QEMU, a popular open-source machine emulator and virtualizer, has just introduced its brand-new major 10 release (after a fourth release‑candidate cycle), bringing a range of noteworthy enhancements, deprecations, and removals.
QEMU’s maintainers continue their long‑running effort to prune rarely‑used subsystems. In light of this, version 10 deprecates the ancient -old-param Arm boot flag and puts the PXA2xx CPU family plus its iwMMXt DSP extension on a short leash—expect full removal in a future cycle. For most users, nothing breaks, yet the payoff is a leaner codebase and less confusion about platform support.
Noteworthy CPU‑ and board‑level enhancements include:
Arm
- The default pointer‑authentication emulator now uses a QEMU‑specific algorithm that is orders of magnitude faster than the spec‑compliant QARMA5 routine. Should you need full conformance, it is still one
-cpu max,pauth-qarma5=onaway. - Secure EL2 timer emulation and new architectural bits—FEAT_AFP, RPRES, and XS—round out the update.
RISC‑V
- A slew of ISA extensions arrive, including Svukte, ssstateen, SHA, Smrnmi, and the delegation of performance counters.
- On the machine’s side, developers can try out Tenstorrent’s Ascalon core or the Xiangshan Nanhu RV64 CPU, while the new
riscv-iommu-sysdevice prepares the ground for advanced I/O isolation.
x86
- Tight loops built from
rep movs/stosand friends finally get a speed‑boost, shaving measurable seconds off boot times in string‑heavy firmware and OS code paths. - Fresh models for Intel Clearwater Forest and Sierra Forest‑v2 keep QEMU aligned with announced Xeon roadmaps.
HPPA (yes, the PA‑RISC enthusiasts get candy too)
- Up to 256 GB of guest RAM is now possible, translation is faster, and a Guardian Service Processor (Diva GSP) PCI device joins the party. Thanks to new Astro LLMIO wiring, graphics aficionados can disable the legacy Artist card and plug in an alternate GPU.
Regarding I/O and storage, high‑throughput guests sometimes swamped the single I/O thread that backed virtio-scsi. QEMU 10 introduces true multiqueue support, letting separate virtqueues fan out onto multiple threads—catching up to the virtio-blk path introduced in QEMU 9.0.
For users of qemu-nbd, an optional handshake timeout knob helps integration tests fail fast rather than hang forever, and the long‑standing daemon‑exit stall is gone.
Mac‑oriented workflows also receive improvements via the new apple-gfx-pci (x86‑64) and apple-gfx-mmio (Apple‑silicon) virtual GPUs that tap into macOS’s ParavirtualizedGraphics.framework. On the Linux side, Intel Gen 11/12 IGD passthrough is markedly more reliable, and VFIO migration gains multi‑FD support.
It’s worth noting also that the virtio-mem hot‑plug balloon now works on s390x mainframe guests, dovetailing with broader virtio‑mem adoption across data‑center fleets. Meanwhile, s390x itself benefits from generation‑17 CPU definitions and the option to bypass the IOMMU for PCI devices when raw speed trumps isolation.
Last but definitely not least, a brand‑new “cpr‑transfer” live‑update mode debuts for administrators who need to roll out QEMU updates without lengthy downtime. The team also fixed a year-old issue where paused VMs could deadlock during migration—a bug that had bitten more than one unlucky operator.
For those interested in the specifics, visit the changelog, which lists all the changes in detail.
The new release can be downloaded from QEMU’s official download page, where users can find the tarball. Prebuilt packages are expected to appear in rolling distributions within days.
