TMS320F28069: eCAN transmit ack bit still set after being cleared

Francesco Salvatori

Part Number: TMS320F28069

Hi,

i have a problem with my eCAN interrupt service routine (mailbox 0 to 30 are receive mailbox, while mailbox 31 is a transmit mailbox):

interrupt void canISR(void)
{
	struct ECAN_REGS ECanShadow;

	ECanShadow.CANRMP.all = ECanaRegs.CANRMP.all;
	ECanShadow.CANTA.all  = ECanaRegs.CANTA.all;

	ECanShadow.CANGIF0.all = ECanaRegs.CANGIF0.all;

	if (!ECanShadow.CANGIF0.bit.GMIF0)
	{
		// ... do something

		// clear all system interrupts flag
		ECanaRegs.CANGIF0.all = ECanShadow.CANGIF0.all;

		// Acknowledge interrupt to recieve more interrupts from PIE group 9
		PieCtrlRegs.PIEACK.all |= PIEACK_GROUP9;
		return;
	}

	const Uint16 mailBoxNum =  ECanShadow.CANGIF0.bit.MIV0;

	if (mailBoxNum != 31)
	{
		// ... do something
	}
	else
	{
		ECanaRegs.CANTA.all = ECanShadow.CANTA.all;	// this should clear ECanaRegs.CANTA.bit.TA31
	}


	// Acknowledge interrupt to recieve more interrupts from PIE group 9
	PieCtrlRegs.PIEACK.all |= PIEACK_GROUP9;
}

It works fine almost always but in some rare cases CANTA.TA31 is not properly cleared, thus no other eCAN interrupt is ever triggered.

I've found out that when this happens:
- the last triggered interrupt is actually a transmitting mailbox interrupt, in fact CANGIF0.MIV0 = 31 and CANTA.TA31 is set
- CANGIF0.GMIF0 is set and no other bit in CANGIF0 is set (actually, sometimes CANGIF0.TCOF is set too, but this should not trigger any interrupt since CANGIM.TCOM = 0)
- all can error registers are 0
- CANTA.TA31 is still set after the assignment ECanaRegs.CANTA.all = ECanShadow.CANTA.all, which is supposed to clear it.

Furthermore, I'm pretty sure that no one starts trasmitting a message while the previous trasmission is still pending, so i guess that for no reason the eCAN module should set TA31 at the same time the cpu is trying to clear it

I haven't been able to reproduce this issue at will.

The following workaround seems to work fine:


ECanaRegs.CANTA.all = ECanShadow.CANTA.all;
	
do
{
	ECanShadow.CANTA.all = ECanaRegs.CANTA.all;

	if (ECanShadow.CANTA.all)
	{
		ECanaRegs.CANTA.all = ECanShadow.CANTA.all;
	}
}
while (ECanShadow.CANTA.all);

Has someone ever encountered an issue like this? Am i missing something?

Thanks!

over 5 years ago

0 Hareesh Janakiraman over 5 years ago

TI__Guru* 95315 points

Have you taken a look at the Debug tips in http://www.ti.com/lit/spra876 ?

0 Hareesh Janakiraman over 5 years ago in reply to Hareesh Janakiraman

TI__Guru* 95315 points

You have only one mailbox configured for transmission (MBX31). How do you initiate transmission for this mailbox? Do you initiate only after clearing the TA31 bit from the previous transmission? If so, that eliminate the possibility of a race condition where the CPU tries to clear TA31 but the CAN module tries to set it. No, I have not heard/seen a problem like that. An instruction cannot fail to execute randomly.

0 Francesco Salvatori over 5 years ago in reply to Hareesh Janakiraman

Prodigy 30 points

I have a global variable canInterfaceTransmitting, which i didn't show in my previous post for the sake of simplicity.

I set canInterfaceTransmitting = 1 when I start transmitting a message, canInterfaceTransmitting = 0 in my interrupt service routine, when i (try to) clear TA31. I initiate transmission only when canInterfaceTransmitting == 0.

Moreover, i have added to my code two variables, one that counts how many times i've initiated a transmission, and the other ther counts how many times the interrupt service routine was called for MBX31, and they always have the same value. I think this should further exclude the possibility of a race condition.

0 Hareesh Janakiraman over 5 years ago in reply to Francesco Salvatori

TI__Guru* 95315 points

Since you are only using MBX31 for transmission, please modify your code to directly clear that bit in CANTA (and not reading CANTA into a shadow and then writing it back to CANTA).

else
    {
        ECanaRegs.CANTA.all = 0x80000000; // this should clear ECanaRegs.CANTA.bit.TA31
    }

0 Francesco Salvatori over 5 years ago in reply to Hareesh Janakiraman

Prodigy 30 points

I'll try. It may take a while (days, I suppose) to check whether this works or not, since I don't know how to reproduce the issue at will, and the issue itself happens very rarely.

Anyway, thanks for your suggestions.

0 Hareesh Janakiraman over 5 years ago in reply to Francesco Salvatori

TI__Guru* 95315 points

Understand. Very difficult to come up with a "solution" when the problem occurs rarely and cannot be reproduced at will. There may be other extraneous factors at play but ,as I said earlier, an instruction must be carried out every time it is executed. Will await your updates.

C2000™︎ microcontrollers

C2000 microcontrollers forum

TMS320F28069: eCAN transmit ack bit still set after being cleared