1 /* SPDX-License-Identifier: GPL-2.0+ */ 2 /* 3 * Read-Copy Update mechanism for mutual exclusion (tree-based version) 4 * Internal non-public definitions. 5 * 6 * Copyright IBM Corporation, 2008 7 * 8 * Author: Ingo Molnar <mingo@elte.hu> 9 * Paul E. McKenney <paulmck@linux.ibm.com> 10 */ 11 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/stop_machine.h> 20 #include <linux/rcu_node_tree.h> 21 22 #include "rcu_segcblist.h" 23 24 /* Communicate arguments to a workqueue handler. */ 25 struct rcu_exp_work { 26 unsigned long rew_s; 27 struct work_struct rew_work; 28 }; 29 30 /* RCU's kthread states for tracing. */ 31 #define RCU_KTHREAD_STOPPED 0 32 #define RCU_KTHREAD_RUNNING 1 33 #define RCU_KTHREAD_WAITING 2 34 #define RCU_KTHREAD_OFFCPU 3 35 #define RCU_KTHREAD_YIELDING 4 36 #define RCU_KTHREAD_MAX 4 37 38 /* 39 * Definition for node within the RCU grace-period-detection hierarchy. 40 */ 41 struct rcu_node { 42 raw_spinlock_t __private lock; /* Root rcu_node's lock protects */ 43 /* some rcu_state fields as well as */ 44 /* following. */ 45 unsigned long gp_seq; /* Track rsp->rcu_gp_seq. */ 46 unsigned long gp_seq_needed; /* Track furthest future GP request. */ 47 unsigned long completedqs; /* All QSes done for this node. */ 48 unsigned long qsmask; /* CPUs or groups that need to switch in */ 49 /* order for current grace period to proceed.*/ 50 /* In leaf rcu_node, each bit corresponds to */ 51 /* an rcu_data structure, otherwise, each */ 52 /* bit corresponds to a child rcu_node */ 53 /* structure. */ 54 unsigned long rcu_gp_init_mask; /* Mask of offline CPUs at GP init. */ 55 unsigned long qsmaskinit; 56 /* Per-GP initial value for qsmask. */ 57 /* Initialized from ->qsmaskinitnext at the */ 58 /* beginning of each grace period. */ 59 unsigned long qsmaskinitnext; 60 /* Online CPUs for next grace period. */ 61 unsigned long expmask; /* CPUs or groups that need to check in */ 62 /* to allow the current expedited GP */ 63 /* to complete. */ 64 unsigned long expmaskinit; 65 /* Per-GP initial values for expmask. */ 66 /* Initialized from ->expmaskinitnext at the */ 67 /* beginning of each expedited GP. */ 68 unsigned long expmaskinitnext; 69 /* Online CPUs for next expedited GP. */ 70 /* Any CPU that has ever been online will */ 71 /* have its bit set. */ 72 unsigned long ffmask; /* Fully functional CPUs. */ 73 unsigned long grpmask; /* Mask to apply to parent qsmask. */ 74 /* Only one bit will be set in this mask. */ 75 int grplo; /* lowest-numbered CPU or group here. */ 76 int grphi; /* highest-numbered CPU or group here. */ 77 u8 grpnum; /* CPU/group number for next level up. */ 78 u8 level; /* root is at level 0. */ 79 bool wait_blkd_tasks;/* Necessary to wait for blocked tasks to */ 80 /* exit RCU read-side critical sections */ 81 /* before propagating offline up the */ 82 /* rcu_node tree? */ 83 struct rcu_node *parent; 84 struct list_head blkd_tasks; 85 /* Tasks blocked in RCU read-side critical */ 86 /* section. Tasks are placed at the head */ 87 /* of this list and age towards the tail. */ 88 struct list_head *gp_tasks; 89 /* Pointer to the first task blocking the */ 90 /* current grace period, or NULL if there */ 91 /* is no such task. */ 92 struct list_head *exp_tasks; 93 /* Pointer to the first task blocking the */ 94 /* current expedited grace period, or NULL */ 95 /* if there is no such task. If there */ 96 /* is no current expedited grace period, */ 97 /* then there can cannot be any such task. */ 98 struct list_head *boost_tasks; 99 /* Pointer to first task that needs to be */ 100 /* priority boosted, or NULL if no priority */ 101 /* boosting is needed for this rcu_node */ 102 /* structure. If there are no tasks */ 103 /* queued on this rcu_node structure that */ 104 /* are blocking the current grace period, */ 105 /* there can be no such task. */ 106 struct rt_mutex boost_mtx; 107 /* Used only for the priority-boosting */ 108 /* side effect, not as a lock. */ 109 unsigned long boost_time; 110 /* When to start boosting (jiffies). */ 111 struct task_struct *boost_kthread_task; 112 /* kthread that takes care of priority */ 113 /* boosting for this rcu_node structure. */ 114 unsigned int boost_kthread_status; 115 /* State of boost_kthread_task for tracing. */ 116 #ifdef CONFIG_RCU_NOCB_CPU 117 struct swait_queue_head nocb_gp_wq[2]; 118 /* Place for rcu_nocb_kthread() to wait GP. */ 119 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ 120 raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp; 121 122 spinlock_t exp_lock ____cacheline_internodealigned_in_smp; 123 unsigned long exp_seq_rq; 124 wait_queue_head_t exp_wq[4]; 125 struct rcu_exp_work rew; 126 bool exp_need_flush; /* Need to flush workitem? */ 127 } ____cacheline_internodealigned_in_smp; 128 129 /* 130 * Bitmasks in an rcu_node cover the interval [grplo, grphi] of CPU IDs, and 131 * are indexed relative to this interval rather than the global CPU ID space. 132 * This generates the bit for a CPU in node-local masks. 133 */ 134 #define leaf_node_cpu_bit(rnp, cpu) (BIT((cpu) - (rnp)->grplo)) 135 136 /* 137 * Union to allow "aggregate OR" operation on the need for a quiescent 138 * state by the normal and expedited grace periods. 139 */ 140 union rcu_noqs { 141 struct { 142 u8 norm; 143 u8 exp; 144 } b; /* Bits. */ 145 u16 s; /* Set of bits, aggregate OR here. */ 146 }; 147 148 /* Per-CPU data for read-copy update. */ 149 struct rcu_data { 150 /* 1) quiescent-state and grace-period handling : */ 151 unsigned long gp_seq; /* Track rsp->rcu_gp_seq counter. */ 152 unsigned long gp_seq_needed; /* Track furthest future GP request. */ 153 union rcu_noqs cpu_no_qs; /* No QSes yet for this CPU. */ 154 bool core_needs_qs; /* Core waits for quiesc state. */ 155 bool beenonline; /* CPU online at least once. */ 156 bool gpwrap; /* Possible ->gp_seq wrap. */ 157 bool exp_deferred_qs; /* This CPU awaiting a deferred QS? */ 158 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */ 159 unsigned long grpmask; /* Mask to apply to leaf qsmask. */ 160 unsigned long ticks_this_gp; /* The number of scheduling-clock */ 161 /* ticks this CPU has handled */ 162 /* during and after the last grace */ 163 /* period it is aware of. */ 164 struct irq_work defer_qs_iw; /* Obtain later scheduler attention. */ 165 bool defer_qs_iw_pending; /* Scheduler attention pending? */ 166 167 /* 2) batch handling */ 168 struct rcu_segcblist cblist; /* Segmented callback list, with */ 169 /* different callbacks waiting for */ 170 /* different grace periods. */ 171 long qlen_last_fqs_check; 172 /* qlen at last check for QS forcing */ 173 unsigned long n_force_qs_snap; 174 /* did other CPU force QS recently? */ 175 long blimit; /* Upper limit on a processed batch */ 176 177 /* 3) dynticks interface. */ 178 int dynticks_snap; /* Per-GP tracking for dynticks. */ 179 long dynticks_nesting; /* Track process nesting level. */ 180 long dynticks_nmi_nesting; /* Track irq/NMI nesting level. */ 181 atomic_t dynticks; /* Even value for idle, else odd. */ 182 bool rcu_need_heavy_qs; /* GP old, so heavy quiescent state! */ 183 bool rcu_urgent_qs; /* GP old need light quiescent state. */ 184 #ifdef CONFIG_RCU_FAST_NO_HZ 185 bool all_lazy; /* All CPU's CBs lazy at idle start? */ 186 unsigned long last_accelerate; /* Last jiffy CBs were accelerated. */ 187 unsigned long last_advance_all; /* Last jiffy CBs were all advanced. */ 188 int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */ 189 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */ 190 191 /* 4) rcu_barrier(), OOM callbacks, and expediting. */ 192 struct rcu_head barrier_head; 193 int exp_dynticks_snap; /* Double-check need for IPI. */ 194 195 /* 5) Callback offloading. */ 196 #ifdef CONFIG_RCU_NOCB_CPU 197 struct swait_queue_head nocb_cb_wq; /* For nocb kthreads to sleep on. */ 198 struct task_struct *nocb_gp_kthread; 199 raw_spinlock_t nocb_lock; /* Guard following pair of fields. */ 200 atomic_t nocb_lock_contended; /* Contention experienced. */ 201 int nocb_defer_wakeup; /* Defer wakeup of nocb_kthread. */ 202 struct timer_list nocb_timer; /* Enforce finite deferral. */ 203 204 /* The following fields are used by GP kthread, hence own cacheline. */ 205 raw_spinlock_t nocb_gp_lock ____cacheline_internodealigned_in_smp; 206 bool nocb_gp_sleep; 207 /* Is the nocb GP thread asleep? */ 208 struct swait_queue_head nocb_gp_wq; /* For nocb kthreads to sleep on. */ 209 bool nocb_cb_sleep; /* Is the nocb CB thread asleep? */ 210 struct task_struct *nocb_cb_kthread; 211 struct rcu_data *nocb_next_cb_rdp; 212 /* Next rcu_data in wakeup chain. */ 213 214 /* The following fields are used by CB kthread, hence new cachline. */ 215 struct rcu_data *nocb_gp_rdp ____cacheline_internodealigned_in_smp; 216 /* GP rdp takes GP-end wakeups. */ 217 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ 218 219 /* 6) RCU priority boosting. */ 220 struct task_struct *rcu_cpu_kthread_task; 221 /* rcuc per-CPU kthread or NULL. */ 222 unsigned int rcu_cpu_kthread_status; 223 char rcu_cpu_has_work; 224 225 /* 7) Diagnostic data, including RCU CPU stall warnings. */ 226 unsigned int softirq_snap; /* Snapshot of softirq activity. */ 227 /* ->rcu_iw* fields protected by leaf rcu_node ->lock. */ 228 struct irq_work rcu_iw; /* Check for non-irq activity. */ 229 bool rcu_iw_pending; /* Is ->rcu_iw pending? */ 230 unsigned long rcu_iw_gp_seq; /* ->gp_seq associated with ->rcu_iw. */ 231 unsigned long rcu_ofl_gp_seq; /* ->gp_seq at last offline. */ 232 short rcu_ofl_gp_flags; /* ->gp_flags at last offline. */ 233 unsigned long rcu_onl_gp_seq; /* ->gp_seq at last online. */ 234 short rcu_onl_gp_flags; /* ->gp_flags at last online. */ 235 unsigned long last_fqs_resched; /* Time of last rcu_resched(). */ 236 237 int cpu; 238 }; 239 240 /* Values for nocb_defer_wakeup field in struct rcu_data. */ 241 #define RCU_NOCB_WAKE_NOT 0 242 #define RCU_NOCB_WAKE 1 243 #define RCU_NOCB_WAKE_FORCE 2 244 245 #define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500)) 246 /* For jiffies_till_first_fqs and */ 247 /* and jiffies_till_next_fqs. */ 248 249 #define RCU_JIFFIES_FQS_DIV 256 /* Very large systems need more */ 250 /* delay between bouts of */ 251 /* quiescent-state forcing. */ 252 253 #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time to take */ 254 /* at least one scheduling clock */ 255 /* irq before ratting on them. */ 256 257 #define rcu_wait(cond) \ 258 do { \ 259 for (;;) { \ 260 set_current_state(TASK_INTERRUPTIBLE); \ 261 if (cond) \ 262 break; \ 263 schedule(); \ 264 } \ 265 __set_current_state(TASK_RUNNING); \ 266 } while (0) 267 268 /* 269 * RCU global state, including node hierarchy. This hierarchy is 270 * represented in "heap" form in a dense array. The root (first level) 271 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second 272 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]), 273 * and the third level in ->node[m+1] and following (->node[m+1] referenced 274 * by ->level[2]). The number of levels is determined by the number of 275 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy" 276 * consisting of a single rcu_node. 277 */ 278 struct rcu_state { 279 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */ 280 struct rcu_node *level[RCU_NUM_LVLS + 1]; 281 /* Hierarchy levels (+1 to */ 282 /* shut bogus gcc warning) */ 283 int ncpus; /* # CPUs seen so far. */ 284 285 /* The following fields are guarded by the root rcu_node's lock. */ 286 287 u8 boost ____cacheline_internodealigned_in_smp; 288 /* Subject to priority boost. */ 289 unsigned long gp_seq; /* Grace-period sequence #. */ 290 struct task_struct *gp_kthread; /* Task for grace periods. */ 291 struct swait_queue_head gp_wq; /* Where GP task waits. */ 292 short gp_flags; /* Commands for GP task. */ 293 short gp_state; /* GP kthread sleep state. */ 294 unsigned long gp_wake_time; /* Last GP kthread wake. */ 295 unsigned long gp_wake_seq; /* ->gp_seq at ^^^. */ 296 297 /* End of fields guarded by root rcu_node's lock. */ 298 299 struct mutex barrier_mutex; /* Guards barrier fields. */ 300 atomic_t barrier_cpu_count; /* # CPUs waiting on. */ 301 struct completion barrier_completion; /* Wake at barrier end. */ 302 unsigned long barrier_sequence; /* ++ at start and end of */ 303 /* rcu_barrier(). */ 304 /* End of fields guarded by barrier_mutex. */ 305 306 struct mutex exp_mutex; /* Serialize expedited GP. */ 307 struct mutex exp_wake_mutex; /* Serialize wakeup. */ 308 unsigned long expedited_sequence; /* Take a ticket. */ 309 atomic_t expedited_need_qs; /* # CPUs left to check in. */ 310 struct swait_queue_head expedited_wq; /* Wait for check-ins. */ 311 int ncpus_snap; /* # CPUs seen last time. */ 312 313 unsigned long jiffies_force_qs; /* Time at which to invoke */ 314 /* force_quiescent_state(). */ 315 unsigned long jiffies_kick_kthreads; /* Time at which to kick */ 316 /* kthreads, if configured. */ 317 unsigned long n_force_qs; /* Number of calls to */ 318 /* force_quiescent_state(). */ 319 unsigned long gp_start; /* Time at which GP started, */ 320 /* but in jiffies. */ 321 unsigned long gp_end; /* Time last GP ended, again */ 322 /* in jiffies. */ 323 unsigned long gp_activity; /* Time of last GP kthread */ 324 /* activity in jiffies. */ 325 unsigned long gp_req_activity; /* Time of last GP request */ 326 /* in jiffies. */ 327 unsigned long jiffies_stall; /* Time at which to check */ 328 /* for CPU stalls. */ 329 unsigned long jiffies_resched; /* Time at which to resched */ 330 /* a reluctant CPU. */ 331 unsigned long n_force_qs_gpstart; /* Snapshot of n_force_qs at */ 332 /* GP start. */ 333 unsigned long gp_max; /* Maximum GP duration in */ 334 /* jiffies. */ 335 const char *name; /* Name of structure. */ 336 char abbr; /* Abbreviated name. */ 337 338 raw_spinlock_t ofl_lock ____cacheline_internodealigned_in_smp; 339 /* Synchronize offline with */ 340 /* GP pre-initialization. */ 341 }; 342 343 /* Values for rcu_state structure's gp_flags field. */ 344 #define RCU_GP_FLAG_INIT 0x1 /* Need grace-period initialization. */ 345 #define RCU_GP_FLAG_FQS 0x2 /* Need grace-period quiescent-state forcing. */ 346 347 /* Values for rcu_state structure's gp_state field. */ 348 #define RCU_GP_IDLE 0 /* Initial state and no GP in progress. */ 349 #define RCU_GP_WAIT_GPS 1 /* Wait for grace-period start. */ 350 #define RCU_GP_DONE_GPS 2 /* Wait done for grace-period start. */ 351 #define RCU_GP_ONOFF 3 /* Grace-period initialization hotplug. */ 352 #define RCU_GP_INIT 4 /* Grace-period initialization. */ 353 #define RCU_GP_WAIT_FQS 5 /* Wait for force-quiescent-state time. */ 354 #define RCU_GP_DOING_FQS 6 /* Wait done for force-quiescent-state time. */ 355 #define RCU_GP_CLEANUP 7 /* Grace-period cleanup started. */ 356 #define RCU_GP_CLEANED 8 /* Grace-period cleanup complete. */ 357 358 static const char * const gp_state_names[] = { 359 "RCU_GP_IDLE", 360 "RCU_GP_WAIT_GPS", 361 "RCU_GP_DONE_GPS", 362 "RCU_GP_ONOFF", 363 "RCU_GP_INIT", 364 "RCU_GP_WAIT_FQS", 365 "RCU_GP_DOING_FQS", 366 "RCU_GP_CLEANUP", 367 "RCU_GP_CLEANED", 368 }; 369 370 /* 371 * In order to export the rcu_state name to the tracing tools, it 372 * needs to be added in the __tracepoint_string section. 373 * This requires defining a separate variable tp_<sname>_varname 374 * that points to the string being used, and this will allow 375 * the tracing userspace tools to be able to decipher the string 376 * address to the matching string. 377 */ 378 #ifdef CONFIG_PREEMPT_RCU 379 #define RCU_ABBR 'p' 380 #define RCU_NAME_RAW "rcu_preempt" 381 #else /* #ifdef CONFIG_PREEMPT_RCU */ 382 #define RCU_ABBR 's' 383 #define RCU_NAME_RAW "rcu_sched" 384 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ 385 #ifndef CONFIG_TRACING 386 #define RCU_NAME RCU_NAME_RAW 387 #else /* #ifdef CONFIG_TRACING */ 388 static char rcu_name[] = RCU_NAME_RAW; 389 static const char *tp_rcu_varname __used __tracepoint_string = rcu_name; 390 #define RCU_NAME rcu_name 391 #endif /* #else #ifdef CONFIG_TRACING */ 392 393 int rcu_dynticks_snap(struct rcu_data *rdp); 394 395 /* Forward declarations for tree_plugin.h */ 396 static void rcu_bootup_announce(void); 397 static void rcu_qs(void); 398 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp); 399 #ifdef CONFIG_HOTPLUG_CPU 400 static bool rcu_preempt_has_tasks(struct rcu_node *rnp); 401 #endif /* #ifdef CONFIG_HOTPLUG_CPU */ 402 static int rcu_print_task_exp_stall(struct rcu_node *rnp); 403 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp); 404 static void rcu_flavor_sched_clock_irq(int user); 405 void call_rcu(struct rcu_head *head, rcu_callback_t func); 406 static void dump_blkd_tasks(struct rcu_node *rnp, int ncheck); 407 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags); 408 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp); 409 static bool rcu_is_callbacks_kthread(void); 410 static void rcu_cpu_kthread_setup(unsigned int cpu); 411 static void __init rcu_spawn_boost_kthreads(void); 412 static void rcu_prepare_kthreads(int cpu); 413 static void rcu_cleanup_after_idle(void); 414 static void rcu_prepare_for_idle(void); 415 static bool rcu_preempt_has_tasks(struct rcu_node *rnp); 416 static bool rcu_preempt_need_deferred_qs(struct task_struct *t); 417 static void rcu_preempt_deferred_qs(struct task_struct *t); 418 static void zero_cpu_stall_ticks(struct rcu_data *rdp); 419 static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp); 420 static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq); 421 static void rcu_init_one_nocb(struct rcu_node *rnp); 422 static void __call_rcu_nocb_wake(struct rcu_data *rdp, bool was_empty, 423 unsigned long flags); 424 static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp); 425 static void do_nocb_deferred_wakeup(struct rcu_data *rdp); 426 static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp); 427 static void rcu_spawn_cpu_nocb_kthread(int cpu); 428 static void __init rcu_spawn_nocb_kthreads(void); 429 static void rcu_nocb_lock(struct rcu_data *rdp); 430 static void rcu_nocb_unlock(struct rcu_data *rdp); 431 static void rcu_nocb_unlock_irqrestore(struct rcu_data *rdp, 432 unsigned long flags); 433 #ifdef CONFIG_RCU_NOCB_CPU 434 static void __init rcu_organize_nocb_kthreads(void); 435 #define rcu_nocb_lock_irqsave(rdp, flags) \ 436 do { \ 437 if (!rcu_segcblist_is_offloaded(&(rdp)->cblist)) { \ 438 local_irq_save(flags); \ 439 } else if (!raw_spin_trylock_irqsave(&(rdp)->nocb_lock, (flags))) {\ 440 atomic_inc(&(rdp)->nocb_lock_contended); \ 441 smp_mb__after_atomic(); /* atomic_inc() before lock. */ \ 442 raw_spin_lock_irqsave(&(rdp)->nocb_lock, (flags)); \ 443 smp_mb__before_atomic(); /* atomic_dec() after lock. */ \ 444 atomic_dec(&(rdp)->nocb_lock_contended); \ 445 } \ 446 } while (0) 447 #else /* #ifdef CONFIG_RCU_NOCB_CPU */ 448 #define rcu_nocb_lock_irqsave(rdp, flags) local_irq_save(flags) 449 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ 450 451 static void rcu_bind_gp_kthread(void); 452 static bool rcu_nohz_full_cpu(void); 453 static void rcu_dynticks_task_enter(void); 454 static void rcu_dynticks_task_exit(void); 455 456 /* Forward declarations for tree_stall.h */ 457 static void record_gp_stall_check_time(void); 458 static void rcu_iw_handler(struct irq_work *iwp); 459 static void check_cpu_stall(struct rcu_data *rdp); 460 static void rcu_check_gp_start_stall(struct rcu_node *rnp, struct rcu_data *rdp, 461 const unsigned long gpssdelay); 462