ChangeLog/10.1

Release schedule: Planning/10.1.

System emulation

Removed features and incompatible changes

Consult the 'Removed features' page for details of suggested replacement functionality.

New deprecated options and features

Consult the "Deprecated Features" chapter of the QEMU System Emulation User's Guide for further details of the deprecations and their suggested replacements.

Arm

• New CPU architectural features emulated:
  • FEAT_SME2
  • FEAT_SME2p1
  • FEAT_SME_B16B16
  • FEAT_SME_F16F16
  • FEAT_SVE_B16B16
  • FEAT_SVE2p1

• The 'highbank' and 'midway' board models have been deprecated
• New board model 'max78000fthr' (Analog Devices max78000 Cortex-M4 board)
• CXL can now be used on the Arm 'virt' board
• The 'virt' board now supports using KVM nested virtualization via 'virtualization=on' with a sufficiently new host kernel

HPPA

• Fix & improve floating-point exception (FPE) emulation in user- and system-mode

RISC-V

ISA and Extensions

• Use tail pseudoinstruction for calling tail
• Fix some RISC-V vector instruction corner cases
• Remove capital 'Z' CPU properties
• Support atomic instruction fetch (Ziccif)
• Add max_satp_mode from host cpu
• Extend and configure PMP region count
• Add a property to set vill bit on reserved usage of vsetvli instruction
• Add Svrsw60t59b extension support

Machines

• Add support for RIMT to virt machine ACPI
• Improve Microchip Polarfire SoC customization
• Add Kunminghu CPU and platform

Fixes and Misc

• Don't allow PMP RLB to bypass rule privileges
• Fix checks on writes to pmpcfg in Smepmp MML mode
• Generate strided vector loads/stores with tcg nodes
• Fix endless translation loop on big endian systems
• Fix write_misa vs aligned next_pc
• KVM CSR fixes
• sstc extension fixes
• Fix zama16b order in isa_edata_arr
• Profile handling fixes
• Extend PMP region up to 64
• Add missing named features
• Fix PPN field of Translation-reponse register
• Fix fcvt.s.bf16 NaN box checking
• Avoid infinite delay of async xmit function
• Fix missing exit TB flow for ldff_trans
• Fix migration failure when aia is configured as aplic-imsic
• Fix MEPC/SEPC bit masking for IALIGN

Device emulation and assignment

Network devices

• passt networking backend

VFIO

• Initial support for CoCo guest-memfd memory backends

virtio

• virtio-gpu: allow injection of virtio-gpu EDID name

9pfs

• Fixes for file descriptor reclaiming algorithm (i.e. when running towards host's allowed limit of max. open file descriptors).

• Additional fixes on use-after-unlink idiom (i.e. client operations on a file descriptor after file has been removed).

Audio

• alsa: Set "try-poll" to false by default
• add float sample endianness converters

Crypto subsystem

• Use of the built-in AES impl is now removed for anything other than the TCG instruction emulator. Any other use of AES now requires QEMU to be built with one of gcrypt, nettle or gnutls as crypto providers.

I/O subsystem

• Most device backends which support TCP connectivity can now support configuration of TCP keepalive parameters

GUI

• spice/dbus: support multi-plane dmabuf
• spice: enable gl=on option for non-local or remote clients
• vdagent: add migration support
• gtk: improve scale handling
• gtk: add keep-aspect-ratio and scale option
• vnc: encoding of updates in non-tight VNC encodings are fixed when the client and server have different endianess
• vnc: encoding of updates in 8/16-bpp is fixed when the client and server have different endianess
• vnc: fix display pause issue after migration

GDBStub

• added support for the qGDBServerVersion used by non-gdb clients

TCG Plugins

• ips plugin has gained configurable scaling and instructions per quantum arguments
• new functions to write register values
• new functions to write to virtual address (see caveats)
• new functions to read/write hardware addresses (see caveats)
• new test plugin called patcher
• sparc64 and alpha targets can now export register details to plugins

Migration

• RDMA live migration starts to support ipv6
• Postcopy preempt mode optimization on sequential memory accesses (e.g. video streaming)
• Initial supports of enabling both multifd and postcopy features (multifd is only enabled during precopy so far)
• Fix VFIO multifd live migration when tcp zerocopy send is enabled
• Precopy downtime optimization by avoiding unnecessary LOG_CLEAR operations

Monitor

QMP

• blockdev-mirror gained a new boolean flag target-is-zero that can be used to optimize mirroring of a sparse file to a destination that already reads as zero. Even without use of this flag, mirroring has been optimized to do less work on portions of the source that read as zero.
• blockdev-backup gained a new option on-cbw-error that allow users to specify how to behave when IO errors occur during copy-before-write operations.

Network

• QEMU now attempts to arrange for larger buffers when using NBD over Unix sockets, which dramatically improves throughput. On MacOS, this works out of the box; but on Linux, the system administrator must make the following changes:

   $ cat /etc/sysctl.d/net-mem-max.conf
   net.core.wmem_max = 2097152
   $ sudo sysctl -p /etc/sysctl.d/net-mem-max.conf

Block jobs

• Non-active block-commit was optimized to keep sparseness
• blockdev-mirror was optimized to do less work with zero blocks
• blockdev-mirror and blockdev-backup gained new options, see QMP section

Build Information

Rust usage

The minimum supported version of Rust is now 1.77.

• On Debian bookworm, it is available in the rustc-web package except for the mips64el architecture; Rust support for mips64el requires Debian trixie or newer.
• On Ubuntu 22.04 and 24.04, it is available in the rust-1.77 package; the RUSTC and RUSTDOC environment variables must point to /usr/bin/rustc-1.77 and /usr/bin/rustdoc-1.77 respectively.

Debian bookworm remains a supported platform for QEMU; support for Rust is still considered experimental and not stable, and it is not recommended to be used for anything other than development.

When Rust is enabled, QEMU now requires Meson 1.8.1.

Host support

• Support for Debian bullseye has been dropped.
• Ninja 1.9 is now required. This version of Ninja was already found on all supported platforms (including Debian bullseye) as of QEMU 10.0.
• Experimental support for compiling to WASM using Emscripten.

Known issues

• See https://gitlab.com/qemu-project/qemu/-/milestones/16