Paul E. McKenney | 621934e | 2006-10-04 02:17:02 -0700 | [diff] [blame] | 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/module.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/slab.h> |
| 34 | #include <linux/smp.h> |
| 35 | #include <linux/srcu.h> |
| 36 | |
| 37 | /** |
| 38 | * init_srcu_struct - initialize a sleep-RCU structure |
| 39 | * @sp: structure to initialize. |
| 40 | * |
| 41 | * Must invoke this on a given srcu_struct before passing that srcu_struct |
| 42 | * to any other function. Each srcu_struct represents a separate domain |
| 43 | * of SRCU protection. |
| 44 | */ |
Alan Stern | e6a9201 | 2006-10-04 02:17:05 -0700 | [diff] [blame] | 45 | int init_srcu_struct(struct srcu_struct *sp) |
Paul E. McKenney | 621934e | 2006-10-04 02:17:02 -0700 | [diff] [blame] | 46 | { |
| 47 | sp->completed = 0; |
Paul E. McKenney | 621934e | 2006-10-04 02:17:02 -0700 | [diff] [blame] | 48 | mutex_init(&sp->mutex); |
Alan Stern | e6a9201 | 2006-10-04 02:17:05 -0700 | [diff] [blame] | 49 | sp->per_cpu_ref = alloc_percpu(struct srcu_struct_array); |
| 50 | return (sp->per_cpu_ref ? 0 : -ENOMEM); |
Paul E. McKenney | 621934e | 2006-10-04 02:17:02 -0700 | [diff] [blame] | 51 | } |
| 52 | |
| 53 | /* |
| 54 | * srcu_readers_active_idx -- returns approximate number of readers |
| 55 | * active on the specified rank of per-CPU counters. |
| 56 | */ |
| 57 | |
| 58 | static int srcu_readers_active_idx(struct srcu_struct *sp, int idx) |
| 59 | { |
| 60 | int cpu; |
| 61 | int sum; |
| 62 | |
| 63 | sum = 0; |
| 64 | for_each_possible_cpu(cpu) |
| 65 | sum += per_cpu_ptr(sp->per_cpu_ref, cpu)->c[idx]; |
| 66 | return sum; |
| 67 | } |
| 68 | |
| 69 | /** |
| 70 | * srcu_readers_active - returns approximate number of readers. |
| 71 | * @sp: which srcu_struct to count active readers (holding srcu_read_lock). |
| 72 | * |
| 73 | * Note that this is not an atomic primitive, and can therefore suffer |
| 74 | * severe errors when invoked on an active srcu_struct. That said, it |
| 75 | * can be useful as an error check at cleanup time. |
| 76 | */ |
| 77 | int srcu_readers_active(struct srcu_struct *sp) |
| 78 | { |
| 79 | return srcu_readers_active_idx(sp, 0) + srcu_readers_active_idx(sp, 1); |
| 80 | } |
| 81 | |
| 82 | /** |
| 83 | * cleanup_srcu_struct - deconstruct a sleep-RCU structure |
| 84 | * @sp: structure to clean up. |
| 85 | * |
| 86 | * Must invoke this after you are finished using a given srcu_struct that |
| 87 | * was initialized via init_srcu_struct(), else you leak memory. |
| 88 | */ |
| 89 | void cleanup_srcu_struct(struct srcu_struct *sp) |
| 90 | { |
| 91 | int sum; |
| 92 | |
| 93 | sum = srcu_readers_active(sp); |
| 94 | WARN_ON(sum); /* Leakage unless caller handles error. */ |
| 95 | if (sum != 0) |
| 96 | return; |
| 97 | free_percpu(sp->per_cpu_ref); |
| 98 | sp->per_cpu_ref = NULL; |
| 99 | } |
| 100 | |
| 101 | /** |
| 102 | * srcu_read_lock - register a new reader for an SRCU-protected structure. |
| 103 | * @sp: srcu_struct in which to register the new reader. |
| 104 | * |
| 105 | * Counts the new reader in the appropriate per-CPU element of the |
| 106 | * srcu_struct. Must be called from process context. |
| 107 | * Returns an index that must be passed to the matching srcu_read_unlock(). |
| 108 | */ |
| 109 | int srcu_read_lock(struct srcu_struct *sp) |
| 110 | { |
| 111 | int idx; |
| 112 | |
| 113 | preempt_disable(); |
| 114 | idx = sp->completed & 0x1; |
| 115 | barrier(); /* ensure compiler looks -once- at sp->completed. */ |
| 116 | per_cpu_ptr(sp->per_cpu_ref, smp_processor_id())->c[idx]++; |
| 117 | srcu_barrier(); /* ensure compiler won't misorder critical section. */ |
| 118 | preempt_enable(); |
| 119 | return idx; |
| 120 | } |
| 121 | |
| 122 | /** |
| 123 | * srcu_read_unlock - unregister a old reader from an SRCU-protected structure. |
| 124 | * @sp: srcu_struct in which to unregister the old reader. |
| 125 | * @idx: return value from corresponding srcu_read_lock(). |
| 126 | * |
| 127 | * Removes the count for the old reader from the appropriate per-CPU |
| 128 | * element of the srcu_struct. Note that this may well be a different |
| 129 | * CPU than that which was incremented by the corresponding srcu_read_lock(). |
| 130 | * Must be called from process context. |
| 131 | */ |
| 132 | void srcu_read_unlock(struct srcu_struct *sp, int idx) |
| 133 | { |
| 134 | preempt_disable(); |
| 135 | srcu_barrier(); /* ensure compiler won't misorder critical section. */ |
| 136 | per_cpu_ptr(sp->per_cpu_ref, smp_processor_id())->c[idx]--; |
| 137 | preempt_enable(); |
| 138 | } |
| 139 | |
| 140 | /** |
| 141 | * synchronize_srcu - wait for prior SRCU read-side critical-section completion |
| 142 | * @sp: srcu_struct with which to synchronize. |
| 143 | * |
| 144 | * Flip the completed counter, and wait for the old count to drain to zero. |
| 145 | * As with classic RCU, the updater must use some separate means of |
| 146 | * synchronizing concurrent updates. Can block; must be called from |
| 147 | * process context. |
| 148 | * |
| 149 | * Note that it is illegal to call synchornize_srcu() from the corresponding |
| 150 | * SRCU read-side critical section; doing so will result in deadlock. |
| 151 | * However, it is perfectly legal to call synchronize_srcu() on one |
| 152 | * srcu_struct from some other srcu_struct's read-side critical section. |
| 153 | */ |
| 154 | void synchronize_srcu(struct srcu_struct *sp) |
| 155 | { |
| 156 | int idx; |
| 157 | |
| 158 | idx = sp->completed; |
| 159 | mutex_lock(&sp->mutex); |
| 160 | |
| 161 | /* |
| 162 | * Check to see if someone else did the work for us while we were |
| 163 | * waiting to acquire the lock. We need -two- advances of |
| 164 | * the counter, not just one. If there was but one, we might have |
| 165 | * shown up -after- our helper's first synchronize_sched(), thus |
| 166 | * having failed to prevent CPU-reordering races with concurrent |
| 167 | * srcu_read_unlock()s on other CPUs (see comment below). So we |
| 168 | * either (1) wait for two or (2) supply the second ourselves. |
| 169 | */ |
| 170 | |
| 171 | if ((sp->completed - idx) >= 2) { |
| 172 | mutex_unlock(&sp->mutex); |
| 173 | return; |
| 174 | } |
| 175 | |
| 176 | synchronize_sched(); /* Force memory barrier on all CPUs. */ |
| 177 | |
| 178 | /* |
| 179 | * The preceding synchronize_sched() ensures that any CPU that |
| 180 | * sees the new value of sp->completed will also see any preceding |
| 181 | * changes to data structures made by this CPU. This prevents |
| 182 | * some other CPU from reordering the accesses in its SRCU |
| 183 | * read-side critical section to precede the corresponding |
| 184 | * srcu_read_lock() -- ensuring that such references will in |
| 185 | * fact be protected. |
| 186 | * |
| 187 | * So it is now safe to do the flip. |
| 188 | */ |
| 189 | |
| 190 | idx = sp->completed & 0x1; |
| 191 | sp->completed++; |
| 192 | |
| 193 | synchronize_sched(); /* Force memory barrier on all CPUs. */ |
| 194 | |
| 195 | /* |
| 196 | * At this point, because of the preceding synchronize_sched(), |
| 197 | * all srcu_read_lock() calls using the old counters have completed. |
| 198 | * Their corresponding critical sections might well be still |
| 199 | * executing, but the srcu_read_lock() primitives themselves |
| 200 | * will have finished executing. |
| 201 | */ |
| 202 | |
| 203 | while (srcu_readers_active_idx(sp, idx)) |
| 204 | schedule_timeout_interruptible(1); |
| 205 | |
| 206 | synchronize_sched(); /* Force memory barrier on all CPUs. */ |
| 207 | |
| 208 | /* |
| 209 | * The preceding synchronize_sched() forces all srcu_read_unlock() |
| 210 | * primitives that were executing concurrently with the preceding |
| 211 | * for_each_possible_cpu() loop to have completed by this point. |
| 212 | * More importantly, it also forces the corresponding SRCU read-side |
| 213 | * critical sections to have also completed, and the corresponding |
| 214 | * references to SRCU-protected data items to be dropped. |
| 215 | * |
| 216 | * Note: |
| 217 | * |
| 218 | * Despite what you might think at first glance, the |
| 219 | * preceding synchronize_sched() -must- be within the |
| 220 | * critical section ended by the following mutex_unlock(). |
| 221 | * Otherwise, a task taking the early exit can race |
| 222 | * with a srcu_read_unlock(), which might have executed |
| 223 | * just before the preceding srcu_readers_active() check, |
| 224 | * and whose CPU might have reordered the srcu_read_unlock() |
| 225 | * with the preceding critical section. In this case, there |
| 226 | * is nothing preventing the synchronize_sched() task that is |
| 227 | * taking the early exit from freeing a data structure that |
| 228 | * is still being referenced (out of order) by the task |
| 229 | * doing the srcu_read_unlock(). |
| 230 | * |
| 231 | * Alternatively, the comparison with "2" on the early exit |
| 232 | * could be changed to "3", but this increases synchronize_srcu() |
| 233 | * latency for bulk loads. So the current code is preferred. |
| 234 | */ |
| 235 | |
| 236 | mutex_unlock(&sp->mutex); |
| 237 | } |
| 238 | |
| 239 | /** |
| 240 | * srcu_batches_completed - return batches completed. |
| 241 | * @sp: srcu_struct on which to report batch completion. |
| 242 | * |
| 243 | * Report the number of batches, correlated with, but not necessarily |
| 244 | * precisely the same as, the number of grace periods that have elapsed. |
| 245 | */ |
| 246 | |
| 247 | long srcu_batches_completed(struct srcu_struct *sp) |
| 248 | { |
| 249 | return sp->completed; |
| 250 | } |
| 251 | |
| 252 | EXPORT_SYMBOL_GPL(init_srcu_struct); |
| 253 | EXPORT_SYMBOL_GPL(cleanup_srcu_struct); |
| 254 | EXPORT_SYMBOL_GPL(srcu_read_lock); |
| 255 | EXPORT_SYMBOL_GPL(srcu_read_unlock); |
| 256 | EXPORT_SYMBOL_GPL(synchronize_srcu); |
| 257 | EXPORT_SYMBOL_GPL(srcu_batches_completed); |
| 258 | EXPORT_SYMBOL_GPL(srcu_readers_active); |