飞牛os，版本1.1.23安装hd530核显驱动失败

[seien210300928]

日志如下

[23:29:11] >>> 开始为虚拟机 [fnOS] 配置 GPU 分区
[23:29:11] 选中物理显卡: Intel(R) HD Graphics 530
[23:29:11] 设备路径: 
[23:29:11] 正在执行: 环境准备...
[23:29:11] [成功] 环境准备: 策略已成功应用。
[23:29:11] 正在执行: 电源检查...
[23:29:12] [成功] 电源检查: 虚拟机已关机。
[23:29:12] 正在执行: 系统优化...
[23:29:12] [成功] 系统优化: MMIO 地址空间配置完成。
[23:29:12] 正在执行: 分配显卡...
[23:29:19] [成功] 分配显卡: GPU 分区已成功创建。
[23:29:19] 正在执行: 驱动安装...
[23:29:19] 正在扫描所有挂载硬盘的分区...
[23:29:22] 扫描完成：共发现 1 个分区，分布在 1 个磁盘上。
[23:29:22] 请手动选择操作系统所在的分区...
[23:30:36] 远程环境初始化完成，临时目录: [fnOS.vhdx] 分区 2 (63.9 GB)
[23:30:36] 正在自动启动虚拟机以初始化网络...
[23:30:40] 正在嗅探虚拟机 IP 地址...
[23:30:40] 获取到 MAC 地址: 00:15:5D:01:45:02，正在匹配 IP...
[23:30:40] 成功获取 IP 地址: 192.168.1.67
[23:31:00] >>> 开始 Linux 部署流程
[23:31:00] 连接参数 - Host: 192.168.1.67, Port: 22, User: seien
[23:31:01] 正在等待虚拟机网络就绪并获取 IP...
[23:31:01] 虚拟机 IP 已获取: 192.168.1.67，正在建立 SSH 连接...
[23:31:02] 远程环境初始化完成，临时目录: /home/seien/exhyperv_deploy
[23:31:02] 正在定位并上传宿主机 GPU 驱动...
[23:32:54] 正在上传 WSL 依赖库...
[23:33:08] 正在下载配置脚本...
[23:33:09] --2026-03-02 23:33:09--  https://raw.githubusercontent.com/Justsenger/ExHyperV/main/src/Linux/script/install_dxgkrnl.sh

[23:33:09] Connecting to 192.168.1.68:7897... 
[23:35:23] failed: Connection timed out.
Retrying.


[23:35:24] --2026-03-02 23:35:24--  (try: 2)  https://raw.githubusercontent.com/Justsenger/ExHyperV/main/src/Linux/script/install_dxgkrnl.sh
Connecting to 192.168.1.68:7897... 
[23:35:24] connected.

[23:35:24] Proxy request sent, awaiting response... 
[23:35:24] 200 OK
Length: 22398 (22K) [text/plain]
Saving to: ‘/home/seien/exhyperv_deploy/install_dxgkrnl.sh’

[23:35:24] 
     0K .......... ....
[23:35:25] ...... .                               100%  580K=0.04s

2026-03-02 23:35:25 (580 KB/s) - ‘/home/seien/exhyperv_deploy/install_dxgkrnl.sh’ saved [22398/22398]


[23:35:25] --2026-03-02 23:35:25--  https://raw.githubusercontent.com/Justsenger/ExHyperV/main/src/Linux/script/configure_system.sh

[23:35:25] Connecting to 192.168.1.68:7897... 
[23:35:25] connected.

[23:35:25] Proxy request sent, awaiting response... 
[23:35:25] 200 OK
Length: 5347 (5.2K) [text/plain]
Saving to: ‘/home/seien/exhyperv_deploy/configure_system.sh’

     0K ..
[23:35:25] ...                                                 100% 11.3M=0s

2026-03-02 23:35:26 (11.3 MB/s) - ‘/home/seien/exhyperv_deploy/configure_system.sh’ saved [5347/5347]


[23:35:25] --2026-03-02 23:35:26--  https://raw.githubusercontent.com/Justsenger/ExHyperV/main/src/Linux/script/setup_graphics.sh

[23:35:25] Connecting to 192.168.1.68:7897... 
[23:35:25] connected.

[23:35:25] Proxy request sent, awaiting response... 
[23:35:26] 200 OK
Length: 4082 (4.0K) [text/plain]
Saving to: ‘/home/seien/exhyperv_deploy/setup_graphics.sh’

[23:35:26] 
     0K ...                                                   100% 30.3M=0s

2026-03-02 23:35:26 (30.3 MB/s) - ‘/home/seien/exhyperv_deploy/setup_graphics.sh’ saved [4082/4082]


[23:35:26] 正在编译安装 dxgkrnl 内核模块 (这可能需要一些时间)...
[23:35:26] ========================================
  dxgkrnl Installation Script
========================================
Initial target kernel: 6.12.18-trim

[23:35:26] Current running kernel: 6.12.18-trim

Basic dependencies are already installed.

========================================
  Checking Kernel Headers
========================================
Target kernel version: 6.12.18-trim
✓ Kernel headers found for 6.12.18-trim

[23:35:26] 
========================================
  Preparing Source Code
========================================
Detected kernel version: 6.12
Using branch: linux-msft-wsl-6.6.y (for kernel 6.6+)
Cloning WSL2-Linux-Kernel repository...

[23:35:26] Cloning into '/tmp/WSL2-Linux-Kernel'...

[23:36:35] Already on 'linux-msft-wsl-6.6.y'

[23:36:35] Your branch is up to date with 'origin/linux-msft-wsl-6.6.y'.

[23:36:35] 
========================================
  Build Information
========================================
Target kernel: 6.12.18-trim
Module version: 308063c46
Source branch: linux-msft-wsl-6.6.y


========================================
  Applying Patches
========================================
Applying patches for 6.6 branch...
  - 0001-Add-a-gpu-pv-support.patch (Shared)

[23:36:36] Checking patch drivers/hv/dxgkrnl/dxgmodule.c...
Applied patch drivers/hv/dxgkrnl/dxgmodule.c cleanly.

[23:36:36]   - 0002-Fix-eventfd_signal.patch

[23:36:37] Checking patch drivers/hv/dxgkrnl/dxgmodule.c...
Applied patch drivers/hv/dxgkrnl/dxgmodule.c cleanly.

[23:36:37] 
========================================
  Installing Module Files

[23:36:37] ========================================
Copying dxgkrnl driver...

[23:36:38] Copying include files...

[23:36:38] Checking eventfd_signal API compatibility...

[23:36:38] ✓ Kernel uses 1-parameter eventfd_signal API, no fix needed
Configuring Makefile...
Creating DKMS configuration...
✓ Module files installed to /usr/src/dxgkrnl-308063c46

========================================
  Building and Installing DKMS Module
========================================
Verifying kernel headers...

[23:36:38] ✓ Headers path: /usr/src/linux-headers-6.12.18-trim

[23:36:38] Removing existing dxgkrnl/308063c46...

[23:36:38] Deleting module dxgkrnl-308063c46 completely from the DKMS tree.

[23:36:38] 
Adding module to DKMS...

[23:36:38] Creating symlink /var/lib/dkms/dxgkrnl/308063c46/source -> /usr/src/dxgkrnl-308063c46

[23:36:38] 
Building module (this may take a few minutes)...

[23:36:38] Sign command: /lib/modules/6.12.18-trim/build/scripts/sign-file

[23:36:38] Signing key: /var/lib/dkms/mok.key
Public certificate (MOK): /var/lib/dkms/mok.pub

[23:36:38] 
Building module:
Cleaning build area...
[23:36:38] make -j2 KERNELRELEASE=6.12.18-trim -C /lib/modules/6.12.18-trim/build M=/var/lib/dkms/dxgkrnl/308063c46/build...
[23:36:40] (bad exit status: 2)

[23:36:40] Error! Bad return status for module build on kernel: 6.12.18-trim (x86_64)
Consult /var/lib/dkms/dxgkrnl/308063c46/build/make.log for more information.

[23:36:40] 
========================================
  Build Failed
========================================
DKMS build failed. Please check the build log for details:
  /var/lib/dkms/dxgkrnl/308063c46/build/make.log

Common issues:
  1. Kernel version too new - patches may not be compatible
  2. Missing dependencies - ensure all build tools are installed
  3. Kernel headers mismatch - verify headers match kernel version

Showing last 50 lines of build log:
----------------------------------------

[23:36:41] DKMS make.log for dxgkrnl-308063c46 for kernel 6.12.18-trim (x86_64)
Mon Mar  2 11:36:39 PM CST 2026
make: Entering directory '/usr/src/linux-headers-6.12.18-trim'
  CC [M]  /var/lib/dkms/dxgkrnl/308063c46/build/dxgmodule.o
  CC [M]  /var/lib/dkms/dxgkrnl/308063c46/build/hmgr.o
/var/lib/dkms/dxgkrnl/308063c46/build/hmgr.c: In function ‘expand_table’:
/var/lib/dkms/dxgkrnl/308063c46/build/hmgr.c:196:13: error: implicit declaration of function ‘vzalloc’; did you mean ‘kzalloc’? [-Werror=implicit-function-declaration]
  196 |             vzalloc(new_table_size * sizeof(struct hmgrentry));
      |             ^~~~~~~
      |             kzalloc
/var/lib/dkms/dxgkrnl/308063c46/build/hmgr.c:195:21: warning: cast to pointer from integer of different size [-Wint-to-pointer-cast]
  195 |         new_entry = (struct hmgrentry *)
      |                     ^
/var/lib/dkms/dxgkrnl/308063c46/build/dxgmodule.c: In function ‘dxg_pci_probe_device’:
/var/lib/dkms/dxgkrnl/308063c46/build/dxgmodule.c:615:43: wa
[23:36:41] rning: suggest braces around empty body in an ‘if’ statement [-Wempty-body]
  615 |                 DXG_TRACE("err: %d",  ret);
      |                                           ^
/var/lib/dkms/dxgkrnl/308063c46/build/hmgr.c:205:17: error: implicit declaration of function ‘vfree’; did you mean ‘kvfree’? [-Werror=implicit-function-declaration]
  205 |                 vfree(table->entry_table);
      |                 ^~~~~
      |                 kvfree
cc1: some warnings being treated as errors
make[2]: *** [scripts/Makefile.build:229: /var/lib/dkms/dxgkrnl/308063c46/build/hmgr.o] Error 1
make[2]: *** Waiting for unfinished jobs....
make[1]: *** [/usr/src/linux-headers-6.12.18-trim/Makefile:1932: /var/lib/dkms/dxgkrnl/308063c46/build] Error 2
make: *** [Makefile:224: __sub-make] Error 2
make: Leaving directory '/usr/src/linux-headers-6.12.18-trim'
----------------------------------------

[23:36:41] [回滚] Linux 部署失败，正在移除 GPU 分区...
[23:36:41] >>> [严重错误] Linux 部署失败: Linux 配置失败: dxgkrnl 编译脚本执行失败，请检查日志。

[Justsenger]

6.12的内核太高了，还没有加入支持

[Justsenger]

Linux 部分需要等一次大规模的适配。

[niuzheng168]

有没有什么文档可以说明以下如何build gpu driver 到新的kernel

看看能不能用Claude把这个干了

我在hyper里的飞牛没有GPU用 好急

[Justsenger]

有没有什么文档可以说明以下如何build gpu driver 到新的kernel

看看能不能用Claude把这个干了

我在hyper里的飞牛没有GPU用 好急

FnOS主要利用GPU做什么？dxgkrnl-6.12-patched.tar.gz 刚刚发布了，调用 CUDA 没问题，但是 Mesa 不行。

[niuzheng168]

有没有什么文档可以说明以下如何build gpu driver 到新的kernel
看看能不能用Claude把这个干了
我在hyper里的飞牛没有GPU用 好急

FnOS主要利用GPU做什么？dxgkrnl-6.12-patched.tar.gz 刚刚发布了，调用 CUDA 没问题，但是 Mesa 不行。

mtphoto 有一些 识别的模型 可以用gpu加速 应该是能跑onnxruntime就行

[niuzheng168]

有没有什么文档可以说明以下如何build gpu driver 到新的kernel
看看能不能用Claude把这个干了
我在hyper里的飞牛没有GPU用 好急

FnOS主要利用GPU做什么？dxgkrnl-6.12-patched.tar.gz 刚刚发布了，调用 CUDA 没问题，但是 Mesa 不行。

尝试了一次，安装卡在了

[21:08:41] [STEP: Configuring Kernel Modules Strategy (vgem & dxgkrnl)...]

[21:08:41]  -> Updating initramfs...

[21:08:41] update-initramfs: Generating /boot/initrd.img-6.12.18-trim

[21:08:53]  -> Creating late-load script...

[21:08:53] Created symlink /etc/systemd/system/graphical.target.wants/load-dxg-late.service → /etc/systemd/system/load-dxg-late.service.

把repo扔给opus-4.6

now I am on the guest ubuntu, and the install driver log can be found /home/largeniu/install_driver.log
but it stuck at [21:08:53] Created symlink /etc/systemd/system/graphical.target.wants/load-dxg-late.service → /etc/systemd/system/load-dxg-late.service.
can you help to check why and fix

agent干了一通，搞定了，我试了vllm+qwen-tts，没问题, 这是agent的summary，现在就是在飞牛里不显示这个显卡，飞牛相册的AI加速用不了，但我觉得这是fnos自己的问题

GPU-PV Driver Setup Summary (Ubuntu 24.04 / FnOS on Hyper-V)

Hardware

• Host GPU: NVIDIA GeForce RTX 3090 Ti
• VM: FnOS (Ubuntu 24.04 based), kernel 6.12.18-trim
• Virtualization: Hyper-V GPU-PV (GPU Partitioning)

What the ExHyperV install script did (steps 1-6)

1. Installed dependencies (git, curl, dkms, build-essential, aria2, etc.)
2. Verified kernel headers for 6.12.18-trim
3. Compiled & installed dxgkrnl module via DKMS (pre-patched 6.12 package)
4. Deployed WSL core libraries (libd3d12.so, libd3d12core.so, libdxcore.so) to /usr/lib/wsl/lib/
5. Copied host NVIDIA driver files to /usr/lib/wsl/drivers/
6. Deployed nvidia-smi to /usr/bin/

Where it broke (step 7)

The script created a systemd service load-dxg-late.service with WantedBy=graphical.target. FnOS never reaches graphical.target, so systemctl start hung indefinitely, blocking steps 8-9.

What we fixed

1. Service unit

Changed from graphical.target to basic.target / sysinit.target, and Type=simple to Type=oneshot:

[Unit]
After=sysinit.target

[Service]
Type=oneshot

[Install]
WantedBy=basic.target

2. Environment variables

Manually completed step 8:

GALLIUM_DRIVER=d3d12
DRI_PRIME=1
LIBVA_DRIVER_NAME=d3d12

3. pci-hyperv module

Was not loaded after reboot, meaning no virtual PCI bus existed for the GPU-PV device. Once loaded (by the system on a subsequent boot), lspci showed Microsoft Corporation Basic Render Driver and dxgkrnl bound to it successfully.

4. Cleanup

Removed /tmp/exhyperv_deploy

Verification

nvidia-smi     → RTX 3090 Ti, Driver 595.97, CUDA 13.2
docker --gpus  → PyTorch inference at ~1M samples/sec

Key Takeaway

The two root causes were:

• graphical.target never activates on headless FnOS → service hung → script never finished
• pci-hyperv module must be loaded to create the virtual PCI bus for GPU-PV device binding

[Justsenger]

@niuzheng168 那么请问实际使用中能为FnOS提供哪些功能的加速呢？综合你的回复后会考虑看是否有必要做一个适配。

[Justsenger]

#194