6.18: vc4-kms-v3d causes reboot-loop on RPi3

[HiassofT]

Describe the bug

Booting the 32bit 6.18-rc7 kernel with vc4-kms-v3d enabled in config.txt on a RPi3B+ results in a reboot-loop during vc4 init/probe.

Commenting out the vc4-kms-v3d dtoverlay in config.txt and adding the cma dtoverlay allows the system to boot but then running "dtoverlay vc4-kms-v3d" again results in a reboot.

Steps to reproduce the behaviour

Install RPiOS lite 32bit
rpi-update to rpi-6.18.y

Device (s)

Raspberry Pi 3 Mod. B+

System

OS: RPiOS Trixie lite 32bit, fully updated
firmware:

[    0.292874] raspberrypi-firmware soc:firmware: Attached to firmware from 2025-08-20T17:04:09, variant start
[    0.302884] raspberrypi-firmware soc:firmware: Firmware hash is cd866525580337c0aee4b25880e1f5f9f674fb24

kernel: 6.18-rc7, githash 515f4d9

[    0.000000] Linux version 6.18.0-rc7-v7+ (runner@runnervmwmpaq) (arm-linux-gnueabihf-gcc (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0, GNU ld (GNU Binutils for Ubuntu) 2.42) #1 SMP Mon Nov 24 14:58:52 UTC 2025

Logs

vc4-kms-v3d loaded via config.txt: kms.txt

vc4-kms-v3d loaded later via dtoverlay: kms-dtoverlay.txt

Additional context

See also #7136 (comment) where @pelwell first mentioned this issue

[6by9]

It's a missing clock request.
Reverting 6526402 clk: bcm: rpi: Turn firmware clock on/off when preparing/unpreparing means it loads vc4 OK.

[pelwell]

Not quite. 919d692 has that commit title, but doesn't revert cleanly. Reverting 6526402 first helps, and the double reversion fixes the reboot loop. I've pushed both reversions.

[6by9]

4adc20b added the clock for v3d to bcm2835-rpi-common.dtsi.
That isn't included in the downstream dtb files, so needs fixing up.

[6by9]

Not quite. 919d692 has that commit title, but doesn't revert cleanly. Reverting 6526402 first helps, and the double reversion fixes the reboot loop. I've pushed both reversions.

Gah, copied the wrong bits. Yes I'd reverted both for my initial test to avoid the conflict.

[6by9]

See #7156