eric_14 escribió:pero eso que comentáis no es para la gpu virtual, no? es para el passthough de siempre
Sí, lo he conseguido hacer funcionar
Lo bueno es que una sola gpu y un solo monitor
Lo primero en kernel activar:
[*] Virtualization --->
<M> KVM for AMD processors support
[*] Virtualization --->
<*> Host kernel accelerator for virtio net
Device Drivers --->
[*] Network device support --->
[*] Network core driver support
<*> Universal TUN/TAP device driver support
[*] Networking support --->
Networking options --->
<*> The IPv6 protocol
<*> 802.1d Ethernet Bridging
Luego qemu lo he compilado así:
ebuild R ] app-emulation/qemu-6.0.0::gentoo USE="aio alsa bzip2 caps curl doc fdt filecaps gtk io-uring jpeg multipath ncurses nfs nls opengl oss pin-upstream-blobs plugins png pulseaudio sdl seccomp slirp spice udev usb usbredir vhost-net vhost-user-fs virgl virtfs vnc xattr zstd -accessibility -capstone -debug -glusterfs -gnutls -infiniband -iscsi -jack -jemalloc -lzo -numa -python -rbd -sasl -sdl-image (-selinux) -smartcard -snappy -ssh -static -static-user -systemtap -test -vde -vte -xen -xfs" PYTHON_TARGETS="python3_8 -python3_7 -python3_9" QEMU_SOFTMMU_TARGETS="arm sparc x86_64 -aarch64 -alpha -avr -cris -hppa -i386 -lm32 -m68k -microblaze -microblazeel -mips -mips64 -mips64el -mipsel -moxie -nios2 -or1k -ppc -ppc64 -riscv32 -riscv64 -rx -s390x -sh4 -sh4eb -sparc64 -tricore -unicore32 -xtensa -xtensaeb" QEMU_USER_TARGETS="x86_64 -aarch64 -aarch64_be -alpha -arm -armeb -cris -hexagon -hppa -i386 -m68k -microblaze -microblazeel -mips -mips64 -mips64el -mipsel -mipsn32 -mipsn32el -nios2 -or1k -ppc -ppc64 -ppc64abi32 -ppc64le -riscv32 -riscv64 -s390x -sh4 -sh4eb -sparc -sparc32plus -sparc64 -xtensa -xtensaeb" 0 KiB
Creo la máquina virtual usando OVMF
Creo dos unidades ópticas una con la ISO de Windows y otra con los drivers para usar el disco SSD en
virtIO <hostdev mode="subsystem" type="pci" managed="yes">
<source>
...
</source>
<rom file="/home/juanpe/Documentos/BIOS.rom"/>
...
</hostdev>
Luego en arranque pones las dos unidades ópticas, en CPU como host-passthrough, socket 1, centro 6 e hilos 2.
Inicio la máquina virtual, no me detecta el SSD Windows, busco controlador que está en la ISO virtIO, amd64 bits → windos 10, ok.
Una vez instalado, apagas y borras de la máquina virtual, la tableta, etc... como en la imagen
Añades usb tu ratón, usb teclado, pci de tú grafica video y audio, mi DAC de audio.
Te bajas tu BIOS de tú grafica y la
parcheas Ahora editamos la máquina virtual para añadir la BIOS
<hostdev mode="subsystem" type="pci" managed="yes">
<source>
...
</source>
<rom file="/home/juanpe/Documentos/BIOS.rom"/>
...
</hostdev>
También añado esto:
<features>
...
<hyperv>
...
<vendor_id state="on" value="thissucks"/> <!-- value can be whatever like "hello", maximum of 12 characters-->
</hyperv>
<hidden state="on"/>
...
<ioapic driver="kvm"/>
</features>
Luego creo los directorios
sudo mkdir /etc/libvirt/hooks/qemu.d
Clono
sudo wget 'https://raw.githubusercontent.com/PassthroughPOST/VFIO-Tools/master/libvirt_hooks/qemu' -O /etc/libvirt/hooks/qemu
Doy permisos ejecución
sudo chmod +x /etc/libvirt/hooks/qemu
Reinicio libvirt
sudo service libvirtd restart
Creo directorios donde pego dos script
sudo mkdir -pv /etc/libvirt/hooks/qemu.d/win10/prepare/begin
sudo mkdir -pv /etc/libvirt/hooks/qemu.d/win10/release/end
el script de prepare/begin/start.sh
#!/bin/bash
# Original script provided at https://github.com/joeknock90/Single-GPU-Passthrough
# You should only use these scripts as templates and write your own as stated in the repository's readme.
# Helpful to read output when debugging
set -x
# I don't need to call that
#systemctl isolate multi-user.target
# Stop the display manager service
systemctl stop sddm.service
# Stop the pulse audio service
systemctl stop pulseaudio.service
# Terminate KDE
kquitapp5 plasmashell || killall plasmashell
killall kwin_x11
# Unbind VTconsoles
echo 0 > /sys/class/vtconsole/vtcon0/bind
echo 0 > /sys/class/vtconsole/vtcon1/bind
# Unbind EFI-Framebuffer
echo efi-framebuffer.0 > /sys/bus/platform/drivers/efi-framebuffer/unbind
# If you have the beep package installed and your motherboard has a speaker then this can provide you
# with some sort of feedback on what's going on.
#enable pcspkr
#modprobe pcspkr
#beep -l 100 -f 1000
# Avoid a Race condition by waiting 5 seconds. This can be calibrated to be shorter or longer if required for your system
sleep 8
# Unload all Nvidia drivers
modprobe -r nvidia_drm
modprobe -r nvidia_uvm
modprove -r nvidia_modeset
modprobe -r nvidia
# snd_hda_intel is used for the HDMI audio
# snd_hda_intel is blacklisted completely on my system, so I don't really need a call to remove it since it's not loaded.
modprobe -r snd_hda_intel
#beep -l 500 -f 1000
nvProbCount=$(lsmod | grep nvidia | wc -l)
if [ $nvProbCount -gt 0 ]
then
# what the fuck!
# let us know via the pc speaker that something got fucked
# beep -f 300 -l 1000
sleep 1000 # wait a second
virsh destroy {your_vm_name} # error, forcefully shut down the vm
exit 1
fi
## Detach the GPU
virsh nodedev-detach pci_0000_09_00_0 # Change pci_0000_25_00_0 to the appropriate PCI address that your GPU resides on.
virsh nodedev-detach pci_0000_09_00_1 # Change pci_0000_25_00_1 to the appropriate PCI address that your GPU audio controller resides on.
# Load VFIO Kernel Module
modprobe vfio-pci
#beep -l 100 -f 1000
#modprobe -r pcspkr
el script release/end/revert.sh
#!/bin/bash
# Original script provided at https://github.com/joeknock90/Single-GPU-Passthrough
# You should only use these scripts as templates and write your own as stated in the repository's readme.
# Helpful to read output when debugging
set -x
# Unload VFIO-PCI Kernel Driver
modprobe -r vfio-pci
modprobe -r vfio_iommu_type1
modprobe -r vfio
# Re-Bind GPU to Nvidia Driver
virsh nodedev-reattach pci_0000_09_00_1 > /dev/null 2>&1
virsh nodedev-reattach pci_0000_09_00_0 > /dev/null 2>&1
# Rebind VT consoles
echo 1 > /sys/class/vtconsole/vtcon0/bind
# Some machines might have more than 1 virtual console. Add a line for each corresponding VTConsole
echo 1 > /sys/class/vtconsole/vtcon1/bind
# Re-Bind EFI-Framebuffer
nvidia-xconfig --query-gpu-info > /dev/null 2>&1
echo "efi-framebuffer.0" > /sys/bus/platform/drivers/efi-framebuffer/bind
# It's blacklisted on my system, no need for me to enable it back.
#modprobe snd_hda_intel
modprobe nvidia_uvm
modprobe nvidia_drm
modprobe nvidia_modeset
modprobe nvidia
# Restart Display Manager
systemctl start sddm.service
kstart5 plasmashell
Hasta aquí no me iba.
Me faltaba un parche en el kernel y unas opciones.En el kernel.
Device Drivers --->
[*] IOMMU Hardware Support --->
Generic IOMMU Pagetable Support ----
[*] AMD IOMMU support
<*> AMD IOMMU Version 2 driver
[*] Support for Intel IOMMU using DMA Remapping Devices
[*] Support for Shared Virtual Memory with Intel IOMMU
[*] Enable Intel DMA Remapping Devices by default
[*] Support for Interrupt Remapping
En el grub editamos
GRUB_CMDLINE_LINUX_DEFAULT="quiet splash iommu=1 amd_iommu=on rd.driver.pre=vfio-pc pcie_acs_override=downstream,multifunction"
Aplicamos parche de gentoo
git clone https://github.com/feniksa/gentoo_ACS_override_patch.git /etc/portage/patches
Y remergemos gentoo-sources
Reinicio y con dos click windows ya es un programa mas de linux
![[beer]](/images/smilies/nuevos2/brindando.gif "brindis")
![[carcajad]](/images/smilies/nuevos/risa_ani2.gif "carcajada")
![[Alaa!]](/images/smilies/nuevos/sorprendido_ani2.gif "Ala!")