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