The EK-RA8D2 carries an on-board SEGGER J-Link OB. Both supported host toolchains talk to that same probe over USB / SWD, so the choice between them is reversible at any time.

SEGGER J-Link / Ozone. Proprietary, free for evaluation, with a polished GUI debugger. This is what the project develops against day to day, and it is the path that can program MRAM.

OpenOCD + arm-none-eabi-gdb. GPL, and needs no vendor download – which is the whole point if you cannot install the SEGGER package. It attaches and debugs, but read the MRAM limitation below before relying on it to flash.

Wrapper scripts for both paths live in scripts/dev/ (flash.sh, debug.sh and ozone.sh for SEGGER; openocd_flash.sh and openocd_debug.sh for the other), and the root justfile wraps those in turn. The OpenOCD board config is scripts/dev/openocd/ek-ra8d2.cfg, which documents its own limits inline.

Gotchas

• OpenOCD has no MRAM flash driver. The RA8D2 boots from MRAM at 0x02000000, and there is no upstream OpenOCD driver for the RA-series MRAM controller. RAM-resident loads work today, including load from gdb into SRAM, but persistent MRAM programming falls back to the SEGGER path. This is the one reason the two paths are not interchangeable.
• Cortex-M85 CPUID. OpenOCD recognises the M85 from 0.12 onward. An older build still attaches, but logs "Cortex-M unknown" – easy to misread as a broken connection.
• The secondary M33 is not configured. The board config declares a single target; debugging the M33 means adding a second one.
• DAP / IDCODE values are unconfirmed. The expected-id entries in the board config were never captured from real silicon, and the cfg carries TODO: confirm notes where that matters.
• Linux needs libusb permissions for the on-board probe. The SEGGER udev rules cover it, or add a generic rule for SEGGER's USB vendor ID 1366.

When in doubt during hardware bring-up, use the SEGGER path.