STM32H523 ICACHE Breaks Dual COM Port Setup

[LtBrain]

Hello all,

I'd like to preface this by saying that I'm very new to embedded firmware design, and I recently implemented tinyUSB on my board. I pretty much followed the Phil's Lab guide on this, and used the examples provided in the repository as the basis of most of my code. This has worked very well for me so far, being able to send my IMU data over USB to my PC. However, today I decided to turn on ICACHE to gain more performance/speed out of my STM32H chip, and in doing so learned that turning on ICACHE (without any changes to my functional codebase) in both 1 way and 2 way modes would just make my COM ports disappear. Normally this shows up:

However, with the enable of ICACHE, I would initially get nothing after flashing and resetting, and then I would hear a Windows sound afterwards, followed by Device Manager refreshing 2-3 times to give me a USB device that said: "Unknown USB Device" with the message "Device Descriptor Request Failed". Reverting back (disabling ICACHE) fixed all problems and my COM ports come back.

I'm using VSCode for my STM32 development (the project was generated in CubeMX with the CMAKE toolchain). I followed their guide on the website for their STM32 extension, and development has been going decently with this setup.

Here's the links to my hardware (PCB) and my firmware (note that the firmware in the hardware repository is out of date and completely unused):

https://github.com/LtBrain/astrum
https://github.com/LtBrain/astrum-firmware

Any help would be greatly appreciated!

[1MAFF]

Did not test it further, but saw hard faults and "Unknown USB Device" messages as well using the cdc & msc port example.
Disabling the msc port (so only 1 cdc port) fixed the hard fault / USB problem. Both caches ICACHE & DCACHE were enabled.
Using STM32H563.

CDC speeds achieved: 3.6Mbit/s with -O0 and 6.2Mbit/s with -O2

[1MAFF]

Dual CDC example works for me - two ports appear and communcation on port 0 is still working.
Caches enabled and -O2.

[HiFiPhile]

Hi, please test increasing the delay loop of this errata workaround:

tinyusb/src/portable/st/stm32_fsdev/dcd_stm32_fsdev.c

Line 396 in 7c3f597

#ifdef CFG_TUSB_FSDEV_32BIT

[1MAFF]

Did you check if the MCU is still running?
In case you are using the STM32 UniqueID for serial number - reading with enabled ICACHE results in hard fault.
For that you need to configure the MPU & read the UID before enabling ICACHE, STM32H5 is special in that regard.
https://community.st.com/t5/stm32-mcus/how-to-obtain-and-use-the-stm32-96-bit-uid/ta-p/621443

Follwing will trigger a hard fault:

size_t board_get_unique_id(uint8_t id[], size_t max_len) {
	(void) max_len;
	volatile uint32_t * stm32_uuid = (volatile uint32_t *) UID_BASE;
	uint32_t *id32 = (uint32_t *) (uintptr_t) id;
	uint8_t const len = 12;

	id32[0] = stm32_uuid[0];
	id32[1] = stm32_uuid[1];
	id32[2] = stm32_uuid[2];

	return len;
}

[HiFiPhile]

Hum that's interesting

[LtBrain]

Did you check if the MCU is still running? In case you are using the STM32 UniqueID for serial number - reading with enabled ICACHE results in hard fault. For that you need to configure the MPU & read the UID before enabling ICACHE, STM32H5 is special in that regard. https://community.st.com/t5/stm32-mcus/how-to-obtain-and-use-the-stm32-96-bit-uid/ta-p/621443

Follwing will trigger a hard fault:

size_t board_get_unique_id(uint8_t id[], size_t max_len) {
	(void) max_len;
	volatile uint32_t * stm32_uuid = (volatile uint32_t *) UID_BASE;
	uint32_t *id32 = (uint32_t *) (uintptr_t) id;
	uint8_t const len = 12;

	id32[0] = stm32_uuid[0];
	id32[1] = stm32_uuid[1];
	id32[2] = stm32_uuid[2];

	return len;
}

I went in and did the UID function (replaced the one shown in here with something a little bit different) before enabling ICACHE in code, this seems to have worked out as I now see both COM ports. Thanks for the links! (Also thanks to everyone else who contributed their time and effort to help out)

[hathach]

should be fixed by #3593