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 #ifdef CONFIG_RCU_FANOUT
  40 #define RCU_FANOUT CONFIG_RCU_FANOUT
  41 #else /* #ifdef CONFIG_RCU_FANOUT */
  42 # ifdef CONFIG_64BIT
  43 # define RCU_FANOUT 64
  44 # else
  45 # define RCU_FANOUT 32
  46 # endif
  47 #endif /* #else #ifdef CONFIG_RCU_FANOUT */
  48
  49 #ifdef CONFIG_RCU_FANOUT_LEAF
  50 #define RCU_FANOUT_LEAF CONFIG_RCU_FANOUT_LEAF
  51 #else /* #ifdef CONFIG_RCU_FANOUT_LEAF */
  52 # ifdef CONFIG_64BIT
  53 # define RCU_FANOUT_LEAF 64
  54 # else
  55 # define RCU_FANOUT_LEAF 32
  56 # endif
  57 #endif /* #else #ifdef CONFIG_RCU_FANOUT_LEAF */
  58
  59 #define RCU_FANOUT_1          (RCU_FANOUT_LEAF)
  60 #define RCU_FANOUT_2          (RCU_FANOUT_1 * RCU_FANOUT)
  61 #define RCU_FANOUT_3          (RCU_FANOUT_2 * RCU_FANOUT)
  62 #define RCU_FANOUT_4          (RCU_FANOUT_3 * RCU_FANOUT)
  63
  64 #if NR_CPUS <= RCU_FANOUT_1
  65 #  define RCU_NUM_LVLS        1
  66 #  define NUM_RCU_LVL_0       1
  67 #  define NUM_RCU_NODES       NUM_RCU_LVL_0
  68 #  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0 }
  69 #  define RCU_NODE_NAME_INIT  { "rcu_node_0" }
  70 #  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0" }
  71 #elif NR_CPUS <= RCU_FANOUT_2
  72 #  define RCU_NUM_LVLS        2
  73 #  define NUM_RCU_LVL_0       1
  74 #  define NUM_RCU_LVL_1       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
  75 #  define NUM_RCU_NODES       (NUM_RCU_LVL_0 + NUM_RCU_LVL_1)
  76 #  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0, NUM_RCU_LVL_1 }
  77 #  define RCU_NODE_NAME_INIT  { "rcu_node_0", "rcu_node_1" }
  78 #  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0", "rcu_node_fqs_1" }
  79 #elif NR_CPUS <= RCU_FANOUT_3
  80 #  define RCU_NUM_LVLS        3
  81 #  define NUM_RCU_LVL_0       1
  82 #  define NUM_RCU_LVL_1       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
  83 #  define NUM_RCU_LVL_2       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
  84 #  define NUM_RCU_NODES       (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2)
  85 #  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0, NUM_RCU_LVL_1, NUM_RCU_LVL_2 }
  86 #  define RCU_NODE_NAME_INIT  { "rcu_node_0", "rcu_node_1", "rcu_node_2" }
  87 #  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0", "rcu_node_fqs_1", "rcu_node_fqs_2" }
  88 #elif NR_CPUS <= RCU_FANOUT_4
  89 #  define RCU_NUM_LVLS        4
  90 #  define NUM_RCU_LVL_0       1
  91 #  define NUM_RCU_LVL_1       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3)
  92 #  define NUM_RCU_LVL_2       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
  93 #  define NUM_RCU_LVL_3       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
  94 #  define NUM_RCU_NODES       (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3)
  95 #  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0, NUM_RCU_LVL_1, NUM_RCU_LVL_2, NUM_RCU_LVL_3 }
  96 #  define RCU_NODE_NAME_INIT  { "rcu_node_0", "rcu_node_1", "rcu_node_2", "rcu_node_3" }
  97 #  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0", "rcu_node_fqs_1", "rcu_node_fqs_2", "rcu_node_fqs_3" }
  98 #else
  99 # error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
 100 #endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
 101
 102 extern int rcu_num_lvls;
 103 extern int rcu_num_nodes;
 104
 105 /*
 106  * Dynticks per-CPU state.
 107  */
 108 struct rcu_dynticks {
 109         long long dynticks_nesting; /* Track irq/process nesting level. */
 110                                     /* Process level is worth LLONG_MAX/2. */
 111         int dynticks_nmi_nesting;   /* Track NMI nesting level. */
 112         atomic_t dynticks;          /* Even value for idle, else odd. */
 113 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
 114         long long dynticks_idle_nesting;
 115                                     /* irq/process nesting level from idle. */
 116         atomic_t dynticks_idle;     /* Even value for idle, else odd. */
 117                                     /*  "Idle" excludes userspace execution. */
 118         unsigned long dynticks_idle_jiffies;
 119                                     /* End of last non-NMI non-idle period. */
 120 #endif /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
 121 #ifdef CONFIG_RCU_FAST_NO_HZ
 122         bool all_lazy;              /* Are all CPU's CBs lazy? */
 123         unsigned long nonlazy_posted;
 124                                     /* # times non-lazy CBs posted to CPU. */
 125         unsigned long nonlazy_posted_snap;
 126                                     /* idle-period nonlazy_posted snapshot. */
 127         unsigned long last_accelerate;
 128                                     /* Last jiffy CBs were accelerated. */
 129         unsigned long last_advance_all;
 130                                     /* Last jiffy CBs were all advanced. */
 131         int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */
 132 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
 133 };
 134
 135 /* RCU's kthread states for tracing. */
 136 #define RCU_KTHREAD_STOPPED  0
 137 #define RCU_KTHREAD_RUNNING  1
 138 #define RCU_KTHREAD_WAITING  2
 139 #define RCU_KTHREAD_OFFCPU   3
 140 #define RCU_KTHREAD_YIELDING 4
 141 #define RCU_KTHREAD_MAX      4
 142
 143 /*
 144  * Definition for node within the RCU grace-period-detection hierarchy.
 145  */
 146 struct rcu_node {
 147         raw_spinlock_t lock;    /* Root rcu_node's lock protects some */
 148                                 /*  rcu_state fields as well as following. */
 149         unsigned long gpnum;    /* Current grace period for this node. */
 150                                 /*  This will either be equal to or one */
 151                                 /*  behind the root rcu_node's gpnum. */
 152         unsigned long completed; /* Last GP completed for this node. */
 153                                 /*  This will either be equal to or one */
 154                                 /*  behind the root rcu_node's gpnum. */
 155         unsigned long qsmask;   /* CPUs or groups that need to switch in */
 156                                 /*  order for current grace period to proceed.*/
 157                                 /*  In leaf rcu_node, each bit corresponds to */
 158                                 /*  an rcu_data structure, otherwise, each */
 159                                 /*  bit corresponds to a child rcu_node */
 160                                 /*  structure. */
 161         unsigned long expmask;  /* Groups that have ->blkd_tasks */
 162                                 /*  elements that need to drain to allow the */
 163                                 /*  current expedited grace period to */
 164                                 /*  complete (only for PREEMPT_RCU). */
 165         unsigned long qsmaskinit;
 166                                 /* Per-GP initial value for qsmask & expmask. */
 167                                 /*  Initialized from ->qsmaskinitnext at the */
 168                                 /*  beginning of each grace period. */
 169         unsigned long qsmaskinitnext;
 170                                 /* Online CPUs for next grace period. */
 171         unsigned long grpmask;  /* Mask to apply to parent qsmask. */
 172                                 /*  Only one bit will be set in this mask. */
 173         int     grplo;          /* lowest-numbered CPU or group here. */
 174         int     grphi;          /* highest-numbered CPU or group here. */
 175         u8      grpnum;         /* CPU/group number for next level up. */
 176         u8      level;          /* root is at level 0. */
 177         bool    wait_blkd_tasks;/* Necessary to wait for blocked tasks to */
 178                                 /*  exit RCU read-side critical sections */
 179                                 /*  before propagating offline up the */
 180                                 /*  rcu_node tree? */
 181         struct rcu_node *parent;
 182         struct list_head blkd_tasks;
 183                                 /* Tasks blocked in RCU read-side critical */
 184                                 /*  section.  Tasks are placed at the head */
 185                                 /*  of this list and age towards the tail. */
 186         struct list_head *gp_tasks;
 187                                 /* Pointer to the first task blocking the */
 188                                 /*  current grace period, or NULL if there */
 189                                 /*  is no such task. */
 190         struct list_head *exp_tasks;
 191                                 /* Pointer to the first task blocking the */
 192                                 /*  current expedited grace period, or NULL */
 193                                 /*  if there is no such task.  If there */
 194                                 /*  is no current expedited grace period, */
 195                                 /*  then there can cannot be any such task. */
 196         struct list_head *boost_tasks;
 197                                 /* Pointer to first task that needs to be */
 198                                 /*  priority boosted, or NULL if no priority */
 199                                 /*  boosting is needed for this rcu_node */
 200                                 /*  structure.  If there are no tasks */
 201                                 /*  queued on this rcu_node structure that */
 202                                 /*  are blocking the current grace period, */
 203                                 /*  there can be no such task. */
 204         struct rt_mutex boost_mtx;
 205                                 /* Used only for the priority-boosting */
 206                                 /*  side effect, not as a lock. */
 207         unsigned long boost_time;
 208                                 /* When to start boosting (jiffies). */
 209         struct task_struct *boost_kthread_task;
 210                                 /* kthread that takes care of priority */
 211                                 /*  boosting for this rcu_node structure. */
 212         unsigned int boost_kthread_status;
 213                                 /* State of boost_kthread_task for tracing. */
 214         unsigned long n_tasks_boosted;
 215                                 /* Total number of tasks boosted. */
 216         unsigned long n_exp_boosts;
 217                                 /* Number of tasks boosted for expedited GP. */
 218         unsigned long n_normal_boosts;
 219                                 /* Number of tasks boosted for normal GP. */
 220         unsigned long n_balk_blkd_tasks;
 221                                 /* Refused to boost: no blocked tasks. */
 222         unsigned long n_balk_exp_gp_tasks;
 223                                 /* Refused to boost: nothing blocking GP. */
 224         unsigned long n_balk_boost_tasks;
 225                                 /* Refused to boost: already boosting. */
 226         unsigned long n_balk_notblocked;
 227                                 /* Refused to boost: RCU RS CS still running. */
 228         unsigned long n_balk_notyet;
 229                                 /* Refused to boost: not yet time. */
 230         unsigned long n_balk_nos;
 231                                 /* Refused to boost: not sure why, though. */
 232                                 /*  This can happen due to race conditions. */
 233 #ifdef CONFIG_RCU_NOCB_CPU
 234         wait_queue_head_t nocb_gp_wq[2];
 235                                 /* Place for rcu_nocb_kthread() to wait GP. */
 236 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
 237         int need_future_gp[2];
 238                                 /* Counts of upcoming no-CB GP requests. */
 239         raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp;
 240 } ____cacheline_internodealigned_in_smp;
 241
 242 /*
 243  * Do a full breadth-first scan of the rcu_node structures for the
 244  * specified rcu_state structure.
 245  */
 246 #define rcu_for_each_node_breadth_first(rsp, rnp) \
 247         for ((rnp) = &(rsp)->node[0]; \
 248              (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
 249
 250 /*
 251  * Do a breadth-first scan of the non-leaf rcu_node structures for the
 252  * specified rcu_state structure.  Note that if there is a singleton
 253  * rcu_node tree with but one rcu_node structure, this loop is a no-op.
 254  */
 255 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
 256         for ((rnp) = &(rsp)->node[0]; \
 257              (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
 258
 259 /*
 260  * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
 261  * structure.  Note that if there is a singleton rcu_node tree with but
 262  * one rcu_node structure, this loop -will- visit the rcu_node structure.
 263  * It is still a leaf node, even if it is also the root node.
 264  */
 265 #define rcu_for_each_leaf_node(rsp, rnp) \
 266         for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
 267              (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
 268
 269 /* Index values for nxttail array in struct rcu_data. */
 270 #define RCU_DONE_TAIL           0       /* Also RCU_WAIT head. */
 271 #define RCU_WAIT_TAIL           1       /* Also RCU_NEXT_READY head. */
 272 #define RCU_NEXT_READY_TAIL     2       /* Also RCU_NEXT head. */
 273 #define RCU_NEXT_TAIL           3
 274 #define RCU_NEXT_SIZE           4
 275
 276 /* Per-CPU data for read-copy update. */
 277 struct rcu_data {
 278         /* 1) quiescent-state and grace-period handling : */
 279         unsigned long   completed;      /* Track rsp->completed gp number */
 280                                         /*  in order to detect GP end. */
 281         unsigned long   gpnum;          /* Highest gp number that this CPU */
 282                                         /*  is aware of having started. */
 283         unsigned long   rcu_qs_ctr_snap;/* Snapshot of rcu_qs_ctr to check */
 284                                         /*  for rcu_all_qs() invocations. */
 285         bool            passed_quiesce; /* User-mode/idle loop etc. */
 286         bool            qs_pending;     /* Core waits for quiesc state. */
 287         bool            beenonline;     /* CPU online at least once. */
 288         bool            gpwrap;         /* Possible gpnum/completed wrap. */
 289         struct rcu_node *mynode;        /* This CPU's leaf of hierarchy */
 290         unsigned long grpmask;          /* Mask to apply to leaf qsmask. */
 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
 296         /* 2) batch handling */
 297         /*
 298          * If nxtlist is not NULL, it is partitioned as follows.
 299          * Any of the partitions might be empty, in which case the
 300          * pointer to that partition will be equal to the pointer for
 301          * the following partition.  When the list is empty, all of
 302          * the nxttail elements point to the ->nxtlist pointer itself,
 303          * which in that case is NULL.
 304          *
 305          * [nxtlist, *nxttail[RCU_DONE_TAIL]):
 306          *      Entries that batch # <= ->completed
 307          *      The grace period for these entries has completed, and
 308          *      the other grace-period-completed entries may be moved
 309          *      here temporarily in rcu_process_callbacks().
 310          * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
 311          *      Entries that batch # <= ->completed - 1: waiting for current GP
 312          * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
 313          *      Entries known to have arrived before current GP ended
 314          * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
 315          *      Entries that might have arrived after current GP ended
 316          *      Note that the value of *nxttail[RCU_NEXT_TAIL] will
 317          *      always be NULL, as this is the end of the list.
 318          */
 319         struct rcu_head *nxtlist;
 320         struct rcu_head **nxttail[RCU_NEXT_SIZE];
 321         unsigned long   nxtcompleted[RCU_NEXT_SIZE];
 322                                         /* grace periods for sublists. */
 323         long            qlen_lazy;      /* # of lazy queued callbacks */
 324         long            qlen;           /* # of queued callbacks, incl lazy */
 325         long            qlen_last_fqs_check;
 326                                         /* qlen at last check for QS forcing */
 327         unsigned long   n_cbs_invoked;  /* count of RCU cbs invoked. */
 328         unsigned long   n_nocbs_invoked; /* count of no-CBs RCU cbs invoked. */
 329         unsigned long   n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
 330         unsigned long   n_cbs_adopted;  /* RCU cbs adopted from dying CPU */
 331         unsigned long   n_force_qs_snap;
 332                                         /* did other CPU force QS recently? */
 333         long            blimit;         /* Upper limit on a processed batch */
 334
 335         /* 3) dynticks interface. */
 336         struct rcu_dynticks *dynticks;  /* Shared per-CPU dynticks state. */
 337         int dynticks_snap;              /* Per-GP tracking for dynticks. */
 338
 339         /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
 340         unsigned long dynticks_fqs;     /* Kicked due to dynticks idle. */
 341         unsigned long offline_fqs;      /* Kicked due to being offline. */
 342         unsigned long cond_resched_completed;
 343                                         /* Grace period that needs help */
 344                                         /*  from cond_resched(). */
 345
 346         /* 5) __rcu_pending() statistics. */
 347         unsigned long n_rcu_pending;    /* rcu_pending() calls since boot. */
 348         unsigned long n_rp_qs_pending;
 349         unsigned long n_rp_report_qs;
 350         unsigned long n_rp_cb_ready;
 351         unsigned long n_rp_cpu_needs_gp;
 352         unsigned long n_rp_gp_completed;
 353         unsigned long n_rp_gp_started;
 354         unsigned long n_rp_nocb_defer_wakeup;
 355         unsigned long n_rp_need_nothing;
 356
 357         /* 6) _rcu_barrier() and OOM callbacks. */
 358         struct rcu_head barrier_head;
 359 #ifdef CONFIG_RCU_FAST_NO_HZ
 360         struct rcu_head oom_head;
 361 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
 362
 363         /* 7) Callback offloading. */
 364 #ifdef CONFIG_RCU_NOCB_CPU
 365         struct rcu_head *nocb_head;     /* CBs waiting for kthread. */
 366         struct rcu_head **nocb_tail;
 367         atomic_long_t nocb_q_count;     /* # CBs waiting for nocb */
 368         atomic_long_t nocb_q_count_lazy; /*  invocation (all stages). */
 369         struct rcu_head *nocb_follower_head; /* CBs ready to invoke. */
 370         struct rcu_head **nocb_follower_tail;
 371         wait_queue_head_t nocb_wq;      /* For nocb kthreads to sleep on. */
 372         struct task_struct *nocb_kthread;
 373         int nocb_defer_wakeup;          /* Defer wakeup of nocb_kthread. */
 374
 375         /* The following fields are used by the leader, hence own cacheline. */
 376         struct rcu_head *nocb_gp_head ____cacheline_internodealigned_in_smp;
 377                                         /* CBs waiting for GP. */
 378         struct rcu_head **nocb_gp_tail;
 379         bool nocb_leader_sleep;         /* Is the nocb leader thread asleep? */
 380         struct rcu_data *nocb_next_follower;
 381                                         /* Next follower in wakeup chain. */
 382
 383         /* The following fields are used by the follower, hence new cachline. */
 384         struct rcu_data *nocb_leader ____cacheline_internodealigned_in_smp;
 385                                         /* Leader CPU takes GP-end wakeups. */
 386 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
 387
 388         /* 8) RCU CPU stall data. */
 389         unsigned int softirq_snap;      /* Snapshot of softirq activity. */
 390
 391         int cpu;
 392         struct rcu_state *rsp;
 393 };
 394
 395 /* Values for fqs_state field in struct rcu_state. */
 396 #define RCU_GP_IDLE             0       /* No grace period in progress. */
 397 #define RCU_GP_INIT             1       /* Grace period being initialized. */
 398 #define RCU_SAVE_DYNTICK        2       /* Need to scan dyntick state. */
 399 #define RCU_FORCE_QS            3       /* Need to force quiescent state. */
 400 #define RCU_SIGNAL_INIT         RCU_SAVE_DYNTICK
 401
 402 /* Values for nocb_defer_wakeup field in struct rcu_data. */
 403 #define RCU_NOGP_WAKE_NOT       0
 404 #define RCU_NOGP_WAKE           1
 405 #define RCU_NOGP_WAKE_FORCE     2
 406
 407 #define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500))
 408                                         /* For jiffies_till_first_fqs and */
 409                                         /*  and jiffies_till_next_fqs. */
 410
 411 #define RCU_JIFFIES_FQS_DIV     256     /* Very large systems need more */
 412                                         /*  delay between bouts of */
 413                                         /*  quiescent-state forcing. */
 414
 415 #define RCU_STALL_RAT_DELAY     2       /* Allow other CPUs time to take */
 416                                         /*  at least one scheduling clock */
 417                                         /*  irq before ratting on them. */
 418
 419 #define rcu_wait(cond)                                                  \
 420 do {                                                                    \
 421         for (;;) {                                                      \
 422                 set_current_state(TASK_INTERRUPTIBLE);                  \
 423                 if (cond)                                               \
 424                         break;                                          \
 425                 schedule();                                             \
 426         }                                                               \
 427         __set_current_state(TASK_RUNNING);                              \
 428 } while (0)
 429
 430 /*
 431  * RCU global state, including node hierarchy.  This hierarchy is
 432  * represented in "heap" form in a dense array.  The root (first level)
 433  * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
 434  * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
 435  * and the third level in ->node[m+1] and following (->node[m+1] referenced
 436  * by ->level[2]).  The number of levels is determined by the number of
 437  * CPUs and by CONFIG_RCU_FANOUT.  Small systems will have a "hierarchy"
 438  * consisting of a single rcu_node.
 439  */
 440 struct rcu_state {
 441         struct rcu_node node[NUM_RCU_NODES];    /* Hierarchy. */
 442         struct rcu_node *level[RCU_NUM_LVLS + 1];
 443                                                 /* Hierarchy levels (+1 to */
 444                                                 /*  shut bogus gcc warning) */
 445         u8 flavor_mask;                         /* bit in flavor mask. */
 446         struct rcu_data __percpu *rda;          /* pointer of percu rcu_data. */
 447         void (*call)(struct rcu_head *head,     /* call_rcu() flavor. */
 448                      void (*func)(struct rcu_head *head));
 449
 450         /* The following fields are guarded by the root rcu_node's lock. */
 451
 452         u8      fqs_state ____cacheline_internodealigned_in_smp;
 453                                                 /* Force QS state. */
 454         u8      boost;                          /* Subject to priority boost. */
 455         unsigned long gpnum;                    /* Current gp number. */
 456         unsigned long completed;                /* # of last completed gp. */
 457         struct task_struct *gp_kthread;         /* Task for grace periods. */
 458         wait_queue_head_t gp_wq;                /* Where GP task waits. */
 459         short gp_flags;                         /* Commands for GP task. */
 460         short gp_state;                         /* GP kthread sleep state. */
 461
 462         /* End of fields guarded by root rcu_node's lock. */
 463
 464         raw_spinlock_t orphan_lock ____cacheline_internodealigned_in_smp;
 465                                                 /* Protect following fields. */
 466         struct rcu_head *orphan_nxtlist;        /* Orphaned callbacks that */
 467                                                 /*  need a grace period. */
 468         struct rcu_head **orphan_nxttail;       /* Tail of above. */
 469         struct rcu_head *orphan_donelist;       /* Orphaned callbacks that */
 470                                                 /*  are ready to invoke. */
 471         struct rcu_head **orphan_donetail;      /* Tail of above. */
 472         long qlen_lazy;                         /* Number of lazy callbacks. */
 473         long qlen;                              /* Total number of callbacks. */
 474         /* End of fields guarded by orphan_lock. */
 475
 476         struct mutex barrier_mutex;             /* Guards barrier fields. */
 477         atomic_t barrier_cpu_count;             /* # CPUs waiting on. */
 478         struct completion barrier_completion;   /* Wake at barrier end. */
 479         unsigned long n_barrier_done;           /* ++ at start and end of */
 480                                                 /*  _rcu_barrier(). */
 481         /* End of fields guarded by barrier_mutex. */
 482
 483         atomic_long_t expedited_start;          /* Starting ticket. */
 484         atomic_long_t expedited_done;           /* Done ticket. */
 485         atomic_long_t expedited_wrap;           /* # near-wrap incidents. */
 486         atomic_long_t expedited_tryfail;        /* # acquisition failures. */
 487         atomic_long_t expedited_workdone1;      /* # done by others #1. */
 488         atomic_long_t expedited_workdone2;      /* # done by others #2. */
 489         atomic_long_t expedited_normal;         /* # fallbacks to normal. */
 490         atomic_long_t expedited_stoppedcpus;    /* # successful stop_cpus. */
 491         atomic_long_t expedited_done_tries;     /* # tries to update _done. */
 492         atomic_long_t expedited_done_lost;      /* # times beaten to _done. */
 493         atomic_long_t expedited_done_exit;      /* # times exited _done loop. */
 494
 495         unsigned long jiffies_force_qs;         /* Time at which to invoke */
 496                                                 /*  force_quiescent_state(). */
 497         unsigned long n_force_qs;               /* Number of calls to */
 498                                                 /*  force_quiescent_state(). */
 499         unsigned long n_force_qs_lh;            /* ~Number of calls leaving */
 500                                                 /*  due to lock unavailable. */
 501         unsigned long n_force_qs_ngp;           /* Number of calls leaving */
 502                                                 /*  due to no GP active. */
 503         unsigned long gp_start;                 /* Time at which GP started, */
 504                                                 /*  but in jiffies. */
 505         unsigned long gp_activity;              /* Time of last GP kthread */
 506                                                 /*  activity in jiffies. */
 507         unsigned long jiffies_stall;            /* Time at which to check */
 508                                                 /*  for CPU stalls. */
 509         unsigned long jiffies_resched;          /* Time at which to resched */
 510                                                 /*  a reluctant CPU. */
 511         unsigned long n_force_qs_gpstart;       /* Snapshot of n_force_qs at */
 512                                                 /*  GP start. */
 513         unsigned long gp_max;                   /* Maximum GP duration in */
 514                                                 /*  jiffies. */
 515         const char *name;                       /* Name of structure. */
 516         char abbr;                              /* Abbreviated name. */
 517         struct list_head flavors;               /* List of RCU flavors. */
 518 };
 519
 520 /* Values for rcu_state structure's gp_flags field. */
 521 #define RCU_GP_FLAG_INIT 0x1    /* Need grace-period initialization. */
 522 #define RCU_GP_FLAG_FQS  0x2    /* Need grace-period quiescent-state forcing. */
 523
 524 /* Values for rcu_state structure's gp_flags field. */
 525 #define RCU_GP_WAIT_INIT 0      /* Initial state. */
 526 #define RCU_GP_WAIT_GPS  1      /* Wait for grace-period start. */
 527 #define RCU_GP_DONE_GPS  2      /* Wait done for grace-period start. */
 528 #define RCU_GP_WAIT_FQS  3      /* Wait for force-quiescent-state time. */
 529 #define RCU_GP_DONE_FQS  4      /* Wait done for force-quiescent-state time. */
 530 #define RCU_GP_CLEANUP   5      /* Grace-period cleanup started. */
 531 #define RCU_GP_CLEANED   6      /* Grace-period cleanup complete. */
 532
 533 extern struct list_head rcu_struct_flavors;
 534
 535 /* Sequence through rcu_state structures for each RCU flavor. */
 536 #define for_each_rcu_flavor(rsp) \
 537         list_for_each_entry((rsp), &rcu_struct_flavors, flavors)
 538
 539 /*
 540  * RCU implementation internal declarations:
 541  */
 542 extern struct rcu_state rcu_sched_state;
 543
 544 extern struct rcu_state rcu_bh_state;
 545
 546 #ifdef CONFIG_PREEMPT_RCU
 547 extern struct rcu_state rcu_preempt_state;
 548 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 549
 550 #ifdef CONFIG_RCU_BOOST
 551 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
 552 DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu);
 553 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
 554 DECLARE_PER_CPU(char, rcu_cpu_has_work);
 555 #endif /* #ifdef CONFIG_RCU_BOOST */
 556
 557 #ifndef RCU_TREE_NONCORE
 558
 559 /* Forward declarations for rcutree_plugin.h */
 560 static void rcu_bootup_announce(void);
 561 static void rcu_preempt_note_context_switch(void);
 562 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
 563 #ifdef CONFIG_HOTPLUG_CPU
 564 static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
 565 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
 566 static void rcu_print_detail_task_stall(struct rcu_state *rsp);
 567 static int rcu_print_task_stall(struct rcu_node *rnp);
 568 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
 569 static void rcu_preempt_check_callbacks(void);
 570 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
 571 static void __init __rcu_init_preempt(void);
 572 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
 573 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
 574 static void invoke_rcu_callbacks_kthread(void);
 575 static bool rcu_is_callbacks_kthread(void);
 576 #ifdef CONFIG_RCU_BOOST
 577 static void rcu_preempt_do_callbacks(void);
 578 static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
 579                                                  struct rcu_node *rnp);
 580 #endif /* #ifdef CONFIG_RCU_BOOST */
 581 static void __init rcu_spawn_boost_kthreads(void);
 582 static void rcu_prepare_kthreads(int cpu);
 583 static void rcu_cleanup_after_idle(void);
 584 static void rcu_prepare_for_idle(void);
 585 static void rcu_idle_count_callbacks_posted(void);
 586 static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
 587 static void print_cpu_stall_info_begin(void);
 588 static void print_cpu_stall_info(struct rcu_state *rsp, int cpu);
 589 static void print_cpu_stall_info_end(void);
 590 static void zero_cpu_stall_ticks(struct rcu_data *rdp);
 591 static void increment_cpu_stall_ticks(void);
 592 static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu);
 593 static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq);
 594 static void rcu_nocb_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp);
 595 static void rcu_init_one_nocb(struct rcu_node *rnp);
 596 static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
 597                             bool lazy, unsigned long flags);
 598 static bool rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp,
 599                                       struct rcu_data *rdp,
 600                                       unsigned long flags);
 601 static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp);
 602 static void do_nocb_deferred_wakeup(struct rcu_data *rdp);
 603 static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp);
 604 static void rcu_spawn_all_nocb_kthreads(int cpu);
 605 static void __init rcu_spawn_nocb_kthreads(void);
 606 #ifdef CONFIG_RCU_NOCB_CPU
 607 static void __init rcu_organize_nocb_kthreads(struct rcu_state *rsp);
 608 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
 609 static void __maybe_unused rcu_kick_nohz_cpu(int cpu);
 610 static bool init_nocb_callback_list(struct rcu_data *rdp);
 611 static void rcu_sysidle_enter(int irq);
 612 static void rcu_sysidle_exit(int irq);
 613 static void rcu_sysidle_check_cpu(struct rcu_data *rdp, bool *isidle,
 614                                   unsigned long *maxj);
 615 static bool is_sysidle_rcu_state(struct rcu_state *rsp);
 616 static void rcu_sysidle_report_gp(struct rcu_state *rsp, int isidle,
 617                                   unsigned long maxj);
 618 static void rcu_bind_gp_kthread(void);
 619 static void rcu_sysidle_init_percpu_data(struct rcu_dynticks *rdtp);
 620 static bool rcu_nohz_full_cpu(struct rcu_state *rsp);
 621 static void rcu_dynticks_task_enter(void);
 622 static void rcu_dynticks_task_exit(void);
 623
 624 #endif /* #ifndef RCU_TREE_NONCORE */
 625
 626 #ifdef CONFIG_RCU_TRACE
 627 /* Read out queue lengths for tracing. */
 628 static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll)
 629 {
 630 #ifdef CONFIG_RCU_NOCB_CPU
 631         *ql = atomic_long_read(&rdp->nocb_q_count);
 632         *qll = atomic_long_read(&rdp->nocb_q_count_lazy);
 633 #else /* #ifdef CONFIG_RCU_NOCB_CPU */
 634         *ql = 0;
 635         *qll = 0;
 636 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
 637 }
 638 #endif /* #ifdef CONFIG_RCU_TRACE */