kernel/srcu.c

   1 /*
   2  * Sleepable 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, 2006
  19  *
  20  * Author: Paul McKenney <paulmck@us.ibm.com>
  21  *
  22  * For detailed explanation of Read-Copy Update mechanism see -
  23  *              Documentation/RCU/ *.txt
  24  *
  25  */
  26
  27 #include <linux/export.h>
  28 #include <linux/mutex.h>
  29 #include <linux/percpu.h>
  30 #include <linux/preempt.h>
  31 #include <linux/rcupdate.h>
  32 #include <linux/sched.h>
  33 #include <linux/smp.h>
  34 #include <linux/delay.h>
  35 #include <linux/srcu.h>
  36
  37 static int init_srcu_struct_fields(struct srcu_struct *sp)
  38 {
  39         sp->completed = 0;
  40         mutex_init(&sp->mutex);
  41         sp->per_cpu_ref = alloc_percpu(struct srcu_struct_array);
  42         return sp->per_cpu_ref ? 0 : -ENOMEM;
  43 }
  44
  45 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  46
  47 int __init_srcu_struct(struct srcu_struct *sp, const char *name,
  48                        struct lock_class_key *key)
  49 {
  50         /* Don't re-initialize a lock while it is held. */
  51         debug_check_no_locks_freed((void *)sp, sizeof(*sp));
  52         lockdep_init_map(&sp->dep_map, name, key, 0);
  53         return init_srcu_struct_fields(sp);
  54 }
  55 EXPORT_SYMBOL_GPL(__init_srcu_struct);
  56
  57 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
  58
  59 /**
  60  * init_srcu_struct - initialize a sleep-RCU structure
  61  * @sp: structure to initialize.
  62  *
  63  * Must invoke this on a given srcu_struct before passing that srcu_struct
  64  * to any other function.  Each srcu_struct represents a separate domain
  65  * of SRCU protection.
  66  */
  67 int init_srcu_struct(struct srcu_struct *sp)
  68 {
  69         return init_srcu_struct_fields(sp);
  70 }
  71 EXPORT_SYMBOL_GPL(init_srcu_struct);
  72
  73 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
  74
  75 /*
  76  * srcu_readers_active_idx -- returns approximate number of readers
  77  *      active on the specified rank of per-CPU counters.
  78  */
  79
  80 static int srcu_readers_active_idx(struct srcu_struct *sp, int idx)
  81 {
  82         int cpu;
  83         int sum;
  84
  85         sum = 0;
  86         for_each_possible_cpu(cpu)
  87                 sum += per_cpu_ptr(sp->per_cpu_ref, cpu)->c[idx];
  88         return sum;
  89 }
  90
  91 /**
  92  * srcu_readers_active - returns approximate number of readers.
  93  * @sp: which srcu_struct to count active readers (holding srcu_read_lock).
  94  *
  95  * Note that this is not an atomic primitive, and can therefore suffer
  96  * severe errors when invoked on an active srcu_struct.  That said, it
  97  * can be useful as an error check at cleanup time.
  98  */
  99 static int srcu_readers_active(struct srcu_struct *sp)
 100 {
 101         return srcu_readers_active_idx(sp, 0) + srcu_readers_active_idx(sp, 1);
 102 }
 103
 104 /**
 105  * cleanup_srcu_struct - deconstruct a sleep-RCU structure
 106  * @sp: structure to clean up.
 107  *
 108  * Must invoke this after you are finished using a given srcu_struct that
 109  * was initialized via init_srcu_struct(), else you leak memory.
 110  */
 111 void cleanup_srcu_struct(struct srcu_struct *sp)
 112 {
 113         int sum;
 114
 115         sum = srcu_readers_active(sp);
 116         WARN_ON(sum);  /* Leakage unless caller handles error. */
 117         if (sum != 0)
 118                 return;
 119         free_percpu(sp->per_cpu_ref);
 120         sp->per_cpu_ref = NULL;
 121 }
 122 EXPORT_SYMBOL_GPL(cleanup_srcu_struct);
 123
 124 /*
 125  * Counts the new reader in the appropriate per-CPU element of the
 126  * srcu_struct.  Must be called from process context.
 127  * Returns an index that must be passed to the matching srcu_read_unlock().
 128  */
 129 int __srcu_read_lock(struct srcu_struct *sp)
 130 {
 131         int idx;
 132
 133         preempt_disable();
 134         idx = sp->completed & 0x1;
 135         barrier();  /* ensure compiler looks -once- at sp->completed. */
 136         per_cpu_ptr(sp->per_cpu_ref, smp_processor_id())->c[idx]++;
 137         srcu_barrier();  /* ensure compiler won't misorder critical section. */
 138         preempt_enable();
 139         return idx;
 140 }
 141 EXPORT_SYMBOL_GPL(__srcu_read_lock);
 142
 143 /*
 144  * Removes the count for the old reader from the appropriate per-CPU
 145  * element of the srcu_struct.  Note that this may well be a different
 146  * CPU than that which was incremented by the corresponding srcu_read_lock().
 147  * Must be called from process context.
 148  */
 149 void __srcu_read_unlock(struct srcu_struct *sp, int idx)
 150 {
 151         preempt_disable();
 152         srcu_barrier();  /* ensure compiler won't misorder critical section. */
 153         per_cpu_ptr(sp->per_cpu_ref, smp_processor_id())->c[idx]--;
 154         preempt_enable();
 155 }
 156 EXPORT_SYMBOL_GPL(__srcu_read_unlock);
 157
 158 /*
 159  * We use an adaptive strategy for synchronize_srcu() and especially for
 160  * synchronize_srcu_expedited().  We spin for a fixed time period
 161  * (defined below) to allow SRCU readers to exit their read-side critical
 162  * sections.  If there are still some readers after 10 microseconds,
 163  * we repeatedly block for 1-millisecond time periods.  This approach
 164  * has done well in testing, so there is no need for a config parameter.
 165  */
 166 #define SYNCHRONIZE_SRCU_READER_DELAY 10
 167
 168 /*
 169  * Helper function for synchronize_srcu() and synchronize_srcu_expedited().
 170  */
 171 static void __synchronize_srcu(struct srcu_struct *sp, void (*sync_func)(void))
 172 {
 173         int idx;
 174
 175         rcu_lockdep_assert(!lock_is_held(&sp->dep_map) &&
 176                            !lock_is_held(&rcu_bh_lock_map) &&
 177                            !lock_is_held(&rcu_lock_map) &&
 178                            !lock_is_held(&rcu_sched_lock_map),
 179                            "Illegal synchronize_srcu() in same-type SRCU (or RCU) read-side critical section");
 180
 181         idx = sp->completed;
 182         mutex_lock(&sp->mutex);
 183
 184         /*
 185          * Check to see if someone else did the work for us while we were
 186          * waiting to acquire the lock.  We need -two- advances of
 187          * the counter, not just one.  If there was but one, we might have
 188          * shown up -after- our helper's first synchronize_sched(), thus
 189          * having failed to prevent CPU-reordering races with concurrent
 190          * srcu_read_unlock()s on other CPUs (see comment below).  So we
 191          * either (1) wait for two or (2) supply the second ourselves.
 192          */
 193
 194         if ((sp->completed - idx) >= 2) {
 195                 mutex_unlock(&sp->mutex);
 196                 return;
 197         }
 198
 199         sync_func();  /* Force memory barrier on all CPUs. */
 200
 201         /*
 202          * The preceding synchronize_sched() ensures that any CPU that
 203          * sees the new value of sp->completed will also see any preceding
 204          * changes to data structures made by this CPU.  This prevents
 205          * some other CPU from reordering the accesses in its SRCU
 206          * read-side critical section to precede the corresponding
 207          * srcu_read_lock() -- ensuring that such references will in
 208          * fact be protected.
 209          *
 210          * So it is now safe to do the flip.
 211          */
 212
 213         idx = sp->completed & 0x1;
 214         sp->completed++;
 215
 216         sync_func();  /* Force memory barrier on all CPUs. */
 217
 218         /*
 219          * At this point, because of the preceding synchronize_sched(),
 220          * all srcu_read_lock() calls using the old counters have completed.
 221          * Their corresponding critical sections might well be still
 222          * executing, but the srcu_read_lock() primitives themselves
 223          * will have finished executing.  We initially give readers
 224          * an arbitrarily chosen 10 microseconds to get out of their
 225          * SRCU read-side critical sections, then loop waiting 1/HZ
 226          * seconds per iteration.  The 10-microsecond value has done
 227          * very well in testing.
 228          */
 229
 230         if (srcu_readers_active_idx(sp, idx))
 231                 udelay(SYNCHRONIZE_SRCU_READER_DELAY);
 232         while (srcu_readers_active_idx(sp, idx))
 233                 schedule_timeout_interruptible(1);
 234
 235         sync_func();  /* Force memory barrier on all CPUs. */
 236
 237         /*
 238          * The preceding synchronize_sched() forces all srcu_read_unlock()
 239          * primitives that were executing concurrently with the preceding
 240          * for_each_possible_cpu() loop to have completed by this point.
 241          * More importantly, it also forces the corresponding SRCU read-side
 242          * critical sections to have also completed, and the corresponding
 243          * references to SRCU-protected data items to be dropped.
 244          *
 245          * Note:
 246          *
 247          *      Despite what you might think at first glance, the
 248          *      preceding synchronize_sched() -must- be within the
 249          *      critical section ended by the following mutex_unlock().
 250          *      Otherwise, a task taking the early exit can race
 251          *      with a srcu_read_unlock(), which might have executed
 252          *      just before the preceding srcu_readers_active() check,
 253          *      and whose CPU might have reordered the srcu_read_unlock()
 254          *      with the preceding critical section.  In this case, there
 255          *      is nothing preventing the synchronize_sched() task that is
 256          *      taking the early exit from freeing a data structure that
 257          *      is still being referenced (out of order) by the task
 258          *      doing the srcu_read_unlock().
 259          *
 260          *      Alternatively, the comparison with "2" on the early exit
 261          *      could be changed to "3", but this increases synchronize_srcu()
 262          *      latency for bulk loads.  So the current code is preferred.
 263          */
 264
 265         mutex_unlock(&sp->mutex);
 266 }
 267
 268 /**
 269  * synchronize_srcu - wait for prior SRCU read-side critical-section completion
 270  * @sp: srcu_struct with which to synchronize.
 271  *
 272  * Flip the completed counter, and wait for the old count to drain to zero.
 273  * As with classic RCU, the updater must use some separate means of
 274  * synchronizing concurrent updates.  Can block; must be called from
 275  * process context.
 276  *
 277  * Note that it is illegal to call synchronize_srcu() from the corresponding
 278  * SRCU read-side critical section; doing so will result in deadlock.
 279  * However, it is perfectly legal to call synchronize_srcu() on one
 280  * srcu_struct from some other srcu_struct's read-side critical section.
 281  */
 282 void synchronize_srcu(struct srcu_struct *sp)
 283 {
 284         __synchronize_srcu(sp, synchronize_sched);
 285 }
 286 EXPORT_SYMBOL_GPL(synchronize_srcu);
 287
 288 /**
 289  * synchronize_srcu_expedited - Brute-force SRCU grace period
 290  * @sp: srcu_struct with which to synchronize.
 291  *
 292  * Wait for an SRCU grace period to elapse, but use a "big hammer"
 293  * approach to force the grace period to end quickly.  This consumes
 294  * significant time on all CPUs and is unfriendly to real-time workloads,
 295  * so is thus not recommended for any sort of common-case code.  In fact,
 296  * if you are using synchronize_srcu_expedited() in a loop, please
 297  * restructure your code to batch your updates, and then use a single
 298  * synchronize_srcu() instead.
 299  *
 300  * Note that it is illegal to call this function while holding any lock
 301  * that is acquired by a CPU-hotplug notifier.  And yes, it is also illegal
 302  * to call this function from a CPU-hotplug notifier.  Failing to observe
 303  * these restriction will result in deadlock.  It is also illegal to call
 304  * synchronize_srcu_expedited() from the corresponding SRCU read-side
 305  * critical section; doing so will result in deadlock.  However, it is
 306  * perfectly legal to call synchronize_srcu_expedited() on one srcu_struct
 307  * from some other srcu_struct's read-side critical section, as long as
 308  * the resulting graph of srcu_structs is acyclic.
 309  */
 310 void synchronize_srcu_expedited(struct srcu_struct *sp)
 311 {
 312         __synchronize_srcu(sp, synchronize_sched_expedited);
 313 }
 314 EXPORT_SYMBOL_GPL(synchronize_srcu_expedited);
 315
 316 /**
 317  * srcu_batches_completed - return batches completed.
 318  * @sp: srcu_struct on which to report batch completion.
 319  *
 320  * Report the number of batches, correlated with, but not necessarily
 321  * precisely the same as, the number of grace periods that have elapsed.
 322  */
 323
 324 long srcu_batches_completed(struct srcu_struct *sp)
 325 {
 326         return sp->completed;
 327 }
 328 EXPORT_SYMBOL_GPL(srcu_batches_completed);