LXR linux/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#  define RCU_EXP_NAME_INIT   { "rcu_node_exp_0" }
  74#elif NR_CPUS <= RCU_FANOUT_2
  75#  define RCU_NUM_LVLS        2
  76#  define NUM_RCU_LVL_0       1
  77#  define NUM_RCU_LVL_1       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
  78#  define NUM_RCU_NODES       (NUM_RCU_LVL_0 + NUM_RCU_LVL_1)
  79#  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0, NUM_RCU_LVL_1 }
  80#  define RCU_NODE_NAME_INIT  { "rcu_node_0", "rcu_node_1" }
  81#  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0", "rcu_node_fqs_1" }
  82#  define RCU_EXP_NAME_INIT   { "rcu_node_exp_0", "rcu_node_exp_1" }
  83#elif NR_CPUS <= RCU_FANOUT_3
  84#  define RCU_NUM_LVLS        3
  85#  define NUM_RCU_LVL_0       1
  86#  define NUM_RCU_LVL_1       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
  87#  define NUM_RCU_LVL_2       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
  88#  define NUM_RCU_NODES       (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2)
  89#  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0, NUM_RCU_LVL_1, NUM_RCU_LVL_2 }
  90#  define RCU_NODE_NAME_INIT  { "rcu_node_0", "rcu_node_1", "rcu_node_2" }
  91#  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0", "rcu_node_fqs_1", "rcu_node_fqs_2" }
  92#  define RCU_EXP_NAME_INIT   { "rcu_node_exp_0", "rcu_node_exp_1", "rcu_node_exp_2" }
  93#elif NR_CPUS <= RCU_FANOUT_4
  94#  define RCU_NUM_LVLS        4
  95#  define NUM_RCU_LVL_0       1
  96#  define NUM_RCU_LVL_1       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3)
  97#  define NUM_RCU_LVL_2       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
  98#  define NUM_RCU_LVL_3       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
  99#  define NUM_RCU_NODES       (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3)
 100#  define NUM_RCU_LVL_INIT    { NUM_RCU_LVL_0, NUM_RCU_LVL_1, NUM_RCU_LVL_2, NUM_RCU_LVL_3 }
 101#  define RCU_NODE_NAME_INIT  { "rcu_node_0", "rcu_node_1", "rcu_node_2", "rcu_node_3" }
 102#  define RCU_FQS_NAME_INIT   { "rcu_node_fqs_0", "rcu_node_fqs_1", "rcu_node_fqs_2", "rcu_node_fqs_3" }
 103#  define RCU_EXP_NAME_INIT   { "rcu_node_exp_0", "rcu_node_exp_1", "rcu_node_exp_2", "rcu_node_exp_3" }
 104#else
 105# error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
 106#endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
 107
 108extern int rcu_num_lvls;
 109extern int rcu_num_nodes;
 110
 111/*
 112 * Dynticks per-CPU state.
 113 */
 114struct rcu_dynticks {
 115        long long dynticks_nesting; /* Track irq/process nesting level. */
 116                                    /* Process level is worth LLONG_MAX/2. */
 117        int dynticks_nmi_nesting;   /* Track NMI nesting level. */
 118        atomic_t dynticks;          /* Even value for idle, else odd. */
 119#ifdef CONFIG_NO_HZ_FULL_SYSIDLE
 120        long long dynticks_idle_nesting;
 121                                    /* irq/process nesting level from idle. */
 122        atomic_t dynticks_idle;     /* Even value for idle, else odd. */
 123                                    /*  "Idle" excludes userspace execution. */
 124        unsigned long dynticks_idle_jiffies;
 125                                    /* End of last non-NMI non-idle period. */
 126#endif /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
 127#ifdef CONFIG_RCU_FAST_NO_HZ
 128        bool all_lazy;              /* Are all CPU's CBs lazy? */
 129        unsigned long nonlazy_posted;
 130                                    /* # times non-lazy CBs posted to CPU. */
 131        unsigned long nonlazy_posted_snap;
 132                                    /* idle-period nonlazy_posted snapshot. */
 133        unsigned long last_accelerate;
 134                                    /* Last jiffy CBs were accelerated. */
 135        unsigned long last_advance_all;
 136                                    /* Last jiffy CBs were all advanced. */
 137        int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */
 138#endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
 139};
 140
 141/* RCU's kthread states for tracing. */
 142#define RCU_KTHREAD_STOPPED  0
 143#define RCU_KTHREAD_RUNNING  1
 144#define RCU_KTHREAD_WAITING  2
 145#define RCU_KTHREAD_OFFCPU   3
 146#define RCU_KTHREAD_YIELDING 4
 147#define RCU_KTHREAD_MAX      4
 148
 149/*
 150 * Definition for node within the RCU grace-period-detection hierarchy.
 151 */
 152struct rcu_node {
 153        raw_spinlock_t __private lock;  /* Root rcu_node's lock protects */
 154                                        /*  some rcu_state fields as well as */
 155                                        /*  following. */
 156        unsigned long gpnum;    /* Current grace period for this node. */
 157                                /*  This will either be equal to or one */
 158                                /*  behind the root rcu_node's gpnum. */
 159        unsigned long completed; /* Last GP completed for this node. */
 160                                /*  This will either be equal to or one */
 161                                /*  behind the root rcu_node's gpnum. */
 162        unsigned long qsmask;   /* CPUs or groups that need to switch in */
 163                                /*  order for current grace period to proceed.*/
 164                                /*  In leaf rcu_node, each bit corresponds to */
 165                                /*  an rcu_data structure, otherwise, each */
 166                                /*  bit corresponds to a child rcu_node */
 167                                /*  structure. */
 168        unsigned long qsmaskinit;
 169                                /* Per-GP initial value for qsmask. */
 170                                /*  Initialized from ->qsmaskinitnext at the */
 171                                /*  beginning of each grace period. */
 172        unsigned long qsmaskinitnext;
 173                                /* Online CPUs for next grace period. */
 174        unsigned long expmask;  /* CPUs or groups that need to check in */
 175                                /*  to allow the current expedited GP */
 176                                /*  to complete. */
 177        unsigned long expmaskinit;
 178                                /* Per-GP initial values for expmask. */
 179                                /*  Initialized from ->expmaskinitnext at the */
 180                                /*  beginning of each expedited GP. */
 181        unsigned long expmaskinitnext;
 182                                /* Online CPUs for next expedited GP. */
 183                                /*  Any CPU that has ever been online will */
 184                                /*  have its bit set. */
 185        unsigned long grpmask;  /* Mask to apply to parent qsmask. */
 186                                /*  Only one bit will be set in this mask. */
 187        int     grplo;          /* lowest-numbered CPU or group here. */
 188        int     grphi;          /* highest-numbered CPU or group here. */
 189        u8      grpnum;         /* CPU/group number for next level up. */
 190        u8      level;          /* root is at level 0. */
 191        bool    wait_blkd_tasks;/* Necessary to wait for blocked tasks to */
 192                                /*  exit RCU read-side critical sections */
 193                                /*  before propagating offline up the */
 194                                /*  rcu_node tree? */
 195        struct rcu_node *parent;
 196        struct list_head blkd_tasks;
 197                                /* Tasks blocked in RCU read-side critical */
 198                                /*  section.  Tasks are placed at the head */
 199                                /*  of this list and age towards the tail. */
 200        struct list_head *gp_tasks;
 201                                /* Pointer to the first task blocking the */
 202                                /*  current grace period, or NULL if there */
 203                                /*  is no such task. */
 204        struct list_head *exp_tasks;
 205                                /* Pointer to the first task blocking the */
 206                                /*  current expedited grace period, or NULL */
 207                                /*  if there is no such task.  If there */
 208                                /*  is no current expedited grace period, */
 209                                /*  then there can cannot be any such task. */
 210        struct list_head *boost_tasks;
 211                                /* Pointer to first task that needs to be */
 212                                /*  priority boosted, or NULL if no priority */
 213                                /*  boosting is needed for this rcu_node */
 214                                /*  structure.  If there are no tasks */
 215                                /*  queued on this rcu_node structure that */
 216                                /*  are blocking the current grace period, */
 217                                /*  there can be no such task. */
 218        struct rt_mutex boost_mtx;
 219                                /* Used only for the priority-boosting */
 220                                /*  side effect, not as a lock. */
 221        unsigned long boost_time;
 222                                /* When to start boosting (jiffies). */
 223        struct task_struct *boost_kthread_task;
 224                                /* kthread that takes care of priority */
 225                                /*  boosting for this rcu_node structure. */
 226        unsigned int boost_kthread_status;
 227                                /* State of boost_kthread_task for tracing. */
 228        unsigned long n_tasks_boosted;
 229                                /* Total number of tasks boosted. */
 230        unsigned long n_exp_boosts;
 231                                /* Number of tasks boosted for expedited GP. */
 232        unsigned long n_normal_boosts;
 233                                /* Number of tasks boosted for normal GP. */
 234        unsigned long n_balk_blkd_tasks;
 235                                /* Refused to boost: no blocked tasks. */
 236        unsigned long n_balk_exp_gp_tasks;
 237                                /* Refused to boost: nothing blocking GP. */
 238        unsigned long n_balk_boost_tasks;
 239                                /* Refused to boost: already boosting. */
 240        unsigned long n_balk_notblocked;
 241                                /* Refused to boost: RCU RS CS still running. */
 242        unsigned long n_balk_notyet;
 243                                /* Refused to boost: not yet time. */
 244        unsigned long n_balk_nos;
 245                                /* Refused to boost: not sure why, though. */
 246                                /*  This can happen due to race conditions. */
 247#ifdef CONFIG_RCU_NOCB_CPU
 248        struct swait_queue_head nocb_gp_wq[2];
 249                                /* Place for rcu_nocb_kthread() to wait GP. */
 250#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
 251        int need_future_gp[2];
 252                                /* Counts of upcoming no-CB GP requests. */
 253        raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp;
 254
 255        struct mutex exp_funnel_mutex ____cacheline_internodealigned_in_smp;
 256} ____cacheline_internodealigned_in_smp;
 257
 258/*
 259 * Do a full breadth-first scan of the rcu_node structures for the
 260 * specified rcu_state structure.
 261 */
 262#define rcu_for_each_node_breadth_first(rsp, rnp) \
 263        for ((rnp) = &(rsp)->node[0]; \
 264             (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
 265
 266/*
 267 * Do a breadth-first scan of the non-leaf rcu_node structures for the
 268 * specified rcu_state structure.  Note that if there is a singleton
 269 * rcu_node tree with but one rcu_node structure, this loop is a no-op.
 270 */
 271#define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
 272        for ((rnp) = &(rsp)->node[0]; \
 273             (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
 274
 275/*
 276 * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
 277 * structure.  Note that if there is a singleton rcu_node tree with but
 278 * one rcu_node structure, this loop -will- visit the rcu_node structure.
 279 * It is still a leaf node, even if it is also the root node.
 280 */
 281#define rcu_for_each_leaf_node(rsp, rnp) \
 282        for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
 283             (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
 284
 285/*
 286 * Union to allow "aggregate OR" operation on the need for a quiescent
 287 * state by the normal and expedited grace periods.
 288 */
 289union rcu_noqs {
 290        struct {
 291                u8 norm;
 292                u8 exp;
 293        } b; /* Bits. */
 294        u16 s; /* Set of bits, aggregate OR here. */
 295};
 296
 297/* Index values for nxttail array in struct rcu_data. */
 298#define RCU_DONE_TAIL           0       /* Also RCU_WAIT head. */
 299#define RCU_WAIT_TAIL           1       /* Also RCU_NEXT_READY head. */
 300#define RCU_NEXT_READY_TAIL     2       /* Also RCU_NEXT head. */
 301#define RCU_NEXT_TAIL           3
 302#define RCU_NEXT_SIZE           4
 303
 304/* Per-CPU data for read-copy update. */
 305struct rcu_data {
 306        /* 1) quiescent-state and grace-period handling : */
 307        unsigned long   completed;      /* Track rsp->completed gp number */
 308                                        /*  in order to detect GP end. */
 309        unsigned long   gpnum;          /* Highest gp number that this CPU */
 310                                        /*  is aware of having started. */
 311        unsigned long   rcu_qs_ctr_snap;/* Snapshot of rcu_qs_ctr to check */
 312                                        /*  for rcu_all_qs() invocations. */
 313        union rcu_noqs  cpu_no_qs;      /* No QSes yet for this CPU. */
 314        bool            core_needs_qs;  /* Core waits for quiesc state. */
 315        bool            beenonline;     /* CPU online at least once. */
 316        bool            gpwrap;         /* Possible gpnum/completed wrap. */
 317        struct rcu_node *mynode;        /* This CPU's leaf of hierarchy */
 318        unsigned long grpmask;          /* Mask to apply to leaf qsmask. */
 319        unsigned long   ticks_this_gp;  /* The number of scheduling-clock */
 320                                        /*  ticks this CPU has handled */
 321                                        /*  during and after the last grace */
 322                                        /* period it is aware of. */
 323
 324        /* 2) batch handling */
 325        /*
 326         * If nxtlist is not NULL, it is partitioned as follows.
 327         * Any of the partitions might be empty, in which case the
 328         * pointer to that partition will be equal to the pointer for
 329         * the following partition.  When the list is empty, all of
 330         * the nxttail elements point to the ->nxtlist pointer itself,
 331         * which in that case is NULL.
 332         *
 333         * [nxtlist, *nxttail[RCU_DONE_TAIL]):
 334         *      Entries that batch # <= ->completed
 335         *      The grace period for these entries has completed, and
 336         *      the other grace-period-completed entries may be moved
 337         *      here temporarily in rcu_process_callbacks().
 338         * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
 339         *      Entries that batch # <= ->completed - 1: waiting for current GP
 340         * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
 341         *      Entries known to have arrived before current GP ended
 342         * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
 343         *      Entries that might have arrived after current GP ended
 344         *      Note that the value of *nxttail[RCU_NEXT_TAIL] will
 345         *      always be NULL, as this is the end of the list.
 346         */
 347        struct rcu_head *nxtlist;
 348        struct rcu_head **nxttail[RCU_NEXT_SIZE];
 349        unsigned long   nxtcompleted[RCU_NEXT_SIZE];
 350                                        /* grace periods for sublists. */
 351        long            qlen_lazy;      /* # of lazy queued callbacks */
 352        long            qlen;           /* # of queued callbacks, incl lazy */
 353        long            qlen_last_fqs_check;
 354                                        /* qlen at last check for QS forcing */
 355        unsigned long   n_cbs_invoked;  /* count of RCU cbs invoked. */
 356        unsigned long   n_nocbs_invoked; /* count of no-CBs RCU cbs invoked. */
 357        unsigned long   n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
 358        unsigned long   n_cbs_adopted;  /* RCU cbs adopted from dying CPU */
 359        unsigned long   n_force_qs_snap;
 360                                        /* did other CPU force QS recently? */
 361        long            blimit;         /* Upper limit on a processed batch */
 362
 363        /* 3) dynticks interface. */
 364        struct rcu_dynticks *dynticks;  /* Shared per-CPU dynticks state. */
 365        int dynticks_snap;              /* Per-GP tracking for dynticks. */
 366
 367        /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
 368        unsigned long dynticks_fqs;     /* Kicked due to dynticks idle. */
 369        unsigned long offline_fqs;      /* Kicked due to being offline. */
 370        unsigned long cond_resched_completed;
 371                                        /* Grace period that needs help */
 372                                        /*  from cond_resched(). */
 373
 374        /* 5) __rcu_pending() statistics. */
 375        unsigned long n_rcu_pending;    /* rcu_pending() calls since boot. */
 376        unsigned long n_rp_core_needs_qs;
 377        unsigned long n_rp_report_qs;
 378        unsigned long n_rp_cb_ready;
 379        unsigned long n_rp_cpu_needs_gp;
 380        unsigned long n_rp_gp_completed;
 381        unsigned long n_rp_gp_started;
 382        unsigned long n_rp_nocb_defer_wakeup;
 383        unsigned long n_rp_need_nothing;
 384
 385        /* 6) _rcu_barrier(), OOM callbacks, and expediting. */
 386        struct rcu_head barrier_head;
 387#ifdef CONFIG_RCU_FAST_NO_HZ
 388        struct rcu_head oom_head;
 389#endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
 390        struct mutex exp_funnel_mutex;
 391        atomic_long_t expedited_workdone0;      /* # done by others #0. */
 392        atomic_long_t expedited_workdone1;      /* # done by others #1. */
 393        atomic_long_t expedited_workdone2;      /* # done by others #2. */
 394        atomic_long_t expedited_workdone3;      /* # done by others #3. */
 395
 396        /* 7) Callback offloading. */
 397#ifdef CONFIG_RCU_NOCB_CPU
 398        struct rcu_head *nocb_head;     /* CBs waiting for kthread. */
 399        struct rcu_head **nocb_tail;
 400        atomic_long_t nocb_q_count;     /* # CBs waiting for nocb */
 401        atomic_long_t nocb_q_count_lazy; /*  invocation (all stages). */
 402        struct rcu_head *nocb_follower_head; /* CBs ready to invoke. */
 403        struct rcu_head **nocb_follower_tail;
 404        struct swait_queue_head nocb_wq; /* For nocb kthreads to sleep on. */
 405        struct task_struct *nocb_kthread;
 406        int nocb_defer_wakeup;          /* Defer wakeup of nocb_kthread. */
 407
 408        /* The following fields are used by the leader, hence own cacheline. */
 409        struct rcu_head *nocb_gp_head ____cacheline_internodealigned_in_smp;
 410                                        /* CBs waiting for GP. */
 411        struct rcu_head **nocb_gp_tail;
 412        bool nocb_leader_sleep;         /* Is the nocb leader thread asleep? */
 413        struct rcu_data *nocb_next_follower;
 414                                        /* Next follower in wakeup chain. */
 415
 416        /* The following fields are used by the follower, hence new cachline. */
 417        struct rcu_data *nocb_leader ____cacheline_internodealigned_in_smp;
 418                                        /* Leader CPU takes GP-end wakeups. */
 419#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
 420
 421        /* 8) RCU CPU stall data. */
 422        unsigned int softirq_snap;      /* Snapshot of softirq activity. */
 423
 424        int cpu;
 425        struct rcu_state *rsp;
 426};
 427
 428/* Values for nocb_defer_wakeup field in struct rcu_data. */
 429#define RCU_NOGP_WAKE_NOT       0
 430#define RCU_NOGP_WAKE           1
 431#define RCU_NOGP_WAKE_FORCE     2
 432
 433#define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500))
 434                                        /* For jiffies_till_first_fqs and */
 435                                        /*  and jiffies_till_next_fqs. */
 436
 437#define RCU_JIFFIES_FQS_DIV     256     /* Very large systems need more */
 438                                        /*  delay between bouts of */
 439                                        /*  quiescent-state forcing. */
 440
 441#define RCU_STALL_RAT_DELAY     2       /* Allow other CPUs time to take */
 442                                        /*  at least one scheduling clock */
 443                                        /*  irq before ratting on them. */
 444
 445#define rcu_wait(cond)                                                  \
 446do {                                                                    \
 447        for (;;) {                                                      \
 448                set_current_state(TASK_INTERRUPTIBLE);                  \
 449                if (cond)                                               \
 450                        break;                                          \
 451                schedule();                                             \
 452        }                                                               \
 453        __set_current_state(TASK_RUNNING);                              \
 454} while (0)
 455
 456/*
 457 * RCU global state, including node hierarchy.  This hierarchy is
 458 * represented in "heap" form in a dense array.  The root (first level)
 459 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
 460 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
 461 * and the third level in ->node[m+1] and following (->node[m+1] referenced
 462 * by ->level[2]).  The number of levels is determined by the number of
 463 * CPUs and by CONFIG_RCU_FANOUT.  Small systems will have a "hierarchy"
 464 * consisting of a single rcu_node.
 465 */
 466struct rcu_state {
 467        struct rcu_node node[NUM_RCU_NODES];    /* Hierarchy. */
 468        struct rcu_node *level[RCU_NUM_LVLS + 1];
 469                                                /* Hierarchy levels (+1 to */
 470                                                /*  shut bogus gcc warning) */
 471        u8 flavor_mask;                         /* bit in flavor mask. */
 472        struct rcu_data __percpu *rda;          /* pointer of percu rcu_data. */
 473        call_rcu_func_t call;                   /* call_rcu() flavor. */
 474        int ncpus;                              /* # CPUs seen so far. */
 475
 476        /* The following fields are guarded by the root rcu_node's lock. */
 477
 478        u8      boost ____cacheline_internodealigned_in_smp;
 479                                                /* Subject to priority boost. */
 480        unsigned long gpnum;                    /* Current gp number. */
 481        unsigned long completed;                /* # of last completed gp. */
 482        struct task_struct *gp_kthread;         /* Task for grace periods. */
 483        struct swait_queue_head gp_wq;          /* Where GP task waits. */
 484        short gp_flags;                         /* Commands for GP task. */
 485        short gp_state;                         /* GP kthread sleep state. */
 486
 487        /* End of fields guarded by root rcu_node's lock. */
 488
 489        raw_spinlock_t orphan_lock ____cacheline_internodealigned_in_smp;
 490                                                /* Protect following fields. */
 491        struct rcu_head *orphan_nxtlist;        /* Orphaned callbacks that */
 492                                                /*  need a grace period. */
 493        struct rcu_head **orphan_nxttail;       /* Tail of above. */
 494        struct rcu_head *orphan_donelist;       /* Orphaned callbacks that */
 495                                                /*  are ready to invoke. */
 496        struct rcu_head **orphan_donetail;      /* Tail of above. */
 497        long qlen_lazy;                         /* Number of lazy callbacks. */
 498        long qlen;                              /* Total number of callbacks. */
 499        /* End of fields guarded by orphan_lock. */
 500
 501        struct mutex barrier_mutex;             /* Guards barrier fields. */
 502        atomic_t barrier_cpu_count;             /* # CPUs waiting on. */
 503        struct completion barrier_completion;   /* Wake at barrier end. */
 504        unsigned long barrier_sequence;         /* ++ at start and end of */
 505                                                /*  _rcu_barrier(). */
 506        /* End of fields guarded by barrier_mutex. */
 507
 508        unsigned long expedited_sequence;       /* Take a ticket. */
 509        atomic_long_t expedited_normal;         /* # fallbacks to normal. */
 510        atomic_t expedited_need_qs;             /* # CPUs left to check in. */
 511        struct swait_queue_head expedited_wq;   /* Wait for check-ins. */
 512        int ncpus_snap;                         /* # CPUs seen last time. */
 513
 514        unsigned long jiffies_force_qs;         /* Time at which to invoke */
 515                                                /*  force_quiescent_state(). */
 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
 553static 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
 564extern 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 */
 573extern struct rcu_state rcu_sched_state;
 574
 575extern struct rcu_state rcu_bh_state;
 576
 577#ifdef CONFIG_PREEMPT_RCU
 578extern struct rcu_state rcu_preempt_state;
 579#endif /* #ifdef CONFIG_PREEMPT_RCU */
 580
 581#ifdef CONFIG_RCU_BOOST
 582DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
 583DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu);
 584DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
 585DECLARE_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 */
 591static void rcu_bootup_announce(void);
 592static void rcu_preempt_note_context_switch(void);
 593static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
 594#ifdef CONFIG_HOTPLUG_CPU
 595static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
 596#endif /* #ifdef CONFIG_HOTPLUG_CPU */
 597static void rcu_print_detail_task_stall(struct rcu_state *rsp);
 598static int rcu_print_task_stall(struct rcu_node *rnp);
 599static int rcu_print_task_exp_stall(struct rcu_node *rnp);
 600static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
 601static void rcu_preempt_check_callbacks(void);
 602void call_rcu(struct rcu_head *head, rcu_callback_t func);
 603static void __init __rcu_init_preempt(void);
 604static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
 605static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
 606static void invoke_rcu_callbacks_kthread(void);
 607static bool rcu_is_callbacks_kthread(void);
 608#ifdef CONFIG_RCU_BOOST
 609static void rcu_preempt_do_callbacks(void);
 610static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
 611                                                 struct rcu_node *rnp);
 612#endif /* #ifdef CONFIG_RCU_BOOST */
 613static void __init rcu_spawn_boost_kthreads(void);
 614static void rcu_prepare_kthreads(int cpu);
 615static void rcu_cleanup_after_idle(void);
 616static void rcu_prepare_for_idle(void);
 617static void rcu_idle_count_callbacks_posted(void);
 618static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
 619static void print_cpu_stall_info_begin(void);
 620static void print_cpu_stall_info(struct rcu_state *rsp, int cpu);
 621static void print_cpu_stall_info_end(void);
 622static void zero_cpu_stall_ticks(struct rcu_data *rdp);
 623static void increment_cpu_stall_ticks(void);
 624static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu);
 625static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq);
 626static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp);
 627static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq);
 628static void rcu_init_one_nocb(struct rcu_node *rnp);
 629static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
 630                            bool lazy, unsigned long flags);
 631static bool rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp,
 632                                      struct rcu_data *rdp,
 633                                      unsigned long flags);
 634static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp);
 635static void do_nocb_deferred_wakeup(struct rcu_data *rdp);
 636static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp);
 637static void rcu_spawn_all_nocb_kthreads(int cpu);
 638static void __init rcu_spawn_nocb_kthreads(void);
 639#ifdef CONFIG_RCU_NOCB_CPU
 640static void __init rcu_organize_nocb_kthreads(struct rcu_state *rsp);
 641#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
 642static void __maybe_unused rcu_kick_nohz_cpu(int cpu);
 643static bool init_nocb_callback_list(struct rcu_data *rdp);
 644static void rcu_sysidle_enter(int irq);
 645static void rcu_sysidle_exit(int irq);
 646static void rcu_sysidle_check_cpu(struct rcu_data *rdp, bool *isidle,
 647                                  unsigned long *maxj);
 648static bool is_sysidle_rcu_state(struct rcu_state *rsp);
 649static void rcu_sysidle_report_gp(struct rcu_state *rsp, int isidle,
 650                                  unsigned long maxj);
 651static void rcu_bind_gp_kthread(void);
 652static void rcu_sysidle_init_percpu_data(struct rcu_dynticks *rdtp);
 653static bool rcu_nohz_full_cpu(struct rcu_state *rsp);
 654static void rcu_dynticks_task_enter(void);
 655static 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. */
 661static 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 */
 699static 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
 705static inline void raw_spin_unlock_rcu_node(struct rcu_node *rnp)
 706{
 707        raw_spin_unlock(&ACCESS_PRIVATE(rnp, lock));
 708}
 709
 710static 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
 716static 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)                      \
 722do {                                                                    \
 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)                 \
 729do {                                                                    \
 730        typecheck(unsigned long, flags);                                \
 731        raw_spin_unlock_irqrestore(&ACCESS_PRIVATE(rnp, lock), flags);  \
 732} while (0)
 733
 734static 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}
 742