Trap for young players - STM32's SPIv2 has data packing

[chrisdf]

I copied board_ILI9341_spi.h from boards/addons/gdisp for my nucleo32 STM32L432 project

Couldn't get the LCD working, so I attached a logic analyser and ever byte that was being written was followed by a 0x00, which was obviously screwing up the init procedure.

Turns out on the newer STM32's "SPI2->DR = data;" is a 16bit write to the DR register, which puts the low byte on the FIFO, followed by the high byte - always zero in this case as it's supposed to be an 8 bit transfer.

This makes the SPI device perform two 8 bit transfers.

I finally found my google-foo this morning after a few wasted hours last night and you need to cast the DR register to a uint8_t when you do the write:

*((uint8_t*)&SPI3->DR) = (uint8_t) data;

The STM32F4's don't do data packing on the DR register, but the newer L4's do. I assume other recent STM32's do as well.

[Joel Bodenmann]

Hello @chrisdf and welcome to the µGFX community!

Thank you very much for sharing your experience here. We appreciate it a lot and I'm sure it will help countless people in the future as well! This is definitely interesting.

[chrisdf]

It was a bit confusing for me, the first half of my project was SPI to an RF module. This isn't an issue when using ChibiOS's SPI driver (that internally uses DMA, not direct register stuff)

The ILI9341 driver doesn't appear to be DMA friendly, only giving the low level single byte method for data transfers so I didn't try it.

Maybe that will be a later task, writing a DMA friendly driver for it. I assume it would just need refactoring the data transfer method to pass an array/offset/length