lib/rwsem.c

   1 /* rwsem.c: R/W semaphores: contention handling functions
   2  *
   3  * Written by David Howells (dhowells@redhat.com).
   4  * Derived from arch/i386/kernel/semaphore.c
   5  */
   6 #include <linux/rwsem.h>
   7 #include <linux/sched.h>
   8 #include <linux/init.h>
   9 #include <linux/module.h>
  10
  11 struct rwsem_waiter {
  12         struct list_head        list;
  13         struct task_struct      *task;
  14         unsigned int            flags;
  15 #define RWSEM_WAITING_FOR_READ  0x00000001
  16 #define RWSEM_WAITING_FOR_WRITE 0x00000002
  17 };
  18
  19 #if RWSEM_DEBUG
  20 #undef rwsemtrace
  21 void rwsemtrace(struct rw_semaphore *sem, const char *str)
  22 {
  23         printk("sem=%p\n",sem);
  24         printk("(sem)=%08lx\n",sem->count);
  25         if (sem->debug)
  26                 printk("[%d] %s({%08lx})\n",current->pid,str,sem->count);
  27 }
  28 #endif
  29
  30 /*
  31  * handle the lock being released whilst there are processes blocked on it that can now run
  32  * - if we come here, then:
  33  *   - the 'active part' of the count (&0x0000ffff) reached zero but has been re-incremented
  34  *   - the 'waiting part' of the count (&0xffff0000) is negative (and will still be so)
  35  *   - there must be someone on the queue
  36  * - the spinlock must be held by the caller
  37  * - woken process blocks are discarded from the list after having task zeroed
  38  * - writers are only woken if wakewrite is non-zero
  39  */
  40 static inline struct rw_semaphore *__rwsem_do_wake(struct rw_semaphore *sem, int wakewrite)
  41 {
  42         struct rwsem_waiter *waiter;
  43         struct task_struct *tsk;
  44         struct list_head *next;
  45         signed long oldcount, woken, loop;
  46
  47         rwsemtrace(sem,"Entering __rwsem_do_wake");
  48
  49         if (!wakewrite)
  50                 goto dont_wake_writers;
  51
  52         /* only wake someone up if we can transition the active part of the count from 0 -> 1 */
  53  try_again:
  54         oldcount = rwsem_atomic_update(RWSEM_ACTIVE_BIAS,sem) - RWSEM_ACTIVE_BIAS;
  55         if (oldcount & RWSEM_ACTIVE_MASK)
  56                 goto undo;
  57
  58         waiter = list_entry(sem->wait_list.next,struct rwsem_waiter,list);
  59
  60         /* try to grant a single write lock if there's a writer at the front of the queue
  61          * - note we leave the 'active part' of the count incremented by 1 and the waiting part
  62          *   incremented by 0x00010000
  63          */
  64         if (!(waiter->flags & RWSEM_WAITING_FOR_WRITE))
  65                 goto readers_only;
  66
  67         list_del(&waiter->list);
  68         tsk = waiter->task;
  69         mb();
  70         waiter->task = NULL;
  71         wake_up_process(tsk);
  72         put_task_struct(tsk);
  73         goto out;
  74
  75         /* don't want to wake any writers */
  76  dont_wake_writers:
  77         waiter = list_entry(sem->wait_list.next,struct rwsem_waiter,list);
  78         if (waiter->flags & RWSEM_WAITING_FOR_WRITE)
  79                 goto out;
  80
  81         /* grant an infinite number of read locks to the readers at the front of the queue
  82          * - note we increment the 'active part' of the count by the number of readers (less one
  83          *   for the activity decrement we've already done) before waking any processes up
  84          */
  85  readers_only:
  86         woken = 0;
  87         do {
  88                 woken++;
  89
  90                 if (waiter->list.next==&sem->wait_list)
  91                         break;
  92
  93                 waiter = list_entry(waiter->list.next,struct rwsem_waiter,list);
  94
  95         } while (waiter->flags & RWSEM_WAITING_FOR_READ);
  96
  97         loop = woken;
  98         woken *= RWSEM_ACTIVE_BIAS-RWSEM_WAITING_BIAS;
  99         woken -= RWSEM_ACTIVE_BIAS;
 100         rwsem_atomic_add(woken,sem);
 101
 102         next = sem->wait_list.next;
 103         for (; loop>0; loop--) {
 104                 waiter = list_entry(next,struct rwsem_waiter,list);
 105                 next = waiter->list.next;
 106                 tsk = waiter->task;
 107                 mb();
 108                 waiter->task = NULL;
 109                 wake_up_process(tsk);
 110                 put_task_struct(tsk);
 111         }
 112
 113         sem->wait_list.next = next;
 114         next->prev = &sem->wait_list;
 115
 116  out:
 117         rwsemtrace(sem,"Leaving __rwsem_do_wake");
 118         return sem;
 119
 120         /* undo the change to count, but check for a transition 1->0 */
 121  undo:
 122         if (rwsem_atomic_update(-RWSEM_ACTIVE_BIAS,sem)!=0)
 123                 goto out;
 124         goto try_again;
 125 }
 126
 127 /*
 128  * wait for a lock to be granted
 129  */
 130 static inline struct rw_semaphore *rwsem_down_failed_common(struct rw_semaphore *sem,
 131                                                                  struct rwsem_waiter *waiter,
 132                                                                  signed long adjustment)
 133 {
 134         struct task_struct *tsk = current;
 135         signed long count;
 136
 137         set_task_state(tsk,TASK_UNINTERRUPTIBLE);
 138
 139         /* set up my own style of waitqueue */
 140         spin_lock(&sem->wait_lock);
 141         waiter->task = tsk;
 142         get_task_struct(tsk);
 143
 144         list_add_tail(&waiter->list,&sem->wait_list);
 145
 146         /* note that we're now waiting on the lock, but no longer actively read-locking */
 147         count = rwsem_atomic_update(adjustment,sem);
 148
 149         /* if there are no longer active locks, wake the front queued process(es) up
 150          * - it might even be this process, since the waker takes a more active part
 151          */
 152         if (!(count & RWSEM_ACTIVE_MASK))
 153                 sem = __rwsem_do_wake(sem,1);
 154
 155         spin_unlock(&sem->wait_lock);
 156
 157         /* wait to be given the lock */
 158         for (;;) {
 159                 if (!waiter->task)
 160                         break;
 161                 schedule();
 162                 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
 163         }
 164
 165         tsk->state = TASK_RUNNING;
 166
 167         return sem;
 168 }
 169
 170 /*
 171  * wait for the read lock to be granted
 172  */
 173 struct rw_semaphore fastcall __sched *rwsem_down_read_failed(struct rw_semaphore *sem)
 174 {
 175         struct rwsem_waiter waiter;
 176
 177         rwsemtrace(sem,"Entering rwsem_down_read_failed");
 178
 179         waiter.flags = RWSEM_WAITING_FOR_READ;
 180         rwsem_down_failed_common(sem,&waiter,RWSEM_WAITING_BIAS-RWSEM_ACTIVE_BIAS);
 181
 182         rwsemtrace(sem,"Leaving rwsem_down_read_failed");
 183         return sem;
 184 }
 185
 186 /*
 187  * wait for the write lock to be granted
 188  */
 189 struct rw_semaphore fastcall __sched *rwsem_down_write_failed(struct rw_semaphore *sem)
 190 {
 191         struct rwsem_waiter waiter;
 192
 193         rwsemtrace(sem,"Entering rwsem_down_write_failed");
 194
 195         waiter.flags = RWSEM_WAITING_FOR_WRITE;
 196         rwsem_down_failed_common(sem,&waiter,-RWSEM_ACTIVE_BIAS);
 197
 198         rwsemtrace(sem,"Leaving rwsem_down_write_failed");
 199         return sem;
 200 }
 201
 202 /*
 203  * handle waking up a waiter on the semaphore
 204  * - up_read has decremented the active part of the count if we come here
 205  */
 206 struct rw_semaphore fastcall *rwsem_wake(struct rw_semaphore *sem)
 207 {
 208         rwsemtrace(sem,"Entering rwsem_wake");
 209
 210         spin_lock(&sem->wait_lock);
 211
 212         /* do nothing if list empty */
 213         if (!list_empty(&sem->wait_list))
 214                 sem = __rwsem_do_wake(sem,1);
 215
 216         spin_unlock(&sem->wait_lock);
 217
 218         rwsemtrace(sem,"Leaving rwsem_wake");
 219
 220         return sem;
 221 }
 222
 223 /*
 224  * downgrade a write lock into a read lock
 225  * - caller incremented waiting part of count, and discovered it to be still negative
 226  * - just wake up any readers at the front of the queue
 227  */
 228 struct rw_semaphore fastcall *rwsem_downgrade_wake(struct rw_semaphore *sem)
 229 {
 230         rwsemtrace(sem,"Entering rwsem_downgrade_wake");
 231
 232         spin_lock(&sem->wait_lock);
 233
 234         /* do nothing if list empty */
 235         if (!list_empty(&sem->wait_list))
 236                 sem = __rwsem_do_wake(sem,0);
 237
 238         spin_unlock(&sem->wait_lock);
 239
 240         rwsemtrace(sem,"Leaving rwsem_downgrade_wake");
 241         return sem;
 242 }
 243
 244 EXPORT_SYMBOL_NOVERS(rwsem_down_read_failed);
 245 EXPORT_SYMBOL_NOVERS(rwsem_down_write_failed);
 246 EXPORT_SYMBOL_NOVERS(rwsem_wake);
 247 EXPORT_SYMBOL_NOVERS(rwsem_downgrade_wake);
 248 #if RWSEM_DEBUG
 249 EXPORT_SYMBOL(rwsemtrace);
 250 #endif