/*
* Copyright (c) 2000 by Hewlett-Packard Company.  All rights reserved.
*
* THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
* OR IMPLIED.  ANY USE IS AT YOUR OWN RISK.
*
* Permission is hereby granted to use or copy this program
* for any purpose,  provided the above notices are retained on all copies.
* Permission to modify the code and to distribute modified code is granted,
* provided the above notices are retained, and a notice that the code was
* modified is included with the above copyright notice.
*/

#include "private/thread_local_alloc.h"
               /* To determine type of tsd impl.       */
               /* Includes private/specific.h          */
               /* if needed.                           */

#if defined(USE_CUSTOM_SPECIFIC)

static const tse invalid_tse = {INVALID_QTID, 0, 0, INVALID_THREADID};
           /* A thread-specific data entry which will never    */
           /* appear valid to a reader.  Used to fill in empty */
           /* cache entries to avoid a check for 0.            */

GC_INNER int GC_key_create_inner(tsd ** key_ptr)
{
   int i;
   int ret;
   tsd * result;

   GC_ASSERT(I_HOLD_LOCK());
   /* A quick alignment check, since we need atomic stores */
   GC_ASSERT((word)(&invalid_tse.next) % sizeof(tse *) == 0);
   result = (tsd *)MALLOC_CLEAR(sizeof(tsd));
   if (NULL == result) return ENOMEM;
   ret = pthread_mutex_init(&result->lock, NULL);
   if (ret != 0) return ret;
   for (i = 0; i < TS_CACHE_SIZE; ++i) {
     result -> cache[i] = (/* no const */ tse *)&invalid_tse;
   }
#   ifdef GC_ASSERTIONS
     for (i = 0; i < TS_HASH_SIZE; ++i) {
       GC_ASSERT(result -> hash[i].p == 0);
     }
#   endif
   *key_ptr = result;
   return 0;
}

GC_INNER int GC_setspecific(tsd * key, void * value)
{
   pthread_t self = pthread_self();
   unsigned hash_val = HASH(self);
   volatile tse * entry;

   GC_ASSERT(I_HOLD_LOCK());
   GC_ASSERT(self != INVALID_THREADID);
   GC_dont_gc++; /* disable GC */
   entry = (volatile tse *)MALLOC_CLEAR(sizeof(tse));
   GC_dont_gc--;
   if (0 == entry) return ENOMEM;

   pthread_mutex_lock(&(key -> lock));
   /* Could easily check for an existing entry here.   */
   entry -> next = key->hash[hash_val].p;
   entry -> thread = self;
   entry -> value = TS_HIDE_VALUE(value);
   GC_ASSERT(entry -> qtid == INVALID_QTID);
   /* There can only be one writer at a time, but this needs to be     */
   /* atomic with respect to concurrent readers.                       */
   AO_store_release(&key->hash[hash_val].ao, (AO_t)entry);
   GC_dirty((/* no volatile */ void *)entry);
   GC_dirty(key->hash + hash_val);
   if (pthread_mutex_unlock(&key->lock) != 0)
     ABORT("pthread_mutex_unlock failed (setspecific)");
   return 0;
}

/* Remove thread-specific data for a given thread.  This function is    */
/* called at fork from the child process for all threads except for the */
/* survived one.  GC_remove_specific() should be called on thread exit. */
GC_INNER void GC_remove_specific_after_fork(tsd * key, pthread_t t)
{
   unsigned hash_val = HASH(t);
   tse *entry;
   tse *prev = NULL;

#   ifdef CAN_HANDLE_FORK
     /* Both GC_setspecific and GC_remove_specific should be called    */
     /* with the allocation lock held to ensure the consistency of     */
     /* the hash table in the forked child.                            */
     GC_ASSERT(I_HOLD_LOCK());
#   endif
   pthread_mutex_lock(&(key -> lock));
   entry = key->hash[hash_val].p;
   while (entry != NULL && !THREAD_EQUAL(entry->thread, t)) {
     prev = entry;
     entry = entry->next;
   }
   /* Invalidate qtid field, since qtids may be reused, and a later    */
   /* cache lookup could otherwise find this entry.                    */
   if (entry != NULL) {
     entry -> qtid = INVALID_QTID;
     if (NULL == prev) {
       key->hash[hash_val].p = entry->next;
       GC_dirty(key->hash + hash_val);
     } else {
       prev->next = entry->next;
       GC_dirty(prev);
     }
     /* Atomic! concurrent accesses still work.        */
     /* They must, since readers don't lock.           */
     /* We shouldn't need a volatile access here,      */
     /* since both this and the preceding write        */
     /* should become visible no later than            */
     /* the pthread_mutex_unlock() call.               */
   }
   /* If we wanted to deallocate the entry, we'd first have to clear   */
   /* any cache entries pointing to it.  That probably requires        */
   /* additional synchronization, since we can't prevent a concurrent  */
   /* cache lookup, which should still be examining deallocated memory.*/
   /* This can only happen if the concurrent access is from another    */
   /* thread, and hence has missed the cache, but still...             */
#   ifdef LINT2
     GC_noop1((word)entry);
#   endif

   /* With GC, we're done, since the pointers from the cache will      */
   /* be overwritten, all local pointers to the entries will be        */
   /* dropped, and the entry will then be reclaimed.                   */
   if (pthread_mutex_unlock(&key->lock) != 0)
     ABORT("pthread_mutex_unlock failed (remove_specific after fork)");
}

/* Note that even the slow path doesn't lock.   */
GC_INNER void * GC_slow_getspecific(tsd * key, word qtid,
                                   tse * volatile * cache_ptr)
{
   pthread_t self = pthread_self();
   tse *entry = key->hash[HASH(self)].p;

   GC_ASSERT(qtid != INVALID_QTID);
   while (entry != NULL && !THREAD_EQUAL(entry->thread, self)) {
     entry = entry -> next;
   }
   if (entry == NULL) return NULL;
   /* Set cache_entry. */
   AO_store(&(entry -> qtid), qtid);
       /* It's safe to do this asynchronously.  Either value   */
       /* is safe, though may produce spurious misses.         */
       /* We're replacing one qtid with another one for the    */
       /* same thread.                                         */
   AO_store((volatile AO_t *)cache_ptr, (AO_t)entry);
   return TS_REVEAL_PTR(entry -> value);
}

#ifdef GC_ASSERTIONS
 /* Check that that all elements of the data structure associated  */
 /* with key are marked.                                           */
 void GC_check_tsd_marks(tsd *key)
 {
   int i;
   tse *p;

   if (!GC_is_marked(GC_base(key))) {
     ABORT("Unmarked thread-specific-data table");
   }
   for (i = 0; i < TS_HASH_SIZE; ++i) {
     for (p = key->hash[i].p; p != 0; p = p -> next) {
       if (!GC_is_marked(GC_base(p))) {
         ABORT_ARG1("Unmarked thread-specific-data entry",
                    " at %p", (void *)p);
       }
     }
   }
   for (i = 0; i < TS_CACHE_SIZE; ++i) {
     p = key -> cache[i];
     if (p != &invalid_tse && !GC_is_marked(GC_base(p))) {
       ABORT_ARG1("Unmarked cached thread-specific-data entry",
                  " at %p", (void *)p);
     }
   }
 }
#endif /* GC_ASSERTIONS */

#endif /* USE_CUSTOM_SPECIFIC */