DMA on QSPI write only works when single stepping through the code

[LKa.1]

I'm using BL-475E-IOTA2 board (STM32L475) which comes with MX25R6435F Flash chip. I'm trying to using Quad SPI to write and read to/from the flash chip.

With free running the code(release mode):

• If I don't use DMA when writing to the flash chip, it can read/write with free running (running in release mode) fine.
• If I use DMA, it always fails when I send a write enable command to the flash chip:

in stm32l4xx_hal_qspi.c, HAL_QSPI_Command() returns HAL_BUSY.

With single stepping through the code, with DMA or no DMA, it writes and reads the flash chip fine.

I tried QSPI clock prescaler of 1,2,3 but it didn't help. Sys clk and HCLK (to DMA) is 80MHz

I also tried adding a 1s delay before and after writing the write enable command to the flash, but it still failed the same way.

What could cause this problem?

[Tesla DeLorean · Accepted solution]

>>What could cause this problem?

NOR Flash devices are relatively slow for Write and Erase cycles, it's a stateful process, and not one tuned to cramming down data.

The write buffer (page) is 256-bytes, writes will need to be decomposed into page aligned, and page sized (max) transactions. The memory will then need to come ready before other non-status commands can be sent.

Make sure the HAL DMA commands actually complete, the functions should return immediatly, and don't block, you'll need to do that via your own waiting/callback mechanics. I doubt DMA is going to be profoundly faster.

[LKa.1]

Thank you! Yep you're right the write and some command operations are slow. After writing data to the flash chip and before calling HAL_QSPI_AutoPolling() to wait for the memory to be ready, I have to wait for QSPI_HandleTypeDef.State to be HAL_QSPI_STATE_READY first (ie when it finishes writing all data).