PCIe passthrough issues with M.2 Google Coral TPU on kernels newer than 6.8.12-13-pve

[MartinJ]

When updating my PVE-host to a kernel newer than 6.8.12-13-pve, PCIe passtrough of a M.2 Google Coral TPU fails as the VM will not start anymore with the following error message:

kvm: -device vfio-pci,host=0000:00:02.0,id=hostpci1,bus=pci.0,addr=0x11:
info: OpRegion detected on Intel display 4680.
kvm: -device vfio-pci,host=0000:0b:00.0,id=hostpci2,bus=pci.0,addr=0x1b:
vfio 0000:0b:00.0: error getting device from group 27: No such device
Verify all devices in group 27 are bound to vfio-<bus> or pci-stub and not already in useTASK ERROR: start failed: QEMU exited with code 1

Reverting to kernel 6.8.12-13-pve solves the issue but as this kernel is not part of the PVE Debian Trixie sources and had to be manaully installed, I do not think that this is a long term solution.
The device is bound to the vfio-pci driver and is the only device in its IOMMU group.

root@pve:~# lspci -k -s 0000:0b:00.0
0b:00.0 System peripheral: Global Unichip Corp. Coral Edge TPU
        Subsystem: Global Unichip Corp. Coral Edge TPU
        Kernel driver in use: vfio-pci
root@pve:~# ls /sys/kernel/iommu_groups/27/devices/
0000:0b:00.0

When the start of the VM fails, I get the following errors in dmesg:

root@pve:~# dmesg | grep -i vfio | tail -20 && dmesg | grep -i "0b:00" | tail -20
[  292.193327] vfio-pci 0000:00:02.0: Invalid PCI ROM header signature: expecting 0xaa55, got 0x0000
[  292.217930] vfio-pci 0000:0b:00.0: enabling device (0000 -> 0002)
[  292.217976] vfio-pci 0000:0b:00.0: resetting
[  292.359924] vfio-pci 0000:0b:00.0: reset done
[  385.470037] vfio-pci 0000:01:00.0: resetting
[  385.575009] vfio-pci 0000:01:00.0: reset done
[  385.587580] vfio-pci 0000:00:02.0: resetting
[  385.694988] vfio-pci 0000:00:02.0: reset done
[  385.719237] vfio-pci 0000:0b:00.0: resetting
[  385.923005] vfio-pci 0000:0b:00.0: reset done
[  385.923063] vfio-pci 0000:0b:00.0: Unable to change power state from D0 to D3hot, device inaccessible
[  387.193323] vfio-pci 0000:01:00.0: resetting
[  387.294999] vfio-pci 0000:01:00.0: reset done
[  387.295042] vfio-pci 0000:01:00.0: Masking broken INTx support
[  387.383824] vfio-pci 0000:00:02.0: resetting
[  387.486950] vfio-pci 0000:00:02.0: reset done
[  387.487459] vfio-pci 0000:00:02.0: Invalid PCI ROM header signature: expecting 0xaa55, got 0x0000
[  387.512402] vfio-pci 0000:0b:00.0: enabling device (0000 -> 0002)
[  387.512453] vfio-pci 0000:0b:00.0: resetting
[  387.655064] vfio-pci 0000:0b:00.0: reset done
[  292.489814] pci 0000:0b:00.0: BAR 2 [mem 0x6003a00000-0x6003afffff 64bit pref]: assigned
[  292.489841] pci 0000:0b:00.0: BAR 0 [mem 0x6003b00000-0x6003b03fff 64bit pref]: assigned
[  385.719237] vfio-pci 0000:0b:00.0: resetting
[  385.923005] vfio-pci 0000:0b:00.0: reset done
[  385.923063] vfio-pci 0000:0b:00.0: Unable to change power state from D0 to D3hot, device inaccessible
[  386.047944] pci 0000:0b:00.0: [1ac1:089a] type 00 class 0x0000ff PCIe Endpoint
[  386.048028] pci 0000:0b:00.0: BAR 0 [mem 0x00000000-0x00003fff 64bit pref]
[  386.048034] pci 0000:0b:00.0: BAR 2 [mem 0x00000000-0x000fffff 64bit pref]
[  386.048472] pci 0000:0b:00.0: Adding to iommu group 27
[  386.049898] pci 0000:0b:00.0: BAR 2 [mem 0x6003a00000-0x6003afffff 64bit pref]: assigned
[  386.049927] pci 0000:0b:00.0: BAR 0 [mem 0x6003b00000-0x6003b03fff 64bit pref]: assigned
[  387.512402] vfio-pci 0000:0b:00.0: enabling device (0000 -> 0002)
[  387.512453] vfio-pci 0000:0b:00.0: resetting
[  387.655064] vfio-pci 0000:0b:00.0: reset done
[  387.782980] pci 0000:0b:00.0: [1ac1:089a] type 00 class 0x0000ff PCIe Endpoint
[  387.783049] pci 0000:0b:00.0: BAR 0 [mem 0x6003b00000-0x6003b03fff 64bit pref]
[  387.783055] pci 0000:0b:00.0: BAR 2 [mem 0x6003a00000-0x6003afffff 64bit pref]
[  387.783442] pci 0000:0b:00.0: Adding to iommu group 27
[  387.784910] pci 0000:0b:00.0: BAR 2 [mem 0x6003a00000-0x6003afffff 64bit pref]: assigned
[  387.784943] pci 0000:0b:00.0: BAR 0 [mem 0x6003b00000-0x6003b03fff 64bit pref]: assigned
root@pve:~#

I suspect that the error about changing the power state is the root cause but I was not able to find a solution to the problem.
Any other passed through PCIe devices (HBA, NIC and GPU) continue to work as expected.