]>
Commit | Line | Data |
---|---|---|
4f999d05 KW |
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" | |
737e150e | 26 | #include "block/aio.h" |
9b34277d | 27 | #include "block/thread-pool.h" |
1de7afc9 | 28 | #include "qemu/main-loop.h" |
0ceb849b | 29 | #include "qemu/atomic.h" |
9a1e9481 | 30 | |
4f999d05 KW |
31 | /***********************************************************/ |
32 | /* bottom halves (can be seen as timers which expire ASAP) */ | |
33 | ||
34 | struct QEMUBH { | |
2f4dc3c1 | 35 | AioContext *ctx; |
4f999d05 KW |
36 | QEMUBHFunc *cb; |
37 | void *opaque; | |
4f999d05 | 38 | QEMUBH *next; |
9b47b17e SW |
39 | bool scheduled; |
40 | bool idle; | |
41 | bool deleted; | |
4f999d05 KW |
42 | }; |
43 | ||
f627aab1 | 44 | QEMUBH *aio_bh_new(AioContext *ctx, QEMUBHFunc *cb, void *opaque) |
4f999d05 KW |
45 | { |
46 | QEMUBH *bh; | |
ee82310f PB |
47 | bh = g_new(QEMUBH, 1); |
48 | *bh = (QEMUBH){ | |
49 | .ctx = ctx, | |
50 | .cb = cb, | |
51 | .opaque = opaque, | |
52 | }; | |
dcc772e2 | 53 | qemu_mutex_lock(&ctx->bh_lock); |
f627aab1 | 54 | bh->next = ctx->first_bh; |
dcc772e2 LPF |
55 | /* Make sure that the members are ready before putting bh into list */ |
56 | smp_wmb(); | |
f627aab1 | 57 | ctx->first_bh = bh; |
dcc772e2 | 58 | qemu_mutex_unlock(&ctx->bh_lock); |
4f999d05 KW |
59 | return bh; |
60 | } | |
61 | ||
dcc772e2 | 62 | /* Multiple occurrences of aio_bh_poll cannot be called concurrently */ |
f627aab1 | 63 | int aio_bh_poll(AioContext *ctx) |
4f999d05 | 64 | { |
7887f620 | 65 | QEMUBH *bh, **bhp, *next; |
4f999d05 | 66 | int ret; |
648fb0ea | 67 | |
f627aab1 | 68 | ctx->walking_bh++; |
4f999d05 KW |
69 | |
70 | ret = 0; | |
f627aab1 | 71 | for (bh = ctx->first_bh; bh; bh = next) { |
dcc772e2 LPF |
72 | /* Make sure that fetching bh happens before accessing its members */ |
73 | smp_read_barrier_depends(); | |
7887f620 | 74 | next = bh->next; |
e8d3b1a2 PB |
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)) { | |
ca96ac44 SH |
82 | /* Idle BHs and the notify BH don't count as progress */ |
83 | if (!bh->idle && bh != ctx->notify_dummy_bh) { | |
4f999d05 | 84 | ret = 1; |
ca96ac44 | 85 | } |
4f999d05 KW |
86 | bh->idle = 0; |
87 | bh->cb(bh->opaque); | |
88 | } | |
89 | } | |
90 | ||
f627aab1 | 91 | ctx->walking_bh--; |
648fb0ea | 92 | |
4f999d05 | 93 | /* remove deleted bhs */ |
f627aab1 | 94 | if (!ctx->walking_bh) { |
dcc772e2 | 95 | qemu_mutex_lock(&ctx->bh_lock); |
f627aab1 | 96 | bhp = &ctx->first_bh; |
648fb0ea KW |
97 | while (*bhp) { |
98 | bh = *bhp; | |
99 | if (bh->deleted) { | |
100 | *bhp = bh->next; | |
101 | g_free(bh); | |
102 | } else { | |
103 | bhp = &bh->next; | |
104 | } | |
105 | } | |
dcc772e2 | 106 | qemu_mutex_unlock(&ctx->bh_lock); |
4f999d05 KW |
107 | } |
108 | ||
109 | return ret; | |
110 | } | |
111 | ||
112 | void qemu_bh_schedule_idle(QEMUBH *bh) | |
113 | { | |
4f999d05 | 114 | bh->idle = 1; |
dcc772e2 LPF |
115 | /* Make sure that idle & any writes needed by the callback are done |
116 | * before the locations are read in the aio_bh_poll. | |
117 | */ | |
e8d3b1a2 | 118 | atomic_mb_set(&bh->scheduled, 1); |
4f999d05 KW |
119 | } |
120 | ||
121 | void qemu_bh_schedule(QEMUBH *bh) | |
122 | { | |
924fe129 SH |
123 | AioContext *ctx; |
124 | ||
924fe129 | 125 | ctx = bh->ctx; |
4f999d05 | 126 | bh->idle = 0; |
e8d3b1a2 | 127 | /* The memory barrier implicit in atomic_xchg makes sure that: |
924fe129 SH |
128 | * 1. idle & any writes needed by the callback are done before the |
129 | * locations are read in the aio_bh_poll. | |
130 | * 2. ctx is loaded before scheduled is set and the callback has a chance | |
131 | * to execute. | |
dcc772e2 | 132 | */ |
e8d3b1a2 PB |
133 | if (atomic_xchg(&bh->scheduled, 1) == 0) { |
134 | aio_notify(ctx); | |
135 | } | |
4f999d05 KW |
136 | } |
137 | ||
dcc772e2 LPF |
138 | |
139 | /* This func is async. | |
140 | */ | |
4f999d05 KW |
141 | void qemu_bh_cancel(QEMUBH *bh) |
142 | { | |
143 | bh->scheduled = 0; | |
144 | } | |
145 | ||
dcc772e2 LPF |
146 | /* This func is async.The bottom half will do the delete action at the finial |
147 | * end. | |
148 | */ | |
4f999d05 KW |
149 | void qemu_bh_delete(QEMUBH *bh) |
150 | { | |
151 | bh->scheduled = 0; | |
152 | bh->deleted = 1; | |
153 | } | |
154 | ||
845ca10d PB |
155 | int64_t |
156 | aio_compute_timeout(AioContext *ctx) | |
4f999d05 | 157 | { |
845ca10d PB |
158 | int64_t deadline; |
159 | int timeout = -1; | |
4f999d05 KW |
160 | QEMUBH *bh; |
161 | ||
f627aab1 | 162 | for (bh = ctx->first_bh; bh; bh = bh->next) { |
4f999d05 KW |
163 | if (!bh->deleted && bh->scheduled) { |
164 | if (bh->idle) { | |
165 | /* idle bottom halves will be polled at least | |
166 | * every 10ms */ | |
845ca10d | 167 | timeout = 10000000; |
4f999d05 KW |
168 | } else { |
169 | /* non-idle bottom halves will be executed | |
170 | * immediately */ | |
845ca10d | 171 | return 0; |
4f999d05 KW |
172 | } |
173 | } | |
174 | } | |
e3713e00 | 175 | |
845ca10d | 176 | deadline = timerlistgroup_deadline_ns(&ctx->tlg); |
533a8cf3 | 177 | if (deadline == 0) { |
845ca10d | 178 | return 0; |
533a8cf3 | 179 | } else { |
845ca10d | 180 | return qemu_soonest_timeout(timeout, deadline); |
533a8cf3 | 181 | } |
845ca10d | 182 | } |
533a8cf3 | 183 | |
845ca10d PB |
184 | static gboolean |
185 | aio_ctx_prepare(GSource *source, gint *timeout) | |
186 | { | |
187 | AioContext *ctx = (AioContext *) source; | |
188 | ||
eabc9779 PB |
189 | atomic_or(&ctx->notify_me, 1); |
190 | ||
845ca10d PB |
191 | /* We assume there is no timeout already supplied */ |
192 | *timeout = qemu_timeout_ns_to_ms(aio_compute_timeout(ctx)); | |
a3462c65 PB |
193 | |
194 | if (aio_prepare(ctx)) { | |
195 | *timeout = 0; | |
196 | } | |
197 | ||
845ca10d | 198 | return *timeout == 0; |
e3713e00 PB |
199 | } |
200 | ||
201 | static gboolean | |
202 | aio_ctx_check(GSource *source) | |
203 | { | |
204 | AioContext *ctx = (AioContext *) source; | |
205 | QEMUBH *bh; | |
206 | ||
eabc9779 | 207 | atomic_and(&ctx->notify_me, ~1); |
05e514b1 | 208 | aio_notify_accept(ctx); |
21a03d17 | 209 | |
e3713e00 PB |
210 | for (bh = ctx->first_bh; bh; bh = bh->next) { |
211 | if (!bh->deleted && bh->scheduled) { | |
212 | return true; | |
213 | } | |
214 | } | |
533a8cf3 | 215 | return aio_pending(ctx) || (timerlistgroup_deadline_ns(&ctx->tlg) == 0); |
e3713e00 PB |
216 | } |
217 | ||
218 | static gboolean | |
219 | aio_ctx_dispatch(GSource *source, | |
220 | GSourceFunc callback, | |
221 | gpointer user_data) | |
222 | { | |
223 | AioContext *ctx = (AioContext *) source; | |
224 | ||
225 | assert(callback == NULL); | |
e4c7e2d1 | 226 | aio_dispatch(ctx); |
e3713e00 PB |
227 | return true; |
228 | } | |
229 | ||
2f4dc3c1 PB |
230 | static void |
231 | aio_ctx_finalize(GSource *source) | |
232 | { | |
233 | AioContext *ctx = (AioContext *) source; | |
234 | ||
ca96ac44 | 235 | qemu_bh_delete(ctx->notify_dummy_bh); |
9b34277d | 236 | thread_pool_free(ctx->thread_pool); |
a076972a SH |
237 | |
238 | qemu_mutex_lock(&ctx->bh_lock); | |
239 | while (ctx->first_bh) { | |
240 | QEMUBH *next = ctx->first_bh->next; | |
241 | ||
242 | /* qemu_bh_delete() must have been called on BHs in this AioContext */ | |
243 | assert(ctx->first_bh->deleted); | |
244 | ||
245 | g_free(ctx->first_bh); | |
246 | ctx->first_bh = next; | |
247 | } | |
248 | qemu_mutex_unlock(&ctx->bh_lock); | |
249 | ||
f2e5dca4 | 250 | aio_set_event_notifier(ctx, &ctx->notifier, NULL); |
2f4dc3c1 | 251 | event_notifier_cleanup(&ctx->notifier); |
98563fc3 | 252 | rfifolock_destroy(&ctx->lock); |
dcc772e2 | 253 | qemu_mutex_destroy(&ctx->bh_lock); |
dae21b98 | 254 | timerlistgroup_deinit(&ctx->tlg); |
2f4dc3c1 PB |
255 | } |
256 | ||
e3713e00 PB |
257 | static GSourceFuncs aio_source_funcs = { |
258 | aio_ctx_prepare, | |
259 | aio_ctx_check, | |
260 | aio_ctx_dispatch, | |
2f4dc3c1 | 261 | aio_ctx_finalize |
e3713e00 PB |
262 | }; |
263 | ||
264 | GSource *aio_get_g_source(AioContext *ctx) | |
265 | { | |
266 | g_source_ref(&ctx->source); | |
267 | return &ctx->source; | |
268 | } | |
a915f4bc | 269 | |
9b34277d SH |
270 | ThreadPool *aio_get_thread_pool(AioContext *ctx) |
271 | { | |
272 | if (!ctx->thread_pool) { | |
273 | ctx->thread_pool = thread_pool_new(ctx); | |
274 | } | |
275 | return ctx->thread_pool; | |
276 | } | |
277 | ||
2f4dc3c1 PB |
278 | void aio_notify(AioContext *ctx) |
279 | { | |
eabc9779 PB |
280 | /* Write e.g. bh->scheduled before reading ctx->notify_me. Pairs |
281 | * with atomic_or in aio_ctx_prepare or atomic_add in aio_poll. | |
282 | */ | |
0ceb849b | 283 | smp_mb(); |
eabc9779 | 284 | if (ctx->notify_me) { |
0ceb849b | 285 | event_notifier_set(&ctx->notifier); |
05e514b1 PB |
286 | atomic_mb_set(&ctx->notified, true); |
287 | } | |
288 | } | |
289 | ||
290 | void aio_notify_accept(AioContext *ctx) | |
291 | { | |
292 | if (atomic_xchg(&ctx->notified, false)) { | |
293 | event_notifier_test_and_clear(&ctx->notifier); | |
0ceb849b | 294 | } |
2f4dc3c1 PB |
295 | } |
296 | ||
d5541d86 AB |
297 | static void aio_timerlist_notify(void *opaque) |
298 | { | |
299 | aio_notify(opaque); | |
300 | } | |
301 | ||
da5e1de9 SH |
302 | static void aio_rfifolock_cb(void *opaque) |
303 | { | |
ca96ac44 SH |
304 | AioContext *ctx = opaque; |
305 | ||
da5e1de9 | 306 | /* Kick owner thread in case they are blocked in aio_poll() */ |
ca96ac44 SH |
307 | qemu_bh_schedule(ctx->notify_dummy_bh); |
308 | } | |
309 | ||
310 | static void notify_dummy_bh(void *opaque) | |
311 | { | |
312 | /* Do nothing, we were invoked just to force the event loop to iterate */ | |
da5e1de9 SH |
313 | } |
314 | ||
21a03d17 PB |
315 | static void event_notifier_dummy_cb(EventNotifier *e) |
316 | { | |
317 | } | |
318 | ||
2f78e491 | 319 | AioContext *aio_context_new(Error **errp) |
f627aab1 | 320 | { |
2f78e491 | 321 | int ret; |
2f4dc3c1 PB |
322 | AioContext *ctx; |
323 | ctx = (AioContext *) g_source_new(&aio_source_funcs, sizeof(AioContext)); | |
2f78e491 CN |
324 | ret = event_notifier_init(&ctx->notifier, false); |
325 | if (ret < 0) { | |
326 | g_source_destroy(&ctx->source); | |
327 | error_setg_errno(errp, -ret, "Failed to initialize event notifier"); | |
328 | return NULL; | |
329 | } | |
fcf5def1 | 330 | g_source_set_can_recurse(&ctx->source, true); |
2f78e491 CN |
331 | aio_set_event_notifier(ctx, &ctx->notifier, |
332 | (EventNotifierHandler *) | |
21a03d17 | 333 | event_notifier_dummy_cb); |
9b34277d | 334 | ctx->thread_pool = NULL; |
dcc772e2 | 335 | qemu_mutex_init(&ctx->bh_lock); |
da5e1de9 | 336 | rfifolock_init(&ctx->lock, aio_rfifolock_cb, ctx); |
d5541d86 | 337 | timerlistgroup_init(&ctx->tlg, aio_timerlist_notify, ctx); |
2f4dc3c1 | 338 | |
ca96ac44 SH |
339 | ctx->notify_dummy_bh = aio_bh_new(ctx, notify_dummy_bh, NULL); |
340 | ||
2f4dc3c1 | 341 | return ctx; |
e3713e00 PB |
342 | } |
343 | ||
344 | void aio_context_ref(AioContext *ctx) | |
345 | { | |
346 | g_source_ref(&ctx->source); | |
347 | } | |
348 | ||
349 | void aio_context_unref(AioContext *ctx) | |
350 | { | |
351 | g_source_unref(&ctx->source); | |
f627aab1 | 352 | } |
98563fc3 SH |
353 | |
354 | void aio_context_acquire(AioContext *ctx) | |
355 | { | |
356 | rfifolock_lock(&ctx->lock); | |
357 | } | |
358 | ||
359 | void aio_context_release(AioContext *ctx) | |
360 | { | |
361 | rfifolock_unlock(&ctx->lock); | |
362 | } |