Segfault when copy from PCIE SSD to USB 3.0 HDD

[matius]

Hi experts,

I have been pulling out my hair over this one. Searched everywhere and have found no solution.
My Proxmox host is a Dell Optiplex Micro 7010 64GB RAM, NVME 512 SSD, 2.5 Inch SATA SSD 512 GB (PCIE Passthrough), USB 3.0 2 TB HDD (USB Passthrough).
I have both the 512 SATA SSD and 2 TB HDD mounted to a Ubuntu VM and am trying to transfer 400 GB of Plex movies from the SSD to the HDD. The Ubuntu server continually shuts down about 30 GB in to the transfer due to a kernel panic on the proxmox host I get this segfault error.
I have detached both disks and mounted them directly to the Proxmox host and tried to transfer the movies it all works fine. Could someone please help me understand why this might be happening on the Ubuntu VM? It is crucial for me to be confident that this server won't randomly crash as soon as I get some USB IO activity.

Apr 15 08:14:38 pve kernel: kvm[4655]: segfault at 100c ip 000059c9ae8e6fd0 sp 00007ffff71f14b0 error 4 in qemu-system-x86_64[59c9ae5ff000+6b8000] likely on CPU 0 (core 0, socket 0)
Apr 15 08:14:38 pve kernel: Code: 46 9b 50 00 8b 04 81 89 42 54 80 7b 10 00 0f 85 d6 00 00 00 8b 05 38 aa 46 01 4c 8b 6b 08 85 c0 0f 85 74 01 00 00 49 8b 45 10 <80> 78 0c 00 74 0a 80 78 01 69 0f 84 10 01 00 00 48 8d 35 a4 cb 4b
Apr 15 08:14:38 pve kernel: vmbr0: port 2(tap100i0) entered disabled state
Apr 15 08:14:38 pve kernel: tap100i0 (unregistering): left allmulticast mode
Apr 15 08:14:38 pve kernel: vmbr0: port 2(tap100i0) entered disabled state
Apr 15 08:14:38 pve kernel: vmbr0v14: port 2(tap100i1) entered disabled state
Apr 15 08:14:38 pve kernel: tap100i1 (unregistering): left allmulticast mode
Apr 15 08:14:38 pve kernel: vmbr0v14: port 2(tap100i1) entered disabled state
Apr 15 08:14:38 pve kernel: vmbr0v13: port 2(tap100i2) entered disabled state
Apr 15 08:14:38 pve kernel: tap100i2 (unregistering): left allmulticast mode
Apr 15 08:14:38 pve kernel: vmbr0v13: port 2(tap100i2) entered disabled state
Apr 15 08:14:38 pve systemd[1]: 100.scope: Deactivated successfully.

 

[Lukas Moravek]

Hi @matius,

With PVE 8.4 there is supported virtiofs, this is the way I would go. In my case USB HDD passthrough also caused problems and I never go back to the this solution. That time found in some post, this is not recommend (can't find it now)

https://www.proxmox.com/en/about/company-details/press-releases/proxmox-virtual-environment-8-4

L.

 

[matius]

Thank you for your quick reply Lukas, are there are downsides to using virtiofs over USB HDD passthrough? Is speed or IOPS impacted?
I am also using a 4TB Samsung T7 SSD for Time Machine over Samba on my Ubuntu VM.
Do you suggest using virtiofs for USB SSDs or should I continue using USB passthrough (I haven't had any problems with it as of yet)?

 

[Lukas Moravek]

Hi @matius,

virtiofs, is in my plan next week to make some tests , including performance. Regarding USB passthrough, it works fine until incread IO operations.

 

[Lukas Moravek]

Hi @matius

Performance test results of Virtiofs.

Host copy (same disk - NMVe, different dir as target) on Proxmox 8.4.1, "Windows.iso" 4,49 GB ~1900 MB/s of transfer rate, this values is as reference. Consider that R/W operation is on same disk.

Host to guest copy (guest VM use same NVMe disk as host)
- First Copy "Windows.iso" 4,49 GB - first half of copy ~350MB/s, than lowered to ~130 MB/s of transfer rate
- Second Copy "Debian-12_9.iso" 4 GB first - half of copy ~350MB/s, than lowered to ~230 MB/s of transfer rate
- Second Copy "Debian-12_10.iso" 4GB - first half of copy ~350MB/s, than lowered to ~240 MB/s of transfer rate

First copy was slower, but the following were significantly faster, there was about five seconds pause between copy.

Also our Lab is based on i5 Gen 12, NMVE SSD, 64 RAM on each node.