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 "qapi/error.h"
  27 #include "qemu-common.h"
  28 #include "block/aio.h"
  29 #include "block/thread-pool.h"
  30 #include "qemu/main-loop.h"
  31 #include "qemu/atomic.h"
  32
  33 /***********************************************************/
  34 /* bottom halves (can be seen as timers which expire ASAP) */
  35
  36 struct QEMUBH {
  37     AioContext *ctx;
  38     QEMUBHFunc *cb;
  39     void *opaque;
  40     QEMUBH *next;
  41     bool scheduled;
  42     bool idle;
  43     bool deleted;
  44 };
  45
  46 QEMUBH *aio_bh_new(AioContext *ctx, QEMUBHFunc *cb, void *opaque)
  47 {
  48     QEMUBH *bh;
  49     bh = g_new(QEMUBH, 1);
  50     *bh = (QEMUBH){
  51         .ctx = ctx,
  52         .cb = cb,
  53         .opaque = opaque,
  54     };
  55     qemu_mutex_lock(&ctx->bh_lock);
  56     bh->next = ctx->first_bh;
  57     /* Make sure that the members are ready before putting bh into list */
  58     smp_wmb();
  59     ctx->first_bh = bh;
  60     qemu_mutex_unlock(&ctx->bh_lock);
  61     return bh;
  62 }
  63
  64 void aio_bh_call(QEMUBH *bh)
  65 {
  66     bh->cb(bh->opaque);
  67 }
  68
  69 /* Multiple occurrences of aio_bh_poll cannot be called concurrently */
  70 int aio_bh_poll(AioContext *ctx)
  71 {
  72     QEMUBH *bh, **bhp, *next;
  73     int ret;
  74
  75     ctx->walking_bh++;
  76
  77     ret = 0;
  78     for (bh = ctx->first_bh; bh; bh = next) {
  79         /* Make sure that fetching bh happens before accessing its members */
  80         smp_read_barrier_depends();
  81         next = bh->next;
  82         /* The atomic_xchg is paired with the one in qemu_bh_schedule.  The
  83          * implicit memory barrier ensures that the callback sees all writes
  84          * done by the scheduling thread.  It also ensures that the scheduling
  85          * thread sees the zero before bh->cb has run, and thus will call
  86          * aio_notify again if necessary.
  87          */
  88         if (!bh->deleted && atomic_xchg(&bh->scheduled, 0)) {
  89             /* Idle BHs and the notify BH don't count as progress */
  90             if (!bh->idle && bh != ctx->notify_dummy_bh) {
  91                 ret = 1;
  92             }
  93             bh->idle = 0;
  94             aio_bh_call(bh);
  95         }
  96     }
  97
  98     ctx->walking_bh--;
  99
 100     /* remove deleted bhs */
 101     if (!ctx->walking_bh) {
 102         qemu_mutex_lock(&ctx->bh_lock);
 103         bhp = &ctx->first_bh;
 104         while (*bhp) {
 105             bh = *bhp;
 106             if (bh->deleted) {
 107                 *bhp = bh->next;
 108                 g_free(bh);
 109             } else {
 110                 bhp = &bh->next;
 111             }
 112         }
 113         qemu_mutex_unlock(&ctx->bh_lock);
 114     }
 115
 116     return ret;
 117 }
 118
 119 void qemu_bh_schedule_idle(QEMUBH *bh)
 120 {
 121     bh->idle = 1;
 122     /* Make sure that idle & any writes needed by the callback are done
 123      * before the locations are read in the aio_bh_poll.
 124      */
 125     atomic_mb_set(&bh->scheduled, 1);
 126 }
 127
 128 void qemu_bh_schedule(QEMUBH *bh)
 129 {
 130     AioContext *ctx;
 131
 132     ctx = bh->ctx;
 133     bh->idle = 0;
 134     /* The memory barrier implicit in atomic_xchg makes sure that:
 135      * 1. idle & any writes needed by the callback are done before the
 136      *    locations are read in the aio_bh_poll.
 137      * 2. ctx is loaded before scheduled is set and the callback has a chance
 138      *    to execute.
 139      */
 140     if (atomic_xchg(&bh->scheduled, 1) == 0) {
 141         aio_notify(ctx);
 142     }
 143 }
 144
 145
 146 /* This func is async.
 147  */
 148 void qemu_bh_cancel(QEMUBH *bh)
 149 {
 150     bh->scheduled = 0;
 151 }
 152
 153 /* This func is async.The bottom half will do the delete action at the finial
 154  * end.
 155  */
 156 void qemu_bh_delete(QEMUBH *bh)
 157 {
 158     bh->scheduled = 0;
 159     bh->deleted = 1;
 160 }
 161
 162 int64_t
 163 aio_compute_timeout(AioContext *ctx)
 164 {
 165     int64_t deadline;
 166     int timeout = -1;
 167     QEMUBH *bh;
 168
 169     for (bh = ctx->first_bh; bh; bh = bh->next) {
 170         if (!bh->deleted && bh->scheduled) {
 171             if (bh->idle) {
 172                 /* idle bottom halves will be polled at least
 173                  * every 10ms */
 174                 timeout = 10000000;
 175             } else {
 176                 /* non-idle bottom halves will be executed
 177                  * immediately */
 178                 return 0;
 179             }
 180         }
 181     }
 182
 183     deadline = timerlistgroup_deadline_ns(&ctx->tlg);
 184     if (deadline == 0) {
 185         return 0;
 186     } else {
 187         return qemu_soonest_timeout(timeout, deadline);
 188     }
 189 }
 190
 191 static gboolean
 192 aio_ctx_prepare(GSource *source, gint    *timeout)
 193 {
 194     AioContext *ctx = (AioContext *) source;
 195
 196     atomic_or(&ctx->notify_me, 1);
 197
 198     /* We assume there is no timeout already supplied */
 199     *timeout = qemu_timeout_ns_to_ms(aio_compute_timeout(ctx));
 200
 201     if (aio_prepare(ctx)) {
 202         *timeout = 0;
 203     }
 204
 205     return *timeout == 0;
 206 }
 207
 208 static gboolean
 209 aio_ctx_check(GSource *source)
 210 {
 211     AioContext *ctx = (AioContext *) source;
 212     QEMUBH *bh;
 213
 214     atomic_and(&ctx->notify_me, ~1);
 215     aio_notify_accept(ctx);
 216
 217     for (bh = ctx->first_bh; bh; bh = bh->next) {
 218         if (!bh->deleted && bh->scheduled) {
 219             return true;
 220         }
 221     }
 222     return aio_pending(ctx) || (timerlistgroup_deadline_ns(&ctx->tlg) == 0);
 223 }
 224
 225 static gboolean
 226 aio_ctx_dispatch(GSource     *source,
 227                  GSourceFunc  callback,
 228                  gpointer     user_data)
 229 {
 230     AioContext *ctx = (AioContext *) source;
 231
 232     assert(callback == NULL);
 233     aio_dispatch(ctx);
 234     return true;
 235 }
 236
 237 static void
 238 aio_ctx_finalize(GSource     *source)
 239 {
 240     AioContext *ctx = (AioContext *) source;
 241
 242     qemu_bh_delete(ctx->notify_dummy_bh);
 243     thread_pool_free(ctx->thread_pool);
 244
 245     qemu_mutex_lock(&ctx->bh_lock);
 246     while (ctx->first_bh) {
 247         QEMUBH *next = ctx->first_bh->next;
 248
 249         /* qemu_bh_delete() must have been called on BHs in this AioContext */
 250         assert(ctx->first_bh->deleted);
 251
 252         g_free(ctx->first_bh);
 253         ctx->first_bh = next;
 254     }
 255     qemu_mutex_unlock(&ctx->bh_lock);
 256
 257     aio_set_event_notifier(ctx, &ctx->notifier, false, NULL);
 258     event_notifier_cleanup(&ctx->notifier);
 259     rfifolock_destroy(&ctx->lock);
 260     qemu_mutex_destroy(&ctx->bh_lock);
 261     timerlistgroup_deinit(&ctx->tlg);
 262 }
 263
 264 static GSourceFuncs aio_source_funcs = {
 265     aio_ctx_prepare,
 266     aio_ctx_check,
 267     aio_ctx_dispatch,
 268     aio_ctx_finalize
 269 };
 270
 271 GSource *aio_get_g_source(AioContext *ctx)
 272 {
 273     g_source_ref(&ctx->source);
 274     return &ctx->source;
 275 }
 276
 277 ThreadPool *aio_get_thread_pool(AioContext *ctx)
 278 {
 279     if (!ctx->thread_pool) {
 280         ctx->thread_pool = thread_pool_new(ctx);
 281     }
 282     return ctx->thread_pool;
 283 }
 284
 285 void aio_notify(AioContext *ctx)
 286 {
 287     /* Write e.g. bh->scheduled before reading ctx->notify_me.  Pairs
 288      * with atomic_or in aio_ctx_prepare or atomic_add in aio_poll.
 289      */
 290     smp_mb();
 291     if (ctx->notify_me) {
 292         event_notifier_set(&ctx->notifier);
 293         atomic_mb_set(&ctx->notified, true);
 294     }
 295 }
 296
 297 void aio_notify_accept(AioContext *ctx)
 298 {
 299     if (atomic_xchg(&ctx->notified, false)) {
 300         event_notifier_test_and_clear(&ctx->notifier);
 301     }
 302 }
 303
 304 static void aio_timerlist_notify(void *opaque)
 305 {
 306     aio_notify(opaque);
 307 }
 308
 309 static void aio_rfifolock_cb(void *opaque)
 310 {
 311     AioContext *ctx = opaque;
 312
 313     /* Kick owner thread in case they are blocked in aio_poll() */
 314     qemu_bh_schedule(ctx->notify_dummy_bh);
 315 }
 316
 317 static void notify_dummy_bh(void *opaque)
 318 {
 319     /* Do nothing, we were invoked just to force the event loop to iterate */
 320 }
 321
 322 static void event_notifier_dummy_cb(EventNotifier *e)
 323 {
 324 }
 325
 326 AioContext *aio_context_new(Error **errp)
 327 {
 328     int ret;
 329     AioContext *ctx;
 330     Error *local_err = NULL;
 331
 332     ctx = (AioContext *) g_source_new(&aio_source_funcs, sizeof(AioContext));
 333     aio_context_setup(ctx, &local_err);
 334     if (local_err) {
 335         error_propagate(errp, local_err);
 336         goto fail;
 337     }
 338     ret = event_notifier_init(&ctx->notifier, false);
 339     if (ret < 0) {
 340         error_setg_errno(errp, -ret, "Failed to initialize event notifier");
 341         goto fail;
 342     }
 343     g_source_set_can_recurse(&ctx->source, true);
 344     aio_set_event_notifier(ctx, &ctx->notifier,
 345                            false,
 346                            (EventNotifierHandler *)
 347                            event_notifier_dummy_cb);
 348     ctx->thread_pool = NULL;
 349     qemu_mutex_init(&ctx->bh_lock);
 350     rfifolock_init(&ctx->lock, aio_rfifolock_cb, ctx);
 351     timerlistgroup_init(&ctx->tlg, aio_timerlist_notify, ctx);
 352
 353     ctx->notify_dummy_bh = aio_bh_new(ctx, notify_dummy_bh, NULL);
 354
 355     return ctx;
 356 fail:
 357     g_source_destroy(&ctx->source);
 358     return NULL;
 359 }
 360
 361 void aio_context_ref(AioContext *ctx)
 362 {
 363     g_source_ref(&ctx->source);
 364 }
 365
 366 void aio_context_unref(AioContext *ctx)
 367 {
 368     g_source_unref(&ctx->source);
 369 }
 370
 371 void aio_context_acquire(AioContext *ctx)
 372 {
 373     rfifolock_lock(&ctx->lock);
 374 }
 375
 376 void aio_context_release(AioContext *ctx)
 377 {
 378     rfifolock_unlock(&ctx->lock);
 379 }