STM32L4 SPI transfer using HAL takes a long time (DMA and IT)

[Reneb]

I use a STM32L496 and I'm implementing the SPI1 port for the communication with an external device.

I have used CubeMX to set things up with SPI_NSS_HARD_OUTPUT and an external pull-up on the SPI1_NSS pin. I have connected SPI1_MOSI to SPI1_MISO to read the transmitted data back.

When using HAL_SPI_TransmitReceive_DMA() or HAL_SPI_TransmitReceive_IT() the data is all send and received. The 32 bits are transferred within about 16us as SPI1_SCK is at 2MHz. SPI1_NSS goes low when __HAL_SPI_ENABLE() is used by HAL_SPI_TransmitReceive_DMA() or HAL_SPI_TransmitReceive_IT(). But after this it takes about 200us before SPI1_NSS goes high when  __HAL_SPI_DISABLE() is used from within HAL_SPI_RxCpltCallback().

Why this long delay?

There is no difference between HAL_SPI_TransmitReceive_DMA() and HAL_SPI_TransmitReceive_IT().

When using SPI_NSS_SOFT instead of SPI_NSS_HARD_OUTPUT and make SPI1_NSS low before the use of HAL_SPI_TransmitReceive_DMA() or HAL_SPI_TransmitReceive_IT(), and high when HAL_SPI_GetState() doesn't show SPI1 is still busy, it also shows a delay of about 200us. Why?

Regards.

[Reneb]

This delay of about 200us is caused by the handling of the interrupts as implemented within HAL.

It partly consists of the useless handling of the Half Complete interrupt (DMA_IT_HT), which can't be disabled, as it is set by HAL_DMA_Start_IT() because HAL_SPI_TransmitReceive_DMA()  makes hdma->XferHalfCpltCallback not equal to NULL.

The largest part is simply the handling of the DMA_IT_TC interrupt within HAL.

My own code (as part of the interrupt handling) took less than 4us.