aio: strengthen memory barriers for bottom half scheduling
[qemu.git] / async.c
1 /*
2 * QEMU System Emulator
3 *
4 * Copyright (c) 2003-2008 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25 #include "qemu-common.h"
26 #include "block/aio.h"
27 #include "block/thread-pool.h"
28 #include "qemu/main-loop.h"
29 #include "qemu/atomic.h"
30
31 /***********************************************************/
32 /* bottom halves (can be seen as timers which expire ASAP) */
33
34 struct QEMUBH {
35 AioContext *ctx;
36 QEMUBHFunc *cb;
37 void *opaque;
38 QEMUBH *next;
39 bool scheduled;
40 bool idle;
41 bool deleted;
42 };
43
44 QEMUBH *aio_bh_new(AioContext *ctx, QEMUBHFunc *cb, void *opaque)
45 {
46 QEMUBH *bh;
47 bh = g_new(QEMUBH, 1);
48 *bh = (QEMUBH){
49 .ctx = ctx,
50 .cb = cb,
51 .opaque = opaque,
52 };
53 qemu_mutex_lock(&ctx->bh_lock);
54 bh->next = ctx->first_bh;
55 /* Make sure that the members are ready before putting bh into list */
56 smp_wmb();
57 ctx->first_bh = bh;
58 qemu_mutex_unlock(&ctx->bh_lock);
59 return bh;
60 }
61
62 /* Multiple occurrences of aio_bh_poll cannot be called concurrently */
63 int aio_bh_poll(AioContext *ctx)
64 {
65 QEMUBH *bh, **bhp, *next;
66 int ret;
67
68 ctx->walking_bh++;
69
70 ret = 0;
71 for (bh = ctx->first_bh; bh; bh = next) {
72 /* Make sure that fetching bh happens before accessing its members */
73 smp_read_barrier_depends();
74 next = bh->next;
75 /* The atomic_xchg is paired with the one in qemu_bh_schedule. The
76 * implicit memory barrier ensures that the callback sees all writes
77 * done by the scheduling thread. It also ensures that the scheduling
78 * thread sees the zero before bh->cb has run, and thus will call
79 * aio_notify again if necessary.
80 */
81 if (!bh->deleted && atomic_xchg(&bh->scheduled, 0)) {
82 if (!bh->idle)
83 ret = 1;
84 bh->idle = 0;
85 bh->cb(bh->opaque);
86 }
87 }
88
89 ctx->walking_bh--;
90
91 /* remove deleted bhs */
92 if (!ctx->walking_bh) {
93 qemu_mutex_lock(&ctx->bh_lock);
94 bhp = &ctx->first_bh;
95 while (*bhp) {
96 bh = *bhp;
97 if (bh->deleted) {
98 *bhp = bh->next;
99 g_free(bh);
100 } else {
101 bhp = &bh->next;
102 }
103 }
104 qemu_mutex_unlock(&ctx->bh_lock);
105 }
106
107 return ret;
108 }
109
110 void qemu_bh_schedule_idle(QEMUBH *bh)
111 {
112 bh->idle = 1;
113 /* Make sure that idle & any writes needed by the callback are done
114 * before the locations are read in the aio_bh_poll.
115 */
116 atomic_mb_set(&bh->scheduled, 1);
117 }
118
119 void qemu_bh_schedule(QEMUBH *bh)
120 {
121 AioContext *ctx;
122
123 ctx = bh->ctx;
124 bh->idle = 0;
125 /* The memory barrier implicit in atomic_xchg makes sure that:
126 * 1. idle & any writes needed by the callback are done before the
127 * locations are read in the aio_bh_poll.
128 * 2. ctx is loaded before scheduled is set and the callback has a chance
129 * to execute.
130 */
131 if (atomic_xchg(&bh->scheduled, 1) == 0) {
132 aio_notify(ctx);
133 }
134 }
135
136
137 /* This func is async.
138 */
139 void qemu_bh_cancel(QEMUBH *bh)
140 {
141 bh->scheduled = 0;
142 }
143
144 /* This func is async.The bottom half will do the delete action at the finial
145 * end.
146 */
147 void qemu_bh_delete(QEMUBH *bh)
148 {
149 bh->scheduled = 0;
150 bh->deleted = 1;
151 }
152
153 int64_t
154 aio_compute_timeout(AioContext *ctx)
155 {
156 int64_t deadline;
157 int timeout = -1;
158 QEMUBH *bh;
159
160 for (bh = ctx->first_bh; bh; bh = bh->next) {
161 if (!bh->deleted && bh->scheduled) {
162 if (bh->idle) {
163 /* idle bottom halves will be polled at least
164 * every 10ms */
165 timeout = 10000000;
166 } else {
167 /* non-idle bottom halves will be executed
168 * immediately */
169 return 0;
170 }
171 }
172 }
173
174 deadline = timerlistgroup_deadline_ns(&ctx->tlg);
175 if (deadline == 0) {
176 return 0;
177 } else {
178 return qemu_soonest_timeout(timeout, deadline);
179 }
180 }
181
182 static gboolean
183 aio_ctx_prepare(GSource *source, gint *timeout)
184 {
185 AioContext *ctx = (AioContext *) source;
186
187 /* We assume there is no timeout already supplied */
188 *timeout = qemu_timeout_ns_to_ms(aio_compute_timeout(ctx));
189
190 if (aio_prepare(ctx)) {
191 *timeout = 0;
192 }
193
194 return *timeout == 0;
195 }
196
197 static gboolean
198 aio_ctx_check(GSource *source)
199 {
200 AioContext *ctx = (AioContext *) source;
201 QEMUBH *bh;
202
203 for (bh = ctx->first_bh; bh; bh = bh->next) {
204 if (!bh->deleted && bh->scheduled) {
205 return true;
206 }
207 }
208 return aio_pending(ctx) || (timerlistgroup_deadline_ns(&ctx->tlg) == 0);
209 }
210
211 static gboolean
212 aio_ctx_dispatch(GSource *source,
213 GSourceFunc callback,
214 gpointer user_data)
215 {
216 AioContext *ctx = (AioContext *) source;
217
218 assert(callback == NULL);
219 aio_dispatch(ctx);
220 return true;
221 }
222
223 static void
224 aio_ctx_finalize(GSource *source)
225 {
226 AioContext *ctx = (AioContext *) source;
227
228 thread_pool_free(ctx->thread_pool);
229 aio_set_event_notifier(ctx, &ctx->notifier, NULL);
230 event_notifier_cleanup(&ctx->notifier);
231 rfifolock_destroy(&ctx->lock);
232 qemu_mutex_destroy(&ctx->bh_lock);
233 g_array_free(ctx->pollfds, TRUE);
234 timerlistgroup_deinit(&ctx->tlg);
235 }
236
237 static GSourceFuncs aio_source_funcs = {
238 aio_ctx_prepare,
239 aio_ctx_check,
240 aio_ctx_dispatch,
241 aio_ctx_finalize
242 };
243
244 GSource *aio_get_g_source(AioContext *ctx)
245 {
246 g_source_ref(&ctx->source);
247 return &ctx->source;
248 }
249
250 ThreadPool *aio_get_thread_pool(AioContext *ctx)
251 {
252 if (!ctx->thread_pool) {
253 ctx->thread_pool = thread_pool_new(ctx);
254 }
255 return ctx->thread_pool;
256 }
257
258 void aio_set_dispatching(AioContext *ctx, bool dispatching)
259 {
260 ctx->dispatching = dispatching;
261 if (!dispatching) {
262 /* Write ctx->dispatching before reading e.g. bh->scheduled.
263 * Optimization: this is only needed when we're entering the "unsafe"
264 * phase where other threads must call event_notifier_set.
265 */
266 smp_mb();
267 }
268 }
269
270 void aio_notify(AioContext *ctx)
271 {
272 /* Write e.g. bh->scheduled before reading ctx->dispatching. */
273 smp_mb();
274 if (!ctx->dispatching) {
275 event_notifier_set(&ctx->notifier);
276 }
277 }
278
279 static void aio_timerlist_notify(void *opaque)
280 {
281 aio_notify(opaque);
282 }
283
284 static void aio_rfifolock_cb(void *opaque)
285 {
286 /* Kick owner thread in case they are blocked in aio_poll() */
287 aio_notify(opaque);
288 }
289
290 AioContext *aio_context_new(Error **errp)
291 {
292 int ret;
293 AioContext *ctx;
294 ctx = (AioContext *) g_source_new(&aio_source_funcs, sizeof(AioContext));
295 ret = event_notifier_init(&ctx->notifier, false);
296 if (ret < 0) {
297 g_source_destroy(&ctx->source);
298 error_setg_errno(errp, -ret, "Failed to initialize event notifier");
299 return NULL;
300 }
301 g_source_set_can_recurse(&ctx->source, true);
302 aio_set_event_notifier(ctx, &ctx->notifier,
303 (EventNotifierHandler *)
304 event_notifier_test_and_clear);
305 ctx->pollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
306 ctx->thread_pool = NULL;
307 qemu_mutex_init(&ctx->bh_lock);
308 rfifolock_init(&ctx->lock, aio_rfifolock_cb, ctx);
309 timerlistgroup_init(&ctx->tlg, aio_timerlist_notify, ctx);
310
311 return ctx;
312 }
313
314 void aio_context_ref(AioContext *ctx)
315 {
316 g_source_ref(&ctx->source);
317 }
318
319 void aio_context_unref(AioContext *ctx)
320 {
321 g_source_unref(&ctx->source);
322 }
323
324 void aio_context_acquire(AioContext *ctx)
325 {
326 rfifolock_lock(&ctx->lock);
327 }
328
329 void aio_context_release(AioContext *ctx)
330 {
331 rfifolock_unlock(&ctx->lock);
332 }