Hello all, I'm setting up a fairly (?) complex oVirt over Gluster setup built around 3 Xeon servers-turned-into-workstations, each doubling as oVirt node + one primary Fedora VM w/ a dedicated passthrough GPU (+audio and a couple of USB root devices). One of the servers seems to have some weird issue w/ the passthrough nVidia GPU that seems to require me to edit the VM iommu (1) and passthrough device (2) command line. I tried using the qemu-cmdline addon to add the missing parameters, but it seems that qemu treats the added parameters as an additional device / iommu instead of editing the existing parameters. So: 1. How can I view the VM qemu command line? 2. Can I somehow manually edit the qemu command line, either directly or by somehow adding parameters in the HE XML file? - Gilboa. [1] iommu: VM XXX is down with error. Exit message: internal error: qemu unexpectedly closed the monitor: 2021-11-05T14:59:44.499366Z qemu-kvm: We need to set caching-mode=on for intel-iommu to enable device assignment with IOMMU protection. [2] GPU: May need to add x-vga=off,

On Mon, Nov 8, 2021 at 5:47 PM Gilboa Davara <gilboad(a)gmail.com&gt; wrote: > > Hello all, > > I'm setting up a fairly (?) complex oVirt over Gluster setup built around 3 Xeon servers-turned-into-workstations, each doubling as oVirt node + one primary Fedora VM w/ a dedicated passthrough GPU (+audio and a couple of USB root devices). > One of the servers seems to have some weird issue w/ the passthrough nVidia GPU that seems to require me to edit the VM iommu (1) and passthrough device (2) command line. > I tried using the qemu-cmdline addon to add the missing parameters, but it seems that qemu treats the added parameters as an additional device / iommu instead of editing the existing parameters. > > So: > 1. How can I view the VM qemu command line? less /var/log/libvirt/qemu/vm-name.log

> 2. Can I somehow manually edit the qemu command line, either directly or by somehow adding parameters in the HE XML file? I think this should be possible via vdsm hook, but hooks are bad. Can you explain what do you want to change? Nir

(Sorry: Pressed send too fast...) 3. I'm facing some odd memory allocation issues (only if the VM is Q35, all works well if the VM is configured as i440fx), which may require me to change additional VM parameters. BTW, ironically, the two other hosts in the same cluster, with older MBs, have zero issues with GPU/Audio/USB passthrough...) - Gilboa On Tue, Nov 9, 2021 at 11:52 AM Gilboa Davara <gilboad(a)gmail.com&gt; wrote: > Hello, > > Many thanks for the prompt reply. > > On Mon, Nov 8, 2021 at 8:52 PM Nir Soffer <nsoffer(a)redhat.com&gt; wrote: > >> On Mon, Nov 8, 2021 at 5:47 PM Gilboa Davara <gilboad(a)gmail.com&gt; wrote: >> > >> > Hello all, >> > >> > I'm setting up a fairly (?) complex oVirt over Gluster setup built >> around 3 Xeon servers-turned-into-workstations, each doubling as oVirt node >> + one primary Fedora VM w/ a dedicated passthrough GPU (+audio and a couple >> of USB root devices). >> > One of the servers seems to have some weird issue w/ the passthrough >> nVidia GPU that seems to require me to edit the VM iommu (1) and >> passthrough device (2) command line. >> > I tried using the qemu-cmdline addon to add the missing parameters, >> but it seems that qemu treats the added parameters as an additional device >> / iommu instead of editing the existing parameters. >> > >> > So: >> > 1. How can I view the VM qemu command line? >> >> less /var/log/libvirt/qemu/vm-name.log >> > > Found it, thanks! > > >> > 2. Can I somehow manually edit the qemu command line, either directly >> or by somehow adding parameters in the HE XML file? >> >> I think this should be possible via vdsm hook, but hooks are bad. >> Can you explain what do you want to change? >> >> Nir >> > > 1. IOMMU: > -device intel-iommu,intremap=on,eim=on > To (Add: caching-mode=on, per qemu log message) > -device intel-iommu,intremap=on,eim=on,caching-mode=on > (2021-11-05T18:42:36.651191Z qemu-kvm: We need to set caching-mode=on for > intel-iommu to enable device assignment with IOMMU protection.) >

> 2. Device: > -device > vfio-pci,host=0000:84:00.0,id=ua-c774115b-c0ee-43a5-97c8-12ac73cb6f3a,bus=pci.10,addr=0x0 > > -device > vfio-pci,host=0000:84:00.1,id=ua-3c7d3b66-57aa-470b-afc8-78383ac9a025,bus=pci.8,addr=0x0 > To (Attach hdmi sound as function of GPU, possibly add x-vga=on): > -device > vfio-pci,host=0000:84:00.0,id=ua-c774115b-c0ee-43a5-97c8-12ac73cb6f3a,bus=pci.8.0,x-vga=on,addr=0x0 > > -device > vfio-pci,host=0000:84:00.1,id=ua-3c7d3b66-57aa-470b-afc8-78383ac9a025,bus=pci.8.1,addr=0x0 > > 3. I'm facing some odd memory allocation issues (only if the VM is Q35, > all works well if the VM is configured as i440fx), which may req > _______________________________________________ Users mailing list -- users(a)ovirt.org To unsubscribe send an email to users-leave(a)ovirt.org Privacy Statement: https://www.ovirt.org/privacy-policy.html oVirt Code of Conduct: https://www.ovirt.org/community/about/community-guidelines/ List Archives: https://lists.ovirt.org/archives/list/users@ovirt.org/message/XKZLE6ZEU7Z...

Many thanks! Somehow my Google-fu skills (and DDG-Fu skills) missed this bug report. I wonder if 4.4.9 will fix the sporadic -ENOMEM (1) that I'm getting under Q35... Thanks, - Gilboa (1) 2021-11-05T13:13:06.290874Z qemu-kvm: -device vfio-pci,host=0000:07:00.0,id=ua-a3048786-da87-425f-83e6-51afbb906423,bus=pci.10,addr=0x0: VFIO_MAP_DMA failed: Cannot allocate memory 2021-11-05T13:13:06.362210Z qemu-kvm: -device vfio-pci,host=0000:07:00.0,id=ua-a3048786-da87-425f-83e6-51afbb906423,bus=pci.10,addr=0x0: VFIO_MAP_DMA failed: Cannot allocate memory 2021-11-05T13:13:06.362609Z qemu-kvm: -device vfio-pci,host=0000:07:00.0,id=ua-a3048786-da87-425f-83e6-51afbb906423,bus=pci.10,addr=0x0: vfio 0000:07:00.0: failed to setup container for group 33: memory listener initialization failed: Region ram-node0: vfio_dma_map(0x55fe05d647f0, 0xc0000, 0x7ff40000, 0x7f821fec0000) = -12 (Cannot allocate memory) 2021-11-05 13:13:06.580+0000: shutting down, reason=failed On Tue, Nov 9, 2021 at 3:29 PM Arik Hadas <ahadas(a)redhat.com&gt; wrote: > > > On Tue, Nov 9, 2021 at 11:57 AM Gilboa Davara <gilboad(a)gmail.com&gt; wrote: > >> (Sorry: Pressed send too fast...) >> 3. I'm facing some odd memory allocation issues (only if the VM is Q35, >> all works well if the VM is configured as i440fx), which may require me to >> change additional VM parameters. >> >> BTW, ironically, the two other hosts in the same cluster, with older >> MBs, have zero issues with GPU/Audio/USB passthrough...) >> >> - Gilboa >> >> On Tue, Nov 9, 2021 at 11:52 AM Gilboa Davara <gilboad(a)gmail.com&gt; wrote: >> >>> Hello, >>> >>> Many thanks for the prompt reply. >>> >>> On Mon, Nov 8, 2021 at 8:52 PM Nir Soffer <nsoffer(a)redhat.com&gt; wrote: >>> >>>> On Mon, Nov 8, 2021 at 5:47 PM Gilboa Davara <gilboad(a)gmail.com&gt; >>>> wrote: >>>> > >>>> > Hello all, >>>> > >>>> > I'm setting up a fairly (?) complex oVirt over Gluster setup built >>>> around 3 Xeon servers-turned-into-workstations, each doubling as oVirt node >>>> + one primary Fedora VM w/ a dedicated passthrough GPU (+audio and a couple >>>> of USB root devices). >>>> > One of the servers seems to have some weird issue w/ the passthrough >>>> nVidia GPU that seems to require me to edit the VM iommu (1) and >>>> passthrough device (2) command line. >>>> > I tried using the qemu-cmdline addon to add the missing parameters, >>>> but it seems that qemu treats the added parameters as an additional device >>>> / iommu instead of editing the existing parameters. >>>> > >>>> > So: >>>> > 1. How can I view the VM qemu command line? >>>> >>>> less /var/log/libvirt/qemu/vm-name.log >>>> >>> >>> Found it, thanks! >>> >>> >>>> > 2. Can I somehow manually edit the qemu command line, either >>>> directly or by somehow adding parameters in the HE XML file? >>>> >>>> I think this should be possible via vdsm hook, but hooks are bad. >>>> Can you explain what do you want to change? >>>> >>>> Nir >>>> >>> >>> 1. IOMMU: >>> -device intel-iommu,intremap=on,eim=on >>> To (Add: caching-mode=on, per qemu log message) >>> -device intel-iommu,intremap=on,eim=on,caching-mode=on >>> (2021-11-05T18:42:36.651191Z qemu-kvm: We need to set caching-mode=on >>> for intel-iommu to enable device assignment with IOMMU protection.) >>> >> > Note that this one should be fixed in the latest ovirt-engine > (ovirt-engine-4.4.9.4) - > https://bugzilla.redhat.com/show_bug.cgi?id=2013752 > > >> >>> 2. Device: >>> -device >>> vfio-pci,host=0000:84:00.0,id=ua-c774115b-c0ee-43a5-97c8-12ac73cb6f3a,bus=pci.10,addr=0x0 >>> >>> -device >>> vfio-pci,host=0000:84:00.1,id=ua-3c7d3b66-57aa-470b-afc8-78383ac9a025,bus=pci.8,addr=0x0 >>> To (Attach hdmi sound as function of GPU, possibly add x-vga=on): >>> -device >>> vfio-pci,host=0000:84:00.0,id=ua-c774115b-c0ee-43a5-97c8-12ac73cb6f3a,bus=pci.8.0,x-vga=on,addr=0x0 >>> >>> -device >>> vfio-pci,host=0000:84:00.1,id=ua-3c7d3b66-57aa-470b-afc8-78383ac9a025,bus=pci.8.1,addr=0x0 >>> >>> 3. I'm facing some odd memory allocation issues (only if the VM is Q35, >>> all works well if the VM is configured as i440fx), which may req >>> >> _______________________________________________ >> Users mailing list -- users(a)ovirt.org >> To unsubscribe send an email to users-leave(a)ovirt.org >> Privacy Statement: https://www.ovirt.org/privacy-policy.html >> oVirt Code of Conduct: >> https://www.ovirt.org/community/about/community-guidelines/ >> List Archives: >> https://lists.ovirt.org/archives/list/users@ovirt.org/message/XKZLE6ZEU7Z... >> >