RCU_NOGP_WAKE_NOT 0 422 #define RCU_NOGP_WAKE 1 423 #define RCU_NOGP_WAKE_FORCE 2 424 425 #define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500)) 426 /* For jiffies_till_first_fqs and */ 427 /* and jiffies_till_next_fqs. */ 428 429 #define RCU_JIFFIES_FQS_DIV 256 /* Very large systems need more */ 430 /* delay between bouts of */ 431 /* quiescent-state forcing. */ 432 433 #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time to take */ 434 /* at least one scheduling clock */ 435 /* irq before ratting on them. */ 436 437 #define rcu_wait(cond) \ 438 do { \ 439 for (;;) { \ 440 set_current_state(TASK_INTERRUPTIBLE); \ 441 if (cond) \ 442 break; \ 443 schedule(); \ 444 } \ 445 __set_current_state(TASK_RUNNING); \ 446 } while (0) 447 448 /* 449 * RCU global state, including node hierarchy. This hierarchy is 450 * represented in "heap" form in a dense array. The root (first level) 451 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second 452 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]), 453 * and the third level in ->node[m+1] and following (->node[m+1] referenced 454 * by ->level[2]). The number of levels is determined by the number of 455 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy" 456 * consisting of a single rcu_node. 457 */ 458 struct rcu_state { 459 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */ 460 struct rcu_node *level[RCU_NUM_LVLS + 1]; 461 /* Hierarchy levels (+1 to */ 462 /* shut bogus gcc warning) */ 463 u8 flavor_mask; /* bit in flavor mask. */ 464 struct rcu_data __percpu *rda; /* pointer of percu rcu_data. */ 465 call_rcu_func_t call; /* call_rcu() flavor. */ 466 int ncpus; /* # CPUs seen so far. */ 467 468 /* The following fields are guarded by the root rcu_node's lock. */ 469 470 u8 boost ____cacheline_internodealigned_in_smp; 471 /* Subject to priority boost. */ 472 unsigned long gpnum; /* Current gp number. */ 473 unsigned long completed; /* # of last completed gp. */ 474 struct task_struct *gp_kthread; /* Task for grace periods. */ 475 wait_queue_head_t gp_wq; /* Where GP task waits. */ 476 short gp_flags; /* Commands for GP task. */ 477 short gp_state; /* GP kthread sleep state. */ 478 479 /* End of fields guarded by root rcu_node's lock. */ 480 481 raw_spinlock_t orphan_lock ____cacheline_internodealigned_in_smp; 482 /* Protect following fields. */ 483 struct rcu_head *orphan_nxtlist; /* Orphaned callbacks that */ 484 /* need a grace period. */ 485 struct rcu_head **orphan_nxttail; /* Tail of above. */ 486 struct rcu_head *orphan_donelist; /* Orphaned callbacks that */ 487 /* are ready to invoke. */ 488 struct rcu_head **orphan_donetail; /* Tail of above. */ 489 long qlen_lazy; /* Number of lazy callbacks. */ 490 long qlen; /* Total number of callbacks. */ 491 /* End of fields guarded by orphan_lock. */ 492 493 struct mutex barrier_mutex; /* Guards barrier fields. */ 494 atomic_t barrier_cpu_count; /* # CPUs waiting on. */ 495 struct completion barrier_completion; /* Wake at barrier end. */ 496 unsigned long barrier_sequence; /* ++ at start and end of */ 497 /* _rcu_barrier(). */ 498 /* End of fields guarded by barrier_mutex. */ 499 500 unsigned long expedited_sequence; /* Take a ticket. */ 501 atomic_long_t expedited_workdone0; /* # done by others #0. */ 502 atomic_long_t expedited_workdone1; /* # done by others #1. */ 503 atomic_long_t expedited_workdone2; /* # done by others #2. */ 504 atomic_long_t expedited_workdone3; /* # done by others #3. */ 505 atomic_long_t expedited_normal; /* # fallbacks to normal. */ 506 atomic_t expedited_need_qs; /* # CPUs left to check in. */ 507 wait_queue_head_t expedited_wq; /* Wait for check-ins. */ 508 int ncpus_snap; /* # CPUs seen last time. */ 509 510 unsigned long jiffies_force_qs; /* Time at which to invoke */ 511 /* force_quiescent_state(). */ 512 unsigned long n_force_qs; /* Number of calls to */ 513 /* force_quiescent_state(). */ 514 unsigned long n_force_qs_lh; /* ~Number of calls leaving */ 515 /* due to lock unavailable. */ 516 unsigned long n_force_qs_ngp; /* Number of calls leaving */ 517 /* due to no GP active. */ 518 unsigned long gp_start; /* Time at which GP started, */ 519 /* but in jiffies. */ 520 unsigned long gp_activity; /* Time of last GP kthread */ 521 /* activity in jiffies. */ 522 unsigned long jiffies_stall; /* Time at which to check */ 523 /* for CPU stalls. */ 524 unsigned long jiffies_resched; /* Time at which to resched */ 525 /* a reluctant CPU. */ 526 unsigned long n_force_qs_gpstart; /* Snapshot of n_force_qs at */ 527 /* GP start. */ 528 unsigned long gp_max; /* Maximum GP duration in */ 529 /* jiffies. */ 530 const char *name; /* Name of structure. */ 531 char abbr; /* Abbreviated name. */ 532 struct list_head flavors; /* List of RCU flavors. */ 533 }; 534 535 /* Values for rcu_state structure's gp_flags field. */ 536 #define RCU_GP_FLAG_INIT 0x1 /* Need grace-period initialization. */ 537 #define RCU_GP_FLAG_FQS 0x2 /* Need grace-period quiescent-state forcing. */ 538 539 /* Values for rcu_state structure's gp_state field. */ 540 #define RCU_GP_IDLE 0 /* Initial state and no GP in progress. */ 541 #define RCU_GP_WAIT_GPS 1 /* Wait for grace-period start. */ 542 #define RCU_GP_DONE_GPS 2 /* Wait done for grace-period start. */ 543 #define RCU_GP_WAIT_FQS 3 /* Wait for force-quiescent-state time. */ 544 #define RCU_GP_DOING_FQS 4 /* Wait done for force-quiescent-state time. */ 545 #define RCU_GP_CLEANUP 5 /* Grace-period cleanup started. */ 546 #define RCU_GP_CLEANED 6 /* Grace-period cleanup complete. */ 547 548 extern struct list_head rcu_struct_flavors; 549 550 /* Sequence through rcu_state structures for each RCU flavor. */ 551 #define for_each_rcu_flavor(rsp) \ 552 list_for_each_entry((rsp), &rcu_struct_flavors, flavors) 553 554 /* 555 * RCU implementation internal declarations: 556 */ 557 extern struct rcu_state rcu_sched_state; 558 559 extern struct rcu_state rcu_bh_state; 560 561 #ifdef CONFIG_PREEMPT_RCU 562 extern struct rcu_state rcu_preempt_state; 563 #endif /* #ifdef CONFIG_PREEMPT_RCU */ 564 565 #ifdef CONFIG_RCU_BOOST 566 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status); 567 DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu); 568 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops); 569 DECLARE_PER_CPU(char, rcu_cpu_has_work); 570 #endif /* #ifdef CONFIG_RCU_BOOST */ 571 572 #ifndef RCU_TREE_NONCORE 573 574 /* Forward declarations for rcutree_plugin.h */ 575 static void rcu_bootup_announce(void); 576 static void rcu_preempt_note_context_switch(void); 577 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp); 578 #ifdef CONFIG_HOTPLUG_CPU 579 static bool rcu_preempt_has_tasks(struct rcu_node *rnp); 580 #endif /* #ifdef CONFIG_HOTPLUG_CPU */ 581 static void rcu_print_detail_task_stall(struct rcu_state *rsp); 582 static int rcu_print_task_stall(struct rcu_node *rnp); 583 static int rcu_print_task_exp_stall(struct rcu_node *rnp); 584 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp); 585 static void rcu_preempt_check_callbacks(void); 586 void call_rcu(struct rcu_head *head, rcu_callback_t func); 587 static void __init __rcu_init_preempt(void); 588 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags); 589 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp); 590 static void invoke_rcu_callbacks_kthread(void); 591 static bool rcu_is_callbacks_kthread(void); 592 #ifdef CONFIG_RCU_BOOST 593 static void rcu_preempt_do_callbacks(void); 594 static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp, 595 struct rcu_node *rnp); 596 #endif /* #ifdef CONFIG_RCU_BOOST */ 597 static void __init rcu_spawn_boost_kthreads(void); 598 static void rcu_prepare_kthreads(int cpu); 599 static void rcu_cleanup_after_idle(void); 600 static void rcu_prepare_for_idle(void); 601 static void rcu_idle_count_callbacks_posted(void); 602 static bool rcu_preempt_has_tasks(struct rcu_node *rnp); 603 static void print_cpu_stall_info_begin(void); 604 static void print_cpu_stall_info(struct rcu_state *rsp, int cpu); 605 static void print_cpu_stall_info_end(void); 606 static void zero_cpu_stall_ticks(struct rcu_data *rdp); 607 static void increment_cpu_stall_ticks(void); 608 static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu); 609 static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq); 610 static void rcu_nocb_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp); 611 static void rcu_init_one_nocb(struct rcu_node *rnp); 612 static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, 613 bool lazy, unsigned long flags); 614 static bool rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp, 615 struct rcu_data *rdp, 616 unsigned long flags); 617 static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp); 618 static void do_nocb_deferred_wakeup(struct rcu_data *rdp); 619 static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp); 620 static void rcu_spawn_all_nocb_kthreads(int cpu); 621 static void __init rcu_spawn_nocb_kthreads(void); 622 #ifdef CONFIG_RCU_NOCB_CPU 623 static void __init rcu_organize_nocb_kthreads(struct rcu_state *rsp); 624 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ 625 static void __maybe_unused rcu_kick_nohz_cpu(int cpu); 626 static bool init_nocb_callback_list(struct rcu_data *rdp); 627 static void rcu_sysidle_enter(int irq); 628 static void rcu_sysidle_exit(int irq); 629 static void rcu_sysidle_check_cpu(struct rcu_data *rdp, bool *isidle, 630 unsigned long *maxj); 631 static bool is_sysidle_rcu_state(struct rcu_state *rsp); 632 static void rcu_sysidle_report_gp(struct rcu_state *rsp, int isidle, 633 unsigned long maxj); 634 static void rcu_bind_gp_kthread(void); 635 static void rcu_sysidle_init_percpu_data(struct rcu_dynticks *rdtp); 636 static bool rcu_nohz_full_cpu(struct rcu_state *rsp); 637 static void rcu_dynticks_task_enter(void); 638 static void rcu_dynticks_task_exit(void); 639 640 #endif /* #ifndef RCU_TREE_NONCORE */ 641 642 #ifdef CONFIG_RCU_TRACE 643 /* Read out queue lengths for tracing. */ 644 static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll) 645 { 646 #ifdef CONFIG_RCU_NOCB_CPU 647 *ql = atomic_long_read(&rdp->nocb_q_count); 648 *qll = atomic_long_read(&rdp->nocb_q_count_lazy); 649 #else /* #ifdef CONFIG_RCU_NOCB_CPU */ 650 *ql = 0; 651 *qll = 0; 652 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ 653 } 654 #endif /* #ifdef CONFIG_RCU_TRACE */ 655 656 /* 657 * Place this after a lock-acquisition primitive to guarantee that 658 * an UNLOCK+LOCK pair act as a full barrier. This guarantee applies 659 * if the UNLOCK and LOCK are executed by the same CPU or if the 660 * UNLOCK and LOCK operate on the same lock variable. 661 */ 662 #ifdef CONFIG_PPC 663 #define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */ 664 #else /* #ifdef CONFIG_PPC */ 665 #define smp_mb__after_unlock_lock() do { } while (0) 666 #endif /* #else #ifdef CONFIG_PPC */ 667