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, IOThreadParamInfo *info, Error **errp)
 235 {
 236     IOThread *iothread = IOTHREAD(obj);
 237     int64_t *field = (void *)iothread + info->offset;
 238
 239     visit_type_int64(v, name, field, errp);
 240 }
 241
 242 static bool iothread_set_param(Object *obj, Visitor *v,
 243         const char *name, IOThreadParamInfo *info, Error **errp)
 244 {
 245     IOThread *iothread = IOTHREAD(obj);
 246     int64_t *field = (void *)iothread + info->offset;
 247     int64_t value;
 248
 249     if (!visit_type_int64(v, name, &value, errp)) {
 250         return false;
 251     }
 252
 253     if (value < 0) {
 254         error_setg(errp, "%s value must be in range [0, %" PRId64 "]",
 255                    info->name, INT64_MAX);
 256         return false;
 257     }
 258
 259     *field = value;
 260
 261     return true;
 262 }
 263
 264 static void iothread_get_poll_param(Object *obj, Visitor *v,
 265         const char *name, void *opaque, Error **errp)
 266 {
 267     IOThreadParamInfo *info = opaque;
 268
 269     iothread_get_param(obj, v, name, info, errp);
 270 }
 271
 272 static void iothread_set_poll_param(Object *obj, Visitor *v,
 273         const char *name, void *opaque, Error **errp)
 274 {
 275     IOThread *iothread = IOTHREAD(obj);
 276     IOThreadParamInfo *info = opaque;
 277
 278     if (!iothread_set_param(obj, v, name, info, 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     IOThreadParamInfo *info = opaque;
 295
 296     iothread_get_param(obj, v, name, info, errp);
 297 }
 298
 299 static void iothread_set_aio_param(Object *obj, Visitor *v,
 300         const char *name, void *opaque, Error **errp)
 301 {
 302     IOThread *iothread = IOTHREAD(obj);
 303     IOThreadParamInfo *info = opaque;
 304
 305     if (!iothread_set_param(obj, v, name, info, errp)) {
 306         return;
 307     }
 308
 309     if (iothread->ctx) {
 310         aio_context_set_aio_params(iothread->ctx,
 311                                    iothread->aio_max_batch,
 312                                    errp);
 313     }
 314 }
 315
 316 static void iothread_class_init(ObjectClass *klass, void *class_data)
 317 {
 318     UserCreatableClass *ucc = USER_CREATABLE_CLASS(klass);
 319     ucc->complete = iothread_complete;
 320
 321     object_class_property_add(klass, "poll-max-ns", "int",
 322                               iothread_get_poll_param,
 323                               iothread_set_poll_param,
 324                               NULL, &poll_max_ns_info);
 325     object_class_property_add(klass, "poll-grow", "int",
 326                               iothread_get_poll_param,
 327                               iothread_set_poll_param,
 328                               NULL, &poll_grow_info);
 329     object_class_property_add(klass, "poll-shrink", "int",
 330                               iothread_get_poll_param,
 331                               iothread_set_poll_param,
 332                               NULL, &poll_shrink_info);
 333     object_class_property_add(klass, "aio-max-batch", "int",
 334                               iothread_get_aio_param,
 335                               iothread_set_aio_param,
 336                               NULL, &aio_max_batch_info);
 337 }
 338
 339 static const TypeInfo iothread_info = {
 340     .name = TYPE_IOTHREAD,
 341     .parent = TYPE_OBJECT,
 342     .class_init = iothread_class_init,
 343     .instance_size = sizeof(IOThread),
 344     .instance_init = iothread_instance_init,
 345     .instance_finalize = iothread_instance_finalize,
 346     .interfaces = (InterfaceInfo[]) {
 347         {TYPE_USER_CREATABLE},
 348         {}
 349     },
 350 };
 351
 352 static void iothread_register_types(void)
 353 {
 354     type_register_static(&iothread_info);
 355 }
 356
 357 type_init(iothread_register_types)
 358
 359 char *iothread_get_id(IOThread *iothread)
 360 {
 361     return g_strdup(object_get_canonical_path_component(OBJECT(iothread)));
 362 }
 363
 364 AioContext *iothread_get_aio_context(IOThread *iothread)
 365 {
 366     return iothread->ctx;
 367 }
 368
 369 static int query_one_iothread(Object *object, void *opaque)
 370 {
 371     IOThreadInfoList ***tail = opaque;
 372     IOThreadInfo *info;
 373     IOThread *iothread;
 374
 375     iothread = (IOThread *)object_dynamic_cast(object, TYPE_IOTHREAD);
 376     if (!iothread) {
 377         return 0;
 378     }
 379
 380     info = g_new0(IOThreadInfo, 1);
 381     info->id = iothread_get_id(iothread);
 382     info->thread_id = iothread->thread_id;
 383     info->poll_max_ns = iothread->poll_max_ns;
 384     info->poll_grow = iothread->poll_grow;
 385     info->poll_shrink = iothread->poll_shrink;
 386     info->aio_max_batch = iothread->aio_max_batch;
 387
 388     QAPI_LIST_APPEND(*tail, info);
 389     return 0;
 390 }
 391
 392 IOThreadInfoList *qmp_query_iothreads(Error **errp)
 393 {
 394     IOThreadInfoList *head = NULL;
 395     IOThreadInfoList **prev = &head;
 396     Object *container = object_get_objects_root();
 397
 398     object_child_foreach(container, query_one_iothread, &prev);
 399     return head;
 400 }
 401
 402 GMainContext *iothread_get_g_main_context(IOThread *iothread)
 403 {
 404     qatomic_set(&iothread->run_gcontext, 1);
 405     aio_notify(iothread->ctx);
 406     return iothread->worker_context;
 407 }
 408
 409 IOThread *iothread_create(const char *id, Error **errp)
 410 {
 411     Object *obj;
 412
 413     obj = object_new_with_props(TYPE_IOTHREAD,
 414                                 object_get_internal_root(),
 415                                 id, errp, NULL);
 416
 417     return IOTHREAD(obj);
 418 }
 419
 420 void iothread_destroy(IOThread *iothread)
 421 {
 422     object_unparent(OBJECT(iothread));
 423 }
 424
 425 /* Lookup IOThread by its id.  Only finds user-created objects, not internal
 426  * iothread_create() objects. */
 427 IOThread *iothread_by_id(const char *id)
 428 {
 429     return IOTHREAD(object_resolve_path_type(id, TYPE_IOTHREAD, NULL));
 430 }
 431
 432 bool qemu_in_iothread(void)
 433 {
 434     return qemu_get_current_aio_context() == qemu_get_aio_context() ?
 435                     false : true;
 436 }