[BUG] 7.1.3/7.1.4 + nvidia-open 610.43.03: plymouthd hangs on quit, holds DRM master -> kwin_wayland 'Device or resource busy' black screen; separate hard freezes ~7s into boot. LTS 6.18.38 unaffected (RTX 5070 Ti)

## Pre-flight checklist

- [x] I have searched existing issues and this is not a duplicate.
- [x] I have read the Contributing Guidelines.
- [x] I have verified the issue is reproducible with the latest available CachyOS kernel (linux-cachyos 7.1.4-1).
- [ ] I have tried to reproduce the issue on Arch Linux's `linux` kernel. *(Not yet tested — see "Upstream check" below.)*

## Upstream / vanilla kernel check

I have not tested with a vanilla/upstream kernel yet. The trigger appears to be the NVIDIA 610.43.03 DRM path on kernel 7.1.x (plymouthd renders the splash through the NVIDIA card and only wedges on 7.1; the identical userspace on linux-cachyos-lts 6.18.38-2 works every time). Possibly related to the known kernel-7.x `screen_info`/fbdev handling problem in 610.43.03 (NVIDIA forum link below). Reporting here because **both stock CachyOS 7.1 kernels are affected out of the box** and another CachyOS user reported the same failure chain on 2026-07-10 (forum link below).

## Kernel variant

linux-cachyos (EEVDF, Clang)

## System information

- CPU: Intel Core Ultra 7 265K (Arrow Lake-S)
- GPU: NVIDIA GeForce RTX 5070 Ti (GB203, rev a1) — 3 monitors on DisplayPort (card1, DP-1/2/3)
- iGPU: Intel Arrow Lake-S graphics (enabled, no outputs connected)
- Motherboard: Gigabyte Z890 AORUS ELITE WIFI7, BIOS F20 (2026-06-11)
- RAM: 64 GB
- Desktop: KDE Plasma (Wayland), plasma-login-manager greeter, Limine bootloader, Plymouth enabled
- Fresh CachyOS install from 2026-07-21
- Packages: `linux-cachyos 7.1.4-1` (also reproduced on `7.1.3-2`), `linux-cachyos-nvidia-open 7.1.4-1`, `nvidia-utils 610.43.03-1`, `plymouth 26.134.222-2.1` (+ cachyos-plymouth-bootanimation 2-3), `linux-cachyos-lts 6.18.38-2`, `linux-cachyos-lts-nvidia-open 6.18.38-2`
- Kernel cmdline: `quiet nowatchdog splash rw rootflags=subvol=/@ root=UUID=` (no explicit `nvidia_drm` parameters)

Full `cachyos-bugreport.sh` output: https://gist.github.com/Snakesblood/9cfa85879c3efee4e2632ddaedce2de4

## Describe the bug

Two failure modes on 7.1.x, five consecutive attempts in one evening (2026-07-22/23), no config changes between attempts:

| Boot | Kernel | Result |
|------|--------|--------|
| 23:54 | 7.1.3-2 | plymouth-quit timeout → kwin_wayland EBUSY loop → black screens |
| 23:58 | 7.1.3-2 | Same |
| 00:04 | 7.1.3-2 | **Hard freeze 8 s into boot**, journal truncated mid-boot |
| 00:08 | 6.18.38-2-lts | Works perfectly (all 3 monitors); ran `pacman -Syu` → 7.1.4-1 |
| 00:11 | 7.1.4-1 | **Hard freeze 7 s into boot**, journal truncated mid-boot |
| 00:13 | 6.18.38-2-lts | Works perfectly |

### Failure mode 1: plymouthd never quits, holds DRM master, compositor locked out

NVIDIA module loads fine, plymouthd starts, but `plymouth-quit.service` times out; plymouthd keeps DRM master on the NVIDIA card and kwin_wayland can never open it (boot at 23:58, identical at 23:54):

```
23:58:14 kernel: NVRM: loading NVIDIA UNIX Open Kernel Module for x86_64 610.43.03 Release Build
23:58:17 systemd-modules-load[252]: Inserted module 'nvidia_drm'
23:58:22 systemd[1]: Starting Terminate Plymouth Boot Screen...
23:58:42 systemd[1]: plymouth-quit.service: start operation timed out. Terminating.
23:58:42 systemd[1]: plymouth-quit.service: Failed with result 'timeout'.
23:58:42 systemd[1]: Failed to start Terminate Plymouth Boot Screen.
23:58:43 kwin_wayland[1084]: No backend specified, automatically choosing drm
23:58:43 kwin_wayland[1084]: Failed to open /dev/dri/card1 device (Device or resource busy)
23:58:44 kwin_wayland[1084]: Failed to open /dev/dri/card1 device (Device or resource busy)
... (repeats indefinitely)
23:58:49 plasma-login-greeter[1128]: There are no outputs - creating placeholder screen
```

The greeter session then sits on black/placeholder screens until power-cycled.

### Failure mode 2: hard freeze ~7–8 s into boot, journal truncated

Two other boots (7.1.3-2 at 00:04, 7.1.4-1 at 00:11) died mid-boot: journal ends abruptly while services were mid-startup, no oops/panic captured, machine unresponsive (power cycle required). Plymouth had behaved normally up to switch-root in both. In the 7.1.4-1 freeze there is **no `NVRM: loading` line at all** before the journal ends — the lockup coincides with NVIDIA module initialization:

```
00:11:32 systemd[1]: Finished Plymouth switch root service.
00:11:35 systemd[1]: Starting Flush Journal to Persistent Storage...
00:11:35 systemd-journald[492]: Time spent on flushing to /var/log/journal/ is 5.337ms for 1718 entries.
00:11:35 systemd[1]: Mounted /boot.

```

### On LTS (same userspace, same driver, same config): works every time

```
00:13:33 kernel: NVRM: loading NVIDIA UNIX Open Kernel Module for x86_64 610.43.03 Release Build
00:13:36 systemd-modules-load[249]: Inserted module 'nvidia_drm'
```

plymouth quits normally, all three DP connectors come up, desktop fully functional.

## Expected behavior

7.1.x boots to the greeter with all three DisplayPort monitors, as 6.18.38-lts does.

## Additional context

- Same failure chain reported by another CachyOS user on 2026-07-10 (also NVIDIA 610, cold boots, intermittent): https://discuss.cachyos.org/t/plymouth-bug-report-july-10th-2026/32740
- Possibly related upstream: 610.43.03 mishandling of the kernel-7.x `screen_info`/fbdev refactor causing display-signal loss: https://forums.developer.nvidia.com/t/nvidia-drm-fbdev-0-with-kernel-7-x-causes-vt-black-screen/373248 (unresolved as of 2026-07-16)
- **Workaround confirmed (2026-07-23):** removing plymouth entirely (packages, mkinitcpio hook, `splash` from cmdline) makes 7.1.4-1 boot normally — NVIDIA module loads, all three DP outputs come up, zero `Device or resource busy` errors. This confirms plymouthd's quit-hang holding DRM master is the direct blocker in failure mode 1. Failure mode 2 (hard freeze) was intermittent (2 of 5 boots), so one clean plymouth-free boot doesn't yet prove it's gone — will update after more cold boots.
- Untested alternative: `nvidia_drm.modeset=1 nvidia_drm.fbdev=1`.
- Happy to provide full journals, coredumps, test patches, or a vanilla-kernel repro if useful.