Regression in BalenaOS 6.4.3: drm.edid_firmware Fails with ENOENT (-2) Due to vc4 Load Before Filesystem Mount

[HakanL]

Description

In BalenaOS 6.4.2 (kernel 6.1.77), the drm.edid_firmware kernel parameter successfully loads an EDID file from /lib/firmware (e.g., rt73.bin.xz as a test), though it fails with an "invalid size" error due to the file’s format. In BalenaOS 6.4.3 (kernel 6.12.1), it fails with "error -2" (ENOENT, "No such file or directory"), indicating the file isn’t found at the time of loading. This regression prevents me from using a custom EDID for my 800x480 TFT display on a Raspberry Pi Compute Module 4. For reference, my hardware doesn't have an EDID eeprom, but in earlier versions of the Linux kernel this worked fine with just the video= parameter, however in 6.12.x this seems to be changed where EDID is expected (I'm not getting anything on my display and 1000s of errors in dmesg, like these vc4-drm gpu: [drm] HDMI Sink doesn't support RGB, something's wrong.).

Analysis of dmesg logs shows that in 6.4.3, the vc4 DRM driver initializes before /lib/firmware is mounted, causing a race condition not present in 6.4.2. I’m using off-the-shelf BalenaOS images and would like this fixed without resorting to a custom build.

Steps to Reproduce

1. Flash a Raspberry Pi CM4 with BalenaOS 6.4.2 or 6.4.3.
2. Edit cmdline.txt to include drm.edid_firmware=rt73.bin.xz (a test file in /lib/firmware; replace with a valid edid.dat for real use).
3. Boot the device and check dmesg for EDID loading messages.

Expected Behavior

• The EDID file should load during boot, as in 6.4.2, applying my 800x480 @ 60Hz display configuration without requiring a manual video= override.
• If the file is invalid (e.g., rt73.bin.xz), it should fail gracefully with a size/format error, not ENOENT.

Actual Behavior

• 6.4.2 (kernel 6.1.77):
  • Filesystem mounts at ~3.2-3.7s (mmcblk0p3, p5, p6).
  • vc4 initializes at ~7.4s, finds rt73.bin.xz, and errors with:

[    7.494731] [drm:edid_load [drm]] ERROR Size of EDID firmware "rt73.bin.xz" is invalid (expected 11264, got 1164)

• 6.4.3 (kernel 6.12.1):
• vc4 initializes at ~0.64-0.92s, before filesystem mounts at ~2.77-3.11s (mmcblk0p2, p5, p6).
• Fails with:

[    0.921901] vc4-drm gpu: Direct firmware load for rt73.bin.xz failed with error -2
[    0.921910] vc4-drm gpu: [drm] ERROR [CONNECTOR:33:HDMI-A-1] Requesting EDID firmware "rt73.bin.xz" failed (err=-2)

• Later (~21.9s), after mounts, it finds the file but flags it as invalid:

[   21.903345] vc4-drm gpu: [drm] ERROR Invalid firmware EDID "rt73.bin.xz"

Root Cause

In 6.4.3, the vc4 driver probes before /lib/firmware is mounted, unlike 6.4.2 where mounts precede vc4 init. This appears to stem from a BalenaOS boot sequence change (e.g., initramfs or mount timing) rather than a kernel change in drm.edid_firmware handling. The faster probe in kernel 6.12.1 may exacerbate this race.

Impact

I rely on a custom EDID (edid.dat) for my 800x480 display. Without this working, I’m forced to:

• Stay on 6.4.2, missing updates.

Request

Can this be fixed in a future BalenaOS release by ensuring /lib/firmware is available before vc4 loads? A potential solution is embedding firmware in the initramfs, as I’d rather not build a custom OS.

Environment

• Device: Raspberry Pi Compute Module 4 Rev 1.1
• BalenaOS Versions: 6.4.2 (kernel 6.1.77), 6.4.3 (kernel 6.12.1)
• Command Line: drm.edid_firmware=rt73.bin.xz video=HDMI-A-1:800x480M-32@60D ...

Logs

Attached:

• dmesg_6.4.2.txt: Successful load with size error.
• dmesg_6.4.3.txt: ENOENT failure due to timing.

Additional Notes

• I tested with rt73.bin.xz (a Wi-Fi firmware in /lib/firmware) to confirm accessibility, not as my real EDID. My actual edid.dat works in 6.4.2 but fails in 6.4.3 due to this issue (file not found, not invalid file).

• Reported as of March 19, 2025.

dmesg (21).txt
dmesg (20).txt

Related Forum post: https://forums.balena.io/t/edid-not-loaded-on-balenaos-6-4-and-later/373899