include/block/aio.h

   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     /*
 212      * State for native Linux AIO.  Uses aio_context_acquire/release for
 213      * locking.
 214      */
 215     struct LinuxAioState *linux_aio;
 216 #endif
 217 #ifdef CONFIG_LINUX_IO_URING
 218     /*
 219      * State for Linux io_uring.  Uses aio_context_acquire/release for
 220      * locking.
 221      */
 222     struct LuringState *linux_io_uring;
 223
 224     /* State for file descriptor monitoring using Linux io_uring */
 225     struct io_uring fdmon_io_uring;
 226     AioHandlerSList submit_list;
 227 #endif
 228
 229     /* TimerLists for calling timers - one per clock type.  Has its own
 230      * locking.
 231      */
 232     QEMUTimerListGroup tlg;
 233
 234     int external_disable_cnt;
 235
 236     /* Number of AioHandlers without .io_poll() */
 237     int poll_disable_cnt;
 238
 239     /* Polling mode parameters */
 240     int64_t poll_ns;        /* current polling time in nanoseconds */
 241     int64_t poll_max_ns;    /* maximum polling time in nanoseconds */
 242     int64_t poll_grow;      /* polling time growth factor */
 243     int64_t poll_shrink;    /* polling time shrink factor */
 244
 245     /* AIO engine parameters */
 246     int64_t aio_max_batch;  /* maximum number of requests in a batch */
 247
 248     /*
 249      * List of handlers participating in userspace polling.  Protected by
 250      * ctx->list_lock.  Iterated and modified mostly by the event loop thread
 251      * from aio_poll() with ctx->list_lock incremented.  aio_set_fd_handler()
 252      * only touches the list to delete nodes if ctx->list_lock's count is zero.
 253      */
 254     AioHandlerList poll_aio_handlers;
 255
 256     /* Are we in polling mode or monitoring file descriptors? */
 257     bool poll_started;
 258
 259     /* epoll(7) state used when built with CONFIG_EPOLL */
 260     int epollfd;
 261
 262     const FDMonOps *fdmon_ops;
 263 };
 264
 265 /**
 266  * aio_context_new: Allocate a new AioContext.
 267  *
 268  * AioContext provide a mini event-loop that can be waited on synchronously.
 269  * They also provide bottom halves, a service to execute a piece of code
 270  * as soon as possible.
 271  */
 272 AioContext *aio_context_new(Error **errp);
 273
 274 /**
 275  * aio_context_ref:
 276  * @ctx: The AioContext to operate on.
 277  *
 278  * Add a reference to an AioContext.
 279  */
 280 void aio_context_ref(AioContext *ctx);
 281
 282 /**
 283  * aio_context_unref:
 284  * @ctx: The AioContext to operate on.
 285  *
 286  * Drop a reference to an AioContext.
 287  */
 288 void aio_context_unref(AioContext *ctx);
 289
 290 /* Take ownership of the AioContext.  If the AioContext will be shared between
 291  * threads, and a thread does not want to be interrupted, it will have to
 292  * take ownership around calls to aio_poll().  Otherwise, aio_poll()
 293  * automatically takes care of calling aio_context_acquire and
 294  * aio_context_release.
 295  *
 296  * Note that this is separate from bdrv_drained_begin/bdrv_drained_end.  A
 297  * thread still has to call those to avoid being interrupted by the guest.
 298  *
 299  * Bottom halves, timers and callbacks can be created or removed without
 300  * acquiring the AioContext.
 301  */
 302 void aio_context_acquire(AioContext *ctx);
 303
 304 /* Relinquish ownership of the AioContext. */
 305 void aio_context_release(AioContext *ctx);
 306
 307 /**
 308  * aio_bh_schedule_oneshot_full: Allocate a new bottom half structure that will
 309  * run only once and as soon as possible.
 310  *
 311  * @name: A human-readable identifier for debugging purposes.
 312  */
 313 void aio_bh_schedule_oneshot_full(AioContext *ctx, QEMUBHFunc *cb, void *opaque,
 314                                   const char *name);
 315
 316 /**
 317  * aio_bh_schedule_oneshot: Allocate a new bottom half structure that will run
 318  * only once and as soon as possible.
 319  *
 320  * A convenience wrapper for aio_bh_schedule_oneshot_full() that uses cb as the
 321  * name string.
 322  */
 323 #define aio_bh_schedule_oneshot(ctx, cb, opaque) \
 324     aio_bh_schedule_oneshot_full((ctx), (cb), (opaque), (stringify(cb)))
 325
 326 /**
 327  * aio_bh_new_full: Allocate a new bottom half structure.
 328  *
 329  * Bottom halves are lightweight callbacks whose invocation is guaranteed
 330  * to be wait-free, thread-safe and signal-safe.  The #QEMUBH structure
 331  * is opaque and must be allocated prior to its use.
 332  *
 333  * @name: A human-readable identifier for debugging purposes.
 334  */
 335 QEMUBH *aio_bh_new_full(AioContext *ctx, QEMUBHFunc *cb, void *opaque,
 336                         const char *name);
 337
 338 /**
 339  * aio_bh_new: Allocate a new bottom half structure
 340  *
 341  * A convenience wrapper for aio_bh_new_full() that uses the cb as the name
 342  * string.
 343  */
 344 #define aio_bh_new(ctx, cb, opaque) \
 345     aio_bh_new_full((ctx), (cb), (opaque), (stringify(cb)))
 346
 347 /**
 348  * aio_notify: Force processing of pending events.
 349  *
 350  * Similar to signaling a condition variable, aio_notify forces
 351  * aio_poll to exit, so that the next call will re-examine pending events.
 352  * The caller of aio_notify will usually call aio_poll again very soon,
 353  * or go through another iteration of the GLib main loop.  Hence, aio_notify
 354  * also has the side effect of recalculating the sets of file descriptors
 355  * that the main loop waits for.
 356  *
 357  * Calling aio_notify is rarely necessary, because for example scheduling
 358  * a bottom half calls it already.
 359  */
 360 void aio_notify(AioContext *ctx);
 361
 362 /**
 363  * aio_notify_accept: Acknowledge receiving an aio_notify.
 364  *
 365  * aio_notify() uses an EventNotifier in order to wake up a sleeping
 366  * aio_poll() or g_main_context_iteration().  Calls to aio_notify() are
 367  * usually rare, but the AioContext has to clear the EventNotifier on
 368  * every aio_poll() or g_main_context_iteration() in order to avoid
 369  * busy waiting.  This event_notifier_test_and_clear() cannot be done
 370  * using the usual aio_context_set_event_notifier(), because it must
 371  * be done before processing all events (file descriptors, bottom halves,
 372  * timers).
 373  *
 374  * aio_notify_accept() is an optimized event_notifier_test_and_clear()
 375  * that is specific to an AioContext's notifier; it is used internally
 376  * to clear the EventNotifier only if aio_notify() had been called.
 377  */
 378 void aio_notify_accept(AioContext *ctx);
 379
 380 /**
 381  * aio_bh_call: Executes callback function of the specified BH.
 382  */
 383 void aio_bh_call(QEMUBH *bh);
 384
 385 /**
 386  * aio_bh_poll: Poll bottom halves for an AioContext.
 387  *
 388  * These are internal functions used by the QEMU main loop.
 389  * And notice that multiple occurrences of aio_bh_poll cannot
 390  * be called concurrently
 391  */
 392 int aio_bh_poll(AioContext *ctx);
 393
 394 /**
 395  * qemu_bh_schedule: Schedule a bottom half.
 396  *
 397  * Scheduling a bottom half interrupts the main loop and causes the
 398  * execution of the callback that was passed to qemu_bh_new.
 399  *
 400  * Bottom halves that are scheduled from a bottom half handler are instantly
 401  * invoked.  This can create an infinite loop if a bottom half handler
 402  * schedules itself.
 403  *
 404  * @bh: The bottom half to be scheduled.
 405  */
 406 void qemu_bh_schedule(QEMUBH *bh);
 407
 408 /**
 409  * qemu_bh_cancel: Cancel execution of a bottom half.
 410  *
 411  * Canceling execution of a bottom half undoes the effect of calls to
 412  * qemu_bh_schedule without freeing its resources yet.  While cancellation
 413  * itself is also wait-free and thread-safe, it can of course race with the
 414  * loop that executes bottom halves unless you are holding the iothread
 415  * mutex.  This makes it mostly useless if you are not holding the mutex.
 416  *
 417  * @bh: The bottom half to be canceled.
 418  */
 419 void qemu_bh_cancel(QEMUBH *bh);
 420
 421 /**
 422  *qemu_bh_delete: Cancel execution of a bottom half and free its resources.
 423  *
 424  * Deleting a bottom half frees the memory that was allocated for it by
 425  * qemu_bh_new.  It also implies canceling the bottom half if it was
 426  * scheduled.
 427  * This func is async. The bottom half will do the delete action at the finial
 428  * end.
 429  *
 430  * @bh: The bottom half to be deleted.
 431  */
 432 void qemu_bh_delete(QEMUBH *bh);
 433
 434 /* Return whether there are any pending callbacks from the GSource
 435  * attached to the AioContext, before g_poll is invoked.
 436  *
 437  * This is used internally in the implementation of the GSource.
 438  */
 439 bool aio_prepare(AioContext *ctx);
 440
 441 /* Return whether there are any pending callbacks from the GSource
 442  * attached to the AioContext, after g_poll is invoked.
 443  *
 444  * This is used internally in the implementation of the GSource.
 445  */
 446 bool aio_pending(AioContext *ctx);
 447
 448 /* Dispatch any pending callbacks from the GSource attached to the AioContext.
 449  *
 450  * This is used internally in the implementation of the GSource.
 451  */
 452 void aio_dispatch(AioContext *ctx);
 453
 454 /* Progress in completing AIO work to occur.  This can issue new pending
 455  * aio as a result of executing I/O completion or bh callbacks.
 456  *
 457  * Return whether any progress was made by executing AIO or bottom half
 458  * handlers.  If @blocking == true, this should always be true except
 459  * if someone called aio_notify.
 460  *
 461  * If there are no pending bottom halves, but there are pending AIO
 462  * operations, it may not be possible to make any progress without
 463  * blocking.  If @blocking is true, this function will wait until one
 464  * or more AIO events have completed, to ensure something has moved
 465  * before returning.
 466  */
 467 bool aio_poll(AioContext *ctx, bool blocking);
 468
 469 /* Register a file descriptor and associated callbacks.  Behaves very similarly
 470  * to qemu_set_fd_handler.  Unlike qemu_set_fd_handler, these callbacks will
 471  * be invoked when using aio_poll().
 472  *
 473  * Code that invokes AIO completion functions should rely on this function
 474  * instead of qemu_set_fd_handler[2].
 475  */
 476 void aio_set_fd_handler(AioContext *ctx,
 477                         int fd,
 478                         bool is_external,
 479                         IOHandler *io_read,
 480                         IOHandler *io_write,
 481                         AioPollFn *io_poll,
 482                         IOHandler *io_poll_ready,
 483                         void *opaque);
 484
 485 /* Set polling begin/end callbacks for a file descriptor that has already been
 486  * registered with aio_set_fd_handler.  Do nothing if the file descriptor is
 487  * not registered.
 488  */
 489 void aio_set_fd_poll(AioContext *ctx, int fd,
 490                      IOHandler *io_poll_begin,
 491                      IOHandler *io_poll_end);
 492
 493 /* Register an event notifier and associated callbacks.  Behaves very similarly
 494  * to event_notifier_set_handler.  Unlike event_notifier_set_handler, these callbacks
 495  * will be invoked when using aio_poll().
 496  *
 497  * Code that invokes AIO completion functions should rely on this function
 498  * instead of event_notifier_set_handler.
 499  */
 500 void aio_set_event_notifier(AioContext *ctx,
 501                             EventNotifier *notifier,
 502                             bool is_external,
 503                             EventNotifierHandler *io_read,
 504                             AioPollFn *io_poll,
 505                             EventNotifierHandler *io_poll_ready);
 506
 507 /* Set polling begin/end callbacks for an event notifier that has already been
 508  * registered with aio_set_event_notifier.  Do nothing if the event notifier is
 509  * not registered.
 510  */
 511 void aio_set_event_notifier_poll(AioContext *ctx,
 512                                  EventNotifier *notifier,
 513                                  EventNotifierHandler *io_poll_begin,
 514                                  EventNotifierHandler *io_poll_end);
 515
 516 /* Return a GSource that lets the main loop poll the file descriptors attached
 517  * to this AioContext.
 518  */
 519 GSource *aio_get_g_source(AioContext *ctx);
 520
 521 /* Return the ThreadPool bound to this AioContext */
 522 struct ThreadPool *aio_get_thread_pool(AioContext *ctx);
 523
 524 /* Setup the LinuxAioState bound to this AioContext */
 525 struct LinuxAioState *aio_setup_linux_aio(AioContext *ctx, Error **errp);
 526
 527 /* Return the LinuxAioState bound to this AioContext */
 528 struct LinuxAioState *aio_get_linux_aio(AioContext *ctx);
 529
 530 /* Setup the LuringState bound to this AioContext */
 531 struct LuringState *aio_setup_linux_io_uring(AioContext *ctx, Error **errp);
 532
 533 /* Return the LuringState bound to this AioContext */
 534 struct LuringState *aio_get_linux_io_uring(AioContext *ctx);
 535 /**
 536  * aio_timer_new_with_attrs:
 537  * @ctx: the aio context
 538  * @type: the clock type
 539  * @scale: the scale
 540  * @attributes: 0, or one to multiple OR'ed QEMU_TIMER_ATTR_<id> values
 541  *              to assign
 542  * @cb: the callback to call on timer expiry
 543  * @opaque: the opaque pointer to pass to the callback
 544  *
 545  * Allocate a new timer (with attributes) attached to the context @ctx.
 546  * The function is responsible for memory allocation.
 547  *
 548  * The preferred interface is aio_timer_init or aio_timer_init_with_attrs.
 549  * Use that unless you really need dynamic memory allocation.
 550  *
 551  * Returns: a pointer to the new timer
 552  */
 553 static inline QEMUTimer *aio_timer_new_with_attrs(AioContext *ctx,
 554                                                   QEMUClockType type,
 555                                                   int scale, int attributes,
 556                                                   QEMUTimerCB *cb, void *opaque)
 557 {
 558     return timer_new_full(&ctx->tlg, type, scale, attributes, cb, opaque);
 559 }
 560
 561 /**
 562  * aio_timer_new:
 563  * @ctx: the aio context
 564  * @type: the clock type
 565  * @scale: the scale
 566  * @cb: the callback to call on timer expiry
 567  * @opaque: the opaque pointer to pass to the callback
 568  *
 569  * Allocate a new timer attached to the context @ctx.
 570  * See aio_timer_new_with_attrs for details.
 571  *
 572  * Returns: a pointer to the new timer
 573  */
 574 static inline QEMUTimer *aio_timer_new(AioContext *ctx, QEMUClockType type,
 575                                        int scale,
 576                                        QEMUTimerCB *cb, void *opaque)
 577 {
 578     return timer_new_full(&ctx->tlg, type, scale, 0, cb, opaque);
 579 }
 580
 581 /**
 582  * aio_timer_init_with_attrs:
 583  * @ctx: the aio context
 584  * @ts: the timer
 585  * @type: the clock type
 586  * @scale: the scale
 587  * @attributes: 0, or one to multiple OR'ed QEMU_TIMER_ATTR_<id> values
 588  *              to assign
 589  * @cb: the callback to call on timer expiry
 590  * @opaque: the opaque pointer to pass to the callback
 591  *
 592  * Initialise a new timer (with attributes) attached to the context @ctx.
 593  * The caller is responsible for memory allocation.
 594  */
 595 static inline void aio_timer_init_with_attrs(AioContext *ctx,
 596                                              QEMUTimer *ts, QEMUClockType type,
 597                                              int scale, int attributes,
 598                                              QEMUTimerCB *cb, void *opaque)
 599 {
 600     timer_init_full(ts, &ctx->tlg, type, scale, attributes, cb, opaque);
 601 }
 602
 603 /**
 604  * aio_timer_init:
 605  * @ctx: the aio context
 606  * @ts: the timer
 607  * @type: the clock type
 608  * @scale: the scale
 609  * @cb: the callback to call on timer expiry
 610  * @opaque: the opaque pointer to pass to the callback
 611  *
 612  * Initialise a new timer attached to the context @ctx.
 613  * See aio_timer_init_with_attrs for details.
 614  */
 615 static inline void aio_timer_init(AioContext *ctx,
 616                                   QEMUTimer *ts, QEMUClockType type,
 617                                   int scale,
 618                                   QEMUTimerCB *cb, void *opaque)
 619 {
 620     timer_init_full(ts, &ctx->tlg, type, scale, 0, cb, opaque);
 621 }
 622
 623 /**
 624  * aio_compute_timeout:
 625  * @ctx: the aio context
 626  *
 627  * Compute the timeout that a blocking aio_poll should use.
 628  */
 629 int64_t aio_compute_timeout(AioContext *ctx);
 630
 631 /**
 632  * aio_disable_external:
 633  * @ctx: the aio context
 634  *
 635  * Disable the further processing of external clients.
 636  */
 637 static inline void aio_disable_external(AioContext *ctx)
 638 {
 639     qatomic_inc(&ctx->external_disable_cnt);
 640 }
 641
 642 /**
 643  * aio_enable_external:
 644  * @ctx: the aio context
 645  *
 646  * Enable the processing of external clients.
 647  */
 648 static inline void aio_enable_external(AioContext *ctx)
 649 {
 650     int old;
 651
 652     old = qatomic_fetch_dec(&ctx->external_disable_cnt);
 653     assert(old > 0);
 654     if (old == 1) {
 655         /* Kick event loop so it re-arms file descriptors */
 656         aio_notify(ctx);
 657     }
 658 }
 659
 660 /**
 661  * aio_external_disabled:
 662  * @ctx: the aio context
 663  *
 664  * Return true if the external clients are disabled.
 665  */
 666 static inline bool aio_external_disabled(AioContext *ctx)
 667 {
 668     return qatomic_read(&ctx->external_disable_cnt);
 669 }
 670
 671 /**
 672  * aio_node_check:
 673  * @ctx: the aio context
 674  * @is_external: Whether or not the checked node is an external event source.
 675  *
 676  * Check if the node's is_external flag is okay to be polled by the ctx at this
 677  * moment. True means green light.
 678  */
 679 static inline bool aio_node_check(AioContext *ctx, bool is_external)
 680 {
 681     return !is_external || !qatomic_read(&ctx->external_disable_cnt);
 682 }
 683
 684 /**
 685  * aio_co_schedule:
 686  * @ctx: the aio context
 687  * @co: the coroutine
 688  *
 689  * Start a coroutine on a remote AioContext.
 690  *
 691  * The coroutine must not be entered by anyone else while aio_co_schedule()
 692  * is active.  In addition the coroutine must have yielded unless ctx
 693  * is the context in which the coroutine is running (i.e. the value of
 694  * qemu_get_current_aio_context() from the coroutine itself).
 695  */
 696 void aio_co_schedule(AioContext *ctx, Coroutine *co);
 697
 698 /**
 699  * aio_co_reschedule_self:
 700  * @new_ctx: the new context
 701  *
 702  * Move the currently running coroutine to new_ctx. If the coroutine is already
 703  * running in new_ctx, do nothing.
 704  */
 705 void coroutine_fn aio_co_reschedule_self(AioContext *new_ctx);
 706
 707 /**
 708  * aio_co_wake:
 709  * @co: the coroutine
 710  *
 711  * Restart a coroutine on the AioContext where it was running last, thus
 712  * preventing coroutines from jumping from one context to another when they
 713  * go to sleep.
 714  *
 715  * aio_co_wake may be executed either in coroutine or non-coroutine
 716  * context.  The coroutine must not be entered by anyone else while
 717  * aio_co_wake() is active.
 718  */
 719 void aio_co_wake(Coroutine *co);
 720
 721 /**
 722  * aio_co_enter:
 723  * @ctx: the context to run the coroutine
 724  * @co: the coroutine to run
 725  *
 726  * Enter a coroutine in the specified AioContext.
 727  */
 728 void aio_co_enter(AioContext *ctx, Coroutine *co);
 729
 730 /**
 731  * Return the AioContext whose event loop runs in the current thread.
 732  *
 733  * If called from an IOThread this will be the IOThread's AioContext.  If
 734  * called from the main thread or with the "big QEMU lock" taken it
 735  * will be the main loop AioContext.
 736  */
 737 AioContext *qemu_get_current_aio_context(void);
 738
 739 void qemu_set_current_aio_context(AioContext *ctx);
 740
 741 /**
 742  * aio_context_setup:
 743  * @ctx: the aio context
 744  *
 745  * Initialize the aio context.
 746  */
 747 void aio_context_setup(AioContext *ctx);
 748
 749 /**
 750  * aio_context_destroy:
 751  * @ctx: the aio context
 752  *
 753  * Destroy the aio context.
 754  */
 755 void aio_context_destroy(AioContext *ctx);
 756
 757 /* Used internally, do not call outside AioContext code */
 758 void aio_context_use_g_source(AioContext *ctx);
 759
 760 /**
 761  * aio_context_set_poll_params:
 762  * @ctx: the aio context
 763  * @max_ns: how long to busy poll for, in nanoseconds
 764  * @grow: polling time growth factor
 765  * @shrink: polling time shrink factor
 766  *
 767  * Poll mode can be disabled by setting poll_max_ns to 0.
 768  */
 769 void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns,
 770                                  int64_t grow, int64_t shrink,
 771                                  Error **errp);
 772
 773 /**
 774  * aio_context_set_aio_params:
 775  * @ctx: the aio context
 776  * @max_batch: maximum number of requests in a batch, 0 means that the
 777  *             engine will use its default
 778  */
 779 void aio_context_set_aio_params(AioContext *ctx, int64_t max_batch,
 780                                 Error **errp);
 781
 782 /**
 783  * aio_context_set_thread_pool_params:
 784  * @ctx: the aio context
 785  * @min: min number of threads to have readily available in the thread pool
 786  * @min: max number of threads the thread pool can contain
 787  */
 788 void aio_context_set_thread_pool_params(AioContext *ctx, int64_t min,
 789                                         int64_t max, Error **errp);
 790 #endif