KVM Forum 2022
who it's for
KVM maintainers, QEMU developers, virtualization security engineers, Linux kernel developers, cloud hypervisor developers, arm64 maintainers
topics
Intel TDX host support; AMD SEV-SNP and SVSM; confidential attestation and encrypted dump; pKVM isolated TEE VMs on x86; SecondaryVM host-spawned VMs; libblkio high-performance block I/O API; eBPF-based QEMU extensions; virtio 1.2 and virtio-net RoCE; vhost vDPA and vfio-user devices; architecture-neutral KVM MMU; IOMMUFD integration in QEMU; live migration and VMM live upgrade optimizations
| title | author | length |
|---|---|---|
| KVM-devirt: Extending KVM to a Zero-Overhead Partition Hypervisor | Liang Deng | 20m |
KVM-devirt extends KVM to run bare-metal partitions that eliminate VM exits and remove stage-2 and IOMMU DMA-remap translations, passing APIC registers… | ||
| Intel Trusted Domain Extensions (TDX) Host Kernel Support | Kai Huang | 27m |
Explains how the kernel actually initializes the Intel TDX module so KVM can create protected VMs: a short sequence — global SIMCALL init on any CPU, per-CPU… | ||
| Exploring I/O Support for Virtualization-Based Trusted Execution Environment | Hao Wu | 25m |
Current direct I/O forces TVM data into unprotected shared memory, requiring decrypt/copy-out and encrypt/copy-in steps that add noticeable latency; the… | ||
| Providing Confidential Guest Services with a Secure VM Service Module on AMD | Thomas Lendacky | 28m |
AMD implements an SVSM running at VMPL0 that provides confidential guest services like secure boot, a vTPM, and controlled live migration; the prototype is a… | ||
| Attestation and Confidential Dump for IBM Secure Execution on Linux | Steffen Eiden, Marc Hartmayer | 25m |
IBM Secure Execution on Linux adds two concrete features: explicit attestation via PVA test that returns ECDH-based measurements so a third party can verify a… | ||
| SEV-SNP Development Status Update | Michael Roth, Ashish Kalra | 34m |
Guest SEV‑SNP support and CPU ID security are upstream as of kernel 5.19; hypervisor patches are at version 6 with a v7 planned. Hardware enforces integrity… | ||
| Allowing an Intel TDX Module to Run Without SEAM | Isaku Yamahata | 23m |
Demonstrates running the Intel TDX module on ordinary hardware by bypassing SEAM loading and measurement: the buildable TDX image is copied into a host… | ||
| Nesting Secure VM Hosts | Janosch Frank | 16m |
Nesting secure hosts on s390 implements nested secure vCPUs with about 300 lines of code and a broader Trusted Entity emulation prototype totalling 2,000 lines… | ||
| No More Turtles: The SecondaryVM Framework - An Alternative to Nested... | Mengmei Ye, Angelo Ruocco | 26m |
Prototype implements a SecondaryVM framework that lets a primary VM request host-spawned secondary VMs using a vsock channel and a host-side daemon, enforced… | ||
| Supporting TEE on x86 Client Platforms with pKVM | Jason Chen | 22m |
Implements pKVM on x86 to run isolated TEE VMs (TVMs) alongside a passthrough primary VM with only minor host-kernel changes, using EPT shadowing for memory… | ||
| Now You See Me, Now You Don't: Splitting pKVM Into Discrete, Mutually Exclusive Addr... | Marc Zyngier | 28m |
pKVM adds a set of mutually exclusive address spaces so each vCPU state and its hypervisor stack live only on one physical CPU at a time, using an extra… | ||
| Fuzzing Host-to Guest Attack Surface in Android Protected KVM | Eugene Rodionov, Will Deacon | 26m |
Android pKVM’s host-to-guest attack surface receives an LKL-based fuzzing harness that feeds mutated virtio PCI/MMIO inputs into front-end drivers; fuzzers run… | ||
| Panel Discussion: KVM-based Virtualization Contributor Q&A | Sean Christopherson, Christoffer Dall, Will Deacon, Susie Li, Paolo Bonzini, Kashyap Chamarthy | 38m |
KVM work shows concrete wins and clear limits: pkvm in Android 13 is shipping and runs parts of the Android runtime compiler, using stage-two translation so… | ||
| Scaling KVM and Its Community | Sean Christopherson | 33m |
KVM now spans six architectures, RISC-V ported in 2021, and contains roughly 150,000 lines of KVM code across 300+ files with over 90,000 lines of tests; in… | ||
| Introducing the libblkio High-performance Block I/O API | Stefan Hajnoczi, Alberto Faria | 36m |
libblkio provides a single C API (library implemented in Rust) that exposes drivers for io_uring, io_uring NVMe CMD, vfio PCI/virtio-blk, vhost-user, and… | ||
| Unifying Confidential Attestation | Tobin Feldman-Fitzthum, Dov Murik | 41m |
Compares SEV-ES, SEV-SNP and Intel TDX measurements and proposes two concrete unification paths: a secure vTPM rooted in hardware (benefit: reuse existing TPM… | ||
| Open-Source QEMU and RTL Co-simulation | Edgar Iglesias | 39m |
AMD‑Xilinx uses QEMU as a fast transaction‑level virtual platform glued to SystemC/TLM and RTL simulators so software boots before silicon. The open source… | ||
| Protected KVM on Arm64: A Technical Deep Dive | Quentin Perret | 50m |
Extension of KVM/arm64 that implements confidential computing on Arm v8.0+ CPUs without extra HW by enabling stage-2 MMU enabled for host so EL2 enforces host… | ||
| AIA Virtualization in KVM RISC-V | Anup Patel | 30m |
Implements RISC-V Advanced Interrupt Architecture virtualization inside KVM with a working proof‑of‑concept on QEMU, OpenSBI, Linux KVM and kvmtool. Key… | ||
| Emulating Hyper-V in 2022 | Vitaly Kuznetsov | 21m |
KVM now implements enlightened MSR bitmap nesting for both Intel and AMD SVM and supports the Hyper-V tweak that avoids Auto‑EOI so modern Windows actually… | ||
| Bringing the Power of eBPF to QEMU | Chen Zhang | 24m |
Adds an Apache-licensed iovisor/ubpf library into QEMU to run eBPF programs in userspace, with an assembler, disassembler, interpreter and JIT for x86-64 and… | ||
| The Challenges of Asynchronous Teardown of KVM Guests | Claudio Imbrenda | 25m |
Large (multi‑TB) KVM guests can take minutes to free memory; a laptop benchmark shows ~13 GB/s so a 16 TB guest needs ~20 minutes to shut down. Multiple… | ||
| Exploring an Architecture-neutral KVM MMU | David Matlack | 28m |
KVM’s TDP MMU gets refactored so about 90% of the tdpmu becomes architecture-neutral and portable to ARM, keeping features like eager page splitting, lockless… | ||
| KVM/ARM at Scale: Improvements to the MMU in the Face of Hardware Growing Pains | Oliver Upton | 19m |
Google Cloud runs a 48 vCPU VM with 192 GB RAM and two‑MiB huge pages to exercise dirty‑tracking; enabling dirty logging initially causes a 99% slowdown and a… | ||
| Hot-resizing of Windows VMs via Hyper-V Dynamic Memory Protocol | Maciej S. Szmigiero | 30m |
A QEMU hv-balloon driver implements the Hyper-V Dynamic Memory protocol to hot-add, hot-remove and balloon Windows VM RAM at single-page granularity; an MVP… | ||
| Exploring Intel Sub-Page Protection in Context of Live Migration | Manish Mishra, Shivam Kumar | 25m |
Live migration integrates Intel sub-page write protection (SPP) into KVM/QEMU to track 128‑byte subpages and avoid retransferring whole 4KB pages; prototype… | ||
| Preserving IOMMU States During Kexec Reboot | Fam Zheng | 33m |
Live-update of the host kernel preserves IOMMU mappings across a kexec by reserving guest RAM in a 2 GB reserved DAX region and allocating IOMMU page tables… | ||
| Deep Optimization of VMM Live Upgrade | Shenming Lu | 14m |
Replaces the naive reuse of live-migration by keeping vhost-user backends running and inheriting their FDs, adding an upgrade-agreement mode that avoids… | ||
| Postcopy Preemption | Peter Xu | 29m |
Introduces QEMU postcopy preemption, which adds a dedicated urgent-page channel, a fast page-resolve thread and per-channel page ownership so urgent requests… | ||
| Bring RoCE (RDMA over Converged Ethernet) Capability to Virtio-net Device | Yongji Xie | 19m |
Extends virtio-net to provide full RoCE by implementing the RDMA datapath in the hypervisor back end rather than in guest drivers or SR‑IOV passthrough, adding… | ||
| All Bark and no Bite: vCPU Stall Detection for KVM Guests | Sebastian Ene | 17m |
A para‑virtual vCPU stall detector implements a per‑vCPU MMIO watchdog: crosvm exports per‑CPU registers, the guest writes periodic heartbeats to a Linux… | ||
| What's in Virtio 1.2 - And What Isn't There | Michael S. Tsirkin | 27m |
Virtio 1.2 ships as a three‑year, pandemic‑delayed spec with nine new device types (persistent memory with DAX support, RPMSG-like GPIO events, IOMMU bypass… | ||
| Keynote Session: KVM Status Update | Paolo Bonzini | 14m |
KVM sees ~2500 commits this cycle with ~200 backported to stable and activity up ~10% year over year; Google contributes ~1000 extra commits. RISC-V support… | ||
| Keynote Session: QEMU Status Update | Alex Bennée | 13m |
Project stats show about 7,500 commits, ~500 pull requests and ~450 developers over the year; Meson replaces most Makefiles and LCI is used to produce flatter… | ||
| The Five Big Problems with Confidential Containers | Christophe de Dinechin | 26m |
Confidential containers are now a CNCF sandbox but face five concrete hurdles: attestation interoperability across Intel TDX/AMD SEV and RATS plus a KBS key… | ||
| Slirp is Dead, Long Live Slirp! A New Approach to User-mode Networking | Stefano Brivio, Alona Paz | 28m |
A new user-mode networking daemon, passt, reimplements libslirp’s job with a purpose-built codebase and strict seccomp profile, shipping as a standalone… | ||
| eBPF-based Extensible Paravirtualization | Luigi Leonardi | 27m |
Framework injects host-injected eBPF into guests via a virtual device and guest agent that loads verified eBPF ELF, writes affinity hints to BPF maps, and… | ||
| Hardware Friendly Vhost vDPA: Towards an Efficient and Migratable Device Model | Si-Wei Liu | 31m |
Proposes a simplified vhost‑vDPA migration model that saves device state as TLV device‑state blobs (device‑level, work‑queue, feature sub‑states) and a backend… | ||
| QEMU & KVM Automated Performance Benchmarking @ SUSE | Dario Faggioli | 29m |
SUSE describes an automated benchmarking framework built on MMTests that runs benchmarks inside and across multiple VMs, creates and installs guest OS images… | ||
| QEMU/KVM Upgrade Test (Stable Guest ABI and in Place Upgrade) | Min Deng | 22m |
Stable guest ABI guarantees a VM’s virtual hardware profile does not change across QEMU/KVM upgrades and the test suite validates live migration types… | ||
| CPU Feature Management: Lessons from Trenches | Soham Ghosh, Shivam Kumar | 27m |
Nutanix runs VMs across heterogeneous hosts and documents concrete failures where libvirt and QEMU disagree: pconfig is added in QEMU 3.1 then removed in 4.0… | ||
| QEMU-Storage-Daemon and libblkio: Exploring New Shores for the QE... | Kevin Wolf, Stefano Garzarella | 30m |
qemu-storage-daemon moves the QEMU block layer into a separate process to provide qemu-storage-daemon exports (NBD, FUSE, vhost-user-blk, vhost-vdpa), enable… | ||
| Virtio Devices Emulation in SPDK Based on Vfio-User Protocol | Changpeng Liu, Xiaodong Liu | 21m |
SPDK implements a virtio PCIe abstraction layer and emulates virtio-blk and virtio-scsi over the vfio-user client/server model, comparing it to vhost-user… | ||
| NVMe Emulation Performance Optimization | Jinhao Fan, Klaus Jensen | 21m |
QEMU NVMe emulation moves from ~30 KIOPS to ~330 KIOPS by implementing shadow doorbell buffers, using ioeventfd to avoid full VM exits, adding a dedicated IO… | ||
| How Fast Can We Go? Booting a Linux VM to Userspace in 100ms and Beyond | Rob Bradford | 33m |
Claims a Linux VM boots to userspace in 100ms and reports 64 ms achieved on a modern desktop; measurement uses a guest-triggered PIO to a debug I/O port plus… | ||
| IOMMUFD Integration in QEMU | Yi Liu, Eric Auger | 27m |
iommufd is a new user API that centralizes IO page‑table management for user memory and is intended to supplant VFIO uAPIs and serve VFIO, vDPA, etc. QEMU adds… | ||
| I2C Multi-master and Controller Slave Mode in QEMU | Klaus Jensen | 25m |
Implemented I2C multi‑master and controller slave mode in QEMU by adding an asynchronous I2C send API and an event callback plus a bottom‑half state machine… | ||
| Lessons Learned: Optimizing KVM Performance for EHR Systems | Jon Kohler | 26m |
KVM changes include a new APIC fastpath handler for Windows IPI delivery, suppression of expensive spec-control MSR reads and a host/guest xsave mask fix, plus… | ||
| Running KubeVirt Workloads with No Additional Privileges | Ľuboslav Pivarč | 19m |
KubeVirt now runs VMs without granting cluster-wide extra privileges by landing a rootless user mode that combines libvirt’s unmanaged networking, offloading… | ||
| Hypervisor.Framework - Virtualization on macOS | Alexander Graf | 31m |
Hypervisor.Framework exposes a tiny macOS kernel shim that drives Intel VMX or ARM EL2 while QEMU uses a userland library to create VM context and create vCPU… | ||
| nothing matches | ||