[TUTORIAL] cant add PCIE 0000:01:00.0 to a VM?

[scyto]

The device can be seen in LSPCI and PCI mapping tool but not in the add device UI? is this a bug? (the device has its own iommu group). can i add it manually in the vm conf file and be ok?

 

[dcsapak]

The device can be seen in LSPCI and PCI mapping tool but not in the add device UI? is this a bug? (the device has its own iommu group). can i add it manually in the vm conf file and be ok?

hi,

no by default we filter some things out in the 'raw device' view to not clutter it too much.

I can see that such ai processors would be handy to have also in that view, so if you can post the output of

lspci -nn

the we can check if we can easily add it too

to work around it, you can set it via the cli/api:

qm set VMID -hostpci0 0000:01:00.0

 

[scyto]

to work around it, you can set it via the cli/api:

cool thanks, i will try that, i added it via the vm.conf for now
it didn't work in an nested vm (proxmox host, treunas vm L1, debian L2 VM) it entertaingly reset the whole server when the driver loaded.
next up debian VM on the proxmox host

 

[scyto]

no by default we filter some things out in the 'raw device' view to not clutter it too much.

i understand, tbh it would be nice to have a UI with checkbox and filters rather than adding 15 pcie devices one by one - but i get why that is such a lowwww prio qol feature
consder this op answered.

 

[scyto]

the we can check if we can easily add it too

01:00.0 Co-processor [0b40]: Hailo Technologies Ltd. Hailo-8 AI Processor [1e60:2864] (rev 01)

note frigate is very picky about firmware and driver versions and generally expects older versions than are shipping
hailo provides a drive deb package on their website and one can compile from source too - like this
seems like the hailo.ai 8 and 8L are the natural successors to coral TPU which seems abandonded by google (do they ever stick with anything)