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_store_release(&waiter->task, NULL);
 120                 wake_up_process(tsk);
 121                 put_task_struct(tsk);
 122         }
 123         INIT_LIST_HEAD(&sem->read_wait);
 124         sem->wait_readers = 0;
 125 }
 126
 127 static inline int writer_trylock(struct ld_semaphore *sem)
 128 {
 129         /*
 130          * Only wake this writer if the active part of the count can be
 131          * transitioned from 0 -> 1
 132          */
 133         long count = atomic_long_add_return(LDSEM_ACTIVE_BIAS, &sem->count);
 134         do {
 135                 if ((count & LDSEM_ACTIVE_MASK) == LDSEM_ACTIVE_BIAS)
 136                         return 1;
 137                 if (atomic_long_try_cmpxchg(&sem->count, &count, count - LDSEM_ACTIVE_BIAS))
 138                         return 0;
 139         } while (1);
 140 }
 141
 142 static void __ldsem_wake_writer(struct ld_semaphore *sem)
 143 {
 144         struct ldsem_waiter *waiter;
 145
 146         waiter = list_entry(sem->write_wait.next, struct ldsem_waiter, list);
 147         wake_up_process(waiter->task);
 148 }
 149
 150 /*
 151  * handle the lock release when processes blocked on it that can now run
 152  * - if we come here from up_xxxx(), then:
 153  *   - the 'active part' of count (&0x0000ffff) reached 0 (but may have changed)
 154  *   - the 'waiting part' of count (&0xffff0000) is -ve (and will still be so)
 155  * - the spinlock must be held by the caller
 156  * - woken process blocks are discarded from the list after having task zeroed
 157  */
 158 static void __ldsem_wake(struct ld_semaphore *sem)
 159 {
 160         if (!list_empty(&sem->write_wait))
 161                 __ldsem_wake_writer(sem);
 162         else if (!list_empty(&sem->read_wait))
 163                 __ldsem_wake_readers(sem);
 164 }
 165
 166 static void ldsem_wake(struct ld_semaphore *sem)
 167 {
 168         unsigned long flags;
 169
 170         raw_spin_lock_irqsave(&sem->wait_lock, flags);
 171         __ldsem_wake(sem);
 172         raw_spin_unlock_irqrestore(&sem->wait_lock, flags);
 173 }
 174
 175 /*
 176  * wait for the read lock to be granted
 177  */
 178 static struct ld_semaphore __sched *
 179 down_read_failed(struct ld_semaphore *sem, long count, long timeout)
 180 {
 181         struct ldsem_waiter waiter;
 182         long adjust = -LDSEM_ACTIVE_BIAS + LDSEM_WAIT_BIAS;
 183
 184         /* set up my own style of waitqueue */
 185         raw_spin_lock_irq(&sem->wait_lock);
 186
 187         /*
 188          * Try to reverse the lock attempt but if the count has changed
 189          * so that reversing fails, check if there are are no waiters,
 190          * and early-out if not
 191          */
 192         do {
 193                 if (atomic_long_try_cmpxchg(&sem->count, &count, count + adjust)) {
 194                         count += adjust;
 195                         break;
 196                 }
 197                 if (count > 0) {
 198                         raw_spin_unlock_irq(&sem->wait_lock);
 199                         return sem;
 200                 }
 201         } while (1);
 202
 203         list_add_tail(&waiter.list, &sem->read_wait);
 204         sem->wait_readers++;
 205
 206         waiter.task = current;
 207         get_task_struct(current);
 208
 209         /* if there are no active locks, wake the new lock owner(s) */
 210         if ((count & LDSEM_ACTIVE_MASK) == 0)
 211                 __ldsem_wake(sem);
 212
 213         raw_spin_unlock_irq(&sem->wait_lock);
 214
 215         /* wait to be given the lock */
 216         for (;;) {
 217                 set_current_state(TASK_UNINTERRUPTIBLE);
 218
 219                 if (!smp_load_acquire(&waiter.task))
 220                         break;
 221                 if (!timeout)
 222                         break;
 223                 timeout = schedule_timeout(timeout);
 224         }
 225
 226         __set_current_state(TASK_RUNNING);
 227
 228         if (!timeout) {
 229                 /*
 230                  * Lock timed out but check if this task was just
 231                  * granted lock ownership - if so, pretend there
 232                  * was no timeout; otherwise, cleanup lock wait.
 233                  */
 234                 raw_spin_lock_irq(&sem->wait_lock);
 235                 if (waiter.task) {
 236                         atomic_long_add_return(-LDSEM_WAIT_BIAS, &sem->count);
 237                         list_del(&waiter.list);
 238                         raw_spin_unlock_irq(&sem->wait_lock);
 239                         put_task_struct(waiter.task);
 240                         return NULL;
 241                 }
 242                 raw_spin_unlock_irq(&sem->wait_lock);
 243         }
 244
 245         return sem;
 246 }
 247
 248 /*
 249  * wait for the write lock to be granted
 250  */
 251 static struct ld_semaphore __sched *
 252 down_write_failed(struct ld_semaphore *sem, long count, long timeout)
 253 {
 254         struct ldsem_waiter waiter;
 255         long adjust = -LDSEM_ACTIVE_BIAS;
 256         int locked = 0;
 257
 258         /* set up my own style of waitqueue */
 259         raw_spin_lock_irq(&sem->wait_lock);
 260
 261         /*
 262          * Try to reverse the lock attempt but if the count has changed
 263          * so that reversing fails, check if the lock is now owned,
 264          * and early-out if so.
 265          */
 266         do {
 267                 if (atomic_long_try_cmpxchg(&sem->count, &count, count + adjust))
 268                         break;
 269                 if ((count & LDSEM_ACTIVE_MASK) == LDSEM_ACTIVE_BIAS) {
 270                         raw_spin_unlock_irq(&sem->wait_lock);
 271                         return sem;
 272                 }
 273         } while (1);
 274
 275         list_add_tail(&waiter.list, &sem->write_wait);
 276
 277         waiter.task = current;
 278
 279         set_current_state(TASK_UNINTERRUPTIBLE);
 280         for (;;) {
 281                 if (!timeout)
 282                         break;
 283                 raw_spin_unlock_irq(&sem->wait_lock);
 284                 timeout = schedule_timeout(timeout);
 285                 raw_spin_lock_irq(&sem->wait_lock);
 286                 set_current_state(TASK_UNINTERRUPTIBLE);
 287                 locked = writer_trylock(sem);
 288                 if (locked)
 289                         break;
 290         }
 291
 292         if (!locked)
 293                 atomic_long_add_return(-LDSEM_WAIT_BIAS, &sem->count);
 294         list_del(&waiter.list);
 295
 296         /*
 297          * In case of timeout, wake up every reader who gave the right of way
 298          * to writer. Prevent separation readers into two groups:
 299          * one that helds semaphore and another that sleeps.
 300          * (in case of no contention with a writer)
 301          */
 302         if (!locked && list_empty(&sem->write_wait))
 303                 __ldsem_wake_readers(sem);
 304
 305         raw_spin_unlock_irq(&sem->wait_lock);
 306
 307         __set_current_state(TASK_RUNNING);
 308
 309         /* lock wait may have timed out */
 310         if (!locked)
 311                 return NULL;
 312         return sem;
 313 }
 314
 315
 316
 317 static int __ldsem_down_read_nested(struct ld_semaphore *sem,
 318                                            int subclass, long timeout)
 319 {
 320         long count;
 321
 322         lockdep_acquire_read(sem, subclass, 0, _RET_IP_);
 323
 324         count = atomic_long_add_return(LDSEM_READ_BIAS, &sem->count);
 325         if (count <= 0) {
 326                 lock_stat(sem, contended);
 327                 if (!down_read_failed(sem, count, timeout)) {
 328                         lockdep_release(sem, 1, _RET_IP_);
 329                         return 0;
 330                 }
 331         }
 332         lock_stat(sem, acquired);
 333         return 1;
 334 }
 335
 336 static int __ldsem_down_write_nested(struct ld_semaphore *sem,
 337                                             int subclass, long timeout)
 338 {
 339         long count;
 340
 341         lockdep_acquire(sem, subclass, 0, _RET_IP_);
 342
 343         count = atomic_long_add_return(LDSEM_WRITE_BIAS, &sem->count);
 344         if ((count & LDSEM_ACTIVE_MASK) != LDSEM_ACTIVE_BIAS) {
 345                 lock_stat(sem, contended);
 346                 if (!down_write_failed(sem, count, timeout)) {
 347                         lockdep_release(sem, 1, _RET_IP_);
 348                         return 0;
 349                 }
 350         }
 351         lock_stat(sem, acquired);
 352         return 1;
 353 }
 354
 355
 356 /*
 357  * lock for reading -- returns 1 if successful, 0 if timed out
 358  */
 359 int __sched ldsem_down_read(struct ld_semaphore *sem, long timeout)
 360 {
 361         might_sleep();
 362         return __ldsem_down_read_nested(sem, 0, timeout);
 363 }
 364
 365 /*
 366  * trylock for reading -- returns 1 if successful, 0 if contention
 367  */
 368 int ldsem_down_read_trylock(struct ld_semaphore *sem)
 369 {
 370         long count = atomic_long_read(&sem->count);
 371
 372         while (count >= 0) {
 373                 if (atomic_long_try_cmpxchg(&sem->count, &count, count + LDSEM_READ_BIAS)) {
 374                         lockdep_acquire_read(sem, 0, 1, _RET_IP_);
 375                         lock_stat(sem, acquired);
 376                         return 1;
 377                 }
 378         }
 379         return 0;
 380 }
 381
 382 /*
 383  * lock for writing -- returns 1 if successful, 0 if timed out
 384  */
 385 int __sched ldsem_down_write(struct ld_semaphore *sem, long timeout)
 386 {
 387         might_sleep();
 388         return __ldsem_down_write_nested(sem, 0, timeout);
 389 }
 390
 391 /*
 392  * trylock for writing -- returns 1 if successful, 0 if contention
 393  */
 394 int ldsem_down_write_trylock(struct ld_semaphore *sem)
 395 {
 396         long count = atomic_long_read(&sem->count);
 397
 398         while ((count & LDSEM_ACTIVE_MASK) == 0) {
 399                 if (atomic_long_try_cmpxchg(&sem->count, &count, count + LDSEM_WRITE_BIAS)) {
 400                         lockdep_acquire(sem, 0, 1, _RET_IP_);
 401                         lock_stat(sem, acquired);
 402                         return 1;
 403                 }
 404         }
 405         return 0;
 406 }
 407
 408 /*
 409  * release a read lock
 410  */
 411 void ldsem_up_read(struct ld_semaphore *sem)
 412 {
 413         long count;
 414
 415         lockdep_release(sem, 1, _RET_IP_);
 416
 417         count = atomic_long_add_return(-LDSEM_READ_BIAS, &sem->count);
 418         if (count < 0 && (count & LDSEM_ACTIVE_MASK) == 0)
 419                 ldsem_wake(sem);
 420 }
 421
 422 /*
 423  * release a write lock
 424  */
 425 void ldsem_up_write(struct ld_semaphore *sem)
 426 {
 427         long count;
 428
 429         lockdep_release(sem, 1, _RET_IP_);
 430
 431         count = atomic_long_add_return(-LDSEM_WRITE_BIAS, &sem->count);
 432         if (count < 0)
 433                 ldsem_wake(sem);
 434 }
 435
 436
 437 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 438
 439 int ldsem_down_read_nested(struct ld_semaphore *sem, int subclass, long timeout)
 440 {
 441         might_sleep();
 442         return __ldsem_down_read_nested(sem, subclass, timeout);
 443 }
 444
 445 int ldsem_down_write_nested(struct ld_semaphore *sem, int subclass,
 446                             long timeout)
 447 {
 448         might_sleep();
 449         return __ldsem_down_write_nested(sem, subclass, timeout);
 450 }
 451
 452 #endif