GPU PassThrough in KVM

其他 2020-02-25 12:50:47 阅读次数: 0

实现步骤

环境

OS：

# cat /etc/redhat-release 
CentOS Linux release 7.3.1611 (Core)
# uname -a
Linux hyhive 3.10.0-514.el7.x86_64 #1 SMP Tue Nov 22 16:42:41 UTC 2016 x86_64 x86_64 x86_64 GNU/Linux

QEMU版本

Compiled against library: libvirt 2.0.0
Using library: libvirt 2.0.0
Using API: QEMU 2.0.0
Running hypervisor: QEMU 2.6.0

操作步骤

1.查找pci设备的硬件信息

pci passthrough需要pci的vendor_id和product_id以及iommu组编号，通常再知道pci设备的类型或者厂商名称时，可以使用lspci查找，如（这里测试有三块Quadro M2000显卡）：

# lspci -nn | grep -i vga
06:00.0 VGA compatible controller [0300]: NVIDIA Corporation GM206GL [Quadro M2000] [10de:1430] (rev a1)
0a:03.0 VGA compatible controller [0300]: Matrox Electronics Systems Ltd. MGA G200eW WPCM450 [102b:0532] (rev 0a)
82:00.0 VGA compatible controller [0300]: NVIDIA Corporation GM206GL [Quadro M2000] [10de:1430] (rev a1)
83:00.0 VGA compatible controller [0300]: NVIDIA Corporation GM206GL [Quadro M2000] [10de:1430] (rev a1)

group:06:00.0       82:00.0          83:00.0
venderid:10de       10de             10de
productid:1430      1430             1430

2.打开iommu

iommu功能需要CPU和系统的支持，硬件上要求CPU支持vt-d技术（不是虚拟化）,并在bios中打开（通常在北桥芯片设置中）

配置vt-d

软件上，需要操作系统的iommu grub标志，添加rd.driver.pre=vfio-pci intel_opmmu=on

# cat /etc/sysconfig/grub 
GRUB_TIMEOUT=5
GRUB_DISTRIBUTOR="$(sed 's, release .*$,,g' /etc/system-release)"
GRUB_DEFAULT=saved
GRUB_DISABLE_SUBMENU=true
GRUB_TERMINAL_OUTPUT="console"
GRUB_CMDLINE_LINUX="crashkernel=auto rd.lvm.lv=centos/root rd.lvm.lv=centos/swap rhgb quiet rd.driver.pre=vfio-pci intel_iommu=on"
GRUB_DISABLE_RECOVERY="true"

重新生成grub：

EFI

grub2-mkconfig -o /boot/efi/EFI/centos/grub.cfg

非EFI

# grub2-mkconfig -o /boot/grub2/grub.cfg 
Generating grub configuration file ...
  WARNING: Not using lvmetad because duplicate PVs were found.
  WARNING: Use multipath or vgimportclone to resolve duplicate PVs?
  WARNING: After duplicates are resolved, run "pvscan --cache" to enable lvmetad.
  WARNING: Not using lvmetad because duplicate PVs were found.
  WARNING: Use multipath or vgimportclone to resolve duplicate PVs?
  WARNING: After duplicates are resolved, run "pvscan --cache" to enable lvmetad.
Found linux image: /boot/vmlinuz-3.10.0-514.el7.x86_64
Found initrd image: /boot/initramfs-3.10.0-514.el7.x86_64.img
Found linux image: /boot/vmlinuz-0-rescue-43b1cb10c00d48b18061c67e6df40752
Found initrd image: /boot/initramfs-0-rescue-43b1cb10c00d48b18061c67e6df40752.img
  WARNING: Not using lvmetad because duplicate PVs were found.
  WARNING: Use multipath or vgimportclone to resolve duplicate PVs?
  WARNING: After duplicates are resolved, run "pvscan --cache" to enable lvmetad.
  WARNING: Not using lvmetad because duplicate PVs were found.
  WARNING: Use multipath or vgimportclone to resolve duplicate PVs?
  WARNING: After duplicates are resolved, run "pvscan --cache" to enable lvmetad.
done

3.配置vfio模块

  1. 屏蔽宿主机驱动（该步骤验证为非必须步骤，当出现异常时尝试添加）

以nvidia显卡为例：

# cat /etc/modprobe.d/blacklist.conf
blacklist nouveau
blacklist nvidia
  1. 添加vfio所需内核模块
# cat /etc/modules-load.d/vfio.conf 
vfio
vfio_iommu_type1
vfio_pci
  1. 配置vfio

将1.1中查询到的vendor_id和product_id以vendor_id:product_id的形式添加到vfio option中，完成vfio配置

检查硬件脚本(可以通过以下脚本检查1.1中得到的id是否有误)

# cat gpu.sh 
#!/bin/bash
shopt -s nullglob
for d in /sys/kernel/iommu_groups/*/devices/*; do 
    n=${d#*/iommu_groups/*}; n=${n%%/*}
    printf 'IOMMU Group %s ' "$n"
    lspci -nns "${d##*/}"
done;

通过检查硬件脚本找到显卡对应的显卡设备和声卡设备，将两个设备id都填入vfio.conf文件中

# ./gpu.sh | grep -i nvidia
IOMMU Group 18 06:00.0 VGA compatible controller [0300]: NVIDIA Corporation GM206GL [Quadro M2000] [10de:1430] (rev a1)
IOMMU Group 18 06:00.1 Audio device [0403]: NVIDIA Corporation Device [10de:0fba] (rev a1)
IOMMU Group 38 82:00.0 VGA compatible controller [0300]: NVIDIA Corporation GM206GL [Quadro M2000] [10de:1430] (rev a1)
IOMMU Group 38 82:00.1 Audio device [0403]: NVIDIA Corporation Device [10de:0fba] (rev a1)
IOMMU Group 39 83:00.0 VGA compatible controller [0300]: NVIDIA Corporation GM206GL [Quadro M2000] [10de:1430] (rev a1)
IOMMU Group 39 83:00.1 Audio device [0403]: NVIDIA Corporation Device [10de:0fba] (rev a1)

从上面可以查到的显卡设备为10de:1430声卡设备为10de:0fba

添加到vfio.conf文件中

# cat /etc/modprobe.d/vfio.conf
options vfio-pci ids=10de:1430,10de:0fba
  1. 重新生成initramfs
# mv /boot/initramfs-$(uname -r).img /boot/initramfs-$(uname -r).img.bak
# dracut -v /boot/initramfs-$(uname -r).img $(uname -r)
Executing: /usr/sbin/dracut -v /boot/initramfs-3.10.0-514.el7.x86_64.img 3.10.0-514.el7.x86_64
dracut module 'busybox' will not be installed, because command 'busybox' could not be found!
dracut module 'dmraid' will not be installed, because command 'dmraid' could not be found!
dracut module 'busybox' will not be installed, because command 'busybox' could not be found!
dracut module 'dmraid' will not be installed, because command 'dmraid' could not be found!
*** Including module: bash ***
*** Including module: nss-softokn ***
*** Including module: i18n ***
*** Including module: network ***
*** Including module: ifcfg ***
*** Including module: drm ***
*** Including module: plymouth ***
*** Inc 大专栏  GPU PassThrough in KVMluding module: dm ***
Skipping udev rule: 64-device-mapper.rules
Skipping udev rule: 60-persistent-storage-dm.rules
Skipping udev rule: 55-dm.rules
*** Including module: kernel-modules ***
*** Including module: lvm ***
Skipping udev rule: 64-device-mapper.rules
Skipping udev rule: 56-lvm.rules
Skipping udev rule: 60-persistent-storage-lvm.rules
*** Including module: resume ***
*** Including module: rootfs-block ***
*** Including module: terminfo ***
*** Including module: udev-rules ***
Skipping udev rule: 91-permissions.rules
*** Including module: biosdevname ***
*** Including module: systemd ***
*** Including module: usrmount ***
*** Including module: base ***
*** Including module: fs-lib ***
*** Including module: shutdown ***
*** Including modules done ***
*** Installing kernel module dependencies and firmware ***
*** Installing kernel module dependencies and firmware done ***
*** Resolving executable dependencies ***
*** Resolving executable dependencies done***
*** Hardlinking files ***
*** Hardlinking files done ***
*** Stripping files ***
*** Stripping files done ***
*** Generating early-microcode cpio image contents ***
*** Constructing GenuineIntel.bin ****
*** Store current command line parameters ***
*** Creating image file ***
*** Creating microcode section ***
*** Created microcode section ***
*** Creating image file done ***
*** Creating initramfs image file '/boot/initramfs-3.10.0-514.el7.x86_64.img' done ***
  1. 重启

验证

重启之后，验证vfio是否生效

# dmesg | grep -i vfio
[    0.000000] Command line: BOOT_IMAGE=/vmlinuz-3.10.0-514.el7.x86_64 root=/dev/mapper/centos-root ro crashkernel=auto rd.lvm.lv=centos/root rd.lvm.lv=centos/swap rhgb quiet rd.driver.pre=vfio-pci intel_iommu=on
[    0.000000] Kernel command line: BOOT_IMAGE=/vmlinuz-3.10.0-514.el7.x86_64 root=/dev/mapper/centos-root ro crashkernel=auto rd.lvm.lv=centos/root rd.lvm.lv=centos/swap rhgb quiet rd.driver.pre=vfio-pci intel_iommu=on
[    8.362177] VFIO - User Level meta-driver version: 0.3
[    8.367249] vfio_pci: add [10de:1430[ffff:ffff]] class 0x000000/00000000
[    8.367314] vfio_pci: add [10de:0fba[ffff:ffff]] class 0x000000/00000000
[ 2144.638305] vfio-pci 0000:06:00.0: enabling device (0000 -> 0003)
[ 2144.638464] vfio_ecap_init: 0000:06:00.0 hiding ecap 0x1e@0x258
[ 2144.638479] vfio_ecap_init: 0000:06:00.0 hiding ecap 0x19@0x900

失败案例：

# dmesg | grep vfio 
[    0.000000] Command line: BOOT_IMAGE=/vmlinuz-3.10.0-514.el7.x86_64 root=/dev/mapper/centos-root ro crashkernel=auto rd.lvm.lv=centos/root rd.lvm.lv=centos/swap rhgb quiet rd.driver.pre=vfio-pci intel_iommu=on
[    0.000000] Kernel command line: BOOT_IMAGE=/vmlinuz-3.10.0-514.el7.x86_64 root=/dev/mapper/centos-root ro crashkernel=auto rd.lvm.lv=centos/root rd.lvm.lv=centos/swap rhgb quiet rd.driver.pre=vfio-pci intel_iommu=on
[   19.034596] vfio-pci: probe of 0000:06:00.0 failed with error -22
[   19.034636] vfio-pci: probe of 0000:82:00.0 failed with error -22
[   19.034654] vfio-pci: probe of 0000:83:00.0 failed with error -22
[   19.034661] vfio_pci: add [10de:1430[ffff:ffff]] class 0x000000/00000000

解决办法：

检查/etc/modprobe.d/vfio.conf 文件，检查主板北桥vt-d设置是否打开

调用GPU

创建虚拟机，并添加pci设备

<hostdev mode='subsystem' type='pci' managed='yes'>
      <driver name='vfio'/>
      <source>
        <address domain='0x0000' bus='0x06' slot='0x00' function='0x0'/>
      </source>
      <address type='pci' domain='0x0000' bus='0x00' slot='0x06' function='0x0'/>
    </hostdev>
    <hostdev mode='subsystem' type='pci' managed='yes'>
      <driver name='vfio'/>
      <source>
        <address domain='0x0000' bus='0x06' slot='0x00' function='0x1'/>
      </source>
      <address type='pci' domain='0x0000' bus='0x00' slot='0x07' function='0x0'/>
    </hostdev>

其中address为1.1中06:00:0

通过virsh domain ***.xml孵化出虚拟机即可。

ATI部分显卡及某些其他PCI卡reset问题

ATI部分显卡和其他某些cpi卡存在reset问题，具体表现为：

  1. 虚拟机重启一次或几次后，pci passthrough失效，需要重启宿主机
  2. 虚拟机重启时蓝屏，然后恢复正常

该问题由于pci reset导致，和pci板卡驱动程序相关，对于kvm则无根治手段，可以采用的方式：

https://curtisshoward.com/post/fixing-amd-gpu-passthrough-reset-issues-in-windows/

qemu-kvm代码中关于该问题的解释：

https://github.com/qemu/qemu/blob/master/hw/vfio/pci-quirks.c#L1697

相关资源

pci passthrough当前比较成熟，以下是部分资料：

  1. VGA Passthrough on virtual machines in CentOS 7
  2. PCI passthrough via OVMF
  3. VGAPassthrough
  4. HOW-TO make dual-boot obsolete using kvm VGA passthrough
  5. create-gaming-virtual-machine-using-vfio-pci-passthrough-kvm

和pci reset相关

  1. pci-quirks
  2. pass-through gpu to guest, but only once?!
  3. vfio pass through only works once after reboot
  4. Fixing AMD GPU Passthrough Reset Issues in Windows

在线测试结果

KVM GPU PASSTHROUGH RESULT

猜你喜欢

转载自www.cnblogs.com/lijianming180/p/12361006.html
KvM
GPU
今日推荐
基于大语言模型的开源知识库问答系统 MaxKB GitHub Star 数量突破 5,000 个！
美国拟限制 AI 大模型出口中国和俄罗斯
苹果将与 OpenAI 达成协议，将 ChatGPT 应用于 iPhone
openKylin 社区生态委员会第六次会议圆满召开
阿里云正式发布通义千问 2.5
Python 3.13 发布首个 Beta：实验性自由线程模式和 JIT、改进交互式解释器
Stack Overflow 拿我的代码去训练 AI 大模型，还封了我的账号
Pop!_OS 的 COSMIC 桌面完成 App Store 上架工作
《2024 年一季度互联网投融资运行情况》研究报告
报告：Django 仍然是 74% 开发者的首选
15 年前上了“FFmpeg 耻辱柱”，今天他还得谢谢咱——腾讯QQPlayer一雪前耻？
TIOBE 5 月榜单：Fortran “复活”进入 Top 10
周排行
记一下去大梅沙的准备（2018-05-26）
Spring 注解 事务
基于HTTP协议的客户端缓存
阿里云rds 备份和还原
[PHP] 几个拖慢 PHP 程序/API 运行速度的点
python 代码风格------------PEP8规则
js控制json生成菜单——自制菜单（一）
将字符串: 'k:1|k1:2|k2:3|k3:4 ' ,处理成 python 字典: {'k':1, 'k1':2, ...}
微信小程序转支付宝小程序
Qt551.窗口滚动条
每日归档
更多
2024-05-13(18)
2024-05-12(0)
2024-05-11(38)
2024-05-10(38)
2024-05-09(35)
2024-05-08(42)
2024-05-07(14)
2024-05-06(40)
2024-05-05(0)
2024-05-04(7)

代码天地 © 2018 codetd.com

境外服务器   最新电视剧2022