GreenDam
如何给NVIDIA的vgpu-kvm驱动打补丁强开vGPU
前言
因为nvidia的vGPU是越来越没有折腾的价值的
新款GPU据说都走上了SR-IOV的道路
我是没有钱买新卡的，现在还在用老款呢
Maxwell、Pascal 和 Volta 这几个老架构也准备放弃支持了
估计是全部要转向用GSP那些东西了
我看互联网上好像没有人说这个驱动应该怎么patch，都是直接给成品的
因为可能是最后同步更新新版本的vGPU了，故此最后做个笔记
反正接下来的版本估计是没有新特性了
后来人如果想接着更新也可以有点参考
本文不包括libvgpu_unlock_rs.so修改配置与伪装的部分
这个部分有开源代码可以直接自己阅读
杂种驱动以及集成libvgpu_unlock_rs.so的包后续再写文章，不然本篇就太长了
关于新款
很久以前咸鱼搞了张3070M碰过一阵子，然后没玩几天卡就去世了
只能说大魔改矿卡寿命堪忧
3070M魔改卡，打起游戏来比我1060还有P106舒服多了
首先要patch后切换gpumode为compute
然后才是去尝试弄vGPU
后面卡去世了就没接着弄了
版本说明
本文使用的vGPU包文件为
NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm
如果版本不同的话可以自己变通一下
解包
反正就是，文件拖下来
给个可执行权限
使用--extract-only --target参数解压
没什么好说的
root@pve:~/vgpu# wget http://fnos.makedie.net.kp:5244/d/volume1/OpenFolder/P106/kvm/17.5/NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm.run
--2025-01-27 14:54:41-- http://fnos.makedie.net.kp:5244/d/volume1/OpenFolder/P106/kvm/17.5/NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm.run
Resolving fnos.makedie.net.kp (fnos.makedie.net.kp)... 175.45.178.214
Connecting to fnos.makedie.net.kp (fnos.makedie.net.kp)|175.45.178.214|:5244... connected.
HTTP request sent, awaiting response... 200 OK
Length: 53064480 (51M) [text/plain]
Saving to: ‘NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm.run’
NVIDIA-Linux-x86_64-550.144.02-vgpu 100%[================================================================>] 50.61M 108MB/s in 0.5s
2025-01-27 14:54:42 (108 MB/s) - ‘NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm.run’ saved [53064480/53064480]
root@pve:~/vgpu# chmod a+x NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm.run
root@pve:~/vgpu# ./NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm.run --extract-only --target NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm
Creating directory NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm
Verifying archive integrity... OK
Uncompressing NVIDIA Accelerated Graphics Driver for Linux-x86_64 550.144.02.............................................................................................................................................................................................................................................................................................................................................................................................................................................................................
patch
nvidia-vgpud
这个文件就在根目录，不需要找
这一步主要是支持TeslaP4之类的显卡，可以将旧配置丢进去直接用
这个文件在NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm/nvidia-vgpud
定位到这里
.text:0000000000402F8A jz loc_403665
此处有一个跳转
查看graph的话是这样的，对部分人来说可能看图更好辨识吧
将此处直接nop掉就行了
保存后会有提示，记得看一眼别改完了忘记保存
然后出问题了排查一整天也没有头绪
Applied 6/6 patch(es)
nv-kernel.o_binary
这个文件在包里有两个，此处指的是
NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm\kernel\nvidia\nv-kernel.o_binary
第一处
我们先看看此处
/*!
* @brief This command is used to check if vGPU Software licensing is supported on
* the system and to fetch the list of feature(s) that can be licensed
* on the system.
* Note: This command also returns a dynamic signature that is used in
* the computation of response signature passed back in a subsequent
* command to enable vGPU software license.
*
* @return :
* NV_OK Success
*
*/
NV_STATUS
subdeviceCtrlCmdGpuGetLicensableFeatures_IMPL
(
Subdevice *pSubdevice,
NV2080_CTRL_GPU_GET_LICENSABLE_FEATURES_PARAMS *pParams
)
{
OBJGPU *pGpu = GPU_RES_GET_GPU(pSubdevice);
NvU32 chipID;
NV_GRID_LICENSE_FEATURE_CODE enabledFeature;
portMemSet(pParams, 0, sizeof(NV2080_CTRL_GPU_GET_LICENSABLE_FEATURES_PARAMS));
if (IS_GRID_LICENSED_FEATURES_DISABLED(pGpu))
{
pParams->isLicenseSupported = NV_FALSE;
}
else
{
pParams->isLicenseSupported = isGridLicenseSupported(pGpu);
}
if (!pParams->isLicenseSupported)
{
NV_PRINTF(LEVEL_INFO,
"vGPU Software Licensing not supported on GPU with PCI Device ID: %x\n",
pGpu->idInfo.PCIDeviceID);
return NV_OK;
}
enabledFeature = s_enabledGridFeatureState[pGpu->gpuInstance].enabledFeature;
chipID = DRF_VAL(_PCI, _DEVID, _DEVICE, pGpu->idInfo.PCIDeviceID);
/* vGPU Software Licensing is supported. Fill in the licensable feature
codes and corresponding license info */
if (IS_GRID_FEATURE_VGPU_SUPPORTED(pGpu))
{
VGPU_STATIC_INFO *pVSI = GPU_GET_STATIC_INFO(pGpu);
if (pVSI != NULL)
{
pParams->licensableFeatures[pParams->licensableFeaturesCount].featureCode = NV_GRID_LICENSE_FEATURE_CODE_VGPU;
pParams->licensableFeatures[pParams->licensableFeaturesCount].isFeatureEnabled = (enabledFeature == NV_GRID_LICENSE_FEATURE_CODE_VGPU);
pParams->licensableFeatures[pParams->licensableFeaturesCount].isLicensed = (s_enabledGridFeatureState[pGpu->gpuInstance].bIsLicensed &&
(enabledFeature == NV_GRID_LICENSE_FEATURE_CODE_VGPU));
#if RMCFG_FEATURE_ENABLED(GRID_LICENSE_MANAGER) && !defined(NV_MODS)
_getLicenseExpiryInfo(pParams, pGpu->gpuId, NV_GRID_LICENSE_FEATURE_CODE_VGPU);
#endif
portMemCopy(pParams->licensableFeatures[pParams->licensableFeaturesCount].licenseInfo,
(NV_GRID_LICENSE_INFO_MAX_LENGTH - 1),
pVSI->licenseEdition,
(NV_GRID_LICENSE_INFO_MAX_LENGTH - 1));
portMemCopy(&pParams->licensableFeatures[pParams->licensableFeaturesCount].productName,
(NV_GRID_LICENSE_INFO_MAX_LENGTH - 1),
pVSI->licensedProductName,
(NV_GRID_LICENSE_INFO_MAX_LENGTH - 1));
pParams->licensableFeaturesCount++;
}
}
if (IS_GRID_FEATURE_QUADRO_SUPPORTED(pGpu) && _isGridFeatureSupportedOnSKU(chipID, NV_GRID_LICENSE_FEATURE_CODE_QUADRO))
{
pParams->licensableFeatures[pParams->licensableFeaturesCount].featureCode = NV_GRID_LICENSE_FEATURE_CODE_QUADRO;
pParams->licensableFeatures[pParams->licensableFeaturesCount].isFeatureEnabled = (enabledFeature == NV_GRID_LICENSE_FEATURE_CODE_QUADRO);
pParams->licensableFeatures[pParams->licensableFeaturesCount].isLicensed = (s_enabledGridFeatureState[pGpu->gpuInstance].bIsLicensed &&
(enabledFeature == NV_GRID_LICENSE_FEATURE_CODE_QUADRO));
#if RMCFG_FEATURE_ENABLED(GRID_LICENSE_MANAGER) && !defined(NV_MODS)
_getLicenseExpiryInfo(pParams, pGpu->gpuId, NV_GRID_LICENSE_FEATURE_CODE_QUADRO);
#endif
portMemCopy(&pParams->licensableFeatures[pParams->licensableFeaturesCount].licenseInfo,
(NV_GRID_LICENSE_INFO_MAX_LENGTH - 1),
NV_GRID_LICENSE_FEATURE_VIRTUAL_WORKSTATION_EDITION,
(NV_GRID_LICENSE_INFO_MAX_LENGTH - 1));
portMemCopy(&pParams->licensableFeatures[pParams->licensableFeaturesCount].productName,
(NV_GRID_LICENSE_INFO_MAX_LENGTH - 1),
NV_GRID_LICENSED_PRODUCT_VWS,
(NV_GRID_LICENSE_INFO_MAX_LENGTH - 1));
pParams->licensableFeaturesCount++;
}
if (IS_GRID_FEATURE_GEFORCE_SUPPORTED(pGpu) && _isGridFeatureSupportedOnSKU(chipID, NV_GRID_LICENSE_FEATURE_CODE_GAMING))
{
pParams->licensableFeatures[pParams->licensableFeaturesCount].featureCode = NV_GRID_LICENSE_FEATURE_CODE_GAMING;
pParams->licensableFeatures[pParams->licensableFeaturesCount].isFeatureEnabled = (enabledFeature == NV_GRID_LICENSE_FEATURE_CODE_GAMING);
pParams->licensableFeatures[pParams->licensableFeaturesCount].isLicensed = (s_enabledGridFeatureState[pGpu->gpuInstance].bIsLicensed &&
(enabledFeature == NV_GRID_LICENSE_FEATURE_CODE_GAMING));
#if RMCFG_FEATURE_ENABLED(GRID_LICENSE_MANAGER) && !defined(NV_MODS)
_getLicenseExpiryInfo(pParams, pGpu->gpuId, NV_GRID_LICENSE_FEATURE_CODE_GAMING);
#endif
portMemCopy(&pParams->licensableFeatures[pParams->licensableFeaturesCount].licenseInfo,
(NV_GRID_LICENSE_INFO_MAX_LENGTH - 1),
NV_GRID_LICENSE_FEATURE_GAMING_EDITION,
(NV_GRID_LICENSE_INFO_MAX_LENGTH - 1));
portMemCopy(&pParams->licensableFeatures[pParams->licensableFeaturesCount].productName,
(NV_GRID_LICENSE_INFO_MAX_LENGTH - 1),
NV_GRID_LICENSED_PRODUCT_GAMING,
(NV_GRID_LICENSE_INFO_MAX_LENGTH - 1));
pParams->licensableFeaturesCount++;
}
if (IS_GRID_FEATURE_COMPUTE_SUPPORTED(pGpu) && _isGridFeatureSupportedOnSKU(chipID, NV_GRID_LICENSE_FEATURE_CODE_COMPUTE))
{
pParams->licensableFeatures[pParams->licensableFeaturesCount].featureCode = NV_GRID_LICENSE_FEATURE_CODE_COMPUTE;
pParams->licensableFeatures[pParams->licensableFeaturesCount].isFeatureEnabled = (enabledFeature == NV_GRID_LICENSE_FEATURE_CODE_COMPUTE);
pParams->licensableFeatures[pParams->licensableFeaturesCount].isLicensed = (s_enabledGridFeatureState[pGpu->gpuInstance].bIsLicensed &&
(enabledFeature == NV_GRID_LICENSE_FEATURE_CODE_COMPUTE));
#if RMCFG_FEATURE_ENABLED(GRID_LICENSE_MANAGER) && !defined(NV_MODS)
_getLicenseExpiryInfo(pParams, pGpu->gpuId, NV_GRID_LICENSE_FEATURE_CODE_COMPUTE);
#endif
portMemCopy(&pParams->licensableFeatures[pParams->licensableFeaturesCount].licenseInfo,
(NV_GRID_LICENSE_INFO_MAX_LENGTH - 1),
NV_GRID_LICENSE_FEATURE_COMPUTE_EDITION,
(NV_GRID_LICENSE_INFO_MAX_LENGTH - 1));
portMemCopy(&pParams->licensableFeatures[pParams->licensableFeaturesCount].productName,
(NV_GRID_LICENSE_INFO_MAX_LENGTH - 1),
NV_GRID_LICENSED_PRODUCT_COMPUTE,
(NV_GRID_LICENSE_INFO_MAX_LENGTH - 1));
pParams->licensableFeaturesCount++;
}
// Fill in the signature - TBD.
return NV_OK;
}
从这很容易找到引用
/* License info string for GPU Passthrough case. */
#define NV_GRID_LICENSE_FEATURE_VIRTUAL_WORKSTATION_EDITION "Quadro-Virtual-DWS,5.0;GRID-Virtual-WS,2.0;GRID-Virtual-WS-Ext,2.0"
我们再看看此处
# integdev_gpu_drv/integ/gpu_drv/stage_rel/drivers/resman/src/kernel/virtualization/grid/grid_features.c
NvBool
isGridLicenseSupported(OBJGPU *pGpu)
{
NvU32 chipID = DRF_VAL(_PCI, _DEVID, _DEVICE, pGpu->idInfo.PCIDeviceID);
NvU32 subID = DRF_VAL(_PCI, _DEVID, _DEVICE, pGpu->idInfo.PCISubDeviceID);
NvBool gridLicenseSupport = NV_FALSE;
if (!(IS_SILICON(pGpu)))
return NV_TRUE;
switch (chipID)
{
case NV_PCI_DEVID_DEVICE_PG402_SKU40 : //Tesla M60
case NV_PCI_DEVID_DEVICE_P2754_SKU200 : //Tesla M6
case NV_PCI_DEVID_DEVICE_PG610_SKU200 : //Tesla P40
case NV_PCI_DEVID_DEVICE_PG414_SKU200 : //Tesla P4
case NV_PCI_DEVID_DEVICE_PG418_SKU200 : //Tesla P6
case NV_PCI_DEVID_DEVICE_PH400_SKU202 : //Tesla P100 PCIE 12GB
case NV_PCI_DEVID_DEVICE_PH400_SKU201 : //Tesla P100 PCIE 16GB
case NV_PCI_DEVID_DEVICE_PH403_SKU201 : //Tesla P100 SXM2 16GB
case NV_PCI_DEVID_DEVICE_PG610_SKU210 : //Tesla P10
case NV_PCI_DEVID_DEVICE_PG503_SKU201 : //Tesla V100 SXM2 16GB
case NV_PCI_DEVID_DEVICE_PG500_SKU200 : //Tesla V100 16GB
case NV_PCI_DEVID_DEVICE_PG502_SKU200 : //Tesla V100 FHHL 16GB
case NV_PCI_DEVID_DEVICE_PG500_SKU202 : //Tesla V100 PCIE 32GB
case NV_PCI_DEVID_DEVICE_PG503_SKU203 : //Tesla V100 SXM2 32GB
case NV_PCI_DEVID_DEVICE_PG183_SKU200 : //Tesla T4
case NV_PCI_DEVID_DEVICE_PG183_SKU205 : //Tesla T4G
case NV_PCI_DEVID_DEVICE_PG150_SKU210 : //GFN TU102 SKU210
case NV_PCI_DEVID_DEVICE_PG150_SKU220_215 : //GFN TU102 SKU220 and SKU215
case NV_PCI_DEVID_DEVICE_PG500_SKU212 : //Tesla V100S PCIe 32GB
case NV_PCI_DEVID_DEVICE_PG506_SKU200_202 : //NVIDIA A100
case NV_PCI_DEVID_DEVICE_P1001_SKU200_210 : //NVIDIA A100 PCIE
case NV_PCI_DEVID_DEVICE_PG133_SKU500 : //NVIDIA RTX A6000
case NV_PCI_DEVID_DEVICE_PG133_SKU200 : //NVIDIA A40
case NV_PCI_DEVID_DEVICE_PG133_SKU215 : //NVIDIA A10
case NV_PCI_DEVID_DEVICE_PG506_SKU210_212 : //NVIDIA A100 SXM4 - SKU 210 and SKU 212
case NV_PCI_DEVID_DEVICE_PG133_SKU210 : //NVIDIA A10G
case NV_PCI_DEVID_DEVICE_PG132_SKU500 : //NVIDIA A5000
case NV_PCI_DEVID_DEVICE_PG132_SKU520 : //NVIDIA A5500
case NV_PCI_DEVID_DEVICE_P1001_SKU230 : //NVIDIA A100 PCIE 80G
case NV_PCI_DEVID_DEVICE_P1004_SKU230 : //NVIDIA A100X PCIE 80G
case NV_PCI_DEVID_DEVICE_P1001_SKU205 : //NVIDIA A30
case NV_PCI_DEVID_DEVICE_P1004_SKU205 : //NVIDIA A30X
case NV_PCI_DEVID_DEVICE_PG171_SKU200_TS1 : //NVIDIA A16 TS1
case NV_PCI_DEVID_DEVICE_PG171_SKU200_PG179_SKU220 : //NVIDIA A16 / A2
case NV_PCI_DEVID_DEVICE_PG133_SKU230 : // NVIDIA A10M
case 0x20BF:
gridLicenseSupport = NV_TRUE;
break;
case NV_PCI_DEVID_DEVICE_TU102_PG150_SKU510_875 : //Quadro RTX 6000 and RTX 8000
case NV_PCI_DEVID_DEVICE_TU102_PG150_SKU231_895 : //Quadro RTX 6000P and RTX 8000P
if (gpuIsGridswOnQuadroSupported(pGpu))
{
gridLicenseSupport = NV_TRUE;
}
break;
case NV_PCI_DEVID_DEVICE_P2405_SKU40 :
// check for the 4-part IDs of GRID M40 and M40-SP
// so that they don't get listed as gridLicenseSupported
if ( (subID != NV_PCI_SUBID_DEVICE_GM107_P2405_SKU40) &&
(subID != NV_PCI_SUBID_DEVICE_GM107_P2405_SKUS40) )
gridLicenseSupport = NV_TRUE; // Tesla M10 pGPU and all its vGPU configs
break;
default:
if (IsVOLTAorBetter(pGpu))
{
#if RMCFG_MODULE_ENABLED(VBIOS)
OBJVBIOS *pVbios = GPU_GET_VBIOS(pGpu);
if (pVbios && ((pVbios->InternalUseFlags3) & (NV_BIT_DATA_INTERNAL_V2_BFLAG3_COMPUTE_ONLY)) &&
(BRANDING_TYPE_VGX == gpuDetectVgxBranding(pGpu)))
{
gridLicenseSupport = NV_TRUE;
}
#endif
}
break;
}
return gridLicenseSupport;
}
所以可以顺藤摸瓜把isGridLicenseSupported找到，然后patch
根据上面的代码，我们很容易就从引用处找到nv032696rm就是要patch的方法
不信我们可以进去看看
这一堆的硬件ID，你跟我不是我都不信好吧
虽然上面的代码已经有些旧了，但是这相似度还是很高的
这个地方也特别好patch
这个地方直接patch成
mov eax,1
retn
对应的16进制patch就是这个地方
00 44 89 E0 41 5C 41 5D 41 5E C3 0F 1F 44 00 00
F3 0F 1E FA 41 56 41 55 53 48 83 ED 20 80 BF DA
07 00 00 00 C7 45 1C 00 00 00 00 8B 9F A0 0A 00
41 56 41 55 53 48 -> B8 01 00 00 00 C3
第二处
第二处地方，因为快吃饭了，简单写写
在nv026424rm这里
这俩地方直接nop掉
.text:000000000051E70B test eax, eax
.text:000000000051E70D jz short loc_51E716
这处将0改为1
.text:000000000051E70F mov dword ptr [rbp+0Ch], 0
结果就是
对应16进制的patch就是这个地方
00 00 49 8B 84 24 B8 01 00 00 48 8D 4D 0C BA 1A
00 00 FF 4C 89 E6 E8 AD 22 C3 02 85 C0 74 07 C7
45 0C 00 00 00 00 44 0F B6 B3 75 47 00 00 45 84
85 C0 74 07 ?? ?? ?? 00 -> 90 90 90 90 ?? ?? ?? 01
安装验证
安装
将两个文件替换掉就可以了
然后使用命令安装
注意这个-m kernel
root@pve:~/vgpu/NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm# ./nvidia-installer -m kernel
验证内核驱动
老样子，输入命令验证，没有报错说明内核驱动的patch没问题
root@pve:~/vgpu/NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm# uname -a
Linux pve 6.8.12-5-pve #1 SMP PREEMPT_DYNAMIC PMX 6.8.12-5 (2024-12-03T10:26Z) x86_64 GNU/Linux
root@pve:~/vgpu/NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm# nvidia-smi
Mon Jan 27 17:31:11 2025
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.144.02 Driver Version: 550.144.02 CUDA Version: N/A |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA GeForce GTX 1060 6GB Off | 00000000:01:00.0 Off | N/A |
| 0% 41C P0 30W / 120W | 29MiB / 6144MiB | 1% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| No running processes found |
+-----------------------------------------------------------------------------------------+
root@pve:~/vgpu/NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm# nvidia-smi vgpu
Mon Jan 27 17:31:24 2025
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 550.144.02 Driver Version: 550.144.02 |
|---------------------------------+------------------------------+------------+
| GPU Name | Bus-Id | GPU-Util |
| vGPU ID Name | VM ID VM Name | vGPU-Util |
|=================================+==============================+============|
| 0 NVIDIA GeForce GTX 106... | 00000000:01:00.0 | 0% |
+---------------------------------+------------------------------+------------+
root@pve:~/vgpu/NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm#
mdev配置修改
但是我们的设备依旧没有mdev出现
这是因为我们的卡没有vGPU配置，如果使用了libvgpu_unlock_rs.so会帮你做这一步
平时你没有感知是因为驱动包已经集成了这个插件
这里直接按那个项目说明手动部署，把配置与so丢到正确的地方
mkdir /etc/vgpu_unlock
touch /etc/vgpu_unlock/profile_override.toml
mkdir /etc/systemd/system/{nvidia-vgpud.service.d,nvidia-vgpu-mgr.service.d}
echo -e "[Service]\nEnvironment=LD_PRELOAD=/opt/vgpu_unlock-rs/target/release/libvgpu_unlock_rs.so" > /etc/systemd/system/nvidia-vgpud.service.d/vgpu_unlock.conf
echo -e "[Service]\nEnvironment=LD_PRELOAD=/opt/vgpu_unlock-rs/target/release/libvgpu_unlock_rs.so" > /etc/systemd/system/nvidia-vgpu-mgr.service.d/vgpu_unlock.conf
cat /etc/systemd/system/{nvidia-vgpud.service.d,nvidia-vgpu-mgr.service.d}/*
[Service]
Environment=LD_PRELOAD=/opt/vgpu_unlock-rs/target/release/libvgpu_unlock_rs.so
[Service]
Environment=LD_PRELOAD=/opt/vgpu_unlock-rs/target/release/libvgpu_unlock_rs.so
创建好目录把libvgpu_unlock_rs.so丢进去
mkdir -p /opt/vgpu_unlock-rs/target/release
cp libvgpu_unlock_rs.so /opt/vgpu_unlock-rs/target/release/
随后重载服务并重启服务
systemctl daemon-reload
systemctl restart {nvidia-vgpud.service,nvidia-vgpu-mgr.service}
这个时候我们的mdev设备就应该出来了
root@pve:~# mdevctl types
0000:01:00.0
nvidia-760
Available instances: 24
Device API: vfio-pci
Name: NVIDIA RTXA5500-1B
Description: num_heads=4, frl_config=45, framebuffer=1024M, max_resolution=5120x2880, max_instance=24
nvidia-761
Available instances: 12
Device API: vfio-pci
Name: NVIDIA RTXA5500-2B
Description: num_heads=4, frl_config=45, framebuffer=2048M, max_resolution=5120x2880, max_instance=12
虚拟机内验证
root@pve:~# uname -a
Linux pve 6.8.12-5-pve #1 SMP PREEMPT_DYNAMIC PMX 6.8.12-5 (2024-12-03T10:26Z) x86_64 GNU/Linux
root@pve:~# nvidia-smi
Mon Jan 27 18:14:22 2025
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.144.02 Driver Version: 550.144.02 CUDA Version: N/A |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA GeForce GTX 1060 6GB On | 00000000:01:00.0 Off | N/A |
| 27% 53C P0 69W / 120W | 1043MiB / 6144MiB | 92% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| 0 N/A N/A 60395 C+G vgpu 1014MiB |
+-----------------------------------------------------------------------------------------+
root@pve:~# nvidia-smi vgpu
Mon Jan 27 18:14:24 2025
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 550.144.02 Driver Version: 550.144.02 |
|---------------------------------+------------------------------+------------+
| GPU Name | Bus-Id | GPU-Util |
| vGPU ID Name | VM ID VM Name | vGPU-Util |
|=================================+==============================+============|
| 0 NVIDIA GeForce GTX 106... | 00000000:01:00.0 | 92% |
| 3251634247 NVIDIA RTXA... | c269... windows10,debug-... | 90% |
+---------------------------------+------------------------------+------------+
示例patch
在附件中我提供了一个patch，作为示例
将NVIDIA_550.144.02_vGPU.patch丢到与NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm.run同目录下
然后执行脚本，它自己会把补丁打上
./NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm.run --apply-patch NVIDIA_550.144.02_vGPU.patch
等待它打补丁即可
root@pve:~/vgpu# ./NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm.run --apply-patch NVIDIA_550.144.02_vGPU.patch
Verifying archive integrity... OK
Uncompressing NVIDIA Accelerated Graphics Driver for Linux-x86_64 550.144.02.............................................................................................................................................................................................................................................................................................................................................................................................................................................................................
patching file nvidia-vgpud
patching file kernel/nvidia/nv-kernel.o_binary
chmod: cannot access './NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm-custom/pkg-history.txt': No such file or directory
Adding files to archive named "NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm-custom.run"...
./makeself-help-script.sh
./libglvnd_install_checker/
./libglvnd_install_checker/glvnd_check
./libglvnd_install_checker/libGLX_installcheck.so.0
./libglvnd_install_checker/egl_dummy_vendor.json
./nvidia-xid-logd
./firmware/
./firmware/gsp_tu10x.bin
./firmware/gsp_ga10x.bin
./nvidia-vgpu-mgr
./systemd/
./systemd/nvidia-sleep.sh
./systemd/system-sleep/
./systemd/system-sleep/nvidia
./.manifest
./kernel/
./kernel/count-lines.mk
./kernel/common/
./kernel/common/inc/
./kernel/common/inc/nvstatuscodes.h
./kernel/common/inc/nv-pci.h
./kernel/common/inc/nvCpuUuid.h
./kernel/common/inc/conftest.h
./kernel/common/inc/nv-msi.h
./kernel/common/inc/nv_uvm_interface.h
./kernel/common/inc/nv-procfs.h
./kernel/common/inc/nv-ioctl.h
./kernel/common/inc/nv-timer.h
./kernel/common/inc/nvmisc.h
./kernel/common/inc/nv-gpu-info.h
./kernel/common/inc/nvtypes.h
./kernel/common/inc/nv-firmware-registry.h
./kernel/common/inc/cpuopsys.h
./kernel/common/inc/nv.h
./kernel/common/inc/nv-pgprot.h
./kernel/common/inc/os/
./kernel/common/inc/os/nv_memory_type.h
./kernel/common/inc/nv-memdbg.h
./kernel/common/inc/nv-pci-types.h
./kernel/common/inc/nv-lock.h
./kernel/common/inc/nvstatus.h
./kernel/common/inc/nv-kernel-interface-api.h
./kernel/common/inc/nv-linux.h
./kernel/common/inc/nvgputypes.h
./kernel/common/inc/nv-kref.h
./kernel/common/inc/nv-kthread-q.h
./kernel/common/inc/nv-chardev-numbers.h
./kernel/common/inc/nv-hash.h
./kernel/common/inc/os-interface.h
./kernel/common/inc/nv-kthread-q-os.h
./kernel/common/inc/rm-gpu-ops.h
./kernel/common/inc/nv-modeset-interface.h
./kernel/common/inc/nv_uvm_types.h
./kernel/common/inc/nv-firmware.h
./kernel/common/inc/nv-retpoline.h
./kernel/common/inc/nv-caps.h
./kernel/common/inc/nv-ioctl-numbers.h
./kernel/common/inc/nv-list-helpers.h
./kernel/common/inc/nvlimits.h
./kernel/common/inc/nv-ioctl-numa.h
./kernel/common/inc/nv_stdarg.h
./kernel/common/inc/nv_speculation_barrier.h
./kernel/common/inc/nv-procfs-utils.h
./kernel/common/inc/nv-hypervisor.h
./kernel/common/inc/nv-time.h
./kernel/common/inc/nv-proto.h
./kernel/common/inc/nv-vgpu-vfio-interface.h
./kernel/common/inc/nv-dmabuf.h
./kernel/common/inc/nv-mm.h
./kernel/conftest.sh
./kernel/header-presence-tests.mk
./kernel/Makefile
./kernel/nvidia/
./kernel/nvidia/nvlink_linux.c
./kernel/nvidia/linux_nvswitch.h
./kernel/nvidia/nv-reg.h
./kernel/nvidia/nv-modeset-interface.c
./kernel/nvidia/nv-memdbg.c
./kernel/nvidia/libspdm_ec.c
./kernel/nvidia/nv-dma.c
./kernel/nvidia/nv-report-err.h
./kernel/nvidia/hal/
./kernel/nvidia/hal/library/
./kernel/nvidia/hal/library/cryptlib/
./kernel/nvidia/hal/library/cryptlib/cryptlib_sm2.h
./kernel/nvidia/hal/library/cryptlib/cryptlib_dh.h
./kernel/nvidia/hal/library/cryptlib/cryptlib_rng.h
./kernel/nvidia/hal/library/cryptlib/cryptlib_ec.h
./kernel/nvidia/hal/library/cryptlib/cryptlib_rsa.h
./kernel/nvidia/hal/library/cryptlib/cryptlib_ecd.h
./kernel/nvidia/hal/library/cryptlib/cryptlib_mac.h
./kernel/nvidia/hal/library/cryptlib/cryptlib_aead.h
./kernel/nvidia/hal/library/cryptlib/cryptlib_cert.h
./kernel/nvidia/hal/library/cryptlib/cryptlib_hkdf.h
./kernel/nvidia/hal/library/cryptlib/cryptlib_hash.h
./kernel/nvidia/libspdm_ecc.c
./kernel/nvidia/ioctl_nvswitch.h
./kernel/nvidia/nv-mmap.c
./kernel/nvidia/i2c_nvswitch.c
./kernel/nvidia/nv-ibmnpu.h
./kernel/nvidia/libspdm_aead.c
./kernel/nvidia/nvlink_export.h
./kernel/nvidia/export_nvswitch.h
./kernel/nvidia/nv_uvm_interface.c
./kernel/nvidia/os-usermap.c
./kernel/nvidia/nv-pci-table.h
./kernel/nvidia/nv_gpu_ops.h
./kernel/nvidia/nv-caps-imex.h
./kernel/nvidia/nvidia.Kbuild
./kernel/nvidia/nvidia-sources.Kbuild
./kernel/nvidia/nv-pat.h
./kernel/nvidia/nv-vm.c
./kernel/nvidia/internal/
./kernel/nvidia/internal/libspdm_lib_config.h
./kernel/nvidia/nv-procfs.c
./kernel/nvidia/nvlink_pci.h
./kernel/nvidia/library/
./kernel/nvidia/library/spdm_lib_config.h
./kernel/nvidia/library/cryptlib.h
./kernel/nvidia/ioctl_common_nvswitch.h
./kernel/nvidia/nv-p2p.h
./kernel/nvidia/libspdm_hkdf.c
./kernel/nvidia/nv-cray.c
./kernel/nvidia/detect-self-hosted.h
./kernel/nvidia/nv-caps.c
./kernel/nvidia/libspdm_sha.c
./kernel/nvidia/nv-rsync.h
./kernel/nvidia/nv-nano-timer.c
./kernel/nvidia/libspdm_hkdf_sha.c
./kernel/nvidia/libspdm_x509.c
./kernel/nvidia/libspdm_internal_crypt_lib.c
./kernel/nvidia/nvlink_caps.h
./kernel/nvidia/nvlink_errors.h
./kernel/nvidia/nv-ibmnpu.c
./kernel/nvidia/libspdm_hmac_sha.c
./kernel/nvidia/nv-usermap.c
./kernel/nvidia/libspdm_shash.c
./kernel/nvidia/nv-report-err.c
./kernel/nvidia/nv-dmabuf.c
./kernel/nvidia/libspdm_rand.c
./kernel/nvidia/nvlink_common.h
./kernel/nvidia/libspdm_rsa_ext.c
./kernel/nvidia/nv-pci.c
./kernel/nvidia/internal_crypt_lib.h
./kernel/nvidia/libspdm_aead_aes_gcm.c
./kernel/nvidia/nv-msi.c
./kernel/nvidia/nvlink_linux.h
./kernel/nvidia/rmp2pdefines.h
./kernel/nvidia/linux_nvswitch.c
./kernel/nvidia/libspdm_rsa.c
./kernel/nvidia/nv-rsync.c
./kernel/nvidia/nv-vtophys.c
./kernel/nvidia/nvlink_caps.c
./kernel/nvidia/nvspdm_cryptlib_extensions.h
./kernel/nvidia/nv-vgpu-vfio-interface.c
./kernel/nvidia/nv-kthread-q.c
./kernel/nvidia/nv-acpi.c
./kernel/nvidia/os-registry.c
./kernel/nvidia/os-mlock.c
./kernel/nvidia/nv-p2p.c
./kernel/nvidia/os-pci.c
./kernel/nvidia/nvlink_proto.h
./kernel/nvidia/os-interface.c
./kernel/nvidia/nv-pat.c
./kernel/nvidia/procfs_nvswitch.c
./kernel/nvidia/nv-kernel.o_binary
./kernel/nvidia/nv-i2c.c
./kernel/nvidia/nvlink_os.h
./kernel/nvidia/nv-pci-table.c
./kernel/nvidia/nv-caps-imex.c
./kernel/nvidia/nv.c
./kernel/dkms.conf
./kernel/nvidia-vgpu-vfio/
./kernel/nvidia-vgpu-vfio/nvidia-vgpu-vfio-sources.Kbuild
./kernel/nvidia-vgpu-vfio/nv-pci-table.c
./kernel/nvidia-vgpu-vfio/nv-vgpu-ioctl.h
./kernel/nvidia-vgpu-vfio/vgpu-ctldev.c
./kernel/nvidia-vgpu-vfio/vgpu-devices.h
./kernel/nvidia-vgpu-vfio/nvidia-vgpu-vfio.h
./kernel/nvidia-vgpu-vfio/vgpu-vfio-mdev.c
./kernel/nvidia-vgpu-vfio/nv-pci-table.h
./kernel/nvidia-vgpu-vfio/nvidia-vgpu-vfio.Kbuild
./kernel/nvidia-vgpu-vfio/vgpu-vfio-pci-core.c
./kernel/nvidia-vgpu-vfio/vgpu-devices.c
./kernel/nvidia-vgpu-vfio/nvidia-vgpu-vfio.c
./kernel/Kbuild
./nvidia-vgpud
./nvidia-debugdump
./init-scripts/
./init-scripts/pre-uninstall
./init-scripts/common.sh
./init-scripts/upstart/
./init-scripts/upstart/nvidia-vgpu-mgr.conf
./init-scripts/upstart/nvidia-vgpud.conf
./init-scripts/systemd/
./init-scripts/systemd/nvidia-vgpu-mgr.service
./init-scripts/systemd/nvidia-vgpud.service
./init-scripts/systemd/nvidia-xid-logd.service
./init-scripts/sysv/
./init-scripts/sysv/nvidia-vgpu-mgr
./init-scripts/sysv/nvidia-xid-logd
./init-scripts/sysv/nvidia-vgpud
./init-scripts/post-install
./init-scripts/pre-unload
./nvidia-smi
./nvidia-installer
./nvidia-modprobe
./kernel-open/
./kernel-open/header-presence-tests.mk
./kernel-open/Kbuild
./kernel-open/Makefile
./kernel-open/count-lines.mk
./kernel-open/nvidia-vgpu-vfio/
./kernel-open/nvidia-vgpu-vfio/vgpu-ctldev.c
./kernel-open/nvidia-vgpu-vfio/vgpu-devices.h
./kernel-open/nvidia-vgpu-vfio/nv-vgpu-ioctl.h
./kernel-open/nvidia-vgpu-vfio/nvidia-vgpu-vfio.Kbuild
./kernel-open/nvidia-vgpu-vfio/vgpu-vfio-mdev.c
./kernel-open/nvidia-vgpu-vfio/nvidia-vgpu-vfio.c
./kernel-open/nvidia-vgpu-vfio/nv-pci-table.c
./kernel-open/nvidia-vgpu-vfio/vgpu-devices.c
./kernel-open/nvidia-vgpu-vfio/nvidia-vgpu-vfio-sources.Kbuild
./kernel-open/nvidia-vgpu-vfio/nv-pci-table.h
./kernel-open/nvidia-vgpu-vfio/nvidia-vgpu-vfio.h
./kernel-open/nvidia-vgpu-vfio/vgpu-vfio-pci-core.c
./kernel-open/nvidia/
./kernel-open/nvidia/os-pci.c
./kernel-open/nvidia/os-mlock.c
./kernel-open/nvidia/nv-p2p.c
./kernel-open/nvidia/libspdm_rsa.c
./kernel-open/nvidia/nv-rsync.c
./kernel-open/nvidia/nv-procfs.c
./kernel-open/nvidia/libspdm_rand.c
./kernel-open/nvidia/linux_nvswitch.h
./kernel-open/nvidia/libspdm_hkdf_sha.c
./kernel-open/nvidia/libspdm_internal_crypt_lib.c
./kernel-open/nvidia/nv-report-err.c
./kernel-open/nvidia/nvidia-sources.Kbuild
./kernel-open/nvidia/nv_uvm_interface.c
./kernel-open/nvidia/nvlink_common.h
./kernel-open/nvidia/rmp2pdefines.h
./kernel-open/nvidia/nvlink_linux.h
./kernel-open/nvidia/nv-pat.c
./kernel-open/nvidia/nvlink_errors.h
./kernel-open/nvidia/libspdm_rsa_ext.c
./kernel-open/nvidia/ioctl_nvswitch.h
./kernel-open/nvidia/nv-i2c.c
./kernel-open/nvidia/nv-usermap.c
./kernel-open/nvidia/libspdm_shash.c
./kernel-open/nvidia/nv.c
./kernel-open/nvidia/internal_crypt_lib.h
./kernel-open/nvidia/nv-kthread-q.c
./kernel-open/nvidia/nvlink_proto.h
./kernel-open/nvidia/nv-memdbg.c
./kernel-open/nvidia/os-interface.c
./kernel-open/nvidia/libspdm_aead_aes_gcm.c
./kernel-open/nvidia/nv-acpi.c
./kernel-open/nvidia/os-registry.c
./kernel-open/nvidia/nvlink_caps.c
./kernel-open/nvidia/nv-pci.c
./kernel-open/nvidia/procfs_nvswitch.c
./kernel-open/nvidia/nv-vtophys.c
./kernel-open/nvidia/nv-caps-imex.c
./kernel-open/nvidia/nv-ibmnpu.h
./kernel-open/nvidia/nv-pci-table.c
./kernel-open/nvidia/ioctl_common_nvswitch.h
./kernel-open/nvidia/nv-msi.c
./kernel-open/nvidia/nvlink_linux.c
./kernel-open/nvidia/nv-pat.h
./kernel-open/nvidia/internal/
./kernel-open/nvidia/internal/libspdm_lib_config.h
./kernel-open/nvidia/nv-mmap.c
./kernel-open/nvidia/nv-vm.c
./kernel-open/nvidia/detect-self-hosted.h
./kernel-open/nvidia/nvlink_export.h
./kernel-open/nvidia/nv-kernel.o_binary
./kernel-open/nvidia/libspdm_ecc.c
./kernel-open/nvidia/nvspdm_cryptlib_extensions.h
./kernel-open/nvidia/nv-report-err.h
./kernel-open/nvidia/nv-rsync.h
./kernel-open/nvidia/linux_nvswitch.c
./kernel-open/nvidia/libspdm_ec.c
./kernel-open/nvidia/i2c_nvswitch.c
./kernel-open/nvidia/nv-p2p.h
./kernel-open/nvidia/libspdm_aead.c
./kernel-open/nvidia/library/
./kernel-open/nvidia/library/spdm_lib_config.h
./kernel-open/nvidia/library/cryptlib.h
./kernel-open/nvidia/nvlink_os.h
./kernel-open/nvidia/nv-modeset-interface.c
./kernel-open/nvidia/nv-nano-timer.c
./kernel-open/nvidia/nvlink_caps.h
./kernel-open/nvidia/nv-caps-imex.h
./kernel-open/nvidia/nv-ibmnpu.c
./kernel-open/nvidia/nv-pci-table.h
./kernel-open/nvidia/export_nvswitch.h
./kernel-open/nvidia/libspdm_sha.c
./kernel-open/nvidia/nv-caps.c
./kernel-open/nvidia/nv-dmabuf.c
./kernel-open/nvidia/nv-reg.h
./kernel-open/nvidia/nv-cray.c
./kernel-open/nvidia/libspdm_hkdf.c
./kernel-open/nvidia/nvlink_pci.h
./kernel-open/nvidia/libspdm_hmac_sha.c
./kernel-open/nvidia/hal/
./kernel-open/nvidia/hal/library/
./kernel-open/nvidia/hal/library/cryptlib/
./kernel-open/nvidia/hal/library/cryptlib/cryptlib_dh.h
./kernel-open/nvidia/hal/library/cryptlib/cryptlib_mac.h
./kernel-open/nvidia/hal/library/cryptlib/cryptlib_ec.h
./kernel-open/nvidia/hal/library/cryptlib/cryptlib_hkdf.h
./kernel-open/nvidia/hal/library/cryptlib/cryptlib_ecd.h
./kernel-open/nvidia/hal/library/cryptlib/cryptlib_hash.h
./kernel-open/nvidia/hal/library/cryptlib/cryptlib_rsa.h
./kernel-open/nvidia/hal/library/cryptlib/cryptlib_cert.h
./kernel-open/nvidia/hal/library/cryptlib/cryptlib_aead.h
./kernel-open/nvidia/hal/library/cryptlib/cryptlib_rng.h
./kernel-open/nvidia/hal/library/cryptlib/cryptlib_sm2.h
./kernel-open/nvidia/nv-dma.c
./kernel-open/nvidia/nvidia.Kbuild
./kernel-open/nvidia/nv-vgpu-vfio-interface.c
./kernel-open/nvidia/libspdm_x509.c
./kernel-open/nvidia/os-usermap.c
./kernel-open/nvidia/nv_gpu_ops.h
./kernel-open/dkms.conf
./kernel-open/common/
./kernel-open/common/inc/
./kernel-open/common/inc/nv-mm.h
./kernel-open/common/inc/nv-hypervisor.h
./kernel-open/common/inc/nv-procfs.h
./kernel-open/common/inc/nv-timer.h
./kernel-open/common/inc/nv-gpu-info.h
./kernel-open/common/inc/nv-ioctl.h
./kernel-open/common/inc/nv-firmware-registry.h
./kernel-open/common/inc/conftest.h
./kernel-open/common/inc/nv-chardev-numbers.h
./kernel-open/common/inc/nv-ioctl-numa.h
./kernel-open/common/inc/nv.h
./kernel-open/common/inc/nvCpuUuid.h
./kernel-open/common/inc/os/
./kernel-open/common/inc/os/nv_memory_type.h
./kernel-open/common/inc/nvmisc.h
./kernel-open/common/inc/nv-memdbg.h
./kernel-open/common/inc/nv-retpoline.h
./kernel-open/common/inc/nvtypes.h
./kernel-open/common/inc/nv-pgprot.h
./kernel-open/common/inc/nv-kernel-interface-api.h
./kernel-open/common/inc/os-interface.h
./kernel-open/common/inc/nv-pci.h
./kernel-open/common/inc/nv_uvm_types.h
./kernel-open/common/inc/nv-ioctl-numbers.h
./kernel-open/common/inc/nv-msi.h
./kernel-open/common/inc/nv-kthread-q.h
./kernel-open/common/inc/cpuopsys.h
./kernel-open/common/inc/nv-lock.h
./kernel-open/common/inc/nv-list-helpers.h
./kernel-open/common/inc/nv-pci-types.h
./kernel-open/common/inc/nv_uvm_interface.h
./kernel-open/common/inc/nv-hash.h
./kernel-open/common/inc/nv-kref.h
./kernel-open/common/inc/nv-firmware.h
./kernel-open/common/inc/nvgputypes.h
./kernel-open/common/inc/nv-procfs-utils.h
./kernel-open/common/inc/rm-gpu-ops.h
./kernel-open/common/inc/nv-linux.h
./kernel-open/common/inc/nvstatus.h
./kernel-open/common/inc/nv-dmabuf.h
./kernel-open/common/inc/nv-caps.h
./kernel-open/common/inc/nv-proto.h
./kernel-open/common/inc/nv-kthread-q-os.h
./kernel-open/common/inc/nv-vgpu-vfio-interface.h
./kernel-open/common/inc/nvstatuscodes.h
./kernel-open/common/inc/nv-time.h
./kernel-open/common/inc/nv_speculation_barrier.h
./kernel-open/common/inc/nvlimits.h
./kernel-open/common/inc/nv-modeset-interface.h
./kernel-open/common/inc/nv_stdarg.h
./kernel-open/conftest.sh
./nvidia-smi.1.gz
./vgpuConfig.xml
./LICENSE
./libnvidia-vgpu.so.550.144.02
./html/
./html/faq.html
./html/selectdriver.html
./html/displaydevicenames.html
./html/ngx.html
./html/profiles.html
./html/depth30.html
./html/gpunames.html
./html/nvidia-persistenced.html
./html/acknowledgements.html
./html/wayland-issues.html
./html/xineramaglx.html
./html/randr14.html
./html/installedcomponents.html
./html/addtlresources.html
./html/minimumrequirements.html
./html/glxsupport.html
./html/newusertips.html
./html/dma_issues.html
./html/powermanagement.html
./html/commonproblems.html
./html/egpu.html
./html/sli.html
./html/gsp.html
./html/flippingubb.html
./html/framelock.html
./html/nvidiasettings.html
./html/introduction.html
./html/nvidia-debugdump.html
./html/procinterface.html
./html/xwayland.html
./html/dynamicboost.html
./html/nvidia-peermem.html
./html/dynamicpowermanagement.html
./html/index.html
./html/i2c.html
./html/primerenderoffload.html
./html/xcompositeextension.html
./html/openglenvvariables.html
./html/configmultxscreens.html
./html/kernel_open.html
./html/knownissues.html
./html/retpoline.html
./html/nvidia-smi.html
./html/editxconfig.html
./html/xrandrextension.html
./html/vdpausupport.html
./html/audiosupport.html
./html/kms.html
./html/supportedchips.html
./html/installdriver.html
./html/gbm.html
./html/configtwinview.html
./html/addressingcapabilities.html
./html/configlaptop.html
./html/installationandconfiguration.html
./html/dpi.html
./html/nvidia-ml.html
./html/programmingmodes.html
./html/xconfigoptions.html
./html/optimus.html
./html/appendices.html
./GRID_LICENSE
./nvidia-modprobe.1.gz
./nvidia-bug-report.sh
./makeself.sh
./libnvidia-ml.so.550.144.02
./is_vgpu_host_package.txt
./libnvidia-vgxcfg.so.550.144.02
./supported-gpus/
./supported-gpus/supported-gpus.json
./supported-gpus/LICENSE
./nvidia-installer.1.gz
./grid-third-party-licenses.txt
./pkg-history.txt
./README.txt
./NVIDIA_Changelog
./README.vgpu
./sriov-manage
-e CRC: 528540921
MD5: 4066672be59420b2f412f626b4e80329
Self-extractible archive "NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm-custom.run" successfully created.
随后就会生成处打好补丁的NVIDIA-Linux-x86_64-550.144.02-vgpu-kvm-custom.run
随后按正常驱动方法安装即可，记得加参数不要忘记了
其他版本
18.0 570.124.03
NVIDIA-Linux-x86_64-570.124.03-vgpu-kvm.run --apply-patch NVIDIA_570.124.03_vGPU.patch
注意这个版本有两个结构体发生了变更，需要修改那个unlock_rs.so才能正常覆盖配置
NVA081_CTRL_CMD_VGPU_CONFIG_GET_VGPU_TYPE_INFO
NV0080_CTRL_CMD_GPU_GET_VIRTUALIZATION_MODE
功能验证如图所示
本版本新增的史诗级功能，支持不同显存大小vGPU
虽然之前拿配置覆盖也能实现
结束语
这回是估计是最后折腾这个vGPU了
新卡看看什么时候有钱买，旧卡感觉也就是这样了
等有钱买新卡了再去挑战一下新卡
我是觉得新卡应该是有搞头的，毕竟那3070M都出mdev了，其他卡说不定也可以呢？