audio/soundtrigger/sound_trigger_hw.c

/*
 * 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 "sound_trigger_hw_flounder"
/*#define LOG_NDEBUG 0*/

#include <errno.h>
#include <fcntl.h>
#include <poll.h>
#include <pthread.h>
#include <sys/ioctl.h>
#include <sys/prctl.h>
#include <cutils/log.h>
#include <cutils/uevent.h>

#include <hardware/hardware.h>
#include <system/sound_trigger.h>
#include <hardware/sound_trigger.h>
#include <tinyalsa/asoundlib.h>

#define FLOUNDER_MIXER_VAD	0
#define FLOUNDER_CTRL_DSP	"VAD Mode"
#define UEVENT_MSG_LEN		1024

#define FLOUNDER_VAD_DEV	"/dev/snd/hwC0D0"

#define FLOUNDER_STREAMING_DELAY_USEC	1000
#define FLOUNDER_STREAMING_READ_RETRY_ATTEMPTS	30
#define FLOUNDER_STREAMING_BUFFER_SIZE	(16 * 1024)

static const struct sound_trigger_properties hw_properties = {
    "The Android Open Source Project", // implementor
    "Volantis OK Google ", // description
    1, // version
    { 0xe780f240, 0xf034, 0x11e3, 0xb79a, { 0x00, 0x02, 0xa5, 0xd5, 0xc5, 0x1b } }, // uuid
    1, // max_sound_models
    1, // max_key_phrases
    1, // max_users
    RECOGNITION_MODE_VOICE_TRIGGER, // recognition_modes
    true, // capture_transition
    0, // max_capture_ms
    false, // concurrent_capture
    false, // trigger_in_event
    0 // power_consumption_mw
};

struct flounder_sound_trigger_device {
    struct sound_trigger_hw_device device;
    sound_model_handle_t model_handle;
    recognition_callback_t recognition_callback;
    void *recognition_cookie;
    sound_model_callback_t sound_model_callback;
    void *sound_model_cookie;
    pthread_t callback_thread;
    pthread_mutex_t lock;
    int send_sock;
    int term_sock;
    int vad_fd;
    struct mixer *mixer;
    struct mixer_ctl *ctl_dsp;
    struct sound_trigger_recognition_config *config;
    int is_streaming;
    int opened;
    char *streaming_buf;
    size_t streaming_buf_read;
    size_t streaming_buf_len;
};

struct rt_codec_cmd {
    size_t number;
    int *buf;
};

enum {
    RT_READ_CODEC_DSP_IOCTL = _IOR('R', 0x04, struct rt_codec_cmd),
    RT_WRITE_CODEC_DSP_IOCTL = _IOW('R', 0x04, struct rt_codec_cmd),
};

// Since there's only ever one sound_trigger_device, keep it as a global so that other people can
// dlopen this lib to get at the streaming audio.
static struct flounder_sound_trigger_device g_stdev = { .lock = PTHREAD_MUTEX_INITIALIZER };

static void stdev_dsp_set_power(struct flounder_sound_trigger_device *stdev,
                                int val)
{
    stdev->is_streaming = 0;
    stdev->streaming_buf_read = 0;
    stdev->streaming_buf_len = 0;
    mixer_ctl_set_value(stdev->ctl_dsp, 0, val);
}

static int stdev_init_mixer(struct flounder_sound_trigger_device *stdev)
{
    int ret = -1;

    stdev->vad_fd = open(FLOUNDER_VAD_DEV, O_RDWR);
    if (stdev->vad_fd < 0) {
        ALOGE("Error opening vad device");
        return ret;
    }

    stdev->mixer = mixer_open(FLOUNDER_MIXER_VAD);
    if (!stdev->mixer)
        goto err;

    stdev->ctl_dsp = mixer_get_ctl_by_name(stdev->mixer, FLOUNDER_CTRL_DSP);
    if (!stdev->ctl_dsp)
        goto err;

    stdev_dsp_set_power(stdev, 0); // Disable DSP at the beginning

    return 0;

err:
    close(stdev->vad_fd);
    if (stdev->mixer)
        mixer_close(stdev->mixer);
    return ret;
}

static void stdev_close_term_sock(struct flounder_sound_trigger_device *stdev)
{
    if (stdev->send_sock >=0) {
        close(stdev->send_sock);
        stdev->send_sock = -1;
    }
    if (stdev->term_sock >=0) {
        close(stdev->term_sock);
        stdev->term_sock = -1;
    }
}

static void stdev_close_mixer(struct flounder_sound_trigger_device *stdev)
{
    if (stdev) {
        stdev_dsp_set_power(stdev, 0);
        mixer_close(stdev->mixer);
        stdev_close_term_sock(stdev);
        close(stdev->vad_fd);
    }
}

static int vad_load_sound_model(struct flounder_sound_trigger_device *stdev,
                                char *buf, size_t len)
{
    struct rt_codec_cmd cmd;
    int ret = 0;

    if (!buf || (len == 0))
        return ret;

    cmd.number = len / sizeof(int);
    cmd.buf = (int *)buf;

    ret = ioctl(stdev->vad_fd, RT_WRITE_CODEC_DSP_IOCTL, &cmd);
    if (ret)
        ALOGE("Error VAD write ioctl");
    return ret;
}

static char *sound_trigger_event_alloc(struct flounder_sound_trigger_device *
                                       stdev)
{
    char *data;
    struct sound_trigger_phrase_recognition_event *event;

    data = (char *)calloc(1,
                    sizeof(struct sound_trigger_phrase_recognition_event));
    if (!data)
        return NULL;

    event = (struct sound_trigger_phrase_recognition_event *)data;
    event->common.status = RECOGNITION_STATUS_SUCCESS;
    event->common.type = SOUND_MODEL_TYPE_KEYPHRASE;
    event->common.model = stdev->model_handle;

    if (stdev->config) {
        unsigned int i;

        event->num_phrases = stdev->config->num_phrases;
        if (event->num_phrases > SOUND_TRIGGER_MAX_PHRASES)
            event->num_phrases = SOUND_TRIGGER_MAX_PHRASES;
        for (i=0; i < event->num_phrases; i++)
            memcpy(&event->phrase_extras[i], &stdev->config->phrases[i],
                   sizeof(struct sound_trigger_phrase_recognition_extra));
    }

    event->num_phrases = 1;
    event->phrase_extras[0].confidence_level = 100;
    event->phrase_extras[0].num_levels = 1;
    event->phrase_extras[0].levels[0].level = 100;
    event->phrase_extras[0].levels[0].user_id = 0;
    // Signify that all the data is comming through streaming, not through the
    // buffer.
    event->common.capture_available = true;

    event->common.audio_config = AUDIO_CONFIG_INITIALIZER;
    event->common.audio_config.sample_rate = 16000;
    event->common.audio_config.channel_mask = AUDIO_CHANNEL_IN_MONO;
    event->common.audio_config.format = AUDIO_FORMAT_PCM_16_BIT;

    return data;
}

// The stdev should be locked when you call this function.
static int fetch_streaming_buffer(struct flounder_sound_trigger_device * stdev)
{
    struct rt_codec_cmd cmd;
    int ret = 0;
    int i;

    cmd.number = FLOUNDER_STREAMING_BUFFER_SIZE / sizeof(int);
    cmd.buf = (int*) stdev->streaming_buf;

    ALOGV("%s: Fetching bytes", __func__);
    for (i = 0; i < FLOUNDER_STREAMING_READ_RETRY_ATTEMPTS; i++) {
        ret = ioctl(stdev->vad_fd, RT_READ_CODEC_DSP_IOCTL, &cmd);
        if (ret == 0) {
            usleep(FLOUNDER_STREAMING_DELAY_USEC);
        } else if (ret < 0) {
            ALOGV("%s: IOCTL failed with code %d", __func__, ret);
            return ret;
        } else {
            // The IOCTL returns the number of int16 samples that were read, so we need to multipy
            // it by 2 .
            ALOGV("%s: IOCTL captured %d samples", __func__, ret);
            stdev->streaming_buf_read = 0;
            stdev->streaming_buf_len = ret << 1;
            return 0;
        }
    }
    ALOGV("%s: Timeout waiting for data from dsp", __func__);
    return 0;
}

static void *callback_thread_loop(void *context)
{
    char msg[UEVENT_MSG_LEN];
    struct flounder_sound_trigger_device *stdev =
               (struct flounder_sound_trigger_device *)context;
    struct pollfd fds[2];
    int exit_sockets[2];
    int err = 0;
    int i, n;

    ALOGI("%s", __func__);
    prctl(PR_SET_NAME, (unsigned long)"sound trigger callback", 0, 0, 0);

    pthread_mutex_lock(&stdev->lock);
    if (stdev->recognition_callback == NULL)
        goto exit;

    if (socketpair(AF_UNIX, SOCK_STREAM, 0, exit_sockets) == -1)
        goto exit;

    stdev_close_term_sock(stdev);
    stdev->send_sock = exit_sockets[0];
    stdev->term_sock = exit_sockets[1];

    memset(fds, 0, 2 * sizeof(struct pollfd));
    fds[0].events = POLLIN;
    fds[0].fd = uevent_open_socket(64*1024, true);
    if (fds[0].fd == -1) {
        ALOGE("Error opening socket for hotplug uevent");
        goto exit;
    }
    fds[1].events = POLLIN;
    fds[1].fd = stdev->term_sock;

    stdev_dsp_set_power(stdev, 1);

    pthread_mutex_unlock(&stdev->lock);

    while (1) {
        err = poll(fds, 2, -1);
        pthread_mutex_lock(&stdev->lock);
        if ((err < 0) || (stdev->recognition_callback == NULL)) {
            ALOGE("Error in hotplug CPU poll: %d", errno);
            break;
        }

        if (fds[0].revents & POLLIN) {
            n = uevent_kernel_multicast_recv(fds[0].fd, msg, UEVENT_MSG_LEN);
            if (n <= 0) {
                pthread_mutex_unlock(&stdev->lock);
                continue;
            }
            for (i=0; i < n;) {
                if (strstr(msg + i, "HOTWORD")) {
                    struct sound_trigger_phrase_recognition_event *event;

                    event = (struct sound_trigger_phrase_recognition_event *)
                            sound_trigger_event_alloc(stdev);
                    if (event) {
                        stdev->is_streaming = 1;
                        ALOGI("%s send callback model %d", __func__,
                              stdev->model_handle);
                        stdev->recognition_callback(&event->common,
                                                    stdev->recognition_cookie);
                        free(event);
                        // Start reading data from the DSP while the upper levels do their thing.
                        if (stdev->config && stdev->config->capture_requested) {
                            fetch_streaming_buffer(stdev);
                        }
                    }
                    goto found;
                }
                i += strlen(msg + i) + 1;
            }
        } else if (fds[1].revents & POLLIN) {
            read(fds[1].fd, &n, sizeof(n)); /* clear the socket */
            ALOGI("%s: Termination message", __func__);
            break;
        } else {
            ALOGI("%s: Message to ignore", __func__);
        }
        pthread_mutex_unlock(&stdev->lock);
    }

found:
    close(fds[0].fd);

exit:
    stdev->recognition_callback = NULL;
    stdev_close_term_sock(stdev);

    if (stdev->config && !stdev->config->capture_requested)
        stdev_dsp_set_power(stdev, 0);

    pthread_mutex_unlock(&stdev->lock);

    return (void *)(long)err;
}

static int stdev_get_properties(const struct sound_trigger_hw_device *dev,
                                struct sound_trigger_properties *properties)
{
    struct flounder_sound_trigger_device *stdev =
                               (struct flounder_sound_trigger_device *)dev;

    ALOGI("%s", __func__);
    if (properties == NULL)
        return -EINVAL;
    memcpy(properties, &hw_properties, sizeof(struct sound_trigger_properties));
    return 0;
}

static int stdev_load_sound_model(const struct sound_trigger_hw_device *dev,
                                  struct sound_trigger_sound_model *sound_model,
                                  sound_model_callback_t callback,
                                  void *cookie,
                                  sound_model_handle_t *handle)
{
    struct flounder_sound_trigger_device *stdev =
                                 (struct flounder_sound_trigger_device *)dev;
    int ret = 0;

    ALOGI("%s", __func__);
    pthread_mutex_lock(&stdev->lock);
    if (handle == NULL || sound_model == NULL) {
        ret = -EINVAL;
        goto exit;
    }

    if (stdev->model_handle == 1) {
        ret = -ENOSYS;
        goto exit;
    }

    ret = vad_load_sound_model(stdev,
                               (char *)sound_model + sound_model->data_offset,
                               sound_model->data_size);
    stdev->model_handle = 1;
    stdev->sound_model_callback = callback;
    stdev->sound_model_cookie = cookie;
    *handle = stdev->model_handle;

exit:
    pthread_mutex_unlock(&stdev->lock);
    return ret;
}

static int stdev_unload_sound_model(const struct sound_trigger_hw_device *dev,
                                    sound_model_handle_t handle)
{
    struct flounder_sound_trigger_device *stdev =
                                   (struct flounder_sound_trigger_device *)dev;
    int status = 0;

    ALOGI("%s handle %d", __func__, handle);
    pthread_mutex_lock(&stdev->lock);
    if (handle != 1) {
        status = -EINVAL;
        goto exit;
    }
    if (stdev->model_handle == 0) {
        status = -ENOSYS;
        goto exit;
    }
    stdev->model_handle = 0;
    free(stdev->config);
    stdev->config = NULL;
    if (stdev->recognition_callback != NULL) {
        stdev->recognition_callback = NULL;
        if (stdev->send_sock >=0)
            write(stdev->send_sock, "T", 1);
        pthread_mutex_unlock(&stdev->lock);

        pthread_join(stdev->callback_thread, (void **)NULL);

        pthread_mutex_lock(&stdev->lock);
    }

exit:
    stdev_dsp_set_power(stdev, 0);

    pthread_mutex_unlock(&stdev->lock);
    return status;
}

static int stdev_start_recognition(const struct sound_trigger_hw_device *dev,
                                   sound_model_handle_t sound_model_handle,
                                   const struct sound_trigger_recognition_config *config,
                                   recognition_callback_t callback,
                                   void *cookie)
{
    struct flounder_sound_trigger_device *stdev =
                                  (struct flounder_sound_trigger_device *)dev;
    int status = 0;

    ALOGI("%s sound model %d", __func__, sound_model_handle);
    pthread_mutex_lock(&stdev->lock);
    if (stdev->model_handle != sound_model_handle) {
        status = -ENOSYS;
        goto exit;
    }
    if (stdev->recognition_callback != NULL) {
        status = -ENOSYS;
        goto exit;
    }

    free(stdev->config);
    stdev->config = NULL;
    if (config) {
        stdev->config = malloc(sizeof(*config));
        if (!stdev->config) {
            status = -ENOMEM;
            goto exit;
        }
        memcpy(stdev->config, config, sizeof(*config));
    }

    stdev_dsp_set_power(stdev, 0);

    stdev->recognition_callback = callback;
    stdev->recognition_cookie = cookie;
    pthread_create(&stdev->callback_thread, (const pthread_attr_t *) NULL,
                        callback_thread_loop, stdev);
exit:
    pthread_mutex_unlock(&stdev->lock);
    return status;
}

static int stdev_stop_recognition(const struct sound_trigger_hw_device *dev,
                                  sound_model_handle_t sound_model_handle)
{
    struct flounder_sound_trigger_device *stdev =
                                 (struct flounder_sound_trigger_device *)dev;
    int status = 0;

    ALOGI("%s sound model %d", __func__, sound_model_handle);
    pthread_mutex_lock(&stdev->lock);
    if (stdev->model_handle != sound_model_handle) {
        status = -ENOSYS;
        goto exit;
    }
    if (stdev->recognition_callback == NULL) {
        status = -ENOSYS;
        goto exit;
    }
    free(stdev->config);
    stdev->config = NULL;
    stdev->recognition_callback = NULL;
    if (stdev->send_sock >=0)
        write(stdev->send_sock, "T", 1);
    pthread_mutex_unlock(&stdev->lock);

    pthread_join(stdev->callback_thread, (void **)NULL);

    pthread_mutex_lock(&stdev->lock);

exit:
    stdev_dsp_set_power(stdev, 0);

    pthread_mutex_unlock(&stdev->lock);
    return status;
}

__attribute__ ((visibility ("default")))
int sound_trigger_open_for_streaming()
{
    struct flounder_sound_trigger_device *stdev = &g_stdev;
    int ret = 0;

    pthread_mutex_lock(&stdev->lock);

    if (!stdev->opened) {
        ALOGE("%s: stdev has not been opened", __func__);
        ret = -EFAULT;
        goto exit;
    }
    if (!stdev->is_streaming) {
        ALOGE("%s: DSP is not currently streaming", __func__);
        ret = -EBUSY;
        goto exit;
    }
    // TODO: Probably want to get something from whoever called us to bind to it/assert that it's a
    // valid connection. Perhaps returning a more
    // meaningful handle would be a good idea as well.
    ret = 1;
exit:
    pthread_mutex_unlock(&stdev->lock);
    return ret;
}

__attribute__ ((visibility ("default")))
size_t sound_trigger_read_samples(int audio_handle, void *buffer, size_t  buffer_len)
{
    struct flounder_sound_trigger_device *stdev = &g_stdev;
    int i;
    size_t ret = 0;

    if (audio_handle <= 0) {
        ALOGE("%s: invalid audio handle", __func__);
        return -EINVAL;
    }

    pthread_mutex_lock(&stdev->lock);

    if (!stdev->opened) {
        ALOGE("%s: stdev has not been opened", __func__);
        ret = -EFAULT;
        goto exit;
    }
    if (!stdev->is_streaming) {
        ALOGE("%s: DSP is not currently streaming", __func__);
        ret = -EINVAL;
        goto exit;
    }

    if (stdev->streaming_buf_read == stdev->streaming_buf_len) {
        ret = fetch_streaming_buffer(stdev);
    }

    if (!ret) {
        ret = stdev->streaming_buf_len - stdev->streaming_buf_read;
        if (ret > buffer_len)
            ret = buffer_len;
        memcpy(buffer, stdev->streaming_buf + stdev->streaming_buf_read, ret);
        stdev->streaming_buf_read += ret;
        ALOGV("%s: Sent %zu bytes to buffer", __func__, ret);
    }

exit:
    pthread_mutex_unlock(&stdev->lock);
    return ret;
}

__attribute__ ((visibility ("default")))
int sound_trigger_close_for_streaming(int audio_handle __unused)
{
    // TODO: Power down the DSP? I think we shouldn't in case we want to open this mic for streaming
    // for the voice search?
    return 0;
}

static int stdev_close(hw_device_t *device)
{
    struct flounder_sound_trigger_device *stdev =
                                (struct flounder_sound_trigger_device *)device;
    int ret = 0;

    pthread_mutex_lock(&stdev->lock);
    if (!stdev->opened) {
        ALOGE("%s: device already closed", __func__);
        ret = -EFAULT;
        goto exit;
    }
    free(stdev->streaming_buf);
    stdev_close_mixer(stdev);
    stdev->model_handle = 0;
    stdev->send_sock = 0;
    stdev->term_sock = 0;
    stdev->opened = false;

exit:
    pthread_mutex_unlock(&stdev->lock);
    return ret;
}

static int stdev_open(const hw_module_t *module, const char *name,
                      hw_device_t **device)
{
    struct flounder_sound_trigger_device *stdev;
    int ret;

    if (strcmp(name, SOUND_TRIGGER_HARDWARE_INTERFACE) != 0)
        return -EINVAL;

    stdev = &g_stdev;
    pthread_mutex_lock(&stdev->lock);

    if (stdev->opened) {
        ALOGE("%s: Only one sountrigger can be opened at a time", __func__);
        ret = -EBUSY;
        goto exit;
    }

    stdev->streaming_buf = malloc(FLOUNDER_STREAMING_BUFFER_SIZE);
    if (!stdev->streaming_buf) {
        ret = -ENOMEM;
        goto exit;
    }

    ret = stdev_init_mixer(stdev);
    if (ret) {
        ALOGE("Error mixer init");
        free(stdev->streaming_buf);
        goto exit;
    }

    stdev->device.common.tag = HARDWARE_DEVICE_TAG;
    stdev->device.common.version = SOUND_TRIGGER_DEVICE_API_VERSION_1_0;
    stdev->device.common.module = (struct hw_module_t *)module;
    stdev->device.common.close = stdev_close;
    stdev->device.get_properties = stdev_get_properties;
    stdev->device.load_sound_model = stdev_load_sound_model;
    stdev->device.unload_sound_model = stdev_unload_sound_model;
    stdev->device.start_recognition = stdev_start_recognition;
    stdev->device.stop_recognition = stdev_stop_recognition;
    stdev->send_sock = stdev->term_sock = -1;
    stdev->streaming_buf_read = 0;
    stdev->streaming_buf_len = 0;
    stdev->opened = true;

    *device = &stdev->device.common; /* same address as stdev */
exit:
    pthread_mutex_unlock(&stdev->lock);
    return ret;
}

static struct hw_module_methods_t hal_module_methods = {
    .open = stdev_open,
};

struct sound_trigger_module HAL_MODULE_INFO_SYM = {
    .common = {
        .tag = HARDWARE_MODULE_TAG,
        .module_api_version = SOUND_TRIGGER_MODULE_API_VERSION_1_0,
        .hal_api_version = HARDWARE_HAL_API_VERSION,
        .id = SOUND_TRIGGER_HARDWARE_MODULE_ID,
        .name = "Default sound trigger HAL",
        .author = "The Android Open Source Project",
        .methods = &hal_module_methods,
    },
};