HVM Dom0: Any unmodified OS as Dom0

### Overview In this presentation, Will Auld from Intel introduces ongoing work—primarily developed by team member Chantal and an Intel team in Shanghai—to enable Hardware Virtual Machine (HVM) Dom0 in the Xen hypervisor. Historically, Xen Dom0 has functioned as a paravirtualized (PV) domain. This architecture created heavy maintenance overhead and friction with upstream operating system communities (particularly Linux kernel maintainers), while also preventing the use of unmodified operating systems lacking source access (such as Microsoft Windows or macOS) as Dom0. The project proposes running Dom0 within a full hardware-assisted virtual machine container using Intel VT technologies.

### Key Topics - **History and Paravirtualization Pushback:** Originally, Xen relied on PV because x86 CPUs lacked hardware virtualization primitives. Even after early hardware extensions (VT-x) arrived around 2006, performance bottlenecks like shadow page tables kept Dom0 in PV mode. Upstream Linux maintainers frequently pushed back on Xen-specific patches, viewing them as an unnecessary maintenance burden. - **HVM vs. PVH Options:** Unlike the PVH model—which runs Dom0 inside a hardware container while keeping the paravirtualized interfaces intact—the full HVM approach aims to leverage standard HVM mechanisms used for unprivileged guest domains (DomU). - **Enabling Unmodified OS as Dom0:** By using full HVM, operating systems without kernel modifications or source access can serve as Dom0, enabling management software, Qemu, and virtio to run in ring 3 alongside native and backend drivers. - **New Usage Models:** Potential benefits include running Windows or macOS as Dom0, enabling trusted execution environments, and offering a streamlined, client-friendly user installation experience akin to a Type-2 hypervisor while remaining a high-performance Type-1 hypervisor.

### Technical Details - **Hardware Acceleration:** The implementation reuses the existing Xen DomU HVM infrastructure, utilizing Extended Page Tables (EPT) to support super pages and eliminating the performance tax of 64-bit x86 system call bouncing through the hypervisor. - **I/O and Interrupt Virtualization:** Direct device assignment and I/O pass-through rely on Intel VT-d to minimize VM exits. For interrupt handling, Dom0 controls the physical I/O APIC and is provided with a virtualized local APIC, while the hypervisor retains control of the physical local APIC. - **Boot Flow and OS Support:** Booting operating systems like Windows requires an EFI boot sequence, whereas Linux follows the established HVM boot flow. - **Current Status and Gaps:** Linux Dom0 support is working, but supporting Windows remains incomplete. Porting Qemu (xen-qemu), userland management tools, and libraries to Windows is still needed, alongside backend driver integration using virtio as a communication channel. During the Q&A, the speaker acknowledged that ACPI power management handling remains unresolved for HVM Dom0, performance benchmarks have not yet been completed, and patches have not yet been posted to the xen-devel mailing list.

### Takeaways HVM Dom0 allows Xen deployments to operate unmodified operating systems as the control domain, resolving long-standing upstream maintenance issues and hardware performance limitations. Intel is seeking community collaboration—particularly regarding userland tools and Qemu porting for Windows—and plans to initiate discussions and share materials on the xen-devel mailing list.