Can Proxmox Sleep ? Z840 resumes from suspend by rebooting (6.17.2-1-pve, dual E5-2543 V4 06.62)

marcusbrius

New Member
Jul 21, 2026
2
0
1
Proxmox VE Suspend/Resume Bug Report - HP Z840 (works normally with 3VMs)

Environment​

Hardware: HP Z840 Workstation, BIOS M60 v02.62 (01/04/2024), Dual Intel Xeon E5-2643 v4, 128 GB RAM, ZFS root, Kingston A2000 NVMe. GPUs: Quadro K2200 (host), GTX 780 Ti, Quadro P4000.

Software​

Proxmox VE, kernel 6.17.2-1-pve, UEFI boot, ZFS root.

Issue Summary​

System successfully enters suspend but always reboots on wake. USB wake functions correctly after enabling Runtime Power Management in BIOS. Linux never logs a resume sequence.

Reproduction​

Run systemctl suspend or echo mem > /sys/power/state. Wake using USB keyboard or power button. System reboots.

Tests Performed​

VMs stopped, VFIO GPU bindings removed, hp_wmi unloaded, pci=realloc enabled, deep and s2idle tested, USB wake verified. No change.

Observed Behavior​

Logs always end with PM: suspend entry (deep). No PM: suspend exit is ever logged. last -x records crash/reboot entries.

Relevant Findings​

pci=realloc removed PCI allocation warnings but did not fix suspend. Removing VFIO bindings from GTX 780 Ti and P4000 did not change behavior. USB wake functions correctly.

Conclusion​

Failure appears to occur before Linux regains control during resume. Suspected platform firmware/ACPI resume issue affecting HP Z840 under Linux.

Extra notes :
Want Proxmox as main OS hosting Windows, and not Windows hosting Linux. (not a server, a workstation, need sleep while away)
 
"Suspend? On a virtualization platform?" : been doing so for more than a decade with hyperV and virtualBox on windows pro. The tests described here are done with no VM nor container running. I really tried isolate the issue...
 
I'm going to tell you what most people won't -- you're doing it wrong. Unless you like being unhappy, you might have to change your expectations.

Proxmox is based on Debian with a custom Ubuntu kernel, yes. But it's not really designed to run like a desktop. It's a hypervisor - most people run it 24/7 or boot it on an as-needed basis. They don't bother with sleep or hibernation, because it's not Windows. It's more of a server distro. Even trying "powersave" to the /proc cpu entries can introduce instability.

People coming from desktop-land want to try putting their drives to sleep. Servers aren't designed to run that way, it's not a mass-market desktop. Change your paradigm. Run your spinners 24/7, on UPS, and they'll usually last for years.

If you want to save power, downsize to fewer/larger drives and put a limited subset of files on SSD - they use less electricity. Instead of running a dual-Xeon, grab an inexpensive mini-pc and/or used laptop that can run your frequently-used stuff 24/7 cheaply without worrying about it.

Now you can try going down the rabbit hole of trying to file ACPI bug reports to the kernel maintainers, and it might be fixed in some unknown period of time. Or it might not.

Or you can start to use it as intended, hibernate your VMs and stop your LXCs, and just power off when you're done. The next boot will resume your hibernated VMs that were set to Autostart. Your choice.
 
  • Like
Reactions: Johannes S and UdoB