iothread.c

   1 /*
   2  * Event loop thread
   3  *
   4  * Copyright Red Hat Inc., 2013
   5  *
   6  * Authors:
   7  *  Stefan Hajnoczi   <stefanha@redhat.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2 or later.
  10  * See the COPYING file in the top-level directory.
  11  *
  12  */
  13
  14 #include "qemu/osdep.h"
  15 #include "qom/object.h"
  16 #include "qom/object_interfaces.h"
  17 #include "qemu/module.h"
  18 #include "block/aio.h"
  19 #include "block/block.h"
  20 #include "sysemu/iothread.h"
  21 #include "qmp-commands.h"
  22 #include "qemu/error-report.h"
  23 #include "qemu/rcu.h"
  24 #include "qemu/main-loop.h"
  25
  26 typedef ObjectClass IOThreadClass;
  27
  28 #define IOTHREAD_GET_CLASS(obj) \
  29    OBJECT_GET_CLASS(IOThreadClass, obj, TYPE_IOTHREAD)
  30 #define IOTHREAD_CLASS(klass) \
  31    OBJECT_CLASS_CHECK(IOThreadClass, klass, TYPE_IOTHREAD)
  32
  33 /* Benchmark results from 2016 on NVMe SSD drives show max polling times around
  34  * 16-32 microseconds yield IOPS improvements for both iodepth=1 and iodepth=32
  35  * workloads.
  36  */
  37 #define IOTHREAD_POLL_MAX_NS_DEFAULT 32768ULL
  38
  39 static __thread IOThread *my_iothread;
  40
  41 AioContext *qemu_get_current_aio_context(void)
  42 {
  43     return my_iothread ? my_iothread->ctx : qemu_get_aio_context();
  44 }
  45
  46 static void *iothread_run(void *opaque)
  47 {
  48     IOThread *iothread = opaque;
  49
  50     rcu_register_thread();
  51
  52     my_iothread = iothread;
  53     qemu_mutex_lock(&iothread->init_done_lock);
  54     iothread->thread_id = qemu_get_thread_id();
  55     qemu_cond_signal(&iothread->init_done_cond);
  56     qemu_mutex_unlock(&iothread->init_done_lock);
  57
  58     while (!atomic_read(&iothread->stopping)) {
  59         aio_poll(iothread->ctx, true);
  60
  61         if (atomic_read(&iothread->worker_context)) {
  62             GMainLoop *loop;
  63
  64             g_main_context_push_thread_default(iothread->worker_context);
  65             iothread->main_loop =
  66                 g_main_loop_new(iothread->worker_context, TRUE);
  67             loop = iothread->main_loop;
  68
  69             g_main_loop_run(iothread->main_loop);
  70             iothread->main_loop = NULL;
  71             g_main_loop_unref(loop);
  72
  73             g_main_context_pop_thread_default(iothread->worker_context);
  74             g_main_context_unref(iothread->worker_context);
  75             iothread->worker_context = NULL;
  76         }
  77     }
  78
  79     rcu_unregister_thread();
  80     return NULL;
  81 }
  82
  83 static int iothread_stop(Object *object, void *opaque)
  84 {
  85     IOThread *iothread;
  86
  87     iothread = (IOThread *)object_dynamic_cast(object, TYPE_IOTHREAD);
  88     if (!iothread || !iothread->ctx) {
  89         return 0;
  90     }
  91     iothread->stopping = true;
  92     aio_notify(iothread->ctx);
  93     if (atomic_read(&iothread->main_loop)) {
  94         g_main_loop_quit(iothread->main_loop);
  95     }
  96     qemu_thread_join(&iothread->thread);
  97     return 0;
  98 }
  99
 100 static void iothread_instance_init(Object *obj)
 101 {
 102     IOThread *iothread = IOTHREAD(obj);
 103
 104     iothread->poll_max_ns = IOTHREAD_POLL_MAX_NS_DEFAULT;
 105 }
 106
 107 static void iothread_instance_finalize(Object *obj)
 108 {
 109     IOThread *iothread = IOTHREAD(obj);
 110
 111     iothread_stop(obj, NULL);
 112     qemu_cond_destroy(&iothread->init_done_cond);
 113     qemu_mutex_destroy(&iothread->init_done_lock);
 114     if (!iothread->ctx) {
 115         return;
 116     }
 117     aio_context_unref(iothread->ctx);
 118 }
 119
 120 static void iothread_complete(UserCreatable *obj, Error **errp)
 121 {
 122     Error *local_error = NULL;
 123     IOThread *iothread = IOTHREAD(obj);
 124     char *name, *thread_name;
 125
 126     iothread->stopping = false;
 127     iothread->thread_id = -1;
 128     iothread->ctx = aio_context_new(&local_error);
 129     if (!iothread->ctx) {
 130         error_propagate(errp, local_error);
 131         return;
 132     }
 133
 134     aio_context_set_poll_params(iothread->ctx,
 135                                 iothread->poll_max_ns,
 136                                 iothread->poll_grow,
 137                                 iothread->poll_shrink,
 138                                 &local_error);
 139     if (local_error) {
 140         error_propagate(errp, local_error);
 141         aio_context_unref(iothread->ctx);
 142         iothread->ctx = NULL;
 143         return;
 144     }
 145
 146     qemu_mutex_init(&iothread->init_done_lock);
 147     qemu_cond_init(&iothread->init_done_cond);
 148     iothread->once = (GOnce) G_ONCE_INIT;
 149
 150     /* This assumes we are called from a thread with useful CPU affinity for us
 151      * to inherit.
 152      */
 153     name = object_get_canonical_path_component(OBJECT(obj));
 154     thread_name = g_strdup_printf("IO %s", name);
 155     qemu_thread_create(&iothread->thread, thread_name, iothread_run,
 156                        iothread, QEMU_THREAD_JOINABLE);
 157     g_free(thread_name);
 158     g_free(name);
 159
 160     /* Wait for initialization to complete */
 161     qemu_mutex_lock(&iothread->init_done_lock);
 162     while (iothread->thread_id == -1) {
 163         qemu_cond_wait(&iothread->init_done_cond,
 164                        &iothread->init_done_lock);
 165     }
 166     qemu_mutex_unlock(&iothread->init_done_lock);
 167 }
 168
 169 typedef struct {
 170     const char *name;
 171     ptrdiff_t offset; /* field's byte offset in IOThread struct */
 172 } PollParamInfo;
 173
 174 static PollParamInfo poll_max_ns_info = {
 175     "poll-max-ns", offsetof(IOThread, poll_max_ns),
 176 };
 177 static PollParamInfo poll_grow_info = {
 178     "poll-grow", offsetof(IOThread, poll_grow),
 179 };
 180 static PollParamInfo poll_shrink_info = {
 181     "poll-shrink", offsetof(IOThread, poll_shrink),
 182 };
 183
 184 static void iothread_get_poll_param(Object *obj, Visitor *v,
 185         const char *name, void *opaque, Error **errp)
 186 {
 187     IOThread *iothread = IOTHREAD(obj);
 188     PollParamInfo *info = opaque;
 189     int64_t *field = (void *)iothread + info->offset;
 190
 191     visit_type_int64(v, name, field, errp);
 192 }
 193
 194 static void iothread_set_poll_param(Object *obj, Visitor *v,
 195         const char *name, void *opaque, Error **errp)
 196 {
 197     IOThread *iothread = IOTHREAD(obj);
 198     PollParamInfo *info = opaque;
 199     int64_t *field = (void *)iothread + info->offset;
 200     Error *local_err = NULL;
 201     int64_t value;
 202
 203     visit_type_int64(v, name, &value, &local_err);
 204     if (local_err) {
 205         goto out;
 206     }
 207
 208     if (value < 0) {
 209         error_setg(&local_err, "%s value must be in range [0, %"PRId64"]",
 210                    info->name, INT64_MAX);
 211         goto out;
 212     }
 213
 214     *field = value;
 215
 216     if (iothread->ctx) {
 217         aio_context_set_poll_params(iothread->ctx,
 218                                     iothread->poll_max_ns,
 219                                     iothread->poll_grow,
 220                                     iothread->poll_shrink,
 221                                     &local_err);
 222     }
 223
 224 out:
 225     error_propagate(errp, local_err);
 226 }
 227
 228 static void iothread_class_init(ObjectClass *klass, void *class_data)
 229 {
 230     UserCreatableClass *ucc = USER_CREATABLE_CLASS(klass);
 231     ucc->complete = iothread_complete;
 232
 233     object_class_property_add(klass, "poll-max-ns", "int",
 234                               iothread_get_poll_param,
 235                               iothread_set_poll_param,
 236                               NULL, &poll_max_ns_info, &error_abort);
 237     object_class_property_add(klass, "poll-grow", "int",
 238                               iothread_get_poll_param,
 239                               iothread_set_poll_param,
 240                               NULL, &poll_grow_info, &error_abort);
 241     object_class_property_add(klass, "poll-shrink", "int",
 242                               iothread_get_poll_param,
 243                               iothread_set_poll_param,
 244                               NULL, &poll_shrink_info, &error_abort);
 245 }
 246
 247 static const TypeInfo iothread_info = {
 248     .name = TYPE_IOTHREAD,
 249     .parent = TYPE_OBJECT,
 250     .class_init = iothread_class_init,
 251     .instance_size = sizeof(IOThread),
 252     .instance_init = iothread_instance_init,
 253     .instance_finalize = iothread_instance_finalize,
 254     .interfaces = (InterfaceInfo[]) {
 255         {TYPE_USER_CREATABLE},
 256         {}
 257     },
 258 };
 259
 260 static void iothread_register_types(void)
 261 {
 262     type_register_static(&iothread_info);
 263 }
 264
 265 type_init(iothread_register_types)
 266
 267 char *iothread_get_id(IOThread *iothread)
 268 {
 269     return object_get_canonical_path_component(OBJECT(iothread));
 270 }
 271
 272 AioContext *iothread_get_aio_context(IOThread *iothread)
 273 {
 274     return iothread->ctx;
 275 }
 276
 277 static int query_one_iothread(Object *object, void *opaque)
 278 {
 279     IOThreadInfoList ***prev = opaque;
 280     IOThreadInfoList *elem;
 281     IOThreadInfo *info;
 282     IOThread *iothread;
 283
 284     iothread = (IOThread *)object_dynamic_cast(object, TYPE_IOTHREAD);
 285     if (!iothread) {
 286         return 0;
 287     }
 288
 289     info = g_new0(IOThreadInfo, 1);
 290     info->id = iothread_get_id(iothread);
 291     info->thread_id = iothread->thread_id;
 292     info->poll_max_ns = iothread->poll_max_ns;
 293     info->poll_grow = iothread->poll_grow;
 294     info->poll_shrink = iothread->poll_shrink;
 295
 296     elem = g_new0(IOThreadInfoList, 1);
 297     elem->value = info;
 298     elem->next = NULL;
 299
 300     **prev = elem;
 301     *prev = &elem->next;
 302     return 0;
 303 }
 304
 305 IOThreadInfoList *qmp_query_iothreads(Error **errp)
 306 {
 307     IOThreadInfoList *head = NULL;
 308     IOThreadInfoList **prev = &head;
 309     Object *container = object_get_objects_root();
 310
 311     object_child_foreach(container, query_one_iothread, &prev);
 312     return head;
 313 }
 314
 315 void iothread_stop_all(void)
 316 {
 317     Object *container = object_get_objects_root();
 318     BlockDriverState *bs;
 319     BdrvNextIterator it;
 320
 321     for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
 322         AioContext *ctx = bdrv_get_aio_context(bs);
 323         if (ctx == qemu_get_aio_context()) {
 324             continue;
 325         }
 326         aio_context_acquire(ctx);
 327         bdrv_set_aio_context(bs, qemu_get_aio_context());
 328         aio_context_release(ctx);
 329     }
 330
 331     object_child_foreach(container, iothread_stop, NULL);
 332 }
 333
 334 static gpointer iothread_g_main_context_init(gpointer opaque)
 335 {
 336     AioContext *ctx;
 337     IOThread *iothread = opaque;
 338     GSource *source;
 339
 340     iothread->worker_context = g_main_context_new();
 341
 342     ctx = iothread_get_aio_context(iothread);
 343     source = aio_get_g_source(ctx);
 344     g_source_attach(source, iothread->worker_context);
 345     g_source_unref(source);
 346
 347     aio_notify(iothread->ctx);
 348     return NULL;
 349 }
 350
 351 GMainContext *iothread_get_g_main_context(IOThread *iothread)
 352 {
 353     g_once(&iothread->once, iothread_g_main_context_init, iothread);
 354
 355     return iothread->worker_context;
 356 }