OpenCCA : An Open Framework to Enable Arm CCA Research

Andrin Bertschi, Shweta Shinde · 2025

Read OpenCCA : An Open Framework to Enable Arm CCA Research by Andrin Bertschi, Shweta Shinde (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-05129.txt.

Introduction

Andrin Bertschi and Shweta Shinde built an open platform for security research in a 2025 paper. Confidential computing lets programs run in sealed enclaves that even the machine's owner cannot peek into. The Arm chip design for this has no shipping hardware yet, which forces researchers to fake it in one-off ways.

Their platform, OpenCCA, runs enclave-style code on a cheap, common Arm board that predates the feature. They adapted the whole software stack, from boot loader to kernel, so the sealed operations behave correctly while running at near-real speed. Timing tests of the enclave life cycle show the platform works for performance studies. Case studies modeled on earlier papers show it can host real research questions. The board costs about two hundred fifty dollars, which lowers the entry bar for the field.

Catalog record

CREATORAndrin Bertschi, Shweta Shinde
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE29 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-05129

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

OpenCCA : An Open Framework to Enable Arm CCA Research Andrin Bertschi, Shweta Shinde Abstract Confidential computing has gained traction across major architectures with Intel TDX, AMD SEV-SNP, and Arm CCA. Unlike TDX and SEV-SNP, a key challenge in researching Arm CCA is the absence of hardware support, forcing researchers to develop ad-hoc prototypes on CCA emulators and non-CCA Arm boards. This approach leads to high barriers to entry or duplicated efforts leading to unsound and inconsistent comparisons. To address this, we present OpenCCA , an open research platform that enables the execution of CCA-bound code on commodity Armv8.2 hardware. By systematically adapting the software stack (including bootloader, firmware, hypervisor, and kernel), OpenCCA emulates CCA operations for performance evaluation while preserving functional correctness. We demonstrate its effectiveness with typical life-cycle measurements and case-studies inspired by prior CCA-based papers on an easily available Arm v8.2 Rockchip board that costs $ 250 \$250 . I Introduction † † footnotetext: arXiv: This is the Systex2025 version of the paper. Intel and AMD, leaders in x86 platforms, enable confidential computing with TDX and SEV-SNP respectively [ 1 , 2 ] . Arm, in 2021, announced Realm Management Extension (RME) which is an optional feature on Armv9A to enable confidential computing architecture (CCA) [ 3 ] . Arm has rolled out support for building CCA-enabled platforms, including emulator and software changes [ 4 , 5 , 6 ] . As witnessed by several recent works in top-tier venues, Arm CCA has already received traction by providing a rich and fertile ecosystem for innovative designs to improve confidential computing [ 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 ] . One fundamental hurdle in using Arm CCA is the lack of hardware support, since no public Arm CPU supports CCA yet. Arm’s system emulator, the Fixed Virtual Platform (FVP) provides functional correctness and instruction counts. But, it is not cycle accurate which hinders any performance measurements. To this end, researchers have resorted to building best-effort performance prototypes where they transplant their CCA-bound implementation to an Arm board that does not have CCA support (e.g., Armv8) and replace the CCA instructions and functionality with dummy operations. Fig. 1: OpenCCA tooling. The RK3588 connects over ethernet to a flash server (Raspberry Pi). It controls a MOSFET and power circuit to flash new firmware and exposes UART access. Our survey shows that of the 19 papers released in the last 4 years, 15 spend significant effort on such transplantation (see Table I ). This not only creates unnecessary work for the authors, it also detracts other researchers from experimenting on Arm CCA due to high barrier for entry. We observe that researchers choose different Arm boards for their performance prototypes. These efforts are focused on the scope of the paper and do not lend to full-fledged prototypes that are reusable by others. Thus researchers are repeating the same efforts across boards, which is not the best use of their time. The fragmentation in choice of boards further makes it challenging to compare the performance across different papers, especially since some boards are 10 years old, cost \$10,000 , or are no longer available (e.g., Juno R2). Worst of all, 4/19 papers omit hardware-based performance evaluation entirely due to lack of CCA support. Our motivation to build OpenCCA stems from: (a) our research experience prototyping early CCA works [ 12 , 27 ] ; (b) our roadblocks in subsequent performance prototypes [ 20 , 22 ] ; (c) personal communications with researchers in the community who want to benchmark performance for CCA-based defenses; and (d) large time gaps between Arm announcements and hardware rollouts (e.g., SEL2 announced in 2017, remains unavailable on commodity boards). Taking inspiration from the transplantation to performance prototype approach from prior efforts on CCA [ 7 , 9 , 10 , 8 , 11 , 12 , 14 , 13 , 17 , 15 , 16 , 24 , 25 , 23 ] , we build OpenCCA ( fig. 1 ) that is capable of executing CCA-bound code on a non-CCA Arm board. Specifically, we aim to enable lift-and-shift from Arm FVP to Arm board. We first surveyed the most suitable boards for our goal (see Table II ) and chose the RK3588 Radxa Rock 5B due to its easy availability, support system, and affordability. We then systematically analyzed the entire stack from bootloader, firmware, hypervisor, drivers, host and guest kernel to identify what aspects need to be adapted to emulate CCA operations—for performance and compatibility. Finally, we added CCA awareness to all these components while carefully selecting non-CCA operations that can best estimate CCA overheads. OpenCCA presents a standard development and measurement framework to evaluate CCA-based solutions, akin to efforts on Intel SGX [ 28 ] and RISC-V [ 29 ] . Similar to OpenSGX [ 28 ] , in its current form OpenCCA does not aim to enforce CCA equivalent security on non-CCA boards. Lessons from ongoing efforts on virtCCA [ 16 ] can be coupled with OpenCCA to address this limitation. Our choice of a commodity board allows researchers to take their approach implemented on Arm FVP and lift-and-shift it to OpenCCA for performance estimates and compatibility. We demonstrate this by showcasing out-of-box use of OpenCCA for reporting typical life-cycle metrics. Next, we run standard benchmarks on FVP and OpenCCA to show that we preserve functionality. Lastly, we then build two representative case-studies of CCA-based designs from prior works, under five hours each. OpenCCA is available at [ 30 ] . TABLE I: Survey of Work on CCA as of February 2025. Column 2-3 show if the work is implemented on Simulation Software and Arm Board respectively, with board architecture version in Column 4. Column 5-6 indicate if it is open-source ( ✓ ), closed ( ✗ ), or not applicable (N/A). Related Work Sim. Board Arch. FVP Board Cage [ 8 ] FVP Juno R2 8.0 ✓ ✓ Shelter [ 7 ] FVP Juno R2 8.0 ✓ ✗ Scrutinizer [ 10 ] FVP Juno R2 8.0 ✓ ✗ ACAI [ 12 ] FVP Zynq UltraScale+ 8.0 ✓ ✗ TZ & CCA [ 11 ] FVP Juno R2 8.0 ✗ ✗ HitchHiker [ 9 ] FVP Juno R2 8.0 ✗ ✗ FortifyPatch [ 23 ] FVP Raspberry Pi 3B 8.0 ✗ ✗ RContainer [ 13 ] FVP RK3399 Firefly 8.0 ✗ ✗ CubeVisor [ 24 ] FVP RK3399 Rock 4B 8.0 ✗ ✗ TwinVisor [ 25 ] FVP HiSilicon Kirin 990 8.2 ✓ ✗ Portal [ 14 ] FVP OrangePi 5 Plus 8.2 ✗ ✗ Des. & Ver. [ 15 ] FVP Neoverse N1 8.2 ✗ ✗ virtCCA [ 16 ] – Undisclosed 8.4 N/A ✗ Sharing [ 17 ] – AmpereOne 8.6 N/A ✗ CPC [ 19 ] FVP SEV SNP (x86) – ✗ ✗ GuaranTEE [ 18 ] FVP – – ✓ N/A Devlore [ 20 ] FVP – – ✗ N/A BarriCCAde [ 21 ] QEMU – – ✗ N/A Aster [ 22 ] QEMU – – ✗ N/A

Read the complete text