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