#include "qom/object_interfaces.h"
#include "qemu/module.h"
#include "block/aio.h"
+#include "block/block.h"
#include "sysemu/iothread.h"
-#include "qmp-commands.h"
+#include "qapi/error.h"
+#include "qapi/qapi-commands-misc.h"
#include "qemu/error-report.h"
#include "qemu/rcu.h"
+#include "qemu/main-loop.h"
typedef ObjectClass IOThreadClass;
#define IOTHREAD_CLASS(klass) \
OBJECT_CLASS_CHECK(IOThreadClass, klass, TYPE_IOTHREAD)
+#ifdef CONFIG_POSIX
+/* Benchmark results from 2016 on NVMe SSD drives show max polling times around
+ * 16-32 microseconds yield IOPS improvements for both iodepth=1 and iodepth=32
+ * workloads.
+ */
+#define IOTHREAD_POLL_MAX_NS_DEFAULT 32768ULL
+#else
+#define IOTHREAD_POLL_MAX_NS_DEFAULT 0ULL
+#endif
+
+static __thread IOThread *my_iothread;
+
+AioContext *qemu_get_current_aio_context(void)
+{
+ return my_iothread ? my_iothread->ctx : qemu_get_aio_context();
+}
+
static void *iothread_run(void *opaque)
{
IOThread *iothread = opaque;
- bool blocking;
rcu_register_thread();
-
- qemu_mutex_lock(&iothread->init_done_lock);
+ /*
+ * g_main_context_push_thread_default() must be called before anything
+ * in this new thread uses glib.
+ */
+ g_main_context_push_thread_default(iothread->worker_context);
+ my_iothread = iothread;
iothread->thread_id = qemu_get_thread_id();
- qemu_cond_signal(&iothread->init_done_cond);
- qemu_mutex_unlock(&iothread->init_done_lock);
-
- while (!iothread->stopping) {
- aio_context_acquire(iothread->ctx);
- blocking = true;
- while (!iothread->stopping && aio_poll(iothread->ctx, blocking)) {
- /* Progress was made, keep going */
- blocking = false;
+ qemu_sem_post(&iothread->init_done_sem);
+
+ while (iothread->running) {
+ /*
+ * Note: from functional-wise the g_main_loop_run() below can
+ * already cover the aio_poll() events, but we can't run the
+ * main loop unconditionally because explicit aio_poll() here
+ * is faster than g_main_loop_run() when we do not need the
+ * gcontext at all (e.g., pure block layer iothreads). In
+ * other words, when we want to run the gcontext with the
+ * iothread we need to pay some performance for functionality.
+ */
+ aio_poll(iothread->ctx, true);
+
+ /*
+ * We must check the running state again in case it was
+ * changed in previous aio_poll()
+ */
+ if (iothread->running && atomic_read(&iothread->run_gcontext)) {
+ g_main_loop_run(iothread->main_loop);
}
- aio_context_release(iothread->ctx);
}
+ g_main_context_pop_thread_default(iothread->worker_context);
rcu_unregister_thread();
return NULL;
}
-static int iothread_stop(Object *object, void *opaque)
+/* Runs in iothread_run() thread */
+static void iothread_stop_bh(void *opaque)
{
- IOThread *iothread;
+ IOThread *iothread = opaque;
- iothread = (IOThread *)object_dynamic_cast(object, TYPE_IOTHREAD);
- if (!iothread || !iothread->ctx) {
- return 0;
+ iothread->running = false; /* stop iothread_run() */
+
+ if (iothread->main_loop) {
+ g_main_loop_quit(iothread->main_loop);
+ }
+}
+
+void iothread_stop(IOThread *iothread)
+{
+ if (!iothread->ctx || iothread->stopping) {
+ return;
}
iothread->stopping = true;
- aio_notify(iothread->ctx);
+ aio_bh_schedule_oneshot(iothread->ctx, iothread_stop_bh, iothread);
qemu_thread_join(&iothread->thread);
- return 0;
+}
+
+static void iothread_instance_init(Object *obj)
+{
+ IOThread *iothread = IOTHREAD(obj);
+
+ iothread->poll_max_ns = IOTHREAD_POLL_MAX_NS_DEFAULT;
+ iothread->thread_id = -1;
+ qemu_sem_init(&iothread->init_done_sem, 0);
+ /* By default, we don't run gcontext */
+ atomic_set(&iothread->run_gcontext, 0);
}
static void iothread_instance_finalize(Object *obj)
{
IOThread *iothread = IOTHREAD(obj);
- iothread_stop(obj, NULL);
- qemu_cond_destroy(&iothread->init_done_cond);
- qemu_mutex_destroy(&iothread->init_done_lock);
- if (!iothread->ctx) {
- return;
+ iothread_stop(iothread);
+
+ /*
+ * Before glib2 2.33.10, there is a glib2 bug that GSource context
+ * pointer may not be cleared even if the context has already been
+ * destroyed (while it should). Here let's free the AIO context
+ * earlier to bypass that glib bug.
+ *
+ * We can remove this comment after the minimum supported glib2
+ * version boosts to 2.33.10. Before that, let's free the
+ * GSources first before destroying any GMainContext.
+ */
+ if (iothread->ctx) {
+ aio_context_unref(iothread->ctx);
+ iothread->ctx = NULL;
+ }
+ if (iothread->worker_context) {
+ g_main_context_unref(iothread->worker_context);
+ iothread->worker_context = NULL;
+ g_main_loop_unref(iothread->main_loop);
+ iothread->main_loop = NULL;
}
- aio_context_unref(iothread->ctx);
+ qemu_sem_destroy(&iothread->init_done_sem);
+}
+
+static void iothread_init_gcontext(IOThread *iothread)
+{
+ GSource *source;
+
+ iothread->worker_context = g_main_context_new();
+ source = aio_get_g_source(iothread_get_aio_context(iothread));
+ g_source_attach(source, iothread->worker_context);
+ g_source_unref(source);
+ iothread->main_loop = g_main_loop_new(iothread->worker_context, TRUE);
}
static void iothread_complete(UserCreatable *obj, Error **errp)
char *name, *thread_name;
iothread->stopping = false;
- iothread->thread_id = -1;
+ iothread->running = true;
iothread->ctx = aio_context_new(&local_error);
if (!iothread->ctx) {
error_propagate(errp, local_error);
return;
}
- qemu_mutex_init(&iothread->init_done_lock);
- qemu_cond_init(&iothread->init_done_cond);
+ /*
+ * Init one GMainContext for the iothread unconditionally, even if
+ * it's not used
+ */
+ iothread_init_gcontext(iothread);
+
+ aio_context_set_poll_params(iothread->ctx,
+ iothread->poll_max_ns,
+ iothread->poll_grow,
+ iothread->poll_shrink,
+ &local_error);
+ if (local_error) {
+ error_propagate(errp, local_error);
+ aio_context_unref(iothread->ctx);
+ iothread->ctx = NULL;
+ return;
+ }
/* This assumes we are called from a thread with useful CPU affinity for us
* to inherit.
g_free(name);
/* Wait for initialization to complete */
- qemu_mutex_lock(&iothread->init_done_lock);
while (iothread->thread_id == -1) {
- qemu_cond_wait(&iothread->init_done_cond,
- &iothread->init_done_lock);
+ qemu_sem_wait(&iothread->init_done_sem);
+ }
+}
+
+typedef struct {
+ const char *name;
+ ptrdiff_t offset; /* field's byte offset in IOThread struct */
+} PollParamInfo;
+
+static PollParamInfo poll_max_ns_info = {
+ "poll-max-ns", offsetof(IOThread, poll_max_ns),
+};
+static PollParamInfo poll_grow_info = {
+ "poll-grow", offsetof(IOThread, poll_grow),
+};
+static PollParamInfo poll_shrink_info = {
+ "poll-shrink", offsetof(IOThread, poll_shrink),
+};
+
+static void iothread_get_poll_param(Object *obj, Visitor *v,
+ const char *name, void *opaque, Error **errp)
+{
+ IOThread *iothread = IOTHREAD(obj);
+ PollParamInfo *info = opaque;
+ int64_t *field = (void *)iothread + info->offset;
+
+ visit_type_int64(v, name, field, errp);
+}
+
+static void iothread_set_poll_param(Object *obj, Visitor *v,
+ const char *name, void *opaque, Error **errp)
+{
+ IOThread *iothread = IOTHREAD(obj);
+ PollParamInfo *info = opaque;
+ int64_t *field = (void *)iothread + info->offset;
+ Error *local_err = NULL;
+ int64_t value;
+
+ visit_type_int64(v, name, &value, &local_err);
+ if (local_err) {
+ goto out;
+ }
+
+ if (value < 0) {
+ error_setg(&local_err, "%s value must be in range [0, %"PRId64"]",
+ info->name, INT64_MAX);
+ goto out;
}
- qemu_mutex_unlock(&iothread->init_done_lock);
+
+ *field = value;
+
+ if (iothread->ctx) {
+ aio_context_set_poll_params(iothread->ctx,
+ iothread->poll_max_ns,
+ iothread->poll_grow,
+ iothread->poll_shrink,
+ &local_err);
+ }
+
+out:
+ error_propagate(errp, local_err);
}
static void iothread_class_init(ObjectClass *klass, void *class_data)
{
UserCreatableClass *ucc = USER_CREATABLE_CLASS(klass);
ucc->complete = iothread_complete;
+
+ object_class_property_add(klass, "poll-max-ns", "int",
+ iothread_get_poll_param,
+ iothread_set_poll_param,
+ NULL, &poll_max_ns_info, &error_abort);
+ object_class_property_add(klass, "poll-grow", "int",
+ iothread_get_poll_param,
+ iothread_set_poll_param,
+ NULL, &poll_grow_info, &error_abort);
+ object_class_property_add(klass, "poll-shrink", "int",
+ iothread_get_poll_param,
+ iothread_set_poll_param,
+ NULL, &poll_shrink_info, &error_abort);
}
static const TypeInfo iothread_info = {
.parent = TYPE_OBJECT,
.class_init = iothread_class_init,
.instance_size = sizeof(IOThread),
+ .instance_init = iothread_instance_init,
.instance_finalize = iothread_instance_finalize,
.interfaces = (InterfaceInfo[]) {
{TYPE_USER_CREATABLE},
info = g_new0(IOThreadInfo, 1);
info->id = iothread_get_id(iothread);
info->thread_id = iothread->thread_id;
+ info->poll_max_ns = iothread->poll_max_ns;
+ info->poll_grow = iothread->poll_grow;
+ info->poll_shrink = iothread->poll_shrink;
elem = g_new0(IOThreadInfoList, 1);
elem->value = info;
return head;
}
-void iothread_stop_all(void)
+GMainContext *iothread_get_g_main_context(IOThread *iothread)
{
- Object *container = object_get_objects_root();
+ atomic_set(&iothread->run_gcontext, 1);
+ aio_notify(iothread->ctx);
+ return iothread->worker_context;
+}
- object_child_foreach(container, iothread_stop, NULL);
+IOThread *iothread_create(const char *id, Error **errp)
+{
+ Object *obj;
+
+ obj = object_new_with_props(TYPE_IOTHREAD,
+ object_get_internal_root(),
+ id, errp, NULL);
+
+ return IOTHREAD(obj);
+}
+
+void iothread_destroy(IOThread *iothread)
+{
+ object_unparent(OBJECT(iothread));
+}
+
+/* Lookup IOThread by its id. Only finds user-created objects, not internal
+ * iothread_create() objects. */
+IOThread *iothread_by_id(const char *id)
+{
+ return IOTHREAD(object_resolve_path_type(id, TYPE_IOTHREAD, NULL));
}