合并主线修改

bspMCU/My_print.c

@@ -37,8 +37,8 @@ int USART_Send(char *_pBuffer, uint32_t _iSize)
     int iRet = HAL_UART_Transmit(ptUartUserData->m_uart, (uint8_t *)_pBuffer, _iSize, 100);
     HAL_GPIO_WritePin(RS485_2_DIR_GPIO_Port, RS485_2_DIR_Pin,GPIO_PIN_RESET);
 //    int iRet = HAL_UART_Transmit_DMA(ptUartUserData->m_uart, (uint8_t *)_pBuffer, _iSize);
-
-    return iRet;
+    if (iRet != HAL_OK) return RD_FAILURE;
+    return (int)_iSize;
 }
 
 /*int Dev01check(char *_pBuffer, uint32_t _iSize)

bspMCU/l_uart.c

@@ -24,7 +24,8 @@ int COM1_Send(char *_pBuffer, uint32_t _iSize)
     HAL_GPIO_WritePin(RS485_2_DIR_GPIO_Port, RS485_2_DIR_Pin,GPIO_PIN_SET);
     int iRet = HAL_UART_Transmit(ptUartUserData->m_uart, (uint8_t *)_pBuffer, _iSize, 100);
     HAL_GPIO_WritePin(RS485_2_DIR_GPIO_Port, RS485_2_DIR_Pin,GPIO_PIN_RESET);
-    return iRet;
+    if (iRet != HAL_OK) return RD_FAILURE;
+    return (int)_iSize;
 }
 
 int COM0_Send(char *_pBuffer, uint32_t _iSize)
@@ -45,7 +46,11 @@ static int l_com_new(lua_State *L)
     if (id == 1 || id == 2)
     {    
         TUartUserData *ptUartUserData = UART_userdata_init(LUACOM_ID, baudrate, 0);
-        if (NULL == ptUartUserData) lua_print("com is already used\n");
+        if (NULL == ptUartUserData) 
+        {
+            lua_print("com is already used\n");
+            return g_ptCom1;
+        }
         g_ptCom1 = rd_ComCreate(NULL, NULL, COM1_Send, ptUartUserData->m_buf_size, ptUartUserData);
         UART_IT_init(g_ptCom1);
         u->m_ptComCtrl = g_ptCom1;        

library/common/rd_mempool.c

@@ -307,6 +307,11 @@ WEAK void* Rd_MemPoolMalloc(size_t size)
 
 WEAK void Rd_MemPoolFree(void* ptr) 
 {
+    if (ptr == NULL)
+    {
+        return;
+    }
+
     if (!is_valid_ptr(ptr)) 
     {
         log_e("[%p] MemPool Free Error: Invalid Pointer %p", __builtin_return_address(0), ptr);

peripheral/include/ymodem.h

@@ -64,6 +64,21 @@ enum rym_code
 #define RYM_MAX_ERRORS    ((size_t)5)
 #endif
 
+/* 发送方向: 收到 NAK / 等应答超时时, 重发同一包的最大次数.
+ * 注意 YMODEM 里 NAK 的语义是"这一包没收好, 请重发", 不是失败退出;
+ * 接收端(尤其是刚建完文件/还没就绪时) 头一两次回 NAK 是正常现象. */
+#ifndef RYM_TX_RETRY_MAX
+#define RYM_TX_RETRY_MAX  10
+#endif
+
+/* 发送方向数据块的大小选择:
+ *   1 = 数据阶段一律用 1K(STX) 块 (标准 YMODEM 发送端行为, lrzsz 的 sz 默认如此).
+ *   0 = 按本块实际字节数选择 SOH(128B)/STX(1K).
+ * 对接某些终端(如 Xshell)时它们只接受 STX 数据块, 收到 SOH 数据块会一直回 NAK. */
+#ifndef RYM_TX_BLOCK_1K
+#define RYM_TX_BLOCK_1K   1
+#endif
+
 enum rym_stage
 {
     RYM_STAGE_NONE = 0,

peripheral/ry_sy.c

@@ -8,6 +8,7 @@
 
 #include "ymodem.h"
 #include "rd_stdio.h"
+#include "ringbuffer.h"
 #include <stdlib.h>
 
 #ifndef YMODEM_PATH_MAX
@@ -18,31 +19,68 @@ struct custom_ctx
 {
     struct rym_ctx parent;
     void  *fp;            /* Rd_Fopen 返回的文件句柄 */
-    int    flen;          /* 剩余待接收字节数 */
+    int    flen;          /* 剩余待接收字节数, <0 表示对端未给长度 */
+    int    expect_index;  /* 下一包应有的序号, 0 表示还没收到数据包 */
     char   fpath[YMODEM_PATH_MAX];
 };
 
 /* ═══════════════════════════════════════════════════════════════════════ *
  *          TComCtrl → rym_device_t 桥接                                   *
- *  读: rd_ComRead (内部自带信号量等待, 无协议回调时直接返回原始字节)          *
- *  写: rd_ComSend                                                         *
+ *  读: 直接掏接收环形缓冲区 (见 serial_rym_read, 不走 rd_ComRead)            *
+ *  写: rd_ComSend (底层回调必须按"实际字节数"返回)                          *
  *  无需额外维护信号量                                                      *
  * ═══════════════════════════════════════════════════════════════════════ */
 
+/*
+ * YMODEM 是原始二进制流, 不能借用 rd_ComRead:
+ *   1) m_pfCheck == NULL 时 rd_ComRead 内部 iPackLen 恒为 1, 每次只返回 1 字节,
+ *      再叠加 ymodem 层的 Rd_Delay(1)=1ms, 吞吐被压到 ~1KB/s (115200 下慢 11 倍);
+ *   2) rd_ComRead 在环空时执行 Rd_SemWait (osWaitForever) 永久阻塞,
+ *      使 ymodem 层所有基于 Rd_GetTime() 的超时变成死代码, 对端一停发就永久卡死.
+ * 这里一次取走环里已到达的全部字节; 无数据返回 0, 由上层用
+ * Rd_GetTime()/Rd_Delay(1) 轮询控制节奏与超时退出.
+ */
 static size_t serial_rym_read(void *opaque, uint8_t *buf, size_t len)
 {
     TComCtrl *ptCom = (TComCtrl *)opaque;
-    if (ptCom == NULL || len == 0) return 0;
+    size_t got = 0;
+
+    if (ptCom == NULL || ptCom->m_stComBuf.m_ptRecv == NULL || buf == NULL || len == 0)
+        return 0;
 
-    int ret = rd_ComRead(ptCom, (char *)buf, (uint32_t)len);
-    return (ret > 0) ? (size_t)ret : 0;
+    while (got < len)
+    {
+        char ch;
+        if (rd_RingbufferGetchar(ptCom->m_stComBuf.m_ptRecv, &ch) != 1)
+            break;
+        buf[got++] = (uint8_t)ch;
+    }
+
+    return got;
 }
 
 static size_t serial_rym_write(void *opaque, const uint8_t *buf, size_t len)
 {
     TComCtrl *ptCom = (TComCtrl *)opaque;
-    if (ptCom == NULL) return 0;
-    return (size_t)rd_ComSend(ptCom, (char *)buf, (uint32_t)len);
+    int ret;
+
+    if (ptCom == NULL || buf == NULL || len == 0) return 0;
+
+    /* 返回实际写出的字节数; 失败返回 0 交给 ymodem 层超时处理,
+     * 不能把负数负值强转成 size_t 大数 */
+    ret = rd_ComSend(ptCom, (char *)buf, (uint32_t)len);
+    if (ret <= 0)
+    {
+        /* 发送失败在 ymodem 层只会表现为"对端不回话", 很难定位, 所以直接报出来:
+         *   ret=-2(RD_NULL): TComCtrl 为空或 m_pfSend 未接
+         *   ret=-1(RD_FAILURE): 底层写失败 (上位机侧常见于 COM 口没打开成功/
+         *                        被别的程序占用, 例如 Xshell 还开着)
+         *   ret=0: 一个字莂也没写出去 */
+        log_e("ymodem TX: rd_ComSend failed (ret=%d, len=%u) -> 字节没发出去",
+              ret, (unsigned)len);
+        return 0;
+    }
+    return (size_t)ret;
 }
 
 static void ymodem_build_device(TComCtrl *ptCom, rym_device_t *dev)
@@ -53,6 +91,50 @@ static void ymodem_build_device(TComCtrl *ptCom, rym_device_t *dev)
     dev->opaque = ptCom;
 }
 
+/*
+ * 会话开始前把接收环里的残留字节丢掉.
+ * 否则上一条命令/上一次会话遗留的 'C' (接收端催包) 或 ACK 会顶到前面,
+ * 使 ymodem 的"发送一个包 → 读一个应答"序列整体错位:
+ *   表现就是传输跑了一半报 -118 (RYM_ERR_ACK), 或者会话结束时收不到正确应答.
+ *
+ * 同时必须把 m_sem_Recv 的计数也清零:
+ *   接收路径 (rd_ComRecvProc) 每收一批数据就 Rd_SemPost(m_sem_Recv),
+ *   而 rd_ComRead 的约定是"先等信号量、再读环".
+ *   serial_rym_read 绕过 rd_ComRead 直接掏环, 信号量只进不出, 计数会顶到上限;
+ *   不清零的话, 会话结束后 c:read() 会"信号量立刻等到、环里却没数据",
+ *   阻塞语义被破坏 (空读/假就绪).
+ */
+static void ymodem_flush_rx(TComCtrl *ptCom)
+{
+    char ch;
+    int guard;
+
+    if (ptCom == NULL)
+        return;
+
+    if (ptCom->m_stComBuf.m_ptRecv != NULL)
+    {
+        while (rd_RingbufferGetchar(ptCom->m_stComBuf.m_ptRecv, &ch) == 1)
+        {
+            /* 丢弃 */
+        }
+    }
+
+    /* 把残留的"有数据"信号也吃掉, 恢复"sem>0 ⇒ 环里有数据"的约定.
+     * 计数有上限(init+1), 正常情况下几次就空; guard 只防异常实现死循环.
+     *
+     * 用 0 ms = 纯试探, 不等. 前提是 Rd_SemWaitTimeout 真的遵守超时:
+     *   FreeRTOS: osSemaphoreAcquire(sem, 0)      ✓ 非阻塞
+     *   Win32   : pthread_cond_timedwait 到期时间=当前 ✓ 立即返回
+     *   Linux   : rd_thread.c 的 WEAK 兜底恒返回 -1 ✓ 不会阻, 但也清不掉计数
+     * (Win32 曾把超时丢掉转成阻塞等待, 导致上位机 ry 在这里挂死, 已修) */
+    for (guard = 0; guard < 1024; guard++)
+    {
+        if (Rd_SemWaitTimeout(&ptCom->m_stComBuf.m_sem_Recv, 0) != 0)
+            break;
+    }
+}
+
 /* ═══════════════════════════════════════════════════════════════════════ *
  *          YMODEM 接收回调 (文件写入)                                      *
  * ═══════════════════════════════════════════════════════════════════════ */
@@ -83,27 +165,52 @@ static enum rym_code _rym_recv_begin(struct rym_ctx *ctx, uint8_t *buf, size_t l
     if (cctx->flen == 0)
         cctx->flen = -1;
 
+    cctx->expect_index = 0;   /* 新文件, 序号重新开始 */
+    log_i("ymodem RX: file '%s', declared size=%d", cctx->fpath, cctx->flen);
+
     return RYM_CODE_ACK;
 }
 
 static enum rym_code _rym_recv_data(struct rym_ctx *ctx, uint8_t *buf, size_t len)
 {
     struct custom_ctx *cctx = (struct custom_ctx *)ctx;
+    uint8_t index = ctx->buf[1];       /* 本包序号 */
+    ssize_t ret;
 
     if (cctx->fp == NULL)
         return RYM_CODE_CAN;
 
-    if (cctx->flen == -1)
+    /* 序号去重: 对端重传的包只能回 ACK, 绝不能再写一次.
+     * 否则每次重传都会把同一段数据追加到文件里, 文件会变成
+     * "原内容重复 N 遍" (实测一个 5 字节的文件被撑到了 1490 字节) */
+    if (cctx->expect_index > 0 && cctx->expect_index <= 0xFF)
     {
-        Rd_Mwrite(buf, len, cctx->fp);
+        if (index == (uint8_t)(cctx->expect_index - 1))
+            return RYM_CODE_ACK;             /* 重复包: 已经写过, 只确认 */
+        if (index != (uint8_t)cctx->expect_index)
+            return RYM_CODE_NAK;             /* 跳包/乱序: 不写也不确认, 等对端重传 */
+    }
+
+    if (cctx->flen < 0)
+    {
+        ret = Rd_Mwrite(buf, len, cctx->fp);  /* 对端没给长度, 来多少写多少 */
     }
     else
     {
-        int wlen = (int)(len > (size_t)cctx->flen ? (size_t)cctx->flen : len);
-        Rd_Mwrite(buf, wlen, cctx->fp);
+        /* 严格按包头声明的长度截断, flen 不会变成负数后被当成"无限写" */
+        int wlen = ((int)len > cctx->flen) ? cctx->flen : (int)len;
         cctx->flen -= wlen;
+        ret = Rd_Mwrite(buf, (size_t)wlen, cctx->fp);
+    }
+
+    if (ret < 0)
+    {
+        log_e("ymodem RX data: write failed (ret=%ld), aborting", (long)ret);
+        return RYM_CODE_CAN;                 /* 写入失败: 中止会话 */
     }
 
+    cctx->expect_index = (int)index + 1;     /* 记录下一包应有的序号 */
+
     return RYM_CODE_ACK;
 }
 
@@ -149,7 +256,10 @@ static enum rym_code _rym_send_begin(struct rym_ctx *ctx, uint8_t *buf, size_t l
 
     RD_MEMSET(buf, 0, len);
     const char *fname = _get_path_lastname(cctx->fpath);
+    /* YMODEM 包头第二字段必须是十进制文件长度字符串 (CRC/Checksum 模式由握手字节 C/NAK 协商);
+     * 这里写 "STX" 之类的标识符是非法的, 接收端会把文件长度解析成 0 */
     RD_SPRINTF((char *)buf, "%s%c%ld", fname, '\0', file_size);
+    log_i("ymodem TX: file '%s' declared size=%ld byte(s)", fname, file_size);
 
     return RYM_CODE_SOH;
 }
@@ -173,7 +283,15 @@ static enum rym_code _rym_send_data(struct rym_ctx *ctx, uint8_t *buf, size_t le
         ctx->stage = RYM_STAGE_FINISHING;
     }
 
+#if RYM_TX_BLOCK_1K
+    /* 数据阶段一律用 1K(STX) 块 + 0x1A 填满, 这是标准 YMODEM 发送端的行为
+     * (lrzsz 的 sz 默认就是 1K 块). 部分接收端(含 Xshell)只接受 STX 数据块,
+     * 收到 SOH 数据块会一直回 NAK; 文件尾不足 1K 时靠包头里的长度截断. */
+    return RYM_CODE_STX;
+#else
+    /* 按本块实际字节数选择 128B/1K 块 (旧行为) */
     return (read_size > 128) ? RYM_CODE_STX : RYM_CODE_SOH;
+#endif
 }
 
 static enum rym_code _rym_send_end(struct rym_ctx *ctx, uint8_t *buf, size_t len)
@@ -205,13 +323,18 @@ int rym_download_file(TComCtrl *ptCom, const char *file_path)
         return RD_FAILURE;
     }
     ctx->fp = NULL;
-    RD_STRNCPY(ctx->fpath, file_path, YMODEM_PATH_MAX);
+    RD_STRNCPY(ctx->fpath, file_path, YMODEM_PATH_MAX - 1);
 
     ymodem_build_device(ptCom, &dev);
+    ymodem_flush_rx(ptCom);
     res = rym_recv_on_device(&ctx->parent, &dev,
                              _rym_recv_begin, _rym_recv_data, _rym_recv_end,
                              10);
 
+    /* 会话期间 ISR 一直在 post m_sem_Recv 而没人 acquire,
+     * 结束时再清一次, 保证后续 rd_ComRead 的阻塞语义正常 */
+    ymodem_flush_rx(ptCom);
+
     if (ctx->fp != NULL)
         Rd_Mclose(ctx->fp);
     RD_FREE(ctx);
@@ -230,13 +353,18 @@ int rym_upload_file(TComCtrl *ptCom, const char *file_path)
         return RD_FAILURE;
     }
     ctx->fp = NULL;
-    RD_STRNCPY(ctx->fpath, file_path, YMODEM_PATH_MAX);
+    RD_STRNCPY(ctx->fpath, file_path, YMODEM_PATH_MAX - 1);
 
     ymodem_build_device(ptCom, &dev);
+    ymodem_flush_rx(ptCom);
     res = rym_send_on_device(&ctx->parent, &dev,
                              _rym_send_begin, _rym_send_data, _rym_send_end,
                              10);
 
+    /* 会话期间 ISR 一直在 post m_sem_Recv 而没人 acquire,
+     * 结束时再清一次, 保证后续 rd_ComRead 的阻塞语义正常 */
+    ymodem_flush_rx(ptCom);
+
     if (ctx->fp != NULL)
         Rd_Mclose(ctx->fp);
     RD_FREE(ctx);

peripheral/ymodem.c

@@ -109,9 +109,16 @@ static ssize_t _rym_read_data(struct rym_ctx *ctx, size_t len)
     while (Rd_GetTime() - start < RYM_WAIT_CHR_MS)
     {
         size_t n = ctx->dev.read(ctx->dev.opaque, buf + readlen, len - readlen);
-        readlen += n;
-        if (readlen >= len)
-            return readlen;
+        if (n > 0)
+        {
+            readlen += n;
+            if (readlen >= len)
+                return readlen;
+            /* 还有数据就继续搬, 不要在这里白等 1ms */
+            continue;
+        }
+
+        /* 只有真的没数据可读时才让出 CPU */
         Rd_Delay(1);
     }
     return readlen;
@@ -142,13 +149,29 @@ static int _rym_send_packet(struct rym_ctx *ctx, enum rym_code code, uint8_t ind
     ctx->buf[packetlen - 2] = (uint8_t)(send_crc >> 8);
     ctx->buf[packetlen - 1] = (uint8_t)send_crc & 0xff;
 
+    /* DEBUG: 发包前确认 CRC + 序号正确 */
+    char hex_buf[16];
+    RD_SNPRINTF(hex_buf, sizeof(hex_buf), "%02X:%02X crc=%04X", index, index_inv, send_crc);
+    log_d("ymodem TX pkt: %s [%uB]", hex_buf, (unsigned)packetlen);
+
     size_t writelen = 0;
-    do
+    uint32_t retry = 0;
+
+    while (writelen < packetlen)
     {
         size_t n = ctx->dev.write(ctx->dev.opaque, ctx->buf + writelen, packetlen - writelen);
+        if (n == 0)
+        {
+            /* 底层写回调返回 0 (未写出任何字节) 时不能原地死循环:
+             * 比如它误把 HAL_OK(0) 当作"已写字节数"返回 */
+            if (++retry >= (uint32_t)(RYM_WAIT_CHR_MS / 10u))
+                return RD_FAILURE;
+            Rd_Delay(1);
+            continue;
+        }
         writelen += n;
+        retry = 0;
     }
-    while (writelen < packetlen);
 
     return RD_SUCCESS;
 }
@@ -160,13 +183,9 @@ static void _rym_putchar(struct rym_ctx *ctx, uint8_t code)
 
 static uint8_t _rym_getchar(struct rym_ctx *ctx)
 {
-    uint8_t c;
-    /* rd_ComRead 内部阻塞等待数据, 此处只需重试即可 */
-    while (ctx->dev.read(ctx->dev.opaque, &c, 1) != 1)
-    {
-        Rd_Delay(1);
-    }
-    return c;
+    /* 带超时的等应答: 不能无限等 (对端无响应时旧实现会把 lua 线程永久卡住),
+     * 超时返回 RYM_CODE_NONE, 由调用方按"应答不对"上报 -RYM_ERR_ACK */
+    return (uint8_t)_rym_read_code(ctx, RYM_WAIT_PKG_MS);
 }
 
 static int _rym_do_handshake(struct rym_ctx *ctx, int tm_sec)
@@ -175,12 +194,13 @@ static int _rym_do_handshake(struct rym_ctx *ctx, int tm_sec)
     size_t i;
     uint16_t recv_crc, cal_crc;
     size_t data_sz = 0;
-    uint32_t start, deadline;
 
     ctx->stage = RYM_STAGE_ESTABLISHING;
+    log_i("ymodem RX handshake: start polling, %d rounds x %u ms", tm_sec, (unsigned)RYM_CHD_INTV_MS);
     for (i = 0; i < (size_t)tm_sec; i++)
     {
         _rym_putchar(ctx, RYM_CODE_C);
+        log_i("ymodem RX handshake: sent 'C' #%u, waiting header...", (unsigned)(i + 1));
         code = _rym_read_code(ctx, RYM_CHD_INTV_MS);
         if (code == RYM_CODE_SOH)
         {
@@ -192,33 +212,51 @@ static int _rym_do_handshake(struct rym_ctx *ctx, int tm_sec)
             data_sz = _RYM_STX_PKG_SZ;
             break;
         }
+        else if (code != RYM_CODE_NONE)
+        {
+            /* 有字节但不是包头: 说明链路通, 问题在协议/对端 */
+            log_w("ymodem RX handshake: got 0x%02X while waiting header", (unsigned)code);
+        }
     }
     if (i == (size_t)tm_sec)
-        return -RYM_ERR_TMO;
-
-    /* 接收剩余数据 */
-    i = 0;
-    start = Rd_GetTime();
-    deadline = start + 100;
-    while (Rd_GetTime() <= deadline && i < (data_sz - 1))
     {
-        i += _rym_read_data(ctx, data_sz - 1);
-        Rd_Delay(5);
+        log_e("ymodem RX handshake: no packet header in %d rounds x %u ms (一个字节都没收到 -> 查端口/接线)",
+              tm_sec, (unsigned)RYM_CHD_INTV_MS);
+        return -RYM_ERR_TMO;
     }
 
+    /* 接收剩余数据: 必须一次读完.
+     * _rym_read_data 每次都是从 ctx->buf + 1 的偏移 0 开始写, 分多次调用会互相覆盖,
+     * 而外层用 i 累加长度会误判为"已收齐"; 原来的 100ms 窗口也和
+     * _rym_read_data 内部的 RYM_WAIT_CHR_MS 超时自相矛盾 (根本来不及). */
+    i = _rym_read_data(ctx, data_sz - 1);
     if (i != (data_sz - 1))
+    {
+        log_e("ymodem RX HANDSHAKE: header short read %u/%u", (unsigned)i, (unsigned)(data_sz - 1));
         return -RYM_ERR_DSZ;
+    }
 
     if (ctx->buf[1] != 0 || ctx->buf[2] != 0xFF)
+    {
+        log_e("ymodem RX HANDSHAKE: bad header seq 0x%02X/0x%02X", ctx->buf[1], ctx->buf[2]);
         return -RYM_ERR_SEQ;
+    }
 
     recv_crc = (uint16_t)(*(ctx->buf + data_sz - 2) << 8) | *(ctx->buf + data_sz - 1);
     cal_crc = CRC16(ctx->buf + 3, data_sz - 5);
     if (recv_crc != cal_crc)
+    {
+        log_e("ymodem RX HANDSHAKE: header crc mismatch recv=0x%04X calc=0x%04X", recv_crc, cal_crc);
         return -RYM_ERR_CRC;
+    }
 
     if (ctx->on_begin && ctx->on_begin(ctx, ctx->buf + 3, data_sz - 5) != RYM_CODE_ACK)
+    {
+        log_e("ymodem RX HANDSHAKE: create/open file failed");
         return -RYM_ERR_CAN;
+    }
+
+    log_i("ymodem RX: handshake done");
 
     return RD_SUCCESS;
 }
@@ -228,33 +266,81 @@ static int _rym_do_send_handshake(struct rym_ctx *ctx, int tm_sec)
     enum rym_code code;
     size_t i;
     size_t data_sz;
+    size_t retry;
     uint8_t getc_ack;
 
     ctx->stage = RYM_STAGE_ESTABLISHING;
     data_sz = _RYM_SOH_PKG_SZ;
 
+    log_i("ymodem TX handshake: waiting for receiver's 'C' (%d rounds x %u ms)...",
+          tm_sec, (unsigned)RYM_CHD_INTV_MS);
     for (i = 0; i < (size_t)tm_sec; i++)
     {
         code = _rym_read_code(ctx, RYM_CHD_INTV_MS);
         if (code == RYM_CODE_C)
             break;
+        if (code != RYM_CODE_NONE)
+        {
+            /* 收到别的字节也报出来: 区分"链路完全没数据"与"有数据但不是 'C'" */
+            log_w("ymodem TX handshake: got 0x%02X while waiting 'C'", (unsigned)code);
+        }
     }
     if (i == (size_t)tm_sec)
+    {
+        log_e("ymodem TX handshake: no 'C' in %d rounds x %u ms (一个字节都没收到 -> 查端口/接线)",
+              tm_sec, (unsigned)RYM_CHD_INTV_MS);
         return -RYM_ERR_TMO;
+    }
 
     if (ctx->on_begin && ctx->on_begin(ctx, ctx->buf + 3, data_sz - 5) != RYM_CODE_SOH)
+    {
+        log_e("ymodem TX handshake: build file header failed");
         return -RYM_ERR_CODE;
+    }
 
-    _rym_send_packet(ctx, RYM_CODE_SOH, 0);
+    /* 发包头. 对端可能还没准备好(比如正在建文件)而回 NAK 或干脆不回,
+     * 这两种都是"重发", 不能当失败退出 */
+    for (retry = 0; ; retry++)
+    {
+        _rym_send_packet(ctx, RYM_CODE_SOH, 0);
 
-    getc_ack = _rym_getchar(ctx);
-    if (getc_ack != RYM_CODE_ACK)
-        return -RYM_ERR_ACK;
+        /* 接收端在等包头期间会反复发 'C', 环里可能残留多个;
+         * 先跳过它们, 否则后面读到的"应答"会整体错位 */
+        do
+        {
+            getc_ack = _rym_getchar(ctx);
+        }
+        while (getc_ack == RYM_CODE_C);
+
+        if (getc_ack == RYM_CODE_ACK)
+            break;
+
+        if (getc_ack == RYM_CODE_CAN)
+        {
+            log_e("ymodem TX handshake: receiver cancelled (CAN)");
+            return -RYM_ERR_CAN;
+        }
+
+        if (retry + 1 >= (size_t)RYM_TX_RETRY_MAX)
+        {
+            log_e("ymodem TX HANDSHAKE: header not ACKed after %u tries, last=0x%02X (0x00=timeout)",
+                  (unsigned)RYM_TX_RETRY_MAX, getc_ack);
+            return -RYM_ERR_ACK;
+        }
+
+        log_w("ymodem TX handshake: header %s -> resend (%u/%u)",
+              (getc_ack == RYM_CODE_NAK) ? "NAKed" : "no reply",
+              (unsigned)(retry + 1), (unsigned)RYM_TX_RETRY_MAX);
+    }
 
     getc_ack = _rym_getchar(ctx);
     if (getc_ack != RYM_CODE_C)
+    {
+        log_e("ymodem TX HANDSHAKE: no 'C' after header ACK, got=0x%02X (0x00=timeout)", getc_ack);
         return -RYM_ERR_ACK;
+    }
 
+    log_i("ymodem TX: handshake done");
     ctx->stage = RYM_STAGE_ESTABLISHED;
     return RD_SUCCESS;
 }
@@ -266,10 +352,16 @@ static int _rym_trans_data(struct rym_ctx *ctx, size_t data_sz, enum rym_code *c
 
     size_t i = _rym_read_data(ctx, tsz);
     if (i != tsz)
+    {
+        log_e("ymodem RX data: short read %u/%u", (unsigned)i, (unsigned)tsz);
         return -RYM_ERR_DSZ;
+    }
 
     if ((ctx->buf[1] + ctx->buf[2]) != 0xFF)
+    {
+        log_e("ymodem RX data: bad seq 0x%02X/0x%02X", ctx->buf[1], ctx->buf[2]);
         return -RYM_ERR_SEQ;
+    }
 
     if (ctx->stage == RYM_STAGE_ESTABLISHED && ctx->buf[1] == 0x00)
     {
@@ -281,7 +373,10 @@ static int _rym_trans_data(struct rym_ctx *ctx, size_t data_sz, enum rym_code *c
 
     recv_crc = (uint16_t)(*(ctx->buf + tsz - 1) << 8) | *(ctx->buf + tsz);
     if (recv_crc != CRC16(ctx->buf + 3, data_sz))
+    {
+        log_e("ymodem RX data: crc mismatch recv=0x%04X", recv_crc);
         return -RYM_ERR_CRC;
+    }
 
     if (ctx->on_data)
         *code = ctx->on_data(ctx, ctx->buf + 3, data_sz);
@@ -314,9 +409,12 @@ static int _rym_do_trans(struct rym_ctx *ctx)
             data_sz = 1024;
             break;
         case RYM_CODE_EOT:
+            log_i("ymodem RX: EOT received, data phase done");
             return RD_SUCCESS;
         default:
             errors++;
+            log_w("ymodem RX: unexpected code 0x%02X (error %u/%u)",
+                  (unsigned)code, (unsigned)errors, (unsigned)RYM_MAX_ERRORS);
             if (errors > RYM_MAX_ERRORS)
                 return -RYM_ERR_CODE;
             _rym_putchar(ctx, RYM_CODE_NAK);
@@ -327,6 +425,8 @@ static int _rym_do_trans(struct rym_ctx *ctx)
         if (err != RD_SUCCESS)
         {
             errors++;
+            log_e("ymodem RX: block error %d (error %u/%u)",
+                  err, (unsigned)errors, (unsigned)RYM_MAX_ERRORS);
             if (errors > RYM_MAX_ERRORS)
                 return err;
             _rym_putchar(ctx, RYM_CODE_NAK);
@@ -356,22 +456,63 @@ static int _rym_do_send_trans(struct rym_ctx *ctx)
     uint32_t index = 1;
     uint8_t getc_ack;
     size_t data_sz = _RYM_STX_PKG_SZ;
+    size_t retry;
 
     while (1)
     {
         if (!ctx->on_data)
             return -RYM_ERR_CODE;
 
+        /* 注意: 一个块只读一次文件. 下面重传时必须直接复用 ctx->buf,
+         * 不能再调 on_data, 否则每重传一次就往后多吃一段数据, 文件就乱了 */
         code = ctx->on_data(ctx, ctx->buf + 3, data_sz - 5);
-        _rym_send_packet(ctx, code, index);
-        index++;
+        log_i("ymodem TX data: block %u sending (code=0x%02X%s)",
+              (unsigned)(uint8_t)index, (unsigned)code,
+              (ctx->stage == RYM_STAGE_FINISHING) ? ", last block" : "");
 
-        getc_ack = _rym_getchar(ctx);
-        if (getc_ack != RYM_CODE_ACK)
-            return -RYM_ERR_ACK;
+        for (retry = 0; ; retry++)
+        {
+            _rym_send_packet(ctx, code, index);
+
+            /* 接收端偶尔会在数据阶段夹发 'C' 催包, 直接跳过, 否则应答会整体错位 */
+            do
+            {
+                getc_ack = _rym_getchar(ctx);
+                if (getc_ack == RYM_CODE_C)
+                    log_w("ymodem TX data: skip stray 'C' while waiting ACK of block %u",
+                          (unsigned)(uint8_t)index);
+            }
+            while (getc_ack == RYM_CODE_C);
+
+            if (getc_ack == RYM_CODE_ACK)
+                break;
+
+            if (getc_ack == RYM_CODE_CAN)
+            {
+                log_e("ymodem TX data: receiver cancelled (CAN) at block %u", (unsigned)(uint8_t)index);
+                return -RYM_ERR_CAN;
+            }
+
+            /* NAK / 超时 = "这一块没收好, 请重发" (复用同一块数据) */
+            if (retry + 1 >= (size_t)RYM_TX_RETRY_MAX)
+            {
+                log_e("ymodem TX DATA PHASE: block %u not ACKed after %u tries, last=0x%02X (0x00=timeout, 0x15=NAK, 0x18=CAN)",
+                      (unsigned)(uint8_t)index, (unsigned)RYM_TX_RETRY_MAX, getc_ack);
+                return -RYM_ERR_ACK;
+            }
+
+            log_w("ymodem TX data: block %u %s -> resend (%u/%u)",
+                  (unsigned)(uint8_t)index, (getc_ack == RYM_CODE_NAK) ? "NAKed" : "no reply",
+                  (unsigned)(retry + 1), (unsigned)RYM_TX_RETRY_MAX);
+        }
+
+        index++;
 
         if (ctx->stage == RYM_STAGE_FINISHING)
+        {
+            log_i("ymodem TX: all data blocks sent (%u block(s))", (unsigned)index - 1);
             break;
+        }
     }
 
     return RD_SUCCESS;
@@ -391,7 +532,10 @@ static int _rym_do_fin(struct rym_ctx *ctx)
     _rym_putchar(ctx, RYM_CODE_NAK);
     code = _rym_read_code(ctx, RYM_WAIT_PKG_MS);
     if (code != RYM_CODE_EOT)
+    {
+        log_e("ymodem RX FIN phase: expected EOT, got 0x%02X (0x00=timeout)", (unsigned)code);
         return -RYM_ERR_CODE;
+    }
 
     _rym_putchar(ctx, RYM_CODE_ACK);
     _rym_putchar(ctx, RYM_CODE_C);
@@ -425,6 +569,7 @@ static int _rym_do_fin(struct rym_ctx *ctx)
 
     ctx->stage = RYM_STAGE_FINISHED;
     _rym_putchar(ctx, RYM_CODE_ACK);
+    log_i("ymodem RX: session finished");
     return RD_SUCCESS;
 }
 
@@ -432,28 +577,55 @@ static int _rym_do_send_fin(struct rym_ctx *ctx)
 {
     size_t data_sz;
     uint8_t index = 0;
+    size_t retry;
     uint8_t getc_ack;
 
     data_sz = _RYM_SOH_PKG_SZ;
 
-    _rym_putchar(ctx, RYM_CODE_EOT);
-    getc_ack = _rym_getchar(ctx);
-    if (getc_ack != RYM_CODE_NAK)
-        return -RYM_ERR_ACK;
+    /* EOT 阶段: 对端回 NAK 表示"请再发一次 EOT", 直接回 ACK 表示收到,
+     * 不回(超时)也重发; 三种情况都不能当失败退出 (有些接收端要两三次才确认) */
+    for (retry = 0; ; retry++)
+    {
+        _rym_putchar(ctx, RYM_CODE_EOT);
+        getc_ack = _rym_getchar(ctx);
 
-    _rym_putchar(ctx, RYM_CODE_EOT);
-    getc_ack = _rym_getchar(ctx);
-    if (getc_ack != RYM_CODE_ACK)
-        return -RYM_ERR_ACK;
+        if (getc_ack == RYM_CODE_ACK)
+            break;
 
-    getc_ack = _rym_getchar(ctx);
-    if (getc_ack != RYM_CODE_C)
-        return -RYM_ERR_ACK;
+        if (getc_ack == RYM_CODE_CAN)
+        {
+            log_e("ymodem TX fin: receiver cancelled (CAN)");
+            return -RYM_ERR_CAN;
+        }
+
+        if (retry + 1 >= (size_t)RYM_TX_RETRY_MAX)
+        {
+            log_e("ymodem TX FIN PHASE: EOT not ACKed after %u tries, last=0x%02X (0x00=timeout, 0x06=ACK, 0x15=NAK)",
+                  (unsigned)RYM_TX_RETRY_MAX, getc_ack);
+            return -RYM_ERR_ACK;
+        }
+
+        log_w("ymodem TX fin: EOT %s -> resend (%u/%u)",
+              (getc_ack == RYM_CODE_NAK) ? "NAKed" : "no reply",
+              (unsigned)(retry + 1), (unsigned)RYM_TX_RETRY_MAX);
+    }
+
+    /* 等接收端的 'C' (表示准备接收下一个文件).
+     * 不把这一步当致命错误: 数据已经全部发完并被 ACK, 部分接收端不发 'C'
+     * 而是直接等结束帧; 因为这个地方返回 -118 会误报"传输失败"并且结束帧发不出去,
+     * 接收端就会一直等下去 (表现就是"不会自动停止") */
+    (void)_rym_getchar(ctx);
 
     if (ctx->on_end && ctx->on_end(ctx, ctx->buf + 3, data_sz - 5) != RYM_CODE_SOH)
         return -RYM_ERR_CODE;
 
     _rym_send_packet(ctx, RYM_CODE_SOH, index);
+
+    /* 结束帧的 ACK 只做尽力而为的清理: 拿到了就不会残留到下一次会话,
+     * 拿不到 (超时/对端已收工) 也不影响本次传输结果 */
+    (void)_rym_getchar(ctx);
+
+    log_i("ymodem TX: session finished");
     ctx->stage = RYM_STAGE_FINISHED;
     return RD_SUCCESS;
 }