xref: /openbmc/qemu/include/block/aio.h (revision b14df228)
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.h"
21 #include "qemu/queue.h"
22 #include "qemu/event_notifier.h"
23 #include "qemu/thread.h"
24 #include "qemu/timer.h"
25 
26 typedef struct BlockAIOCB BlockAIOCB;
27 typedef void BlockCompletionFunc(void *opaque, int ret);
28 
29 typedef struct AIOCBInfo {
30     void (*cancel_async)(BlockAIOCB *acb);
31     AioContext *(*get_aio_context)(BlockAIOCB *acb);
32     size_t aiocb_size;
33 } AIOCBInfo;
34 
35 struct BlockAIOCB {
36     const AIOCBInfo *aiocb_info;
37     BlockDriverState *bs;
38     BlockCompletionFunc *cb;
39     void *opaque;
40     int refcnt;
41 };
42 
43 void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
44                    BlockCompletionFunc *cb, void *opaque);
45 void qemu_aio_unref(void *p);
46 void qemu_aio_ref(void *p);
47 
48 typedef struct AioHandler AioHandler;
49 typedef QLIST_HEAD(, AioHandler) AioHandlerList;
50 typedef void QEMUBHFunc(void *opaque);
51 typedef bool AioPollFn(void *opaque);
52 typedef void IOHandler(void *opaque);
53 
54 struct Coroutine;
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     /* The list of registered AIO handlers.  Protected by ctx->list_lock. */
131     AioHandlerList aio_handlers;
132 
133     /* The list of AIO handlers to be deleted.  Protected by ctx->list_lock. */
134     AioHandlerList deleted_aio_handlers;
135 
136     /* Used to avoid unnecessary event_notifier_set calls in aio_notify;
137      * only written from the AioContext home thread, or under the BQL in
138      * the case of the main AioContext.  However, it is read from any
139      * thread so it is still accessed with atomic primitives.
140      *
141      * If this field is 0, everything (file descriptors, bottom halves,
142      * timers) will be re-evaluated before the next blocking poll() or
143      * io_uring wait; therefore, the event_notifier_set call can be
144      * skipped.  If it is non-zero, you may need to wake up a concurrent
145      * aio_poll or the glib main event loop, making event_notifier_set
146      * necessary.
147      *
148      * Bit 0 is reserved for GSource usage of the AioContext, and is 1
149      * between a call to aio_ctx_prepare and the next call to aio_ctx_check.
150      * Bits 1-31 simply count the number of active calls to aio_poll
151      * that are in the prepare or poll phase.
152      *
153      * The GSource and aio_poll must use a different mechanism because
154      * there is no certainty that a call to GSource's prepare callback
155      * (via g_main_context_prepare) is indeed followed by check and
156      * dispatch.  It's not clear whether this would be a bug, but let's
157      * play safe and allow it---it will just cause extra calls to
158      * event_notifier_set until the next call to dispatch.
159      *
160      * Instead, the aio_poll calls include both the prepare and the
161      * dispatch phase, hence a simple counter is enough for them.
162      */
163     uint32_t notify_me;
164 
165     /* A lock to protect between QEMUBH and AioHandler adders and deleter,
166      * and to ensure that no callbacks are removed while we're walking and
167      * dispatching them.
168      */
169     QemuLockCnt list_lock;
170 
171     /* Bottom Halves pending aio_bh_poll() processing */
172     BHList bh_list;
173 
174     /* Chained BH list slices for each nested aio_bh_poll() call */
175     QSIMPLEQ_HEAD(, BHListSlice) bh_slice_list;
176 
177     /* Used by aio_notify.
178      *
179      * "notified" is used to avoid expensive event_notifier_test_and_clear
180      * calls.  When it is clear, the EventNotifier is clear, or one thread
181      * is going to clear "notified" before processing more events.  False
182      * positives are possible, i.e. "notified" could be set even though the
183      * EventNotifier is clear.
184      *
185      * Note that event_notifier_set *cannot* be optimized the same way.  For
186      * more information on the problem that would result, see "#ifdef BUG2"
187      * in the docs/aio_notify_accept.promela formal model.
188      */
189     bool notified;
190     EventNotifier notifier;
191 
192     QSLIST_HEAD(, Coroutine) scheduled_coroutines;
193     QEMUBH *co_schedule_bh;
194 
195     int thread_pool_min;
196     int thread_pool_max;
197     /* Thread pool for performing work and receiving completion callbacks.
198      * Has its own locking.
199      */
200     struct ThreadPool *thread_pool;
201 
202 #ifdef CONFIG_LINUX_AIO
203     /*
204      * State for native Linux AIO.  Uses aio_context_acquire/release for
205      * locking.
206      */
207     struct LinuxAioState *linux_aio;
208 #endif
209 #ifdef CONFIG_LINUX_IO_URING
210     /*
211      * State for Linux io_uring.  Uses aio_context_acquire/release for
212      * locking.
213      */
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 /* Set polling begin/end callbacks for a file descriptor that has already been
478  * registered with aio_set_fd_handler.  Do nothing if the file descriptor is
479  * not registered.
480  */
481 void aio_set_fd_poll(AioContext *ctx, int fd,
482                      IOHandler *io_poll_begin,
483                      IOHandler *io_poll_end);
484 
485 /* Register an event notifier and associated callbacks.  Behaves very similarly
486  * to event_notifier_set_handler.  Unlike event_notifier_set_handler, these callbacks
487  * will be invoked when using aio_poll().
488  *
489  * Code that invokes AIO completion functions should rely on this function
490  * instead of event_notifier_set_handler.
491  */
492 void aio_set_event_notifier(AioContext *ctx,
493                             EventNotifier *notifier,
494                             bool is_external,
495                             EventNotifierHandler *io_read,
496                             AioPollFn *io_poll,
497                             EventNotifierHandler *io_poll_ready);
498 
499 /* Set polling begin/end callbacks for an event notifier that has already been
500  * registered with aio_set_event_notifier.  Do nothing if the event notifier is
501  * not registered.
502  */
503 void aio_set_event_notifier_poll(AioContext *ctx,
504                                  EventNotifier *notifier,
505                                  EventNotifierHandler *io_poll_begin,
506                                  EventNotifierHandler *io_poll_end);
507 
508 /* Return a GSource that lets the main loop poll the file descriptors attached
509  * to this AioContext.
510  */
511 GSource *aio_get_g_source(AioContext *ctx);
512 
513 /* Return the ThreadPool bound to this AioContext */
514 struct ThreadPool *aio_get_thread_pool(AioContext *ctx);
515 
516 /* Setup the LinuxAioState bound to this AioContext */
517 struct LinuxAioState *aio_setup_linux_aio(AioContext *ctx, Error **errp);
518 
519 /* Return the LinuxAioState bound to this AioContext */
520 struct LinuxAioState *aio_get_linux_aio(AioContext *ctx);
521 
522 /* Setup the LuringState bound to this AioContext */
523 struct LuringState *aio_setup_linux_io_uring(AioContext *ctx, Error **errp);
524 
525 /* Return the LuringState bound to this AioContext */
526 struct LuringState *aio_get_linux_io_uring(AioContext *ctx);
527 /**
528  * aio_timer_new_with_attrs:
529  * @ctx: the aio context
530  * @type: the clock type
531  * @scale: the scale
532  * @attributes: 0, or one to multiple OR'ed QEMU_TIMER_ATTR_<id> values
533  *              to assign
534  * @cb: the callback to call on timer expiry
535  * @opaque: the opaque pointer to pass to the callback
536  *
537  * Allocate a new timer (with attributes) attached to the context @ctx.
538  * The function is responsible for memory allocation.
539  *
540  * The preferred interface is aio_timer_init or aio_timer_init_with_attrs.
541  * Use that unless you really need dynamic memory allocation.
542  *
543  * Returns: a pointer to the new timer
544  */
545 static inline QEMUTimer *aio_timer_new_with_attrs(AioContext *ctx,
546                                                   QEMUClockType type,
547                                                   int scale, int attributes,
548                                                   QEMUTimerCB *cb, void *opaque)
549 {
550     return timer_new_full(&ctx->tlg, type, scale, attributes, cb, opaque);
551 }
552 
553 /**
554  * aio_timer_new:
555  * @ctx: the aio context
556  * @type: the clock type
557  * @scale: the scale
558  * @cb: the callback to call on timer expiry
559  * @opaque: the opaque pointer to pass to the callback
560  *
561  * Allocate a new timer attached to the context @ctx.
562  * See aio_timer_new_with_attrs for details.
563  *
564  * Returns: a pointer to the new timer
565  */
566 static inline QEMUTimer *aio_timer_new(AioContext *ctx, QEMUClockType type,
567                                        int scale,
568                                        QEMUTimerCB *cb, void *opaque)
569 {
570     return timer_new_full(&ctx->tlg, type, scale, 0, cb, opaque);
571 }
572 
573 /**
574  * aio_timer_init_with_attrs:
575  * @ctx: the aio context
576  * @ts: the timer
577  * @type: the clock type
578  * @scale: the scale
579  * @attributes: 0, or one to multiple OR'ed QEMU_TIMER_ATTR_<id> values
580  *              to assign
581  * @cb: the callback to call on timer expiry
582  * @opaque: the opaque pointer to pass to the callback
583  *
584  * Initialise a new timer (with attributes) attached to the context @ctx.
585  * The caller is responsible for memory allocation.
586  */
587 static inline void aio_timer_init_with_attrs(AioContext *ctx,
588                                              QEMUTimer *ts, QEMUClockType type,
589                                              int scale, int attributes,
590                                              QEMUTimerCB *cb, void *opaque)
591 {
592     timer_init_full(ts, &ctx->tlg, type, scale, attributes, cb, opaque);
593 }
594 
595 /**
596  * aio_timer_init:
597  * @ctx: the aio context
598  * @ts: the timer
599  * @type: the clock type
600  * @scale: the scale
601  * @cb: the callback to call on timer expiry
602  * @opaque: the opaque pointer to pass to the callback
603  *
604  * Initialise a new timer attached to the context @ctx.
605  * See aio_timer_init_with_attrs for details.
606  */
607 static inline void aio_timer_init(AioContext *ctx,
608                                   QEMUTimer *ts, QEMUClockType type,
609                                   int scale,
610                                   QEMUTimerCB *cb, void *opaque)
611 {
612     timer_init_full(ts, &ctx->tlg, type, scale, 0, cb, opaque);
613 }
614 
615 /**
616  * aio_compute_timeout:
617  * @ctx: the aio context
618  *
619  * Compute the timeout that a blocking aio_poll should use.
620  */
621 int64_t aio_compute_timeout(AioContext *ctx);
622 
623 /**
624  * aio_disable_external:
625  * @ctx: the aio context
626  *
627  * Disable the further processing of external clients.
628  */
629 static inline void aio_disable_external(AioContext *ctx)
630 {
631     qatomic_inc(&ctx->external_disable_cnt);
632 }
633 
634 /**
635  * aio_enable_external:
636  * @ctx: the aio context
637  *
638  * Enable the processing of external clients.
639  */
640 static inline void aio_enable_external(AioContext *ctx)
641 {
642     int old;
643 
644     old = qatomic_fetch_dec(&ctx->external_disable_cnt);
645     assert(old > 0);
646     if (old == 1) {
647         /* Kick event loop so it re-arms file descriptors */
648         aio_notify(ctx);
649     }
650 }
651 
652 /**
653  * aio_external_disabled:
654  * @ctx: the aio context
655  *
656  * Return true if the external clients are disabled.
657  */
658 static inline bool aio_external_disabled(AioContext *ctx)
659 {
660     return qatomic_read(&ctx->external_disable_cnt);
661 }
662 
663 /**
664  * aio_node_check:
665  * @ctx: the aio context
666  * @is_external: Whether or not the checked node is an external event source.
667  *
668  * Check if the node's is_external flag is okay to be polled by the ctx at this
669  * moment. True means green light.
670  */
671 static inline bool aio_node_check(AioContext *ctx, bool is_external)
672 {
673     return !is_external || !qatomic_read(&ctx->external_disable_cnt);
674 }
675 
676 /**
677  * aio_co_schedule:
678  * @ctx: the aio context
679  * @co: the coroutine
680  *
681  * Start a coroutine on a remote AioContext.
682  *
683  * The coroutine must not be entered by anyone else while aio_co_schedule()
684  * is active.  In addition the coroutine must have yielded unless ctx
685  * is the context in which the coroutine is running (i.e. the value of
686  * qemu_get_current_aio_context() from the coroutine itself).
687  */
688 void aio_co_schedule(AioContext *ctx, struct Coroutine *co);
689 
690 /**
691  * aio_co_reschedule_self:
692  * @new_ctx: the new context
693  *
694  * Move the currently running coroutine to new_ctx. If the coroutine is already
695  * running in new_ctx, do nothing.
696  */
697 void coroutine_fn aio_co_reschedule_self(AioContext *new_ctx);
698 
699 /**
700  * aio_co_wake:
701  * @co: the coroutine
702  *
703  * Restart a coroutine on the AioContext where it was running last, thus
704  * preventing coroutines from jumping from one context to another when they
705  * go to sleep.
706  *
707  * aio_co_wake may be executed either in coroutine or non-coroutine
708  * context.  The coroutine must not be entered by anyone else while
709  * aio_co_wake() is active.
710  */
711 void aio_co_wake(struct Coroutine *co);
712 
713 /**
714  * aio_co_enter:
715  * @ctx: the context to run the coroutine
716  * @co: the coroutine to run
717  *
718  * Enter a coroutine in the specified AioContext.
719  */
720 void aio_co_enter(AioContext *ctx, struct Coroutine *co);
721 
722 /**
723  * Return the AioContext whose event loop runs in the current thread.
724  *
725  * If called from an IOThread this will be the IOThread's AioContext.  If
726  * called from the main thread or with the "big QEMU lock" taken it
727  * will be the main loop AioContext.
728  */
729 AioContext *qemu_get_current_aio_context(void);
730 
731 void qemu_set_current_aio_context(AioContext *ctx);
732 
733 /**
734  * aio_context_setup:
735  * @ctx: the aio context
736  *
737  * Initialize the aio context.
738  */
739 void aio_context_setup(AioContext *ctx);
740 
741 /**
742  * aio_context_destroy:
743  * @ctx: the aio context
744  *
745  * Destroy the aio context.
746  */
747 void aio_context_destroy(AioContext *ctx);
748 
749 /* Used internally, do not call outside AioContext code */
750 void aio_context_use_g_source(AioContext *ctx);
751 
752 /**
753  * aio_context_set_poll_params:
754  * @ctx: the aio context
755  * @max_ns: how long to busy poll for, in nanoseconds
756  * @grow: polling time growth factor
757  * @shrink: polling time shrink factor
758  *
759  * Poll mode can be disabled by setting poll_max_ns to 0.
760  */
761 void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns,
762                                  int64_t grow, int64_t shrink,
763                                  Error **errp);
764 
765 /**
766  * aio_context_set_aio_params:
767  * @ctx: the aio context
768  * @max_batch: maximum number of requests in a batch, 0 means that the
769  *             engine will use its default
770  */
771 void aio_context_set_aio_params(AioContext *ctx, int64_t max_batch,
772                                 Error **errp);
773 
774 /**
775  * aio_context_set_thread_pool_params:
776  * @ctx: the aio context
777  * @min: min number of threads to have readily available in the thread pool
778  * @min: max number of threads the thread pool can contain
779  */
780 void aio_context_set_thread_pool_params(AioContext *ctx, int64_t min,
781                                         int64_t max, Error **errp);
782 #endif
783