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 and CONFIG_RCU_FANOUT.
  33  * In theory, it should be possible to add more levels straightforwardly.
  34  * In practice, this did work well going from three levels to four.
  35  * Of course, your mileage may vary.
  36  */
  37 #define MAX_RCU_LVLS 4
  38 #if CONFIG_RCU_FANOUT > 16
  39 #define RCU_FANOUT_LEAF       16
  40 #else /* #if CONFIG_RCU_FANOUT > 16 */
  41 #define RCU_FANOUT_LEAF       (CONFIG_RCU_FANOUT)
  42 #endif /* #else #if CONFIG_RCU_FANOUT > 16 */
  43 #define RCU_FANOUT_1          (RCU_FANOUT_LEAF)
  44 #define RCU_FANOUT_2          (RCU_FANOUT_1 * CONFIG_RCU_FANOUT)
  45 #define RCU_FANOUT_3          (RCU_FANOUT_2 * CONFIG_RCU_FANOUT)
  46 #define RCU_FANOUT_4          (RCU_FANOUT_3 * CONFIG_RCU_FANOUT)
  47
  48 #if NR_CPUS <= RCU_FANOUT_1
  49 #  define NUM_RCU_LVLS        1
  50 #  define NUM_RCU_LVL_0       1
  51 #  define NUM_RCU_LVL_1       (NR_CPUS)
  52 #  define NUM_RCU_LVL_2       0
  53 #  define NUM_RCU_LVL_3       0
  54 #  define NUM_RCU_LVL_4       0
  55 #elif NR_CPUS <= RCU_FANOUT_2
  56 #  define NUM_RCU_LVLS        2
  57 #  define NUM_RCU_LVL_0       1
  58 #  define NUM_RCU_LVL_1       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
  59 #  define NUM_RCU_LVL_2       (NR_CPUS)
  60 #  define NUM_RCU_LVL_3       0
  61 #  define NUM_RCU_LVL_4       0
  62 #elif NR_CPUS <= RCU_FANOUT_3
  63 #  define NUM_RCU_LVLS        3
  64 #  define NUM_RCU_LVL_0       1
  65 #  define NUM_RCU_LVL_1       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
  66 #  define NUM_RCU_LVL_2       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
  67 #  define NUM_RCU_LVL_3       (NR_CPUS)
  68 #  define NUM_RCU_LVL_4       0
  69 #elif NR_CPUS <= RCU_FANOUT_4
  70 #  define NUM_RCU_LVLS        4
  71 #  define NUM_RCU_LVL_0       1
  72 #  define NUM_RCU_LVL_1       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3)
  73 #  define NUM_RCU_LVL_2       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
  74 #  define NUM_RCU_LVL_3       DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
  75 #  define NUM_RCU_LVL_4       (NR_CPUS)
  76 #else
  77 # error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
  78 #endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
  79
  80 #define RCU_SUM (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3 + NUM_RCU_LVL_4)
  81 #define NUM_RCU_NODES (RCU_SUM - NR_CPUS)
  82
  83 /*
  84  * Dynticks per-CPU state.
  85  */
  86 struct rcu_dynticks {
  87         int dynticks_nesting;   /* Track irq/process nesting level. */
  88         int dynticks_nmi_nesting; /* Track NMI nesting level. */
  89         atomic_t dynticks;      /* Even value for dynticks-idle, else odd. */
  90 };
  91
  92 /*
  93  * Definition for node within the RCU grace-period-detection hierarchy.
  94  */
  95 struct rcu_node {
  96         raw_spinlock_t lock;    /* Root rcu_node's lock protects some */
  97                                 /*  rcu_state fields as well as following. */
  98         unsigned long gpnum;    /* Current grace period for this node. */
  99                                 /*  This will either be equal to or one */
 100                                 /*  behind the root rcu_node's gpnum. */
 101         unsigned long completed; /* Last GP completed for this node. */
 102                                 /*  This will either be equal to or one */
 103                                 /*  behind the root rcu_node's gpnum. */
 104         unsigned long qsmask;   /* CPUs or groups that need to switch in */
 105                                 /*  order for current grace period to proceed.*/
 106                                 /*  In leaf rcu_node, each bit corresponds to */
 107                                 /*  an rcu_data structure, otherwise, each */
 108                                 /*  bit corresponds to a child rcu_node */
 109                                 /*  structure. */
 110         unsigned long expmask;  /* Groups that have ->blkd_tasks */
 111                                 /*  elements that need to drain to allow the */
 112                                 /*  current expedited grace period to */
 113                                 /*  complete (only for TREE_PREEMPT_RCU). */
 114         unsigned long wakemask; /* CPUs whose kthread needs to be awakened. */
 115         unsigned long qsmaskinit;
 116                                 /* Per-GP initial value for qsmask & expmask. */
 117         unsigned long grpmask;  /* Mask to apply to parent qsmask. */
 118                                 /*  Only one bit will be set in this mask. */
 119         int     grplo;          /* lowest-numbered CPU or group here. */
 120         int     grphi;          /* highest-numbered CPU or group here. */
 121         u8      grpnum;         /* CPU/group number for next level up. */
 122         u8      level;          /* root is at level 0. */
 123         struct rcu_node *parent;
 124         struct list_head blkd_tasks;
 125                                 /* Tasks blocked in RCU read-side critical */
 126                                 /*  section.  Tasks are placed at the head */
 127                                 /*  of this list and age towards the tail. */
 128         struct list_head *gp_tasks;
 129                                 /* Pointer to the first task blocking the */
 130                                 /*  current grace period, or NULL if there */
 131                                 /*  is no such task. */
 132         struct list_head *exp_tasks;
 133                                 /* Pointer to the first task blocking the */
 134                                 /*  current expedited grace period, or NULL */
 135                                 /*  if there is no such task.  If there */
 136                                 /*  is no current expedited grace period, */
 137                                 /*  then there can cannot be any such task. */
 138 #ifdef CONFIG_RCU_BOOST
 139         struct list_head *boost_tasks;
 140                                 /* Pointer to first task that needs to be */
 141                                 /*  priority boosted, or NULL if no priority */
 142                                 /*  boosting is needed for this rcu_node */
 143                                 /*  structure.  If there are no tasks */
 144                                 /*  queued on this rcu_node structure that */
 145                                 /*  are blocking the current grace period, */
 146                                 /*  there can be no such task. */
 147         unsigned long boost_time;
 148                                 /* When to start boosting (jiffies). */
 149         struct task_struct *boost_kthread_task;
 150                                 /* kthread that takes care of priority */
 151                                 /*  boosting for this rcu_node structure. */
 152         wait_queue_head_t boost_wq;
 153                                 /* Wait queue on which to park the boost */
 154                                 /*  kthread. */
 155 #endif /* #ifdef CONFIG_RCU_BOOST */
 156         struct task_struct *node_kthread_task;
 157                                 /* kthread that takes care of this rcu_node */
 158                                 /*  structure, for example, awakening the */
 159                                 /*  per-CPU kthreads as needed. */
 160         wait_queue_head_t node_wq;
 161                                 /* Wait queue on which to park the per-node */
 162                                 /*  kthread. */
 163 } ____cacheline_internodealigned_in_smp;
 164
 165 /*
 166  * Do a full breadth-first scan of the rcu_node structures for the
 167  * specified rcu_state structure.
 168  */
 169 #define rcu_for_each_node_breadth_first(rsp, rnp) \
 170         for ((rnp) = &(rsp)->node[0]; \
 171              (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
 172
 173 /*
 174  * Do a breadth-first scan of the non-leaf rcu_node structures for the
 175  * specified rcu_state structure.  Note that if there is a singleton
 176  * rcu_node tree with but one rcu_node structure, this loop is a no-op.
 177  */
 178 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
 179         for ((rnp) = &(rsp)->node[0]; \
 180              (rnp) < (rsp)->level[NUM_RCU_LVLS - 1]; (rnp)++)
 181
 182 /*
 183  * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
 184  * structure.  Note that if there is a singleton rcu_node tree with but
 185  * one rcu_node structure, this loop -will- visit the rcu_node structure.
 186  * It is still a leaf node, even if it is also the root node.
 187  */
 188 #define rcu_for_each_leaf_node(rsp, rnp) \
 189         for ((rnp) = (rsp)->level[NUM_RCU_LVLS - 1]; \
 190              (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
 191
 192 /* Index values for nxttail array in struct rcu_data. */
 193 #define RCU_DONE_TAIL           0       /* Also RCU_WAIT head. */
 194 #define RCU_WAIT_TAIL           1       /* Also RCU_NEXT_READY head. */
 195 #define RCU_NEXT_READY_TAIL     2       /* Also RCU_NEXT head. */
 196 #define RCU_NEXT_TAIL           3
 197 #define RCU_NEXT_SIZE           4
 198
 199 /* Per-CPU data for read-copy update. */
 200 struct rcu_data {
 201         /* 1) quiescent-state and grace-period handling : */
 202         unsigned long   completed;      /* Track rsp->completed gp number */
 203                                         /*  in order to detect GP end. */
 204         unsigned long   gpnum;          /* Highest gp number that this CPU */
 205                                         /*  is aware of having started. */
 206         unsigned long   passed_quiesc_completed;
 207                                         /* Value of completed at time of qs. */
 208         bool            passed_quiesc;  /* User-mode/idle loop etc. */
 209         bool            qs_pending;     /* Core waits for quiesc state. */
 210         bool            beenonline;     /* CPU online at least once. */
 211         bool            preemptable;    /* Preemptable RCU? */
 212         struct rcu_node *mynode;        /* This CPU's leaf of hierarchy */
 213         unsigned long grpmask;          /* Mask to apply to leaf qsmask. */
 214
 215         /* 2) batch handling */
 216         /*
 217          * If nxtlist is not NULL, it is partitioned as follows.
 218          * Any of the partitions might be empty, in which case the
 219          * pointer to that partition will be equal to the pointer for
 220          * the following partition.  When the list is empty, all of
 221          * the nxttail elements point to the ->nxtlist pointer itself,
 222          * which in that case is NULL.
 223          *
 224          * [nxtlist, *nxttail[RCU_DONE_TAIL]):
 225          *      Entries that batch # <= ->completed
 226          *      The grace period for these entries has completed, and
 227          *      the other grace-period-completed entries may be moved
 228          *      here temporarily in rcu_process_callbacks().
 229          * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
 230          *      Entries that batch # <= ->completed - 1: waiting for current GP
 231          * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
 232          *      Entries known to have arrived before current GP ended
 233          * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
 234          *      Entries that might have arrived after current GP ended
 235          *      Note that the value of *nxttail[RCU_NEXT_TAIL] will
 236          *      always be NULL, as this is the end of the list.
 237          */
 238         struct rcu_head *nxtlist;
 239         struct rcu_head **nxttail[RCU_NEXT_SIZE];
 240         long            qlen;           /* # of queued callbacks */
 241         long            qlen_last_fqs_check;
 242                                         /* qlen at last check for QS forcing */
 243         unsigned long   n_cbs_invoked;  /* count of RCU cbs invoked. */
 244         unsigned long   n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
 245         unsigned long   n_cbs_adopted;  /* RCU cbs adopted from dying CPU */
 246         unsigned long   n_force_qs_snap;
 247                                         /* did other CPU force QS recently? */
 248         long            blimit;         /* Upper limit on a processed batch */
 249
 250 #ifdef CONFIG_NO_HZ
 251         /* 3) dynticks interface. */
 252         struct rcu_dynticks *dynticks;  /* Shared per-CPU dynticks state. */
 253         int dynticks_snap;              /* Per-GP tracking for dynticks. */
 254 #endif /* #ifdef CONFIG_NO_HZ */
 255
 256         /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
 257 #ifdef CONFIG_NO_HZ
 258         unsigned long dynticks_fqs;     /* Kicked due to dynticks idle. */
 259 #endif /* #ifdef CONFIG_NO_HZ */
 260         unsigned long offline_fqs;      /* Kicked due to being offline. */
 261         unsigned long resched_ipi;      /* Sent a resched IPI. */
 262
 263         /* 5) __rcu_pending() statistics. */
 264         unsigned long n_rcu_pending;    /* rcu_pending() calls since boot. */
 265         unsigned long n_rp_qs_pending;
 266         unsigned long n_rp_report_qs;
 267         unsigned long n_rp_cb_ready;
 268         unsigned long n_rp_cpu_needs_gp;
 269         unsigned long n_rp_gp_completed;
 270         unsigned long n_rp_gp_started;
 271         unsigned long n_rp_need_fqs;
 272         unsigned long n_rp_need_nothing;
 273
 274         int cpu;
 275 };
 276
 277 /* Values for signaled field in struct rcu_state. */
 278 #define RCU_GP_IDLE             0       /* No grace period in progress. */
 279 #define RCU_GP_INIT             1       /* Grace period being initialized. */
 280 #define RCU_SAVE_DYNTICK        2       /* Need to scan dyntick state. */
 281 #define RCU_FORCE_QS            3       /* Need to force quiescent state. */
 282 #ifdef CONFIG_NO_HZ
 283 #define RCU_SIGNAL_INIT         RCU_SAVE_DYNTICK
 284 #else /* #ifdef CONFIG_NO_HZ */
 285 #define RCU_SIGNAL_INIT         RCU_FORCE_QS
 286 #endif /* #else #ifdef CONFIG_NO_HZ */
 287
 288 #define RCU_JIFFIES_TILL_FORCE_QS        3      /* for rsp->jiffies_force_qs */
 289
 290 #ifdef CONFIG_PROVE_RCU
 291 #define RCU_STALL_DELAY_DELTA          (5 * HZ)
 292 #else
 293 #define RCU_STALL_DELAY_DELTA          0
 294 #endif
 295
 296 #define RCU_SECONDS_TILL_STALL_CHECK   (CONFIG_RCU_CPU_STALL_TIMEOUT * HZ + \
 297                                         RCU_STALL_DELAY_DELTA)
 298                                                 /* for rsp->jiffies_stall */
 299 #define RCU_SECONDS_TILL_STALL_RECHECK (3 * RCU_SECONDS_TILL_STALL_CHECK + 30)
 300                                                 /* for rsp->jiffies_stall */
 301 #define RCU_STALL_RAT_DELAY             2       /* Allow other CPUs time */
 302                                                 /*  to take at least one */
 303                                                 /*  scheduling clock irq */
 304                                                 /*  before ratting on them. */
 305
 306
 307 /*
 308  * RCU global state, including node hierarchy.  This hierarchy is
 309  * represented in "heap" form in a dense array.  The root (first level)
 310  * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
 311  * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
 312  * and the third level in ->node[m+1] and following (->node[m+1] referenced
 313  * by ->level[2]).  The number of levels is determined by the number of
 314  * CPUs and by CONFIG_RCU_FANOUT.  Small systems will have a "hierarchy"
 315  * consisting of a single rcu_node.
 316  */
 317 struct rcu_state {
 318         struct rcu_node node[NUM_RCU_NODES];    /* Hierarchy. */
 319         struct rcu_node *level[NUM_RCU_LVLS];   /* Hierarchy levels. */
 320         u32 levelcnt[MAX_RCU_LVLS + 1];         /* # nodes in each level. */
 321         u8 levelspread[NUM_RCU_LVLS];           /* kids/node in each level. */
 322         struct rcu_data __percpu *rda;          /* pointer of percu rcu_data. */
 323
 324         /* The following fields are guarded by the root rcu_node's lock. */
 325
 326         u8      signaled ____cacheline_internodealigned_in_smp;
 327                                                 /* Force QS state. */
 328         u8      fqs_active;                     /* force_quiescent_state() */
 329                                                 /*  is running. */
 330         u8      fqs_need_gp;                    /* A CPU was prevented from */
 331                                                 /*  starting a new grace */
 332                                                 /*  period because */
 333                                                 /*  force_quiescent_state() */
 334                                                 /*  was running. */
 335         unsigned long gpnum;                    /* Current gp number. */
 336         unsigned long completed;                /* # of last completed gp. */
 337
 338         /* End of fields guarded by root rcu_node's lock. */
 339
 340         raw_spinlock_t onofflock;               /* exclude on/offline and */
 341                                                 /*  starting new GP. */
 342         raw_spinlock_t fqslock;                 /* Only one task forcing */
 343                                                 /*  quiescent states. */
 344         unsigned long jiffies_force_qs;         /* Time at which to invoke */
 345                                                 /*  force_quiescent_state(). */
 346         unsigned long n_force_qs;               /* Number of calls to */
 347                                                 /*  force_quiescent_state(). */
 348         unsigned long n_force_qs_lh;            /* ~Number of calls leaving */
 349                                                 /*  due to lock unavailable. */
 350         unsigned long n_force_qs_ngp;           /* Number of calls leaving */
 351                                                 /*  due to no GP active. */
 352         unsigned long gp_start;                 /* Time at which GP started, */
 353                                                 /*  but in jiffies. */
 354         unsigned long jiffies_stall;            /* Time at which to check */
 355                                                 /*  for CPU stalls. */
 356         char *name;                             /* Name of structure. */
 357 };
 358
 359 /* Return values for rcu_preempt_offline_tasks(). */
 360
 361 #define RCU_OFL_TASKS_NORM_GP   0x1             /* Tasks blocking normal */
 362                                                 /*  GP were moved to root. */
 363 #define RCU_OFL_TASKS_EXP_GP    0x2             /* Tasks blocking expedited */
 364                                                 /*  GP were moved to root. */
 365
 366 /*
 367  * RCU implementation internal declarations:
 368  */
 369 extern struct rcu_state rcu_sched_state;
 370 DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
 371
 372 extern struct rcu_state rcu_bh_state;
 373 DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
 374
 375 #ifdef CONFIG_TREE_PREEMPT_RCU
 376 extern struct rcu_state rcu_preempt_state;
 377 DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
 378 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
 379
 380 #ifndef RCU_TREE_NONCORE
 381
 382 /* Forward declarations for rcutree_plugin.h */
 383 static void rcu_bootup_announce(void);
 384 long rcu_batches_completed(void);
 385 static void rcu_preempt_note_context_switch(int cpu);
 386 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
 387 #ifdef CONFIG_HOTPLUG_CPU
 388 static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp,
 389                                       unsigned long flags);
 390 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
 391 static void rcu_print_detail_task_stall(struct rcu_state *rsp);
 392 static void rcu_print_task_stall(struct rcu_node *rnp);
 393 static void rcu_preempt_stall_reset(void);
 394 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
 395 #ifdef CONFIG_HOTPLUG_CPU
 396 static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
 397                                      struct rcu_node *rnp,
 398                                      struct rcu_data *rdp);
 399 static void rcu_preempt_offline_cpu(int cpu);
 400 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
 401 static void rcu_preempt_check_callbacks(int cpu);
 402 static void rcu_preempt_process_callbacks(void);
 403 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
 404 #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU)
 405 static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp);
 406 #endif /* #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) */
 407 static int rcu_preempt_pending(int cpu);
 408 static int rcu_preempt_needs_cpu(int cpu);
 409 static void __cpuinit rcu_preempt_init_percpu_data(int cpu);
 410 static void rcu_preempt_send_cbs_to_online(void);
 411 static void __init __rcu_init_preempt(void);
 412 static void rcu_needs_cpu_flush(void);
 413 static void __init rcu_init_boost_waitqueue(struct rcu_node *rnp);
 414 static void rcu_initiate_boost(struct rcu_node *rnp);
 415 static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp,
 416                                           cpumask_var_t cm);
 417 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
 418 static int __cpuinit rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
 419                                                  struct rcu_node *rnp,
 420                                                  int rnp_index);
 421 #ifdef CONFIG_HOTPLUG_CPU
 422 static void rcu_stop_boost_kthread(struct rcu_node *rnp);
 423 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
 424
 425 #endif /* #ifndef RCU_TREE_NONCORE */