forked from apache/brpc
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathkey.cpp
473 lines (428 loc) · 15 KB
/
key.cpp
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
// bthread - A M:N threading library to make applications more concurrent.
// Copyright (c) 2014 Baidu.com, Inc. All Rights Reserved
// Author: Ge,Jun (gejun@baidu.com)
// Date: Sun Aug 3 12:46:15 CST 2014
#include <pthread.h>
#include "base/macros.h"
#include "base/atomicops.h"
#include "bvar/passive_status.h"
#include "bthread/errno.h" // EAGAIN
#include "bthread/task_group.h" // TaskGroup
// Implement bthread_key_t related functions
namespace bthread {
class KeyTable;
// defined in task_group.cpp
extern __thread TaskGroup* tls_task_group;
extern __thread LocalStorage tls_bls;
static __thread bool tls_ever_created_keytable = false;
// We keep thread specific data in a two-level array. The top-level array
// contains at most KEY_1STLEVEL_SIZE pointers to dynamically allocated
// arrays of at most KEY_2NDLEVEL_SIZE data pointers. Many applications
// may just occupy one or two second level array, thus this machanism keeps
// memory footprint smaller and we can change KEY_1STLEVEL_SIZE to a
// bigger number more freely. The tradeoff is an additional memory indirection:
// negligible at most time.
static const uint32_t KEY_2NDLEVEL_SIZE = 32;
// Notice that we're trying to make the memory of second level and first
// level both 256 bytes to make memory allocator happier.
static const uint32_t KEY_1STLEVEL_SIZE = 31;
// Max tls in one thread, currently the value is 992 which should be enough
// for most projects throughout baidu.
static const uint32_t KEYS_MAX = KEY_2NDLEVEL_SIZE * KEY_1STLEVEL_SIZE;
// destructors/version of TLS.
struct KeyInfo {
uint32_t version;
void (*dtor)(void*, const void*);
const void* dtor_args;
};
static KeyInfo s_key_info[KEYS_MAX] = {};
// For allocating keys.
static pthread_mutex_t s_key_mutex = PTHREAD_MUTEX_INITIALIZER;
static size_t nfreekey = 0;
static size_t nkey = 0;
static uint32_t s_free_keys[KEYS_MAX];
// Stats.
static base::static_atomic<size_t> nkeytable = BASE_STATIC_ATOMIC_INIT(0);
static base::static_atomic<size_t> nsubkeytable = BASE_STATIC_ATOMIC_INIT(0);
// The second-level array.
// Align with cacheline to avoid false sharing.
class BAIDU_CACHELINE_ALIGNMENT SubKeyTable {
public:
SubKeyTable() {
memset(_data, 0, sizeof(_data));
nsubkeytable.fetch_add(1, base::memory_order_relaxed);
}
// NOTE: Call clear first.
~SubKeyTable() {
nsubkeytable.fetch_sub(1, base::memory_order_relaxed);
}
void clear(uint32_t offset) {
for (uint32_t i = 0; i < KEY_2NDLEVEL_SIZE; ++i) {
void* p = _data[i].ptr;
if (p) {
// Set the position to NULL before calling dtor which may set
// the position again.
_data[i].ptr = NULL;
KeyInfo info = bthread::s_key_info[offset + i];
if (info.dtor && _data[i].version == info.version) {
info.dtor(p, info.dtor_args);
}
}
}
}
bool cleared() const {
// We need to iterate again to check if every slot is empty. An
// alternative is remember if set_data() was called during clear.
for (uint32_t i = 0; i < KEY_2NDLEVEL_SIZE; ++i) {
if (_data[i].ptr) {
return false;
}
}
return true;
}
inline void* get_data(uint32_t index, uint32_t version) const {
if (_data[index].version == version) {
return _data[index].ptr;
}
return NULL;
}
inline void set_data(uint32_t index, uint32_t version, void* data) {
_data[index].version = version;
_data[index].ptr = data;
}
private:
struct Data {
uint32_t version;
void* ptr;
};
Data _data[KEY_2NDLEVEL_SIZE];
};
// The first-level array.
// Align with cacheline to avoid false sharing.
class BAIDU_CACHELINE_ALIGNMENT KeyTable {
public:
KeyTable() : next(NULL) {
memset(_subs, 0, sizeof(_subs));
nkeytable.fetch_add(1, base::memory_order_relaxed);
}
~KeyTable() {
nkeytable.fetch_sub(1, base::memory_order_relaxed);
for (int ntry = 0; ntry < PTHREAD_DESTRUCTOR_ITERATIONS; ++ntry) {
for (uint32_t i = 0; i < KEY_1STLEVEL_SIZE; ++i) {
if (_subs[i]) {
_subs[i]->clear(i * KEY_2NDLEVEL_SIZE);
}
}
bool all_cleared = true;
for (uint32_t i = 0; i < KEY_1STLEVEL_SIZE; ++i) {
if (_subs[i] != NULL && !_subs[i]->cleared()) {
all_cleared = false;
break;
}
}
if (all_cleared) {
for (uint32_t i = 0; i < KEY_1STLEVEL_SIZE; ++i) {
delete _subs[i];
}
return;
}
}
LOG(ERROR) << "Fail to destroy all objects in KeyTable[" << this << ']';
}
inline void* get_data(bthread_key_t key) const {
const uint32_t subidx = key.index / KEY_2NDLEVEL_SIZE;
if (subidx < KEY_1STLEVEL_SIZE) {
const SubKeyTable* sub_kt = _subs[subidx];
if (sub_kt) {
return sub_kt->get_data(
key.index - subidx * KEY_2NDLEVEL_SIZE, key.version);
}
}
return NULL;
}
inline int set_data(bthread_key_t key, void* data) {
const uint32_t subidx = key.index / KEY_2NDLEVEL_SIZE;
if (subidx < KEY_1STLEVEL_SIZE &&
key.version == s_key_info[key.index].version) {
SubKeyTable* sub_kt = _subs[subidx];
if (sub_kt == NULL) {
sub_kt = new (std::nothrow) SubKeyTable;
if (NULL == sub_kt) {
return ENOMEM;
}
_subs[subidx] = sub_kt;
}
sub_kt->set_data(key.index - subidx * KEY_2NDLEVEL_SIZE,
key.version, data);
return 0;
}
CHECK(false) << "bthread_setspecific is called on invalid " << key;
return EINVAL;
}
public:
KeyTable* next;
private:
SubKeyTable* _subs[KEY_1STLEVEL_SIZE];
};
static KeyTable* borrow_keytable(bthread_keytable_pool_t* pool) {
if (pool != NULL && pool->free_keytables) {
BAIDU_SCOPED_LOCK(pool->mutex);
KeyTable* p = (KeyTable*)pool->free_keytables;
if (p) {
pool->free_keytables = p->next;
return p;
}
}
return NULL;
}
// Referenced in task_group.cpp, must be extern.
// Caller of this function must hold the KeyTable
void return_keytable(bthread_keytable_pool_t* pool, KeyTable* kt) {
if (NULL == kt) {
return;
}
if (pool == NULL) {
delete kt;
return;
}
std::unique_lock<pthread_mutex_t> mu(pool->mutex);
if (pool->destroyed) {
mu.unlock();
delete kt;
return;
}
kt->next = (KeyTable*)pool->free_keytables;
pool->free_keytables = kt;
}
static void cleanup_pthread(void*) {
KeyTable* kt = tls_bls.keytable;
if (kt) {
delete kt;
// After deletion: tls may be set during deletion.
tls_bls.keytable = NULL;
}
}
static void arg_as_dtor(void* data, const void* arg) {
typedef void (*KeyDtor)(void*);
return ((KeyDtor)arg)(data);
}
static int get_key_count(void*) {
BAIDU_SCOPED_LOCK(bthread::s_key_mutex);
return (int)nkey - (int)nfreekey;
}
static size_t get_keytable_count(void*) {
return nkeytable.load(base::memory_order_relaxed);
}
static size_t get_keytable_memory(void*) {
const size_t n = nkeytable.load(base::memory_order_relaxed);
const size_t nsub = nsubkeytable.load(base::memory_order_relaxed);
return n * sizeof(KeyTable) + nsub * sizeof(SubKeyTable);
}
static bvar::PassiveStatus<int> s_bthread_key_count(
"bthread_key_count", get_key_count, NULL);
static bvar::PassiveStatus<size_t> s_bthread_keytable_count(
"bthread_keytable_count", get_keytable_count, NULL);
static bvar::PassiveStatus<size_t> s_bthread_keytable_memory(
"bthread_keytable_memory", get_keytable_memory, NULL);
} // namespace bthread
extern "C" {
int bthread_keytable_pool_init(bthread_keytable_pool_t* pool) {
if (pool == NULL) {
LOG(ERROR) << "Param[pool] is NULL";
return EINVAL;
}
pthread_mutex_init(&pool->mutex, NULL);
pool->free_keytables = NULL;
pool->destroyed = 0;
return 0;
}
int bthread_keytable_pool_destroy(bthread_keytable_pool_t* pool) {
if (pool == NULL) {
LOG(ERROR) << "Param[pool] is NULL";
return EINVAL;
}
bthread::KeyTable* saved_free_keytables = NULL;
{
BAIDU_SCOPED_LOCK(pool->mutex);
if (pool->free_keytables) {
saved_free_keytables = (bthread::KeyTable*)pool->free_keytables;
pool->free_keytables = NULL;
}
pool->destroyed = 1;
}
// Cheat get/setspecific and destroy the keytables.
bthread::TaskGroup* const g = bthread::tls_task_group;
bthread::KeyTable* old_kt = bthread::tls_bls.keytable;
while (saved_free_keytables) {
bthread::KeyTable* kt = saved_free_keytables;
saved_free_keytables = kt->next;
bthread::tls_bls.keytable = kt;
if (g) {
g->current_task()->local_storage.keytable = kt;
}
delete kt;
if (old_kt == kt) {
old_kt = NULL;
}
}
bthread::tls_bls.keytable = old_kt;
if (g) {
g->current_task()->local_storage.keytable = old_kt;
}
// TODO: return_keytable may race with this function, we don't destroy
// the mutex right now.
// pthread_mutex_destroy(&pool->mutex);
return 0;
}
int bthread_keytable_pool_getstat(bthread_keytable_pool_t* pool,
bthread_keytable_pool_stat_t* stat) {
if (pool == NULL || stat == NULL) {
LOG(ERROR) << "Param[pool] or Param[stat] is NULL";
return EINVAL;
}
std::unique_lock<pthread_mutex_t> mu(pool->mutex);
size_t count = 0;
bthread::KeyTable* p = (bthread::KeyTable*)pool->free_keytables;
for (; p; p = p->next, ++count) {}
stat->nfree = count;
return 0;
}
// TODO: this is not strict `reserve' because we only check #free.
// Currently there's no way to track KeyTables that may be returned
// to the pool in future.
void bthread_keytable_pool_reserve(bthread_keytable_pool_t* pool,
size_t nfree,
bthread_key_t key,
void* ctor(const void*),
const void* ctor_args) {
if (pool == NULL) {
LOG(ERROR) << "Param[pool] is NULL";
return;
}
bthread_keytable_pool_stat_t stat;
if (bthread_keytable_pool_getstat(pool, &stat) != 0) {
LOG(ERROR) << "Fail to getstat of pool=" << pool;
return;
}
for (size_t i = stat.nfree; i < nfree; ++i) {
bthread::KeyTable* kt = new (std::nothrow) bthread::KeyTable;
if (kt == NULL) {
break;
}
void* data = ctor(ctor_args);
if (data) {
kt->set_data(key, data);
} // else append kt w/o data.
std::unique_lock<pthread_mutex_t> mu(pool->mutex);
if (pool->destroyed) {
mu.unlock();
delete kt;
break;
}
kt->next = (bthread::KeyTable*)pool->free_keytables;
pool->free_keytables = kt;
if (data == NULL) {
break;
}
}
}
int bthread_key_create2(bthread_key_t* key,
void (*dtor)(void*, const void*),
const void* dtor_args) __THROW {
uint32_t index = 0;
{
BAIDU_SCOPED_LOCK(bthread::s_key_mutex);
if (bthread::nfreekey > 0) {
index = bthread::s_free_keys[--bthread::nfreekey];
} else if (bthread::nkey < bthread::KEYS_MAX) {
index = bthread::nkey++;
} else {
return EAGAIN; // what pthread_key_create returns in this case.
}
}
bthread::s_key_info[index].dtor = dtor;
bthread::s_key_info[index].dtor_args = dtor_args;
key->index = index;
key->version = bthread::s_key_info[index].version;
if (key->version == 0) {
++bthread::s_key_info[index].version;
++key->version;
}
return 0;
}
int bthread_key_create(bthread_key_t* key, void (*dtor)(void*)) __THROW {
if (dtor == NULL) {
return bthread_key_create2(key, NULL, NULL);
} else {
return bthread_key_create2(key, bthread::arg_as_dtor, (const void*)dtor);
}
}
int bthread_key_delete(bthread_key_t key) __THROW {
if (key.index < bthread::KEYS_MAX &&
key.version == bthread::s_key_info[key.index].version) {
BAIDU_SCOPED_LOCK(bthread::s_key_mutex);
if (key.version == bthread::s_key_info[key.index].version) {
if (++bthread::s_key_info[key.index].version == 0) {
++bthread::s_key_info[key.index].version;
}
bthread::s_key_info[key.index].dtor = NULL;
bthread::s_key_info[key.index].dtor_args = NULL;
bthread::s_free_keys[bthread::nfreekey++] = key.index;
return 0;
}
}
CHECK(false) << "bthread_key_delete is called on invalid " << key;
return EINVAL;
}
// NOTE: Can't borrow_keytable in bthread_setspecific, otherwise following
// memory leak may occur:
// -> bthread_getspecific fails to borrow_keytable and returns NULL.
// -> bthread_setspecific succeeds to borrow_keytable and overwrites old data
// at the position with newly created data, the old data is leaked.
int bthread_setspecific(bthread_key_t key, void* data) __THROW {
bthread::KeyTable* kt = bthread::tls_bls.keytable;
if (NULL == kt) {
kt = new (std::nothrow) bthread::KeyTable;
if (NULL == kt) {
return ENOMEM;
}
bthread::tls_bls.keytable = kt;
bthread::TaskGroup* const g = bthread::tls_task_group;
if (g) {
g->current_task()->local_storage.keytable = kt;
}
if (!bthread::tls_ever_created_keytable) {
bthread::tls_ever_created_keytable = true;
CHECK_EQ(0, base::thread_atexit(bthread::cleanup_pthread, NULL));
}
}
return kt->set_data(key, data);
}
void* bthread_getspecific(bthread_key_t key) __THROW {
bthread::KeyTable* kt = bthread::tls_bls.keytable;
if (kt) {
return kt->get_data(key);
}
bthread::TaskGroup* const g = bthread::tls_task_group;
if (g) {
bthread::TaskMeta* const task = g->current_task();
kt = bthread::borrow_keytable(task->attr.keytable_pool);
if (kt) {
g->current_task()->local_storage.keytable = kt;
bthread::tls_bls.keytable = kt;
return kt->get_data(key);
}
}
return NULL;
}
void bthread_assign_data(void* data) __THROW {
bthread::tls_bls.assigned_data = data;
bthread::TaskGroup* const g = bthread::tls_task_group;
if (g) {
g->current_task()->local_storage.assigned_data = data;
}
}
void* bthread_get_assigned_data() __THROW {
return bthread::tls_bls.assigned_data;
}
} // extern "C"