Wiki Home

Hardware virtualization

Definition

Hardware virtualization runs several operating systems on one machine, each in a virtual machine (VM) that believes it owns the hardware, under a hypervisor that arbitrates the real resources. Processors expose it through a dedicated, more privileged mode and a second stage of address translation: the guest OS translates virtual addresses to what it thinks are physical addresses, and the hypervisor’s stage-two tables translate those into the actual physical addresses, so each VM sees only the memory the hypervisor grants it. The classical conditions a processor must meet for this to be efficient and safe were stated by Popek and Goldberg (1974). On Arm the hypervisor runs at exception level EL2, beneath the EL3 firmware, and an IOMMU (called the SMMU on Arm) applies the same address-translation control to DMA-capable peripherals so a device cannot read memory outside the VM it is assigned to (Dall and Nieh, 2014).

A virtualization-based trusted execution environment reuses this machinery for isolation rather than multiplexing. A small, trusted hypervisor makes one VM an enclave whose memory and devices are walled off from a larger, untrusted hypervisor and the host OS that run alongside it. Because the isolation is enforced at EL2 through stage-two translation and the IOMMU, an enclave can be built without changing the most privileged firmware at EL3, and the enclave can run a stock OS and stock device drivers. This is the alternative substrate to a trusted execution environment built on a two-world split such as Arm TrustZone.

Papers that use this concept

• ASGARD: Protecting On-Device Deep Neural Networks with Virtualization-Based Trusted Execution Environments — builds on-device model protection on a virtualization-based TEE, extending a VM enclave over an integrated accelerator through secure I/O passthrough so isolation is enforced at the hypervisor level without touching the EL3 secure monitor.
• PAL*M: Property Attestation for Large Generative Models — runs operations inside a confidential virtual machine (Intel TDX), a TEE that places an entire VM behind the isolation built on hardware virtualization.

See also

• Trusted execution environment
• Model partitioning across a TEE and an accelerator

References

• Dall, C. and Nieh, J. “KVM/ARM: The Design and Implementation of the Linux ARM Hypervisor.” International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS), 2014.
• Popek, G. J. and Goldberg, R. P. “Formal Requirements for Virtualizable Third Generation Architectures.” Communications of the ACM, vol. 17, no. 7, pp. 412-421, 1974.