mirror of
https://gitlab.freedesktop.org/pulseaudio/pulseaudio.git
synced 2025-10-29 05:40:23 -04:00
This patch replaces every occurrence of ')\n{' with ') {'.
Command used for this:
find . -type d \( -name ffmpeg \) -prune -o \
-regex '\(.*\.[hc]\|.*\.cc\)' \
-a -not -name core-util.c -a -not \
-name adrian-aec.c -a -not -name g711.c \
-exec sed -i -e '/)$/{N;s/)\n{$/) {/}' {} \;
The excluded files are mirrored files from external sources.
900 lines
28 KiB
C
900 lines
28 KiB
C
/***
|
|
This file is part of PulseAudio.
|
|
|
|
Copyright 2009,2010 Daniel Mack <daniel@caiaq.de>
|
|
|
|
PulseAudio is free software; you can redistribute it and/or modify
|
|
it under the terms of the GNU Lesser General Public License as published
|
|
by the Free Software Foundation; either version 2.1 of the License,
|
|
or (at your option) any later version.
|
|
|
|
PulseAudio 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
|
|
General Public License for more details.
|
|
|
|
You should have received a copy of the GNU Lesser General Public License
|
|
along with PulseAudio; if not, write to the Free Software
|
|
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
|
|
USA.
|
|
***/
|
|
|
|
/* TODO:
|
|
- implement hardware volume controls
|
|
- handle audio device stream format changes (will require changes to the core)
|
|
- add an "off" mode that removes all sinks and sources
|
|
*/
|
|
|
|
#ifdef HAVE_CONFIG_H
|
|
#include <config.h>
|
|
#endif
|
|
|
|
#include <pulse/xmalloc.h>
|
|
|
|
#include <pulsecore/sink.h>
|
|
#include <pulsecore/source.h>
|
|
#include <pulsecore/module.h>
|
|
#include <pulsecore/sample-util.h>
|
|
#include <pulsecore/core-util.h>
|
|
#include <pulsecore/modargs.h>
|
|
#include <pulsecore/log.h>
|
|
#include <pulsecore/macro.h>
|
|
#include <pulsecore/llist.h>
|
|
#include <pulsecore/card.h>
|
|
#include <pulsecore/strbuf.h>
|
|
#include <pulsecore/thread.h>
|
|
#include <pulsecore/thread-mq.h>
|
|
#include <pulsecore/i18n.h>
|
|
|
|
#include <CoreAudio/CoreAudio.h>
|
|
#include <CoreAudio/CoreAudioTypes.h>
|
|
#include <CoreAudio/AudioHardware.h>
|
|
|
|
#include "module-coreaudio-device-symdef.h"
|
|
|
|
#define DEFAULT_FRAMES_PER_IOPROC 512
|
|
|
|
PA_MODULE_AUTHOR("Daniel Mack");
|
|
PA_MODULE_DESCRIPTION("CoreAudio device");
|
|
PA_MODULE_VERSION(PACKAGE_VERSION);
|
|
PA_MODULE_LOAD_ONCE(FALSE);
|
|
PA_MODULE_USAGE("object_id=<the CoreAudio device id> "
|
|
"ioproc_frames=<audio frames per IOProc call> ");
|
|
|
|
static const char* const valid_modargs[] = {
|
|
"object_id",
|
|
"ioproc_frames",
|
|
NULL
|
|
};
|
|
|
|
enum {
|
|
CA_MESSAGE_RENDER = PA_SINK_MESSAGE_MAX,
|
|
};
|
|
|
|
typedef struct coreaudio_sink coreaudio_sink;
|
|
typedef struct coreaudio_source coreaudio_source;
|
|
|
|
struct userdata {
|
|
AudioObjectID object_id;
|
|
AudioDeviceIOProcID proc_id;
|
|
|
|
pa_thread_mq thread_mq;
|
|
pa_asyncmsgq *async_msgq;
|
|
|
|
pa_rtpoll *rtpoll;
|
|
pa_thread *thread;
|
|
|
|
pa_module *module;
|
|
pa_card *card;
|
|
pa_bool_t running;
|
|
|
|
char *device_name, *vendor_name;
|
|
|
|
const AudioBufferList *render_input_data;
|
|
AudioBufferList *render_output_data;
|
|
|
|
AudioStreamBasicDescription stream_description;
|
|
|
|
PA_LLIST_HEAD(coreaudio_sink, sinks);
|
|
PA_LLIST_HEAD(coreaudio_source, sources);
|
|
};
|
|
|
|
struct coreaudio_sink {
|
|
pa_sink *pa_sink;
|
|
struct userdata *userdata;
|
|
|
|
char *name;
|
|
unsigned int channel_idx;
|
|
pa_bool_t active;
|
|
|
|
pa_channel_map map;
|
|
pa_sample_spec ss;
|
|
|
|
PA_LLIST_FIELDS(coreaudio_sink);
|
|
};
|
|
|
|
struct coreaudio_source {
|
|
pa_source *pa_source;
|
|
struct userdata *userdata;
|
|
|
|
char *name;
|
|
unsigned int channel_idx;
|
|
pa_bool_t active;
|
|
|
|
pa_channel_map map;
|
|
pa_sample_spec ss;
|
|
|
|
PA_LLIST_FIELDS(coreaudio_source);
|
|
};
|
|
|
|
static int card_set_profile(pa_card *c, pa_card_profile *new_profile) {
|
|
return 0;
|
|
}
|
|
|
|
static OSStatus io_render_proc (AudioDeviceID device,
|
|
const AudioTimeStamp *now,
|
|
const AudioBufferList *inputData,
|
|
const AudioTimeStamp *inputTime,
|
|
AudioBufferList *outputData,
|
|
const AudioTimeStamp *outputTime,
|
|
void *clientData) {
|
|
struct userdata *u = clientData;
|
|
|
|
pa_assert(u);
|
|
pa_assert(device == u->object_id);
|
|
|
|
u->render_input_data = inputData;
|
|
u->render_output_data = outputData;
|
|
|
|
if (u->sinks)
|
|
pa_assert_se(pa_asyncmsgq_send(u->async_msgq, PA_MSGOBJECT(u->sinks->pa_sink),
|
|
CA_MESSAGE_RENDER, NULL, 0, NULL) == 0);
|
|
|
|
if (u->sources)
|
|
pa_assert_se(pa_asyncmsgq_send(u->async_msgq, PA_MSGOBJECT(u->sources->pa_source),
|
|
CA_MESSAGE_RENDER, NULL, 0, NULL) == 0);
|
|
|
|
return 0;
|
|
}
|
|
|
|
static OSStatus ca_stream_format_changed(AudioObjectID objectID,
|
|
UInt32 numberAddresses,
|
|
const AudioObjectPropertyAddress addresses[],
|
|
void *clientData) {
|
|
struct userdata *u = clientData;
|
|
UInt32 i;
|
|
|
|
pa_assert(u);
|
|
|
|
/* REVISIT: PA can't currently handle external format change requests.
|
|
* Hence, we set the original format back in this callback to avoid horrible audio artefacts.
|
|
* The device settings will appear to be 'locked' for any application as long as the PA daemon is running.
|
|
* Once we're able to propagate such events up in the core, this needs to be changed. */
|
|
|
|
for (i = 0; i < numberAddresses; i++)
|
|
AudioObjectSetPropertyData(objectID, addresses + i, 0, NULL, sizeof(u->stream_description), &u->stream_description);
|
|
|
|
return 0;
|
|
}
|
|
|
|
static pa_usec_t get_latency_us(pa_object *o) {
|
|
struct userdata *u;
|
|
pa_sample_spec *ss;
|
|
bool is_source;
|
|
UInt32 v = 0, total = 0;
|
|
UInt32 err, size = sizeof(v);
|
|
AudioObjectPropertyAddress property_address;
|
|
AudioObjectID stream_id;
|
|
|
|
if (pa_sink_isinstance(o)) {
|
|
coreaudio_sink *sink = PA_SINK(o)->userdata;
|
|
|
|
u = sink->userdata;
|
|
ss = &sink->ss;
|
|
is_source = FALSE;
|
|
} else if (pa_source_isinstance(o)) {
|
|
coreaudio_source *source = PA_SOURCE(o)->userdata;
|
|
|
|
u = source->userdata;
|
|
ss = &source->ss;
|
|
is_source = TRUE;
|
|
} else
|
|
pa_assert_not_reached();
|
|
|
|
pa_assert(u);
|
|
|
|
property_address.mScope = is_source ? kAudioDevicePropertyScopeInput : kAudioDevicePropertyScopeOutput;
|
|
property_address.mElement = kAudioObjectPropertyElementMaster;
|
|
|
|
/* get the device latency */
|
|
property_address.mSelector = kAudioDevicePropertyLatency;
|
|
size = sizeof(v);
|
|
err = AudioObjectGetPropertyData(u->object_id, &property_address, 0, NULL, &size, &v);
|
|
if (!err)
|
|
total += v;
|
|
else
|
|
pa_log_warn("Failed to get device latency: %i", err);
|
|
|
|
/* the IOProc buffer size */
|
|
property_address.mSelector = kAudioDevicePropertyBufferFrameSize;
|
|
size = sizeof(v);
|
|
err = AudioObjectGetPropertyData(u->object_id, &property_address, 0, NULL, &size, &v);
|
|
if (!err)
|
|
total += v;
|
|
else
|
|
pa_log_warn("Failed to get buffer frame size: %i", err);
|
|
|
|
/* IOProc safety offset - this value is the same for both directions, hence we divide it by 2 */
|
|
property_address.mSelector = kAudioDevicePropertySafetyOffset;
|
|
size = sizeof(v);
|
|
err = AudioObjectGetPropertyData(u->object_id, &property_address, 0, NULL, &size, &v);
|
|
if (!err)
|
|
total += v / 2;
|
|
else
|
|
pa_log_warn("Failed to get safety offset: %i", err);
|
|
|
|
/* get the stream latency.
|
|
* FIXME: this assumes the stream latency is the same for all streams */
|
|
property_address.mSelector = kAudioDevicePropertyStreams;
|
|
size = sizeof(stream_id);
|
|
err = AudioObjectGetPropertyData(u->object_id, &property_address, 0, NULL, &size, &stream_id);
|
|
if (!err) {
|
|
property_address.mSelector = kAudioStreamPropertyLatency;
|
|
property_address.mScope = kAudioObjectPropertyScopeGlobal;
|
|
size = sizeof(v);
|
|
err = AudioObjectGetPropertyData(stream_id, &property_address, 0, NULL, &size, &v);
|
|
if (!err)
|
|
total += v;
|
|
else
|
|
pa_log_warn("Failed to get stream latency: %i", err);
|
|
} else
|
|
pa_log_warn("Failed to get streams: %i", err);
|
|
|
|
return pa_bytes_to_usec(total * pa_frame_size(ss), ss);
|
|
}
|
|
|
|
static void ca_device_check_device_state(struct userdata *u) {
|
|
coreaudio_sink *ca_sink;
|
|
coreaudio_source *ca_source;
|
|
pa_bool_t active = FALSE;
|
|
|
|
pa_assert(u);
|
|
|
|
for (ca_sink = u->sinks; ca_sink; ca_sink = ca_sink->next)
|
|
if (ca_sink->active)
|
|
active = TRUE;
|
|
|
|
for (ca_source = u->sources; ca_source; ca_source = ca_source->next)
|
|
if (ca_source->active)
|
|
active = TRUE;
|
|
|
|
if (active && !u->running)
|
|
AudioDeviceStart(u->object_id, u->proc_id);
|
|
else if (!active && u->running)
|
|
AudioDeviceStop(u->object_id, u->proc_id);
|
|
|
|
u->running = active;
|
|
}
|
|
|
|
static int sink_process_msg(pa_msgobject *o, int code, void *data, int64_t offset, pa_memchunk *chunk) {
|
|
coreaudio_sink *sink = PA_SINK(o)->userdata;
|
|
struct userdata *u = sink->userdata;
|
|
unsigned int i;
|
|
pa_memchunk audio_chunk;
|
|
|
|
switch (code) {
|
|
case CA_MESSAGE_RENDER: {
|
|
/* audio out */
|
|
for (i = 0; i < u->render_output_data->mNumberBuffers; i++) {
|
|
AudioBuffer *buf = u->render_output_data->mBuffers + i;
|
|
|
|
pa_assert(sink);
|
|
|
|
if (PA_SINK_IS_OPENED(sink->pa_sink->thread_info.state)) {
|
|
audio_chunk.memblock = pa_memblock_new_fixed(u->module->core->mempool, buf->mData, buf->mDataByteSize, FALSE);
|
|
audio_chunk.length = buf->mDataByteSize;
|
|
audio_chunk.index = 0;
|
|
|
|
pa_sink_render_into_full(sink->pa_sink, &audio_chunk);
|
|
pa_memblock_unref_fixed(audio_chunk.memblock);
|
|
}
|
|
|
|
sink = sink->next;
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
case PA_SINK_MESSAGE_GET_LATENCY: {
|
|
*((pa_usec_t *) data) = get_latency_us(PA_OBJECT(o));
|
|
return 0;
|
|
}
|
|
}
|
|
|
|
return pa_sink_process_msg(o, code, data, offset, chunk);
|
|
}
|
|
|
|
static int source_process_msg(pa_msgobject *o, int code, void *data, int64_t offset, pa_memchunk *chunk) {
|
|
coreaudio_source *source = PA_SOURCE(o)->userdata;
|
|
struct userdata *u = source->userdata;
|
|
unsigned int i;
|
|
pa_memchunk audio_chunk;
|
|
|
|
switch (code) {
|
|
case CA_MESSAGE_RENDER: {
|
|
/* audio in */
|
|
for (i = 0; i < u->render_input_data->mNumberBuffers; i++) {
|
|
const AudioBuffer *buf = u->render_input_data->mBuffers + i;
|
|
|
|
pa_assert(source);
|
|
|
|
if (PA_SOURCE_IS_OPENED(source->pa_source->thread_info.state)) {
|
|
audio_chunk.memblock = pa_memblock_new_fixed(u->module->core->mempool, buf->mData, buf->mDataByteSize, TRUE);
|
|
audio_chunk.length = buf->mDataByteSize;
|
|
audio_chunk.index = 0;
|
|
|
|
pa_source_post(source->pa_source, &audio_chunk);
|
|
pa_memblock_unref_fixed(audio_chunk.memblock);
|
|
}
|
|
|
|
source = source->next;
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
case PA_SOURCE_MESSAGE_GET_LATENCY: {
|
|
*((pa_usec_t *) data) = get_latency_us(PA_OBJECT(o));
|
|
return 0;
|
|
}
|
|
}
|
|
|
|
return pa_source_process_msg(o, code, data, offset, chunk);;
|
|
}
|
|
|
|
static int ca_sink_set_state(pa_sink *s, pa_sink_state_t state) {
|
|
coreaudio_sink *sink = s->userdata;
|
|
|
|
switch (state) {
|
|
case PA_SINK_SUSPENDED:
|
|
case PA_SINK_IDLE:
|
|
sink->active = FALSE;
|
|
break;
|
|
|
|
case PA_SINK_RUNNING:
|
|
sink->active = TRUE;
|
|
break;
|
|
|
|
case PA_SINK_UNLINKED:
|
|
case PA_SINK_INIT:
|
|
case PA_SINK_INVALID_STATE:
|
|
;
|
|
}
|
|
|
|
ca_device_check_device_state(sink->userdata);
|
|
|
|
return 0;
|
|
}
|
|
|
|
static int ca_device_create_sink(pa_module *m, AudioBuffer *buf, int channel_idx) {
|
|
OSStatus err;
|
|
UInt32 size;
|
|
struct userdata *u = m->userdata;
|
|
pa_sink_new_data new_data;
|
|
pa_sink_flags_t flags = PA_SINK_LATENCY | PA_SINK_HARDWARE;
|
|
coreaudio_sink *ca_sink;
|
|
pa_sink *sink;
|
|
unsigned int i;
|
|
char tmp[255];
|
|
pa_strbuf *strbuf;
|
|
AudioObjectPropertyAddress property_address;
|
|
|
|
ca_sink = pa_xnew0(coreaudio_sink, 1);
|
|
ca_sink->map.channels = buf->mNumberChannels;
|
|
ca_sink->ss.channels = buf->mNumberChannels;
|
|
ca_sink->channel_idx = channel_idx;
|
|
|
|
/* build a name for this stream */
|
|
strbuf = pa_strbuf_new();
|
|
|
|
for (i = 0; i < buf->mNumberChannels; i++) {
|
|
property_address.mSelector = kAudioObjectPropertyElementName;
|
|
property_address.mScope = kAudioDevicePropertyScopeOutput;
|
|
property_address.mElement = channel_idx + i + 1;
|
|
size = sizeof(tmp);
|
|
err = AudioObjectGetPropertyData(u->object_id, &property_address, 0, NULL, &size, tmp);
|
|
if (err || !strlen(tmp))
|
|
snprintf(tmp, sizeof(tmp), "Channel %d", (int) property_address.mElement);
|
|
|
|
if (i > 0)
|
|
pa_strbuf_puts(strbuf, ", ");
|
|
|
|
pa_strbuf_puts(strbuf, tmp);
|
|
}
|
|
|
|
ca_sink->name = pa_strbuf_tostring_free(strbuf);
|
|
|
|
pa_log_debug("Stream name is >%s<", ca_sink->name);
|
|
|
|
/* default to mono streams */
|
|
for (i = 0; i < ca_sink->map.channels; i++)
|
|
ca_sink->map.map[i] = PA_CHANNEL_POSITION_MONO;
|
|
|
|
if (buf->mNumberChannels == 2) {
|
|
ca_sink->map.map[0] = PA_CHANNEL_POSITION_LEFT;
|
|
ca_sink->map.map[1] = PA_CHANNEL_POSITION_RIGHT;
|
|
}
|
|
|
|
ca_sink->ss.rate = u->stream_description.mSampleRate;
|
|
ca_sink->ss.format = PA_SAMPLE_FLOAT32LE;
|
|
|
|
pa_sink_new_data_init(&new_data);
|
|
new_data.card = u->card;
|
|
new_data.driver = __FILE__;
|
|
new_data.module = u->module;
|
|
new_data.namereg_fail = FALSE;
|
|
pa_sink_new_data_set_name(&new_data, ca_sink->name);
|
|
pa_sink_new_data_set_channel_map(&new_data, &ca_sink->map);
|
|
pa_sink_new_data_set_sample_spec(&new_data, &ca_sink->ss);
|
|
pa_proplist_sets(new_data.proplist, PA_PROP_DEVICE_STRING, u->device_name);
|
|
pa_proplist_sets(new_data.proplist, PA_PROP_DEVICE_PRODUCT_NAME, u->device_name);
|
|
pa_proplist_sets(new_data.proplist, PA_PROP_DEVICE_DESCRIPTION, u->device_name);
|
|
pa_proplist_sets(new_data.proplist, PA_PROP_DEVICE_ACCESS_MODE, "mmap");
|
|
pa_proplist_sets(new_data.proplist, PA_PROP_DEVICE_CLASS, "sound");
|
|
pa_proplist_sets(new_data.proplist, PA_PROP_DEVICE_API, "CoreAudio");
|
|
pa_proplist_setf(new_data.proplist, PA_PROP_DEVICE_BUFFERING_BUFFER_SIZE, "%lu", (unsigned long) buf->mDataByteSize);
|
|
|
|
if (u->vendor_name)
|
|
pa_proplist_sets(new_data.proplist, PA_PROP_DEVICE_VENDOR_NAME, u->vendor_name);
|
|
|
|
sink = pa_sink_new(m->core, &new_data, flags);
|
|
pa_sink_new_data_done(&new_data);
|
|
|
|
if (!sink) {
|
|
pa_log("unable to create sink.");
|
|
return -1;
|
|
}
|
|
|
|
sink->parent.process_msg = sink_process_msg;
|
|
sink->userdata = ca_sink;
|
|
sink->set_state = ca_sink_set_state;
|
|
|
|
pa_sink_set_asyncmsgq(sink, u->thread_mq.inq);
|
|
pa_sink_set_rtpoll(sink, u->rtpoll);
|
|
|
|
ca_sink->pa_sink = sink;
|
|
ca_sink->userdata = u;
|
|
|
|
PA_LLIST_PREPEND(coreaudio_sink, u->sinks, ca_sink);
|
|
|
|
return 0;
|
|
}
|
|
|
|
static int ca_source_set_state(pa_source *s, pa_source_state_t state) {
|
|
coreaudio_source *source = s->userdata;
|
|
|
|
switch (state) {
|
|
case PA_SOURCE_SUSPENDED:
|
|
case PA_SOURCE_IDLE:
|
|
source->active = FALSE;
|
|
break;
|
|
|
|
case PA_SOURCE_RUNNING:
|
|
source->active = TRUE;
|
|
break;
|
|
|
|
case PA_SOURCE_UNLINKED:
|
|
case PA_SOURCE_INIT:
|
|
case PA_SOURCE_INVALID_STATE:
|
|
;
|
|
}
|
|
|
|
ca_device_check_device_state(source->userdata);
|
|
|
|
return 0;
|
|
}
|
|
|
|
static int ca_device_create_source(pa_module *m, AudioBuffer *buf, int channel_idx) {
|
|
OSStatus err;
|
|
UInt32 size;
|
|
struct userdata *u = m->userdata;
|
|
pa_source_new_data new_data;
|
|
pa_source_flags_t flags = PA_SOURCE_LATENCY | PA_SOURCE_HARDWARE;
|
|
coreaudio_source *ca_source;
|
|
pa_source *source;
|
|
unsigned int i;
|
|
char tmp[255];
|
|
pa_strbuf *strbuf;
|
|
AudioObjectPropertyAddress property_address;
|
|
|
|
ca_source = pa_xnew0(coreaudio_source, 1);
|
|
ca_source->map.channels = buf->mNumberChannels;
|
|
ca_source->ss.channels = buf->mNumberChannels;
|
|
ca_source->channel_idx = channel_idx;
|
|
|
|
/* build a name for this stream */
|
|
strbuf = pa_strbuf_new();
|
|
|
|
for (i = 0; i < buf->mNumberChannels; i++) {
|
|
property_address.mSelector = kAudioObjectPropertyElementName;
|
|
property_address.mScope = kAudioDevicePropertyScopeInput;
|
|
property_address.mElement = channel_idx + i + 1;
|
|
size = sizeof(tmp);
|
|
err = AudioObjectGetPropertyData(u->object_id, &property_address, 0, NULL, &size, tmp);
|
|
if (err || !strlen(tmp))
|
|
snprintf(tmp, sizeof(tmp), "Channel %d", (int) property_address.mElement);
|
|
|
|
if (i > 0)
|
|
pa_strbuf_puts(strbuf, ", ");
|
|
|
|
pa_strbuf_puts(strbuf, tmp);
|
|
}
|
|
|
|
ca_source->name = pa_strbuf_tostring_free(strbuf);
|
|
|
|
pa_log_debug("Stream name is >%s<", ca_source->name);
|
|
|
|
/* default to mono streams */
|
|
for (i = 0; i < ca_source->map.channels; i++)
|
|
ca_source->map.map[i] = PA_CHANNEL_POSITION_MONO;
|
|
|
|
if (buf->mNumberChannels == 2) {
|
|
ca_source->map.map[0] = PA_CHANNEL_POSITION_LEFT;
|
|
ca_source->map.map[1] = PA_CHANNEL_POSITION_RIGHT;
|
|
}
|
|
|
|
ca_source->ss.rate = u->stream_description.mSampleRate;
|
|
ca_source->ss.format = PA_SAMPLE_FLOAT32LE;
|
|
|
|
pa_source_new_data_init(&new_data);
|
|
new_data.card = u->card;
|
|
new_data.driver = __FILE__;
|
|
new_data.module = u->module;
|
|
new_data.namereg_fail = FALSE;
|
|
pa_source_new_data_set_name(&new_data, ca_source->name);
|
|
pa_source_new_data_set_channel_map(&new_data, &ca_source->map);
|
|
pa_source_new_data_set_sample_spec(&new_data, &ca_source->ss);
|
|
pa_proplist_sets(new_data.proplist, PA_PROP_DEVICE_STRING, u->device_name);
|
|
pa_proplist_sets(new_data.proplist, PA_PROP_DEVICE_PRODUCT_NAME, u->device_name);
|
|
pa_proplist_sets(new_data.proplist, PA_PROP_DEVICE_DESCRIPTION, u->device_name);
|
|
pa_proplist_sets(new_data.proplist, PA_PROP_DEVICE_ACCESS_MODE, "mmap");
|
|
pa_proplist_sets(new_data.proplist, PA_PROP_DEVICE_CLASS, "sound");
|
|
pa_proplist_sets(new_data.proplist, PA_PROP_DEVICE_API, "CoreAudio");
|
|
pa_proplist_setf(new_data.proplist, PA_PROP_DEVICE_BUFFERING_BUFFER_SIZE, "%lu", (unsigned long) buf->mDataByteSize);
|
|
|
|
if (u->vendor_name)
|
|
pa_proplist_sets(new_data.proplist, PA_PROP_DEVICE_VENDOR_NAME, u->vendor_name);
|
|
|
|
source = pa_source_new(m->core, &new_data, flags);
|
|
pa_source_new_data_done(&new_data);
|
|
|
|
if (!source) {
|
|
pa_log("unable to create source.");
|
|
return -1;
|
|
}
|
|
|
|
source->parent.process_msg = source_process_msg;
|
|
source->userdata = ca_source;
|
|
source->set_state = ca_source_set_state;
|
|
|
|
pa_source_set_asyncmsgq(source, u->thread_mq.inq);
|
|
pa_source_set_rtpoll(source, u->rtpoll);
|
|
|
|
ca_source->pa_source = source;
|
|
ca_source->userdata = u;
|
|
|
|
PA_LLIST_PREPEND(coreaudio_source, u->sources, ca_source);
|
|
|
|
return 0;
|
|
}
|
|
|
|
static int ca_device_create_streams(pa_module *m, bool direction_in) {
|
|
OSStatus err;
|
|
UInt32 size, i, channel_idx;
|
|
struct userdata *u = m->userdata;
|
|
AudioBufferList *buffer_list;
|
|
AudioObjectPropertyAddress property_address;
|
|
|
|
property_address.mScope = direction_in ? kAudioDevicePropertyScopeInput : kAudioDevicePropertyScopeOutput;
|
|
property_address.mElement = kAudioObjectPropertyElementMaster;
|
|
|
|
/* get current stream format */
|
|
size = sizeof(AudioStreamBasicDescription);
|
|
property_address.mSelector = kAudioDevicePropertyStreamFormat;
|
|
err = AudioObjectGetPropertyData(u->object_id, &property_address, 0, NULL, &size, &u->stream_description);
|
|
if (err) {
|
|
/* no appropriate streams found - silently bail. */
|
|
return -1;
|
|
}
|
|
|
|
if (u->stream_description.mFormatID != kAudioFormatLinearPCM) {
|
|
pa_log("Unsupported audio format '%c%c%c%c'",
|
|
(char) (u->stream_description.mFormatID >> 24),
|
|
(char) (u->stream_description.mFormatID >> 16) & 0xff,
|
|
(char) (u->stream_description.mFormatID >> 8) & 0xff,
|
|
(char) (u->stream_description.mFormatID & 0xff));
|
|
return -1;
|
|
}
|
|
|
|
/* get stream configuration */
|
|
size = 0;
|
|
property_address.mSelector = kAudioDevicePropertyStreamConfiguration;
|
|
err = AudioObjectGetPropertyDataSize(u->object_id, &property_address, 0, NULL, &size);
|
|
if (err) {
|
|
pa_log("Failed to get kAudioDevicePropertyStreamConfiguration (%s).", direction_in ? "input" : "output");
|
|
return -1;
|
|
}
|
|
|
|
if (!size)
|
|
return 0;
|
|
|
|
buffer_list = (AudioBufferList *) pa_xmalloc(size);
|
|
err = AudioObjectGetPropertyData(u->object_id, &property_address, 0, NULL, &size, buffer_list);
|
|
|
|
if (!err) {
|
|
pa_log_debug("Sample rate: %f", u->stream_description.mSampleRate);
|
|
pa_log_debug("%d bytes per packet", (unsigned int) u->stream_description.mBytesPerPacket);
|
|
pa_log_debug("%d frames per packet", (unsigned int) u->stream_description.mFramesPerPacket);
|
|
pa_log_debug("%d bytes per frame", (unsigned int) u->stream_description.mBytesPerFrame);
|
|
pa_log_debug("%d channels per frame", (unsigned int) u->stream_description.mChannelsPerFrame);
|
|
pa_log_debug("%d bits per channel", (unsigned int) u->stream_description.mBitsPerChannel);
|
|
|
|
for (channel_idx = 0, i = 0; i < buffer_list->mNumberBuffers; i++) {
|
|
AudioBuffer *buf = buffer_list->mBuffers + i;
|
|
|
|
if (direction_in)
|
|
ca_device_create_source(m, buf, channel_idx);
|
|
else
|
|
ca_device_create_sink(m, buf, channel_idx);
|
|
|
|
channel_idx += buf->mNumberChannels;
|
|
}
|
|
}
|
|
|
|
pa_xfree(buffer_list);
|
|
return 0;
|
|
}
|
|
|
|
static void thread_func(void *userdata) {
|
|
struct userdata *u = userdata;
|
|
|
|
pa_assert(u);
|
|
pa_assert(u->module);
|
|
pa_assert(u->module->core);
|
|
|
|
pa_log_debug("Thread starting up");
|
|
|
|
if (u->module->core->realtime_scheduling)
|
|
pa_make_realtime(u->module->core->realtime_priority);
|
|
|
|
pa_thread_mq_install(&u->thread_mq);
|
|
|
|
for (;;) {
|
|
coreaudio_sink *ca_sink;
|
|
int ret;
|
|
|
|
PA_LLIST_FOREACH(ca_sink, u->sinks) {
|
|
if (PA_UNLIKELY(ca_sink->pa_sink->thread_info.rewind_requested))
|
|
pa_sink_process_rewind(ca_sink->pa_sink, 0);
|
|
}
|
|
|
|
ret = pa_rtpoll_run(u->rtpoll, TRUE);
|
|
|
|
if (ret < 0)
|
|
goto fail;
|
|
|
|
if (ret == 0)
|
|
goto finish;
|
|
}
|
|
|
|
fail:
|
|
/* If this was no regular exit from the loop we have to continue
|
|
* processing messages until we received PA_MESSAGE_SHUTDOWN */
|
|
pa_asyncmsgq_post(u->thread_mq.outq, PA_MSGOBJECT(u->module->core), PA_CORE_MESSAGE_UNLOAD_MODULE, u->module, 0, NULL, NULL);
|
|
pa_asyncmsgq_wait_for(u->thread_mq.inq, PA_MESSAGE_SHUTDOWN);
|
|
|
|
finish:
|
|
pa_log_debug("Thread shutting down");
|
|
}
|
|
|
|
int pa__init(pa_module *m) {
|
|
OSStatus err;
|
|
UInt32 size, frames;
|
|
struct userdata *u = NULL;
|
|
pa_modargs *ma = NULL;
|
|
char tmp[64];
|
|
pa_card_new_data card_new_data;
|
|
pa_card_profile *p;
|
|
coreaudio_sink *ca_sink;
|
|
coreaudio_source *ca_source;
|
|
AudioObjectPropertyAddress property_address;
|
|
|
|
pa_assert(m);
|
|
|
|
if (!(ma = pa_modargs_new(m->argument, valid_modargs))) {
|
|
pa_log("Failed to parse module arguments.");
|
|
goto fail;
|
|
}
|
|
|
|
u = pa_xnew0(struct userdata, 1);
|
|
u->module = m;
|
|
m->userdata = u;
|
|
|
|
if (pa_modargs_get_value_u32(ma, "object_id", (unsigned int *) &u->object_id) != 0) {
|
|
pa_log("Failed to parse object_id argument.");
|
|
goto fail;
|
|
}
|
|
|
|
property_address.mScope = kAudioObjectPropertyScopeGlobal;
|
|
property_address.mElement = kAudioObjectPropertyElementMaster;
|
|
|
|
/* get device product name */
|
|
property_address.mSelector = kAudioDevicePropertyDeviceName;
|
|
size = sizeof(tmp);
|
|
err = AudioObjectGetPropertyData(u->object_id, &property_address, 0, NULL, &size, tmp);
|
|
if (err) {
|
|
pa_log("Failed to get kAudioDevicePropertyDeviceName (err = %08x).", (int) err);
|
|
goto fail;
|
|
}
|
|
|
|
u->device_name = pa_xstrdup(tmp);
|
|
|
|
pa_card_new_data_init(&card_new_data);
|
|
pa_proplist_sets(card_new_data.proplist, PA_PROP_DEVICE_STRING, tmp);
|
|
card_new_data.driver = __FILE__;
|
|
pa_card_new_data_set_name(&card_new_data, tmp);
|
|
pa_log_info("Initializing module for CoreAudio device '%s' (id %d)", tmp, (unsigned int) u->object_id);
|
|
|
|
/* get device vendor name (may fail) */
|
|
property_address.mSelector = kAudioDevicePropertyDeviceManufacturer;
|
|
size = sizeof(tmp);
|
|
err = AudioObjectGetPropertyData(u->object_id, &property_address, 0, NULL, &size, tmp);
|
|
if (!err)
|
|
u->vendor_name = pa_xstrdup(tmp);
|
|
|
|
/* add on profile */
|
|
p = pa_card_profile_new("on", _("On"), 0);
|
|
pa_hashmap_put(card_new_data.profiles, p->name, p);
|
|
|
|
/* create the card object */
|
|
u->card = pa_card_new(m->core, &card_new_data);
|
|
if (!u->card) {
|
|
pa_log("Unable to create card.\n");
|
|
goto fail;
|
|
}
|
|
|
|
pa_card_new_data_done(&card_new_data);
|
|
u->card->userdata = u;
|
|
u->card->set_profile = card_set_profile;
|
|
|
|
u->rtpoll = pa_rtpoll_new();
|
|
pa_thread_mq_init(&u->thread_mq, m->core->mainloop, u->rtpoll);
|
|
u->async_msgq = pa_asyncmsgq_new(0);
|
|
pa_rtpoll_item_new_asyncmsgq_read(u->rtpoll, PA_RTPOLL_EARLY-1, u->async_msgq);
|
|
|
|
PA_LLIST_HEAD_INIT(coreaudio_sink, u->sinks);
|
|
|
|
/* create sinks */
|
|
ca_device_create_streams(m, FALSE);
|
|
|
|
/* create sources */
|
|
ca_device_create_streams(m, TRUE);
|
|
|
|
/* create the message thread */
|
|
if (!(u->thread = pa_thread_new(u->device_name, thread_func, u))) {
|
|
pa_log("Failed to create thread.");
|
|
goto fail;
|
|
}
|
|
|
|
/* register notification callback for stream format changes */
|
|
property_address.mSelector = kAudioDevicePropertyStreamFormat;
|
|
property_address.mScope = kAudioObjectPropertyScopeGlobal;
|
|
property_address.mElement = kAudioObjectPropertyElementMaster;
|
|
|
|
AudioObjectAddPropertyListener(u->object_id, &property_address, ca_stream_format_changed, u);
|
|
|
|
/* set number of frames in IOProc */
|
|
frames = DEFAULT_FRAMES_PER_IOPROC;
|
|
pa_modargs_get_value_u32(ma, "ioproc_frames", (unsigned int *) &frames);
|
|
|
|
property_address.mSelector = kAudioDevicePropertyBufferFrameSize;
|
|
AudioObjectSetPropertyData(u->object_id, &property_address, 0, NULL, sizeof(frames), &frames);
|
|
pa_log_debug("%u frames per IOProc\n", (unsigned int) frames);
|
|
|
|
/* create one ioproc for both directions */
|
|
err = AudioDeviceCreateIOProcID(u->object_id, io_render_proc, u, &u->proc_id);
|
|
if (err) {
|
|
pa_log("AudioDeviceCreateIOProcID() failed (err = %08x\n).", (int) err);
|
|
goto fail;
|
|
}
|
|
|
|
for (ca_sink = u->sinks; ca_sink; ca_sink = ca_sink->next)
|
|
pa_sink_put(ca_sink->pa_sink);
|
|
|
|
for (ca_source = u->sources; ca_source; ca_source = ca_source->next)
|
|
pa_source_put(ca_source->pa_source);
|
|
|
|
pa_modargs_free(ma);
|
|
|
|
return 0;
|
|
|
|
fail:
|
|
if (u)
|
|
pa__done(m);
|
|
|
|
if (ma)
|
|
pa_modargs_free(ma);
|
|
|
|
return -1;
|
|
}
|
|
|
|
void pa__done(pa_module *m) {
|
|
struct userdata *u;
|
|
coreaudio_sink *ca_sink;
|
|
coreaudio_source *ca_source;
|
|
AudioObjectPropertyAddress property_address;
|
|
|
|
pa_assert(m);
|
|
|
|
u = m->userdata;
|
|
pa_assert(u);
|
|
|
|
/* unlink sinks */
|
|
for (ca_sink = u->sinks; ca_sink; ca_sink = ca_sink->next)
|
|
if (ca_sink->pa_sink)
|
|
pa_sink_unlink(ca_sink->pa_sink);
|
|
|
|
/* unlink sources */
|
|
for (ca_source = u->sources; ca_source; ca_source = ca_source->next)
|
|
if (ca_source->pa_source)
|
|
pa_source_unlink(ca_source->pa_source);
|
|
|
|
if (u->thread) {
|
|
pa_asyncmsgq_send(u->thread_mq.inq, NULL, PA_MESSAGE_SHUTDOWN, NULL, 0, NULL);
|
|
pa_thread_free(u->thread);
|
|
pa_thread_mq_done(&u->thread_mq);
|
|
pa_asyncmsgq_unref(u->async_msgq);
|
|
}
|
|
|
|
/* free sinks */
|
|
for (ca_sink = u->sinks; ca_sink;) {
|
|
coreaudio_sink *next = ca_sink->next;
|
|
|
|
if (ca_sink->pa_sink)
|
|
pa_sink_unref(ca_sink->pa_sink);
|
|
|
|
pa_xfree(ca_sink->name);
|
|
pa_xfree(ca_sink);
|
|
ca_sink = next;
|
|
}
|
|
|
|
/* free sources */
|
|
for (ca_source = u->sources; ca_source;) {
|
|
coreaudio_source *next = ca_source->next;
|
|
|
|
if (ca_source->pa_source)
|
|
pa_source_unref(ca_source->pa_source);
|
|
|
|
pa_xfree(ca_source->name);
|
|
pa_xfree(ca_source);
|
|
ca_source = next;
|
|
}
|
|
|
|
if (u->proc_id) {
|
|
AudioDeviceStop(u->object_id, u->proc_id);
|
|
AudioDeviceDestroyIOProcID(u->object_id, u->proc_id);
|
|
}
|
|
|
|
property_address.mSelector = kAudioDevicePropertyStreamFormat;
|
|
property_address.mScope = kAudioObjectPropertyScopeGlobal;
|
|
property_address.mElement = kAudioObjectPropertyElementMaster;
|
|
|
|
AudioObjectRemovePropertyListener(kAudioObjectSystemObject, &property_address, ca_stream_format_changed, u);
|
|
|
|
pa_xfree(u->device_name);
|
|
pa_xfree(u->vendor_name);
|
|
pa_rtpoll_free(u->rtpoll);
|
|
pa_card_free(u->card);
|
|
|
|
pa_xfree(u);
|
|
}
|