kernel/rcu/tree.h

   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
  39 #define MAX_RCU_LVLS 4
  40
  41 #ifdef CONFIG_RCU_FANOUT
  42 #define RCU_FANOUT CONFIG_RCU_FANOUT
  43 #else /* #ifdef CONFIG_RCU_FANOUT */
  44 # ifdef CONFIG_64BIT
  45 # define RCU_FANOUT 64
  46 # else
  47 # define RCU_FANOUT 32
  48 # endif
  49 #endif /* #else #ifdef CONFIG_RCU_FANOUT */
  50
  51 #ifdef CONFIG_RCU_FANOUT_LEAF
  52 #define RCU_FANOUT_LEAF CONFIG_RCU_FANOUT_LEAF
  53 #else /* #ifdef CONFIG_RCU_FANOUT_LEAF */
  54 # ifdef CONFIG_64BIT
  55 # define RCU_FANOUT_LEAF 64
  56 # else
  57 # define RCU_FANOUT_LEAF 32
  58 # endif
  59 #endif /* #else #ifdef CONFIG_RCU_FANOUT_LEAF */
  60
  61 #define RCU_FANOUT_1          (RCU_FANOUT_LEAF)
  62 #define RCU_FANOUT_2          (RCU_FANOUT_1 * RCU_FANOUT)
  63 #define RCU_FANOUT_3          (RCU_FANOUT_2 * RCU_FANOUT)
  64 #define RCU_FANOUT_4          (RCU_FANOUT_3 * RCU_FANOUT)
  65
  66 #if NR_CPUS <= RCU_FANOUT_1
  67 #  define RCU_NUM_LVLS        1
  68 #  define NUM_RCU_LVL_0       1
  69 #  define NUM_RCU_LVL_1       (NR_CPUS)
  70 #  define NUM_RCU_LVL_2       0
  71 #  define NUM_RCU_LVL_3       0
  72 #  define NUM_RCU_LVL_4       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_LVL_2       (NR_CPUS)
  78 #  define NUM_RCU_LVL_3       0
  79 #  define NUM_RCU_LVL_4       0
  80 #elif NR_CPUS <= RCU_FANOUT_3
  81 #  define RCU_NUM_LVLS        3
  82 #  define NUM_RCU_LVL_0       1
  83 #  define NUM_RCU_LVL_1       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
  84 #  define NUM_RCU_LVL_2       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
  85 #  define NUM_RCU_LVL_3       (NR_CPUS)
  86 #  define NUM_RCU_LVL_4       0
  87 #elif NR_CPUS <= RCU_FANOUT_4
  88 #  define RCU_NUM_LVLS        4
  89 #  define NUM_RCU_LVL_0       1
  90 #  define NUM_RCU_LVL_1       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3)
  91 #  define NUM_RCU_LVL_2       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
  92 #  define NUM_RCU_LVL_3       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
  93 #  define NUM_RCU_LVL_4       (NR_CPUS)
  94 #else
  95 # error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
  96 #endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
  97
  98 #define RCU_SUM (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3 + NUM_RCU_LVL_4)
  99 #define NUM_RCU_NODES (RCU_SUM - NR_CPUS)
 100
 101 extern int rcu_num_lvls;
 102 extern int rcu_num_nodes;
 103
 104 /*
 105  * Dynticks per-CPU state.
 106  */
 107 struct rcu_dynticks {
 108         long long dynticks_nesting; /* Track irq/process nesting level. */
 109                                     /* Process level is worth LLONG_MAX/2. */
 110         int dynticks_nmi_nesting;   /* Track NMI nesting level. */
 111         atomic_t dynticks;          /* Even value for idle, else odd. */
 112 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
 113         long long dynticks_idle_nesting;
 114                                     /* irq/process nesting level from idle. */
 115         atomic_t dynticks_idle;     /* Even value for idle, else odd. */
 116                                     /*  "Idle" excludes userspace execution. */
 117         unsigned long dynticks_idle_jiffies;
 118                                     /* End of last non-NMI non-idle period. */
 119 #endif /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
 120 #ifdef CONFIG_RCU_FAST_NO_HZ
 121         bool all_lazy;              /* Are all CPU's CBs lazy? */
 122         unsigned long nonlazy_posted;
 123                                     /* # times non-lazy CBs posted to CPU. */
 124         unsigned long nonlazy_posted_snap;
 125                                     /* idle-period nonlazy_posted snapshot. */
 126         unsigned long last_accelerate;
 127                                     /* Last jiffy CBs were accelerated. */
 128         unsigned long last_advance_all;
 129                                     /* Last jiffy CBs were all advanced. */
 130         int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */
 131 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
 132 };
 133
 134 /* RCU's kthread states for tracing. */
 135 #define RCU_KTHREAD_STOPPED  0
 136 #define RCU_KTHREAD_RUNNING  1
 137 #define RCU_KTHREAD_WAITING  2
 138 #define RCU_KTHREAD_OFFCPU   3
 139 #define RCU_KTHREAD_YIELDING 4
 140 #define RCU_KTHREAD_MAX      4
 141
 142 /*
 143  * Definition for node within the RCU grace-period-detection hierarchy.
 144  */
 145 struct rcu_node {
 146         raw_spinlock_t lock;    /* Root rcu_node's lock protects some */
 147                                 /*  rcu_state fields as well as following. */
 148         unsigned long gpnum;    /* Current grace period for this node. */
 149                                 /*  This will either be equal to or one */
 150                                 /*  behind the root rcu_node's gpnum. */
 151         unsigned long completed; /* Last GP completed for this node. */
 152                                 /*  This will either be equal to or one */
 153                                 /*  behind the root rcu_node's gpnum. */
 154         unsigned long qsmask;   /* CPUs or groups that need to switch in */
 155                                 /*  order for current grace period to proceed.*/
 156                                 /*  In leaf rcu_node, each bit corresponds to */
 157                                 /*  an rcu_data structure, otherwise, each */
 158                                 /*  bit corresponds to a child rcu_node */
 159                                 /*  structure. */
 160         unsigned long expmask;  /* Groups that have ->blkd_tasks */
 161                                 /*  elements that need to drain to allow the */
 162                                 /*  current expedited grace period to */
 163                                 /*  complete (only for PREEMPT_RCU). */
 164         unsigned long qsmaskinit;
 165                                 /* Per-GP initial value for qsmask & expmask. */
 166                                 /*  Initialized from ->qsmaskinitnext at the */
 167                                 /*  beginning of each grace period. */
 168         unsigned long qsmaskinitnext;
 169                                 /* Online CPUs for next grace period. */
 170         unsigned long grpmask;  /* Mask to apply to parent qsmask. */
 171                                 /*  Only one bit will be set in this mask. */
 172         int     grplo;          /* lowest-numbered CPU or group here. */
 173         int     grphi;          /* highest-numbered CPU or group here. */
 174         u8      grpnum;         /* CPU/group number for next level up. */
 175         u8      level;          /* root is at level 0. */
 176         bool    wait_blkd_tasks;/* Necessary to wait for blocked tasks to */
 177                                 /*  exit RCU read-side critical sections */
 178                                 /*  before propagating offline up the */
 179                                 /*  rcu_node tree? */
 180         struct rcu_node *parent;
 181         struct list_head blkd_tasks;
 182                                 /* Tasks blocked in RCU read-side critical */
 183                                 /*  section.  Tasks are placed at the head */
 184                                 /*  of this list and age towards the tail. */
 185         struct list_head *gp_tasks;
 186                                 /* Pointer to the first task blocking the */
 187                                 /*  current grace period, or NULL if there */
 188                                 /*  is no such task. */
 189         struct list_head *exp_tasks;
 190                                 /* Pointer to the first task blocking the */
 191                                 /*  current expedited grace period, or NULL */
 192                                 /*  if there is no such task.  If there */
 193                                 /*  is no current expedited grace period, */
 194                                 /*  then there can cannot be any such task. */
 195         struct list_head *boost_tasks;
 196                                 /* Pointer to first task that needs to be */
 197                                 /*  priority boosted, or NULL if no priority */
 198                                 /*  boosting is needed for this rcu_node */
 199                                 /*  structure.  If there are no tasks */
 200                                 /*  queued on this rcu_node structure that */
 201                                 /*  are blocking the current grace period, */
 202                                 /*  there can be no such task. */
 203         struct rt_mutex boost_mtx;
 204                                 /* Used only for the priority-boosting */
 205                                 /*  side effect, not as a lock. */
 206         unsigned long boost_time;
 207                                 /* When to start boosting (jiffies). */
 208         struct task_struct *boost_kthread_task;
 209                                 /* kthread that takes care of priority */
 210                                 /*  boosting for this rcu_node structure. */
 211         unsigned int boost_kthread_status;
 212                                 /* State of boost_kthread_task for tracing. */
 213         unsigned long n_tasks_boosted;
 214                                 /* Total number of tasks boosted. */
 215         unsigned long n_exp_boosts;
 216                                 /* Number of tasks boosted for expedited GP. */
 217         unsigned long n_normal_boosts;
 218                                 /* Number of tasks boosted for normal GP. */
 219         unsigned long n_balk_blkd_tasks;
 220                                 /* Refused to boost: no blocked tasks. */
 221         unsigned long n_balk_exp_gp_tasks;
 222                                 /* Refused to boost: nothing blocking GP. */
 223         unsigned long n_balk_boost_tasks;
 224                                 /* Refused to boost: already boosting. */
 225         unsigned long n_balk_notblocked;
 226                                 /* Refused to boost: RCU RS CS still running. */
 227         unsigned long n_balk_notyet;
 228                                 /* Refused to boost: not yet time. */
 229         unsigned long n_balk_nos;
 230                                 /* Refused to boost: not sure why, though. */
 231                                 /*  This can happen due to race conditions. */
 232 #ifdef CONFIG_RCU_NOCB_CPU
 233         wait_queue_head_t nocb_gp_wq[2];
 234                                 /* Place for rcu_nocb_kthread() to wait GP. */
 235 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
 236         int need_future_gp[2];
 237                                 /* Counts of upcoming no-CB GP requests. */
 238         raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp;
 239 } ____cacheline_internodealigned_in_smp;
 240
 241 /*
 242  * Do a full breadth-first scan of the rcu_node structures for the
 243  * specified rcu_state structure.
 244  */
 245 #define rcu_for_each_node_breadth_first(rsp, rnp) \
 246         for ((rnp) = &(rsp)->node[0]; \
 247              (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
 248
 249 /*
 250  * Do a breadth-first scan of the non-leaf rcu_node structures for the
 251  * specified rcu_state structure.  Note that if there is a singleton
 252  * rcu_node tree with but one rcu_node structure, this loop is a no-op.
 253  */
 254 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
 255         for ((rnp) = &(rsp)->node[0]; \
 256              (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
 257
 258 /*
 259  * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
 260  * structure.  Note that if there is a singleton rcu_node tree with but
 261  * one rcu_node structure, this loop -will- visit the rcu_node structure.
 262  * It is still a leaf node, even if it is also the root node.
 263  */
 264 #define rcu_for_each_leaf_node(rsp, rnp) \
 265         for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
 266              (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
 267
 268 /* Index values for nxttail array in struct rcu_data. */
 269 #define RCU_DONE_TAIL           0       /* Also RCU_WAIT head. */
 270 #define RCU_WAIT_TAIL           1       /* Also RCU_NEXT_READY head. */
 271 #define RCU_NEXT_READY_TAIL     2       /* Also RCU_NEXT head. */
 272 #define RCU_NEXT_TAIL           3
 273 #define RCU_NEXT_SIZE           4
 274
 275 /* Per-CPU data for read-copy update. */
 276 struct rcu_data {
 277         /* 1) quiescent-state and grace-period handling : */
 278         unsigned long   completed;      /* Track rsp->completed gp number */
 279                                         /*  in order to detect GP end. */
 280         unsigned long   gpnum;          /* Highest gp number that this CPU */
 281                                         /*  is aware of having started. */
 282         unsigned long   rcu_qs_ctr_snap;/* Snapshot of rcu_qs_ctr to check */
 283                                         /*  for rcu_all_qs() invocations. */
 284         bool            passed_quiesce; /* User-mode/idle loop etc. */
 285         bool            qs_pending;     /* Core waits for quiesc state. */
 286         bool            beenonline;     /* CPU online at least once. */
 287         bool            gpwrap;         /* Possible gpnum/completed wrap. */
 288         struct rcu_node *mynode;        /* This CPU's leaf of hierarchy */
 289         unsigned long grpmask;          /* Mask to apply to leaf qsmask. */
 290 #ifdef CONFIG_RCU_CPU_STALL_INFO
 291         unsigned long   ticks_this_gp;  /* The number of scheduling-clock */
 292                                         /*  ticks this CPU has handled */
 293                                         /*  during and after the last grace */
 294                                         /* period it is aware of. */
 295 #endif /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
 296
 297         /* 2) batch handling */
 298         /*
 299          * If nxtlist is not NULL, it is partitioned as follows.
 300          * Any of the partitions might be empty, in which case the
 301          * pointer to that partition will be equal to the pointer for
 302          * the following partition.  When the list is empty, all of
 303          * the nxttail elements point to the ->nxtlist pointer itself,
 304          * which in that case is NULL.
 305          *
 306          * [nxtlist, *nxttail[RCU_DONE_TAIL]):
 307          *      Entries that batch # <= ->completed
 308          *      The grace period for these entries has completed, and
 309          *      the other grace-period-completed entries may be moved
 310          *      here temporarily in rcu_process_callbacks().
 311          * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
 312          *      Entries that batch # <= ->completed - 1: waiting for current GP
 313          * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
 314          *      Entries known to have arrived before current GP ended
 315          * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
 316          *      Entries that might have arrived after current GP ended
 317          *      Note that the value of *nxttail[RCU_NEXT_TAIL] will
 318          *      always be NULL, as this is the end of the list.
 319          */
 320         struct rcu_head *nxtlist;
 321         struct rcu_head **nxttail[RCU_NEXT_SIZE];
 322         unsigned long   nxtcompleted[RCU_NEXT_SIZE];
 323                                         /* grace periods for sublists. */
 324         long            qlen_lazy;      /* # of lazy queued callbacks */
 325         long            qlen;           /* # of queued callbacks, incl lazy */
 326         long            qlen_last_fqs_check;
 327                                         /* qlen at last check for QS forcing */
 328         unsigned long   n_cbs_invoked;  /* count of RCU cbs invoked. */
 329         unsigned long   n_nocbs_invoked; /* count of no-CBs RCU cbs invoked. */
 330         unsigned long   n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
 331         unsigned long   n_cbs_adopted;  /* RCU cbs adopted from dying CPU */
 332         unsigned long   n_force_qs_snap;
 333                                         /* did other CPU force QS recently? */
 334         long            blimit;         /* Upper limit on a processed batch */
 335
 336         /* 3) dynticks interface. */
 337         struct rcu_dynticks *dynticks;  /* Shared per-CPU dynticks state. */
 338         int dynticks_snap;              /* Per-GP tracking for dynticks. */
 339
 340         /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
 341         unsigned long dynticks_fqs;     /* Kicked due to dynticks idle. */
 342         unsigned long offline_fqs;      /* Kicked due to being offline. */
 343         unsigned long cond_resched_completed;
 344                                         /* Grace period that needs help */
 345                                         /*  from cond_resched(). */
 346
 347         /* 5) __rcu_pending() statistics. */
 348         unsigned long n_rcu_pending;    /* rcu_pending() calls since boot. */
 349         unsigned long n_rp_qs_pending;
 350         unsigned long n_rp_report_qs;
 351         unsigned long n_rp_cb_ready;
 352         unsigned long n_rp_cpu_needs_gp;
 353         unsigned long n_rp_gp_completed;
 354         unsigned long n_rp_gp_started;
 355         unsigned long n_rp_nocb_defer_wakeup;
 356         unsigned long n_rp_need_nothing;
 357
 358         /* 6) _rcu_barrier() and OOM callbacks. */
 359         struct rcu_head barrier_head;
 360 #ifdef CONFIG_RCU_FAST_NO_HZ
 361         struct rcu_head oom_head;
 362 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
 363
 364         /* 7) Callback offloading. */
 365 #ifdef CONFIG_RCU_NOCB_CPU
 366         struct rcu_head *nocb_head;     /* CBs waiting for kthread. */
 367         struct rcu_head **nocb_tail;
 368         atomic_long_t nocb_q_count;     /* # CBs waiting for nocb */
 369         atomic_long_t nocb_q_count_lazy; /*  invocation (all stages). */
 370         struct rcu_head *nocb_follower_head; /* CBs ready to invoke. */
 371         struct rcu_head **nocb_follower_tail;
 372         wait_queue_head_t nocb_wq;      /* For nocb kthreads to sleep on. */
 373         struct task_struct *nocb_kthread;
 374         int nocb_defer_wakeup;          /* Defer wakeup of nocb_kthread. */
 375
 376         /* The following fields are used by the leader, hence own cacheline. */
 377         struct rcu_head *nocb_gp_head ____cacheline_internodealigned_in_smp;
 378                                         /* CBs waiting for GP. */
 379         struct rcu_head **nocb_gp_tail;
 380         bool nocb_leader_sleep;         /* Is the nocb leader thread asleep? */
 381         struct rcu_data *nocb_next_follower;
 382                                         /* Next follower in wakeup chain. */
 383
 384         /* The following fields are used by the follower, hence new cachline. */
 385         struct rcu_data *nocb_leader ____cacheline_internodealigned_in_smp;
 386                                         /* Leader CPU takes GP-end wakeups. */
 387 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
 388
 389         /* 8) RCU CPU stall data. */
 390 #ifdef CONFIG_RCU_CPU_STALL_INFO
 391         unsigned int softirq_snap;      /* Snapshot of softirq activity. */
 392 #endif /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
 393
 394         int cpu;
 395         struct rcu_state *rsp;
 396 };
 397
 398 /* Values for fqs_state field in struct rcu_state. */
 399 #define RCU_GP_IDLE             0       /* No grace period in progress. */
 400 #define RCU_GP_INIT             1       /* Grace period being initialized. */
 401 #define RCU_SAVE_DYNTICK        2       /* Need to scan dyntick state. */
 402 #define RCU_FORCE_QS            3       /* Need to force quiescent state. */
 403 #define RCU_SIGNAL_INIT         RCU_SAVE_DYNTICK
 404
 405 /* Values for nocb_defer_wakeup field in struct rcu_data. */
 406 #define RCU_NOGP_WAKE_NOT       0
 407 #define RCU_NOGP_WAKE           1
 408 #define RCU_NOGP_WAKE_FORCE     2
 409
 410 #define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500))
 411                                         /* For jiffies_till_first_fqs and */
 412                                         /*  and jiffies_till_next_fqs. */
 413
 414 #define RCU_JIFFIES_FQS_DIV     256     /* Very large systems need more */
 415                                         /*  delay between bouts of */
 416                                         /*  quiescent-state forcing. */
 417
 418 #define RCU_STALL_RAT_DELAY     2       /* Allow other CPUs time to take */
 419                                         /*  at least one scheduling clock */
 420                                         /*  irq before ratting on them. */
 421
 422 #define rcu_wait(cond)                                                  \
 423 do {                                                                    \
 424         for (;;) {                                                      \
 425                 set_current_state(TASK_INTERRUPTIBLE);                  \
 426                 if (cond)                                               \
 427                         break;                                          \
 428                 schedule();                                             \
 429         }                                                               \
 430         __set_current_state(TASK_RUNNING);                              \
 431 } while (0)
 432
 433 /*
 434  * RCU global state, including node hierarchy.  This hierarchy is
 435  * represented in "heap" form in a dense array.  The root (first level)
 436  * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
 437  * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
 438  * and the third level in ->node[m+1] and following (->node[m+1] referenced
 439  * by ->level[2]).  The number of levels is determined by the number of
 440  * CPUs and by CONFIG_RCU_FANOUT.  Small systems will have a "hierarchy"
 441  * consisting of a single rcu_node.
 442  */
 443 struct rcu_state {
 444         struct rcu_node node[NUM_RCU_NODES];    /* Hierarchy. */
 445         struct rcu_node *level[RCU_NUM_LVLS];   /* Hierarchy levels. */
 446         u32 levelcnt[MAX_RCU_LVLS + 1];         /* # nodes in each level. */
 447         u8 levelspread[RCU_NUM_LVLS];           /* kids/node in each level. */
 448         u8 flavor_mask;                         /* bit in flavor mask. */
 449         struct rcu_data __percpu *rda;          /* pointer of percu rcu_data. */
 450         void (*call)(struct rcu_head *head,     /* call_rcu() flavor. */
 451                      void (*func)(struct rcu_head *head));
 452
 453         /* The following fields are guarded by the root rcu_node's lock. */
 454
 455         u8      fqs_state ____cacheline_internodealigned_in_smp;
 456                                                 /* Force QS state. */
 457         u8      boost;                          /* Subject to priority boost. */
 458         unsigned long gpnum;                    /* Current gp number. */
 459         unsigned long completed;                /* # of last completed gp. */
 460         struct task_struct *gp_kthread;         /* Task for grace periods. */
 461         wait_queue_head_t gp_wq;                /* Where GP task waits. */
 462         short gp_flags;                         /* Commands for GP task. */
 463         short gp_state;                         /* GP kthread sleep state. */
 464
 465         /* End of fields guarded by root rcu_node's lock. */
 466
 467         raw_spinlock_t orphan_lock ____cacheline_internodealigned_in_smp;
 468                                                 /* Protect following fields. */
 469         struct rcu_head *orphan_nxtlist;        /* Orphaned callbacks that */
 470                                                 /*  need a grace period. */
 471         struct rcu_head **orphan_nxttail;       /* Tail of above. */
 472         struct rcu_head *orphan_donelist;       /* Orphaned callbacks that */
 473                                                 /*  are ready to invoke. */
 474         struct rcu_head **orphan_donetail;      /* Tail of above. */
 475         long qlen_lazy;                         /* Number of lazy callbacks. */
 476         long qlen;                              /* Total number of callbacks. */
 477         /* End of fields guarded by orphan_lock. */
 478
 479         struct mutex barrier_mutex;             /* Guards barrier fields. */
 480         atomic_t barrier_cpu_count;             /* # CPUs waiting on. */
 481         struct completion barrier_completion;   /* Wake at barrier end. */
 482         unsigned long n_barrier_done;           /* ++ at start and end of */
 483                                                 /*  _rcu_barrier(). */
 484         /* End of fields guarded by barrier_mutex. */
 485
 486         atomic_long_t expedited_start;          /* Starting ticket. */
 487         atomic_long_t expedited_done;           /* Done ticket. */
 488         atomic_long_t expedited_wrap;           /* # near-wrap incidents. */
 489         atomic_long_t expedited_tryfail;        /* # acquisition failures. */
 490         atomic_long_t expedited_workdone1;      /* # done by others #1. */
 491         atomic_long_t expedited_workdone2;      /* # done by others #2. */
 492         atomic_long_t expedited_normal;         /* # fallbacks to normal. */
 493         atomic_long_t expedited_stoppedcpus;    /* # successful stop_cpus. */
 494         atomic_long_t expedited_done_tries;     /* # tries to update _done. */
 495         atomic_long_t expedited_done_lost;      /* # times beaten to _done. */
 496         atomic_long_t expedited_done_exit;      /* # times exited _done loop. */
 497
 498         unsigned long jiffies_force_qs;         /* Time at which to invoke */
 499                                                 /*  force_quiescent_state(). */
 500         unsigned long n_force_qs;               /* Number of calls to */
 501                                                 /*  force_quiescent_state(). */
 502         unsigned long n_force_qs_lh;            /* ~Number of calls leaving */
 503                                                 /*  due to lock unavailable. */
 504         unsigned long n_force_qs_ngp;           /* Number of calls leaving */
 505                                                 /*  due to no GP active. */
 506         unsigned long gp_start;                 /* Time at which GP started, */
 507                                                 /*  but in jiffies. */
 508         unsigned long gp_activity;              /* Time of last GP kthread */
 509                                                 /*  activity in jiffies. */
 510         unsigned long jiffies_stall;            /* Time at which to check */
 511                                                 /*  for CPU stalls. */
 512         unsigned long jiffies_resched;          /* Time at which to resched */
 513                                                 /*  a reluctant CPU. */
 514         unsigned long n_force_qs_gpstart;       /* Snapshot of n_force_qs at */
 515                                                 /*  GP start. */
 516         unsigned long gp_max;                   /* Maximum GP duration in */
 517                                                 /*  jiffies. */
 518         const char *name;                       /* Name of structure. */
 519         char abbr;                              /* Abbreviated name. */
 520         struct list_head flavors;               /* List of RCU flavors. */
 521 };
 522
 523 /* Values for rcu_state structure's gp_flags field. */
 524 #define RCU_GP_FLAG_INIT 0x1    /* Need grace-period initialization. */
 525 #define RCU_GP_FLAG_FQS  0x2    /* Need grace-period quiescent-state forcing. */
 526
 527 /* Values for rcu_state structure's gp_flags field. */
 528 #define RCU_GP_WAIT_INIT 0      /* Initial state. */
 529 #define RCU_GP_WAIT_GPS  1      /* Wait for grace-period start. */
 530 #define RCU_GP_WAIT_FQS  2      /* Wait for force-quiescent-state time. */
 531
 532 extern struct list_head rcu_struct_flavors;
 533
 534 /* Sequence through rcu_state structures for each RCU flavor. */
 535 #define for_each_rcu_flavor(rsp) \
 536         list_for_each_entry((rsp), &rcu_struct_flavors, flavors)
 537
 538 /*
 539  * RCU implementation internal declarations:
 540  */
 541 extern struct rcu_state rcu_sched_state;
 542
 543 extern struct rcu_state rcu_bh_state;
 544
 545 #ifdef CONFIG_PREEMPT_RCU
 546 extern struct rcu_state rcu_preempt_state;
 547 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 548
 549 #ifdef CONFIG_RCU_BOOST
 550 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
 551 DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu);
 552 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
 553 DECLARE_PER_CPU(char, rcu_cpu_has_work);
 554 #endif /* #ifdef CONFIG_RCU_BOOST */
 555
 556 #ifndef RCU_TREE_NONCORE
 557
 558 /* Forward declarations for rcutree_plugin.h */
 559 static void rcu_bootup_announce(void);
 560 static void rcu_preempt_note_context_switch(void);
 561 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
 562 #ifdef CONFIG_HOTPLUG_CPU
 563 static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
 564 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
 565 static void rcu_print_detail_task_stall(struct rcu_state *rsp);
 566 static int rcu_print_task_stall(struct rcu_node *rnp);
 567 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
 568 static void rcu_preempt_check_callbacks(void);
 569 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
 570 static void __init __rcu_init_preempt(void);
 571 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
 572 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
 573 static void invoke_rcu_callbacks_kthread(void);
 574 static bool rcu_is_callbacks_kthread(void);
 575 #ifdef CONFIG_RCU_BOOST
 576 static void rcu_preempt_do_callbacks(void);
 577 static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
 578                                                  struct rcu_node *rnp);
 579 #endif /* #ifdef CONFIG_RCU_BOOST */
 580 static void __init rcu_spawn_boost_kthreads(void);
 581 static void rcu_prepare_kthreads(int cpu);
 582 static void rcu_cleanup_after_idle(void);
 583 static void rcu_prepare_for_idle(void);
 584 static void rcu_idle_count_callbacks_posted(void);
 585 static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
 586 static void print_cpu_stall_info_begin(void);
 587 static void print_cpu_stall_info(struct rcu_state *rsp, int cpu);
 588 static void print_cpu_stall_info_end(void);
 589 static void zero_cpu_stall_ticks(struct rcu_data *rdp);
 590 static void increment_cpu_stall_ticks(void);
 591 static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu);
 592 static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq);
 593 static void rcu_nocb_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp);
 594 static void rcu_init_one_nocb(struct rcu_node *rnp);
 595 static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
 596                             bool lazy, unsigned long flags);
 597 static bool rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp,
 598                                       struct rcu_data *rdp,
 599                                       unsigned long flags);
 600 static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp);
 601 static void do_nocb_deferred_wakeup(struct rcu_data *rdp);
 602 static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp);
 603 static void rcu_spawn_all_nocb_kthreads(int cpu);
 604 static void __init rcu_spawn_nocb_kthreads(void);
 605 #ifdef CONFIG_RCU_NOCB_CPU
 606 static void __init rcu_organize_nocb_kthreads(struct rcu_state *rsp);
 607 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
 608 static void __maybe_unused rcu_kick_nohz_cpu(int cpu);
 609 static bool init_nocb_callback_list(struct rcu_data *rdp);
 610 static void rcu_sysidle_enter(int irq);
 611 static void rcu_sysidle_exit(int irq);
 612 static void rcu_sysidle_check_cpu(struct rcu_data *rdp, bool *isidle,
 613                                   unsigned long *maxj);
 614 static bool is_sysidle_rcu_state(struct rcu_state *rsp);
 615 static void rcu_sysidle_report_gp(struct rcu_state *rsp, int isidle,
 616                                   unsigned long maxj);
 617 static void rcu_bind_gp_kthread(void);
 618 static void rcu_sysidle_init_percpu_data(struct rcu_dynticks *rdtp);
 619 static bool rcu_nohz_full_cpu(struct rcu_state *rsp);
 620 static void rcu_dynticks_task_enter(void);
 621 static void rcu_dynticks_task_exit(void);
 622
 623 #endif /* #ifndef RCU_TREE_NONCORE */
 624
 625 #ifdef CONFIG_RCU_TRACE
 626 /* Read out queue lengths for tracing. */
 627 static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll)
 628 {
 629 #ifdef CONFIG_RCU_NOCB_CPU
 630         *ql = atomic_long_read(&rdp->nocb_q_count);
 631         *qll = atomic_long_read(&rdp->nocb_q_count_lazy);
 632 #else /* #ifdef CONFIG_RCU_NOCB_CPU */
 633         *ql = 0;
 634         *qll = 0;
 635 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
 636 }
 637 #endif /* #ifdef CONFIG_RCU_TRACE */