XenGT: A software based Intel Graphics Virtualization Solution
summary
XenGT implements a mediated pass-through for Intel integrated GPUs, running native guest drivers while a kernel VDG manager partitions the global GTT with ballooning, validates PPGTT shadow page tables and mediates tail-register updates for command buffers. It schedules render engines with a 16 ms round-robin quantum, isolates per-VM graphics aperture regions, and achieves near-native single-VM performance and a two-VM combined throughput roughly equal to one VM. Code targets Haswell (Gen4) integrated graphics and is published on GitHub.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 0:54Agenda
- 1:29Graphics
- 2:50Emulation
- 3:21Driver Model
- 4:11Device Pass Through
- 4:31DD Architecture
- 9:02GPU Interface
- 9:57Memory Space
- 12:04PPD
- 12:45Command Buffer
- 14:45Render Engine
- 15:59Display Engine
- 17:37Performance
- 18:20Single VM Performance
related talks
33mGPU Para-Virtualization on Xen
36mCurrent State of Graphics Virtualization Upstream
22mAMD-vIOMMU: A Hardware-assisted Virtual IOMMU Technology
30mNested Virtualization (X86) Part II
46mState of Intel SGX in Linux
35mEverything Great about Upstream Graphics
30mNested Virtualization (X86) Part I
46mGraphic Testing Without Hardware: Discovering the Power of VKMS!