Recent content by kiler129

OP marked it as they said they replaced the SSD with a Sandisk one ;) I will caution you before using "disable_idle_d3" - it is a bit of a band-aid that may cause other issues as well. Some devices reasonably expect d3 support, especially if it was advertised/probed as supported earlier during...

Another datapoint from me here - the Samsung 990 EVO Plus definitely doesn't work in passthrough, even on a server platform: [ 1293.609288] vfio-pci 0000:46:00.0: resetting [ 1293.671772] vfio-pci 0000:46:00.0: reset done [ 1293.673440] vfio-pci 0000:46:00.0: Unable to change power state from...

See edits below - the crash is just a red herring of BAR 0 being unable to be assigned: =================ORIGINAL POST================= After platform migration to an EPYC 9004 system I am also unable to boot any VMs with PCIe passthrough. With ReBAR enabled in the BIOS, the stacktrace looks...

I have a very peculiar issue which I never saw before, and which took me forever to even narrow down to PVE host. My server has a single interface that carries multiple VLANs. Then VMs are attached to particular VLANs. My network config (here limited to two VLANs) is quite simple: auto ensfp0...

Thank you! I added some more details and expanded the performance tweaks. (if you don't see 3 posts wait as I triggered moderator approval :D). Indeed, this is what motivated me to write and update this piece. It's extremely hard to truly understand and even more so implement hypervisor tuning...

That's most likely not it - one VM that does this has 1GB assigned but the other is 32GB. The bigger one uses static huge pages while the small one does not. Anything I can try to debug this? I'm guessing it's not the VFIO modules itself as these are tied to the kernel. I can try reverting EDK...

Hm, did I miss something? I was under impression I'm not using any obsolete packages, besides not being at 7.4 yet: # pveversion pve-manager/7.3-6/723bb6ec (running kernel: 6.1.15-1-pve) root@hv-chi:~# apt update Hit:1 http://ftp.us.debian.org/debian bullseye InRelease Get:2...

@Chris: all my Win10 VMs started doing this recently and it only affects OVMF ones. It will shows "Guest did not initialized the display yet", the standard resolution change appears and switches to black screen. When they start they will run for some time and just crash silently becoming...

Use proper storage Using a VirtIO Single with a disk backed by a physical SATA drive is pretty much required to achieve predictible performance. Windows (at least 10) doesn't play well with full SCSI LUN passthru (i.e. with full SMART etc visible in the guest), but works with block device...

VM Resources Isolation This is by far the hardest part to achieve, and the more I dug into it the harder it seemed. There are conceptually four categories of isolation, in the order of increasing difficultiness: Isolating vCPUs from other VM's loads (this is solved by CPU pinning, see above)...

This post is going to be pretty long too long to fit in a single post, but it represents a summary and lessons learned over ~3 weeks of experiments. This post is a half-tutorial and half-RFC so maybe PVE can be improved, as well as a half-tutorial how to actually achieve good results. This...

Backup service wouldn't help you if you want to recover a production system quickly. Backup is more a DR solution. I think it's simply because Linux module uses compression algo which isn't supported by GRUB module, and even if booted in UEFI mode it still relies on GRUB to provide rootfs mounting.

FYI: this is still broken. On v7.3-1 the rescue boot will only print "Compression algorithm 84 is not supported" when trying to run the rescue boot.

I have a peculiar issue with one of my test machines. After dropping an additional NVMe into it Proxmox (7.3-1) "didn't boot". ...or I should say stopped responding over the network. The KVM turned out to be useless. I decided to plug in a physical monitor and a physical keyboard. Well, this...

Thank you for great links and amazing writeup on your website. Oh, yes, that's not even a question with chiplet designs - pinning is necessary to avoid cross-IOD latency. New Proxmox has a handy shortcut for that, but it will let vCPU workers roam between physical cores according to the Linux...