xref: /openbmc/qemu/include/block/aio.h (revision d044b7c3)
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 #ifdef CONFIG_LINUX_IO_URING
18 #include <liburing.h>
19 #endif
20 #include "qemu/coroutine-core.h"
21 #include "qemu/queue.h"
22 #include "qemu/event_notifier.h"
23 #include "qemu/thread.h"
24 #include "qemu/timer.h"
25 #include "block/graph-lock.h"
26 
27 typedef struct BlockAIOCB BlockAIOCB;
28 typedef void BlockCompletionFunc(void *opaque, int ret);
29 
30 typedef struct AIOCBInfo {
31     void (*cancel_async)(BlockAIOCB *acb);
32     AioContext *(*get_aio_context)(BlockAIOCB *acb);
33     size_t aiocb_size;
34 } AIOCBInfo;
35 
36 struct BlockAIOCB {
37     const AIOCBInfo *aiocb_info;
38     BlockDriverState *bs;
39     BlockCompletionFunc *cb;
40     void *opaque;
41     int refcnt;
42 };
43 
44 void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
45                    BlockCompletionFunc *cb, void *opaque);
46 void qemu_aio_unref(void *p);
47 void qemu_aio_ref(void *p);
48 
49 typedef struct AioHandler AioHandler;
50 typedef QLIST_HEAD(, AioHandler) AioHandlerList;
51 typedef void QEMUBHFunc(void *opaque);
52 typedef bool AioPollFn(void *opaque);
53 typedef void IOHandler(void *opaque);
54 
55 struct ThreadPool;
56 struct LinuxAioState;
57 struct LuringState;
58 
59 /* Is polling disabled? */
60 bool aio_poll_disabled(AioContext *ctx);
61 
62 /* Callbacks for file descriptor monitoring implementations */
63 typedef struct {
64     /*
65      * update:
66      * @ctx: the AioContext
67      * @old_node: the existing handler or NULL if this file descriptor is being
68      *            monitored for the first time
69      * @new_node: the new handler or NULL if this file descriptor is being
70      *            removed
71      *
72      * Add/remove/modify a monitored file descriptor.
73      *
74      * Called with ctx->list_lock acquired.
75      */
76     void (*update)(AioContext *ctx, AioHandler *old_node, AioHandler *new_node);
77 
78     /*
79      * wait:
80      * @ctx: the AioContext
81      * @ready_list: list for handlers that become ready
82      * @timeout: maximum duration to wait, in nanoseconds
83      *
84      * Wait for file descriptors to become ready and place them on ready_list.
85      *
86      * Called with ctx->list_lock incremented but not locked.
87      *
88      * Returns: number of ready file descriptors.
89      */
90     int (*wait)(AioContext *ctx, AioHandlerList *ready_list, int64_t timeout);
91 
92     /*
93      * need_wait:
94      * @ctx: the AioContext
95      *
96      * Tell aio_poll() when to stop userspace polling early because ->wait()
97      * has fds ready.
98      *
99      * File descriptor monitoring implementations that cannot poll fd readiness
100      * from userspace should use aio_poll_disabled() here.  This ensures that
101      * file descriptors are not starved by handlers that frequently make
102      * progress via userspace polling.
103      *
104      * Returns: true if ->wait() should be called, false otherwise.
105      */
106     bool (*need_wait)(AioContext *ctx);
107 } FDMonOps;
108 
109 /*
110  * Each aio_bh_poll() call carves off a slice of the BH list, so that newly
111  * scheduled BHs are not processed until the next aio_bh_poll() call.  All
112  * active aio_bh_poll() calls chain their slices together in a list, so that
113  * nested aio_bh_poll() calls process all scheduled bottom halves.
114  */
115 typedef QSLIST_HEAD(, QEMUBH) BHList;
116 typedef struct BHListSlice BHListSlice;
117 struct BHListSlice {
118     BHList bh_list;
119     QSIMPLEQ_ENTRY(BHListSlice) next;
120 };
121 
122 typedef QSLIST_HEAD(, AioHandler) AioHandlerSList;
123 
124 struct AioContext {
125     GSource source;
126 
127     /* Used by AioContext users to protect from multi-threaded access.  */
128     QemuRecMutex lock;
129 
130     /*
131      * Keep track of readers and writers of the block layer graph.
132      * This is essential to avoid performing additions and removal
133      * of nodes and edges from block graph while some
134      * other thread is traversing it.
135      */
136     BdrvGraphRWlock *bdrv_graph;
137 
138     /* The list of registered AIO handlers.  Protected by ctx->list_lock. */
139     AioHandlerList aio_handlers;
140 
141     /* The list of AIO handlers to be deleted.  Protected by ctx->list_lock. */
142     AioHandlerList deleted_aio_handlers;
143 
144     /* Used to avoid unnecessary event_notifier_set calls in aio_notify;
145      * only written from the AioContext home thread, or under the BQL in
146      * the case of the main AioContext.  However, it is read from any
147      * thread so it is still accessed with atomic primitives.
148      *
149      * If this field is 0, everything (file descriptors, bottom halves,
150      * timers) will be re-evaluated before the next blocking poll() or
151      * io_uring wait; therefore, the event_notifier_set call can be
152      * skipped.  If it is non-zero, you may need to wake up a concurrent
153      * aio_poll or the glib main event loop, making event_notifier_set
154      * necessary.
155      *
156      * Bit 0 is reserved for GSource usage of the AioContext, and is 1
157      * between a call to aio_ctx_prepare and the next call to aio_ctx_check.
158      * Bits 1-31 simply count the number of active calls to aio_poll
159      * that are in the prepare or poll phase.
160      *
161      * The GSource and aio_poll must use a different mechanism because
162      * there is no certainty that a call to GSource's prepare callback
163      * (via g_main_context_prepare) is indeed followed by check and
164      * dispatch.  It's not clear whether this would be a bug, but let's
165      * play safe and allow it---it will just cause extra calls to
166      * event_notifier_set until the next call to dispatch.
167      *
168      * Instead, the aio_poll calls include both the prepare and the
169      * dispatch phase, hence a simple counter is enough for them.
170      */
171     uint32_t notify_me;
172 
173     /* A lock to protect between QEMUBH and AioHandler adders and deleter,
174      * and to ensure that no callbacks are removed while we're walking and
175      * dispatching them.
176      */
177     QemuLockCnt list_lock;
178 
179     /* Bottom Halves pending aio_bh_poll() processing */
180     BHList bh_list;
181 
182     /* Chained BH list slices for each nested aio_bh_poll() call */
183     QSIMPLEQ_HEAD(, BHListSlice) bh_slice_list;
184 
185     /* Used by aio_notify.
186      *
187      * "notified" is used to avoid expensive event_notifier_test_and_clear
188      * calls.  When it is clear, the EventNotifier is clear, or one thread
189      * is going to clear "notified" before processing more events.  False
190      * positives are possible, i.e. "notified" could be set even though the
191      * EventNotifier is clear.
192      *
193      * Note that event_notifier_set *cannot* be optimized the same way.  For
194      * more information on the problem that would result, see "#ifdef BUG2"
195      * in the docs/aio_notify_accept.promela formal model.
196      */
197     bool notified;
198     EventNotifier notifier;
199 
200     QSLIST_HEAD(, Coroutine) scheduled_coroutines;
201     QEMUBH *co_schedule_bh;
202 
203     int thread_pool_min;
204     int thread_pool_max;
205     /* Thread pool for performing work and receiving completion callbacks.
206      * Has its own locking.
207      */
208     struct ThreadPool *thread_pool;
209 
210 #ifdef CONFIG_LINUX_AIO
211     struct LinuxAioState *linux_aio;
212 #endif
213 #ifdef CONFIG_LINUX_IO_URING
214     struct LuringState *linux_io_uring;
215 
216     /* State for file descriptor monitoring using Linux io_uring */
217     struct io_uring fdmon_io_uring;
218     AioHandlerSList submit_list;
219 #endif
220 
221     /* TimerLists for calling timers - one per clock type.  Has its own
222      * locking.
223      */
224     QEMUTimerListGroup tlg;
225 
226     int external_disable_cnt;
227 
228     /* Number of AioHandlers without .io_poll() */
229     int poll_disable_cnt;
230 
231     /* Polling mode parameters */
232     int64_t poll_ns;        /* current polling time in nanoseconds */
233     int64_t poll_max_ns;    /* maximum polling time in nanoseconds */
234     int64_t poll_grow;      /* polling time growth factor */
235     int64_t poll_shrink;    /* polling time shrink factor */
236 
237     /* AIO engine parameters */
238     int64_t aio_max_batch;  /* maximum number of requests in a batch */
239 
240     /*
241      * List of handlers participating in userspace polling.  Protected by
242      * ctx->list_lock.  Iterated and modified mostly by the event loop thread
243      * from aio_poll() with ctx->list_lock incremented.  aio_set_fd_handler()
244      * only touches the list to delete nodes if ctx->list_lock's count is zero.
245      */
246     AioHandlerList poll_aio_handlers;
247 
248     /* Are we in polling mode or monitoring file descriptors? */
249     bool poll_started;
250 
251     /* epoll(7) state used when built with CONFIG_EPOLL */
252     int epollfd;
253 
254     const FDMonOps *fdmon_ops;
255 };
256 
257 /**
258  * aio_context_new: Allocate a new AioContext.
259  *
260  * AioContext provide a mini event-loop that can be waited on synchronously.
261  * They also provide bottom halves, a service to execute a piece of code
262  * as soon as possible.
263  */
264 AioContext *aio_context_new(Error **errp);
265 
266 /**
267  * aio_context_ref:
268  * @ctx: The AioContext to operate on.
269  *
270  * Add a reference to an AioContext.
271  */
272 void aio_context_ref(AioContext *ctx);
273 
274 /**
275  * aio_context_unref:
276  * @ctx: The AioContext to operate on.
277  *
278  * Drop a reference to an AioContext.
279  */
280 void aio_context_unref(AioContext *ctx);
281 
282 /* Take ownership of the AioContext.  If the AioContext will be shared between
283  * threads, and a thread does not want to be interrupted, it will have to
284  * take ownership around calls to aio_poll().  Otherwise, aio_poll()
285  * automatically takes care of calling aio_context_acquire and
286  * aio_context_release.
287  *
288  * Note that this is separate from bdrv_drained_begin/bdrv_drained_end.  A
289  * thread still has to call those to avoid being interrupted by the guest.
290  *
291  * Bottom halves, timers and callbacks can be created or removed without
292  * acquiring the AioContext.
293  */
294 void aio_context_acquire(AioContext *ctx);
295 
296 /* Relinquish ownership of the AioContext. */
297 void aio_context_release(AioContext *ctx);
298 
299 /**
300  * aio_bh_schedule_oneshot_full: Allocate a new bottom half structure that will
301  * run only once and as soon as possible.
302  *
303  * @name: A human-readable identifier for debugging purposes.
304  */
305 void aio_bh_schedule_oneshot_full(AioContext *ctx, QEMUBHFunc *cb, void *opaque,
306                                   const char *name);
307 
308 /**
309  * aio_bh_schedule_oneshot: Allocate a new bottom half structure that will run
310  * only once and as soon as possible.
311  *
312  * A convenience wrapper for aio_bh_schedule_oneshot_full() that uses cb as the
313  * name string.
314  */
315 #define aio_bh_schedule_oneshot(ctx, cb, opaque) \
316     aio_bh_schedule_oneshot_full((ctx), (cb), (opaque), (stringify(cb)))
317 
318 /**
319  * aio_bh_new_full: Allocate a new bottom half structure.
320  *
321  * Bottom halves are lightweight callbacks whose invocation is guaranteed
322  * to be wait-free, thread-safe and signal-safe.  The #QEMUBH structure
323  * is opaque and must be allocated prior to its use.
324  *
325  * @name: A human-readable identifier for debugging purposes.
326  */
327 QEMUBH *aio_bh_new_full(AioContext *ctx, QEMUBHFunc *cb, void *opaque,
328                         const char *name);
329 
330 /**
331  * aio_bh_new: Allocate a new bottom half structure
332  *
333  * A convenience wrapper for aio_bh_new_full() that uses the cb as the name
334  * string.
335  */
336 #define aio_bh_new(ctx, cb, opaque) \
337     aio_bh_new_full((ctx), (cb), (opaque), (stringify(cb)))
338 
339 /**
340  * aio_notify: Force processing of pending events.
341  *
342  * Similar to signaling a condition variable, aio_notify forces
343  * aio_poll to exit, so that the next call will re-examine pending events.
344  * The caller of aio_notify will usually call aio_poll again very soon,
345  * or go through another iteration of the GLib main loop.  Hence, aio_notify
346  * also has the side effect of recalculating the sets of file descriptors
347  * that the main loop waits for.
348  *
349  * Calling aio_notify is rarely necessary, because for example scheduling
350  * a bottom half calls it already.
351  */
352 void aio_notify(AioContext *ctx);
353 
354 /**
355  * aio_notify_accept: Acknowledge receiving an aio_notify.
356  *
357  * aio_notify() uses an EventNotifier in order to wake up a sleeping
358  * aio_poll() or g_main_context_iteration().  Calls to aio_notify() are
359  * usually rare, but the AioContext has to clear the EventNotifier on
360  * every aio_poll() or g_main_context_iteration() in order to avoid
361  * busy waiting.  This event_notifier_test_and_clear() cannot be done
362  * using the usual aio_context_set_event_notifier(), because it must
363  * be done before processing all events (file descriptors, bottom halves,
364  * timers).
365  *
366  * aio_notify_accept() is an optimized event_notifier_test_and_clear()
367  * that is specific to an AioContext's notifier; it is used internally
368  * to clear the EventNotifier only if aio_notify() had been called.
369  */
370 void aio_notify_accept(AioContext *ctx);
371 
372 /**
373  * aio_bh_call: Executes callback function of the specified BH.
374  */
375 void aio_bh_call(QEMUBH *bh);
376 
377 /**
378  * aio_bh_poll: Poll bottom halves for an AioContext.
379  *
380  * These are internal functions used by the QEMU main loop.
381  * And notice that multiple occurrences of aio_bh_poll cannot
382  * be called concurrently
383  */
384 int aio_bh_poll(AioContext *ctx);
385 
386 /**
387  * qemu_bh_schedule: Schedule a bottom half.
388  *
389  * Scheduling a bottom half interrupts the main loop and causes the
390  * execution of the callback that was passed to qemu_bh_new.
391  *
392  * Bottom halves that are scheduled from a bottom half handler are instantly
393  * invoked.  This can create an infinite loop if a bottom half handler
394  * schedules itself.
395  *
396  * @bh: The bottom half to be scheduled.
397  */
398 void qemu_bh_schedule(QEMUBH *bh);
399 
400 /**
401  * qemu_bh_cancel: Cancel execution of a bottom half.
402  *
403  * Canceling execution of a bottom half undoes the effect of calls to
404  * qemu_bh_schedule without freeing its resources yet.  While cancellation
405  * itself is also wait-free and thread-safe, it can of course race with the
406  * loop that executes bottom halves unless you are holding the iothread
407  * mutex.  This makes it mostly useless if you are not holding the mutex.
408  *
409  * @bh: The bottom half to be canceled.
410  */
411 void qemu_bh_cancel(QEMUBH *bh);
412 
413 /**
414  *qemu_bh_delete: Cancel execution of a bottom half and free its resources.
415  *
416  * Deleting a bottom half frees the memory that was allocated for it by
417  * qemu_bh_new.  It also implies canceling the bottom half if it was
418  * scheduled.
419  * This func is async. The bottom half will do the delete action at the finial
420  * end.
421  *
422  * @bh: The bottom half to be deleted.
423  */
424 void qemu_bh_delete(QEMUBH *bh);
425 
426 /* Return whether there are any pending callbacks from the GSource
427  * attached to the AioContext, before g_poll is invoked.
428  *
429  * This is used internally in the implementation of the GSource.
430  */
431 bool aio_prepare(AioContext *ctx);
432 
433 /* Return whether there are any pending callbacks from the GSource
434  * attached to the AioContext, after g_poll is invoked.
435  *
436  * This is used internally in the implementation of the GSource.
437  */
438 bool aio_pending(AioContext *ctx);
439 
440 /* Dispatch any pending callbacks from the GSource attached to the AioContext.
441  *
442  * This is used internally in the implementation of the GSource.
443  */
444 void aio_dispatch(AioContext *ctx);
445 
446 /* Progress in completing AIO work to occur.  This can issue new pending
447  * aio as a result of executing I/O completion or bh callbacks.
448  *
449  * Return whether any progress was made by executing AIO or bottom half
450  * handlers.  If @blocking == true, this should always be true except
451  * if someone called aio_notify.
452  *
453  * If there are no pending bottom halves, but there are pending AIO
454  * operations, it may not be possible to make any progress without
455  * blocking.  If @blocking is true, this function will wait until one
456  * or more AIO events have completed, to ensure something has moved
457  * before returning.
458  */
459 bool aio_poll(AioContext *ctx, bool blocking);
460 
461 /* Register a file descriptor and associated callbacks.  Behaves very similarly
462  * to qemu_set_fd_handler.  Unlike qemu_set_fd_handler, these callbacks will
463  * be invoked when using aio_poll().
464  *
465  * Code that invokes AIO completion functions should rely on this function
466  * instead of qemu_set_fd_handler[2].
467  */
468 void aio_set_fd_handler(AioContext *ctx,
469                         int fd,
470                         bool is_external,
471                         IOHandler *io_read,
472                         IOHandler *io_write,
473                         AioPollFn *io_poll,
474                         IOHandler *io_poll_ready,
475                         void *opaque);
476 
477 /* Register an event notifier and associated callbacks.  Behaves very similarly
478  * to event_notifier_set_handler.  Unlike event_notifier_set_handler, these callbacks
479  * will be invoked when using aio_poll().
480  *
481  * Code that invokes AIO completion functions should rely on this function
482  * instead of event_notifier_set_handler.
483  */
484 void aio_set_event_notifier(AioContext *ctx,
485                             EventNotifier *notifier,
486                             bool is_external,
487                             EventNotifierHandler *io_read,
488                             AioPollFn *io_poll,
489                             EventNotifierHandler *io_poll_ready);
490 
491 /* Set polling begin/end callbacks for an event notifier that has already been
492  * registered with aio_set_event_notifier.  Do nothing if the event notifier is
493  * not registered.
494  */
495 void aio_set_event_notifier_poll(AioContext *ctx,
496                                  EventNotifier *notifier,
497                                  EventNotifierHandler *io_poll_begin,
498                                  EventNotifierHandler *io_poll_end);
499 
500 /* Return a GSource that lets the main loop poll the file descriptors attached
501  * to this AioContext.
502  */
503 GSource *aio_get_g_source(AioContext *ctx);
504 
505 /* Return the ThreadPool bound to this AioContext */
506 struct ThreadPool *aio_get_thread_pool(AioContext *ctx);
507 
508 /* Setup the LinuxAioState bound to this AioContext */
509 struct LinuxAioState *aio_setup_linux_aio(AioContext *ctx, Error **errp);
510 
511 /* Return the LinuxAioState bound to this AioContext */
512 struct LinuxAioState *aio_get_linux_aio(AioContext *ctx);
513 
514 /* Setup the LuringState bound to this AioContext */
515 struct LuringState *aio_setup_linux_io_uring(AioContext *ctx, Error **errp);
516 
517 /* Return the LuringState bound to this AioContext */
518 struct LuringState *aio_get_linux_io_uring(AioContext *ctx);
519 /**
520  * aio_timer_new_with_attrs:
521  * @ctx: the aio context
522  * @type: the clock type
523  * @scale: the scale
524  * @attributes: 0, or one to multiple OR'ed QEMU_TIMER_ATTR_<id> values
525  *              to assign
526  * @cb: the callback to call on timer expiry
527  * @opaque: the opaque pointer to pass to the callback
528  *
529  * Allocate a new timer (with attributes) attached to the context @ctx.
530  * The function is responsible for memory allocation.
531  *
532  * The preferred interface is aio_timer_init or aio_timer_init_with_attrs.
533  * Use that unless you really need dynamic memory allocation.
534  *
535  * Returns: a pointer to the new timer
536  */
537 static inline QEMUTimer *aio_timer_new_with_attrs(AioContext *ctx,
538                                                   QEMUClockType type,
539                                                   int scale, int attributes,
540                                                   QEMUTimerCB *cb, void *opaque)
541 {
542     return timer_new_full(&ctx->tlg, type, scale, attributes, cb, opaque);
543 }
544 
545 /**
546  * aio_timer_new:
547  * @ctx: the aio context
548  * @type: the clock type
549  * @scale: the scale
550  * @cb: the callback to call on timer expiry
551  * @opaque: the opaque pointer to pass to the callback
552  *
553  * Allocate a new timer attached to the context @ctx.
554  * See aio_timer_new_with_attrs for details.
555  *
556  * Returns: a pointer to the new timer
557  */
558 static inline QEMUTimer *aio_timer_new(AioContext *ctx, QEMUClockType type,
559                                        int scale,
560                                        QEMUTimerCB *cb, void *opaque)
561 {
562     return timer_new_full(&ctx->tlg, type, scale, 0, cb, opaque);
563 }
564 
565 /**
566  * aio_timer_init_with_attrs:
567  * @ctx: the aio context
568  * @ts: the timer
569  * @type: the clock type
570  * @scale: the scale
571  * @attributes: 0, or one to multiple OR'ed QEMU_TIMER_ATTR_<id> values
572  *              to assign
573  * @cb: the callback to call on timer expiry
574  * @opaque: the opaque pointer to pass to the callback
575  *
576  * Initialise a new timer (with attributes) attached to the context @ctx.
577  * The caller is responsible for memory allocation.
578  */
579 static inline void aio_timer_init_with_attrs(AioContext *ctx,
580                                              QEMUTimer *ts, QEMUClockType type,
581                                              int scale, int attributes,
582                                              QEMUTimerCB *cb, void *opaque)
583 {
584     timer_init_full(ts, &ctx->tlg, type, scale, attributes, cb, opaque);
585 }
586 
587 /**
588  * aio_timer_init:
589  * @ctx: the aio context
590  * @ts: the timer
591  * @type: the clock type
592  * @scale: the scale
593  * @cb: the callback to call on timer expiry
594  * @opaque: the opaque pointer to pass to the callback
595  *
596  * Initialise a new timer attached to the context @ctx.
597  * See aio_timer_init_with_attrs for details.
598  */
599 static inline void aio_timer_init(AioContext *ctx,
600                                   QEMUTimer *ts, QEMUClockType type,
601                                   int scale,
602                                   QEMUTimerCB *cb, void *opaque)
603 {
604     timer_init_full(ts, &ctx->tlg, type, scale, 0, cb, opaque);
605 }
606 
607 /**
608  * aio_compute_timeout:
609  * @ctx: the aio context
610  *
611  * Compute the timeout that a blocking aio_poll should use.
612  */
613 int64_t aio_compute_timeout(AioContext *ctx);
614 
615 /**
616  * aio_disable_external:
617  * @ctx: the aio context
618  *
619  * Disable the further processing of external clients.
620  */
621 static inline void aio_disable_external(AioContext *ctx)
622 {
623     qatomic_inc(&ctx->external_disable_cnt);
624 }
625 
626 /**
627  * aio_enable_external:
628  * @ctx: the aio context
629  *
630  * Enable the processing of external clients.
631  */
632 static inline void aio_enable_external(AioContext *ctx)
633 {
634     int old;
635 
636     old = qatomic_fetch_dec(&ctx->external_disable_cnt);
637     assert(old > 0);
638     if (old == 1) {
639         /* Kick event loop so it re-arms file descriptors */
640         aio_notify(ctx);
641     }
642 }
643 
644 /**
645  * aio_external_disabled:
646  * @ctx: the aio context
647  *
648  * Return true if the external clients are disabled.
649  */
650 static inline bool aio_external_disabled(AioContext *ctx)
651 {
652     return qatomic_read(&ctx->external_disable_cnt);
653 }
654 
655 /**
656  * aio_node_check:
657  * @ctx: the aio context
658  * @is_external: Whether or not the checked node is an external event source.
659  *
660  * Check if the node's is_external flag is okay to be polled by the ctx at this
661  * moment. True means green light.
662  */
663 static inline bool aio_node_check(AioContext *ctx, bool is_external)
664 {
665     return !is_external || !qatomic_read(&ctx->external_disable_cnt);
666 }
667 
668 /**
669  * aio_co_schedule:
670  * @ctx: the aio context
671  * @co: the coroutine
672  *
673  * Start a coroutine on a remote AioContext.
674  *
675  * The coroutine must not be entered by anyone else while aio_co_schedule()
676  * is active.  In addition the coroutine must have yielded unless ctx
677  * is the context in which the coroutine is running (i.e. the value of
678  * qemu_get_current_aio_context() from the coroutine itself).
679  */
680 void aio_co_schedule(AioContext *ctx, Coroutine *co);
681 
682 /**
683  * aio_co_reschedule_self:
684  * @new_ctx: the new context
685  *
686  * Move the currently running coroutine to new_ctx. If the coroutine is already
687  * running in new_ctx, do nothing.
688  */
689 void coroutine_fn aio_co_reschedule_self(AioContext *new_ctx);
690 
691 /**
692  * aio_co_wake:
693  * @co: the coroutine
694  *
695  * Restart a coroutine on the AioContext where it was running last, thus
696  * preventing coroutines from jumping from one context to another when they
697  * go to sleep.
698  *
699  * aio_co_wake may be executed either in coroutine or non-coroutine
700  * context.  The coroutine must not be entered by anyone else while
701  * aio_co_wake() is active.
702  */
703 void aio_co_wake(Coroutine *co);
704 
705 /**
706  * aio_co_enter:
707  * @ctx: the context to run the coroutine
708  * @co: the coroutine to run
709  *
710  * Enter a coroutine in the specified AioContext.
711  */
712 void aio_co_enter(AioContext *ctx, Coroutine *co);
713 
714 /**
715  * Return the AioContext whose event loop runs in the current thread.
716  *
717  * If called from an IOThread this will be the IOThread's AioContext.  If
718  * called from the main thread or with the "big QEMU lock" taken it
719  * will be the main loop AioContext.
720  */
721 AioContext *qemu_get_current_aio_context(void);
722 
723 void qemu_set_current_aio_context(AioContext *ctx);
724 
725 /**
726  * aio_context_setup:
727  * @ctx: the aio context
728  *
729  * Initialize the aio context.
730  */
731 void aio_context_setup(AioContext *ctx);
732 
733 /**
734  * aio_context_destroy:
735  * @ctx: the aio context
736  *
737  * Destroy the aio context.
738  */
739 void aio_context_destroy(AioContext *ctx);
740 
741 /* Used internally, do not call outside AioContext code */
742 void aio_context_use_g_source(AioContext *ctx);
743 
744 /**
745  * aio_context_set_poll_params:
746  * @ctx: the aio context
747  * @max_ns: how long to busy poll for, in nanoseconds
748  * @grow: polling time growth factor
749  * @shrink: polling time shrink factor
750  *
751  * Poll mode can be disabled by setting poll_max_ns to 0.
752  */
753 void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns,
754                                  int64_t grow, int64_t shrink,
755                                  Error **errp);
756 
757 /**
758  * aio_context_set_aio_params:
759  * @ctx: the aio context
760  * @max_batch: maximum number of requests in a batch, 0 means that the
761  *             engine will use its default
762  */
763 void aio_context_set_aio_params(AioContext *ctx, int64_t max_batch,
764                                 Error **errp);
765 
766 /**
767  * aio_context_set_thread_pool_params:
768  * @ctx: the aio context
769  * @min: min number of threads to have readily available in the thread pool
770  * @min: max number of threads the thread pool can contain
771  */
772 void aio_context_set_thread_pool_params(AioContext *ctx, int64_t min,
773                                         int64_t max, Error **errp);
774 #endif
775