Intel GPU透传问题：i915驱动加载失败求助

Hey there, let's tackle your Intel GPU passthrough issue step by step. First off, to answer your core question: Intel integrated GPU passthrough is absolutely feasible now—the 2015 limitations you referenced are mostly a thing of the past, as long as your hardware supports VT-d/IOMMU and you configure things correctly.

Let's Diagnose the "Bridge Device Not Found" Error #

That [drm:i915_driver_load [i915]] *ERROR* bridge device not found message is the key here. Intel integrated GPUs don't exist in isolation—they're tied to a PCI Express Root Port (bridge device) on your motherboard. If you only pass through the GPU itself (00:02.0) without its parent bridge, the guest OS's i915 driver can't initialize the GPU properly.

Fix Steps to Try #

1. Verify Host IOMMU Configuration & PCI Grouping #

First, make sure your host's IOMMU is properly set up (even if kvm-ok says KVM works):

• Check if intel_iommu=on is in your kernel command line:

  cat /proc/cmdline
  

  If not, add it to /etc/default/grub (modify GRUB_CMDLINE_LINUX_DEFAULT), run sudo update-grub, and reboot.
• Find the IOMMU group your GPU belongs to—this will show all devices tied to the same group (which may include the bridge you need to pass through):

  find /sys/kernel/iommu_groups/ -type l | grep -i 00:02.0
  

  For example, if your GPU is in group 2, list all devices in that group:

  ls -l /sys/kernel/iommu_groups/2/devices/
  

  Note down the bridge device (usually something like 00:00.0 or 00:01.0).

2. Switch from pci_stub to vfio-pci #

pci_stub is a basic placeholder driver—vfio-pci is designed specifically for PCI passthrough and handles device isolation better:

• Unload the i915 driver and bind your GPU (and its bridge) to vfio-pci:
  First, get your GPU's vendor/device ID with lspci -nn | grep VGA (looks like 8086:1912):

  # Unload i915 (may need to stop X server first by switching to a TTY)
  sudo modprobe -r i915
  # Bind GPU to vfio-pci
  echo "8086 1912" | sudo tee /sys/bus/pci/drivers/vfio-pci/new_id
  echo "00:02.0" | sudo tee /sys/bus/pci/devices/0000:00:02.0/driver/unbind
  echo "0000:00:02.0" | sudo tee /sys/bus/pci/drivers/vfio-pci/bind
  # Repeat the last three steps for the bridge device you found earlier
  

• Make sure vfio modules load on boot: create /etc/modules-load.d/vfio.conf with:

  vfio
  vfio-pci
  vfio_iommu_type1
  

3. Update VirtualBox VM Configuration #

You need to pass through both the GPU and its parent bridge to the VM:

• First, detach any existing PCI attachments if needed:

  VBoxManage modifyvm "my-vm" --pciattach 00:02.0@none
  

• Attach the bridge device first, then the GPU (use unused PCI addresses in the guest, like 01:04.0 for the bridge and 01:05.0 for the GPU):

  VBoxManage modifyvm "my-vm" --pciattach BRIDGE_DEVICE@01:04.0
  VBoxManage modifyvm "my-vm" --pciattach 00:02.0@01:05.0
  

• Also ensure the VM has IOMMU and hardware virtualization enabled:

  VBoxManage modifyvm "my-vm" --ioapic on
  VBoxManage modifyvm "my-vm" --hwvirtex on
  VBoxManage modifyvm "my-vm" --nestedpaging on
  

4. Guest OS (openSUSE Leap 42.3) Tweaks #

openSUSE Leap 42.3 is quite old (2017 release, kernel 4.4.x), which might have limited support for newer Intel GPUs:

• Try upgrading to a newer openSUSE release (like Leap 15.5) if possible—this will give you a much newer kernel with better i915 support.
• If you must stick with 42.3, force the i915 driver to load by adding i915.force_probe=YOUR_GPU_DEVICE_ID to the guest's kernel command line (replace YOUR_GPU_DEVICE_ID with the 4-digit ID from lspci -nn, e.g., 1912).

Final Notes #

Double-check that your motherboard's BIOS/UEFI has VT-d (or Intel IOMMU) enabled—this is a critical prerequisite that's easy to miss. Also, make sure no other host processes are using the GPU (like your desktop environment) before binding it to vfio-pci; you may need to switch to a TTY session to unload i915 successfully.

内容的提问来源于stack exchange，提问作者Jan