{{ message }}
forked from python/cpython
-
Notifications
You must be signed in to change notification settings - Fork 1
Expand file tree
/
Copy pathqsbr.c
More file actions
289 lines (249 loc) · 9.2 KB
/
Copy pathqsbr.c
File metadata and controls
289 lines (249 loc) · 9.2 KB
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
/*
* Implementation of safe memory reclamation scheme using
* quiescent states.
*
* This is derived from the "GUS" safe memory reclamation technique
* in FreeBSD written by Jeffrey Roberson. It is heavily modified. Any bugs
* in this code are likely due to the modifications.
*
* The original copyright is preserved below.
*
* Copyright (c) 2019,2020 Jeffrey Roberson <jeff@FreeBSD.org>
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice unmodified, this list of conditions, and the following
* disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#include "Python.h"
#include "pycore_initconfig.h" // _PyStatus_NO_MEMORY()
#include "pycore_lock.h" // PyMutex_Lock()
#include "pycore_qsbr.h"
#include "pycore_pystate.h" // _PyThreadState_GET()
// Starting size of the array of qsbr thread states
#define MIN_ARRAY_SIZE 8
// For _Py_qsbr_deferred_advance(): the number of deferrals before advancing
// the write sequence.
#define QSBR_DEFERRED_LIMIT 10
// Allocate a QSBR thread state from the freelist
static struct _qsbr_thread_state *
qsbr_allocate(struct _qsbr_shared *shared)
{
struct _qsbr_thread_state *qsbr = shared->freelist;
if (qsbr == NULL) {
return NULL;
}
shared->freelist = qsbr->freelist_next;
qsbr->freelist_next = NULL;
qsbr->shared = shared;
qsbr->allocated = true;
return qsbr;
}
// Initialize (or reintialize) the freelist of QSBR thread states
static void
initialize_new_array(struct _qsbr_shared *shared)
{
for (Py_ssize_t i = 0; i != shared->size; i++) {
struct _qsbr_thread_state *qsbr = &shared->array[i].qsbr;
if (qsbr->tstate != NULL) {
// Update the thread state pointer to its QSBR state
_PyThreadStateImpl *tstate = (_PyThreadStateImpl *)qsbr->tstate;
tstate->qsbr = qsbr;
}
if (!qsbr->allocated) {
// Push to freelist
qsbr->freelist_next = shared->freelist;
shared->freelist = qsbr;
}
}
}
// Grow the array of QSBR thread states. Returns 0 on success, -1 on failure.
static int
grow_thread_array(struct _qsbr_shared *shared)
{
Py_ssize_t new_size = shared->size * 2;
if (new_size < MIN_ARRAY_SIZE) {
new_size = MIN_ARRAY_SIZE;
}
struct _qsbr_pad *array = PyMem_RawCalloc(new_size, sizeof(*array));
if (array == NULL) {
return -1;
}
struct _qsbr_pad *old = shared->array;
if (old != NULL) {
memcpy(array, shared->array, shared->size * sizeof(*array));
}
shared->array = array;
shared->size = new_size;
shared->freelist = NULL;
initialize_new_array(shared);
PyMem_RawFree(old);
return 0;
}
uint64_t
_Py_qsbr_advance(struct _qsbr_shared *shared)
{
// NOTE: with 64-bit sequence numbers, we don't have to worry too much
// about the wr_seq getting too far ahead of rd_seq, but if we ever use
// 32-bit sequence numbers, we'll need to be more careful.
return _Py_atomic_add_uint64(&shared->wr_seq, QSBR_INCR) + QSBR_INCR;
}
uint64_t
_Py_qsbr_deferred_advance(struct _qsbr_thread_state *qsbr)
{
if (++qsbr->deferrals < QSBR_DEFERRED_LIMIT) {
return _Py_qsbr_shared_current(qsbr->shared) + QSBR_INCR;
}
qsbr->deferrals = 0;
return _Py_qsbr_advance(qsbr->shared);
}
static uint64_t
qsbr_poll_scan(struct _qsbr_shared *shared)
{
// Synchronize with store in _Py_qsbr_attach(). We need to ensure that
// the reads from each thread's sequence number are not reordered to see
// earlier "offline" states.
_Py_atomic_fence_seq_cst();
// Compute the minimum sequence number of all attached threads
uint64_t min_seq = _Py_atomic_load_uint64(&shared->wr_seq);
struct _qsbr_pad *array = shared->array;
for (Py_ssize_t i = 0, size = shared->size; i != size; i++) {
struct _qsbr_thread_state *qsbr = &array[i].qsbr;
uint64_t seq = _Py_atomic_load_uint64(&qsbr->seq);
if (seq != QSBR_OFFLINE && QSBR_LT(seq, min_seq)) {
min_seq = seq;
}
}
// Update the shared read sequence
uint64_t rd_seq = _Py_atomic_load_uint64(&shared->rd_seq);
if (QSBR_LT(rd_seq, min_seq)) {
// It's okay if the compare-exchange failed: another thread updated it
(void)_Py_atomic_compare_exchange_uint64(&shared->rd_seq, &rd_seq, min_seq);
rd_seq = min_seq;
}
return rd_seq;
}
bool
_Py_qsbr_poll(struct _qsbr_thread_state *qsbr, uint64_t goal)
{
assert(_Py_atomic_load_int_relaxed(&_PyThreadState_GET()->state) == _Py_THREAD_ATTACHED);
if (_Py_qbsr_goal_reached(qsbr, goal)) {
return true;
}
uint64_t rd_seq = qsbr_poll_scan(qsbr->shared);
return QSBR_LEQ(goal, rd_seq);
}
void
_Py_qsbr_attach(struct _qsbr_thread_state *qsbr)
{
assert(qsbr->seq == 0 && "already attached");
uint64_t seq = _Py_qsbr_shared_current(qsbr->shared);
_Py_atomic_store_uint64(&qsbr->seq, seq); // needs seq_cst
}
void
_Py_qsbr_detach(struct _qsbr_thread_state *qsbr)
{
assert(qsbr->seq != 0 && "already detached");
_Py_atomic_store_uint64_release(&qsbr->seq, QSBR_OFFLINE);
}
Py_ssize_t
_Py_qsbr_reserve(PyInterpreterState *interp)
{
struct _qsbr_shared *shared = &interp->qsbr;
PyMutex_Lock(&shared->mutex);
// Try allocating from our internal freelist
struct _qsbr_thread_state *qsbr = qsbr_allocate(shared);
// If there are no free entries, we pause all threads, grow the array,
// and update the pointers in PyThreadState to entries in the new array.
if (qsbr == NULL) {
_PyEval_StopTheWorld(interp);
if (grow_thread_array(shared) == 0) {
qsbr = qsbr_allocate(shared);
}
_PyEval_StartTheWorld(interp);
}
PyMutex_Unlock(&shared->mutex);
if (qsbr == NULL) {
return -1;
}
// Return an index rather than the pointer because the array may be
// resized and the pointer invalidated.
return (struct _qsbr_pad *)qsbr - shared->array;
}
void
_Py_qsbr_register(_PyThreadStateImpl *tstate, PyInterpreterState *interp,
Py_ssize_t index)
{
// Associate the QSBR state with the thread state
struct _qsbr_shared *shared = &interp->qsbr;
PyMutex_Lock(&shared->mutex);
struct _qsbr_thread_state *qsbr = &interp->qsbr.array[index].qsbr;
assert(qsbr->allocated && qsbr->tstate == NULL);
qsbr->tstate = (PyThreadState *)tstate;
tstate->qsbr = qsbr;
PyMutex_Unlock(&shared->mutex);
}
void
_Py_qsbr_unregister(PyThreadState *tstate)
{
struct _qsbr_shared *shared = &tstate->interp->qsbr;
struct _PyThreadStateImpl *tstate_imp = (_PyThreadStateImpl*) tstate;
// gh-119369: GIL must be released (if held) to prevent deadlocks, because
// we might not have an active tstate, which means that blocking on PyMutex
// locks will not implicitly release the GIL.
assert(!tstate->_status.holds_gil);
PyMutex_Lock(&shared->mutex);
// NOTE: we must load (or reload) the thread state's qbsr inside the mutex
// because the array may have been resized (changing tstate->qsbr) while
// we waited to acquire the mutex.
struct _qsbr_thread_state *qsbr = tstate_imp->qsbr;
assert(qsbr->seq == 0 && "thread state must be detached");
assert(qsbr->allocated && qsbr->tstate == tstate);
tstate_imp->qsbr = NULL;
qsbr->tstate = NULL;
qsbr->allocated = false;
qsbr->freelist_next = shared->freelist;
shared->freelist = qsbr;
PyMutex_Unlock(&shared->mutex);
}
void
_Py_qsbr_fini(PyInterpreterState *interp)
{
struct _qsbr_shared *shared = &interp->qsbr;
PyMem_RawFree(shared->array);
shared->array = NULL;
shared->size = 0;
shared->freelist = NULL;
}
void
_Py_qsbr_after_fork(_PyThreadStateImpl *tstate)
{
struct _qsbr_thread_state *this_qsbr = tstate->qsbr;
struct _qsbr_shared *shared = this_qsbr->shared;
_PyMutex_at_fork_reinit(&shared->mutex);
for (Py_ssize_t i = 0; i != shared->size; i++) {
struct _qsbr_thread_state *qsbr = &shared->array[i].qsbr;
if (qsbr != this_qsbr && qsbr->allocated) {
qsbr->tstate = NULL;
qsbr->allocated = false;
qsbr->freelist_next = shared->freelist;
shared->freelist = qsbr;
}
}
}
You can’t perform that action at this time.