Introduction
Confidential computing has emerged as a critical requirement for cloud security, providing isolated execution environments (Confidential Virtual Machines, or CVMs) that protect sensitive data from even the host infrastructure. While Arm's upcoming Confidential Compute Architecture (CCA) for Armv9-A addresses this, a significant deployment gap exists: millions of legacy Armv8-A servers currently lack these hardware-native features. virtCCA bridges this gap by leveraging existing TrustZone technology to implement a software/firmware architecture compatible with Arm CCA specifications.
The virtCCA Architecture
The core of virtCCA is the TrustZone Management Monitor (TMM), a minimal trusted software layer that emulates the functionality of the CCA’s Realm Management Monitor (RMM). To accommodate varying hardware capabilities across Armv8-A generations, virtCCA supports two distinct modes:
- virtCCA-SEL2: Designed for Armv8.4+ hardware, this mode utilizes the Secure EL2 (S-EL2) extension to provide hardware-assisted virtualization for secure worlds.
- virtCCA-EL3: For older hardware lacking S-EL2, the TMM resides at the highest privilege level (EL3). In this configuration, the TMM emulates virtualization features in software, allowing CVMs to run in a "bare-metal" fashion at S-EL1.
Interface Compatibility and Design
A primary design goal of virtCCA is "drop-in" compatibility with the Arm CCA standard. This ensures that the existing software ecosystem—including KVM hypervisors and guest operating systems—can transition to native Armv9-A hardware in the future with minimal friction. The system implements several critical interfaces, achieving high compatibility with management, service, and power control interfaces, though it introduces new mechanisms for memory mapping to handle TrustZone's specific constraints.
Beyond API compatibility, virtCCA optimizes memory management to overcome TrustZone's hardware limitations. While standard TrustZone often supports only a few memory regions, the TMM manages a large contiguous block of secure memory using custom buddy and slab allocators. Furthermore, the architecture is NUMA-aware, ensuring memory is allocated close to the physical CPU cores to minimize latency.
Performance Evaluation
The researchers evaluated virtCCA on Arm v8.4 servers and Kunpeng 920 platforms using a mix of micro-benchmarks and real-world server workloads (Redis, MySQL, Nginx).
- Computational Overhead: The system demonstrates remarkably low overhead for CPU-bound tasks. On the nbench suite, virtCCA-SEL2 introduced only a 1.8% overhead, while virtCCA-EL3 achieved an even lower 0.8% overhead due to its bare-metal execution style.
- I/O and Latency: For I/O-intensive workloads, the overhead for virtCCA-SEL2 remained under 29.7%. In Redis testing, latency increases were kept within 35% of the baseline.
- NUMA Impact: Implementing NUMA-aware memory allocation proved critical, resulting in a 66.7% improvement in I/O performance compared to non-aware configurations.
Conclusion
virtCCA demonstrates that advanced confidential computing features can be successfully backported to legacy hardware through clever firmware design. By maintaining a small Trusted Computing Base (TCB) of approximately 3.2K to 4K lines of code, virtCCA provides a secure, performant, and API-compatible bridge to the future of Arm confidential computing. While side-channel protections remain outside the current scope, virtCCA offers a viable path for immediate deployment of CVMs on existing Arm infrastructure.
Summary updated