https://bugzilla.kernel.org/show_bug.cgi?id=221657

Artem S. Tashkinov ([email protected]) changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[email protected]

--- Comment #3 from Artem S. Tashkinov ([email protected]) ---
(In reply to Timothée Monney from comment #2)
> (In reply to Artem S. Tashkinov from comment #1)
> > This looks like a BIOS/HW issue. No way the Linux kernel can be responsible
> > for your HW spontaneously deciding to power on itself.
> > 
> > The latest BIOS is 7D32vHN, you could give it a try.
> 
> I gave it a try but it didn't change anythings.
> 
> Should I maybe go on others forum to report this ? Do you know which one
> would be appropriate ?

You could ask on:

 * https://hardforum.com/forums/motherboards.4/
 * https://forums.anandtech.com/forums/motherboards.6/
 * https://www.reddit.com/r/techsupport/
 * https://www.reddit.com/r/pchelp/

Also here's what an LLM has to say:

This could still be firmware, ACPI, or a hardware wake/power-button issue. If
the machine reaches true S5/off, Linux should no longer be able to “decide” to
power it back on. But Linux can influence the shutdown path and ACPI state, so
a kernel/driver/firmware interaction could make the board reboot instead of
entering S5 cleanly.

What can cause a motherboard to turn itself on:

### 1. It is not actually reaching true S5/off

This is the big one. `poweroff` asks the kernel/systemd/firmware to transition
to ACPI soft-off. If ACPI shutdown fails or firmware interprets the final
transition as reboot, it can look like “the board turns itself on.”

The fact that:

* `sudo halt` leaves a black screen
* `poweroff` / `shutdown -h now` restart
* physical button also restarts

suggests the final ACPI power-state transition is broken or being overridden,
not merely a userspace command problem.

### 2. Firmware wake sources

On MSI boards, check **Settings → Advanced → Wake Up Event Setup**. MSI’s own
Wake-on-LAN docs mention `ErP Ready` and `Resume By PCI-E/Networking Device` as
relevant BIOS settings for allowing wake from off/sleep states. ([MSI][1])

Likely candidates:

* **Resume by PCI-E / PCIe device**
* **Resume by USB device**
* **Wake by RTC alarm**
* **Wake by PS/2 keyboard/mouse**
* **Wake by onboard LAN / Intel NIC**
* **Power on by keyboard/mouse**
* **PME event wake**
* **Thunderbolt/USB-C wake**, if present
* **Wake from onboard Wi-Fi/Bluetooth module**

Even with Wake-on-LAN “disabled” in the OS, firmware-level PCIe/PME wake may
still exist.

### 3. Network adapter or PCIe PME wake

The RTX 3090, NIC, Wi-Fi module, USB controllers, and PCI bridges can all be
wake-capable devices. Linux exposes some ACPI wake sources under
`/proc/acpi/wakeup`; Arch’s power-management docs note that wake triggers
include things like power buttons, WOL, and device-triggered wake events, and
that there is no single unified interface for every possible trigger. ([Arch
Wiki][2])

Useful check:

```bash
cat /proc/acpi/wakeup
```

Then temporarily disable suspicious enabled devices, for example:

```bash
echo XHCI | sudo tee /proc/acpi/wakeup
echo RP01 | sudo tee /proc/acpi/wakeup
echo GLAN | sudo tee /proc/acpi/wakeup
```

Names vary by board.

Also check NIC wake state:

```bash
sudo ethtool eno1 | grep Wake-on
sudo ethtool -s eno1 wol d
```

Replace `eno1` with the actual interface.

### 4. Front-panel power switch fault

Your “short circuit” suspicion is reasonable. A flaky case power switch or
front-panel header can absolutely cause immediate power-on.

Best test:

1. Shut down.
2. Disconnect the case **PWR_SW** connector from the motherboard.
3. Start the machine by briefly shorting the two power-switch pins with a
screwdriver.
4. Shut it down again.

If it now stays off, the case switch/wiring is suspect.

Also inspect:

* front-panel header polarity/placement
* reset switch accidentally on power pins
* damaged case wiring
* stuck physical button
* motherboard standoff short
* metal debris behind the board

### 5. PSU / standby rail weirdness

ATX systems are never fully dead while connected to AC. The motherboard remains
powered via **5VSB** and can respond to wake events. A bad PSU or unstable 5VSB
rail can create strange power behavior, though I would rank this below
firmware/wake-source/front-panel causes.

Testing with another PSU is worthwhile if BIOS wake options and PWR_SW
isolation do not explain it.

### 6. AC power restore behavior not actually applying

They said they set AC loss behavior to “Power Off,” which is good, but I would
still clear CMOS and re-set it. Firmware settings can get weird after BIOS
updates.

Try:

* clear CMOS
* load optimized defaults
* disable all wake events
* set AC restore to **Power Off**
* enable ErP
* save, power-cycle, retest

### 7. NVIDIA / kernel interaction during shutdown

The NVIDIA driver probably cannot “wake” the board from true off, but it can
contribute to a broken shutdown path: GPU driver unload, PCIe power management,
runtime PM, or ACPI handoff could leave the platform in a bad transition state.

Given they tested `nomodeset` and disabled `nvidia-powerd`, that weakens but
does not eliminate the possibility.

A good isolation test:

```bash
sudo systemctl isolate rescue.target
sudo modprobe -r nvidia_drm nvidia_modeset nvidia_uvm nvidia
sudo poweroff
```

Or boot a live USB with no NVIDIA proprietary driver and test shutdown.

### 8. “Poweroff” vs “reboot” path confusion

Try forcing systemd/kernel behavior:

```bash
sudo systemctl poweroff --force --force
```

Also test from firmware/installer/live environment:

* boot Ubuntu live USB
* do not install NVIDIA driver
* power off from live session
* test whether it restarts

If it restarts even from UEFI setup screen or a live USB, it is almost
certainly firmware/hardware/wake-source rather than Ubuntu userspace.

### My ranking

Most likely:

1. **Firmware wake source / PCIe PME / USB wake**
2. **ACPI S5 transition bug between kernel and BIOS**
3. **Front-panel power switch or wiring fault**
4. **PSU standby rail / electrical issue**
5. **NVIDIA driver interfering with shutdown path**

So yes, your hardware suspicion is valid, especially the power button theory.
But I would not rule out Linux entirely. The cleaner statement is: Linux
probably cannot power it on from a true off state, but Linux/ACPI/driver
behavior can prevent it from ever reaching that true off state.

[1]: https://www.msi.com/support/technical_details/MB_Wake_On_LAN "How to set
up Wake-On-Lan (WOL) - msi Motherboard"
[2]: https://wiki.archlinux.org/title/Power_management/Wakeup_triggers "Power
management/Wakeup triggers"

-- 
You may reply to this email to add a comment.

You are receiving this mail because:
You are watching the assignee of the bug.

_______________________________________________
acpi-bugzilla mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/acpi-bugzilla

Reply via email to