| /* |
| * Copyright (C) 2013-2016 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 "msm8974_platform" |
| /*#define LOG_NDEBUG 0*/ |
| #define LOG_NDDEBUG 0 |
| |
| #include <stdlib.h> |
| #include <dlfcn.h> |
| #include <cutils/log.h> |
| #include <cutils/str_parms.h> |
| #include <cutils/properties.h> |
| #include <audio_hw.h> |
| #include <platform_api.h> |
| #include "platform.h" |
| #include "audio_extn.h" |
| #include <linux/msm_audio.h> |
| |
| #define MIXER_XML_DEFAULT_PATH "/system/etc/mixer_paths.xml" |
| #define MIXER_XML_BASE_STRING "/system/etc/mixer_paths" |
| #define TOMTOM_8226_SND_CARD_NAME "msm8226-tomtom-snd-card" |
| #define TOMTOM_MIXER_FILE_SUFFIX "wcd9330" |
| |
| #define LIB_ACDB_LOADER "libacdbloader.so" |
| #define AUDIO_DATA_BLOCK_MIXER_CTL "HDMI EDID" |
| #define CVD_VERSION_MIXER_CTL "CVD Version" |
| |
| |
| /* |
| * This file will have a maximum of 38 bytes: |
| * |
| * 4 bytes: number of audio blocks |
| * 4 bytes: total length of Short Audio Descriptor (SAD) blocks |
| * Maximum 10 * 3 bytes: SAD blocks |
| */ |
| #define MAX_SAD_BLOCKS 10 |
| #define SAD_BLOCK_SIZE 3 |
| |
| #define MAX_CVD_VERSION_STRING_SIZE 100 |
| |
| /* EDID format ID for LPCM audio */ |
| #define EDID_FORMAT_LPCM 1 |
| |
| /* Retry for delay in FW loading*/ |
| #define RETRY_NUMBER 10 |
| #define RETRY_US 500000 |
| #define MAX_SND_CARD 8 |
| |
| #define MAX_SND_CARD_NAME_LEN 31 |
| |
| #define DEFAULT_APP_TYPE_RX_PATH 0x11130 |
| |
| #define TOSTRING_(x) #x |
| #define TOSTRING(x) TOSTRING_(x) |
| |
| struct audio_block_header |
| { |
| int reserved; |
| int length; |
| }; |
| |
| enum { |
| CAL_MODE_SEND = 0x1, |
| CAL_MODE_PERSIST = 0x2, |
| CAL_MODE_RTAC = 0x4 |
| }; |
| |
| #define PLATFORM_CONFIG_KEY_OPERATOR_INFO "operator_info" |
| |
| struct operator_info { |
| struct listnode list; |
| char *name; |
| char *mccmnc; |
| }; |
| |
| struct operator_specific_device { |
| struct listnode list; |
| char *operator; |
| char *mixer_path; |
| int acdb_id; |
| }; |
| |
| static struct listnode operator_info_list; |
| static struct listnode *operator_specific_device_table[SND_DEVICE_MAX]; |
| |
| /* Audio calibration related functions */ |
| typedef void (*acdb_deallocate_t)(); |
| typedef int (*acdb_init_v2_cvd_t)(char *, char *, int); |
| typedef int (*acdb_init_v2_t)(char *); |
| typedef int (*acdb_init_t)(); |
| typedef void (*acdb_send_audio_cal_t)(int, int); |
| typedef void (*acdb_send_voice_cal_t)(int, int); |
| typedef int (*acdb_reload_vocvoltable_t)(int); |
| typedef int (*acdb_send_gain_dep_cal_t)(int, int, int, int, int); |
| |
| /* Audio calibration related functions */ |
| struct platform_data { |
| struct audio_device *adev; |
| bool fluence_in_spkr_mode; |
| bool fluence_in_voice_call; |
| bool fluence_in_voice_comm; |
| bool fluence_in_voice_rec; |
| /* 0 = no fluence, 1 = fluence, 2 = fluence pro */ |
| int fluence_type; |
| int source_mic_type; |
| bool speaker_lr_swap; |
| |
| void *acdb_handle; |
| acdb_deallocate_t acdb_deallocate; |
| acdb_send_audio_cal_t acdb_send_audio_cal; |
| acdb_send_voice_cal_t acdb_send_voice_cal; |
| acdb_reload_vocvoltable_t acdb_reload_vocvoltable; |
| acdb_send_gain_dep_cal_t acdb_send_gain_dep_cal; |
| struct csd_data *csd; |
| char ec_ref_mixer_path[64]; |
| |
| char *snd_card_name; |
| int max_vol_index; |
| int max_mic_count; |
| |
| void *hw_info; |
| }; |
| |
| static int pcm_device_table[AUDIO_USECASE_MAX][2] = { |
| [USECASE_AUDIO_PLAYBACK_DEEP_BUFFER] = {DEEP_BUFFER_PCM_DEVICE, |
| DEEP_BUFFER_PCM_DEVICE}, |
| [USECASE_AUDIO_PLAYBACK_LOW_LATENCY] = {LOWLATENCY_PCM_DEVICE, |
| LOWLATENCY_PCM_DEVICE}, |
| [USECASE_AUDIO_PLAYBACK_MULTI_CH] = {MULTIMEDIA2_PCM_DEVICE, |
| MULTIMEDIA2_PCM_DEVICE}, |
| [USECASE_AUDIO_PLAYBACK_OFFLOAD] = {PLAYBACK_OFFLOAD_DEVICE, |
| PLAYBACK_OFFLOAD_DEVICE}, |
| [USECASE_AUDIO_PLAYBACK_TTS] = {MULTIMEDIA2_PCM_DEVICE, |
| MULTIMEDIA2_PCM_DEVICE}, |
| [USECASE_AUDIO_PLAYBACK_ULL] = {MULTIMEDIA3_PCM_DEVICE, |
| MULTIMEDIA3_PCM_DEVICE}, |
| |
| [USECASE_AUDIO_RECORD] = {AUDIO_RECORD_PCM_DEVICE, |
| AUDIO_RECORD_PCM_DEVICE}, |
| [USECASE_AUDIO_RECORD_LOW_LATENCY] = {LOWLATENCY_PCM_DEVICE, |
| LOWLATENCY_PCM_DEVICE}, |
| |
| [USECASE_VOICE_CALL] = {VOICE_CALL_PCM_DEVICE, |
| VOICE_CALL_PCM_DEVICE}, |
| [USECASE_VOICE2_CALL] = {VOICE2_CALL_PCM_DEVICE, VOICE2_CALL_PCM_DEVICE}, |
| [USECASE_VOLTE_CALL] = {VOLTE_CALL_PCM_DEVICE, VOLTE_CALL_PCM_DEVICE}, |
| [USECASE_QCHAT_CALL] = {QCHAT_CALL_PCM_DEVICE, QCHAT_CALL_PCM_DEVICE}, |
| [USECASE_VOWLAN_CALL] = {VOWLAN_CALL_PCM_DEVICE, VOWLAN_CALL_PCM_DEVICE}, |
| [USECASE_VOICEMMODE1_CALL] = {VOICEMMODE1_CALL_PCM_DEVICE, |
| VOICEMMODE1_CALL_PCM_DEVICE}, |
| [USECASE_VOICEMMODE2_CALL] = {VOICEMMODE2_CALL_PCM_DEVICE, |
| VOICEMMODE2_CALL_PCM_DEVICE}, |
| |
| [USECASE_INCALL_REC_UPLINK] = {AUDIO_RECORD_PCM_DEVICE, |
| AUDIO_RECORD_PCM_DEVICE}, |
| [USECASE_INCALL_REC_DOWNLINK] = {AUDIO_RECORD_PCM_DEVICE, |
| AUDIO_RECORD_PCM_DEVICE}, |
| [USECASE_INCALL_REC_UPLINK_AND_DOWNLINK] = {AUDIO_RECORD_PCM_DEVICE, |
| AUDIO_RECORD_PCM_DEVICE}, |
| [USECASE_AUDIO_HFP_SCO] = {HFP_PCM_RX, HFP_SCO_RX}, |
| |
| [USECASE_AUDIO_SPKR_CALIB_RX] = {SPKR_PROT_CALIB_RX_PCM_DEVICE, -1}, |
| [USECASE_AUDIO_SPKR_CALIB_TX] = {-1, SPKR_PROT_CALIB_TX_PCM_DEVICE}, |
| |
| [USECASE_AUDIO_PLAYBACK_AFE_PROXY] = {AFE_PROXY_PLAYBACK_PCM_DEVICE, |
| AFE_PROXY_RECORD_PCM_DEVICE}, |
| [USECASE_AUDIO_RECORD_AFE_PROXY] = {AFE_PROXY_PLAYBACK_PCM_DEVICE, |
| AFE_PROXY_RECORD_PCM_DEVICE}, |
| [USECASE_AUDIO_DSM_FEEDBACK] = {QUAT_MI2S_PCM_DEVICE, QUAT_MI2S_PCM_DEVICE}, |
| |
| }; |
| |
| /* Array to store sound devices */ |
| static const char * const device_table[SND_DEVICE_MAX] = { |
| [SND_DEVICE_NONE] = "none", |
| /* Playback sound devices */ |
| [SND_DEVICE_OUT_HANDSET] = "handset", |
| [SND_DEVICE_OUT_SPEAKER] = "speaker", |
| [SND_DEVICE_OUT_SPEAKER_REVERSE] = "speaker-reverse", |
| [SND_DEVICE_OUT_SPEAKER_SAFE] = "speaker-safe", |
| [SND_DEVICE_OUT_HEADPHONES] = "headphones", |
| [SND_DEVICE_OUT_LINE] = "line", |
| [SND_DEVICE_OUT_SPEAKER_AND_HEADPHONES] = "speaker-and-headphones", |
| [SND_DEVICE_OUT_SPEAKER_SAFE_AND_HEADPHONES] = "speaker-safe-and-headphones", |
| [SND_DEVICE_OUT_SPEAKER_AND_LINE] = "speaker-and-line", |
| [SND_DEVICE_OUT_SPEAKER_SAFE_AND_LINE] = "speaker-safe-and-line", |
| [SND_DEVICE_OUT_VOICE_HANDSET] = "voice-handset", |
| [SND_DEVICE_OUT_VOICE_HAC_HANDSET] = "voice-hac-handset", |
| [SND_DEVICE_OUT_VOICE_SPEAKER] = "voice-speaker", |
| [SND_DEVICE_OUT_VOICE_HEADPHONES] = "voice-headphones", |
| [SND_DEVICE_OUT_VOICE_LINE] = "voice-line", |
| [SND_DEVICE_OUT_HDMI] = "hdmi", |
| [SND_DEVICE_OUT_SPEAKER_AND_HDMI] = "speaker-and-hdmi", |
| [SND_DEVICE_OUT_BT_SCO] = "bt-sco-headset", |
| [SND_DEVICE_OUT_BT_SCO_WB] = "bt-sco-headset-wb", |
| [SND_DEVICE_OUT_VOICE_HANDSET_TMUS] = "voice-handset-tmus", |
| [SND_DEVICE_OUT_VOICE_TTY_FULL_HEADPHONES] = "voice-tty-full-headphones", |
| [SND_DEVICE_OUT_VOICE_TTY_VCO_HEADPHONES] = "voice-tty-vco-headphones", |
| [SND_DEVICE_OUT_VOICE_TTY_HCO_HANDSET] = "voice-tty-hco-handset", |
| [SND_DEVICE_OUT_VOICE_TX] = "voice-tx", |
| [SND_DEVICE_OUT_SPEAKER_PROTECTED] = "speaker-protected", |
| [SND_DEVICE_OUT_VOICE_SPEAKER_PROTECTED] = "voice-speaker-protected", |
| [SND_DEVICE_OUT_VOICE_SPEAKER_HFP] = "voice-speaker-hfp", |
| |
| /* Capture sound devices */ |
| [SND_DEVICE_IN_HANDSET_MIC] = "handset-mic", |
| [SND_DEVICE_IN_HANDSET_MIC_AEC] = "handset-mic", |
| [SND_DEVICE_IN_HANDSET_MIC_NS] = "handset-mic", |
| [SND_DEVICE_IN_HANDSET_MIC_AEC_NS] = "handset-mic", |
| [SND_DEVICE_IN_HANDSET_DMIC] = "dmic-endfire", |
| [SND_DEVICE_IN_HANDSET_DMIC_AEC] = "dmic-endfire", |
| [SND_DEVICE_IN_HANDSET_DMIC_NS] = "dmic-endfire", |
| [SND_DEVICE_IN_HANDSET_DMIC_AEC_NS] = "dmic-endfire", |
| [SND_DEVICE_IN_HANDSET_DMIC_STEREO] = "dmic-endfire", |
| |
| [SND_DEVICE_IN_SPEAKER_MIC] = "speaker-mic", |
| [SND_DEVICE_IN_SPEAKER_MIC_AEC] = "speaker-mic", |
| [SND_DEVICE_IN_SPEAKER_MIC_NS] = "speaker-mic", |
| [SND_DEVICE_IN_SPEAKER_MIC_AEC_NS] = "speaker-mic", |
| [SND_DEVICE_IN_SPEAKER_DMIC] = "speaker-dmic-endfire", |
| [SND_DEVICE_IN_SPEAKER_DMIC_AEC] = "speaker-dmic-endfire", |
| [SND_DEVICE_IN_SPEAKER_DMIC_NS] = "speaker-dmic-endfire", |
| [SND_DEVICE_IN_SPEAKER_DMIC_AEC_NS] = "speaker-dmic-endfire", |
| [SND_DEVICE_IN_SPEAKER_DMIC_STEREO] = "speaker-dmic-endfire", |
| |
| [SND_DEVICE_IN_HEADSET_MIC] = "headset-mic", |
| [SND_DEVICE_IN_HEADSET_MIC_AEC] = "headset-mic", |
| |
| [SND_DEVICE_IN_HDMI_MIC] = "hdmi-mic", |
| [SND_DEVICE_IN_BT_SCO_MIC] = "bt-sco-mic", |
| [SND_DEVICE_IN_BT_SCO_MIC_NREC] = "bt-sco-mic", |
| [SND_DEVICE_IN_BT_SCO_MIC_WB] = "bt-sco-mic-wb", |
| [SND_DEVICE_IN_BT_SCO_MIC_WB_NREC] = "bt-sco-mic-wb", |
| [SND_DEVICE_IN_CAMCORDER_MIC] = "camcorder-mic", |
| |
| [SND_DEVICE_IN_VOICE_DMIC] = "voice-dmic-ef", |
| [SND_DEVICE_IN_VOICE_DMIC_TMUS] = "voice-dmic-ef-tmus", |
| [SND_DEVICE_IN_VOICE_SPEAKER_MIC] = "voice-speaker-mic", |
| [SND_DEVICE_IN_VOICE_SPEAKER_DMIC] = "voice-speaker-dmic-ef", |
| [SND_DEVICE_IN_VOICE_SPEAKER_MIC_HFP] = "voice-speaker-mic-hfp", |
| [SND_DEVICE_IN_VOICE_HEADSET_MIC] = "voice-headset-mic", |
| [SND_DEVICE_IN_VOICE_TTY_FULL_HEADSET_MIC] = "voice-tty-full-headset-mic", |
| [SND_DEVICE_IN_VOICE_TTY_VCO_HANDSET_MIC] = "voice-tty-vco-handset-mic", |
| [SND_DEVICE_IN_VOICE_TTY_HCO_HEADSET_MIC] = "voice-tty-hco-headset-mic", |
| |
| [SND_DEVICE_IN_VOICE_REC_MIC] = "voice-rec-mic", |
| [SND_DEVICE_IN_VOICE_REC_MIC_NS] = "voice-rec-mic", |
| [SND_DEVICE_IN_VOICE_REC_DMIC_STEREO] = "voice-rec-dmic-ef", |
| [SND_DEVICE_IN_VOICE_REC_DMIC_FLUENCE] = "voice-rec-dmic-ef-fluence", |
| [SND_DEVICE_IN_VOICE_REC_HEADSET_MIC] = "headset-mic", |
| |
| [SND_DEVICE_IN_UNPROCESSED_MIC] = "voice-rec-mic", |
| [SND_DEVICE_IN_UNPROCESSED_HEADSET_MIC] = "headset-mic", |
| |
| [SND_DEVICE_IN_VOICE_RX] = "voice-rx", |
| |
| [SND_DEVICE_IN_THREE_MIC] = "three-mic", |
| [SND_DEVICE_IN_QUAD_MIC] = "quad-mic", |
| [SND_DEVICE_IN_CAPTURE_VI_FEEDBACK] = "vi-feedback", |
| [SND_DEVICE_IN_HANDSET_TMIC] = "three-mic", |
| [SND_DEVICE_IN_HANDSET_QMIC] = "quad-mic", |
| }; |
| |
| /* ACDB IDs (audio DSP path configuration IDs) for each sound device */ |
| static int acdb_device_table[SND_DEVICE_MAX] = { |
| [SND_DEVICE_NONE] = -1, |
| [SND_DEVICE_OUT_HANDSET] = 7, |
| [SND_DEVICE_OUT_SPEAKER] = 15, |
| [SND_DEVICE_OUT_SPEAKER_REVERSE] = 15, |
| [SND_DEVICE_OUT_SPEAKER_SAFE] = 15, |
| [SND_DEVICE_OUT_HEADPHONES] = 10, |
| [SND_DEVICE_OUT_LINE] = 77, |
| [SND_DEVICE_OUT_SPEAKER_AND_HEADPHONES] = 10, |
| [SND_DEVICE_OUT_SPEAKER_SAFE_AND_HEADPHONES] = 10, |
| [SND_DEVICE_OUT_SPEAKER_AND_LINE] = 77, |
| [SND_DEVICE_OUT_SPEAKER_SAFE_AND_LINE] = 77, |
| [SND_DEVICE_OUT_VOICE_HANDSET] = ACDB_ID_VOICE_HANDSET, |
| [SND_DEVICE_OUT_VOICE_SPEAKER] = ACDB_ID_VOICE_SPEAKER, |
| [SND_DEVICE_OUT_VOICE_HAC_HANDSET] = 53, |
| [SND_DEVICE_OUT_VOICE_HEADPHONES] = 10, |
| [SND_DEVICE_OUT_VOICE_LINE] = 77, |
| [SND_DEVICE_OUT_HDMI] = 18, |
| [SND_DEVICE_OUT_SPEAKER_AND_HDMI] = 15, |
| [SND_DEVICE_OUT_BT_SCO] = 22, |
| [SND_DEVICE_OUT_BT_SCO_WB] = 39, |
| [SND_DEVICE_OUT_VOICE_HANDSET_TMUS] = ACDB_ID_VOICE_HANDSET_TMUS, |
| [SND_DEVICE_OUT_VOICE_TTY_FULL_HEADPHONES] = 17, |
| [SND_DEVICE_OUT_VOICE_TTY_VCO_HEADPHONES] = 17, |
| [SND_DEVICE_OUT_VOICE_TTY_HCO_HANDSET] = 37, |
| [SND_DEVICE_OUT_VOICE_TX] = 45, |
| [SND_DEVICE_OUT_SPEAKER_PROTECTED] = 124, |
| [SND_DEVICE_OUT_VOICE_SPEAKER_PROTECTED] = 101, |
| [SND_DEVICE_OUT_VOICE_SPEAKER_HFP] = ACDB_ID_VOICE_SPEAKER, |
| |
| [SND_DEVICE_IN_HANDSET_MIC] = 4, |
| [SND_DEVICE_IN_HANDSET_MIC_AEC] = 106, |
| [SND_DEVICE_IN_HANDSET_MIC_NS] = 107, |
| [SND_DEVICE_IN_HANDSET_MIC_AEC_NS] = 108, |
| [SND_DEVICE_IN_HANDSET_DMIC] = 41, |
| [SND_DEVICE_IN_HANDSET_DMIC_AEC] = 109, |
| [SND_DEVICE_IN_HANDSET_DMIC_NS] = 110, |
| [SND_DEVICE_IN_HANDSET_DMIC_AEC_NS] = 111, |
| [SND_DEVICE_IN_HANDSET_DMIC_STEREO] = 34, |
| |
| [SND_DEVICE_IN_SPEAKER_MIC] = 11, |
| [SND_DEVICE_IN_SPEAKER_MIC_AEC] = 112, |
| [SND_DEVICE_IN_SPEAKER_MIC_NS] = 113, |
| [SND_DEVICE_IN_SPEAKER_MIC_AEC_NS] = 114, |
| [SND_DEVICE_IN_SPEAKER_DMIC] = 43, |
| [SND_DEVICE_IN_SPEAKER_DMIC_AEC] = 115, |
| [SND_DEVICE_IN_SPEAKER_DMIC_NS] = 116, |
| [SND_DEVICE_IN_SPEAKER_DMIC_AEC_NS] = 117, |
| [SND_DEVICE_IN_SPEAKER_DMIC_STEREO] = 35, |
| |
| [SND_DEVICE_IN_HEADSET_MIC] = ACDB_ID_HEADSET_MIC_AEC, |
| [SND_DEVICE_IN_HEADSET_MIC_AEC] = ACDB_ID_HEADSET_MIC_AEC, |
| |
| [SND_DEVICE_IN_HDMI_MIC] = 4, |
| [SND_DEVICE_IN_BT_SCO_MIC] = 21, |
| [SND_DEVICE_IN_BT_SCO_MIC_NREC] = 21, |
| [SND_DEVICE_IN_BT_SCO_MIC_WB] = 38, |
| [SND_DEVICE_IN_BT_SCO_MIC_WB_NREC] = 38, |
| [SND_DEVICE_IN_CAMCORDER_MIC] = 61, |
| |
| [SND_DEVICE_IN_VOICE_DMIC] = 41, |
| [SND_DEVICE_IN_VOICE_DMIC_TMUS] = ACDB_ID_VOICE_DMIC_EF_TMUS, |
| [SND_DEVICE_IN_VOICE_SPEAKER_MIC] = 11, |
| [SND_DEVICE_IN_VOICE_SPEAKER_MIC_HFP] = 11, |
| [SND_DEVICE_IN_VOICE_SPEAKER_DMIC] = 43, |
| [SND_DEVICE_IN_VOICE_HEADSET_MIC] = ACDB_ID_HEADSET_MIC_AEC, |
| [SND_DEVICE_IN_VOICE_TTY_FULL_HEADSET_MIC] = 16, |
| [SND_DEVICE_IN_VOICE_TTY_VCO_HANDSET_MIC] = 36, |
| [SND_DEVICE_IN_VOICE_TTY_HCO_HEADSET_MIC] = 16, |
| |
| [SND_DEVICE_IN_VOICE_REC_MIC] = ACDB_ID_VOICE_REC_MIC, |
| [SND_DEVICE_IN_VOICE_REC_MIC_NS] = 113, |
| [SND_DEVICE_IN_VOICE_REC_DMIC_STEREO] = 35, |
| [SND_DEVICE_IN_VOICE_REC_DMIC_FLUENCE] = 43, |
| [SND_DEVICE_IN_VOICE_REC_HEADSET_MIC] = ACDB_ID_HEADSET_MIC_AEC, |
| |
| [SND_DEVICE_IN_UNPROCESSED_MIC] = ACDB_ID_VOICE_REC_MIC, |
| [SND_DEVICE_IN_UNPROCESSED_HEADSET_MIC] = ACDB_ID_HEADSET_MIC_AEC, |
| |
| [SND_DEVICE_IN_VOICE_RX] = 44, |
| |
| [SND_DEVICE_IN_THREE_MIC] = 46, |
| [SND_DEVICE_IN_QUAD_MIC] = 46, |
| [SND_DEVICE_IN_CAPTURE_VI_FEEDBACK] = 102, |
| [SND_DEVICE_IN_HANDSET_TMIC] = 125, |
| [SND_DEVICE_IN_HANDSET_QMIC] = 125, |
| }; |
| |
| struct name_to_index { |
| char name[100]; |
| unsigned int index; |
| }; |
| |
| #define TO_NAME_INDEX(X) #X, X |
| |
| /* Used to get index from parsed string */ |
| static const struct name_to_index snd_device_name_index[SND_DEVICE_MAX] = { |
| /* out */ |
| {TO_NAME_INDEX(SND_DEVICE_OUT_HANDSET)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_SPEAKER)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_SPEAKER_REVERSE)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_SPEAKER_SAFE)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_HEADPHONES)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_LINE)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_SPEAKER_AND_HEADPHONES)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_SPEAKER_SAFE_AND_HEADPHONES)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_SPEAKER_AND_LINE)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_SPEAKER_SAFE_AND_LINE)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_VOICE_HANDSET)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_VOICE_SPEAKER)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_VOICE_SPEAKER_HFP)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_VOICE_HEADPHONES)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_VOICE_LINE)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_HDMI)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_SPEAKER_AND_HDMI)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_BT_SCO)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_BT_SCO_WB)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_VOICE_HANDSET_TMUS)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_VOICE_HAC_HANDSET)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_VOICE_TTY_FULL_HEADPHONES)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_VOICE_TTY_VCO_HEADPHONES)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_VOICE_TTY_HCO_HANDSET)}, |
| |
| /* in */ |
| {TO_NAME_INDEX(SND_DEVICE_OUT_SPEAKER_PROTECTED)}, |
| {TO_NAME_INDEX(SND_DEVICE_OUT_VOICE_SPEAKER_PROTECTED)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_HANDSET_MIC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_HANDSET_MIC_AEC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_HANDSET_MIC_NS)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_HANDSET_MIC_AEC_NS)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_HANDSET_DMIC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_HANDSET_DMIC_AEC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_HANDSET_DMIC_NS)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_HANDSET_DMIC_AEC_NS)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_HANDSET_DMIC_STEREO)}, |
| |
| {TO_NAME_INDEX(SND_DEVICE_IN_SPEAKER_MIC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_SPEAKER_MIC_AEC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_SPEAKER_MIC_NS)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_SPEAKER_MIC_AEC_NS)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_SPEAKER_DMIC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_SPEAKER_DMIC_AEC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_SPEAKER_DMIC_NS)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_SPEAKER_DMIC_AEC_NS)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_SPEAKER_DMIC_STEREO)}, |
| |
| {TO_NAME_INDEX(SND_DEVICE_IN_HEADSET_MIC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_HEADSET_MIC_AEC)}, |
| |
| {TO_NAME_INDEX(SND_DEVICE_IN_HDMI_MIC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_BT_SCO_MIC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_BT_SCO_MIC_NREC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_BT_SCO_MIC_WB)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_BT_SCO_MIC_WB_NREC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_CAMCORDER_MIC)}, |
| |
| {TO_NAME_INDEX(SND_DEVICE_IN_VOICE_DMIC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_VOICE_DMIC_TMUS)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_VOICE_SPEAKER_MIC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_VOICE_SPEAKER_MIC_HFP)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_VOICE_SPEAKER_DMIC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_VOICE_HEADSET_MIC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_VOICE_TTY_FULL_HEADSET_MIC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_VOICE_TTY_VCO_HANDSET_MIC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_VOICE_TTY_HCO_HEADSET_MIC)}, |
| |
| {TO_NAME_INDEX(SND_DEVICE_IN_VOICE_REC_MIC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_VOICE_REC_MIC_NS)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_VOICE_REC_DMIC_STEREO)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_VOICE_REC_DMIC_FLUENCE)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_VOICE_REC_HEADSET_MIC)}, |
| |
| {TO_NAME_INDEX(SND_DEVICE_IN_UNPROCESSED_MIC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_UNPROCESSED_HEADSET_MIC)}, |
| |
| {TO_NAME_INDEX(SND_DEVICE_IN_THREE_MIC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_QUAD_MIC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_CAPTURE_VI_FEEDBACK)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_HANDSET_TMIC)}, |
| {TO_NAME_INDEX(SND_DEVICE_IN_HANDSET_QMIC)}, |
| }; |
| |
| static char * backend_tag_table[SND_DEVICE_MAX] = {0}; |
| static char * hw_interface_table[SND_DEVICE_MAX] = {0}; |
| |
| static const struct name_to_index usecase_name_index[AUDIO_USECASE_MAX] = { |
| {TO_NAME_INDEX(USECASE_AUDIO_PLAYBACK_DEEP_BUFFER)}, |
| {TO_NAME_INDEX(USECASE_AUDIO_PLAYBACK_LOW_LATENCY)}, |
| {TO_NAME_INDEX(USECASE_AUDIO_PLAYBACK_MULTI_CH)}, |
| {TO_NAME_INDEX(USECASE_AUDIO_PLAYBACK_OFFLOAD)}, |
| {TO_NAME_INDEX(USECASE_AUDIO_PLAYBACK_TTS)}, |
| {TO_NAME_INDEX(USECASE_AUDIO_PLAYBACK_ULL)}, |
| {TO_NAME_INDEX(USECASE_AUDIO_RECORD)}, |
| {TO_NAME_INDEX(USECASE_AUDIO_RECORD_LOW_LATENCY)}, |
| {TO_NAME_INDEX(USECASE_VOICE_CALL)}, |
| {TO_NAME_INDEX(USECASE_VOICE2_CALL)}, |
| {TO_NAME_INDEX(USECASE_VOLTE_CALL)}, |
| {TO_NAME_INDEX(USECASE_QCHAT_CALL)}, |
| {TO_NAME_INDEX(USECASE_VOWLAN_CALL)}, |
| {TO_NAME_INDEX(USECASE_INCALL_REC_UPLINK)}, |
| {TO_NAME_INDEX(USECASE_INCALL_REC_DOWNLINK)}, |
| {TO_NAME_INDEX(USECASE_INCALL_REC_UPLINK_AND_DOWNLINK)}, |
| {TO_NAME_INDEX(USECASE_AUDIO_HFP_SCO)}, |
| }; |
| |
| #define DEEP_BUFFER_PLATFORM_DELAY (29*1000LL) |
| #define LOW_LATENCY_PLATFORM_DELAY (13*1000LL) |
| |
| static pthread_once_t check_op_once_ctl = PTHREAD_ONCE_INIT; |
| static bool is_tmus = false; |
| |
| static void check_operator() |
| { |
| char value[PROPERTY_VALUE_MAX]; |
| int mccmnc; |
| property_get("gsm.sim.operator.numeric",value,"0"); |
| mccmnc = atoi(value); |
| ALOGD("%s: tmus mccmnc %d", __func__, mccmnc); |
| switch(mccmnc) { |
| /* TMUS MCC(310), MNC(490, 260, 026) */ |
| case 310490: |
| case 310260: |
| case 310026: |
| /* Add new TMUS MNC(800, 660, 580, 310, 270, 250, 240, 230, 220, 210, 200, 160) */ |
| case 310800: |
| case 310660: |
| case 310580: |
| case 310310: |
| case 310270: |
| case 310250: |
| case 310240: |
| case 310230: |
| case 310220: |
| case 310210: |
| case 310200: |
| case 310160: |
| is_tmus = true; |
| break; |
| } |
| } |
| |
| bool is_operator_tmus() |
| { |
| pthread_once(&check_op_once_ctl, check_operator); |
| return is_tmus; |
| } |
| |
| static char *get_current_operator() |
| { |
| struct listnode *node; |
| struct operator_info *info_item; |
| char mccmnc[PROPERTY_VALUE_MAX]; |
| char *ret = NULL; |
| |
| property_get("gsm.sim.operator.numeric",mccmnc,"0"); |
| |
| list_for_each(node, &operator_info_list) { |
| info_item = node_to_item(node, struct operator_info, list); |
| if (strstr(info_item->mccmnc, mccmnc) != NULL) { |
| ret = info_item->name; |
| } |
| } |
| |
| return ret; |
| } |
| |
| static struct operator_specific_device *get_operator_specific_device(snd_device_t snd_device) |
| { |
| struct listnode *node; |
| struct operator_specific_device *ret = NULL; |
| struct operator_specific_device *device_item; |
| char *operator_name; |
| |
| operator_name = get_current_operator(); |
| if (operator_name == NULL) |
| return ret; |
| |
| list_for_each(node, operator_specific_device_table[snd_device]) { |
| device_item = node_to_item(node, struct operator_specific_device, list); |
| if (strcmp(operator_name, device_item->operator) == 0) { |
| ret = device_item; |
| } |
| } |
| |
| return ret; |
| } |
| |
| |
| static int get_operator_specific_device_acdb_id(snd_device_t snd_device) |
| { |
| struct operator_specific_device *device; |
| int ret = acdb_device_table[snd_device]; |
| |
| device = get_operator_specific_device(snd_device); |
| if (device != NULL) |
| ret = device->acdb_id; |
| |
| return ret; |
| } |
| |
| static const char *get_operator_specific_device_mixer_path(snd_device_t snd_device) |
| { |
| struct operator_specific_device *device; |
| const char *ret = device_table[snd_device]; |
| |
| device = get_operator_specific_device(snd_device); |
| if (device != NULL) |
| ret = device->mixer_path; |
| |
| return ret; |
| } |
| |
| bool platform_send_gain_dep_cal(void *platform, int level) |
| { |
| bool ret_val = false; |
| struct platform_data *my_data = (struct platform_data *)platform; |
| struct audio_device *adev = my_data->adev; |
| int acdb_dev_id, app_type; |
| int acdb_dev_type = MSM_SNDDEV_CAP_RX; |
| int mode = CAL_MODE_RTAC; |
| struct listnode *node; |
| struct audio_usecase *usecase; |
| |
| if (my_data->acdb_send_gain_dep_cal == NULL) { |
| ALOGE("%s: dlsym error for acdb_send_gain_dep_cal", __func__); |
| return ret_val; |
| } |
| |
| if (!voice_is_in_call(adev)) { |
| ALOGV("%s: Not Voice call usecase, apply new cal for level %d", |
| __func__, level); |
| app_type = DEFAULT_APP_TYPE_RX_PATH; |
| |
| // find the current active sound device |
| list_for_each(node, &adev->usecase_list) { |
| usecase = node_to_item(node, struct audio_usecase, list); |
| |
| if (usecase != NULL && |
| usecase->type == PCM_PLAYBACK && |
| (usecase->stream.out->devices == AUDIO_DEVICE_OUT_SPEAKER)) { |
| |
| ALOGV("%s: out device is %d", __func__, usecase->out_snd_device); |
| if (audio_extn_spkr_prot_is_enabled()) { |
| acdb_dev_id = audio_extn_spkr_prot_get_acdb_id(usecase->out_snd_device); |
| } else { |
| acdb_dev_id = acdb_device_table[usecase->out_snd_device]; |
| } |
| |
| if (!my_data->acdb_send_gain_dep_cal(acdb_dev_id, app_type, |
| acdb_dev_type, mode, level)) { |
| // set ret_val true if at least one calibration is set successfully |
| ret_val = true; |
| } else { |
| ALOGE("%s: my_data->acdb_send_gain_dep_cal failed ", __func__); |
| } |
| } else { |
| ALOGW("%s: Usecase list is empty", __func__); |
| } |
| } |
| } else { |
| ALOGW("%s: Voice call in progress .. ignore setting new cal", |
| __func__); |
| } |
| return ret_val; |
| } |
| |
| void platform_set_echo_reference(struct audio_device *adev, bool enable, audio_devices_t out_device) |
| { |
| struct platform_data *my_data = (struct platform_data *)adev->platform; |
| snd_device_t snd_device = SND_DEVICE_NONE; |
| |
| if (strcmp(my_data->ec_ref_mixer_path, "")) { |
| ALOGV("%s: diabling %s", __func__, my_data->ec_ref_mixer_path); |
| audio_route_reset_and_update_path(adev->audio_route, my_data->ec_ref_mixer_path); |
| } |
| |
| if (enable) { |
| strcpy(my_data->ec_ref_mixer_path, "echo-reference"); |
| if (out_device != AUDIO_DEVICE_NONE) { |
| snd_device = platform_get_output_snd_device(adev->platform, out_device); |
| platform_add_backend_name(adev->platform, my_data->ec_ref_mixer_path, snd_device); |
| } |
| |
| ALOGV("%s: enabling %s", __func__, my_data->ec_ref_mixer_path); |
| audio_route_apply_and_update_path(adev->audio_route, my_data->ec_ref_mixer_path); |
| } |
| } |
| |
| static struct csd_data *open_csd_client(bool i2s_ext_modem) |
| { |
| struct csd_data *csd = calloc(1, sizeof(struct csd_data)); |
| |
| csd->csd_client = dlopen(LIB_CSD_CLIENT, RTLD_NOW); |
| if (csd->csd_client == NULL) { |
| ALOGE("%s: DLOPEN failed for %s", __func__, LIB_CSD_CLIENT); |
| goto error; |
| } else { |
| ALOGV("%s: DLOPEN successful for %s", __func__, LIB_CSD_CLIENT); |
| |
| csd->deinit = (deinit_t)dlsym(csd->csd_client, |
| "csd_client_deinit"); |
| if (csd->deinit == NULL) { |
| ALOGE("%s: dlsym error %s for csd_client_deinit", __func__, |
| dlerror()); |
| goto error; |
| } |
| csd->disable_device = (disable_device_t)dlsym(csd->csd_client, |
| "csd_client_disable_device"); |
| if (csd->disable_device == NULL) { |
| ALOGE("%s: dlsym error %s for csd_client_disable_device", |
| __func__, dlerror()); |
| goto error; |
| } |
| csd->enable_device_config = (enable_device_config_t)dlsym(csd->csd_client, |
| "csd_client_enable_device_config"); |
| if (csd->enable_device_config == NULL) { |
| ALOGE("%s: dlsym error %s for csd_client_enable_device_config", |
| __func__, dlerror()); |
| goto error; |
| } |
| csd->enable_device = (enable_device_t)dlsym(csd->csd_client, |
| "csd_client_enable_device"); |
| if (csd->enable_device == NULL) { |
| ALOGE("%s: dlsym error %s for csd_client_enable_device", |
| __func__, dlerror()); |
| goto error; |
| } |
| csd->start_voice = (start_voice_t)dlsym(csd->csd_client, |
| "csd_client_start_voice"); |
| if (csd->start_voice == NULL) { |
| ALOGE("%s: dlsym error %s for csd_client_start_voice", |
| __func__, dlerror()); |
| goto error; |
| } |
| csd->stop_voice = (stop_voice_t)dlsym(csd->csd_client, |
| "csd_client_stop_voice"); |
| if (csd->stop_voice == NULL) { |
| ALOGE("%s: dlsym error %s for csd_client_stop_voice", |
| __func__, dlerror()); |
| goto error; |
| } |
| csd->volume = (volume_t)dlsym(csd->csd_client, |
| "csd_client_volume"); |
| if (csd->volume == NULL) { |
| ALOGE("%s: dlsym error %s for csd_client_volume", |
| __func__, dlerror()); |
| goto error; |
| } |
| csd->mic_mute = (mic_mute_t)dlsym(csd->csd_client, |
| "csd_client_mic_mute"); |
| if (csd->mic_mute == NULL) { |
| ALOGE("%s: dlsym error %s for csd_client_mic_mute", |
| __func__, dlerror()); |
| goto error; |
| } |
| csd->slow_talk = (slow_talk_t)dlsym(csd->csd_client, |
| "csd_client_slow_talk"); |
| if (csd->slow_talk == NULL) { |
| ALOGE("%s: dlsym error %s for csd_client_slow_talk", |
| __func__, dlerror()); |
| goto error; |
| } |
| csd->start_playback = (start_playback_t)dlsym(csd->csd_client, |
| "csd_client_start_playback"); |
| if (csd->start_playback == NULL) { |
| ALOGE("%s: dlsym error %s for csd_client_start_playback", |
| __func__, dlerror()); |
| goto error; |
| } |
| csd->stop_playback = (stop_playback_t)dlsym(csd->csd_client, |
| "csd_client_stop_playback"); |
| if (csd->stop_playback == NULL) { |
| ALOGE("%s: dlsym error %s for csd_client_stop_playback", |
| __func__, dlerror()); |
| goto error; |
| } |
| csd->start_record = (start_record_t)dlsym(csd->csd_client, |
| "csd_client_start_record"); |
| if (csd->start_record == NULL) { |
| ALOGE("%s: dlsym error %s for csd_client_start_record", |
| __func__, dlerror()); |
| goto error; |
| } |
| csd->stop_record = (stop_record_t)dlsym(csd->csd_client, |
| "csd_client_stop_record"); |
| if (csd->stop_record == NULL) { |
| ALOGE("%s: dlsym error %s for csd_client_stop_record", |
| __func__, dlerror()); |
| goto error; |
| } |
| |
| csd->get_sample_rate = (get_sample_rate_t)dlsym(csd->csd_client, |
| "csd_client_get_sample_rate"); |
| if (csd->get_sample_rate == NULL) { |
| ALOGE("%s: dlsym error %s for csd_client_get_sample_rate", |
| __func__, dlerror()); |
| |
| goto error; |
| } |
| |
| csd->init = (init_t)dlsym(csd->csd_client, "csd_client_init"); |
| |
| if (csd->init == NULL) { |
| ALOGE("%s: dlsym error %s for csd_client_init", |
| __func__, dlerror()); |
| goto error; |
| } else { |
| csd->init(i2s_ext_modem); |
| } |
| } |
| return csd; |
| |
| error: |
| free(csd); |
| csd = NULL; |
| return csd; |
| } |
| |
| void close_csd_client(struct csd_data *csd) |
| { |
| if (csd != NULL) { |
| csd->deinit(); |
| dlclose(csd->csd_client); |
| free(csd); |
| csd = NULL; |
| } |
| } |
| |
| static void platform_csd_init(struct platform_data *my_data) |
| { |
| #ifdef PLATFORM_MSM8084 |
| int32_t modems, (*count_modems)(void); |
| const char *name = "libdetectmodem.so"; |
| const char *func = "count_modems"; |
| const char *error; |
| |
| my_data->csd = NULL; |
| |
| void *lib = dlopen(name, RTLD_NOW); |
| error = dlerror(); |
| if (!lib) { |
| ALOGE("%s: could not find %s: %s", __func__, name, error); |
| return; |
| } |
| |
| count_modems = NULL; |
| *(void **)(&count_modems) = dlsym(lib, func); |
| error = dlerror(); |
| if (!count_modems) { |
| ALOGE("%s: could not find symbol %s in %s: %s", |
| __func__, func, name, error); |
| goto done; |
| } |
| |
| modems = count_modems(); |
| if (modems < 0) { |
| ALOGE("%s: count_modems failed\n", __func__); |
| goto done; |
| } |
| |
| ALOGD("%s: num_modems %d\n", __func__, modems); |
| if (modems > 0) |
| my_data->csd = open_csd_client(false /*is_i2s_ext_modem*/); |
| |
| done: |
| dlclose(lib); |
| #else |
| my_data->csd = NULL; |
| #endif |
| } |
| |
| static void set_platform_defaults(struct platform_data * my_data) |
| { |
| int32_t dev; |
| for (dev = 0; dev < SND_DEVICE_MAX; dev++) { |
| backend_tag_table[dev] = NULL; |
| hw_interface_table[dev] = NULL; |
| operator_specific_device_table[dev] = NULL; |
| } |
| |
| // To overwrite these go to the audio_platform_info.xml file. |
| backend_tag_table[SND_DEVICE_IN_BT_SCO_MIC] = strdup("bt-sco"); |
| backend_tag_table[SND_DEVICE_IN_BT_SCO_MIC_NREC] = strdup("bt-sco"); |
| backend_tag_table[SND_DEVICE_OUT_BT_SCO] = strdup("bt-sco"); |
| backend_tag_table[SND_DEVICE_OUT_HDMI] = strdup("hdmi"); |
| backend_tag_table[SND_DEVICE_OUT_SPEAKER_AND_HDMI] = strdup("speaker-and-hdmi"); |
| backend_tag_table[SND_DEVICE_OUT_BT_SCO_WB] = strdup("bt-sco-wb"); |
| backend_tag_table[SND_DEVICE_IN_BT_SCO_MIC_WB] = strdup("bt-sco-wb"); |
| backend_tag_table[SND_DEVICE_IN_BT_SCO_MIC_WB_NREC] = strdup("bt-sco-wb"); |
| backend_tag_table[SND_DEVICE_OUT_VOICE_TX] = strdup("afe-proxy"); |
| backend_tag_table[SND_DEVICE_IN_VOICE_RX] = strdup("afe-proxy"); |
| |
| hw_interface_table[SND_DEVICE_OUT_HANDSET] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_SPEAKER] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_SPEAKER_REVERSE] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_SPEAKER_SAFE] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_HEADPHONES] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_LINE] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_SPEAKER_AND_HEADPHONES] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_SPEAKER_SAFE_AND_HEADPHONES] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_SPEAKER_AND_LINE] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_SPEAKER_SAFE_AND_LINE] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_VOICE_HANDSET] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_VOICE_HAC_HANDSET] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_VOICE_SPEAKER] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_VOICE_HEADPHONES] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_VOICE_LINE] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_HDMI] = strdup("HDMI_RX"); |
| hw_interface_table[SND_DEVICE_OUT_SPEAKER_AND_HDMI] = strdup("SLIMBUS_0_RX-and-HDMI_RX"); |
| hw_interface_table[SND_DEVICE_OUT_BT_SCO] = strdup("SEC_AUX_PCM_RX"); |
| hw_interface_table[SND_DEVICE_OUT_BT_SCO_WB] = strdup("SEC_AUX_PCM_RX"); |
| hw_interface_table[SND_DEVICE_OUT_VOICE_HANDSET_TMUS] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_VOICE_TTY_FULL_HEADPHONES] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_VOICE_TTY_VCO_HEADPHONES] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_VOICE_TTY_HCO_HANDSET] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_VOICE_TX] = strdup("AFE_PCM_RX"); |
| hw_interface_table[SND_DEVICE_OUT_SPEAKER_PROTECTED] = strdup("SLIMBUS_0_RX"); |
| hw_interface_table[SND_DEVICE_OUT_VOICE_SPEAKER_PROTECTED] = strdup("SLIMBUS_0_RX"); |
| |
| my_data->max_mic_count = PLATFORM_DEFAULT_MIC_COUNT; |
| } |
| |
| void get_cvd_version(char *cvd_version, struct audio_device *adev) |
| { |
| struct mixer_ctl *ctl; |
| int count; |
| int ret = 0; |
| |
| ctl = mixer_get_ctl_by_name(adev->mixer, CVD_VERSION_MIXER_CTL); |
| if (!ctl) { |
| ALOGE("%s: Could not get ctl for mixer cmd - %s", __func__, CVD_VERSION_MIXER_CTL); |
| goto done; |
| } |
| mixer_ctl_update(ctl); |
| |
| count = mixer_ctl_get_num_values(ctl); |
| if (count > MAX_CVD_VERSION_STRING_SIZE) |
| count = MAX_CVD_VERSION_STRING_SIZE - 1; |
| |
| ret = mixer_ctl_get_array(ctl, cvd_version, count); |
| if (ret != 0) { |
| ALOGE("%s: ERROR! mixer_ctl_get_array() failed to get CVD Version", __func__); |
| goto done; |
| } |
| |
| done: |
| return; |
| } |
| |
| void *platform_init(struct audio_device *adev) |
| { |
| char value[PROPERTY_VALUE_MAX]; |
| struct platform_data *my_data = NULL; |
| int retry_num = 0, snd_card_num = 0, key = 0, ret = 0; |
| bool dual_mic_config = false, use_default_mixer_path = true; |
| const char *snd_card_name; |
| char *cvd_version = NULL; |
| char *snd_internal_name = NULL; |
| char *tmp = NULL; |
| char mixer_xml_file[MIXER_PATH_MAX_LENGTH]= {0}; |
| char platform_info_file[MIXER_PATH_MAX_LENGTH]= {0}; |
| struct snd_card_split *snd_split_handle = NULL; |
| my_data = calloc(1, sizeof(struct platform_data)); |
| |
| my_data->adev = adev; |
| |
| list_init(&operator_info_list); |
| |
| set_platform_defaults(my_data); |
| |
| while (snd_card_num < MAX_SND_CARD) { |
| adev->mixer = mixer_open(snd_card_num); |
| |
| while (!adev->mixer && retry_num < RETRY_NUMBER) { |
| usleep(RETRY_US); |
| adev->mixer = mixer_open(snd_card_num); |
| retry_num++; |
| } |
| |
| if (!adev->mixer) { |
| ALOGE("%s: Unable to open the mixer card: %d", __func__, |
| snd_card_num); |
| retry_num = 0; |
| snd_card_num++; |
| continue; |
| } |
| |
| snd_card_name = mixer_get_name(adev->mixer); |
| my_data->hw_info = hw_info_init(snd_card_name); |
| |
| audio_extn_set_snd_card_split(snd_card_name); |
| snd_split_handle = audio_extn_get_snd_card_split(); |
| |
| /* Get the codec internal name from the sound card and/or form factor |
| * name and form the mixer paths and platfor info file name dynamically. |
| * This is generic way of picking any codec and forma factor name based |
| * mixer and platform info files in future with no code change. |
| |
| * current code extends and looks for any of the exteneded mixer path and |
| * platform info file present based on codec and form factor. |
| |
| * order of picking appropriate file is |
| * <i> mixer_paths_<codec_name>_<form_factor>.xml, if file not present |
| * <ii> mixer_paths_<codec_name>.xml, if file not present |
| * <iii> mixer_paths.xml |
| |
| * same order is followed for audio_platform_info.xml as well |
| */ |
| |
| // need to carryforward old file name |
| if (!strncmp(snd_card_name, TOMTOM_8226_SND_CARD_NAME, |
| sizeof(TOMTOM_8226_SND_CARD_NAME))) { |
| snprintf(mixer_xml_file, sizeof(mixer_xml_file), "%s_%s.xml", |
| MIXER_XML_BASE_STRING, TOMTOM_MIXER_FILE_SUFFIX ); |
| } else { |
| |
| snprintf(mixer_xml_file, sizeof(mixer_xml_file), "%s_%s_%s.xml", |
| MIXER_XML_BASE_STRING, snd_split_handle->snd_card, |
| snd_split_handle->form_factor); |
| |
| if (F_OK != access(mixer_xml_file, 0)) { |
| memset(mixer_xml_file, 0, sizeof(mixer_xml_file)); |
| snprintf(mixer_xml_file, sizeof(mixer_xml_file), "%s_%s.xml", |
| MIXER_XML_BASE_STRING, snd_split_handle->snd_card); |
| |
| if (F_OK != access(mixer_xml_file, 0)) { |
| memset(mixer_xml_file, 0, sizeof(mixer_xml_file)); |
| strlcpy(mixer_xml_file, MIXER_XML_DEFAULT_PATH, MIXER_PATH_MAX_LENGTH); |
| } |
| } |
| |
| snprintf(platform_info_file, sizeof(platform_info_file), "%s_%s_%s.xml", |
| PLATFORM_INFO_XML_BASE_STRING, snd_split_handle->snd_card, |
| snd_split_handle->form_factor); |
| |
| if (F_OK != access(platform_info_file, 0)) { |
| memset(platform_info_file, 0, sizeof(platform_info_file)); |
| snprintf(platform_info_file, sizeof(platform_info_file), "%s_%s.xml", |
| PLATFORM_INFO_XML_BASE_STRING, snd_split_handle->snd_card); |
| |
| if (F_OK != access(platform_info_file, 0)) { |
| memset(platform_info_file, 0, sizeof(platform_info_file)); |
| strlcpy(platform_info_file, PLATFORM_INFO_XML_PATH, MIXER_PATH_MAX_LENGTH); |
| } |
| } |
| } |
| |
| /* Initialize platform specific ids and/or backends*/ |
| platform_info_init(platform_info_file, my_data); |
| |
| /* validate the sound card name |
| * my_data->snd_card_name can contain |
| * <a> complete sound card name, i.e. <device>-<codec>-<form_factor>-snd-card |
| * example: msm8994-tomtom-mtp-snd-card |
| * <b> or sub string of the card name, i.e. <device>-<codec> |
| * example: msm8994-tomtom |
| * so use length of my_data->snd_card_name for comparision |
| */ |
| |
| if (my_data->snd_card_name != NULL && |
| strncmp(snd_card_name, my_data->snd_card_name, strlen(my_data->snd_card_name)) != 0) { |
| ALOGI("%s: found valid sound card %s, but not primary sound card %s", |
| __func__, snd_card_name, my_data->snd_card_name); |
| retry_num = 0; |
| snd_card_num++; |
| hw_info_deinit(my_data->hw_info); |
| my_data->hw_info = NULL; |
| continue; |
| } |
| ALOGI("%s: found sound card %s, primary sound card expeted is %s", |
| __func__, snd_card_name, my_data->snd_card_name); |
| |
| ALOGD("%s: Loading mixer file: %s", __func__, mixer_xml_file); |
| adev->audio_route = audio_route_init(snd_card_num, mixer_xml_file); |
| |
| if (!adev->audio_route) { |
| ALOGE("%s: Failed to init audio route controls, aborting.", __func__); |
| goto init_failed; |
| } |
| adev->snd_card = snd_card_num; |
| ALOGD("%s: Opened sound card:%d", __func__, snd_card_num); |
| break; |
| } |
| |
| if (snd_card_num >= MAX_SND_CARD) { |
| ALOGE("%s: Unable to find correct sound card, aborting.", __func__); |
| goto init_failed; |
| } |
| |
| //set max volume step for voice call |
| property_get("ro.config.vc_call_vol_steps", value, TOSTRING(MAX_VOL_INDEX)); |
| my_data->max_vol_index = atoi(value); |
| |
| property_get("persist.audio.dualmic.config",value,""); |
| if (!strcmp("endfire", value)) { |
| dual_mic_config = true; |
| } |
| |
| my_data->source_mic_type = SOURCE_DUAL_MIC; |
| |
| my_data->fluence_in_spkr_mode = false; |
| my_data->fluence_in_voice_call = false; |
| my_data->fluence_in_voice_comm = false; |
| my_data->fluence_in_voice_rec = false; |
| |
| property_get("ro.qc.sdk.audio.fluencetype", value, "none"); |
| if (!strcmp("fluencepro", value)) { |
| my_data->fluence_type = FLUENCE_PRO_ENABLE; |
| } else if (!strcmp("fluence", value) || (dual_mic_config)) { |
| my_data->fluence_type = FLUENCE_ENABLE; |
| } else if (!strcmp("none", value)) { |
| my_data->fluence_type = FLUENCE_DISABLE; |
| } |
| |
| if (my_data->fluence_type != FLUENCE_DISABLE) { |
| property_get("persist.audio.fluence.voicecall",value,""); |
| if (!strcmp("true", value)) { |
| my_data->fluence_in_voice_call = true; |
| } |
| |
| property_get("persist.audio.fluence.voicecomm",value,""); |
| if (!strcmp("true", value)) { |
| my_data->fluence_in_voice_comm = true; |
| } |
| |
| property_get("persist.audio.fluence.voicerec",value,""); |
| if (!strcmp("true", value)) { |
| my_data->fluence_in_voice_rec = true; |
| } |
| |
| property_get("persist.audio.fluence.speaker",value,""); |
| if (!strcmp("true", value)) { |
| my_data->fluence_in_spkr_mode = true; |
| } |
| } |
| |
| // support max to mono, example if max count is 3, usecase supports Three, dual and mono mic |
| switch (my_data->max_mic_count) { |
| case 4: |
| my_data->source_mic_type |= SOURCE_QUAD_MIC; |
| case 3: |
| my_data->source_mic_type |= SOURCE_THREE_MIC; |
| case 2: |
| my_data->source_mic_type |= SOURCE_DUAL_MIC; |
| case 1: |
| my_data->source_mic_type |= SOURCE_MONO_MIC; |
| break; |
| default: |
| ALOGE("%s: max_mic_count (%d), is not supported, setting to default", |
| __func__, my_data->max_mic_count); |
| my_data->source_mic_type = SOURCE_MONO_MIC|SOURCE_DUAL_MIC; |
| break; |
| } |
| |
| ALOGV("%s: Fluence_Type(%d) max_mic_count(%d) mic_type(0x%x) fluence_in_voice_call(%d)" |
| " fluence_in_voice_comm(%d) fluence_in_voice_rec(%d) fluence_in_spkr_mode(%d) ", |
| __func__, my_data->fluence_type, my_data->max_mic_count, my_data->source_mic_type, |
| my_data->fluence_in_voice_call, my_data->fluence_in_voice_comm, |
| my_data->fluence_in_voice_rec, my_data->fluence_in_spkr_mode); |
| |
| my_data->acdb_handle = dlopen(LIB_ACDB_LOADER, RTLD_NOW); |
| if (my_data->acdb_handle == NULL) { |
| ALOGE("%s: DLOPEN failed for %s", __func__, LIB_ACDB_LOADER); |
| } else { |
| ALOGV("%s: DLOPEN successful for %s", __func__, LIB_ACDB_LOADER); |
| my_data->acdb_deallocate = (acdb_deallocate_t)dlsym(my_data->acdb_handle, |
| "acdb_loader_deallocate_ACDB"); |
| if (!my_data->acdb_deallocate) |
| ALOGE("%s: Could not find the symbol acdb_loader_deallocate_ACDB from %s", |
| __func__, LIB_ACDB_LOADER); |
| |
| my_data->acdb_send_audio_cal = (acdb_send_audio_cal_t)dlsym(my_data->acdb_handle, |
| "acdb_loader_send_audio_cal"); |
| if (!my_data->acdb_send_audio_cal) |
| ALOGE("%s: Could not find the symbol acdb_send_audio_cal from %s", |
| __func__, LIB_ACDB_LOADER); |
| |
| my_data->acdb_send_voice_cal = (acdb_send_voice_cal_t)dlsym(my_data->acdb_handle, |
| "acdb_loader_send_voice_cal"); |
| if (!my_data->acdb_send_voice_cal) |
| ALOGE("%s: Could not find the symbol acdb_loader_send_voice_cal from %s", |
| __func__, LIB_ACDB_LOADER); |
| |
| my_data->acdb_reload_vocvoltable = (acdb_reload_vocvoltable_t)dlsym(my_data->acdb_handle, |
| "acdb_loader_reload_vocvoltable"); |
| if (!my_data->acdb_reload_vocvoltable) |
| ALOGE("%s: Could not find the symbol acdb_loader_reload_vocvoltable from %s", |
| __func__, LIB_ACDB_LOADER); |
| |
| my_data->acdb_send_gain_dep_cal = (acdb_send_gain_dep_cal_t)dlsym(my_data->acdb_handle, |
| "acdb_loader_send_gain_dep_cal"); |
| if (!my_data->acdb_send_gain_dep_cal) |
| ALOGV("%s: Could not find the symbol acdb_loader_send_gain_dep_cal from %s", |
| __func__, LIB_ACDB_LOADER); |
| |
| #if defined (PLATFORM_MSM8994) || (PLATFORM_MSM8996) |
| acdb_init_v2_cvd_t acdb_init; |
| acdb_init = (acdb_init_v2_cvd_t)dlsym(my_data->acdb_handle, |
| "acdb_loader_init_v2"); |
| if (acdb_init == NULL) { |
| ALOGE("%s: dlsym error %s for acdb_loader_init_v2", __func__, dlerror()); |
| goto acdb_init_fail; |
| } |
| |
| cvd_version = calloc(1, MAX_CVD_VERSION_STRING_SIZE); |
| get_cvd_version(cvd_version, adev); |
| if (!cvd_version) |
| ALOGE("failed to allocate cvd_version"); |
| else |
| acdb_init((char *)snd_card_name, cvd_version, 0); |
| free(cvd_version); |
| #elif defined (PLATFORM_MSM8084) |
| acdb_init_v2_t acdb_init; |
| acdb_init = (acdb_init_v2_t)dlsym(my_data->acdb_handle, |
| "acdb_loader_init_v2"); |
| if (acdb_init == NULL) { |
| ALOGE("%s: dlsym error %s for acdb_loader_init_v2", __func__, dlerror()); |
| goto acdb_init_fail; |
| } |
| acdb_init((char *)snd_card_name); |
| #else |
| acdb_init_t acdb_init; |
| acdb_init = (acdb_init_t)dlsym(my_data->acdb_handle, |
| "acdb_loader_init_ACDB"); |
| if (acdb_init == NULL) |
| ALOGE("%s: dlsym error %s for acdb_loader_init_ACDB", __func__, dlerror()); |
| else |
| acdb_init(); |
| #endif |
| } |
| |
| acdb_init_fail: |
| |
| audio_extn_spkr_prot_init(adev); |
| |
| audio_extn_hwdep_cal_send(adev->snd_card, my_data->acdb_handle); |
| |
| /* load csd client */ |
| platform_csd_init(my_data); |
| |
| return my_data; |
| |
| init_failed: |
| if (my_data) |
| free(my_data); |
| return NULL; |
| } |
| |
| void platform_deinit(void *platform) |
| { |
| int32_t dev; |
| struct operator_info *info_item; |
| struct operator_specific_device *device_item; |
| struct listnode *node; |
| |
| struct platform_data *my_data = (struct platform_data *)platform; |
| close_csd_client(my_data->csd); |
| |
| hw_info_deinit(my_data->hw_info); |
| |
| for (dev = 0; dev < SND_DEVICE_MAX; dev++) { |
| if (backend_tag_table[dev]) |
| free(backend_tag_table[dev]); |
| if (hw_interface_table[dev]) |
| free(hw_interface_table[dev]); |
| if (operator_specific_device_table[dev]) { |
| while (!list_empty(operator_specific_device_table[dev])) { |
| node = list_head(operator_specific_device_table[dev]); |
| list_remove(node); |
| device_item = node_to_item(node, struct operator_specific_device, list); |
| free(device_item->operator); |
| free(device_item->mixer_path); |
| free(device_item); |
| } |
| free(operator_specific_device_table[dev]); |
| } |
| } |
| |
| if (my_data->snd_card_name) |
| free(my_data->snd_card_name); |
| |
| while (!list_empty(&operator_info_list)) { |
| node = list_head(&operator_info_list); |
| list_remove(node); |
| info_item = node_to_item(node, struct operator_info, list); |
| free(info_item->name); |
| free(info_item->mccmnc); |
| free(info_item); |
| } |
| |
| free(platform); |
| } |
| |
| const char *platform_get_snd_device_name(snd_device_t snd_device) |
| { |
| if (snd_device >= SND_DEVICE_MIN && snd_device < SND_DEVICE_MAX) { |
| if (operator_specific_device_table[snd_device] != NULL) { |
| return get_operator_specific_device_mixer_path(snd_device); |
| } |
| return device_table[snd_device]; |
| } else |
| return "none"; |
| } |
| |
| int platform_get_snd_device_name_extn(void *platform, snd_device_t snd_device, |
| char *device_name) |
| { |
| struct platform_data *my_data = (struct platform_data *)platform; |
| |
| if (platform == NULL || device_name == NULL) { |
| ALOGW("%s: something wrong, use legacy get_snd_device name", __func__); |
| device_name = platform_get_snd_device_name(snd_device); |
| } else if (snd_device >= SND_DEVICE_MIN && snd_device < SND_DEVICE_MAX) { |
| if (operator_specific_device_table[snd_device] != NULL) { |
| strlcpy(device_name, get_operator_specific_device_mixer_path(snd_device), |
| DEVICE_NAME_MAX_SIZE); |
| } else { |
| strlcpy(device_name, device_table[snd_device], DEVICE_NAME_MAX_SIZE); |
| } |
| hw_info_append_hw_type(my_data->hw_info, snd_device, device_name); |
| } else { |
| strlcpy(device_name, "none", DEVICE_NAME_MAX_SIZE); |
| } |
| |
| return 0; |
| } |
| |
| void platform_add_backend_name(void *platform, char *mixer_path, |
| snd_device_t snd_device) |
| { |
| struct platform_data *my_data = (struct platform_data *)platform; |
| |
| if ((snd_device < SND_DEVICE_MIN) || (snd_device >= SND_DEVICE_MAX)) { |
| ALOGE("%s: Invalid snd_device = %d", __func__, snd_device); |
| return; |
| } |
| |
| const char * suffix = backend_tag_table[snd_device]; |
| |
| if (suffix != NULL) { |
| strcat(mixer_path, " "); |
| strcat(mixer_path, suffix); |
| } |
| } |
| |
| bool platform_check_backends_match(snd_device_t snd_device1, snd_device_t snd_device2) |
| { |
| bool result = true; |
| |
| ALOGV("%s: snd_device1 = %s, snd_device2 = %s", __func__, |
| platform_get_snd_device_name(snd_device1), |
| platform_get_snd_device_name(snd_device2)); |
| |
| if ((snd_device1 < SND_DEVICE_MIN) || (snd_device1 >= SND_DEVICE_MAX)) { |
| ALOGE("%s: Invalid snd_device = %s", __func__, |
| platform_get_snd_device_name(snd_device1)); |
| return false; |
| } |
| if ((snd_device2 < SND_DEVICE_MIN) || (snd_device2 >= SND_DEVICE_MAX)) { |
| ALOGE("%s: Invalid snd_device = %s", __func__, |
| platform_get_snd_device_name(snd_device2)); |
| return false; |
| } |
| const char * be_itf1 = hw_interface_table[snd_device1]; |
| const char * be_itf2 = hw_interface_table[snd_device2]; |
| |
| if (NULL != be_itf1 && NULL != be_itf2) { |
| if ((NULL == strstr(be_itf2, be_itf1)) && (NULL == strstr(be_itf1, be_itf2))) |
| result = false; |
| } |
| |
| ALOGV("%s: be_itf1 = %s, be_itf2 = %s, match %d", __func__, be_itf1, be_itf2, result); |
| return result; |
| } |
| |
| int platform_get_pcm_device_id(audio_usecase_t usecase, int device_type) |
| { |
| int device_id; |
| if (device_type == PCM_PLAYBACK) |
| device_id = pcm_device_table[usecase][0]; |
| else |
| device_id = pcm_device_table[usecase][1]; |
| return device_id; |
| } |
| |
| static int find_index(const struct name_to_index * table, int32_t len, |
| const char * name) |
| { |
| int ret = 0; |
| int32_t i; |
| |
| if (table == NULL) { |
| ALOGE("%s: table is NULL", __func__); |
| ret = -ENODEV; |
| goto done; |
| } |
| |
| if (name == NULL) { |
| ALOGE("null key"); |
| ret = -ENODEV; |
| goto done; |
| } |
| |
| for (i=0; i < len; i++) { |
| if (!strcmp(table[i].name, name)) { |
| ret = table[i].index; |
| goto done; |
| } |
| } |
| ALOGE("%s: Could not find index for name = %s", |
| __func__, name); |
| ret = -ENODEV; |
| done: |
| return ret; |
| } |
| |
| int platform_get_snd_device_index(char *device_name) |
| { |
| return find_index(snd_device_name_index, SND_DEVICE_MAX, device_name); |
| } |
| |
| int platform_get_usecase_index(const char *usecase_name) |
| { |
| return find_index(usecase_name_index, AUDIO_USECASE_MAX, usecase_name); |
| } |
| |
| void platform_add_operator_specific_device(snd_device_t snd_device, |
| const char *operator, |
| const char *mixer_path, |
| unsigned int acdb_id) |
| { |
| struct operator_specific_device *device; |
| |
| if (operator_specific_device_table[snd_device] == NULL) { |
| operator_specific_device_table[snd_device] = |
| (struct listnode *)calloc(1, sizeof(struct listnode)); |
| list_init(operator_specific_device_table[snd_device]); |
| } |
| |
| device = (struct operator_specific_device *)calloc(1, sizeof(struct operator_specific_device)); |
| |
| device->operator = strdup(operator); |
| device->mixer_path = strdup(mixer_path); |
| device->acdb_id = acdb_id; |
| |
| list_add_tail(operator_specific_device_table[snd_device], &device->list); |
| |
| ALOGD("%s: device[%s] -> operator[%s] mixer_path[%s] acdb_id[%d]", __func__, |
| platform_get_snd_device_name(snd_device), operator, mixer_path, acdb_id); |
| |
| } |
| |
| int platform_set_snd_device_acdb_id(snd_device_t snd_device, unsigned int acdb_id) |
| { |
| int ret = 0; |
| |
| if ((snd_device < SND_DEVICE_MIN) || (snd_device >= SND_DEVICE_MAX)) { |
| ALOGE("%s: Invalid snd_device = %d", |
| __func__, snd_device); |
| ret = -EINVAL; |
| goto done; |
| } |
| |
| ALOGV("%s: acdb_device_table[%s]: old = %d new = %d", __func__, |
| platform_get_snd_device_name(snd_device), acdb_device_table[snd_device], acdb_id); |
| acdb_device_table[snd_device] = acdb_id; |
| done: |
| return ret; |
| } |
| |
| int platform_get_snd_device_acdb_id(snd_device_t snd_device) |
| { |
| if ((snd_device < SND_DEVICE_MIN) || (snd_device >= SND_DEVICE_MAX)) { |
| ALOGE("%s: Invalid snd_device = %d", __func__, snd_device); |
| return -EINVAL; |
| } |
| |
| if (operator_specific_device_table[snd_device] != NULL) |
| return get_operator_specific_device_acdb_id(snd_device); |
| else |
| return acdb_device_table[snd_device]; |
| } |
| |
| int platform_send_audio_calibration(void *platform, snd_device_t snd_device) |
| { |
| struct platform_data *my_data = (struct platform_data *)platform; |
| int acdb_dev_id, acdb_dev_type; |
| |
| acdb_dev_id = acdb_device_table[audio_extn_get_spkr_prot_snd_device(snd_device)]; |
| if (acdb_dev_id < 0) { |
| ALOGE("%s: Could not find acdb id for device(%d)", |
| __func__, snd_device); |
| return -EINVAL; |
| } |
| if (my_data->acdb_send_audio_cal) { |
| ALOGV("%s: sending audio calibration for snd_device(%d) acdb_id(%d)", |
| __func__, snd_device, acdb_dev_id); |
| if (snd_device >= SND_DEVICE_OUT_BEGIN && |
| snd_device < SND_DEVICE_OUT_END) |
| acdb_dev_type = ACDB_DEV_TYPE_OUT; |
| else |
| acdb_dev_type = ACDB_DEV_TYPE_IN; |
| my_data->acdb_send_audio_cal(acdb_dev_id, acdb_dev_type); |
| } |
| return 0; |
| } |
| |
| int platform_switch_voice_call_device_pre(void *platform) |
| { |
| struct platform_data *my_data = (struct platform_data *)platform; |
| int ret = 0; |
| |
| if (my_data->csd != NULL && |
| voice_is_in_call(my_data->adev)) { |
| /* This must be called before disabling mixer controls on APQ side */ |
| ret = my_data->csd->disable_device(); |
| if (ret < 0) { |
| ALOGE("%s: csd_client_disable_device, failed, error %d", |
| __func__, ret); |
| } |
| } |
| return ret; |
| } |
| |
| int platform_switch_voice_call_enable_device_config(void *platform, |
| snd_device_t out_snd_device, |
| snd_device_t in_snd_device) |
| { |
| struct platform_data *my_data = (struct platform_data *)platform; |
| int acdb_rx_id, acdb_tx_id; |
| int ret = 0; |
| |
| if (my_data->csd == NULL) |
| return ret; |
| |
| if (out_snd_device == SND_DEVICE_OUT_VOICE_SPEAKER && |
| audio_extn_spkr_prot_is_enabled()) |
| acdb_rx_id = platform_get_snd_device_acdb_id(SND_DEVICE_OUT_SPEAKER_PROTECTED); |
| else |
| acdb_rx_id = platform_get_snd_device_acdb_id(out_snd_device); |
| |
| acdb_tx_id = platform_get_snd_device_acdb_id(in_snd_device); |
| |
| if (acdb_rx_id > 0 && acdb_tx_id > 0) { |
| ret = my_data->csd->enable_device_config(acdb_rx_id, acdb_tx_id); |
| if (ret < 0) { |
| ALOGE("%s: csd_enable_device_config, failed, error %d", |
| __func__, ret); |
| } |
| } else { |
| ALOGE("%s: Incorrect ACDB IDs (rx: %d tx: %d)", __func__, |
| acdb_rx_id, acdb_tx_id); |
| } |
| |
| return ret; |
| } |
| |
| int platform_switch_voice_call_device_post(void *platform, |
| snd_device_t out_snd_device, |
| snd_device_t in_snd_device) |
| { |
| struct platform_data *my_data = (struct platform_data *)platform; |
| int acdb_rx_id, acdb_tx_id; |
| |
| if (my_data->acdb_send_voice_cal == NULL) { |
| ALOGE("%s: dlsym error for acdb_send_voice_call", __func__); |
| } else { |
| if (out_snd_device == SND_DEVICE_OUT_VOICE_SPEAKER && |
| audio_extn_spkr_prot_is_enabled()) |
| out_snd_device = SND_DEVICE_OUT_VOICE_SPEAKER_PROTECTED; |
| |
| acdb_rx_id = platform_get_snd_device_acdb_id(out_snd_device); |
| acdb_tx_id = platform_get_snd_device_acdb_id(in_snd_device); |
| |
| if (acdb_rx_id > 0 && acdb_tx_id > 0) |
| my_data->acdb_send_voice_cal(acdb_rx_id, acdb_tx_id); |
| else |
| ALOGE("%s: Incorrect ACDB IDs (rx: %d tx: %d)", __func__, |
| acdb_rx_id, acdb_tx_id); |
| } |
| |
| return 0; |
| } |
| |
| int platform_switch_voice_call_usecase_route_post(void *platform, |
| snd_device_t out_snd_device, |
| snd_device_t in_snd_device) |
| { |
| struct platform_data *my_data = (struct platform_data *)platform; |
| int acdb_rx_id, acdb_tx_id; |
| int ret = 0; |
| |
| if (my_data->csd == NULL) |
| return ret; |
| |
| if (out_snd_device == SND_DEVICE_OUT_VOICE_SPEAKER && |
| audio_extn_spkr_prot_is_enabled()) |
| acdb_rx_id = platform_get_snd_device_acdb_id(SND_DEVICE_OUT_VOICE_SPEAKER_PROTECTED); |
| else |
| acdb_rx_id = platform_get_snd_device_acdb_id(out_snd_device); |
| |
| acdb_tx_id = platform_get_snd_device_acdb_id(in_snd_device); |
| |
| if (acdb_rx_id > 0 && acdb_tx_id > 0) { |
| ret = my_data->csd->enable_device(acdb_rx_id, acdb_tx_id, |
| my_data->adev->acdb_settings); |
| if (ret < 0) { |
| ALOGE("%s: csd_enable_device, failed, error %d", __func__, ret); |
| } |
| } else { |
| ALOGE("%s: Incorrect ACDB IDs (rx: %d tx: %d)", __func__, |
| acdb_rx_id, acdb_tx_id); |
| } |
| |
| return ret; |
| } |
| |
| int platform_start_voice_call(void *platform, uint32_t vsid) |
| { |
| struct platform_data *my_data = (struct platform_data *)platform; |
| int ret = 0; |
| |
| if (my_data->csd != NULL) { |
| ret = my_data->csd->start_voice(vsid); |
| if (ret < 0) { |
| ALOGE("%s: csd_start_voice error %d\n", __func__, ret); |
| } |
| } |
| return ret; |
| } |
| |
| int platform_stop_voice_call(void *platform, uint32_t vsid) |
| { |
| struct platform_data *my_data = (struct platform_data *)platform; |
| int ret = 0; |
| |
| if (my_data->csd != NULL) { |
| ret = my_data->csd->stop_voice(vsid); |
| if (ret < 0) { |
| ALOGE("%s: csd_stop_voice error %d\n", __func__, ret); |
| } |
| } |
| return ret; |
| } |
| |
| int platform_get_sample_rate(void *platform, uint32_t *rate) |
| { |
| struct platform_data *my_data = (struct platform_data *)platform; |
| int ret = 0; |
| |
| if (my_data->csd != NULL) { |
| ret = my_data->csd->get_sample_rate(rate); |
| if (ret < 0) { |
| ALOGE("%s: csd_get_sample_rate error %d\n", __func__, ret); |
| } |
| } |
| return ret; |
| } |
| |
| void platform_set_speaker_gain_in_combo(struct audio_device *adev, |
| snd_device_t snd_device, |
| bool enable) |
| { |
| const char* name; |
| switch (snd_device) { |
| case SND_DEVICE_OUT_SPEAKER_AND_HEADPHONES: |
| if (enable) |
| name = "spkr-gain-in-headphone-combo"; |
| else |
| name = "speaker-gain-default"; |
| break; |
| case SND_DEVICE_OUT_SPEAKER_AND_LINE: |
| if (enable) |
| name = "spkr-gain-in-line-combo"; |
| else |
| name = "speaker-gain-default"; |
| break; |
| case SND_DEVICE_OUT_SPEAKER_SAFE_AND_HEADPHONES: |
| if (enable) |
| name = "spkr-safe-gain-in-headphone-combo"; |
| else |
| name = "speaker-safe-gain-default"; |
| break; |
| case SND_DEVICE_OUT_SPEAKER_SAFE_AND_LINE: |
| if (enable) |
| name = "spkr-safe-gain-in-line-combo"; |
| else |
| name = "speaker-safe-gain-default"; |
| break; |
| default: |
| return; |
| } |
| |
| audio_route_apply_and_update_path(adev->audio_route, name); |
| } |
| |
| int platform_set_voice_volume(void *platform, int volume) |
| { |
| struct platform_data *my_data = (struct platform_data *)platform; |
| struct audio_device *adev = my_data->adev; |
| struct mixer_ctl *ctl; |
| const char *mixer_ctl_name = "Voice Rx Gain"; |
| int vol_index = 0, ret = 0; |
| uint32_t set_values[ ] = {0, |
| ALL_SESSION_VSID, |
| DEFAULT_VOLUME_RAMP_DURATION_MS}; |
| |
| // Voice volume levels are mapped to adsp volume levels as follows. |
| // 100 -> 5, 80 -> 4, 60 -> 3, 40 -> 2, 20 -> 1 0 -> 0 |
| // But this values don't changed in kernel. So, below change is need. |
| vol_index = (int)percent_to_index(volume, MIN_VOL_INDEX, my_data->max_vol_index); |
| set_values[0] = vol_index; |
| |
| ctl = mixer_get_ctl_by_name(adev->mixer, mixer_ctl_name); |
| if (!ctl) { |
| ALOGE("%s: Could not get ctl for mixer cmd - %s", |
| __func__, mixer_ctl_name); |
| return -EINVAL; |
| } |
| ALOGV("Setting voice volume index: %d", set_values[0]); |
| mixer_ctl_set_array(ctl, set_values, ARRAY_SIZE(set_values)); |
| |
| if (my_data->csd != NULL) { |
| ret = my_data->csd->volume(ALL_SESSION_VSID, volume, |
| DEFAULT_VOLUME_RAMP_DURATION_MS); |
| if (ret < 0) { |
| ALOGE("%s: csd_volume error %d", __func__, ret); |
| } |
| } |
| return ret; |
| } |
| |
| int platform_set_mic_mute(void *platform, bool state) |
| { |
| struct platform_data *my_data = (struct platform_data *)platform; |
| struct audio_device *adev = my_data->adev; |
| struct mixer_ctl *ctl; |
| const char *mixer_ctl_name = "Voice Tx Mute"; |
| int ret = 0; |
| uint32_t set_values[ ] = {0, |
| ALL_SESSION_VSID, |
| DEFAULT_MUTE_RAMP_DURATION_MS}; |
| |
| if (adev->mode != AUDIO_MODE_IN_CALL && |
| adev->mode != AUDIO_MODE_IN_COMMUNICATION) |
| return 0; |
| |
| if (adev->enable_hfp) |
| mixer_ctl_name = "HFP Tx Mute"; |
| |
| set_values[0] = state; |
| ctl = mixer_get_ctl_by_name(adev->mixer, mixer_ctl_name); |
| if (!ctl) { |
| ALOGE("%s: Could not get ctl for mixer cmd - %s", |
| __func__, mixer_ctl_name); |
| return -EINVAL; |
| } |
| ALOGV("Setting voice mute state: %d", state); |
| mixer_ctl_set_array(ctl, set_values, ARRAY_SIZE(set_values)); |
| |
| if (my_data->csd != NULL) { |
| ret = my_data->csd->mic_mute(ALL_SESSION_VSID, state, |
| DEFAULT_MUTE_RAMP_DURATION_MS); |
| if (ret < 0) { |
| ALOGE("%s: csd_mic_mute error %d", __func__, ret); |
| } |
| } |
| return ret; |
| } |
| |
| int platform_set_device_mute(void *platform, bool state, char *dir) |
| { |
| struct platform_data *my_data = (struct platform_data *)platform; |
| struct audio_device *adev = my_data->adev; |
| struct mixer_ctl *ctl; |
| char *mixer_ctl_name = NULL; |
| int ret = 0; |
| uint32_t set_values[ ] = {0, |
| ALL_SESSION_VSID, |
| 0}; |
| if(dir == NULL) { |
| ALOGE("%s: Invalid direction:%s", __func__, dir); |
| return -EINVAL; |
| } |
| |
| if (!strncmp("rx", dir, sizeof("rx"))) { |
| mixer_ctl_name = "Voice Rx Device Mute"; |
| } else if (!strncmp("tx", dir, sizeof("tx"))) { |
| mixer_ctl_name = "Voice Tx Device Mute"; |
| } else { |
| return -EINVAL; |
| } |
| |
| set_values[0] = state; |
| ctl = mixer_get_ctl_by_name(adev->mixer, mixer_ctl_name); |
| if (!ctl) { |
| ALOGE("%s: Could not get ctl for mixer cmd - %s", |
| __func__, mixer_ctl_name); |
| return -EINVAL; |
| } |
| |
| ALOGV("%s: Setting device mute state: %d, mixer ctrl:%s", |
| __func__,state, mixer_ctl_name); |
| mixer_ctl_set_array(ctl, set_values, ARRAY_SIZE(set_values)); |
| |
| return ret; |
| } |
| |
| bool platform_can_split_snd_device(snd_device_t snd_device, |
| int *num_devices, |
| snd_device_t *new_snd_devices) |
| { |
| bool status = false; |
| |
| if (NULL == num_devices || NULL == new_snd_devices) { |
| ALOGE("%s: NULL pointer ..", __func__); |
| return false; |
| } |
| |
| /* |
| * If wired headset/headphones/line devices share the same backend |
| * with speaker/earpiece this routine returns false. |
| */ |
| if (snd_device == SND_DEVICE_OUT_SPEAKER_AND_HEADPHONES && |
| !platform_check_backends_match(SND_DEVICE_OUT_SPEAKER, SND_DEVICE_OUT_HEADPHONES)) { |
| *num_devices = 2; |
| new_snd_devices[0] = SND_DEVICE_OUT_SPEAKER; |
| new_snd_devices[1] = SND_DEVICE_OUT_HEADPHONES; |
| status = true; |
| } else if (snd_device == SND_DEVICE_OUT_SPEAKER_AND_LINE && |
| !platform_check_backends_match(SND_DEVICE_OUT_SPEAKER, SND_DEVICE_OUT_LINE)) { |
| *num_devices = 2; |
| new_snd_devices[0] = SND_DEVICE_OUT_SPEAKER; |
| new_snd_devices[1] = SND_DEVICE_OUT_LINE; |
| status = true; |
| } else if (snd_device == SND_DEVICE_OUT_SPEAKER_SAFE_AND_HEADPHONES && |
| !platform_check_backends_match(SND_DEVICE_OUT_SPEAKER_SAFE, SND_DEVICE_OUT_HEADPHONES)) { |
| *num_devices = 2; |
| new_snd_devices[0] = SND_DEVICE_OUT_SPEAKER_SAFE; |
| new_snd_devices[1] = SND_DEVICE_OUT_HEADPHONES; |
| status = true; |
| } else if (snd_device == SND_DEVICE_OUT_SPEAKER_SAFE_AND_LINE && |
| !platform_check_backends_match(SND_DEVICE_OUT_SPEAKER_SAFE, SND_DEVICE_OUT_LINE)) { |
| *num_devices = 2; |
| new_snd_devices[0] = SND_DEVICE_OUT_SPEAKER_SAFE; |
| new_snd_devices[1] = SND_DEVICE_OUT_LINE; |
| status = true; |
| } |
| return status; |
| } |
| |
| snd_device_t platform_get_output_snd_device(void *platform, audio_devices_t devices) |
| { |
| struct platform_data *my_data = (struct platform_data *)platform; |
| struct audio_device *adev = my_data->adev; |
| audio_mode_t mode = adev->mode; |
| snd_device_t snd_device = SND_DEVICE_NONE; |
| |
| ALOGV("%s: enter: output devices(%#x)", __func__, devices); |
| if (devices == AUDIO_DEVICE_NONE || |
| devices & AUDIO_DEVICE_BIT_IN) { |
| ALOGV("%s: Invalid output devices (%#x)", __func__, devices); |
| goto exit; |
| } |
| |
| if (popcount(devices) == 2) { |
| if (devices == (AUDIO_DEVICE_OUT_WIRED_HEADPHONE | |
| AUDIO_DEVICE_OUT_SPEAKER) || |
| devices == (AUDIO_DEVICE_OUT_WIRED_HEADSET | |
| AUDIO_DEVICE_OUT_SPEAKER)) { |
| snd_device = SND_DEVICE_OUT_SPEAKER_AND_HEADPHONES; |
| } else if (devices == (AUDIO_DEVICE_OUT_LINE | |
| AUDIO_DEVICE_OUT_SPEAKER)) { |
| snd_device = SND_DEVICE_OUT_SPEAKER_AND_LINE; |
| } else if (devices == (AUDIO_DEVICE_OUT_WIRED_HEADPHONE | |
| AUDIO_DEVICE_OUT_SPEAKER_SAFE) || |
| devices == (AUDIO_DEVICE_OUT_WIRED_HEADSET | |
| AUDIO_DEVICE_OUT_SPEAKER_SAFE)) { |
| snd_device = SND_DEVICE_OUT_SPEAKER_SAFE_AND_HEADPHONES; |
| } else if (devices == (AUDIO_DEVICE_OUT_LINE | |
| AUDIO_DEVICE_OUT_SPEAKER_SAFE)) { |
| snd_device = SND_DEVICE_OUT_SPEAKER_SAFE_AND_LINE; |
| } else if (devices == (AUDIO_DEVICE_OUT_AUX_DIGITAL | |
| AUDIO_DEVICE_OUT_SPEAKER)) { |
| snd_device = SND_DEVICE_OUT_SPEAKER_AND_HDMI; |
| } else { |
| ALOGE("%s: Invalid combo device(%#x)", __func__, devices); |
| goto exit; |
| } |
| if (snd_device != SND_DEVICE_NONE) { |
| goto exit; |
| } |
| } |
| |
| if (popcount(devices) != 1) { |
| ALOGE("%s: Invalid output devices(%#x)", __func__, devices); |
| goto exit; |
| } |
| |
| if (voice_is_in_call(adev) || adev->enable_voicerx) { |
| if (devices & AUDIO_DEVICE_OUT_WIRED_HEADPHONE || |
| devices & AUDIO_DEVICE_OUT_WIRED_HEADSET || |
| devices & AUDIO_DEVICE_OUT_LINE) { |
| if (voice_is_in_call(adev) && |
| (adev->voice.tty_mode == TTY_MODE_FULL)) |
| snd_device = SND_DEVICE_OUT_VOICE_TTY_FULL_HEADPHONES; |
| else if (voice_is_in_call(adev) && |
| (adev->voice.tty_mode == TTY_MODE_VCO)) |
| snd_device = SND_DEVICE_OUT_VOICE_TTY_VCO_HEADPHONES; |
| else if (voice_is_in_call(adev) && |
| (adev->voice.tty_mode == TTY_MODE_HCO)) |
| snd_device = SND_DEVICE_OUT_VOICE_TTY_HCO_HANDSET; |
| else { |
| if (devices & AUDIO_DEVICE_OUT_LINE) |
| snd_device = SND_DEVICE_OUT_VOICE_LINE; |
| else |
| snd_device = SND_DEVICE_OUT_VOICE_HEADPHONES; |
| } |
| } else if (devices & AUDIO_DEVICE_OUT_ALL_SCO) { |
| if (adev->bt_wb_speech_enabled) { |
| snd_device = SND_DEVICE_OUT_BT_SCO_WB; |
| } else { |
| snd_device = SND_DEVICE_OUT_BT_SCO; |
| } |
| } else if (devices & (AUDIO_DEVICE_OUT_SPEAKER | AUDIO_DEVICE_OUT_SPEAKER_SAFE)) { |
| if (!adev->enable_hfp) { |
| snd_device = SND_DEVICE_OUT_VOICE_SPEAKER; |
| } else { |
| snd_device = SND_DEVICE_OUT_VOICE_SPEAKER_HFP; |
| } |
| } else if (devices & AUDIO_DEVICE_OUT_EARPIECE) { |
| if(adev->voice.hac) |
| snd_device = SND_DEVICE_OUT_VOICE_HAC_HANDSET; |
| else if (is_operator_tmus()) |
| snd_device = SND_DEVICE_OUT_VOICE_HANDSET_TMUS; |
| else |
| snd_device = SND_DEVICE_OUT_VOICE_HANDSET; |
| } else if (devices & AUDIO_DEVICE_OUT_TELEPHONY_TX) |
| snd_device = SND_DEVICE_OUT_VOICE_TX; |
| |
| if (snd_device != SND_DEVICE_NONE) { |
| goto exit; |
| } |
| } |
| |
| if (devices & AUDIO_DEVICE_OUT_WIRED_HEADPHONE || |
| devices & AUDIO_DEVICE_OUT_WIRED_HEADSET) { |
| snd_device = SND_DEVICE_OUT_HEADPHONES; |
| } else if (devices & AUDIO_DEVICE_OUT_LINE) { |
| snd_device = SND_DEVICE_OUT_LINE; |
| } else if (devices & AUDIO_DEVICE_OUT_SPEAKER_SAFE) { |
| snd_device = SND_DEVICE_OUT_SPEAKER_SAFE; |
| } else if (devices & AUDIO_DEVICE_OUT_SPEAKER) { |
| if (my_data->speaker_lr_swap) |
| snd_device = SND_DEVICE_OUT_SPEAKER_REVERSE; |
| else |
| snd_device = SND_DEVICE_OUT_SPEAKER; |
| } else if (devices & AUDIO_DEVICE_OUT_ALL_SCO) { |
| if (adev->bt_wb_speech_enabled) { |
| snd_device = SND_DEVICE_OUT_BT_SCO_WB; |
| } else { |
| snd_device = SND_DEVICE_OUT_BT_SCO; |
| } |
| } else if (devices & AUDIO_DEVICE_OUT_AUX_DIGITAL) { |
| snd_device = SND_DEVICE_OUT_HDMI ; |
| } else if (devices & AUDIO_DEVICE_OUT_EARPIECE) { |
| /*HAC support for voice-ish audio (eg visual voicemail)*/ |
| if(adev->voice.hac) |
| snd_device = SND_DEVICE_OUT_VOICE_HAC_HANDSET; |
| else |
| snd_device = SND_DEVICE_OUT_HANDSET; |
| } else { |
| ALOGE("%s: Unknown device(s) %#x", __func__, devices); |
| } |
| exit: |
| ALOGV("%s: exit: snd_device(%s)", __func__, device_table[snd_device]); |
| return snd_device; |
| } |
| |
| snd_device_t platform_get_input_snd_device(void *platform, audio_devices_t out_device) |
| { |
| struct platform_data *my_data = (struct platform_data *)platform; |
| struct audio_device *adev = my_data->adev; |
| audio_source_t source = (adev->active_input == NULL) ? |
| AUDIO_SOURCE_DEFAULT : adev->active_input->source; |
| |
| audio_mode_t mode = adev->mode; |
| audio_devices_t in_device = ((adev->active_input == NULL) ? |
| AUDIO_DEVICE_NONE : adev->active_input->device) |
| & ~AUDIO_DEVICE_BIT_IN; |
| audio_channel_mask_t channel_mask = (adev->active_input == NULL) ? |
| AUDIO_CHANNEL_IN_MONO : adev->active_input->channel_mask; |
| snd_device_t snd_device = SND_DEVICE_NONE; |
| int channel_count = popcount(channel_mask); |
| |
| ALOGV("%s: enter: out_device(%#x) in_device(%#x) channel_count (%d) channel_mask (0x%x)", |
| __func__, out_device, in_device, channel_count, channel_mask); |
| if ((out_device != AUDIO_DEVICE_NONE) && voice_is_in_call(adev)) { |
| if (adev->voice.tty_mode != TTY_MODE_OFF) { |
| if (out_device & AUDIO_DEVICE_OUT_WIRED_HEADPHONE || |
| out_device & AUDIO_DEVICE_OUT_WIRED_HEADSET || |
| out_device & AUDIO_DEVICE_OUT_LINE) { |
| switch (adev->voice.tty_mode) { |
| case TTY_MODE_FULL: |
| snd_device = SND_DEVICE_IN_VOICE_TTY_FULL_HEADSET_MIC; |
| break; |
| case TTY_MODE_VCO: |
| snd_device = SND_DEVICE_IN_VOICE_TTY_VCO_HANDSET_MIC; |
| break; |
| case TTY_MODE_HCO: |
| snd_device = SND_DEVICE_IN_VOICE_TTY_HCO_HEADSET_MIC; |
| break; |
| default: |
| ALOGE("%s: Invalid TTY mode (%#x)", __func__, adev->voice.tty_mode); |
| } |
| goto exit; |
| } |
| } |
| if (out_device & AUDIO_DEVICE_OUT_EARPIECE) { |
| if (my_data->fluence_in_voice_call == false) { |
| snd_device = SND_DEVICE_IN_HANDSET_MIC; |
| } else { |
| if (is_operator_tmus()) |
| snd_device = SND_DEVICE_IN_VOICE_DMIC_TMUS; |
| else |
| snd_device = SND_DEVICE_IN_VOICE_DMIC; |
| } |
| } else if (out_device & AUDIO_DEVICE_OUT_WIRED_HEADSET) { |
| snd_device = SND_DEVICE_IN_VOICE_HEADSET_MIC; |
| } else if (out_device & AUDIO_DEVICE_OUT_ALL_SCO) { |
| if (adev->bt_wb_speech_enabled) { |
| if (adev->bluetooth_nrec) |
| snd_device = SND_DEVICE_IN_BT_SCO_MIC_WB_NREC; |
| else |
| snd_device = SND_DEVICE_IN_BT_SCO_MIC_WB; |
| } else { |
| if (adev->bluetooth_nrec) |
| snd_device = SND_DEVICE_IN_BT_SCO_MIC_NREC; |
| else |
| snd_device = SND_DEVICE_IN_BT_SCO_MIC; |
| } |
| } else if (out_device & AUDIO_DEVICE_OUT_SPEAKER || |
| out_device & AUDIO_DEVICE_OUT_SPEAKER_SAFE || |
| out_device & AUDIO_DEVICE_OUT_WIRED_HEADPHONE || |
| out_device & AUDIO_DEVICE_OUT_LINE) { |
| if (my_data->fluence_in_voice_call && my_data->fluence_in_spkr_mode) { |
| if (my_data->source_mic_type & SOURCE_DUAL_MIC) { |
| snd_device = SND_DEVICE_IN_VOICE_SPEAKER_DMIC; |
| } else { |
| snd_device = SND_DEVICE_IN_VOICE_SPEAKER_MIC; |
| } |
| } |
| |
| //select default |
| if (snd_device == SND_DEVICE_NONE) { |
| if (!adev->enable_hfp) { |
| snd_device = SND_DEVICE_IN_VOICE_SPEAKER_MIC; |
| } else { |
| snd_device = SND_DEVICE_IN_VOICE_SPEAKER_MIC_HFP; |
| platform_set_echo_reference(adev, true, out_device); |
| } |
| } |
| } else if (out_device & AUDIO_DEVICE_OUT_TELEPHONY_TX) { |
| snd_device = SND_DEVICE_IN_VOICE_RX; |
| } |
| } else if (source == AUDIO_SOURCE_CAMCORDER) { |
| if (in_device & AUDIO_DEVICE_IN_BUILTIN_MIC || |
| in_device & AUDIO_DEVICE_IN_BACK_MIC) { |
| snd_device = SND_DEVICE_IN_CAMCORDER_MIC; |
| } |
| } else if (source == AUDIO_SOURCE_VOICE_RECOGNITION) { |
| if (in_device & AUDIO_DEVICE_IN_BUILTIN_MIC) { |
| if (my_data->fluence_in_voice_rec && channel_count == 1) { |
| if ((my_data->fluence_type == FLUENCE_PRO_ENABLE) && |
| (my_data->source_mic_type & SOURCE_QUAD_MIC)) { |
| snd_device = SND_DEVICE_IN_HANDSET_QMIC; |
| } else if ((my_data->fluence_type == FLUENCE_PRO_ENABLE) && |
| (my_data->source_mic_type & SOURCE_THREE_MIC)) { |
| snd_device = SND_DEVICE_IN_HANDSET_TMIC; |
| } else if (((my_data->fluence_type == FLUENCE_PRO_ENABLE) || |
| (my_data->fluence_type == FLUENCE_ENABLE)) && |
| (my_data->source_mic_type & SOURCE_DUAL_MIC)) { |
| snd_device = SND_DEVICE_IN_VOICE_REC_DMIC_FLUENCE; |
| } |
| platform_set_echo_reference(adev, true, out_device); |
| } else if ((channel_mask == AUDIO_CHANNEL_IN_FRONT_BACK) && |
| (my_data->source_mic_type & SOURCE_DUAL_MIC)) { |
| snd_device = SND_DEVICE_IN_VOICE_REC_DMIC_STEREO; |
| } else if (((int)channel_mask == AUDIO_CHANNEL_INDEX_MASK_3) && |
| (my_data->source_mic_type & SOURCE_THREE_MIC)) { |
| snd_device = SND_DEVICE_IN_THREE_MIC; |
| } else if (((int)channel_mask == AUDIO_CHANNEL_INDEX_MASK_4) && |
| (my_data->source_mic_type & SOURCE_QUAD_MIC)) { |
| snd_device = SND_DEVICE_IN_QUAD_MIC; |
| } |
| if (snd_device == SND_DEVICE_NONE) { |
| if (adev->active_input->enable_ns) |
| snd_device = SND_DEVICE_IN_VOICE_REC_MIC_NS; |
| else |
| snd_device = SND_DEVICE_IN_VOICE_REC_MIC; |
| } |
| } else if (in_device & AUDIO_DEVICE_IN_WIRED_HEADSET) { |
| snd_device = SND_DEVICE_IN_VOICE_REC_HEADSET_MIC; |
| } |
| } else if (source == AUDIO_SOURCE_UNPROCESSED) { |
| if (in_device & AUDIO_DEVICE_IN_BUILTIN_MIC) { |
| snd_device = SND_DEVICE_IN_UNPROCESSED_MIC; |
| } else if (in_device & AUDIO_DEVICE_IN_WIRED_HEADSET) { |
| snd_device = SND_DEVICE_IN_UNPROCESSED_HEADSET_MIC; |
| } |
| } else if (source == AUDIO_SOURCE_VOICE_COMMUNICATION || |
| mode == AUDIO_MODE_IN_COMMUNICATION) { |
| if (out_device & (AUDIO_DEVICE_OUT_SPEAKER | AUDIO_DEVICE_OUT_SPEAKER_SAFE)) |
| in_device = AUDIO_DEVICE_IN_BACK_MIC; |
| if (adev->active_input) { |
| if (adev->active_input->enable_aec && |
| adev->active_input->enable_ns) { |
| if (in_device & AUDIO_DEVICE_IN_BACK_MIC) { |
| if (my_data->fluence_in_spkr_mode && |
| my_data->fluence_in_voice_comm && |
| (my_data->source_mic_type & SOURCE_DUAL_MIC)) { |
| snd_device = SND_DEVICE_IN_SPEAKER_DMIC_AEC_NS; |
| } else { |
| snd_device = SND_DEVICE_IN_SPEAKER_MIC_AEC_NS; |
| } |
| } else if (in_device & AUDIO_DEVICE_IN_BUILTIN_MIC) { |
| if (my_data->fluence_in_voice_comm && |
| (my_data->source_mic_type & SOURCE_DUAL_MIC)) { |
| snd_device = SND_DEVICE_IN_HANDSET_DMIC_AEC_NS; |
| } else { |
| snd_device = SND_DEVICE_IN_HANDSET_MIC_AEC_NS; |
| } |
| } else if (in_device & AUDIO_DEVICE_IN_WIRED_HEADSET) { |
| snd_device = SND_DEVICE_IN_HEADSET_MIC_AEC; |
| } |
| platform_set_echo_reference(adev, true, out_device); |
| } else if (adev->active_input->enable_aec) { |
| if (in_device & AUDIO_DEVICE_IN_BACK_MIC) { |
| if (my_data->fluence_in_spkr_mode && |
| my_data->fluence_in_voice_comm && |
| (my_data->source_mic_type & SOURCE_DUAL_MIC)) { |
| snd_device = SND_DEVICE_IN_SPEAKER_DMIC_AEC; |
| } else { |
| snd_device = SND_DEVICE_IN_SPEAKER_MIC_AEC; |
| } |
| } else if (in_device & AUDIO_DEVICE_IN_BUILTIN_MIC) { |
| if (my_data->fluence_in_voice_comm && |
| (my_data->source_mic_type & SOURCE_DUAL_MIC)) { |
| snd_device = SND_DEVICE_IN_HANDSET_DMIC_AEC; |
| } else { |
| snd_device = SND_DEVICE_IN_HANDSET_MIC_AEC; |
| } |
| } else if (in_device & AUDIO_DEVICE_IN_WIRED_HEADSET) { |
| snd_device = SND_DEVICE_IN_HEADSET_MIC_AEC; |
| } |
| platform_set_echo_reference(adev, true, out_device); |
| } else if (adev->active_input->enable_ns) { |
| if (in_device & AUDIO_DEVICE_IN_BACK_MIC) { |
| if (my_data->fluence_in_spkr_mode && |
| my_data->fluence_in_voice_comm && |
| (my_data->source_mic_type & SOURCE_DUAL_MIC)) { |
| snd_device = SND_DEVICE_IN_SPEAKER_DMIC_NS; |
| } else { |
| snd_device = SND_DEVICE_IN_SPEAKER_MIC_NS; |
| } |
| } else if (in_device & AUDIO_DEVICE_IN_BUILTIN_MIC) { |
| if (my_data->fluence_in_voice_comm && |
| (my_data->source_mic_type & SOURCE_DUAL_MIC)) { |
| snd_device = SND_DEVICE_IN_HANDSET_DMIC_NS; |
| } else { |
| snd_device = SND_DEVICE_IN_HANDSET_MIC_NS; |
| } |
| } |
| } |
| } |
| } else if (source == AUDIO_SOURCE_DEFAULT) { |
| goto exit; |
| } |
| |
| |
| if (snd_device != SND_DEVICE_NONE) { |
| goto exit; |
| } |
| |
| if (in_device != AUDIO_DEVICE_NONE && |
| !(in_device & AUDIO_DEVICE_IN_VOICE_CALL) && |
| !(in_device & AUDIO_DEVICE_IN_COMMUNICATION)) { |
| if (in_device & AUDIO_DEVICE_IN_BUILTIN_MIC) { |
| if ((my_data->source_mic_type & SOURCE_QUAD_MIC) && |
| (int)channel_mask == AUDIO_CHANNEL_INDEX_MASK_4) { |
| snd_device = SND_DEVICE_IN_QUAD_MIC; |
| } else if ((my_data->source_mic_type & SOURCE_THREE_MIC) && |
| (int)channel_mask == AUDIO_CHANNEL_INDEX_MASK_3) { |
| snd_device = SND_DEVICE_IN_THREE_MIC; |
| } else if ((my_data->source_mic_type & SOURCE_DUAL_MIC) && |
| channel_count == 2) { |
| snd_device = SND_DEVICE_IN_HANDSET_DMIC_STEREO; |
| } else if ((my_data->source_mic_type & SOURCE_MONO_MIC) && |
| channel_count == 1) { |
| snd_device = SND_DEVICE_IN_HANDSET_MIC; |
| } else { |
| ALOGE("%s: something wrong (1): source type (%d) channel_count (%d) .." |
| " channel mask (0x%x) no combination found .. setting to mono", __func__, |
| my_data->source_mic_type, channel_count, channel_mask); |
| snd_device = SND_DEVICE_IN_HANDSET_MIC; |
| } |
| } else if (in_device & AUDIO_DEVICE_IN_BACK_MIC) { |
| if ((my_data->source_mic_type & SOURCE_DUAL_MIC) && |
| channel_count == 2) { |
| snd_device = SND_DEVICE_IN_SPEAKER_DMIC_STEREO; |
| } else if ((my_data->source_mic_type & SOURCE_MONO_MIC) && |
| channel_count == 1) { |
| snd_device = SND_DEVICE_IN_SPEAKER_MIC; |
| } else { |
| ALOGE("%s: something wrong (2): source type (%d) channel_count (%d) .." |
| " no combination found .. setting to mono", __func__, |
| my_data->source_mic_type, channel_count); |
| snd_device = SND_DEVICE_IN_SPEAKER_MIC; |
| } |
| } else if (in_device & AUDIO_DEVICE_IN_WIRED_HEADSET) { |
| snd_device = SND_DEVICE_IN_HEADSET_MIC; |
| } else if (in_device & AUDIO_DEVICE_IN_BLUETOOTH_SCO_HEADSET) { |
| if (adev->bt_wb_speech_enabled) { |
| if (adev->bluetooth_nrec) |
| snd_device = SND_DEVICE_IN_BT_SCO_MIC_WB_NREC; |
| else |
| snd_device = SND_DEVICE_IN_BT_SCO_MIC_WB; |
| } else { |
| if (adev->bluetooth_nrec) |
| snd_device = SND_DEVICE_IN_BT_SCO_MIC_NREC; |
| else |
| snd_device = SND_DEVICE_IN_BT_SCO_MIC; |
| } |
| } else if (in_device & AUDIO_DEVICE_IN_AUX_DIGITAL) { |
| snd_device = SND_DEVICE_IN_HDMI_MIC; |
| } else { |
| ALOGE("%s: Unknown input device(s) %#x", __func__, in_device); |
| ALOGW("%s: Using default handset-mic", __func__); |
| snd_device = SND_DEVICE_IN_HANDSET_MIC; |
| } |
| } else { |
| if (out_device & AUDIO_DEVICE_OUT_EARPIECE) { |
| snd_device = SND_DEVICE_IN_HANDSET_MIC; |
| } else if (out_device & AUDIO_DEVICE_OUT_WIRED_HEADSET) { |
| snd_device = SND_DEVICE_IN_HEADSET_MIC; |
| } else if (out_device & AUDIO_DEVICE_OUT_SPEAKER || |
| out_device & AUDIO_DEVICE_OUT_SPEAKER_SAFE || |
| out_device & AUDIO_DEVICE_OUT_WIRED_HEADPHONE || |
| out_device & AUDIO_DEVICE_OUT_LINE) { |
| if ((my_data->source_mic_type & SOURCE_DUAL_MIC) && |
| channel_count == 2) { |
| snd_device = SND_DEVICE_IN_SPEAKER_DMIC_STEREO; |
| } else if ((my_data->source_mic_type & SOURCE_MONO_MIC) && |
| channel_count == 1) { |
| snd_device = SND_DEVICE_IN_SPEAKER_MIC; |
| } else { |
| ALOGE("%s: something wrong (3): source type (%d) channel_count (%d) .." |
| " no combination found .. setting to mono", __func__, |
| my_data->source_mic_type, channel_count); |
| snd_device = SND_DEVICE_IN_SPEAKER_MIC; |
| } |
| } else if (out_device & AUDIO_DEVICE_OUT_BLUETOOTH_SCO_HEADSET) { |
| if (adev->bt_wb_speech_enabled) { |
| if (adev->bluetooth_nrec) |
| snd_device = SND_DEVICE_IN_BT_SCO_MIC_WB_NREC; |
| else |
| snd_device = SND_DEVICE_IN_BT_SCO_MIC_WB; |
| } else { |
| if (adev->bluetooth_nrec) |
| snd_device = SND_DEVICE_IN_BT_SCO_MIC_NREC; |
| else |
| snd_device = SND_DEVICE_IN_BT_SCO_MIC; |
| } |
| } else if (out_device & AUDIO_DEVICE_OUT_AUX_DIGITAL) { |
| snd_device = SND_DEVICE_IN_HDMI_MIC; |
| } else { |
| ALOGE("%s: Unknown output device(s) %#x", __func__, out_device); |
| ALOGW("%s: Using default handset-mic", __func__); |
| snd_device = SND_DEVICE_IN_HANDSET_MIC; |
| } |
| } |
| exit: |
| ALOGV("%s: exit: in_snd_device(%s)", __func__, device_table[snd_device]); |
| return snd_device; |
| } |
| |
| int platform_set_hdmi_channels(void *platform, int channel_count) |
| { |
| struct platform_data *my_data = (struct platform_data *)platform; |
| struct audio_device *adev = my_data->adev; |
| struct mixer_ctl *ctl; |
| const char *channel_cnt_str = NULL; |
| const char *mixer_ctl_name = "HDMI_RX Channels"; |
| switch (channel_count) { |
| case 8: |
| channel_cnt_str = "Eight"; break; |
| case 7: |
| channel_cnt_str = "Seven"; break; |
| case 6: |
| channel_cnt_str = "Six"; break; |
| case 5: |
| channel_cnt_str = "Five"; break; |
| case 4: |
| channel_cnt_str = "Four"; break; |
| case 3: |
| channel_cnt_str = "Three"; break; |
| default: |
| channel_cnt_str = "Two"; break; |
| } |
| ctl = mixer_get_ctl_by_name(adev->mixer, mixer_ctl_name); |
| if (!ctl) { |
| ALOGE("%s: Could not get ctl for mixer cmd - %s", |
| __func__, mixer_ctl_name); |
| return -EINVAL; |
| } |
| ALOGV("HDMI channel count: %s", channel_cnt_str); |
| mixer_ctl_set_enum_by_string(ctl, channel_cnt_str); |
| return 0; |
| } |
| |
| int platform_edid_get_max_channels(void *platform) |
| { |
| struct platform_data *my_data = (struct platform_data *)platform; |
| struct audio_device *adev = my_data->adev; |
| char block[MAX_SAD_BLOCKS * SAD_BLOCK_SIZE]; |
| char *sad = block; |
| int num_audio_blocks; |
| int channel_count; |
| int max_channels = 0; |
| int i, ret, count; |
| |
| struct mixer_ctl *ctl; |
| |
| ctl = mixer_get_ctl_by_name(adev->mixer, AUDIO_DATA_BLOCK_MIXER_CTL); |
| if (!ctl) { |
| ALOGE("%s: Could not get ctl for mixer cmd - %s", |
| __func__, AUDIO_DATA_BLOCK_MIXER_CTL); |
| return 0; |
| } |
| |
| mixer_ctl_update(ctl); |
| |
| count = mixer_ctl_get_num_values(ctl); |
| |
| /* Read SAD blocks, clamping the maximum size for safety */ |
| if (count > (int)sizeof(block)) |
| count = (int)sizeof(block); |
| |
| ret = mixer_ctl_get_array(ctl, block, count); |
| if (ret != 0) { |
| ALOGE("%s: mixer_ctl_get_array() failed to get EDID info", __func__); |
| return 0; |
| } |
| |
| /* Calculate the number of SAD blocks */ |
| num_audio_blocks = count / SAD_BLOCK_SIZE; |
| |
| for (i = 0; i < num_audio_blocks; i++) { |
| /* Only consider LPCM blocks */ |
| if ((sad[0] >> 3) != EDID_FORMAT_LPCM) { |
| sad += 3; |
| continue; |
| } |
| |
| channel_count = (sad[0] & 0x7) + 1; |
| if (channel_count > max_channels) |
| max_channels = channel_count; |
| |
| /* Advance to next block */ |
| sad += 3; |
| } |
| |
| return max_channels; |
| } |
| |
| int platform_set_incall_recording_session_id(void *platform, |
| uint32_t session_id, int rec_mode) |
| { |
| int ret = 0; |
| struct platform_data *my_data = (struct platform_data *)platform; |
| struct audio_device *adev = my_data->adev; |
| struct mixer_ctl *ctl; |
| const char *mixer_ctl_name = "Voc VSID"; |
| int num_ctl_values; |
| int i; |
| |
| ctl = mixer_get_ctl_by_name(adev->mixer, mixer_ctl_name); |
| if (!ctl) { |
| ALOGE("%s: Could not get ctl for mixer cmd - %s", |
| __func__, mixer_ctl_name); |
| ret = -EINVAL; |
| } else { |
| num_ctl_values = mixer_ctl_get_num_values(ctl); |
| for (i = 0; i < num_ctl_values; i++) { |
| if (mixer_ctl_set_value(ctl, i, session_id)) { |
| ALOGV("Error: invalid session_id: %x", session_id); |
| ret = -EINVAL; |
| break; |
| } |
| } |
| } |
| |
| if (my_data->csd != NULL) { |
| ret = my_data->csd->start_record(ALL_SESSION_VSID, rec_mode); |
| if (ret < 0) { |
| ALOGE("%s: csd_client_start_record failed, error %d", |
| __func__, ret); |
| } |
| } |
| |
| return ret; |
| } |
| |
| int platform_stop_incall_recording_usecase(void *platform) |
| { |
| int ret = 0; |
| struct platform_data *my_data = (struct platform_data *)platform; |
| |
| if (my_data->csd != NULL) { |
| ret = my_data->csd->stop_record(ALL_SESSION_VSID); |
| if (ret < 0) { |
| ALOGE("%s: csd_client_stop_record failed, error %d", |
| __func__, ret); |
| } |
| } |
| |
| return ret; |
| } |
| |
| int platform_start_incall_music_usecase(void *platform) |
| { |
| int ret = 0; |
| struct platform_data *my_data = (struct platform_data *)platform; |
| |
| if (my_data->csd != NULL) { |
| ret = my_data->csd->start_playback(ALL_SESSION_VSID); |
| if (ret < 0) { |
| ALOGE("%s: csd_client_start_playback failed, error %d", |
| __func__, ret); |
| } |
| } |
| |
| return ret; |
| } |
| |
| int platform_stop_incall_music_usecase(void *platform) |
| { |
| int ret = 0; |
| struct platform_data *my_data = (struct platform_data *)platform; |
| |
| if (my_data->csd != NULL) { |
| ret = my_data->csd->stop_playback(ALL_SESSION_VSID); |
| if (ret < 0) { |
| ALOGE("%s: csd_client_stop_playback failed, error %d", |
| __func__, ret); |
| } |
| } |
| |
| return ret; |
| } |
| |
| int platform_set_parameters(void *platform, struct str_parms *parms) |
| { |
| struct platform_data *my_data = (struct platform_data *)platform; |
| char value[128]; |
| char *kv_pairs = str_parms_to_str(parms); |
| int ret = 0, err; |
| |
| if (kv_pairs == NULL) { |
| ret = -EINVAL; |
| ALOGE("%s: key-value pair is NULL",__func__); |
| goto done; |
| } |
| |
| ALOGV("%s: enter: %s", __func__, kv_pairs); |
| |
| err = str_parms_get_str(parms, PLATFORM_CONFIG_KEY_SOUNDCARD_NAME, |
| value, sizeof(value)); |
| if (err >= 0) { |
| str_parms_del(parms, PLATFORM_CONFIG_KEY_SOUNDCARD_NAME); |
| my_data->snd_card_name = strdup(value); |
| ALOGV("%s: sound card name %s", __func__, my_data->snd_card_name); |
| } |
| |
| err = str_parms_get_str(parms, PLATFORM_CONFIG_KEY_OPERATOR_INFO, |
| value, sizeof(value)); |
| if (err >= 0) { |
| struct operator_info *info; |
| char *str = value; |
| char *name; |
| |
| str_parms_del(parms, PLATFORM_CONFIG_KEY_OPERATOR_INFO); |
| info = (struct operator_info *)calloc(1, sizeof(struct operator_info)); |
| name = strtok(str, ";"); |
| info->name = strdup(name); |
| info->mccmnc = strdup(str + strlen(name) + 1); |
| |
| list_add_tail(&operator_info_list, &info->list); |
| ALOGV("%s: add operator[%s] mccmnc[%s]", __func__, info->name, info->mccmnc); |
| } |
| |
| memset(value, 0, sizeof(value)); |
| err = str_parms_get_str(parms, PLATFORM_CONFIG_KEY_MAX_MIC_COUNT, |
| value, sizeof(value)); |
| if (err >= 0) { |
| str_parms_del(parms, PLATFORM_CONFIG_KEY_MAX_MIC_COUNT); |
| my_data->max_mic_count = atoi(value); |
| ALOGV("%s: max_mic_count %s/%d", __func__, value, my_data->max_mic_count); |
| } |
| |
| done: |
| ALOGV("%s: exit with code(%d)", __func__, ret); |
| if (kv_pairs != NULL) |
| free(kv_pairs); |
| |
| return ret; |
| } |
| |
| /* Delay in Us */ |
| int64_t platform_render_latency(audio_usecase_t usecase) |
| { |
| switch (usecase) { |
| case USECASE_AUDIO_PLAYBACK_DEEP_BUFFER: |
| return DEEP_BUFFER_PLATFORM_DELAY; |
| case USECASE_AUDIO_PLAYBACK_LOW_LATENCY: |
| return LOW_LATENCY_PLATFORM_DELAY; |
| default: |
| return 0; |
| } |
| } |
| |
| int platform_set_snd_device_backend(snd_device_t device, const char *backend_tag, |
| const char * hw_interface) |
| { |
| int ret = 0; |
| |
| if ((device < SND_DEVICE_MIN) || (device >= SND_DEVICE_MAX)) { |
| ALOGE("%s: Invalid snd_device = %d", |
| __func__, device); |
| ret = -EINVAL; |
| goto done; |
| } |
| |
| ALOGV("%s: backend_tag_table[%s]: old = %s new = %s", __func__, |
| platform_get_snd_device_name(device), |
| backend_tag_table[device] != NULL ? backend_tag_table[device]: "null", backend_tag); |
| if (backend_tag_table[device]) { |
| free(backend_tag_table[device]); |
| } |
| backend_tag_table[device] = strdup(backend_tag); |
| |
| if (hw_interface != NULL) { |
| if (hw_interface_table[device]) |
| free(hw_interface_table[device]); |
| ALOGV("%s: hw_interface_table[%d] = %s", __func__, device, hw_interface); |
| hw_interface_table[device] = strdup(hw_interface); |
| } |
| done: |
| return ret; |
| } |
| |
| int platform_set_usecase_pcm_id(audio_usecase_t usecase, int32_t type, int32_t pcm_id) |
| { |
| int ret = 0; |
| if ((usecase <= USECASE_INVALID) || (usecase >= AUDIO_USECASE_MAX)) { |
| ALOGE("%s: invalid usecase case idx %d", __func__, usecase); |
| ret = -EINVAL; |
| goto done; |
| } |
| |
| if ((type != 0) && (type != 1)) { |
| ALOGE("%s: invalid usecase type", __func__); |
| ret = -EINVAL; |
| } |
| ALOGV("%s: pcm_device_table[%d][%d] = %d", __func__, usecase, type, pcm_id); |
| pcm_device_table[usecase][type] = pcm_id; |
| done: |
| return ret; |
| } |
| |
| #define DEFAULT_NOMINAL_SPEAKER_GAIN 20 |
| int ramp_speaker_gain(struct audio_device *adev, bool ramp_up, int target_ramp_up_gain) { |
| // backup_gain: gain to try to set in case of an error during ramp |
| int start_gain, end_gain, step, backup_gain, i; |
| bool error = false; |
| const struct mixer_ctl *ctl; |
| const char *mixer_ctl_name_gain_left = "Left Speaker Gain"; |
| const char *mixer_ctl_name_gain_right = "Right Speaker Gain"; |
| struct mixer_ctl *ctl_left = mixer_get_ctl_by_name(adev->mixer, mixer_ctl_name_gain_left); |
| struct mixer_ctl *ctl_right = mixer_get_ctl_by_name(adev->mixer, mixer_ctl_name_gain_right); |
| if (!ctl_left || !ctl_right) { |
| ALOGE("%s: Could not get ctl for mixer cmd - %s or %s, not applying speaker gain ramp", |
| __func__, mixer_ctl_name_gain_left, mixer_ctl_name_gain_right); |
| return -EINVAL; |
| } else if ((mixer_ctl_get_num_values(ctl_left) != 1) |
| || (mixer_ctl_get_num_values(ctl_right) != 1)) { |
| ALOGE("%s: Unexpected num values for mixer cmd - %s or %s, not applying speaker gain ramp", |
| __func__, mixer_ctl_name_gain_left, mixer_ctl_name_gain_right); |
| return -EINVAL; |
| } |
| if (ramp_up) { |
| start_gain = 0; |
| end_gain = target_ramp_up_gain > 0 ? target_ramp_up_gain : DEFAULT_NOMINAL_SPEAKER_GAIN; |
| step = +1; |
| backup_gain = end_gain; |
| } else { |
| // using same gain on left and right |
| const int left_gain = mixer_ctl_get_value(ctl_left, 0); |
| start_gain = left_gain > 0 ? left_gain : DEFAULT_NOMINAL_SPEAKER_GAIN; |
| end_gain = 0; |
| step = -1; |
| backup_gain = start_gain; |
| } |
| for (i = start_gain ; i != (end_gain + step) ; i += step) { |
| //ALOGV("setting speaker gain to %d", i); |
| if (mixer_ctl_set_value(ctl_left, 0, i)) { |
| ALOGE("%s: error setting %s to %d during gain ramp", |
| __func__, mixer_ctl_name_gain_left, i); |
| error = true; |
| break; |
| } |
| if (mixer_ctl_set_value(ctl_right, 0, i)) { |
| ALOGE("%s: error setting %s to %d during gain ramp", |
| __func__, mixer_ctl_name_gain_right, i); |
| error = true; |
| break; |
| } |
| usleep(1000); |
| } |
| if (error) { |
| // an error occured during the ramp, let's still try to go back to a safe volume |
| if (mixer_ctl_set_value(ctl_left, 0, backup_gain)) { |
| ALOGE("%s: error restoring left gain to %d", __func__, backup_gain); |
| } |
| if (mixer_ctl_set_value(ctl_right, 0, backup_gain)) { |
| ALOGE("%s: error restoring right gain to %d", __func__, backup_gain); |
| } |
| } |
| return start_gain; |
| } |
| |
| int platform_swap_lr_channels(struct audio_device *adev, bool swap_channels) |
| { |
| // only update if there is active pcm playback on speaker |
| struct audio_usecase *usecase; |
| struct listnode *node; |
| struct platform_data *my_data = (struct platform_data *)adev->platform; |
| |
| if (my_data->speaker_lr_swap != swap_channels) { |
| my_data->speaker_lr_swap = swap_channels; |
| |
| list_for_each(node, &adev->usecase_list) { |
| usecase = node_to_item(node, struct audio_usecase, list); |
| if (usecase->type == PCM_PLAYBACK && |
| usecase->stream.out->devices & AUDIO_DEVICE_OUT_SPEAKER) { |
| /* |
| * If acdb tuning is different for SPEAKER_REVERSE, it is must |
| * to perform device switch to disable the current backend to |
| * enable it with new acdb data. |
| */ |
| if (acdb_device_table[SND_DEVICE_OUT_SPEAKER] != |
| acdb_device_table[SND_DEVICE_OUT_SPEAKER_REVERSE]) { |
| const int initial_skpr_gain = ramp_speaker_gain(adev, false /*ramp_up*/, -1); |
| select_devices(adev, usecase->id); |
| if (initial_skpr_gain != -EINVAL) { |
| ramp_speaker_gain(adev, true /*ramp_up*/, initial_skpr_gain); |
| } |
| } else { |
| const char *mixer_path; |
| if (swap_channels) { |
| mixer_path = platform_get_snd_device_name(SND_DEVICE_OUT_SPEAKER_REVERSE); |
| audio_route_apply_and_update_path(adev->audio_route, mixer_path); |
| } else { |
| mixer_path = platform_get_snd_device_name(SND_DEVICE_OUT_SPEAKER); |
| audio_route_apply_and_update_path(adev->audio_route, mixer_path); |
| } |
| } |
| break; |
| } |
| } |
| } |
| return 0; |
| } |