dGPU re-enumerates after suspend/resume in integrated mode (zombie state)

[heymuun]

Describe the bug
After switching to integrated mode with EnvyControl, the NVIDIA dGPU is correctly disabled at boot. However, after suspend/resume, the BIOS re‑enumerates the NVIDIA GPU and exposes it to the kernel again. This leaves the device in a “zombie” state (present in PCI tree, but unbound to any driver), which wastes power and can cause instability.

---

To Reproduce
Steps to reproduce the behavior:

1. Run sudo envycontrol -s integrated
2. Reboot → dGPU is disabled (not visible in lspci, renderer is AMD iGPU).
3. Suspend the system.
4. Resume → kernel logs show NVIDIA PCI devices (0000:01:00.0 and 0000:01:00.1) being re‑added.
5. dGPU remains unbound, but present in PCI tree.

---

Expected behavior
The dGPU should remain disabled after suspend/resume, with no PCI re‑enumeration. EnvyControl should either:

• Automatically remove the device again after resume, or
• Bind it to a safe driver (e.g. Nouveau) to prevent unmanaged zombie state.

---

System Information:

• Model: ASUS TUF Gaming A15 FA507UV
• Distro: Fedora Linux 43 (GNOME, previously KDE Plasma)
• Kernel: 6.17.11-300.fc43.x86_64
• DE/WM and Display Manager: GNOME 45 with GDM
• EnvyControl version: 3.5.2
• Nvidia driver version: None installed (no matching packages in dnf list installed | grep nvidia)
• lspci output (before suspend):
  (no NVIDIA devices listed)
• lspci output (after suspend/resume):
  pci 0000:01:00.0: [10de:28e0] VGA compatible controller pci 0000:01:00.1: [10de:22be] Audio device
• Kernel logs (after resume):
  pcieport 0000:00:01.1: Slot(0): Card present pci 0000:01:00.0: [10de:28e0] type 00 class 0x030000 PCIe Legacy Endpoint pci 0000:01:00.1: [10de:22be] type 00 class 0x040300 PCIe Endpoint input: HDA NVidia HDMI/DP,pcm=3 ...

---

Screenshot: Basically I closed my laptop lid DURING integrated mode to check ZOMBIE dgpu wake up

Additional context

• glxinfo confirms AMD Radeon 780M iGPU is the active renderer.
• lspci shows no NVIDIA device before suspend, but devices reappear after resume.
• journalctl -k logs confirm PCI hotplug events for NVIDIA GPU/audio controller after resume.
• Workarounds tested: binding to Nouveau (safe hold) vs. systemd sleep hook to auto‑remove PCI device.

Suggested fix
Add a systemd sleep hook in EnvyControl that automatically re‑removes the dGPU after resume, or provide an option to bind it to Nouveau for safe handling of BIOS wakeups.

[bayasdev]

@heymuun does the power consumption increase after the dGPU comes back?

[heymuun]

@heymuun does the power consumption increase after the dGPU comes back?

Yes. Not just the power, but also the temperature near the gpu gets very hot.