aboutsummaryrefslogtreecommitdiff
path: root/util/qemu-coroutine-lock.c
blob: b44b5d55ebad8b8b6c4b8d7e41030ae527308c45 (plain)
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
/*
 * coroutine queues and locks
 *
 * Copyright (c) 2011 Kevin Wolf <kwolf@redhat.com>
 *
 * Permission is hereby granted, free of charge, to any person obtaining a copy
 * of this software and associated documentation files (the "Software"), to deal
 * in the Software without restriction, including without limitation the rights
 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 * copies of the Software, and to permit persons to whom the Software is
 * furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
 * THE SOFTWARE.
 *
 * The lock-free mutex implementation is based on OSv
 * (core/lfmutex.cc, include/lockfree/mutex.hh).
 * Copyright (C) 2013 Cloudius Systems, Ltd.
 */

#include "qemu/osdep.h"
#include "qemu-common.h"
#include "qemu/coroutine.h"
#include "qemu/coroutine_int.h"
#include "qemu/processor.h"
#include "qemu/queue.h"
#include "block/aio.h"
#include "trace.h"

void qemu_co_queue_init(CoQueue *queue)
{
    QSIMPLEQ_INIT(&queue->entries);
}

void coroutine_fn qemu_co_queue_wait(CoQueue *queue, CoMutex *mutex)
{
    Coroutine *self = qemu_coroutine_self();
    QSIMPLEQ_INSERT_TAIL(&queue->entries, self, co_queue_next);

    if (mutex) {
        qemu_co_mutex_unlock(mutex);
    }

    /* There is no race condition here.  Other threads will call
     * aio_co_schedule on our AioContext, which can reenter this
     * coroutine but only after this yield and after the main loop
     * has gone through the next iteration.
     */
    qemu_coroutine_yield();
    assert(qemu_in_coroutine());

    /* TODO: OSv implements wait morphing here, where the wakeup
     * primitive automatically places the woken coroutine on the
     * mutex's queue.  This avoids the thundering herd effect.
     */
    if (mutex) {
        qemu_co_mutex_lock(mutex);
    }
}

/**
 * qemu_co_queue_run_restart:
 *
 * Enter each coroutine that was previously marked for restart by
 * qemu_co_queue_next() or qemu_co_queue_restart_all().  This function is
 * invoked by the core coroutine code when the current coroutine yields or
 * terminates.
 */
void qemu_co_queue_run_restart(Coroutine *co)
{
    Coroutine *next;
    QSIMPLEQ_HEAD(, Coroutine) tmp_queue_wakeup =
        QSIMPLEQ_HEAD_INITIALIZER(tmp_queue_wakeup);

    trace_qemu_co_queue_run_restart(co);

    /* Because "co" has yielded, any coroutine that we wakeup can resume it.
     * If this happens and "co" terminates, co->co_queue_wakeup becomes
     * invalid memory.  Therefore, use a temporary queue and do not touch
     * the "co" coroutine as soon as you enter another one.
     *
     * In its turn resumed "co" can pupulate "co_queue_wakeup" queue with
     * new coroutines to be woken up.  The caller, who has resumed "co",
     * will be responsible for traversing the same queue, which may cause
     * a different wakeup order but not any missing wakeups.
     */
    QSIMPLEQ_CONCAT(&tmp_queue_wakeup, &co->co_queue_wakeup);

    while ((next = QSIMPLEQ_FIRST(&tmp_queue_wakeup))) {
        QSIMPLEQ_REMOVE_HEAD(&tmp_queue_wakeup, co_queue_next);
        qemu_coroutine_enter(next);
    }
}

static bool qemu_co_queue_do_restart(CoQueue *queue, bool single)
{
    Coroutine *next;

    if (QSIMPLEQ_EMPTY(&queue->entries)) {
        return false;
    }

    while ((next = QSIMPLEQ_FIRST(&queue->entries)) != NULL) {
        QSIMPLEQ_REMOVE_HEAD(&queue->entries, co_queue_next);
        aio_co_wake(next);
        if (single) {
            break;
        }
    }
    return true;
}

bool coroutine_fn qemu_co_queue_next(CoQueue *queue)
{
    assert(qemu_in_coroutine());
    return qemu_co_queue_do_restart(queue, true);
}

void coroutine_fn qemu_co_queue_restart_all(CoQueue *queue)
{
    assert(qemu_in_coroutine());
    qemu_co_queue_do_restart(queue, false);
}

bool qemu_co_enter_next(CoQueue *queue)
{
    Coroutine *next;

    next = QSIMPLEQ_FIRST(&queue->entries);
    if (!next) {
        return false;
    }

    QSIMPLEQ_REMOVE_HEAD(&queue->entries, co_queue_next);
    qemu_coroutine_enter(next);
    return true;
}

bool qemu_co_queue_empty(CoQueue *queue)
{
    return QSIMPLEQ_FIRST(&queue->entries) == NULL;
}

/* The wait records are handled with a multiple-producer, single-consumer
 * lock-free queue.  There cannot be two concurrent pop_waiter() calls
 * because pop_waiter() can only be called while mutex->handoff is zero.
 * This can happen in three cases:
 * - in qemu_co_mutex_unlock, before the hand-off protocol has started.
 *   In this case, qemu_co_mutex_lock will see mutex->handoff == 0 and
 *   not take part in the handoff.
 * - in qemu_co_mutex_lock, if it steals the hand-off responsibility from
 *   qemu_co_mutex_unlock.  In this case, qemu_co_mutex_unlock will fail
 *   the cmpxchg (it will see either 0 or the next sequence value) and
 *   exit.  The next hand-off cannot begin until qemu_co_mutex_lock has
 *   woken up someone.
 * - in qemu_co_mutex_unlock, if it takes the hand-off token itself.
 *   In this case another iteration starts with mutex->handoff == 0;
 *   a concurrent qemu_co_mutex_lock will fail the cmpxchg, and
 *   qemu_co_mutex_unlock will go back to case (1).
 *
 * The following functions manage this queue.
 */
typedef struct CoWaitRecord {
    Coroutine *co;
    QSLIST_ENTRY(CoWaitRecord) next;
} CoWaitRecord;

static void push_waiter(CoMutex *mutex, CoWaitRecord *w)
{
    w->co = qemu_coroutine_self();
    QSLIST_INSERT_HEAD_ATOMIC(&mutex->from_push, w, next);
}

static void move_waiters(CoMutex *mutex)
{
    QSLIST_HEAD(, CoWaitRecord) reversed;
    QSLIST_MOVE_ATOMIC(&reversed, &mutex->from_push);
    while (!QSLIST_EMPTY(&reversed)) {
        CoWaitRecord *w = QSLIST_FIRST(&reversed);
        QSLIST_REMOVE_HEAD(&reversed, next);
        QSLIST_INSERT_HEAD(&mutex->to_pop, w, next);
    }
}

static CoWaitRecord *pop_waiter(CoMutex *mutex)
{
    CoWaitRecord *w;

    if (QSLIST_EMPTY(&mutex->to_pop)) {
        move_waiters(mutex);
        if (QSLIST_EMPTY(&mutex->to_pop)) {
            return NULL;
        }
    }
    w = QSLIST_FIRST(&mutex->to_pop);
    QSLIST_REMOVE_HEAD(&mutex->to_pop, next);
    return w;
}

static bool has_waiters(CoMutex *mutex)
{
    return QSLIST_EMPTY(&mutex->to_pop) || QSLIST_EMPTY(&mutex->from_push);
}

void qemu_co_mutex_init(CoMutex *mutex)
{
    memset(mutex, 0, sizeof(*mutex));
}

static void coroutine_fn qemu_co_mutex_wake(CoMutex *mutex, Coroutine *co)
{
    /* Read co before co->ctx; pairs with smp_wmb() in
     * qemu_coroutine_enter().
     */
    smp_read_barrier_depends();
    mutex->ctx = co->ctx;
    aio_co_wake(co);
}

static void coroutine_fn qemu_co_mutex_lock_slowpath(AioContext *ctx,
                                                     CoMutex *mutex)
{
    Coroutine *self = qemu_coroutine_self();
    CoWaitRecord w;
    unsigned old_handoff;

    trace_qemu_co_mutex_lock_entry(mutex, self);
    w.co = self;
    push_waiter(mutex, &w);

    /* This is the "Responsibility Hand-Off" protocol; a lock() picks from
     * a concurrent unlock() the responsibility of waking somebody up.
     */
    old_handoff = atomic_mb_read(&mutex->handoff);
    if (old_handoff &&
        has_waiters(mutex) &&
        atomic_cmpxchg(&mutex->handoff, old_handoff, 0) == old_handoff) {
        /* There can be no concurrent pops, because there can be only
         * one active handoff at a time.
         */
        CoWaitRecord *to_wake = pop_waiter(mutex);
        Coroutine *co = to_wake->co;
        if (co == self) {
            /* We got the lock ourselves!  */
            assert(to_wake == &w);
            mutex->ctx = ctx;
            return;
        }

        qemu_co_mutex_wake(mutex, co);
    }

    qemu_coroutine_yield();
    trace_qemu_co_mutex_lock_return(mutex, self);
}

void coroutine_fn qemu_co_mutex_lock(CoMutex *mutex)
{
    AioContext *ctx = qemu_get_current_aio_context();
    Coroutine *self = qemu_coroutine_self();
    int waiters, i;

    /* Running a very small critical section on pthread_mutex_t and CoMutex
     * shows that pthread_mutex_t is much faster because it doesn't actually
     * go to sleep.  What happens is that the critical section is shorter
     * than the latency of entering the kernel and thus FUTEX_WAIT always
     * fails.  With CoMutex there is no such latency but you still want to
     * avoid wait and wakeup.  So introduce it artificially.
     */
    i = 0;
retry_fast_path:
    waiters = atomic_cmpxchg(&mutex->locked, 0, 1);
    if (waiters != 0) {
        while (waiters == 1 && ++i < 1000) {
            if (atomic_read(&mutex->ctx) == ctx) {
                break;
            }
            if (atomic_read(&mutex->locked) == 0) {
                goto retry_fast_path;
            }
            cpu_relax();
        }
        waiters = atomic_fetch_inc(&mutex->locked);
    }

    if (waiters == 0) {
        /* Uncontended.  */
        trace_qemu_co_mutex_lock_uncontended(mutex, self);
        mutex->ctx = ctx;
    } else {
        qemu_co_mutex_lock_slowpath(ctx, mutex);
    }
    mutex->holder = self;
    self->locks_held++;
}

void coroutine_fn qemu_co_mutex_unlock(CoMutex *mutex)
{
    Coroutine *self = qemu_coroutine_self();

    trace_qemu_co_mutex_unlock_entry(mutex, self);

    assert(mutex->locked);
    assert(mutex->holder == self);
    assert(qemu_in_coroutine());

    mutex->ctx = NULL;
    mutex->holder = NULL;
    self->locks_held--;
    if (atomic_fetch_dec(&mutex->locked) == 1) {
        /* No waiting qemu_co_mutex_lock().  Pfew, that was easy!  */
        return;
    }

    for (;;) {
        CoWaitRecord *to_wake = pop_waiter(mutex);
        unsigned our_handoff;

        if (to_wake) {
            qemu_co_mutex_wake(mutex, to_wake->co);
            break;
        }

        /* Some concurrent lock() is in progress (we know this because
         * mutex->locked was >1) but it hasn't yet put itself on the wait
         * queue.  Pick a sequence number for the handoff protocol (not 0).
         */
        if (++mutex->sequence == 0) {
            mutex->sequence = 1;
        }

        our_handoff = mutex->sequence;
        atomic_mb_set(&mutex->handoff, our_handoff);
        if (!has_waiters(mutex)) {
            /* The concurrent lock has not added itself yet, so it
             * will be able to pick our handoff.
             */
            break;
        }

        /* Try to do the handoff protocol ourselves; if somebody else has
         * already taken it, however, we're done and they're responsible.
         */
        if (atomic_cmpxchg(&mutex->handoff, our_handoff, 0) != our_handoff) {
            break;
        }
    }

    trace_qemu_co_mutex_unlock_return(mutex, self);
}

void qemu_co_rwlock_init(CoRwlock *lock)
{
    memset(lock, 0, sizeof(*lock));
    qemu_co_queue_init(&lock->queue);
    qemu_co_mutex_init(&lock->mutex);
}

void qemu_co_rwlock_rdlock(CoRwlock *lock)
{
    Coroutine *self = qemu_coroutine_self();

    qemu_co_mutex_lock(&lock->mutex);
    /* For fairness, wait if a writer is in line.  */
    while (lock->pending_writer) {
        qemu_co_queue_wait(&lock->queue, &lock->mutex);
    }
    lock->reader++;
    qemu_co_mutex_unlock(&lock->mutex);

    /* The rest of the read-side critical section is run without the mutex.  */
    self->locks_held++;
}

void qemu_co_rwlock_unlock(CoRwlock *lock)
{
    Coroutine *self = qemu_coroutine_self();

    assert(qemu_in_coroutine());
    if (!lock->reader) {
        /* The critical section started in qemu_co_rwlock_wrlock.  */
        qemu_co_queue_restart_all(&lock->queue);
    } else {
        self->locks_held--;

        qemu_co_mutex_lock(&lock->mutex);
        lock->reader--;
        assert(lock->reader >= 0);
        /* Wakeup only one waiting writer */
        if (!lock->reader) {
            qemu_co_queue_next(&lock->queue);
        }
    }
    qemu_co_mutex_unlock(&lock->mutex);
}

void qemu_co_rwlock_wrlock(CoRwlock *lock)
{
    qemu_co_mutex_lock(&lock->mutex);
    lock->pending_writer++;
    while (lock->reader) {
        qemu_co_queue_wait(&lock->queue, &lock->mutex);
    }
    lock->pending_writer--;

    /* The rest of the write-side critical section is run with
     * the mutex taken, so that lock->reader remains zero.
     * There is no need to update self->locks_held.
     */
}