]>
Commit | Line | Data |
---|---|---|
a76bab49 AL |
1 | /* |
2 | * QEMU aio implementation | |
3 | * | |
4 | * Copyright IBM, Corp. 2008 | |
5 | * | |
6 | * Authors: | |
7 | * Anthony Liguori <aliguori@us.ibm.com> | |
8 | * | |
9 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
10 | * the COPYING file in the top-level directory. | |
11 | * | |
12 | */ | |
13 | ||
14 | #ifndef QEMU_AIO_H | |
15 | #define QEMU_AIO_H | |
16 | ||
17 | #include "qemu-common.h" | |
1de7afc9 PB |
18 | #include "qemu/queue.h" |
19 | #include "qemu/event_notifier.h" | |
dcc772e2 | 20 | #include "qemu/thread.h" |
dae21b98 | 21 | #include "qemu/timer.h" |
a76bab49 | 22 | |
7c84b1b8 | 23 | typedef struct BlockAIOCB BlockAIOCB; |
097310b5 | 24 | typedef void BlockCompletionFunc(void *opaque, int ret); |
85e8dab1 | 25 | |
d7331bed | 26 | typedef struct AIOCBInfo { |
7c84b1b8 MA |
27 | void (*cancel_async)(BlockAIOCB *acb); |
28 | AioContext *(*get_aio_context)(BlockAIOCB *acb); | |
8c82e9a4 | 29 | size_t aiocb_size; |
d7331bed | 30 | } AIOCBInfo; |
85e8dab1 | 31 | |
7c84b1b8 | 32 | struct BlockAIOCB { |
d7331bed | 33 | const AIOCBInfo *aiocb_info; |
85e8dab1 | 34 | BlockDriverState *bs; |
097310b5 | 35 | BlockCompletionFunc *cb; |
85e8dab1 | 36 | void *opaque; |
f197fe2b | 37 | int refcnt; |
85e8dab1 PB |
38 | }; |
39 | ||
d7331bed | 40 | void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs, |
097310b5 | 41 | BlockCompletionFunc *cb, void *opaque); |
8007429a | 42 | void qemu_aio_unref(void *p); |
f197fe2b | 43 | void qemu_aio_ref(void *p); |
85e8dab1 | 44 | |
f627aab1 PB |
45 | typedef struct AioHandler AioHandler; |
46 | typedef void QEMUBHFunc(void *opaque); | |
47 | typedef void IOHandler(void *opaque); | |
48 | ||
0187f5c9 PB |
49 | struct ThreadPool; |
50 | struct LinuxAioState; | |
51 | ||
6a1751b7 | 52 | struct AioContext { |
e3713e00 PB |
53 | GSource source; |
54 | ||
98563fc3 | 55 | /* Protects all fields from multi-threaded access */ |
3fe71223 | 56 | QemuRecMutex lock; |
98563fc3 | 57 | |
a915f4bc PB |
58 | /* The list of registered AIO handlers */ |
59 | QLIST_HEAD(, AioHandler) aio_handlers; | |
60 | ||
61 | /* This is a simple lock used to protect the aio_handlers list. | |
62 | * Specifically, it's used to ensure that no callbacks are removed while | |
63 | * we're walking and dispatching callbacks. | |
64 | */ | |
65 | int walking_handlers; | |
66 | ||
eabc9779 PB |
67 | /* Used to avoid unnecessary event_notifier_set calls in aio_notify; |
68 | * accessed with atomic primitives. If this field is 0, everything | |
69 | * (file descriptors, bottom halves, timers) will be re-evaluated | |
70 | * before the next blocking poll(), thus the event_notifier_set call | |
71 | * can be skipped. If it is non-zero, you may need to wake up a | |
72 | * concurrent aio_poll or the glib main event loop, making | |
73 | * event_notifier_set necessary. | |
74 | * | |
75 | * Bit 0 is reserved for GSource usage of the AioContext, and is 1 | |
54a16a63 | 76 | * between a call to aio_ctx_prepare and the next call to aio_ctx_check. |
eabc9779 PB |
77 | * Bits 1-31 simply count the number of active calls to aio_poll |
78 | * that are in the prepare or poll phase. | |
79 | * | |
80 | * The GSource and aio_poll must use a different mechanism because | |
81 | * there is no certainty that a call to GSource's prepare callback | |
82 | * (via g_main_context_prepare) is indeed followed by check and | |
83 | * dispatch. It's not clear whether this would be a bug, but let's | |
84 | * play safe and allow it---it will just cause extra calls to | |
85 | * event_notifier_set until the next call to dispatch. | |
86 | * | |
87 | * Instead, the aio_poll calls include both the prepare and the | |
88 | * dispatch phase, hence a simple counter is enough for them. | |
0ceb849b | 89 | */ |
eabc9779 | 90 | uint32_t notify_me; |
0ceb849b | 91 | |
dcc772e2 LPF |
92 | /* lock to protect between bh's adders and deleter */ |
93 | QemuMutex bh_lock; | |
0ceb849b | 94 | |
f627aab1 PB |
95 | /* Anchor of the list of Bottom Halves belonging to the context */ |
96 | struct QEMUBH *first_bh; | |
97 | ||
98 | /* A simple lock used to protect the first_bh list, and ensure that | |
99 | * no callbacks are removed while we're walking and dispatching callbacks. | |
100 | */ | |
101 | int walking_bh; | |
2f4dc3c1 | 102 | |
05e514b1 PB |
103 | /* Used by aio_notify. |
104 | * | |
105 | * "notified" is used to avoid expensive event_notifier_test_and_clear | |
106 | * calls. When it is clear, the EventNotifier is clear, or one thread | |
107 | * is going to clear "notified" before processing more events. False | |
108 | * positives are possible, i.e. "notified" could be set even though the | |
109 | * EventNotifier is clear. | |
110 | * | |
111 | * Note that event_notifier_set *cannot* be optimized the same way. For | |
112 | * more information on the problem that would result, see "#ifdef BUG2" | |
113 | * in the docs/aio_notify_accept.promela formal model. | |
114 | */ | |
115 | bool notified; | |
2f4dc3c1 | 116 | EventNotifier notifier; |
6b5f8762 | 117 | |
9b34277d SH |
118 | /* Thread pool for performing work and receiving completion callbacks */ |
119 | struct ThreadPool *thread_pool; | |
dae21b98 | 120 | |
0187f5c9 PB |
121 | #ifdef CONFIG_LINUX_AIO |
122 | /* State for native Linux AIO. Uses aio_context_acquire/release for | |
123 | * locking. | |
124 | */ | |
125 | struct LinuxAioState *linux_aio; | |
126 | #endif | |
127 | ||
dae21b98 AB |
128 | /* TimerLists for calling timers - one per clock type */ |
129 | QEMUTimerListGroup tlg; | |
c1e1e5fa FZ |
130 | |
131 | int external_disable_cnt; | |
fbe3fc5c FZ |
132 | |
133 | /* epoll(7) state used when built with CONFIG_EPOLL */ | |
134 | int epollfd; | |
135 | bool epoll_enabled; | |
136 | bool epoll_available; | |
6a1751b7 | 137 | }; |
f627aab1 | 138 | |
f627aab1 PB |
139 | /** |
140 | * aio_context_new: Allocate a new AioContext. | |
141 | * | |
142 | * AioContext provide a mini event-loop that can be waited on synchronously. | |
143 | * They also provide bottom halves, a service to execute a piece of code | |
144 | * as soon as possible. | |
145 | */ | |
2f78e491 | 146 | AioContext *aio_context_new(Error **errp); |
f627aab1 | 147 | |
e3713e00 PB |
148 | /** |
149 | * aio_context_ref: | |
150 | * @ctx: The AioContext to operate on. | |
151 | * | |
152 | * Add a reference to an AioContext. | |
153 | */ | |
154 | void aio_context_ref(AioContext *ctx); | |
155 | ||
156 | /** | |
157 | * aio_context_unref: | |
158 | * @ctx: The AioContext to operate on. | |
159 | * | |
160 | * Drop a reference to an AioContext. | |
161 | */ | |
162 | void aio_context_unref(AioContext *ctx); | |
163 | ||
98563fc3 | 164 | /* Take ownership of the AioContext. If the AioContext will be shared between |
49110174 PB |
165 | * threads, and a thread does not want to be interrupted, it will have to |
166 | * take ownership around calls to aio_poll(). Otherwise, aio_poll() | |
167 | * automatically takes care of calling aio_context_acquire and | |
168 | * aio_context_release. | |
98563fc3 | 169 | * |
49110174 PB |
170 | * Access to timers and BHs from a thread that has not acquired AioContext |
171 | * is possible. Access to callbacks for now must be done while the AioContext | |
172 | * is owned by the thread (FIXME). | |
98563fc3 SH |
173 | */ |
174 | void aio_context_acquire(AioContext *ctx); | |
175 | ||
176 | /* Relinquish ownership of the AioContext. */ | |
177 | void aio_context_release(AioContext *ctx); | |
178 | ||
5b8bb359 PB |
179 | /** |
180 | * aio_bh_schedule_oneshot: Allocate a new bottom half structure that will run | |
181 | * only once and as soon as possible. | |
182 | */ | |
183 | void aio_bh_schedule_oneshot(AioContext *ctx, QEMUBHFunc *cb, void *opaque); | |
184 | ||
f627aab1 PB |
185 | /** |
186 | * aio_bh_new: Allocate a new bottom half structure. | |
187 | * | |
188 | * Bottom halves are lightweight callbacks whose invocation is guaranteed | |
189 | * to be wait-free, thread-safe and signal-safe. The #QEMUBH structure | |
190 | * is opaque and must be allocated prior to its use. | |
191 | */ | |
192 | QEMUBH *aio_bh_new(AioContext *ctx, QEMUBHFunc *cb, void *opaque); | |
193 | ||
2f4dc3c1 PB |
194 | /** |
195 | * aio_notify: Force processing of pending events. | |
196 | * | |
197 | * Similar to signaling a condition variable, aio_notify forces | |
722f8d90 YB |
198 | * aio_poll to exit, so that the next call will re-examine pending events. |
199 | * The caller of aio_notify will usually call aio_poll again very soon, | |
2f4dc3c1 PB |
200 | * or go through another iteration of the GLib main loop. Hence, aio_notify |
201 | * also has the side effect of recalculating the sets of file descriptors | |
202 | * that the main loop waits for. | |
203 | * | |
204 | * Calling aio_notify is rarely necessary, because for example scheduling | |
205 | * a bottom half calls it already. | |
206 | */ | |
207 | void aio_notify(AioContext *ctx); | |
208 | ||
05e514b1 PB |
209 | /** |
210 | * aio_notify_accept: Acknowledge receiving an aio_notify. | |
211 | * | |
212 | * aio_notify() uses an EventNotifier in order to wake up a sleeping | |
213 | * aio_poll() or g_main_context_iteration(). Calls to aio_notify() are | |
214 | * usually rare, but the AioContext has to clear the EventNotifier on | |
215 | * every aio_poll() or g_main_context_iteration() in order to avoid | |
216 | * busy waiting. This event_notifier_test_and_clear() cannot be done | |
217 | * using the usual aio_context_set_event_notifier(), because it must | |
218 | * be done before processing all events (file descriptors, bottom halves, | |
219 | * timers). | |
220 | * | |
221 | * aio_notify_accept() is an optimized event_notifier_test_and_clear() | |
222 | * that is specific to an AioContext's notifier; it is used internally | |
223 | * to clear the EventNotifier only if aio_notify() had been called. | |
224 | */ | |
225 | void aio_notify_accept(AioContext *ctx); | |
226 | ||
df281b80 PD |
227 | /** |
228 | * aio_bh_call: Executes callback function of the specified BH. | |
229 | */ | |
230 | void aio_bh_call(QEMUBH *bh); | |
231 | ||
f627aab1 PB |
232 | /** |
233 | * aio_bh_poll: Poll bottom halves for an AioContext. | |
234 | * | |
235 | * These are internal functions used by the QEMU main loop. | |
dcc772e2 LPF |
236 | * And notice that multiple occurrences of aio_bh_poll cannot |
237 | * be called concurrently | |
f627aab1 PB |
238 | */ |
239 | int aio_bh_poll(AioContext *ctx); | |
f627aab1 PB |
240 | |
241 | /** | |
242 | * qemu_bh_schedule: Schedule a bottom half. | |
243 | * | |
244 | * Scheduling a bottom half interrupts the main loop and causes the | |
245 | * execution of the callback that was passed to qemu_bh_new. | |
246 | * | |
247 | * Bottom halves that are scheduled from a bottom half handler are instantly | |
248 | * invoked. This can create an infinite loop if a bottom half handler | |
249 | * schedules itself. | |
250 | * | |
251 | * @bh: The bottom half to be scheduled. | |
252 | */ | |
253 | void qemu_bh_schedule(QEMUBH *bh); | |
254 | ||
255 | /** | |
256 | * qemu_bh_cancel: Cancel execution of a bottom half. | |
257 | * | |
258 | * Canceling execution of a bottom half undoes the effect of calls to | |
259 | * qemu_bh_schedule without freeing its resources yet. While cancellation | |
260 | * itself is also wait-free and thread-safe, it can of course race with the | |
261 | * loop that executes bottom halves unless you are holding the iothread | |
262 | * mutex. This makes it mostly useless if you are not holding the mutex. | |
263 | * | |
264 | * @bh: The bottom half to be canceled. | |
265 | */ | |
266 | void qemu_bh_cancel(QEMUBH *bh); | |
267 | ||
268 | /** | |
269 | *qemu_bh_delete: Cancel execution of a bottom half and free its resources. | |
270 | * | |
271 | * Deleting a bottom half frees the memory that was allocated for it by | |
272 | * qemu_bh_new. It also implies canceling the bottom half if it was | |
273 | * scheduled. | |
dcc772e2 LPF |
274 | * This func is async. The bottom half will do the delete action at the finial |
275 | * end. | |
f627aab1 PB |
276 | * |
277 | * @bh: The bottom half to be deleted. | |
278 | */ | |
279 | void qemu_bh_delete(QEMUBH *bh); | |
280 | ||
cd9ba1eb | 281 | /* Return whether there are any pending callbacks from the GSource |
a3462c65 PB |
282 | * attached to the AioContext, before g_poll is invoked. |
283 | * | |
284 | * This is used internally in the implementation of the GSource. | |
285 | */ | |
286 | bool aio_prepare(AioContext *ctx); | |
287 | ||
288 | /* Return whether there are any pending callbacks from the GSource | |
289 | * attached to the AioContext, after g_poll is invoked. | |
cd9ba1eb PB |
290 | * |
291 | * This is used internally in the implementation of the GSource. | |
292 | */ | |
293 | bool aio_pending(AioContext *ctx); | |
294 | ||
e4c7e2d1 PB |
295 | /* Dispatch any pending callbacks from the GSource attached to the AioContext. |
296 | * | |
297 | * This is used internally in the implementation of the GSource. | |
721671ad SH |
298 | * |
299 | * @dispatch_fds: true to process fds, false to skip them | |
300 | * (can be used as an optimization by callers that know there | |
301 | * are no fds ready) | |
e4c7e2d1 | 302 | */ |
721671ad | 303 | bool aio_dispatch(AioContext *ctx, bool dispatch_fds); |
e4c7e2d1 | 304 | |
7c0628b2 PB |
305 | /* Progress in completing AIO work to occur. This can issue new pending |
306 | * aio as a result of executing I/O completion or bh callbacks. | |
bcdc1857 | 307 | * |
acfb23ad PB |
308 | * Return whether any progress was made by executing AIO or bottom half |
309 | * handlers. If @blocking == true, this should always be true except | |
310 | * if someone called aio_notify. | |
7c0628b2 PB |
311 | * |
312 | * If there are no pending bottom halves, but there are pending AIO | |
313 | * operations, it may not be possible to make any progress without | |
314 | * blocking. If @blocking is true, this function will wait until one | |
315 | * or more AIO events have completed, to ensure something has moved | |
316 | * before returning. | |
7c0628b2 PB |
317 | */ |
318 | bool aio_poll(AioContext *ctx, bool blocking); | |
a76bab49 AL |
319 | |
320 | /* Register a file descriptor and associated callbacks. Behaves very similarly | |
6484e422 | 321 | * to qemu_set_fd_handler. Unlike qemu_set_fd_handler, these callbacks will |
87f68d31 | 322 | * be invoked when using aio_poll(). |
a76bab49 AL |
323 | * |
324 | * Code that invokes AIO completion functions should rely on this function | |
325 | * instead of qemu_set_fd_handler[2]. | |
326 | */ | |
a915f4bc PB |
327 | void aio_set_fd_handler(AioContext *ctx, |
328 | int fd, | |
dca21ef2 | 329 | bool is_external, |
a915f4bc PB |
330 | IOHandler *io_read, |
331 | IOHandler *io_write, | |
a915f4bc | 332 | void *opaque); |
9958c351 PB |
333 | |
334 | /* Register an event notifier and associated callbacks. Behaves very similarly | |
335 | * to event_notifier_set_handler. Unlike event_notifier_set_handler, these callbacks | |
87f68d31 | 336 | * will be invoked when using aio_poll(). |
9958c351 PB |
337 | * |
338 | * Code that invokes AIO completion functions should rely on this function | |
339 | * instead of event_notifier_set_handler. | |
340 | */ | |
a915f4bc PB |
341 | void aio_set_event_notifier(AioContext *ctx, |
342 | EventNotifier *notifier, | |
dca21ef2 | 343 | bool is_external, |
f2e5dca4 | 344 | EventNotifierHandler *io_read); |
a915f4bc | 345 | |
e3713e00 PB |
346 | /* Return a GSource that lets the main loop poll the file descriptors attached |
347 | * to this AioContext. | |
348 | */ | |
349 | GSource *aio_get_g_source(AioContext *ctx); | |
350 | ||
9b34277d SH |
351 | /* Return the ThreadPool bound to this AioContext */ |
352 | struct ThreadPool *aio_get_thread_pool(AioContext *ctx); | |
353 | ||
0187f5c9 PB |
354 | /* Return the LinuxAioState bound to this AioContext */ |
355 | struct LinuxAioState *aio_get_linux_aio(AioContext *ctx); | |
356 | ||
4e29e831 AB |
357 | /** |
358 | * aio_timer_new: | |
359 | * @ctx: the aio context | |
360 | * @type: the clock type | |
361 | * @scale: the scale | |
362 | * @cb: the callback to call on timer expiry | |
363 | * @opaque: the opaque pointer to pass to the callback | |
364 | * | |
365 | * Allocate a new timer attached to the context @ctx. | |
366 | * The function is responsible for memory allocation. | |
367 | * | |
368 | * The preferred interface is aio_timer_init. Use that | |
369 | * unless you really need dynamic memory allocation. | |
370 | * | |
371 | * Returns: a pointer to the new timer | |
372 | */ | |
373 | static inline QEMUTimer *aio_timer_new(AioContext *ctx, QEMUClockType type, | |
374 | int scale, | |
375 | QEMUTimerCB *cb, void *opaque) | |
376 | { | |
377 | return timer_new_tl(ctx->tlg.tl[type], scale, cb, opaque); | |
378 | } | |
379 | ||
380 | /** | |
381 | * aio_timer_init: | |
382 | * @ctx: the aio context | |
383 | * @ts: the timer | |
384 | * @type: the clock type | |
385 | * @scale: the scale | |
386 | * @cb: the callback to call on timer expiry | |
387 | * @opaque: the opaque pointer to pass to the callback | |
388 | * | |
389 | * Initialise a new timer attached to the context @ctx. | |
390 | * The caller is responsible for memory allocation. | |
391 | */ | |
392 | static inline void aio_timer_init(AioContext *ctx, | |
393 | QEMUTimer *ts, QEMUClockType type, | |
394 | int scale, | |
395 | QEMUTimerCB *cb, void *opaque) | |
396 | { | |
f186aa97 | 397 | timer_init_tl(ts, ctx->tlg.tl[type], scale, cb, opaque); |
4e29e831 AB |
398 | } |
399 | ||
845ca10d PB |
400 | /** |
401 | * aio_compute_timeout: | |
402 | * @ctx: the aio context | |
403 | * | |
404 | * Compute the timeout that a blocking aio_poll should use. | |
405 | */ | |
406 | int64_t aio_compute_timeout(AioContext *ctx); | |
407 | ||
c1e1e5fa FZ |
408 | /** |
409 | * aio_disable_external: | |
410 | * @ctx: the aio context | |
411 | * | |
412 | * Disable the further processing of external clients. | |
413 | */ | |
414 | static inline void aio_disable_external(AioContext *ctx) | |
415 | { | |
416 | atomic_inc(&ctx->external_disable_cnt); | |
417 | } | |
418 | ||
419 | /** | |
420 | * aio_enable_external: | |
421 | * @ctx: the aio context | |
422 | * | |
423 | * Enable the processing of external clients. | |
424 | */ | |
425 | static inline void aio_enable_external(AioContext *ctx) | |
426 | { | |
427 | assert(ctx->external_disable_cnt > 0); | |
428 | atomic_dec(&ctx->external_disable_cnt); | |
429 | } | |
430 | ||
5ceb9e39 FZ |
431 | /** |
432 | * aio_external_disabled: | |
433 | * @ctx: the aio context | |
434 | * | |
435 | * Return true if the external clients are disabled. | |
436 | */ | |
437 | static inline bool aio_external_disabled(AioContext *ctx) | |
438 | { | |
439 | return atomic_read(&ctx->external_disable_cnt); | |
440 | } | |
441 | ||
c1e1e5fa FZ |
442 | /** |
443 | * aio_node_check: | |
444 | * @ctx: the aio context | |
445 | * @is_external: Whether or not the checked node is an external event source. | |
446 | * | |
447 | * Check if the node's is_external flag is okay to be polled by the ctx at this | |
448 | * moment. True means green light. | |
449 | */ | |
450 | static inline bool aio_node_check(AioContext *ctx, bool is_external) | |
451 | { | |
452 | return !is_external || !atomic_read(&ctx->external_disable_cnt); | |
453 | } | |
454 | ||
e4370165 PB |
455 | /** |
456 | * Return the AioContext whose event loop runs in the current thread. | |
457 | * | |
458 | * If called from an IOThread this will be the IOThread's AioContext. If | |
459 | * called from another thread it will be the main loop AioContext. | |
460 | */ | |
461 | AioContext *qemu_get_current_aio_context(void); | |
462 | ||
463 | /** | |
464 | * @ctx: the aio context | |
465 | * | |
466 | * Return whether we are running in the I/O thread that manages @ctx. | |
467 | */ | |
468 | static inline bool aio_context_in_iothread(AioContext *ctx) | |
469 | { | |
470 | return ctx == qemu_get_current_aio_context(); | |
471 | } | |
472 | ||
37fcee5d FZ |
473 | /** |
474 | * aio_context_setup: | |
475 | * @ctx: the aio context | |
476 | * | |
477 | * Initialize the aio context. | |
478 | */ | |
7e003465 | 479 | void aio_context_setup(AioContext *ctx); |
37fcee5d | 480 | |
a76bab49 | 481 | #endif |