async.c

   1 /*
   2  * QEMU System Emulator
   3  *
   4  * Copyright (c) 2003-2008 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24
  25 #include "qemu-common.h"
  26 #include "block/aio.h"
  27 #include "block/thread-pool.h"
  28 #include "qemu/main-loop.h"
  29 #include "qemu/atomic.h"
  30
  31 /***********************************************************/
  32 /* bottom halves (can be seen as timers which expire ASAP) */
  33
  34 struct QEMUBH {
  35     AioContext *ctx;
  36     QEMUBHFunc *cb;
  37     void *opaque;
  38     QEMUBH *next;
  39     bool scheduled;
  40     bool idle;
  41     bool deleted;
  42 };
  43
  44 QEMUBH *aio_bh_new(AioContext *ctx, QEMUBHFunc *cb, void *opaque)
  45 {
  46     QEMUBH *bh;
  47     bh = g_new(QEMUBH, 1);
  48     *bh = (QEMUBH){
  49         .ctx = ctx,
  50         .cb = cb,
  51         .opaque = opaque,
  52     };
  53     qemu_mutex_lock(&ctx->bh_lock);
  54     bh->next = ctx->first_bh;
  55     /* Make sure that the members are ready before putting bh into list */
  56     smp_wmb();
  57     ctx->first_bh = bh;
  58     qemu_mutex_unlock(&ctx->bh_lock);
  59     return bh;
  60 }
  61
  62 /* Multiple occurrences of aio_bh_poll cannot be called concurrently */
  63 int aio_bh_poll(AioContext *ctx)
  64 {
  65     QEMUBH *bh, **bhp, *next;
  66     int ret;
  67
  68     ctx->walking_bh++;
  69
  70     ret = 0;
  71     for (bh = ctx->first_bh; bh; bh = next) {
  72         /* Make sure that fetching bh happens before accessing its members */
  73         smp_read_barrier_depends();
  74         next = bh->next;
  75         /* The atomic_xchg is paired with the one in qemu_bh_schedule.  The
  76          * implicit memory barrier ensures that the callback sees all writes
  77          * done by the scheduling thread.  It also ensures that the scheduling
  78          * thread sees the zero before bh->cb has run, and thus will call
  79          * aio_notify again if necessary.
  80          */
  81         if (!bh->deleted && atomic_xchg(&bh->scheduled, 0)) {
  82             /* Idle BHs and the notify BH don't count as progress */
  83             if (!bh->idle && bh != ctx->notify_dummy_bh) {
  84                 ret = 1;
  85             }
  86             bh->idle = 0;
  87             bh->cb(bh->opaque);
  88         }
  89     }
  90
  91     ctx->walking_bh--;
  92
  93     /* remove deleted bhs */
  94     if (!ctx->walking_bh) {
  95         qemu_mutex_lock(&ctx->bh_lock);
  96         bhp = &ctx->first_bh;
  97         while (*bhp) {
  98             bh = *bhp;
  99             if (bh->deleted) {
 100                 *bhp = bh->next;
 101                 g_free(bh);
 102             } else {
 103                 bhp = &bh->next;
 104             }
 105         }
 106         qemu_mutex_unlock(&ctx->bh_lock);
 107     }
 108
 109     return ret;
 110 }
 111
 112 void qemu_bh_schedule_idle(QEMUBH *bh)
 113 {
 114     bh->idle = 1;
 115     /* Make sure that idle & any writes needed by the callback are done
 116      * before the locations are read in the aio_bh_poll.
 117      */
 118     atomic_mb_set(&bh->scheduled, 1);
 119 }
 120
 121 void qemu_bh_schedule(QEMUBH *bh)
 122 {
 123     AioContext *ctx;
 124
 125     ctx = bh->ctx;
 126     bh->idle = 0;
 127     /* The memory barrier implicit in atomic_xchg makes sure that:
 128      * 1. idle & any writes needed by the callback are done before the
 129      *    locations are read in the aio_bh_poll.
 130      * 2. ctx is loaded before scheduled is set and the callback has a chance
 131      *    to execute.
 132      */
 133     if (atomic_xchg(&bh->scheduled, 1) == 0) {
 134         aio_notify(ctx);
 135     }
 136 }
 137
 138
 139 /* This func is async.
 140  */
 141 void qemu_bh_cancel(QEMUBH *bh)
 142 {
 143     bh->scheduled = 0;
 144 }
 145
 146 /* This func is async.The bottom half will do the delete action at the finial
 147  * end.
 148  */
 149 void qemu_bh_delete(QEMUBH *bh)
 150 {
 151     bh->scheduled = 0;
 152     bh->deleted = 1;
 153 }
 154
 155 int64_t
 156 aio_compute_timeout(AioContext *ctx)
 157 {
 158     int64_t deadline;
 159     int timeout = -1;
 160     QEMUBH *bh;
 161
 162     for (bh = ctx->first_bh; bh; bh = bh->next) {
 163         if (!bh->deleted && bh->scheduled) {
 164             if (bh->idle) {
 165                 /* idle bottom halves will be polled at least
 166                  * every 10ms */
 167                 timeout = 10000000;
 168             } else {
 169                 /* non-idle bottom halves will be executed
 170                  * immediately */
 171                 return 0;
 172             }
 173         }
 174     }
 175
 176     deadline = timerlistgroup_deadline_ns(&ctx->tlg);
 177     if (deadline == 0) {
 178         return 0;
 179     } else {
 180         return qemu_soonest_timeout(timeout, deadline);
 181     }
 182 }
 183
 184 static gboolean
 185 aio_ctx_prepare(GSource *source, gint    *timeout)
 186 {
 187     AioContext *ctx = (AioContext *) source;
 188
 189     atomic_or(&ctx->notify_me, 1);
 190
 191     /* We assume there is no timeout already supplied */
 192     *timeout = qemu_timeout_ns_to_ms(aio_compute_timeout(ctx));
 193
 194     if (aio_prepare(ctx)) {
 195         *timeout = 0;
 196     }
 197
 198     return *timeout == 0;
 199 }
 200
 201 static gboolean
 202 aio_ctx_check(GSource *source)
 203 {
 204     AioContext *ctx = (AioContext *) source;
 205     QEMUBH *bh;
 206
 207     atomic_and(&ctx->notify_me, ~1);
 208     aio_notify_accept(ctx);
 209
 210     for (bh = ctx->first_bh; bh; bh = bh->next) {
 211         if (!bh->deleted && bh->scheduled) {
 212             return true;
 213         }
 214     }
 215     return aio_pending(ctx) || (timerlistgroup_deadline_ns(&ctx->tlg) == 0);
 216 }
 217
 218 static gboolean
 219 aio_ctx_dispatch(GSource     *source,
 220                  GSourceFunc  callback,
 221                  gpointer     user_data)
 222 {
 223     AioContext *ctx = (AioContext *) source;
 224
 225     assert(callback == NULL);
 226     aio_dispatch(ctx);
 227     return true;
 228 }
 229
 230 static void
 231 aio_ctx_finalize(GSource     *source)
 232 {
 233     AioContext *ctx = (AioContext *) source;
 234
 235     qemu_bh_delete(ctx->notify_dummy_bh);
 236     thread_pool_free(ctx->thread_pool);
 237
 238     qemu_mutex_lock(&ctx->bh_lock);
 239     while (ctx->first_bh) {
 240         QEMUBH *next = ctx->first_bh->next;
 241
 242         /* qemu_bh_delete() must have been called on BHs in this AioContext */
 243         assert(ctx->first_bh->deleted);
 244
 245         g_free(ctx->first_bh);
 246         ctx->first_bh = next;
 247     }
 248     qemu_mutex_unlock(&ctx->bh_lock);
 249
 250     aio_set_event_notifier(ctx, &ctx->notifier, false, NULL);
 251     event_notifier_cleanup(&ctx->notifier);
 252     rfifolock_destroy(&ctx->lock);
 253     qemu_mutex_destroy(&ctx->bh_lock);
 254     timerlistgroup_deinit(&ctx->tlg);
 255 }
 256
 257 static GSourceFuncs aio_source_funcs = {
 258     aio_ctx_prepare,
 259     aio_ctx_check,
 260     aio_ctx_dispatch,
 261     aio_ctx_finalize
 262 };
 263
 264 GSource *aio_get_g_source(AioContext *ctx)
 265 {
 266     g_source_ref(&ctx->source);
 267     return &ctx->source;
 268 }
 269
 270 ThreadPool *aio_get_thread_pool(AioContext *ctx)
 271 {
 272     if (!ctx->thread_pool) {
 273         ctx->thread_pool = thread_pool_new(ctx);
 274     }
 275     return ctx->thread_pool;
 276 }
 277
 278 void aio_notify(AioContext *ctx)
 279 {
 280     /* Write e.g. bh->scheduled before reading ctx->notify_me.  Pairs
 281      * with atomic_or in aio_ctx_prepare or atomic_add in aio_poll.
 282      */
 283     smp_mb();
 284     if (ctx->notify_me) {
 285         event_notifier_set(&ctx->notifier);
 286         atomic_mb_set(&ctx->notified, true);
 287     }
 288 }
 289
 290 void aio_notify_accept(AioContext *ctx)
 291 {
 292     if (atomic_xchg(&ctx->notified, false)) {
 293         event_notifier_test_and_clear(&ctx->notifier);
 294     }
 295 }
 296
 297 static void aio_timerlist_notify(void *opaque)
 298 {
 299     aio_notify(opaque);
 300 }
 301
 302 static void aio_rfifolock_cb(void *opaque)
 303 {
 304     AioContext *ctx = opaque;
 305
 306     /* Kick owner thread in case they are blocked in aio_poll() */
 307     qemu_bh_schedule(ctx->notify_dummy_bh);
 308 }
 309
 310 static void notify_dummy_bh(void *opaque)
 311 {
 312     /* Do nothing, we were invoked just to force the event loop to iterate */
 313 }
 314
 315 static void event_notifier_dummy_cb(EventNotifier *e)
 316 {
 317 }
 318
 319 AioContext *aio_context_new(Error **errp)
 320 {
 321     int ret;
 322     AioContext *ctx;
 323     ctx = (AioContext *) g_source_new(&aio_source_funcs, sizeof(AioContext));
 324     ret = event_notifier_init(&ctx->notifier, false);
 325     if (ret < 0) {
 326         g_source_destroy(&ctx->source);
 327         error_setg_errno(errp, -ret, "Failed to initialize event notifier");
 328         return NULL;
 329     }
 330     g_source_set_can_recurse(&ctx->source, true);
 331     aio_set_event_notifier(ctx, &ctx->notifier,
 332                            false,
 333                            (EventNotifierHandler *)
 334                            event_notifier_dummy_cb);
 335     ctx->thread_pool = NULL;
 336     qemu_mutex_init(&ctx->bh_lock);
 337     rfifolock_init(&ctx->lock, aio_rfifolock_cb, ctx);
 338     timerlistgroup_init(&ctx->tlg, aio_timerlist_notify, ctx);
 339
 340     ctx->notify_dummy_bh = aio_bh_new(ctx, notify_dummy_bh, NULL);
 341
 342     return ctx;
 343 }
 344
 345 void aio_context_ref(AioContext *ctx)
 346 {
 347     g_source_ref(&ctx->source);
 348 }
 349
 350 void aio_context_unref(AioContext *ctx)
 351 {
 352     g_source_unref(&ctx->source);
 353 }
 354
 355 void aio_context_acquire(AioContext *ctx)
 356 {
 357     rfifolock_lock(&ctx->lock);
 358 }
 359
 360 void aio_context_release(AioContext *ctx)
 361 {
 362     rfifolock_unlock(&ctx->lock);
 363 }