]>
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 | ||
d38ea87a | 25 | #include "qemu/osdep.h" |
da34e65c | 26 | #include "qapi/error.h" |
4f999d05 | 27 | #include "qemu-common.h" |
737e150e | 28 | #include "block/aio.h" |
9b34277d | 29 | #include "block/thread-pool.h" |
1de7afc9 | 30 | #include "qemu/main-loop.h" |
0ceb849b | 31 | #include "qemu/atomic.h" |
9a1e9481 | 32 | |
4f999d05 KW |
33 | /***********************************************************/ |
34 | /* bottom halves (can be seen as timers which expire ASAP) */ | |
35 | ||
36 | struct QEMUBH { | |
2f4dc3c1 | 37 | AioContext *ctx; |
4f999d05 KW |
38 | QEMUBHFunc *cb; |
39 | void *opaque; | |
4f999d05 | 40 | QEMUBH *next; |
9b47b17e SW |
41 | bool scheduled; |
42 | bool idle; | |
43 | bool deleted; | |
4f999d05 KW |
44 | }; |
45 | ||
f627aab1 | 46 | QEMUBH *aio_bh_new(AioContext *ctx, QEMUBHFunc *cb, void *opaque) |
4f999d05 KW |
47 | { |
48 | QEMUBH *bh; | |
ee82310f PB |
49 | bh = g_new(QEMUBH, 1); |
50 | *bh = (QEMUBH){ | |
51 | .ctx = ctx, | |
52 | .cb = cb, | |
53 | .opaque = opaque, | |
54 | }; | |
dcc772e2 | 55 | qemu_mutex_lock(&ctx->bh_lock); |
f627aab1 | 56 | bh->next = ctx->first_bh; |
dcc772e2 LPF |
57 | /* Make sure that the members are ready before putting bh into list */ |
58 | smp_wmb(); | |
f627aab1 | 59 | ctx->first_bh = bh; |
dcc772e2 | 60 | qemu_mutex_unlock(&ctx->bh_lock); |
4f999d05 KW |
61 | return bh; |
62 | } | |
63 | ||
df281b80 PD |
64 | void aio_bh_call(QEMUBH *bh) |
65 | { | |
66 | bh->cb(bh->opaque); | |
67 | } | |
68 | ||
dcc772e2 | 69 | /* Multiple occurrences of aio_bh_poll cannot be called concurrently */ |
f627aab1 | 70 | int aio_bh_poll(AioContext *ctx) |
4f999d05 | 71 | { |
7887f620 | 72 | QEMUBH *bh, **bhp, *next; |
4f999d05 | 73 | int ret; |
648fb0ea | 74 | |
f627aab1 | 75 | ctx->walking_bh++; |
4f999d05 KW |
76 | |
77 | ret = 0; | |
f627aab1 | 78 | for (bh = ctx->first_bh; bh; bh = next) { |
dcc772e2 LPF |
79 | /* Make sure that fetching bh happens before accessing its members */ |
80 | smp_read_barrier_depends(); | |
7887f620 | 81 | next = bh->next; |
e8d3b1a2 PB |
82 | /* The atomic_xchg is paired with the one in qemu_bh_schedule. The |
83 | * implicit memory barrier ensures that the callback sees all writes | |
84 | * done by the scheduling thread. It also ensures that the scheduling | |
85 | * thread sees the zero before bh->cb has run, and thus will call | |
86 | * aio_notify again if necessary. | |
87 | */ | |
88 | if (!bh->deleted && atomic_xchg(&bh->scheduled, 0)) { | |
ca96ac44 SH |
89 | /* Idle BHs and the notify BH don't count as progress */ |
90 | if (!bh->idle && bh != ctx->notify_dummy_bh) { | |
4f999d05 | 91 | ret = 1; |
ca96ac44 | 92 | } |
4f999d05 | 93 | bh->idle = 0; |
df281b80 | 94 | aio_bh_call(bh); |
4f999d05 KW |
95 | } |
96 | } | |
97 | ||
f627aab1 | 98 | ctx->walking_bh--; |
648fb0ea | 99 | |
4f999d05 | 100 | /* remove deleted bhs */ |
f627aab1 | 101 | if (!ctx->walking_bh) { |
dcc772e2 | 102 | qemu_mutex_lock(&ctx->bh_lock); |
f627aab1 | 103 | bhp = &ctx->first_bh; |
648fb0ea KW |
104 | while (*bhp) { |
105 | bh = *bhp; | |
106 | if (bh->deleted) { | |
107 | *bhp = bh->next; | |
108 | g_free(bh); | |
109 | } else { | |
110 | bhp = &bh->next; | |
111 | } | |
112 | } | |
dcc772e2 | 113 | qemu_mutex_unlock(&ctx->bh_lock); |
4f999d05 KW |
114 | } |
115 | ||
116 | return ret; | |
117 | } | |
118 | ||
119 | void qemu_bh_schedule_idle(QEMUBH *bh) | |
120 | { | |
4f999d05 | 121 | bh->idle = 1; |
dcc772e2 LPF |
122 | /* Make sure that idle & any writes needed by the callback are done |
123 | * before the locations are read in the aio_bh_poll. | |
124 | */ | |
e8d3b1a2 | 125 | atomic_mb_set(&bh->scheduled, 1); |
4f999d05 KW |
126 | } |
127 | ||
128 | void qemu_bh_schedule(QEMUBH *bh) | |
129 | { | |
924fe129 SH |
130 | AioContext *ctx; |
131 | ||
924fe129 | 132 | ctx = bh->ctx; |
4f999d05 | 133 | bh->idle = 0; |
e8d3b1a2 | 134 | /* The memory barrier implicit in atomic_xchg makes sure that: |
924fe129 SH |
135 | * 1. idle & any writes needed by the callback are done before the |
136 | * locations are read in the aio_bh_poll. | |
137 | * 2. ctx is loaded before scheduled is set and the callback has a chance | |
138 | * to execute. | |
dcc772e2 | 139 | */ |
e8d3b1a2 PB |
140 | if (atomic_xchg(&bh->scheduled, 1) == 0) { |
141 | aio_notify(ctx); | |
142 | } | |
4f999d05 KW |
143 | } |
144 | ||
dcc772e2 LPF |
145 | |
146 | /* This func is async. | |
147 | */ | |
4f999d05 KW |
148 | void qemu_bh_cancel(QEMUBH *bh) |
149 | { | |
150 | bh->scheduled = 0; | |
151 | } | |
152 | ||
dcc772e2 LPF |
153 | /* This func is async.The bottom half will do the delete action at the finial |
154 | * end. | |
155 | */ | |
4f999d05 KW |
156 | void qemu_bh_delete(QEMUBH *bh) |
157 | { | |
158 | bh->scheduled = 0; | |
159 | bh->deleted = 1; | |
160 | } | |
161 | ||
845ca10d PB |
162 | int64_t |
163 | aio_compute_timeout(AioContext *ctx) | |
4f999d05 | 164 | { |
845ca10d PB |
165 | int64_t deadline; |
166 | int timeout = -1; | |
4f999d05 KW |
167 | QEMUBH *bh; |
168 | ||
f627aab1 | 169 | for (bh = ctx->first_bh; bh; bh = bh->next) { |
4f999d05 KW |
170 | if (!bh->deleted && bh->scheduled) { |
171 | if (bh->idle) { | |
172 | /* idle bottom halves will be polled at least | |
173 | * every 10ms */ | |
845ca10d | 174 | timeout = 10000000; |
4f999d05 KW |
175 | } else { |
176 | /* non-idle bottom halves will be executed | |
177 | * immediately */ | |
845ca10d | 178 | return 0; |
4f999d05 KW |
179 | } |
180 | } | |
181 | } | |
e3713e00 | 182 | |
845ca10d | 183 | deadline = timerlistgroup_deadline_ns(&ctx->tlg); |
533a8cf3 | 184 | if (deadline == 0) { |
845ca10d | 185 | return 0; |
533a8cf3 | 186 | } else { |
845ca10d | 187 | return qemu_soonest_timeout(timeout, deadline); |
533a8cf3 | 188 | } |
845ca10d | 189 | } |
533a8cf3 | 190 | |
845ca10d PB |
191 | static gboolean |
192 | aio_ctx_prepare(GSource *source, gint *timeout) | |
193 | { | |
194 | AioContext *ctx = (AioContext *) source; | |
195 | ||
eabc9779 PB |
196 | atomic_or(&ctx->notify_me, 1); |
197 | ||
845ca10d PB |
198 | /* We assume there is no timeout already supplied */ |
199 | *timeout = qemu_timeout_ns_to_ms(aio_compute_timeout(ctx)); | |
a3462c65 PB |
200 | |
201 | if (aio_prepare(ctx)) { | |
202 | *timeout = 0; | |
203 | } | |
204 | ||
845ca10d | 205 | return *timeout == 0; |
e3713e00 PB |
206 | } |
207 | ||
208 | static gboolean | |
209 | aio_ctx_check(GSource *source) | |
210 | { | |
211 | AioContext *ctx = (AioContext *) source; | |
212 | QEMUBH *bh; | |
213 | ||
eabc9779 | 214 | atomic_and(&ctx->notify_me, ~1); |
05e514b1 | 215 | aio_notify_accept(ctx); |
21a03d17 | 216 | |
e3713e00 PB |
217 | for (bh = ctx->first_bh; bh; bh = bh->next) { |
218 | if (!bh->deleted && bh->scheduled) { | |
219 | return true; | |
220 | } | |
221 | } | |
533a8cf3 | 222 | return aio_pending(ctx) || (timerlistgroup_deadline_ns(&ctx->tlg) == 0); |
e3713e00 PB |
223 | } |
224 | ||
225 | static gboolean | |
226 | aio_ctx_dispatch(GSource *source, | |
227 | GSourceFunc callback, | |
228 | gpointer user_data) | |
229 | { | |
230 | AioContext *ctx = (AioContext *) source; | |
231 | ||
232 | assert(callback == NULL); | |
e4c7e2d1 | 233 | aio_dispatch(ctx); |
e3713e00 PB |
234 | return true; |
235 | } | |
236 | ||
2f4dc3c1 PB |
237 | static void |
238 | aio_ctx_finalize(GSource *source) | |
239 | { | |
240 | AioContext *ctx = (AioContext *) source; | |
241 | ||
ca96ac44 | 242 | qemu_bh_delete(ctx->notify_dummy_bh); |
9b34277d | 243 | thread_pool_free(ctx->thread_pool); |
a076972a SH |
244 | |
245 | qemu_mutex_lock(&ctx->bh_lock); | |
246 | while (ctx->first_bh) { | |
247 | QEMUBH *next = ctx->first_bh->next; | |
248 | ||
249 | /* qemu_bh_delete() must have been called on BHs in this AioContext */ | |
250 | assert(ctx->first_bh->deleted); | |
251 | ||
252 | g_free(ctx->first_bh); | |
253 | ctx->first_bh = next; | |
254 | } | |
255 | qemu_mutex_unlock(&ctx->bh_lock); | |
256 | ||
dca21ef2 | 257 | aio_set_event_notifier(ctx, &ctx->notifier, false, NULL); |
2f4dc3c1 | 258 | event_notifier_cleanup(&ctx->notifier); |
98563fc3 | 259 | rfifolock_destroy(&ctx->lock); |
dcc772e2 | 260 | qemu_mutex_destroy(&ctx->bh_lock); |
dae21b98 | 261 | timerlistgroup_deinit(&ctx->tlg); |
2f4dc3c1 PB |
262 | } |
263 | ||
e3713e00 PB |
264 | static GSourceFuncs aio_source_funcs = { |
265 | aio_ctx_prepare, | |
266 | aio_ctx_check, | |
267 | aio_ctx_dispatch, | |
2f4dc3c1 | 268 | aio_ctx_finalize |
e3713e00 PB |
269 | }; |
270 | ||
271 | GSource *aio_get_g_source(AioContext *ctx) | |
272 | { | |
273 | g_source_ref(&ctx->source); | |
274 | return &ctx->source; | |
275 | } | |
a915f4bc | 276 | |
9b34277d SH |
277 | ThreadPool *aio_get_thread_pool(AioContext *ctx) |
278 | { | |
279 | if (!ctx->thread_pool) { | |
280 | ctx->thread_pool = thread_pool_new(ctx); | |
281 | } | |
282 | return ctx->thread_pool; | |
283 | } | |
284 | ||
2f4dc3c1 PB |
285 | void aio_notify(AioContext *ctx) |
286 | { | |
eabc9779 PB |
287 | /* Write e.g. bh->scheduled before reading ctx->notify_me. Pairs |
288 | * with atomic_or in aio_ctx_prepare or atomic_add in aio_poll. | |
289 | */ | |
0ceb849b | 290 | smp_mb(); |
eabc9779 | 291 | if (ctx->notify_me) { |
0ceb849b | 292 | event_notifier_set(&ctx->notifier); |
05e514b1 PB |
293 | atomic_mb_set(&ctx->notified, true); |
294 | } | |
295 | } | |
296 | ||
297 | void aio_notify_accept(AioContext *ctx) | |
298 | { | |
299 | if (atomic_xchg(&ctx->notified, false)) { | |
300 | event_notifier_test_and_clear(&ctx->notifier); | |
0ceb849b | 301 | } |
2f4dc3c1 PB |
302 | } |
303 | ||
d5541d86 AB |
304 | static void aio_timerlist_notify(void *opaque) |
305 | { | |
306 | aio_notify(opaque); | |
307 | } | |
308 | ||
da5e1de9 SH |
309 | static void aio_rfifolock_cb(void *opaque) |
310 | { | |
ca96ac44 SH |
311 | AioContext *ctx = opaque; |
312 | ||
da5e1de9 | 313 | /* Kick owner thread in case they are blocked in aio_poll() */ |
ca96ac44 SH |
314 | qemu_bh_schedule(ctx->notify_dummy_bh); |
315 | } | |
316 | ||
317 | static void notify_dummy_bh(void *opaque) | |
318 | { | |
319 | /* Do nothing, we were invoked just to force the event loop to iterate */ | |
da5e1de9 SH |
320 | } |
321 | ||
21a03d17 PB |
322 | static void event_notifier_dummy_cb(EventNotifier *e) |
323 | { | |
324 | } | |
325 | ||
2f78e491 | 326 | AioContext *aio_context_new(Error **errp) |
f627aab1 | 327 | { |
2f78e491 | 328 | int ret; |
2f4dc3c1 | 329 | AioContext *ctx; |
37fcee5d FZ |
330 | Error *local_err = NULL; |
331 | ||
2f4dc3c1 | 332 | ctx = (AioContext *) g_source_new(&aio_source_funcs, sizeof(AioContext)); |
37fcee5d FZ |
333 | aio_context_setup(ctx, &local_err); |
334 | if (local_err) { | |
335 | error_propagate(errp, local_err); | |
336 | goto fail; | |
337 | } | |
2f78e491 CN |
338 | ret = event_notifier_init(&ctx->notifier, false); |
339 | if (ret < 0) { | |
2f78e491 | 340 | error_setg_errno(errp, -ret, "Failed to initialize event notifier"); |
37fcee5d | 341 | goto fail; |
2f78e491 | 342 | } |
fcf5def1 | 343 | g_source_set_can_recurse(&ctx->source, true); |
2f78e491 | 344 | aio_set_event_notifier(ctx, &ctx->notifier, |
dca21ef2 | 345 | false, |
2f78e491 | 346 | (EventNotifierHandler *) |
21a03d17 | 347 | event_notifier_dummy_cb); |
9b34277d | 348 | ctx->thread_pool = NULL; |
dcc772e2 | 349 | qemu_mutex_init(&ctx->bh_lock); |
da5e1de9 | 350 | rfifolock_init(&ctx->lock, aio_rfifolock_cb, ctx); |
d5541d86 | 351 | timerlistgroup_init(&ctx->tlg, aio_timerlist_notify, ctx); |
2f4dc3c1 | 352 | |
ca96ac44 SH |
353 | ctx->notify_dummy_bh = aio_bh_new(ctx, notify_dummy_bh, NULL); |
354 | ||
2f4dc3c1 | 355 | return ctx; |
37fcee5d FZ |
356 | fail: |
357 | g_source_destroy(&ctx->source); | |
358 | return NULL; | |
e3713e00 PB |
359 | } |
360 | ||
361 | void aio_context_ref(AioContext *ctx) | |
362 | { | |
363 | g_source_ref(&ctx->source); | |
364 | } | |
365 | ||
366 | void aio_context_unref(AioContext *ctx) | |
367 | { | |
368 | g_source_unref(&ctx->source); | |
f627aab1 | 369 | } |
98563fc3 SH |
370 | |
371 | void aio_context_acquire(AioContext *ctx) | |
372 | { | |
373 | rfifolock_lock(&ctx->lock); | |
374 | } | |
375 | ||
376 | void aio_context_release(AioContext *ctx) | |
377 | { | |
378 | rfifolock_unlock(&ctx->lock); | |
379 | } |