This message was deleted.
# harvestera
adamant-kite-43734
06/13/2026, 11:13 PMThis message was deleted.
p
prehistoric-morning-49258
06/16/2026, 12:15 PMit sounds like basically consumer/homelab hw (cpu/mobo) limitation. like might be possible to hack around depends on reqs, but it's not surprising they physically don't support secure pcie virt. it's more like "HIGHLY recommend avoiding v1.8 if wanting to do secure multi-tenant pcie virt on NUClab hw"?
t
thousands-advantage-10804
06/16/2026, 4:15 PMagreed.
b
brave-park-93691
06/16/2026, 5:00 PMI'm on Intel Xeon + L40S with v1.8.0, and my IOMMU groups are fine. But I'm hitting the same controller bugs: duplicate
permittedHostDevicesbrave-park-93691
06/16/2026, 5:00 PMThese reproduce on Intel as well, so they seem separate from the IOMMU group change. So there may be two distinct problems here: the IOMMU grouping (hardware/kernel behavior) and the pcidevices-controller bugs (which reproduce across hardware).
t
thousands-advantage-10804
06/16/2026, 5:01 PMI have seen that as well. rebooting usually fixes that.
b
brave-park-93691
06/16/2026, 5:12 PMThat’s a fair workaround. The reason I’m digging deeper is that in a more production-leaning setup a reboot is costlier
brave-park-93691
06/16/2026, 5:12 PMIt means draining the node and restarting every VM on it just to re-attach a single GPU. So Im hoping to understand the root cause well enough to avoid the reboot
t
thousands-advantage-10804
06/16/2026, 5:12 PMagreed that reboots are not ideal. Open a Github issue?
2 Views