[PVE-User] VM Start timeout with pci passthrough

[leesteken at pm.me]

‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐
On Friday, March 20, 2020 12:46 PM, Mark Adams via pve-user <pve-user at pve.proxmox.com> wrote:

> Hi All,
>
> I am having the issue that is detailed in this forum post:
>
> https://forum.proxmox.com/threads/vm-start-timeout-with-pci-gpu.45843/
>
> I thought I would take it to the mailing list to see if anyone here has any
> ideas?
>
> VM's boot fine the first time the machine starts up, and even a few reboots
> work fine - but eventually the timeout start happening.
>
> If you start the vm manually it works fine.
>
> Also if you lower the ram allocation (for instance from 64 -> 32) then it
> also seems to boot fine.
>
> Any thoughts on this very welcome!

When doing PCI passthrough, KVM needs to allocate and pin (and zero out) all of the memory,
because of possible DMA from PCI devices.
This used to take quite some time (minutes), and although it has been improved,
maybe it still does. If you reduce the RAM of the VM, does it also start quicker?
If you use (1GB) hugepages, does the VM start quicker?

In my case, booting a VM with 32GB on a AMD Zen+ machine only takes seconds.
Does your system have multiple NUMA nodes or sockets? Maybe, if the amount is more than
a single node, it takes longer? What platform are you using?

Maybe the Proxmox developers can tell you where the time-out is specified,
or make the time-out dependent on the amount of RAM (if that is the cause)?

kind regards, Arjen