Changelog:
System emulation
Incompatible changes
The aio=native option to "-drive" now requires the cache=none option, instead of silently disabling itself for other cache modes. The newly invalid combination had been warning since QEMU 2.3.
Specifying block device parameter aio=native is now an error on POSIX systems if qemu is compiled without libaio support. The newly invalid combination had been warning since QEMU 2.3.
The experimental x-drive option for the sdhci-pci device has been removed. Instead of passing a drive directly to the SD controller device you now must create an SD card object (which will automatically be plugged into the SD controller), so "-device sdhci-pci,x-drive=mydrive -drive id=mydrive,[...]" becomes "-device sdhci-pci -device sd-card,drive=mydrive -drive id=mydrive,[...]".
The s390-virtio machine has been removed.
Machine types pc-q35-1.4, pc-q35-1.5, pc-q35-1.6, pc-q35-1.7, pc-q35-2.0, pc-q35-2.1, pc-q35-2.2 and pc-q35-2.3 have been removed.
The "virt" machine type's flash device has changed when TrustZone is active ("-machine virt,secure=on"). The first flash device is only available in secure memory, while the second is available in non-secure memory too.
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.
The original qcow2 image encryption is fatally flawed, and support for it will be disabled entirely from the system emulators. It'll remain available only in command line tools qemu-img, qemu-io, qemu-nbd to facilitate data liberation. It is recommended to use 'qemu-img convert' to convert qcow2 encrypted images to uncrypted ones. The new LUKS encryption driver can provide a secure replacement if raw files are acceptable, while a future release will integrate luks into qcow2 natively.
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.
ARM
Support for a separate EL3 address space
System mode supports BE8 and BE32. Note that qemu-system-arm can emulate both big-endian and little-endian guests (unlike user-mode emulation which has separate qemu-arm and qemu-armeb binaries).
Support for the SETEND instruction, used most notably on Raspbian through the arm-mem library (previously known as libcofi).
Faster boot thanks to DMA support in fw_cfg
The "virt" machine type supports a virtual power button and the "system_powerdown" monitor command
The "virt" machine type supports configuring network cards with -nic in addition to -netdev
The RAM limit for the "virt" machine type is now 255GB
The "xlnz-zynqmp" machine type now includes SPI controllers
The "xlnx-ep108" machine type now supports SPI flash
New partial Raspberry Pi 2 emulation with "raspi2" machine type. For now, it can boot older releases of Windows and Raspbian, but lacks a number of devices including USB.
New palmetto-bmc machine type using the new, partial ASPEED AST2400 SoC implementation
KVM
Support for guest debugging (software and hardware breakpoints, single step) on AArch64
MIPS
Support for FPU and MSA in KVM guests
Support for R6 Virtual Processors
Initial support for Cluster Power Controller and Global Configuration Registers allowing the guest to control the start of Virtual Processors
Support for Inter-Thread Communication Unit
Support for MAAR registers in P5600 CPU
PowerPC
Improved support for migration of g3beige and mac99 machines
Fix serial ports for g3beige and mac99 machines (OpenBIOS)
The gdb stub supports the VSX instruction set extensions
pSeries
pSeries machine types starting at pseries-2.6 use XHCI as the USB host controller instead of OHCI
Support for more hypercalls (H_SET_SPRG0, H_SET_DABR, H_SET_XDABR and H_PAGE_INIT)
Support for EEH on assigned PCI devices can use the normal spapr-pci-host-bridge instead of the special spapr-pci-vfio-host-bridge.
s390
Fixes and improvements in s390x PCI support
Support for hotplug of s390x cpus via cpu-add
Support for booting from virtio-scsi devices in the s390-ccw bios
SH
SPARC
sun4m: Fix for ldstub instruction resolves several 32-bit Solaris bugs (MUTEX_HELD hang, libC error, Java WebStart segfault)
sun4u: FreeBSD 10.3+ can now run under qemu-system-sparc64 in -nographic mode
TileGX
Tricore
Support for context management, illegal opcode and opd traps
Support for FPU instructions
x86
TCG
Support for the XSAVE/XSAVEOPT, MPX, FSGSBASE and PKE features
KVM
Support for "split irqchip". In this mode, QEMU emulates the IOAPIC, PIC (i8259) and PIT (i8254) devices while leaving the local APIC emulation to the kernel. This mode reduces the attack surface of KVM.
Support for the new PKU feature found in some Skylake processors
Support for migrating the TSC rate
Xen
Q35
Support resume (S3)
Support for legacy Windows guests (XP/2003)
Device emulation and assignment
New IPMI emulation subsystem. QEMU can now emulate an internal BMC or attach to an external BMC simulator such as OpenIPMI's lanserv. IPMI however is not yet exposed in SMBIOS and ACPI tables (do we want to docume?)
FIXME: what's the state of nvdimm?
ACPI
The floppy disk controller's characteristics are now exposed in the ACPI tables, which makes it possible to use floppies on Windows together with UEFI firmware.
Block devices
The floppy disk consk or an empty disk to a 2.88 MB disk
Improved compatibility of the SD device model with various operating systems and firmwares
The NVMe device supports the "bootindex" property.
The SDHCI device supports reset.
ivshmem
No longer available on hosts lacking eventfd(2), because inter-vm interrupts don't work there
New devices ivshmem-plain and ivshmem-doorbell, fully backwards compatible for guests, notable differences to ivshmem:
PCI revision is 1 instead of 0
ivshmem role=master becomes master=on, role=peer becomes master=off
ivshmem x-memdev=ID becomes ivshmem-plain memdev=ID
ivshmem shm=NAME,size=SZ becomes ivshmem-plain memdev=ID, with -object memory-backend-file,id=ID,mem-path=/dev/mem/NAME,size=SZ,share
ivshmem chardev=ID becomes ivshmem-doorbell,chardev=ID
Property ioeventfd defaults to on instead of off
ivshmem-plain never has MSI-X capability, and ivshmem-doorbell always has MSI-X capability
Device ivshmem is deprecated, and its experimental property x-memdev is gone
Interrupting a peer that reuses an unplugged peer's ID works again (broken in v1.2.0)
Unplug no longer destroys the character device, for consistency with other devices
The funny "no shared memory, yet" state is no longer guest-visible, and can no longer fail or mess up migration
Guests may require PCI revision 1 to make sure they're not exposed to the funny state
docs/specs/ivshmem-spec.txt rewritten for completeness and accuracy.
SCSI
Support for the LSI SAS1068 HBA (also known as "MPT Fusion"). Note that some operating systems will not recognize disks attached to this adapter, unless the disks are assigned a world-wide name (WWN).
PCI/PCIe
PCIe Multi-root support (using the new pxb-pcie root-compex)
USB
MTP: initial support for events
VFIO
Support for AMD XGBE platform passthrough
New sysfsdev property provides a more general way to specify the device to attach to.
Provided PCI option ROMs are fixed to include the same vendor and device id as the device exposed to the guest. This facilitates changing the ids of the devices.
virtio
Performance improvements via optimized vring accesses
The balloon driver statistics now include the amount of available memory (corresponding to "Available" in /proc/meminfo for Linux guests).
Character devices
The socket character device backend can now enable TLS over TCP connections, acting either as a TLS server:
$QEMU -object tls-creds-x509,id=tls0,dir=$HOME/.pki/qemutls,endpoint=server \
-chardev socket,id=s0,host=127.0.0.1,port=9000,tls-creds=tls0,server \
-device isa-serial,chardev=s0 \
...other args...
or a TLS client:
$QEMU -object tls-creds-x509,id=tls0,dir=$HOME/.pki/qemutls,endpoint=client \
-chardev socket,id=s0,host=127.0.0.1,port=9000,tls-creds=tls0 \
-device isa-serial,chardev=s0 \
...other args...
If operating in server mode, the same set of TLS credentials can be used for both character devices and the VNC server
All character devices can have their output logged to a plain file
$QEMU -chardev stdio,id=mon0,logfile=monitor.log \
-mon chardev=mon0 \
...other args...
will result in logging of all output on the HMP monitor. The logappend parameter controls whether the file is truncated at startup, defaulting to append.
GUI
SDL2 and SPICE now support OpenGL and virgl. For SPICE, Unix sockets are the only usable transport when OpenGL is enabled.
The "-vnc" and "-display vnc" options support ipv4=off and ipv6=off. Previously, only "ipv4" and "ipv6" were available.
Support getting input events directly from linux evdev devices, using "-object input-linux,id=$name,evdev=/dev/input/event$nr"
Support for ncurses on Windows.
Monitor
Support for a new "detach" option to "dump-guest-memory". The option dumps memory in the background. Progress can be queried using the new commands "info dump" (human monitor) and "query-dump" (QMP), as well as through the QMP event DUMP_COMPLETED.
Support for a new command "input-send-event" replacing the previous experimental command "x-input-send-event".
The human monitor command "drive_add -n" allows creating block devices that do not have a BlockBackend (similar to QMP blockdev-add).
Migration
Postcopy is not experimental anymore; the x-postcopy-ram capability was renamed to postcopy-ram.
Network
SLIRP now supports IPv6 for ICMP, UDP, TCP and TFTP.
mirror filter which can mirror traffic from netdev to socket chardev, vice versa.
redirector filter which can redirect traffic from netdev to socket chardev, vice versa.
Secret passing system
There is a new standard mechanism for securely passing secret credentials to QEMU, which will be used in combination with other subsystems. For example, network block device passwords, block device decryption passphrases, or TLS private key passwords can all use the same mechanism.
Passing credentials inline (insecure, only for developer testing)
$QEMU -object secret,id=sec0,data=letmein
Passing credentials via a plain file
$QEMU -object secret,id=sec0,file=mypassword.txt
Passing credentials via a base64 encoded file
$QEMU -object secret,id=sec0,file=mypassword.txt,format=base64
Passing credentials inline, encrypted with a master key (recommended for management apps)
$QEMU -object secret,id=master0,file=mykey.b64,format=base64 \
-object secret,id=sec0,data=[base64 ciphertext],\
keyid=master0,iv=[base64 IV],format=base64
TLS credential handling
It is now possible to use encrypted TLS private keys with credentials for TLS servers/clients in QEMU. The password for unlocking the private key is provided by a secret object whose id is specified via the passwordid' property
$QEMU -object secret,id=tlskey0,file=mypassword.txt \
-object tls-creds-x509,id=tls0,dir=$HOME/.pki/qemutls,endpoint=server,passwordid=tlskey0 \
...other args...
Block devices
Block device throttling now support specifying a burst length as well. While previously the burst could only be specified as a total number of IOPS (e.g. 10000 IOPS), more complex specifications such as "10000 IOPS for 10 seconds" are now possible. Note that, because of the implementation of the algorithm, a guest that is allowed "10000 IOPS for 10 seconds" will also be allowed to perform for example 5000 IOPS for 20 seconds.
The curl block device driver now supports HTTP authentication and HTTP proxy authentication via the new properties 'username', 'password-secret', 'proxy-username' and 'proxy-password-secret'.
$QEMU -object secret,id=sec0,file=password.txt \
-object secret,id=sec1,file=proxy-password.txt \
-drive driver=http,host=localhost,port=443,username=fred,password-secret=sec0,proxy-username=bob,proxy-password-secret=sec1 \
...other args...
The RBD block device driver can now use the secret object type to securely receive the authentication password without exposing it in the command line args
$QEMU -object secret,id=sec0,file=password.b64,format=base64 \
-drive driver=rbd,filename=rbd:pool/image:id=myname:auth_supported=cephx,password-secret=sec0 \
...other args...
The iSCSI block device driver can now use the secret object type to securely receive the authentication password without exposing it in the command line args
$QEMU -object secret,id=sec0,file=password.txt \
-iscsi user=fred,password-secret=sec0 \
-drive file=iscsi://192.168.122.1:3260/iqn.2013-12.com.example%3Aiscsi-chap-netpool/1
NB this syntax requires that all iSCSI backed drives use the same password
The qemu-io tool gained support for new '--object' and '--image-opts' arguments. The --object argument allows 'secret' and 'tls-creds-x509' objects to be defined for use in association with a block device backend. The '--image-opts' argument instructs qemu-io to parse the image string as a set of image options, instead of a plain filename. For example, to connect qemu-io to an NBD server using TLS
qemu-io -c "read 0 512" \
--object tls-creds-x509,id=tls0,dir=$HOME/.pki/qemutls,endpoint=client \
--image-opts driver=nbd,host=localhost,port=10809,tls-creds=tls0
The qemu-nbd tool gained support for new '--object' and '--image-opts' arguments. The --object argument allows 'secret' and 'tls-creds-x509' objects to be defined for use in association with a block device backend or the NBD server. The '--image-opts' argument instructs qemu-io to parse the image string as a set of image options, instead of a plain filename. For example, to connect qemu-nbd to an HTTP server with authentication and export it over NBD using TLS
qemu-nbd --readonly \
--object secret,id=sec0,file=passwd.txt \
--object tls-creds-x509,id=tls0,dir=$HOME/.pki/qemutls,endpoint=server \
--image-opts driver=http,url=http://some.random.host/some/image,username=fred,password-secret=sec0
The qemu-img tool gained support for new '--object' and '--image-opts' arguments. The --object argument allows 'secret' and 'tls-creds-x509' objects to be defined for use in association with a block device backend or the NBD server. The '--image-opts' argument instructs qemu-io to parse the image string as a set of image options, instead of a plain filename. For example, to a remote HTTP server with authentication
qemu-img info --object secret,id=sec0,file=passwd.txt \
--image-opts driver=http,url=http://some.random.host/some/image,username=fred,password-secret=sec0
Support for deleting snapshots on Sheepdog devices.
The NBD client and server now support use of TLS. When enabled, the server will mandate that the client also enable TLS and drop any client which attempts to continue in plain text. To run a qemu-nbd server with TLS:
qemu-nbd --object tls-creds-x509,id=tls0,dir=$HOME/.pki/qemutls,endpoint=server \
--tls-creds tls0 \
/path/to/disk/image
To connect to a server that requires TLS with qemu-img:
qemu-img info --object tls-creds-x509,id=tls0,dir=$HOME/.pki/qemutls,endpoint=client \
--image-opts driver=nbd,host=localhost,port=10809,tls-creds=tls0
To start a VM pointing to the NBD server
$QEMU -object tls-creds-x509,id=tls0,dir=$HOME/.pki/qemutls,endpoint=client \
-drive driver=nbd,host=localhost,port=10809,tls-creds=tls0 \
...other args...
The NBD server gained support for specifying an export name. When the client negotiates use of the new style NBD protocol the default export name is "". The --exportname argument allows this to be customized:
qemu-nbd --exportname myvol /path/to/myvol.qcow2
QEMU gained support for volumes formatted with the LUKSv1 data format. To format a new LUKS volume
qemu-img create -f luks \
--object secret,id=sec0,file=passphrase.txt \
-o key-secret=sec0 \
demo.luks 10G
To boot a guest from a LUKS volume:
$QEMU -object secret,id=sec0,file=passphrase.txt \
-drive driver=luks,key-secret=sec0,file=demo.luks \
...other args...
The LUKS implementation is intended to be compatible with that used by cryptsetup/dm-crypt, so it should be possible to use disk images interchangeably between them. The only caveat is that some less common cipher/hash algorithms are not yet supported by QEMU. It is also not yet possible to manage key-slots with qemu-img.
TCG
Record/replay support extended to cover character devices.
Tracing
The "stderr" tracing backend was replaced by the "log" tracing backend, which is now the default. This backend prints tracing messages to the destination specified with the "-D" option.
In addition to the existing "-trace file=...", tracepoints can be enabled using "-trace [enable=]...". The new option also supports globbing, as in "-trace bdrv_aio_*".
In addition to the existing "-trace file=...", tracepoints can be enabling using "-d trace:...". This option also supports globbing, as in "-d trace:bdrv_aio_*".
When using "-daemonize", the "-D" option also provides the file to which QEMU's stderr output will be redirected.
TCG supports a new "-dfilter" option to limit exec, out_asm, op and op_opt logging to a range of guest physical addresses. ARM also applies the filter to in_asm logging; this will be extended to other targets in future releases (FIXME: probably should do it now instead...)
A "%d" substring in the log file name is replaced with QEMU's pid.
User-mode emulation
The default CPU for ppc64 and ppc64le is now POWER8
a6bc21154c