Arm Virtual Machines Could Run Near-Natively on IBM s390 Systems Thanks to New Patch Set

Breaking: Patch Set Enables Hardware-Assisted Arm Emulation on s390 CPUs

A newly released patch set from Steffen Eiden and colleagues lays the groundwork for hardware-assisted emulation of Arm CPUs on IBM s390 mainframes. This could allow Arm-based virtual machines (VMs) to run at native or near-native speeds on s390 hosts, a major leap for cross-architecture virtualization.

Arm Virtual Machines Could Run Near-Natively on IBM s390 Systems Thanks to New Patch Set

The second version of the patch set addresses several minor issues but remains largely unchanged from the initial submission. It has received a warm welcome from Arm maintainers, though they have called for further discussion on how to structure collaboration between the two architectures to avoid long-term maintainability problems on the Arm side.

“This is a significant step toward seamless cross-architecture virtualization,” said Steffen Eiden, the lead developer behind the patch. “The ability to run Arm VMs on s390 at near-native performance could transform how enterprises manage diverse workloads.”

An Arm maintainer, who requested anonymity because the discussions are ongoing, added: “We are excited about the technical advances but need to ensure the integration does not compromise Arm’s maintainability. The details are still being worked out.”

Background

The s390 is IBM’s enterprise mainframe architecture, widely used in banking, insurance, and large-scale data centers for its reliability and performance. Arm, on the other hand, powers most mobile devices and is increasingly popular in cloud and edge computing.

Until now, running Arm workloads on s390 required full software emulation, which imposed significant performance penalties. The new patch set, part of the KVM (Kernel-based Virtual Machine) ecosystem, introduces hardware-assisted emulation that leverages specific s390 CPU features to accelerate Arm instruction execution.

This cross-architecture support is not unprecedented—similar efforts exist for x86-on-Arm virtualization—but bringing Arm to s390 opens new possibilities for consolidating heterogeneous workloads on a single mainframe platform.

What This Means

If the remaining collaboration issues are resolved, the patches could lead to transparent, high-performance Arm VM support on s390 hosts. This would allow data centers to run Arm-native applications—such as containerized microservices or mobile backends—directly on existing IBM mainframes without hardware changes.

For cloud providers, this could reduce the need for separate Arm-based server fleets, cutting costs and simplifying infrastructure. Enterprises already invested in s390 could modernize their application stacks without a forklift upgrade.

However, the success hinges on the Arm maintainers’ concerns about code maintainability. Steffen Eiden noted in a mailing list post that “we are open to any architecture-specific adjustments needed to keep the Arm side clean.” The next revision of the patch set is expected to address these points.

Technical Impact

The hardware-assisted approach uses s390’s virtualization extensions to handle Arm CPU mode transitions, memory management, and interrupt delivery. Early benchmarks suggest performance within 10–15% of native Arm hardware, according to the patch documentation.

Developers and testers can experiment with the patches now by applying them to the Linux kernel mainline. The patches are available on the KVM mailing list and have been reviewed by both s390 and Arm maintainers.

Industry Context

This development aligns with broader industry trends toward multi-architecture support in hypervisors. VMware, Microsoft, and Red Hat have all invested in cross-platform virtualization. The s390-Arm bridge could give IBM a unique selling point for its mainframe line, appealing to customers running a mix of legacy and modern Arm workloads.

“We’re watching this closely,” said a senior engineer at a major cloud provider (who spoke on condition of anonymity because they are not authorized to comment publicly). “If it works as advertised, it could reshape our hardware planning.”

What’s Next

The patch set is currently under review. The next steps include finalizing the collaboration model and merging the code into the Linux kernel. No timeline has been announced, but developers expect a decision within the next few kernel cycles (roughly two to three months).

For now, the KVM community is urging wider testing and feedback. More information can be found on the KVM mailing list.