2 * QEMU aio implementation
4 * Copyright IBM, Corp. 2008
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
17 #include "qemu/typedefs.h"
18 #include "qemu-common.h"
19 #include "qemu/queue.h"
20 #include "qemu/event_notifier.h"
21 #include "qemu/thread.h"
22 #include "qemu/rfifolock.h"
23 #include "qemu/timer.h"
25 typedef struct BlockAIOCB BlockAIOCB;
26 typedef void BlockCompletionFunc(void *opaque, int ret);
28 typedef struct AIOCBInfo {
29 void (*cancel_async)(BlockAIOCB *acb);
30 AioContext *(*get_aio_context)(BlockAIOCB *acb);
35 const AIOCBInfo *aiocb_info;
37 BlockCompletionFunc *cb;
42 void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
43 BlockCompletionFunc *cb, void *opaque);
44 void qemu_aio_unref(void *p);
45 void qemu_aio_ref(void *p);
47 typedef struct AioHandler AioHandler;
48 typedef void QEMUBHFunc(void *opaque);
49 typedef void IOHandler(void *opaque);
54 /* Protects all fields from multi-threaded access */
57 /* The list of registered AIO handlers */
58 QLIST_HEAD(, AioHandler) aio_handlers;
60 /* This is a simple lock used to protect the aio_handlers list.
61 * Specifically, it's used to ensure that no callbacks are removed while
62 * we're walking and dispatching callbacks.
66 /* Used to avoid unnecessary event_notifier_set calls in aio_notify.
67 * Writes protected by lock or BQL, reads are lockless.
71 /* lock to protect between bh's adders and deleter */
74 /* Anchor of the list of Bottom Halves belonging to the context */
75 struct QEMUBH *first_bh;
77 /* A simple lock used to protect the first_bh list, and ensure that
78 * no callbacks are removed while we're walking and dispatching callbacks.
82 /* Used for aio_notify. */
83 EventNotifier notifier;
85 /* GPollFDs for aio_poll() */
88 /* Thread pool for performing work and receiving completion callbacks */
89 struct ThreadPool *thread_pool;
91 /* TimerLists for calling timers - one per clock type */
92 QEMUTimerListGroup tlg;
95 /* Used internally to synchronize aio_poll against qemu_bh_schedule. */
96 void aio_set_dispatching(AioContext *ctx, bool dispatching);
99 * aio_context_new: Allocate a new AioContext.
101 * AioContext provide a mini event-loop that can be waited on synchronously.
102 * They also provide bottom halves, a service to execute a piece of code
103 * as soon as possible.
105 AioContext *aio_context_new(Error **errp);
109 * @ctx: The AioContext to operate on.
111 * Add a reference to an AioContext.
113 void aio_context_ref(AioContext *ctx);
117 * @ctx: The AioContext to operate on.
119 * Drop a reference to an AioContext.
121 void aio_context_unref(AioContext *ctx);
123 /* Take ownership of the AioContext. If the AioContext will be shared between
124 * threads, a thread must have ownership when calling aio_poll().
126 * Note that multiple threads calling aio_poll() means timers, BHs, and
127 * callbacks may be invoked from a different thread than they were registered
128 * from. Therefore, code must use AioContext acquire/release or use
129 * fine-grained synchronization to protect shared state if other threads will
130 * be accessing it simultaneously.
132 void aio_context_acquire(AioContext *ctx);
134 /* Relinquish ownership of the AioContext. */
135 void aio_context_release(AioContext *ctx);
138 * aio_bh_new: Allocate a new bottom half structure.
140 * Bottom halves are lightweight callbacks whose invocation is guaranteed
141 * to be wait-free, thread-safe and signal-safe. The #QEMUBH structure
142 * is opaque and must be allocated prior to its use.
144 QEMUBH *aio_bh_new(AioContext *ctx, QEMUBHFunc *cb, void *opaque);
147 * aio_notify: Force processing of pending events.
149 * Similar to signaling a condition variable, aio_notify forces
150 * aio_wait to exit, so that the next call will re-examine pending events.
151 * The caller of aio_notify will usually call aio_wait again very soon,
152 * or go through another iteration of the GLib main loop. Hence, aio_notify
153 * also has the side effect of recalculating the sets of file descriptors
154 * that the main loop waits for.
156 * Calling aio_notify is rarely necessary, because for example scheduling
157 * a bottom half calls it already.
159 void aio_notify(AioContext *ctx);
162 * aio_bh_poll: Poll bottom halves for an AioContext.
164 * These are internal functions used by the QEMU main loop.
165 * And notice that multiple occurrences of aio_bh_poll cannot
166 * be called concurrently
168 int aio_bh_poll(AioContext *ctx);
171 * qemu_bh_schedule: Schedule a bottom half.
173 * Scheduling a bottom half interrupts the main loop and causes the
174 * execution of the callback that was passed to qemu_bh_new.
176 * Bottom halves that are scheduled from a bottom half handler are instantly
177 * invoked. This can create an infinite loop if a bottom half handler
180 * @bh: The bottom half to be scheduled.
182 void qemu_bh_schedule(QEMUBH *bh);
185 * qemu_bh_cancel: Cancel execution of a bottom half.
187 * Canceling execution of a bottom half undoes the effect of calls to
188 * qemu_bh_schedule without freeing its resources yet. While cancellation
189 * itself is also wait-free and thread-safe, it can of course race with the
190 * loop that executes bottom halves unless you are holding the iothread
191 * mutex. This makes it mostly useless if you are not holding the mutex.
193 * @bh: The bottom half to be canceled.
195 void qemu_bh_cancel(QEMUBH *bh);
198 *qemu_bh_delete: Cancel execution of a bottom half and free its resources.
200 * Deleting a bottom half frees the memory that was allocated for it by
201 * qemu_bh_new. It also implies canceling the bottom half if it was
203 * This func is async. The bottom half will do the delete action at the finial
206 * @bh: The bottom half to be deleted.
208 void qemu_bh_delete(QEMUBH *bh);
210 /* Return whether there are any pending callbacks from the GSource
211 * attached to the AioContext, before g_poll is invoked.
213 * This is used internally in the implementation of the GSource.
215 bool aio_prepare(AioContext *ctx);
217 /* Return whether there are any pending callbacks from the GSource
218 * attached to the AioContext, after g_poll is invoked.
220 * This is used internally in the implementation of the GSource.
222 bool aio_pending(AioContext *ctx);
224 /* Dispatch any pending callbacks from the GSource attached to the AioContext.
226 * This is used internally in the implementation of the GSource.
228 bool aio_dispatch(AioContext *ctx);
230 /* Progress in completing AIO work to occur. This can issue new pending
231 * aio as a result of executing I/O completion or bh callbacks.
233 * Return whether any progress was made by executing AIO or bottom half
234 * handlers. If @blocking == true, this should always be true except
235 * if someone called aio_notify.
237 * If there are no pending bottom halves, but there are pending AIO
238 * operations, it may not be possible to make any progress without
239 * blocking. If @blocking is true, this function will wait until one
240 * or more AIO events have completed, to ensure something has moved
243 bool aio_poll(AioContext *ctx, bool blocking);
245 /* Register a file descriptor and associated callbacks. Behaves very similarly
246 * to qemu_set_fd_handler2. Unlike qemu_set_fd_handler2, these callbacks will
247 * be invoked when using aio_poll().
249 * Code that invokes AIO completion functions should rely on this function
250 * instead of qemu_set_fd_handler[2].
252 void aio_set_fd_handler(AioContext *ctx,
258 /* Register an event notifier and associated callbacks. Behaves very similarly
259 * to event_notifier_set_handler. Unlike event_notifier_set_handler, these callbacks
260 * will be invoked when using aio_poll().
262 * Code that invokes AIO completion functions should rely on this function
263 * instead of event_notifier_set_handler.
265 void aio_set_event_notifier(AioContext *ctx,
266 EventNotifier *notifier,
267 EventNotifierHandler *io_read);
269 /* Return a GSource that lets the main loop poll the file descriptors attached
270 * to this AioContext.
272 GSource *aio_get_g_source(AioContext *ctx);
274 /* Return the ThreadPool bound to this AioContext */
275 struct ThreadPool *aio_get_thread_pool(AioContext *ctx);
279 * @ctx: the aio context
280 * @type: the clock type
282 * @cb: the callback to call on timer expiry
283 * @opaque: the opaque pointer to pass to the callback
285 * Allocate a new timer attached to the context @ctx.
286 * The function is responsible for memory allocation.
288 * The preferred interface is aio_timer_init. Use that
289 * unless you really need dynamic memory allocation.
291 * Returns: a pointer to the new timer
293 static inline QEMUTimer *aio_timer_new(AioContext *ctx, QEMUClockType type,
295 QEMUTimerCB *cb, void *opaque)
297 return timer_new_tl(ctx->tlg.tl[type], scale, cb, opaque);
302 * @ctx: the aio context
304 * @type: the clock type
306 * @cb: the callback to call on timer expiry
307 * @opaque: the opaque pointer to pass to the callback
309 * Initialise a new timer attached to the context @ctx.
310 * The caller is responsible for memory allocation.
312 static inline void aio_timer_init(AioContext *ctx,
313 QEMUTimer *ts, QEMUClockType type,
315 QEMUTimerCB *cb, void *opaque)
317 timer_init(ts, ctx->tlg.tl[type], scale, cb, opaque);
321 * aio_compute_timeout:
322 * @ctx: the aio context
324 * Compute the timeout that a blocking aio_poll should use.
326 int64_t aio_compute_timeout(AioContext *ctx);