Sequentiality broken

[wangnick]

ACAN 2.0.0 on Teensy 3.2 (Arduino 1.8.12, Teensyduino 1.51), direct attached TJA1050, mTransmitBufferSize 16.
Repeating every 10 seconds to quickly send 41 CAN frames.
After a few packages of 41 CAN frames, sequentiality is broken.
Sender: CanJsonTest.ino.txt
Receiver (ACAN2515 2.0.2 on Arduino Nano with MCP2515/TJA1050 for what its worth):
acantest01.ino.txt
Receiver output: CanTest.Reception.log
On the wire, and at the receivers TJA1050 digital output towards the MCP:

Note: Increasing mTransmitBuffer to contain all 41 frames avoids the issue, so the problem seems to be some race condition with my application calling can.tryToSend in a very tight loop whilst the transmit buffer is full.

Btw, thanks for the great library. I've built an MQTT CAN hub including a firmware reflasher over CAN for Arduino and Teensy 3/4 on top of it ...

[wangnick]

The issue seems a race condition indeed. I believe it is caused by mailbox 15 completing its transmission and becoming inactive (and scheduling an interrupt) whilst in ACAN::tryToSend past noInterrupts() but before retrieving FLEXCANb_MBn_CS. In this case the calling of the message_isr() is postponed, and rather the current tryToSend CANMessage frame overtakes the ones pending in the transfer buffer into mailbox 15. Upon reenabling interrupts() the postponed message_isr() is called, but now mailbox 15 is not FLEXCAN_MB_CODE_TX_INACTIVE anymore ...

The issue is resolved (tested) by not trying in ACAN::tryToSend to find an available mailbox when the transmit buffer is not empty:

  }else{ // Data
    noInterrupts () ;
    if (!mTransmitBufferCount) { // Don't compete with the transmit buffer for an inactive mailbox (race condition)
      //--- Find an available mailbox
      for (uint32_t index = mFirstTxMailBoxIndex ; (index < MB_COUNT) && !sent ; index++) {
        const uint32_t code = FLEXCAN_get_code (FLEXCANb_MBn_CS (mFlexcanBaseAddress, index)) ;
        if (code == FLEXCAN_MB_CODE_TX_INACTIVE) {
          writeTxRegisters (inMessage, index);
          sent = true ;
        }
      }
    }
    //--- If buffered messages ahead of us awaiting transmission, or no mailboxes available, try to buffer it
    if (!sent) {

Note: In acan-t4 the issue is already resolved in a similar way.

[pierremolinaro]

Thank you very for your message and your bug fix. I will make soon a new release of the ACAN library. Best regards, Pierre Molinaro

…

Le 4 mars 2020 à 23:16, wangnick ***@***.***> a écrit : Issue is resolved in ACAN::tryToSend by not trying to find an available mailbox where the transmit buffer is not empty: }else{ // Data noInterrupts () ; if (!mTransmitBufferCount) { // Don't compete with the transmit buffer for an inactive mailbox (race condition) //--- Find an available mailbox for (uint32_t index = mFirstTxMailBoxIndex ; (index < MB_COUNT) && !sent ; index++) { const uint32_t code = FLEXCAN_get_code (FLEXCANb_MBn_CS (mFlexcanBaseAddress, index)) ; if (code == FLEXCAN_MB_CODE_TX_INACTIVE) { writeTxRegisters (inMessage, index); sent = true ; } } } //--- If buffered messages ahead of us awaiting transmission, or no mailboxes available, try to buffer it if (!sent) { — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1?email_source=notifications&email_token=AEWKZVCRTRK3736CMWC46TLRF3HNLA5CNFSM4LBQQ6Q2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEN2UUIA#issuecomment-594889248>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEWKZVBP46OF52ZIXIQYG6TRF3HNLANCNFSM4LBQQ6QQ>.