LXR linux/kernel/rcutree.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, write to the Free Software
  17 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  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/irq_work.h>
  31
  32/*
  33 * Define shape of hierarchy based on NR_CPUS, CONFIG_RCU_FANOUT, and
  34 * CONFIG_RCU_FANOUT_LEAF.
  35 * In theory, it should be possible to add more levels straightforwardly.
  36 * In practice, this did work well going from three levels to four.
  37 * Of course, your mileage may vary.
  38 */
  39#define MAX_RCU_LVLS 4
  40#define RCU_FANOUT_1          (CONFIG_RCU_FANOUT_LEAF)
  41#define RCU_FANOUT_2          (RCU_FANOUT_1 * CONFIG_RCU_FANOUT)
  42#define RCU_FANOUT_3          (RCU_FANOUT_2 * CONFIG_RCU_FANOUT)
  43#define RCU_FANOUT_4          (RCU_FANOUT_3 * CONFIG_RCU_FANOUT)
  44
  45#if NR_CPUS <= RCU_FANOUT_1
  46#  define RCU_NUM_LVLS        1
  47#  define NUM_RCU_LVL_0       1
  48#  define NUM_RCU_LVL_1       (NR_CPUS)
  49#  define NUM_RCU_LVL_2       0
  50#  define NUM_RCU_LVL_3       0
  51#  define NUM_RCU_LVL_4       0
  52#elif NR_CPUS <= RCU_FANOUT_2
  53#  define RCU_NUM_LVLS        2
  54#  define NUM_RCU_LVL_0       1
  55#  define NUM_RCU_LVL_1       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
  56#  define NUM_RCU_LVL_2       (NR_CPUS)
  57#  define NUM_RCU_LVL_3       0
  58#  define NUM_RCU_LVL_4       0
  59#elif NR_CPUS <= RCU_FANOUT_3
  60#  define RCU_NUM_LVLS        3
  61#  define NUM_RCU_LVL_0       1
  62#  define NUM_RCU_LVL_1       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
  63#  define NUM_RCU_LVL_2       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
  64#  define NUM_RCU_LVL_3       (NR_CPUS)
  65#  define NUM_RCU_LVL_4       0
  66#elif NR_CPUS <= RCU_FANOUT_4
  67#  define RCU_NUM_LVLS        4
  68#  define NUM_RCU_LVL_0       1
  69#  define NUM_RCU_LVL_1       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3)
  70#  define NUM_RCU_LVL_2       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
  71#  define NUM_RCU_LVL_3       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
  72#  define NUM_RCU_LVL_4       (NR_CPUS)
  73#else
  74# error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
  75#endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
  76
  77#define RCU_SUM (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3 + NUM_RCU_LVL_4)
  78#define NUM_RCU_NODES (RCU_SUM - NR_CPUS)
  79
  80extern int rcu_num_lvls;
  81extern int rcu_num_nodes;
  82
  83/*
  84 * Dynticks per-CPU state.
  85 */
  86struct rcu_dynticks {
  87        long long dynticks_nesting; /* Track irq/process nesting level. */
  88                                    /* Process level is worth LLONG_MAX/2. */
  89        int dynticks_nmi_nesting;   /* Track NMI nesting level. */
  90        atomic_t dynticks;          /* Even value for idle, else odd. */
  91#ifdef CONFIG_RCU_FAST_NO_HZ
  92        bool all_lazy;              /* Are all CPU's CBs lazy? */
  93        unsigned long nonlazy_posted;
  94                                    /* # times non-lazy CBs posted to CPU. */
  95        unsigned long nonlazy_posted_snap;
  96                                    /* idle-period nonlazy_posted snapshot. */
  97        unsigned long last_accelerate;
  98                                    /* Last jiffy CBs were accelerated. */
  99        int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */
 100#endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
 101};
 102
 103/* RCU's kthread states for tracing. */
 104#define RCU_KTHREAD_STOPPED  0
 105#define RCU_KTHREAD_RUNNING  1
 106#define RCU_KTHREAD_WAITING  2
 107#define RCU_KTHREAD_OFFCPU   3
 108#define RCU_KTHREAD_YIELDING 4
 109#define RCU_KTHREAD_MAX      4
 110
 111/*
 112 * Definition for node within the RCU grace-period-detection hierarchy.
 113 */
 114struct rcu_node {
 115        raw_spinlock_t lock;    /* Root rcu_node's lock protects some */
 116                                /*  rcu_state fields as well as following. */
 117        unsigned long gpnum;    /* Current grace period for this node. */
 118                                /*  This will either be equal to or one */
 119                                /*  behind the root rcu_node's gpnum. */
 120        unsigned long completed; /* Last GP completed for this node. */
 121                                /*  This will either be equal to or one */
 122                                /*  behind the root rcu_node's gpnum. */
 123        unsigned long qsmask;   /* CPUs or groups that need to switch in */
 124                                /*  order for current grace period to proceed.*/
 125                                /*  In leaf rcu_node, each bit corresponds to */
 126                                /*  an rcu_data structure, otherwise, each */
 127                                /*  bit corresponds to a child rcu_node */
 128                                /*  structure. */
 129        unsigned long expmask;  /* Groups that have ->blkd_tasks */
 130                                /*  elements that need to drain to allow the */
 131                                /*  current expedited grace period to */
 132                                /*  complete (only for TREE_PREEMPT_RCU). */
 133        unsigned long qsmaskinit;
 134                                /* Per-GP initial value for qsmask & expmask. */
 135        unsigned long grpmask;  /* Mask to apply to parent qsmask. */
 136                                /*  Only one bit will be set in this mask. */
 137        int     grplo;          /* lowest-numbered CPU or group here. */
 138        int     grphi;          /* highest-numbered CPU or group here. */
 139        u8      grpnum;         /* CPU/group number for next level up. */
 140        u8      level;          /* root is at level 0. */
 141        struct rcu_node *parent;
 142        struct list_head blkd_tasks;
 143                                /* Tasks blocked in RCU read-side critical */
 144                                /*  section.  Tasks are placed at the head */
 145                                /*  of this list and age towards the tail. */
 146        struct list_head *gp_tasks;
 147                                /* Pointer to the first task blocking the */
 148                                /*  current grace period, or NULL if there */
 149                                /*  is no such task. */
 150        struct list_head *exp_tasks;
 151                                /* Pointer to the first task blocking the */
 152                                /*  current expedited grace period, or NULL */
 153                                /*  if there is no such task.  If there */
 154                                /*  is no current expedited grace period, */
 155                                /*  then there can cannot be any such task. */
 156#ifdef CONFIG_RCU_BOOST
 157        struct list_head *boost_tasks;
 158                                /* Pointer to first task that needs to be */
 159                                /*  priority boosted, or NULL if no priority */
 160                                /*  boosting is needed for this rcu_node */
 161                                /*  structure.  If there are no tasks */
 162                                /*  queued on this rcu_node structure that */
 163                                /*  are blocking the current grace period, */
 164                                /*  there can be no such task. */
 165        unsigned long boost_time;
 166                                /* When to start boosting (jiffies). */
 167        struct task_struct *boost_kthread_task;
 168                                /* kthread that takes care of priority */
 169                                /*  boosting for this rcu_node structure. */
 170        unsigned int boost_kthread_status;
 171                                /* State of boost_kthread_task for tracing. */
 172        unsigned long n_tasks_boosted;
 173                                /* Total number of tasks boosted. */
 174        unsigned long n_exp_boosts;
 175                                /* Number of tasks boosted for expedited GP. */
 176        unsigned long n_normal_boosts;
 177                                /* Number of tasks boosted for normal GP. */
 178        unsigned long n_balk_blkd_tasks;
 179                                /* Refused to boost: no blocked tasks. */
 180        unsigned long n_balk_exp_gp_tasks;
 181                                /* Refused to boost: nothing blocking GP. */
 182        unsigned long n_balk_boost_tasks;
 183                                /* Refused to boost: already boosting. */
 184        unsigned long n_balk_notblocked;
 185                                /* Refused to boost: RCU RS CS still running. */
 186        unsigned long n_balk_notyet;
 187                                /* Refused to boost: not yet time. */
 188        unsigned long n_balk_nos;
 189                                /* Refused to boost: not sure why, though. */
 190                                /*  This can happen due to race conditions. */
 191#endif /* #ifdef CONFIG_RCU_BOOST */
 192#ifdef CONFIG_RCU_NOCB_CPU
 193        wait_queue_head_t nocb_gp_wq[2];
 194                                /* Place for rcu_nocb_kthread() to wait GP. */
 195#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
 196        int need_future_gp[2];
 197                                /* Counts of upcoming no-CB GP requests. */
 198        raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp;
 199} ____cacheline_internodealigned_in_smp;
 200
 201/*
 202 * Do a full breadth-first scan of the rcu_node structures for the
 203 * specified rcu_state structure.
 204 */
 205#define rcu_for_each_node_breadth_first(rsp, rnp) \
 206        for ((rnp) = &(rsp)->node[0]; \
 207             (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
 208
 209/*
 210 * Do a breadth-first scan of the non-leaf rcu_node structures for the
 211 * specified rcu_state structure.  Note that if there is a singleton
 212 * rcu_node tree with but one rcu_node structure, this loop is a no-op.
 213 */
 214#define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
 215        for ((rnp) = &(rsp)->node[0]; \
 216             (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
 217
 218/*
 219 * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
 220 * structure.  Note that if there is a singleton rcu_node tree with but
 221 * one rcu_node structure, this loop -will- visit the rcu_node structure.
 222 * It is still a leaf node, even if it is also the root node.
 223 */
 224#define rcu_for_each_leaf_node(rsp, rnp) \
 225        for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
 226             (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
 227
 228/* Index values for nxttail array in struct rcu_data. */
 229#define RCU_DONE_TAIL           0       /* Also RCU_WAIT head. */
 230#define RCU_WAIT_TAIL           1       /* Also RCU_NEXT_READY head. */
 231#define RCU_NEXT_READY_TAIL     2       /* Also RCU_NEXT head. */
 232#define RCU_NEXT_TAIL           3
 233#define RCU_NEXT_SIZE           4
 234
 235/* Per-CPU data for read-copy update. */
 236struct rcu_data {
 237        /* 1) quiescent-state and grace-period handling : */
 238        unsigned long   completed;      /* Track rsp->completed gp number */
 239                                        /*  in order to detect GP end. */
 240        unsigned long   gpnum;          /* Highest gp number that this CPU */
 241                                        /*  is aware of having started. */
 242        bool            passed_quiesce; /* User-mode/idle loop etc. */
 243        bool            qs_pending;     /* Core waits for quiesc state. */
 244        bool            beenonline;     /* CPU online at least once. */
 245        bool            preemptible;    /* Preemptible RCU? */
 246        struct rcu_node *mynode;        /* This CPU's leaf of hierarchy */
 247        unsigned long grpmask;          /* Mask to apply to leaf qsmask. */
 248#ifdef CONFIG_RCU_CPU_STALL_INFO
 249        unsigned long   ticks_this_gp;  /* The number of scheduling-clock */
 250                                        /*  ticks this CPU has handled */
 251                                        /*  during and after the last grace */
 252                                        /* period it is aware of. */
 253#endif /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
 254
 255        /* 2) batch handling */
 256        /*
 257         * If nxtlist is not NULL, it is partitioned as follows.
 258         * Any of the partitions might be empty, in which case the
 259         * pointer to that partition will be equal to the pointer for
 260         * the following partition.  When the list is empty, all of
 261         * the nxttail elements point to the ->nxtlist pointer itself,
 262         * which in that case is NULL.
 263         *
 264         * [nxtlist, *nxttail[RCU_DONE_TAIL]):
 265         *      Entries that batch # <= ->completed
 266         *      The grace period for these entries has completed, and
 267         *      the other grace-period-completed entries may be moved
 268         *      here temporarily in rcu_process_callbacks().
 269         * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
 270         *      Entries that batch # <= ->completed - 1: waiting for current GP
 271         * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
 272         *      Entries known to have arrived before current GP ended
 273         * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
 274         *      Entries that might have arrived after current GP ended
 275         *      Note that the value of *nxttail[RCU_NEXT_TAIL] will
 276         *      always be NULL, as this is the end of the list.
 277         */
 278        struct rcu_head *nxtlist;
 279        struct rcu_head **nxttail[RCU_NEXT_SIZE];
 280        unsigned long   nxtcompleted[RCU_NEXT_SIZE];
 281                                        /* grace periods for sublists. */
 282        long            qlen_lazy;      /* # of lazy queued callbacks */
 283        long            qlen;           /* # of queued callbacks, incl lazy */
 284        long            qlen_last_fqs_check;
 285                                        /* qlen at last check for QS forcing */
 286        unsigned long   n_cbs_invoked;  /* count of RCU cbs invoked. */
 287        unsigned long   n_nocbs_invoked; /* count of no-CBs RCU cbs invoked. */
 288        unsigned long   n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
 289        unsigned long   n_cbs_adopted;  /* RCU cbs adopted from dying CPU */
 290        unsigned long   n_force_qs_snap;
 291                                        /* did other CPU force QS recently? */
 292        long            blimit;         /* Upper limit on a processed batch */
 293
 294        /* 3) dynticks interface. */
 295        struct rcu_dynticks *dynticks;  /* Shared per-CPU dynticks state. */
 296        int dynticks_snap;              /* Per-GP tracking for dynticks. */
 297
 298        /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
 299        unsigned long dynticks_fqs;     /* Kicked due to dynticks idle. */
 300        unsigned long offline_fqs;      /* Kicked due to being offline. */
 301
 302        /* 5) __rcu_pending() statistics. */
 303        unsigned long n_rcu_pending;    /* rcu_pending() calls since boot. */
 304        unsigned long n_rp_qs_pending;
 305        unsigned long n_rp_report_qs;
 306        unsigned long n_rp_cb_ready;
 307        unsigned long n_rp_cpu_needs_gp;
 308        unsigned long n_rp_gp_completed;
 309        unsigned long n_rp_gp_started;
 310        unsigned long n_rp_need_nothing;
 311
 312        /* 6) _rcu_barrier() and OOM callbacks. */
 313        struct rcu_head barrier_head;
 314#ifdef CONFIG_RCU_FAST_NO_HZ
 315        struct rcu_head oom_head;
 316#endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
 317
 318        /* 7) Callback offloading. */
 319#ifdef CONFIG_RCU_NOCB_CPU
 320        struct rcu_head *nocb_head;     /* CBs waiting for kthread. */
 321        struct rcu_head **nocb_tail;
 322        atomic_long_t nocb_q_count;     /* # CBs waiting for kthread */
 323        atomic_long_t nocb_q_count_lazy; /*  (approximate). */
 324        int nocb_p_count;               /* # CBs being invoked by kthread */
 325        int nocb_p_count_lazy;          /*  (approximate). */
 326        wait_queue_head_t nocb_wq;      /* For nocb kthreads to sleep on. */
 327        struct task_struct *nocb_kthread;
 328#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
 329
 330        /* 8) RCU CPU stall data. */
 331#ifdef CONFIG_RCU_CPU_STALL_INFO
 332        unsigned int softirq_snap;      /* Snapshot of softirq activity. */
 333#endif /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
 334
 335        int cpu;
 336        struct rcu_state *rsp;
 337};
 338
 339/* Values for fqs_state field in struct rcu_state. */
 340#define RCU_GP_IDLE             0       /* No grace period in progress. */
 341#define RCU_GP_INIT             1       /* Grace period being initialized. */
 342#define RCU_SAVE_DYNTICK        2       /* Need to scan dyntick state. */
 343#define RCU_FORCE_QS            3       /* Need to force quiescent state. */
 344#define RCU_SIGNAL_INIT         RCU_SAVE_DYNTICK
 345
 346#define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500))
 347                                        /* For jiffies_till_first_fqs and */
 348                                        /*  and jiffies_till_next_fqs. */
 349
 350#define RCU_JIFFIES_FQS_DIV     256     /* Very large systems need more */
 351                                        /*  delay between bouts of */
 352                                        /*  quiescent-state forcing. */
 353
 354#define RCU_STALL_RAT_DELAY     2       /* Allow other CPUs time to take */
 355                                        /*  at least one scheduling clock */
 356                                        /*  irq before ratting on them. */
 357
 358#define rcu_wait(cond)                                                  \
 359do {                                                                    \
 360        for (;;) {                                                      \
 361                set_current_state(TASK_INTERRUPTIBLE);                  \
 362                if (cond)                                               \
 363                        break;                                          \
 364                schedule();                                             \
 365        }                                                               \
 366        __set_current_state(TASK_RUNNING);                              \
 367} while (0)
 368
 369/*
 370 * RCU global state, including node hierarchy.  This hierarchy is
 371 * represented in "heap" form in a dense array.  The root (first level)
 372 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
 373 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
 374 * and the third level in ->node[m+1] and following (->node[m+1] referenced
 375 * by ->level[2]).  The number of levels is determined by the number of
 376 * CPUs and by CONFIG_RCU_FANOUT.  Small systems will have a "hierarchy"
 377 * consisting of a single rcu_node.
 378 */
 379struct rcu_state {
 380        struct rcu_node node[NUM_RCU_NODES];    /* Hierarchy. */
 381        struct rcu_node *level[RCU_NUM_LVLS];   /* Hierarchy levels. */
 382        u32 levelcnt[MAX_RCU_LVLS + 1];         /* # nodes in each level. */
 383        u8 levelspread[RCU_NUM_LVLS];           /* kids/node in each level. */
 384        struct rcu_data __percpu *rda;          /* pointer of percu rcu_data. */
 385        void (*call)(struct rcu_head *head,     /* call_rcu() flavor. */
 386                     void (*func)(struct rcu_head *head));
 387
 388        /* The following fields are guarded by the root rcu_node's lock. */
 389
 390        u8      fqs_state ____cacheline_internodealigned_in_smp;
 391                                                /* Force QS state. */
 392        u8      boost;                          /* Subject to priority boost. */
 393        unsigned long gpnum;                    /* Current gp number. */
 394        unsigned long completed;                /* # of last completed gp. */
 395        struct task_struct *gp_kthread;         /* Task for grace periods. */
 396        wait_queue_head_t gp_wq;                /* Where GP task waits. */
 397        int gp_flags;                           /* Commands for GP task. */
 398
 399        /* End of fields guarded by root rcu_node's lock. */
 400
 401        raw_spinlock_t orphan_lock ____cacheline_internodealigned_in_smp;
 402                                                /* Protect following fields. */
 403        struct rcu_head *orphan_nxtlist;        /* Orphaned callbacks that */
 404                                                /*  need a grace period. */
 405        struct rcu_head **orphan_nxttail;       /* Tail of above. */
 406        struct rcu_head *orphan_donelist;       /* Orphaned callbacks that */
 407                                                /*  are ready to invoke. */
 408        struct rcu_head **orphan_donetail;      /* Tail of above. */
 409        long qlen_lazy;                         /* Number of lazy callbacks. */
 410        long qlen;                              /* Total number of callbacks. */
 411        /* End of fields guarded by orphan_lock. */
 412
 413        struct mutex onoff_mutex;               /* Coordinate hotplug & GPs. */
 414
 415        struct mutex barrier_mutex;             /* Guards barrier fields. */
 416        atomic_t barrier_cpu_count;             /* # CPUs waiting on. */
 417        struct completion barrier_completion;   /* Wake at barrier end. */
 418        unsigned long n_barrier_done;           /* ++ at start and end of */
 419                                                /*  _rcu_barrier(). */
 420        /* End of fields guarded by barrier_mutex. */
 421
 422        atomic_long_t expedited_start;          /* Starting ticket. */
 423        atomic_long_t expedited_done;           /* Done ticket. */
 424        atomic_long_t expedited_wrap;           /* # near-wrap incidents. */
 425        atomic_long_t expedited_tryfail;        /* # acquisition failures. */
 426        atomic_long_t expedited_workdone1;      /* # done by others #1. */
 427        atomic_long_t expedited_workdone2;      /* # done by others #2. */
 428        atomic_long_t expedited_normal;         /* # fallbacks to normal. */
 429        atomic_long_t expedited_stoppedcpus;    /* # successful stop_cpus. */
 430        atomic_long_t expedited_done_tries;     /* # tries to update _done. */
 431        atomic_long_t expedited_done_lost;      /* # times beaten to _done. */
 432        atomic_long_t expedited_done_exit;      /* # times exited _done loop. */
 433
 434        unsigned long jiffies_force_qs;         /* Time at which to invoke */
 435                                                /*  force_quiescent_state(). */
 436        unsigned long n_force_qs;               /* Number of calls to */
 437                                                /*  force_quiescent_state(). */
 438        unsigned long n_force_qs_lh;            /* ~Number of calls leaving */
 439                                                /*  due to lock unavailable. */
 440        unsigned long n_force_qs_ngp;           /* Number of calls leaving */
 441                                                /*  due to no GP active. */
 442        unsigned long gp_start;                 /* Time at which GP started, */
 443                                                /*  but in jiffies. */
 444        unsigned long gp_activity;              /* Time of last GP kthread */
 445                                                /*  activity in jiffies. */
 446        unsigned long jiffies_stall;            /* Time at which to check */
 447                                                /*  for CPU stalls. */
 448        unsigned long jiffies_resched;          /* Time at which to resched */
 449                                                /*  a reluctant CPU. */
 450        unsigned long gp_max;                   /* Maximum GP duration in */
 451                                                /*  jiffies. */
 452        char *name;                             /* Name of structure. */
 453        char abbr;                              /* Abbreviated name. */
 454        struct list_head flavors;               /* List of RCU flavors. */
 455        struct irq_work wakeup_work;            /* Postponed wakeups */
 456};
 457
 458/* Values for rcu_state structure's gp_flags field. */
 459#define RCU_GP_FLAG_INIT 0x1    /* Need grace-period initialization. */
 460#define RCU_GP_FLAG_FQS  0x2    /* Need grace-period quiescent-state forcing. */
 461
 462extern struct list_head rcu_struct_flavors;
 463
 464/* Sequence through rcu_state structures for each RCU flavor. */
 465#define for_each_rcu_flavor(rsp) \
 466        list_for_each_entry((rsp), &rcu_struct_flavors, flavors)
 467
 468/* Return values for rcu_preempt_offline_tasks(). */
 469
 470#define RCU_OFL_TASKS_NORM_GP   0x1             /* Tasks blocking normal */
 471                                                /*  GP were moved to root. */
 472#define RCU_OFL_TASKS_EXP_GP    0x2             /* Tasks blocking expedited */
 473                                                /*  GP were moved to root. */
 474
 475/*
 476 * RCU implementation internal declarations:
 477 */
 478extern struct rcu_state rcu_sched_state;
 479DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
 480
 481extern struct rcu_state rcu_bh_state;
 482DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
 483
 484#ifdef CONFIG_TREE_PREEMPT_RCU
 485extern struct rcu_state rcu_preempt_state;
 486DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
 487#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
 488
 489#ifdef CONFIG_RCU_BOOST
 490DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
 491DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu);
 492DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
 493DECLARE_PER_CPU(char, rcu_cpu_has_work);
 494#endif /* #ifdef CONFIG_RCU_BOOST */
 495
 496#ifndef RCU_TREE_NONCORE
 497
 498/* Forward declarations for rcutree_plugin.h */
 499static void rcu_bootup_announce(void);
 500long rcu_batches_completed(void);
 501static void rcu_preempt_note_context_switch(int cpu);
 502static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
 503#ifdef CONFIG_HOTPLUG_CPU
 504static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp,
 505                                      unsigned long flags);
 506#endif /* #ifdef CONFIG_HOTPLUG_CPU */
 507static void rcu_print_detail_task_stall(struct rcu_state *rsp);
 508static int rcu_print_task_stall(struct rcu_node *rnp);
 509static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
 510#ifdef CONFIG_HOTPLUG_CPU
 511static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
 512                                     struct rcu_node *rnp,
 513                                     struct rcu_data *rdp);
 514#endif /* #ifdef CONFIG_HOTPLUG_CPU */
 515static void rcu_preempt_check_callbacks(int cpu);
 516void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
 517#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU)
 518static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
 519                               bool wake);
 520#endif /* #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) */
 521static void __init __rcu_init_preempt(void);
 522static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
 523static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
 524static void invoke_rcu_callbacks_kthread(void);
 525static bool rcu_is_callbacks_kthread(void);
 526#ifdef CONFIG_RCU_BOOST
 527static void rcu_preempt_do_callbacks(void);
 528static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
 529                                                 struct rcu_node *rnp);
 530#endif /* #ifdef CONFIG_RCU_BOOST */
 531static void __init rcu_spawn_boost_kthreads(void);
 532static void rcu_prepare_kthreads(int cpu);
 533static void rcu_cleanup_after_idle(int cpu);
 534static void rcu_prepare_for_idle(int cpu);
 535static void rcu_idle_count_callbacks_posted(void);
 536static void print_cpu_stall_info_begin(void);
 537static void print_cpu_stall_info(struct rcu_state *rsp, int cpu);
 538static void print_cpu_stall_info_end(void);
 539static void zero_cpu_stall_ticks(struct rcu_data *rdp);
 540static void increment_cpu_stall_ticks(void);
 541static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu);
 542static int rcu_nocb_needs_gp(struct rcu_state *rsp);
 543static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq);
 544static void rcu_nocb_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp);
 545static void rcu_init_one_nocb(struct rcu_node *rnp);
 546static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
 547                            bool lazy);
 548static bool rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp,
 549                                      struct rcu_data *rdp);
 550static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp);
 551static void rcu_spawn_all_nocb_kthreads(int cpu);
 552static void __init rcu_spawn_nocb_kthreads(void);
 553static void rcu_kick_nohz_cpu(int cpu);
 554static bool init_nocb_callback_list(struct rcu_data *rdp);
 555
 556#endif /* #ifndef RCU_TREE_NONCORE */
 557
 558#ifdef CONFIG_RCU_TRACE
 559#ifdef CONFIG_RCU_NOCB_CPU
 560/* Sum up queue lengths for tracing. */
 561static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll)
 562{
 563        *ql = atomic_long_read(&rdp->nocb_q_count) + rdp->nocb_p_count;
 564        *qll = atomic_long_read(&rdp->nocb_q_count_lazy) + rdp->nocb_p_count_lazy;
 565}
 566#else /* #ifdef CONFIG_RCU_NOCB_CPU */
 567static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll)
 568{
 569        *ql = 0;
 570        *qll = 0;
 571}
 572#endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
 573#endif /* #ifdef CONFIG_RCU_TRACE */
 574