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 #include <linux/swait.h>
  31 #include <linux/stop_machine.h>
  32
  33 /*
  34  * Define shape of hierarchy based on NR_CPUS, CONFIG_RCU_FANOUT, and
  35  * CONFIG_RCU_FANOUT_LEAF.
  36  * In theory, it should be possible to add more levels straightforwardly.
  37  * In practice, this did work well going from three levels to four.
  38  * Of course, your mileage may vary.
  39  */
  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_NODES       NUM_RCU_LVL_0
  70 #  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0 }
  71 #  define RCU_NODE_NAME_INIT  { "rcu_node_0" }
  72 #  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_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 #elif NR_CPUS <= RCU_FANOUT_3
  82 #  define RCU_NUM_LVLS        3
  83 #  define NUM_RCU_LVL_0       1
  84 #  define NUM_RCU_LVL_1       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
  85 #  define NUM_RCU_LVL_2       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
  86 #  define NUM_RCU_NODES       (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2)
  87 #  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0, NUM_RCU_LVL_1, NUM_RCU_LVL_2 }
  88 #  define RCU_NODE_NAME_INIT  { "rcu_node_0", "rcu_node_1", "rcu_node_2" }
  89 #  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0", "rcu_node_fqs_1", "rcu_node_fqs_2" }
  90 #elif NR_CPUS <= RCU_FANOUT_4
  91 #  define RCU_NUM_LVLS        4
  92 #  define NUM_RCU_LVL_0       1
  93 #  define NUM_RCU_LVL_1       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3)
  94 #  define NUM_RCU_LVL_2       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
  95 #  define NUM_RCU_LVL_3       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
  96 #  define NUM_RCU_NODES       (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3)
  97 #  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0, NUM_RCU_LVL_1, NUM_RCU_LVL_2, NUM_RCU_LVL_3 }
  98 #  define RCU_NODE_NAME_INIT  { "rcu_node_0", "rcu_node_1", "rcu_node_2", "rcu_node_3" }
  99 #  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0", "rcu_node_fqs_1", "rcu_node_fqs_2", "rcu_node_fqs_3" }
 100 #else
 101 # error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
 102 #endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
 103
 104 extern int rcu_num_lvls;
 105 extern int rcu_num_nodes;
 106
 107 /*
 108  * Dynticks per-CPU state.
 109  */
 110 struct rcu_dynticks {
 111         long long dynticks_nesting; /* Track irq/process nesting level. */
 112                                     /* Process level is worth LLONG_MAX/2. */
 113         int dynticks_nmi_nesting;   /* Track NMI nesting level. */
 114         atomic_t dynticks;          /* Even value for idle, else odd. */
 115 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
 116         long long dynticks_idle_nesting;
 117                                     /* irq/process nesting level from idle. */
 118         atomic_t dynticks_idle;     /* Even value for idle, else odd. */
 119                                     /*  "Idle" excludes userspace execution. */
 120         unsigned long dynticks_idle_jiffies;
 121                                     /* End of last non-NMI non-idle period. */
 122 #endif /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
 123 #ifdef CONFIG_RCU_FAST_NO_HZ
 124         bool all_lazy;              /* Are all CPU's CBs lazy? */
 125         unsigned long nonlazy_posted;
 126                                     /* # times non-lazy CBs posted to CPU. */
 127         unsigned long nonlazy_posted_snap;
 128                                     /* idle-period nonlazy_posted snapshot. */
 129         unsigned long last_accelerate;
 130                                     /* Last jiffy CBs were accelerated. */
 131         unsigned long last_advance_all;
 132                                     /* Last jiffy CBs were all advanced. */
 133         int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */
 134 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
 135 };
 136
 137 /* RCU's kthread states for tracing. */
 138 #define RCU_KTHREAD_STOPPED  0
 139 #define RCU_KTHREAD_RUNNING  1
 140 #define RCU_KTHREAD_WAITING  2
 141 #define RCU_KTHREAD_OFFCPU   3
 142 #define RCU_KTHREAD_YIELDING 4
 143 #define RCU_KTHREAD_MAX      4
 144
 145 /*
 146  * Definition for node within the RCU grace-period-detection hierarchy.
 147  */
 148 struct rcu_node {
 149         raw_spinlock_t __private lock;  /* Root rcu_node's lock protects */
 150                                         /*  some rcu_state fields as well as */
 151                                         /*  following. */
 152         unsigned long gpnum;    /* Current grace period for this node. */
 153                                 /*  This will either be equal to or one */
 154                                 /*  behind the root rcu_node's gpnum. */
 155         unsigned long completed; /* Last GP completed for this node. */
 156                                 /*  This will either be equal to or one */
 157                                 /*  behind the root rcu_node's gpnum. */
 158         unsigned long qsmask;   /* CPUs or groups that need to switch in */
 159                                 /*  order for current grace period to proceed.*/
 160                                 /*  In leaf rcu_node, each bit corresponds to */
 161                                 /*  an rcu_data structure, otherwise, each */
 162                                 /*  bit corresponds to a child rcu_node */
 163                                 /*  structure. */
 164         unsigned long qsmaskinit;
 165                                 /* Per-GP initial value for qsmask. */
 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 expmask;  /* CPUs or groups that need to check in */
 171                                 /*  to allow the current expedited GP */
 172                                 /*  to complete. */
 173         unsigned long expmaskinit;
 174                                 /* Per-GP initial values for expmask. */
 175                                 /*  Initialized from ->expmaskinitnext at the */
 176                                 /*  beginning of each expedited GP. */
 177         unsigned long expmaskinitnext;
 178                                 /* Online CPUs for next expedited GP. */
 179                                 /*  Any CPU that has ever been online will */
 180                                 /*  have its bit set. */
 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         struct swait_queue_head 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         spinlock_t exp_lock ____cacheline_internodealigned_in_smp;
 252         unsigned long exp_seq_rq;
 253         wait_queue_head_t exp_wq[4];
 254 } ____cacheline_internodealigned_in_smp;
 255
 256 /*
 257  * Do a full breadth-first scan of the rcu_node structures for the
 258  * specified rcu_state structure.
 259  */
 260 #define rcu_for_each_node_breadth_first(rsp, rnp) \
 261         for ((rnp) = &(rsp)->node[0]; \
 262              (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
 263
 264 /*
 265  * Do a breadth-first scan of the non-leaf rcu_node structures for the
 266  * specified rcu_state structure.  Note that if there is a singleton
 267  * rcu_node tree with but one rcu_node structure, this loop is a no-op.
 268  */
 269 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
 270         for ((rnp) = &(rsp)->node[0]; \
 271              (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
 272
 273 /*
 274  * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
 275  * structure.  Note that if there is a singleton rcu_node tree with but
 276  * one rcu_node structure, this loop -will- visit the rcu_node structure.
 277  * It is still a leaf node, even if it is also the root node.
 278  */
 279 #define rcu_for_each_leaf_node(rsp, rnp) \
 280         for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
 281              (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
 282
 283 /*
 284  * Union to allow "aggregate OR" operation on the need for a quiescent
 285  * state by the normal and expedited grace periods.
 286  */
 287 union rcu_noqs {
 288         struct {
 289                 u8 norm;
 290                 u8 exp;
 291         } b; /* Bits. */
 292         u16 s; /* Set of bits, aggregate OR here. */
 293 };
 294
 295 /* Index values for nxttail array in struct rcu_data. */
 296 #define RCU_DONE_TAIL           0       /* Also RCU_WAIT head. */
 297 #define RCU_WAIT_TAIL           1       /* Also RCU_NEXT_READY head. */
 298 #define RCU_NEXT_READY_TAIL     2       /* Also RCU_NEXT head. */
 299 #define RCU_NEXT_TAIL           3
 300 #define RCU_NEXT_SIZE           4
 301
 302 /* Per-CPU data for read-copy update. */
 303 struct rcu_data {
 304         /* 1) quiescent-state and grace-period handling : */
 305         unsigned long   completed;      /* Track rsp->completed gp number */
 306                                         /*  in order to detect GP end. */
 307         unsigned long   gpnum;          /* Highest gp number that this CPU */
 308                                         /*  is aware of having started. */
 309         unsigned long   rcu_qs_ctr_snap;/* Snapshot of rcu_qs_ctr to check */
 310                                         /*  for rcu_all_qs() invocations. */
 311         union rcu_noqs  cpu_no_qs;      /* No QSes yet for this CPU. */
 312         bool            core_needs_qs;  /* Core waits for quiesc state. */
 313         bool            beenonline;     /* CPU online at least once. */
 314         bool            gpwrap;         /* Possible gpnum/completed wrap. */
 315         struct rcu_node *mynode;        /* This CPU's leaf of hierarchy */
 316         unsigned long grpmask;          /* Mask to apply to leaf qsmask. */
 317         unsigned long   ticks_this_gp;  /* The number of scheduling-clock */
 318                                         /*  ticks this CPU has handled */
 319                                         /*  during and after the last grace */
 320                                         /* period it is aware of. */
 321
 322         /* 2) batch handling */
 323         /*
 324          * If nxtlist is not NULL, it is partitioned as follows.
 325          * Any of the partitions might be empty, in which case the
 326          * pointer to that partition will be equal to the pointer for
 327          * the following partition.  When the list is empty, all of
 328          * the nxttail elements point to the ->nxtlist pointer itself,
 329          * which in that case is NULL.
 330          *
 331          * [nxtlist, *nxttail[RCU_DONE_TAIL]):
 332          *      Entries that batch # <= ->completed
 333          *      The grace period for these entries has completed, and
 334          *      the other grace-period-completed entries may be moved
 335          *      here temporarily in rcu_process_callbacks().
 336          * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
 337          *      Entries that batch # <= ->completed - 1: waiting for current GP
 338          * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
 339          *      Entries known to have arrived before current GP ended
 340          * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
 341          *      Entries that might have arrived after current GP ended
 342          *      Note that the value of *nxttail[RCU_NEXT_TAIL] will
 343          *      always be NULL, as this is the end of the list.
 344          */
 345         struct rcu_head *nxtlist;
 346         struct rcu_head **nxttail[RCU_NEXT_SIZE];
 347         unsigned long   nxtcompleted[RCU_NEXT_SIZE];
 348                                         /* grace periods for sublists. */
 349         long            qlen_lazy;      /* # of lazy queued callbacks */
 350         long            qlen;           /* # of queued callbacks, incl lazy */
 351         long            qlen_last_fqs_check;
 352                                         /* qlen at last check for QS forcing */
 353         unsigned long   n_cbs_invoked;  /* count of RCU cbs invoked. */
 354         unsigned long   n_nocbs_invoked; /* count of no-CBs RCU cbs invoked. */
 355         unsigned long   n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
 356         unsigned long   n_cbs_adopted;  /* RCU cbs adopted from dying CPU */
 357         unsigned long   n_force_qs_snap;
 358                                         /* did other CPU force QS recently? */
 359         long            blimit;         /* Upper limit on a processed batch */
 360
 361         /* 3) dynticks interface. */
 362         struct rcu_dynticks *dynticks;  /* Shared per-CPU dynticks state. */
 363         int dynticks_snap;              /* Per-GP tracking for dynticks. */
 364
 365         /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
 366         unsigned long dynticks_fqs;     /* Kicked due to dynticks idle. */
 367         unsigned long offline_fqs;      /* Kicked due to being offline. */
 368         unsigned long cond_resched_completed;
 369                                         /* Grace period that needs help */
 370                                         /*  from cond_resched(). */
 371
 372         /* 5) __rcu_pending() statistics. */
 373         unsigned long n_rcu_pending;    /* rcu_pending() calls since boot. */
 374         unsigned long n_rp_core_needs_qs;
 375         unsigned long n_rp_report_qs;
 376         unsigned long n_rp_cb_ready;
 377         unsigned long n_rp_cpu_needs_gp;
 378         unsigned long n_rp_gp_completed;
 379         unsigned long n_rp_gp_started;
 380         unsigned long n_rp_nocb_defer_wakeup;
 381         unsigned long n_rp_need_nothing;
 382
 383         /* 6) _rcu_barrier(), OOM callbacks, and expediting. */
 384         struct rcu_head barrier_head;
 385 #ifdef CONFIG_RCU_FAST_NO_HZ
 386         struct rcu_head oom_head;
 387 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
 388         atomic_long_t exp_workdone1;    /* # done by others #1. */
 389         atomic_long_t exp_workdone2;    /* # done by others #2. */
 390         atomic_long_t exp_workdone3;    /* # done by others #3. */
 391
 392         /* 7) Callback offloading. */
 393 #ifdef CONFIG_RCU_NOCB_CPU
 394         struct rcu_head *nocb_head;     /* CBs waiting for kthread. */
 395         struct rcu_head **nocb_tail;
 396         atomic_long_t nocb_q_count;     /* # CBs waiting for nocb */
 397         atomic_long_t nocb_q_count_lazy; /*  invocation (all stages). */
 398         struct rcu_head *nocb_follower_head; /* CBs ready to invoke. */
 399         struct rcu_head **nocb_follower_tail;
 400         struct swait_queue_head nocb_wq; /* For nocb kthreads to sleep on. */
 401         struct task_struct *nocb_kthread;
 402         int nocb_defer_wakeup;          /* Defer wakeup of nocb_kthread. */
 403
 404         /* The following fields are used by the leader, hence own cacheline. */
 405         struct rcu_head *nocb_gp_head ____cacheline_internodealigned_in_smp;
 406                                         /* CBs waiting for GP. */
 407         struct rcu_head **nocb_gp_tail;
 408         bool nocb_leader_sleep;         /* Is the nocb leader thread asleep? */
 409         struct rcu_data *nocb_next_follower;
 410                                         /* Next follower in wakeup chain. */
 411
 412         /* The following fields are used by the follower, hence new cachline. */
 413         struct rcu_data *nocb_leader ____cacheline_internodealigned_in_smp;
 414                                         /* Leader CPU takes GP-end wakeups. */
 415 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
 416
 417         /* 8) RCU CPU stall data. */
 418         unsigned int softirq_snap;      /* Snapshot of softirq activity. */
 419
 420         int cpu;
 421         struct rcu_state *rsp;
 422 };
 423
 424 /* Values for nocb_defer_wakeup field in struct rcu_data. */
 425 #define RCU_NOGP_WAKE_NOT       0
 426 #define RCU_NOGP_WAKE           1
 427 #define RCU_NOGP_WAKE_FORCE     2
 428
 429 #define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500))
 430                                         /* For jiffies_till_first_fqs and */
 431                                         /*  and jiffies_till_next_fqs. */
 432
 433 #define RCU_JIFFIES_FQS_DIV     256     /* Very large systems need more */
 434                                         /*  delay between bouts of */
 435                                         /*  quiescent-state forcing. */
 436
 437 #define RCU_STALL_RAT_DELAY     2       /* Allow other CPUs time to take */
 438                                         /*  at least one scheduling clock */
 439                                         /*  irq before ratting on them. */
 440
 441 #define rcu_wait(cond)                                                  \
 442 do {                                                                    \
 443         for (;;) {                                                      \
 444                 set_current_state(TASK_INTERRUPTIBLE);                  \
 445                 if (cond)                                               \
 446                         break;                                          \
 447                 schedule();                                             \
 448         }                                                               \
 449         __set_current_state(TASK_RUNNING);                              \
 450 } while (0)
 451
 452 /*
 453  * RCU global state, including node hierarchy.  This hierarchy is
 454  * represented in "heap" form in a dense array.  The root (first level)
 455  * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
 456  * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
 457  * and the third level in ->node[m+1] and following (->node[m+1] referenced
 458  * by ->level[2]).  The number of levels is determined by the number of
 459  * CPUs and by CONFIG_RCU_FANOUT.  Small systems will have a "hierarchy"
 460  * consisting of a single rcu_node.
 461  */
 462 struct rcu_state {
 463         struct rcu_node node[NUM_RCU_NODES];    /* Hierarchy. */
 464         struct rcu_node *level[RCU_NUM_LVLS + 1];
 465                                                 /* Hierarchy levels (+1 to */
 466                                                 /*  shut bogus gcc warning) */
 467         u8 flavor_mask;                         /* bit in flavor mask. */
 468         struct rcu_data __percpu *rda;          /* pointer of percu rcu_data. */
 469         call_rcu_func_t call;                   /* call_rcu() flavor. */
 470         int ncpus;                              /* # CPUs seen so far. */
 471
 472         /* The following fields are guarded by the root rcu_node's lock. */
 473
 474         u8      boost ____cacheline_internodealigned_in_smp;
 475                                                 /* Subject to priority boost. */
 476         unsigned long gpnum;                    /* Current gp number. */
 477         unsigned long completed;                /* # of last completed gp. */
 478         struct task_struct *gp_kthread;         /* Task for grace periods. */
 479         struct swait_queue_head gp_wq;          /* Where GP task waits. */
 480         short gp_flags;                         /* Commands for GP task. */
 481         short gp_state;                         /* GP kthread sleep state. */
 482
 483         /* End of fields guarded by root rcu_node's lock. */
 484
 485         raw_spinlock_t orphan_lock ____cacheline_internodealigned_in_smp;
 486                                                 /* Protect following fields. */
 487         struct rcu_head *orphan_nxtlist;        /* Orphaned callbacks that */
 488                                                 /*  need a grace period. */
 489         struct rcu_head **orphan_nxttail;       /* Tail of above. */
 490         struct rcu_head *orphan_donelist;       /* Orphaned callbacks that */
 491                                                 /*  are ready to invoke. */
 492         struct rcu_head **orphan_donetail;      /* Tail of above. */
 493         long qlen_lazy;                         /* Number of lazy callbacks. */
 494         long qlen;                              /* Total number of callbacks. */
 495         /* End of fields guarded by orphan_lock. */
 496
 497         struct mutex barrier_mutex;             /* Guards barrier fields. */
 498         atomic_t barrier_cpu_count;             /* # CPUs waiting on. */
 499         struct completion barrier_completion;   /* Wake at barrier end. */
 500         unsigned long barrier_sequence;         /* ++ at start and end of */
 501                                                 /*  _rcu_barrier(). */
 502         /* End of fields guarded by barrier_mutex. */
 503
 504         struct mutex exp_mutex;                 /* Serialize expedited GP. */
 505         struct mutex exp_wake_mutex;            /* Serialize wakeup. */
 506         unsigned long expedited_sequence;       /* Take a ticket. */
 507         atomic_long_t expedited_normal;         /* # fallbacks to normal. */
 508         atomic_t expedited_need_qs;             /* # CPUs left to check in. */
 509         struct swait_queue_head expedited_wq;   /* Wait for check-ins. */
 510         int ncpus_snap;                         /* # CPUs seen last time. */
 511
 512         unsigned long jiffies_force_qs;         /* Time at which to invoke */
 513                                                 /*  force_quiescent_state(). */
 514         unsigned long jiffies_kick_kthreads;    /* Time at which to kick */
 515                                                 /*  kthreads, if configured. */
 516         unsigned long n_force_qs;               /* Number of calls to */
 517                                                 /*  force_quiescent_state(). */
 518         unsigned long n_force_qs_lh;            /* ~Number of calls leaving */
 519                                                 /*  due to lock unavailable. */
 520         unsigned long n_force_qs_ngp;           /* Number of calls leaving */
 521                                                 /*  due to no GP active. */
 522         unsigned long gp_start;                 /* Time at which GP started, */
 523                                                 /*  but in jiffies. */
 524         unsigned long gp_activity;              /* Time of last GP kthread */
 525                                                 /*  activity in jiffies. */
 526         unsigned long jiffies_stall;            /* Time at which to check */
 527                                                 /*  for CPU stalls. */
 528         unsigned long jiffies_resched;          /* Time at which to resched */
 529                                                 /*  a reluctant CPU. */
 530         unsigned long n_force_qs_gpstart;       /* Snapshot of n_force_qs at */
 531                                                 /*  GP start. */
 532         unsigned long gp_max;                   /* Maximum GP duration in */
 533                                                 /*  jiffies. */
 534         const char *name;                       /* Name of structure. */
 535         char abbr;                              /* Abbreviated name. */
 536         struct list_head flavors;               /* List of RCU flavors. */
 537 };
 538
 539 /* Values for rcu_state structure's gp_flags field. */
 540 #define RCU_GP_FLAG_INIT 0x1    /* Need grace-period initialization. */
 541 #define RCU_GP_FLAG_FQS  0x2    /* Need grace-period quiescent-state forcing. */
 542
 543 /* Values for rcu_state structure's gp_state field. */
 544 #define RCU_GP_IDLE      0      /* Initial state and no GP in progress. */
 545 #define RCU_GP_WAIT_GPS  1      /* Wait for grace-period start. */
 546 #define RCU_GP_DONE_GPS  2      /* Wait done for grace-period start. */
 547 #define RCU_GP_WAIT_FQS  3      /* Wait for force-quiescent-state time. */
 548 #define RCU_GP_DOING_FQS 4      /* Wait done for force-quiescent-state time. */
 549 #define RCU_GP_CLEANUP   5      /* Grace-period cleanup started. */
 550 #define RCU_GP_CLEANED   6      /* Grace-period cleanup complete. */
 551
 552 #ifndef RCU_TREE_NONCORE
 553 static const char * const gp_state_names[] = {
 554         "RCU_GP_IDLE",
 555         "RCU_GP_WAIT_GPS",
 556         "RCU_GP_DONE_GPS",
 557         "RCU_GP_WAIT_FQS",
 558         "RCU_GP_DOING_FQS",
 559         "RCU_GP_CLEANUP",
 560         "RCU_GP_CLEANED",
 561 };
 562 #endif /* #ifndef RCU_TREE_NONCORE */
 563
 564 extern struct list_head rcu_struct_flavors;
 565
 566 /* Sequence through rcu_state structures for each RCU flavor. */
 567 #define for_each_rcu_flavor(rsp) \
 568         list_for_each_entry((rsp), &rcu_struct_flavors, flavors)
 569
 570 /*
 571  * RCU implementation internal declarations:
 572  */
 573 extern struct rcu_state rcu_sched_state;
 574
 575 extern struct rcu_state rcu_bh_state;
 576
 577 #ifdef CONFIG_PREEMPT_RCU
 578 extern struct rcu_state rcu_preempt_state;
 579 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 580
 581 #ifdef CONFIG_RCU_BOOST
 582 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
 583 DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu);
 584 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
 585 DECLARE_PER_CPU(char, rcu_cpu_has_work);
 586 #endif /* #ifdef CONFIG_RCU_BOOST */
 587
 588 #ifndef RCU_TREE_NONCORE
 589
 590 /* Forward declarations for rcutree_plugin.h */
 591 static void rcu_bootup_announce(void);
 592 static void rcu_preempt_note_context_switch(void);
 593 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
 594 #ifdef CONFIG_HOTPLUG_CPU
 595 static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
 596 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
 597 static void rcu_print_detail_task_stall(struct rcu_state *rsp);
 598 static int rcu_print_task_stall(struct rcu_node *rnp);
 599 static int rcu_print_task_exp_stall(struct rcu_node *rnp);
 600 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
 601 static void rcu_preempt_check_callbacks(void);
 602 void call_rcu(struct rcu_head *head, rcu_callback_t func);
 603 static void __init __rcu_init_preempt(void);
 604 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
 605 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
 606 static void invoke_rcu_callbacks_kthread(void);
 607 static bool rcu_is_callbacks_kthread(void);
 608 #ifdef CONFIG_RCU_BOOST
 609 static void rcu_preempt_do_callbacks(void);
 610 static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
 611                                                  struct rcu_node *rnp);
 612 #endif /* #ifdef CONFIG_RCU_BOOST */
 613 static void __init rcu_spawn_boost_kthreads(void);
 614 static void rcu_prepare_kthreads(int cpu);
 615 static void rcu_cleanup_after_idle(void);
 616 static void rcu_prepare_for_idle(void);
 617 static void rcu_idle_count_callbacks_posted(void);
 618 static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
 619 static void print_cpu_stall_info_begin(void);
 620 static void print_cpu_stall_info(struct rcu_state *rsp, int cpu);
 621 static void print_cpu_stall_info_end(void);
 622 static void zero_cpu_stall_ticks(struct rcu_data *rdp);
 623 static void increment_cpu_stall_ticks(void);
 624 static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu);
 625 static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq);
 626 static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp);
 627 static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq);
 628 static void rcu_init_one_nocb(struct rcu_node *rnp);
 629 static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
 630                             bool lazy, unsigned long flags);
 631 static bool rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp,
 632                                       struct rcu_data *rdp,
 633                                       unsigned long flags);
 634 static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp);
 635 static void do_nocb_deferred_wakeup(struct rcu_data *rdp);
 636 static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp);
 637 static void rcu_spawn_all_nocb_kthreads(int cpu);
 638 static void __init rcu_spawn_nocb_kthreads(void);
 639 #ifdef CONFIG_RCU_NOCB_CPU
 640 static void __init rcu_organize_nocb_kthreads(struct rcu_state *rsp);
 641 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
 642 static void __maybe_unused rcu_kick_nohz_cpu(int cpu);
 643 static bool init_nocb_callback_list(struct rcu_data *rdp);
 644 static void rcu_sysidle_enter(int irq);
 645 static void rcu_sysidle_exit(int irq);
 646 static void rcu_sysidle_check_cpu(struct rcu_data *rdp, bool *isidle,
 647                                   unsigned long *maxj);
 648 static bool is_sysidle_rcu_state(struct rcu_state *rsp);
 649 static void rcu_sysidle_report_gp(struct rcu_state *rsp, int isidle,
 650                                   unsigned long maxj);
 651 static void rcu_bind_gp_kthread(void);
 652 static void rcu_sysidle_init_percpu_data(struct rcu_dynticks *rdtp);
 653 static bool rcu_nohz_full_cpu(struct rcu_state *rsp);
 654 static void rcu_dynticks_task_enter(void);
 655 static void rcu_dynticks_task_exit(void);
 656
 657 #endif /* #ifndef RCU_TREE_NONCORE */
 658
 659 #ifdef CONFIG_RCU_TRACE
 660 /* Read out queue lengths for tracing. */
 661 static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll)
 662 {
 663 #ifdef CONFIG_RCU_NOCB_CPU
 664         *ql = atomic_long_read(&rdp->nocb_q_count);
 665         *qll = atomic_long_read(&rdp->nocb_q_count_lazy);
 666 #else /* #ifdef CONFIG_RCU_NOCB_CPU */
 667         *ql = 0;
 668         *qll = 0;
 669 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
 670 }
 671 #endif /* #ifdef CONFIG_RCU_TRACE */
 672
 673 /*
 674  * Place this after a lock-acquisition primitive to guarantee that
 675  * an UNLOCK+LOCK pair act as a full barrier.  This guarantee applies
 676  * if the UNLOCK and LOCK are executed by the same CPU or if the
 677  * UNLOCK and LOCK operate on the same lock variable.
 678  */
 679 #ifdef CONFIG_PPC
 680 #define smp_mb__after_unlock_lock()     smp_mb()  /* Full ordering for lock. */
 681 #else /* #ifdef CONFIG_PPC */
 682 #define smp_mb__after_unlock_lock()     do { } while (0)
 683 #endif /* #else #ifdef CONFIG_PPC */
 684
 685 /*
 686  * Wrappers for the rcu_node::lock acquire and release.
 687  *
 688  * Because the rcu_nodes form a tree, the tree traversal locking will observe
 689  * different lock values, this in turn means that an UNLOCK of one level
 690  * followed by a LOCK of another level does not imply a full memory barrier;
 691  * and most importantly transitivity is lost.
 692  *
 693  * In order to restore full ordering between tree levels, augment the regular
 694  * lock acquire functions with smp_mb__after_unlock_lock().
 695  *
 696  * As ->lock of struct rcu_node is a __private field, therefore one should use
 697  * these wrappers rather than directly call raw_spin_{lock,unlock}* on ->lock.
 698  */
 699 static inline void raw_spin_lock_rcu_node(struct rcu_node *rnp)
 700 {
 701         raw_spin_lock(&ACCESS_PRIVATE(rnp, lock));
 702         smp_mb__after_unlock_lock();
 703 }
 704
 705 static inline void raw_spin_unlock_rcu_node(struct rcu_node *rnp)
 706 {
 707         raw_spin_unlock(&ACCESS_PRIVATE(rnp, lock));
 708 }
 709
 710 static inline void raw_spin_lock_irq_rcu_node(struct rcu_node *rnp)
 711 {
 712         raw_spin_lock_irq(&ACCESS_PRIVATE(rnp, lock));
 713         smp_mb__after_unlock_lock();
 714 }
 715
 716 static inline void raw_spin_unlock_irq_rcu_node(struct rcu_node *rnp)
 717 {
 718         raw_spin_unlock_irq(&ACCESS_PRIVATE(rnp, lock));
 719 }
 720
 721 #define raw_spin_lock_irqsave_rcu_node(rnp, flags)                      \
 722 do {                                                                    \
 723         typecheck(unsigned long, flags);                                \
 724         raw_spin_lock_irqsave(&ACCESS_PRIVATE(rnp, lock), flags);       \
 725         smp_mb__after_unlock_lock();                                    \
 726 } while (0)
 727
 728 #define raw_spin_unlock_irqrestore_rcu_node(rnp, flags)                 \
 729 do {                                                                    \
 730         typecheck(unsigned long, flags);                                \
 731         raw_spin_unlock_irqrestore(&ACCESS_PRIVATE(rnp, lock), flags);  \
 732 } while (0)
 733
 734 static inline bool raw_spin_trylock_rcu_node(struct rcu_node *rnp)
 735 {
 736         bool locked = raw_spin_trylock(&ACCESS_PRIVATE(rnp, lock));
 737
 738         if (locked)
 739                 smp_mb__after_unlock_lock();
 740         return locked;
 741 }