hugepage allocation failed (startup after a shutdown)

[ferociousmilkyway]

Hello,

I've recently begun using the "hugepages: any" setting in my pci passthrough VM's, and I'm seeing improved DPC latency results and general performance.

VM's will always start from a fresh boot with with this flag, but after a shutdown and start up of a "hugepages: any" VM, I often see the following error:

start failed: hugepage allocation failed at /usr/share/perl5/PVE/QemuServer/Memory.pm line 532.

Once this occurs, the VM will not start unless I remove the "hugepages" conf setting, or reboot the whole VM server. My system supports and allocates 1GB hugepages with the "any" setting.

I have no supporting GRUB statements, and besides this issue, it's working great.

It seems that KVM/QEMU is not releasing the hugepages on shutdown?

Any help greatly appreciated! Thanks for the awesome OS.

 

[dcsapak]

we write 0 to /sys/kernel/mm/hugepages/hugepages-${SIZE}/nr_hugepages which should release them after the vm is shutdown
of course you could be unlucky and the memory fragmentation does not allow for a 1gb hugepage to be allocated again

 

[AlexLup]

Got the same experience! Works at first, all good, then it stops, need to reboot the pve node..

 

[soupdiver]

Same here... is there any way to get around this?
Funny thing is, sometime I can restart the VM dozen of times and sometimes not even once