Problems: STM32H743ZI two SPIs to receive data with DMA

[MSun.5]

Hello,

I use STM32H743ZI two SPIs with DMA to collect data. The data are OK when only one SPI with DMA is activated. However, when I use two SPIs with DMA to collect data simultaneously, the data collected are not correct.

The DMA setup:Codes to collect data using SPI with DMA:I use SPIs with DMA to collect a triangular signal (16 bit) at ~600KSPS, and save and transfer the data to data[]. If I only use SPI2 with DMA, the collection data[] is,The collection is good.

If I only use SPI3 with DMA, the collection data[] is,The collection is good.

However, if I use both SPI2 with DMA and SPI3 with DMA to collect the signal simultaneously, the collection data[] is,

The collections from both SPI2 and SPI3 are distorted.

[T J]

what is the SPI slave ?

there are 2 slaves on 1 SPI bus ?

how do you account for the nSS pins ? .. there should be two nSS pins

as far as I know, in the older ST processors the automatic nSS functionality didn't work.

certainly not with DMA. maybe it is resolved on the H7.

anyhow, nSS pins, how do you address that ?

[AvaTar]

Not doing HAL, but looks like two different SPI units, with different DMA channels.

The HAL/Cube code, I think, is quite pathetic in handling concurrency.

I suppose one channel blocks out handling of the receive data for the second channel long enough that the next data overwrite the current one.

[waclawek.jan]

> I suppose one channel blocks out handling of the receive data for the second channel l

That isn't very feasible unless the SPIs run at extreme rates, and even then I'm not quite sure this could happen with two SPI/DMAs only.

Can't see more through the haze in my crystall ball.

JW

[AvaTar]

> That isn't very feasible unless the SPIs run at extreme rates, and even then I'm not quite sure this could happen with two SPI/DMAs only.

Not much overlap required for occasional data corruption. Or perhaps two external devices triggered at the same time.

[OP] > I use STM32H743ZI two SPIs with DMA to collect data.

I would have used one SPI interface, and manually managed NSS signals.

> Can't see more through the haze in my crystall ball.

I very much agree.