WIP: BareSlice: Extending Arm CCA to Support Bare-Metal Confidential Virtual Machines

ABSTRACT

The Arm architecture is becoming popular in cloud computing due to its efficiency, low power consumption, and rich ecosystem, making it highly suitable for data centers. For example, Nvidia introduced the Grace CPU, a 144-core Arm architecture processor, to serve as the foundation for next-generation data centers. However, as cloud infrastructures grow, the use of complex hypervisors to manage tenant environments introduces security concerns. In 2021, Arm released the Confidential Computing Architecture (CCA), which provides secure execution environments for confidential VMs and removes the hypervisor from the Trusted Computing Base. Despite CCA’s improvements, confidential VMs still face security and performance challenges due to the shared nature of physical resources. While bare-metal cloud services offer a promising solution by allocating directly-accessible physical resources to each tenant, the substantial costs of dedicating hardware to each tenant has limited its widespread adoption.

In this paper, we present BARESLICE, a novel bare-metal cloud system that supports multiple untrusted tenants on shared bare-metal hardware. BARESLICE assigns dedicated physical resources to each tenant, which we call a slice, by leveraging the hardware features of CCA to ensure robust isolation and create a secure computing environment free from virtualization overhead. We analyze the security of BARESLICE to show its effectiveness against common attacks. Additionally, our performance assessment illustrates that BARESLICE achieves bare-metal performance levels across several real-world applications like Apache and Memcached. BARESLICE achieves a combination of enhanced security and high efficiency, providing a compelling solution for modern confidential cloud computing challenges.

Yuxin Hu headshot

 

Yuxin Hu is a Ph.D. student in Computer Science at Vanderbilt University. His research focuses on system security, particularly in Trusted Execution Environments (TEE), with an emphasis on ARM TrustZone and ARM Confidential Compute Architecture (CCA).

License: CC-3.0
Submitted by Regan Williams on