## Summary

After a suspend/resume cycle with the lid closed and an external Dell U2723QE attached to the laptop's built-in HDMI port, the i915 driver on `linux-omarchy 7.2.5-3` fails to restore the display state (`Restoring old state failed with -22`). From that point on, **every** atomic modeset on `HDMI-A-2` is rejected by the kernel with `EINVAL`, including the 640x480 and 720x400 fallbacks that aquamarine tries.

Hyprland still reports the monitor as connected and enabled, but with mode `0x0@60` at position `0x0` (overlapping `eDP-1`). Workspaces 1-5 are pinned to that monitor via workspace rules, so they stay assigned to it and every window on them becomes **invisible**: the windows exist in `hyprctl clients`, but there is no physical output rendering them, and they are not folded back onto the laptop panel either.

A `hyprctl reload` (which forces a fresh modeset) fully recovers the output at 3840x2160@30 without a reboot, so the hardware and the cable are fine.

## System details

| Item | Value |
|---|---|
| Machine | Lenovo ThinkPad T490 (20N3S4EU06), BIOS N2IETA7P 1.85 (2026-04-07) |
| CPU | Intel Core i5-8365U (Whiskey Lake) |
| GPU | Intel UHD Graphics 620 (WhiskeyLake-U GT2, 8086:3ea0), driver `i915` |
| External monitor | Dell U2723QE (27", 3840x2160) on the laptop's native HDMI port, seen as `HDMI-A-2`. HDMI 1.4 link, so 4K is limited to 30 Hz |
| Internal panel | eDP-1, Chimei Innolux 0x14D4, 1920x1080@60, scale 1.25 |
| Omarchy | 4.0.4-1 |
| Kernel | `linux-omarchy 7.2.5-3` (`7.2.5-3-omarchy`) |
| Hyprland | 0.56.2-2 (commit efb50993) |
| aquamarine | 0.15.0-2 |
| mesa | 26.2.2-1 |
| linux-firmware | 20260810-2 |
| Suspend type | S3 (`deep`) via logind lid switch |
| USB-C | a generic USB hub (0bda:5483, Realtek) with keyboard, webcam and r8152 Ethernet is attached to the USB-C port. No display goes through it |

These packages were updated on 2026-09-15 (kernel 7.1.9 → 7.2.3/7.2.5, aquamarine 0.14.0 → 0.15.0, mesa 26.2.1 → 26.2.2). The same docked setup worked across suspend/resume before that update.

## Steps to reproduce

1. Dock the T490: Dell U2723QE on the built-in HDMI port, lid closed, external keyboard on USB-C.
2. `monitors.lua` (relevant part):
   ```lua
   hl.monitor({ output = "", mode = "preferred", position = "auto", scale = 1.25 })
   hl.monitor({ output = "desc:Dell Inc. DELL U2723QE ...", mode = "preferred", position = "0x0", scale = 2 })
   for ws = 1, 5 do hl.workspace_rule({ workspace = tostring(ws), monitor = "desc:Dell Inc. DELL U2723QE ...", default = (ws == 1) }) end
   for ws = 6, 10 do hl.workspace_rule({ workspace = tostring(ws), monitor = "eDP-1", default = (ws == 6) }) end
   ```
3. Open the lid briefly and close it again, so logind triggers two suspends a few seconds apart (this is what happened here, see the timeline). A single suspend/resume with the lid closed may also be enough; I have not isolated that yet.
4. Wake the machine.

**Expected:** the Dell comes back at 3840x2160@30 and shows workspaces 1-5.

**Actual:** the Dell stays black. `hyprctl monitors` shows it as enabled with mode `0x0`. Windows on workspaces 1-5 cannot be seen on any screen.

## Timeline (journal, current boot)

```
09:34:26 systemd-logind: Lid closed.
09:34:26 systemd-logind: Suspending...
09:34:38 kernel: PM: suspend entry (deep)
09:35:10 systemd-logind: Lid opened.
09:35:10 kernel: ACPI: PM: Waking up from system sleep state S3
09:35:10 kernel: i915 0000:00:02.0: [drm] *ERROR* [CONNECTOR:135:DP-2][ENCODER:134:DDI C/PHY C][DPRX] Failed to enable link training
09:35:15 systemd-logind: Lid closed.
09:35:35 systemd-logind: Suspending...
09:35:35 kernel: PM: suspend entry (deep)
09:35:42 kernel: ACPI: PM: Waking up from system sleep state S3
09:35:42 kernel: i915 0000:00:02.0: [drm] *ERROR* Restoring old state failed with -22
09:35:42 kernel: PM: suspend exit
```

Note that `DP-2` has nothing connected (`/sys/class/drm/card1-DP-2/status` = `disconnected`), yet i915 attempted link training on it during the first resume. The second resume is the one that leaves `HDMI-A-2` unusable.

## Hyprland / aquamarine log after the resume

aquamarine walks down the whole mode list and the kernel rejects every mode, with and without `max_bpc`:

```
DEBUG from aquamarine ]: Swapchain: Reconfigured a swapchain to [Vector2D: x: 3840, y: 2160] XR24 of length 3
DEBUG from aquamarine ]: atomic drm request: failed to commit: Invalid argument, flags: ATOMIC_ALLOW_MODESET ATOMIC_TEST_ONLY
WARN  from aquamarine ]: drm: atomic commit failed with max_bpc set, retrying without max_bpc
DEBUG from aquamarine ]: atomic drm request: failed to commit: Invalid argument, flags: ATOMIC_ALLOW_MODESET ATOMIC_TEST_ONLY
... (same for every mode in the list) ...
DEBUG from aquamarine ]: Swapchain: Reconfigured a swapchain to [Vector2D: x: 640, y: 480] XR24 of length 3
DEBUG from aquamarine ]: atomic drm request: failed to commit: Invalid argument, flags: ATOMIC_ALLOW_MODESET ATOMIC_TEST_ONLY
DEBUG from aquamarine ]: Swapchain: Reconfigured a swapchain to [Vector2D: x: 720, y: 400] XR24 of length 3
DEBUG from aquamarine ]: atomic drm request: failed to commit: Invalid argument, flags: ATOMIC_ALLOW_MODESET ATOMIC_TEST_ONLY
DEBUG from aquamarine ]: drm: Modesetting HDMI-A-2 with [email protected]
ERR   from aquamarine ]: atomic drm request: failed to commit: Invalid argument, flags: ATOMIC_ALLOW_MODESET PAGE_FLIP_EVENT
```

This repeats in a loop (`Disabling output HDMI-A-2` → `Modesetting HDMI-A-2 with [email protected]` → `failed to commit`) until the reload.

## State while broken

```
$ hyprctl monitors
Monitor eDP-1 (ID 0):
	[email protected] at 0x0
	active workspace: 6 (6)
	scale: 1.25
	dpmsStatus: 1
	disabled: false

Monitor HDMI-A-2 (ID 1):
	[email protected] at 0x0          <-- no mode, same position as eDP-1
	description: Dell Inc. DELL U2723QE
	active workspace: 1 (1)
	scale: 1
	dpmsStatus: 1
	disabled: false
	availableModes: [email protected] ... [email protected]

$ hyprctl clients -j | jq -r '.[] | "\(.workspace.id)\t\(.monitor)\t\(.class)"'
1	1	kitty        <-- on HDMI-A-2, invisible
1	1	kitty        <-- on HDMI-A-2, invisible
6	0	chromium
6	0	kitty

$ cat /sys/class/drm/card1-HDMI-A-2/status /sys/class/drm/card1-HDMI-A-2/enabled
connected
enabled
```

## Recovery

```
$ hyprctl reload
$ hyprctl monitors -j | jq -r '.[] | "\(.name)\t\(.width)x\(.height)@\(.refreshRate)\t\(.x),\(.y)\tscale=\(.scale)"'
HDMI-A-2	[email protected]	0,0	scale=2
eDP-1	[email protected]	1920,0	scale=1.25
```

The reload produced one more transient kernel error (`i915: *ERROR* Atomic update failure on pipe B`) but the output has been stable since. Unplugging and replugging the HDMI cable would presumably have the same effect.

## Why I think this belongs here

- The kernel is Omarchy's own `linux-omarchy` package, and the regression appeared right after the 2026-09-15 update to the 7.2 series. The `Restoring old state failed with -22` on resume points at i915 in that kernel.
- Independently of the kernel bug, Hyprland/aquamarine leave the output in a half-alive state (`enabled`, `0x0`, workspaces still bound to it) instead of either retrying the modeset after resume or treating the output as gone so the pinned workspaces fold back onto `eDP-1`. A user ends up with windows they cannot reach on any screen and no hint of why. If a re-modeset after resume (what `hyprctl reload` does) could be triggered automatically by Omarchy's sleep hooks, that would make this failure self-healing.

Happy to test kernel parameters, an LTS kernel, or a patched aquamarine, and to attach the full `omarchy-debug` output on request (it is 1100+ lines and includes network details, so I left it out of the body).