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         if (!bh->deleted && bh->scheduled) {
  76             bh->scheduled = 0;
  77             /* Paired with write barrier in bh schedule to ensure reading for
  78              * idle & callbacks coming after bh's scheduling.
  79              */
  80             smp_rmb();
  81             if (!bh->idle)
  82                 ret = 1;
  83             bh->idle = 0;
  84             bh->cb(bh->opaque);
  85         }
  86     }
  87
  88     ctx->walking_bh--;
  89
  90     /* remove deleted bhs */
  91     if (!ctx->walking_bh) {
  92         qemu_mutex_lock(&ctx->bh_lock);
  93         bhp = &ctx->first_bh;
  94         while (*bhp) {
  95             bh = *bhp;
  96             if (bh->deleted) {
  97                 *bhp = bh->next;
  98                 g_free(bh);
  99             } else {
 100                 bhp = &bh->next;
 101             }
 102         }
 103         qemu_mutex_unlock(&ctx->bh_lock);
 104     }
 105
 106     return ret;
 107 }
 108
 109 void qemu_bh_schedule_idle(QEMUBH *bh)
 110 {
 111     if (bh->scheduled)
 112         return;
 113     bh->idle = 1;
 114     /* Make sure that idle & any writes needed by the callback are done
 115      * before the locations are read in the aio_bh_poll.
 116      */
 117     smp_wmb();
 118     bh->scheduled = 1;
 119 }
 120
 121 void qemu_bh_schedule(QEMUBH *bh)
 122 {
 123     AioContext *ctx;
 124
 125     if (bh->scheduled)
 126         return;
 127     ctx = bh->ctx;
 128     bh->idle = 0;
 129     /* Make sure that:
 130      * 1. idle & any writes needed by the callback are done before the
 131      *    locations are read in the aio_bh_poll.
 132      * 2. ctx is loaded before scheduled is set and the callback has a chance
 133      *    to execute.
 134      */
 135     smp_mb();
 136     bh->scheduled = 1;
 137     aio_notify(ctx);
 138 }
 139
 140
 141 /* This func is async.
 142  */
 143 void qemu_bh_cancel(QEMUBH *bh)
 144 {
 145     bh->scheduled = 0;
 146 }
 147
 148 /* This func is async.The bottom half will do the delete action at the finial
 149  * end.
 150  */
 151 void qemu_bh_delete(QEMUBH *bh)
 152 {
 153     bh->scheduled = 0;
 154     bh->deleted = 1;
 155 }
 156
 157 int64_t
 158 aio_compute_timeout(AioContext *ctx)
 159 {
 160     int64_t deadline;
 161     int timeout = -1;
 162     QEMUBH *bh;
 163
 164     for (bh = ctx->first_bh; bh; bh = bh->next) {
 165         if (!bh->deleted && bh->scheduled) {
 166             if (bh->idle) {
 167                 /* idle bottom halves will be polled at least
 168                  * every 10ms */
 169                 timeout = 10000000;
 170             } else {
 171                 /* non-idle bottom halves will be executed
 172                  * immediately */
 173                 return 0;
 174             }
 175         }
 176     }
 177
 178     deadline = timerlistgroup_deadline_ns(&ctx->tlg);
 179     if (deadline == 0) {
 180         return 0;
 181     } else {
 182         return qemu_soonest_timeout(timeout, deadline);
 183     }
 184 }
 185
 186 static gboolean
 187 aio_ctx_prepare(GSource *source, gint    *timeout)
 188 {
 189     AioContext *ctx = (AioContext *) source;
 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     for (bh = ctx->first_bh; bh; bh = bh->next) {
 208         if (!bh->deleted && bh->scheduled) {
 209             return true;
 210         }
 211     }
 212     return aio_pending(ctx) || (timerlistgroup_deadline_ns(&ctx->tlg) == 0);
 213 }
 214
 215 static gboolean
 216 aio_ctx_dispatch(GSource     *source,
 217                  GSourceFunc  callback,
 218                  gpointer     user_data)
 219 {
 220     AioContext *ctx = (AioContext *) source;
 221
 222     assert(callback == NULL);
 223     aio_dispatch(ctx);
 224     return true;
 225 }
 226
 227 static void
 228 aio_ctx_finalize(GSource     *source)
 229 {
 230     AioContext *ctx = (AioContext *) source;
 231
 232     thread_pool_free(ctx->thread_pool);
 233     aio_set_event_notifier(ctx, &ctx->notifier, NULL);
 234     event_notifier_cleanup(&ctx->notifier);
 235     rfifolock_destroy(&ctx->lock);
 236     qemu_mutex_destroy(&ctx->bh_lock);
 237     g_array_free(ctx->pollfds, TRUE);
 238     timerlistgroup_deinit(&ctx->tlg);
 239 }
 240
 241 static GSourceFuncs aio_source_funcs = {
 242     aio_ctx_prepare,
 243     aio_ctx_check,
 244     aio_ctx_dispatch,
 245     aio_ctx_finalize
 246 };
 247
 248 GSource *aio_get_g_source(AioContext *ctx)
 249 {
 250     g_source_ref(&ctx->source);
 251     return &ctx->source;
 252 }
 253
 254 ThreadPool *aio_get_thread_pool(AioContext *ctx)
 255 {
 256     if (!ctx->thread_pool) {
 257         ctx->thread_pool = thread_pool_new(ctx);
 258     }
 259     return ctx->thread_pool;
 260 }
 261
 262 void aio_set_dispatching(AioContext *ctx, bool dispatching)
 263 {
 264     ctx->dispatching = dispatching;
 265     if (!dispatching) {
 266         /* Write ctx->dispatching before reading e.g. bh->scheduled.
 267          * Optimization: this is only needed when we're entering the "unsafe"
 268          * phase where other threads must call event_notifier_set.
 269          */
 270         smp_mb();
 271     }
 272 }
 273
 274 void aio_notify(AioContext *ctx)
 275 {
 276     /* Write e.g. bh->scheduled before reading ctx->dispatching.  */
 277     smp_mb();
 278     if (!ctx->dispatching) {
 279         event_notifier_set(&ctx->notifier);
 280     }
 281 }
 282
 283 static void aio_timerlist_notify(void *opaque)
 284 {
 285     aio_notify(opaque);
 286 }
 287
 288 static void aio_rfifolock_cb(void *opaque)
 289 {
 290     /* Kick owner thread in case they are blocked in aio_poll() */
 291     aio_notify(opaque);
 292 }
 293
 294 AioContext *aio_context_new(Error **errp)
 295 {
 296     int ret;
 297     AioContext *ctx;
 298     ctx = (AioContext *) g_source_new(&aio_source_funcs, sizeof(AioContext));
 299     ret = event_notifier_init(&ctx->notifier, false);
 300     if (ret < 0) {
 301         g_source_destroy(&ctx->source);
 302         error_setg_errno(errp, -ret, "Failed to initialize event notifier");
 303         return NULL;
 304     }
 305     g_source_set_can_recurse(&ctx->source, true);
 306     aio_set_event_notifier(ctx, &ctx->notifier,
 307                            (EventNotifierHandler *)
 308                            event_notifier_test_and_clear);
 309     ctx->pollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
 310     ctx->thread_pool = NULL;
 311     qemu_mutex_init(&ctx->bh_lock);
 312     rfifolock_init(&ctx->lock, aio_rfifolock_cb, ctx);
 313     timerlistgroup_init(&ctx->tlg, aio_timerlist_notify, ctx);
 314
 315     return ctx;
 316 }
 317
 318 void aio_context_ref(AioContext *ctx)
 319 {
 320     g_source_ref(&ctx->source);
 321 }
 322
 323 void aio_context_unref(AioContext *ctx)
 324 {
 325     g_source_unref(&ctx->source);
 326 }
 327
 328 void aio_context_acquire(AioContext *ctx)
 329 {
 330     rfifolock_lock(&ctx->lock);
 331 }
 332
 333 void aio_context_release(AioContext *ctx)
 334 {
 335     rfifolock_unlock(&ctx->lock);
 336 }