The handling of the floppy device controller is different between <2.4 and >=2.4 machine types that use the Q35 chipset (e.g. "-M pc-q35-2.3" vs. "-M pc-q35-2.4"). This can cause problems if you are defining floppy drives with command-line options such as "-global isa-fdc.driveA=id.
The ARM 'virt' board default interface type has changed from IDE to virtio. This means that some incorrect command lines that we previously silently accepted will now fail with an error message like "qemu-system-arm: -drive file=img.qcow2,id=foo: Drive 'foo' is already in use because it has been automatically connected to another device (did you need 'if=none' in the drive options?)". As the error message suggests, you should add "if=none" to the -drive option to fix this.
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.
Host floppy device pass-through (block driver "host_floppy") is deprecated, and will be dropped in a future release.
Block device parameter aio=native has no effect without cache.direct=on. 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 configure option --disable-guest-base is unneeded and will be removed in a future release.
Major fixes to the implementation of floating point exceptions.
New board model xlnx-ep108
Support for ACPI v5.1 tables in the "-M virt" board.
Support for instantiation sysbus devices from the command line (using "-device") in the "-M virt" board.
Emulation of the stream ID for MSI writes, for use in GICv3.
The "virt" board default drive type is now virtio; this means that drives created with if=virtio or with no if= specification will be created as drives plugged into a virtio-blk-pci device. Short form options like -hda will also create this kind of drive. (Note that at time of writing Linux only supports the virt board's PCI controller for 32-bit ARM kernels; support has not yet made it into the 64-bit kernels. So 64-bit guests will need to continue using long command lines and virtio-mmio for now.) Unfortunately this means that some old command lines will need to change -- see the "incompatible changes" section above for details.
More accurate emulation of the dp8393x network card and RC4030 DMA/IOMMU controller.
Support for microMIPS32 R6 emulation (enabled in new "-cpu mips32r6-generic")
Support for unaligned R6 and MSA memory accesses in TCG
Support for XPA in MIPS32 and LPA in MIPS64 (eXtended and Large Physical Addressing) emulation
Support for MIPS UHI semihosting
The default RAM size for the pseries machine is 512 MB.
Support of PCI device hotplug on SPAPR (pSeries).
Channel I/O is now available when running with TCG. Thus, the default machine for qemu-system-s390x is now s390-ccw.
Several other fixes for TCG (emulation) mode.
Extended name and UUID in STSI 3.2.2 information block
Support for reading/writing guest memory while holding the IPTE lock under KVM, including access register mode
Various cleanups in the s390-virtio and virtio-ccw transports
Support for diag288 watchdog (KVM only).
Support for vector registers
Add virtio-1 specific ccws to virtio-ccw (SET_REV and v1 version of SET_VQ)
Revision 1 (and therefore virtio 1) is not yet enabled, however.
The s390-ccw-virtio machine is now versioned; the first versioned machine is s390-ccw-virtio-2.4
The s390-ccw alias has been removed
Optimizations to code generated by TCG.
Fix SunOS 4.1.4 boot on sun4m with OpenBIOS
Improvements to system management mode emulation, including support for high SMRAM and TSEG on machines using the Q35 chipset.
q35 machine types starting with pc-q35-2.4.0 do not have a floppy disk controller. It will be created if you use "-drive if=floppy", "-fda" or "-fdb" to add a floppy disk drive, or it can be created if necessary with "-device isa-fdc".
q35 now implements the TCO watchdog. Unlike real hardware, the watchdog is disabled when the virtual machine boot, so as to let existing firmware run with new QEMU. This may change in the future, but the change would be restricted to new machine types.
Support for MMIO operations outside the "big QEMU lock". For now, this only applies to the ACPI PM timer, which can alone improve performance substantially for very large Windows guests as long as they do not span multiple NUMA nodes in the host. For guests that span multiple NUMA nodes more kernel changes are required.
Support for system management mode (requires Linux 4.2).
When running under KVM, CPUID information includes the ARAT ("Always running APIC timer") bit
Device emulation and assignment
Support for memory hot-unplug.
S3/S4 states can be disabled for boards using the Q35 machine type via "-global ICH9_LPC.disable_s3=on" and"-global ICH9_LPC.disable_s4=on"
Minimal support in NVMe emulation for the NVME_VOLATILE_WRITE_CACHE feature.
The infamous floppy device controller is not added to Q35 boards if not explicitly requested and no floppy drives are specified with "-drive if=floppy".
I/O throttling now supports groups, so I/O limits can be shared among multiple drives (use -drive throttling.group=<name>).
Performance of the parallels image format block driver has been improved
Improved support for flow control in virtio-serial.
Fix macio CDROM detection for PPC
Fix macio data corruption bug under Darwin/OS X PPC
non-32bit register reads to the AHCI device work correctly now.
Numerous NCQ fixes and adjustments, principally relating to migration and pause/resume.
AHCI drives now support rerror=stop and werror=stop fully.
AHCI devices may now be migrated, allowing default Q35 configurations to be migrated.
AHCI CDROM devices now report the correct signature, which fixes a boot hang in some experimental OVMF versions with SATA support.
AHCI ICC bits are now correctly ignored, fixing an OpenBSD boot hang.
CVE-2015-5154: Correct buffer overrun.
Support for virtio-keyboard, virtio-mouse, virtio-tablet.
Support for the "rocker" L2 switch device.
scsi-generic now supports migration.
Support for extra PCI root buses using PCI expander bridge devices. Unlike PCI-PCI bridges, a PCI expander bridge's bus can be associated with a NUMA node, allowing the guest OS to recognize the proximity of a device to RAM and CPUs.
Support for TPM 2
Support for resetting AMD Bonaire and Hawaii GPUs
Platform device passthrough support for Calxeda xgmac devices
Support for up to 1024 queues.
Support for ioeventfd in virtio-mmio.
Support for virtio 1 standard:
For virtio-pci, Virtio 1 mode can now be enabled by specifying disable-modern=false. To suppress the legacy interface and create a non-transitional device, specify disable-legacy=true additionally. All devices now support this mode.
This support is still experimental: it functions fine in normal usage and so should be appropriate for guest driver development, but there are known issues around migration, guest out of memory handling, and performance.
When used with virtio 1, virtio blk no longer supports scsi command passthrough. Please use virtio scsi instead.
New virtio-gpu device (only supports accelerated 2D for now)
New virtio-input-host, virtio-keyboard, virtio-mouse and virtio-tablet devices (and corresponding virtio-*-pci devices for use on a PCI bus).
Support for cross-endian vhost (i.e. little-endian host and big-endian guest, or vice versa).
vhost can now be enabled even if MSI-X is not
virtio-balloon can tell the guest that it should deflate the balloon on OOM conditions.
Support for virtio-vga, a VGA device that also supports the virtio-gpu interface.
Improved support for flow control in spice-char.
Support for OpenGL-based display rendering in the SDL2 and GTK+ backend. This is preparatory work for 3D acceleration.
Improvements to the Cocoa front-end, fixing full-screen mode and adding a list of consoles to the View menu.
The two extra keys in Brazilian 107-key keyboards are now usable.
qmp: New MIGRATION event to communicate change in the migration state
Support for compression of RAM data using multiple threads for compression and decompression (using migration capability "compress" and migration parameters "compress_threads", "compress_level" and "decompress_threads").
Support for multi-queue vhost-user backends.
Block devices in system emulation
The BLOCK_IMAGE_CORRUPTED event has a new "node-name" field.
Block device mirroring supports concurrent unmap (aka discard or trim) operations on the source device and can create a thin-provisioned image in this case.
Block device mirroring can use discard or "write zero" operations to speed up copying of large zero regions.
A longhand version of -global (-global driver=DRIVER,property=PROP,value=VAL) is introduced, to set properties globally for devices that have a period in their name. The older syntax -global DRIVER.PROP=VAL did not allow this.
New option -fw_cfg to pass arbitrary binary data to the guest.
New command line option "-icount sleep=no". The option will run emulation at the maximum possible speed: every time the CPU would go to sleep, the virtual clock will move to the next timer deadline. For virtual machines that have no other sources of non-determinism (e.g. asynchronous block I/O, character devices or user input) this will also make execution deterministic.
Block devices and tools
The "null" block device now has a new "latency-ns" option to delay the answer from the block device.
The iscsi driver can use the target's FUA capabilities to greatly improve roundtrip times in write-through caching modes (cache.writeback=off). These modes are recommended when the storage has a non-volatile (battery-backed) cache.
Parallels format driver now supports image creation and write to the image. Performance is significantly improved.
qcow2 performance improvements.
qemu-io supports encrypted qcow2 images (which are deprecated).
Obsolete audio backends have been removed: esd (superseded by pulseaudio), winwave (superseded by dsound), fmod (not compatible with the GPL)
Support for building a .msi file with the Windows version of the guest agent ("make msi"). This requires msitools.
qemu-ga implements guest-get-fsinfo and guest-network-get-interfaces on Windows too
The default CPU for qemu-sh4 and qemu-sh4eb is the sh7785.
QEMU now requires a minimum glib version of 2.22. (In particular, we will no longer build on a stock RHEL5 or Centos 5 system.)
QEMU can now optionally be linked against tcmalloc.
QEMU now compiles using clang 3.5 without warnings, which includes disabling GCC features not supported by clang.
QEMU now compiles with ICC.
libepoxy is required to compile QEMU with OpenGL support.
Building on Mac OS X versions earlier than 10.5 is no longer supported.
Sound on Windows now requires DirectSound (the old 'winwave' default audio backend has been dropped).
SDL audio only works with SDL 1.x.
Problems with QEMU for Windows and builds from newer versions of MinGW-w64, see this thread on qemu-devel (this is not a regression, other versions are also affected)