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/osdep.h"
  26 #include "qemu-common.h"
  27 #include "block/aio.h"
  28 #include "block/thread-pool.h"
  29 #include "qemu/main-loop.h"
  30 #include "qemu/atomic.h"
  31
  32 /***********************************************************/
  33 /* bottom halves (can be seen as timers which expire ASAP) */
  34
  35 struct QEMUBH {
  36     AioContext *ctx;
  37     QEMUBHFunc *cb;
  38     void *opaque;
  39     QEMUBH *next;
  40     bool scheduled;
  41     bool idle;
  42     bool deleted;
  43 };
  44
  45 QEMUBH *aio_bh_new(AioContext *ctx, QEMUBHFunc *cb, void *opaque)
  46 {
  47     QEMUBH *bh;
  48     bh = g_new(QEMUBH, 1);
  49     *bh = (QEMUBH){
  50         .ctx = ctx,
  51         .cb = cb,
  52         .opaque = opaque,
  53     };
  54     qemu_mutex_lock(&ctx->bh_lock);
  55     bh->next = ctx->first_bh;
  56     /* Make sure that the members are ready before putting bh into list */
  57     smp_wmb();
  58     ctx->first_bh = bh;
  59     qemu_mutex_unlock(&ctx->bh_lock);
  60     return bh;
  61 }
  62
  63 void aio_bh_call(QEMUBH *bh)
  64 {
  65     bh->cb(bh->opaque);
  66 }
  67
  68 /* Multiple occurrences of aio_bh_poll cannot be called concurrently */
  69 int aio_bh_poll(AioContext *ctx)
  70 {
  71     QEMUBH *bh, **bhp, *next;
  72     int ret;
  73
  74     ctx->walking_bh++;
  75
  76     ret = 0;
  77     for (bh = ctx->first_bh; bh; bh = next) {
  78         /* Make sure that fetching bh happens before accessing its members */
  79         smp_read_barrier_depends();
  80         next = bh->next;
  81         /* The atomic_xchg is paired with the one in qemu_bh_schedule.  The
  82          * implicit memory barrier ensures that the callback sees all writes
  83          * done by the scheduling thread.  It also ensures that the scheduling
  84          * thread sees the zero before bh->cb has run, and thus will call
  85          * aio_notify again if necessary.
  86          */
  87         if (!bh->deleted && atomic_xchg(&bh->scheduled, 0)) {
  88             /* Idle BHs and the notify BH don't count as progress */
  89             if (!bh->idle && bh != ctx->notify_dummy_bh) {
  90                 ret = 1;
  91             }
  92             bh->idle = 0;
  93             aio_bh_call(bh);
  94         }
  95     }
  96
  97     ctx->walking_bh--;
  98
  99     /* remove deleted bhs */
 100     if (!ctx->walking_bh) {
 101         qemu_mutex_lock(&ctx->bh_lock);
 102         bhp = &ctx->first_bh;
 103         while (*bhp) {
 104             bh = *bhp;
 105             if (bh->deleted) {
 106                 *bhp = bh->next;
 107                 g_free(bh);
 108             } else {
 109                 bhp = &bh->next;
 110             }
 111         }
 112         qemu_mutex_unlock(&ctx->bh_lock);
 113     }
 114
 115     return ret;
 116 }
 117
 118 void qemu_bh_schedule_idle(QEMUBH *bh)
 119 {
 120     bh->idle = 1;
 121     /* Make sure that idle & any writes needed by the callback are done
 122      * before the locations are read in the aio_bh_poll.
 123      */
 124     atomic_mb_set(&bh->scheduled, 1);
 125 }
 126
 127 void qemu_bh_schedule(QEMUBH *bh)
 128 {
 129     AioContext *ctx;
 130
 131     ctx = bh->ctx;
 132     bh->idle = 0;
 133     /* The memory barrier implicit in atomic_xchg makes sure that:
 134      * 1. idle & any writes needed by the callback are done before the
 135      *    locations are read in the aio_bh_poll.
 136      * 2. ctx is loaded before scheduled is set and the callback has a chance
 137      *    to execute.
 138      */
 139     if (atomic_xchg(&bh->scheduled, 1) == 0) {
 140         aio_notify(ctx);
 141     }
 142 }
 143
 144
 145 /* This func is async.
 146  */
 147 void qemu_bh_cancel(QEMUBH *bh)
 148 {
 149     bh->scheduled = 0;
 150 }
 151
 152 /* This func is async.The bottom half will do the delete action at the finial
 153  * end.
 154  */
 155 void qemu_bh_delete(QEMUBH *bh)
 156 {
 157     bh->scheduled = 0;
 158     bh->deleted = 1;
 159 }
 160
 161 int64_t
 162 aio_compute_timeout(AioContext *ctx)
 163 {
 164     int64_t deadline;
 165     int timeout = -1;
 166     QEMUBH *bh;
 167
 168     for (bh = ctx->first_bh; bh; bh = bh->next) {
 169         if (!bh->deleted && bh->scheduled) {
 170             if (bh->idle) {
 171                 /* idle bottom halves will be polled at least
 172                  * every 10ms */
 173                 timeout = 10000000;
 174             } else {
 175                 /* non-idle bottom halves will be executed
 176                  * immediately */
 177                 return 0;
 178             }
 179         }
 180     }
 181
 182     deadline = timerlistgroup_deadline_ns(&ctx->tlg);
 183     if (deadline == 0) {
 184         return 0;
 185     } else {
 186         return qemu_soonest_timeout(timeout, deadline);
 187     }
 188 }
 189
 190 static gboolean
 191 aio_ctx_prepare(GSource *source, gint    *timeout)
 192 {
 193     AioContext *ctx = (AioContext *) source;
 194
 195     atomic_or(&ctx->notify_me, 1);
 196
 197     /* We assume there is no timeout already supplied */
 198     *timeout = qemu_timeout_ns_to_ms(aio_compute_timeout(ctx));
 199
 200     if (aio_prepare(ctx)) {
 201         *timeout = 0;
 202     }
 203
 204     return *timeout == 0;
 205 }
 206
 207 static gboolean
 208 aio_ctx_check(GSource *source)
 209 {
 210     AioContext *ctx = (AioContext *) source;
 211     QEMUBH *bh;
 212
 213     atomic_and(&ctx->notify_me, ~1);
 214     aio_notify_accept(ctx);
 215
 216     for (bh = ctx->first_bh; bh; bh = bh->next) {
 217         if (!bh->deleted && bh->scheduled) {
 218             return true;
 219         }
 220     }
 221     return aio_pending(ctx) || (timerlistgroup_deadline_ns(&ctx->tlg) == 0);
 222 }
 223
 224 static gboolean
 225 aio_ctx_dispatch(GSource     *source,
 226                  GSourceFunc  callback,
 227                  gpointer     user_data)
 228 {
 229     AioContext *ctx = (AioContext *) source;
 230
 231     assert(callback == NULL);
 232     aio_dispatch(ctx);
 233     return true;
 234 }
 235
 236 static void
 237 aio_ctx_finalize(GSource     *source)
 238 {
 239     AioContext *ctx = (AioContext *) source;
 240
 241     qemu_bh_delete(ctx->notify_dummy_bh);
 242     thread_pool_free(ctx->thread_pool);
 243
 244     qemu_mutex_lock(&ctx->bh_lock);
 245     while (ctx->first_bh) {
 246         QEMUBH *next = ctx->first_bh->next;
 247
 248         /* qemu_bh_delete() must have been called on BHs in this AioContext */
 249         assert(ctx->first_bh->deleted);
 250
 251         g_free(ctx->first_bh);
 252         ctx->first_bh = next;
 253     }
 254     qemu_mutex_unlock(&ctx->bh_lock);
 255
 256     aio_set_event_notifier(ctx, &ctx->notifier, false, NULL);
 257     event_notifier_cleanup(&ctx->notifier);
 258     rfifolock_destroy(&ctx->lock);
 259     qemu_mutex_destroy(&ctx->bh_lock);
 260     timerlistgroup_deinit(&ctx->tlg);
 261 }
 262
 263 static GSourceFuncs aio_source_funcs = {
 264     aio_ctx_prepare,
 265     aio_ctx_check,
 266     aio_ctx_dispatch,
 267     aio_ctx_finalize
 268 };
 269
 270 GSource *aio_get_g_source(AioContext *ctx)
 271 {
 272     g_source_ref(&ctx->source);
 273     return &ctx->source;
 274 }
 275
 276 ThreadPool *aio_get_thread_pool(AioContext *ctx)
 277 {
 278     if (!ctx->thread_pool) {
 279         ctx->thread_pool = thread_pool_new(ctx);
 280     }
 281     return ctx->thread_pool;
 282 }
 283
 284 void aio_notify(AioContext *ctx)
 285 {
 286     /* Write e.g. bh->scheduled before reading ctx->notify_me.  Pairs
 287      * with atomic_or in aio_ctx_prepare or atomic_add in aio_poll.
 288      */
 289     smp_mb();
 290     if (ctx->notify_me) {
 291         event_notifier_set(&ctx->notifier);
 292         atomic_mb_set(&ctx->notified, true);
 293     }
 294 }
 295
 296 void aio_notify_accept(AioContext *ctx)
 297 {
 298     if (atomic_xchg(&ctx->notified, false)) {
 299         event_notifier_test_and_clear(&ctx->notifier);
 300     }
 301 }
 302
 303 static void aio_timerlist_notify(void *opaque)
 304 {
 305     aio_notify(opaque);
 306 }
 307
 308 static void aio_rfifolock_cb(void *opaque)
 309 {
 310     AioContext *ctx = opaque;
 311
 312     /* Kick owner thread in case they are blocked in aio_poll() */
 313     qemu_bh_schedule(ctx->notify_dummy_bh);
 314 }
 315
 316 static void notify_dummy_bh(void *opaque)
 317 {
 318     /* Do nothing, we were invoked just to force the event loop to iterate */
 319 }
 320
 321 static void event_notifier_dummy_cb(EventNotifier *e)
 322 {
 323 }
 324
 325 AioContext *aio_context_new(Error **errp)
 326 {
 327     int ret;
 328     AioContext *ctx;
 329     Error *local_err = NULL;
 330
 331     ctx = (AioContext *) g_source_new(&aio_source_funcs, sizeof(AioContext));
 332     aio_context_setup(ctx, &local_err);
 333     if (local_err) {
 334         error_propagate(errp, local_err);
 335         goto fail;
 336     }
 337     ret = event_notifier_init(&ctx->notifier, false);
 338     if (ret < 0) {
 339         error_setg_errno(errp, -ret, "Failed to initialize event notifier");
 340         goto fail;
 341     }
 342     g_source_set_can_recurse(&ctx->source, true);
 343     aio_set_event_notifier(ctx, &ctx->notifier,
 344                            false,
 345                            (EventNotifierHandler *)
 346                            event_notifier_dummy_cb);
 347     ctx->thread_pool = NULL;
 348     qemu_mutex_init(&ctx->bh_lock);
 349     rfifolock_init(&ctx->lock, aio_rfifolock_cb, ctx);
 350     timerlistgroup_init(&ctx->tlg, aio_timerlist_notify, ctx);
 351
 352     ctx->notify_dummy_bh = aio_bh_new(ctx, notify_dummy_bh, NULL);
 353
 354     return ctx;
 355 fail:
 356     g_source_destroy(&ctx->source);
 357     return NULL;
 358 }
 359
 360 void aio_context_ref(AioContext *ctx)
 361 {
 362     g_source_ref(&ctx->source);
 363 }
 364
 365 void aio_context_unref(AioContext *ctx)
 366 {
 367     g_source_unref(&ctx->source);
 368 }
 369
 370 void aio_context_acquire(AioContext *ctx)
 371 {
 372     rfifolock_lock(&ctx->lock);
 373 }
 374
 375 void aio_context_release(AioContext *ctx)
 376 {
 377     rfifolock_unlock(&ctx->lock);
 378 }