Regression in 6.17.13-1/2-pve: DMAR faults with VFIO when intel_iommu=on iommu=pt (works on 6.17.9-1-pve)

[Jonathan50543]

Hello,

I would like to report what appears to be a kernel regression affecting VFIO passthrough.

Summary​

On my system VFIO passthrough works correctly with kernel 6.17.9-1-pve, but after upgrading to 6.17.13-1-pve or 6.17.13-2-pve, DMAR faults start appearing after the VM runs for some time.

Downgrading back to 6.17.9-1-pve makes the issue disappear completely, suggesting a regression introduced between these kernel versions.

Kernel parameters​

The host kernel is booted with:

intel_iommu=on iommu=pt

Symptoms​

After the VM has been running for a while, the host begins logging repeated DMAR faults such as:

[ 8955.704107] DMAR: DRHD: handling fault status reg 3
[ 8955.704115] DMAR: [DMA Write NO_PASID] Request device [00:0d.0] fault addr 0x64d87000 [fault reason 0x05] PTE Write access is not set
[ 8955.704146] DMAR: DRHD: handling fault status reg 3
[ 8955.704150] DMAR: [DMA Write NO_PASID] Request device [00:0d.0] fault addr 0x64d8a000 [fault reason 0x05] PTE Write access is not set
[ 8955.704194] DMAR: DRHD: handling fault status reg 3
[ 8955.704197] DMAR: [DMA Write NO_PASID] Request device [00:0d.0] fault addr 0x64d8c000 [fault reason 0x05] PTE Write access is not set
[ 8961.574179] dmar_fault: 2057 callbacks suppressed
[ 8961.574185] DMAR: DRHD: handling fault status reg 3
[ 8961.574190] DMAR: [DMA Read NO_PASID] Request device [00:0d.0] fault addr 0x408b5000 [fault reason 0x06] PTE Read access is not set
[ 8963.079753] DMAR: DRHD: handling fault status reg 3
[ 8963.079761] DMAR: [DMA Read NO_PASID] Request device [05:00.4] fault addr 0x40a65000 [fault reason 0x06] PTE Read access is not set
[ 8963.080493] DMAR: DRHD: handling fault status reg 3
[ 8963.080497] DMAR: [DMA Read NO_PASID] Request device [05:00.4] fault addr 0x40a65000 [fault reason 0x06] PTE Read access is not set
[ 9037.575995] dmar_fault: 29 callbacks suppressed
[ 9037.575999] DMAR: DRHD: handling fault status reg 3
[ 9037.576003] DMAR: [DMA Read NO_PASID] Request device [05:00.4] fault addr 0x40bd0000 [fault reason 0x06] PTE Read access is not set

Two devices attached to the VM appear in the fault logs.

Hardware​

Platform: Intel N305 (Alder Lake-N)

Devices passed through via VFIO:

00:0d.0 USB controller [0c03]: Intel Corporation Alder Lake-N Thunderbolt 4 USB Controller [8086:464e]
05:00.4 Ethernet controller [0200]: Mellanox Technologies MT27800 Family [ConnectX-5 Virtual Function] [15b3:1018]

VM configuration (relevant parts)​

BIOS:

SeaBIOS

CPU configuration:

cpu: host,flags=-nested-virt;-virt-ssbd;-amd-ssbd;-amd-no-ssb;-pdpe1gb;-hv-tlbflush;-hv-evmcs

Passthrough devices:

hostpci1: 0000:00:0d,pcie=1,rombar=0
hostpci2: 0000:05:00.4,pcie=1,rombar=0

Guest​

The guest is running a 6.12 kernel.

Regression confirmation​

• 6.17.9-1-pve → no DMAR faults, passthrough stable
• 6.17.13-1-pve → DMAR faults appear
• 6.17.13-2-pve → same issue
• Downgrading back to 6.17.9-1-pve → issue disappears

Additional info​

To avoid exposing unrelated hardware details, I have only included the relevant log excerpts and device information above.
If more targeted debugging information is required, I can provide it.

Thanks.

 

[cybertorture]

Same story with SATA passthrough

Kernel parameters​

intel_iommu=on iommu=pt consoleblank=600 nouveau.blacklist=1 i915.max_vfs=7 module_blacklist=xe

Mar 16 05:38:05 cyberserver kernel: dmar_fault: 21963 callbacks suppressed
Mar 16 05:38:05 cyberserver kernel: DMAR: DRHD: handling fault status reg 3
Mar 16 05:38:05 cyberserver kernel: DMAR: [DMA Write NO_PASID] Request device [00:17.0] fault addr 0x463a9000 [fault reason 0x05] PTE Write access is not set
Mar 16 05:38:05 cyberserver kernel: DMAR: DRHD: handling fault status reg 3
Mar 16 05:38:05 cyberserver kernel: DMAR: [DMA Write NO_PASID] Request device [00:17.0] fault addr 0x463a9000 [fault reason 0x05] PTE Write access is not set
Mar 16 05:38:05 cyberserver kernel: DMAR: DRHD: handling fault status reg 3
Mar 16 05:38:05 cyberserver kernel: DMAR: [DMA Write NO_PASID] Request device [00:17.0] fault addr 0x463a9000 [fault reason 0x05] PTE Write access is not set
Mar 16 05:38:05 cyberserver kernel: DMAR: DRHD: handling fault status reg 3

VM configuration​

hostpci0: mapping=SATA
machine: q35

cat /etc/pve/mapping/pci.cfg

SATA
        map id=8086:7ae2,iommugroup=8,node=cyberserver,path=0000:00:17.0,subsystem-id=1458:b005

VM guest

Version:
25.10.2.1 - Goldeye

Kernel versions 6.17.13-(1 and 2)-pve does have a problem. Version 6.17.9-1-pve everything is working as intended.

Cheers

 

[dcsapak]

can you post more of the journal, maybe it does contain a hint?

is it actually broken or does everything work like usual with just the addition of the new log lines?

 

[Jonathan50543]

can you post more of the journal, maybe it does contain a hint?

is it actually broken or does everything work like usual with just the addition of the new log lines?

What kind of journal would you like? From the host or the guest?

It did break the device connection in the guest, as my disks were shown disconnected and the CX5 NIC unreachable.