esp-idf/components/driver/mcpwm/mcpwm_fault.c

302 lines
11 KiB
C
Raw Normal View History

2022-05-28 05:03:05 -04:00
/*
* SPDX-FileCopyrightText: 2022 Espressif Systems (Shanghai) CO LTD
*
* SPDX-License-Identifier: Apache-2.0
*/
#include <stdlib.h>
#include <stdarg.h>
#include <sys/cdefs.h>
#include "sdkconfig.h"
#if CONFIG_MCPWM_ENABLE_DEBUG_LOG
// The local log level must be defined before including esp_log.h
// Set the maximum log level for this source file
#define LOG_LOCAL_LEVEL ESP_LOG_DEBUG
#endif
#include "freertos/FreeRTOS.h"
#include "esp_attr.h"
#include "esp_check.h"
#include "esp_err.h"
#include "esp_log.h"
#include "esp_memory_utils.h"
#include "soc/soc_caps.h"
#include "soc/mcpwm_periph.h"
#include "hal/mcpwm_ll.h"
#include "driver/mcpwm_fault.h"
#include "driver/gpio.h"
#include "mcpwm_private.h"
static const char *TAG = "mcpwm";
static void mcpwm_gpio_fault_default_isr(void *args);
static esp_err_t mcpwm_del_gpio_fault(mcpwm_fault_handle_t fault);
static esp_err_t mcpwm_del_soft_fault(mcpwm_fault_handle_t fault);
static esp_err_t mcpwm_gpio_fault_register_to_group(mcpwm_gpio_fault_t *fault, int group_id)
{
mcpwm_group_t *group = mcpwm_acquire_group_handle(group_id);
ESP_RETURN_ON_FALSE(group, ESP_ERR_NO_MEM, TAG, "no mem for group (%d)", group_id);
int fault_id = -1;
portENTER_CRITICAL(&group->spinlock);
for (int i = 0; i < SOC_MCPWM_GPIO_FAULTS_PER_GROUP; i++) {
if (!group->gpio_faults[i]) {
fault_id = i;
group->gpio_faults[i] = fault;
break;
}
}
portEXIT_CRITICAL(&group->spinlock);
if (fault_id < 0) {
mcpwm_release_group_handle(group);
group = NULL;
} else {
fault->base.group = group;
fault->fault_id = fault_id;
}
ESP_RETURN_ON_FALSE(fault_id >= 0, ESP_ERR_NOT_FOUND, TAG, "no free gpio fault in group (%d)", group_id);
return ESP_OK;
}
static void mcpwm_gpio_fault_unregister_from_group(mcpwm_gpio_fault_t *fault)
{
mcpwm_group_t *group = fault->base.group;
int fault_id = fault->fault_id;
portENTER_CRITICAL(&group->spinlock);
group->gpio_faults[fault_id] = NULL;
portEXIT_CRITICAL(&group->spinlock);
// fault has a reference on group, release it now
mcpwm_release_group_handle(group);
}
static esp_err_t mcpwm_gpio_fault_destory(mcpwm_gpio_fault_t *fault)
{
if (fault->intr) {
ESP_RETURN_ON_ERROR(esp_intr_free(fault->intr), TAG, "uninstall interrupt service failed");
}
if (fault->base.group) {
mcpwm_gpio_fault_unregister_from_group(fault);
}
free(fault);
return ESP_OK;
}
esp_err_t mcpwm_new_gpio_fault(const mcpwm_gpio_fault_config_t *config, mcpwm_fault_handle_t *ret_fault)
{
#if CONFIG_MCPWM_ENABLE_DEBUG_LOG
esp_log_level_set(TAG, ESP_LOG_DEBUG);
#endif
esp_err_t ret = ESP_OK;
mcpwm_gpio_fault_t *fault = NULL;
ESP_GOTO_ON_FALSE(config && ret_fault, ESP_ERR_INVALID_ARG, err, TAG, "invalid argument");
ESP_GOTO_ON_FALSE(config->group_id < SOC_MCPWM_GROUPS && config->group_id >= 0, ESP_ERR_INVALID_ARG,
err, TAG, "invalid group ID:%d", config->group_id);
fault = heap_caps_calloc(1, sizeof(mcpwm_gpio_fault_t), MCPWM_MEM_ALLOC_CAPS);
ESP_GOTO_ON_FALSE(fault, ESP_ERR_NO_MEM, err, TAG, "no mem for gpio fault");
ESP_GOTO_ON_ERROR(mcpwm_gpio_fault_register_to_group(fault, config->group_id), err, TAG, "register gpio fault failed");
mcpwm_group_t *group = fault->base.group;
int group_id = group->group_id;
mcpwm_hal_context_t *hal = &group->hal;
int fault_id = fault->fault_id;
// GPIO configuration
gpio_config_t gpio_conf = {
.intr_type = GPIO_INTR_DISABLE,
.mode = GPIO_MODE_INPUT | (config->flags.io_loop_back ? GPIO_MODE_OUTPUT : 0), // also enable the output path if `io_loop_back` is enabled
.pin_bit_mask = (1ULL << config->gpio_num),
.pull_down_en = config->flags.pull_down,
.pull_up_en = config->flags.pull_up,
};
ESP_GOTO_ON_ERROR(gpio_config(&gpio_conf), err, TAG, "config fault GPIO failed");
esp_rom_gpio_connect_in_signal(config->gpio_num, mcpwm_periph_signals.groups[group_id].gpio_faults[fault_id].fault_sig, 0);
// set fault detection polarity
// different gpio faults share the same config register, using a group level spin lock
portENTER_CRITICAL(&group->spinlock);
mcpwm_ll_fault_set_active_level(hal->dev, fault_id, config->flags.active_level);
portEXIT_CRITICAL(&group->spinlock);
// enable fault detection
mcpwm_ll_fault_enable_detection(hal->dev, fault_id, true);
// fill in other operator members
fault->base.type = MCPWM_FAULT_TYPE_GPIO;
fault->gpio_num = config->gpio_num;
fault->base.del = mcpwm_del_gpio_fault;
*ret_fault = &fault->base;
ESP_LOGD(TAG, "new gpio fault (%d,%d) at %p, GPIO: %d", group_id, fault_id, fault, config->gpio_num);
return ESP_OK;
err:
if (fault) {
mcpwm_gpio_fault_destory(fault);
}
return ret;
}
static esp_err_t mcpwm_del_gpio_fault(mcpwm_fault_handle_t fault)
{
mcpwm_gpio_fault_t *gpio_fault = __containerof(fault, mcpwm_gpio_fault_t, base);
mcpwm_group_t *group = fault->group;
mcpwm_hal_context_t *hal = &group->hal;
int fault_id = gpio_fault->fault_id;
ESP_LOGD(TAG, "del GPIO fault (%d,%d)", group->group_id, fault_id);
gpio_reset_pin(gpio_fault->gpio_num);
portENTER_CRITICAL(&group->spinlock);
mcpwm_ll_intr_enable(hal->dev, MCPWM_LL_EVENT_FAULT_MASK(fault_id), false);
mcpwm_ll_intr_clear_status(hal->dev, MCPWM_LL_EVENT_FAULT_MASK(fault_id));
portEXIT_CRITICAL(&group->spinlock);
// disable fault detection
mcpwm_ll_fault_enable_detection(hal->dev, fault_id, false);
// recycle memory resource
ESP_RETURN_ON_ERROR(mcpwm_gpio_fault_destory(gpio_fault), TAG, "destory GPIO fault failed");
return ESP_OK;
}
esp_err_t mcpwm_new_soft_fault(const mcpwm_soft_fault_config_t *config, mcpwm_fault_handle_t *ret_fault)
{
esp_err_t ret = ESP_OK;
mcpwm_soft_fault_t *soft_fault = NULL;
ESP_GOTO_ON_FALSE(config && ret_fault, ESP_ERR_INVALID_ARG, err, TAG, "invalid argument");
soft_fault = heap_caps_calloc(1, sizeof(mcpwm_soft_fault_t), MCPWM_MEM_ALLOC_CAPS);
ESP_GOTO_ON_FALSE(soft_fault, ESP_ERR_NO_MEM, err, TAG, "no mem for soft fault");
// fill in other fault members
soft_fault->base.type = MCPWM_FAULT_TYPE_SOFT;
soft_fault->base.del = mcpwm_del_soft_fault;
*ret_fault = &soft_fault->base;
ESP_LOGD(TAG, "new soft fault at %p", soft_fault);
return ESP_OK;
err:
2022-07-28 01:05:18 -04:00
// soft_fault must be NULL in the error handling path, and it's a determined behaviour to free a NULL pointer in esp-idf
free(soft_fault);
2022-05-28 05:03:05 -04:00
return ret;
}
static esp_err_t mcpwm_del_soft_fault(mcpwm_fault_handle_t fault)
{
mcpwm_soft_fault_t *soft_fault = __containerof(fault, mcpwm_soft_fault_t, base);
ESP_LOGD(TAG, "del soft fault %p", soft_fault);
free(soft_fault);
return ESP_OK;
}
esp_err_t mcpwm_soft_fault_activate(mcpwm_fault_handle_t fault)
{
ESP_RETURN_ON_FALSE(fault, ESP_ERR_INVALID_ARG, TAG, "invalid argument");
ESP_RETURN_ON_FALSE(fault->type == MCPWM_FAULT_TYPE_SOFT, ESP_ERR_INVALID_ARG, TAG, "not a valid soft fault");
mcpwm_group_t *group = fault->group;
mcpwm_soft_fault_t *soft_fault = __containerof(fault, mcpwm_soft_fault_t, base);
mcpwm_oper_t *operator = soft_fault->operator;
ESP_RETURN_ON_FALSE(operator, ESP_ERR_INVALID_STATE, TAG, "no operator is assigned to the fault");
switch (operator->brake_mode_on_soft_fault) {
case MCPWM_OPER_BRAKE_MODE_CBC:
mcpwm_ll_brake_trigger_soft_cbc(group->hal.dev, operator->oper_id);
break;
case MCPWM_OPER_BRAKE_MODE_OST:
mcpwm_ll_brake_trigger_soft_ost(group->hal.dev, operator->oper_id);
break;
default:
ESP_RETURN_ON_FALSE(false, ESP_ERR_INVALID_STATE, TAG, "unknown brake mode:%d", operator->brake_mode_on_soft_fault);
break;
}
return ESP_OK;
}
esp_err_t mcpwm_del_fault(mcpwm_fault_handle_t fault)
{
ESP_RETURN_ON_FALSE(fault, ESP_ERR_INVALID_ARG, TAG, "invalid argument");
return fault->del(fault);
}
esp_err_t mcpwm_fault_register_event_callbacks(mcpwm_fault_handle_t fault, const mcpwm_fault_event_callbacks_t *cbs, void *user_data)
{
ESP_RETURN_ON_FALSE(fault && cbs, ESP_ERR_INVALID_ARG, TAG, "invalid argument");
ESP_RETURN_ON_FALSE(fault->type == MCPWM_FAULT_TYPE_GPIO, ESP_ERR_INVALID_ARG, TAG, "only gpio fault can register event callback");
mcpwm_gpio_fault_t *gpio_fault = __containerof(fault, mcpwm_gpio_fault_t, base);
mcpwm_group_t *group = fault->group;
int group_id = group->group_id;
mcpwm_hal_context_t *hal = &group->hal;
int fault_id = gpio_fault->fault_id;
#if CONFIG_MCWPM_ISR_IRAM_SAFE
if (cbs->on_fault_enter) {
ESP_RETURN_ON_FALSE(esp_ptr_in_iram(cbs->on_fault_enter), ESP_ERR_INVALID_ARG, TAG, "on_fault_enter callback not in IRAM");
}
if (cbs->on_fault_exit) {
ESP_RETURN_ON_FALSE(esp_ptr_in_iram(cbs->on_fault_exit), ESP_ERR_INVALID_ARG, TAG, "on_fault_exit callback not in IRAM");
}
if (user_data) {
ESP_RETURN_ON_FALSE(esp_ptr_internal(user_data), ESP_ERR_INVALID_ARG, TAG, "user context not in internal RAM");
}
#endif
// lazy install interrupt service
if (!gpio_fault->intr) {
// we want the interrupt service to be enabled after allocation successfully
2022-05-28 05:03:05 -04:00
int isr_flags = MCPWM_INTR_ALLOC_FLAG & ~ESP_INTR_FLAG_INTRDISABLED;
ESP_RETURN_ON_ERROR(esp_intr_alloc_intrstatus(mcpwm_periph_signals.groups[group_id].irq_id, isr_flags,
(uint32_t)mcpwm_ll_intr_get_status_reg(hal->dev), MCPWM_LL_EVENT_FAULT_MASK(fault_id),
mcpwm_gpio_fault_default_isr, gpio_fault, &gpio_fault->intr), TAG, "install interrupt service for gpio fault failed");
}
// different mcpwm events share the same interrupt control register
portENTER_CRITICAL(&group->spinlock);
mcpwm_ll_intr_enable(hal->dev, MCPWM_LL_EVENT_FAULT_ENTER(fault_id), cbs->on_fault_enter != NULL);
mcpwm_ll_intr_enable(hal->dev, MCPWM_LL_EVENT_FAULT_EXIT(fault_id), cbs->on_fault_exit != NULL);
portEXIT_CRITICAL(&group->spinlock);
gpio_fault->on_fault_enter = cbs->on_fault_enter;
gpio_fault->on_fault_exit = cbs->on_fault_exit;
gpio_fault->user_data = user_data;
return ESP_OK;
}
static void IRAM_ATTR mcpwm_gpio_fault_default_isr(void *args)
{
mcpwm_gpio_fault_t *fault = (mcpwm_gpio_fault_t *)args;
mcpwm_group_t *group = fault->base.group;
mcpwm_hal_context_t *hal = &group->hal;
int fault_id = fault->fault_id;
bool need_yield = false;
uint32_t status = mcpwm_ll_intr_get_status(hal->dev);
mcpwm_ll_intr_clear_status(hal->dev, status & MCPWM_LL_EVENT_FAULT_MASK(fault_id));
mcpwm_fault_event_data_t edata = {
// TODO
};
if (status & MCPWM_LL_EVENT_FAULT_ENTER(fault_id)) {
mcpwm_fault_event_cb_t cb = fault->on_fault_enter;
if (cb) {
if (cb(&fault->base, &edata, fault->user_data)) {
need_yield = true;
}
}
}
if (status & MCPWM_LL_EVENT_FAULT_EXIT(fault_id)) {
mcpwm_fault_event_cb_t cb = fault->on_fault_exit;
if (cb) {
if (cb(&fault->base, &edata, fault->user_data)) {
need_yield = true;
}
}
}
if (need_yield) {
portYIELD_FROM_ISR();
}
}