[BUG] Poweroff returns to power-on on ASUS ROG STRIX X870E-E GAMING WIFI7 NEO (ErP On workaround; S5 wake source suspected)

## Pre-flight checklist

- [x] I have searched existing issues and this is not a duplicate. (Related-but-closed: #529, #897.)
- [x] I have read the Contributing Guidelines.
- [x] I have verified the issue is reproducible with the latest available CachyOS kernel. (`7.2.3-1-cachyos`, current stable.)
- [ ] I have tried to reproduce the issue on Arch Linux's `linux` kernel. — **Not yet done.** This is the open triage question: I believe the bug is firmware-side (it disappears with BIOS ErP On regardless of OS), but I have not yet confirmed whether it's kernel-specific. Happy to test `linux-cachyos-zen` and Arch `linux` and post the result.

## Upstream / vanilla kernel check

- I have **not** tested with a vanilla/upstream kernel yet. Based on the ErP behavior this points to firmware (BIOS) rather than the kernel, but I'll confirm with a vanilla-kernel test and update this issue.

## Kernel variant

- `linux-cachyos (EEVDF, Clang)` — running `7.2.3-1-cachyos` (Clang 22.1.8, PREEMPT_DYNAMIC), booted via Limine.

## System information (cachyos-bugreport.sh)

The script requires root and this box's sudo needs a password I don't have from the SSH account I used, so I couldn't capture its output. Provided `inxi -Fxxxz` manually below instead. I can attach a full bugreport if you tell me how to run it non-interactively / I can add the host to my sudoers.

## Manual system information

- **Kernel version:** `7.2.3-1-cachyos` (x86_64, 64-bit, clang 22.1.8, clocksource tsc)
- **CPU:** AMD Ryzen 9 9950X3D (16c/32t, Zen 5)
- **GPU:** NVIDIA GB203 [GeForce RTX 5080] — driver `nvidia` 610.57.04
- **Board / firmware:** ASUSTeK **ROG STRIX X870E-E GAMING WIFI7 NEO**, Rev 1.xx, UEFI, AMI **BIOS 1002, dated 07/13/2026**
- **RAM:** 30 GiB
- **NICs:** RTL8126 5GbE (`r8169`), MT7927 Wi-Fi 7 (`mt7925e`)
- **DE/init:** KWin Wayland (Xwayland), systemd, btrfs subvol `@`
- **SCX scheduler:** none active (`scx_loader.service` disabled)

## Bug description

`systemctl poweroff` completes cleanly all the way to ACPI S5, but the board powers itself back on within seconds to ~1 minute, booting straight to the Limine boot manager. The machine behaves as if it rebooted. Behavior is **intermittent** — often the *second* consecutive poweroff sticks. Setting **BIOS ErP Ready → On (S4+S5)** stops it entirely; the bug only occurs with ErP Disabled.

## Steps to reproduce

1. BIOS → APM Configuration → **ErP Ready: Disabled** (or S5-only)
2. From a logged-in CachyOS session, run `systemctl poweroff`
3. OS shuts down fully to S5; wait a few seconds to ~1 minute
4. Observe: board powers back on by itself (intermittent — repeat to see the auto-on; the next poweroff usually sticks)

## Expected behavior

`systemctl poweroff` leaves the system off (S5). No automatic power-on.

## Actual behavior

After a clean S5 power-off, the board powers back on by itself within ~1 minute and boots normally. Repeatable with ErP Disabled; **not** reproducible with ErP On (multiple cycles tested).

## Logs / stack traces

From the journal of a failing boot, the shutdown is clean to the last second:

- All filesystems unmounted, `systemd-poweroff.service` reached, `Reached target Shutdown`, no oops/panic, `nowatchdog` in cmdline.
- No kernel errors in the final moments before S5.
- No resume/S4 evidence in the following boot — this is a genuine power-on, not a resume.
- `last -x reboot` shows a string of ~1-minute "boot sessions" — each is an auto power-on after a failed poweroff; the following real boot is minutes later.
- While running, `/proc/acpi/wakeup` shows ~25 wake sources armed for S4/S5: `XHCI0/1/2`, PCIe `GPP0/2/3/7/8`, 10GbE upstream ports, USB hub ports.
- Boot cmdline: `quiet nowatchdog splash rw rootflags=subvol=/@ ... reboot=acpi`. (`reboot=` only affects the warm-reset path and has **no** effect on this issue — confirmed.)
- BIOS Memory Context Restore / Power Down Enable toggling: no effect (only made POST slower).

## Hypothesis

With ErP Disabled, the board leaves a PCIe PME/USB wake source armed at S5 (5V standby present); a spurious wake event during the power-off transition makes the firmware power the system back on. That fits the "ErP On fixes it" data and points to firmware. However, the kernel's handling of PME wake enables at shutdown could be contributing (e.g. not clearing PME_En before the S5 transition, or an interaction with CachyOS ACPI/ASUS patches — cf. the suspicion in the now-closed #529). A kernel-side triage is welcome.

## Additional system information

- Also reported on the ASUS ROG forum: ROG STRIX X870E-E GAMING WIFI7 NEO — poweroff auto-powers-back-on (fixed by ErP On).
- The user's only concern with the ErP On workaround is loss of Wake-on-LAN / standby wake, which they use for remote access — hence the ask for a proper fix rather than living with ErP On permanently.
- Willing to run any requested kernel/parameter test (zen, Arch `linux`, `pcie_port_pm=off`, per-source wake disabling) and post results.

Start by reproducing the shutdown with ErP Disabled and compare CachyOS, linux-cachyos-zen, and Arch linux kernels; record the journal and /proc/acpi/wakeup state. Try the mentioned pcie_port_pm=off and per-source wake disabling tests. Done means determining whether the behavior is kernel-specific or firmware-side, with results posted for the ASUS board and ErP settings.

Written by the indexing model from the issue text.