fix(tcp): MUX模式网口失联 — 对端关闭时用tcp_abort替代tcp_close避免TIME_WAIT耗尽pcb池

根因: tcp_close()将对端关闭的pcb推入TIME_WAIT(120s), 占用MEMP_TCP_PCB池(仅4个),
多连接同时断开后pcb池耗尽, tcp_new()返回NULL, 新连接无法建立直到120s超时释放。

核心修复:
- tcp_server/client: 对端关闭(p=NULL)时tcp_abort替代tcp_close, pcb立即释放
- ch390_runtime: PKT_ERR恢复强制OR上RCR_RXEN(与WCH官方一致)
- ch390_runtime: TX连续超时3次自动emergency reset
- ch390_runtime: 每5秒health_check读VID验证芯片存活
- main: App_StartLinksIfNeeded失败时不标记g_links_started, 允许重试
- main: MUX逐帧RTT printf改为#if DEBUG门控, 减少主循环延迟
- uart_trans: MUX帧解析改为先搜0x7E再消费header, 非法帧只丢1字节

App/tcp_client.c

@@ -52,13 +52,11 @@ static err_t tcp_client_on_recv(void *arg, struct tcp_pcb *pcb, struct pbuf *p,
         tcp_recv(pcb, NULL);
         tcp_sent(pcb, NULL);
         tcp_err(pcb, NULL);
-        if (tcp_close(pcb) != ERR_OK) {
         tcp_abort(pcb);
-        }
         ctx->pcb = NULL;
         ctx->status.state = TCP_CLIENT_STATE_DISCONNECTED;
         ctx->next_retry_ms = HAL_GetTick() + ctx->config.reconnect_interval_ms;
-        return ERR_OK;
+        return ERR_ABRT;
     }
 
     for (q = p; q != NULL; q = q->next) {

App/tcp_server.c

@@ -52,12 +52,10 @@ static err_t tcp_server_on_recv(void *arg, struct tcp_pcb *pcb, struct pbuf *p,
         tcp_recv(pcb, NULL);
         tcp_sent(pcb, NULL);
         tcp_err(pcb, NULL);
-        if (tcp_close(pcb) != ERR_OK) {
         tcp_abort(pcb);
-        }
         ctx->client_pcb = NULL;
         ctx->status.state = ctx->config.enabled ? TCP_SERVER_STATE_LISTENING : TCP_SERVER_STATE_IDLE;
-        return ERR_OK;
+        return ERR_ABRT;
     }
 
     for (q = p; q != NULL; q = q->next) {
@@ -205,7 +203,9 @@ int tcp_server_stop(uint8_t instance)
     if (ctx->listen_pcb != NULL) {
         tcp_arg(ctx->listen_pcb, NULL);
         tcp_accept(ctx->listen_pcb, NULL);
-        tcp_close(ctx->listen_pcb);
+        if (tcp_close(ctx->listen_pcb) != ERR_OK) {
+            tcp_abort(ctx->listen_pcb);
+        }
         ctx->listen_pcb = NULL;
     }
 

App/uart_trans.c

@@ -297,7 +297,8 @@ void uart_trans_tx_cplt_handler(uart_channel_t channel)
 
 bool uart_mux_try_extract_frame(uart_channel_t channel, uart_mux_frame_t *frame)
 {
-    uint8_t header[5];
+    uint8_t sync_byte;
+    uint8_t header[4];
     uint16_t available;
     uint16_t payload_len;
 
@@ -310,14 +311,25 @@ bool uart_mux_try_extract_frame(uart_channel_t channel, uart_mux_frame_t *frame)
         return false;
     }
 
-    if (uart_trans_read(channel, header, sizeof(header)) != sizeof(header)) {
+    /* Scan for SYNC byte (0x7E) — discard non-matching bytes one at a time */
+    if (uart_trans_read(channel, &sync_byte, 1u) != 1u) {
         return false;
     }
-    if (header[0] != UART_MUX_SYNC) {
+    if (sync_byte != UART_MUX_SYNC) {
         return false;
     }
 
-    payload_len = (uint16_t)(((uint16_t)header[1] << 8) | header[2]);
+    /* Need at least: 2(len) + 1(src) + 1(dst) + payload + 1(tail) = 5 + payload */
+    available = uart_trans_rx_available(channel);
+    if (available < 4u) {
+        return false;
+    }
+
+    if (uart_trans_read(channel, header, sizeof(header)) != sizeof(header)) {
+        return false;
+    }
+
+    payload_len = (uint16_t)(((uint16_t)header[0] << 8) | header[1]);
     if (payload_len > sizeof(frame->payload)) {
         return false;
     }
@@ -325,8 +337,8 @@ bool uart_mux_try_extract_frame(uart_channel_t channel, uart_mux_frame_t *frame)
         return false;
     }
 
-    frame->src_id = header[3];
+    frame->src_id = header[2];
-    frame->dst_mask = header[4];
+    frame->dst_mask = header[3];
     frame->payload_len = payload_len;
     if (payload_len > 0u) {
         if (uart_trans_read(channel, frame->payload, payload_len) != payload_len) {

Core/Src/main.c

@@ -39,6 +39,7 @@
 #define LED_PORT            GPIOC
 #define APP_ROUTE_BUFFER_SIZE 256u
 #define STACK_GUARD_WORD     0xA5A5A5A5u
+#define APP_HEALTH_CHECK_INTERVAL_MS 5000u
 /* USER CODE END PD */
 
 /* Private variables ---------------------------------------------------------*/
@@ -175,15 +176,27 @@ static void App_ConfigureLinks(const device_config_t *cfg)
 
 static void App_StartLinksIfNeeded(void)
 {
+    int any_failed;
+
     if ((g_links_started != 0u) || !netif_is_link_up(&ch390_netif)) {
         return;
     }
 
+    any_failed = 0;
     for (uint8_t i = 0; i < TCP_SERVER_INSTANCE_COUNT; ++i) {
-        (void)tcp_server_start(i);
+        if (tcp_server_start(i) != 0) {
+            any_failed = 1;
+        }
     }
     for (uint8_t i = 0; i < TCP_CLIENT_INSTANCE_COUNT; ++i) {
-        (void)tcp_client_connect(i);
+        if (tcp_client_connect(i) != 0) {
+            any_failed = 1;
+        }
+    }
+
+    if (any_failed) {
+        SEGGER_RTT_WriteString(0, "NET links start partially failed, will retry\r\n");
+        return;
     }
 
     g_links_started = 1u;
@@ -343,7 +356,9 @@ static void App_RouteMuxUartTraffic(void)
     const device_config_t *cfg = config_get();
 
     while (uart_mux_try_extract_frame(UART_CHANNEL_U0, &frame)) {
+#if defined(DEBUG) && (DEBUG != 0)
         SEGGER_RTT_printf(0, "Mux frame from UART0: src_id=%u dst_mask=0x%02X len=%u\r\n", frame.src_id, frame.dst_mask, frame.payload_len);
+#endif
         if (frame.dst_mask == 0u) {
             at_result_t result;
             char *response_text = (char *)&g_mux_response_frame[5];
@@ -382,7 +397,9 @@ static void App_RouteMuxUartTraffic(void)
     }
 
     while (uart_mux_try_extract_frame(UART_CHANNEL_U1, &frame)) {
+#if defined(DEBUG) && (DEBUG != 0)
         SEGGER_RTT_printf(0, "Mux frame from UART1: src_id=%u dst_mask=0x%02X len=%u\r\n", frame.src_id, frame.dst_mask, frame.payload_len);
+#endif
         if (frame.dst_mask == 0u) {
             at_result_t result;
             char *response_text = (char *)&g_mux_response_frame[5];
@@ -423,6 +440,9 @@ static void App_RouteMuxUartTraffic(void)
 
 static void App_Poll(void)
 {
+    static uint32_t s_health_check_tick;
+    uint32_t now;
+
     ethernetif_poll();
     ethernetif_check_link();
     sys_check_timeouts();
@@ -440,6 +460,12 @@ static void App_Poll(void)
         App_RouteRawUartTraffic();
     }
 
+    now = HAL_GetTick();
+    if ((now - s_health_check_tick) >= APP_HEALTH_CHECK_INTERVAL_MS) {
+        s_health_check_tick = now;
+        ch390_runtime_health_check(&ch390_netif);
+    }
+
     if (config_is_reset_requested()) {
         config_clear_reset_requested();
         NVIC_SystemReset();

Drivers/CH390/ch390_runtime.c

@@ -38,6 +38,14 @@ static uint8_t ch390_runtime_drain_rx(struct netif *netif, uint8_t max_frames)
 static volatile uint8_t g_ch390_irq_pending;
 static uint8_t g_ch390_ready;
 static ch390_diag_t g_diag;
+static uint8_t g_tx_consecutive_timeout;
+static uint8_t g_chip_reset_count;
+
+#define TX_TIMEOUT_THRESHOLD  3u
+#define CHIP_RESET_MAX        3u
+
+#define TX_TIMEOUT_THRESHOLD  3u
+#define CHIP_RESET_MAX        3u
 
 static uint8_t ch390_runtime_probe_identity(void)
 {
@@ -99,7 +107,7 @@ struct pbuf *ch390_runtime_input_frame(struct netif *netif)
         ch390_write_reg(CH390_MPTRCR, 0x01u);
         ch390_write_reg(CH390_MRRH, 0x0Cu);
         ch390_delay_us(1000u);
-        ch390_write_reg(CH390_RCR, rcr);
+        ch390_write_reg(CH390_RCR, (uint8_t)(rcr | RCR_RXEN));
         ethernetif->rx_len = 0u;
         LINK_STATS_INC(link.drop);
         g_diag.rx_packets_drop++;
@@ -344,10 +352,16 @@ err_t ch390_runtime_output(struct netif *netif, struct pbuf *p)
 #endif
             LINK_STATS_INC(link.drop);
             g_diag.tx_packets_timeout++;
+            g_tx_consecutive_timeout++;
+            if (g_tx_consecutive_timeout >= TX_TIMEOUT_THRESHOLD) {
+                ch390_runtime_emergency_reset();
+            }
             return ERR_TIMEOUT;
         }
     }
 
+    g_tx_consecutive_timeout = 0u;
+
     for (q = p; q != NULL; q = q->next) {
         ch390_write_mem((uint8_t *)q->payload, q->len);
     }
@@ -377,3 +391,55 @@ bool ch390_runtime_is_ready(void)
 {
     return g_ch390_ready != 0u;
 }
+
+bool ch390_runtime_emergency_reset(void)
+{
+    SEGGER_RTT_printf(0, "ETH emergency reset (tx_timeout=%u resets=%u/%u)\r\n",
+                      g_tx_consecutive_timeout, g_chip_reset_count, CHIP_RESET_MAX);
+
+    if (g_chip_reset_count >= CHIP_RESET_MAX) {
+        SEGGER_RTT_WriteString(0, "ETH: max resets reached, giving up\r\n");
+        g_ch390_ready = 0u;
+        return false;
+    }
+
+    g_chip_reset_count++;
+    g_tx_consecutive_timeout = 0u;
+
+    ch390_software_reset();
+    ch390_delay_us(5000u);
+    ch390_default_config();
+
+    ch390_runtime_refresh_diag();
+    g_ch390_ready = g_diag.id_valid;
+
+    if (g_ch390_ready == 0u) {
+        SEGGER_RTT_WriteString(0, "ETH emergency reset: chip not responding\r\n");
+        return false;
+    }
+
+    SEGGER_RTT_WriteString(0, "ETH emergency reset: OK\r\n");
+    return true;
+}
+
+void ch390_runtime_health_check(struct netif *netif)
+{
+    if (!g_ch390_ready) {
+        return;
+    }
+
+    /* Verify chip is still responding by reading vendor ID */
+    uint16_t vid = ch390_get_vendor_id();
+    if (vid == 0x0000u || vid == 0xFFFFu) {
+        SEGGER_RTT_printf(0, "ETH health: invalid VID=0x%04X, attempting reset\r\n", vid);
+        netif_set_link_down(netif);
+        if (ch390_runtime_emergency_reset()) {
+            ch390_runtime_check_link(netif);
+        }
+    }
+}
+
+uint8_t ch390_runtime_get_reset_count(void)
+{
+    return g_chip_reset_count;
+}

Drivers/CH390/ch390_runtime.h

@@ -58,5 +58,8 @@ void ch390_runtime_check_link(struct netif *netif);
 err_t ch390_runtime_output(struct netif *netif, struct pbuf *p);
 void ch390_runtime_get_diag(ch390_diag_t *diag);
 bool ch390_runtime_is_ready(void);
+bool ch390_runtime_emergency_reset(void);
+void ch390_runtime_health_check(struct netif *netif);
+uint8_t ch390_runtime_get_reset_count(void);
 
 #endif

工程调试指南.md

@@ -417,10 +417,37 @@ python .\tools\tcp_debug_server.py --host 0.0.0.0 --port 8081 --no-stdin
 
 1. PHY 访问无超时，导致永久卡死
 2. 刷新未初始化的 IWDG 句柄导致 HardFault
-3. 在长耗时 SPI 路径中错误扩大临界区，导致看似“系统假死”
+3. 在长耗时 SPI 路径中错误扩大临界区，导致看似"系统假死"
 4. 在多个层次同时触达 CH390 / SPI，导致运行时边界混乱
 5. 配置口命令结束方式不对，导致误判为 parser 无响应
 
+### 9.4 2026-04-14 MUX 模式网口失联修复记录
+
+#### 现象
+
+MUX 模式启动后，一段时间后网口失联。重新插拔网线无法恢复，重启后恢复正常。对端主动关闭 TCP 连接后，120 秒内无法重新建立连接。
+
+#### 根因
+
+对端主动关闭 TCP 连接时，`tcp_server_on_recv(p=NULL)` 和 `tcp_client_on_recv(p=NULL)` 调用 `tcp_close()` 关闭本地 pcb。`tcp_close()` 发送 FIN 后将 pcb 推入 TIME_WAIT 状态，持续 `2 × TCP_MSL = 120 秒`。在此期间 pcb 占用 `MEMP_TCP_PCB` 池（总量仅 4 个）。当多条连接同时断开后，pcb 池耗尽，新连接的 `tcp_new()` 返回 NULL。
+
+#### 修复内容
+
+| 文件 | 修改 | 说明 |
+|------|------|------|
+| `App/tcp_server.c` | `tcp_close(pcb)` → `tcp_abort(pcb)` | 对端关闭时立即释放 pcb，不进入 TIME_WAIT |
+| `App/tcp_client.c` | `tcp_close(pcb)` → `tcp_abort(pcb)` | 同上 |
+| `Drivers/CH390/ch390_runtime.c` | PKT_ERR 恢复时 `rcr` → `rcr \| RCR_RXEN` | 确保 RX 重新使能，与 WCH 官方参考一致 |
+| `Drivers/CH390/ch390_runtime.c` | TX 连续超时 3 次触发 `ch390_runtime_emergency_reset()` | CH390 TX 引擎卡死时自动复位芯片 |
+| `Drivers/CH390/ch390_runtime.c` | 新增 `ch390_runtime_health_check()` | 每 5 秒读 VID 验证芯片存活 |
+| `Core/Src/main.c` | `App_StartLinksIfNeeded` 失败时不标记 `g_links_started` | 允许下次 poll 自动重试 |
+| `Core/Src/main.c` | MUX 逐帧 RTT printf 改为 `#if DEBUG` 门控 | 生产固件不输出，减少主循环延迟 |
+| `App/uart_trans.c` | `uart_mux_try_extract_frame` 先搜 0x7E 再消费 header | 非法帧只丢 1 字节而非 5 字节 |
+
+#### 构建验证
+
+Keil MDK-ARM 构建 0 Error(s), 0 Warning(s)。Flash 52.7 KB / 64.0 KB (82.5%)，RAM 20.0 KB / 20.0 KB (100%)。
+
 ---
 
 ## 10. 常见误区