Using STM32G4 with Zephyr and USB Networking

[JayBeavers]

I am trying to stand up a project using an STM32G4 using Zephyr with USB Device Next Networking. I have a similar project up and running on an nRF52. I'm getting the error on my STM32 project that upon system reset, I get a hard fault in z_cstart() -> ... -> statics_init() -> sys_heap_init(). I've isolated this to be a problem when I turn on CONFIG_NET_L2_ETHERNET=y.  If I set CONFIG_DEBUG=y I see a similar failure upon the call to z_bss_zero().

From what I can tell, the errors are consistent with an error initializing the soc memory.  This is a custom board with a G473CC soc.  I've made a custom zephyr dts for this board with #include <st/g4/stm32g473Xc.dtsi> so afaik the memory definitions should be correct.

I haven't set any memory protections in the option bytes, is there perhaps a default protection in place that comes from the factory that may be causing an issue?

[JayBeavers]

For whatever reason, I don't have the same memory initialization issues when I use a Nucleo-H533RE.  I can't use my Nucleo-G431RB for testing, as it doesn't expose the USB port.

All said though, I never see the Zephyr USB-Device-Next stack initialize on the Nucleo-H533RE.  When I use a Nordic nRF52840-DK, I can see the USB stack come online with:

usbd: msg_cb: USBD message: VBUS ready

However I don't get the same from my H533RE.  From what I can tell, USBD never "really" starts up, though I don't see warnings or errors to say why.

[Sarra.S]

Hello @JayBeavers, 

Could you share the memory regions defined in your DTS file and linker script? 

[JayBeavers]

&flash0 {
	partitions {
		compatible = "fixed-partitions";
		#address-cells = <1>;
		#size-cells = <1>;

		slot0_partition: partition@0 {
			label = "image-0";
			reg = <0x00000000 DT_SIZE_K(252)>;
		};
		/* Set 4Kb of storage at the end of the 256Kb of flash */
		storage_partition: partition@3f000 {
			label = "storage";
			reg = <0x0003f000 DT_SIZE_K(4)>;
		};
	};
};

I'm using the default linker scripts as generated by west's cmake export. 

[JayBeavers]

Any further thoughts?