Clarification on accessing the bootloader about qmk_firmware

[sigprof]

The intended fuse settings for the atmel-dfu and qmk-dfu bootloaders are described in the ISP flashing guide; however, that guide does not give a detailed explanation of those settings and the corresponding hardware configuration:

• hfuse = 0x99 (the BOOTRST fuse is not programmed), or 0xD9 if you also want to disable JTAG immediately (but the QMK startup code would disable JTAG during early startup anyway).
• efuse = 0xF3 (the HWBE fuse is programmed).
• The PE2/HWB pin is pulled down (preferably with a resistor, so that flashing a firmware which tries to use the E2 pin as output won't overload the pin, but a direct connection to GND could work too — apparently the reference Pro Micro design have it connected like that).
• The RESET pin must not have a capacitor connected, so that RESET does not get pulled low when power is applied.

The original atmel-dfu bootloader cannot handle the configuration with BOOTRST enabled (the chip would be stuck in the bootloader). There is some code in qmk-dfu which attempts to handle BOOTRST, but apparently that code does not work correctly. Instead of using BOOTRST, those bootloaders rely on HWBE (and a somewhat underdocumented detail that the state of the HWB pin does not affect the behavior of the power-on reset).

The requirement to pull the E2 pin down prevents the usage of that pin in the matrix, but you can still use that pin for some tasks that are compatible with the pull-down — e.g., lots of ATmega32U4-based boards use that pin as WS2812_DI_PIN.

[MaxJ345]

That pin (PE2/HWB) is currently not connected to anything in my design. That's probably the source of my problem.

My RESET pin is connected to VCC via a pull-up resistor. Is that fine?

[sigprof]

Yes, having the PE2/HWB pin floating won't work if you intend to enable the HWBE fuse; however, your efuse value also does not enable HWBE.

Having a pull-up on the RESET pin is fine; but some people also add a capacitor “just for a good measure” (and, e.g., STM32 datasheets recommend having a capacitor on the NRST pin to protect against spurious resets), and adding that capacitor breaks startup when the HWBE fuse is enabled (it effectively converts the power-on reset, which normally ignores the state of the HWB pin, into an external reset, which uses the HWB pin).

The caterina bootloader probably does not need the external PE2/HWB connection, because with that bootloader you need to disable the HWBE fuse and enable the BOOTRST fuse instead; however, that bootloader is very annoying for users, so you should not use it if you have any other options.

Maybe it could be possible to fix the lufa-dfu/qmk-dfu code to work correctly with the BOOTRST configuration, but that would require modifying the LUFA code (the bootloader sources actually live there).

[MaxJ345]

I shorted pin 33 (PE2/HWB) to ground. I set the high fuse to 0x99 and extended fuse to 0xF3. My results were:

• Physical reset (shorting the RESET pin to ground) --> bootloader mode
• QK_BOOT keycode --> bootloader mode

That solves my bootloader issue!

What is BOOTRST? What is HWBE? How are they related?

[sigprof]

These are fuse bits, described in the ATmega32U4 datasheet:



(the documentation is not ideal — e.g., it mentions the ALE name for the HWB pin, but ATmega32U4 does not support the ALE function for that pin — that function is present in some other chips though).

[MaxJ345]

Thank you for the help!