blob: 0ba2f4f82946259f76231f20b7ccd2d7eb071e44 [file] [log] [blame]
/*
* Copyright (c) 2014-2021, The Linux Foundation. All rights reserved.
* Not a Contribution.
*
* Copyright (C) 2014 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 "audio_hw_utils"
/* #define LOG_NDEBUG 0 */
#include <inttypes.h>
#include <errno.h>
#include <cutils/properties.h>
#include <cutils/config_utils.h>
#include <stdlib.h>
#include <dlfcn.h>
#include <cutils/str_parms.h>
#include <log/log.h>
#include <cutils/misc.h>
#include <unistd.h>
#include <sys/ioctl.h>
#include "audio_hw.h"
#include "platform.h"
#include "platform_api.h"
#include "audio_extn.h"
#include "voice_extn.h"
#include "voice.h"
#include <sound/compress_params.h>
#include <sound/compress_offload.h>
#include <sound/devdep_params.h>
#include <tinycompress/tinycompress.h>
#ifdef DYNAMIC_LOG_ENABLED
#include <log_xml_parser.h>
#define LOG_MASK HAL_MOD_FILE_UTILS
#include <log_utils.h>
#endif
#ifdef AUDIO_EXTERNAL_HDMI_ENABLED
#include "audio_parsers.h"
#endif
#define AUDIO_IO_POLICY_VENDOR_CONFIG_FILE_NAME "audio_io_policy.conf"
#define AUDIO_OUTPUT_POLICY_VENDOR_CONFIG_FILE_NAME "audio_output_policy.conf"
#define OUTPUTS_TAG "outputs"
#define INPUTS_TAG "inputs"
#define DYNAMIC_VALUE_TAG "dynamic"
#define FLAGS_TAG "flags"
#define PROFILES_TAG "profile"
#define FORMATS_TAG "formats"
#define SAMPLING_RATES_TAG "sampling_rates"
#define BIT_WIDTH_TAG "bit_width"
#define APP_TYPE_TAG "app_type"
#define STRING_TO_ENUM(string) { #string, string }
#define ARRAY_SIZE(a) (sizeof(a) / sizeof((a)[0]))
#define BASE_TABLE_SIZE 64
#define MAX_BASEINDEX_LEN 256
#ifndef SND_AUDIOCODEC_TRUEHD
#define SND_AUDIOCODEC_TRUEHD 0x00000023
#endif
#define APP_TYPE_VOIP_AUDIO 0x1113A
#ifdef AUDIO_EXTERNAL_HDMI_ENABLED
#define PROFESSIONAL (1<<0) /* 0 = consumer, 1 = professional */
#define NON_LPCM (1<<1) /* 0 = audio, 1 = non-audio */
#define SR_44100 (0<<0) /* 44.1kHz */
#define SR_NOTID (1<<0) /* non indicated */
#define SR_48000 (2<<0) /* 48kHz */
#define SR_32000 (3<<0) /* 32kHz */
#define SR_22050 (4<<0) /* 22.05kHz */
#define SR_24000 (6<<0) /* 24kHz */
#define SR_88200 (8<<0) /* 88.2kHz */
#define SR_96000 (10<<0) /* 96kHz */
#define SR_176400 (12<<0) /* 176.4kHz */
#define SR_192000 (14<<0) /* 192kHz */
#endif
/* ToDo: Check and update a proper value in msec */
#define COMPRESS_OFFLOAD_PLAYBACK_LATENCY 50
#define PCM_OFFLOAD_PLAYBACK_DSP_PATHDELAY 62
#define PCM_OFFLOAD_PLAYBACK_LATENCY PCM_OFFLOAD_PLAYBACK_DSP_PATHDELAY
#ifndef MAX_CHANNELS_SUPPORTED
#define MAX_CHANNELS_SUPPORTED 8
#endif
#ifdef __LP64__
#define VNDK_FWK_LIB_PATH "/vendor/lib64/libqti_vndfwk_detect.so"
#else
#define VNDK_FWK_LIB_PATH "/vendor/lib/libqti_vndfwk_detect.so"
#endif
typedef struct vndkfwk_s {
void *lib_handle;
int (*isVendorEnhancedFwk)(void);
int (*getVendorEnhancedInfo)(void);
const char *lib_name;
} vndkfwk_t;
static vndkfwk_t mVndkFwk = {
NULL, NULL, NULL, VNDK_FWK_LIB_PATH};
typedef struct {
const char *id_string;
const int value;
} mixer_config_lookup;
struct string_to_enum {
const char *name;
uint32_t value;
};
const struct string_to_enum s_flag_name_to_enum_table[] = {
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_DIRECT),
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_PRIMARY),
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_FAST),
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_RAW),
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_DEEP_BUFFER),
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_COMPRESS_OFFLOAD),
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_NON_BLOCKING),
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_HW_AV_SYNC),
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_INCALL_MUSIC),
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_COMPRESS_PASSTHROUGH),
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_TIMESTAMP),
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_VOIP_RX),
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_BD),
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_INTERACTIVE),
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_MEDIA),
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_SYS_NOTIFICATION),
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_NAV_GUIDANCE),
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_PHONE),
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_FRONT_PASSENGER),
STRING_TO_ENUM(AUDIO_OUTPUT_FLAG_REAR_SEAT),
STRING_TO_ENUM(AUDIO_INPUT_FLAG_NONE),
STRING_TO_ENUM(AUDIO_INPUT_FLAG_FAST),
STRING_TO_ENUM(AUDIO_INPUT_FLAG_HW_HOTWORD),
STRING_TO_ENUM(AUDIO_INPUT_FLAG_RAW),
STRING_TO_ENUM(AUDIO_INPUT_FLAG_SYNC),
STRING_TO_ENUM(AUDIO_INPUT_FLAG_TIMESTAMP),
STRING_TO_ENUM(AUDIO_INPUT_FLAG_COMPRESS),
STRING_TO_ENUM(AUDIO_INPUT_FLAG_PASSTHROUGH),
STRING_TO_ENUM(AUDIO_INPUT_FLAG_MMAP_NOIRQ),
STRING_TO_ENUM(AUDIO_INPUT_FLAG_VOIP_TX),
STRING_TO_ENUM(AUDIO_INPUT_FLAG_HW_AV_SYNC),
STRING_TO_ENUM(AUDIO_INPUT_FLAG_DIRECT),
};
const struct string_to_enum s_format_name_to_enum_table[] = {
STRING_TO_ENUM(AUDIO_FORMAT_PCM_8_BIT),
STRING_TO_ENUM(AUDIO_FORMAT_PCM_16_BIT),
STRING_TO_ENUM(AUDIO_FORMAT_PCM_24_BIT_PACKED),
STRING_TO_ENUM(AUDIO_FORMAT_PCM_8_24_BIT),
STRING_TO_ENUM(AUDIO_FORMAT_PCM_32_BIT),
STRING_TO_ENUM(AUDIO_FORMAT_MP3),
STRING_TO_ENUM(AUDIO_FORMAT_AAC),
STRING_TO_ENUM(AUDIO_FORMAT_VORBIS),
STRING_TO_ENUM(AUDIO_FORMAT_AMR_NB),
STRING_TO_ENUM(AUDIO_FORMAT_AMR_WB),
STRING_TO_ENUM(AUDIO_FORMAT_AC3),
STRING_TO_ENUM(AUDIO_FORMAT_E_AC3),
STRING_TO_ENUM(AUDIO_FORMAT_DTS),
STRING_TO_ENUM(AUDIO_FORMAT_DTS_HD),
STRING_TO_ENUM(AUDIO_FORMAT_DOLBY_TRUEHD),
STRING_TO_ENUM(AUDIO_FORMAT_IEC61937),
STRING_TO_ENUM(AUDIO_FORMAT_E_AC3_JOC),
STRING_TO_ENUM(AUDIO_FORMAT_WMA),
STRING_TO_ENUM(AUDIO_FORMAT_WMA_PRO),
STRING_TO_ENUM(AUDIO_FORMAT_AAC_ADIF),
STRING_TO_ENUM(AUDIO_FORMAT_AMR_WB_PLUS),
STRING_TO_ENUM(AUDIO_FORMAT_EVRC),
STRING_TO_ENUM(AUDIO_FORMAT_EVRCB),
STRING_TO_ENUM(AUDIO_FORMAT_EVRCWB),
STRING_TO_ENUM(AUDIO_FORMAT_QCELP),
STRING_TO_ENUM(AUDIO_FORMAT_MP2),
STRING_TO_ENUM(AUDIO_FORMAT_EVRCNW),
STRING_TO_ENUM(AUDIO_FORMAT_FLAC),
STRING_TO_ENUM(AUDIO_FORMAT_ALAC),
STRING_TO_ENUM(AUDIO_FORMAT_APE),
STRING_TO_ENUM(AUDIO_FORMAT_AAC_LC),
STRING_TO_ENUM(AUDIO_FORMAT_AAC_HE_V1),
STRING_TO_ENUM(AUDIO_FORMAT_AAC_HE_V2),
STRING_TO_ENUM(AUDIO_FORMAT_AAC_ADTS),
STRING_TO_ENUM(AUDIO_FORMAT_AAC_ADTS_LC),
STRING_TO_ENUM(AUDIO_FORMAT_AAC_ADTS_HE_V1),
STRING_TO_ENUM(AUDIO_FORMAT_AAC_ADTS_HE_V2),
STRING_TO_ENUM(AUDIO_FORMAT_DSD),
STRING_TO_ENUM(AUDIO_FORMAT_AAC_LATM),
STRING_TO_ENUM(AUDIO_FORMAT_AAC_LATM_LC),
STRING_TO_ENUM(AUDIO_FORMAT_AAC_LATM_HE_V1),
STRING_TO_ENUM(AUDIO_FORMAT_AAC_LATM_HE_V2),
STRING_TO_ENUM(AUDIO_FORMAT_APTX),
};
/* payload structure avt_device drift query */
struct audio_avt_device_drift_stats {
uint32_t minor_version;
/* Indicates the device interface direction as either
* source (Tx) or sink (Rx).
*/
uint16_t device_direction;
/* Reference timer for drift accumulation and time stamp information.
* currently it only support AFE_REF_TIMER_TYPE_AVTIMER
*/
uint16_t reference_timer;
struct audio_avt_device_drift_param drift_param;
} __attribute__((packed));
static char bTable[BASE_TABLE_SIZE] = {
'A','B','C','D','E','F','G','H','I','J','K','L',
'M','N','O','P','Q','R','S','T','U','V','W','X',
'Y','Z','a','b','c','d','e','f','g','h','i','j',
'k','l','m','n','o','p','q','r','s','t','u','v',
'w','x','y','z','0','1','2','3','4','5','6','7',
'8','9','+','/'
};
static uint32_t string_to_enum(const struct string_to_enum *table, size_t size,
const char *name)
{
size_t i;
for (i = 0; i < size; i++) {
if (strcmp(table[i].name, name) == 0) {
ALOGV("%s found %s", __func__, table[i].name);
return table[i].value;
}
}
ALOGE("%s cound not find %s", __func__, name);
return 0;
}
static audio_io_flags_t parse_flag_names(char *name)
{
uint32_t flag = 0;
audio_io_flags_t io_flags;
char *last_r;
char *flag_name = strtok_r(name, "|", &last_r);
while (flag_name != NULL) {
if (strlen(flag_name) != 0) {
flag |= string_to_enum(s_flag_name_to_enum_table,
ARRAY_SIZE(s_flag_name_to_enum_table),
flag_name);
}
flag_name = strtok_r(NULL, "|", &last_r);
}
ALOGV("parse_flag_names: flag - %x", flag);
io_flags.in_flags = (audio_input_flags_t)flag;
io_flags.out_flags = (audio_output_flags_t)flag;
return io_flags;
}
static void parse_format_names(char *name, struct streams_io_cfg *s_info)
{
struct stream_format *sf_info = NULL;
char *last_r;
char *str = strtok_r(name, "|", &last_r);
if (str != NULL && strcmp(str, DYNAMIC_VALUE_TAG) == 0)
return;
list_init(&s_info->format_list);
while (str != NULL) {
audio_format_t format = (audio_format_t)string_to_enum(s_format_name_to_enum_table,
ARRAY_SIZE(s_format_name_to_enum_table), str);
ALOGV("%s: format - %d", __func__, format);
if (format != 0) {
sf_info = (struct stream_format *)calloc(1, sizeof(struct stream_format));
if (sf_info == NULL)
break; /* return whatever was parsed */
sf_info->format = format;
list_add_tail(&s_info->format_list, &sf_info->list);
}
str = strtok_r(NULL, "|", &last_r);
}
}
static void parse_sample_rate_names(char *name, struct streams_io_cfg *s_info)
{
struct stream_sample_rate *ss_info = NULL;
uint32_t sample_rate = 48000;
char *last_r;
char *str = strtok_r(name, "|", &last_r);
if (str != NULL && 0 == strcmp(str, DYNAMIC_VALUE_TAG))
return;
list_init(&s_info->sample_rate_list);
while (str != NULL) {
sample_rate = (uint32_t)strtol(str, (char **)NULL, 10);
ALOGV("%s: sample_rate - %d", __func__, sample_rate);
if (0 != sample_rate) {
ss_info = (struct stream_sample_rate *)calloc(1, sizeof(struct stream_sample_rate));
if (!ss_info) {
ALOGE("%s: memory allocation failure", __func__);
return;
}
ss_info->sample_rate = sample_rate;
list_add_tail(&s_info->sample_rate_list, &ss_info->list);
}
str = strtok_r(NULL, "|", &last_r);
}
}
static int parse_bit_width_names(char *name)
{
int bit_width = 16;
char *last_r;
char *str = strtok_r(name, "|", &last_r);
if (str != NULL && strcmp(str, DYNAMIC_VALUE_TAG))
bit_width = (int)strtol(str, (char **)NULL, 10);
ALOGV("%s: bit_width - %d", __func__, bit_width);
return bit_width;
}
static int parse_app_type_names(void *platform, char *name)
{
int app_type = platform_get_default_app_type(platform);
char *last_r;
char *str = strtok_r(name, "|", &last_r);
if (str != NULL && strcmp(str, DYNAMIC_VALUE_TAG))
app_type = (int)strtol(str, (char **)NULL, 10);
ALOGV("%s: app_type - %d", __func__, app_type);
return app_type;
}
static void update_streams_cfg_list(cnode *root, void *platform,
struct listnode *streams_cfg_list)
{
cnode *node = root->first_child;
struct streams_io_cfg *s_info;
ALOGV("%s", __func__);
s_info = (struct streams_io_cfg *)calloc(1, sizeof(struct streams_io_cfg));
if (!s_info) {
ALOGE("failed to allocate mem for s_info list element");
return;
}
while (node) {
if (strcmp(node->name, FLAGS_TAG) == 0) {
s_info->flags = parse_flag_names((char *)node->value);
} else if (strcmp(node->name, PROFILES_TAG) == 0) {
strlcpy(s_info->profile, (char *)node->value, sizeof(s_info->profile));
} else if (strcmp(node->name, FORMATS_TAG) == 0) {
parse_format_names((char *)node->value, s_info);
} else if (strcmp(node->name, SAMPLING_RATES_TAG) == 0) {
s_info->app_type_cfg.sample_rate = CODEC_BACKEND_DEFAULT_SAMPLE_RATE;
parse_sample_rate_names((char *)node->value, s_info);
} else if (strcmp(node->name, BIT_WIDTH_TAG) == 0) {
s_info->app_type_cfg.bit_width = parse_bit_width_names((char *)node->value);
} else if (strcmp(node->name, APP_TYPE_TAG) == 0) {
s_info->app_type_cfg.app_type = parse_app_type_names(platform, (char *)node->value);
}
node = node->next;
}
list_add_tail(streams_cfg_list, &s_info->list);
}
static void load_cfg_list(cnode *root, void *platform,
struct listnode *streams_output_cfg_list,
struct listnode *streams_input_cfg_list)
{
cnode *node = NULL;
node = config_find(root, OUTPUTS_TAG);
if (node != NULL) {
node = node->first_child;
while (node) {
ALOGV("%s: loading output %s", __func__, node->name);
update_streams_cfg_list(node, platform, streams_output_cfg_list);
node = node->next;
}
} else {
ALOGI("%s: could not load output, node is NULL", __func__);
}
node = config_find(root, INPUTS_TAG);
if (node != NULL) {
node = node->first_child;
while (node) {
ALOGV("%s: loading input %s", __func__, node->name);
update_streams_cfg_list(node, platform, streams_input_cfg_list);
node = node->next;
}
} else {
ALOGI("%s: could not load input, node is NULL", __func__);
}
}
static void send_app_type_cfg(void *platform, struct mixer *mixer,
struct listnode *streams_output_cfg_list,
struct listnode *streams_input_cfg_list)
{
size_t app_type_cfg[MAX_LENGTH_MIXER_CONTROL_IN_INT] = {0};
int length = 0, i, num_app_types = 0;
struct listnode *node;
bool update;
struct mixer_ctl *ctl = NULL;
const char *mixer_ctl_name = "App Type Config";
struct streams_io_cfg *s_info = NULL;
uint32_t target_bit_width = 0;
if (!mixer) {
ALOGE("%s: mixer is null",__func__);
return;
}
ctl = mixer_get_ctl_by_name(mixer, mixer_ctl_name);
if (!ctl) {
ALOGE("%s: Could not get ctl for mixer cmd - %s",__func__, mixer_ctl_name);
return;
}
app_type_cfg[length++] = num_app_types;
if (list_empty(streams_output_cfg_list)) {
app_type_cfg[length++] = platform_get_default_app_type_v2(platform, PCM_PLAYBACK);
app_type_cfg[length++] = 48000;
app_type_cfg[length++] = 16;
num_app_types += 1;
}
if (list_empty(streams_input_cfg_list)) {
app_type_cfg[length++] = platform_get_default_app_type_v2(platform, PCM_CAPTURE);
app_type_cfg[length++] = 48000;
app_type_cfg[length++] = 16;
num_app_types += 1;
}
/* get target bit width for ADM enforce mode */
target_bit_width = adev_get_dsp_bit_width_enforce_mode();
list_for_each(node, streams_output_cfg_list) {
s_info = node_to_item(node, struct streams_io_cfg, list);
update = true;
for (i=0; i<length; i=i+3) {
if (app_type_cfg[i+1] == 0)
break;
else if (app_type_cfg[i+1] == (size_t)s_info->app_type_cfg.app_type) {
if (app_type_cfg[i+2] < (size_t)s_info->app_type_cfg.sample_rate)
app_type_cfg[i+2] = s_info->app_type_cfg.sample_rate;
if (app_type_cfg[i+3] < (size_t)s_info->app_type_cfg.bit_width)
app_type_cfg[i+3] = s_info->app_type_cfg.bit_width;
/* ADM bit width = max(enforce_bit_width, bit_width from s_info */
if (audio_extn_is_dsp_bit_width_enforce_mode_supported(s_info->flags.out_flags) &&
(target_bit_width > app_type_cfg[i+3]))
app_type_cfg[i+3] = target_bit_width;
update = false;
break;
}
}
if (update && ((length + 3) <= MAX_LENGTH_MIXER_CONTROL_IN_INT)) {
num_app_types += 1;
app_type_cfg[length++] = s_info->app_type_cfg.app_type;
app_type_cfg[length++] = s_info->app_type_cfg.sample_rate;
app_type_cfg[length] = s_info->app_type_cfg.bit_width;
if (audio_extn_is_dsp_bit_width_enforce_mode_supported(s_info->flags.out_flags) &&
(target_bit_width > app_type_cfg[length]))
app_type_cfg[length] = target_bit_width;
length++;
}
}
list_for_each(node, streams_input_cfg_list) {
s_info = node_to_item(node, struct streams_io_cfg, list);
update = true;
for (i=0; i<length; i=i+3) {
if (app_type_cfg[i+1] == 0)
break;
else if (app_type_cfg[i+1] == (size_t)s_info->app_type_cfg.app_type) {
if (app_type_cfg[i+2] < (size_t)s_info->app_type_cfg.sample_rate)
app_type_cfg[i+2] = s_info->app_type_cfg.sample_rate;
if (app_type_cfg[i+3] < (size_t)s_info->app_type_cfg.bit_width)
app_type_cfg[i+3] = s_info->app_type_cfg.bit_width;
update = false;
break;
}
}
if (update && ((length + 3) <= MAX_LENGTH_MIXER_CONTROL_IN_INT)) {
num_app_types += 1;
app_type_cfg[length++] = s_info->app_type_cfg.app_type;
app_type_cfg[length++] = s_info->app_type_cfg.sample_rate;
app_type_cfg[length++] = s_info->app_type_cfg.bit_width;
}
}
ALOGV("%s: num_app_types: %d", __func__, num_app_types);
if (num_app_types) {
app_type_cfg[0] = num_app_types;
mixer_ctl_set_array(ctl, app_type_cfg, length);
}
}
/* Function to retrieve audio vendor configs path */
void audio_get_vendor_config_path (char* config_file_path, int path_size)
{
char vendor_sku[PROPERTY_VALUE_MAX] = {'\0'};
if (property_get("ro.boot.product.vendor.sku", vendor_sku, "") <= 0) {
#ifdef LINUX_ENABLED
/* Audio configs are stored in /etc */
snprintf(config_file_path, path_size, "%s", "/etc");
#else
/* Audio configs are stored in /vendor/etc */
snprintf(config_file_path, path_size, "%s", "/vendor/etc");
#endif
} else {
/* Audio configs are stored in /vendor/etc/audio/sku_${vendor_sku} */
snprintf(config_file_path, path_size,
"%s%s", "/vendor/etc/audio/sku_", vendor_sku);
}
}
void audio_extn_utils_update_streams_cfg_lists(void *platform,
struct mixer *mixer,
struct listnode *streams_output_cfg_list,
struct listnode *streams_input_cfg_list)
{
cnode *root;
char *data = NULL;
char vendor_config_path[VENDOR_CONFIG_PATH_MAX_LENGTH];
char audio_io_policy_file[VENDOR_CONFIG_FILE_MAX_LENGTH];
char audio_output_policy_file[VENDOR_CONFIG_FILE_MAX_LENGTH];
ALOGV("%s", __func__);
list_init(streams_output_cfg_list);
list_init(streams_input_cfg_list);
root = config_node("", "");
if (root == NULL) {
ALOGE("cfg_list, NULL config root");
return;
}
/* Get path for audio configuration files in vendor */
audio_get_vendor_config_path(vendor_config_path,
sizeof(vendor_config_path));
/* Get path for audio_io_policy_file in vendor */
snprintf(audio_io_policy_file, sizeof(audio_io_policy_file),
"%s/%s", vendor_config_path, AUDIO_IO_POLICY_VENDOR_CONFIG_FILE_NAME);
/* Load audio_io_policy_file from vendor */
data = (char *)load_file(audio_io_policy_file, NULL);
if (data == NULL) {
ALOGD("%s: failed to open io config file(%s), trying older config file",
__func__, audio_io_policy_file);
/* Get path for audio_output_policy_file in vendor */
snprintf(audio_output_policy_file, sizeof(audio_output_policy_file),
"%s/%s", vendor_config_path,
AUDIO_OUTPUT_POLICY_VENDOR_CONFIG_FILE_NAME);
/* Load audio_output_policy_file from vendor */
data = (char *)load_file(audio_output_policy_file, NULL);
if (data == NULL) {
send_app_type_cfg(platform, mixer,
streams_output_cfg_list,
streams_input_cfg_list);
ALOGE("%s: could not load io policy config!", __func__);
free(root);
return;
}
}
config_load(root, data);
load_cfg_list(root, platform, streams_output_cfg_list,
streams_input_cfg_list);
send_app_type_cfg(platform, mixer, streams_output_cfg_list,
streams_input_cfg_list);
config_free(root);
free(root);
free(data);
}
static void audio_extn_utils_dump_streams_cfg_list(
struct listnode *streams_cfg_list)
{
struct listnode *node_i, *node_j;
struct streams_io_cfg *s_info;
struct stream_format *sf_info;
struct stream_sample_rate *ss_info;
list_for_each(node_i, streams_cfg_list) {
s_info = node_to_item(node_i, struct streams_io_cfg, list);
ALOGV("%s: flags-%d, sample_rate-%d, bit_width-%d, app_type-%d",
__func__, s_info->flags.out_flags, s_info->app_type_cfg.sample_rate,
s_info->app_type_cfg.bit_width, s_info->app_type_cfg.app_type);
list_for_each(node_j, &s_info->format_list) {
sf_info = node_to_item(node_j, struct stream_format, list);
ALOGV("format-%x", sf_info->format);
}
list_for_each(node_j, &s_info->sample_rate_list) {
ss_info = node_to_item(node_j, struct stream_sample_rate, list);
ALOGV("sample rate-%d", ss_info->sample_rate);
}
}
}
void audio_extn_utils_dump_streams_cfg_lists(
struct listnode *streams_output_cfg_list,
struct listnode *streams_input_cfg_list)
{
ALOGV("%s", __func__);
audio_extn_utils_dump_streams_cfg_list(streams_output_cfg_list);
audio_extn_utils_dump_streams_cfg_list(streams_input_cfg_list);
}
static void audio_extn_utils_release_streams_cfg_list(
struct listnode *streams_cfg_list)
{
struct listnode *node_i, *node_j;
struct streams_io_cfg *s_info;
ALOGV("%s", __func__);
while (!list_empty(streams_cfg_list)) {
node_i = list_head(streams_cfg_list);
s_info = node_to_item(node_i, struct streams_io_cfg, list);
while (!list_empty(&s_info->format_list)) {
node_j = list_head(&s_info->format_list);
list_remove(node_j);
free(node_to_item(node_j, struct stream_format, list));
}
while (!list_empty(&s_info->sample_rate_list)) {
node_j = list_head(&s_info->sample_rate_list);
list_remove(node_j);
free(node_to_item(node_j, struct stream_sample_rate, list));
}
list_remove(node_i);
free(node_to_item(node_i, struct streams_io_cfg, list));
}
}
void audio_extn_utils_release_streams_cfg_lists(
struct listnode *streams_output_cfg_list,
struct listnode *streams_input_cfg_list)
{
ALOGV("%s", __func__);
audio_extn_utils_release_streams_cfg_list(streams_output_cfg_list);
audio_extn_utils_release_streams_cfg_list(streams_input_cfg_list);
}
static bool set_app_type_cfg(struct streams_io_cfg *s_info,
struct stream_app_type_cfg *app_type_cfg,
uint32_t sample_rate, uint32_t bit_width)
{
struct listnode *node_i;
struct stream_sample_rate *ss_info;
list_for_each(node_i, &s_info->sample_rate_list) {
ss_info = node_to_item(node_i, struct stream_sample_rate, list);
if ((sample_rate <= ss_info->sample_rate) &&
(bit_width == s_info->app_type_cfg.bit_width)) {
app_type_cfg->app_type = s_info->app_type_cfg.app_type;
app_type_cfg->sample_rate = ss_info->sample_rate;
app_type_cfg->bit_width = s_info->app_type_cfg.bit_width;
ALOGV("%s app_type_cfg->app_type %d, app_type_cfg->sample_rate %d, app_type_cfg->bit_width %d",
__func__, app_type_cfg->app_type, app_type_cfg->sample_rate, app_type_cfg->bit_width);
return true;
}
}
/*
* Reiterate through the list assuming dafault sample rate.
* Handles scenario where input sample rate is higher
* than all sample rates in list for the input bit width.
*/
sample_rate = CODEC_BACKEND_DEFAULT_SAMPLE_RATE;
list_for_each(node_i, &s_info->sample_rate_list) {
ss_info = node_to_item(node_i, struct stream_sample_rate, list);
if ((sample_rate <= ss_info->sample_rate) &&
(bit_width == s_info->app_type_cfg.bit_width)) {
app_type_cfg->app_type = s_info->app_type_cfg.app_type;
app_type_cfg->sample_rate = sample_rate;
app_type_cfg->bit_width = s_info->app_type_cfg.bit_width;
ALOGV("%s Assuming sample rate. app_type_cfg->app_type %d, app_type_cfg->sample_rate %d, app_type_cfg->bit_width %d",
__func__, app_type_cfg->app_type, app_type_cfg->sample_rate, app_type_cfg->bit_width);
return true;
}
}
return false;
}
void audio_extn_utils_update_stream_input_app_type_cfg(void *platform,
struct listnode *streams_input_cfg_list,
struct listnode *devices __unused,
audio_input_flags_t flags,
audio_format_t format,
uint32_t sample_rate,
uint32_t bit_width,
char* profile,
struct stream_app_type_cfg *app_type_cfg)
{
struct listnode *node_i, *node_j;
struct streams_io_cfg *s_info;
struct stream_format *sf_info;
ALOGV("%s: flags: 0x%x, format: 0x%x sample_rate %d, profile %s",
__func__, flags, format, sample_rate, profile);
list_for_each(node_i, streams_input_cfg_list) {
s_info = node_to_item(node_i, struct streams_io_cfg, list);
/* Along with flags do profile matching if set at either end.*/
if (s_info->flags.in_flags == flags &&
((profile[0] == '\0' && s_info->profile[0] == '\0') ||
strncmp(s_info->profile, profile, sizeof(s_info->profile)) == 0)) {
list_for_each(node_j, &s_info->format_list) {
sf_info = node_to_item(node_j, struct stream_format, list);
if (sf_info->format == format) {
if (set_app_type_cfg(s_info, app_type_cfg, sample_rate, bit_width))
return;
}
}
}
}
ALOGW("%s: App type could not be selected. Falling back to default", __func__);
app_type_cfg->app_type = platform_get_default_app_type_v2(platform, PCM_CAPTURE);
app_type_cfg->sample_rate = CODEC_BACKEND_DEFAULT_SAMPLE_RATE;
app_type_cfg->bit_width = 16;
}
void audio_extn_utils_update_stream_output_app_type_cfg(void *platform,
struct listnode *streams_output_cfg_list,
struct listnode *devices,
audio_output_flags_t flags,
audio_format_t format,
uint32_t sample_rate,
uint32_t bit_width,
audio_channel_mask_t channel_mask,
char *profile,
struct stream_app_type_cfg *app_type_cfg)
{
struct listnode *node_i, *node_j;
struct streams_io_cfg *s_info;
struct stream_format *sf_info;
char value[PROPERTY_VALUE_MAX] = {0};
if (compare_device_type(devices, AUDIO_DEVICE_OUT_SPEAKER)) {
int bw = platform_get_snd_device_bit_width(SND_DEVICE_OUT_SPEAKER);
if ((-ENOSYS != bw) && (bit_width > (uint32_t)bw))
bit_width = (uint32_t)bw;
else if (-ENOSYS == bw)
bit_width = CODEC_BACKEND_DEFAULT_BIT_WIDTH;
sample_rate = DEFAULT_OUTPUT_SAMPLING_RATE;
ALOGI("%s Allowing 24 and above bits playback on speaker ONLY at default sampling rate", __func__);
}
property_get("vendor.audio.playback.mch.downsample",value,"");
if (!strncmp("true", value, sizeof("true"))) {
if ((popcount(channel_mask) > 2) &&
(sample_rate > CODEC_BACKEND_DEFAULT_SAMPLE_RATE) &&
!(flags & (audio_output_flags_t)AUDIO_OUTPUT_FLAG_COMPRESS_PASSTHROUGH)) {
sample_rate = CODEC_BACKEND_DEFAULT_SAMPLE_RATE;
ALOGD("%s: MCH session defaulting sample rate to %d",
__func__, sample_rate);
}
}
/* Set sampling rate to 176.4 for DSD64
* and 352.8Khz for DSD128.
* Set Bit Width to 16. output will be 16 bit
* post DoP in ASM.
*/
if ((flags & (audio_output_flags_t)AUDIO_OUTPUT_FLAG_COMPRESS_PASSTHROUGH) &&
(format == AUDIO_FORMAT_DSD)) {
bit_width = 16;
if (sample_rate == INPUT_SAMPLING_RATE_DSD64)
sample_rate = OUTPUT_SAMPLING_RATE_DSD64;
else if (sample_rate == INPUT_SAMPLING_RATE_DSD128)
sample_rate = OUTPUT_SAMPLING_RATE_DSD128;
}
ALOGV("%s: flags: %x, format: %x sample_rate %d, profile %s, app_type %d",
__func__, flags, format, sample_rate, profile, app_type_cfg->app_type);
list_for_each(node_i, streams_output_cfg_list) {
s_info = node_to_item(node_i, struct streams_io_cfg, list);
/* Along with flags do profile matching if set at either end.*/
if (s_info->flags.out_flags == flags &&
((profile[0] == '\0' && s_info->profile[0] == '\0') ||
strncmp(s_info->profile, profile, sizeof(s_info->profile)) == 0)) {
list_for_each(node_j, &s_info->format_list) {
sf_info = node_to_item(node_j, struct stream_format, list);
if (sf_info->format == format) {
if (set_app_type_cfg(s_info, app_type_cfg, sample_rate, bit_width))
return;
}
}
}
}
list_for_each(node_i, streams_output_cfg_list) {
s_info = node_to_item(node_i, struct streams_io_cfg, list);
if (s_info->flags.out_flags == AUDIO_OUTPUT_FLAG_PRIMARY) {
ALOGV("Compatible output profile not found.");
app_type_cfg->app_type = s_info->app_type_cfg.app_type;
app_type_cfg->sample_rate = s_info->app_type_cfg.sample_rate;
app_type_cfg->bit_width = s_info->app_type_cfg.bit_width;
ALOGV("%s Default to primary output: App type: %d sample_rate %d",
__func__, s_info->app_type_cfg.app_type, app_type_cfg->sample_rate);
return;
}
}
ALOGW("%s: App type could not be selected. Falling back to default", __func__);
app_type_cfg->app_type = platform_get_default_app_type(platform);
app_type_cfg->sample_rate = CODEC_BACKEND_DEFAULT_SAMPLE_RATE;
app_type_cfg->bit_width = 16;
}
static bool audio_is_this_native_usecase(struct audio_usecase *uc)
{
bool native_usecase = false;
struct stream_out *out = (struct stream_out*) uc->stream.out;
if (PCM_PLAYBACK == uc->type && out != NULL &&
NATIVE_AUDIO_MODE_INVALID != platform_get_native_support() &&
is_offload_usecase(uc->id) &&
(out->sample_rate == OUTPUT_SAMPLING_RATE_44100))
native_usecase = true;
return native_usecase;
}
bool audio_extn_is_dsp_bit_width_enforce_mode_supported(audio_output_flags_t flags)
{
/* DSP bitwidth enforce mode for ADM and AFE:
* includes:
* deep buffer, low latency, direct pcm and offload.
* excludes:
* ull(raw+fast), VOIP.
*/
if ((flags & AUDIO_OUTPUT_FLAG_VOIP_RX) ||
((flags & AUDIO_OUTPUT_FLAG_RAW) &&
(flags & AUDIO_OUTPUT_FLAG_FAST)))
return false;
if ((flags & AUDIO_OUTPUT_FLAG_COMPRESS_OFFLOAD) ||
(flags & AUDIO_OUTPUT_FLAG_DIRECT) ||
(flags & AUDIO_OUTPUT_FLAG_DEEP_BUFFER) ||
(flags & AUDIO_OUTPUT_FLAG_PRIMARY))
return true;
else
return false;
}
static inline bool audio_is_vr_mode_on(struct audio_device *(__attribute__((unused)) adev))
{
return adev->vr_audio_mode_enabled;
}
static void audio_extn_btsco_get_sample_rate(int snd_device, int *sample_rate)
{
switch (snd_device) {
case SND_DEVICE_OUT_BT_SCO:
case SND_DEVICE_IN_BT_SCO_MIC:
case SND_DEVICE_IN_BT_SCO_MIC_NREC:
*sample_rate = 8000;
break;
case SND_DEVICE_OUT_BT_SCO_WB:
case SND_DEVICE_IN_BT_SCO_MIC_WB:
case SND_DEVICE_IN_BT_SCO_MIC_WB_NREC:
*sample_rate = 16000;
break;
default:
ALOGD("%s:Not a BT SCO device, need not update sampling rate\n", __func__);
break;
}
}
void audio_extn_utils_update_stream_app_type_cfg_for_usecase(
struct audio_device *adev,
struct audio_usecase *usecase)
{
ALOGV("%s", __func__);
switch(usecase->type) {
case PCM_PLAYBACK:
audio_extn_utils_update_stream_output_app_type_cfg(adev->platform,
&adev->streams_output_cfg_list,
&usecase->stream.out->device_list,
usecase->stream.out->flags,
usecase->stream.out->hal_op_format,
usecase->stream.out->sample_rate,
usecase->stream.out->bit_width,
usecase->stream.out->channel_mask,
usecase->stream.out->profile,
&usecase->stream.out->app_type_cfg);
ALOGV("%s Selected apptype: %d", __func__, usecase->stream.out->app_type_cfg.app_type);
break;
case PCM_CAPTURE:
if (usecase->id == USECASE_AUDIO_RECORD_VOIP)
usecase->stream.in->app_type_cfg.app_type = APP_TYPE_VOIP_AUDIO;
else
audio_extn_utils_update_stream_input_app_type_cfg(adev->platform,
&adev->streams_input_cfg_list,
&usecase->stream.in->device_list,
usecase->stream.in->flags,
usecase->stream.in->format,
usecase->stream.in->sample_rate,
usecase->stream.in->bit_width,
usecase->stream.in->profile,
&usecase->stream.in->app_type_cfg);
ALOGV("%s Selected apptype: %d", __func__, usecase->stream.in->app_type_cfg.app_type);
break;
case TRANSCODE_LOOPBACK_RX :
audio_extn_utils_update_stream_output_app_type_cfg(adev->platform,
&adev->streams_output_cfg_list,
&usecase->stream.inout->out_config.device_list,
0,
usecase->stream.inout->out_config.format,
usecase->stream.inout->out_config.sample_rate,
usecase->stream.inout->out_config.bit_width,
usecase->stream.inout->out_config.channel_mask,
usecase->stream.inout->profile,
&usecase->stream.inout->out_app_type_cfg);
ALOGV("%s Selected apptype: %d", __func__, usecase->stream.inout->out_app_type_cfg.app_type);
break;
case PCM_HFP_CALL:
switch (usecase->id) {
case USECASE_AUDIO_HFP_SCO:
case USECASE_AUDIO_HFP_SCO_WB:
audio_extn_btsco_get_sample_rate(usecase->out_snd_device,
&usecase->out_app_type_cfg.sample_rate);
usecase->in_app_type_cfg.sample_rate = usecase->out_app_type_cfg.sample_rate;
break;
case USECASE_AUDIO_HFP_SCO_DOWNLINK:
case USECASE_AUDIO_HFP_SCO_WB_DOWNLINK:
audio_extn_btsco_get_sample_rate(usecase->in_snd_device,
&usecase->in_app_type_cfg.sample_rate);
usecase->out_app_type_cfg.sample_rate = usecase->in_app_type_cfg.sample_rate;
break;
default:
ALOGE("%s: usecase id (%d) not supported, use default sample rate",
__func__, usecase->id);
usecase->in_app_type_cfg.sample_rate = CODEC_BACKEND_DEFAULT_SAMPLE_RATE;
usecase->out_app_type_cfg.sample_rate = CODEC_BACKEND_DEFAULT_SAMPLE_RATE;
break;
}
/* update out_app_type_cfg */
usecase->out_app_type_cfg.bit_width =
platform_get_snd_device_bit_width(usecase->out_snd_device);
usecase->out_app_type_cfg.app_type =
platform_get_default_app_type_v2(adev->platform, PCM_PLAYBACK);
/* update in_app_type_cfg */
usecase->in_app_type_cfg.bit_width =
platform_get_snd_device_bit_width(usecase->in_snd_device);
usecase->in_app_type_cfg.app_type =
platform_get_default_app_type_v2(adev->platform, PCM_CAPTURE);
ALOGV("%s Selected apptype: playback %d capture %d",
__func__, usecase->out_app_type_cfg.app_type, usecase->in_app_type_cfg.app_type);
break;
default:
ALOGE("%s: app type cfg not supported for usecase type (%d)",
__func__, usecase->type);
}
}
static int set_stream_app_type_mixer_ctrl(struct audio_device *adev,
int pcm_device_id, int app_type,
int acdb_dev_id, int sample_rate,
int stream_type,
snd_device_t snd_device)
{
char mixer_ctl_name[MAX_LENGTH_MIXER_CONTROL_IN_INT];
struct mixer_ctl *ctl;
int app_type_cfg[MAX_LENGTH_MIXER_CONTROL_IN_INT], len = 0, rc = 0;
int snd_device_be_idx = -1;
if (stream_type == PCM_PLAYBACK) {
snprintf(mixer_ctl_name, sizeof(mixer_ctl_name),
"Audio Stream %d App Type Cfg", pcm_device_id);
} else if (stream_type == PCM_CAPTURE) {
snprintf(mixer_ctl_name, sizeof(mixer_ctl_name),
"Audio Stream Capture %d App Type Cfg", pcm_device_id);
}
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);
rc = -EINVAL;
goto exit;
}
app_type_cfg[len++] = app_type;
app_type_cfg[len++] = acdb_dev_id;
app_type_cfg[len++] = sample_rate;
snd_device_be_idx = platform_get_snd_device_backend_index(snd_device);
if (snd_device_be_idx > 0)
app_type_cfg[len++] = snd_device_be_idx;
ALOGV("%s: stream type %d app_type %d, acdb_dev_id %d "
"sample rate %d, snd_device_be_idx %d",
__func__, stream_type, app_type, acdb_dev_id, sample_rate,
snd_device_be_idx);
mixer_ctl_set_array(ctl, app_type_cfg, len);
exit:
return rc;
}
static int audio_extn_utils_send_app_type_cfg_hfp(struct audio_device *adev,
struct audio_usecase *usecase)
{
int pcm_device_id, acdb_dev_id = 0, snd_device = usecase->out_snd_device;
int32_t sample_rate = DEFAULT_OUTPUT_SAMPLING_RATE;
int app_type = 0, rc = 0;
bool is_bus_dev_usecase = false;
ALOGV("%s", __func__);
if (usecase->type != PCM_HFP_CALL) {
ALOGV("%s: not a HFP path, no need to cfg app type", __func__);
rc = 0;
goto exit_send_app_type_cfg;
}
if ((usecase->id != USECASE_AUDIO_HFP_SCO) &&
(usecase->id != USECASE_AUDIO_HFP_SCO_WB) &&
(usecase->id != USECASE_AUDIO_HFP_SCO_DOWNLINK) &&
(usecase->id != USECASE_AUDIO_HFP_SCO_WB_DOWNLINK)) {
ALOGV("%s: a usecase where app type cfg is not required", __func__);
rc = 0;
goto exit_send_app_type_cfg;
}
if (compare_device_type(&usecase->device_list, AUDIO_DEVICE_OUT_BUS))
is_bus_dev_usecase = true;
snd_device = usecase->out_snd_device;
pcm_device_id = platform_get_pcm_device_id(usecase->id, PCM_PLAYBACK);
acdb_dev_id = platform_get_snd_device_acdb_id(snd_device);
if (acdb_dev_id < 0) {
ALOGE("%s: Couldn't get the acdb dev id", __func__);
rc = -EINVAL;
goto exit_send_app_type_cfg;
}
if (usecase->type == PCM_HFP_CALL) {
/* config HFP session:1 playback path */
if (is_bus_dev_usecase) {
app_type = usecase->out_app_type_cfg.app_type;
sample_rate= usecase->out_app_type_cfg.sample_rate;
} else {
snd_device = SND_DEVICE_NONE; // use legacy behavior
app_type = platform_get_default_app_type_v2(adev->platform, PCM_PLAYBACK);
sample_rate= CODEC_BACKEND_DEFAULT_SAMPLE_RATE;
}
rc = set_stream_app_type_mixer_ctrl(adev, pcm_device_id, app_type,
acdb_dev_id, sample_rate,
PCM_PLAYBACK,
snd_device);
if (rc < 0)
goto exit_send_app_type_cfg;
/* config HFP session:1 capture path */
if (is_bus_dev_usecase) {
snd_device = usecase->in_snd_device;
pcm_device_id = platform_get_pcm_device_id(usecase->id, PCM_CAPTURE);
acdb_dev_id = platform_get_snd_device_acdb_id(snd_device);
if (acdb_dev_id < 0) {
ALOGE("%s: Couldn't get the acdb dev id", __func__);
rc = -EINVAL;
goto exit_send_app_type_cfg;
}
app_type = usecase->in_app_type_cfg.app_type;
sample_rate= usecase->in_app_type_cfg.sample_rate;
} else {
snd_device = SND_DEVICE_NONE; // use legacy behavior
app_type = platform_get_default_app_type_v2(adev->platform, PCM_CAPTURE);
}
rc = set_stream_app_type_mixer_ctrl(adev, pcm_device_id, app_type,
acdb_dev_id, sample_rate,
PCM_CAPTURE,
snd_device);
if (rc < 0)
goto exit_send_app_type_cfg;
if (is_bus_dev_usecase)
goto exit_send_app_type_cfg;
/* config HFP session:2 capture path */
pcm_device_id = HFP_ASM_RX_TX;
snd_device = usecase->in_snd_device;
acdb_dev_id = platform_get_snd_device_acdb_id(snd_device);
if (acdb_dev_id <= 0) {
ALOGE("%s: Couldn't get the acdb dev id", __func__);
rc = -EINVAL;
goto exit_send_app_type_cfg;
}
app_type = platform_get_default_app_type_v2(adev->platform, PCM_CAPTURE);
rc = set_stream_app_type_mixer_ctrl(adev, pcm_device_id, app_type,
acdb_dev_id, sample_rate, PCM_CAPTURE,
SND_DEVICE_NONE);
if (rc < 0)
goto exit_send_app_type_cfg;
/* config HFP session:2 playback path */
app_type = platform_get_default_app_type_v2(adev->platform, PCM_PLAYBACK);
rc = set_stream_app_type_mixer_ctrl(adev, pcm_device_id, app_type,
acdb_dev_id, sample_rate,
PCM_PLAYBACK, SND_DEVICE_NONE);
if (rc < 0)
goto exit_send_app_type_cfg;
}
rc = 0;
exit_send_app_type_cfg:
return rc;
}
int audio_extn_utils_get_app_sample_rate_for_device(
struct audio_device *adev,
struct audio_usecase *usecase, int snd_device)
{
char value[PROPERTY_VALUE_MAX] = {0};
int sample_rate = DEFAULT_OUTPUT_SAMPLING_RATE;
if ((usecase->type == PCM_PLAYBACK) && (usecase->stream.out != NULL)) {
property_get("vendor.audio.playback.mch.downsample",value,"");
if (!strncmp("true", value, sizeof("true"))) {
if ((popcount(usecase->stream.out->channel_mask) > 2) &&
(usecase->stream.out->app_type_cfg.sample_rate > CODEC_BACKEND_DEFAULT_SAMPLE_RATE) &&
!(usecase->stream.out->flags &
(audio_output_flags_t)AUDIO_OUTPUT_FLAG_COMPRESS_PASSTHROUGH))
sample_rate = CODEC_BACKEND_DEFAULT_SAMPLE_RATE;
}
if (usecase->id == USECASE_AUDIO_PLAYBACK_VOIP) {
usecase->stream.out->app_type_cfg.sample_rate = usecase->stream.out->sample_rate;
} else if ((snd_device == SND_DEVICE_OUT_HDMI ||
snd_device == SND_DEVICE_OUT_USB_HEADSET ||
snd_device == SND_DEVICE_OUT_DISPLAY_PORT) &&
(usecase->stream.out->sample_rate >= OUTPUT_SAMPLING_RATE_44100)) {
/*
* To best utlize DSP, check if the stream sample rate is supported/multiple of
* configured device sample rate, if not update the COPP rate to be equal to the
* device sample rate, else open COPP at stream sample rate
*/
platform_check_and_update_copp_sample_rate(adev->platform, snd_device,
usecase->stream.out->sample_rate,
&usecase->stream.out->app_type_cfg.sample_rate);
} else if (((snd_device != SND_DEVICE_OUT_HEADPHONES_44_1 &&
!audio_is_this_native_usecase(usecase)) &&
usecase->stream.out->sample_rate == OUTPUT_SAMPLING_RATE_44100) ||
(usecase->stream.out->sample_rate < OUTPUT_SAMPLING_RATE_44100)) {
/* Reset to default if no native stream is active*/
usecase->stream.out->app_type_cfg.sample_rate = DEFAULT_OUTPUT_SAMPLING_RATE;
} else if (snd_device == SND_DEVICE_OUT_BT_A2DP) {
/*
* For a2dp playback get encoder sampling rate and set copp sampling rate,
* for bit width use the stream param only.
*/
audio_extn_a2dp_get_enc_sample_rate(&usecase->stream.out->app_type_cfg.sample_rate);
ALOGI("%s using %d sample rate rate for A2DP CoPP",
__func__, usecase->stream.out->app_type_cfg.sample_rate);
} else if (compare_device_type(&usecase->stream.out->device_list,
AUDIO_DEVICE_OUT_SPEAKER)) {
usecase->stream.out->app_type_cfg.sample_rate = DEFAULT_OUTPUT_SAMPLING_RATE;
}
audio_extn_btsco_get_sample_rate(snd_device, &usecase->stream.out->app_type_cfg.sample_rate);
sample_rate = usecase->stream.out->app_type_cfg.sample_rate;
if (((usecase->stream.out->format == AUDIO_FORMAT_E_AC3) ||
(usecase->stream.out->format == AUDIO_FORMAT_E_AC3_JOC) ||
(usecase->stream.out->format == AUDIO_FORMAT_DOLBY_TRUEHD))
&& audio_extn_passthru_is_passthrough_stream(usecase->stream.out)
&& !audio_extn_passthru_is_convert_supported(adev, usecase->stream.out)) {
sample_rate = sample_rate * 4;
if (sample_rate > HDMI_PASSTHROUGH_MAX_SAMPLE_RATE)
sample_rate = HDMI_PASSTHROUGH_MAX_SAMPLE_RATE;
}
} else if (usecase->type == PCM_CAPTURE) {
if (usecase->stream.in != NULL) {
if (usecase->id == USECASE_AUDIO_RECORD_VOIP)
usecase->stream.in->app_type_cfg.sample_rate = usecase->stream.in->sample_rate;
if (voice_is_in_call_rec_stream(usecase->stream.in)) {
audio_extn_btsco_get_sample_rate(usecase->in_snd_device,
&usecase->stream.in->app_type_cfg.sample_rate);
} else {
audio_extn_btsco_get_sample_rate(snd_device,
&usecase->stream.in->app_type_cfg.sample_rate);
}
sample_rate = usecase->stream.in->app_type_cfg.sample_rate;
} else if (usecase->id == USECASE_AUDIO_SPKR_CALIB_TX) {
if ((property_get("vendor.audio.spkr_prot.tx.sampling_rate", value, NULL) > 0))
sample_rate = atoi(value);
else
sample_rate = SAMPLE_RATE_8000;
}
} else if (usecase->type == TRANSCODE_LOOPBACK_RX) {
sample_rate = usecase->stream.inout->out_config.sample_rate;
}
return sample_rate;
}
static int send_app_type_cfg_for_device(struct audio_device *adev,
struct audio_usecase *usecase,
int split_snd_device)
{
char mixer_ctl_name[MAX_LENGTH_MIXER_CONTROL_IN_INT];
size_t app_type_cfg[MAX_LENGTH_MIXER_CONTROL_IN_INT] = {0};
int len = 0, rc;
struct mixer_ctl *ctl;
int pcm_device_id = 0, acdb_dev_id, app_type;
int snd_device = split_snd_device, snd_device_be_idx = -1;
int32_t sample_rate = DEFAULT_OUTPUT_SAMPLING_RATE;
struct streams_io_cfg *s_info = NULL;
struct listnode *node = NULL;
int bd_app_type = 0;
ALOGV("%s: usecase->out_snd_device %s, usecase->in_snd_device %s, split_snd_device %s",
__func__, platform_get_snd_device_name(usecase->out_snd_device),
platform_get_snd_device_name(usecase->in_snd_device),
platform_get_snd_device_name(split_snd_device));
if (usecase->type != PCM_PLAYBACK && usecase->type != PCM_CAPTURE &&
usecase->type != TRANSCODE_LOOPBACK_RX) {
ALOGE("%s: not a playback/capture path, no need to cfg app type", __func__);
rc = 0;
goto exit_send_app_type_cfg;
}
if ((usecase->id != USECASE_AUDIO_PLAYBACK_DEEP_BUFFER) &&
(usecase->id != USECASE_AUDIO_PLAYBACK_WITH_HAPTICS) &&
(usecase->id != USECASE_AUDIO_PLAYBACK_LOW_LATENCY) &&
(usecase->id != USECASE_AUDIO_PLAYBACK_MULTI_CH) &&
(usecase->id != USECASE_AUDIO_PLAYBACK_ULL) &&
(usecase->id != USECASE_AUDIO_PLAYBACK_VOIP) &&
(usecase->id != USECASE_AUDIO_TRANSCODE_LOOPBACK_RX) &&
(!is_interactive_usecase(usecase->id)) &&
(!is_offload_usecase(usecase->id)) &&
(usecase->type != PCM_CAPTURE) &&
(!audio_extn_auto_hal_is_bus_device_usecase(usecase->id))) {
ALOGV("%s: a rx/tx/loopback path where app type cfg is not required %d", __func__, usecase->id);
rc = 0;
goto exit_send_app_type_cfg;
}
//if VR is active then only send the mixer control
if (usecase->id == USECASE_AUDIO_PLAYBACK_ULL && !audio_is_vr_mode_on(adev)) {
ALOGI("ULL doesnt need sending app type cfg, returning");
rc = 0;
goto exit_send_app_type_cfg;
}
if (usecase->type == PCM_PLAYBACK || usecase->type == TRANSCODE_LOOPBACK_RX) {
pcm_device_id = platform_get_pcm_device_id(usecase->id, PCM_PLAYBACK);
snprintf(mixer_ctl_name, sizeof(mixer_ctl_name),
"Audio Stream %d App Type Cfg", pcm_device_id);
} else if (usecase->type == PCM_CAPTURE) {
pcm_device_id = platform_get_pcm_device_id(usecase->id, PCM_CAPTURE);
snprintf(mixer_ctl_name, sizeof(mixer_ctl_name),
"Audio Stream Capture %d App Type Cfg", pcm_device_id);
}
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);
rc = -EINVAL;
goto exit_send_app_type_cfg;
}
snd_device = platform_get_spkr_prot_snd_device(snd_device);
if (voice_is_in_call_rec_stream(usecase->stream.in) && usecase->type == PCM_CAPTURE) {
snd_device_t voice_device = voice_get_incall_rec_snd_device(usecase->in_snd_device);
acdb_dev_id = platform_get_snd_device_acdb_id(voice_device);
ALOGV("acdb id for voice call use case %d", acdb_dev_id);
} else {
acdb_dev_id = platform_get_snd_device_acdb_id(snd_device);
}
if (acdb_dev_id <= 0) {
ALOGE("%s: Couldn't get the acdb dev id", __func__);
rc = -EINVAL;
goto exit_send_app_type_cfg;
}
snd_device_be_idx = platform_get_snd_device_backend_index(snd_device);
if (snd_device_be_idx < 0) {
ALOGE("%s: Couldn't get the backend index for snd device %s ret=%d",
__func__, platform_get_snd_device_name(snd_device),
snd_device_be_idx);
}
sample_rate = audio_extn_utils_get_app_sample_rate_for_device(adev, usecase, snd_device);
if ((usecase->type == PCM_PLAYBACK) && (usecase->stream.out != NULL)) {
/* Interactive streams are supported with only direct app type id.
* Get Direct profile app type and use it for interactive streams
*/
list_for_each(node, &adev->streams_output_cfg_list) {
s_info = node_to_item(node, struct streams_io_cfg, list);
if (s_info->flags.out_flags == (audio_output_flags_t)(AUDIO_OUTPUT_FLAG_BD |
AUDIO_OUTPUT_FLAG_DIRECT_PCM |
AUDIO_OUTPUT_FLAG_DIRECT))
bd_app_type = s_info->app_type_cfg.app_type;
}
if (usecase->stream.out->flags == (audio_output_flags_t)AUDIO_OUTPUT_FLAG_INTERACTIVE)
app_type = bd_app_type;
else
app_type = usecase->stream.out->app_type_cfg.app_type;
app_type_cfg[len++] = app_type;
app_type_cfg[len++] = acdb_dev_id;
app_type_cfg[len++] = sample_rate;
if (snd_device_be_idx > 0)
app_type_cfg[len++] = snd_device_be_idx;
ALOGI("%s PLAYBACK app_type %d, acdb_dev_id %d, sample_rate %d, snd_device_be_idx %d",
__func__, app_type, acdb_dev_id, sample_rate, snd_device_be_idx);
} else if ((usecase->type == PCM_CAPTURE) && (usecase->stream.in != NULL)) {
app_type = usecase->stream.in->app_type_cfg.app_type;
app_type_cfg[len++] = app_type;
app_type_cfg[len++] = acdb_dev_id;
app_type_cfg[len++] = sample_rate;
if (snd_device_be_idx > 0)
app_type_cfg[len++] = snd_device_be_idx;
ALOGI("%s CAPTURE app_type %d, acdb_dev_id %d, sample_rate %d, snd_device_be_idx %d",
__func__, app_type, acdb_dev_id, sample_rate, snd_device_be_idx);
} else {
app_type = platform_get_default_app_type_v2(adev->platform, usecase->type);
if(usecase->type == TRANSCODE_LOOPBACK_RX) {
app_type = usecase->stream.inout->out_app_type_cfg.app_type;
}
app_type_cfg[len++] = app_type;
app_type_cfg[len++] = acdb_dev_id;
app_type_cfg[len++] = sample_rate;
if (snd_device_be_idx > 0)
app_type_cfg[len++] = snd_device_be_idx;
ALOGI("%s default app_type %d, acdb_dev_id %d, sample_rate %d, snd_device_be_idx %d",
__func__, app_type, acdb_dev_id, sample_rate, snd_device_be_idx);
}
if(ctl)
mixer_ctl_set_array(ctl, app_type_cfg, len);
rc = 0;
exit_send_app_type_cfg:
return rc;
}
static int audio_extn_utils_check_input_parameters(uint32_t sample_rate,
audio_format_t format,
int channel_count)
{
int ret = 0;
if (((format != AUDIO_FORMAT_PCM_16_BIT) && (format != AUDIO_FORMAT_PCM_8_24_BIT) &&
(format != AUDIO_FORMAT_PCM_24_BIT_PACKED) && (format != AUDIO_FORMAT_PCM_32_BIT) &&
(format != AUDIO_FORMAT_PCM_FLOAT)) &&
!voice_extn_compress_voip_is_format_supported(format) &&
!audio_extn_compr_cap_format_supported(format) &&
!audio_extn_cin_format_supported(format))
ret = -EINVAL;
switch (channel_count) {
case 1:
case 2:
case 3:
case 4:
case 6:
case 8:
break;
default:
ret = -EINVAL;
}
switch (sample_rate) {
case 8000:
case 11025:
case 12000:
case 16000:
case 22050:
case 24000:
case 32000:
case 44100:
case 48000:
case 88200:
case 96000:
case 176400:
case 192000:
break;
default:
ret = -EINVAL;
}
return ret;
}
static inline uint32_t audio_extn_utils_nearest_multiple(uint32_t num, uint32_t multiplier)
{
uint32_t remainder = 0;
if (!multiplier)
return num;
remainder = num % multiplier;
if (remainder)
num += (multiplier - remainder);
return num;
}
static inline uint32_t audio_extn_utils_lcm(uint32_t num1, uint32_t num2)
{
uint32_t high = num1, low = num2, temp = 0;
if (!num1 || !num2)
return 0;
if (num1 < num2) {
high = num2;
low = num1;
}
while (low != 0) {
temp = low;
low = high % low;
high = temp;
}
return (num1 * num2)/high;
}
int audio_extn_utils_send_app_type_cfg(struct audio_device *adev,
struct audio_usecase *usecase)
{
int i, num_devices = 0;
snd_device_t new_snd_devices[SND_DEVICE_OUT_END] = {0};
snd_device_t in_snd_device = usecase->in_snd_device;
int rc = 0;
if (usecase->type == PCM_HFP_CALL) {
return audio_extn_utils_send_app_type_cfg_hfp(adev, usecase);
}
switch (usecase->type) {
case PCM_PLAYBACK:
case TRANSCODE_LOOPBACK_RX:
ALOGD("%s: usecase->out_snd_device %s",
__func__, platform_get_snd_device_name(usecase->out_snd_device));
/* check for out combo device */
if (platform_split_snd_device(adev->platform,
usecase->out_snd_device,
&num_devices, new_snd_devices)) {
new_snd_devices[0] = usecase->out_snd_device;
num_devices = 1;
}
break;
case PCM_CAPTURE:
ALOGD("%s: usecase->in_snd_device %s",
__func__, platform_get_snd_device_name(usecase->in_snd_device));
if (voice_is_in_call_rec_stream(usecase->stream.in)) {
in_snd_device = voice_get_incall_rec_backend_device(usecase->stream.in);
}
/* check for in combo device */
if (platform_split_snd_device(adev->platform,
in_snd_device,
&num_devices, new_snd_devices)) {
new_snd_devices[0] = in_snd_device;
num_devices = 1;
}
break;
default:
ALOGI("%s: not a playback/capture path, no need to cfg app type", __func__);
rc = 0;
break;
}
for (i = 0; i < num_devices; i++) {
rc = send_app_type_cfg_for_device(adev, usecase, new_snd_devices[i]);
if (rc)
break;
}
return rc;
}
int read_line_from_file(const char *path, char *buf, size_t count)
{
char * fgets_ret;
FILE * fd;
int rv;
fd = fopen(path, "r");
if (fd == NULL)
return -1;
fgets_ret = fgets(buf, (int)count, fd);
if (NULL != fgets_ret) {
rv = (int)strlen(buf);
} else {
rv = ferror(fd);
}
fclose(fd);
return rv;
}
/*Translates ALSA formats to AOSP PCM formats*/
audio_format_t alsa_format_to_hal(uint32_t alsa_format)
{
audio_format_t format;
switch(alsa_format) {
case SNDRV_PCM_FORMAT_S16_LE:
format = AUDIO_FORMAT_PCM_16_BIT;
break;
case SNDRV_PCM_FORMAT_S24_3LE:
format = AUDIO_FORMAT_PCM_24_BIT_PACKED;
break;
case SNDRV_PCM_FORMAT_S24_LE:
format = AUDIO_FORMAT_PCM_8_24_BIT;
break;
case SNDRV_PCM_FORMAT_S32_LE:
format = AUDIO_FORMAT_PCM_32_BIT;
break;
default:
ALOGW("Incorrect ALSA format");
format = AUDIO_FORMAT_INVALID;
}
return format;
}
/*Translates hal format (AOSP) to alsa formats*/
uint32_t hal_format_to_alsa(audio_format_t hal_format)
{
uint32_t alsa_format;
switch (hal_format) {
case AUDIO_FORMAT_PCM_32_BIT: {
if (platform_supports_true_32bit())
alsa_format = SNDRV_PCM_FORMAT_S32_LE;
else
alsa_format = SNDRV_PCM_FORMAT_S24_3LE;
}
break;
case AUDIO_FORMAT_PCM_8_BIT:
alsa_format = SNDRV_PCM_FORMAT_S8;
break;
case AUDIO_FORMAT_PCM_24_BIT_PACKED:
alsa_format = SNDRV_PCM_FORMAT_S24_3LE;
break;
case AUDIO_FORMAT_PCM_8_24_BIT: {
if (platform_supports_true_32bit())
alsa_format = SNDRV_PCM_FORMAT_S32_LE;
else
alsa_format = SNDRV_PCM_FORMAT_S24_3LE;
}
break;
case AUDIO_FORMAT_PCM_FLOAT:
alsa_format = SNDRV_PCM_FORMAT_S24_3LE;
break;
default:
case AUDIO_FORMAT_PCM_16_BIT:
alsa_format = SNDRV_PCM_FORMAT_S16_LE;
break;
}
return alsa_format;
}
/*Translates PCM formats to AOSP formats*/
audio_format_t pcm_format_to_hal(uint32_t pcm_format)
{
audio_format_t format = AUDIO_FORMAT_INVALID;
switch(pcm_format) {
case PCM_FORMAT_S16_LE:
format = AUDIO_FORMAT_PCM_16_BIT;
break;
case PCM_FORMAT_S24_3LE:
format = AUDIO_FORMAT_PCM_24_BIT_PACKED;
break;
case PCM_FORMAT_S24_LE:
format = AUDIO_FORMAT_PCM_8_24_BIT;
break;
case PCM_FORMAT_S32_LE:
format = AUDIO_FORMAT_PCM_32_BIT;
break;
default:
ALOGW("Incorrect PCM format");
format = AUDIO_FORMAT_INVALID;
}
return format;
}
/*Translates hal format (AOSP) to alsa formats*/
uint32_t hal_format_to_pcm(audio_format_t hal_format)
{
uint32_t pcm_format;
switch (hal_format) {
case AUDIO_FORMAT_PCM_32_BIT:
case AUDIO_FORMAT_PCM_8_24_BIT:
case AUDIO_FORMAT_PCM_FLOAT: {
if (platform_supports_true_32bit())
pcm_format = PCM_FORMAT_S32_LE;
else
pcm_format = PCM_FORMAT_S24_3LE;
}
break;
case AUDIO_FORMAT_PCM_8_BIT:
pcm_format = PCM_FORMAT_S8;
break;
case AUDIO_FORMAT_PCM_24_BIT_PACKED:
pcm_format = PCM_FORMAT_S24_3LE;
break;
default:
case AUDIO_FORMAT_PCM_16_BIT:
pcm_format = PCM_FORMAT_S16_LE;
break;
}
return pcm_format;
}
uint32_t get_alsa_fragment_size(uint32_t bytes_per_sample,
uint32_t sample_rate,
uint32_t noOfChannels,
int64_t duration_ms)
{
uint32_t fragment_size = 0;
uint32_t pcm_offload_time = PCM_OFFLOAD_BUFFER_DURATION;
if (duration_ms >= MIN_OFFLOAD_BUFFER_DURATION_MS && duration_ms <= MAX_OFFLOAD_BUFFER_DURATION_MS)
pcm_offload_time = duration_ms;
fragment_size = (pcm_offload_time
* sample_rate
* bytes_per_sample
* noOfChannels)/1000;
if (fragment_size < MIN_PCM_OFFLOAD_FRAGMENT_SIZE)
fragment_size = MIN_PCM_OFFLOAD_FRAGMENT_SIZE;
else if (fragment_size > MAX_PCM_OFFLOAD_FRAGMENT_SIZE)
fragment_size = MAX_PCM_OFFLOAD_FRAGMENT_SIZE;
/*To have same PCM samples for all channels, the buffer size requires to
*be multiple of (number of channels * bytes per sample)
*For writes to succeed, the buffer must be written at address which is multiple of 32
*/
fragment_size = ALIGN(fragment_size, (bytes_per_sample * noOfChannels * 32));
ALOGI("PCM offload Fragment size to %d bytes", fragment_size);
return fragment_size;
}
/* Calculates the fragment size required to configure compress session.
* Based on the alsa format selected, decide if conversion is needed in
* HAL ( e.g. convert AUDIO_FORMAT_PCM_FLOAT input format to
* AUDIO_FORMAT_PCM_24_BIT_PACKED before writing to the compress driver.
*/
void audio_extn_utils_update_direct_pcm_fragment_size(struct stream_out *out)
{
audio_format_t dst_format = out->hal_op_format;
audio_format_t src_format = out->hal_ip_format;
uint32_t hal_op_bytes_per_sample = audio_bytes_per_sample(dst_format);
uint32_t hal_ip_bytes_per_sample = audio_bytes_per_sample(src_format);
out->compr_config.fragment_size =
get_alsa_fragment_size(hal_op_bytes_per_sample,
out->sample_rate,
popcount(out->channel_mask),
out->info.duration_us / 1000);
if ((src_format != dst_format) &&
hal_op_bytes_per_sample != hal_ip_bytes_per_sample) {
out->hal_fragment_size =
((out->compr_config.fragment_size * hal_ip_bytes_per_sample) /
hal_op_bytes_per_sample);
ALOGI("enable conversion hal_input_fragment_size is %d src_format %x dst_format %x",
out->hal_fragment_size, src_format, dst_format);
} else {
out->hal_fragment_size = out->compr_config.fragment_size;
}
}
/* converts pcm format 24_8 to 8_24 inplace */
size_t audio_extn_utils_convert_format_24_8_to_8_24(void *buf, size_t bytes)
{
size_t i = 0;
int *int_buf_stream = buf;
if ((bytes % 4) != 0) {
ALOGE("%s: wrong inout buffer! ... is not 32 bit aligned ", __func__);
return -EINVAL;
}
for (; i < (bytes / 4); i++)
int_buf_stream[i] >>= 8;
return bytes;
}
#ifdef AUDIO_GKI_ENABLED
int get_snd_codec_id(audio_format_t format)
{
int id = 0;
switch (format & AUDIO_FORMAT_MAIN_MASK) {
case AUDIO_FORMAT_MP3:
id = SND_AUDIOCODEC_MP3;
break;
case AUDIO_FORMAT_AAC:
id = SND_AUDIOCODEC_AAC;
break;
case AUDIO_FORMAT_AAC_ADTS:
id = SND_AUDIOCODEC_AAC;
break;
case AUDIO_FORMAT_AAC_LATM:
id = SND_AUDIOCODEC_AAC;
break;
case AUDIO_FORMAT_PCM:
id = SND_AUDIOCODEC_PCM;
break;
case AUDIO_FORMAT_FLAC:
case AUDIO_FORMAT_ALAC:
case AUDIO_FORMAT_APE:
case AUDIO_FORMAT_VORBIS:
case AUDIO_FORMAT_WMA:
case AUDIO_FORMAT_WMA_PRO:
case AUDIO_FORMAT_DSD:
case AUDIO_FORMAT_APTX:
id = SND_AUDIOCODEC_BESPOKE;
break;
case AUDIO_FORMAT_MP2:
id = SND_AUDIOCODEC_MP2;
break;
case AUDIO_FORMAT_AC3:
id = SND_AUDIOCODEC_AC3;
break;
case AUDIO_FORMAT_E_AC3:
case AUDIO_FORMAT_E_AC3_JOC:
id = SND_AUDIOCODEC_EAC3;
break;
case AUDIO_FORMAT_DTS:
case AUDIO_FORMAT_DTS_HD:
id = SND_AUDIOCODEC_DTS;
break;
case AUDIO_FORMAT_DOLBY_TRUEHD:
id = SND_AUDIOCODEC_TRUEHD;
break;
case AUDIO_FORMAT_IEC61937:
id = SND_AUDIOCODEC_IEC61937;
break;
default:
ALOGE("%s: Unsupported audio format :%x", __func__, format);
}
return id;
}
#else
int get_snd_codec_id(audio_format_t format)
{
int id = 0;
switch (format & AUDIO_FORMAT_MAIN_MASK) {
case AUDIO_FORMAT_MP3:
id = SND_AUDIOCODEC_MP3;
break;
case AUDIO_FORMAT_AAC:
id = SND_AUDIOCODEC_AAC;
break;
case AUDIO_FORMAT_AAC_ADTS:
id = SND_AUDIOCODEC_AAC;
break;
case AUDIO_FORMAT_AAC_LATM:
id = SND_AUDIOCODEC_AAC;
break;
case AUDIO_FORMAT_PCM:
id = SND_AUDIOCODEC_PCM;
break;
case AUDIO_FORMAT_FLAC:
id = SND_AUDIOCODEC_FLAC;
break;
case AUDIO_FORMAT_ALAC:
id = SND_AUDIOCODEC_ALAC;
break;
case AUDIO_FORMAT_APE:
id = SND_AUDIOCODEC_APE;
break;
case AUDIO_FORMAT_VORBIS:
id = SND_AUDIOCODEC_VORBIS;
break;
case AUDIO_FORMAT_WMA:
id = SND_AUDIOCODEC_WMA;
break;
case AUDIO_FORMAT_WMA_PRO:
id = SND_AUDIOCODEC_WMA_PRO;
break;
case AUDIO_FORMAT_MP2:
id = SND_AUDIOCODEC_MP2;
break;
case AUDIO_FORMAT_AC3:
id = SND_AUDIOCODEC_AC3;
break;
case AUDIO_FORMAT_E_AC3:
case AUDIO_FORMAT_E_AC3_JOC:
id = SND_AUDIOCODEC_EAC3;
break;
case AUDIO_FORMAT_DTS:
case AUDIO_FORMAT_DTS_HD:
id = SND_AUDIOCODEC_DTS;
break;
case AUDIO_FORMAT_DOLBY_TRUEHD:
id = SND_AUDIOCODEC_TRUEHD;
break;
case AUDIO_FORMAT_IEC61937:
id = SND_AUDIOCODEC_IEC61937;
break;
case AUDIO_FORMAT_DSD:
id = SND_AUDIOCODEC_DSD;
break;
case AUDIO_FORMAT_APTX:
id = SND_AUDIOCODEC_APTX;
break;
default:
ALOGE("%s: Unsupported audio format :%x", __func__, format);
}
return id;
}
#endif
void audio_extn_utils_send_audio_calibration(struct audio_device *adev,
struct audio_usecase *usecase)
{
int type = usecase->type;
if (type == PCM_PLAYBACK && usecase->stream.out != NULL) {
platform_send_audio_calibration(adev->platform, usecase,
usecase->stream.out->app_type_cfg.app_type);
} else if (type == PCM_CAPTURE && usecase->stream.in != NULL) {
platform_send_audio_calibration(adev->platform, usecase,
usecase->stream.in->app_type_cfg.app_type);
} else if ((type == PCM_HFP_CALL) || (type == PCM_CAPTURE) ||
(type == TRANSCODE_LOOPBACK_RX && usecase->stream.inout != NULL)) {
platform_send_audio_calibration(adev->platform, usecase,
platform_get_default_app_type_v2(adev->platform, usecase->type));
} else {
/* No need to send audio calibration for voice and voip call usecases */
if ((type != VOICE_CALL) && (type != VOIP_CALL))
ALOGW("%s: No audio calibration for usecase type = %d", __func__, type);
}
}
// Base64 Encode and Decode
// Not all features supported. This must be used only with following conditions.
// Decode Modes: Support with and without padding
// CRLF not handling. So no CRLF in string to decode.
// Encode Modes: Supports only padding
int b64decode(char *inp, int ilen, uint8_t* outp)
{
int i, j, k, ii, num;
int rem, pcnt;
uint32_t res=0;
uint8_t getIndex[MAX_BASEINDEX_LEN];
uint8_t tmp, cflag;
if(inp == NULL || outp == NULL || ilen <= 0) {
ALOGE("[%s] received NULL pointer or zero length",__func__);
return -1;
}
memset(getIndex, MAX_BASEINDEX_LEN-1, sizeof(getIndex));
for(i=0;i<BASE_TABLE_SIZE;i++) {
getIndex[(uint8_t)bTable[i]] = (uint8_t)i;
}
getIndex[(uint8_t)'=']=0;
j=0;k=0;
num = ilen/4;
rem = ilen%4;
if(rem==0)
num = num-1;
cflag=0;
for(i=0; i<num; i++) {
res=0;
for(ii=0;ii<4;ii++) {
res = res << 6;
tmp = getIndex[(uint8_t)inp[j++]];
res = res | tmp;
cflag = cflag | tmp;
}
outp[k++] = (res >> 16)&0xFF;
outp[k++] = (res >> 8)&0xFF;
outp[k++] = res & 0xFF;
}
// Handle last bytes special
pcnt=0;
if(rem == 0) {
//With padding or full data
res = 0;
for(ii=0;ii<4;ii++) {
if(inp[j] == '=')
pcnt++;
res = res << 6;
tmp = getIndex[(uint8_t)inp[j++]];
res = res | tmp;
cflag = cflag | tmp;
}
outp[k++] = res >> 16;
if(pcnt == 2)
goto done;
outp[k++] = (res>>8)&0xFF;
if(pcnt == 1)
goto done;
outp[k++] = res&0xFF;
} else {
//without padding
res = 0;
for(i=0;i<rem;i++) {
res = res << 6;
tmp = getIndex[(uint8_t)inp[j++]];
res = res | tmp;
cflag = cflag | tmp;
}
for(i=rem;i<4;i++) {
res = res << 6;
pcnt++;
}
outp[k++] = res >> 16;
if(pcnt == 2)
goto done;
outp[k++] = (res>>8)&0xFF;
if(pcnt == 1)
goto done;
outp[k++] = res&0xFF;
}
done:
if(cflag == 0xFF) {
ALOGE("[%s] base64 decode failed. Invalid character found %s",
__func__, inp);
return 0;
}
return k;
}
int b64encode(uint8_t *inp, int ilen, char* outp)
{
int i,j,k, num;
int rem=0;
uint32_t res=0;
if(inp == NULL || outp == NULL || ilen<=0) {
ALOGE("[%s] received NULL pointer or zero input length",__func__);
return -1;
}
num = ilen/3;
rem = ilen%3;
j=0;k=0;
for(i=0; i<num; i++) {
//prepare index
res = inp[j++]<<16;
res = res | inp[j++]<<8;
res = res | inp[j++];
//get output map from index
outp[k++] = (char) bTable[(res>>18)&0x3F];
outp[k++] = (char) bTable[(res>>12)&0x3F];
outp[k++] = (char) bTable[(res>>6)&0x3F];
outp[k++] = (char) bTable[res&0x3F];
}
switch(rem) {
case 1:
res = inp[j++]<<16;
outp[k++] = (char) bTable[res>>18];
outp[k++] = (char) bTable[(res>>12)&0x3F];
//outp[k++] = '=';
//outp[k++] = '=';
break;
case 2:
res = inp[j++]<<16;
res = res | inp[j++]<<8;
outp[k++] = (char) bTable[res>>18];
outp[k++] = (char) bTable[(res>>12)&0x3F];
outp[k++] = (char) bTable[(res>>6)&0x3F];
//outp[k++] = '=';
break;
default:
break;
}
outp[k] = '\0';
return k;
}
int audio_extn_utils_get_codec_version(const char *snd_card_name,
int card_num,
char *codec_version)
{
char procfs_path[50];
FILE *fp;
if (strstr(snd_card_name, "tasha")) {
snprintf(procfs_path, sizeof(procfs_path),
"/proc/asound/card%d/codecs/tasha/version", card_num);
if ((fp = fopen(procfs_path, "r")) != NULL) {
fgets(codec_version, CODEC_VERSION_MAX_LENGTH, fp);
fclose(fp);
} else {
ALOGE("%s: ERROR. cannot open %s", __func__, procfs_path);
return -ENOENT;
}
ALOGD("%s: codec version %s", __func__, codec_version);
}
return 0;
}
int audio_extn_utils_get_codec_variant(int card_num,
char *codec_variant)
{
char procfs_path[50];
FILE *fp;
snprintf(procfs_path, sizeof(procfs_path),
"/proc/asound/card%d/codecs/wcd938x/variant", card_num);
if ((fp = fopen(procfs_path, "r")) == NULL) {
snprintf(procfs_path, sizeof(procfs_path),
"/proc/asound/card%d/codecs/wcd937x/variant", card_num);
if ((fp = fopen(procfs_path, "r")) == NULL) {
ALOGE("%s: ERROR. cannot open %s", __func__, procfs_path);
return -ENOENT;
}
}
fgets(codec_variant, CODEC_VARIANT_MAX_LENGTH, fp);
fclose(fp);
ALOGD("%s: codec variant is %s", __func__, codec_variant);
return 0;
}
#ifdef AUDIO_EXTERNAL_HDMI_ENABLED
void get_default_compressed_channel_status(
unsigned char *channel_status)
{
memset(channel_status,0,24);
/* block start bit in preamble bit 3 */
channel_status[0] |= PROFESSIONAL;
//compre out
channel_status[0] |= NON_LPCM;
// sample rate; fixed 48K for default/transcode
channel_status[3] |= SR_48000;
}
int32_t get_compressed_channel_status(void *audio_stream_data,
uint32_t audio_frame_size,
unsigned char *channel_status,
enum audio_parser_code_type codec_type)
// codec_type - AUDIO_PARSER_CODEC_AC3
// - AUDIO_PARSER_CODEC_DTS
{
unsigned char *stream;
int ret = 0;
stream = (unsigned char *)audio_stream_data;
if (audio_stream_data == NULL || audio_frame_size == 0) {
ALOGW("no buffer to get channel status, return default for compress");
get_default_compressed_channel_status(channel_status);
return ret;
}
memset(channel_status,0,24);
if(init_audio_parser(stream, audio_frame_size, codec_type) == -1)
{
ALOGE("init audio parser failed");
return -1;
}
ret = get_channel_status(channel_status, codec_type);
return ret;
}
void get_lpcm_channel_status(uint32_t sampleRate,
unsigned char *channel_status)
{
int32_t status = 0;
memset(channel_status,0,24);
/* block start bit in preamble bit 3 */
channel_status[0] |= PROFESSIONAL;
//LPCM OUT
channel_status[0] &= ~NON_LPCM;
switch (sampleRate) {
case 8000:
case 11025:
case 12000:
case 16000:
case 22050:
channel_status[3] |= SR_NOTID;
break;
case 24000:
channel_status[3] |= SR_24000;
break;
case 32000:
channel_status[3] |= SR_32000;
break;
case 44100:
channel_status[3] |= SR_44100;
break;
case 48000:
channel_status[3] |= SR_48000;
break;
case 88200:
channel_status[3] |= SR_88200;
break;
case 96000:
channel_status[3] |= SR_96000;
break;
case 176400:
channel_status[3] |= SR_176400;
break;
case 192000:
channel_status[3] |= SR_192000;
break;
default:
ALOGV("Invalid sample_rate %u\n", sampleRate);
status = -1;
break;
}
}
void audio_utils_set_hdmi_channel_status(struct stream_out *out, char * buffer, size_t bytes)
{
unsigned char channel_status[24]={0};
struct snd_aes_iec958 iec958;
const char *mixer_ctl_name = "IEC958 Playback PCM Stream";
struct mixer_ctl *ctl;
ALOGV("%s: buffer %s bytes %zd", __func__, buffer, bytes);
if (audio_extn_utils_is_dolby_format(out->format) &&
/*TODO:Extend code to support DTS passthrough*/
/*set compressed channel status bits*/
audio_extn_passthru_is_passthrough_stream(out) &&
audio_extn_is_hdmi_passthru_enabled()) {
get_compressed_channel_status(buffer, bytes, channel_status, AUDIO_PARSER_CODEC_AC3);
} else {
/*set channel status bit for LPCM*/
get_lpcm_channel_status(out->sample_rate, channel_status);
}
memcpy(iec958.status, channel_status,sizeof(iec958.status));
ctl = mixer_get_ctl_by_name(out->dev->mixer, mixer_ctl_name);
if (!ctl) {
ALOGE("%s: Could not get ctl for mixer cmd - %s",
__func__, mixer_ctl_name);
return;
}
if (mixer_ctl_set_array(ctl, &iec958, sizeof(iec958)) < 0) {
ALOGE("%s: Could not set channel status for ext HDMI ",
__func__);
return;
}
}
#endif
int audio_extn_utils_get_avt_device_drift(
struct audio_usecase *usecase,
struct audio_avt_device_drift_param *drift_param)
{
int ret = 0, count = 0;
char avt_device_drift_mixer_ctl_name[MIXER_PATH_MAX_LENGTH] = {0};
const char *backend = NULL;
struct mixer_ctl *ctl = NULL;
struct audio_avt_device_drift_stats drift_stats;
struct audio_device *adev = NULL;
if (usecase != NULL && usecase->type == PCM_PLAYBACK) {
backend = platform_get_snd_device_backend_interface(usecase->out_snd_device);
if (!backend) {
ALOGE("%s: Unsupported device %d", __func__,
get_device_types(&usecase->stream.out->device_list));
ret = -EINVAL;
goto done;
}
strlcpy(avt_device_drift_mixer_ctl_name,
backend,
MIXER_PATH_MAX_LENGTH);
count = strlen(backend);
if (MIXER_PATH_MAX_LENGTH - count > 0) {
strlcat(&avt_device_drift_mixer_ctl_name[count],
" DRIFT",
MIXER_PATH_MAX_LENGTH - count);
} else {
ret = -EINVAL;
goto done;
}
} else {
ALOGE("%s: Invalid usecase",__func__);
ret = -EINVAL;
goto done;
}
adev = usecase->stream.out->dev;
ctl = mixer_get_ctl_by_name(adev->mixer, avt_device_drift_mixer_ctl_name);
if (!ctl) {
ALOGE("%s: Could not get ctl for mixer cmd - %s",
__func__, avt_device_drift_mixer_ctl_name);
ret = -EINVAL;
goto done;
}
ALOGV("%s: Getting AV Timer vs Device Drift mixer ctrl name %s", __func__,
avt_device_drift_mixer_ctl_name);
mixer_ctl_update(ctl);
count = mixer_ctl_get_num_values(ctl);
if (count != sizeof(struct audio_avt_device_drift_stats)) {
ALOGE("%s: mixer_ctl_get_num_values() invalid drift_stats data size",
__func__);
ret = -EINVAL;
goto done;
}
ret = mixer_ctl_get_array(ctl, (void *)&drift_stats, count);
if (ret != 0) {
ALOGE("%s: mixer_ctl_get_array() failed to get drift_stats Params",
__func__);
ret = -EINVAL;
goto done;
}
memcpy(drift_param, &drift_stats.drift_param,
sizeof(struct audio_avt_device_drift_param));
done:
return ret;
}
#ifdef SNDRV_COMPRESS_PATH_DELAY
int audio_extn_utils_compress_get_dsp_latency(struct stream_out *out)
{
int ret = -EINVAL;
struct snd_compr_metadata metadata;
int delay_ms = COMPRESS_OFFLOAD_PLAYBACK_LATENCY;
/* override the latency for pcm offload use case */
if ((out->flags & AUDIO_OUTPUT_FLAG_DIRECT) &&
!(out->flags & AUDIO_OUTPUT_FLAG_COMPRESS_OFFLOAD)) {
delay_ms = PCM_OFFLOAD_PLAYBACK_LATENCY;
}
if (property_get_bool("vendor.audio.playback.dsp.pathdelay", false)) {
ALOGD("%s:: Quering DSP delay %d",__func__, __LINE__);
if (!(is_offload_usecase(out->usecase))) {
ALOGE("%s:: not supported for non offload session", __func__);
goto exit;
}
if (!out->compr) {
ALOGD("%s:: Invalid compress handle,returning default dsp latency",
__func__);
goto exit;
}
metadata.key = SNDRV_COMPRESS_PATH_DELAY;
ret = compress_get_metadata(out->compr, &metadata);
if(ret) {
ALOGE("%s::error %s", __func__, compress_get_error(out->compr));
goto exit;
}
delay_ms = metadata.value[0] / 1000; /*convert to ms*/
} else {
ALOGD("%s:: Using Fix DSP delay",__func__);
}
exit:
ALOGD("%s:: delay in ms is %d",__func__, delay_ms);
return delay_ms;
}
#else
int audio_extn_utils_compress_get_dsp_latency(struct stream_out *out __unused)
{
return COMPRESS_OFFLOAD_PLAYBACK_LATENCY;
}
#endif
#ifdef SNDRV_COMPRESS_RENDER_MODE
int audio_extn_utils_compress_set_render_mode(struct stream_out *out)
{
struct snd_compr_metadata metadata;
int ret = -EINVAL;
if (!(is_offload_usecase(out->usecase))) {
ALOGE("%s:: not supported for non offload session", __func__);
goto exit;
}
if (!out->compr) {
ALOGD("%s:: Invalid compress handle",
__func__);
goto exit;
}
ALOGD("%s:: render mode %d", __func__, out->render_mode);
metadata.key = SNDRV_COMPRESS_RENDER_MODE;
if (out->render_mode == RENDER_MODE_AUDIO_MASTER) {
metadata.value[0] = SNDRV_COMPRESS_RENDER_MODE_AUDIO_MASTER;
} else if (out->render_mode == RENDER_MODE_AUDIO_STC_MASTER) {
metadata.value[0] = SNDRV_COMPRESS_RENDER_MODE_STC_MASTER;
} else {
ret = 0;
goto exit;
}
ret = compress_set_metadata(out->compr, &metadata);
if(ret) {
ALOGE("%s::error %s", __func__, compress_get_error(out->compr));
}
exit:
return ret;
}
#else
int audio_extn_utils_compress_set_render_mode(struct stream_out *out __unused)
{
ALOGD("%s:: configuring render mode not supported", __func__);
return 0;
}
#endif
#ifdef SNDRV_COMPRESS_CLK_REC_MODE
int audio_extn_utils_compress_set_clk_rec_mode(
struct audio_usecase *usecase)
{
struct snd_compr_metadata metadata;
struct stream_out *out = NULL;
int ret = -EINVAL;
if (usecase == NULL || usecase->type != PCM_PLAYBACK) {
ALOGE("%s:: Invalid use case", __func__);
goto exit;
}
out = usecase->stream.out;
if (!out) {
ALOGE("%s:: invalid stream", __func__);
goto exit;
}
if (!is_offload_usecase(out->usecase)) {
ALOGE("%s:: not supported for non offload session", __func__);
goto exit;
}
if (out->render_mode != RENDER_MODE_AUDIO_STC_MASTER) {
ALOGD("%s:: clk recovery is only supported in STC render mode",
__func__);
ret = 0;
goto exit;
}
if (!out->compr) {
ALOGD("%s:: Invalid compress handle",
__func__);
goto exit;
}
metadata.key = SNDRV_COMPRESS_CLK_REC_MODE;
switch(usecase->out_snd_device) {
case SND_DEVICE_OUT_HDMI:
case SND_DEVICE_OUT_SPEAKER_AND_HDMI:
case SND_DEVICE_OUT_DISPLAY_PORT:
case SND_DEVICE_OUT_SPEAKER_AND_DISPLAY_PORT:
metadata.value[0] = SNDRV_COMPRESS_CLK_REC_MODE_NONE;
break;
default:
metadata.value[0] = SNDRV_COMPRESS_CLK_REC_MODE_AUTO;
break;
}
ALOGD("%s:: clk recovery mode %d",__func__, metadata.value[0]);
ret = compress_set_metadata(out->compr, &metadata);
if(ret) {
ALOGE("%s::error %s", __func__, compress_get_error(out->compr));
}
exit:
return ret;
}
#else
int audio_extn_utils_compress_set_clk_rec_mode(
struct audio_usecase *usecase __unused)
{
ALOGD("%s:: configuring render mode not supported", __func__);
return 0;
}
#endif
#ifdef SNDRV_COMPRESS_RENDER_WINDOW
int audio_extn_utils_compress_set_render_window(
struct stream_out *out,
struct audio_out_render_window_param *render_window)
{
struct snd_compr_metadata metadata;
int ret = -EINVAL;
if(render_window == NULL) {
ALOGE("%s:: Invalid render_window", __func__);
goto exit;
}
ALOGD("%s:: render window start 0x%"PRIx64" end 0x%"PRIx64"",
__func__,render_window->render_ws, render_window->render_we);
if (!is_offload_usecase(out->usecase)) {
ALOGE("%s:: not supported for non offload session", __func__);
goto exit;
}
if ((out->render_mode != RENDER_MODE_AUDIO_MASTER) &&
(out->render_mode != RENDER_MODE_AUDIO_STC_MASTER)) {
ALOGD("%s:: only supported in timestamp mode, current "
"render mode mode %d", __func__, out->render_mode);
goto exit;
}
if (!out->compr) {
ALOGW("%s:: offload session not yet opened,"
"render window will be configure later", __func__);
/* store render window to reconfigure in start_output_stream() */
goto exit;
}
metadata.key = SNDRV_COMPRESS_RENDER_WINDOW;
/*render window start value */
metadata.value[0] = 0xFFFFFFFF & render_window->render_ws; /* lsb */
metadata.value[1] = \
(0xFFFFFFFF00000000 & render_window->render_ws) >> 32; /* msb*/
/*render window end value */
metadata.value[2] = 0xFFFFFFFF & render_window->render_we; /* lsb */
metadata.value[3] = \
(0xFFFFFFFF00000000 & render_window->render_we) >> 32; /* msb*/
ret = compress_set_metadata(out->compr, &metadata);
if(ret) {
ALOGE("%s::error %s", __func__, compress_get_error(out->compr));
}
exit:
return ret;
}
#else
int audio_extn_utils_compress_set_render_window(
struct stream_out *out __unused,
struct audio_out_render_window_param *render_window __unused)
{
ALOGD("%s:: configuring render window not supported", __func__);
return 0;
}
#endif
#ifdef SNDRV_COMPRESS_START_DELAY
int audio_extn_utils_compress_set_start_delay(
struct stream_out *out,
struct audio_out_start_delay_param *delay_param)
{
struct snd_compr_metadata metadata;
int ret = -EINVAL;
if(delay_param == NULL) {
ALOGE("%s:: Invalid delay_param", __func__);
goto exit;
}
ALOGD("%s:: render start delay 0x%"PRIx64" ", __func__,
delay_param->start_delay);
if (!is_offload_usecase(out->usecase)) {
ALOGE("%s:: not supported for non offload session", __func__);
goto exit;
}
if ((out->render_mode != RENDER_MODE_AUDIO_MASTER) &&
(out->render_mode != RENDER_MODE_AUDIO_STC_MASTER)) {
ALOGD("%s:: only supported in timestamp mode, current "
"render mode mode %d", __func__, out->render_mode);
goto exit;
}
if (!out->compr) {
ALOGW("%s:: offload session not yet opened,"
"start delay will be configure later", __func__);
goto exit;
}
metadata.key = SNDRV_COMPRESS_START_DELAY;
metadata.value[0] = 0xFFFFFFFF & delay_param->start_delay; /* lsb */
metadata.value[1] = \
(0xFFFFFFFF00000000 & delay_param->start_delay) >> 32; /* msb*/
ret = compress_set_metadata(out->compr, &metadata);
if(ret) {
ALOGE("%s::error %s", __func__, compress_get_error(out->compr));
}
exit:
return ret;
}
#else
int audio_extn_utils_compress_set_start_delay(
struct stream_out *out __unused,
struct audio_out_start_delay_param *delay_param __unused)
{
ALOGD("%s:: configuring render window not supported", __func__);
return 0;
}
#endif
#ifdef SNDRV_COMPRESS_DSP_POSITION
int audio_extn_utils_compress_get_dsp_presentation_pos(struct stream_out *out,
uint64_t *frames, struct timespec *timestamp, int32_t clock_id)
{
int ret = -EINVAL;
uint64_t *val = NULL;
uint64_t time = 0;
struct snd_compr_metadata metadata;
ALOGV("%s:: Quering DSP position with clock id %d",__func__, clock_id);
metadata.key = SNDRV_COMPRESS_DSP_POSITION;
metadata.value[0] = clock_id;
ret = compress_get_metadata(out->compr, &metadata);
if (ret) {
ALOGE("%s::error %s", __func__, compress_get_error(out->compr));
ret = -errno;
goto exit;
}
val = (uint64_t *)&metadata.value[1];
*frames = *val;
time = *(val + 1);
timestamp->tv_sec = time / 1000000;
timestamp->tv_nsec = (time % 1000000)*1000;
exit:
return ret;
}
#else
int audio_extn_utils_compress_get_dsp_presentation_pos(struct stream_out *out __unused,
uint64_t *frames __unused, struct timespec *timestamp __unused,
int32_t clock_id __unused)
{
ALOGD("%s:: dsp presentation position not supported", __func__);
return 0;
}
#endif
#ifdef SNDRV_PCM_IOCTL_DSP_POSITION
int audio_extn_utils_pcm_get_dsp_presentation_pos(struct stream_out *out,
uint64_t *frames, struct timespec *timestamp, int32_t clock_id)
{
int ret = -EINVAL;
uint64_t time = 0;
struct snd_pcm_prsnt_position prsnt_position;
ALOGV("%s:: Quering DSP position with clock id %d",__func__, clock_id);
prsnt_position.clock_id = clock_id;
ret = pcm_ioctl(out->pcm, SNDRV_PCM_IOCTL_DSP_POSITION, &prsnt_position);
if (ret) {
ALOGE("%s::error %d", __func__, ret);
ret = -EIO;
goto exit;
}
*frames = prsnt_position.frames;
time = prsnt_position.timestamp;
timestamp->tv_sec = time / 1000000;
timestamp->tv_nsec = (time % 1000000)*1000;
exit:
return ret;
}
#else
int audio_extn_utils_pcm_get_dsp_presentation_pos(struct stream_out *out __unused,
uint64_t *frames __unused, struct timespec *timestamp __unused,
int32_t clock_id __unused)
{
ALOGD("%s:: dsp presentation position not supported", __func__);
return 0;
}
#endif
#define MAX_SND_CARD 8
#define RETRY_US 1000000
#define RETRY_NUMBER 40
#define PLATFORM_INFO_XML_PATH "audio_platform_info.xml"
#define PLATFORM_INFO_XML_BASE_STRING "audio_platform_info"
bool audio_extn_utils_resolve_config_file(char file_name[MIXER_PATH_MAX_LENGTH])
{
char full_config_path[MIXER_PATH_MAX_LENGTH];
char vendor_config_path[VENDOR_CONFIG_PATH_MAX_LENGTH];
/* Get path for audio configuration files in vendor */
audio_get_vendor_config_path(vendor_config_path,
sizeof(vendor_config_path));
snprintf(full_config_path, sizeof(full_config_path),
"%s/%s", vendor_config_path, file_name);
if (F_OK == access(full_config_path, 0)) {
strlcpy(file_name, full_config_path, MIXER_PATH_MAX_LENGTH);
return true;
}
return false;
}
/* platform_info_file should be size 'MIXER_PATH_MAX_LENGTH' */
int audio_extn_utils_get_platform_info(const char* snd_card_name, char* platform_info_file)
{
if (NULL == snd_card_name) {
return -1;
}
struct snd_card_split *snd_split_handle = NULL;
int ret = 0;
audio_extn_set_snd_card_split(snd_card_name);
snd_split_handle = audio_extn_get_snd_card_split();
snprintf(platform_info_file, MIXER_PATH_MAX_LENGTH, "%s_%s_%s.xml",
PLATFORM_INFO_XML_BASE_STRING, snd_split_handle->snd_card,
snd_split_handle->form_factor);
if (!audio_extn_utils_resolve_config_file(platform_info_file)) {
memset(platform_info_file, 0, MIXER_PATH_MAX_LENGTH);
snprintf(platform_info_file, MIXER_PATH_MAX_LENGTH, "%s_%s.xml",
PLATFORM_INFO_XML_BASE_STRING, snd_split_handle->snd_card);
if (!audio_extn_utils_resolve_config_file(platform_info_file)) {
memset(platform_info_file, 0, MIXER_PATH_MAX_LENGTH);
strlcpy(platform_info_file, PLATFORM_INFO_XML_PATH, MIXER_PATH_MAX_LENGTH);
ret = audio_extn_utils_resolve_config_file(platform_info_file) ? 0 : -1;
}
}
return ret;
}
int audio_extn_utils_get_snd_card_num()
{
int snd_card_num = 0;
struct mixer *mixer = NULL;
snd_card_num = audio_extn_utils_open_snd_mixer(&mixer);
if (mixer)
mixer_close(mixer);
return snd_card_num;
}
int audio_extn_utils_open_snd_mixer(struct mixer **mixer_handle)
{
void *hw_info = NULL;
struct mixer *mixer = NULL;
int retry_num = 0;
int snd_card_num = 0;
char* snd_card_name = NULL;
int snd_card_detection_info[MAX_SND_CARD] = {0};
if (!mixer_handle) {
ALOGE("invalid mixer handle");
return -1;
}
*mixer_handle = NULL;
/*
* Try with all the sound cards ( 0 to 7 ) and if none of them were detected
* sleep for 1 sec and try detections with sound card 0 again.
* If sound card gets detected, check if it is relevant, if not check with the
* other sound cards. To ensure that the irrelevant sound card is not check again,
* we maintain it in min_snd_card_num.
*/
while (retry_num < RETRY_NUMBER) {
snd_card_num = 0;
while (snd_card_num < MAX_SND_CARD) {
if (snd_card_detection_info[snd_card_num] == 0) {
mixer = mixer_open(snd_card_num);
if (!mixer)
snd_card_num++;
else
break;
} else
snd_card_num++;
}
if (!mixer) {
usleep(RETRY_US);
retry_num++;
ALOGD("%s: retry, retry_num %d", __func__, retry_num);
continue;
}
snd_card_name = strdup(mixer_get_name(mixer));
if (!snd_card_name) {
ALOGE("failed to allocate memory for snd_card_name\n");
mixer_close(mixer);
return -1;
}
ALOGD("%s: snd_card_name: %s", __func__, snd_card_name);
snd_card_detection_info[snd_card_num] = 1;
hw_info = hw_info_init(snd_card_name);
if (hw_info) {
ALOGD("%s: Opened sound card:%d", __func__, snd_card_num);
break;
}
ALOGE("%s: Failed to init hardware info, snd_card_num:%d", __func__, snd_card_num);
free(snd_card_name);
snd_card_name = NULL;
mixer_close(mixer);
mixer = NULL;
}
if (snd_card_name)
free(snd_card_name);
if (hw_info)
hw_info_deinit(hw_info);
if (retry_num >= RETRY_NUMBER) {
ALOGE("%s: Unable to find correct sound card, aborting.", __func__);
return -1;
}
if (mixer)
*mixer_handle = mixer;
return snd_card_num;
}
void audio_extn_utils_close_snd_mixer(struct mixer *mixer)
{
if (mixer)
mixer_close(mixer);
}
#ifdef SNDRV_COMPRESS_ENABLE_ADJUST_SESSION_CLOCK
int audio_extn_utils_compress_enable_drift_correction(
struct stream_out *out,
struct audio_out_enable_drift_correction *drift)
{
struct snd_compr_metadata metadata;
int ret = -EINVAL;
if(drift == NULL) {
ALOGE("%s:: Invalid param", __func__);
goto exit;
}
ALOGD("%s:: drift enable %d", __func__,drift->enable);
if (!is_offload_usecase(out->usecase)) {
ALOGE("%s:: not supported for non offload session", __func__);
goto exit;
}
if (!out->compr) {
ALOGW("%s:: offload session not yet opened,"
"start delay will be configure later", __func__);
goto exit;
}
metadata.key = SNDRV_COMPRESS_ENABLE_ADJUST_SESSION_CLOCK;
metadata.value[0] = drift->enable;
out->drift_correction_enabled = drift->enable;
ret = compress_set_metadata(out->compr, &metadata);
if(ret) {
ALOGE("%s::error %s", __func__, compress_get_error(out->compr));
out->drift_correction_enabled = false;
}
exit:
return ret;
}
#else
int audio_extn_utils_compress_enable_drift_correction(
struct stream_out *out __unused,
struct audio_out_enable_drift_correction *drift __unused)
{
ALOGD("%s:: configuring drift enablement not supported", __func__);
return 0;
}
#endif
#ifdef SNDRV_COMPRESS_ADJUST_SESSION_CLOCK
int audio_extn_utils_compress_correct_drift(
struct stream_out *out,
struct audio_out_correct_drift *drift_param)
{
struct snd_compr_metadata metadata;
int ret = -EINVAL;
if (drift_param == NULL) {
ALOGE("%s:: Invalid drift_param", __func__);
goto exit;
}
ALOGD("%s:: adjust time 0x%"PRIx64" ", __func__,
drift_param->adjust_time);
if (!is_offload_usecase(out->usecase)) {
ALOGE("%s:: not supported for non offload session", __func__);
goto exit;
}
if (!out->compr) {
ALOGW("%s:: offload session not yet opened", __func__);
goto exit;
}
if (!out->drift_correction_enabled) {
ALOGE("%s:: drift correction not enabled", __func__);
goto exit;
}
metadata.key = SNDRV_COMPRESS_ADJUST_SESSION_CLOCK;
metadata.value[0] = 0xFFFFFFFF & drift_param->adjust_time; /* lsb */
metadata.value[1] = \
(0xFFFFFFFF00000000 & drift_param->adjust_time) >> 32; /* msb*/
ret = compress_set_metadata(out->compr, &metadata);
if(ret)
ALOGE("%s::error %s", __func__, compress_get_error(out->compr));
exit:
return ret;
}
#else
int audio_extn_utils_compress_correct_drift(
struct stream_out *out __unused,
struct audio_out_correct_drift *drift_param __unused)
{
ALOGD("%s:: setting adjust clock not supported", __func__);
return 0;
}
#endif
int audio_extn_utils_set_channel_map(
struct stream_out *out,
struct audio_out_channel_map_param *channel_map_param)
{
int ret = -EINVAL, i = 0;
int channels = audio_channel_count_from_out_mask(out->channel_mask);
if (channel_map_param == NULL) {
ALOGE("%s:: Invalid channel_map", __func__);
goto exit;
}
if (channel_map_param->channels != channels) {
ALOGE("%s:: Channels(%d) does not match stream channels(%d)",
__func__, channel_map_param->channels, channels);
goto exit;
}
for ( i = 0; i < channels; i++) {
ALOGV("%s:: channel_map[%d]- %d", __func__, i, channel_map_param->channel_map[i]);
out->channel_map_param.channel_map[i] = channel_map_param->channel_map[i];
}
ret = 0;
exit:
return ret;
}
int audio_extn_utils_set_pan_scale_params(
struct stream_out *out,
struct mix_matrix_params *mm_params)
{
int ret = -EINVAL, i = 0, j = 0;
if (mm_params == NULL || out == NULL) {
ALOGE("%s:: Invalid mix matrix or out param", __func__);
goto exit;
}
if (mm_params->num_output_channels > MAX_CHANNELS_SUPPORTED ||
mm_params->num_output_channels <= 0 ||
mm_params->num_input_channels > MAX_CHANNELS_SUPPORTED ||
mm_params->num_input_channels <= 0)
goto exit;
out->pan_scale_params.num_output_channels = mm_params->num_output_channels;
out->pan_scale_params.num_input_channels = mm_params->num_input_channels;
out->pan_scale_params.has_output_channel_map =
mm_params->has_output_channel_map;
for (i = 0; i < mm_params->num_output_channels; i++)
out->pan_scale_params.output_channel_map[i] =
mm_params->output_channel_map[i];
out->pan_scale_params.has_input_channel_map =
mm_params->has_input_channel_map;
for (i = 0; i < mm_params->num_input_channels; i++)
out->pan_scale_params.input_channel_map[i] =
mm_params->input_channel_map[i];
out->pan_scale_params.has_mixer_coeffs = mm_params->has_mixer_coeffs;
for (i = 0; i < mm_params->num_output_channels; i++)
for (j = 0; j < mm_params->num_input_channels; j++) {
//Convert the channel coefficient gains in Q14 format
out->pan_scale_params.mixer_coeffs[i][j] =
mm_params->mixer_coeffs[i][j] * (2 << 13);
}
ret = platform_set_stream_pan_scale_params(out->dev->platform,
out->pcm_device_id,
out->pan_scale_params);
exit:
return ret;
}
int audio_extn_utils_set_downmix_params(
struct stream_out *out,
struct mix_matrix_params *mm_params)
{
int ret = -EINVAL, i = 0, j = 0;
struct audio_usecase *usecase = NULL;
if (mm_params == NULL || out == NULL) {
ALOGE("%s:: Invalid mix matrix or out param", __func__);
goto exit;
}
if (mm_params->num_output_channels > MAX_CHANNELS_SUPPORTED ||
mm_params->num_output_channels <= 0 ||
mm_params->num_input_channels > MAX_CHANNELS_SUPPORTED ||
mm_params->num_input_channels <= 0)
goto exit;
usecase = get_usecase_from_list(out->dev, out->usecase);
if (!usecase) {
ALOGE("%s: Get usecase list failed!", __func__);
goto exit;
}
out->downmix_params.num_output_channels = mm_params->num_output_channels;
out->downmix_params.num_input_channels = mm_params->num_input_channels;
out->downmix_params.has_output_channel_map =
mm_params->has_output_channel_map;
for (i = 0; i < mm_params->num_output_channels; i++) {
out->downmix_params.output_channel_map[i] =
mm_params->output_channel_map[i];
}
out->downmix_params.has_input_channel_map =
mm_params->has_input_channel_map;
for (i = 0; i < mm_params->num_input_channels; i++)
out->downmix_params.input_channel_map[i] =
mm_params->input_channel_map[i];
out->downmix_params.has_mixer_coeffs = mm_params->has_mixer_coeffs;
for (i = 0; i < mm_params->num_output_channels; i++)
for (j = 0; j < mm_params->num_input_channels; j++) {
//Convert the channel coefficient gains in Q14 format
out->downmix_params.mixer_coeffs[i][j] =
mm_params->mixer_coeffs[i][j] * (2 << 13);
}
ret = platform_set_stream_downmix_params(out->dev->platform,
out->pcm_device_id,
usecase->out_snd_device,
out->downmix_params);
exit:
return ret;
}
bool audio_extn_utils_is_dolby_format(audio_format_t format)
{
if (format == AUDIO_FORMAT_AC3 ||
format == AUDIO_FORMAT_E_AC3 ||
format == AUDIO_FORMAT_E_AC3_JOC)
return true;
else
return false;
}
int audio_extn_utils_get_bit_width_from_string(const char *id_string)
{
int i;
const mixer_config_lookup mixer_bitwidth_config[] = {{"S24_3LE", 24},
{"S32_LE", 32},
{"S24_LE", 24},
{"S16_LE", 16}};
int num_configs = sizeof(mixer_bitwidth_config) / sizeof(mixer_bitwidth_config[0]);
for (i = 0; i < num_configs; i++) {
if (!strcmp(id_string, mixer_bitwidth_config[i].id_string))
return mixer_bitwidth_config[i].value;
}
return -EINVAL;
}
int audio_extn_utils_get_sample_rate_from_string(const char *id_string)
{
int i;
const mixer_config_lookup mixer_samplerate_config[] = {{"KHZ_8", 8000},
{"KHZ_11P025", 11025},
{"KHZ_16", 16000},
{"KHZ_22P05", 22050},
{"KHZ_32", 32000},
{"KHZ_48", 48000},
{"KHZ_96", 96000},
{"KHZ_144", 144000},
{"KHZ_192", 192000},
{"KHZ_384", 384000},
{"KHZ_44P1", 44100},
{"KHZ_88P2", 88200},
{"KHZ_176P4", 176400},
{"KHZ_352P8", 352800}};
int num_configs = sizeof(mixer_samplerate_config) / sizeof(mixer_samplerate_config[0]);
for (i = 0; i < num_configs; i++) {
if (!strcmp(id_string, mixer_samplerate_config[i].id_string))
return mixer_samplerate_config[i].value;
}
return -EINVAL;
}
int audio_extn_utils_get_channels_from_string(const char *id_string)
{
int i;
const mixer_config_lookup mixer_channels_config[] = {{"One", 1},
{"Two", 2},
{"Three",3},
{"Four", 4},
{"Five", 5},
{"Six", 6},
{"Seven", 7},
{"Eight", 8}};
int num_configs = sizeof(mixer_channels_config) / sizeof(mixer_channels_config[0]);
for (i = 0; i < num_configs; i++) {
if (!strcmp(id_string, mixer_channels_config[i].id_string))
return mixer_channels_config[i].value;
}
return -EINVAL;
}
void audio_extn_utils_release_snd_device(snd_device_t snd_device)
{
audio_extn_dev_arbi_release(snd_device);
audio_extn_sound_trigger_update_device_status(snd_device,
ST_EVENT_SND_DEVICE_FREE);
audio_extn_listen_update_device_status(snd_device,
LISTEN_EVENT_SND_DEVICE_FREE);
}
int audio_extn_utils_get_license_params(
const struct audio_device *adev,
struct audio_license_params *license_params)
{
if(!license_params)
return -EINVAL;
return platform_get_license_by_product(adev->platform,
(const char*)license_params->product, &license_params->key, license_params->license);
}
int audio_extn_utils_send_app_type_gain(struct audio_device *adev,
int app_type,
int *gain)
{
int gain_cfg[4];
const char *mixer_ctl_name = "App Type Gain";
struct mixer_ctl *ctl;
ctl = mixer_get_ctl_by_name(adev->mixer, mixer_ctl_name);
if (!ctl) {
ALOGE("%s: Could not get volume ctl mixer %s", __func__,
mixer_ctl_name);
return -EINVAL;
}
gain_cfg[0] = 0;
gain_cfg[1] = app_type;
gain_cfg[2] = gain[0];
gain_cfg[3] = gain[1];
ALOGV("%s app_type %d l(%d) r(%d)", __func__, app_type, gain[0], gain[1]);
return mixer_ctl_set_array(ctl, gain_cfg,
sizeof(gain_cfg)/sizeof(gain_cfg[0]));
}
static void vndk_fwk_init()
{
if (mVndkFwk.lib_handle != NULL)
return;
mVndkFwk.lib_handle = dlopen(VNDK_FWK_LIB_PATH, RTLD_NOW);
if (mVndkFwk.lib_handle == NULL) {
ALOGW("%s: failed to dlopen VNDK_FWK_LIB %s", __func__, strerror(errno));
return;
}
*(void **)(&mVndkFwk.isVendorEnhancedFwk) =
dlsym(mVndkFwk.lib_handle, "isRunningWithVendorEnhancedFramework");
if (mVndkFwk.isVendorEnhancedFwk == NULL) {
ALOGW("%s: dlsym failed %s", __func__, strerror(errno));
if (mVndkFwk.lib_handle) {
dlclose(mVndkFwk.lib_handle);
mVndkFwk.lib_handle = NULL;
}
return;
}
*(void **)(&mVndkFwk.getVendorEnhancedInfo) =
dlsym(mVndkFwk.lib_handle, "getVendorEnhancedInfo");
if (mVndkFwk.getVendorEnhancedInfo == NULL) {
ALOGW("%s: dlsym failed %s", __func__, strerror(errno));
if (mVndkFwk.lib_handle) {
dlclose(mVndkFwk.lib_handle);
mVndkFwk.lib_handle = NULL;
}
}
return;
}
bool audio_extn_utils_is_vendor_enhanced_fwk()
{
static int is_vendor_enhanced_fwk = -EINVAL;
if (is_vendor_enhanced_fwk != -EINVAL)
return (bool)is_vendor_enhanced_fwk;
vndk_fwk_init();
if (mVndkFwk.isVendorEnhancedFwk != NULL) {
is_vendor_enhanced_fwk = mVndkFwk.isVendorEnhancedFwk();
ALOGW("%s: is_vendor_enhanced_fwk %d", __func__, is_vendor_enhanced_fwk);
} else {
is_vendor_enhanced_fwk = 0;
ALOGW("%s: default to non enhanced_fwk config", __func__);
}
return (bool)is_vendor_enhanced_fwk;
}
int audio_extn_utils_get_vendor_enhanced_info()
{
static int vendor_enhanced_info = -EINVAL;
if (vendor_enhanced_info != -EINVAL)
return vendor_enhanced_info;
vndk_fwk_init();
if (mVndkFwk.getVendorEnhancedInfo != NULL) {
vendor_enhanced_info = mVndkFwk.getVendorEnhancedInfo();
ALOGW("%s: vendor_enhanced_info 0x%x", __func__, vendor_enhanced_info);
} else {
vendor_enhanced_info = 0x0;
ALOGW("%s: default to vendor_enhanced_info 0x0", __func__);
}
return vendor_enhanced_info;
}
int audio_extn_utils_get_perf_mode_flag(void)
{
#ifdef COMPRESSED_PERF_MODE_FLAG
return COMPRESSED_PERF_MODE_FLAG;
#else
return 0;
#endif
}
size_t audio_extn_utils_get_input_buffer_size(uint32_t sample_rate,
audio_format_t format,
int channel_count,
int64_t duration_ms,
bool is_low_latency)
{
size_t size = 0;
size_t capture_duration = AUDIO_CAPTURE_PERIOD_DURATION_MSEC;
uint32_t bytes_per_period_sample = 0;
if (audio_extn_utils_check_input_parameters(sample_rate, format, channel_count) != 0)
return 0;
if (duration_ms >= MIN_OFFLOAD_BUFFER_DURATION_MS && duration_ms <= MAX_OFFLOAD_BUFFER_DURATION_MS)
capture_duration = duration_ms;
size = (sample_rate * capture_duration) / 1000;
if (is_low_latency)
size = LOW_LATENCY_CAPTURE_PERIOD_SIZE;
bytes_per_period_sample = audio_bytes_per_sample(format) * channel_count;
size *= bytes_per_period_sample;
/* make sure the size is multiple of 32 bytes and additionally multiple of
* the frame_size (required for 24bit samples and non-power-of-2 channel counts)
* At 48 kHz mono 16-bit PCM:
* 5.000 ms = 240 frames = 15*16*1*2 = 480, a whole multiple of 32 (15)
* 3.333 ms = 160 frames = 10*16*1*2 = 320, a whole multiple of 32 (10)
*
* The loop reaches result within 32 iterations, as initial size is
* already a multiple of frame_size
*/
size = audio_extn_utils_nearest_multiple(size, audio_extn_utils_lcm(32, bytes_per_period_sample));
return size;
}
int audio_extn_utils_hash_fn(void *key)
{
return (int)key;
}
bool audio_extn_utils_hash_eq(void *key1, void *key2)
{
return (key1 == key2);
}