VirtualBox 7.2.4 issues with e1000 reset adapter and virtio-net freeze vm

[mbauer]

Hello everyone,

I'm using Debian 13 (kernel 6.12.43+deb13-amd64) on the physical host, which uses an AMD Ryzen 7 5700G processor.

My virtual machines are also using the Debian 13 OS (kernel 6.12.43+deb13-amd64).

The VirtualBox version is 7.2.4. Basically, I'm performing load tests between the VMs in a database rotation and others as nodes in a Kubernetes cluster.

I performed tests with a trocar chipset (ICH9, PIIX3), enabling Nested VT-x/AMD-V, Paravirtualization Interface as KVM, and also as the default.
I tested the network type Intel PRO/1000 MT DESKTOP (82540EM) e Intel PRO/1000 T Server (82543GC)

I disabled offloads as some forums suggest using the command:
ethtool -K enp0s3 gso off tso off rxvlan off txvlan off gro off tx off rx off sg off

And the problem below persists:
[qua nov 26 09:25:31 2025] e1000 0000:00:03.0 enp0s3: NETDEV WATCHDOG: CPU: 1: transmit queue 0 timed out 5004 ms
[qua nov 26 09:25:31 2025] e1000 0000:00:03.0 enp0s3: Reset adapter
[qua nov 26 09:25:33 2025] e1000: enp0s3 NIC Link is Up 1000 Mbps Full Duplex, Flow Control: RX

When testing the Virtio-Net network, a "head up" error occurs, causing the VM to freeze and requiring a restart to resume operation.

I've seen reports suggesting using PCnet-FAST III, but limiting it to a maximum of 100MB isn't ideal.

This issue occurs on all 3 physical machines and their VMs at random times.

Any help would be greatly appreciated.

[klaus]

Clearly something we don't see, and that makes it hard to investigate without a simple recipe for reproduction.

BTW, using PCNet doesn't limit the throughput to 100Mbps (just like using e1000 doesn't limit it to 1Gbps). That said, the design of PCNet tends to be less optimized and therefore needs more CPU time which usually limits the throughput to a lower value than e1000.