## Summary

On a hybrid-graphics laptop where the HDMI port is wired to the NVIDIA dGPU, resuming from `deep` suspend leaves the external monitor permanently corrupted (pink screen, or black backlit). The dGPU's display engine cannot complete a modeset after resume until reboot. Switching to `s2idle` works around it.

## Hardware

- MSI GL65 9SEK (MS-16U5), Intel Core i7-9750H
- iGPU: Intel UHD 630 (`i915`, drives internal `eDP-1`)
- dGPU: NVIDIA RTX 2060 Mobile (TU106), `nvidia-open` 610.57.04 — **HDMI-A-1 is wired to the dGPU** (verified via `/sys/class/drm/card1-HDMI-A-1`)
- Omarchy 4.0.4-1, Hyprland 0.56.2, Linux 7.2.3-1-omarchy

## Reproduction

1. Boot normally (default `mem_sleep` = `deep`)
2. `systemctl suspend`, resume
3. External monitor shows a pink screen (or black backlit) — internal panel is fine

100% reproducible on this machine, including on the first suspend after a fresh boot.

## Evidence

Journal from resume:

```
[drm:nv_drm_atomic_commit [nvidia_drm]] *ERROR* [nvidia-drm] Failed to apply atomic modeset. Error code: -11
[drm:nv_drm_atomic_commit [nvidia_drm]] *ERROR* [nvidia-drm] Flip event timeout on head 0
NVRM: Xid (PCI:0000:01:00): 13, Graphics Exception: Shader Program Header 11 Error
NVRM: Xid (PCI:0000:01:00): 13, pid=..., name=Hyprland, Graphics Exception: channel 0x00000005
```

After this failure the output is unrecoverable in-session: `hyprctl reload`, a full DPMS off/on cycle, disabling/re-enabling the output, `hl.dsp.force_renderer_reload()`, and a VT switch round-trip all fail; every subsequent modeset attempt on `HDMI-A-1` hits `Flip event timeout on head 0` / atomic modeset `-11` again.

Related observations:

- The `nvidia-suspend`/`nvidia-resume`/`nvidia-hibernate` services are a no-op on this driver: `/proc/driver/nvidia/suspend` does not exist in 610.57.04, and `nvidia-sleep.sh` exits early in that case.
- `NVreg_PreserveVideoMemoryAllocations=0` (set by Omarchy, cf. reverted #5519 and discussion #5500) is required here: with the preserve parameter set and no procfs suspend interface, the driver aborts sleep with `pci_pm_freeze(): nv_pmops_freeze returns -5`.

## Workaround that works

`mem_sleep_default=s2idle` on the kernel cmdline. With s2idle:

- No Xid 13, no GPU corruption; the desktop on the external monitor returns after ~4s of `Flip event timeout` retries and stays stable.
- `s2idle` is properly supported on this CPU (`cat /sys/power/mem_sleep` → `[s2idle] deep` after the change), and hibernation is unaffected.

## Suggestion

Consider either:

- documenting this in the manual for laptops where an output is wired to the dGPU, and/or
- shipping an opt-in setup command (e.g. `omarchy setup nvidia-sleep-fix`) that adds the `mem_sleep_default=s2idle` limine-entry-tool drop-in:

```bash
echo 'KERNEL_CMDLINE[default]+=" mem_sleep_default=s2idle"' | sudo tee /etc/limine-entry-tool.d/mem-sleep.conf
sudo limine-mkinitcpio
```

## Full context

- `omarchy debug --no-sudo --print` output: attached (omarchy-debug.log)
- Suspected root cause is in the NVIDIA driver (dGPU VRAM/display state not preserved across S3 without the procfs suspend path), but Omarchy users on MSI/ASUS-style hybrids will keep hitting this.