gstreamer/subprojects/gst-plugins-bad/sys/nvcodec/gstcudaipcsrc.cpp
Seungha Yang 6fa405c1e2 cudaipcclient: Protect IPC handle import/close with global lock
Protect import/close with signle lock to avoid importing a IPC handle
while it's being closed by another cudaipcsrc from other thread.
Also fixing cuda context leak

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/5230>
2023-08-25 10:06:58 +00:00

570 lines
17 KiB
C++

/* GStreamer
* Copyright (C) 2023 Seungha Yang <seungha@centricular.com>
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Library General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This library is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Library General Public License for more details.
*
* You should have received a copy of the GNU Library General Public
* License along with this library; if not, write to the
* Free Software Foundation, Inc., 51 Franklin St, Fifth Floor,
* Boston, MA 02110-1301, USA.
*/
/**
* SECTION:element-cudaipcsrc
* @title: cudaipcsrc
* @short_description: CUDA Inter Process Communication (IPC) src
*
* cudaipcsrc imports CUDA memory exported by peer cudaipcsrc element
*
* ## Example launch line
* ```
* gst-launch-1.0 cudaipcsrc ! queue ! cudadownload ! videoconvert ! queue ! autovideosink
* ```
*
* Since: 1.24
*/
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif
#include "gstcudaipcsrc.h"
#include "gstcudaformat.h"
#include <mutex>
#include <string>
#ifdef G_OS_WIN32
#include "gstcudaipcclient_win32.h"
#else
#include "gstcudaipcclient_unix.h"
#endif
GST_DEBUG_CATEGORY_STATIC (gst_cuda_ipc_src_debug);
#define GST_CAT_DEFAULT gst_cuda_ipc_src_debug
static GstStaticPadTemplate src_template = GST_STATIC_PAD_TEMPLATE ("src",
GST_PAD_SRC,
GST_PAD_ALWAYS,
GST_STATIC_CAPS (GST_VIDEO_CAPS_MAKE_WITH_FEATURES
(GST_CAPS_FEATURE_MEMORY_CUDA_MEMORY,
GST_CUDA_FORMATS)));
enum
{
PROP_0,
PROP_DEVICE_ID,
PROP_ADDRESS,
PROP_PROCESSING_DEADLINE,
PROP_IO_MODE,
PROP_CONN_TIMEOUT,
PROP_BUFFER_SIZE,
};
#define DEFAULT_DEVICE_ID -1
#ifdef G_OS_WIN32
#define DEFAULT_ADDRESS "\\\\.\\pipe\\gst.cuda.ipc"
#else
#define DEFAULT_ADDRESS "/tmp/gst.cuda.ipc"
#endif
#define DEFAULT_PROCESSING_DEADLINE (20 * GST_MSECOND)
#define DEFAULT_IO_MODE GST_CUDA_IPC_IO_COPY
#define DEFAULT_CONN_TIMEOUT 5
#define DEFAULT_BUFFER_SIZE 3
/* *INDENT-OFF* */
struct GstCudaIpcSrcPrivate
{
GstCudaContext *context = nullptr;
GstCudaStream *stream = nullptr;
GstCudaIpcClient *client = nullptr;
GstCaps *caps = nullptr;
GstVideoInfo info;
std::mutex lock;
bool flushing = false;
/* properties */
gint device_id = DEFAULT_DEVICE_ID;
std::string address = DEFAULT_ADDRESS;
GstClockTime processing_deadline = DEFAULT_PROCESSING_DEADLINE;
GstCudaIpcIOMode io_mode = DEFAULT_IO_MODE;
guint conn_timeout = DEFAULT_CONN_TIMEOUT;
guint buffer_size = DEFAULT_BUFFER_SIZE;
};
/* *INDENT-ON* */
struct _GstCudaIpcSrc
{
GstBaseSrc parent;
GstCudaIpcSrcPrivate *priv;
};
static void gst_cuda_ipc_src_finalize (GObject * object);
static void gst_cuda_ipc_src_set_property (GObject * object,
guint prop_id, const GValue * value, GParamSpec * pspec);
static void gst_win32_video_src_get_property (GObject * object, guint prop_id,
GValue * value, GParamSpec * pspec);
static GstClock *gst_cuda_ipc_src_provide_clock (GstElement * elem);
static void gst_cuda_ipc_src_set_context (GstElement * elem,
GstContext * context);
static gboolean gst_cuda_ipc_src_start (GstBaseSrc * src);
static gboolean gst_cuda_ipc_src_stop (GstBaseSrc * src);
static gboolean gst_cuda_ipc_src_unlock (GstBaseSrc * src);
static gboolean gst_cuda_ipc_src_unlock_stop (GstBaseSrc * src);
static gboolean gst_cuda_ipc_src_query (GstBaseSrc * src, GstQuery * query);
static GstCaps *gst_cuda_ipc_src_get_caps (GstBaseSrc * src, GstCaps * filter);
static GstCaps *gst_cuda_ipc_src_fixate (GstBaseSrc * src, GstCaps * caps);
static GstFlowReturn gst_cuda_ipc_src_create (GstBaseSrc * src, guint64 offset,
guint size, GstBuffer ** buf);
#define gst_cuda_ipc_src_parent_class parent_class
G_DEFINE_TYPE (GstCudaIpcSrc, gst_cuda_ipc_src, GST_TYPE_BASE_SRC);
static void
gst_cuda_ipc_src_class_init (GstCudaIpcSrcClass * klass)
{
GObjectClass *object_class = G_OBJECT_CLASS (klass);
GstElementClass *element_class = GST_ELEMENT_CLASS (klass);
GstBaseSrcClass *src_class = GST_BASE_SRC_CLASS (klass);
object_class->finalize = gst_cuda_ipc_src_finalize;
object_class->set_property = gst_cuda_ipc_src_set_property;
object_class->get_property = gst_win32_video_src_get_property;
g_object_class_install_property (object_class, PROP_DEVICE_ID,
g_param_spec_int ("cuda-device-id", "CUDA Device ID",
"CUDA device id to use (-1 = auto)", -1, G_MAXINT, DEFAULT_DEVICE_ID,
(GParamFlags) (G_PARAM_READWRITE | GST_PARAM_MUTABLE_READY |
G_PARAM_STATIC_STRINGS)));
g_object_class_install_property (object_class, PROP_ADDRESS,
g_param_spec_string ("address", "Address",
"Server address. Specifies name of WIN32 named pipe "
"or unix domain socket path on Linux",
DEFAULT_ADDRESS, (GParamFlags) (G_PARAM_READWRITE |
G_PARAM_STATIC_STRINGS | GST_PARAM_MUTABLE_READY)));
g_object_class_install_property (object_class, PROP_PROCESSING_DEADLINE,
g_param_spec_uint64 ("processing-deadline", "Processing deadline",
"Maximum processing time for a buffer in nanoseconds", 0, G_MAXUINT64,
DEFAULT_PROCESSING_DEADLINE, (GParamFlags) (G_PARAM_READWRITE |
G_PARAM_STATIC_STRINGS | GST_PARAM_MUTABLE_PLAYING)));
g_object_class_install_property (object_class, PROP_IO_MODE,
g_param_spec_enum ("io-mode", "IO Mode",
"Memory I/O mode to use. This option will be ignored if the selected "
"IPC mode is mmap",
GST_TYPE_CUDA_IPC_IO_MODE, DEFAULT_IO_MODE,
(GParamFlags) (G_PARAM_READWRITE | GST_PARAM_MUTABLE_READY |
G_PARAM_STATIC_STRINGS)));
g_object_class_install_property (object_class, PROP_CONN_TIMEOUT,
g_param_spec_uint ("connection-timeout", "Connection Timeout",
"Connection timeout in seconds (0 = never timeout)", 0, G_MAXINT,
DEFAULT_CONN_TIMEOUT,
(GParamFlags) (G_PARAM_READWRITE | GST_PARAM_MUTABLE_READY |
G_PARAM_STATIC_STRINGS)));
g_object_class_install_property (object_class, PROP_BUFFER_SIZE,
g_param_spec_uint ("buffer-size", "Buffer Size",
"Size of internal buffer", 1, G_MAXINT,
DEFAULT_BUFFER_SIZE,
(GParamFlags) (G_PARAM_READWRITE | GST_PARAM_MUTABLE_READY |
G_PARAM_STATIC_STRINGS)));
gst_element_class_set_static_metadata (element_class,
"CUDA IPC Src", "Source/Video",
"Receive CUDA memory from the cudaipcsrc element",
"Seungha Yang <seungha@centricular.com>");
gst_element_class_add_static_pad_template (element_class, &src_template);
element_class->provide_clock =
GST_DEBUG_FUNCPTR (gst_cuda_ipc_src_provide_clock);
element_class->set_context = GST_DEBUG_FUNCPTR (gst_cuda_ipc_src_set_context);
src_class->start = GST_DEBUG_FUNCPTR (gst_cuda_ipc_src_start);
src_class->stop = GST_DEBUG_FUNCPTR (gst_cuda_ipc_src_stop);
src_class->unlock = GST_DEBUG_FUNCPTR (gst_cuda_ipc_src_unlock);
src_class->unlock_stop = GST_DEBUG_FUNCPTR (gst_cuda_ipc_src_unlock_stop);
src_class->query = GST_DEBUG_FUNCPTR (gst_cuda_ipc_src_query);
src_class->get_caps = GST_DEBUG_FUNCPTR (gst_cuda_ipc_src_get_caps);
src_class->fixate = GST_DEBUG_FUNCPTR (gst_cuda_ipc_src_fixate);
src_class->create = GST_DEBUG_FUNCPTR (gst_cuda_ipc_src_create);
GST_DEBUG_CATEGORY_INIT (gst_cuda_ipc_src_debug, "cudaipcsrc",
0, "cudaipcsrc");
gst_type_mark_as_plugin_api (GST_TYPE_CUDA_IPC_IO_MODE,
(GstPluginAPIFlags) 0);
}
static void
gst_cuda_ipc_src_init (GstCudaIpcSrc * self)
{
gst_base_src_set_format (GST_BASE_SRC (self), GST_FORMAT_TIME);
gst_base_src_set_live (GST_BASE_SRC (self), TRUE);
self->priv = new GstCudaIpcSrcPrivate ();
GST_OBJECT_FLAG_SET (self, GST_ELEMENT_FLAG_PROVIDE_CLOCK);
GST_OBJECT_FLAG_SET (self, GST_ELEMENT_FLAG_REQUIRE_CLOCK);
}
static void
gst_cuda_ipc_src_finalize (GObject * object)
{
GstCudaIpcSrc *self = GST_CUDA_IPC_SRC (object);
delete self->priv;
G_OBJECT_CLASS (parent_class)->finalize (object);
}
static void
gst_cuda_ipc_src_set_property (GObject * object, guint prop_id,
const GValue * value, GParamSpec * pspec)
{
GstCudaIpcSrc *self = GST_CUDA_IPC_SRC (object);
GstCudaIpcSrcPrivate *priv = self->priv;
std::unique_lock < std::mutex > lk (priv->lock);
switch (prop_id) {
case PROP_DEVICE_ID:
priv->device_id = g_value_get_int (value);
break;
case PROP_ADDRESS:
{
const gchar *address = g_value_get_string (value);
priv->address.clear ();
if (address)
priv->address = address;
break;
}
case PROP_PROCESSING_DEADLINE:
{
GstClockTime prev_val, new_val;
prev_val = priv->processing_deadline;
new_val = g_value_get_uint64 (value);
priv->processing_deadline = new_val;
if (prev_val != new_val) {
lk.unlock ();
GST_DEBUG_OBJECT (self, "Posting latency message");
gst_element_post_message (GST_ELEMENT_CAST (self),
gst_message_new_latency (GST_OBJECT_CAST (self)));
}
break;
}
case PROP_IO_MODE:
priv->io_mode = (GstCudaIpcIOMode) g_value_get_enum (value);
break;
case PROP_CONN_TIMEOUT:
priv->conn_timeout = g_value_get_uint (value);
break;
case PROP_BUFFER_SIZE:
priv->buffer_size = g_value_get_uint (value);
break;
default:
G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
break;
}
}
static void
gst_win32_video_src_get_property (GObject * object, guint prop_id,
GValue * value, GParamSpec * pspec)
{
GstCudaIpcSrc *self = GST_CUDA_IPC_SRC (object);
GstCudaIpcSrcPrivate *priv = self->priv;
std::lock_guard < std::mutex > lk (priv->lock);
switch (prop_id) {
case PROP_DEVICE_ID:
g_value_set_int (value, priv->device_id);
break;
case PROP_ADDRESS:
g_value_set_string (value, priv->address.c_str ());
break;
case PROP_PROCESSING_DEADLINE:
g_value_set_uint64 (value, priv->processing_deadline);
break;
case PROP_IO_MODE:
g_value_set_enum (value, priv->io_mode);
break;
case PROP_CONN_TIMEOUT:
g_value_set_uint (value, priv->conn_timeout);
break;
case PROP_BUFFER_SIZE:
g_value_set_uint (value, priv->buffer_size);
break;
default:
G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
break;
}
}
static GstClock *
gst_cuda_ipc_src_provide_clock (GstElement * elem)
{
return gst_system_clock_obtain ();
}
static void
gst_cuda_ipc_src_set_context (GstElement * elem, GstContext * context)
{
GstCudaIpcSrc *self = GST_CUDA_IPC_SRC (elem);
GstCudaIpcSrcPrivate *priv = self->priv;
gst_cuda_handle_set_context (elem, context, priv->device_id, &priv->context);
GST_ELEMENT_CLASS (parent_class)->set_context (elem, context);
}
static gboolean
gst_cuda_ipc_src_start (GstBaseSrc * src)
{
GstCudaIpcSrc *self = GST_CUDA_IPC_SRC (src);
GstCudaIpcSrcPrivate *priv = self->priv;
GST_DEBUG_OBJECT (self, "Start");
if (!gst_cuda_ensure_element_context (GST_ELEMENT_CAST (self),
priv->device_id, &priv->context)) {
GST_ERROR_OBJECT (self, "Couldn't get CUDA context");
return FALSE;
}
priv->stream = gst_cuda_stream_new (priv->context);
std::lock_guard < std::mutex > lk (priv->lock);
priv->client = gst_cuda_ipc_client_new (priv->address.c_str (), priv->context,
priv->stream, priv->io_mode, priv->conn_timeout, priv->buffer_size - 1);
return TRUE;
}
static gboolean
gst_cuda_ipc_src_stop (GstBaseSrc * src)
{
GstCudaIpcSrc *self = GST_CUDA_IPC_SRC (src);
GstCudaIpcSrcPrivate *priv = self->priv;
std::lock_guard < std::mutex > lk (priv->lock);
GST_DEBUG_OBJECT (self, "Stop");
if (priv->client)
gst_cuda_ipc_client_stop (priv->client);
gst_clear_object (&priv->client);
gst_clear_cuda_stream (&priv->stream);
gst_clear_object (&priv->context);
gst_clear_caps (&priv->caps);
return TRUE;
}
static gboolean
gst_cuda_ipc_src_unlock (GstBaseSrc * src)
{
GstCudaIpcSrc *self = GST_CUDA_IPC_SRC (src);
GstCudaIpcSrcPrivate *priv = self->priv;
GST_DEBUG_OBJECT (self, "Unlock");
std::lock_guard < std::mutex > lk (priv->lock);
priv->flushing = true;
if (priv->client)
gst_cuda_ipc_client_set_flushing (priv->client, true);
GST_DEBUG_OBJECT (self, "Unlocked");
return TRUE;
}
static gboolean
gst_cuda_ipc_src_unlock_stop (GstBaseSrc * src)
{
GstCudaIpcSrc *self = GST_CUDA_IPC_SRC (src);
GstCudaIpcSrcPrivate *priv = self->priv;
GST_DEBUG_OBJECT (self, "Unlock stop");
std::lock_guard < std::mutex > lk (priv->lock);
priv->flushing = false;
if (priv->client)
gst_cuda_ipc_client_set_flushing (priv->client, false);
GST_DEBUG_OBJECT (self, "Unlock stopped");
return TRUE;
}
static gboolean
gst_cuda_ipc_src_query (GstBaseSrc * src, GstQuery * query)
{
GstCudaIpcSrc *self = GST_CUDA_IPC_SRC (src);
GstCudaIpcSrcPrivate *priv = self->priv;
switch (GST_QUERY_TYPE (query)) {
case GST_QUERY_LATENCY:
{
std::lock_guard < std::mutex > lk (priv->lock);
if (GST_CLOCK_TIME_IS_VALID (priv->processing_deadline)) {
gst_query_set_latency (query, TRUE, priv->processing_deadline,
GST_CLOCK_TIME_NONE);
} else {
gst_query_set_latency (query, TRUE, 0, 0);
}
return TRUE;
}
case GST_QUERY_CONTEXT:
if (gst_cuda_handle_context_query (GST_ELEMENT (self), query,
priv->context)) {
return TRUE;
}
break;
default:
break;
}
return GST_BASE_SRC_CLASS (parent_class)->query (src, query);
}
static GstCaps *
gst_cuda_ipc_src_get_caps (GstBaseSrc * src, GstCaps * filter)
{
GstCudaIpcSrc *self = GST_CUDA_IPC_SRC (src);
GstCudaIpcSrcPrivate *priv = self->priv;
GstCudaIpcClient *client = nullptr;
GstCaps *caps = nullptr;
GST_DEBUG_OBJECT (self, "Get caps");
priv->lock.lock ();
if (priv->caps)
caps = gst_caps_ref (priv->caps);
else if (priv->client)
client = (GstCudaIpcClient *) gst_object_ref (priv->client);
priv->lock.unlock ();
if (!caps && client)
caps = gst_cuda_ipc_client_get_caps (priv->client);
if (!caps)
caps = gst_pad_get_pad_template_caps (GST_BASE_SRC_PAD (src));
if (filter) {
GstCaps *tmp = gst_caps_intersect_full (filter,
caps, GST_CAPS_INTERSECT_FIRST);
gst_caps_unref (caps);
caps = tmp;
}
gst_clear_object (&client);
GST_DEBUG_OBJECT (self, "Returning caps %" GST_PTR_FORMAT, caps);
return caps;
}
static GstCaps *
gst_cuda_ipc_src_fixate (GstBaseSrc * src, GstCaps * caps)
{
/* We don't negotiate with server. In here, we do fixate resolution to
* 320 x 240 (same as default of videotestsrc) which makes a little more
* sense than 1x1 */
caps = gst_caps_make_writable (caps);
for (guint i = 0; i < gst_caps_get_size (caps); i++) {
GstStructure *s = gst_caps_get_structure (caps, i);
gst_structure_fixate_field_nearest_int (s, "width", 320);
gst_structure_fixate_field_nearest_int (s, "height", 240);
}
return gst_caps_fixate (caps);
}
static GstFlowReturn
gst_cuda_ipc_src_create (GstBaseSrc * src, guint64 offset, guint size,
GstBuffer ** buf)
{
GstCudaIpcSrc *self = GST_CUDA_IPC_SRC (src);
GstCudaIpcSrcPrivate *priv = self->priv;
GstFlowReturn ret;
GstSample *sample = nullptr;
GstCaps *caps;
GstClock *clock;
bool is_system_clock = true;
GstClockTime pts;
GstClockTime base_time;
GstClockTime now_system;
GstClockTime now_gst;
GstClockTime remote_pts;
GstBuffer *buffer;
ret = gst_cuda_ipc_client_run (priv->client);
if (ret != GST_FLOW_OK)
return ret;
ret = gst_cuda_ipc_client_get_sample (priv->client, &sample);
if (ret != GST_FLOW_OK)
return ret;
now_system = gst_util_get_timestamp ();
clock = gst_element_get_clock (GST_ELEMENT_CAST (self));
now_gst = gst_clock_get_time (clock);
base_time = GST_ELEMENT_CAST (self)->base_time;
is_system_clock = gst_cuda_ipc_clock_is_system (clock);
gst_object_unref (clock);
buffer = gst_sample_get_buffer (sample);
remote_pts = GST_BUFFER_PTS (buffer);
if (!is_system_clock) {
GstClockTimeDiff now_pts = now_gst - base_time + remote_pts - now_system;
if (now_pts >= 0)
pts = now_pts;
else
pts = 0;
} else {
if (remote_pts >= base_time) {
pts = remote_pts - base_time;
} else {
GST_WARNING_OBJECT (self,
"Remote clock is smaller than our base time, remote %"
GST_TIME_FORMAT ", base_time %" GST_TIME_FORMAT,
GST_TIME_ARGS (remote_pts), GST_TIME_ARGS (base_time));
pts = 0;
}
}
GST_BUFFER_PTS (buffer) = pts;
std::unique_lock < std::mutex > lk (priv->lock);
caps = gst_sample_get_caps (sample);
if (!priv->caps || !gst_caps_is_equal (priv->caps, caps)) {
gst_caps_replace (&priv->caps, caps);
lk.unlock ();
gst_base_src_set_caps (src, priv->caps);
}
*buf = gst_buffer_ref (buffer);
gst_sample_unref (sample);
return GST_FLOW_OK;
}