TX message guaranteed delivery (#421)

* wait for tx slots before clearing nak

* fix bootstub

* Fixed misra

* Cleanup

* Added bulk write test to test USB NAK on bulk CAN messages

* Added automated bulk tx test

* Fixed linter

* Fixed latency test influence

* Added timeout to python API

* Disabled can write timeout in bulk write test

Co-authored-by: Robbe <[email protected]>

board/drivers/can.h

@@ -28,6 +28,7 @@ void can_set_forwarding(int from, int to);
 
 void can_init(uint8_t can_number);
 void can_init_all(void);
+bool can_tx_check_min_slots_free(uint32_t min);
 void can_send(CAN_FIFOMailBox_TypeDef *to_push, uint8_t bus_number, bool skip_tx_hook);
 bool can_pop(can_ring *q, CAN_FIFOMailBox_TypeDef *elem);
 
@@ -107,6 +108,20 @@ bool can_push(can_ring *q, CAN_FIFOMailBox_TypeDef *elem) {
   return ret;
 }
 
+uint32_t can_slots_empty(can_ring *q) {
+  uint32_t ret = 0;
+
+  ENTER_CRITICAL();
+  if (q->w_ptr >= q->r_ptr) {
+    ret = q->fifo_size - 1U - q->w_ptr + q->r_ptr;
+  } else {
+    ret = q->r_ptr - q->w_ptr - 1U;
+  }
+  EXIT_CRITICAL();
+
+  return ret;
+}
+
 void can_clear(can_ring *q) {
   ENTER_CRITICAL();
   q->w_ptr = 0;
@@ -317,6 +332,10 @@ void process_can(uint8_t can_number) {
         CAN->sTxMailBox[0].TDHR = to_send.RDHR;
         CAN->sTxMailBox[0].TDTR = to_send.RDTR;
         CAN->sTxMailBox[0].TIR = to_send.RIR;
+
+        if (can_tx_check_min_slots_free(MAX_CAN_MSGS_PER_BULK_TRANSFER)) {
+          usb_outep3_resume_if_paused();
+        }
       }
     }
 
@@ -405,6 +424,14 @@ void CAN3_TX_IRQ_Handler(void) { process_can(2); }
 void CAN3_RX0_IRQ_Handler(void) { can_rx(2); }
 void CAN3_SCE_IRQ_Handler(void) { can_sce(CAN3); }
 
+bool can_tx_check_min_slots_free(uint32_t min) {
+  return
+    (can_slots_empty(&can_tx1_q) >= min) &&
+    (can_slots_empty(&can_tx2_q) >= min) &&
+    (can_slots_empty(&can_tx3_q) >= min) &&
+    (can_slots_empty(&can_txgmlan_q) >= min);
+}
+
 void can_send(CAN_FIFOMailBox_TypeDef *to_push, uint8_t bus_number, bool skip_tx_hook) {
   if (skip_tx_hook || safety_tx_hook(to_push) != 0) {
     if (bus_number < BUS_MAX) {

board/drivers/usb.h

@@ -23,12 +23,16 @@ typedef union _USB_Setup {
 }
 USB_Setup_TypeDef;
 
+#define MAX_CAN_MSGS_PER_BULK_TRANSFER 4U
+
 void usb_init(void);
 int usb_cb_control_msg(USB_Setup_TypeDef *setup, uint8_t *resp, bool hardwired);
 int usb_cb_ep1_in(void *usbdata, int len, bool hardwired);
 void usb_cb_ep2_out(void *usbdata, int len, bool hardwired);
 void usb_cb_ep3_out(void *usbdata, int len, bool hardwired);
+void usb_cb_ep3_out_complete(void);
 void usb_cb_enumeration_complete(void);
+void usb_outep3_resume_if_paused(void);
 
 // **** supporting defines ****
 
@@ -380,6 +384,7 @@ USB_Setup_TypeDef setup;
 uint8_t usbdata[0x100];
 uint8_t* ep0_txdata = NULL;
 uint16_t ep0_txlen = 0;
+bool outep3_processing = false;
 
 // Store the current interface alt setting.
 int current_int0_alt_setting = 0;
@@ -744,6 +749,7 @@ void usb_irqhandler(void) {
       }
 
       if (endpoint == 3) {
+        outep3_processing = true;
         usb_cb_ep3_out(usbdata, len, 1);
       }
     } else if (status == STS_SETUP_UPDT) {
@@ -816,15 +822,17 @@ void usb_irqhandler(void) {
       #ifdef DEBUG_USB
         puts("  OUT3 PACKET XFRC\n");
       #endif
-      USBx_OUTEP(3)->DOEPTSIZ = (1U << 19) | 0x40U;
-      USBx_OUTEP(3)->DOEPCTL |= USB_OTG_DOEPCTL_EPENA | USB_OTG_DOEPCTL_CNAK;
+      // NAK cleared by process_can (if tx buffers have room)
+      outep3_processing = false;
+      usb_cb_ep3_out_complete();
     } else if ((USBx_OUTEP(3)->DOEPINT & 0x2000) != 0) {
       #ifdef DEBUG_USB
         puts("  OUT3 PACKET WTF\n");
       #endif
       // if NAK was set trigger this, unknown interrupt
-      USBx_OUTEP(3)->DOEPTSIZ = (1U << 19) | 0x40U;
-      USBx_OUTEP(3)->DOEPCTL |= USB_OTG_DOEPCTL_CNAK;
+      // TODO: why was this here? fires when TX buffers when we can't clear NAK
+      // USBx_OUTEP(3)->DOEPTSIZ = (1U << 19) | 0x40U;
+      // USBx_OUTEP(3)->DOEPCTL |= USB_OTG_DOEPCTL_CNAK;
     } else if ((USBx_OUTEP(3)->DOEPINT) != 0) {
       puts("OUTEP3 error ");
       puth(USBx_OUTEP(3)->DOEPINT);
@@ -932,6 +940,13 @@ void usb_irqhandler(void) {
   //USBx->GINTMSK = 0xFFFFFFFF & ~(USB_OTG_GINTMSK_NPTXFEM | USB_OTG_GINTMSK_PTXFEM | USB_OTG_GINTSTS_SOF | USB_OTG_GINTSTS_EOPF);
 }
 
+void usb_outep3_resume_if_paused() {
+  if (!outep3_processing && (USBx_OUTEP(3)->DOEPCTL & USB_OTG_DOEPCTL_NAKSTS) != 0) {
+    USBx_OUTEP(3)->DOEPTSIZ = (1U << 19) | 0x40U;
+    USBx_OUTEP(3)->DOEPCTL |= USB_OTG_DOEPCTL_EPENA | USB_OTG_DOEPCTL_CNAK;
+  }
+}
+
 void OTG_FS_IRQ_Handler(void) {
   NVIC_DisableIRQ(OTG_FS_IRQn);
   //__disable_irq();

board/main.c

@@ -235,6 +235,12 @@ void usb_cb_ep3_out(void *usbdata, int len, bool hardwired) {
   }
 }
 
+void usb_cb_ep3_out_complete() {
+  if (can_tx_check_min_slots_free(MAX_CAN_MSGS_PER_BULK_TRANSFER)) {
+    usb_outep3_resume_if_paused();
+  }
+}
+
 void usb_cb_enumeration_complete() {
   puts("USB enumeration complete\n");
   is_enumerated = 1;

board/pedal/main.c

@@ -76,6 +76,7 @@ void usb_cb_ep3_out(void *usbdata, int len, bool hardwired) {
   UNUSED(len);
   UNUSED(hardwired);
 }
+void usb_cb_ep3_out_complete(void) {}
 void usb_cb_enumeration_complete(void) {}
 
 int usb_cb_control_msg(USB_Setup_TypeDef *setup, uint8_t *resp, bool hardwired) {

board/spi_flasher.h

@@ -110,6 +110,7 @@ void usb_cb_ep3_out(void *usbdata, int len, bool hardwired) {
   UNUSED(len);
   UNUSED(hardwired);
 }
+void usb_cb_ep3_out_complete(void) {}
 
 int is_enumerated = 0;
 void usb_cb_enumeration_complete(void) {

python/__init__.py

@@ -478,7 +478,12 @@ def set_uart_callback(self, uart, install):
 
   # ******************* can *******************
 
-  def can_send_many(self, arr):
+  # The panda will NAK CAN writes when there is CAN congestion. 
+  # libusb will try to send it again, with a max timeout.
+  # Timeout is in ms. If set to 0, the timeout is infinite.
+  CAN_SEND_TIMEOUT_MS = 10
+
+  def can_send_many(self, arr, timeout=CAN_SEND_TIMEOUT_MS):
     snds = []
     transmit = 1
     extended = 4
@@ -501,13 +506,13 @@ def can_send_many(self, arr):
           for s in snds:
             self._handle.bulkWrite(3, s)
         else:
-          self._handle.bulkWrite(3, b''.join(snds))
+          self._handle.bulkWrite(3, b''.join(snds), timeout=timeout)
         break
       except (usb1.USBErrorIO, usb1.USBErrorOverflow):
         print("CAN: BAD SEND MANY, RETRYING")
 
-  def can_send(self, addr, dat, bus):
-    self.can_send_many([[addr, None, dat, bus]])
+  def can_send(self, addr, dat, bus, timeout=CAN_SEND_TIMEOUT_MS):
+    self.can_send_many([[addr, None, dat, bus]], timeout=timeout)
 
   def can_recv(self):
     dat = bytearray()

tests/automated/7_can_loopback.py

@@ -1,6 +1,7 @@
 import os
 import time
 import random
+import threading
 from panda import Panda
 from nose.tools import assert_equal, assert_less, assert_greater
 from .helpers import panda_jungle, start_heartbeat_thread, reset_pandas, time_many_sends, test_all_pandas, test_all_gen2_pandas, clear_can_buffers, panda_connect_and_init
@@ -200,3 +201,44 @@ def test(p_send, p_recv):
   finally:
     # Set back to silent mode
     p.set_safety_mode(Panda.SAFETY_SILENT)
+
+@test_all_pandas
+@panda_connect_and_init
+def test_bulk_write(p):
+  # The TX buffers on pandas is 0x100 in length.
+  NUM_MESSAGES_PER_BUS = 10000
+
+  def flood_tx(panda):
+    print('Sending!')
+    msg = b"\xaa"*4
+    packet = [[0xaa, None, msg, 0], [0xaa, None, msg, 1], [0xaa, None, msg, 2]] * NUM_MESSAGES_PER_BUS
+
+    # Disable timeout
+    panda.can_send_many(packet, timeout=0)
+    print(f"Done sending {3*NUM_MESSAGES_PER_BUS} messages!")
+
+  # Start heartbeat
+  start_heartbeat_thread(p)
+
+  # Set safety mode and power saving
+  p.set_safety_mode(Panda.SAFETY_ALLOUTPUT)
+  p.set_power_save(False)
+
+  # Start transmisson
+  threading.Thread(target=flood_tx, args=(p,)).start()
+
+  # Receive as much as we can in a few second time period
+  rx = []
+  old_len = 0
+  start_time = time.time()
+  while time.time() - start_time < 2 or len(rx) > old_len:
+    old_len = len(rx)
+    rx.extend(panda_jungle.can_recv())
+  print(f"Received {len(rx)} messages")
+
+  # All messages should have been received
+  if len(rx) != 3*NUM_MESSAGES_PER_BUS:
+    Exception("Did not receive all messages!")
+
+  # Set back to silent mode
+  p.set_safety_mode(Panda.SAFETY_SILENT)

tests/bulk_write_test.py

@@ -0,0 +1,38 @@
+#!/usr/bin/env python3
+import time
+import threading
+
+from panda import Panda
+
+# The TX buffers on pandas is 0x100 in length.
+NUM_MESSAGES_PER_BUS = 10000
+
+def flood_tx(panda):
+  print('Sending!')
+  msg = b"\xaa"*4
+  packet = [[0xaa, None, msg, 0], [0xaa, None, msg, 1], [0xaa, None, msg, 2]] * NUM_MESSAGES_PER_BUS
+  panda.can_send_many(packet)
+  print(f"Done sending {3*NUM_MESSAGES_PER_BUS} messages!")
+
+if __name__ == "__main__":
+  serials = Panda.list()
+  if len(serials) != 2:
+    raise Exception("Connect two pandas to perform this test!")
+
+  sender = Panda(serials[0])
+  receiver = Panda(serials[1])
+
+  sender.set_safety_mode(Panda.SAFETY_ALLOUTPUT)
+  receiver.set_safety_mode(Panda.SAFETY_ALLOUTPUT)
+
+  # Start transmisson
+  threading.Thread(target=flood_tx, args=(sender,)).start()
+
+  # Receive as much as we can in a few second time period
+  rx = []
+  old_len = 0
+  start_time = time.time()
+  while time.time() - start_time < 2 or len(rx) > old_len:
+    old_len = len(rx)
+    rx.extend(receiver.can_recv())
+  print(f"Received {len(rx)} messages")