include/linux/rcutree.h

   1 /*
   2  * Read-Copy Update mechanism for mutual exclusion (tree-based version)
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  17  *
  18  * Copyright IBM Corporation, 2008
  19  *
  20  * Author: Dipankar Sarma <dipankar@in.ibm.com>
  21  *         Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical algorithm
  22  *
  23  * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
  24  * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
  25  *
  26  * For detailed explanation of Read-Copy Update mechanism see -
  27  *      Documentation/RCU
  28  */
  29
  30 #ifndef __LINUX_RCUTREE_H
  31 #define __LINUX_RCUTREE_H
  32
  33 #include <linux/cache.h>
  34 #include <linux/spinlock.h>
  35 #include <linux/threads.h>
  36 #include <linux/percpu.h>
  37 #include <linux/cpumask.h>
  38 #include <linux/seqlock.h>
  39
  40 /*
  41  * Define shape of hierarchy based on NR_CPUS and CONFIG_RCU_FANOUT.
  42  * In theory, it should be possible to add more levels straightforwardly.
  43  * In practice, this has not been tested, so there is probably some
  44  * bug somewhere.
  45  */
  46 #define MAX_RCU_LVLS 3
  47 #define RCU_FANOUT            (CONFIG_RCU_FANOUT)
  48 #define RCU_FANOUT_SQ         (RCU_FANOUT * RCU_FANOUT)
  49 #define RCU_FANOUT_CUBE       (RCU_FANOUT_SQ * RCU_FANOUT)
  50
  51 #if NR_CPUS <= RCU_FANOUT
  52 #  define NUM_RCU_LVLS        1
  53 #  define NUM_RCU_LVL_0       1
  54 #  define NUM_RCU_LVL_1       (NR_CPUS)
  55 #  define NUM_RCU_LVL_2       0
  56 #  define NUM_RCU_LVL_3       0
  57 #elif NR_CPUS <= RCU_FANOUT_SQ
  58 #  define NUM_RCU_LVLS        2
  59 #  define NUM_RCU_LVL_0       1
  60 #  define NUM_RCU_LVL_1       (((NR_CPUS) + RCU_FANOUT - 1) / RCU_FANOUT)
  61 #  define NUM_RCU_LVL_2       (NR_CPUS)
  62 #  define NUM_RCU_LVL_3       0
  63 #elif NR_CPUS <= RCU_FANOUT_CUBE
  64 #  define NUM_RCU_LVLS        3
  65 #  define NUM_RCU_LVL_0       1
  66 #  define NUM_RCU_LVL_1       (((NR_CPUS) + RCU_FANOUT_SQ - 1) / RCU_FANOUT_SQ)
  67 #  define NUM_RCU_LVL_2       (((NR_CPUS) + (RCU_FANOUT) - 1) / (RCU_FANOUT))
  68 #  define NUM_RCU_LVL_3       NR_CPUS
  69 #else
  70 # error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
  71 #endif /* #if (NR_CPUS) <= RCU_FANOUT */
  72
  73 #define RCU_SUM (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3)
  74 #define NUM_RCU_NODES (RCU_SUM - NR_CPUS)
  75
  76 /*
  77  * Dynticks per-CPU state.
  78  */
  79 struct rcu_dynticks {
  80         int dynticks_nesting;   /* Track nesting level, sort of. */
  81         int dynticks;           /* Even value for dynticks-idle, else odd. */
  82         int dynticks_nmi;       /* Even value for either dynticks-idle or */
  83                                 /*  not in nmi handler, else odd.  So this */
  84                                 /*  remains even for nmi from irq handler. */
  85 };
  86
  87 /*
  88  * Definition for node within the RCU grace-period-detection hierarchy.
  89  */
  90 struct rcu_node {
  91         spinlock_t lock;
  92         unsigned long qsmask;   /* CPUs or groups that need to switch in */
  93                                 /*  order for current grace period to proceed.*/
  94         unsigned long qsmaskinit;
  95                                 /* Per-GP initialization for qsmask. */
  96         unsigned long grpmask;  /* Mask to apply to parent qsmask. */
  97         int     grplo;          /* lowest-numbered CPU or group here. */
  98         int     grphi;          /* highest-numbered CPU or group here. */
  99         u8      grpnum;         /* CPU/group number for next level up. */
 100         u8      level;          /* root is at level 0. */
 101         struct rcu_node *parent;
 102 } ____cacheline_internodealigned_in_smp;
 103
 104 /* Index values for nxttail array in struct rcu_data. */
 105 #define RCU_DONE_TAIL           0       /* Also RCU_WAIT head. */
 106 #define RCU_WAIT_TAIL           1       /* Also RCU_NEXT_READY head. */
 107 #define RCU_NEXT_READY_TAIL     2       /* Also RCU_NEXT head. */
 108 #define RCU_NEXT_TAIL           3
 109 #define RCU_NEXT_SIZE           4
 110
 111 /* Per-CPU data for read-copy update. */
 112 struct rcu_data {
 113         /* 1) quiescent-state and grace-period handling : */
 114         long            completed;      /* Track rsp->completed gp number */
 115                                         /*  in order to detect GP end. */
 116         long            gpnum;          /* Highest gp number that this CPU */
 117                                         /*  is aware of having started. */
 118         long            passed_quiesc_completed;
 119                                         /* Value of completed at time of qs. */
 120         bool            passed_quiesc;  /* User-mode/idle loop etc. */
 121         bool            qs_pending;     /* Core waits for quiesc state. */
 122         bool            beenonline;     /* CPU online at least once. */
 123         struct rcu_node *mynode;        /* This CPU's leaf of hierarchy */
 124         unsigned long grpmask;          /* Mask to apply to leaf qsmask. */
 125
 126         /* 2) batch handling */
 127         /*
 128          * If nxtlist is not NULL, it is partitioned as follows.
 129          * Any of the partitions might be empty, in which case the
 130          * pointer to that partition will be equal to the pointer for
 131          * the following partition.  When the list is empty, all of
 132          * the nxttail elements point to nxtlist, which is NULL.
 133          *
 134          * [*nxttail[RCU_NEXT_READY_TAIL], NULL = *nxttail[RCU_NEXT_TAIL]):
 135          *      Entries that might have arrived after current GP ended
 136          * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
 137          *      Entries known to have arrived before current GP ended
 138          * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
 139          *      Entries that batch # <= ->completed - 1: waiting for current GP
 140          * [nxtlist, *nxttail[RCU_DONE_TAIL]):
 141          *      Entries that batch # <= ->completed
 142          *      The grace period for these entries has completed, and
 143          *      the other grace-period-completed entries may be moved
 144          *      here temporarily in rcu_process_callbacks().
 145          */
 146         struct rcu_head *nxtlist;
 147         struct rcu_head **nxttail[RCU_NEXT_SIZE];
 148         long            qlen;           /* # of queued callbacks */
 149         long            blimit;         /* Upper limit on a processed batch */
 150
 151 #ifdef CONFIG_NO_HZ
 152         /* 3) dynticks interface. */
 153         struct rcu_dynticks *dynticks;  /* Shared per-CPU dynticks state. */
 154         int dynticks_snap;              /* Per-GP tracking for dynticks. */
 155         int dynticks_nmi_snap;          /* Per-GP tracking for dynticks_nmi. */
 156 #endif /* #ifdef CONFIG_NO_HZ */
 157
 158         /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
 159 #ifdef CONFIG_NO_HZ
 160         unsigned long dynticks_fqs;     /* Kicked due to dynticks idle. */
 161 #endif /* #ifdef CONFIG_NO_HZ */
 162         unsigned long offline_fqs;      /* Kicked due to being offline. */
 163         unsigned long resched_ipi;      /* Sent a resched IPI. */
 164
 165         /* 5) state to allow this CPU to force_quiescent_state on others */
 166         long n_rcu_pending;             /* rcu_pending() calls since boot. */
 167         long n_rcu_pending_force_qs;    /* when to force quiescent states. */
 168
 169         int cpu;
 170 };
 171
 172 /* Values for signaled field in struct rcu_state. */
 173 #define RCU_GP_INIT             0       /* Grace period being initialized. */
 174 #define RCU_SAVE_DYNTICK        1       /* Need to scan dyntick state. */
 175 #define RCU_FORCE_QS            2       /* Need to force quiescent state. */
 176 #ifdef CONFIG_NO_HZ
 177 #define RCU_SIGNAL_INIT         RCU_SAVE_DYNTICK
 178 #else /* #ifdef CONFIG_NO_HZ */
 179 #define RCU_SIGNAL_INIT         RCU_FORCE_QS
 180 #endif /* #else #ifdef CONFIG_NO_HZ */
 181
 182 #define RCU_JIFFIES_TILL_FORCE_QS        3      /* for rsp->jiffies_force_qs */
 183 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
 184 #define RCU_SECONDS_TILL_STALL_CHECK   (10 * HZ)  /* for rsp->jiffies_stall */
 185 #define RCU_SECONDS_TILL_STALL_RECHECK (30 * HZ)  /* for rsp->jiffies_stall */
 186 #define RCU_STALL_RAT_DELAY             2         /* Allow other CPUs time */
 187                                                   /*  to take at least one */
 188                                                   /*  scheduling clock irq */
 189                                                   /*  before ratting on them. */
 190
 191 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
 192
 193 /*
 194  * RCU global state, including node hierarchy.  This hierarchy is
 195  * represented in "heap" form in a dense array.  The root (first level)
 196  * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
 197  * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
 198  * and the third level in ->node[m+1] and following (->node[m+1] referenced
 199  * by ->level[2]).  The number of levels is determined by the number of
 200  * CPUs and by CONFIG_RCU_FANOUT.  Small systems will have a "hierarchy"
 201  * consisting of a single rcu_node.
 202  */
 203 struct rcu_state {
 204         struct rcu_node node[NUM_RCU_NODES];    /* Hierarchy. */
 205         struct rcu_node *level[NUM_RCU_LVLS];   /* Hierarchy levels. */
 206         u32 levelcnt[MAX_RCU_LVLS + 1];         /* # nodes in each level. */
 207         u8 levelspread[NUM_RCU_LVLS];           /* kids/node in each level. */
 208         struct rcu_data *rda[NR_CPUS];          /* array of rdp pointers. */
 209
 210         /* The following fields are guarded by the root rcu_node's lock. */
 211
 212         u8      signaled ____cacheline_internodealigned_in_smp;
 213                                                 /* Force QS state. */
 214         long    gpnum;                          /* Current gp number. */
 215         long    completed;                      /* # of last completed gp. */
 216         spinlock_t onofflock;                   /* exclude on/offline and */
 217                                                 /*  starting new GP. */
 218         spinlock_t fqslock;                     /* Only one task forcing */
 219                                                 /*  quiescent states. */
 220         unsigned long jiffies_force_qs;         /* Time at which to invoke */
 221                                                 /*  force_quiescent_state(). */
 222         unsigned long n_force_qs;               /* Number of calls to */
 223                                                 /*  force_quiescent_state(). */
 224         unsigned long n_force_qs_lh;            /* ~Number of calls leaving */
 225                                                 /*  due to lock unavailable. */
 226         unsigned long n_force_qs_ngp;           /* Number of calls leaving */
 227                                                 /*  due to no GP active. */
 228 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
 229         unsigned long gp_start;                 /* Time at which GP started, */
 230                                                 /*  but in jiffies. */
 231         unsigned long jiffies_stall;            /* Time at which to check */
 232                                                 /*  for CPU stalls. */
 233 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
 234 #ifdef CONFIG_NO_HZ
 235         long dynticks_completed;                /* Value of completed @ snap. */
 236 #endif /* #ifdef CONFIG_NO_HZ */
 237 };
 238
 239 extern struct rcu_state rcu_state;
 240 DECLARE_PER_CPU(struct rcu_data, rcu_data);
 241
 242 extern struct rcu_state rcu_bh_state;
 243 DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
 244
 245 /*
 246  * Increment the quiescent state counter.
 247  * The counter is a bit degenerated: We do not need to know
 248  * how many quiescent states passed, just if there was at least
 249  * one since the start of the grace period. Thus just a flag.
 250  */
 251 static inline void rcu_qsctr_inc(int cpu)
 252 {
 253         struct rcu_data *rdp = &per_cpu(rcu_data, cpu);
 254         rdp->passed_quiesc = 1;
 255         rdp->passed_quiesc_completed = rdp->completed;
 256 }
 257 static inline void rcu_bh_qsctr_inc(int cpu)
 258 {
 259         struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu);
 260         rdp->passed_quiesc = 1;
 261         rdp->passed_quiesc_completed = rdp->completed;
 262 }
 263
 264 extern int rcu_pending(int cpu);
 265 extern int rcu_needs_cpu(int cpu);
 266
 267 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 268 extern struct lockdep_map rcu_lock_map;
 269 # define rcu_read_acquire()     \
 270                         lock_acquire(&rcu_lock_map, 0, 0, 2, 1, NULL, _THIS_IP_)
 271 # define rcu_read_release()     lock_release(&rcu_lock_map, 1, _THIS_IP_)
 272 #else
 273 # define rcu_read_acquire()     do { } while (0)
 274 # define rcu_read_release()     do { } while (0)
 275 #endif
 276
 277 static inline void __rcu_read_lock(void)
 278 {
 279         preempt_disable();
 280         __acquire(RCU);
 281         rcu_read_acquire();
 282 }
 283 static inline void __rcu_read_unlock(void)
 284 {
 285         rcu_read_release();
 286         __release(RCU);
 287         preempt_enable();
 288 }
 289 static inline void __rcu_read_lock_bh(void)
 290 {
 291         local_bh_disable();
 292         __acquire(RCU_BH);
 293         rcu_read_acquire();
 294 }
 295 static inline void __rcu_read_unlock_bh(void)
 296 {
 297         rcu_read_release();
 298         __release(RCU_BH);
 299         local_bh_enable();
 300 }
 301
 302 #define __synchronize_sched() synchronize_rcu()
 303
 304 #define call_rcu_sched(head, func) call_rcu(head, func)
 305
 306 static inline void rcu_init_sched(void)
 307 {
 308 }
 309
 310 extern void __rcu_init(void);
 311 extern void rcu_check_callbacks(int cpu, int user);
 312 extern void rcu_restart_cpu(int cpu);
 313
 314 extern long rcu_batches_completed(void);
 315 extern long rcu_batches_completed_bh(void);
 316
 317 #ifdef CONFIG_NO_HZ
 318 void rcu_enter_nohz(void);
 319 void rcu_exit_nohz(void);
 320 #else /* CONFIG_NO_HZ */
 321 static inline void rcu_enter_nohz(void)
 322 {
 323 }
 324 static inline void rcu_exit_nohz(void)
 325 {
 326 }
 327 #endif /* CONFIG_NO_HZ */
 328
 329 /* A context switch is a grace period for rcutree. */
 330 static inline int rcu_blocking_is_gp(void)
 331 {
 332         return num_online_cpus() == 1;
 333 }
 334
 335 #endif /* __LINUX_RCUTREE_H */