Merge "Return error instead of unloading existing model" into nyc-dev

    AutoMutex lock(mLock);

    if (mModels.size() >= mDescriptor.properties.max_sound_models) {

        /* Make space for a keyphrase sound model by first trying to swap out a previously loaded

         * keyphrase sound model, or if needed, another sound model. This decision would optimally

         * happen in SoundTriggerHelper, but is happening here because state tracking isn't good

         * enough in SoundTriggerHelper to ensure that state is consistent between it and the HAL,

         * nor does sufficient error handling exist to recover from inconsistencies.

         * Once that exists:

         * TODO: we should return an error instead of unloading a previous sound model here.

         */

        if (mModels.size() == 0) {

            return INVALID_OPERATION;

        }

        if (sound_model->type == SOUND_MODEL_TYPE_KEYPHRASE) {

            ALOGW("loadSoundModel() max number of models exceeded %d making room for a new one",

                  mDescriptor.properties.max_sound_models);

            sound_model_handle_t unload_handle = mModels.valueAt(0)->mHandle;

            for (size_t i = 0; i < mModels.size(); i++) {

                if (mModels.valueAt(i)->mType == SOUND_MODEL_TYPE_KEYPHRASE) {

                    unload_handle = mModels.keyAt(i);

                    break;

                }

            }

            unloadSoundModel_l(unload_handle);

        } else {

        ALOGW("loadSoundModel(): Not loading, max number of models (%d) would be exceeded",

              mDescriptor.properties.max_sound_models);

        return INVALID_OPERATION;

    }

    }

    status_t status = mHwDevice->load_sound_model(mHwDevice, sound_model,

                                                  SoundTriggerHwService::soundModelCallback,