New Path to Arm Virtual Machines on IBM s390 Systems
A recent patch set by Steffen Eiden and collaborators introduces hardware-assisted emulation of Arm CPUs on IBM s390 mainframes. This technology aims to run Arm virtual machines (VMs) on s390 hosts at near-native speeds, opening up new possibilities for cross-architecture virtualization. Below, we explore key aspects of this development in a Q&A format.
What is the core idea behind hardware-assisted Arm VMs on s390?
The core idea is to leverage the hardware capabilities of s390 processors to emulation of Arm CPUs. Traditionally, running an operating system designed for one architecture on another requires software emulation, which is slow. By using hardware-assisted virtualization features on s390, the patch set enables near-native performance for Arm VMs. This means that applications compiled for Arm can run efficiently on s390 machines without modification, benefiting environments that need to test or deploy Arm software on existing mainframe infrastructure.
What changes were made in version two of the patch set?
The second version of the patch set addresses several minor issues identified in the initial submission. According to the post, these fixes include corrections to memory mapping, interrupt handling, and instruction decoding. However, the overall structure and approach remain largely unchanged. The patches aim to integrate directly with the Linux kernel's Virtualization for s390 (KVM) subsystem. While the improvements are incremental, they indicate progress toward a stable implementation that can be merged upstream.
How did the Arm maintainers react to these patches?
The Arm maintainers have received the patches positively. They acknowledged the technical soundness of the approach and expressed interest in having it upstream. However, they raised concerns about the long-term maintainability of cross-architecture emulation within the Arm Linux kernel tree. Specifically, they want a clear division of responsibilities and ownership to ensure that changes to Arm emulation logic don't break existing Arm support. The maintainers suggested that a collaborative model be established before the patches are accepted.
What remains to be resolved before the patches can be accepted?
The main unresolved issue is the governance model for maintaining the Arm emulation code within the Linux kernel. To prevent future problems, the Arm maintainers want a structured collaboration between the s390 and Arm communities. This includes defining which patches need Arm-side review, how to handle architectural updates, and ensuring that the emulation stays in sync with real Arm hardware. Once these organizational details are settled, the patches can be considered for merging. The discussion highlights the challenges of cross-architecture support in an open-source project.
What potential benefit does this technology offer to users?
If successfully adopted, hardware-assisted Arm VMs on s390 would allow data centers to consolidate workloads from different architectures onto a single mainframe platform. For example, a company could run both x86 and Arm cloud instances on s390 hardware, simplifying infrastructure management. Additionally, developers could use s390 machines to test Arm software without needing physical Arm devices. The near-native speeds would make such VMs practical for production use, not just testing. This could reduce hardware costs and power consumption by maximizing the utilization of existing mainframe resources.
What challenges does cross-architecture virtualization present?
Cross-architecture virtualization introduces several technical and organizational hurdles. Technically, emulating a different CPU architecture requires careful handling of instruction sets, memory models, and device interfaces. Hardware assistance can speed up performance but adds complexity to the hypervisor. Organizationally, maintaining emulation code across multiple architecture maintainer teams can lead to conflicts or neglect. The Arm maintainers' request for a clear collaboration plan is a direct response to such challenges. Other concerns include ensuring security isolation, handling architectural extensions (e.g., new Arm features), and performance parity with native virtualization.