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 17737e150eSPaolo Bonzini #include "qemu-common.h" 181de7afc9SPaolo Bonzini #include "qemu/queue.h" 191de7afc9SPaolo Bonzini #include "qemu/event_notifier.h" 20*dcc772e2SLiu Ping Fan #include "qemu/thread.h" 21737e150eSPaolo Bonzini 22737e150eSPaolo Bonzini typedef struct BlockDriverAIOCB BlockDriverAIOCB; 23737e150eSPaolo Bonzini typedef void BlockDriverCompletionFunc(void *opaque, int ret); 24737e150eSPaolo Bonzini 25737e150eSPaolo Bonzini typedef struct AIOCBInfo { 26737e150eSPaolo Bonzini void (*cancel)(BlockDriverAIOCB *acb); 27737e150eSPaolo Bonzini size_t aiocb_size; 28737e150eSPaolo Bonzini } AIOCBInfo; 29737e150eSPaolo Bonzini 30737e150eSPaolo Bonzini struct BlockDriverAIOCB { 31737e150eSPaolo Bonzini const AIOCBInfo *aiocb_info; 32737e150eSPaolo Bonzini BlockDriverState *bs; 33737e150eSPaolo Bonzini BlockDriverCompletionFunc *cb; 34737e150eSPaolo Bonzini void *opaque; 35737e150eSPaolo Bonzini }; 36737e150eSPaolo Bonzini 37737e150eSPaolo Bonzini void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs, 38737e150eSPaolo Bonzini BlockDriverCompletionFunc *cb, void *opaque); 39737e150eSPaolo Bonzini void qemu_aio_release(void *p); 40737e150eSPaolo Bonzini 41737e150eSPaolo Bonzini typedef struct AioHandler AioHandler; 42737e150eSPaolo Bonzini typedef void QEMUBHFunc(void *opaque); 43737e150eSPaolo Bonzini typedef void IOHandler(void *opaque); 44737e150eSPaolo Bonzini 45737e150eSPaolo Bonzini typedef struct AioContext { 46737e150eSPaolo Bonzini GSource source; 47737e150eSPaolo Bonzini 48737e150eSPaolo Bonzini /* The list of registered AIO handlers */ 49737e150eSPaolo Bonzini QLIST_HEAD(, AioHandler) aio_handlers; 50737e150eSPaolo Bonzini 51737e150eSPaolo Bonzini /* This is a simple lock used to protect the aio_handlers list. 52737e150eSPaolo Bonzini * Specifically, it's used to ensure that no callbacks are removed while 53737e150eSPaolo Bonzini * we're walking and dispatching callbacks. 54737e150eSPaolo Bonzini */ 55737e150eSPaolo Bonzini int walking_handlers; 56737e150eSPaolo Bonzini 57*dcc772e2SLiu Ping Fan /* lock to protect between bh's adders and deleter */ 58*dcc772e2SLiu Ping Fan QemuMutex bh_lock; 59737e150eSPaolo Bonzini /* Anchor of the list of Bottom Halves belonging to the context */ 60737e150eSPaolo Bonzini struct QEMUBH *first_bh; 61737e150eSPaolo Bonzini 62737e150eSPaolo Bonzini /* A simple lock used to protect the first_bh list, and ensure that 63737e150eSPaolo Bonzini * no callbacks are removed while we're walking and dispatching callbacks. 64737e150eSPaolo Bonzini */ 65737e150eSPaolo Bonzini int walking_bh; 66737e150eSPaolo Bonzini 67737e150eSPaolo Bonzini /* Used for aio_notify. */ 68737e150eSPaolo Bonzini EventNotifier notifier; 696b5f8762SStefan Hajnoczi 706b5f8762SStefan Hajnoczi /* GPollFDs for aio_poll() */ 716b5f8762SStefan Hajnoczi GArray *pollfds; 729b34277dSStefan Hajnoczi 739b34277dSStefan Hajnoczi /* Thread pool for performing work and receiving completion callbacks */ 749b34277dSStefan Hajnoczi struct ThreadPool *thread_pool; 75737e150eSPaolo Bonzini } AioContext; 76737e150eSPaolo Bonzini 77737e150eSPaolo Bonzini /* Returns 1 if there are still outstanding AIO requests; 0 otherwise */ 78737e150eSPaolo Bonzini typedef int (AioFlushEventNotifierHandler)(EventNotifier *e); 79737e150eSPaolo Bonzini 80737e150eSPaolo Bonzini /** 81737e150eSPaolo Bonzini * aio_context_new: Allocate a new AioContext. 82737e150eSPaolo Bonzini * 83737e150eSPaolo Bonzini * AioContext provide a mini event-loop that can be waited on synchronously. 84737e150eSPaolo Bonzini * They also provide bottom halves, a service to execute a piece of code 85737e150eSPaolo Bonzini * as soon as possible. 86737e150eSPaolo Bonzini */ 87737e150eSPaolo Bonzini AioContext *aio_context_new(void); 88737e150eSPaolo Bonzini 89737e150eSPaolo Bonzini /** 90737e150eSPaolo Bonzini * aio_context_ref: 91737e150eSPaolo Bonzini * @ctx: The AioContext to operate on. 92737e150eSPaolo Bonzini * 93737e150eSPaolo Bonzini * Add a reference to an AioContext. 94737e150eSPaolo Bonzini */ 95737e150eSPaolo Bonzini void aio_context_ref(AioContext *ctx); 96737e150eSPaolo Bonzini 97737e150eSPaolo Bonzini /** 98737e150eSPaolo Bonzini * aio_context_unref: 99737e150eSPaolo Bonzini * @ctx: The AioContext to operate on. 100737e150eSPaolo Bonzini * 101737e150eSPaolo Bonzini * Drop a reference to an AioContext. 102737e150eSPaolo Bonzini */ 103737e150eSPaolo Bonzini void aio_context_unref(AioContext *ctx); 104737e150eSPaolo Bonzini 105737e150eSPaolo Bonzini /** 106737e150eSPaolo Bonzini * aio_bh_new: Allocate a new bottom half structure. 107737e150eSPaolo Bonzini * 108737e150eSPaolo Bonzini * Bottom halves are lightweight callbacks whose invocation is guaranteed 109737e150eSPaolo Bonzini * to be wait-free, thread-safe and signal-safe. The #QEMUBH structure 110737e150eSPaolo Bonzini * is opaque and must be allocated prior to its use. 111737e150eSPaolo Bonzini */ 112737e150eSPaolo Bonzini QEMUBH *aio_bh_new(AioContext *ctx, QEMUBHFunc *cb, void *opaque); 113737e150eSPaolo Bonzini 114737e150eSPaolo Bonzini /** 115737e150eSPaolo Bonzini * aio_notify: Force processing of pending events. 116737e150eSPaolo Bonzini * 117737e150eSPaolo Bonzini * Similar to signaling a condition variable, aio_notify forces 118737e150eSPaolo Bonzini * aio_wait to exit, so that the next call will re-examine pending events. 119737e150eSPaolo Bonzini * The caller of aio_notify will usually call aio_wait again very soon, 120737e150eSPaolo Bonzini * or go through another iteration of the GLib main loop. Hence, aio_notify 121737e150eSPaolo Bonzini * also has the side effect of recalculating the sets of file descriptors 122737e150eSPaolo Bonzini * that the main loop waits for. 123737e150eSPaolo Bonzini * 124737e150eSPaolo Bonzini * Calling aio_notify is rarely necessary, because for example scheduling 125737e150eSPaolo Bonzini * a bottom half calls it already. 126737e150eSPaolo Bonzini */ 127737e150eSPaolo Bonzini void aio_notify(AioContext *ctx); 128737e150eSPaolo Bonzini 129737e150eSPaolo Bonzini /** 130737e150eSPaolo Bonzini * aio_bh_poll: Poll bottom halves for an AioContext. 131737e150eSPaolo Bonzini * 132737e150eSPaolo Bonzini * These are internal functions used by the QEMU main loop. 133*dcc772e2SLiu Ping Fan * And notice that multiple occurrences of aio_bh_poll cannot 134*dcc772e2SLiu Ping Fan * be called concurrently 135737e150eSPaolo Bonzini */ 136737e150eSPaolo Bonzini int aio_bh_poll(AioContext *ctx); 137737e150eSPaolo Bonzini 138737e150eSPaolo Bonzini /** 139737e150eSPaolo Bonzini * qemu_bh_schedule: Schedule a bottom half. 140737e150eSPaolo Bonzini * 141737e150eSPaolo Bonzini * Scheduling a bottom half interrupts the main loop and causes the 142737e150eSPaolo Bonzini * execution of the callback that was passed to qemu_bh_new. 143737e150eSPaolo Bonzini * 144737e150eSPaolo Bonzini * Bottom halves that are scheduled from a bottom half handler are instantly 145737e150eSPaolo Bonzini * invoked. This can create an infinite loop if a bottom half handler 146737e150eSPaolo Bonzini * schedules itself. 147737e150eSPaolo Bonzini * 148737e150eSPaolo Bonzini * @bh: The bottom half to be scheduled. 149737e150eSPaolo Bonzini */ 150737e150eSPaolo Bonzini void qemu_bh_schedule(QEMUBH *bh); 151737e150eSPaolo Bonzini 152737e150eSPaolo Bonzini /** 153737e150eSPaolo Bonzini * qemu_bh_cancel: Cancel execution of a bottom half. 154737e150eSPaolo Bonzini * 155737e150eSPaolo Bonzini * Canceling execution of a bottom half undoes the effect of calls to 156737e150eSPaolo Bonzini * qemu_bh_schedule without freeing its resources yet. While cancellation 157737e150eSPaolo Bonzini * itself is also wait-free and thread-safe, it can of course race with the 158737e150eSPaolo Bonzini * loop that executes bottom halves unless you are holding the iothread 159737e150eSPaolo Bonzini * mutex. This makes it mostly useless if you are not holding the mutex. 160737e150eSPaolo Bonzini * 161737e150eSPaolo Bonzini * @bh: The bottom half to be canceled. 162737e150eSPaolo Bonzini */ 163737e150eSPaolo Bonzini void qemu_bh_cancel(QEMUBH *bh); 164737e150eSPaolo Bonzini 165737e150eSPaolo Bonzini /** 166737e150eSPaolo Bonzini *qemu_bh_delete: Cancel execution of a bottom half and free its resources. 167737e150eSPaolo Bonzini * 168737e150eSPaolo Bonzini * Deleting a bottom half frees the memory that was allocated for it by 169737e150eSPaolo Bonzini * qemu_bh_new. It also implies canceling the bottom half if it was 170737e150eSPaolo Bonzini * scheduled. 171*dcc772e2SLiu Ping Fan * This func is async. The bottom half will do the delete action at the finial 172*dcc772e2SLiu Ping Fan * end. 173737e150eSPaolo Bonzini * 174737e150eSPaolo Bonzini * @bh: The bottom half to be deleted. 175737e150eSPaolo Bonzini */ 176737e150eSPaolo Bonzini void qemu_bh_delete(QEMUBH *bh); 177737e150eSPaolo Bonzini 178737e150eSPaolo Bonzini /* Return whether there are any pending callbacks from the GSource 179737e150eSPaolo Bonzini * attached to the AioContext. 180737e150eSPaolo Bonzini * 181737e150eSPaolo Bonzini * This is used internally in the implementation of the GSource. 182737e150eSPaolo Bonzini */ 183737e150eSPaolo Bonzini bool aio_pending(AioContext *ctx); 184737e150eSPaolo Bonzini 185737e150eSPaolo Bonzini /* Progress in completing AIO work to occur. This can issue new pending 186737e150eSPaolo Bonzini * aio as a result of executing I/O completion or bh callbacks. 187737e150eSPaolo Bonzini * 188737e150eSPaolo Bonzini * If there is no pending AIO operation or completion (bottom half), 1892ea9b58fSKevin Wolf * return false. If there are pending AIO operations of bottom halves, 1902ea9b58fSKevin Wolf * return true. 191737e150eSPaolo Bonzini * 192737e150eSPaolo Bonzini * If there are no pending bottom halves, but there are pending AIO 193737e150eSPaolo Bonzini * operations, it may not be possible to make any progress without 194737e150eSPaolo Bonzini * blocking. If @blocking is true, this function will wait until one 195737e150eSPaolo Bonzini * or more AIO events have completed, to ensure something has moved 196737e150eSPaolo Bonzini * before returning. 197737e150eSPaolo Bonzini */ 198737e150eSPaolo Bonzini bool aio_poll(AioContext *ctx, bool blocking); 199737e150eSPaolo Bonzini 200737e150eSPaolo Bonzini #ifdef CONFIG_POSIX 201737e150eSPaolo Bonzini /* Returns 1 if there are still outstanding AIO requests; 0 otherwise */ 202737e150eSPaolo Bonzini typedef int (AioFlushHandler)(void *opaque); 203737e150eSPaolo Bonzini 204737e150eSPaolo Bonzini /* Register a file descriptor and associated callbacks. Behaves very similarly 205737e150eSPaolo Bonzini * to qemu_set_fd_handler2. Unlike qemu_set_fd_handler2, these callbacks will 206737e150eSPaolo Bonzini * be invoked when using qemu_aio_wait(). 207737e150eSPaolo Bonzini * 208737e150eSPaolo Bonzini * Code that invokes AIO completion functions should rely on this function 209737e150eSPaolo Bonzini * instead of qemu_set_fd_handler[2]. 210737e150eSPaolo Bonzini */ 211737e150eSPaolo Bonzini void aio_set_fd_handler(AioContext *ctx, 212737e150eSPaolo Bonzini int fd, 213737e150eSPaolo Bonzini IOHandler *io_read, 214737e150eSPaolo Bonzini IOHandler *io_write, 215737e150eSPaolo Bonzini AioFlushHandler *io_flush, 216737e150eSPaolo Bonzini void *opaque); 217737e150eSPaolo Bonzini #endif 218737e150eSPaolo Bonzini 219737e150eSPaolo Bonzini /* Register an event notifier and associated callbacks. Behaves very similarly 220737e150eSPaolo Bonzini * to event_notifier_set_handler. Unlike event_notifier_set_handler, these callbacks 221737e150eSPaolo Bonzini * will be invoked when using qemu_aio_wait(). 222737e150eSPaolo Bonzini * 223737e150eSPaolo Bonzini * Code that invokes AIO completion functions should rely on this function 224737e150eSPaolo Bonzini * instead of event_notifier_set_handler. 225737e150eSPaolo Bonzini */ 226737e150eSPaolo Bonzini void aio_set_event_notifier(AioContext *ctx, 227737e150eSPaolo Bonzini EventNotifier *notifier, 228737e150eSPaolo Bonzini EventNotifierHandler *io_read, 229737e150eSPaolo Bonzini AioFlushEventNotifierHandler *io_flush); 230737e150eSPaolo Bonzini 231737e150eSPaolo Bonzini /* Return a GSource that lets the main loop poll the file descriptors attached 232737e150eSPaolo Bonzini * to this AioContext. 233737e150eSPaolo Bonzini */ 234737e150eSPaolo Bonzini GSource *aio_get_g_source(AioContext *ctx); 235737e150eSPaolo Bonzini 2369b34277dSStefan Hajnoczi /* Return the ThreadPool bound to this AioContext */ 2379b34277dSStefan Hajnoczi struct ThreadPool *aio_get_thread_pool(AioContext *ctx); 2389b34277dSStefan Hajnoczi 239737e150eSPaolo Bonzini /* Functions to operate on the main QEMU AioContext. */ 240737e150eSPaolo Bonzini 241737e150eSPaolo Bonzini bool qemu_aio_wait(void); 242737e150eSPaolo Bonzini void qemu_aio_set_event_notifier(EventNotifier *notifier, 243737e150eSPaolo Bonzini EventNotifierHandler *io_read, 244737e150eSPaolo Bonzini AioFlushEventNotifierHandler *io_flush); 245737e150eSPaolo Bonzini 246737e150eSPaolo Bonzini #ifdef CONFIG_POSIX 247737e150eSPaolo Bonzini void qemu_aio_set_fd_handler(int fd, 248737e150eSPaolo Bonzini IOHandler *io_read, 249737e150eSPaolo Bonzini IOHandler *io_write, 250737e150eSPaolo Bonzini AioFlushHandler *io_flush, 251737e150eSPaolo Bonzini void *opaque); 252737e150eSPaolo Bonzini #endif 253737e150eSPaolo Bonzini 254737e150eSPaolo Bonzini #endif 255