https://bugzilla.kernel.org/show_bug.cgi?id=219009 --- Comment #17 from Ben Hirlston (ozonehelix@xxxxxxxxx) --- (In reply to blake from comment #16) > Created attachment 306799 [details] > attachment-12508-0.html > > At least for me, I haven't noticed any performance hit. These systems are > all headless though. > > On Sat, Aug 31, 2024, at 6:51 AM, bugzilla-daemon@xxxxxxxxxx wrote: > > https://bugzilla.kernel.org/show_bug.cgi?id=219009 > > > > --- Comment #14 from h4ck3r (michal.litwinczuk@xxxxx) --- > > (In reply to blake from comment #13) > > > I recently experienced this. I built a proxmox cluster with 7950x. Every > > > node that I tested on would hard reset with no logs when a VM was doing > > > nested virtualization. > > > > > > Our CI testing uses VMs, and putting the CI in a VM itself makes it > pretty > > > easy to reproduce, just takes some time. > > > > > > Setting kvm_amd.vls=0 seems to have resolved the issue, we had zero node > > > resets today and I was trying to force them. > > > > > > Kernel is Proxmox's 6.8.12-1-pve. > > > > > > Thanks, > > > Blake > > > > Disabling vls does help with crashes, but has too much performance penalty. > > In my case it would lock gpu utilization at 40% max. (proxmox win10/11 > hyperv > > enabled) > > > > -- > > You may reply to this email to add a comment. > > > > You are receiving this mail because: > > You are on the CC list for the bug. the performance hit for me hasn't been that bad but disabling vls is the same as disabling svm for the vm its more like telling the VM the CPU is incapable of nested virtualization that is what I observed -- You may reply to this email to add a comment. You are receiving this mail because: You are watching the assignee of the bug.
