/* ** ** Copyright 2007, The Android Open Source Project ** ** Licensed under the Apache License, Version 2.0 (the "License"); ** you may not use this file except in compliance with the License. ** You may obtain a copy of the License at ** ** http://www.apache.org/licenses/LICENSE-2.0 ** ** Unless required by applicable law or agreed to in writing, software ** distributed under the License is distributed on an "AS IS" BASIS, ** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. ** See the License for the specific language governing permissions and ** limitations under the License. */ #define LOG_TAG "IAudioFlinger" //#define LOG_NDEBUG 0 #include <utils/Log.h> #include <stdint.h> #include <sys/types.h> #include <binder/Parcel.h> #include <media/IAudioFlinger.h> namespace android { enum { CREATE_TRACK = IBinder::FIRST_CALL_TRANSACTION, OPEN_RECORD, SAMPLE_RATE, CHANNEL_COUNT, FORMAT, FRAME_COUNT, LATENCY, SET_MASTER_VOLUME, SET_MASTER_MUTE, MASTER_VOLUME, MASTER_MUTE, SET_STREAM_VOLUME, SET_STREAM_MUTE, STREAM_VOLUME, STREAM_MUTE, SET_MODE, SET_MIC_MUTE, GET_MIC_MUTE, SET_PARAMETERS, GET_PARAMETERS, REGISTER_CLIENT, GET_INPUTBUFFERSIZE, OPEN_OUTPUT, OPEN_DUPLICATE_OUTPUT, CLOSE_OUTPUT, SUSPEND_OUTPUT, RESTORE_OUTPUT, OPEN_INPUT, CLOSE_INPUT, SET_STREAM_OUTPUT, SET_VOICE_VOLUME, GET_RENDER_POSITION, GET_INPUT_FRAMES_LOST, NEW_AUDIO_SESSION_ID, ACQUIRE_AUDIO_SESSION_ID, RELEASE_AUDIO_SESSION_ID, QUERY_NUM_EFFECTS, QUERY_EFFECT, GET_EFFECT_DESCRIPTOR, CREATE_EFFECT, MOVE_EFFECTS, LOAD_HW_MODULE }; class BpAudioFlinger : public BpInterface<IAudioFlinger> { public: BpAudioFlinger(const sp<IBinder>& impl) : BpInterface<IAudioFlinger>(impl) { } virtual sp<IAudioTrack> createTrack( pid_t pid, audio_stream_type_t streamType, uint32_t sampleRate, audio_format_t format, uint32_t channelMask, int frameCount, track_flags_t flags, const sp<IMemory>& sharedBuffer, audio_io_handle_t output, pid_t tid, int *sessionId, status_t *status) { Parcel data, reply; sp<IAudioTrack> track; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32(pid); data.writeInt32((int32_t) streamType); data.writeInt32(sampleRate); data.writeInt32(format); data.writeInt32(channelMask); data.writeInt32(frameCount); data.writeInt32((int32_t) flags); data.writeStrongBinder(sharedBuffer->asBinder()); data.writeInt32((int32_t) output); data.writeInt32((int32_t) tid); int lSessionId = 0; if (sessionId != NULL) { lSessionId = *sessionId; } data.writeInt32(lSessionId); status_t lStatus = remote()->transact(CREATE_TRACK, data, &reply); if (lStatus != NO_ERROR) { ALOGE("createTrack error: %s", strerror(-lStatus)); } else { lSessionId = reply.readInt32(); if (sessionId != NULL) { *sessionId = lSessionId; } lStatus = reply.readInt32(); track = interface_cast<IAudioTrack>(reply.readStrongBinder()); } if (status) { *status = lStatus; } return track; } virtual sp<IAudioRecord> openRecord( pid_t pid, audio_io_handle_t input, uint32_t sampleRate, audio_format_t format, uint32_t channelMask, int frameCount, track_flags_t flags, int *sessionId, status_t *status) { Parcel data, reply; sp<IAudioRecord> record; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32(pid); data.writeInt32((int32_t) input); data.writeInt32(sampleRate); data.writeInt32(format); data.writeInt32(channelMask); data.writeInt32(frameCount); data.writeInt32(flags); int lSessionId = 0; if (sessionId != NULL) { lSessionId = *sessionId; } data.writeInt32(lSessionId); status_t lStatus = remote()->transact(OPEN_RECORD, data, &reply); if (lStatus != NO_ERROR) { ALOGE("openRecord error: %s", strerror(-lStatus)); } else { lSessionId = reply.readInt32(); if (sessionId != NULL) { *sessionId = lSessionId; } lStatus = reply.readInt32(); record = interface_cast<IAudioRecord>(reply.readStrongBinder()); } if (status) { *status = lStatus; } return record; } virtual uint32_t sampleRate(audio_io_handle_t output) const { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32((int32_t) output); remote()->transact(SAMPLE_RATE, data, &reply); return reply.readInt32(); } virtual int channelCount(audio_io_handle_t output) const { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32((int32_t) output); remote()->transact(CHANNEL_COUNT, data, &reply); return reply.readInt32(); } virtual audio_format_t format(audio_io_handle_t output) const { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32((int32_t) output); remote()->transact(FORMAT, data, &reply); return (audio_format_t) reply.readInt32(); } virtual size_t frameCount(audio_io_handle_t output) const { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32((int32_t) output); remote()->transact(FRAME_COUNT, data, &reply); return reply.readInt32(); } virtual uint32_t latency(audio_io_handle_t output) const { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32((int32_t) output); remote()->transact(LATENCY, data, &reply); return reply.readInt32(); } virtual status_t setMasterVolume(float value) { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeFloat(value); remote()->transact(SET_MASTER_VOLUME, data, &reply); return reply.readInt32(); } virtual status_t setMasterMute(bool muted) { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32(muted); remote()->transact(SET_MASTER_MUTE, data, &reply); return reply.readInt32(); } virtual float masterVolume() const { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); remote()->transact(MASTER_VOLUME, data, &reply); return reply.readFloat(); } virtual bool masterMute() const { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); remote()->transact(MASTER_MUTE, data, &reply); return reply.readInt32(); } virtual status_t setStreamVolume(audio_stream_type_t stream, float value, audio_io_handle_t output) { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32((int32_t) stream); data.writeFloat(value); data.writeInt32((int32_t) output); remote()->transact(SET_STREAM_VOLUME, data, &reply); return reply.readInt32(); } virtual status_t setStreamMute(audio_stream_type_t stream, bool muted) { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32((int32_t) stream); data.writeInt32(muted); remote()->transact(SET_STREAM_MUTE, data, &reply); return reply.readInt32(); } virtual float streamVolume(audio_stream_type_t stream, audio_io_handle_t output) const { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32((int32_t) stream); data.writeInt32((int32_t) output); remote()->transact(STREAM_VOLUME, data, &reply); return reply.readFloat(); } virtual bool streamMute(audio_stream_type_t stream) const { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32((int32_t) stream); remote()->transact(STREAM_MUTE, data, &reply); return reply.readInt32(); } virtual status_t setMode(audio_mode_t mode) { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32(mode); remote()->transact(SET_MODE, data, &reply); return reply.readInt32(); } virtual status_t setMicMute(bool state) { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32(state); remote()->transact(SET_MIC_MUTE, data, &reply); return reply.readInt32(); } virtual bool getMicMute() const { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); remote()->transact(GET_MIC_MUTE, data, &reply); return reply.readInt32(); } virtual status_t setParameters(audio_io_handle_t ioHandle, const String8& keyValuePairs) { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32((int32_t) ioHandle); data.writeString8(keyValuePairs); remote()->transact(SET_PARAMETERS, data, &reply); return reply.readInt32(); } virtual String8 getParameters(audio_io_handle_t ioHandle, const String8& keys) const { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32((int32_t) ioHandle); data.writeString8(keys); remote()->transact(GET_PARAMETERS, data, &reply); return reply.readString8(); } virtual void registerClient(const sp<IAudioFlingerClient>& client) { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeStrongBinder(client->asBinder()); remote()->transact(REGISTER_CLIENT, data, &reply); } virtual size_t getInputBufferSize(uint32_t sampleRate, audio_format_t format, int channelCount) const { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32(sampleRate); data.writeInt32(format); data.writeInt32(channelCount); remote()->transact(GET_INPUTBUFFERSIZE, data, &reply); return reply.readInt32(); } virtual audio_io_handle_t openOutput(audio_module_handle_t module, audio_devices_t *pDevices, uint32_t *pSamplingRate, audio_format_t *pFormat, audio_channel_mask_t *pChannelMask, uint32_t *pLatencyMs, audio_output_flags_t flags) { Parcel data, reply; audio_devices_t devices = pDevices ? *pDevices : (audio_devices_t)0; uint32_t samplingRate = pSamplingRate ? *pSamplingRate : 0; audio_format_t format = pFormat ? *pFormat : AUDIO_FORMAT_DEFAULT; audio_channel_mask_t channelMask = pChannelMask ? *pChannelMask : (audio_channel_mask_t)0; uint32_t latency = pLatencyMs ? *pLatencyMs : 0; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32(module); data.writeInt32(devices); data.writeInt32(samplingRate); data.writeInt32(format); data.writeInt32(channelMask); data.writeInt32(latency); data.writeInt32((int32_t) flags); remote()->transact(OPEN_OUTPUT, data, &reply); audio_io_handle_t output = (audio_io_handle_t) reply.readInt32(); ALOGV("openOutput() returned output, %d", output); devices = (audio_devices_t)reply.readInt32(); if (pDevices) *pDevices = devices; samplingRate = reply.readInt32(); if (pSamplingRate) *pSamplingRate = samplingRate; format = (audio_format_t) reply.readInt32(); if (pFormat) *pFormat = format; channelMask = (audio_channel_mask_t)reply.readInt32(); if (pChannelMask) *pChannelMask = channelMask; latency = reply.readInt32(); if (pLatencyMs) *pLatencyMs = latency; return output; } virtual audio_io_handle_t openDuplicateOutput(audio_io_handle_t output1, audio_io_handle_t output2) { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32((int32_t) output1); data.writeInt32((int32_t) output2); remote()->transact(OPEN_DUPLICATE_OUTPUT, data, &reply); return (audio_io_handle_t) reply.readInt32(); } virtual status_t closeOutput(audio_io_handle_t output) { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32((int32_t) output); remote()->transact(CLOSE_OUTPUT, data, &reply); return reply.readInt32(); } virtual status_t suspendOutput(audio_io_handle_t output) { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32((int32_t) output); remote()->transact(SUSPEND_OUTPUT, data, &reply); return reply.readInt32(); } virtual status_t restoreOutput(audio_io_handle_t output) { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32((int32_t) output); remote()->transact(RESTORE_OUTPUT, data, &reply); return reply.readInt32(); } virtual audio_io_handle_t openInput(audio_module_handle_t module, audio_devices_t *pDevices, uint32_t *pSamplingRate, audio_format_t *pFormat, audio_channel_mask_t *pChannelMask) { Parcel data, reply; audio_devices_t devices = pDevices ? *pDevices : (audio_devices_t)0; uint32_t samplingRate = pSamplingRate ? *pSamplingRate : 0; audio_format_t format = pFormat ? *pFormat : AUDIO_FORMAT_DEFAULT; audio_channel_mask_t channelMask = pChannelMask ? *pChannelMask : (audio_channel_mask_t)0; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32(module); data.writeInt32(devices); data.writeInt32(samplingRate); data.writeInt32(format); data.writeInt32(channelMask); remote()->transact(OPEN_INPUT, data, &reply); audio_io_handle_t input = (audio_io_handle_t) reply.readInt32(); devices = (audio_devices_t)reply.readInt32(); if (pDevices) *pDevices = devices; samplingRate = reply.readInt32(); if (pSamplingRate) *pSamplingRate = samplingRate; format = (audio_format_t) reply.readInt32(); if (pFormat) *pFormat = format; channelMask = (audio_channel_mask_t)reply.readInt32(); if (pChannelMask) *pChannelMask = channelMask; return input; } virtual status_t closeInput(int input) { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32(input); remote()->transact(CLOSE_INPUT, data, &reply); return reply.readInt32(); } virtual status_t setStreamOutput(audio_stream_type_t stream, audio_io_handle_t output) { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32((int32_t) stream); data.writeInt32((int32_t) output); remote()->transact(SET_STREAM_OUTPUT, data, &reply); return reply.readInt32(); } virtual status_t setVoiceVolume(float volume) { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeFloat(volume); remote()->transact(SET_VOICE_VOLUME, data, &reply); return reply.readInt32(); } virtual status_t getRenderPosition(uint32_t *halFrames, uint32_t *dspFrames, audio_io_handle_t output) const { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32((int32_t) output); remote()->transact(GET_RENDER_POSITION, data, &reply); status_t status = reply.readInt32(); if (status == NO_ERROR) { uint32_t tmp = reply.readInt32(); if (halFrames) { *halFrames = tmp; } tmp = reply.readInt32(); if (dspFrames) { *dspFrames = tmp; } } return status; } virtual unsigned int getInputFramesLost(audio_io_handle_t ioHandle) const { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32((int32_t) ioHandle); remote()->transact(GET_INPUT_FRAMES_LOST, data, &reply); return reply.readInt32(); } virtual int newAudioSessionId() { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); status_t status = remote()->transact(NEW_AUDIO_SESSION_ID, data, &reply); int id = 0; if (status == NO_ERROR) { id = reply.readInt32(); } return id; } virtual void acquireAudioSessionId(int audioSession) { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32(audioSession); remote()->transact(ACQUIRE_AUDIO_SESSION_ID, data, &reply); } virtual void releaseAudioSessionId(int audioSession) { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32(audioSession); remote()->transact(RELEASE_AUDIO_SESSION_ID, data, &reply); } virtual status_t queryNumberEffects(uint32_t *numEffects) const { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); status_t status = remote()->transact(QUERY_NUM_EFFECTS, data, &reply); if (status != NO_ERROR) { return status; } status = reply.readInt32(); if (status != NO_ERROR) { return status; } if (numEffects != NULL) { *numEffects = (uint32_t)reply.readInt32(); } return NO_ERROR; } virtual status_t queryEffect(uint32_t index, effect_descriptor_t *pDescriptor) const { if (pDescriptor == NULL) { return BAD_VALUE; } Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32(index); status_t status = remote()->transact(QUERY_EFFECT, data, &reply); if (status != NO_ERROR) { return status; } status = reply.readInt32(); if (status != NO_ERROR) { return status; } reply.read(pDescriptor, sizeof(effect_descriptor_t)); return NO_ERROR; } virtual status_t getEffectDescriptor(const effect_uuid_t *pUuid, effect_descriptor_t *pDescriptor) const { if (pUuid == NULL || pDescriptor == NULL) { return BAD_VALUE; } Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.write(pUuid, sizeof(effect_uuid_t)); status_t status = remote()->transact(GET_EFFECT_DESCRIPTOR, data, &reply); if (status != NO_ERROR) { return status; } status = reply.readInt32(); if (status != NO_ERROR) { return status; } reply.read(pDescriptor, sizeof(effect_descriptor_t)); return NO_ERROR; } virtual sp<IEffect> createEffect(pid_t pid, effect_descriptor_t *pDesc, const sp<IEffectClient>& client, int32_t priority, audio_io_handle_t output, int sessionId, status_t *status, int *id, int *enabled) { Parcel data, reply; sp<IEffect> effect; if (pDesc == NULL) { return effect; if (status) { *status = BAD_VALUE; } } data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32(pid); data.write(pDesc, sizeof(effect_descriptor_t)); data.writeStrongBinder(client->asBinder()); data.writeInt32(priority); data.writeInt32((int32_t) output); data.writeInt32(sessionId); status_t lStatus = remote()->transact(CREATE_EFFECT, data, &reply); if (lStatus != NO_ERROR) { ALOGE("createEffect error: %s", strerror(-lStatus)); } else { lStatus = reply.readInt32(); int tmp = reply.readInt32(); if (id) { *id = tmp; } tmp = reply.readInt32(); if (enabled != NULL) { *enabled = tmp; } effect = interface_cast<IEffect>(reply.readStrongBinder()); reply.read(pDesc, sizeof(effect_descriptor_t)); } if (status) { *status = lStatus; } return effect; } virtual status_t moveEffects(int session, audio_io_handle_t srcOutput, audio_io_handle_t dstOutput) { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeInt32(session); data.writeInt32((int32_t) srcOutput); data.writeInt32((int32_t) dstOutput); remote()->transact(MOVE_EFFECTS, data, &reply); return reply.readInt32(); } virtual audio_module_handle_t loadHwModule(const char *name) { Parcel data, reply; data.writeInterfaceToken(IAudioFlinger::getInterfaceDescriptor()); data.writeCString(name); remote()->transact(LOAD_HW_MODULE, data, &reply); return (audio_module_handle_t) reply.readInt32(); } }; IMPLEMENT_META_INTERFACE(AudioFlinger, "android.media.IAudioFlinger"); // ---------------------------------------------------------------------- status_t BnAudioFlinger::onTransact( uint32_t code, const Parcel& data, Parcel* reply, uint32_t flags) { switch (code) { case CREATE_TRACK: { CHECK_INTERFACE(IAudioFlinger, data, reply); pid_t pid = data.readInt32(); int streamType = data.readInt32(); uint32_t sampleRate = data.readInt32(); audio_format_t format = (audio_format_t) data.readInt32(); int channelCount = data.readInt32(); size_t bufferCount = data.readInt32(); track_flags_t flags = (track_flags_t) data.readInt32(); sp<IMemory> buffer = interface_cast<IMemory>(data.readStrongBinder()); audio_io_handle_t output = (audio_io_handle_t) data.readInt32(); pid_t tid = (pid_t) data.readInt32(); int sessionId = data.readInt32(); status_t status; sp<IAudioTrack> track = createTrack(pid, (audio_stream_type_t) streamType, sampleRate, format, channelCount, bufferCount, flags, buffer, output, tid, &sessionId, &status); reply->writeInt32(sessionId); reply->writeInt32(status); reply->writeStrongBinder(track->asBinder()); return NO_ERROR; } break; case OPEN_RECORD: { CHECK_INTERFACE(IAudioFlinger, data, reply); pid_t pid = data.readInt32(); audio_io_handle_t input = (audio_io_handle_t) data.readInt32(); uint32_t sampleRate = data.readInt32(); audio_format_t format = (audio_format_t) data.readInt32(); int channelCount = data.readInt32(); size_t bufferCount = data.readInt32(); track_flags_t flags = (track_flags_t) data.readInt32(); int sessionId = data.readInt32(); status_t status; sp<IAudioRecord> record = openRecord(pid, input, sampleRate, format, channelCount, bufferCount, flags, &sessionId, &status); reply->writeInt32(sessionId); reply->writeInt32(status); reply->writeStrongBinder(record->asBinder()); return NO_ERROR; } break; case SAMPLE_RATE: { CHECK_INTERFACE(IAudioFlinger, data, reply); reply->writeInt32( sampleRate((audio_io_handle_t) data.readInt32()) ); return NO_ERROR; } break; case CHANNEL_COUNT: { CHECK_INTERFACE(IAudioFlinger, data, reply); reply->writeInt32( channelCount((audio_io_handle_t) data.readInt32()) ); return NO_ERROR; } break; case FORMAT: { CHECK_INTERFACE(IAudioFlinger, data, reply); reply->writeInt32( format((audio_io_handle_t) data.readInt32()) ); return NO_ERROR; } break; case FRAME_COUNT: { CHECK_INTERFACE(IAudioFlinger, data, reply); reply->writeInt32( frameCount((audio_io_handle_t) data.readInt32()) ); return NO_ERROR; } break; case LATENCY: { CHECK_INTERFACE(IAudioFlinger, data, reply); reply->writeInt32( latency((audio_io_handle_t) data.readInt32()) ); return NO_ERROR; } break; case SET_MASTER_VOLUME: { CHECK_INTERFACE(IAudioFlinger, data, reply); reply->writeInt32( setMasterVolume(data.readFloat()) ); return NO_ERROR; } break; case SET_MASTER_MUTE: { CHECK_INTERFACE(IAudioFlinger, data, reply); reply->writeInt32( setMasterMute(data.readInt32()) ); return NO_ERROR; } break; case MASTER_VOLUME: { CHECK_INTERFACE(IAudioFlinger, data, reply); reply->writeFloat( masterVolume() ); return NO_ERROR; } break; case MASTER_MUTE: { CHECK_INTERFACE(IAudioFlinger, data, reply); reply->writeInt32( masterMute() ); return NO_ERROR; } break; case SET_STREAM_VOLUME: { CHECK_INTERFACE(IAudioFlinger, data, reply); int stream = data.readInt32(); float volume = data.readFloat(); audio_io_handle_t output = (audio_io_handle_t) data.readInt32(); reply->writeInt32( setStreamVolume((audio_stream_type_t) stream, volume, output) ); return NO_ERROR; } break; case SET_STREAM_MUTE: { CHECK_INTERFACE(IAudioFlinger, data, reply); int stream = data.readInt32(); reply->writeInt32( setStreamMute((audio_stream_type_t) stream, data.readInt32()) ); return NO_ERROR; } break; case STREAM_VOLUME: { CHECK_INTERFACE(IAudioFlinger, data, reply); int stream = data.readInt32(); int output = data.readInt32(); reply->writeFloat( streamVolume((audio_stream_type_t) stream, output) ); return NO_ERROR; } break; case STREAM_MUTE: { CHECK_INTERFACE(IAudioFlinger, data, reply); int stream = data.readInt32(); reply->writeInt32( streamMute((audio_stream_type_t) stream) ); return NO_ERROR; } break; case SET_MODE: { CHECK_INTERFACE(IAudioFlinger, data, reply); audio_mode_t mode = (audio_mode_t) data.readInt32(); reply->writeInt32( setMode(mode) ); return NO_ERROR; } break; case SET_MIC_MUTE: { CHECK_INTERFACE(IAudioFlinger, data, reply); int state = data.readInt32(); reply->writeInt32( setMicMute(state) ); return NO_ERROR; } break; case GET_MIC_MUTE: { CHECK_INTERFACE(IAudioFlinger, data, reply); reply->writeInt32( getMicMute() ); return NO_ERROR; } break; case SET_PARAMETERS: { CHECK_INTERFACE(IAudioFlinger, data, reply); audio_io_handle_t ioHandle = (audio_io_handle_t) data.readInt32(); String8 keyValuePairs(data.readString8()); reply->writeInt32(setParameters(ioHandle, keyValuePairs)); return NO_ERROR; } break; case GET_PARAMETERS: { CHECK_INTERFACE(IAudioFlinger, data, reply); audio_io_handle_t ioHandle = (audio_io_handle_t) data.readInt32(); String8 keys(data.readString8()); reply->writeString8(getParameters(ioHandle, keys)); return NO_ERROR; } break; case REGISTER_CLIENT: { CHECK_INTERFACE(IAudioFlinger, data, reply); sp<IAudioFlingerClient> client = interface_cast<IAudioFlingerClient>(data.readStrongBinder()); registerClient(client); return NO_ERROR; } break; case GET_INPUTBUFFERSIZE: { CHECK_INTERFACE(IAudioFlinger, data, reply); uint32_t sampleRate = data.readInt32(); audio_format_t format = (audio_format_t) data.readInt32(); int channelCount = data.readInt32(); reply->writeInt32( getInputBufferSize(sampleRate, format, channelCount) ); return NO_ERROR; } break; case OPEN_OUTPUT: { CHECK_INTERFACE(IAudioFlinger, data, reply); audio_module_handle_t module = (audio_module_handle_t)data.readInt32(); audio_devices_t devices = (audio_devices_t)data.readInt32(); uint32_t samplingRate = data.readInt32(); audio_format_t format = (audio_format_t) data.readInt32(); audio_channel_mask_t channelMask = (audio_channel_mask_t)data.readInt32(); uint32_t latency = data.readInt32(); audio_output_flags_t flags = (audio_output_flags_t) data.readInt32(); audio_io_handle_t output = openOutput(module, &devices, &samplingRate, &format, &channelMask, &latency, flags); ALOGV("OPEN_OUTPUT output, %p", output); reply->writeInt32((int32_t) output); reply->writeInt32(devices); reply->writeInt32(samplingRate); reply->writeInt32(format); reply->writeInt32(channelMask); reply->writeInt32(latency); return NO_ERROR; } break; case OPEN_DUPLICATE_OUTPUT: { CHECK_INTERFACE(IAudioFlinger, data, reply); audio_io_handle_t output1 = (audio_io_handle_t) data.readInt32(); audio_io_handle_t output2 = (audio_io_handle_t) data.readInt32(); reply->writeInt32((int32_t) openDuplicateOutput(output1, output2)); return NO_ERROR; } break; case CLOSE_OUTPUT: { CHECK_INTERFACE(IAudioFlinger, data, reply); reply->writeInt32(closeOutput((audio_io_handle_t) data.readInt32())); return NO_ERROR; } break; case SUSPEND_OUTPUT: { CHECK_INTERFACE(IAudioFlinger, data, reply); reply->writeInt32(suspendOutput((audio_io_handle_t) data.readInt32())); return NO_ERROR; } break; case RESTORE_OUTPUT: { CHECK_INTERFACE(IAudioFlinger, data, reply); reply->writeInt32(restoreOutput((audio_io_handle_t) data.readInt32())); return NO_ERROR; } break; case OPEN_INPUT: { CHECK_INTERFACE(IAudioFlinger, data, reply); audio_module_handle_t module = (audio_module_handle_t)data.readInt32(); audio_devices_t devices = (audio_devices_t)data.readInt32(); uint32_t samplingRate = data.readInt32(); audio_format_t format = (audio_format_t) data.readInt32(); audio_channel_mask_t channelMask = (audio_channel_mask_t)data.readInt32(); audio_io_handle_t input = openInput(module, &devices, &samplingRate, &format, &channelMask); reply->writeInt32((int32_t) input); reply->writeInt32(devices); reply->writeInt32(samplingRate); reply->writeInt32(format); reply->writeInt32(channelMask); return NO_ERROR; } break; case CLOSE_INPUT: { CHECK_INTERFACE(IAudioFlinger, data, reply); reply->writeInt32(closeInput((audio_io_handle_t) data.readInt32())); return NO_ERROR; } break; case SET_STREAM_OUTPUT: { CHECK_INTERFACE(IAudioFlinger, data, reply); uint32_t stream = data.readInt32(); audio_io_handle_t output = (audio_io_handle_t) data.readInt32(); reply->writeInt32(setStreamOutput((audio_stream_type_t) stream, output)); return NO_ERROR; } break; case SET_VOICE_VOLUME: { CHECK_INTERFACE(IAudioFlinger, data, reply); float volume = data.readFloat(); reply->writeInt32( setVoiceVolume(volume) ); return NO_ERROR; } break; case GET_RENDER_POSITION: { CHECK_INTERFACE(IAudioFlinger, data, reply); audio_io_handle_t output = (audio_io_handle_t) data.readInt32(); uint32_t halFrames; uint32_t dspFrames; status_t status = getRenderPosition(&halFrames, &dspFrames, output); reply->writeInt32(status); if (status == NO_ERROR) { reply->writeInt32(halFrames); reply->writeInt32(dspFrames); } return NO_ERROR; } case GET_INPUT_FRAMES_LOST: { CHECK_INTERFACE(IAudioFlinger, data, reply); audio_io_handle_t ioHandle = (audio_io_handle_t) data.readInt32(); reply->writeInt32(getInputFramesLost(ioHandle)); return NO_ERROR; } break; case NEW_AUDIO_SESSION_ID: { CHECK_INTERFACE(IAudioFlinger, data, reply); reply->writeInt32(newAudioSessionId()); return NO_ERROR; } break; case ACQUIRE_AUDIO_SESSION_ID: { CHECK_INTERFACE(IAudioFlinger, data, reply); int audioSession = data.readInt32(); acquireAudioSessionId(audioSession); return NO_ERROR; } break; case RELEASE_AUDIO_SESSION_ID: { CHECK_INTERFACE(IAudioFlinger, data, reply); int audioSession = data.readInt32(); releaseAudioSessionId(audioSession); return NO_ERROR; } break; case QUERY_NUM_EFFECTS: { CHECK_INTERFACE(IAudioFlinger, data, reply); uint32_t numEffects; status_t status = queryNumberEffects(&numEffects); reply->writeInt32(status); if (status == NO_ERROR) { reply->writeInt32((int32_t)numEffects); } return NO_ERROR; } case QUERY_EFFECT: { CHECK_INTERFACE(IAudioFlinger, data, reply); effect_descriptor_t desc; status_t status = queryEffect(data.readInt32(), &desc); reply->writeInt32(status); if (status == NO_ERROR) { reply->write(&desc, sizeof(effect_descriptor_t)); } return NO_ERROR; } case GET_EFFECT_DESCRIPTOR: { CHECK_INTERFACE(IAudioFlinger, data, reply); effect_uuid_t uuid; data.read(&uuid, sizeof(effect_uuid_t)); effect_descriptor_t desc; status_t status = getEffectDescriptor(&uuid, &desc); reply->writeInt32(status); if (status == NO_ERROR) { reply->write(&desc, sizeof(effect_descriptor_t)); } return NO_ERROR; } case CREATE_EFFECT: { CHECK_INTERFACE(IAudioFlinger, data, reply); pid_t pid = data.readInt32(); effect_descriptor_t desc; data.read(&desc, sizeof(effect_descriptor_t)); sp<IEffectClient> client = interface_cast<IEffectClient>(data.readStrongBinder()); int32_t priority = data.readInt32(); audio_io_handle_t output = (audio_io_handle_t) data.readInt32(); int sessionId = data.readInt32(); status_t status; int id; int enabled; sp<IEffect> effect = createEffect(pid, &desc, client, priority, output, sessionId, &status, &id, &enabled); reply->writeInt32(status); reply->writeInt32(id); reply->writeInt32(enabled); reply->writeStrongBinder(effect->asBinder()); reply->write(&desc, sizeof(effect_descriptor_t)); return NO_ERROR; } break; case MOVE_EFFECTS: { CHECK_INTERFACE(IAudioFlinger, data, reply); int session = data.readInt32(); audio_io_handle_t srcOutput = (audio_io_handle_t) data.readInt32(); audio_io_handle_t dstOutput = (audio_io_handle_t) data.readInt32(); reply->writeInt32(moveEffects(session, srcOutput, dstOutput)); return NO_ERROR; } break; case LOAD_HW_MODULE: { CHECK_INTERFACE(IAudioFlinger, data, reply); reply->writeInt32(loadHwModule(data.readCString())); return NO_ERROR; } break; default: return BBinder::onTransact(code, data, reply, flags); } } // ---------------------------------------------------------------------------- }; // namespace android