iothread.c

   1 /*
   2  * Event loop thread
   3  *
   4  * Copyright Red Hat Inc., 2013, 2020
   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 "qapi/error.h"
  22 #include "qapi/qapi-commands-misc.h"
  23 #include "qemu/error-report.h"
  24 #include "qemu/rcu.h"
  25 #include "qemu/main-loop.h"
  26
  27 typedef ObjectClass IOThreadClass;
  28
  29 DECLARE_CLASS_CHECKERS(IOThreadClass, IOTHREAD,
  30                        TYPE_IOTHREAD)
  31
  32 #ifdef CONFIG_POSIX
  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 #else
  39 #define IOTHREAD_POLL_MAX_NS_DEFAULT 0ULL
  40 #endif
  41
  42 static void *iothread_run(void *opaque)
  43 {
  44     IOThread *iothread = opaque;
  45
  46     rcu_register_thread();
  47     /*
  48      * g_main_context_push_thread_default() must be called before anything
  49      * in this new thread uses glib.
  50      */
  51     g_main_context_push_thread_default(iothread->worker_context);
  52     qemu_set_current_aio_context(iothread->ctx);
  53     iothread->thread_id = qemu_get_thread_id();
  54     qemu_sem_post(&iothread->init_done_sem);
  55
  56     while (iothread->running) {
  57         /*
  58          * Note: from functional-wise the g_main_loop_run() below can
  59          * already cover the aio_poll() events, but we can't run the
  60          * main loop unconditionally because explicit aio_poll() here
  61          * is faster than g_main_loop_run() when we do not need the
  62          * gcontext at all (e.g., pure block layer iothreads).  In
  63          * other words, when we want to run the gcontext with the
  64          * iothread we need to pay some performance for functionality.
  65          */
  66         aio_poll(iothread->ctx, true);
  67
  68         /*
  69          * We must check the running state again in case it was
  70          * changed in previous aio_poll()
  71          */
  72         if (iothread->running && qatomic_read(&iothread->run_gcontext)) {
  73             g_main_loop_run(iothread->main_loop);
  74         }
  75     }
  76
  77     g_main_context_pop_thread_default(iothread->worker_context);
  78     rcu_unregister_thread();
  79     return NULL;
  80 }
  81
  82 /* Runs in iothread_run() thread */
  83 static void iothread_stop_bh(void *opaque)
  84 {
  85     IOThread *iothread = opaque;
  86
  87     iothread->running = false; /* stop iothread_run() */
  88
  89     if (iothread->main_loop) {
  90         g_main_loop_quit(iothread->main_loop);
  91     }
  92 }
  93
  94 void iothread_stop(IOThread *iothread)
  95 {
  96     if (!iothread->ctx || iothread->stopping) {
  97         return;
  98     }
  99     iothread->stopping = true;
 100     aio_bh_schedule_oneshot(iothread->ctx, iothread_stop_bh, iothread);
 101     qemu_thread_join(&iothread->thread);
 102 }
 103
 104 static void iothread_instance_init(Object *obj)
 105 {
 106     IOThread *iothread = IOTHREAD(obj);
 107
 108     iothread->poll_max_ns = IOTHREAD_POLL_MAX_NS_DEFAULT;
 109     iothread->thread_id = -1;
 110     qemu_sem_init(&iothread->init_done_sem, 0);
 111     /* By default, we don't run gcontext */
 112     qatomic_set(&iothread->run_gcontext, 0);
 113 }
 114
 115 static void iothread_instance_finalize(Object *obj)
 116 {
 117     IOThread *iothread = IOTHREAD(obj);
 118
 119     iothread_stop(iothread);
 120
 121     /*
 122      * Before glib2 2.33.10, there is a glib2 bug that GSource context
 123      * pointer may not be cleared even if the context has already been
 124      * destroyed (while it should).  Here let's free the AIO context
 125      * earlier to bypass that glib bug.
 126      *
 127      * We can remove this comment after the minimum supported glib2
 128      * version boosts to 2.33.10.  Before that, let's free the
 129      * GSources first before destroying any GMainContext.
 130      */
 131     if (iothread->ctx) {
 132         aio_context_unref(iothread->ctx);
 133         iothread->ctx = NULL;
 134     }
 135     if (iothread->worker_context) {
 136         g_main_context_unref(iothread->worker_context);
 137         iothread->worker_context = NULL;
 138         g_main_loop_unref(iothread->main_loop);
 139         iothread->main_loop = NULL;
 140     }
 141     qemu_sem_destroy(&iothread->init_done_sem);
 142 }
 143
 144 static void iothread_init_gcontext(IOThread *iothread)
 145 {
 146     GSource *source;
 147
 148     iothread->worker_context = g_main_context_new();
 149     source = aio_get_g_source(iothread_get_aio_context(iothread));
 150     g_source_attach(source, iothread->worker_context);
 151     g_source_unref(source);
 152     iothread->main_loop = g_main_loop_new(iothread->worker_context, TRUE);
 153 }
 154
 155 static void iothread_set_aio_context_params(IOThread *iothread, Error **errp)
 156 {
 157     ERRP_GUARD();
 158
 159     aio_context_set_poll_params(iothread->ctx,
 160                                 iothread->poll_max_ns,
 161                                 iothread->poll_grow,
 162                                 iothread->poll_shrink,
 163                                 errp);
 164     if (*errp) {
 165         return;
 166     }
 167
 168     aio_context_set_aio_params(iothread->ctx,
 169                                iothread->aio_max_batch,
 170                                errp);
 171 }
 172
 173 static void iothread_complete(UserCreatable *obj, Error **errp)
 174 {
 175     Error *local_error = NULL;
 176     IOThread *iothread = IOTHREAD(obj);
 177     char *thread_name;
 178
 179     iothread->stopping = false;
 180     iothread->running = true;
 181     iothread->ctx = aio_context_new(errp);
 182     if (!iothread->ctx) {
 183         return;
 184     }
 185
 186     /*
 187      * Init one GMainContext for the iothread unconditionally, even if
 188      * it's not used
 189      */
 190     iothread_init_gcontext(iothread);
 191
 192     iothread_set_aio_context_params(iothread, &local_error);
 193     if (local_error) {
 194         error_propagate(errp, local_error);
 195         aio_context_unref(iothread->ctx);
 196         iothread->ctx = NULL;
 197         return;
 198     }
 199
 200     /* This assumes we are called from a thread with useful CPU affinity for us
 201      * to inherit.
 202      */
 203     thread_name = g_strdup_printf("IO %s",
 204                         object_get_canonical_path_component(OBJECT(obj)));
 205     qemu_thread_create(&iothread->thread, thread_name, iothread_run,
 206                        iothread, QEMU_THREAD_JOINABLE);
 207     g_free(thread_name);
 208
 209     /* Wait for initialization to complete */
 210     while (iothread->thread_id == -1) {
 211         qemu_sem_wait(&iothread->init_done_sem);
 212     }
 213 }
 214
 215 typedef struct {
 216     const char *name;
 217     ptrdiff_t offset; /* field's byte offset in IOThread struct */
 218 } IOThreadParamInfo;
 219
 220 static IOThreadParamInfo poll_max_ns_info = {
 221     "poll-max-ns", offsetof(IOThread, poll_max_ns),
 222 };
 223 static IOThreadParamInfo poll_grow_info = {
 224     "poll-grow", offsetof(IOThread, poll_grow),
 225 };
 226 static IOThreadParamInfo poll_shrink_info = {
 227     "poll-shrink", offsetof(IOThread, poll_shrink),
 228 };
 229 static IOThreadParamInfo aio_max_batch_info = {
 230     "aio-max-batch", offsetof(IOThread, aio_max_batch),
 231 };
 232
 233 static void iothread_get_param(Object *obj, Visitor *v,
 234         const char *name, void *opaque, Error **errp)
 235 {
 236     IOThread *iothread = IOTHREAD(obj);
 237     IOThreadParamInfo *info = opaque;
 238     int64_t *field = (void *)iothread + info->offset;
 239
 240     visit_type_int64(v, name, field, errp);
 241 }
 242
 243 static bool iothread_set_param(Object *obj, Visitor *v,
 244         const char *name, void *opaque, Error **errp)
 245 {
 246     IOThread *iothread = IOTHREAD(obj);
 247     IOThreadParamInfo *info = opaque;
 248     int64_t *field = (void *)iothread + info->offset;
 249     int64_t value;
 250
 251     if (!visit_type_int64(v, name, &value, errp)) {
 252         return false;
 253     }
 254
 255     if (value < 0) {
 256         error_setg(errp, "%s value must be in range [0, %" PRId64 "]",
 257                    info->name, INT64_MAX);
 258         return false;
 259     }
 260
 261     *field = value;
 262
 263     return true;
 264 }
 265
 266 static void iothread_get_poll_param(Object *obj, Visitor *v,
 267         const char *name, void *opaque, Error **errp)
 268 {
 269
 270     iothread_get_param(obj, v, name, opaque, errp);
 271 }
 272
 273 static void iothread_set_poll_param(Object *obj, Visitor *v,
 274         const char *name, void *opaque, Error **errp)
 275 {
 276     IOThread *iothread = IOTHREAD(obj);
 277
 278     if (!iothread_set_param(obj, v, name, opaque, errp)) {
 279         return;
 280     }
 281
 282     if (iothread->ctx) {
 283         aio_context_set_poll_params(iothread->ctx,
 284                                     iothread->poll_max_ns,
 285                                     iothread->poll_grow,
 286                                     iothread->poll_shrink,
 287                                     errp);
 288     }
 289 }
 290
 291 static void iothread_get_aio_param(Object *obj, Visitor *v,
 292         const char *name, void *opaque, Error **errp)
 293 {
 294
 295     iothread_get_param(obj, v, name, opaque, errp);
 296 }
 297
 298 static void iothread_set_aio_param(Object *obj, Visitor *v,
 299         const char *name, void *opaque, Error **errp)
 300 {
 301     IOThread *iothread = IOTHREAD(obj);
 302
 303     if (!iothread_set_param(obj, v, name, opaque, errp)) {
 304         return;
 305     }
 306
 307     if (iothread->ctx) {
 308         aio_context_set_aio_params(iothread->ctx,
 309                                    iothread->aio_max_batch,
 310                                    errp);
 311     }
 312 }
 313
 314 static void iothread_class_init(ObjectClass *klass, void *class_data)
 315 {
 316     UserCreatableClass *ucc = USER_CREATABLE_CLASS(klass);
 317     ucc->complete = iothread_complete;
 318
 319     object_class_property_add(klass, "poll-max-ns", "int",
 320                               iothread_get_poll_param,
 321                               iothread_set_poll_param,
 322                               NULL, &poll_max_ns_info);
 323     object_class_property_add(klass, "poll-grow", "int",
 324                               iothread_get_poll_param,
 325                               iothread_set_poll_param,
 326                               NULL, &poll_grow_info);
 327     object_class_property_add(klass, "poll-shrink", "int",
 328                               iothread_get_poll_param,
 329                               iothread_set_poll_param,
 330                               NULL, &poll_shrink_info);
 331     object_class_property_add(klass, "aio-max-batch", "int",
 332                               iothread_get_aio_param,
 333                               iothread_set_aio_param,
 334                               NULL, &aio_max_batch_info);
 335 }
 336
 337 static const TypeInfo iothread_info = {
 338     .name = TYPE_IOTHREAD,
 339     .parent = TYPE_OBJECT,
 340     .class_init = iothread_class_init,
 341     .instance_size = sizeof(IOThread),
 342     .instance_init = iothread_instance_init,
 343     .instance_finalize = iothread_instance_finalize,
 344     .interfaces = (InterfaceInfo[]) {
 345         {TYPE_USER_CREATABLE},
 346         {}
 347     },
 348 };
 349
 350 static void iothread_register_types(void)
 351 {
 352     type_register_static(&iothread_info);
 353 }
 354
 355 type_init(iothread_register_types)
 356
 357 char *iothread_get_id(IOThread *iothread)
 358 {
 359     return g_strdup(object_get_canonical_path_component(OBJECT(iothread)));
 360 }
 361
 362 AioContext *iothread_get_aio_context(IOThread *iothread)
 363 {
 364     return iothread->ctx;
 365 }
 366
 367 static int query_one_iothread(Object *object, void *opaque)
 368 {
 369     IOThreadInfoList ***tail = opaque;
 370     IOThreadInfo *info;
 371     IOThread *iothread;
 372
 373     iothread = (IOThread *)object_dynamic_cast(object, TYPE_IOTHREAD);
 374     if (!iothread) {
 375         return 0;
 376     }
 377
 378     info = g_new0(IOThreadInfo, 1);
 379     info->id = iothread_get_id(iothread);
 380     info->thread_id = iothread->thread_id;
 381     info->poll_max_ns = iothread->poll_max_ns;
 382     info->poll_grow = iothread->poll_grow;
 383     info->poll_shrink = iothread->poll_shrink;
 384     info->aio_max_batch = iothread->aio_max_batch;
 385
 386     QAPI_LIST_APPEND(*tail, info);
 387     return 0;
 388 }
 389
 390 IOThreadInfoList *qmp_query_iothreads(Error **errp)
 391 {
 392     IOThreadInfoList *head = NULL;
 393     IOThreadInfoList **prev = &head;
 394     Object *container = object_get_objects_root();
 395
 396     object_child_foreach(container, query_one_iothread, &prev);
 397     return head;
 398 }
 399
 400 GMainContext *iothread_get_g_main_context(IOThread *iothread)
 401 {
 402     qatomic_set(&iothread->run_gcontext, 1);
 403     aio_notify(iothread->ctx);
 404     return iothread->worker_context;
 405 }
 406
 407 IOThread *iothread_create(const char *id, Error **errp)
 408 {
 409     Object *obj;
 410
 411     obj = object_new_with_props(TYPE_IOTHREAD,
 412                                 object_get_internal_root(),
 413                                 id, errp, NULL);
 414
 415     return IOTHREAD(obj);
 416 }
 417
 418 void iothread_destroy(IOThread *iothread)
 419 {
 420     object_unparent(OBJECT(iothread));
 421 }
 422
 423 /* Lookup IOThread by its id.  Only finds user-created objects, not internal
 424  * iothread_create() objects. */
 425 IOThread *iothread_by_id(const char *id)
 426 {
 427     return IOTHREAD(object_resolve_path_type(id, TYPE_IOTHREAD, NULL));
 428 }
 429
 430 bool qemu_in_iothread(void)
 431 {
 432     return qemu_get_current_aio_context() == qemu_get_aio_context() ?
 433                     false : true;
 434 }