- OtaCoordinator lock/unlock: 移除共享 m_locked 标志,改为传参传递 take 结果,消除多任务并发 lock() 超时覆盖持锁状态导致 mutex 永久死锁的问题 - SPI appPing: 发送帧填充至 16 字节防止首字节丢失,增大轮询间隔至 10ms, 增大接收缓冲区避免字节偏移解析失败 - SPI init: 先创建 mutex 再使用 LockGuard,消除竞态窗口 - 压制 WiFi 框架 INFO 日志 (pp/net80211/wifi/wifi_init),减少启动打印 - 降级 7 条调试日志为 DBG 级别(UART0 初始化、冷启动缓存清理、互斥锁/重连 任务/回调/netif 创建) - GD32 版本号读取增加 5 次重试上限和 OTA 繁忙时跳过逻辑 - 移除 forwardFirmware 中已过时注释
447 lines
15 KiB
C++
447 lines
15 KiB
C++
/*
|
||
* ota_coordinator.cpp - 全局 OTA 协调器实现
|
||
*
|
||
* OtaCoordinator 是系统级 OTA 状态的唯一权威来源。
|
||
* 传输层(WiFi/BLE)通过本协调器发起所有 OTA 操作。
|
||
*/
|
||
#include "ota_coordinator.h"
|
||
#include "esp32_ota_core.h"
|
||
#include "relay/ota_relay.h"
|
||
#include "monitor/debug_log.h"
|
||
|
||
#define TAG "OTA_COORD"
|
||
|
||
// ============================================================
|
||
// 单例
|
||
// ============================================================
|
||
OtaCoordinator& OtaCoordinator::instance()
|
||
{
|
||
static OtaCoordinator inst;
|
||
return inst;
|
||
}
|
||
|
||
// ============================================================
|
||
// 线程安全辅助
|
||
// ============================================================
|
||
bool OtaCoordinator::lock()
|
||
{
|
||
// ★ 修复: 返回 take 结果,由调用方传回 unlock()。
|
||
// 旧实现用成员 bool m_locked 记录 take 结果,多任务并发 lock() 时
|
||
// 超时方会把 m_locked 置 false,覆盖持锁方的 true,导致 unlock() 跳过 give,
|
||
// mutex 永久死锁 → 后续 take 超时触发 vTaskPriorityDisinherit assert → PANIC。
|
||
if (!m_mutex) return false;
|
||
bool ok = (xSemaphoreTake(m_mutex, kLockTimeout) == pdTRUE);
|
||
if (!ok) {
|
||
DEBUG_WARN(TAG, "lock() 超时 (%u ms),本次状态变更被跳过",
|
||
(unsigned)pdTICKS_TO_MS(kLockTimeout));
|
||
}
|
||
return ok;
|
||
}
|
||
|
||
void OtaCoordinator::unlock(bool handle)
|
||
{
|
||
// ★ 修复: 仅在对应的 lock() 成功 take 时才 give。
|
||
if (handle && m_mutex) {
|
||
xSemaphoreGive(m_mutex);
|
||
}
|
||
}
|
||
|
||
// ============================================================
|
||
// 卡死自愈检查(必须在持锁状态下调用)
|
||
// ============================================================
|
||
void OtaCoordinator::watchdogCheckLocked()
|
||
{
|
||
// 仅对"忙"状态(可能卡死的)做检查
|
||
switch (m_sysState) {
|
||
case OTA_SYS_GD32_STORING:
|
||
case OTA_SYS_GD32_RELAYING:
|
||
case OTA_SYS_ESP32_RECEIVING:
|
||
case OTA_SYS_ESP32_FINALIZING:
|
||
case OTA_SYS_ESP32_REBOOTING:
|
||
break;
|
||
default:
|
||
return; // IDLE/STORED/COMPLETE/ERROR 无需检查
|
||
}
|
||
|
||
if (m_busySinceTick == 0) {
|
||
// 未记录起始时刻(理论上不应发生),保守不触发
|
||
return;
|
||
}
|
||
|
||
TickType_t elapsed = xTaskGetTickCount() - m_busySinceTick;
|
||
if (elapsed < kBusyWatchdogTicks) {
|
||
return; // 未超时
|
||
}
|
||
|
||
// ★ 超时自愈:强制转 ERROR。
|
||
// ERROR 不在任何 beginXxx() 的拒绝列表中,下次请求会自动从 ERROR 恢复。
|
||
// 清理由各子系统自行负责(abortCurrent 中会按 ERROR 不做实际清理,仅复位状态)。
|
||
DEBUG_ERR(TAG, "忙状态 %d 卡死超过 %" PRIu32 " ms,强制自愈 → ERROR",
|
||
(int)m_sysState, (uint32_t)pdTICKS_TO_MS(kBusyWatchdogTicks));
|
||
m_sysState = OTA_SYS_ERROR;
|
||
m_busySinceTick = 0;
|
||
}
|
||
|
||
// ============================================================
|
||
// 初始化
|
||
// ============================================================
|
||
void OtaCoordinator::init()
|
||
{
|
||
if (!m_mutex) {
|
||
m_mutex = xSemaphoreCreateMutex();
|
||
if (m_mutex) {
|
||
DEBUG_INFO(TAG, "OTA 全局协调器已初始化");
|
||
} else {
|
||
DEBUG_ERR(TAG, "协调器 mutex 创建失败");
|
||
}
|
||
}
|
||
m_sysState = OTA_SYS_IDLE;
|
||
m_gd32StorageDone = false;
|
||
m_busySinceTick = 0;
|
||
|
||
// ★ 重启恢复:如果 NVS 里有完整固件元数据,开机时立即进入 STORED 状态,
|
||
// 否则用户重启后看到的状态是 IDLE 但 NVS 里其实有未中继的固件,前后不一致。
|
||
ota_meta_t boot_meta;
|
||
if (OtaRelay::metaLoad(&boot_meta) == ESP_OK &&
|
||
boot_meta.total_size > 0 &&
|
||
boot_meta.stored_size >= boot_meta.total_size) {
|
||
m_sysState = OTA_SYS_GD32_STORED;
|
||
m_gd32StorageDone = true;
|
||
DEBUG_INFO(TAG, "启动恢复: NVS 检测到完整 GD32 固件 (%" PRIu32 " 字节), → GD32_STORED",
|
||
boot_meta.total_size);
|
||
}
|
||
}
|
||
|
||
// ============================================================
|
||
// GD32 存储阶段
|
||
// ============================================================
|
||
esp_err_t OtaCoordinator::beginGd32Storage(uint32_t total_size)
|
||
{
|
||
bool h = lock();
|
||
|
||
// ★ 先做卡死自愈检查:若之前忙状态卡死,自动转 ERROR 继续后续流程
|
||
if (h) watchdogCheckLocked();
|
||
|
||
// 系统忙检查:仅拒绝真正"在传输"的活跃状态
|
||
// STORED / COMPLETE 是"待中继/已中继"的非活跃状态 → 允许新存储覆盖
|
||
if (h && (m_sysState == OTA_SYS_ESP32_RECEIVING ||
|
||
m_sysState == OTA_SYS_ESP32_FINALIZING ||
|
||
m_sysState == OTA_SYS_ESP32_REBOOTING ||
|
||
m_sysState == OTA_SYS_GD32_STORING ||
|
||
m_sysState == OTA_SYS_GD32_RELAYING)) {
|
||
DEBUG_WARN(TAG, "系统忙 (state=%d),拒绝 GD32 存储请求", (int)m_sysState);
|
||
unlock(h);
|
||
return ESP_ERR_INVALID_STATE;
|
||
}
|
||
|
||
// 如果之前已有存储但未中继(STORED)或刚中继完(COMPLETE),先清理再开始新存储
|
||
if (h && (m_sysState == OTA_SYS_GD32_STORED || m_sysState == OTA_SYS_GD32_COMPLETE)) {
|
||
DEBUG_INFO(TAG, "清理旧 GD32 存储状态,开始新存储 (size=%" PRIu32 ")", total_size);
|
||
OtaRelay::metaClear();
|
||
}
|
||
|
||
if (h) {
|
||
m_sysState = OTA_SYS_GD32_STORING;
|
||
m_gd32StorageDone = false;
|
||
m_busySinceTick = xTaskGetTickCount(); // ★ 记录忙状态起始时刻
|
||
DEBUG_INFO(TAG, "GD32 固件存储开始: %" PRIu32 " 字节, 系统状态 → GD32_STORING", total_size);
|
||
}
|
||
unlock(h);
|
||
return h ? ESP_OK : ESP_ERR_INVALID_STATE;
|
||
}
|
||
|
||
void OtaCoordinator::completeGd32Storage()
|
||
{
|
||
bool h = lock();
|
||
if (h && m_sysState == OTA_SYS_GD32_STORING) {
|
||
m_sysState = OTA_SYS_GD32_STORED;
|
||
m_gd32StorageDone = true;
|
||
m_busySinceTick = 0; // ★ STORED 是非忙状态
|
||
DEBUG_INFO(TAG, "GD32 固件存储完成, 系统状态 → GD32_STORED");
|
||
} else if (h) {
|
||
DEBUG_WARN(TAG, "completeGd32Storage 在非预期状态: %d", (int)m_sysState);
|
||
}
|
||
unlock(h);
|
||
}
|
||
|
||
// ============================================================
|
||
// GD32 中继阶段
|
||
// ============================================================
|
||
esp_err_t OtaCoordinator::triggerGd32Relay()
|
||
{
|
||
bool h = lock();
|
||
|
||
// ★ 卡死自愈检查
|
||
if (h) watchdogCheckLocked();
|
||
|
||
// 校验:必须处于"已存储"或"已中继完成"状态
|
||
if (h && m_sysState != OTA_SYS_GD32_STORED && m_sysState != OTA_SYS_IDLE) {
|
||
if (m_sysState == OTA_SYS_GD32_STORING) {
|
||
DEBUG_WARN(TAG, "GD32 正在存储中,拒绝触发中继");
|
||
unlock(h);
|
||
return ESP_ERR_INVALID_STATE;
|
||
}
|
||
if (m_sysState == OTA_SYS_GD32_RELAYING) {
|
||
DEBUG_WARN(TAG, "GD32 中继已在进行中");
|
||
unlock(h);
|
||
return ESP_ERR_INVALID_STATE;
|
||
}
|
||
if (m_sysState == OTA_SYS_ESP32_RECEIVING ||
|
||
m_sysState == OTA_SYS_ESP32_FINALIZING ||
|
||
m_sysState == OTA_SYS_ESP32_REBOOTING) {
|
||
DEBUG_WARN(TAG, "ESP32 OTA 进行中,拒绝中继触发");
|
||
unlock(h);
|
||
return ESP_ERR_INVALID_STATE;
|
||
}
|
||
}
|
||
|
||
// 验证固件是否确实已存储(通过 NVS 元数据)
|
||
ota_meta_t meta;
|
||
if (h && (OtaRelay::metaLoad(&meta) != ESP_OK ||
|
||
meta.total_size == 0 ||
|
||
meta.stored_size < meta.total_size)) {
|
||
DEBUG_WARN(TAG, "GD32 固件未完整存储,拒绝触发中继");
|
||
unlock(h);
|
||
return ESP_ERR_INVALID_STATE;
|
||
}
|
||
|
||
// 检查 OtaRelay 是否空闲
|
||
if (h && OtaRelay::getState() != RELAY_IDLE &&
|
||
OtaRelay::getState() != RELAY_DONE &&
|
||
OtaRelay::getState() != RELAY_ERROR) {
|
||
DEBUG_WARN(TAG, "OtaRelay 正忙 (state=%d),拒绝触发", (int)OtaRelay::getState());
|
||
unlock(h);
|
||
return ESP_ERR_INVALID_STATE;
|
||
}
|
||
|
||
// 向 OtaRelay task 发送启动消息
|
||
if (h && !OtaRelay::dispatchQueue) {
|
||
DEBUG_ERR(TAG, "OtaRelay dispatchQueue 未就绪");
|
||
unlock(h);
|
||
return ESP_ERR_INVALID_STATE;
|
||
}
|
||
|
||
if (!h) {
|
||
unlock(h);
|
||
return ESP_ERR_INVALID_STATE;
|
||
}
|
||
|
||
m_sysState = OTA_SYS_GD32_RELAYING;
|
||
m_busySinceTick = xTaskGetTickCount(); // ★ 记录忙状态起始时刻
|
||
DEBUG_INFO(TAG, "触发 GD32 中继, 系统状态 → GD32_RELAYING");
|
||
|
||
uint32_t cmd = 1;
|
||
BaseType_t ret = xQueueSend(OtaRelay::dispatchQueue, &cmd, pdMS_TO_TICKS(200));
|
||
if (ret != pdTRUE) {
|
||
DEBUG_ERR(TAG, "向 OtaRelay 发送启动消息失败");
|
||
m_sysState = OTA_SYS_GD32_STORED;
|
||
m_busySinceTick = 0; // 回退到 STORED(非忙)
|
||
unlock(h);
|
||
return ESP_FAIL;
|
||
}
|
||
|
||
unlock(h);
|
||
return ESP_OK;
|
||
}
|
||
|
||
void OtaCoordinator::notifyRelayStarted()
|
||
{
|
||
bool h = lock();
|
||
if (h && m_sysState != OTA_SYS_GD32_RELAYING) {
|
||
m_sysState = OTA_SYS_GD32_RELAYING;
|
||
m_busySinceTick = xTaskGetTickCount(); // ★ 记录起始时刻
|
||
DEBUG_INFO(TAG, "中继已开始, 系统状态 → GD32_RELAYING");
|
||
}
|
||
unlock(h);
|
||
}
|
||
|
||
void OtaCoordinator::notifyRelayComplete(bool success)
|
||
{
|
||
bool h = lock();
|
||
if (h) {
|
||
if (success) {
|
||
m_sysState = OTA_SYS_GD32_COMPLETE;
|
||
m_gd32StorageDone = false;
|
||
m_busySinceTick = 0; // ★ COMPLETE 非忙
|
||
DEBUG_INFO(TAG, "GD32 中继完成, 系统状态 → GD32_COMPLETE");
|
||
} else {
|
||
m_sysState = OTA_SYS_ERROR;
|
||
m_busySinceTick = 0; // ★ ERROR 非忙(可被后续请求恢复)
|
||
DEBUG_ERR(TAG, "GD32 中继失败, 系统状态 → ERROR");
|
||
}
|
||
}
|
||
unlock(h);
|
||
}
|
||
|
||
// ============================================================
|
||
// ESP32 OTA 阶段
|
||
// ============================================================
|
||
esp_err_t OtaCoordinator::beginEsp32Ota(uint32_t total_size)
|
||
{
|
||
bool h = lock();
|
||
|
||
// ★ 卡死自愈检查
|
||
if (h) watchdogCheckLocked();
|
||
|
||
// 系统忙检查
|
||
if (h && (m_sysState == OTA_SYS_GD32_STORING ||
|
||
m_sysState == OTA_SYS_GD32_RELAYING)) {
|
||
DEBUG_WARN(TAG, "GD32 正在 %s,拒绝 ESP32 OTA",
|
||
m_sysState == OTA_SYS_GD32_STORING ? "存储" : "中继");
|
||
unlock(h);
|
||
return ESP_ERR_INVALID_STATE;
|
||
}
|
||
|
||
esp_err_t err = ESP_ERR_INVALID_STATE;
|
||
if (h) {
|
||
// 委托 Esp32OtaCore 处理分区/互斥逻辑
|
||
err = Esp32OtaCore::instance().begin(total_size);
|
||
if (err == ESP_OK) {
|
||
m_sysState = OTA_SYS_ESP32_RECEIVING;
|
||
m_busySinceTick = xTaskGetTickCount(); // ★ 记录起始时刻
|
||
DEBUG_INFO(TAG, "ESP32 OTA 开始: %" PRIu32 " 字节, 系统状态 → ESP32_RECEIVING", total_size);
|
||
} else {
|
||
DEBUG_WARN(TAG, "ESP32 OTA begin 失败: %s", esp_err_to_name(err));
|
||
}
|
||
}
|
||
|
||
unlock(h);
|
||
return err;
|
||
}
|
||
|
||
void OtaCoordinator::notifyEsp32ReceiveComplete()
|
||
{
|
||
bool h = lock();
|
||
if (h && m_sysState == OTA_SYS_ESP32_RECEIVING) {
|
||
m_sysState = OTA_SYS_ESP32_FINALIZING;
|
||
DEBUG_INFO(TAG, "ESP32 固件接收完毕, 系统状态 → ESP32_FINALIZING");
|
||
}
|
||
unlock(h);
|
||
}
|
||
|
||
void OtaCoordinator::notifyEsp32FinalizeComplete()
|
||
{
|
||
bool h = lock();
|
||
if (h) {
|
||
m_sysState = OTA_SYS_ESP32_REBOOTING;
|
||
DEBUG_INFO(TAG, "ESP32 OTA 收尾完成, 系统状态 → ESP32_REBOOTING");
|
||
}
|
||
unlock(h);
|
||
}
|
||
|
||
void OtaCoordinator::notifyEsp32Rebooting()
|
||
{
|
||
bool h = lock();
|
||
if (h) {
|
||
m_sysState = OTA_SYS_ESP32_REBOOTING;
|
||
DEBUG_INFO(TAG, "ESP32 即将重启, 系统状态 → ESP32_REBOOTING");
|
||
}
|
||
unlock(h);
|
||
}
|
||
|
||
// ============================================================
|
||
// 中止
|
||
// ============================================================
|
||
void OtaCoordinator::abortCurrent()
|
||
{
|
||
bool h = lock();
|
||
if (h) {
|
||
DEBUG_INFO(TAG, "中止当前 OTA 操作 (state=%d)", (int)m_sysState);
|
||
|
||
if (m_sysState == OTA_SYS_GD32_STORING) {
|
||
OtaRelay::metaClear();
|
||
} else if (m_sysState == OTA_SYS_GD32_RELAYING) {
|
||
OtaRelay::instance().reset();
|
||
} else if (m_sysState == OTA_SYS_ESP32_RECEIVING ||
|
||
m_sysState == OTA_SYS_ESP32_FINALIZING) {
|
||
Esp32OtaCore::instance().abort();
|
||
}
|
||
|
||
m_sysState = OTA_SYS_IDLE;
|
||
m_gd32StorageDone = false;
|
||
m_busySinceTick = 0; // ★ 离开忙状态
|
||
DEBUG_INFO(TAG, "已中止, 系统状态 → IDLE");
|
||
}
|
||
unlock(h);
|
||
}
|
||
|
||
// ============================================================
|
||
// 状态查询
|
||
// ============================================================
|
||
ota_system_state_t OtaCoordinator::getSystemState() const
|
||
{
|
||
// 对于纯读取,短暂加锁保证一致性
|
||
OtaCoordinator* self = const_cast<OtaCoordinator*>(this);
|
||
bool h = self->lock();
|
||
ota_system_state_t s = m_sysState;
|
||
self->unlock(h);
|
||
return s;
|
||
}
|
||
|
||
bool OtaCoordinator::isIdle() const
|
||
{
|
||
return (getSystemState() == OTA_SYS_IDLE ||
|
||
getSystemState() == OTA_SYS_GD32_STORED ||
|
||
getSystemState() == OTA_SYS_GD32_COMPLETE);
|
||
}
|
||
|
||
bool OtaCoordinator::isGd32Busy() const
|
||
{
|
||
ota_system_state_t s = getSystemState();
|
||
return (s == OTA_SYS_GD32_STORING ||
|
||
s == OTA_SYS_GD32_RELAYING);
|
||
}
|
||
|
||
bool OtaCoordinator::isEsp32Busy() const
|
||
{
|
||
ota_system_state_t s = getSystemState();
|
||
return (s == OTA_SYS_ESP32_RECEIVING ||
|
||
s == OTA_SYS_ESP32_FINALIZING ||
|
||
s == OTA_SYS_ESP32_REBOOTING);
|
||
}
|
||
|
||
const char* OtaCoordinator::getSystemStateString() const
|
||
{
|
||
switch (getSystemState()) {
|
||
case OTA_SYS_IDLE: return "IDLE";
|
||
case OTA_SYS_GD32_STORING: return "GD32_STORING";
|
||
case OTA_SYS_GD32_STORED: return "GD32_STORED";
|
||
case OTA_SYS_GD32_RELAYING: return "GD32_RELAYING";
|
||
case OTA_SYS_GD32_COMPLETE: return "GD32_COMPLETE";
|
||
case OTA_SYS_ESP32_RECEIVING: return "ESP32_RECEIVING";
|
||
case OTA_SYS_ESP32_FINALIZING: return "ESP32_FINALIZING";
|
||
case OTA_SYS_ESP32_REBOOTING: return "ESP32_REBOOTING";
|
||
case OTA_SYS_ERROR: return "ERROR";
|
||
default: return "UNKNOWN";
|
||
}
|
||
}
|
||
|
||
bool OtaCoordinator::getGd32AppAlive() const
|
||
{
|
||
return OtaRelay::getGd32AppAlive();
|
||
}
|
||
|
||
bool OtaCoordinator::isGd32StorageDone() const
|
||
{
|
||
OtaCoordinator* self = const_cast<OtaCoordinator*>(this);
|
||
bool h = self->lock();
|
||
bool done = m_gd32StorageDone;
|
||
self->unlock(h);
|
||
return done;
|
||
}
|
||
|
||
// ============================================================
|
||
// 完全复位
|
||
// ============================================================
|
||
void OtaCoordinator::reset()
|
||
{
|
||
bool h = lock();
|
||
if (h) {
|
||
m_sysState = OTA_SYS_IDLE;
|
||
m_gd32StorageDone = false;
|
||
m_busySinceTick = 0; // ★ 复位
|
||
DEBUG_INFO(TAG, "协调器已完全复位");
|
||
}
|
||
unlock(h);
|
||
}
|