xref: /openbmc/qemu/include/block/aio.h (revision 51e47cf8)
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 #include "hw/qdev-core.h"
27 
28 
29 typedef struct BlockAIOCB BlockAIOCB;
30 typedef void BlockCompletionFunc(void *opaque, int ret);
31 
32 typedef struct AIOCBInfo {
33     void (*cancel_async)(BlockAIOCB *acb);
34     AioContext *(*get_aio_context)(BlockAIOCB *acb);
35     size_t aiocb_size;
36 } AIOCBInfo;
37 
38 struct BlockAIOCB {
39     const AIOCBInfo *aiocb_info;
40     BlockDriverState *bs;
41     BlockCompletionFunc *cb;
42     void *opaque;
43     int refcnt;
44 };
45 
46 void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
47                    BlockCompletionFunc *cb, void *opaque);
48 void qemu_aio_unref(void *p);
49 void qemu_aio_ref(void *p);
50 
51 typedef struct AioHandler AioHandler;
52 typedef QLIST_HEAD(, AioHandler) AioHandlerList;
53 typedef void QEMUBHFunc(void *opaque);
54 typedef bool AioPollFn(void *opaque);
55 typedef void IOHandler(void *opaque);
56 
57 struct ThreadPool;
58 struct LinuxAioState;
59 struct LuringState;
60 
61 /* Is polling disabled? */
62 bool aio_poll_disabled(AioContext *ctx);
63 
64 /* Callbacks for file descriptor monitoring implementations */
65 typedef struct {
66     /*
67      * update:
68      * @ctx: the AioContext
69      * @old_node: the existing handler or NULL if this file descriptor is being
70      *            monitored for the first time
71      * @new_node: the new handler or NULL if this file descriptor is being
72      *            removed
73      *
74      * Add/remove/modify a monitored file descriptor.
75      *
76      * Called with ctx->list_lock acquired.
77      */
78     void (*update)(AioContext *ctx, AioHandler *old_node, AioHandler *new_node);
79 
80     /*
81      * wait:
82      * @ctx: the AioContext
83      * @ready_list: list for handlers that become ready
84      * @timeout: maximum duration to wait, in nanoseconds
85      *
86      * Wait for file descriptors to become ready and place them on ready_list.
87      *
88      * Called with ctx->list_lock incremented but not locked.
89      *
90      * Returns: number of ready file descriptors.
91      */
92     int (*wait)(AioContext *ctx, AioHandlerList *ready_list, int64_t timeout);
93 
94     /*
95      * need_wait:
96      * @ctx: the AioContext
97      *
98      * Tell aio_poll() when to stop userspace polling early because ->wait()
99      * has fds ready.
100      *
101      * File descriptor monitoring implementations that cannot poll fd readiness
102      * from userspace should use aio_poll_disabled() here.  This ensures that
103      * file descriptors are not starved by handlers that frequently make
104      * progress via userspace polling.
105      *
106      * Returns: true if ->wait() should be called, false otherwise.
107      */
108     bool (*need_wait)(AioContext *ctx);
109 } FDMonOps;
110 
111 /*
112  * Each aio_bh_poll() call carves off a slice of the BH list, so that newly
113  * scheduled BHs are not processed until the next aio_bh_poll() call.  All
114  * active aio_bh_poll() calls chain their slices together in a list, so that
115  * nested aio_bh_poll() calls process all scheduled bottom halves.
116  */
117 typedef QSLIST_HEAD(, QEMUBH) BHList;
118 typedef struct BHListSlice BHListSlice;
119 struct BHListSlice {
120     BHList bh_list;
121     QSIMPLEQ_ENTRY(BHListSlice) next;
122 };
123 
124 typedef QSLIST_HEAD(, AioHandler) AioHandlerSList;
125 
126 struct AioContext {
127     GSource source;
128 
129     /* Used by AioContext users to protect from multi-threaded access.  */
130     QemuRecMutex lock;
131 
132     /*
133      * Keep track of readers and writers of the block layer graph.
134      * This is essential to avoid performing additions and removal
135      * of nodes and edges from block graph while some
136      * other thread is traversing it.
137      */
138     BdrvGraphRWlock *bdrv_graph;
139 
140     /* The list of registered AIO handlers.  Protected by ctx->list_lock. */
141     AioHandlerList aio_handlers;
142 
143     /* The list of AIO handlers to be deleted.  Protected by ctx->list_lock. */
144     AioHandlerList deleted_aio_handlers;
145 
146     /* Used to avoid unnecessary event_notifier_set calls in aio_notify;
147      * only written from the AioContext home thread, or under the BQL in
148      * the case of the main AioContext.  However, it is read from any
149      * thread so it is still accessed with atomic primitives.
150      *
151      * If this field is 0, everything (file descriptors, bottom halves,
152      * timers) will be re-evaluated before the next blocking poll() or
153      * io_uring wait; therefore, the event_notifier_set call can be
154      * skipped.  If it is non-zero, you may need to wake up a concurrent
155      * aio_poll or the glib main event loop, making event_notifier_set
156      * necessary.
157      *
158      * Bit 0 is reserved for GSource usage of the AioContext, and is 1
159      * between a call to aio_ctx_prepare and the next call to aio_ctx_check.
160      * Bits 1-31 simply count the number of active calls to aio_poll
161      * that are in the prepare or poll phase.
162      *
163      * The GSource and aio_poll must use a different mechanism because
164      * there is no certainty that a call to GSource's prepare callback
165      * (via g_main_context_prepare) is indeed followed by check and
166      * dispatch.  It's not clear whether this would be a bug, but let's
167      * play safe and allow it---it will just cause extra calls to
168      * event_notifier_set until the next call to dispatch.
169      *
170      * Instead, the aio_poll calls include both the prepare and the
171      * dispatch phase, hence a simple counter is enough for them.
172      */
173     uint32_t notify_me;
174 
175     /* A lock to protect between QEMUBH and AioHandler adders and deleter,
176      * and to ensure that no callbacks are removed while we're walking and
177      * dispatching them.
178      */
179     QemuLockCnt list_lock;
180 
181     /* Bottom Halves pending aio_bh_poll() processing */
182     BHList bh_list;
183 
184     /* Chained BH list slices for each nested aio_bh_poll() call */
185     QSIMPLEQ_HEAD(, BHListSlice) bh_slice_list;
186 
187     /* Used by aio_notify.
188      *
189      * "notified" is used to avoid expensive event_notifier_test_and_clear
190      * calls.  When it is clear, the EventNotifier is clear, or one thread
191      * is going to clear "notified" before processing more events.  False
192      * positives are possible, i.e. "notified" could be set even though the
193      * EventNotifier is clear.
194      *
195      * Note that event_notifier_set *cannot* be optimized the same way.  For
196      * more information on the problem that would result, see "#ifdef BUG2"
197      * in the docs/aio_notify_accept.promela formal model.
198      */
199     bool notified;
200     EventNotifier notifier;
201 
202     QSLIST_HEAD(, Coroutine) scheduled_coroutines;
203     QEMUBH *co_schedule_bh;
204 
205     int thread_pool_min;
206     int thread_pool_max;
207     /* Thread pool for performing work and receiving completion callbacks.
208      * Has its own locking.
209      */
210     struct ThreadPool *thread_pool;
211 
212 #ifdef CONFIG_LINUX_AIO
213     struct LinuxAioState *linux_aio;
214 #endif
215 #ifdef CONFIG_LINUX_IO_URING
216     struct LuringState *linux_io_uring;
217 
218     /* State for file descriptor monitoring using Linux io_uring */
219     struct io_uring fdmon_io_uring;
220     AioHandlerSList submit_list;
221 #endif
222 
223     /* TimerLists for calling timers - one per clock type.  Has its own
224      * locking.
225      */
226     QEMUTimerListGroup tlg;
227 
228     int external_disable_cnt;
229 
230     /* Number of AioHandlers without .io_poll() */
231     int poll_disable_cnt;
232 
233     /* Polling mode parameters */
234     int64_t poll_ns;        /* current polling time in nanoseconds */
235     int64_t poll_max_ns;    /* maximum polling time in nanoseconds */
236     int64_t poll_grow;      /* polling time growth factor */
237     int64_t poll_shrink;    /* polling time shrink factor */
238 
239     /* AIO engine parameters */
240     int64_t aio_max_batch;  /* maximum number of requests in a batch */
241 
242     /*
243      * List of handlers participating in userspace polling.  Protected by
244      * ctx->list_lock.  Iterated and modified mostly by the event loop thread
245      * from aio_poll() with ctx->list_lock incremented.  aio_set_fd_handler()
246      * only touches the list to delete nodes if ctx->list_lock's count is zero.
247      */
248     AioHandlerList poll_aio_handlers;
249 
250     /* Are we in polling mode or monitoring file descriptors? */
251     bool poll_started;
252 
253     /* epoll(7) state used when built with CONFIG_EPOLL */
254     int epollfd;
255 
256     const FDMonOps *fdmon_ops;
257 };
258 
259 /**
260  * aio_context_new: Allocate a new AioContext.
261  *
262  * AioContext provide a mini event-loop that can be waited on synchronously.
263  * They also provide bottom halves, a service to execute a piece of code
264  * as soon as possible.
265  */
266 AioContext *aio_context_new(Error **errp);
267 
268 /**
269  * aio_context_ref:
270  * @ctx: The AioContext to operate on.
271  *
272  * Add a reference to an AioContext.
273  */
274 void aio_context_ref(AioContext *ctx);
275 
276 /**
277  * aio_context_unref:
278  * @ctx: The AioContext to operate on.
279  *
280  * Drop a reference to an AioContext.
281  */
282 void aio_context_unref(AioContext *ctx);
283 
284 /* Take ownership of the AioContext.  If the AioContext will be shared between
285  * threads, and a thread does not want to be interrupted, it will have to
286  * take ownership around calls to aio_poll().  Otherwise, aio_poll()
287  * automatically takes care of calling aio_context_acquire and
288  * aio_context_release.
289  *
290  * Note that this is separate from bdrv_drained_begin/bdrv_drained_end.  A
291  * thread still has to call those to avoid being interrupted by the guest.
292  *
293  * Bottom halves, timers and callbacks can be created or removed without
294  * acquiring the AioContext.
295  */
296 void aio_context_acquire(AioContext *ctx);
297 
298 /* Relinquish ownership of the AioContext. */
299 void aio_context_release(AioContext *ctx);
300 
301 /**
302  * aio_bh_schedule_oneshot_full: Allocate a new bottom half structure that will
303  * run only once and as soon as possible.
304  *
305  * @name: A human-readable identifier for debugging purposes.
306  */
307 void aio_bh_schedule_oneshot_full(AioContext *ctx, QEMUBHFunc *cb, void *opaque,
308                                   const char *name);
309 
310 /**
311  * aio_bh_schedule_oneshot: Allocate a new bottom half structure that will run
312  * only once and as soon as possible.
313  *
314  * A convenience wrapper for aio_bh_schedule_oneshot_full() that uses cb as the
315  * name string.
316  */
317 #define aio_bh_schedule_oneshot(ctx, cb, opaque) \
318     aio_bh_schedule_oneshot_full((ctx), (cb), (opaque), (stringify(cb)))
319 
320 /**
321  * aio_bh_new_full: Allocate a new bottom half structure.
322  *
323  * Bottom halves are lightweight callbacks whose invocation is guaranteed
324  * to be wait-free, thread-safe and signal-safe.  The #QEMUBH structure
325  * is opaque and must be allocated prior to its use.
326  *
327  * @name: A human-readable identifier for debugging purposes.
328  * @reentrancy_guard: A guard set when entering a cb to prevent
329  * device-reentrancy issues
330  */
331 QEMUBH *aio_bh_new_full(AioContext *ctx, QEMUBHFunc *cb, void *opaque,
332                         const char *name, MemReentrancyGuard *reentrancy_guard);
333 
334 /**
335  * aio_bh_new: Allocate a new bottom half structure
336  *
337  * A convenience wrapper for aio_bh_new_full() that uses the cb as the name
338  * string.
339  */
340 #define aio_bh_new(ctx, cb, opaque) \
341     aio_bh_new_full((ctx), (cb), (opaque), (stringify(cb)), NULL)
342 
343 /**
344  * aio_bh_new_guarded: Allocate a new bottom half structure with a
345  * reentrancy_guard
346  *
347  * A convenience wrapper for aio_bh_new_full() that uses the cb as the name
348  * string.
349  */
350 #define aio_bh_new_guarded(ctx, cb, opaque, guard) \
351     aio_bh_new_full((ctx), (cb), (opaque), (stringify(cb)), guard)
352 
353 /**
354  * aio_notify: Force processing of pending events.
355  *
356  * Similar to signaling a condition variable, aio_notify forces
357  * aio_poll to exit, so that the next call will re-examine pending events.
358  * The caller of aio_notify will usually call aio_poll again very soon,
359  * or go through another iteration of the GLib main loop.  Hence, aio_notify
360  * also has the side effect of recalculating the sets of file descriptors
361  * that the main loop waits for.
362  *
363  * Calling aio_notify is rarely necessary, because for example scheduling
364  * a bottom half calls it already.
365  */
366 void aio_notify(AioContext *ctx);
367 
368 /**
369  * aio_notify_accept: Acknowledge receiving an aio_notify.
370  *
371  * aio_notify() uses an EventNotifier in order to wake up a sleeping
372  * aio_poll() or g_main_context_iteration().  Calls to aio_notify() are
373  * usually rare, but the AioContext has to clear the EventNotifier on
374  * every aio_poll() or g_main_context_iteration() in order to avoid
375  * busy waiting.  This event_notifier_test_and_clear() cannot be done
376  * using the usual aio_context_set_event_notifier(), because it must
377  * be done before processing all events (file descriptors, bottom halves,
378  * timers).
379  *
380  * aio_notify_accept() is an optimized event_notifier_test_and_clear()
381  * that is specific to an AioContext's notifier; it is used internally
382  * to clear the EventNotifier only if aio_notify() had been called.
383  */
384 void aio_notify_accept(AioContext *ctx);
385 
386 /**
387  * aio_bh_call: Executes callback function of the specified BH.
388  */
389 void aio_bh_call(QEMUBH *bh);
390 
391 /**
392  * aio_bh_poll: Poll bottom halves for an AioContext.
393  *
394  * These are internal functions used by the QEMU main loop.
395  * And notice that multiple occurrences of aio_bh_poll cannot
396  * be called concurrently
397  */
398 int aio_bh_poll(AioContext *ctx);
399 
400 /**
401  * qemu_bh_schedule: Schedule a bottom half.
402  *
403  * Scheduling a bottom half interrupts the main loop and causes the
404  * execution of the callback that was passed to qemu_bh_new.
405  *
406  * Bottom halves that are scheduled from a bottom half handler are instantly
407  * invoked.  This can create an infinite loop if a bottom half handler
408  * schedules itself.
409  *
410  * @bh: The bottom half to be scheduled.
411  */
412 void qemu_bh_schedule(QEMUBH *bh);
413 
414 /**
415  * qemu_bh_cancel: Cancel execution of a bottom half.
416  *
417  * Canceling execution of a bottom half undoes the effect of calls to
418  * qemu_bh_schedule without freeing its resources yet.  While cancellation
419  * itself is also wait-free and thread-safe, it can of course race with the
420  * loop that executes bottom halves unless you are holding the iothread
421  * mutex.  This makes it mostly useless if you are not holding the mutex.
422  *
423  * @bh: The bottom half to be canceled.
424  */
425 void qemu_bh_cancel(QEMUBH *bh);
426 
427 /**
428  *qemu_bh_delete: Cancel execution of a bottom half and free its resources.
429  *
430  * Deleting a bottom half frees the memory that was allocated for it by
431  * qemu_bh_new.  It also implies canceling the bottom half if it was
432  * scheduled.
433  * This func is async. The bottom half will do the delete action at the finial
434  * end.
435  *
436  * @bh: The bottom half to be deleted.
437  */
438 void qemu_bh_delete(QEMUBH *bh);
439 
440 /* Return whether there are any pending callbacks from the GSource
441  * attached to the AioContext, before g_poll is invoked.
442  *
443  * This is used internally in the implementation of the GSource.
444  */
445 bool aio_prepare(AioContext *ctx);
446 
447 /* Return whether there are any pending callbacks from the GSource
448  * attached to the AioContext, after g_poll is invoked.
449  *
450  * This is used internally in the implementation of the GSource.
451  */
452 bool aio_pending(AioContext *ctx);
453 
454 /* Dispatch any pending callbacks from the GSource attached to the AioContext.
455  *
456  * This is used internally in the implementation of the GSource.
457  */
458 void aio_dispatch(AioContext *ctx);
459 
460 /* Progress in completing AIO work to occur.  This can issue new pending
461  * aio as a result of executing I/O completion or bh callbacks.
462  *
463  * Return whether any progress was made by executing AIO or bottom half
464  * handlers.  If @blocking == true, this should always be true except
465  * if someone called aio_notify.
466  *
467  * If there are no pending bottom halves, but there are pending AIO
468  * operations, it may not be possible to make any progress without
469  * blocking.  If @blocking is true, this function will wait until one
470  * or more AIO events have completed, to ensure something has moved
471  * before returning.
472  */
473 bool aio_poll(AioContext *ctx, bool blocking);
474 
475 /* Register a file descriptor and associated callbacks.  Behaves very similarly
476  * to qemu_set_fd_handler.  Unlike qemu_set_fd_handler, these callbacks will
477  * be invoked when using aio_poll().
478  *
479  * Code that invokes AIO completion functions should rely on this function
480  * instead of qemu_set_fd_handler[2].
481  */
482 void aio_set_fd_handler(AioContext *ctx,
483                         int fd,
484                         bool is_external,
485                         IOHandler *io_read,
486                         IOHandler *io_write,
487                         AioPollFn *io_poll,
488                         IOHandler *io_poll_ready,
489                         void *opaque);
490 
491 /* Register an event notifier and associated callbacks.  Behaves very similarly
492  * to event_notifier_set_handler.  Unlike event_notifier_set_handler, these callbacks
493  * will be invoked when using aio_poll().
494  *
495  * Code that invokes AIO completion functions should rely on this function
496  * instead of event_notifier_set_handler.
497  */
498 void aio_set_event_notifier(AioContext *ctx,
499                             EventNotifier *notifier,
500                             bool is_external,
501                             EventNotifierHandler *io_read,
502                             AioPollFn *io_poll,
503                             EventNotifierHandler *io_poll_ready);
504 
505 /* Set polling begin/end callbacks for an event notifier that has already been
506  * registered with aio_set_event_notifier.  Do nothing if the event notifier is
507  * not registered.
508  */
509 void aio_set_event_notifier_poll(AioContext *ctx,
510                                  EventNotifier *notifier,
511                                  EventNotifierHandler *io_poll_begin,
512                                  EventNotifierHandler *io_poll_end);
513 
514 /* Return a GSource that lets the main loop poll the file descriptors attached
515  * to this AioContext.
516  */
517 GSource *aio_get_g_source(AioContext *ctx);
518 
519 /* Return the ThreadPool bound to this AioContext */
520 struct ThreadPool *aio_get_thread_pool(AioContext *ctx);
521 
522 /* Setup the LinuxAioState bound to this AioContext */
523 struct LinuxAioState *aio_setup_linux_aio(AioContext *ctx, Error **errp);
524 
525 /* Return the LinuxAioState bound to this AioContext */
526 struct LinuxAioState *aio_get_linux_aio(AioContext *ctx);
527 
528 /* Setup the LuringState bound to this AioContext */
529 struct LuringState *aio_setup_linux_io_uring(AioContext *ctx, Error **errp);
530 
531 /* Return the LuringState bound to this AioContext */
532 struct LuringState *aio_get_linux_io_uring(AioContext *ctx);
533 /**
534  * aio_timer_new_with_attrs:
535  * @ctx: the aio context
536  * @type: the clock type
537  * @scale: the scale
538  * @attributes: 0, or one to multiple OR'ed QEMU_TIMER_ATTR_<id> values
539  *              to assign
540  * @cb: the callback to call on timer expiry
541  * @opaque: the opaque pointer to pass to the callback
542  *
543  * Allocate a new timer (with attributes) attached to the context @ctx.
544  * The function is responsible for memory allocation.
545  *
546  * The preferred interface is aio_timer_init or aio_timer_init_with_attrs.
547  * Use that unless you really need dynamic memory allocation.
548  *
549  * Returns: a pointer to the new timer
550  */
551 static inline QEMUTimer *aio_timer_new_with_attrs(AioContext *ctx,
552                                                   QEMUClockType type,
553                                                   int scale, int attributes,
554                                                   QEMUTimerCB *cb, void *opaque)
555 {
556     return timer_new_full(&ctx->tlg, type, scale, attributes, cb, opaque);
557 }
558 
559 /**
560  * aio_timer_new:
561  * @ctx: the aio context
562  * @type: the clock type
563  * @scale: the scale
564  * @cb: the callback to call on timer expiry
565  * @opaque: the opaque pointer to pass to the callback
566  *
567  * Allocate a new timer attached to the context @ctx.
568  * See aio_timer_new_with_attrs for details.
569  *
570  * Returns: a pointer to the new timer
571  */
572 static inline QEMUTimer *aio_timer_new(AioContext *ctx, QEMUClockType type,
573                                        int scale,
574                                        QEMUTimerCB *cb, void *opaque)
575 {
576     return timer_new_full(&ctx->tlg, type, scale, 0, cb, opaque);
577 }
578 
579 /**
580  * aio_timer_init_with_attrs:
581  * @ctx: the aio context
582  * @ts: the timer
583  * @type: the clock type
584  * @scale: the scale
585  * @attributes: 0, or one to multiple OR'ed QEMU_TIMER_ATTR_<id> values
586  *              to assign
587  * @cb: the callback to call on timer expiry
588  * @opaque: the opaque pointer to pass to the callback
589  *
590  * Initialise a new timer (with attributes) attached to the context @ctx.
591  * The caller is responsible for memory allocation.
592  */
593 static inline void aio_timer_init_with_attrs(AioContext *ctx,
594                                              QEMUTimer *ts, QEMUClockType type,
595                                              int scale, int attributes,
596                                              QEMUTimerCB *cb, void *opaque)
597 {
598     timer_init_full(ts, &ctx->tlg, type, scale, attributes, cb, opaque);
599 }
600 
601 /**
602  * aio_timer_init:
603  * @ctx: the aio context
604  * @ts: the timer
605  * @type: the clock type
606  * @scale: the scale
607  * @cb: the callback to call on timer expiry
608  * @opaque: the opaque pointer to pass to the callback
609  *
610  * Initialise a new timer attached to the context @ctx.
611  * See aio_timer_init_with_attrs for details.
612  */
613 static inline void aio_timer_init(AioContext *ctx,
614                                   QEMUTimer *ts, QEMUClockType type,
615                                   int scale,
616                                   QEMUTimerCB *cb, void *opaque)
617 {
618     timer_init_full(ts, &ctx->tlg, type, scale, 0, cb, opaque);
619 }
620 
621 /**
622  * aio_compute_timeout:
623  * @ctx: the aio context
624  *
625  * Compute the timeout that a blocking aio_poll should use.
626  */
627 int64_t aio_compute_timeout(AioContext *ctx);
628 
629 /**
630  * aio_disable_external:
631  * @ctx: the aio context
632  *
633  * Disable the further processing of external clients.
634  */
635 static inline void aio_disable_external(AioContext *ctx)
636 {
637     qatomic_inc(&ctx->external_disable_cnt);
638 }
639 
640 /**
641  * aio_enable_external:
642  * @ctx: the aio context
643  *
644  * Enable the processing of external clients.
645  */
646 static inline void aio_enable_external(AioContext *ctx)
647 {
648     int old;
649 
650     old = qatomic_fetch_dec(&ctx->external_disable_cnt);
651     assert(old > 0);
652     if (old == 1) {
653         /* Kick event loop so it re-arms file descriptors */
654         aio_notify(ctx);
655     }
656 }
657 
658 /**
659  * aio_external_disabled:
660  * @ctx: the aio context
661  *
662  * Return true if the external clients are disabled.
663  */
664 static inline bool aio_external_disabled(AioContext *ctx)
665 {
666     return qatomic_read(&ctx->external_disable_cnt);
667 }
668 
669 /**
670  * aio_node_check:
671  * @ctx: the aio context
672  * @is_external: Whether or not the checked node is an external event source.
673  *
674  * Check if the node's is_external flag is okay to be polled by the ctx at this
675  * moment. True means green light.
676  */
677 static inline bool aio_node_check(AioContext *ctx, bool is_external)
678 {
679     return !is_external || !qatomic_read(&ctx->external_disable_cnt);
680 }
681 
682 /**
683  * aio_co_schedule:
684  * @ctx: the aio context
685  * @co: the coroutine
686  *
687  * Start a coroutine on a remote AioContext.
688  *
689  * The coroutine must not be entered by anyone else while aio_co_schedule()
690  * is active.  In addition the coroutine must have yielded unless ctx
691  * is the context in which the coroutine is running (i.e. the value of
692  * qemu_get_current_aio_context() from the coroutine itself).
693  */
694 void aio_co_schedule(AioContext *ctx, Coroutine *co);
695 
696 /**
697  * aio_co_reschedule_self:
698  * @new_ctx: the new context
699  *
700  * Move the currently running coroutine to new_ctx. If the coroutine is already
701  * running in new_ctx, do nothing.
702  */
703 void coroutine_fn aio_co_reschedule_self(AioContext *new_ctx);
704 
705 /**
706  * aio_co_wake:
707  * @co: the coroutine
708  *
709  * Restart a coroutine on the AioContext where it was running last, thus
710  * preventing coroutines from jumping from one context to another when they
711  * go to sleep.
712  *
713  * aio_co_wake may be executed either in coroutine or non-coroutine
714  * context.  The coroutine must not be entered by anyone else while
715  * aio_co_wake() is active.
716  */
717 void aio_co_wake(Coroutine *co);
718 
719 /**
720  * aio_co_enter:
721  * @ctx: the context to run the coroutine
722  * @co: the coroutine to run
723  *
724  * Enter a coroutine in the specified AioContext.
725  */
726 void aio_co_enter(AioContext *ctx, Coroutine *co);
727 
728 /**
729  * Return the AioContext whose event loop runs in the current thread.
730  *
731  * If called from an IOThread this will be the IOThread's AioContext.  If
732  * called from the main thread or with the "big QEMU lock" taken it
733  * will be the main loop AioContext.
734  */
735 AioContext *qemu_get_current_aio_context(void);
736 
737 void qemu_set_current_aio_context(AioContext *ctx);
738 
739 /**
740  * aio_context_setup:
741  * @ctx: the aio context
742  *
743  * Initialize the aio context.
744  */
745 void aio_context_setup(AioContext *ctx);
746 
747 /**
748  * aio_context_destroy:
749  * @ctx: the aio context
750  *
751  * Destroy the aio context.
752  */
753 void aio_context_destroy(AioContext *ctx);
754 
755 /* Used internally, do not call outside AioContext code */
756 void aio_context_use_g_source(AioContext *ctx);
757 
758 /**
759  * aio_context_set_poll_params:
760  * @ctx: the aio context
761  * @max_ns: how long to busy poll for, in nanoseconds
762  * @grow: polling time growth factor
763  * @shrink: polling time shrink factor
764  *
765  * Poll mode can be disabled by setting poll_max_ns to 0.
766  */
767 void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns,
768                                  int64_t grow, int64_t shrink,
769                                  Error **errp);
770 
771 /**
772  * aio_context_set_aio_params:
773  * @ctx: the aio context
774  * @max_batch: maximum number of requests in a batch, 0 means that the
775  *             engine will use its default
776  */
777 void aio_context_set_aio_params(AioContext *ctx, int64_t max_batch,
778                                 Error **errp);
779 
780 /**
781  * aio_context_set_thread_pool_params:
782  * @ctx: the aio context
783  * @min: min number of threads to have readily available in the thread pool
784  * @min: max number of threads the thread pool can contain
785  */
786 void aio_context_set_thread_pool_params(AioContext *ctx, int64_t min,
787                                         int64_t max, Error **errp);
788 #endif
789