cdc_uart: support databits, stopbits and parity setup

It's possible for the Windows CDC-ACM driver to ignore the IN endpoint
for long periods of time - multiple frames - if the host application
doesn't consume uart RX data. Boost buffer sizes to compensate.

Also prevent usb_thread from potentially being idle for a tick when
there's work to do.

CMakeLists.txt

@@ -17,6 +17,7 @@ add_executable(picoprobe
         src/cdc_uart.c
         src/get_serial.c
         src/sw_dp_pio.c
+        src/tusb_edpt_handler.c
 )
 
 target_sources(picoprobe PRIVATE

src/cdc_uart.c

@@ -34,8 +34,9 @@
 TaskHandle_t uart_taskhandle;
 TickType_t last_wake, interval = 100;
 
-static uint8_t tx_buf[CFG_TUD_CDC_TX_BUFSIZE];
-static uint8_t rx_buf[CFG_TUD_CDC_RX_BUFSIZE];
+/* Max 1 FIFO worth of data */
+static uint8_t tx_buf[32];
+static uint8_t rx_buf[32];
 // Actually s^-1 so 25ms
 #define DEBOUNCE_MS 40
 static uint debounce_ticks = 5;
@@ -59,6 +60,7 @@ void cdc_uart_init(void) {
 void cdc_task(void)
 {
     static int was_connected = 0;
+    static uint cdc_tx_oe = 0;
     uint rx_len = 0;
 
     // Consume uart fifo regardless even if not connected
@@ -77,6 +79,9 @@ void cdc_task(void)
           rx_led_debounce = debounce_ticks;
 #endif
           written = MIN(tud_cdc_write_available(), rx_len);
+          if (rx_len > written)
+              cdc_tx_oe++;
+
           if (written > 0) {
             tud_cdc_write(rx_buf, written);
             tud_cdc_write_flush();
@@ -113,6 +118,7 @@ void cdc_task(void)
     } else if (was_connected) {
       tud_cdc_write_clear();
       was_connected = 0;
+      cdc_tx_oe = 0;
     }
 }
 
@@ -131,6 +137,8 @@ void cdc_thread(void *ptr)
 
 void tud_cdc_line_coding_cb(uint8_t itf, cdc_line_coding_t const* line_coding)
 {
+  uart_parity_t parity;
+  uint data_bits, stop_bits;
   /* Set the tick thread interval to the amount of time it takes to
    * fill up half a FIFO. Millis is too coarse for integer divide.
    */
@@ -145,6 +153,51 @@ void tud_cdc_line_coding_cb(uint8_t itf, cdc_line_coding_t const* line_coding)
   tud_cdc_write_clear();
   tud_cdc_read_flush();
   uart_init(PICOPROBE_UART_INTERFACE, line_coding->bit_rate);
+
+  switch (line_coding->parity) {
+  case CDC_LINE_CODING_PARITY_ODD:
+    parity = UART_PARITY_ODD;
+    break;
+  case CDC_LINE_CODING_PARITY_EVEN:
+    parity = UART_PARITY_EVEN;
+    break;
+  default:
+    picoprobe_info("invalid parity setting %u\n", line_coding->parity);
+    /* fallthrough */
+  case CDC_LINE_CODING_PARITY_NONE:
+    parity = UART_PARITY_NONE;
+    break;
+  }
+
+  switch (line_coding->data_bits) {
+  case 5:
+  case 6:
+  case 7:
+  case 8:
+    data_bits = line_coding->data_bits;
+    break;
+  default:
+    picoprobe_info("invalid data bits setting: %u\n", line_coding->data_bits);
+    data_bits = 8;
+    break;
+  }
+
+  /* The PL011 only supports 1 or 2 stop bits. 1.5 stop bits is translated to 2,
+   * which is safer than the alternative. */
+  switch (line_coding->stop_bits) {
+  case CDC_LINE_CONDING_STOP_BITS_1_5:
+  case CDC_LINE_CONDING_STOP_BITS_2:
+    stop_bits = 2;
+  break;
+  default:
+    picoprobe_info("invalid stop bits setting: %u\n", line_coding->stop_bits);
+    /* fallthrough */
+  case CDC_LINE_CONDING_STOP_BITS_1:
+    stop_bits = 1;
+  break;
+  }
+
+  uart_set_format(PICOPROBE_UART_INTERFACE, data_bits, stop_bits, parity);
   vTaskResume(uart_taskhandle);
 }
 

src/main.c

@@ -39,6 +39,7 @@
 #include "cdc_uart.h"
 #include "get_serial.h"
 #include "led.h"
+#include "tusb_edpt_handler.h"
 #include "DAP.h"
 
 // UART0 for Picoprobe debug
@@ -53,10 +54,12 @@ static uint8_t RxDataBuffer[CFG_TUD_HID_EP_BUFSIZE];
 #define TUD_TASK_PRIO  (tskIDLE_PRIORITY + 2)
 #define DAP_TASK_PRIO  (tskIDLE_PRIORITY + 1)
 
-static TaskHandle_t dap_taskhandle, tud_taskhandle;
+TaskHandle_t dap_taskhandle, tud_taskhandle;
 
 void usb_thread(void *ptr)
 {
+    TickType_t wake;
+    wake = xTaskGetTickCount();
     do {
         tud_task();
 #ifdef PICOPROBE_USB_CONNECTED_LED
@@ -65,8 +68,9 @@ void usb_thread(void *ptr)
         else
             gpio_put(PICOPROBE_USB_CONNECTED_LED, 0);
 #endif
-        // Trivial delay to save power
-        vTaskDelay(1);
+        // Go to sleep for up to a tick if nothing to do
+        if (!tud_task_event_ready())
+            xTaskDelayUntil(&wake, 1);
     } while (1);
 }
 
@@ -75,22 +79,6 @@ void usb_thread(void *ptr)
 #define tud_vendor_flush(x) ((void)0)
 #endif
 
-void dap_thread(void *ptr)
-{
-    uint32_t resp_len;
-    do {
-        if (tud_vendor_available()) {
-            tud_vendor_read(RxDataBuffer, sizeof(RxDataBuffer));
-            resp_len = DAP_ProcessCommand(RxDataBuffer, TxDataBuffer);
-            tud_vendor_write(TxDataBuffer, resp_len);
-            tud_vendor_flush();
-        } else {
-            // Trivial delay to save power
-            vTaskDelay(1);
-        }
-    } while (1);
-}
-
 int main(void) {
 
     board_init();

src/tusb_config.h

@@ -68,8 +68,15 @@
 #define CFG_TUD_MIDI            0
 #define CFG_TUD_VENDOR          1
 
-#define CFG_TUD_CDC_RX_BUFSIZE 64
-#define CFG_TUD_CDC_TX_BUFSIZE 64
+/*
+ * TX bufsize (actually UART RX) is oversized because the Windows CDC-ACM
+ * driver submits a grand total of _one_ URB at any one time.
+ * This means the application must consume the data before the next IN token
+ * is issued. At high datarates this leads to huge variation in instantaneous
+ * throughput on USB, so a large runway is needed.
+ */
+#define CFG_TUD_CDC_RX_BUFSIZE 128
+#define CFG_TUD_CDC_TX_BUFSIZE 4096
 
 #define CFG_TUD_VENDOR_RX_BUFSIZE 8192
 #define CFG_TUD_VENDOR_TX_BUFSIZE 8192

src/tusb_edpt_handler.c

@@ -0,0 +1,218 @@
+/**
+ * Copyright (c) 2023 Raspberry Pi (Trading) Ltd.
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#include "tusb_edpt_handler.h"
+#include "DAP.h"
+
+static uint8_t itf_num;
+static uint8_t _rhport;
+
+volatile uint32_t _resp_len;
+
+uint8_t _out_ep_addr;
+uint8_t _in_ep_addr;
+
+buffer_t USBRequestBuffer; 
+buffer_t USBResponseBuffer;
+
+void dap_edpt_init(void) {
+
+}
+    
+void dap_edpt_reset(uint8_t __unused rhport)
+{
+    itf_num = 0;
+}
+
+uint16_t dap_edpt_open(uint8_t __unused rhport, tusb_desc_interface_t const *itf_desc, uint16_t max_len)
+{   
+          
+    TU_VERIFY(TUSB_CLASS_VENDOR_SPECIFIC == itf_desc->bInterfaceClass &&
+              PICOPROBE_INTERFACE_SUBCLASS == itf_desc->bInterfaceSubClass &&
+              PICOPROBE_INTERFACE_PROTOCOL == itf_desc->bInterfaceProtocol, 0);
+
+    //  Initialise circular buffer indices
+    USBResponseBuffer.packet_wr_idx = 0;
+    USBResponseBuffer.packet_rd_idx = 0;
+    USBRequestBuffer.packet_wr_idx = 0;
+    USBRequestBuffer.packet_rd_idx = 0;
+
+    // Initialse full/empty flags
+    USBResponseBuffer.wasFull = false;
+    USBResponseBuffer.wasEmpty = true;
+    USBRequestBuffer.wasFull = false;
+    USBRequestBuffer.wasEmpty = true;
+
+    uint16_t const drv_len = sizeof(tusb_desc_interface_t) + (itf_desc->bNumEndpoints * sizeof(tusb_desc_endpoint_t));
+    TU_VERIFY(max_len >= drv_len, 0);
+    itf_num = itf_desc->bInterfaceNumber;
+
+    // Initialising the OUT endpoint
+
+    tusb_desc_endpoint_t *edpt_desc = (tusb_desc_endpoint_t *) (itf_desc + 1);
+    uint8_t ep_addr = edpt_desc->bEndpointAddress;
+
+    _out_ep_addr = ep_addr;
+
+    // The OUT endpoint requires a call to usbd_edpt_xfer to initialise the endpoint, giving tinyUSB a buffer to consume when a transfer occurs at the endpoint
+    usbd_edpt_open(rhport, edpt_desc);
+    usbd_edpt_xfer(rhport, ep_addr, &(USBRequestBuffer.data[USBRequestBuffer.packet_wr_idx][0]), DAP_PACKET_SIZE);
+
+    // Initiliasing the IN endpoint
+
+    edpt_desc++;
+    ep_addr = edpt_desc->bEndpointAddress;
+
+    _in_ep_addr = ep_addr;
+
+    // The IN endpoint doesn't need a transfer to initialise it, as this will be done by the main loop of dap_thread
+    usbd_edpt_open(rhport, edpt_desc);
+
+    return drv_len;
+
+}
+
+bool dap_edpt_control_xfer_cb(uint8_t __unused rhport, uint8_t stage, tusb_control_request_t const *request)
+{   
+    return false;
+}
+
+// Manage USBResponseBuffer (request) write and USBRequestBuffer (response) read indices
+bool dap_edpt_xfer_cb(uint8_t __unused rhport, uint8_t ep_addr, xfer_result_t result, uint32_t xferred_bytes)
+{   
+    const uint8_t ep_dir = tu_edpt_dir(ep_addr);
+
+    if(ep_dir == TUSB_DIR_IN)
+    {   
+        if(xferred_bytes >= 0u && xferred_bytes <= DAP_PACKET_SIZE)
+        {   
+            USBResponseBuffer.packet_rd_idx = (USBResponseBuffer.packet_rd_idx + 1) % DAP_PACKET_COUNT;
+
+            // This checks that the buffer was not empty in DAP thread, which means the next buffer was not queued up for the in endpoint callback
+            // So, queue up the buffer at the new read index, since we expect read to catch up to write at this point.
+            // It is possible for the read index to be multiple spaces behind the write index (if the USB callbacks are lagging behind dap thread), 
+            // so we account for this by only setting wasEmpty to true if the next callback will empty the buffer
+            if(!USBResponseBuffer.wasEmpty)
+            {   
+                usbd_edpt_xfer(rhport, ep_addr, &(USBResponseBuffer.data[USBResponseBuffer.packet_rd_idx][0]), (uint16_t) _resp_len);
+                USBResponseBuffer.wasEmpty = ((USBResponseBuffer.packet_rd_idx + 1) % DAP_PACKET_COUNT == USBResponseBuffer.packet_wr_idx);
+            }        
+
+            //  Wake up DAP thread after processing the callback
+            vTaskResume(dap_taskhandle);
+            return true;
+        }
+        
+        return false;
+
+    } else if(ep_dir == TUSB_DIR_OUT)    {
+
+        if(xferred_bytes >= 0u && xferred_bytes <= DAP_PACKET_SIZE)
+        {   
+            // Only queue the next buffer in the out callback if the buffer is not full
+            // If full, we set the wasFull flag, which will be checked by dap thread                       
+            if(!buffer_full(&USBRequestBuffer))
+            {   
+                USBRequestBuffer.packet_wr_idx = (USBRequestBuffer.packet_wr_idx + 1) % DAP_PACKET_COUNT; 
+                usbd_edpt_xfer(rhport, ep_addr, &(USBRequestBuffer.data[USBRequestBuffer.packet_wr_idx][0]), DAP_PACKET_SIZE);
+                USBRequestBuffer.wasFull = false; 
+            }
+            else {
+                USBRequestBuffer.wasFull = true;
+            }
+
+            //  Wake up DAP thread after processing the callback
+            vTaskResume(dap_taskhandle);
+            return true;
+        }
+
+        return false;
+    }
+    else return false;
+}
+
+void dap_thread(void *ptr)
+{
+    uint8_t DAPRequestBuffer[DAP_PACKET_SIZE];
+    uint8_t DAPResponseBuffer[DAP_PACKET_SIZE];
+
+    do
+    {   
+        while(USBRequestBuffer.packet_rd_idx != USBRequestBuffer.packet_wr_idx)
+        {   
+            // Read a single packet from the USB buffer into the DAP Request buffer
+            memcpy(DAPRequestBuffer, &(USBRequestBuffer.data[USBRequestBuffer.packet_rd_idx]), DAP_PACKET_SIZE);
+            USBRequestBuffer.packet_rd_idx = (USBRequestBuffer.packet_rd_idx + 1) % DAP_PACKET_COUNT; 
+
+            // If the buffer was full in the out callback, we need to queue up another buffer for the endpoint to consume, now that we know there is space in the buffer.
+            if(USBRequestBuffer.wasFull)
+            {   
+                vTaskSuspendAll(); // Suspend the scheduler to safely update the write index
+                USBRequestBuffer.packet_wr_idx = (USBRequestBuffer.packet_wr_idx + 1) % DAP_PACKET_COUNT;
+                usbd_edpt_xfer(_rhport, _out_ep_addr, &(USBRequestBuffer.data[USBRequestBuffer.packet_wr_idx][0]), DAP_PACKET_SIZE);
+                USBRequestBuffer.wasFull = false;
+                xTaskResumeAll();
+            }
+
+            _resp_len = DAP_ProcessCommand(DAPRequestBuffer, DAPResponseBuffer);
+
+
+            //  Suspend the scheduler to avoid stale values/race conditions between threads
+            vTaskSuspendAll();
+
+            if(buffer_empty(&USBResponseBuffer))
+            {
+                memcpy(&(USBResponseBuffer.data[USBResponseBuffer.packet_wr_idx]), DAPResponseBuffer, (uint16_t) _resp_len); 
+                USBResponseBuffer.packet_wr_idx = (USBResponseBuffer.packet_wr_idx + 1) % DAP_PACKET_COUNT; 
+
+                usbd_edpt_xfer(_rhport, _in_ep_addr, &(USBResponseBuffer.data[USBResponseBuffer.packet_rd_idx][0]), (uint16_t) _resp_len);
+            } else {
+
+                memcpy(&(USBResponseBuffer.data[USBResponseBuffer.packet_wr_idx]), DAPResponseBuffer, (uint16_t) _resp_len); 
+                USBResponseBuffer.packet_wr_idx = (USBResponseBuffer.packet_wr_idx + 1) % DAP_PACKET_COUNT; 
+                
+                // The In callback needs to check this flag to know when to queue up the next buffer.
+                USBResponseBuffer.wasEmpty = false;
+            }
+            xTaskResumeAll();
+        }
+        
+        // Suspend DAP thread until it is awoken by a USB thread callback
+        vTaskSuspend(dap_taskhandle);
+
+    } while (1);
+    
+}
+
+usbd_class_driver_t const _dap_edpt_driver =
+{
+    .init = dap_edpt_init,
+    .reset = dap_edpt_reset,
+    .open = dap_edpt_open,
+    .control_xfer_cb = dap_edpt_control_xfer_cb,
+    .xfer_cb = dap_edpt_xfer_cb,
+    .sof = NULL,
+    #if CFG_TUSB_DEBUG >= 2
+        .name = "PICOPROBE ENDPOINT"
+    #endif
+};
+
+// Add the custom driver to the tinyUSB stack
+usbd_class_driver_t const *usbd_app_driver_get_cb(uint8_t *driver_count)
+{
+    *driver_count = 1;
+    return &_dap_edpt_driver;
+}
+
+bool buffer_full(buffer_t *buffer)
+{
+    return ((buffer->packet_wr_idx + 1) % DAP_PACKET_COUNT == buffer->packet_rd_idx);
+}
+
+bool buffer_empty(buffer_t *buffer)
+{
+    return (buffer->packet_wr_idx == buffer->packet_rd_idx);
+}

src/tusb_edpt_handler.h

@@ -0,0 +1,41 @@
+/**
+ * Copyright (c) 2023 Raspberry Pi (Trading) Ltd.
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#ifndef TUSB_EDPT_HANDLER_H
+#define TUSB_EDPT_HANDLER_H
+
+#include "tusb.h"
+
+#include "device/usbd_pvt.h"
+#include "DAP_config.h"
+
+#define PICOPROBE_INTERFACE_SUBCLASS 0x00
+#define PICOPROBE_INTERFACE_PROTOCOL 0x00
+
+typedef struct {
+    uint8_t data[DAP_PACKET_COUNT][DAP_PACKET_SIZE];  
+    volatile uint32_t packet_wr_idx;
+    volatile uint32_t packet_rd_idx;
+    volatile bool wasEmpty;
+    volatile bool wasFull;
+} buffer_t;
+
+extern TaskHandle_t dap_taskhandle, tud_taskhandle;
+
+/* Main DAP loop */
+void dap_thread(void *ptr);
+
+/* Endpoint Handling */
+void picoprobe_edpt_init(void);
+uint16_t picoprobe_edpt_open(uint8_t __unused rhport, tusb_desc_interface_t const *itf_desc, uint16_t max_len);
+bool picoprobe_edpt_control_xfer_cb(uint8_t __unused rhport, uint8_t stage,  tusb_control_request_t const *request);
+bool picoprobe_edpt_xfer_cb(uint8_t __unused rhport, uint8_t ep_addr, xfer_result_t result, uint32_t xferred_bytes);
+
+/* Helper Functions */
+bool buffer_full(buffer_t *buffer);
+bool buffer_empty(buffer_t *buffer);
+
+#endif

src/usb_descriptors.c

@@ -242,4 +242,4 @@ TU_VERIFY_STATIC(sizeof(desc_ms_os_20) == MS_OS_20_DESC_LEN, "Incorrect size");
 uint8_t const * tud_descriptor_bos_cb(void)
 {
   return desc_bos;
-}
+}