Virtualized GPU Tile Allocation via Mediated Pass-Through
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current GPU virtualization technologies face challenges in efficiently sharing graphics processing resources among multiple virtual machines while maintaining performance and secure isolation, particularly in scenarios where multiple virtual machines need to access performance-critical resources without hypervisor intervention.
Innovation Solution
The implementation of a virtualized graphics processing architecture that allows multiple virtual machines to share a physical GPU by using mediated pass-through and trap-and-emulation for privileged operations, with each VM running a native graphics driver and direct access to performance-critical resources, while employing a GPU scheduler to manage resource allocation and context switching to ensure efficient and secure usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If hardware support for translation is used, then translation speed is improved, but the GPU can be passed through to only one VM
Solution Approach 1:
The patent introduces a translation lookaside buffer (TLB) as an intermediary component between the GPU and multiple VMs. The TLB caches address translations and allows the GPU to be shared among multiple VMs while maintaining hardware-accelerated translation performance. This mediator enables multiple VMs to access the GPU simultaneously without requiring full hardware pass-through to each individual VM.
2Adaptability or versatility
If software approach with shadow structures is used, then multiple VMs can share physical GPU, but translation performance deteriorates
Solution Approach 1:
The patent segments the address translation function into two parts: a hardware TLB for fast caching of translations and software shadow structures for maintaining translation accuracy across multiple VMs. The TLB is divided into per-VM portions and a shared portion, allowing each VM to have dedicated fast translation capability while sharing the overall translation infrastructure. This segmentation maintains both multi-VM support and high translation performance.
3Productivity
If direct DMA access is provided to multiple VMs, then device performance is improved, but secure isolation among VMs deteriorates
Solution Approach 1:
The patent implements local quality by providing each VM with its own dedicated portion of the TLB that stores translations specific to that VM's address space. This per-VM TLB portion ensures secure isolation because each VM's direct DMA accesses are translated using only its own cached translations, preventing access to other VMs' memory spaces. Meanwhile, the shared TLB portion enables efficient resource sharing when translations are identical across VMs.
Data Source
AI summary
An apparatus and method for provisioning virtualized tile-based graphics processing circuitry. For example, one embodiment of an apparatus comprises: processing resources to process commands including graphics commands and generate results; resource partitioning circuitry to partition the processing resources into a plurality of tiles in accordance with a specified tile-based resource allocation policy; and graphics virtualization circuitry to perform tile-based allocation of the processing resources to a plurality of virtual machines in accordance with a specified virtualization policy, the virtual machines to be executed in a virtualized execution environment.