drivers/tty/tty_ldsem.c

   1 /*
   2  * Ldisc rw semaphore
   3  *
   4  * The ldisc semaphore is semantically a rw_semaphore but which enforces
   5  * an alternate policy, namely:
   6  *   1) Supports lock wait timeouts
   7  *   2) Write waiter has priority
   8  *   3) Downgrading is not supported
   9  *
  10  * Implementation notes:
  11  *   1) Upper half of semaphore count is a wait count (differs from rwsem
  12  *      in that rwsem normalizes the upper half to the wait bias)
  13  *   2) Lacks overflow checking
  14  *
  15  * The generic counting was copied and modified from include/asm-generic/rwsem.h
  16  * by Paul Mackerras <paulus@samba.org>.
  17  *
  18  * The scheduling policy was copied and modified from lib/rwsem.c
  19  * Written by David Howells (dhowells@redhat.com).
  20  *
  21  * This implementation incorporates the write lock stealing work of
  22  * Michel Lespinasse <walken@google.com>.
  23  *
  24  * Copyright (C) 2013 Peter Hurley <peter@hurleysoftware.com>
  25  *
  26  * This file may be redistributed under the terms of the GNU General Public
  27  * License v2.
  28  */
  29
  30 #include <linux/list.h>
  31 #include <linux/spinlock.h>
  32 #include <linux/atomic.h>
  33 #include <linux/tty.h>
  34 #include <linux/sched.h>
  35
  36
  37 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  38 # define __acq(l, s, t, r, c, n, i)             \
  39                                 lock_acquire(&(l)->dep_map, s, t, r, c, n, i)
  40 # define __rel(l, n, i)                         \
  41                                 lock_release(&(l)->dep_map, n, i)
  42 #define lockdep_acquire(l, s, t, i)             __acq(l, s, t, 0, 1, NULL, i)
  43 #define lockdep_acquire_nest(l, s, t, n, i)     __acq(l, s, t, 0, 1, n, i)
  44 #define lockdep_acquire_read(l, s, t, i)        __acq(l, s, t, 1, 1, NULL, i)
  45 #define lockdep_release(l, n, i)                __rel(l, n, i)
  46 #else
  47 # define lockdep_acquire(l, s, t, i)            do { } while (0)
  48 # define lockdep_acquire_nest(l, s, t, n, i)    do { } while (0)
  49 # define lockdep_acquire_read(l, s, t, i)       do { } while (0)
  50 # define lockdep_release(l, n, i)               do { } while (0)
  51 #endif
  52
  53 #ifdef CONFIG_LOCK_STAT
  54 # define lock_stat(_lock, stat)         lock_##stat(&(_lock)->dep_map, _RET_IP_)
  55 #else
  56 # define lock_stat(_lock, stat)         do { } while (0)
  57 #endif
  58
  59
  60 #if BITS_PER_LONG == 64
  61 # define LDSEM_ACTIVE_MASK      0xffffffffL
  62 #else
  63 # define LDSEM_ACTIVE_MASK      0x0000ffffL
  64 #endif
  65
  66 #define LDSEM_UNLOCKED          0L
  67 #define LDSEM_ACTIVE_BIAS       1L
  68 #define LDSEM_WAIT_BIAS         (-LDSEM_ACTIVE_MASK-1)
  69 #define LDSEM_READ_BIAS         LDSEM_ACTIVE_BIAS
  70 #define LDSEM_WRITE_BIAS        (LDSEM_WAIT_BIAS + LDSEM_ACTIVE_BIAS)
  71
  72 struct ldsem_waiter {
  73         struct list_head list;
  74         struct task_struct *task;
  75 };
  76
  77 static inline long ldsem_atomic_update(long delta, struct ld_semaphore *sem)
  78 {
  79         return atomic_long_add_return(delta, (atomic_long_t *)&sem->count);
  80 }
  81
  82 /*
  83  * ldsem_cmpxchg() updates @*old with the last-known sem->count value.
  84  * Returns 1 if count was successfully changed; @*old will have @new value.
  85  * Returns 0 if count was not changed; @*old will have most recent sem->count
  86  */
  87 static inline int ldsem_cmpxchg(long *old, long new, struct ld_semaphore *sem)
  88 {
  89         long tmp = atomic_long_cmpxchg(&sem->count, *old, new);
  90         if (tmp == *old) {
  91                 *old = new;
  92                 return 1;
  93         } else {
  94                 *old = tmp;
  95                 return 0;
  96         }
  97 }
  98
  99 /*
 100  * Initialize an ldsem:
 101  */
 102 void __init_ldsem(struct ld_semaphore *sem, const char *name,
 103                   struct lock_class_key *key)
 104 {
 105 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 106         /*
 107          * Make sure we are not reinitializing a held semaphore:
 108          */
 109         debug_check_no_locks_freed((void *)sem, sizeof(*sem));
 110         lockdep_init_map(&sem->dep_map, name, key, 0);
 111 #endif
 112         sem->count = LDSEM_UNLOCKED;
 113         sem->wait_readers = 0;
 114         raw_spin_lock_init(&sem->wait_lock);
 115         INIT_LIST_HEAD(&sem->read_wait);
 116         INIT_LIST_HEAD(&sem->write_wait);
 117 }
 118
 119 static void __ldsem_wake_readers(struct ld_semaphore *sem)
 120 {
 121         struct ldsem_waiter *waiter, *next;
 122         struct task_struct *tsk;
 123         long adjust, count;
 124
 125         /* Try to grant read locks to all readers on the read wait list.
 126          * Note the 'active part' of the count is incremented by
 127          * the number of readers before waking any processes up.
 128          */
 129         adjust = sem->wait_readers * (LDSEM_ACTIVE_BIAS - LDSEM_WAIT_BIAS);
 130         count = ldsem_atomic_update(adjust, sem);
 131         do {
 132                 if (count > 0)
 133                         break;
 134                 if (ldsem_cmpxchg(&count, count - adjust, sem))
 135                         return;
 136         } while (1);
 137
 138         list_for_each_entry_safe(waiter, next, &sem->read_wait, list) {
 139                 tsk = waiter->task;
 140                 smp_store_release(&waiter->task, NULL);
 141                 wake_up_process(tsk);
 142                 put_task_struct(tsk);
 143         }
 144         INIT_LIST_HEAD(&sem->read_wait);
 145         sem->wait_readers = 0;
 146 }
 147
 148 static inline int writer_trylock(struct ld_semaphore *sem)
 149 {
 150         /* only wake this writer if the active part of the count can be
 151          * transitioned from 0 -> 1
 152          */
 153         long count = ldsem_atomic_update(LDSEM_ACTIVE_BIAS, sem);
 154         do {
 155                 if ((count & LDSEM_ACTIVE_MASK) == LDSEM_ACTIVE_BIAS)
 156                         return 1;
 157                 if (ldsem_cmpxchg(&count, count - LDSEM_ACTIVE_BIAS, sem))
 158                         return 0;
 159         } while (1);
 160 }
 161
 162 static void __ldsem_wake_writer(struct ld_semaphore *sem)
 163 {
 164         struct ldsem_waiter *waiter;
 165
 166         waiter = list_entry(sem->write_wait.next, struct ldsem_waiter, list);
 167         wake_up_process(waiter->task);
 168 }
 169
 170 /*
 171  * handle the lock release when processes blocked on it that can now run
 172  * - if we come here from up_xxxx(), then:
 173  *   - the 'active part' of count (&0x0000ffff) reached 0 (but may have changed)
 174  *   - the 'waiting part' of count (&0xffff0000) is -ve (and will still be so)
 175  * - the spinlock must be held by the caller
 176  * - woken process blocks are discarded from the list after having task zeroed
 177  */
 178 static void __ldsem_wake(struct ld_semaphore *sem)
 179 {
 180         if (!list_empty(&sem->write_wait))
 181                 __ldsem_wake_writer(sem);
 182         else if (!list_empty(&sem->read_wait))
 183                 __ldsem_wake_readers(sem);
 184 }
 185
 186 static void ldsem_wake(struct ld_semaphore *sem)
 187 {
 188         unsigned long flags;
 189
 190         raw_spin_lock_irqsave(&sem->wait_lock, flags);
 191         __ldsem_wake(sem);
 192         raw_spin_unlock_irqrestore(&sem->wait_lock, flags);
 193 }
 194
 195 /*
 196  * wait for the read lock to be granted
 197  */
 198 static struct ld_semaphore __sched *
 199 down_read_failed(struct ld_semaphore *sem, long count, long timeout)
 200 {
 201         struct ldsem_waiter waiter;
 202         struct task_struct *tsk = current;
 203         long adjust = -LDSEM_ACTIVE_BIAS + LDSEM_WAIT_BIAS;
 204
 205         /* set up my own style of waitqueue */
 206         raw_spin_lock_irq(&sem->wait_lock);
 207
 208         /* Try to reverse the lock attempt but if the count has changed
 209          * so that reversing fails, check if there are are no waiters,
 210          * and early-out if not */
 211         do {
 212                 if (ldsem_cmpxchg(&count, count + adjust, sem))
 213                         break;
 214                 if (count > 0) {
 215                         raw_spin_unlock_irq(&sem->wait_lock);
 216                         return sem;
 217                 }
 218         } while (1);
 219
 220         list_add_tail(&waiter.list, &sem->read_wait);
 221         sem->wait_readers++;
 222
 223         waiter.task = tsk;
 224         get_task_struct(tsk);
 225
 226         /* if there are no active locks, wake the new lock owner(s) */
 227         if ((count & LDSEM_ACTIVE_MASK) == 0)
 228                 __ldsem_wake(sem);
 229
 230         raw_spin_unlock_irq(&sem->wait_lock);
 231
 232         /* wait to be given the lock */
 233         for (;;) {
 234                 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
 235
 236                 if (!smp_load_acquire(&waiter.task))
 237                         break;
 238                 if (!timeout)
 239                         break;
 240                 timeout = schedule_timeout(timeout);
 241         }
 242
 243         __set_task_state(tsk, TASK_RUNNING);
 244
 245         if (!timeout) {
 246                 /* lock timed out but check if this task was just
 247                  * granted lock ownership - if so, pretend there
 248                  * was no timeout; otherwise, cleanup lock wait */
 249                 raw_spin_lock_irq(&sem->wait_lock);
 250                 if (waiter.task) {
 251                         ldsem_atomic_update(-LDSEM_WAIT_BIAS, sem);
 252                         list_del(&waiter.list);
 253                         raw_spin_unlock_irq(&sem->wait_lock);
 254                         put_task_struct(waiter.task);
 255                         return NULL;
 256                 }
 257                 raw_spin_unlock_irq(&sem->wait_lock);
 258         }
 259
 260         return sem;
 261 }
 262
 263 /*
 264  * wait for the write lock to be granted
 265  */
 266 static struct ld_semaphore __sched *
 267 down_write_failed(struct ld_semaphore *sem, long count, long timeout)
 268 {
 269         struct ldsem_waiter waiter;
 270         struct task_struct *tsk = current;
 271         long adjust = -LDSEM_ACTIVE_BIAS;
 272         int locked = 0;
 273
 274         /* set up my own style of waitqueue */
 275         raw_spin_lock_irq(&sem->wait_lock);
 276
 277         /* Try to reverse the lock attempt but if the count has changed
 278          * so that reversing fails, check if the lock is now owned,
 279          * and early-out if so */
 280         do {
 281                 if (ldsem_cmpxchg(&count, count + adjust, sem))
 282                         break;
 283                 if ((count & LDSEM_ACTIVE_MASK) == LDSEM_ACTIVE_BIAS) {
 284                         raw_spin_unlock_irq(&sem->wait_lock);
 285                         return sem;
 286                 }
 287         } while (1);
 288
 289         list_add_tail(&waiter.list, &sem->write_wait);
 290
 291         waiter.task = tsk;
 292
 293         set_task_state(tsk, TASK_UNINTERRUPTIBLE);
 294         for (;;) {
 295                 if (!timeout)
 296                         break;
 297                 raw_spin_unlock_irq(&sem->wait_lock);
 298                 timeout = schedule_timeout(timeout);
 299                 raw_spin_lock_irq(&sem->wait_lock);
 300                 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
 301                 locked = writer_trylock(sem);
 302                 if (locked)
 303                         break;
 304         }
 305
 306         if (!locked)
 307                 ldsem_atomic_update(-LDSEM_WAIT_BIAS, sem);
 308         list_del(&waiter.list);
 309
 310         /*
 311          * In case of timeout, wake up every reader who gave the right of way
 312          * to writer. Prevent separation readers into two groups:
 313          * one that helds semaphore and another that sleeps.
 314          * (in case of no contention with a writer)
 315          */
 316         if (!locked && list_empty(&sem->write_wait))
 317                 __ldsem_wake_readers(sem);
 318
 319         raw_spin_unlock_irq(&sem->wait_lock);
 320
 321         __set_task_state(tsk, TASK_RUNNING);
 322
 323         /* lock wait may have timed out */
 324         if (!locked)
 325                 return NULL;
 326         return sem;
 327 }
 328
 329
 330
 331 static int __ldsem_down_read_nested(struct ld_semaphore *sem,
 332                                            int subclass, long timeout)
 333 {
 334         long count;
 335
 336         lockdep_acquire_read(sem, subclass, 0, _RET_IP_);
 337
 338         count = ldsem_atomic_update(LDSEM_READ_BIAS, sem);
 339         if (count <= 0) {
 340                 lock_stat(sem, contended);
 341                 if (!down_read_failed(sem, count, timeout)) {
 342                         lockdep_release(sem, 1, _RET_IP_);
 343                         return 0;
 344                 }
 345         }
 346         lock_stat(sem, acquired);
 347         return 1;
 348 }
 349
 350 static int __ldsem_down_write_nested(struct ld_semaphore *sem,
 351                                             int subclass, long timeout)
 352 {
 353         long count;
 354
 355         lockdep_acquire(sem, subclass, 0, _RET_IP_);
 356
 357         count = ldsem_atomic_update(LDSEM_WRITE_BIAS, sem);
 358         if ((count & LDSEM_ACTIVE_MASK) != LDSEM_ACTIVE_BIAS) {
 359                 lock_stat(sem, contended);
 360                 if (!down_write_failed(sem, count, timeout)) {
 361                         lockdep_release(sem, 1, _RET_IP_);
 362                         return 0;
 363                 }
 364         }
 365         lock_stat(sem, acquired);
 366         return 1;
 367 }
 368
 369
 370 /*
 371  * lock for reading -- returns 1 if successful, 0 if timed out
 372  */
 373 int __sched ldsem_down_read(struct ld_semaphore *sem, long timeout)
 374 {
 375         might_sleep();
 376         return __ldsem_down_read_nested(sem, 0, timeout);
 377 }
 378
 379 /*
 380  * trylock for reading -- returns 1 if successful, 0 if contention
 381  */
 382 int ldsem_down_read_trylock(struct ld_semaphore *sem)
 383 {
 384         long count = sem->count;
 385
 386         while (count >= 0) {
 387                 if (ldsem_cmpxchg(&count, count + LDSEM_READ_BIAS, sem)) {
 388                         lockdep_acquire_read(sem, 0, 1, _RET_IP_);
 389                         lock_stat(sem, acquired);
 390                         return 1;
 391                 }
 392         }
 393         return 0;
 394 }
 395
 396 /*
 397  * lock for writing -- returns 1 if successful, 0 if timed out
 398  */
 399 int __sched ldsem_down_write(struct ld_semaphore *sem, long timeout)
 400 {
 401         might_sleep();
 402         return __ldsem_down_write_nested(sem, 0, timeout);
 403 }
 404
 405 /*
 406  * trylock for writing -- returns 1 if successful, 0 if contention
 407  */
 408 int ldsem_down_write_trylock(struct ld_semaphore *sem)
 409 {
 410         long count = sem->count;
 411
 412         while ((count & LDSEM_ACTIVE_MASK) == 0) {
 413                 if (ldsem_cmpxchg(&count, count + LDSEM_WRITE_BIAS, sem)) {
 414                         lockdep_acquire(sem, 0, 1, _RET_IP_);
 415                         lock_stat(sem, acquired);
 416                         return 1;
 417                 }
 418         }
 419         return 0;
 420 }
 421
 422 /*
 423  * release a read lock
 424  */
 425 void ldsem_up_read(struct ld_semaphore *sem)
 426 {
 427         long count;
 428
 429         lockdep_release(sem, 1, _RET_IP_);
 430
 431         count = ldsem_atomic_update(-LDSEM_READ_BIAS, sem);
 432         if (count < 0 && (count & LDSEM_ACTIVE_MASK) == 0)
 433                 ldsem_wake(sem);
 434 }
 435
 436 /*
 437  * release a write lock
 438  */
 439 void ldsem_up_write(struct ld_semaphore *sem)
 440 {
 441         long count;
 442
 443         lockdep_release(sem, 1, _RET_IP_);
 444
 445         count = ldsem_atomic_update(-LDSEM_WRITE_BIAS, sem);
 446         if (count < 0)
 447                 ldsem_wake(sem);
 448 }
 449
 450
 451 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 452
 453 int ldsem_down_read_nested(struct ld_semaphore *sem, int subclass, long timeout)
 454 {
 455         might_sleep();
 456         return __ldsem_down_read_nested(sem, subclass, timeout);
 457 }
 458
 459 int ldsem_down_write_nested(struct ld_semaphore *sem, int subclass,
 460                             long timeout)
 461 {
 462         might_sleep();
 463         return __ldsem_down_write_nested(sem, subclass, timeout);
 464 }
 465
 466 #endif