2 * This is a reimplementation of a subset of the pthread_getspecific/setspecific
3 * interface. This appears to outperform the standard linuxthreads one
4 * by a significant margin.
5 * The major restriction is that each thread may only make a single
6 * pthread_setspecific call on a single key. (The current data structure
7 * doesn't really require that. The restriction should be easily removable.)
8 * We don't currently support the destruction functions, though that
10 * We also currently assume that only one pthread_setspecific call
11 * can be executed at a time, though that assumption would be easy to remove
17 /* Called during key creation or setspecific. */
18 /* For the GC we already hold lock. */
19 /* Currently allocated objects leak on thread exit. */
20 /* That's hard to fix, but OK if we allocate garbage */
21 /* collected memory. */
22 #define MALLOC_CLEAR(n) GC_INTERNAL_MALLOC(n, NORMAL)
23 #define PREFIXED(name) GC_##name
25 #define TS_CACHE_SIZE 1024
26 #define CACHE_HASH(n) (((((long)n) >> 8) ^ (long)n) & (TS_CACHE_SIZE - 1))
27 #define TS_HASH_SIZE 1024
28 #define HASH(n) (((((long)n) >> 8) ^ (long)n) & (TS_HASH_SIZE - 1))
30 /* An entry describing a thread-specific value for a given thread. */
31 /* All such accessible structures preserve the invariant that if either */
32 /* thread is a valid pthread id or qtid is a valid "quick tread id" */
33 /* for a thread, then value holds the corresponding thread specific */
34 /* value. This invariant must be preserved at ALL times, since */
35 /* asynchronous reads are allowed. */
36 typedef struct thread_specific_entry
{
37 unsigned long qtid
; /* quick thread id, only for cache */
39 struct thread_specific_entry
*next
;
44 /* We represent each thread-specific datum as two tables. The first is */
45 /* a cache, indexed by a "quick thread identifier". The "quick" thread */
46 /* identifier is an easy to compute value, which is guaranteed to */
47 /* determine the thread, though a thread may correspond to more than */
48 /* one value. We typically use the address of a page in the stack. */
49 /* The second is a hash table, indexed by pthread_self(). It is used */
50 /* only as a backup. */
52 /* Return the "quick thread id". Default version. Assumes page size, */
53 /* or at least thread stack separation, is at least 4K. */
54 /* Must be defined so that it never returns 0. (Page 0 can't really */
55 /* be part of any stack, since that would make 0 a valid stack pointer.)*/
56 static __inline__
unsigned long quick_thread_id() {
58 return (unsigned long)(&dummy
) >> 12;
61 #define INVALID_QTID ((unsigned long)0)
62 #define INVALID_THREADID ((pthread_t)0)
64 typedef struct thread_specific_data
{
65 tse
* volatile cache
[TS_CACHE_SIZE
];
66 /* A faster index to the hash table */
67 tse
* hash
[TS_HASH_SIZE
];
71 typedef tsd
* PREFIXED(key_t
);
73 extern int PREFIXED(key_create
) (tsd
** key_ptr
, void (* destructor
)(void *));
75 extern int PREFIXED(setspecific
) (tsd
* key
, void * value
);
77 extern void PREFIXED(remove_specific
) (tsd
* key
);
79 /* An internal version of getspecific that assumes a cache miss. */
80 void * PREFIXED(slow_getspecific
) (tsd
* key
, unsigned long qtid
,
81 tse
* volatile * cache_entry
);
83 static __inline__
void * PREFIXED(getspecific
) (tsd
* key
) {
84 long qtid
= quick_thread_id();
85 unsigned hash_val
= CACHE_HASH(qtid
);
86 tse
* volatile * entry_ptr
= key
-> cache
+ hash_val
;
87 tse
* entry
= *entry_ptr
; /* Must be loaded only once. */
88 if (EXPECT(entry
-> qtid
== qtid
, 1)) {
89 GC_ASSERT(entry
-> thread
== pthread_self());
90 return entry
-> value
;
92 return PREFIXED(slow_getspecific
) (key
, qtid
, entry_ptr
);