When using the e1000 emulation in bhyve with Windows guests, it's possible to get a lockup where a single vCPU spins at 100%. This can be reproduced within 10 minutes or so by running iperf3 on Windows to the host. In normal use, the lockup can take hours or even weeks, but seems to eventually occur. A workaround is to use the virtio network drivers on Windows with the virtio-net emulation on bhyve.
Has anyone run into this recently, or has this been fixed upstream? I ran into what this describes today (using a build form Feb 2018) while copying files over RDP to a VM ~300 MB of files over a local network was good enough to hang the VM. Even the VNC console was unresponsive after this and htop shows 100% CPU. My bhyve VM is set to 2 virtual cpus. Let me know if there is anything I can do to help triage/fix this. I can confirm virtio does not have this problem, have copied a few GBs of data over RDP without any issues.
ran into the same problem on a 12.2-RELEASE-p6 system. Windows sometimes crashes after a few minutes, and bhyve exiting shows: rdmsr to register 0x3a on vcpu 0 e82545: TSO hdrlen too small for vlan insertion (0 vs 12) -- dropped e82545: TSO hdrlen too small for vlan insertion (0 vs 12) -- dropped Assertion failed: (dtype == ntype), function e82545_transmit, file /usr/src/usr.sbin/bhyve/pci_e82545.c, line 1137. Abort trap rdmsr to register 0x3a on vcpu 0 Segmentation fault
I believe this was fixed with this commit: "bhyve: Update TX descriptor base address and host mapping on change" https://github.com/freebsd/freebsd-src/commit/60dc6bee This change was MFCd to stable/12, but later than the 12.2 release.