xref: /openbmc/qemu/include/block/aio.h (revision 68ba85cecc7a46ceb66c2f4b5e2165546821d062)
1737e150eSPaolo Bonzini /*
2737e150eSPaolo Bonzini  * QEMU aio implementation
3737e150eSPaolo Bonzini  *
4737e150eSPaolo Bonzini  * Copyright IBM, Corp. 2008
5737e150eSPaolo Bonzini  *
6737e150eSPaolo Bonzini  * Authors:
7737e150eSPaolo Bonzini  *  Anthony Liguori   <aliguori@us.ibm.com>
8737e150eSPaolo Bonzini  *
9737e150eSPaolo Bonzini  * This work is licensed under the terms of the GNU GPL, version 2.  See
10737e150eSPaolo Bonzini  * the COPYING file in the top-level directory.
11737e150eSPaolo Bonzini  *
12737e150eSPaolo Bonzini  */
13737e150eSPaolo Bonzini 
14737e150eSPaolo Bonzini #ifndef QEMU_AIO_H
15737e150eSPaolo Bonzini #define QEMU_AIO_H
16737e150eSPaolo Bonzini 
1773fd282eSStefan Hajnoczi #ifdef CONFIG_LINUX_IO_URING
1873fd282eSStefan Hajnoczi #include <liburing.h>
1973fd282eSStefan Hajnoczi #endif
20*68ba85ceSMarkus Armbruster #include "qemu/coroutine-core.h"
211de7afc9SPaolo Bonzini #include "qemu/queue.h"
221de7afc9SPaolo Bonzini #include "qemu/event_notifier.h"
23dcc772e2SLiu Ping Fan #include "qemu/thread.h"
24dae21b98SAlex Bligh #include "qemu/timer.h"
25aead9dc9SPaolo Bonzini #include "block/graph-lock.h"
26737e150eSPaolo Bonzini 
277c84b1b8SMarkus Armbruster typedef struct BlockAIOCB BlockAIOCB;
28097310b5SMarkus Armbruster typedef void BlockCompletionFunc(void *opaque, int ret);
29737e150eSPaolo Bonzini 
30737e150eSPaolo Bonzini typedef struct AIOCBInfo {
317c84b1b8SMarkus Armbruster     void (*cancel_async)(BlockAIOCB *acb);
327c84b1b8SMarkus Armbruster     AioContext *(*get_aio_context)(BlockAIOCB *acb);
33737e150eSPaolo Bonzini     size_t aiocb_size;
34737e150eSPaolo Bonzini } AIOCBInfo;
35737e150eSPaolo Bonzini 
367c84b1b8SMarkus Armbruster struct BlockAIOCB {
37737e150eSPaolo Bonzini     const AIOCBInfo *aiocb_info;
38737e150eSPaolo Bonzini     BlockDriverState *bs;
39097310b5SMarkus Armbruster     BlockCompletionFunc *cb;
40737e150eSPaolo Bonzini     void *opaque;
41f197fe2bSFam Zheng     int refcnt;
42737e150eSPaolo Bonzini };
43737e150eSPaolo Bonzini 
44737e150eSPaolo Bonzini void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
45097310b5SMarkus Armbruster                    BlockCompletionFunc *cb, void *opaque);
468007429aSFam Zheng void qemu_aio_unref(void *p);
47f197fe2bSFam Zheng void qemu_aio_ref(void *p);
48737e150eSPaolo Bonzini 
49737e150eSPaolo Bonzini typedef struct AioHandler AioHandler;
504749079cSStefan Hajnoczi typedef QLIST_HEAD(, AioHandler) AioHandlerList;
51737e150eSPaolo Bonzini typedef void QEMUBHFunc(void *opaque);
52f6a51c84SStefan Hajnoczi typedef bool AioPollFn(void *opaque);
53737e150eSPaolo Bonzini typedef void IOHandler(void *opaque);
54737e150eSPaolo Bonzini 
550c330a73SPaolo Bonzini struct Coroutine;
560187f5c9SPaolo Bonzini struct ThreadPool;
570187f5c9SPaolo Bonzini struct LinuxAioState;
586663a0a3SAarushi Mehta struct LuringState;
590187f5c9SPaolo Bonzini 
60aa38e19fSStefan Hajnoczi /* Is polling disabled? */
61aa38e19fSStefan Hajnoczi bool aio_poll_disabled(AioContext *ctx);
62aa38e19fSStefan Hajnoczi 
631f050a46SStefan Hajnoczi /* Callbacks for file descriptor monitoring implementations */
641f050a46SStefan Hajnoczi typedef struct {
651f050a46SStefan Hajnoczi     /*
661f050a46SStefan Hajnoczi      * update:
671f050a46SStefan Hajnoczi      * @ctx: the AioContext
68b321051cSStefan Hajnoczi      * @old_node: the existing handler or NULL if this file descriptor is being
69b321051cSStefan Hajnoczi      *            monitored for the first time
70b321051cSStefan Hajnoczi      * @new_node: the new handler or NULL if this file descriptor is being
71b321051cSStefan Hajnoczi      *            removed
721f050a46SStefan Hajnoczi      *
73b321051cSStefan Hajnoczi      * Add/remove/modify a monitored file descriptor.
741f050a46SStefan Hajnoczi      *
751f050a46SStefan Hajnoczi      * Called with ctx->list_lock acquired.
761f050a46SStefan Hajnoczi      */
77b321051cSStefan Hajnoczi     void (*update)(AioContext *ctx, AioHandler *old_node, AioHandler *new_node);
781f050a46SStefan Hajnoczi 
791f050a46SStefan Hajnoczi     /*
801f050a46SStefan Hajnoczi      * wait:
811f050a46SStefan Hajnoczi      * @ctx: the AioContext
821f050a46SStefan Hajnoczi      * @ready_list: list for handlers that become ready
831f050a46SStefan Hajnoczi      * @timeout: maximum duration to wait, in nanoseconds
841f050a46SStefan Hajnoczi      *
851f050a46SStefan Hajnoczi      * Wait for file descriptors to become ready and place them on ready_list.
861f050a46SStefan Hajnoczi      *
871f050a46SStefan Hajnoczi      * Called with ctx->list_lock incremented but not locked.
881f050a46SStefan Hajnoczi      *
891f050a46SStefan Hajnoczi      * Returns: number of ready file descriptors.
901f050a46SStefan Hajnoczi      */
911f050a46SStefan Hajnoczi     int (*wait)(AioContext *ctx, AioHandlerList *ready_list, int64_t timeout);
92aa38e19fSStefan Hajnoczi 
93aa38e19fSStefan Hajnoczi     /*
94aa38e19fSStefan Hajnoczi      * need_wait:
95aa38e19fSStefan Hajnoczi      * @ctx: the AioContext
96aa38e19fSStefan Hajnoczi      *
97aa38e19fSStefan Hajnoczi      * Tell aio_poll() when to stop userspace polling early because ->wait()
98aa38e19fSStefan Hajnoczi      * has fds ready.
99aa38e19fSStefan Hajnoczi      *
100aa38e19fSStefan Hajnoczi      * File descriptor monitoring implementations that cannot poll fd readiness
101aa38e19fSStefan Hajnoczi      * from userspace should use aio_poll_disabled() here.  This ensures that
102aa38e19fSStefan Hajnoczi      * file descriptors are not starved by handlers that frequently make
103aa38e19fSStefan Hajnoczi      * progress via userspace polling.
104aa38e19fSStefan Hajnoczi      *
105aa38e19fSStefan Hajnoczi      * Returns: true if ->wait() should be called, false otherwise.
106aa38e19fSStefan Hajnoczi      */
107aa38e19fSStefan Hajnoczi     bool (*need_wait)(AioContext *ctx);
1081f050a46SStefan Hajnoczi } FDMonOps;
1091f050a46SStefan Hajnoczi 
1108c6b0356SStefan Hajnoczi /*
1118c6b0356SStefan Hajnoczi  * Each aio_bh_poll() call carves off a slice of the BH list, so that newly
1128c6b0356SStefan Hajnoczi  * scheduled BHs are not processed until the next aio_bh_poll() call.  All
1138c6b0356SStefan Hajnoczi  * active aio_bh_poll() calls chain their slices together in a list, so that
1148c6b0356SStefan Hajnoczi  * nested aio_bh_poll() calls process all scheduled bottom halves.
1158c6b0356SStefan Hajnoczi  */
1168c6b0356SStefan Hajnoczi typedef QSLIST_HEAD(, QEMUBH) BHList;
1178c6b0356SStefan Hajnoczi typedef struct BHListSlice BHListSlice;
1188c6b0356SStefan Hajnoczi struct BHListSlice {
1198c6b0356SStefan Hajnoczi     BHList bh_list;
1208c6b0356SStefan Hajnoczi     QSIMPLEQ_ENTRY(BHListSlice) next;
1218c6b0356SStefan Hajnoczi };
1228c6b0356SStefan Hajnoczi 
12373fd282eSStefan Hajnoczi typedef QSLIST_HEAD(, AioHandler) AioHandlerSList;
12473fd282eSStefan Hajnoczi 
1256a1751b7SAlex Bligh struct AioContext {
126737e150eSPaolo Bonzini     GSource source;
127737e150eSPaolo Bonzini 
1287c690fd1SPaolo Bonzini     /* Used by AioContext users to protect from multi-threaded access.  */
1293fe71223SPaolo Bonzini     QemuRecMutex lock;
13098563fc3SStefan Hajnoczi 
131aead9dc9SPaolo Bonzini     /*
132aead9dc9SPaolo Bonzini      * Keep track of readers and writers of the block layer graph.
133aead9dc9SPaolo Bonzini      * This is essential to avoid performing additions and removal
134aead9dc9SPaolo Bonzini      * of nodes and edges from block graph while some
135aead9dc9SPaolo Bonzini      * other thread is traversing it.
136aead9dc9SPaolo Bonzini      */
137aead9dc9SPaolo Bonzini     BdrvGraphRWlock *bdrv_graph;
138aead9dc9SPaolo Bonzini 
1397c690fd1SPaolo Bonzini     /* The list of registered AIO handlers.  Protected by ctx->list_lock. */
1404749079cSStefan Hajnoczi     AioHandlerList aio_handlers;
1414749079cSStefan Hajnoczi 
1424749079cSStefan Hajnoczi     /* The list of AIO handlers to be deleted.  Protected by ctx->list_lock. */
1434749079cSStefan Hajnoczi     AioHandlerList deleted_aio_handlers;
144737e150eSPaolo Bonzini 
145eabc9779SPaolo Bonzini     /* Used to avoid unnecessary event_notifier_set calls in aio_notify;
1463c18a92dSPaolo Bonzini      * only written from the AioContext home thread, or under the BQL in
1473c18a92dSPaolo Bonzini      * the case of the main AioContext.  However, it is read from any
1483c18a92dSPaolo Bonzini      * thread so it is still accessed with atomic primitives.
1493c18a92dSPaolo Bonzini      *
1503c18a92dSPaolo Bonzini      * If this field is 0, everything (file descriptors, bottom halves,
1513c18a92dSPaolo Bonzini      * timers) will be re-evaluated before the next blocking poll() or
1523c18a92dSPaolo Bonzini      * io_uring wait; therefore, the event_notifier_set call can be
1533c18a92dSPaolo Bonzini      * skipped.  If it is non-zero, you may need to wake up a concurrent
1543c18a92dSPaolo Bonzini      * aio_poll or the glib main event loop, making event_notifier_set
1553c18a92dSPaolo Bonzini      * necessary.
156eabc9779SPaolo Bonzini      *
157eabc9779SPaolo Bonzini      * Bit 0 is reserved for GSource usage of the AioContext, and is 1
15854a16a63SCao jin      * between a call to aio_ctx_prepare and the next call to aio_ctx_check.
159eabc9779SPaolo Bonzini      * Bits 1-31 simply count the number of active calls to aio_poll
160eabc9779SPaolo Bonzini      * that are in the prepare or poll phase.
161eabc9779SPaolo Bonzini      *
162eabc9779SPaolo Bonzini      * The GSource and aio_poll must use a different mechanism because
163eabc9779SPaolo Bonzini      * there is no certainty that a call to GSource's prepare callback
164eabc9779SPaolo Bonzini      * (via g_main_context_prepare) is indeed followed by check and
165eabc9779SPaolo Bonzini      * dispatch.  It's not clear whether this would be a bug, but let's
166eabc9779SPaolo Bonzini      * play safe and allow it---it will just cause extra calls to
167eabc9779SPaolo Bonzini      * event_notifier_set until the next call to dispatch.
168eabc9779SPaolo Bonzini      *
169eabc9779SPaolo Bonzini      * Instead, the aio_poll calls include both the prepare and the
170eabc9779SPaolo Bonzini      * dispatch phase, hence a simple counter is enough for them.
1710ceb849bSPaolo Bonzini      */
172eabc9779SPaolo Bonzini     uint32_t notify_me;
1730ceb849bSPaolo Bonzini 
1747c690fd1SPaolo Bonzini     /* A lock to protect between QEMUBH and AioHandler adders and deleter,
1757c690fd1SPaolo Bonzini      * and to ensure that no callbacks are removed while we're walking and
1767c690fd1SPaolo Bonzini      * dispatching them.
177d7c99a12SPaolo Bonzini      */
178d7c99a12SPaolo Bonzini     QemuLockCnt list_lock;
1790ceb849bSPaolo Bonzini 
1808c6b0356SStefan Hajnoczi     /* Bottom Halves pending aio_bh_poll() processing */
1818c6b0356SStefan Hajnoczi     BHList bh_list;
1828c6b0356SStefan Hajnoczi 
1838c6b0356SStefan Hajnoczi     /* Chained BH list slices for each nested aio_bh_poll() call */
1848c6b0356SStefan Hajnoczi     QSIMPLEQ_HEAD(, BHListSlice) bh_slice_list;
185737e150eSPaolo Bonzini 
18605e514b1SPaolo Bonzini     /* Used by aio_notify.
18705e514b1SPaolo Bonzini      *
18805e514b1SPaolo Bonzini      * "notified" is used to avoid expensive event_notifier_test_and_clear
18905e514b1SPaolo Bonzini      * calls.  When it is clear, the EventNotifier is clear, or one thread
19005e514b1SPaolo Bonzini      * is going to clear "notified" before processing more events.  False
19105e514b1SPaolo Bonzini      * positives are possible, i.e. "notified" could be set even though the
19205e514b1SPaolo Bonzini      * EventNotifier is clear.
19305e514b1SPaolo Bonzini      *
19405e514b1SPaolo Bonzini      * Note that event_notifier_set *cannot* be optimized the same way.  For
19505e514b1SPaolo Bonzini      * more information on the problem that would result, see "#ifdef BUG2"
19605e514b1SPaolo Bonzini      * in the docs/aio_notify_accept.promela formal model.
19705e514b1SPaolo Bonzini      */
19805e514b1SPaolo Bonzini     bool notified;
199737e150eSPaolo Bonzini     EventNotifier notifier;
2006b5f8762SStefan Hajnoczi 
2010c330a73SPaolo Bonzini     QSLIST_HEAD(, Coroutine) scheduled_coroutines;
2020c330a73SPaolo Bonzini     QEMUBH *co_schedule_bh;
2030c330a73SPaolo Bonzini 
20471ad4713SNicolas Saenz Julienne     int thread_pool_min;
20571ad4713SNicolas Saenz Julienne     int thread_pool_max;
2067c690fd1SPaolo Bonzini     /* Thread pool for performing work and receiving completion callbacks.
2077c690fd1SPaolo Bonzini      * Has its own locking.
2087c690fd1SPaolo Bonzini      */
2099b34277dSStefan Hajnoczi     struct ThreadPool *thread_pool;
210dae21b98SAlex Bligh 
2110187f5c9SPaolo Bonzini #ifdef CONFIG_LINUX_AIO
2126663a0a3SAarushi Mehta     /*
2136663a0a3SAarushi Mehta      * State for native Linux AIO.  Uses aio_context_acquire/release for
2140187f5c9SPaolo Bonzini      * locking.
2150187f5c9SPaolo Bonzini      */
2160187f5c9SPaolo Bonzini     struct LinuxAioState *linux_aio;
2170187f5c9SPaolo Bonzini #endif
2186663a0a3SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING
2196663a0a3SAarushi Mehta     /*
2206663a0a3SAarushi Mehta      * State for Linux io_uring.  Uses aio_context_acquire/release for
2216663a0a3SAarushi Mehta      * locking.
2226663a0a3SAarushi Mehta      */
2236663a0a3SAarushi Mehta     struct LuringState *linux_io_uring;
22473fd282eSStefan Hajnoczi 
22573fd282eSStefan Hajnoczi     /* State for file descriptor monitoring using Linux io_uring */
22673fd282eSStefan Hajnoczi     struct io_uring fdmon_io_uring;
22773fd282eSStefan Hajnoczi     AioHandlerSList submit_list;
2286663a0a3SAarushi Mehta #endif
2290187f5c9SPaolo Bonzini 
2307c690fd1SPaolo Bonzini     /* TimerLists for calling timers - one per clock type.  Has its own
2317c690fd1SPaolo Bonzini      * locking.
2327c690fd1SPaolo Bonzini      */
233dae21b98SAlex Bligh     QEMUTimerListGroup tlg;
234c1e1e5faSFam Zheng 
235c1e1e5faSFam Zheng     int external_disable_cnt;
236fbe3fc5cSFam Zheng 
2374a1cba38SStefan Hajnoczi     /* Number of AioHandlers without .io_poll() */
2384a1cba38SStefan Hajnoczi     int poll_disable_cnt;
2394a1cba38SStefan Hajnoczi 
24082a41186SStefan Hajnoczi     /* Polling mode parameters */
24182a41186SStefan Hajnoczi     int64_t poll_ns;        /* current polling time in nanoseconds */
24282a41186SStefan Hajnoczi     int64_t poll_max_ns;    /* maximum polling time in nanoseconds */
24382a41186SStefan Hajnoczi     int64_t poll_grow;      /* polling time growth factor */
24482a41186SStefan Hajnoczi     int64_t poll_shrink;    /* polling time shrink factor */
2454a1cba38SStefan Hajnoczi 
2461793ad02SStefano Garzarella     /* AIO engine parameters */
2471793ad02SStefano Garzarella     int64_t aio_max_batch;  /* maximum number of requests in a batch */
2481793ad02SStefano Garzarella 
249d37d0e36SStefan Hajnoczi     /*
250d37d0e36SStefan Hajnoczi      * List of handlers participating in userspace polling.  Protected by
251d37d0e36SStefan Hajnoczi      * ctx->list_lock.  Iterated and modified mostly by the event loop thread
252d37d0e36SStefan Hajnoczi      * from aio_poll() with ctx->list_lock incremented.  aio_set_fd_handler()
253d37d0e36SStefan Hajnoczi      * only touches the list to delete nodes if ctx->list_lock's count is zero.
254d37d0e36SStefan Hajnoczi      */
255d37d0e36SStefan Hajnoczi     AioHandlerList poll_aio_handlers;
256d37d0e36SStefan Hajnoczi 
257684e508cSStefan Hajnoczi     /* Are we in polling mode or monitoring file descriptors? */
258684e508cSStefan Hajnoczi     bool poll_started;
259684e508cSStefan Hajnoczi 
260fbe3fc5cSFam Zheng     /* epoll(7) state used when built with CONFIG_EPOLL */
261fbe3fc5cSFam Zheng     int epollfd;
2621f050a46SStefan Hajnoczi 
2631f050a46SStefan Hajnoczi     const FDMonOps *fdmon_ops;
2646a1751b7SAlex Bligh };
265737e150eSPaolo Bonzini 
266737e150eSPaolo Bonzini /**
267737e150eSPaolo Bonzini  * aio_context_new: Allocate a new AioContext.
268737e150eSPaolo Bonzini  *
269737e150eSPaolo Bonzini  * AioContext provide a mini event-loop that can be waited on synchronously.
270737e150eSPaolo Bonzini  * They also provide bottom halves, a service to execute a piece of code
271737e150eSPaolo Bonzini  * as soon as possible.
272737e150eSPaolo Bonzini  */
2732f78e491SChrysostomos Nanakos AioContext *aio_context_new(Error **errp);
274737e150eSPaolo Bonzini 
275737e150eSPaolo Bonzini /**
276737e150eSPaolo Bonzini  * aio_context_ref:
277737e150eSPaolo Bonzini  * @ctx: The AioContext to operate on.
278737e150eSPaolo Bonzini  *
279737e150eSPaolo Bonzini  * Add a reference to an AioContext.
280737e150eSPaolo Bonzini  */
281737e150eSPaolo Bonzini void aio_context_ref(AioContext *ctx);
282737e150eSPaolo Bonzini 
283737e150eSPaolo Bonzini /**
284737e150eSPaolo Bonzini  * aio_context_unref:
285737e150eSPaolo Bonzini  * @ctx: The AioContext to operate on.
286737e150eSPaolo Bonzini  *
287737e150eSPaolo Bonzini  * Drop a reference to an AioContext.
288737e150eSPaolo Bonzini  */
289737e150eSPaolo Bonzini void aio_context_unref(AioContext *ctx);
290737e150eSPaolo Bonzini 
29198563fc3SStefan Hajnoczi /* Take ownership of the AioContext.  If the AioContext will be shared between
29249110174SPaolo Bonzini  * threads, and a thread does not want to be interrupted, it will have to
29349110174SPaolo Bonzini  * take ownership around calls to aio_poll().  Otherwise, aio_poll()
29449110174SPaolo Bonzini  * automatically takes care of calling aio_context_acquire and
29549110174SPaolo Bonzini  * aio_context_release.
29698563fc3SStefan Hajnoczi  *
2977c690fd1SPaolo Bonzini  * Note that this is separate from bdrv_drained_begin/bdrv_drained_end.  A
2987c690fd1SPaolo Bonzini  * thread still has to call those to avoid being interrupted by the guest.
2997c690fd1SPaolo Bonzini  *
3007c690fd1SPaolo Bonzini  * Bottom halves, timers and callbacks can be created or removed without
3017c690fd1SPaolo Bonzini  * acquiring the AioContext.
30298563fc3SStefan Hajnoczi  */
30398563fc3SStefan Hajnoczi void aio_context_acquire(AioContext *ctx);
30498563fc3SStefan Hajnoczi 
30598563fc3SStefan Hajnoczi /* Relinquish ownership of the AioContext. */
30698563fc3SStefan Hajnoczi void aio_context_release(AioContext *ctx);
30798563fc3SStefan Hajnoczi 
308737e150eSPaolo Bonzini /**
3090f08586cSStefan Hajnoczi  * aio_bh_schedule_oneshot_full: Allocate a new bottom half structure that will
3100f08586cSStefan Hajnoczi  * run only once and as soon as possible.
3110f08586cSStefan Hajnoczi  *
3120f08586cSStefan Hajnoczi  * @name: A human-readable identifier for debugging purposes.
3135b8bb359SPaolo Bonzini  */
3140f08586cSStefan Hajnoczi void aio_bh_schedule_oneshot_full(AioContext *ctx, QEMUBHFunc *cb, void *opaque,
3150f08586cSStefan Hajnoczi                                   const char *name);
3165b8bb359SPaolo Bonzini 
3175b8bb359SPaolo Bonzini /**
3180f08586cSStefan Hajnoczi  * aio_bh_schedule_oneshot: Allocate a new bottom half structure that will run
3190f08586cSStefan Hajnoczi  * only once and as soon as possible.
3200f08586cSStefan Hajnoczi  *
3210f08586cSStefan Hajnoczi  * A convenience wrapper for aio_bh_schedule_oneshot_full() that uses cb as the
3220f08586cSStefan Hajnoczi  * name string.
3230f08586cSStefan Hajnoczi  */
3240f08586cSStefan Hajnoczi #define aio_bh_schedule_oneshot(ctx, cb, opaque) \
3250f08586cSStefan Hajnoczi     aio_bh_schedule_oneshot_full((ctx), (cb), (opaque), (stringify(cb)))
3260f08586cSStefan Hajnoczi 
3270f08586cSStefan Hajnoczi /**
3280f08586cSStefan Hajnoczi  * aio_bh_new_full: Allocate a new bottom half structure.
329737e150eSPaolo Bonzini  *
330737e150eSPaolo Bonzini  * Bottom halves are lightweight callbacks whose invocation is guaranteed
331737e150eSPaolo Bonzini  * to be wait-free, thread-safe and signal-safe.  The #QEMUBH structure
332737e150eSPaolo Bonzini  * is opaque and must be allocated prior to its use.
3330f08586cSStefan Hajnoczi  *
3340f08586cSStefan Hajnoczi  * @name: A human-readable identifier for debugging purposes.
335737e150eSPaolo Bonzini  */
3360f08586cSStefan Hajnoczi QEMUBH *aio_bh_new_full(AioContext *ctx, QEMUBHFunc *cb, void *opaque,
3370f08586cSStefan Hajnoczi                         const char *name);
3380f08586cSStefan Hajnoczi 
3390f08586cSStefan Hajnoczi /**
3400f08586cSStefan Hajnoczi  * aio_bh_new: Allocate a new bottom half structure
3410f08586cSStefan Hajnoczi  *
3420f08586cSStefan Hajnoczi  * A convenience wrapper for aio_bh_new_full() that uses the cb as the name
3430f08586cSStefan Hajnoczi  * string.
3440f08586cSStefan Hajnoczi  */
3450f08586cSStefan Hajnoczi #define aio_bh_new(ctx, cb, opaque) \
3460f08586cSStefan Hajnoczi     aio_bh_new_full((ctx), (cb), (opaque), (stringify(cb)))
347737e150eSPaolo Bonzini 
348737e150eSPaolo Bonzini /**
349737e150eSPaolo Bonzini  * aio_notify: Force processing of pending events.
350737e150eSPaolo Bonzini  *
351737e150eSPaolo Bonzini  * Similar to signaling a condition variable, aio_notify forces
352722f8d90SYaowei Bai  * aio_poll to exit, so that the next call will re-examine pending events.
353722f8d90SYaowei Bai  * The caller of aio_notify will usually call aio_poll again very soon,
354737e150eSPaolo Bonzini  * or go through another iteration of the GLib main loop.  Hence, aio_notify
355737e150eSPaolo Bonzini  * also has the side effect of recalculating the sets of file descriptors
356737e150eSPaolo Bonzini  * that the main loop waits for.
357737e150eSPaolo Bonzini  *
358737e150eSPaolo Bonzini  * Calling aio_notify is rarely necessary, because for example scheduling
359737e150eSPaolo Bonzini  * a bottom half calls it already.
360737e150eSPaolo Bonzini  */
361737e150eSPaolo Bonzini void aio_notify(AioContext *ctx);
362737e150eSPaolo Bonzini 
363737e150eSPaolo Bonzini /**
36405e514b1SPaolo Bonzini  * aio_notify_accept: Acknowledge receiving an aio_notify.
36505e514b1SPaolo Bonzini  *
36605e514b1SPaolo Bonzini  * aio_notify() uses an EventNotifier in order to wake up a sleeping
36705e514b1SPaolo Bonzini  * aio_poll() or g_main_context_iteration().  Calls to aio_notify() are
36805e514b1SPaolo Bonzini  * usually rare, but the AioContext has to clear the EventNotifier on
36905e514b1SPaolo Bonzini  * every aio_poll() or g_main_context_iteration() in order to avoid
37005e514b1SPaolo Bonzini  * busy waiting.  This event_notifier_test_and_clear() cannot be done
37105e514b1SPaolo Bonzini  * using the usual aio_context_set_event_notifier(), because it must
37205e514b1SPaolo Bonzini  * be done before processing all events (file descriptors, bottom halves,
37305e514b1SPaolo Bonzini  * timers).
37405e514b1SPaolo Bonzini  *
37505e514b1SPaolo Bonzini  * aio_notify_accept() is an optimized event_notifier_test_and_clear()
37605e514b1SPaolo Bonzini  * that is specific to an AioContext's notifier; it is used internally
37705e514b1SPaolo Bonzini  * to clear the EventNotifier only if aio_notify() had been called.
37805e514b1SPaolo Bonzini  */
37905e514b1SPaolo Bonzini void aio_notify_accept(AioContext *ctx);
38005e514b1SPaolo Bonzini 
38105e514b1SPaolo Bonzini /**
382df281b80SPavel Dovgalyuk  * aio_bh_call: Executes callback function of the specified BH.
383df281b80SPavel Dovgalyuk  */
384df281b80SPavel Dovgalyuk void aio_bh_call(QEMUBH *bh);
385df281b80SPavel Dovgalyuk 
386df281b80SPavel Dovgalyuk /**
387737e150eSPaolo Bonzini  * aio_bh_poll: Poll bottom halves for an AioContext.
388737e150eSPaolo Bonzini  *
389737e150eSPaolo Bonzini  * These are internal functions used by the QEMU main loop.
390dcc772e2SLiu Ping Fan  * And notice that multiple occurrences of aio_bh_poll cannot
391dcc772e2SLiu Ping Fan  * be called concurrently
392737e150eSPaolo Bonzini  */
393737e150eSPaolo Bonzini int aio_bh_poll(AioContext *ctx);
394737e150eSPaolo Bonzini 
395737e150eSPaolo Bonzini /**
396737e150eSPaolo Bonzini  * qemu_bh_schedule: Schedule a bottom half.
397737e150eSPaolo Bonzini  *
398737e150eSPaolo Bonzini  * Scheduling a bottom half interrupts the main loop and causes the
399737e150eSPaolo Bonzini  * execution of the callback that was passed to qemu_bh_new.
400737e150eSPaolo Bonzini  *
401737e150eSPaolo Bonzini  * Bottom halves that are scheduled from a bottom half handler are instantly
402737e150eSPaolo Bonzini  * invoked.  This can create an infinite loop if a bottom half handler
403737e150eSPaolo Bonzini  * schedules itself.
404737e150eSPaolo Bonzini  *
405737e150eSPaolo Bonzini  * @bh: The bottom half to be scheduled.
406737e150eSPaolo Bonzini  */
407737e150eSPaolo Bonzini void qemu_bh_schedule(QEMUBH *bh);
408737e150eSPaolo Bonzini 
409737e150eSPaolo Bonzini /**
410737e150eSPaolo Bonzini  * qemu_bh_cancel: Cancel execution of a bottom half.
411737e150eSPaolo Bonzini  *
412737e150eSPaolo Bonzini  * Canceling execution of a bottom half undoes the effect of calls to
413737e150eSPaolo Bonzini  * qemu_bh_schedule without freeing its resources yet.  While cancellation
414737e150eSPaolo Bonzini  * itself is also wait-free and thread-safe, it can of course race with the
415737e150eSPaolo Bonzini  * loop that executes bottom halves unless you are holding the iothread
416737e150eSPaolo Bonzini  * mutex.  This makes it mostly useless if you are not holding the mutex.
417737e150eSPaolo Bonzini  *
418737e150eSPaolo Bonzini  * @bh: The bottom half to be canceled.
419737e150eSPaolo Bonzini  */
420737e150eSPaolo Bonzini void qemu_bh_cancel(QEMUBH *bh);
421737e150eSPaolo Bonzini 
422737e150eSPaolo Bonzini /**
423737e150eSPaolo Bonzini  *qemu_bh_delete: Cancel execution of a bottom half and free its resources.
424737e150eSPaolo Bonzini  *
425737e150eSPaolo Bonzini  * Deleting a bottom half frees the memory that was allocated for it by
426737e150eSPaolo Bonzini  * qemu_bh_new.  It also implies canceling the bottom half if it was
427737e150eSPaolo Bonzini  * scheduled.
428dcc772e2SLiu Ping Fan  * This func is async. The bottom half will do the delete action at the finial
429dcc772e2SLiu Ping Fan  * end.
430737e150eSPaolo Bonzini  *
431737e150eSPaolo Bonzini  * @bh: The bottom half to be deleted.
432737e150eSPaolo Bonzini  */
433737e150eSPaolo Bonzini void qemu_bh_delete(QEMUBH *bh);
434737e150eSPaolo Bonzini 
435737e150eSPaolo Bonzini /* Return whether there are any pending callbacks from the GSource
436a3462c65SPaolo Bonzini  * attached to the AioContext, before g_poll is invoked.
437a3462c65SPaolo Bonzini  *
438a3462c65SPaolo Bonzini  * This is used internally in the implementation of the GSource.
439a3462c65SPaolo Bonzini  */
440a3462c65SPaolo Bonzini bool aio_prepare(AioContext *ctx);
441a3462c65SPaolo Bonzini 
442a3462c65SPaolo Bonzini /* Return whether there are any pending callbacks from the GSource
443a3462c65SPaolo Bonzini  * attached to the AioContext, after g_poll is invoked.
444737e150eSPaolo Bonzini  *
445737e150eSPaolo Bonzini  * This is used internally in the implementation of the GSource.
446737e150eSPaolo Bonzini  */
447737e150eSPaolo Bonzini bool aio_pending(AioContext *ctx);
448737e150eSPaolo Bonzini 
449e4c7e2d1SPaolo Bonzini /* Dispatch any pending callbacks from the GSource attached to the AioContext.
450e4c7e2d1SPaolo Bonzini  *
451e4c7e2d1SPaolo Bonzini  * This is used internally in the implementation of the GSource.
452e4c7e2d1SPaolo Bonzini  */
453a153bf52SPaolo Bonzini void aio_dispatch(AioContext *ctx);
454e4c7e2d1SPaolo Bonzini 
455737e150eSPaolo Bonzini /* Progress in completing AIO work to occur.  This can issue new pending
456737e150eSPaolo Bonzini  * aio as a result of executing I/O completion or bh callbacks.
457737e150eSPaolo Bonzini  *
458acfb23adSPaolo Bonzini  * Return whether any progress was made by executing AIO or bottom half
459acfb23adSPaolo Bonzini  * handlers.  If @blocking == true, this should always be true except
460acfb23adSPaolo Bonzini  * if someone called aio_notify.
461737e150eSPaolo Bonzini  *
462737e150eSPaolo Bonzini  * If there are no pending bottom halves, but there are pending AIO
463737e150eSPaolo Bonzini  * operations, it may not be possible to make any progress without
464737e150eSPaolo Bonzini  * blocking.  If @blocking is true, this function will wait until one
465737e150eSPaolo Bonzini  * or more AIO events have completed, to ensure something has moved
466737e150eSPaolo Bonzini  * before returning.
467737e150eSPaolo Bonzini  */
468737e150eSPaolo Bonzini bool aio_poll(AioContext *ctx, bool blocking);
469737e150eSPaolo Bonzini 
470737e150eSPaolo Bonzini /* Register a file descriptor and associated callbacks.  Behaves very similarly
4716484e422SFam Zheng  * to qemu_set_fd_handler.  Unlike qemu_set_fd_handler, these callbacks will
47287f68d31SPaolo Bonzini  * be invoked when using aio_poll().
473737e150eSPaolo Bonzini  *
474737e150eSPaolo Bonzini  * Code that invokes AIO completion functions should rely on this function
475737e150eSPaolo Bonzini  * instead of qemu_set_fd_handler[2].
476737e150eSPaolo Bonzini  */
477737e150eSPaolo Bonzini void aio_set_fd_handler(AioContext *ctx,
478737e150eSPaolo Bonzini                         int fd,
479dca21ef2SFam Zheng                         bool is_external,
480737e150eSPaolo Bonzini                         IOHandler *io_read,
481737e150eSPaolo Bonzini                         IOHandler *io_write,
482f6a51c84SStefan Hajnoczi                         AioPollFn *io_poll,
483826cc324SStefan Hajnoczi                         IOHandler *io_poll_ready,
484737e150eSPaolo Bonzini                         void *opaque);
485737e150eSPaolo Bonzini 
486684e508cSStefan Hajnoczi /* Set polling begin/end callbacks for a file descriptor that has already been
487684e508cSStefan Hajnoczi  * registered with aio_set_fd_handler.  Do nothing if the file descriptor is
488684e508cSStefan Hajnoczi  * not registered.
489684e508cSStefan Hajnoczi  */
490684e508cSStefan Hajnoczi void aio_set_fd_poll(AioContext *ctx, int fd,
491684e508cSStefan Hajnoczi                      IOHandler *io_poll_begin,
492684e508cSStefan Hajnoczi                      IOHandler *io_poll_end);
493684e508cSStefan Hajnoczi 
494737e150eSPaolo Bonzini /* Register an event notifier and associated callbacks.  Behaves very similarly
495737e150eSPaolo Bonzini  * to event_notifier_set_handler.  Unlike event_notifier_set_handler, these callbacks
49687f68d31SPaolo Bonzini  * will be invoked when using aio_poll().
497737e150eSPaolo Bonzini  *
498737e150eSPaolo Bonzini  * Code that invokes AIO completion functions should rely on this function
499737e150eSPaolo Bonzini  * instead of event_notifier_set_handler.
500737e150eSPaolo Bonzini  */
501737e150eSPaolo Bonzini void aio_set_event_notifier(AioContext *ctx,
502737e150eSPaolo Bonzini                             EventNotifier *notifier,
503dca21ef2SFam Zheng                             bool is_external,
504f6a51c84SStefan Hajnoczi                             EventNotifierHandler *io_read,
505826cc324SStefan Hajnoczi                             AioPollFn *io_poll,
506826cc324SStefan Hajnoczi                             EventNotifierHandler *io_poll_ready);
507737e150eSPaolo Bonzini 
508684e508cSStefan Hajnoczi /* Set polling begin/end callbacks for an event notifier that has already been
509684e508cSStefan Hajnoczi  * registered with aio_set_event_notifier.  Do nothing if the event notifier is
510684e508cSStefan Hajnoczi  * not registered.
511684e508cSStefan Hajnoczi  */
512684e508cSStefan Hajnoczi void aio_set_event_notifier_poll(AioContext *ctx,
513684e508cSStefan Hajnoczi                                  EventNotifier *notifier,
514684e508cSStefan Hajnoczi                                  EventNotifierHandler *io_poll_begin,
515684e508cSStefan Hajnoczi                                  EventNotifierHandler *io_poll_end);
516684e508cSStefan Hajnoczi 
517737e150eSPaolo Bonzini /* Return a GSource that lets the main loop poll the file descriptors attached
518737e150eSPaolo Bonzini  * to this AioContext.
519737e150eSPaolo Bonzini  */
520737e150eSPaolo Bonzini GSource *aio_get_g_source(AioContext *ctx);
521737e150eSPaolo Bonzini 
5229b34277dSStefan Hajnoczi /* Return the ThreadPool bound to this AioContext */
5239b34277dSStefan Hajnoczi struct ThreadPool *aio_get_thread_pool(AioContext *ctx);
5249b34277dSStefan Hajnoczi 
525ed6e2161SNishanth Aravamudan /* Setup the LinuxAioState bound to this AioContext */
526ed6e2161SNishanth Aravamudan struct LinuxAioState *aio_setup_linux_aio(AioContext *ctx, Error **errp);
527ed6e2161SNishanth Aravamudan 
5280187f5c9SPaolo Bonzini /* Return the LinuxAioState bound to this AioContext */
5290187f5c9SPaolo Bonzini struct LinuxAioState *aio_get_linux_aio(AioContext *ctx);
5300187f5c9SPaolo Bonzini 
5316663a0a3SAarushi Mehta /* Setup the LuringState bound to this AioContext */
5326663a0a3SAarushi Mehta struct LuringState *aio_setup_linux_io_uring(AioContext *ctx, Error **errp);
5336663a0a3SAarushi Mehta 
5346663a0a3SAarushi Mehta /* Return the LuringState bound to this AioContext */
5356663a0a3SAarushi Mehta struct LuringState *aio_get_linux_io_uring(AioContext *ctx);
5364e29e831SAlex Bligh /**
53789a603a0SArtem Pisarenko  * aio_timer_new_with_attrs:
53889a603a0SArtem Pisarenko  * @ctx: the aio context
53989a603a0SArtem Pisarenko  * @type: the clock type
54089a603a0SArtem Pisarenko  * @scale: the scale
54189a603a0SArtem Pisarenko  * @attributes: 0, or one to multiple OR'ed QEMU_TIMER_ATTR_<id> values
54289a603a0SArtem Pisarenko  *              to assign
54389a603a0SArtem Pisarenko  * @cb: the callback to call on timer expiry
54489a603a0SArtem Pisarenko  * @opaque: the opaque pointer to pass to the callback
54589a603a0SArtem Pisarenko  *
54689a603a0SArtem Pisarenko  * Allocate a new timer (with attributes) attached to the context @ctx.
54789a603a0SArtem Pisarenko  * The function is responsible for memory allocation.
54889a603a0SArtem Pisarenko  *
54989a603a0SArtem Pisarenko  * The preferred interface is aio_timer_init or aio_timer_init_with_attrs.
55089a603a0SArtem Pisarenko  * Use that unless you really need dynamic memory allocation.
55189a603a0SArtem Pisarenko  *
55289a603a0SArtem Pisarenko  * Returns: a pointer to the new timer
55389a603a0SArtem Pisarenko  */
55489a603a0SArtem Pisarenko static inline QEMUTimer *aio_timer_new_with_attrs(AioContext *ctx,
55589a603a0SArtem Pisarenko                                                   QEMUClockType type,
55689a603a0SArtem Pisarenko                                                   int scale, int attributes,
55789a603a0SArtem Pisarenko                                                   QEMUTimerCB *cb, void *opaque)
55889a603a0SArtem Pisarenko {
55989a603a0SArtem Pisarenko     return timer_new_full(&ctx->tlg, type, scale, attributes, cb, opaque);
56089a603a0SArtem Pisarenko }
56189a603a0SArtem Pisarenko 
56289a603a0SArtem Pisarenko /**
5634e29e831SAlex Bligh  * aio_timer_new:
5644e29e831SAlex Bligh  * @ctx: the aio context
5654e29e831SAlex Bligh  * @type: the clock type
5664e29e831SAlex Bligh  * @scale: the scale
5674e29e831SAlex Bligh  * @cb: the callback to call on timer expiry
5684e29e831SAlex Bligh  * @opaque: the opaque pointer to pass to the callback
5694e29e831SAlex Bligh  *
5704e29e831SAlex Bligh  * Allocate a new timer attached to the context @ctx.
57189a603a0SArtem Pisarenko  * See aio_timer_new_with_attrs for details.
5724e29e831SAlex Bligh  *
5734e29e831SAlex Bligh  * Returns: a pointer to the new timer
5744e29e831SAlex Bligh  */
5754e29e831SAlex Bligh static inline QEMUTimer *aio_timer_new(AioContext *ctx, QEMUClockType type,
5764e29e831SAlex Bligh                                        int scale,
5774e29e831SAlex Bligh                                        QEMUTimerCB *cb, void *opaque)
5784e29e831SAlex Bligh {
57989a603a0SArtem Pisarenko     return timer_new_full(&ctx->tlg, type, scale, 0, cb, opaque);
58089a603a0SArtem Pisarenko }
58189a603a0SArtem Pisarenko 
58289a603a0SArtem Pisarenko /**
58389a603a0SArtem Pisarenko  * aio_timer_init_with_attrs:
58489a603a0SArtem Pisarenko  * @ctx: the aio context
58589a603a0SArtem Pisarenko  * @ts: the timer
58689a603a0SArtem Pisarenko  * @type: the clock type
58789a603a0SArtem Pisarenko  * @scale: the scale
58889a603a0SArtem Pisarenko  * @attributes: 0, or one to multiple OR'ed QEMU_TIMER_ATTR_<id> values
58989a603a0SArtem Pisarenko  *              to assign
59089a603a0SArtem Pisarenko  * @cb: the callback to call on timer expiry
59189a603a0SArtem Pisarenko  * @opaque: the opaque pointer to pass to the callback
59289a603a0SArtem Pisarenko  *
59389a603a0SArtem Pisarenko  * Initialise a new timer (with attributes) attached to the context @ctx.
59489a603a0SArtem Pisarenko  * The caller is responsible for memory allocation.
59589a603a0SArtem Pisarenko  */
59689a603a0SArtem Pisarenko static inline void aio_timer_init_with_attrs(AioContext *ctx,
59789a603a0SArtem Pisarenko                                              QEMUTimer *ts, QEMUClockType type,
59889a603a0SArtem Pisarenko                                              int scale, int attributes,
59989a603a0SArtem Pisarenko                                              QEMUTimerCB *cb, void *opaque)
60089a603a0SArtem Pisarenko {
60189a603a0SArtem Pisarenko     timer_init_full(ts, &ctx->tlg, type, scale, attributes, cb, opaque);
6024e29e831SAlex Bligh }
6034e29e831SAlex Bligh 
6044e29e831SAlex Bligh /**
6054e29e831SAlex Bligh  * aio_timer_init:
6064e29e831SAlex Bligh  * @ctx: the aio context
6074e29e831SAlex Bligh  * @ts: the timer
6084e29e831SAlex Bligh  * @type: the clock type
6094e29e831SAlex Bligh  * @scale: the scale
6104e29e831SAlex Bligh  * @cb: the callback to call on timer expiry
6114e29e831SAlex Bligh  * @opaque: the opaque pointer to pass to the callback
6124e29e831SAlex Bligh  *
6134e29e831SAlex Bligh  * Initialise a new timer attached to the context @ctx.
61489a603a0SArtem Pisarenko  * See aio_timer_init_with_attrs for details.
6154e29e831SAlex Bligh  */
6164e29e831SAlex Bligh static inline void aio_timer_init(AioContext *ctx,
6174e29e831SAlex Bligh                                   QEMUTimer *ts, QEMUClockType type,
6184e29e831SAlex Bligh                                   int scale,
6194e29e831SAlex Bligh                                   QEMUTimerCB *cb, void *opaque)
6204e29e831SAlex Bligh {
62189a603a0SArtem Pisarenko     timer_init_full(ts, &ctx->tlg, type, scale, 0, cb, opaque);
6224e29e831SAlex Bligh }
6234e29e831SAlex Bligh 
624845ca10dSPaolo Bonzini /**
625845ca10dSPaolo Bonzini  * aio_compute_timeout:
626845ca10dSPaolo Bonzini  * @ctx: the aio context
627845ca10dSPaolo Bonzini  *
628845ca10dSPaolo Bonzini  * Compute the timeout that a blocking aio_poll should use.
629845ca10dSPaolo Bonzini  */
630845ca10dSPaolo Bonzini int64_t aio_compute_timeout(AioContext *ctx);
631845ca10dSPaolo Bonzini 
632c1e1e5faSFam Zheng /**
633c1e1e5faSFam Zheng  * aio_disable_external:
634c1e1e5faSFam Zheng  * @ctx: the aio context
635c1e1e5faSFam Zheng  *
636c1e1e5faSFam Zheng  * Disable the further processing of external clients.
637c1e1e5faSFam Zheng  */
638c1e1e5faSFam Zheng static inline void aio_disable_external(AioContext *ctx)
639c1e1e5faSFam Zheng {
640d73415a3SStefan Hajnoczi     qatomic_inc(&ctx->external_disable_cnt);
641c1e1e5faSFam Zheng }
642c1e1e5faSFam Zheng 
643c1e1e5faSFam Zheng /**
644c1e1e5faSFam Zheng  * aio_enable_external:
645c1e1e5faSFam Zheng  * @ctx: the aio context
646c1e1e5faSFam Zheng  *
647c1e1e5faSFam Zheng  * Enable the processing of external clients.
648c1e1e5faSFam Zheng  */
649c1e1e5faSFam Zheng static inline void aio_enable_external(AioContext *ctx)
650c1e1e5faSFam Zheng {
651321d1dbaSStefan Hajnoczi     int old;
652321d1dbaSStefan Hajnoczi 
653d73415a3SStefan Hajnoczi     old = qatomic_fetch_dec(&ctx->external_disable_cnt);
654321d1dbaSStefan Hajnoczi     assert(old > 0);
655321d1dbaSStefan Hajnoczi     if (old == 1) {
656321d1dbaSStefan Hajnoczi         /* Kick event loop so it re-arms file descriptors */
657321d1dbaSStefan Hajnoczi         aio_notify(ctx);
658321d1dbaSStefan Hajnoczi     }
659c1e1e5faSFam Zheng }
660c1e1e5faSFam Zheng 
661c1e1e5faSFam Zheng /**
6625ceb9e39SFam Zheng  * aio_external_disabled:
6635ceb9e39SFam Zheng  * @ctx: the aio context
6645ceb9e39SFam Zheng  *
6655ceb9e39SFam Zheng  * Return true if the external clients are disabled.
6665ceb9e39SFam Zheng  */
6675ceb9e39SFam Zheng static inline bool aio_external_disabled(AioContext *ctx)
6685ceb9e39SFam Zheng {
669d73415a3SStefan Hajnoczi     return qatomic_read(&ctx->external_disable_cnt);
6705ceb9e39SFam Zheng }
6715ceb9e39SFam Zheng 
6725ceb9e39SFam Zheng /**
673c1e1e5faSFam Zheng  * aio_node_check:
674c1e1e5faSFam Zheng  * @ctx: the aio context
675c1e1e5faSFam Zheng  * @is_external: Whether or not the checked node is an external event source.
676c1e1e5faSFam Zheng  *
677c1e1e5faSFam Zheng  * Check if the node's is_external flag is okay to be polled by the ctx at this
678c1e1e5faSFam Zheng  * moment. True means green light.
679c1e1e5faSFam Zheng  */
680c1e1e5faSFam Zheng static inline bool aio_node_check(AioContext *ctx, bool is_external)
681c1e1e5faSFam Zheng {
682d73415a3SStefan Hajnoczi     return !is_external || !qatomic_read(&ctx->external_disable_cnt);
683c1e1e5faSFam Zheng }
684c1e1e5faSFam Zheng 
68537fcee5dSFam Zheng /**
6860c330a73SPaolo Bonzini  * aio_co_schedule:
6870c330a73SPaolo Bonzini  * @ctx: the aio context
6880c330a73SPaolo Bonzini  * @co: the coroutine
6890c330a73SPaolo Bonzini  *
6900c330a73SPaolo Bonzini  * Start a coroutine on a remote AioContext.
6910c330a73SPaolo Bonzini  *
6920c330a73SPaolo Bonzini  * The coroutine must not be entered by anyone else while aio_co_schedule()
6930c330a73SPaolo Bonzini  * is active.  In addition the coroutine must have yielded unless ctx
6940c330a73SPaolo Bonzini  * is the context in which the coroutine is running (i.e. the value of
6950c330a73SPaolo Bonzini  * qemu_get_current_aio_context() from the coroutine itself).
6960c330a73SPaolo Bonzini  */
6970c330a73SPaolo Bonzini void aio_co_schedule(AioContext *ctx, struct Coroutine *co);
6980c330a73SPaolo Bonzini 
6990c330a73SPaolo Bonzini /**
70026b0b698SKevin Wolf  * aio_co_reschedule_self:
70126b0b698SKevin Wolf  * @new_ctx: the new context
70226b0b698SKevin Wolf  *
70326b0b698SKevin Wolf  * Move the currently running coroutine to new_ctx. If the coroutine is already
70426b0b698SKevin Wolf  * running in new_ctx, do nothing.
70526b0b698SKevin Wolf  */
70626b0b698SKevin Wolf void coroutine_fn aio_co_reschedule_self(AioContext *new_ctx);
70726b0b698SKevin Wolf 
70826b0b698SKevin Wolf /**
7090c330a73SPaolo Bonzini  * aio_co_wake:
7100c330a73SPaolo Bonzini  * @co: the coroutine
7110c330a73SPaolo Bonzini  *
7120c330a73SPaolo Bonzini  * Restart a coroutine on the AioContext where it was running last, thus
7130c330a73SPaolo Bonzini  * preventing coroutines from jumping from one context to another when they
7140c330a73SPaolo Bonzini  * go to sleep.
7150c330a73SPaolo Bonzini  *
7160c330a73SPaolo Bonzini  * aio_co_wake may be executed either in coroutine or non-coroutine
7170c330a73SPaolo Bonzini  * context.  The coroutine must not be entered by anyone else while
7180c330a73SPaolo Bonzini  * aio_co_wake() is active.
7190c330a73SPaolo Bonzini  */
7200c330a73SPaolo Bonzini void aio_co_wake(struct Coroutine *co);
7210c330a73SPaolo Bonzini 
7220c330a73SPaolo Bonzini /**
7238865852eSFam Zheng  * aio_co_enter:
7248865852eSFam Zheng  * @ctx: the context to run the coroutine
7258865852eSFam Zheng  * @co: the coroutine to run
7268865852eSFam Zheng  *
7278865852eSFam Zheng  * Enter a coroutine in the specified AioContext.
7288865852eSFam Zheng  */
7298865852eSFam Zheng void aio_co_enter(AioContext *ctx, struct Coroutine *co);
7308865852eSFam Zheng 
7318865852eSFam Zheng /**
732e4370165SPaolo Bonzini  * Return the AioContext whose event loop runs in the current thread.
733e4370165SPaolo Bonzini  *
734e4370165SPaolo Bonzini  * If called from an IOThread this will be the IOThread's AioContext.  If
7355f50be9bSPaolo Bonzini  * called from the main thread or with the "big QEMU lock" taken it
7365f50be9bSPaolo Bonzini  * will be the main loop AioContext.
737e4370165SPaolo Bonzini  */
738e4370165SPaolo Bonzini AioContext *qemu_get_current_aio_context(void);
739e4370165SPaolo Bonzini 
7405f50be9bSPaolo Bonzini void qemu_set_current_aio_context(AioContext *ctx);
7415f50be9bSPaolo Bonzini 
742e4370165SPaolo Bonzini /**
74337fcee5dSFam Zheng  * aio_context_setup:
74437fcee5dSFam Zheng  * @ctx: the aio context
74537fcee5dSFam Zheng  *
74637fcee5dSFam Zheng  * Initialize the aio context.
74737fcee5dSFam Zheng  */
7487e003465SCao jin void aio_context_setup(AioContext *ctx);
74937fcee5dSFam Zheng 
7504a1cba38SStefan Hajnoczi /**
751cd0a6d2bSJie Wang  * aio_context_destroy:
752cd0a6d2bSJie Wang  * @ctx: the aio context
753cd0a6d2bSJie Wang  *
754cd0a6d2bSJie Wang  * Destroy the aio context.
755cd0a6d2bSJie Wang  */
756cd0a6d2bSJie Wang void aio_context_destroy(AioContext *ctx);
757cd0a6d2bSJie Wang 
758ba607ca8SStefan Hajnoczi /* Used internally, do not call outside AioContext code */
759ba607ca8SStefan Hajnoczi void aio_context_use_g_source(AioContext *ctx);
760ba607ca8SStefan Hajnoczi 
761cd0a6d2bSJie Wang /**
7624a1cba38SStefan Hajnoczi  * aio_context_set_poll_params:
7634a1cba38SStefan Hajnoczi  * @ctx: the aio context
7644a1cba38SStefan Hajnoczi  * @max_ns: how long to busy poll for, in nanoseconds
76582a41186SStefan Hajnoczi  * @grow: polling time growth factor
76682a41186SStefan Hajnoczi  * @shrink: polling time shrink factor
7674a1cba38SStefan Hajnoczi  *
7684a1cba38SStefan Hajnoczi  * Poll mode can be disabled by setting poll_max_ns to 0.
7694a1cba38SStefan Hajnoczi  */
7704a1cba38SStefan Hajnoczi void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns,
77182a41186SStefan Hajnoczi                                  int64_t grow, int64_t shrink,
7724a1cba38SStefan Hajnoczi                                  Error **errp);
7734a1cba38SStefan Hajnoczi 
7741793ad02SStefano Garzarella /**
7751793ad02SStefano Garzarella  * aio_context_set_aio_params:
7761793ad02SStefano Garzarella  * @ctx: the aio context
7771793ad02SStefano Garzarella  * @max_batch: maximum number of requests in a batch, 0 means that the
7781793ad02SStefano Garzarella  *             engine will use its default
7791793ad02SStefano Garzarella  */
7801793ad02SStefano Garzarella void aio_context_set_aio_params(AioContext *ctx, int64_t max_batch,
7811793ad02SStefano Garzarella                                 Error **errp);
7821793ad02SStefano Garzarella 
78371ad4713SNicolas Saenz Julienne /**
78471ad4713SNicolas Saenz Julienne  * aio_context_set_thread_pool_params:
78571ad4713SNicolas Saenz Julienne  * @ctx: the aio context
78671ad4713SNicolas Saenz Julienne  * @min: min number of threads to have readily available in the thread pool
78771ad4713SNicolas Saenz Julienne  * @min: max number of threads the thread pool can contain
78871ad4713SNicolas Saenz Julienne  */
78971ad4713SNicolas Saenz Julienne void aio_context_set_thread_pool_params(AioContext *ctx, int64_t min,
79071ad4713SNicolas Saenz Julienne                                         int64_t max, Error **errp);
791737e150eSPaolo Bonzini #endif
792