RTOS/EK-TM4C123GXL: SPI driver

Richard Bland

Expert 2050 points

Part Number: EK-TM4C123GXL

Tool/software: TI-RTOS

Hi,

I am implementing a SPI slave receiver. The RTOS driver manages a SPI transaction consisting of a fixed number of data-frames.

In my situation I am receiving an unknown number of frames arriving at unknown times.

I am wondering if it is OK to set the transaction.count to 1 and loop. I have modified the spiloopback example to show what I mean.....

	/* Initialize SPI handle with slave mode */
	SPI_Params_init(&slaveSpiParams);
	slaveSpiParams.mode = SPI_SLAVE;
	slaveSpi = SPI_open(Board_SPI1, &slaveSpiParams);
	if (slaveSpi == NULL) {
		System_abort("Error initializing SPI\n");
	}
	else {
		System_printf("SPI initialized\n");
	}
	
	/* Initialize slave SPI transaction structure */
	slaveTransaction.count = 1;
	slaveTransaction.txBuf = NULL;
	slaveTransaction.rxBuf = (Ptr)slaveRxBuffer;
	
	while(1) {
		/* Initiate SPI transfer */
		transferOK = SPI_transfer(slaveSpi, &slaveTransaction);

		myRxBuffer[inPtr++] = slaveRxBuffer[0];
		
		if(transferOK) {
			/* Print contents of slave receive buffer */
			System_printf("Slave: %s\n", slaveRxBuffer);
		}
		else {
			System_printf("Unsuccessful slave SPI transfer");
		}
	}

Will this be OK or is there a better way of approaching it?

Thanks,

Richard

over 7 years ago

0 Emmanuel Trinidad - Alicea over 7 years ago

TI__Genius 10260 points

Hi Richard,

The approach above will likely satisfy scenarios where bytes of data are transferred infrequently but will possibly have issues when transferring large amounts of data continuously. The reason is that each time you call SPI_transfer() the DMA is configured to perform the data transfer from memory to SPI TX & from SPI RX to memory; then the task is blocked until the transfer is completed. While this does not take a huge amount of time/cycles there is still the possibility that data might be missed between the end of a transfer & setting up the next.

Is there a protocol between master & slave? For example, if the master sends 1 byte which is the amount of data it will send next; the slave can then receive/parse it & prepare for the next, larger, transfer. Having some sort of protocol/handshaking would help alleviate the problem with large continuous transfers.

Let me know if this helps,

-- Emmanuel

0 Richard Bland over 7 years ago in reply to Emmanuel Trinidad - Alicea

Expert 2050 points

Hi Emmanuel,
Thanks yes that's very helpful.
In an earlier project that did not use RTOS, I set up the SPI with uDMA and ping-pong buffering. This was very successful and worked flawlessly with 16bit frames at 5MHz. It was continuously clocked and the protocol exchanged 0s when there was nothing to say. Parsing was easy: throw away 0s, the first non-zero word is the length of the message to follow.
I wanted to set up something similar for this project using RTOS but could not see a way to setup the SPI driver to achieve a "free-running" exchange.
The issue is how much time have I got between SPI_transfer() unblocking/finishing and the next SPI_transfer() call?
If the SSI port is setup to buffer 8 16bit words in it's FIFO and the interface is clocked at 1MHz this is 128uS - plenty of time to empty the rxBuffer. But perhaps SPI_transfer() clears the FIFO on entry?
How do I get to see the code for SPI_transfer()?
I could arrange for a hardware handshake but this adds complexity which I like to avoid!
Many thanks - and Happy Christmas!
Richard

0 cb1_mobile over 7 years ago in reply to Emmanuel Trinidad - Alicea

Guru 117855 points

Emmanuel,

May I memorialize your post as, "quite caring, nicely detailed - terrific!" You've gone (far) beyond simple opinion - have made the effort to "justify your direction" - which makes your guidance especially compelling.

You've identified the poster's "real needs" - which insures a better end result - and assists (others here) as well... Bravo!

0 Richard Bland over 7 years ago in reply to cb1_mobile

Expert 2050 points

cb_1
Whilst I entirely agree with your sentiment, I do not find your post appropriate.
This forum is for technical not emotional support.

0 cb1_mobile over 7 years ago in reply to Richard Bland

Guru 117855 points

Thank you for your "opinion" Richard. May I ask "How you justify such "strong" opinion?"

With the "demise of "LIKE" - there is (little) means left to identify and register appreciation for such, "special efforts."

Clearly Emmanuel's writing demanded such "recognition."

In closing - is the "Emptiness of your complaint" not SO WELL SIGNALED by your, "Happy Christmas" offering? (or is that "technical?")

0 ToddMullanix over 7 years ago in reply to cb1_mobile

TI__Guru* 96960 points

Hi Richard,

We have an enhancement request to do some type of ping-pong buffering with SPI but there is no time-frame for it. Note: you can still use your scheme and just change the code to use Hwi instead of the driverlib to plug the vector table.

Todd

0 Robert Adsett over 7 years ago in reply to Richard Bland

Guru 27665 points

Oh noes, someone complimented someone on the internets! Let me nip that in the bud.

Robert

Let me echo the original compliment, well deserved.

Arm-based microcontrollers

Arm-based microcontrollers forum

RTOS/EK-TM4C123GXL: SPI driver