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 2026b0b698SKevin Wolf #include "qemu/coroutine.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" 25737e150eSPaolo Bonzini 267c84b1b8SMarkus Armbruster typedef struct BlockAIOCB BlockAIOCB; 27097310b5SMarkus Armbruster typedef void BlockCompletionFunc(void *opaque, int ret); 28737e150eSPaolo Bonzini 29737e150eSPaolo Bonzini typedef struct AIOCBInfo { 307c84b1b8SMarkus Armbruster void (*cancel_async)(BlockAIOCB *acb); 317c84b1b8SMarkus Armbruster AioContext *(*get_aio_context)(BlockAIOCB *acb); 32737e150eSPaolo Bonzini size_t aiocb_size; 33737e150eSPaolo Bonzini } AIOCBInfo; 34737e150eSPaolo Bonzini 357c84b1b8SMarkus Armbruster struct BlockAIOCB { 36737e150eSPaolo Bonzini const AIOCBInfo *aiocb_info; 37737e150eSPaolo Bonzini BlockDriverState *bs; 38097310b5SMarkus Armbruster BlockCompletionFunc *cb; 39737e150eSPaolo Bonzini void *opaque; 40f197fe2bSFam Zheng int refcnt; 41737e150eSPaolo Bonzini }; 42737e150eSPaolo Bonzini 43737e150eSPaolo Bonzini void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs, 44097310b5SMarkus Armbruster BlockCompletionFunc *cb, void *opaque); 458007429aSFam Zheng void qemu_aio_unref(void *p); 46f197fe2bSFam Zheng void qemu_aio_ref(void *p); 47737e150eSPaolo Bonzini 48737e150eSPaolo Bonzini typedef struct AioHandler AioHandler; 494749079cSStefan Hajnoczi typedef QLIST_HEAD(, AioHandler) AioHandlerList; 50737e150eSPaolo Bonzini typedef void QEMUBHFunc(void *opaque); 51f6a51c84SStefan Hajnoczi typedef bool AioPollFn(void *opaque); 52737e150eSPaolo Bonzini typedef void IOHandler(void *opaque); 53737e150eSPaolo Bonzini 540c330a73SPaolo Bonzini struct Coroutine; 550187f5c9SPaolo Bonzini struct ThreadPool; 560187f5c9SPaolo Bonzini struct LinuxAioState; 576663a0a3SAarushi Mehta struct LuringState; 580187f5c9SPaolo Bonzini 59aa38e19fSStefan Hajnoczi /* Is polling disabled? */ 60aa38e19fSStefan Hajnoczi bool aio_poll_disabled(AioContext *ctx); 61aa38e19fSStefan Hajnoczi 621f050a46SStefan Hajnoczi /* Callbacks for file descriptor monitoring implementations */ 631f050a46SStefan Hajnoczi typedef struct { 641f050a46SStefan Hajnoczi /* 651f050a46SStefan Hajnoczi * update: 661f050a46SStefan Hajnoczi * @ctx: the AioContext 67b321051cSStefan Hajnoczi * @old_node: the existing handler or NULL if this file descriptor is being 68b321051cSStefan Hajnoczi * monitored for the first time 69b321051cSStefan Hajnoczi * @new_node: the new handler or NULL if this file descriptor is being 70b321051cSStefan Hajnoczi * removed 711f050a46SStefan Hajnoczi * 72b321051cSStefan Hajnoczi * Add/remove/modify a monitored file descriptor. 731f050a46SStefan Hajnoczi * 741f050a46SStefan Hajnoczi * Called with ctx->list_lock acquired. 751f050a46SStefan Hajnoczi */ 76b321051cSStefan Hajnoczi void (*update)(AioContext *ctx, AioHandler *old_node, AioHandler *new_node); 771f050a46SStefan Hajnoczi 781f050a46SStefan Hajnoczi /* 791f050a46SStefan Hajnoczi * wait: 801f050a46SStefan Hajnoczi * @ctx: the AioContext 811f050a46SStefan Hajnoczi * @ready_list: list for handlers that become ready 821f050a46SStefan Hajnoczi * @timeout: maximum duration to wait, in nanoseconds 831f050a46SStefan Hajnoczi * 841f050a46SStefan Hajnoczi * Wait for file descriptors to become ready and place them on ready_list. 851f050a46SStefan Hajnoczi * 861f050a46SStefan Hajnoczi * Called with ctx->list_lock incremented but not locked. 871f050a46SStefan Hajnoczi * 881f050a46SStefan Hajnoczi * Returns: number of ready file descriptors. 891f050a46SStefan Hajnoczi */ 901f050a46SStefan Hajnoczi int (*wait)(AioContext *ctx, AioHandlerList *ready_list, int64_t timeout); 91aa38e19fSStefan Hajnoczi 92aa38e19fSStefan Hajnoczi /* 93aa38e19fSStefan Hajnoczi * need_wait: 94aa38e19fSStefan Hajnoczi * @ctx: the AioContext 95aa38e19fSStefan Hajnoczi * 96aa38e19fSStefan Hajnoczi * Tell aio_poll() when to stop userspace polling early because ->wait() 97aa38e19fSStefan Hajnoczi * has fds ready. 98aa38e19fSStefan Hajnoczi * 99aa38e19fSStefan Hajnoczi * File descriptor monitoring implementations that cannot poll fd readiness 100aa38e19fSStefan Hajnoczi * from userspace should use aio_poll_disabled() here. This ensures that 101aa38e19fSStefan Hajnoczi * file descriptors are not starved by handlers that frequently make 102aa38e19fSStefan Hajnoczi * progress via userspace polling. 103aa38e19fSStefan Hajnoczi * 104aa38e19fSStefan Hajnoczi * Returns: true if ->wait() should be called, false otherwise. 105aa38e19fSStefan Hajnoczi */ 106aa38e19fSStefan Hajnoczi bool (*need_wait)(AioContext *ctx); 1071f050a46SStefan Hajnoczi } FDMonOps; 1081f050a46SStefan Hajnoczi 1098c6b0356SStefan Hajnoczi /* 1108c6b0356SStefan Hajnoczi * Each aio_bh_poll() call carves off a slice of the BH list, so that newly 1118c6b0356SStefan Hajnoczi * scheduled BHs are not processed until the next aio_bh_poll() call. All 1128c6b0356SStefan Hajnoczi * active aio_bh_poll() calls chain their slices together in a list, so that 1138c6b0356SStefan Hajnoczi * nested aio_bh_poll() calls process all scheduled bottom halves. 1148c6b0356SStefan Hajnoczi */ 1158c6b0356SStefan Hajnoczi typedef QSLIST_HEAD(, QEMUBH) BHList; 1168c6b0356SStefan Hajnoczi typedef struct BHListSlice BHListSlice; 1178c6b0356SStefan Hajnoczi struct BHListSlice { 1188c6b0356SStefan Hajnoczi BHList bh_list; 1198c6b0356SStefan Hajnoczi QSIMPLEQ_ENTRY(BHListSlice) next; 1208c6b0356SStefan Hajnoczi }; 1218c6b0356SStefan Hajnoczi 12273fd282eSStefan Hajnoczi typedef QSLIST_HEAD(, AioHandler) AioHandlerSList; 12373fd282eSStefan Hajnoczi 1246a1751b7SAlex Bligh struct AioContext { 125737e150eSPaolo Bonzini GSource source; 126737e150eSPaolo Bonzini 1277c690fd1SPaolo Bonzini /* Used by AioContext users to protect from multi-threaded access. */ 1283fe71223SPaolo Bonzini QemuRecMutex lock; 12998563fc3SStefan Hajnoczi 1307c690fd1SPaolo Bonzini /* The list of registered AIO handlers. Protected by ctx->list_lock. */ 1314749079cSStefan Hajnoczi AioHandlerList aio_handlers; 1324749079cSStefan Hajnoczi 1334749079cSStefan Hajnoczi /* The list of AIO handlers to be deleted. Protected by ctx->list_lock. */ 1344749079cSStefan Hajnoczi AioHandlerList deleted_aio_handlers; 135737e150eSPaolo Bonzini 136eabc9779SPaolo Bonzini /* Used to avoid unnecessary event_notifier_set calls in aio_notify; 1373c18a92dSPaolo Bonzini * only written from the AioContext home thread, or under the BQL in 1383c18a92dSPaolo Bonzini * the case of the main AioContext. However, it is read from any 1393c18a92dSPaolo Bonzini * thread so it is still accessed with atomic primitives. 1403c18a92dSPaolo Bonzini * 1413c18a92dSPaolo Bonzini * If this field is 0, everything (file descriptors, bottom halves, 1423c18a92dSPaolo Bonzini * timers) will be re-evaluated before the next blocking poll() or 1433c18a92dSPaolo Bonzini * io_uring wait; therefore, the event_notifier_set call can be 1443c18a92dSPaolo Bonzini * skipped. If it is non-zero, you may need to wake up a concurrent 1453c18a92dSPaolo Bonzini * aio_poll or the glib main event loop, making event_notifier_set 1463c18a92dSPaolo Bonzini * necessary. 147eabc9779SPaolo Bonzini * 148eabc9779SPaolo Bonzini * Bit 0 is reserved for GSource usage of the AioContext, and is 1 14954a16a63SCao jin * between a call to aio_ctx_prepare and the next call to aio_ctx_check. 150eabc9779SPaolo Bonzini * Bits 1-31 simply count the number of active calls to aio_poll 151eabc9779SPaolo Bonzini * that are in the prepare or poll phase. 152eabc9779SPaolo Bonzini * 153eabc9779SPaolo Bonzini * The GSource and aio_poll must use a different mechanism because 154eabc9779SPaolo Bonzini * there is no certainty that a call to GSource's prepare callback 155eabc9779SPaolo Bonzini * (via g_main_context_prepare) is indeed followed by check and 156eabc9779SPaolo Bonzini * dispatch. It's not clear whether this would be a bug, but let's 157eabc9779SPaolo Bonzini * play safe and allow it---it will just cause extra calls to 158eabc9779SPaolo Bonzini * event_notifier_set until the next call to dispatch. 159eabc9779SPaolo Bonzini * 160eabc9779SPaolo Bonzini * Instead, the aio_poll calls include both the prepare and the 161eabc9779SPaolo Bonzini * dispatch phase, hence a simple counter is enough for them. 1620ceb849bSPaolo Bonzini */ 163eabc9779SPaolo Bonzini uint32_t notify_me; 1640ceb849bSPaolo Bonzini 1657c690fd1SPaolo Bonzini /* A lock to protect between QEMUBH and AioHandler adders and deleter, 1667c690fd1SPaolo Bonzini * and to ensure that no callbacks are removed while we're walking and 1677c690fd1SPaolo Bonzini * dispatching them. 168d7c99a12SPaolo Bonzini */ 169d7c99a12SPaolo Bonzini QemuLockCnt list_lock; 1700ceb849bSPaolo Bonzini 1718c6b0356SStefan Hajnoczi /* Bottom Halves pending aio_bh_poll() processing */ 1728c6b0356SStefan Hajnoczi BHList bh_list; 1738c6b0356SStefan Hajnoczi 1748c6b0356SStefan Hajnoczi /* Chained BH list slices for each nested aio_bh_poll() call */ 1758c6b0356SStefan Hajnoczi QSIMPLEQ_HEAD(, BHListSlice) bh_slice_list; 176737e150eSPaolo Bonzini 17705e514b1SPaolo Bonzini /* Used by aio_notify. 17805e514b1SPaolo Bonzini * 17905e514b1SPaolo Bonzini * "notified" is used to avoid expensive event_notifier_test_and_clear 18005e514b1SPaolo Bonzini * calls. When it is clear, the EventNotifier is clear, or one thread 18105e514b1SPaolo Bonzini * is going to clear "notified" before processing more events. False 18205e514b1SPaolo Bonzini * positives are possible, i.e. "notified" could be set even though the 18305e514b1SPaolo Bonzini * EventNotifier is clear. 18405e514b1SPaolo Bonzini * 18505e514b1SPaolo Bonzini * Note that event_notifier_set *cannot* be optimized the same way. For 18605e514b1SPaolo Bonzini * more information on the problem that would result, see "#ifdef BUG2" 18705e514b1SPaolo Bonzini * in the docs/aio_notify_accept.promela formal model. 18805e514b1SPaolo Bonzini */ 18905e514b1SPaolo Bonzini bool notified; 190737e150eSPaolo Bonzini EventNotifier notifier; 1916b5f8762SStefan Hajnoczi 1920c330a73SPaolo Bonzini QSLIST_HEAD(, Coroutine) scheduled_coroutines; 1930c330a73SPaolo Bonzini QEMUBH *co_schedule_bh; 1940c330a73SPaolo Bonzini 195*71ad4713SNicolas Saenz Julienne int thread_pool_min; 196*71ad4713SNicolas Saenz Julienne int thread_pool_max; 1977c690fd1SPaolo Bonzini /* Thread pool for performing work and receiving completion callbacks. 1987c690fd1SPaolo Bonzini * Has its own locking. 1997c690fd1SPaolo Bonzini */ 2009b34277dSStefan Hajnoczi struct ThreadPool *thread_pool; 201dae21b98SAlex Bligh 2020187f5c9SPaolo Bonzini #ifdef CONFIG_LINUX_AIO 2036663a0a3SAarushi Mehta /* 2046663a0a3SAarushi Mehta * State for native Linux AIO. Uses aio_context_acquire/release for 2050187f5c9SPaolo Bonzini * locking. 2060187f5c9SPaolo Bonzini */ 2070187f5c9SPaolo Bonzini struct LinuxAioState *linux_aio; 2080187f5c9SPaolo Bonzini #endif 2096663a0a3SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2106663a0a3SAarushi Mehta /* 2116663a0a3SAarushi Mehta * State for Linux io_uring. Uses aio_context_acquire/release for 2126663a0a3SAarushi Mehta * locking. 2136663a0a3SAarushi Mehta */ 2146663a0a3SAarushi Mehta struct LuringState *linux_io_uring; 21573fd282eSStefan Hajnoczi 21673fd282eSStefan Hajnoczi /* State for file descriptor monitoring using Linux io_uring */ 21773fd282eSStefan Hajnoczi struct io_uring fdmon_io_uring; 21873fd282eSStefan Hajnoczi AioHandlerSList submit_list; 2196663a0a3SAarushi Mehta #endif 2200187f5c9SPaolo Bonzini 2217c690fd1SPaolo Bonzini /* TimerLists for calling timers - one per clock type. Has its own 2227c690fd1SPaolo Bonzini * locking. 2237c690fd1SPaolo Bonzini */ 224dae21b98SAlex Bligh QEMUTimerListGroup tlg; 225c1e1e5faSFam Zheng 226c1e1e5faSFam Zheng int external_disable_cnt; 227fbe3fc5cSFam Zheng 2284a1cba38SStefan Hajnoczi /* Number of AioHandlers without .io_poll() */ 2294a1cba38SStefan Hajnoczi int poll_disable_cnt; 2304a1cba38SStefan Hajnoczi 23182a41186SStefan Hajnoczi /* Polling mode parameters */ 23282a41186SStefan Hajnoczi int64_t poll_ns; /* current polling time in nanoseconds */ 23382a41186SStefan Hajnoczi int64_t poll_max_ns; /* maximum polling time in nanoseconds */ 23482a41186SStefan Hajnoczi int64_t poll_grow; /* polling time growth factor */ 23582a41186SStefan Hajnoczi int64_t poll_shrink; /* polling time shrink factor */ 2364a1cba38SStefan Hajnoczi 2371793ad02SStefano Garzarella /* AIO engine parameters */ 2381793ad02SStefano Garzarella int64_t aio_max_batch; /* maximum number of requests in a batch */ 2391793ad02SStefano Garzarella 240d37d0e36SStefan Hajnoczi /* 241d37d0e36SStefan Hajnoczi * List of handlers participating in userspace polling. Protected by 242d37d0e36SStefan Hajnoczi * ctx->list_lock. Iterated and modified mostly by the event loop thread 243d37d0e36SStefan Hajnoczi * from aio_poll() with ctx->list_lock incremented. aio_set_fd_handler() 244d37d0e36SStefan Hajnoczi * only touches the list to delete nodes if ctx->list_lock's count is zero. 245d37d0e36SStefan Hajnoczi */ 246d37d0e36SStefan Hajnoczi AioHandlerList poll_aio_handlers; 247d37d0e36SStefan Hajnoczi 248684e508cSStefan Hajnoczi /* Are we in polling mode or monitoring file descriptors? */ 249684e508cSStefan Hajnoczi bool poll_started; 250684e508cSStefan Hajnoczi 251fbe3fc5cSFam Zheng /* epoll(7) state used when built with CONFIG_EPOLL */ 252fbe3fc5cSFam Zheng int epollfd; 2531f050a46SStefan Hajnoczi 2541f050a46SStefan Hajnoczi const FDMonOps *fdmon_ops; 2556a1751b7SAlex Bligh }; 256737e150eSPaolo Bonzini 257737e150eSPaolo Bonzini /** 258737e150eSPaolo Bonzini * aio_context_new: Allocate a new AioContext. 259737e150eSPaolo Bonzini * 260737e150eSPaolo Bonzini * AioContext provide a mini event-loop that can be waited on synchronously. 261737e150eSPaolo Bonzini * They also provide bottom halves, a service to execute a piece of code 262737e150eSPaolo Bonzini * as soon as possible. 263737e150eSPaolo Bonzini */ 2642f78e491SChrysostomos Nanakos AioContext *aio_context_new(Error **errp); 265737e150eSPaolo Bonzini 266737e150eSPaolo Bonzini /** 267737e150eSPaolo Bonzini * aio_context_ref: 268737e150eSPaolo Bonzini * @ctx: The AioContext to operate on. 269737e150eSPaolo Bonzini * 270737e150eSPaolo Bonzini * Add a reference to an AioContext. 271737e150eSPaolo Bonzini */ 272737e150eSPaolo Bonzini void aio_context_ref(AioContext *ctx); 273737e150eSPaolo Bonzini 274737e150eSPaolo Bonzini /** 275737e150eSPaolo Bonzini * aio_context_unref: 276737e150eSPaolo Bonzini * @ctx: The AioContext to operate on. 277737e150eSPaolo Bonzini * 278737e150eSPaolo Bonzini * Drop a reference to an AioContext. 279737e150eSPaolo Bonzini */ 280737e150eSPaolo Bonzini void aio_context_unref(AioContext *ctx); 281737e150eSPaolo Bonzini 28298563fc3SStefan Hajnoczi /* Take ownership of the AioContext. If the AioContext will be shared between 28349110174SPaolo Bonzini * threads, and a thread does not want to be interrupted, it will have to 28449110174SPaolo Bonzini * take ownership around calls to aio_poll(). Otherwise, aio_poll() 28549110174SPaolo Bonzini * automatically takes care of calling aio_context_acquire and 28649110174SPaolo Bonzini * aio_context_release. 28798563fc3SStefan Hajnoczi * 2887c690fd1SPaolo Bonzini * Note that this is separate from bdrv_drained_begin/bdrv_drained_end. A 2897c690fd1SPaolo Bonzini * thread still has to call those to avoid being interrupted by the guest. 2907c690fd1SPaolo Bonzini * 2917c690fd1SPaolo Bonzini * Bottom halves, timers and callbacks can be created or removed without 2927c690fd1SPaolo Bonzini * acquiring the AioContext. 29398563fc3SStefan Hajnoczi */ 29498563fc3SStefan Hajnoczi void aio_context_acquire(AioContext *ctx); 29598563fc3SStefan Hajnoczi 29698563fc3SStefan Hajnoczi /* Relinquish ownership of the AioContext. */ 29798563fc3SStefan Hajnoczi void aio_context_release(AioContext *ctx); 29898563fc3SStefan Hajnoczi 299737e150eSPaolo Bonzini /** 3000f08586cSStefan Hajnoczi * aio_bh_schedule_oneshot_full: Allocate a new bottom half structure that will 3010f08586cSStefan Hajnoczi * run only once and as soon as possible. 3020f08586cSStefan Hajnoczi * 3030f08586cSStefan Hajnoczi * @name: A human-readable identifier for debugging purposes. 3045b8bb359SPaolo Bonzini */ 3050f08586cSStefan Hajnoczi void aio_bh_schedule_oneshot_full(AioContext *ctx, QEMUBHFunc *cb, void *opaque, 3060f08586cSStefan Hajnoczi const char *name); 3075b8bb359SPaolo Bonzini 3085b8bb359SPaolo Bonzini /** 3090f08586cSStefan Hajnoczi * aio_bh_schedule_oneshot: Allocate a new bottom half structure that will run 3100f08586cSStefan Hajnoczi * only once and as soon as possible. 3110f08586cSStefan Hajnoczi * 3120f08586cSStefan Hajnoczi * A convenience wrapper for aio_bh_schedule_oneshot_full() that uses cb as the 3130f08586cSStefan Hajnoczi * name string. 3140f08586cSStefan Hajnoczi */ 3150f08586cSStefan Hajnoczi #define aio_bh_schedule_oneshot(ctx, cb, opaque) \ 3160f08586cSStefan Hajnoczi aio_bh_schedule_oneshot_full((ctx), (cb), (opaque), (stringify(cb))) 3170f08586cSStefan Hajnoczi 3180f08586cSStefan Hajnoczi /** 3190f08586cSStefan Hajnoczi * aio_bh_new_full: Allocate a new bottom half structure. 320737e150eSPaolo Bonzini * 321737e150eSPaolo Bonzini * Bottom halves are lightweight callbacks whose invocation is guaranteed 322737e150eSPaolo Bonzini * to be wait-free, thread-safe and signal-safe. The #QEMUBH structure 323737e150eSPaolo Bonzini * is opaque and must be allocated prior to its use. 3240f08586cSStefan Hajnoczi * 3250f08586cSStefan Hajnoczi * @name: A human-readable identifier for debugging purposes. 326737e150eSPaolo Bonzini */ 3270f08586cSStefan Hajnoczi QEMUBH *aio_bh_new_full(AioContext *ctx, QEMUBHFunc *cb, void *opaque, 3280f08586cSStefan Hajnoczi const char *name); 3290f08586cSStefan Hajnoczi 3300f08586cSStefan Hajnoczi /** 3310f08586cSStefan Hajnoczi * aio_bh_new: Allocate a new bottom half structure 3320f08586cSStefan Hajnoczi * 3330f08586cSStefan Hajnoczi * A convenience wrapper for aio_bh_new_full() that uses the cb as the name 3340f08586cSStefan Hajnoczi * string. 3350f08586cSStefan Hajnoczi */ 3360f08586cSStefan Hajnoczi #define aio_bh_new(ctx, cb, opaque) \ 3370f08586cSStefan Hajnoczi aio_bh_new_full((ctx), (cb), (opaque), (stringify(cb))) 338737e150eSPaolo Bonzini 339737e150eSPaolo Bonzini /** 340737e150eSPaolo Bonzini * aio_notify: Force processing of pending events. 341737e150eSPaolo Bonzini * 342737e150eSPaolo Bonzini * Similar to signaling a condition variable, aio_notify forces 343722f8d90SYaowei Bai * aio_poll to exit, so that the next call will re-examine pending events. 344722f8d90SYaowei Bai * The caller of aio_notify will usually call aio_poll again very soon, 345737e150eSPaolo Bonzini * or go through another iteration of the GLib main loop. Hence, aio_notify 346737e150eSPaolo Bonzini * also has the side effect of recalculating the sets of file descriptors 347737e150eSPaolo Bonzini * that the main loop waits for. 348737e150eSPaolo Bonzini * 349737e150eSPaolo Bonzini * Calling aio_notify is rarely necessary, because for example scheduling 350737e150eSPaolo Bonzini * a bottom half calls it already. 351737e150eSPaolo Bonzini */ 352737e150eSPaolo Bonzini void aio_notify(AioContext *ctx); 353737e150eSPaolo Bonzini 354737e150eSPaolo Bonzini /** 35505e514b1SPaolo Bonzini * aio_notify_accept: Acknowledge receiving an aio_notify. 35605e514b1SPaolo Bonzini * 35705e514b1SPaolo Bonzini * aio_notify() uses an EventNotifier in order to wake up a sleeping 35805e514b1SPaolo Bonzini * aio_poll() or g_main_context_iteration(). Calls to aio_notify() are 35905e514b1SPaolo Bonzini * usually rare, but the AioContext has to clear the EventNotifier on 36005e514b1SPaolo Bonzini * every aio_poll() or g_main_context_iteration() in order to avoid 36105e514b1SPaolo Bonzini * busy waiting. This event_notifier_test_and_clear() cannot be done 36205e514b1SPaolo Bonzini * using the usual aio_context_set_event_notifier(), because it must 36305e514b1SPaolo Bonzini * be done before processing all events (file descriptors, bottom halves, 36405e514b1SPaolo Bonzini * timers). 36505e514b1SPaolo Bonzini * 36605e514b1SPaolo Bonzini * aio_notify_accept() is an optimized event_notifier_test_and_clear() 36705e514b1SPaolo Bonzini * that is specific to an AioContext's notifier; it is used internally 36805e514b1SPaolo Bonzini * to clear the EventNotifier only if aio_notify() had been called. 36905e514b1SPaolo Bonzini */ 37005e514b1SPaolo Bonzini void aio_notify_accept(AioContext *ctx); 37105e514b1SPaolo Bonzini 37205e514b1SPaolo Bonzini /** 373df281b80SPavel Dovgalyuk * aio_bh_call: Executes callback function of the specified BH. 374df281b80SPavel Dovgalyuk */ 375df281b80SPavel Dovgalyuk void aio_bh_call(QEMUBH *bh); 376df281b80SPavel Dovgalyuk 377df281b80SPavel Dovgalyuk /** 378737e150eSPaolo Bonzini * aio_bh_poll: Poll bottom halves for an AioContext. 379737e150eSPaolo Bonzini * 380737e150eSPaolo Bonzini * These are internal functions used by the QEMU main loop. 381dcc772e2SLiu Ping Fan * And notice that multiple occurrences of aio_bh_poll cannot 382dcc772e2SLiu Ping Fan * be called concurrently 383737e150eSPaolo Bonzini */ 384737e150eSPaolo Bonzini int aio_bh_poll(AioContext *ctx); 385737e150eSPaolo Bonzini 386737e150eSPaolo Bonzini /** 387737e150eSPaolo Bonzini * qemu_bh_schedule: Schedule a bottom half. 388737e150eSPaolo Bonzini * 389737e150eSPaolo Bonzini * Scheduling a bottom half interrupts the main loop and causes the 390737e150eSPaolo Bonzini * execution of the callback that was passed to qemu_bh_new. 391737e150eSPaolo Bonzini * 392737e150eSPaolo Bonzini * Bottom halves that are scheduled from a bottom half handler are instantly 393737e150eSPaolo Bonzini * invoked. This can create an infinite loop if a bottom half handler 394737e150eSPaolo Bonzini * schedules itself. 395737e150eSPaolo Bonzini * 396737e150eSPaolo Bonzini * @bh: The bottom half to be scheduled. 397737e150eSPaolo Bonzini */ 398737e150eSPaolo Bonzini void qemu_bh_schedule(QEMUBH *bh); 399737e150eSPaolo Bonzini 400737e150eSPaolo Bonzini /** 401737e150eSPaolo Bonzini * qemu_bh_cancel: Cancel execution of a bottom half. 402737e150eSPaolo Bonzini * 403737e150eSPaolo Bonzini * Canceling execution of a bottom half undoes the effect of calls to 404737e150eSPaolo Bonzini * qemu_bh_schedule without freeing its resources yet. While cancellation 405737e150eSPaolo Bonzini * itself is also wait-free and thread-safe, it can of course race with the 406737e150eSPaolo Bonzini * loop that executes bottom halves unless you are holding the iothread 407737e150eSPaolo Bonzini * mutex. This makes it mostly useless if you are not holding the mutex. 408737e150eSPaolo Bonzini * 409737e150eSPaolo Bonzini * @bh: The bottom half to be canceled. 410737e150eSPaolo Bonzini */ 411737e150eSPaolo Bonzini void qemu_bh_cancel(QEMUBH *bh); 412737e150eSPaolo Bonzini 413737e150eSPaolo Bonzini /** 414737e150eSPaolo Bonzini *qemu_bh_delete: Cancel execution of a bottom half and free its resources. 415737e150eSPaolo Bonzini * 416737e150eSPaolo Bonzini * Deleting a bottom half frees the memory that was allocated for it by 417737e150eSPaolo Bonzini * qemu_bh_new. It also implies canceling the bottom half if it was 418737e150eSPaolo Bonzini * scheduled. 419dcc772e2SLiu Ping Fan * This func is async. The bottom half will do the delete action at the finial 420dcc772e2SLiu Ping Fan * end. 421737e150eSPaolo Bonzini * 422737e150eSPaolo Bonzini * @bh: The bottom half to be deleted. 423737e150eSPaolo Bonzini */ 424737e150eSPaolo Bonzini void qemu_bh_delete(QEMUBH *bh); 425737e150eSPaolo Bonzini 426737e150eSPaolo Bonzini /* Return whether there are any pending callbacks from the GSource 427a3462c65SPaolo Bonzini * attached to the AioContext, before g_poll is invoked. 428a3462c65SPaolo Bonzini * 429a3462c65SPaolo Bonzini * This is used internally in the implementation of the GSource. 430a3462c65SPaolo Bonzini */ 431a3462c65SPaolo Bonzini bool aio_prepare(AioContext *ctx); 432a3462c65SPaolo Bonzini 433a3462c65SPaolo Bonzini /* Return whether there are any pending callbacks from the GSource 434a3462c65SPaolo Bonzini * attached to the AioContext, after g_poll is invoked. 435737e150eSPaolo Bonzini * 436737e150eSPaolo Bonzini * This is used internally in the implementation of the GSource. 437737e150eSPaolo Bonzini */ 438737e150eSPaolo Bonzini bool aio_pending(AioContext *ctx); 439737e150eSPaolo Bonzini 440e4c7e2d1SPaolo Bonzini /* Dispatch any pending callbacks from the GSource attached to the AioContext. 441e4c7e2d1SPaolo Bonzini * 442e4c7e2d1SPaolo Bonzini * This is used internally in the implementation of the GSource. 443e4c7e2d1SPaolo Bonzini */ 444a153bf52SPaolo Bonzini void aio_dispatch(AioContext *ctx); 445e4c7e2d1SPaolo Bonzini 446737e150eSPaolo Bonzini /* Progress in completing AIO work to occur. This can issue new pending 447737e150eSPaolo Bonzini * aio as a result of executing I/O completion or bh callbacks. 448737e150eSPaolo Bonzini * 449acfb23adSPaolo Bonzini * Return whether any progress was made by executing AIO or bottom half 450acfb23adSPaolo Bonzini * handlers. If @blocking == true, this should always be true except 451acfb23adSPaolo Bonzini * if someone called aio_notify. 452737e150eSPaolo Bonzini * 453737e150eSPaolo Bonzini * If there are no pending bottom halves, but there are pending AIO 454737e150eSPaolo Bonzini * operations, it may not be possible to make any progress without 455737e150eSPaolo Bonzini * blocking. If @blocking is true, this function will wait until one 456737e150eSPaolo Bonzini * or more AIO events have completed, to ensure something has moved 457737e150eSPaolo Bonzini * before returning. 458737e150eSPaolo Bonzini */ 459737e150eSPaolo Bonzini bool aio_poll(AioContext *ctx, bool blocking); 460737e150eSPaolo Bonzini 461737e150eSPaolo Bonzini /* Register a file descriptor and associated callbacks. Behaves very similarly 4626484e422SFam Zheng * to qemu_set_fd_handler. Unlike qemu_set_fd_handler, these callbacks will 46387f68d31SPaolo Bonzini * be invoked when using aio_poll(). 464737e150eSPaolo Bonzini * 465737e150eSPaolo Bonzini * Code that invokes AIO completion functions should rely on this function 466737e150eSPaolo Bonzini * instead of qemu_set_fd_handler[2]. 467737e150eSPaolo Bonzini */ 468737e150eSPaolo Bonzini void aio_set_fd_handler(AioContext *ctx, 469737e150eSPaolo Bonzini int fd, 470dca21ef2SFam Zheng bool is_external, 471737e150eSPaolo Bonzini IOHandler *io_read, 472737e150eSPaolo Bonzini IOHandler *io_write, 473f6a51c84SStefan Hajnoczi AioPollFn *io_poll, 474826cc324SStefan Hajnoczi IOHandler *io_poll_ready, 475737e150eSPaolo Bonzini void *opaque); 476737e150eSPaolo Bonzini 477684e508cSStefan Hajnoczi /* Set polling begin/end callbacks for a file descriptor that has already been 478684e508cSStefan Hajnoczi * registered with aio_set_fd_handler. Do nothing if the file descriptor is 479684e508cSStefan Hajnoczi * not registered. 480684e508cSStefan Hajnoczi */ 481684e508cSStefan Hajnoczi void aio_set_fd_poll(AioContext *ctx, int fd, 482684e508cSStefan Hajnoczi IOHandler *io_poll_begin, 483684e508cSStefan Hajnoczi IOHandler *io_poll_end); 484684e508cSStefan Hajnoczi 485737e150eSPaolo Bonzini /* Register an event notifier and associated callbacks. Behaves very similarly 486737e150eSPaolo Bonzini * to event_notifier_set_handler. Unlike event_notifier_set_handler, these callbacks 48787f68d31SPaolo Bonzini * will be invoked when using aio_poll(). 488737e150eSPaolo Bonzini * 489737e150eSPaolo Bonzini * Code that invokes AIO completion functions should rely on this function 490737e150eSPaolo Bonzini * instead of event_notifier_set_handler. 491737e150eSPaolo Bonzini */ 492737e150eSPaolo Bonzini void aio_set_event_notifier(AioContext *ctx, 493737e150eSPaolo Bonzini EventNotifier *notifier, 494dca21ef2SFam Zheng bool is_external, 495f6a51c84SStefan Hajnoczi EventNotifierHandler *io_read, 496826cc324SStefan Hajnoczi AioPollFn *io_poll, 497826cc324SStefan Hajnoczi EventNotifierHandler *io_poll_ready); 498737e150eSPaolo Bonzini 499684e508cSStefan Hajnoczi /* Set polling begin/end callbacks for an event notifier that has already been 500684e508cSStefan Hajnoczi * registered with aio_set_event_notifier. Do nothing if the event notifier is 501684e508cSStefan Hajnoczi * not registered. 502684e508cSStefan Hajnoczi */ 503684e508cSStefan Hajnoczi void aio_set_event_notifier_poll(AioContext *ctx, 504684e508cSStefan Hajnoczi EventNotifier *notifier, 505684e508cSStefan Hajnoczi EventNotifierHandler *io_poll_begin, 506684e508cSStefan Hajnoczi EventNotifierHandler *io_poll_end); 507684e508cSStefan Hajnoczi 508737e150eSPaolo Bonzini /* Return a GSource that lets the main loop poll the file descriptors attached 509737e150eSPaolo Bonzini * to this AioContext. 510737e150eSPaolo Bonzini */ 511737e150eSPaolo Bonzini GSource *aio_get_g_source(AioContext *ctx); 512737e150eSPaolo Bonzini 5139b34277dSStefan Hajnoczi /* Return the ThreadPool bound to this AioContext */ 5149b34277dSStefan Hajnoczi struct ThreadPool *aio_get_thread_pool(AioContext *ctx); 5159b34277dSStefan Hajnoczi 516ed6e2161SNishanth Aravamudan /* Setup the LinuxAioState bound to this AioContext */ 517ed6e2161SNishanth Aravamudan struct LinuxAioState *aio_setup_linux_aio(AioContext *ctx, Error **errp); 518ed6e2161SNishanth Aravamudan 5190187f5c9SPaolo Bonzini /* Return the LinuxAioState bound to this AioContext */ 5200187f5c9SPaolo Bonzini struct LinuxAioState *aio_get_linux_aio(AioContext *ctx); 5210187f5c9SPaolo Bonzini 5226663a0a3SAarushi Mehta /* Setup the LuringState bound to this AioContext */ 5236663a0a3SAarushi Mehta struct LuringState *aio_setup_linux_io_uring(AioContext *ctx, Error **errp); 5246663a0a3SAarushi Mehta 5256663a0a3SAarushi Mehta /* Return the LuringState bound to this AioContext */ 5266663a0a3SAarushi Mehta struct LuringState *aio_get_linux_io_uring(AioContext *ctx); 5274e29e831SAlex Bligh /** 52889a603a0SArtem Pisarenko * aio_timer_new_with_attrs: 52989a603a0SArtem Pisarenko * @ctx: the aio context 53089a603a0SArtem Pisarenko * @type: the clock type 53189a603a0SArtem Pisarenko * @scale: the scale 53289a603a0SArtem Pisarenko * @attributes: 0, or one to multiple OR'ed QEMU_TIMER_ATTR_<id> values 53389a603a0SArtem Pisarenko * to assign 53489a603a0SArtem Pisarenko * @cb: the callback to call on timer expiry 53589a603a0SArtem Pisarenko * @opaque: the opaque pointer to pass to the callback 53689a603a0SArtem Pisarenko * 53789a603a0SArtem Pisarenko * Allocate a new timer (with attributes) attached to the context @ctx. 53889a603a0SArtem Pisarenko * The function is responsible for memory allocation. 53989a603a0SArtem Pisarenko * 54089a603a0SArtem Pisarenko * The preferred interface is aio_timer_init or aio_timer_init_with_attrs. 54189a603a0SArtem Pisarenko * Use that unless you really need dynamic memory allocation. 54289a603a0SArtem Pisarenko * 54389a603a0SArtem Pisarenko * Returns: a pointer to the new timer 54489a603a0SArtem Pisarenko */ 54589a603a0SArtem Pisarenko static inline QEMUTimer *aio_timer_new_with_attrs(AioContext *ctx, 54689a603a0SArtem Pisarenko QEMUClockType type, 54789a603a0SArtem Pisarenko int scale, int attributes, 54889a603a0SArtem Pisarenko QEMUTimerCB *cb, void *opaque) 54989a603a0SArtem Pisarenko { 55089a603a0SArtem Pisarenko return timer_new_full(&ctx->tlg, type, scale, attributes, cb, opaque); 55189a603a0SArtem Pisarenko } 55289a603a0SArtem Pisarenko 55389a603a0SArtem Pisarenko /** 5544e29e831SAlex Bligh * aio_timer_new: 5554e29e831SAlex Bligh * @ctx: the aio context 5564e29e831SAlex Bligh * @type: the clock type 5574e29e831SAlex Bligh * @scale: the scale 5584e29e831SAlex Bligh * @cb: the callback to call on timer expiry 5594e29e831SAlex Bligh * @opaque: the opaque pointer to pass to the callback 5604e29e831SAlex Bligh * 5614e29e831SAlex Bligh * Allocate a new timer attached to the context @ctx. 56289a603a0SArtem Pisarenko * See aio_timer_new_with_attrs for details. 5634e29e831SAlex Bligh * 5644e29e831SAlex Bligh * Returns: a pointer to the new timer 5654e29e831SAlex Bligh */ 5664e29e831SAlex Bligh static inline QEMUTimer *aio_timer_new(AioContext *ctx, QEMUClockType type, 5674e29e831SAlex Bligh int scale, 5684e29e831SAlex Bligh QEMUTimerCB *cb, void *opaque) 5694e29e831SAlex Bligh { 57089a603a0SArtem Pisarenko return timer_new_full(&ctx->tlg, type, scale, 0, cb, opaque); 57189a603a0SArtem Pisarenko } 57289a603a0SArtem Pisarenko 57389a603a0SArtem Pisarenko /** 57489a603a0SArtem Pisarenko * aio_timer_init_with_attrs: 57589a603a0SArtem Pisarenko * @ctx: the aio context 57689a603a0SArtem Pisarenko * @ts: the timer 57789a603a0SArtem Pisarenko * @type: the clock type 57889a603a0SArtem Pisarenko * @scale: the scale 57989a603a0SArtem Pisarenko * @attributes: 0, or one to multiple OR'ed QEMU_TIMER_ATTR_<id> values 58089a603a0SArtem Pisarenko * to assign 58189a603a0SArtem Pisarenko * @cb: the callback to call on timer expiry 58289a603a0SArtem Pisarenko * @opaque: the opaque pointer to pass to the callback 58389a603a0SArtem Pisarenko * 58489a603a0SArtem Pisarenko * Initialise a new timer (with attributes) attached to the context @ctx. 58589a603a0SArtem Pisarenko * The caller is responsible for memory allocation. 58689a603a0SArtem Pisarenko */ 58789a603a0SArtem Pisarenko static inline void aio_timer_init_with_attrs(AioContext *ctx, 58889a603a0SArtem Pisarenko QEMUTimer *ts, QEMUClockType type, 58989a603a0SArtem Pisarenko int scale, int attributes, 59089a603a0SArtem Pisarenko QEMUTimerCB *cb, void *opaque) 59189a603a0SArtem Pisarenko { 59289a603a0SArtem Pisarenko timer_init_full(ts, &ctx->tlg, type, scale, attributes, cb, opaque); 5934e29e831SAlex Bligh } 5944e29e831SAlex Bligh 5954e29e831SAlex Bligh /** 5964e29e831SAlex Bligh * aio_timer_init: 5974e29e831SAlex Bligh * @ctx: the aio context 5984e29e831SAlex Bligh * @ts: the timer 5994e29e831SAlex Bligh * @type: the clock type 6004e29e831SAlex Bligh * @scale: the scale 6014e29e831SAlex Bligh * @cb: the callback to call on timer expiry 6024e29e831SAlex Bligh * @opaque: the opaque pointer to pass to the callback 6034e29e831SAlex Bligh * 6044e29e831SAlex Bligh * Initialise a new timer attached to the context @ctx. 60589a603a0SArtem Pisarenko * See aio_timer_init_with_attrs for details. 6064e29e831SAlex Bligh */ 6074e29e831SAlex Bligh static inline void aio_timer_init(AioContext *ctx, 6084e29e831SAlex Bligh QEMUTimer *ts, QEMUClockType type, 6094e29e831SAlex Bligh int scale, 6104e29e831SAlex Bligh QEMUTimerCB *cb, void *opaque) 6114e29e831SAlex Bligh { 61289a603a0SArtem Pisarenko timer_init_full(ts, &ctx->tlg, type, scale, 0, cb, opaque); 6134e29e831SAlex Bligh } 6144e29e831SAlex Bligh 615845ca10dSPaolo Bonzini /** 616845ca10dSPaolo Bonzini * aio_compute_timeout: 617845ca10dSPaolo Bonzini * @ctx: the aio context 618845ca10dSPaolo Bonzini * 619845ca10dSPaolo Bonzini * Compute the timeout that a blocking aio_poll should use. 620845ca10dSPaolo Bonzini */ 621845ca10dSPaolo Bonzini int64_t aio_compute_timeout(AioContext *ctx); 622845ca10dSPaolo Bonzini 623c1e1e5faSFam Zheng /** 624c1e1e5faSFam Zheng * aio_disable_external: 625c1e1e5faSFam Zheng * @ctx: the aio context 626c1e1e5faSFam Zheng * 627c1e1e5faSFam Zheng * Disable the further processing of external clients. 628c1e1e5faSFam Zheng */ 629c1e1e5faSFam Zheng static inline void aio_disable_external(AioContext *ctx) 630c1e1e5faSFam Zheng { 631d73415a3SStefan Hajnoczi qatomic_inc(&ctx->external_disable_cnt); 632c1e1e5faSFam Zheng } 633c1e1e5faSFam Zheng 634c1e1e5faSFam Zheng /** 635c1e1e5faSFam Zheng * aio_enable_external: 636c1e1e5faSFam Zheng * @ctx: the aio context 637c1e1e5faSFam Zheng * 638c1e1e5faSFam Zheng * Enable the processing of external clients. 639c1e1e5faSFam Zheng */ 640c1e1e5faSFam Zheng static inline void aio_enable_external(AioContext *ctx) 641c1e1e5faSFam Zheng { 642321d1dbaSStefan Hajnoczi int old; 643321d1dbaSStefan Hajnoczi 644d73415a3SStefan Hajnoczi old = qatomic_fetch_dec(&ctx->external_disable_cnt); 645321d1dbaSStefan Hajnoczi assert(old > 0); 646321d1dbaSStefan Hajnoczi if (old == 1) { 647321d1dbaSStefan Hajnoczi /* Kick event loop so it re-arms file descriptors */ 648321d1dbaSStefan Hajnoczi aio_notify(ctx); 649321d1dbaSStefan Hajnoczi } 650c1e1e5faSFam Zheng } 651c1e1e5faSFam Zheng 652c1e1e5faSFam Zheng /** 6535ceb9e39SFam Zheng * aio_external_disabled: 6545ceb9e39SFam Zheng * @ctx: the aio context 6555ceb9e39SFam Zheng * 6565ceb9e39SFam Zheng * Return true if the external clients are disabled. 6575ceb9e39SFam Zheng */ 6585ceb9e39SFam Zheng static inline bool aio_external_disabled(AioContext *ctx) 6595ceb9e39SFam Zheng { 660d73415a3SStefan Hajnoczi return qatomic_read(&ctx->external_disable_cnt); 6615ceb9e39SFam Zheng } 6625ceb9e39SFam Zheng 6635ceb9e39SFam Zheng /** 664c1e1e5faSFam Zheng * aio_node_check: 665c1e1e5faSFam Zheng * @ctx: the aio context 666c1e1e5faSFam Zheng * @is_external: Whether or not the checked node is an external event source. 667c1e1e5faSFam Zheng * 668c1e1e5faSFam Zheng * Check if the node's is_external flag is okay to be polled by the ctx at this 669c1e1e5faSFam Zheng * moment. True means green light. 670c1e1e5faSFam Zheng */ 671c1e1e5faSFam Zheng static inline bool aio_node_check(AioContext *ctx, bool is_external) 672c1e1e5faSFam Zheng { 673d73415a3SStefan Hajnoczi return !is_external || !qatomic_read(&ctx->external_disable_cnt); 674c1e1e5faSFam Zheng } 675c1e1e5faSFam Zheng 67637fcee5dSFam Zheng /** 6770c330a73SPaolo Bonzini * aio_co_schedule: 6780c330a73SPaolo Bonzini * @ctx: the aio context 6790c330a73SPaolo Bonzini * @co: the coroutine 6800c330a73SPaolo Bonzini * 6810c330a73SPaolo Bonzini * Start a coroutine on a remote AioContext. 6820c330a73SPaolo Bonzini * 6830c330a73SPaolo Bonzini * The coroutine must not be entered by anyone else while aio_co_schedule() 6840c330a73SPaolo Bonzini * is active. In addition the coroutine must have yielded unless ctx 6850c330a73SPaolo Bonzini * is the context in which the coroutine is running (i.e. the value of 6860c330a73SPaolo Bonzini * qemu_get_current_aio_context() from the coroutine itself). 6870c330a73SPaolo Bonzini */ 6880c330a73SPaolo Bonzini void aio_co_schedule(AioContext *ctx, struct Coroutine *co); 6890c330a73SPaolo Bonzini 6900c330a73SPaolo Bonzini /** 69126b0b698SKevin Wolf * aio_co_reschedule_self: 69226b0b698SKevin Wolf * @new_ctx: the new context 69326b0b698SKevin Wolf * 69426b0b698SKevin Wolf * Move the currently running coroutine to new_ctx. If the coroutine is already 69526b0b698SKevin Wolf * running in new_ctx, do nothing. 69626b0b698SKevin Wolf */ 69726b0b698SKevin Wolf void coroutine_fn aio_co_reschedule_self(AioContext *new_ctx); 69826b0b698SKevin Wolf 69926b0b698SKevin Wolf /** 7000c330a73SPaolo Bonzini * aio_co_wake: 7010c330a73SPaolo Bonzini * @co: the coroutine 7020c330a73SPaolo Bonzini * 7030c330a73SPaolo Bonzini * Restart a coroutine on the AioContext where it was running last, thus 7040c330a73SPaolo Bonzini * preventing coroutines from jumping from one context to another when they 7050c330a73SPaolo Bonzini * go to sleep. 7060c330a73SPaolo Bonzini * 7070c330a73SPaolo Bonzini * aio_co_wake may be executed either in coroutine or non-coroutine 7080c330a73SPaolo Bonzini * context. The coroutine must not be entered by anyone else while 7090c330a73SPaolo Bonzini * aio_co_wake() is active. 7100c330a73SPaolo Bonzini */ 7110c330a73SPaolo Bonzini void aio_co_wake(struct Coroutine *co); 7120c330a73SPaolo Bonzini 7130c330a73SPaolo Bonzini /** 7148865852eSFam Zheng * aio_co_enter: 7158865852eSFam Zheng * @ctx: the context to run the coroutine 7168865852eSFam Zheng * @co: the coroutine to run 7178865852eSFam Zheng * 7188865852eSFam Zheng * Enter a coroutine in the specified AioContext. 7198865852eSFam Zheng */ 7208865852eSFam Zheng void aio_co_enter(AioContext *ctx, struct Coroutine *co); 7218865852eSFam Zheng 7228865852eSFam Zheng /** 723e4370165SPaolo Bonzini * Return the AioContext whose event loop runs in the current thread. 724e4370165SPaolo Bonzini * 725e4370165SPaolo Bonzini * If called from an IOThread this will be the IOThread's AioContext. If 7265f50be9bSPaolo Bonzini * called from the main thread or with the "big QEMU lock" taken it 7275f50be9bSPaolo Bonzini * will be the main loop AioContext. 728e4370165SPaolo Bonzini */ 729e4370165SPaolo Bonzini AioContext *qemu_get_current_aio_context(void); 730e4370165SPaolo Bonzini 7315f50be9bSPaolo Bonzini void qemu_set_current_aio_context(AioContext *ctx); 7325f50be9bSPaolo Bonzini 733e4370165SPaolo Bonzini /** 73437fcee5dSFam Zheng * aio_context_setup: 73537fcee5dSFam Zheng * @ctx: the aio context 73637fcee5dSFam Zheng * 73737fcee5dSFam Zheng * Initialize the aio context. 73837fcee5dSFam Zheng */ 7397e003465SCao jin void aio_context_setup(AioContext *ctx); 74037fcee5dSFam Zheng 7414a1cba38SStefan Hajnoczi /** 742cd0a6d2bSJie Wang * aio_context_destroy: 743cd0a6d2bSJie Wang * @ctx: the aio context 744cd0a6d2bSJie Wang * 745cd0a6d2bSJie Wang * Destroy the aio context. 746cd0a6d2bSJie Wang */ 747cd0a6d2bSJie Wang void aio_context_destroy(AioContext *ctx); 748cd0a6d2bSJie Wang 749ba607ca8SStefan Hajnoczi /* Used internally, do not call outside AioContext code */ 750ba607ca8SStefan Hajnoczi void aio_context_use_g_source(AioContext *ctx); 751ba607ca8SStefan Hajnoczi 752cd0a6d2bSJie Wang /** 7534a1cba38SStefan Hajnoczi * aio_context_set_poll_params: 7544a1cba38SStefan Hajnoczi * @ctx: the aio context 7554a1cba38SStefan Hajnoczi * @max_ns: how long to busy poll for, in nanoseconds 75682a41186SStefan Hajnoczi * @grow: polling time growth factor 75782a41186SStefan Hajnoczi * @shrink: polling time shrink factor 7584a1cba38SStefan Hajnoczi * 7594a1cba38SStefan Hajnoczi * Poll mode can be disabled by setting poll_max_ns to 0. 7604a1cba38SStefan Hajnoczi */ 7614a1cba38SStefan Hajnoczi void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns, 76282a41186SStefan Hajnoczi int64_t grow, int64_t shrink, 7634a1cba38SStefan Hajnoczi Error **errp); 7644a1cba38SStefan Hajnoczi 7651793ad02SStefano Garzarella /** 7661793ad02SStefano Garzarella * aio_context_set_aio_params: 7671793ad02SStefano Garzarella * @ctx: the aio context 7681793ad02SStefano Garzarella * @max_batch: maximum number of requests in a batch, 0 means that the 7691793ad02SStefano Garzarella * engine will use its default 7701793ad02SStefano Garzarella */ 7711793ad02SStefano Garzarella void aio_context_set_aio_params(AioContext *ctx, int64_t max_batch, 7721793ad02SStefano Garzarella Error **errp); 7731793ad02SStefano Garzarella 774*71ad4713SNicolas Saenz Julienne /** 775*71ad4713SNicolas Saenz Julienne * aio_context_set_thread_pool_params: 776*71ad4713SNicolas Saenz Julienne * @ctx: the aio context 777*71ad4713SNicolas Saenz Julienne * @min: min number of threads to have readily available in the thread pool 778*71ad4713SNicolas Saenz Julienne * @min: max number of threads the thread pool can contain 779*71ad4713SNicolas Saenz Julienne */ 780*71ad4713SNicolas Saenz Julienne void aio_context_set_thread_pool_params(AioContext *ctx, int64_t min, 781*71ad4713SNicolas Saenz Julienne int64_t max, Error **errp); 782737e150eSPaolo Bonzini #endif 783