include/linux/rcupdate.h

   1 /*
   2  * Read-Copy Update mechanism for mutual exclusion
   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 (C) IBM Corporation, 2001
  19  *
  20  * Author: Dipankar Sarma <dipankar@in.ibm.com>
  21  *
  22  * Based on the original work by Paul McKenney <paul.mckenney@us.ibm.com>
  23  * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
  24  * Papers:
  25  * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
  26  * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
  27  *
  28  * For detailed explanation of Read-Copy Update mechanism see -
  29  *              http://lse.sourceforge.net/locking/rcupdate.html
  30  *
  31  */
  32
  33 #ifndef __LINUX_RCUPDATE_H
  34 #define __LINUX_RCUPDATE_H
  35
  36 #ifdef __KERNEL__
  37
  38 #include <linux/cache.h>
  39 #include <linux/spinlock.h>
  40 #include <linux/threads.h>
  41 #include <linux/percpu.h>
  42 #include <linux/cpumask.h>
  43 #include <linux/seqlock.h>
  44
  45 /**
  46  * struct rcu_head - callback structure for use with RCU
  47  * @next: next update requests in a list
  48  * @func: actual update function to call after the grace period.
  49  */
  50 struct rcu_head {
  51         struct rcu_head *next;
  52         void (*func)(struct rcu_head *head);
  53 };
  54
  55 #define RCU_HEAD_INIT   { .next = NULL, .func = NULL }
  56 #define RCU_HEAD(head) struct rcu_head head = RCU_HEAD_INIT
  57 #define INIT_RCU_HEAD(ptr) do { \
  58        (ptr)->next = NULL; (ptr)->func = NULL; \
  59 } while (0)
  60
  61
  62
  63 /* Global control variables for rcupdate callback mechanism. */
  64 struct rcu_ctrlblk {
  65         long    cur;            /* Current batch number.                      */
  66         long    completed;      /* Number of the last completed batch         */
  67         int     next_pending;   /* Is the next batch already waiting?         */
  68 } ____cacheline_maxaligned_in_smp;
  69
  70 /* Is batch a before batch b ? */
  71 static inline int rcu_batch_before(long a, long b)
  72 {
  73         return (a - b) < 0;
  74 }
  75
  76 /* Is batch a after batch b ? */
  77 static inline int rcu_batch_after(long a, long b)
  78 {
  79         return (a - b) > 0;
  80 }
  81
  82 /*
  83  * Per-CPU data for Read-Copy UPdate.
  84  * nxtlist - new callbacks are added here
  85  * curlist - current batch for which quiescent cycle started if any
  86  */
  87 struct rcu_data {
  88         /* 1) quiescent state handling : */
  89         long            quiescbatch;     /* Batch # for grace period */
  90         int             passed_quiesc;   /* User-mode/idle loop etc. */
  91         int             qs_pending;      /* core waits for quiesc state */
  92
  93         /* 2) batch handling */
  94         long            batch;           /* Batch # for current RCU batch */
  95         struct rcu_head *nxtlist;
  96         struct rcu_head **nxttail;
  97         long            count; /* # of queued items */
  98         struct rcu_head *curlist;
  99         struct rcu_head **curtail;
 100         struct rcu_head *donelist;
 101         struct rcu_head **donetail;
 102         int cpu;
 103         struct rcu_head barrier;
 104 };
 105
 106 DECLARE_PER_CPU(struct rcu_data, rcu_data);
 107 DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
 108 extern struct rcu_ctrlblk rcu_ctrlblk;
 109 extern struct rcu_ctrlblk rcu_bh_ctrlblk;
 110
 111 /*
 112  * Increment the quiescent state counter.
 113  * The counter is a bit degenerated: We do not need to know
 114  * how many quiescent states passed, just if there was at least
 115  * one since the start of the grace period. Thus just a flag.
 116  */
 117 static inline void rcu_qsctr_inc(int cpu)
 118 {
 119         struct rcu_data *rdp = &per_cpu(rcu_data, cpu);
 120         rdp->passed_quiesc = 1;
 121 }
 122 static inline void rcu_bh_qsctr_inc(int cpu)
 123 {
 124         struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu);
 125         rdp->passed_quiesc = 1;
 126 }
 127
 128 static inline int __rcu_pending(struct rcu_ctrlblk *rcp,
 129                                                 struct rcu_data *rdp)
 130 {
 131         /* This cpu has pending rcu entries and the grace period
 132          * for them has completed.
 133          */
 134         if (rdp->curlist && !rcu_batch_before(rcp->completed, rdp->batch))
 135                 return 1;
 136
 137         /* This cpu has no pending entries, but there are new entries */
 138         if (!rdp->curlist && rdp->nxtlist)
 139                 return 1;
 140
 141         /* This cpu has finished callbacks to invoke */
 142         if (rdp->donelist)
 143                 return 1;
 144
 145         /* The rcu core waits for a quiescent state from the cpu */
 146         if (rdp->quiescbatch != rcp->cur || rdp->qs_pending)
 147                 return 1;
 148
 149         /* nothing to do */
 150         return 0;
 151 }
 152
 153 static inline int rcu_pending(int cpu)
 154 {
 155         return __rcu_pending(&rcu_ctrlblk, &per_cpu(rcu_data, cpu)) ||
 156                 __rcu_pending(&rcu_bh_ctrlblk, &per_cpu(rcu_bh_data, cpu));
 157 }
 158
 159 /**
 160  * rcu_read_lock - mark the beginning of an RCU read-side critical section.
 161  *
 162  * When synchronize_rcu() is invoked on one CPU while other CPUs
 163  * are within RCU read-side critical sections, then the
 164  * synchronize_rcu() is guaranteed to block until after all the other
 165  * CPUs exit their critical sections.  Similarly, if call_rcu() is invoked
 166  * on one CPU while other CPUs are within RCU read-side critical
 167  * sections, invocation of the corresponding RCU callback is deferred
 168  * until after the all the other CPUs exit their critical sections.
 169  *
 170  * Note, however, that RCU callbacks are permitted to run concurrently
 171  * with RCU read-side critical sections.  One way that this can happen
 172  * is via the following sequence of events: (1) CPU 0 enters an RCU
 173  * read-side critical section, (2) CPU 1 invokes call_rcu() to register
 174  * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
 175  * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
 176  * callback is invoked.  This is legal, because the RCU read-side critical
 177  * section that was running concurrently with the call_rcu() (and which
 178  * therefore might be referencing something that the corresponding RCU
 179  * callback would free up) has completed before the corresponding
 180  * RCU callback is invoked.
 181  *
 182  * RCU read-side critical sections may be nested.  Any deferred actions
 183  * will be deferred until the outermost RCU read-side critical section
 184  * completes.
 185  *
 186  * It is illegal to block while in an RCU read-side critical section.
 187  */
 188 #define rcu_read_lock()         preempt_disable()
 189
 190 /**
 191  * rcu_read_unlock - marks the end of an RCU read-side critical section.
 192  *
 193  * See rcu_read_lock() for more information.
 194  */
 195 #define rcu_read_unlock()       preempt_enable()
 196
 197 /*
 198  * So where is rcu_write_lock()?  It does not exist, as there is no
 199  * way for writers to lock out RCU readers.  This is a feature, not
 200  * a bug -- this property is what provides RCU's performance benefits.
 201  * Of course, writers must coordinate with each other.  The normal
 202  * spinlock primitives work well for this, but any other technique may be
 203  * used as well.  RCU does not care how the writers keep out of each
 204  * others' way, as long as they do so.
 205  */
 206
 207 /**
 208  * rcu_read_lock_bh - mark the beginning of a softirq-only RCU critical section
 209  *
 210  * This is equivalent of rcu_read_lock(), but to be used when updates
 211  * are being done using call_rcu_bh(). Since call_rcu_bh() callbacks
 212  * consider completion of a softirq handler to be a quiescent state,
 213  * a process in RCU read-side critical section must be protected by
 214  * disabling softirqs. Read-side critical sections in interrupt context
 215  * can use just rcu_read_lock().
 216  *
 217  */
 218 #define rcu_read_lock_bh()      local_bh_disable()
 219
 220 /*
 221  * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section
 222  *
 223  * See rcu_read_lock_bh() for more information.
 224  */
 225 #define rcu_read_unlock_bh()    local_bh_enable()
 226
 227 /**
 228  * rcu_dereference - fetch an RCU-protected pointer in an
 229  * RCU read-side critical section.  This pointer may later
 230  * be safely dereferenced.
 231  *
 232  * Inserts memory barriers on architectures that require them
 233  * (currently only the Alpha), and, more importantly, documents
 234  * exactly which pointers are protected by RCU.
 235  */
 236
 237 #define rcu_dereference(p)     ({ \
 238                                 typeof(p) _________p1 = p; \
 239                                 smp_read_barrier_depends(); \
 240                                 (_________p1); \
 241                                 })
 242
 243 /**
 244  * rcu_assign_pointer - assign (publicize) a pointer to a newly
 245  * initialized structure that will be dereferenced by RCU read-side
 246  * critical sections.  Returns the value assigned.
 247  *
 248  * Inserts memory barriers on architectures that require them
 249  * (pretty much all of them other than x86), and also prevents
 250  * the compiler from reordering the code that initializes the
 251  * structure after the pointer assignment.  More importantly, this
 252  * call documents which pointers will be dereferenced by RCU read-side
 253  * code.
 254  */
 255
 256 #define rcu_assign_pointer(p, v)        ({ \
 257                                                 smp_wmb(); \
 258                                                 (p) = (v); \
 259                                         })
 260
 261 /**
 262  * synchronize_sched - block until all CPUs have exited any non-preemptive
 263  * kernel code sequences.
 264  *
 265  * This means that all preempt_disable code sequences, including NMI and
 266  * hardware-interrupt handlers, in progress on entry will have completed
 267  * before this primitive returns.  However, this does not guarantee that
 268  * softirq handlers will have completed, since in some kernels
 269  *
 270  * This primitive provides the guarantees made by the (deprecated)
 271  * synchronize_kernel() API.  In contrast, synchronize_rcu() only
 272  * guarantees that rcu_read_lock() sections will have completed.
 273  */
 274 #define synchronize_sched() synchronize_rcu()
 275
 276 extern void rcu_init(void);
 277 extern void rcu_check_callbacks(int cpu, int user);
 278 extern void rcu_restart_cpu(int cpu);
 279 extern long rcu_batches_completed(void);
 280
 281 /* Exported interfaces */
 282 extern void FASTCALL(call_rcu(struct rcu_head *head,
 283                                 void (*func)(struct rcu_head *head)));
 284 extern void FASTCALL(call_rcu_bh(struct rcu_head *head,
 285                                 void (*func)(struct rcu_head *head)));
 286 extern __deprecated_for_modules void synchronize_kernel(void);
 287 extern void synchronize_rcu(void);
 288 void synchronize_idle(void);
 289 extern void rcu_barrier(void);
 290
 291 #endif /* __KERNEL__ */
 292 #endif /* __LINUX_RCUPDATE_H */