RTX 4050 Laptop (Lenovo LOQ 15ARP10E): dGPU signal on platform GPIO 4 wakes system from s2idle when HDMI is connected

On Lenovo laptops where the HDMI port is wired to the NVIDIA dGPU, suspending with an external HDMI monitor connected makes the system wake again 5–7 seconds later. The wake comes from AMD GPIO pin 4, whose ACPI _EVT handler notifies the NVIDIA GPU (\_SB.PCI0.GPP0.PEGP, 0x81). Without a monitor on the dGPU output, the system stays asleep.

The same pin and the same _EVT target have now been confirmed on three Lenovo models (LOQ 15ARP10E, Legion Pro 5 16ARX8, and an 83LT). For the Legion, the AMD kernel maintainer’s assessment on LKML was that the NVIDIA stack appears to be asserting this GPIO during suspend, and that it should be fixed there rather than with a kernel quirk.

System

  • Lenovo LOQ 15ARP10E (83S0), board LNVNB161216, BIOS SUCN18WW
  • Ryzen 7 7735HS + RTX 4050 Laptop; internal panel on the iGPU, HDMI on the dGPU
  • Fedora 44, kernel 7.2.11-200.fc44
  • Driver 615.78.08 (also seen with 615.71.09), RPM Fusion akmod; UseKernelSuspendNotifiers: 1, PreserveVideoMemoryAllocations: 1
  • mem_sleep: [s2idle] only

Steps to reproduce

  1. Connect an external monitor to the HDMI port (dGPU).
  2. echo 1 | sudo tee /sys/power/pm_debug_messages
  3. systemctl suspend
  4. The system resumes after ~5–7 s. The kernel log shows GPIO 4 is active, either via the EC/SCI path or directly via IRQ 7 (pinctrl_amd) when acpi.ec_no_wakeup=1 is set.

Expected: the system stays in s2idle until a user wake event.

ACPI context (SSDT22)
_AEI: GpioInt (Edge, ActiveBoth, ExclusiveAndWake, PullNone, …) { 0x0004 }
_EVT: Case (0x04) { … Notify (\_SB.PCI0.GPP0.PEGP, 0x81) }

nvidia-bug-report.log.gz (1.2 MB)

Impact: besides wasting power (~60 wake cycles per night), each spurious cycle occasionally hangs the machine mid-transition: 6 hard hangs in about 330 cycles over 15 days, each needing a forced power-off.

Workaround: gpiolib_acpi.ignore_wake=AMDI0030:00@4. With it, the first full night was a single 9 h 51 min suspend woken only by the keyboard.

The attached nvidia-bug-report.log.gz was generated with the workaround active (serial numbers and network identifiers redacted). I can provide one captured right after a spurious wake (workaround removed) if that helps.

Related