Bug 2559096

1. Please describe the problem:

On an ASUS ROG Zephyrus G14 GA403UI with an AMD Ryzen 9 8945HS, AMD IOMMU IVMD exclusion ranges are exposed by Linux as IOMMU_RESV_DIRECT regions. This prevents VFIO from assigning an otherwise properly isolated NVIDIA RTX 4070 Laptop GPU to a QEMU/KVM guest.

Hardware:

ASUS ROG Zephyrus G14 GA403UI_GA403UI
BIOS: GA403UI.311, release date 06/09/2025
CPU: AMD Ryzen 9 8945HS

Discrete GPU:
01:00.0 [10de:2860] NVIDIA AD106M [GeForce RTX 4070 Max-Q / Mobile]

GPU audio:
01:00.1 [10de:22bd] NVIDIA AD106M High Definition Audio Controller

Integrated GPU:
65:00.0 [1002:1900] AMD HawkPoint1

AMD-Vi is enabled and operational.

The NVIDIA GPU and its audio function are isolated together in IOMMU group 14, with no unrelated devices.

For IOMMU group 14:

$ cat /sys/kernel/iommu_groups/14/reserved_regions

0x0000000097d36000 0x0000000097d37fff direct
0x00000000fee00000 0x00000000feefffff msi
0x000000fd00000000 0x000000ffffffffff reserved

The same 0x97d36000-0x97d37fff direct region is exposed for many other IOMMU groups on this system.

The ACPI IVRS table contains two IVMD type 0x22 entries corresponding to adjacent 4 KiB regions at:

0x97d36000
0x97d37000

Relevant raw IVRS bytes:

00C0: 48 00 00 00 22 01 00 01 22 08 20 00 00 00 FF 0F
00D0: 00 00 00 00 00 00 00 00 00 70 D3 97 00 00 00 00
00E0: 01 00 00 00 00 00 00 00 22 08 20 00 00 00 FF 0F
00F0: 00 00 00 00 00 00 00 00 00 60 D3 97 00 00 00 00
0100: 01 00 00 00 00 00 00 00

The GPU can be detached successfully from its normal host drivers and bound to vfio-pci:

01:00.0: nvidia -> vfio-pci
01:00.1: snd_hda_intel -> vfio-pci

Manual reattach also works correctly.

However, when QEMU/VFIO attempts to use the IOMMU group, the kernel reports:

vfio-pci 0000:01:00.0: Firmware has requested this device have a 1:1 IOMMU mapping, rejecting configuring the device without a 1:1 mapping. Contact your platform vendor.

QEMU/libvirt then fails with:

vfio 0000:01:00.0: failed to setup container for group 14:
Failed to set group container: Invalid argument

After the VM start fails, libvirt correctly rolls the devices back to:

01:00.0 -> nvidia
01:00.1 -> snd_hda_intel

Therefore, PCI detach/bind itself works. The failure occurs when VFIO attempts to establish ownership/DMA mapping for the IOMMU group.

No ACS override, allow_unsafe_interrupts, vfio-pci.ids=, or other IOMMU bypass/workaround is being used.


2. What is the Version-Release number of the kernel:

Current kernel:

7.2.11-200.fc44.x86_64

$ uname -a

Linux fedora 7.2.11-200.fc44.x86_64 #1 SMP PREEMPT_DYNAMIC Thu Oct 15 15:58:58 UTC 2026 x86_64 GNU/Linux


3. Did it work previously in Fedora? If so, what kernel version did the issue
   *first* appear? Old kernels are available for download at
   https://koji.fedoraproject.org/koji/packageinfo?packageID=8 :

No known working Fedora kernel.

The issue was reproduced with:

7.2.8-200.fc44.x86_64
7.2.11-200.fc44.x86_64

With both kernels, /sys/kernel/iommu_groups/14/reserved_regions contains:

0x0000000097d36000 0x0000000097d37fff direct

and VFIO fails when QEMU attempts to use the GPU group.


4. Can you reproduce this issue? If so, please provide the steps to reproduce
   the issue below:

Yes, it is reproducible.

1. Boot Fedora 44 normally with AMD-Vi enabled.

2. Verify that the NVIDIA GPU and audio function are isolated in IOMMU group 14:

01:00.0 NVIDIA RTX 4070 Laptop GPU
01:00.1 NVIDIA HD Audio

3. Verify the reserved region:

cat /sys/kernel/iommu_groups/14/reserved_regions

The output includes:

0x0000000097d36000 0x0000000097d37fff direct

4. Stop users of the NVIDIA GPU:

sudo systemctl stop gdm
sudo systemctl stop nvidia-powerd.service

Verify that:

sudo fuser -v /dev/nvidia*

shows no remaining users.

5. Start a libvirt VM configured with both NVIDIA PCI functions as managed host devices:

<hostdev mode='subsystem' type='pci' managed='yes'>
  <driver name='vfio'/>
  <source>
    <address domain='0x0000' bus='0x01' slot='0x00' function='0x0'/>
  </source>
</hostdev>

<hostdev mode='subsystem' type='pci' managed='yes'>
  <driver name='vfio'/>
  <source>
    <address domain='0x0000' bus='0x01' slot='0x00' function='0x1'/>
  </source>
</hostdev>

6. Start the VM:

virsh -c qemu:///system start Win11

The GPU is detached and bound to vfio-pci, but VM startup fails.

7. Kernel log contains:

vfio-pci 0000:01:00.0: Firmware has requested this device have a 1:1 IOMMU mapping, rejecting configuring the device without a 1:1 mapping. Contact your platform vendor.

8. QEMU/libvirt reports:

vfio 0000:01:00.0: failed to setup container for group 14:
Failed to set group container: Invalid argument

9. Libvirt then rolls the devices back successfully to nvidia and snd_hda_intel.


5. Does this problem occur with the latest Rawhide kernel? To install the
   Rawhide kernel, run ``sudo dnf install fedora-repos-rawhide`` followed by
   ``sudo dnf update --enablerepo=rawhide kernel``:

Not tested with the latest Rawhide kernel.

I have reproduced the issue with Fedora 44 kernels 7.2.8-200.fc44.x86_64 and 7.2.11-200.fc44.x86_64.

I can test a Rawhide kernel or a proposed AMD IOMMU patch if requested.


6. Are you running any modules that not shipped with directly Fedora's kernel?:

Yes.

The proprietary NVIDIA kernel modules from RPM Fusion are installed through akmods:

akmod-nvidia-615.71.09-3.fc44.x86_64

For the currently running kernel:

kmod-nvidia-7.2.11-200.fc44.x86_64-615.71.09-3.fc44.x86_64

The NVIDIA modules are used normally by the host, but the VFIO failure occurs after the GPU has been successfully detached from the NVIDIA driver and bound to vfio-pci.

No out-of-tree IOMMU or VFIO modules are installed.


7. 7. Please attach the kernel logs. You can get the complete kernel log
   for a boot with ``journalctl --no-hostname -k > dmesg.txt``. If the
   issue occurred on a previous boot, use the journalctl ``-b`` flag.

The issue was reproduced again on the current boot with kernel:

7.2.11-200.fc44.x86_64

Immediately after reproducing the VFIO failure I collected:

- complete kernel log: dmesg-vfio.txt
- filtered libvirt/QEMU/VFIO journal: vfio-libvirt-filtered.txt
- virtqemud journal: virtqemud-vfio.txt

The reproduced QEMU/libvirt failure was:

vfio 0000:01:00.0: failed to setup container for group 14:
Failed to set group container: Invalid argument

The host recovered correctly after the failed VM start:

01:00.0 -> nvidia
01:00.1 -> snd_hda_intel

and nvidia-smi worked normally again.

I can also provide, if useful:

- complete ACPI IVRS dump
- decoded ivrs.dsl
- full lspci -nnk output
- full IOMMU group topology
- reserved_regions output for all groups
- libvirt domain XML

Reproducible: Always