Blackwell (RTX 5060 Ti, sm_120): shutdown hangs ~2 min in nv_pci_shutdown → RestoreConsole → unixCallVideoBIOS (soft lockup on CPU)

NVIDIA Open GPU Kernel Modules Version

610.43.02

Operating System and Version

Description: Debian GNU/Linux 13 (trixie)

Kernel Release

Linux pclinux 6.12.94+deb13-amd64 #1 SMP PREEMPT_DYNAMIC Debian 6.12.94-1 (2026-06-20) x86_64 GNU/Linux

Please confirm you are running a stable release kernel (e.g. not a -rc). We do not accept bug reports for unreleased kernels.

• I am running on a stable kernel release.

Build Command

N/A — driver was installed as a prebuilt package from NVIDIA's official Debian repository (nvidia-open 610.43.02), not built from source.

Terminal output/Build Log

N/A — not a build issue. The driver loads correctly and the system runs fine. The bug is a shutdown hang. Full call trace from kdump is in the "Describe the bug" / "More Info" sections below.

More Info

Hardware:

• CPU: Intel i9-12900KF (no integrated graphics)
• Motherboard: ASUS TUF GAMING B760-PLUS WIFI D4, BIOS 1825 (2025-10-09)
• RAM: 32 GB DDR4
• GPU: ASUS DUAL GeForce RTX 5060 Ti OC 16GB GDDR7 (GB206, sm_120)
• Boot: legacy MBR, dual-boot with Windows 10 (Windows shuts down cleanly).

Software:

• Debian 13 (Trixie), kernel 6.12.94+deb13-amd64
• nvidia-open 610.43.02 (DKMS), from NVIDIA's official Debian repository
• CUDA 13.3
• KDE Plasma on Wayland via SDDM
• Kernel cmdline: ro quiet splash nvidia-drm.modeset=1
• nvidia, nvidia_modeset, nvidia_uvm, nvidia_drm loaded from initramfs.

Describe the bug:
System shutdown consistently hangs for ~2 minutes 9 seconds inside nv_pci_shutdown, before the motherboard finally powers off. 100% reproducible on every shutdown/reboot.

A kdump captured during the hang shows a soft lockup on CPU#8 with PID 1 (systemd-shutdown) stuck for 26+ seconds in the following call chain:

kernel_power_off
device_shutdown
pci_device_shutdown
nv_pci_shutdown [nvidia]
nv_pci_remove_helper [nvidia]
nv_drm_remove.cold [nvidia_drm]
nv_drm_dev_unload [nvidia_drm]
ReleaseOwnership [nvidia_modeset]
nvKmsIoctl [nvidia_modeset]
nvkms_ioctl_from_kapi [nvidia_modeset]
ReleaseOwnership [nvidia_modeset]
RestoreConsole [nvidia_modeset]
nvFreeCoreChannelEvo [nvidia_modeset]
nvRmVTSwitch [nvidia_modeset]
nvRmApiControl [nvidia_modeset]
nvkms_call_rm [nvidia_modeset]
Nv04ControlKernel [nvidia]
_nv04ControlWithSecInfo [nvidia]
rmapiControlWithSecInfoTls [nvidia]
rmapiControlWithSecInfo [nvidia]
_rmapiRmControl [nvidia]
serverControl [nvidia]
gpuresControl_IMPL [nvidia]
resControl_IMPL [nvidia]
deviceCtrlCmdOsUnixVTSwitch_IMPL [nvidia]
unixCallVideoBIOS [nvidia]
nv_vbios_call [nvidia]
X86EMU_exec [nvidia]
x86emuOp_in_word_AX_DX [nvidia] <- stuck here (first kdump)
os_io_read_dword [nvidia] <- reading I/O port 0x600c
x86emuOp_out_word_DX_AX [nvidia] <- stuck here (second kdump)
os_io_write_dword [nvidia] <- writing I/O port 0x6008

The driver is calling into the VBIOS via the x86 emulator (X86EMU_exec) to perform a VT-switch / console restore during PCI shutdown, and the I/O port reads/writes to the GPU's BAR I/O space never return.

Two independent kdumps from separate reboots show the exact same call chain, with the only difference being whether the lockup is captured at a port read (0x600c) or a port write (0x6008). The watchdog detects the soft lockup at 26 s, but the system continues to retry until the power-off eventually completes (~2 min 9 s total).

To reproduce:

1. Boot Debian 13 with nvidia-open 610.43.02 on a Blackwell GPU (RTX 5060 Ti, sm_120) and nvidia-drm.modeset=1.
2. Log in to a KDE Plasma Wayland session via SDDM.
3. Issue any normal shutdown (KDE menu, systemctl poweroff, shutdown -h now).
4. The screen turns off but the motherboard does not cut power.
5. After ~2 min 9 s, power finally cuts.

Things ruled out:

• Not SDDM (cleanly stops in ~100 ms before this point).
• Not disks, swap, or filesystem unmount (all complete cleanly in the log before nv_pci_shutdown is entered).
• Not the network.
• Not RANDSTRUCT (kernel built with CONFIG_RANDSTRUCT_NONE=y, unlike issue #1033).
• Not GSP: tested with NVreg_EnableGpuFirmware=0 and the hang is identical.
• Not hardware/VBIOS/BIOS: Windows shuts down cleanly on the same machine (dual boot).
• The CPU is i9-12900KF (no iGPU), so the RTX 5060 Ti is the only display path and the only target for console restore.

The system is otherwise stable during normal use; the bug only manifests at shutdown/reboot.

The proprietary driver does not support Blackwell sm_120 (only nvidia-open does), so a direct comparison with the proprietary driver of the same version is not possible.

nvidia-bug-report.log.gz attached.

nvidia-bug-report.log.gz

Start by reproducing shutdown on the listed Debian 13, RTX 5060 Ti, and nvidia-open 610.43.02 setup, then inspect nvidia-bug-report.log.gz and compare the two kdump call traces through nv_pci_shutdown, RestoreConsole, and unixCallVideoBIOS. Done means shutdown no longer hangs for about two minutes or triggers the reported soft lockup.