machine: Drop use of DEFAULT_RAM_SIZE in help text
[qemu/cris-port.git] / block / linux-aio.c
blobc991443c5d5cfbb0c1edb468c9090e76f67a3572
1 /*
2 * Linux native AIO support.
4 * Copyright (C) 2009 IBM, Corp.
5 * Copyright (C) 2009 Red Hat, Inc.
7 * This work is licensed under the terms of the GNU GPL, version 2 or later.
8 * See the COPYING file in the top-level directory.
9 */
10 #include "qemu-common.h"
11 #include "block/aio.h"
12 #include "qemu/queue.h"
13 #include "block/raw-aio.h"
14 #include "qemu/event_notifier.h"
16 #include <libaio.h>
19 * Queue size (per-device).
21 * XXX: eventually we need to communicate this to the guest and/or make it
22 * tunable by the guest. If we get more outstanding requests at a time
23 * than this we will get EAGAIN from io_submit which is communicated to
24 * the guest as an I/O error.
26 #define MAX_EVENTS 128
28 #define MAX_QUEUED_IO 128
30 struct qemu_laiocb {
31 BlockAIOCB common;
32 struct qemu_laio_state *ctx;
33 struct iocb iocb;
34 ssize_t ret;
35 size_t nbytes;
36 QEMUIOVector *qiov;
37 bool is_read;
38 QSIMPLEQ_ENTRY(qemu_laiocb) next;
41 typedef struct {
42 int plugged;
43 unsigned int n;
44 bool blocked;
45 QSIMPLEQ_HEAD(, qemu_laiocb) pending;
46 } LaioQueue;
48 struct qemu_laio_state {
49 io_context_t ctx;
50 EventNotifier e;
52 /* io queue for submit at batch */
53 LaioQueue io_q;
55 /* I/O completion processing */
56 QEMUBH *completion_bh;
57 struct io_event events[MAX_EVENTS];
58 int event_idx;
59 int event_max;
62 static void ioq_submit(struct qemu_laio_state *s);
64 static inline ssize_t io_event_ret(struct io_event *ev)
66 return (ssize_t)(((uint64_t)ev->res2 << 32) | ev->res);
70 * Completes an AIO request (calls the callback and frees the ACB).
72 static void qemu_laio_process_completion(struct qemu_laio_state *s,
73 struct qemu_laiocb *laiocb)
75 int ret;
77 ret = laiocb->ret;
78 if (ret != -ECANCELED) {
79 if (ret == laiocb->nbytes) {
80 ret = 0;
81 } else if (ret >= 0) {
82 /* Short reads mean EOF, pad with zeros. */
83 if (laiocb->is_read) {
84 qemu_iovec_memset(laiocb->qiov, ret, 0,
85 laiocb->qiov->size - ret);
86 } else {
87 ret = -EINVAL;
91 laiocb->common.cb(laiocb->common.opaque, ret);
93 qemu_aio_unref(laiocb);
96 /* The completion BH fetches completed I/O requests and invokes their
97 * callbacks.
99 * The function is somewhat tricky because it supports nested event loops, for
100 * example when a request callback invokes aio_poll(). In order to do this,
101 * the completion events array and index are kept in qemu_laio_state. The BH
102 * reschedules itself as long as there are completions pending so it will
103 * either be called again in a nested event loop or will be called after all
104 * events have been completed. When there are no events left to complete, the
105 * BH returns without rescheduling.
107 static void qemu_laio_completion_bh(void *opaque)
109 struct qemu_laio_state *s = opaque;
111 /* Fetch more completion events when empty */
112 if (s->event_idx == s->event_max) {
113 do {
114 struct timespec ts = { 0 };
115 s->event_max = io_getevents(s->ctx, MAX_EVENTS, MAX_EVENTS,
116 s->events, &ts);
117 } while (s->event_max == -EINTR);
119 s->event_idx = 0;
120 if (s->event_max <= 0) {
121 s->event_max = 0;
122 return; /* no more events */
126 /* Reschedule so nested event loops see currently pending completions */
127 qemu_bh_schedule(s->completion_bh);
129 /* Process completion events */
130 while (s->event_idx < s->event_max) {
131 struct iocb *iocb = s->events[s->event_idx].obj;
132 struct qemu_laiocb *laiocb =
133 container_of(iocb, struct qemu_laiocb, iocb);
135 laiocb->ret = io_event_ret(&s->events[s->event_idx]);
136 s->event_idx++;
138 qemu_laio_process_completion(s, laiocb);
141 if (!s->io_q.plugged && !QSIMPLEQ_EMPTY(&s->io_q.pending)) {
142 ioq_submit(s);
146 static void qemu_laio_completion_cb(EventNotifier *e)
148 struct qemu_laio_state *s = container_of(e, struct qemu_laio_state, e);
150 if (event_notifier_test_and_clear(&s->e)) {
151 qemu_bh_schedule(s->completion_bh);
155 static void laio_cancel(BlockAIOCB *blockacb)
157 struct qemu_laiocb *laiocb = (struct qemu_laiocb *)blockacb;
158 struct io_event event;
159 int ret;
161 if (laiocb->ret != -EINPROGRESS) {
162 return;
164 ret = io_cancel(laiocb->ctx->ctx, &laiocb->iocb, &event);
165 laiocb->ret = -ECANCELED;
166 if (ret != 0) {
167 /* iocb is not cancelled, cb will be called by the event loop later */
168 return;
171 laiocb->common.cb(laiocb->common.opaque, laiocb->ret);
174 static const AIOCBInfo laio_aiocb_info = {
175 .aiocb_size = sizeof(struct qemu_laiocb),
176 .cancel_async = laio_cancel,
179 static void ioq_init(LaioQueue *io_q)
181 QSIMPLEQ_INIT(&io_q->pending);
182 io_q->plugged = 0;
183 io_q->n = 0;
184 io_q->blocked = false;
187 static void ioq_submit(struct qemu_laio_state *s)
189 int ret, len;
190 struct qemu_laiocb *aiocb;
191 struct iocb *iocbs[MAX_QUEUED_IO];
192 QSIMPLEQ_HEAD(, qemu_laiocb) completed;
194 do {
195 len = 0;
196 QSIMPLEQ_FOREACH(aiocb, &s->io_q.pending, next) {
197 iocbs[len++] = &aiocb->iocb;
198 if (len == MAX_QUEUED_IO) {
199 break;
203 ret = io_submit(s->ctx, len, iocbs);
204 if (ret == -EAGAIN) {
205 break;
207 if (ret < 0) {
208 abort();
211 s->io_q.n -= ret;
212 aiocb = container_of(iocbs[ret - 1], struct qemu_laiocb, iocb);
213 QSIMPLEQ_SPLIT_AFTER(&s->io_q.pending, aiocb, next, &completed);
214 } while (ret == len && !QSIMPLEQ_EMPTY(&s->io_q.pending));
215 s->io_q.blocked = (s->io_q.n > 0);
218 void laio_io_plug(BlockDriverState *bs, void *aio_ctx)
220 struct qemu_laio_state *s = aio_ctx;
222 s->io_q.plugged++;
225 void laio_io_unplug(BlockDriverState *bs, void *aio_ctx, bool unplug)
227 struct qemu_laio_state *s = aio_ctx;
229 assert(s->io_q.plugged > 0 || !unplug);
231 if (unplug && --s->io_q.plugged > 0) {
232 return;
235 if (!s->io_q.blocked && !QSIMPLEQ_EMPTY(&s->io_q.pending)) {
236 ioq_submit(s);
240 BlockAIOCB *laio_submit(BlockDriverState *bs, void *aio_ctx, int fd,
241 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
242 BlockCompletionFunc *cb, void *opaque, int type)
244 struct qemu_laio_state *s = aio_ctx;
245 struct qemu_laiocb *laiocb;
246 struct iocb *iocbs;
247 off_t offset = sector_num * 512;
249 laiocb = qemu_aio_get(&laio_aiocb_info, bs, cb, opaque);
250 laiocb->nbytes = nb_sectors * 512;
251 laiocb->ctx = s;
252 laiocb->ret = -EINPROGRESS;
253 laiocb->is_read = (type == QEMU_AIO_READ);
254 laiocb->qiov = qiov;
256 iocbs = &laiocb->iocb;
258 switch (type) {
259 case QEMU_AIO_WRITE:
260 io_prep_pwritev(iocbs, fd, qiov->iov, qiov->niov, offset);
261 break;
262 case QEMU_AIO_READ:
263 io_prep_preadv(iocbs, fd, qiov->iov, qiov->niov, offset);
264 break;
265 /* Currently Linux kernel does not support other operations */
266 default:
267 fprintf(stderr, "%s: invalid AIO request type 0x%x.\n",
268 __func__, type);
269 goto out_free_aiocb;
271 io_set_eventfd(&laiocb->iocb, event_notifier_get_fd(&s->e));
273 QSIMPLEQ_INSERT_TAIL(&s->io_q.pending, laiocb, next);
274 s->io_q.n++;
275 if (!s->io_q.blocked &&
276 (!s->io_q.plugged || s->io_q.n >= MAX_QUEUED_IO)) {
277 ioq_submit(s);
279 return &laiocb->common;
281 out_free_aiocb:
282 qemu_aio_unref(laiocb);
283 return NULL;
286 void laio_detach_aio_context(void *s_, AioContext *old_context)
288 struct qemu_laio_state *s = s_;
290 aio_set_event_notifier(old_context, &s->e, NULL);
291 qemu_bh_delete(s->completion_bh);
294 void laio_attach_aio_context(void *s_, AioContext *new_context)
296 struct qemu_laio_state *s = s_;
298 s->completion_bh = aio_bh_new(new_context, qemu_laio_completion_bh, s);
299 aio_set_event_notifier(new_context, &s->e, qemu_laio_completion_cb);
302 void *laio_init(void)
304 struct qemu_laio_state *s;
306 s = g_malloc0(sizeof(*s));
307 if (event_notifier_init(&s->e, false) < 0) {
308 goto out_free_state;
311 if (io_setup(MAX_EVENTS, &s->ctx) != 0) {
312 goto out_close_efd;
315 ioq_init(&s->io_q);
317 return s;
319 out_close_efd:
320 event_notifier_cleanup(&s->e);
321 out_free_state:
322 g_free(s);
323 return NULL;
326 void laio_cleanup(void *s_)
328 struct qemu_laio_state *s = s_;
330 event_notifier_cleanup(&s->e);
332 if (io_destroy(s->ctx) != 0) {
333 fprintf(stderr, "%s: destroy AIO context %p failed\n",
334 __func__, &s->ctx);
336 g_free(s);