Question on SAI DMA buffer

[Jansunil]

Hi Team,

I am using a the SAI Peripheral from the STM32L5 family to read data from a master device. I am employing the DMA peripheral to read the data from SAI. I have a buffer of 32kb and I am using the STM board as a slave device. I am using the HAL_SAI_RxCpltCallback() to detect the buffer full status. However, I see that the DMA data is getting overwritten by the time it reaches the Rx complete callback function, unless I stop the DMA transfer inside the Rx complete callback (which is not preferred for the use case that I am working on). This is seen only for a certain frequencies of FS signal.

Here are my debug steps:

STEP-1: Start SAI DMA transfer

STEP-2: Wait for control to get into the HAL_SAI_RxCpltCallback()

STEP-3: Read the data from the DMA buffer within the HAL_SAI_RxCpltCallback()- Here's where I see the data is being overwritten when I read it.

DMA Settings: Circular Mode, 32 bit word transfer

SAI Settings: 32 bit data * 4 slots , slave mode.

Are there any suggestions to circumvent this issue? Any leads on this would be of help. Thanks in advance.

Regards,

Janani Sunil.

[LCE]

Check if that STM32's DMA supports double buffer mode (DBM bit, usually set in DMA's CR register).

If it does support DBM, then you activate that, and in half-complete callback you change the currently unused (check CR's CT bit) destination address register M0AR / M1AR.

If it does not support DBM, you should choose the buffer size more wisely: 32 kB takes a lot time to work on, even if it is only copying data, so you need a smaller buffer which can be worked on in the time it takes new data to arrive (SAI sampling period).

[Jansunil]

Hi @Community member​ - Thank you for the suggestion. However, I have a question on DBM.

I have activated DBM by calling the HAL_DMAEx_MultiBufferStart_IT(). I have submitted two local buffers :

uint32_t buffer1[400];

uint32_t buffer2[400];

Upon the processor's entry to HAL_SAI_RxHalfCpltCallback() after filling up half of the buffer, I see that both the buffers (ie buffer1 and buffer2) are filled up completely.

How is the callback supposed to work in this case?- Does the processor access the half transfer complete callback after receiving every 200 slots of data?

Thanks,

Janani Sunil

[AScha.3]

• dont use : HAL_DMAEx_MultiBufferStart_IT
• just set DMA Settings: Circular Mode
• enable callbacks (in Cube -> project manager -> advanced s. -> SAI enable
• now you get 2 callbacks: half... and full...
• process the calling -> but in time, before the other callback coming.
• thats all.

[Jansunil]

Hi @AScha.3​ - Thank you for your reply.

The above steps work perfect for the single buffer mode.

In case I want to employ a DBM, without using the HAL_DMAEx_MultiBufferStart_IT(), which other way would I tell the processor that the DMA needs to work in double buffer mode? The HAL_DMAEx_MultiBufferStart_IT() accepts two memory addresses with which I can configure both the buffers.

Thanks,

Janani Sunil

[AScha.3]

DMA in Circular Mode can do " double buffer " anyway.

(is the "usual" way to work with dma for continuous data transfers )

the buffer is used in 2 half blocks, thats why you have to use/process both blocks:

1. HAL_SAI_TxHalfCpltCallback() - or ..Rx... for receiving data
2. HAL_SAI_TxCpltCallback()

so set the buffer 2x size , what as a data-block size you want.

memory addresses are simply : block start, block 1/2 size , block end.

understand now ?

[LCE]

AScha.3 is correct, DBM is not needed in most cases.

The Half- / Complete splitting is probably good enough.