drivers/tty/tty_ldsem.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Ldisc rw semaphore
   4  *
   5  * The ldisc semaphore is semantically a rw_semaphore but which enforces
   6  * an alternate policy, namely:
   7  *   1) Supports lock wait timeouts
   8  *   2) Write waiter has priority
   9  *   3) Downgrading is not supported
  10  *
  11  * Implementation notes:
  12  *   1) Upper half of semaphore count is a wait count (differs from rwsem
  13  *      in that rwsem normalizes the upper half to the wait bias)
  14  *   2) Lacks overflow checking
  15  *
  16  * The generic counting was copied and modified from include/asm-generic/rwsem.h
  17  * by Paul Mackerras <paulus@samba.org>.
  18  *
  19  * The scheduling policy was copied and modified from lib/rwsem.c
  20  * Written by David Howells (dhowells@redhat.com).
  21  *
  22  * This implementation incorporates the write lock stealing work of
  23  * Michel Lespinasse <walken@google.com>.
  24  *
  25  * Copyright (C) 2013 Peter Hurley <peter@hurleysoftware.com>
  26  */
  27
  28 #include <linux/list.h>
  29 #include <linux/spinlock.h>
  30 #include <linux/atomic.h>
  31 #include <linux/tty.h>
  32 #include <linux/sched.h>
  33 #include <linux/sched/debug.h>
  34 #include <linux/sched/task.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 /*
  78  * Initialize an ldsem:
  79  */
  80 void __init_ldsem(struct ld_semaphore *sem, const char *name,
  81                   struct lock_class_key *key)
  82 {
  83 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  84         /*
  85          * Make sure we are not reinitializing a held semaphore:
  86          */
  87         debug_check_no_locks_freed((void *)sem, sizeof(*sem));
  88         lockdep_init_map(&sem->dep_map, name, key, 0);
  89 #endif
  90         atomic_long_set(&sem->count, LDSEM_UNLOCKED);
  91         sem->wait_readers = 0;
  92         raw_spin_lock_init(&sem->wait_lock);
  93         INIT_LIST_HEAD(&sem->read_wait);
  94         INIT_LIST_HEAD(&sem->write_wait);
  95 }
  96
  97 static void __ldsem_wake_readers(struct ld_semaphore *sem)
  98 {
  99         struct ldsem_waiter *waiter, *next;
 100         struct task_struct *tsk;
 101         long adjust, count;
 102
 103         /*
 104          * Try to grant read locks to all readers on the read wait list.
 105          * Note the 'active part' of the count is incremented by
 106          * the number of readers before waking any processes up.
 107          */
 108         adjust = sem->wait_readers * (LDSEM_ACTIVE_BIAS - LDSEM_WAIT_BIAS);
 109         count = atomic_long_add_return(adjust, &sem->count);
 110         do {
 111                 if (count > 0)
 112                         break;
 113                 if (atomic_long_try_cmpxchg(&sem->count, &count, count - adjust))
 114                         return;
 115         } while (1);
 116
 117         list_for_each_entry_safe(waiter, next, &sem->read_wait, list) {
 118                 tsk = waiter->task;
 119                 smp_mb();
 120                 waiter->task = NULL;
 121                 wake_up_process(tsk);
 122                 put_task_struct(tsk);
 123         }
 124         INIT_LIST_HEAD(&sem->read_wait);
 125         sem->wait_readers = 0;
 126 }
 127
 128 static inline int writer_trylock(struct ld_semaphore *sem)
 129 {
 130         /*
 131          * Only wake this writer if the active part of the count can be
 132          * transitioned from 0 -> 1
 133          */
 134         long count = atomic_long_add_return(LDSEM_ACTIVE_BIAS, &sem->count);
 135         do {
 136                 if ((count & LDSEM_ACTIVE_MASK) == LDSEM_ACTIVE_BIAS)
 137                         return 1;
 138                 if (atomic_long_try_cmpxchg(&sem->count, &count, count - LDSEM_ACTIVE_BIAS))
 139                         return 0;
 140         } while (1);
 141 }
 142
 143 static void __ldsem_wake_writer(struct ld_semaphore *sem)
 144 {
 145         struct ldsem_waiter *waiter;
 146
 147         waiter = list_entry(sem->write_wait.next, struct ldsem_waiter, list);
 148         wake_up_process(waiter->task);
 149 }
 150
 151 /*
 152  * handle the lock release when processes blocked on it that can now run
 153  * - if we come here from up_xxxx(), then:
 154  *   - the 'active part' of count (&0x0000ffff) reached 0 (but may have changed)
 155  *   - the 'waiting part' of count (&0xffff0000) is -ve (and will still be so)
 156  * - the spinlock must be held by the caller
 157  * - woken process blocks are discarded from the list after having task zeroed
 158  */
 159 static void __ldsem_wake(struct ld_semaphore *sem)
 160 {
 161         if (!list_empty(&sem->write_wait))
 162                 __ldsem_wake_writer(sem);
 163         else if (!list_empty(&sem->read_wait))
 164                 __ldsem_wake_readers(sem);
 165 }
 166
 167 static void ldsem_wake(struct ld_semaphore *sem)
 168 {
 169         unsigned long flags;
 170
 171         raw_spin_lock_irqsave(&sem->wait_lock, flags);
 172         __ldsem_wake(sem);
 173         raw_spin_unlock_irqrestore(&sem->wait_lock, flags);
 174 }
 175
 176 /*
 177  * wait for the read lock to be granted
 178  */
 179 static struct ld_semaphore __sched *
 180 down_read_failed(struct ld_semaphore *sem, long count, long timeout)
 181 {
 182         struct ldsem_waiter waiter;
 183         long adjust = -LDSEM_ACTIVE_BIAS + LDSEM_WAIT_BIAS;
 184
 185         /* set up my own style of waitqueue */
 186         raw_spin_lock_irq(&sem->wait_lock);
 187
 188         /*
 189          * Try to reverse the lock attempt but if the count has changed
 190          * so that reversing fails, check if there are are no waiters,
 191          * and early-out if not
 192          */
 193         do {
 194                 if (atomic_long_try_cmpxchg(&sem->count, &count, count + adjust)) {
 195                         count += adjust;
 196                         break;
 197                 }
 198                 if (count > 0) {
 199                         raw_spin_unlock_irq(&sem->wait_lock);
 200                         return sem;
 201                 }
 202         } while (1);
 203
 204         list_add_tail(&waiter.list, &sem->read_wait);
 205         sem->wait_readers++;
 206
 207         waiter.task = current;
 208         get_task_struct(current);
 209
 210         /* if there are no active locks, wake the new lock owner(s) */
 211         if ((count & LDSEM_ACTIVE_MASK) == 0)
 212                 __ldsem_wake(sem);
 213
 214         raw_spin_unlock_irq(&sem->wait_lock);
 215
 216         /* wait to be given the lock */
 217         for (;;) {
 218                 set_current_state(TASK_UNINTERRUPTIBLE);
 219
 220                 if (!waiter.task)
 221                         break;
 222                 if (!timeout)
 223                         break;
 224                 timeout = schedule_timeout(timeout);
 225         }
 226
 227         __set_current_state(TASK_RUNNING);
 228
 229         if (!timeout) {
 230                 /*
 231                  * Lock timed out but check if this task was just
 232                  * granted lock ownership - if so, pretend there
 233                  * was no timeout; otherwise, cleanup lock wait.
 234                  */
 235                 raw_spin_lock_irq(&sem->wait_lock);
 236                 if (waiter.task) {
 237                         atomic_long_add_return(-LDSEM_WAIT_BIAS, &sem->count);
 238                         list_del(&waiter.list);
 239                         raw_spin_unlock_irq(&sem->wait_lock);
 240                         put_task_struct(waiter.task);
 241                         return NULL;
 242                 }
 243                 raw_spin_unlock_irq(&sem->wait_lock);
 244         }
 245
 246         return sem;
 247 }
 248
 249 /*
 250  * wait for the write lock to be granted
 251  */
 252 static struct ld_semaphore __sched *
 253 down_write_failed(struct ld_semaphore *sem, long count, long timeout)
 254 {
 255         struct ldsem_waiter waiter;
 256         long adjust = -LDSEM_ACTIVE_BIAS;
 257         int locked = 0;
 258
 259         /* set up my own style of waitqueue */
 260         raw_spin_lock_irq(&sem->wait_lock);
 261
 262         /*
 263          * Try to reverse the lock attempt but if the count has changed
 264          * so that reversing fails, check if the lock is now owned,
 265          * and early-out if so.
 266          */
 267         do {
 268                 if (atomic_long_try_cmpxchg(&sem->count, &count, count + adjust))
 269                         break;
 270                 if ((count & LDSEM_ACTIVE_MASK) == LDSEM_ACTIVE_BIAS) {
 271                         raw_spin_unlock_irq(&sem->wait_lock);
 272                         return sem;
 273                 }
 274         } while (1);
 275
 276         list_add_tail(&waiter.list, &sem->write_wait);
 277
 278         waiter.task = current;
 279
 280         set_current_state(TASK_UNINTERRUPTIBLE);
 281         for (;;) {
 282                 if (!timeout)
 283                         break;
 284                 raw_spin_unlock_irq(&sem->wait_lock);
 285                 timeout = schedule_timeout(timeout);
 286                 raw_spin_lock_irq(&sem->wait_lock);
 287                 set_current_state(TASK_UNINTERRUPTIBLE);
 288                 locked = writer_trylock(sem);
 289                 if (locked)
 290                         break;
 291         }
 292
 293         if (!locked)
 294                 atomic_long_add_return(-LDSEM_WAIT_BIAS, &sem->count);
 295         list_del(&waiter.list);
 296
 297         /*
 298          * In case of timeout, wake up every reader who gave the right of way
 299          * to writer. Prevent separation readers into two groups:
 300          * one that helds semaphore and another that sleeps.
 301          * (in case of no contention with a writer)
 302          */
 303         if (!locked && list_empty(&sem->write_wait))
 304                 __ldsem_wake_readers(sem);
 305
 306         raw_spin_unlock_irq(&sem->wait_lock);
 307
 308         __set_current_state(TASK_RUNNING);
 309
 310         /* lock wait may have timed out */
 311         if (!locked)
 312                 return NULL;
 313         return sem;
 314 }
 315
 316
 317
 318 static int __ldsem_down_read_nested(struct ld_semaphore *sem,
 319                                            int subclass, long timeout)
 320 {
 321         long count;
 322
 323         lockdep_acquire_read(sem, subclass, 0, _RET_IP_);
 324
 325         count = atomic_long_add_return(LDSEM_READ_BIAS, &sem->count);
 326         if (count <= 0) {
 327                 lock_stat(sem, contended);
 328                 if (!down_read_failed(sem, count, timeout)) {
 329                         lockdep_release(sem, 1, _RET_IP_);
 330                         return 0;
 331                 }
 332         }
 333         lock_stat(sem, acquired);
 334         return 1;
 335 }
 336
 337 static int __ldsem_down_write_nested(struct ld_semaphore *sem,
 338                                             int subclass, long timeout)
 339 {
 340         long count;
 341
 342         lockdep_acquire(sem, subclass, 0, _RET_IP_);
 343
 344         count = atomic_long_add_return(LDSEM_WRITE_BIAS, &sem->count);
 345         if ((count & LDSEM_ACTIVE_MASK) != LDSEM_ACTIVE_BIAS) {
 346                 lock_stat(sem, contended);
 347                 if (!down_write_failed(sem, count, timeout)) {
 348                         lockdep_release(sem, 1, _RET_IP_);
 349                         return 0;
 350                 }
 351         }
 352         lock_stat(sem, acquired);
 353         return 1;
 354 }
 355
 356
 357 /*
 358  * lock for reading -- returns 1 if successful, 0 if timed out
 359  */
 360 int __sched ldsem_down_read(struct ld_semaphore *sem, long timeout)
 361 {
 362         might_sleep();
 363         return __ldsem_down_read_nested(sem, 0, timeout);
 364 }
 365
 366 /*
 367  * trylock for reading -- returns 1 if successful, 0 if contention
 368  */
 369 int ldsem_down_read_trylock(struct ld_semaphore *sem)
 370 {
 371         long count = atomic_long_read(&sem->count);
 372
 373         while (count >= 0) {
 374                 if (atomic_long_try_cmpxchg(&sem->count, &count, count + LDSEM_READ_BIAS)) {
 375                         lockdep_acquire_read(sem, 0, 1, _RET_IP_);
 376                         lock_stat(sem, acquired);
 377                         return 1;
 378                 }
 379         }
 380         return 0;
 381 }
 382
 383 /*
 384  * lock for writing -- returns 1 if successful, 0 if timed out
 385  */
 386 int __sched ldsem_down_write(struct ld_semaphore *sem, long timeout)
 387 {
 388         might_sleep();
 389         return __ldsem_down_write_nested(sem, 0, timeout);
 390 }
 391
 392 /*
 393  * trylock for writing -- returns 1 if successful, 0 if contention
 394  */
 395 int ldsem_down_write_trylock(struct ld_semaphore *sem)
 396 {
 397         long count = atomic_long_read(&sem->count);
 398
 399         while ((count & LDSEM_ACTIVE_MASK) == 0) {
 400                 if (atomic_long_try_cmpxchg(&sem->count, &count, count + LDSEM_WRITE_BIAS)) {
 401                         lockdep_acquire(sem, 0, 1, _RET_IP_);
 402                         lock_stat(sem, acquired);
 403                         return 1;
 404                 }
 405         }
 406         return 0;
 407 }
 408
 409 /*
 410  * release a read lock
 411  */
 412 void ldsem_up_read(struct ld_semaphore *sem)
 413 {
 414         long count;
 415
 416         lockdep_release(sem, 1, _RET_IP_);
 417
 418         count = atomic_long_add_return(-LDSEM_READ_BIAS, &sem->count);
 419         if (count < 0 && (count & LDSEM_ACTIVE_MASK) == 0)
 420                 ldsem_wake(sem);
 421 }
 422
 423 /*
 424  * release a write lock
 425  */
 426 void ldsem_up_write(struct ld_semaphore *sem)
 427 {
 428         long count;
 429
 430         lockdep_release(sem, 1, _RET_IP_);
 431
 432         count = atomic_long_add_return(-LDSEM_WRITE_BIAS, &sem->count);
 433         if (count < 0)
 434                 ldsem_wake(sem);
 435 }
 436
 437
 438 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 439
 440 int ldsem_down_read_nested(struct ld_semaphore *sem, int subclass, long timeout)
 441 {
 442         might_sleep();
 443         return __ldsem_down_read_nested(sem, subclass, timeout);
 444 }
 445
 446 int ldsem_down_write_nested(struct ld_semaphore *sem, int subclass,
 447                             long timeout)
 448 {
 449         might_sleep();
 450         return __ldsem_down_write_nested(sem, subclass, timeout);
 451 }
 452
 453 #endif