Xen Summit 2026

Nearly three years after Broadcom's takeover of VMware, the community around Xen and XCP-ng looks different. This is the weather report for that community, not the product roadmap.
We will walk the climate: the wave of VMware refugees arriving on the forum and in production, where they get stuck and what pulls them through; who is actually contributing to an open type-1 hypervisor, and why the reviewer bottleneck matters more than any single feature; and the widening ecosystem of partners, integrators, and distributions now betting on Xen.
We will look at why Xen still exists when the hyperscalers moved on, what "digital sovereignty" really means once you get past the slogans, and where the project is heading, from the datacenter out to Arm, RISC-V, and the safety-certified edge.
This one is for the room that lives in this code every day: a step back from your own corner of the tree to the state of the whole community and ecosystem around it. Expect an honest forecast, clouds included.

Status update of Xen-related work for XCP-ng.
Will introduce some parts of Xen-related roadmap for XCP-ng, status update on on-going projects, ...

XenServer has been working on Nested Virtulaisation, along the lines of the plans laid out at XenSummit 2025, San Jose.

This is a progress report of how far we've got, the issues encountered, and what's next.

TODO: Roger, as a maintainer, can explain how to contribute to Xen.

SVM AVIC is feature missing in upstream Xen code base. This session covers the journey of enabling AVIC support on AMD x86 systems.

The Xen hypervisor needs to manage several resources for each guest, each of those requiring the hypervisor to allocate memory for this task. In order to limit the amount of memory needed, most resources are limited, e.g. via guest configuration.
There is an alternative approach: instead of limiting each resource individually, manage all resources via a common resource management, allowing more flexibility for the guests.
This presentation is a followup of a related design session at the Xen summit 2022.

Xen Argo provides a robust mechanism for inter-domain communication, offering strong isolation compared to traditional shared memory approaches. However, its message-based API creates a portability barrier for standard POSIX applications. In this session, I will present the realization of a new Linux kernel module that solves this challenge by implementing a complete VSOCK transport layer over Xen Argo. By exposing Argo through the standard AF_VSOCK interface, we enable existing socket-based applications to run over Argo channels, supporting both datagrams and reliable data streams (SOCK_STREAM) without code modification.
Building upon initial designs, we will dive deep into the kernel-level challenges to achieve a fully functional TCP-like stream over Argo's asynchronous rings. Key technical topics include:

Realizing the Connection-Oriented Handshake: Mapping the POSIX TCP state machine (SYN/SYN-ACK) onto Argo, including dynamic ring allocation with asymmetric hypervisor permissions to enable bidirectional traffic.

Socket multiplexing: allowing concurrent logical connections over a single ring, bypassing the strict 1:1 port-to-ring limitation.

The Xen livepatch framework provides a way to redirect calls to functions at runtime. It is useful when a running system needs to be updated without being stopped, for example to apply security fixes.

This session explores the possibility of extending this mechanism into a more tracing/debugging-friendly feature. Such an extension could allow more flexible use of existing Xen tracing tools such as xentrace. We will showcase this idea by sharing our recent experience using xentrace, and discuss how a callback-based extension could make similar investigations easier in general.

We would also like to discuss possible future extensions inspired by Linux tracing mechanisms such as kprobes/ftrace/eBPF which could make the feature more useful in practice.

Xen makes use of various techniques to conditionally remove code from the final binary. From Kconfig, to preprocessor tricks and DCE. This session explains a number of exotic key techniques used at AMD to drastically reduce code footprint with minimal changes to the source code. It highlights the set of constraints and requirements behind the new techniques and specific examples where they apply, with metrics showing the gains.

Future Arm CPUs are evolving toward systems where Virtualization Host Extensions (VHEs) are always enabled. This talk shows how Xen can adapt to the next generation of Arm hardware while continuing to support today’s programmable-VHE and non-VHE systems.
We will walk through the engineering work required to make Xen bootable, and upstreamable on VHE-always platforms. The challenge goes far beyond early boot. We’ll cover VHE changes assumptions around EL2 bring-up, system-register access, MMU setup, exception and trap routing, timers, interrupts, and host/guest world switching.
Attendees will learn how Xen detects and selects VHE policy during boot and secondary CPU bring-up; how vhe=on/off/auto supports different platform capabilities; and how EL1/EL2 system-register abstractions help reduce fragile VHE-specific conditionals. We will also cover the areas most likely to expose hidden assumptions, including MMU state, timer delivery, GICv2/GICv3 virtualization, and VHE-sensitive save/restore paths.
Finally, we will discuss the path from an initial homogeneous Dom0-boot prototype with basic timer and interrupt support to product-quality SMP, testing, and platform validation.

The RISC-V port of the Xen hypervisor has been progressing over the past year. This session gives a practical status update on where the port stands today, covering the two main use cases: dom0less deployments for embedded and safety-critical workloads, and dom0 with the xl toolstack for more traditional virtualization scenarios.

We will walk through what works, what is still in progress, and what the remaining gaps are with a live demo to show the port running.