Solved Bhyve and Windows Server 2025

T

tennea9

Hello everyone,

Has anyone managed to install Windows Server 2025 in a VM using Bhyve?
I'm currently unable to do so, despite preparing the ISO image to bypass the usual checks. It works with 11 but not with 2025.
Is there a documented procedure available anywhere?

Thank you very much,
 
bhyve supports Windows desktop versions Vista, 7, 8/8.1/8.2, 10 and 11, and Windows Server versions 2008/2008R2, 2012/2012R2, 2016, 2019, 2022, and 2025.
 
Hi, can I ask how is your experience with Windows server 2025 running on bhyve vm? How is your hardware configuration? How many cpu cores do you give to vm, and how many memories and how much disk space?
 
Actually, I'm using it on a custom server:
H11dsi-NT rev. 2.0 with 2 × AMD 7601 CPUs.
I didn't modify the ISO.

It works fine, but only on servers compatible with version 2025.
If you can install it on a physical server, it should also work with bhyve.

For example, on OVH "SYS-4" hosted servers, it doesn't work.
I can only install 2022 on those SYS-4.

I always set at least 2 vCPUs for any Windows installation with bhyve.
RAM depends on the usage of the Windows Server:
  • For Core, 4 GB is a good starting point.
  • For Desktop, 8 GB or more is recommended.
 
So I sat up half an evening working with Claude and ChatGPT on this, and this was the short transcript of the conversation. For reference. I was trying trying to run a Windows Server 2025 guest inside a FreeBSD 15.1-RELEASE-p2 machine. I used a raw file as the backing for the vm image (on a ZFS stripe of 4 x consumer SSDs) with a recordsize=32k. The bhyve script I used:

bhyve -c sockets=1,cores=1,threads=1 \
-m 20G \
-w -H -u \
-s 0,hostbridge \
-s 1,lpc \
-s 2,virtio-net,tap4,mac="${mac}" \
-s 3,nvme,/kvm/"${vm}"/disk0,nocache \
-s 4,virtio-rnd \
-s 29,fbuf,tcp=10.11.12.134:5900,w=1024,h=768,wait -s 30,xhci,tablet \
-l bootrom,/usr/local/share/uefi-firmware/BHYVE_UEFI.fd,/kvm/"${vm}"/vars.fd,fwcfg="qemu" \
"${vm}"

The problem I had was that the host CPU was always relatively high compared with the guest (~5%)
1 vCPU 20%
2 vCPU 90%
4 vCPU 160%
8 vCPU 300%

Here's the transcript, I hope it's accurate and perhaps useful to someone in the bhyve team:

Key conclusion reached:
- The bhyve investigation established repeated guest reads of virtual HPET register 0xFED000F0, going through vmm_emulate_instruction()/vhpet_mmio_read().
- The Windows physical HPET device was disabled and BCD showed useplatformclock No, but the bhyve virtual HPET traffic remained at roughly 9,000–9,600 reads/sec.
- The proposed Server 2025 bhyve command uses UEFI, virtio networking, NVMe storage, VNC framebuffer, xHCI tablet, and 20 GB RAM; two vCPUs were suggested as an optional improvement over one.

Important DTrace evidence:
- vhpet_mmio_read at 0xFED000F0 repeatedly occurred.
- One sample showed 0xFED000F0 = 53,770 reads, versus 0xFED00100 = 2,711 and 0xFED00108 = 2,711.
- Another sample showed roughly 9,000–9,600 reads/sec at 0xFED000F0.
- Stack traces showed vmm_emulate_instruction -> vm_run and vhpet_mmio_read.
- vmm_emulate_instruction received arg1=0xfed000f0.
- FreeBSD source showed bhyve's virtual HPET is exposed at VHPET_BASE = 0xFED00000 and advertised through ACPI.

Windows checks:
- High precision event timer ACPI\PNP0103\0 was initially OK.
- It was disabled and subsequently reported Status Error.
- bcdedit /enum "{current}" showed useplatformclock No.
- The intention was to reboot and then move on rather than patch bhyve.
 
You must log in or register to reply here.
Back
Top