guard pages are now inserted after guest RAM, to guard against guest-triggered buffer overflow attacks
The mips32r5-generic CPU was renamed to P5600
Host floppy device pass-through (block driver “host_floppy”) has been removed; it is still possible to use them just like any other device file, however, a medium change will no longer be passed through to the guest
Future incompatible changes
Three options are using different names on the command line and in configuration file. In particular:
The "acpi" configuration file section matches command-line option "acpitable";
The "boot-opts" configuration file section matches command-line option "boot";
The "smp-opts" configuration file section matches command-line option "smp".
-readconfig will standardize on the name for the command line option.
Behavior of automatic calculation of SMP topology when some SMP topology options for -smp are omitted (sockets, cores, threads) will change in the future. If guest ABI needs to be preserved on upgrades while using the SMP topology options, users should either set set all options explicitly (sockets, cores, threads), or omit all of them.
Image encryption is fatally flawed, and will be dropped entirely. It'll remain available only in qemu-img, so you can use 'qemu-img convert' to convert encrypted images to uncrypted ones.
Block device parameter aio=native has no effect without cache.direct=on. It will be made an error.
Block device parameter aio=native has no effect if qemu is compiled without libaio support. It will be made an error.
A few devices will be configured with explicit properties instead of implicitly. Unlikely to affect users; for the full list, see the 2.3 ChangeLog.
QMP command blockdev-add is still a work in progress. It doesn't support all block drivers, it lacks a matching blockdev-del, and more. It might change incompatibly.
The s390-virtio machine has been deprecated for 2.5; it will be removed in 2.6. s390x users should switch to the (default) s390-ccw-virtio machine.
Changes to device "sdhci-pci" will make migration between old and new versions impossible.
We intend to drop support for running QEMU on MacOSX 10.5 hosts in the QEMU 2.6 release, unless somebody who uses it wishes to step forward and help us with regular testing.
The "virt" machine type supports passing SMBIOS to the firmware.
Semihosting support on AArch64
New i.MX31 SoC.
The ZynqMP and Allwinner A10 platforms support AHCI.
Support for VGICv3 in KVM
Support for GICv3 in the ACPI tables.
The "virt" machine now has a second PCIe MMIO region of 512GB in size in high memory. Note that older 32-bit ARM Linux kernels built without CONFIG_LPAE have a bug where the presence of this region in high memory causes them to refuse to use the PCIe controller at all. In this case you can either reconfigure your kernel with CONFIG_LPAE=y, or pass QEMU the "-machine highmem=off" option to disable the use of high memory for PCIe. The kernel bug is expected to be fixed in Linux kernel release 4.4.
The mips32r5-generic CPU was renamed to P5600
Improvements to MIPS R6 emulation
Support for memory hotplug
The shipped version of SLOF includes GPT support.
Using VFIO doesn't need spapr-pci-vfio-host-bridge anymore.
Improve ability to boot MacOS 9 (based upon GSoC project "Implement support for Mac OS 9 in QEMU " by Cormac O'Brien)
Storage keys are migrated.
New "info skeys" command in HMP to dump the storage key for a given address.
Support for virtio 1 in the virtio-ccw devices.
A maximum virtio-ccw revision can be specified via the "max_revision" property: max_revision=0 may be used to enforce usage of legacy virtio mode.
Support for boot from El Torito iso images on virtio-blk has been added.
sun4u: Fix EBus device enumeration under FreeBSD SPARC64 (OpenBIOS)
The emulated IOMMU (VT-d) supports devices behind a bridge
QEMU will warn when using a "-cpu" model that includes unsupported features. These features are disabled automatically, just like in previous versions of QEMU
/machine/icc-bridge was removed from the QOM tree. Software relying on icc-bridge to find CPU objects should use the "qom_path" field of "query-cpus" QMP command
CPU models and features
Haswell and Broadwell CPU models now include ABM
Cache information passthrough (which was enabled by default on "-cpu host") is now disabled by default
ABM, POPCNT, and SSE4a are not enabled in the default CPU models (qemu64, qemu32) anymore, as many hosts don't support it
RDTSCP was removed from AMD CPU models, as current KVM versions can't expose RDTSCP to guests in AMD hosts
New Intel memory instructions (clflushopt/clwb/pcommit) are now supported
TCG now supports Debug Extensions (CR4.DE)
Support for Hyper-V-compatible reporting of crashes.
Support for passthrough of Intel integrated GPUs.
Device emulation and assignment
fw_cfg supports a DMA interface on ARM and x86. This interface makes -kernel/-initrd much faster if supported by the firmware. SeaBIOS supports the DMA interface starting with release 1.9.0 (commit 06316c9d). The UEFI guest fw for ARM VMs (known as ArmVirtQemu or AAVMF) supports the DMA interface starting with git commit 953bcbcc / SVN r18545.
AHCI ATAPI PIO transfers greater than one sector are fixed. This helps on OVMF, which utilizes such transfers.
Zero byte count limits will no longer hang ATAPI drives.
ATAPI PIO reads, when the byte count limit is set to the ATAPI block size, are now asynchronous. This should help to prevent hangs due to missing CDROM media mounted via NFS.
scsi-generic devices now can be used together with a "readonly=on" backend.
Shared memory device
A sample implementation of the ivshmem client and ivhsmem server is included with QEMU.
ivshmem supports a new "x-memdev" property that uses a memory object for the backend. This makes it possible to use ivshmem with a hugetlbfs backend. This property is experimental and may be removed in future release in favour of an alternative ivshmem device.
Use kvm irqfd for msi notifications
Protocol change, native long -> LE int64_t
hotplug support added for multifunction devices. To use, add all functions except function 0 first. Adding function 0 exposes the device to the guest. Request removal on function 0. On guest acknowledge, all functions are ejected together.
virtio-gpu now supports 3D mode
vhost-user now supports live migration. client changes are required to enable this. When used with an old client without migration support, vhost-user will now block migration (instead of failing silently)
vhost-user now supports multi-queue. Use queues=# to enable this. client changes are required to enable this mode. When used with an old client without multi-queue support, device will automatically fall back on using a single pair of queues.
vhost-user protocol now includes protocol feature negotiation, including multiple new messages. When used with old clients, all new messages are automatically disabled.
vhost-user no longer sents the RESET_OWNER message on device stop. The only QEMU version that sent it was 2.4, the message is now officially deprecated.
migration now works when virtio 1 is enabled for virtio-pci
For virtio-pci, virtio 1 performance on kvm on Intel CPUs has been improved (on kernel 4.4 and up).
a new flag modern-pio-notify can be used to enable PIO for notifications in virtio 1 mode, to improve performance for host kernels older than 4.4, and processors without EPT support.
virtio devices can now be placed on the pci express bus
vhost is no longer disabled when guest does not use MSI-X. The vhostforce flag is no longer required.
in virtio 1 mode, scsi passthrough is now disabled for virtio blk
Please note that for virtio-pci, the modern (virtio 1) interface is still disabled by default. To enable, set the flag disable-modern=off.
New syntax for enabling TLS in the VNC server:
Equivalent to -vnc hostname:0,tls: -object tls-creds-anon,id=tls0,endpoint=server -vnc hostname:0,tls-creds=tls0
Equivalent to -vnc hostname:0,tls,x509=/path/to/certs: -object tls-creds-x509,id=tls0,endpoint=server,dir=/path/to/certs,verify-peer=no -vnc hostname:0,tls-creds=tls0
Equivalent to -vnc hostname:0,tls,x509verify=/path/to/certs: -object tls-creds-x509,id=tls0,endpoint=server,dir=/path/to/certs,verify-peer=yes -vnc hostname:0,tls-creds=tls0
The Cocoa GUI does not have show an 'open image file' dialog box anymore even if QEMU is started without arguments
Thu curses GUI supports 256 colors and line graphics.
Support for network filters. Currently, the only filter objects are "filter-buffer", which batches packets every N microseconds, and "filter-dump", which can be used to log the network traffic in a file. Filters are attached to a netdev device using e.g. "-object filter-buffer,id=filter,netdev=net0,queue=rx,interval=1000" (which creates a 1ms filter-buffer).
Block devices in system emulation
Improved system emulation performance for targets with software TLBs (e.g. SPARC).
The HMP “change” command (QMP's “blockdev-change-medium”) now allows you to change the read-only mode of the device (e.g. when inserting a read-only floppy disk image into a previously R/W drive)
Fine-grained control over a block device's tray with the new QMP commands “blockdev-open-tray”, “blockdev-close-tray”, “x-blockdev-insert-medium”, and “x-blockdev-remove-medium” (the latter two are experimental for now)
New “reopen” command in qemu-io
block-dirty-bitmap-add and block-dirty-bitmap-clear transaction actions have been added to now fully support (transient) incremental bitmap usage and management.
QMP transactions now support a "completion-mode" parameter which controls the completion behavior of jobs launched by transactions, which will allow them to fail together. See the bitmaps.md documentation for how this affects incremental backups.
Block I/O accounting can now report average queue depth, min/avg/max latency, and failed/invalid request counts
qcow2 learnt a new option cache-clean-interval, which allows to free unused cache entries after some time.
An experimental QMP command x-blockdev-del was added as a completement for the (also still experimental) blockdev-add command.
A new QMP command blockdev-snapshot that allows creating a snapshot using as overlay an image previously opened with blockdev-add. This allows opening the overlay image with arbitrary run-time options, solving one of the limitations of blockdev-snapshot-sync.
It is now possible to open an image without its backing file by specifying the empty string as a backing file reference when opening the image. This is useful for creating snapshots, since images opened with blockdev-add are not supposed to have a backing file before the blockdev-snapshot operation.
Host CD-ROM support now works on Mac OS X hosts
Host floppy support has been removed (it was deprecated in QEMU 2.3)
The temporary "x-data-plane=on/off" option for virtio-blk device is removed now, all users are requested to use the canonical "-object iothread,id=<id> -device virtio-blk,iothread=<id>,..." syntax.
Add an optional qemu-ga.conf system configuration
Support for dumping the configuration current file with --dump-conf
Win32 support for guest-set-user-password
New command guest-exec
The configure option --disable-guest-base has been removed.
libcacard has been moved to a standalone project, hosted at git://anongit.freedesktop.org/spice/libcacard. The libcacard library from QEMU 2.4 can also be used to build QEMU 2.5.
virtio-gpu 3D support requires virglrenderer.
SDL audio only works with SDL 1.x.
64-bit QEMU might crash on Windows (problems with stack unwinding, depends on build environment, possible fix)
QEMU's configure script fails with pdksh from OpenBSD (see bug #1525682. Using another shell with configure should work.