Finally It Comes: A KMS Driver For QEMU-KVM

[phoronix]

Phoronix: Finally It Comes: A KMS Driver For QEMU-KVM

While Linux KVM virtualization works well for many, one of the areas where the Kernel-based Virtual Machine and its QEMU integration have lagged behind other virtualization solutions like VirtualBox and VMware is in terms of its 2D/3D support within guests. The KVM-QEMU situation is slightly more positive today though with the introduction of a basic KMS (kernel mode-setting) driver for KVM-QEMU riding in the Linux kernel...

http://www.phoronix.com/vr.php?view=OTM0Nw

[thefirstm]

Does this mean that there is now KMS support for those old Cirrus framebuffers? I once had a PC with one of those.

[schmidtbag]

i don't know very much about the hardware level of virtualization, but why couldn't intel, amd, and nvidia create something like vt-x or amd-v for GPUs? those 2 instruction sets have made incomparable performance differences. if GPUs were to get a similar feature, then virtualization can become a practicality and a reality for mac and linux users looking to run their games, itunes, photoshop, video editors, etc. sure there would be a performance hit from the host OS consuming the resources first, but i would rather lose a little performance than have to dual boot.

does anyone know why this hasn't been done or why it wouldn't be possible?

[curaga]

So Michael, any benchmarks coming?

[liam]

i don't know very much about the hardware level of virtualization, but why couldn't intel, amd, and nvidia create something like vt-x or amd-v for GPUs? those 2 instruction sets have made incomparable performance differences. if GPUs were to get a similar feature, then virtualization can become a practicality and a reality for mac and linux users looking to run their games, itunes, photoshop, video editors, etc. sure there would be a performance hit from the host OS consuming the resources first, but i would rather lose a little performance than have to dual boot.

does anyone know why this hasn't been done or why it wouldn't be possible?

This already exists. It's known as pci passthrough( Intels is called vt-d,iirc; amd has the same thing but I don't recall its name. The problem is finding chips and software support(pretty sure kvm/qemu allows it though).

[tettamanti]

Does this mean that there is now KMS support for those old Cirrus framebuffers? I once had a PC with one of those.

No, the driver is limited to the virtual device emulated by QEMU.

[schmidtbag]

This already exists. It's known as pci passthrough( Intels is called vt-d,iirc; amd has the same thing but I don't recall its name. The problem is finding chips and software support(pretty sure kvm/qemu allows it though).

i never noticed virtualbox having this. virtualbox uses fake 3d drivers that cooperate with the host's GPU, but it doesn't appear to be a direct connection.

[Ansla]

This already exists. It's known as pci passthrough( Intels is called vt-d,iirc; amd has the same thing but I don't recall its name. The problem is finding chips and software support(pretty sure kvm/qemu allows it though).

pci passthrough is not very useful for graphics as the virtual machine must have exclusive control over the device (the host can not use it)

[liam]

pci passthrough is not very useful for graphics as the virtual machine must have exclusive control over the device (the host can not use it)

Graphics is one of the explicit uses for passthrough, but I'd have to imagine that the iommu takes care of this as long as thereis sufficient dram.