QEMU(1)

📦

version 11.1.3 · 8 versions

✨ QEMU ✨

okay so welcome to the QEMU manual page, bestie — this is part 1 of 26, so grab a snack 🍓, we're gonna be here a while. This is the qemu page: QEMU User Documentation, and it's genuinely iconic.

✨ NAME ✨

qemu — QEMU User Documentation. that's it, that's the name 🫶

✨ SYNOPSIS ✨

qemu-system-x86_64 [options] [disk_image]

✨ DESCRIPTION ✨

so basically the QEMU PC System emulator simulates all these peripherals, like, a whole computer's worth of them 💅:

i440FX host PCI bridge and PIIX3 PCI to ISA bridge
Cirrus CLGD 5446 PCI VGA card or a dummy VGA card with Bochs VESA extensions (hardware level, including all non standard modes)
PS/2 mouse and keyboard
2 PCI IDE interfaces with hard disk and CD-ROM support
Floppy disk
PCI and ISA network adapters
Serial ports
IPMI BMC, either and internal or external one
Creative SoundBlaster 16 sound card
ENSONIQ AudioPCI ES1370 sound card
Intel 82801AA AC97 Audio compatible sound card
Intel HD Audio Controller and HDA codec
Adlib (OPL2) - Yamaha YM3812 compatible chip
Gravis Ultrasound GF1 sound card
CS4231A compatible sound card
PC speaker
PCI UHCI, OHCI, EHCI or XHCI USB controller and a virtual USB-1.1 hub.

SMP is supported with a large number of virtual CPUs — and the upper limit is configuration dependent, so no cap on the vibe, only on the config 😭

QEMU uses the PC BIOS from the Seabios project and the Plex86/Bochs LGPL VGA BIOS.

QEMU uses YM3812 emulation by Tatsuyuki Satoh.

QEMU uses GUS emulation (GUSEMU32 <http://www.deinmeister.de/gusemu/>) by Tibor "TS" Schütz. credit where credit's due, periodt 👑

ngl, a little heads up: by default, GUS shares IRQ(7) with parallel ports — so QEMU must be told to not have parallel ports to have working GUS. that's the whole thing, that's the rule.

qemu-system-x86_64 dos.img -device gus -parallel none

Alternatively:

qemu-system-x86_64 dos.img -device gus,irq=5

Or some other unclaimed IRQ. slay, your call 💅

CS4231A is the chip used in Windows Sound System and GUSMAX products — lowkey a fun fact, keep it.

The PC speaker audio device can be configured using the pcspk-audiodev machine property, i.e.

qemu-system-x86_64 some.img -audiodev <backend>,id=<name> -machine pcspk-audiodev=<name>

🎀 Machine-specific options 🎀

It supports the following machine-specific options:

x-south-bridge=PIIX3|piix4-isa (Experimental option to select a particular south bridge. Default: PIIX3)

bestie it literally says experimental so treat it with respect ✨

✨ OPTIONS ✨

disk_image is a raw hard disk image for IDE hard disk 0. Some targets do not need a disk image — no disk, no problem 🦋

so here's a thing that trips everyone up: when you're dealing with options parameters as arbitrary strings containing commas, such as in "file=my,file" and "string=a,b", it's necessary to double the commas. for instance, "-fw_cfg name=z,string=a,,b" will be parsed as "-fw_cfg name=z,string=a,b". double the comma, save a life 💀

🌈 Standard options 🌈

-h — Display help and exit
-version — Display version information and exit
-machine [type=]name[,prop=value[,...]] — Select the emulated machine by name. Use -machine help to list available machines.

For architectures which aim to support live migration compatibility across releases, each release will introduce a new versioned machine type. For example, the 2.8.0 release introduced machine types "pc-i440fx-2.8" and "pc-q35-2.8" for the x86_64/i686 architectures.

To allow live migration of guests from QEMU version 2.8.0, to QEMU version 2.9.0, the 2.9.0 version must support the "pc-i440fx-2.8" and "pc-q35-2.8" machines too. To allow users live migrating VMs to skip multiple intermediate releases when upgrading, new releases of QEMU will support machine types from many previous versions. the legacy support is honestly so valid of them 🫶

Supported machine properties are:

accel=accels1[:accels2[:...]] — This is used to enable an accelerator. Depending on the target architecture, kvm, xen, hvf, nitro, nvmm, whpx, mshv or tcg can be available. By default, tcg is used. If there is more than one accelerator specified, the next one is used if the previous one fails to initialize.
vmport=on|off|auto — Enables emulation of VMWare IO port, for vmmouse etc. auto says to select the value based on accel and i8042. For accel=xen or i8042=off the default is off otherwise the default is on.
dump-guest-core=on|off — Include guest memory in a core dump. The default is on.
mem-merge=on|off — Enables or disables memory merge support. This feature, when supported by the host, de-duplicates identical memory pages among VMs instances (enabled by default).
aes-key-wrap=on|off — Enables or disables AES key wrapping support on s390-ccw hosts. This feature controls whether AES wrapping keys will be created to allow execution of AES cryptographic functions. The default is on.
dea-key-wrap=on|off — Enables or disables DEA key wrapping support on s390-ccw hosts. This feature controls whether DEA wrapping keys will be created to allow execution of DEA cryptographic functions. The default is on.
nvdimm=on|off — Enables or disables NVDIMM support. The default is off.
confidential-guest-support=<id> — confidential guest support object to use. The default is none.
memory-encryption=<id> — The alias of confidential-guest-support. Recommend to use confidential-guest-support.
hmat=on|off — Enables or disables ACPI Heterogeneous Memory Attribute Table (HMAT) support. The default is off.
spcr=on|off — Enables or disables ACPI Serial Port Console Redirection Table (SPCR) support. The default is on.
aux-ram-share=on|off — Allocate auxiliary guest RAM as an anonymous file that is shareable with an external process. This option applies to memory allocated as a side effect of creating various devices. It does not apply to memory-backend-objects, whether explicitly specified on the command line, or implicitly created by the -m command line option. The default is off.

To use the cpr-transfer migration mode, you must set aux-ram-share=on. no cap, that's mandatory 🫡

memory-backend='id' — An alternative to legacy -mem-path and mem-prealloc options. Allows to use a memory backend as main RAM.

For example:

-object memory-backend-file,id=pc.ram,size=512M,mem-path=/hugetlbfs,prealloc=on,share=on
-machine memory-backend=pc.ram
-m 512M

Migration compatibility note:

as backend id one shall use value of 'default-ram-id', advertised by machine type (available via query-machines QMP command), if migration to/from old QEMU (<5.0) is expected.
for machine types 4.0 and older, user shall use x-use-canonical-path-for-ramblock-id=off backend option if migration to/from old QEMU (<5.0) is expected.

For example:

-object memory-backend-ram,id=pc.ram,size=512M,x-use-canonical-path-for-ramblock-id=off
-machine memory-backend=pc.ram
-m 512M

cxl-fmw.0.targets.0=firsttarget,cxl-fmw.0.targets.1=secondtarget,cxl-fmw.0.size=size[,cxl-fmw.0.interleave-granularity=granularity] — Define a CXL Fixed Memory Window (CFMW).

Described in the CXL 2.0 ECN: CEDT CFMWS & QTG _DSM.

They are regions of Host Physical Addresses (HPA) on a system which may be interleaved across one or more CXL host bridges. The system software will assign particular devices into these windows and configure the downstream Host-managed Device Memory (HDM) decoders in root ports, switch ports and devices appropriately to meet the interleave requirements before enabling the memory devices.

targets.X=target provides the mapping to CXL host bridges which may be identified by the id provided in the -device entry. Multiple entries are needed to specify all the targets when the fixed memory window represents interleaved memory. X is the target index from 0.

size=size sets the size of the CFMW. This must be a multiple of 256MiB. The region will be aligned to 256MiB but the location is platform and configuration dependent.

interleave-granularity=granularity sets the granularity of interleave. Default 256 (bytes). Only 256, 512, 1k, 2k, 4k, 8k and 16k granularities supported.

Example:

-machine cxl-fmw.0.targets.0=cxl.0,cxl-fmw.0.targets.1=cxl.1,cxl-fmw.0.size=128G,cxl-fmw.0.interleave-granularity=512

sgx-epc.0.memdev=<memid>,sgx-epc.0.node=<numaid> — Define an SGX EPC section.
smp-cache.0.cache=cachename,smp-cache.0.topology=topologylevel — Define cache properties for SMP system.

cache=cachename specifies the cache that the properties will be applied on. This field is the combination of cache level and cache type. It supports l1d (L1 data cache), l1i (L1 instruction cache), l2 (L2 unified cache) and l3 (L3 unified cache).

topology=topologylevel sets the cache topology level. It accepts CPU topology levels including core, module, cluster, die, socket, book, drawer and a special value default. If default is set, then the cache topology will follow the architecture's default cache topology model. If another topology level is set, the cache will be shared at corresponding CPU topology level. For example, topology=core makes the cache shared by all threads within a core. The omitting cache will default to using the default level.

The default cache topology model for an i386 PC machine is as follows: l1d, l1i, and l2 caches are per core, while the l3 cache is per die.

Example:

-machine smp-cache.0.cache=l1d,smp-cache.0.topology=core,smp-cache.1.cache=l1i,smp-cache.1.topology=core

-M — as -machine.
-cpu model — Select CPU model (-cpu help for list and additional feature selection)

and bestie, we are so not done yet — more where that came from 🌸

-accel name[,prop=value[,...]] — okay so this is the one that turns on an accelerator, aka the thing that makes your guest go zoom 💨. Depending on your target architecture you might have kvm, xen, hvf, nitro, nvmm, whpx, mshv or tcg available. Default is tcg (yeah, the pure-emulation one — no shame, it works everywhere). And here's the cute bit: if you list more than one accelerator, the next one gets used if the previous one fails to initialize. Built-in fallback chain, love that for you 🫶
- igd-passthru=on|off — only matters when Xen is in use. Controls whether Intel integrated graphics devices can be passed through to the guest. Default is off.
- kernel-irqchip=on|off|split — controls KVM in-kernel irqchip support. The default is full acceleration of the interrupt controllers ✨. On x86, split irqchip reduces the kernel attack surface — at a performance cost for non-MSI interrupts. Disabling the in-kernel irqchip completely is not recommended except for debugging purposes (fr, don't do it casually).
- kvm-shadow-mem=size — defines the size of the KVM shadow MMU.
- one-insn-per-tb=on|off — makes the TCG accelerator put only one guest instruction into each translation block. This slows down emulation a lot, but it can be useful in some situations — like when you're trying to analyse the logs produced by the -d option. iykyk 🔍
- split-wx=on|off — controls the use of split w^x mapping for the TCG code generation buffer. Some operating systems require this to be enabled, and in that case it defaults on. On other operating systems it defaults off, but you can enable it for testing or debugging.
- tb-size=n — controls the size (in MiB) of the TCG translation block cache. 🧠
- thread=single|multi — controls the number of TCG threads. When TCG is multi-threaded there's one thread per vCPU, so it takes advantage of additional host cores. The default is to enable multi-threading where both the back-end and front-ends support it and no incompatible TCG features have been enabled (e.g. icount/replay).
- dirty-ring-size=n — when the KVM accelerator is used, this controls the size of the per-vCPU dirty page ring buffer (the number of entries for each vCPU). It should be a value that's a power of two, and it should be 1024 or bigger (but still less than the maximum value that the kernel supports). 4096 could be a good initial value if you have no idea which is the best. Set this value to 0 to disable the feature. By default this feature is disabled (dirty-ring-size=0). When enabled, KVM will instead record dirty pages in a bitmap.
- eager-split-size=n — KVM implements dirty page logging at PAGE_SIZE granularity, and enabling dirty-logging on a huge-page requires breaking it into PAGE_SIZE pages in the first place. KVM on ARM does this splitting lazily by default. There are performance benefits in doing the huge-page split eagerly, especially in situations where TLBI costs associated with break-before-make sequences are considerable and also if guest workloads are read intensive 👀. The size here specifies how many pages to break at a time and needs to be a valid block size, which is 1GiB/2MiB/4KiB, 32MiB/16KiB and 512MiB/64KiB for 4KiB/16KiB/64KiB PAGE_SIZE respectively. Be wary of specifying a higher size — it will have an impact on the memory 🫠. By default this feature is disabled (eager-split-size=0).
- notify-vmexit=run|internal-error|disable,notify-window=n — enables or disables notify VM exit support on an x86 host and specifies the corresponding notify window to trigger the VM exit if enabled. The run option enables the feature — it does nothing and continues if the exit happens. The internal-error option enables the feature — it raises an internal error 💀. The disable option doesn't enable the feature. This feature can mitigate the CPU stuck issue that happens because event windows don't open up for a specified amount of time (i.e. notify-window). Default: notify-vmexit=run,notify-window=0.
- device=path — sets the path to the KVM device node. Defaults to /dev/kvm. This option can be used to pass the KVM device to use via a file descriptor by setting the value to /dev/fdset/NN.
- hyperv=on|off|auto — for the WHPX backend, determines whether to enable Hyper-V enlightenments. On x86_64, Hyper-V enlightenments are on by default. On AArch64, they're off by default.

-smp — the "how many CPUs and how are they stacked" option, and honestly it's giving architecture degree 🎓✨

-smp
[[cpus=]n][,maxcpus=maxcpus][,drawers=drawers][,books=books][,sockets=sockets][,dies=dies][,clusters=clusters][,modules=modules][,cores=cores][,threads=threads]

So this simulates an SMP system with n CPUs initially present on the machine type board. On boards that support CPU hotplug, the optional maxcpus parameter can be set to enable further CPUs to be added at runtime 🔥. When both parameters are omitted, the maximum number of CPUs will be calculated from the provided topology members and the initial CPU count will match the maximum number. When only one of them is given, then the omitted one gets set to its counterpart's value. Both parameters may be specified, but the maximum number of CPUs must be equal to or greater than the initial CPU count. The product of the CPU topology hierarchy must equal the maximum number of CPUs. Both parameters are subject to an upper limit determined by the specific machine type you chose.

To control reporting of CPU topology information, you can specify values for the topology parameters. Machines may only support a subset of the parameters, and different machines may support different subsets, which vary depending on the capacity of the corresponding CPU targets. So for a particular machine type board, an expected topology hierarchy can be defined through the supported sub-option. Unsupported parameters can also be provided in addition to the sub-option, but their values must be set as 1 for the purpose of correct parsing 👑

Either the initial CPU count, or at least one of the topology parameters, must be specified. The specified parameters must be greater than zero — explicit configuration like cpus=0 is not allowed. Values for any omitted parameters will be computed from the ones you gave.

For example, the following sub-option defines a CPU topology hierarchy (2 sockets totally on the machine, 2 cores per socket, 2 threads per core) for a machine that only supports sockets/cores/threads. Some members of the option can be omitted but their values will be automatically computed:

-smp 8,sockets=2,cores=2,threads=2,maxcpus=8

The following sub-option defines a CPU topology hierarchy (2 sockets totally on the machine, 2 dies per socket, 2 modules per die, 2 cores per module, 2 threads per core) for PC machines which support sockets/dies/modules/cores/threads. Some members of the option can be omitted but their values will be automatically computed:

-smp 32,sockets=2,dies=2,modules=2,cores=2,threads=2,maxcpus=32

The following sub-option defines a CPU topology hierarchy (2 sockets totally on the machine, 2 clusters per socket, 2 cores per cluster, 2 threads per core) for ARM virt machines which support sockets/clusters/cores/threads. Some members of the option can be omitted but their values will be automatically computed:

-smp 16,sockets=2,clusters=2,cores=2,threads=2,maxcpus=16

Historically, preference was given to the coarsest topology parameters when computing missing values (ie sockets preferred over cores, which were preferred over threads). However, this behaviour is considered liable to change 👀. Prior to 6.2 the preference was sockets over cores over threads. Since 6.2 the preference is cores over sockets over threads.

For example, the following option defines a machine board with 2 sockets of 1 core before 6.2 and 1 socket of 2 cores after 6.2:

-smp 2

Note: the cluster topology will only be generated in ACPI and exposed to the guest if it's explicitly specified in -smp. No sneaky auto-clusters, bestie 💅

-numa — the NUMA machine, and it comes in so many flavors 🍓

-numa node[,mem=size][,cpus=firstcpu[-lastcpu]][,nodeid=node][,initiator=initiator]

-numa node[,memdev=id][,cpus=firstcpu[-lastcpu]][,nodeid=node][,initiator=initiator]

-numa dist,src=source,dst=destination,val=distance

-numa cpu,node-id=node[,socket-id=x][,core-id=y][,thread-id=z]

-numa hmat-lb,initiator=node,target=node,hierarchy=hierarchy,data-type=type[,latency=lat][,bandwidth=bw]

-numa hmat-cache,node-id=node,size=size,level=level[,associativity=str][,policy=str][,line=size]

What it does: define a NUMA node and assign RAM and VCPUs to it 💖. Set the NUMA distance from a source node to a destination node. And set the ACPI Heterogeneous Memory Attributes for the given nodes.

Legacy VCPU assignment uses the cpus option, where firstcpu and lastcpu are CPU indexes. Each cpus option represents a contiguous range of CPU indexes (or a single VCPU if lastcpu is omitted). A non-contiguous set of VCPUs can be represented by providing multiple cpus options. If cpus is omitted on all nodes, VCPUs are automatically split between them — the machine just handles it, main character energy ✨

For example, the following option assigns VCPUs 0, 1, 2 and 5 to a NUMA node:

-numa node,cpus=0-2,cpus=5

🦋 `cpu` — the new-and-improved way to place CPUs

okay so cpu is a whole new alternative to the cpus option, and the way it works is that it uses the socket-id|core-id|thread-id properties to assign CPU objects to a node, using the CPU's topology layout properties. The set of properties is machine specific, and it depends on the machine type / smp options you're running — so don't come for me if your box has different ones. You can query the whole set with the hotpluggable-cpus monitor command, iykyk. The node-id property tells it which node the CPU object gets assigned to, and it's required — the node has to be declared with the node option before you use it with cpu. no skipping steps bestie.

For example:

-M pc \
-smp 1,sockets=2,maxcpus=2 \
-numa node,nodeid=0 -numa node,nodeid=1 \
-numa cpu,node-id=0,socket-id=0 -numa cpu,node-id=1,socket-id=1

🍓 `memdev` — the one you should actually be using

memdev assigns RAM from a given memory backend device to a node. And like… it is recommended that you use memdev over the legacy mem option, periodt. Why? Because memdev gives you better performance and more control over the backend's RAM — for example, the prealloc parameter of -memory-backend-ram lets you preallocate memory. obsessed.

🧚 the legacy `mem` situation

For compatibility reasons, the legacy mem option is supported in 5.0 and older machine types. But heads up: mem and memdev are mutually exclusive. If one node uses memdev, all the rest of the nodes have to use memdev too, and vice versa. No mixing, that's the rule 💅.

Also, users must specify memory for all NUMA nodes via memdev (or the legacy mem if it's available to you). And in QEMU 5.2, the support for -numa node without memory specified was removed. gone, reduced to atoms 😭.

🫶 `initiator` — the bestie node

initiator is an additional option that points to an initiator NUMA node that has the best performance (like, the lowest latency or the largest bandwidth) to this NUMA node. Note that you can only set this option when the machine property hmat is set to on.

The following example creates a machine with 2 NUMA nodes — node 0 has a CPU, node 1 has only memory, and its initiator is node 0. Note that because node 0 has a CPU, by default the initiator of node 0 is itself and must be itself:

-machine hmat=on \
-m 2G,slots=2,maxmem=4G \
-object memory-backend-ram,size=1G,id=m0 \
-object memory-backend-ram,size=1G,id=m1 \
-numa node,nodeid=0,memdev=m0 \
-numa node,nodeid=1,memdev=m1,initiator=0 \
-smp 2,sockets=2,maxcpus=2  \
-numa cpu,node-id=0,socket-id=0 \
-numa cpu,node-id=0,socket-id=1

📏 distances — the math part, stay with me

source and destination are NUMA node IDs. distance is the NUMA distance from source to destination. The distance from a node to itself is always 10. If any pair of nodes is given a distance, then all pairs must be given distances. Although — when distances are only given in one direction for each pair of nodes, then the distances in the opposite directions are assumed to be the same. But if an asymmetrical pair of distances is given for even one node pair, then all node pairs must be provided distance values for both directions, even when they're symmetrical. And when a node is unreachable from another node, set that pair's distance to 255.

lowkey important: the -numa option doesn't allocate any of the specified resources, it just assigns existing resources to NUMA nodes. So you still have to use the -m and -smp options to allocate RAM and VCPUs respectively. don't forget or it's giving nothing.

🌈 `hmat-lb` — latency and bandwidth, but fancy

Use hmat-lb to set System Locality Latency and Bandwidth Information between initiator and target NUMA nodes in the ACPI Heterogeneous Attribute Memory Table (HMAT). The initiator NUMA node can create memory requests, and usually it has one or more processors. The target NUMA node contains addressable memory.

In the hmat-lb option, node are NUMA node IDs. hierarchy is the memory hierarchy of the target NUMA node: if hierarchy is memory, the structure represents the memory performance; if hierarchy is first-level|second-level|third-level, this structure represents aggregated performance of memory side caches for each domain. The type of data-type is the type of data represented by this structure instance — if hierarchy is memory, data-type is access|read|write latency or access|read|write bandwidth of the target memory; if hierarchy is first-level|second-level|third-level, data-type is access|read|write hit latency or access|read|write hit bandwidth of the target memory side cache.

lat is the latency value in nanoseconds. bw is the bandwidth value — the possible value and units are NUM[M|G|T], meaning the bandwidth value is NUM byte per second (or MiB/s, GiB/s or TiB/s depending on the suffix you used). Note that if the latency or bandwidth value is 0, it means the corresponding latency or bandwidth information is not provided.

🎀 `hmat-cache` — cache info, but make it structured

In the hmat-cache option, node-id is the NUMA-id the memory belongs to. size is the size of the memory side cache in bytes. level is the cache level described in this structure — note that cache level 0 should not be used with the hmat-cache option 🚫. associativity is the cache associativity, and the possible values are none/direct(direct-mapped)/complex(complex cache indexing). policy is the write policy. line is the cache Line size in bytes.

For example, the following options describe 2 NUMA nodes. Node 0 has 2 cpus and a ram, node 1 has only a ram. The processors in node 0 access memory in node 0 with access-latency 5 nanoseconds, and access-bandwidth is 200 MiB/s; the processors in NUMA node 0 access memory in NUMA node 1 with access-latency 10 nanoseconds, and access-bandwidth is 100 MiB/s. And for memory side cache information, NUMA node 0 and 1 both have 1 level memory cache, size is 10KiB, policy is write-back, and the cache Line size is 8 bytes:

-machine hmat=on \
-m 2G \
-object memory-backend-ram,size=1G,id=m0 \
-object memory-backend-ram,size=1G,id=m1 \
-smp 2,sockets=2,maxcpus=2 \
-numa node,nodeid=0,memdev=m0 \
-numa node,nodeid=1,memdev=m1,initiator=0 \
-numa cpu,node-id=0,socket-id=0 \
-numa cpu,node-id=0,socket-id=1 \
-numa hmat-lb,initiator=0,target=0,hierarchy=memory,data-type=access-latency,latency=5 \
-numa hmat-lb,initiator=0,target=0,hierarchy=memory,data-type=access-bandwidth,bandwidth=200M \
-numa hmat-lb,initiator=0,target=1,hierarchy=memory,data-type=access-latency,latency=10 \
-numa hmat-lb,initiator=0,target=1,hierarchy=memory,data-type=access-bandwidth,bandwidth=100M \
-numa hmat-cache,node-id=0,size=10K,level=1,associativity=direct,policy=write-back,line=8 \
-numa hmat-cache,node-id=1,size=10K,level=1,associativity=direct,policy=write-back,line=8

💅 `-add-fd fd=fd,set=set[,opaque=opaque]`

Adds a file descriptor to an fd set. The valid options are:

fd=fd — this one defines the file descriptor of which a duplicate gets added to the fd set. And no, the file descriptor cannot be stdin, stdout, or stderr. hard no 🙈.
set=set — this defines the ID of the fd set you're adding the file descriptor to.
opaque=opaque — this defines a free-form string you can use to describe the fd. cute little label, love that for you.

You can open an image using pre-opened file descriptors from an fd set, like this:

qemu-system-x86_64 \
 -add-fd fd=3,set=2,opaque="rdwr:/path/to/file" \
 -add-fd fd=4,set=2,opaque="rdonly:/path/to/file" \
 -drive file=/dev/fdset/2,index=0,media=disk

⭐ `-set group.id.arg=value`

Sets the parameter arg for item id of type group. short, sweet, iconic.

🔥 `-global driver.prop=value` / `-global driver=driver,property=property,value=value`

Sets the default value of the driver's property prop to value, e.g.:

qemu-system-x86_64 -global ide-hd.physical_block_size=4096 disk-image.img

In particular, you can use this to set driver properties for devices which are created automatically by the machine model. If you want to create a device that is not created automatically and set properties on it, use -device instead, bestie.

-global driver.prop=value is shorthand for -global driver=driver,property=prop,value=value. And the longhand syntax works even when driver contains a dot. slay.

🧚 `-boot`

-boot [order=drives][,once=drives][,menu=on|off][,splash=sp_name][,splash-time=sp_time][,reboot-timeout=rb_timeout][,strict=on|off]

Specifies the boot order drives as a string of drive letters. The valid drive letters depend on the target architecture. The x86 PC uses: a, b (floppy 1 and 2), c (first hard disk), d (first CD-ROM), n-p (Etherboot from network adapter 1-4). Hard disk boot is the default. If you want a particular boot order to apply only on the first startup, specify it via once. Note that the order or once parameter should not be used together with the bootindex property of devices, since firmware implementations normally do not support both at the same time. pick a lane 😭.

Interactive boot menus/prompts can be enabled via menu=on, as far as firmware/BIOS supports them. The default is non-interactive boot.

A splash picture can be passed to the bios, letting the user show it as a logo, when the option splash=sp_name is given and menu=on, if the firmware/BIOS supports them. Currently SeaBIOS for x86 systems supports it. Limitation: the splash file has to be a jpeg file or a BMP file in 24 BPP format (true color). The resolution should be supported by the SVGA mode, so the recommended sizes are 320x240, 640x480, 800x640.

A timeout can be passed to the bios — the guest will pause for rb_timeout ms when boot fails, then reboot. If rb_timeout is '-1', the guest will not reboot, and qemu passes '-1' to the bios by default. Currently SeaBIOS for x86 systems supports it.

Do a strict boot via strict=on, as far as firmware/BIOS supports it. This only affects things when the boot priority is changed by bootindex options. The default is non-strict boot.

# try to boot from network first, then from hard disk
qemu-system-x86_64 -boot order=nc
# boot from CD-ROM first, switch back to default order after reboot
qemu-system-x86_64 -boot once=d
# boot with a splash picture for 5 seconds.
qemu-system-x86_64 -boot menu=on,splash=/root/boot.bmp,splash-time=5000

Note: the legacy format -boot drives is still supported, but its use is discouraged because it may be removed from future versions. 💅

-m [size=]megs[,slots=n,maxmem=size] — sets guest startup RAM to megs megabytes. Default is 128 MiB. You can add a suffix of M or G to mean megabytes or gigabytes respectively. The optional slots and maxmem pair sets the amount of hotpluggable memory slots and the maximum amount of memory. Heads up: maxmem must be aligned to the page size. 🧠

For example, this command line sets guest startup RAM to 1GiB, creates 3 slots to hotplug extra memory, and sets the maximum memory the guest can reach to 4GiB:

qemu-system-x86_64 -m 1G,slots=3,maxmem=4G

If slots and maxmem are not specified, memory hotplug won't be enabled and the guest startup RAM will never increase.

-mem-path path — allocate guest RAM from a temporarily created file in path.
-mem-prealloc — preallocate memory when using -mem-path.
-k language — use keyboard layout language (for example fr for French). This option is only needed where it is not easy to get raw PC keycodes — like on Macs, with some X11 servers, or with a VNC or curses display. You don't normally need it on PC/Linux or PC/Windows hosts. ✨

The available layouts are:

ar  de-ch  es  fo     fr-ca  hu  ja  mk     no  pt-br  sv
da  en-gb  et  fr     fr-ch  is  lt  nl     pl  ru     th
de  en-us  fi  fr-be  hr     it  lv  nl-be  pt  sl     tr

The default is en-us.

-audio [driver=]driver[,model=value][,prop=value[,...]] — if the model option is specified, -audio is a shortcut for configuring both the guest audio hardware and the host audio backend in one go. The guest hardware model can be set with model=modelname. Use model=help to list the available device types. 🎧

The following two examples do exactly the same thing, showing how -audio can shorten the command line:

qemu-system-x86_64 -audiodev pa,id=pa -device sb16,audiodev=pa
qemu-system-x86_64 -audio pa,model=sb16

If the model option is not specified, -audio configures a default audio backend that will be used whenever the audiodev property is not set on a device or machine. In particular, -audio none ensures that no audio is produced even for machines that have embedded sound hardware.

In both cases, the driver option is the same as with the corresponding -audiodev option below. Use driver=help to list the available drivers.

-audiodev [driver=]driver,id=id[,prop=value[,...]] — adds a new audio backend driver identified by id. 🔊

If no audio backend is specified, QEMU will attempt to select a default one. The -display option may influence which backend is selected.

There are global and driver specific properties. Some values can be set differently for input and output, and they're marked with in|out.. You can set the input's property with in.prop and the output's property with out.prop. For example:

-audiodev alsa,id=example,in.frequency=44110,out.frequency=8000
-audiodev alsa,id=example,out.channels=1 # leaves in.channels unspecified

NOTE: parameter validation is known to be incomplete. In many cases, specifying an invalid option causes QEMU to print an error message and continue emulation without sound. 😭

Valid global options are:

id=identifier — identifies the audio backend.
timer-period=period — sets the timer period used by the audio subsystem in microseconds. Default is 10000 (10 ms).
in|out.mixing-engine=on|off — use QEMU's mixing engine to mix all streams inside QEMU and convert audio formats when not supported by the backend. When off, fixed-settings must be off too. Note that disabling this option means the selected backend must support multiple streams and the audio formats used by the virtual cards, otherwise you'll get no sound. It's not recommended to disable this option unless you want to use 5.1 or 7.1 audio, as the mixing engine only supports mono and stereo audio. Default is on.
in|out.fixed-settings=on|off — use fixed settings for host audio. When off, it will change based on how the guest opens the sound card. In this case you must not specify frequency, channels or format. Default is on.
in|out.frequency=frequency — specify the frequency to use when using fixed-settings. Default is 44100Hz.
in|out.channels=channels — specify the number of channels to use when using fixed-settings. Default is 2 (stereo).
in|out.format=format — specify the sample format to use when using fixed-settings. Valid values are: s8, s16, s32, u8, u16, u32, f32. Default is s16.
in|out.voices=voices — specify the number of voices to use. Default is 1.
in|out.buffer-length=usecs — sets the size of the buffer in microseconds.

-audiodev none,id=id[,prop=value[,...]] — creates a dummy backend that discards all outputs. This backend has no backend specific properties. 🙈
-audiodev alsa,id=id[,prop=value[,...]] — creates a backend using ALSA. This backend is only available on Linux. 🐧

ALSA specific options are:

in|out.dev=device — specify the ALSA device to use for input and/or output. Default is default.
in|out.period-length=usecs — sets the period length in microseconds.
in|out.try-poll=on|off — attempt to use poll mode with the device. Default is off.
threshold=threshold — threshold (in microseconds) when playback starts. Default is 0.

-audiodev coreaudio,id=id[,prop=value[,...]] — creates a backend using Apple's Core Audio. This backend is only available on Mac OS and only supports playback. 🍎

Core Audio specific options are:

in|out.buffer-count=count — sets the count of the buffers.

-audiodev dbus,id=id[,prop=value[,...]] — creates a D-Bus backend. It must be associated with the display (as -display dbus,audiodev=id). (Since 7.0) 🚌

D-Bus specific options are:

nsamples — number of samples per read/write (default to 480, 10ms at 48kHz) (Since 10.0).

-audiodev dsound,id=id[,prop=value[,...]] — creates a backend using Microsoft's DirectSound. This backend is only available on Windows and only supports playback. 🪟

DirectSound specific options are:

latency=usecs — add extra usecs microseconds latency to playback. Default is 10000 (10 ms).

-audiodev oss,id=id[,prop=value[,...]] — creates a backend using OSS. This backend is available on most Unix-like systems. 🐚

OSS specific options are:

in|out.dev=device — specify the file name of the OSS device to use. Default is /dev/dsp.
in|out.buffer-count=count — sets the count of the buffers.
in|out.try-poll=on|off — attempt to use poll mode with the device. Default is on.
try-mmap=on|off — try using memory mapped device access. Default is off.
exclusive=on|off — open the device in exclusive mode (vmix won't work in this case). Default is off.
dsp-policy=policy — sets the timing policy (between 0 and 10, where a smaller number means smaller latency but higher CPU usage). Use -1 to use buffer sizes specified by buffer and buffer-count. This option is ignored if you do not have OSS 4. Default is 5.

-audiodev pa,id=id[,prop=value[,...]] — creates a backend using PulseAudio. This backend is available on most systems. 🌊

PulseAudio specific options are:

server=server — sets the PulseAudio server to connect to.
in|out.name=sink — use the specified source/sink for recording/playback.
in|out.latency=usecs — desired latency in microseconds. The PulseAudio server will try to honor this value but actual latencies may be lower or higher.

-audiodev pipewire,id=id[,prop=value[,...]] — creates a backend using PipeWire. This backend is available on most systems. 🧵

PipeWire specific options are:

in|out.latency=usecs — desired latency in microseconds.
in|out.name=sink — use the specified source/sink for recording/playback.
in|out.stream-name — specify the name of pipewire stream.

-audiodev sdl,id=id[,prop=value[,...]] — creates a backend using SDL. This backend is available on most systems, but you should use your platform's native backend if possible. 🎮

SDL specific options are:

in|out.buffer-count=count — sets the count of the buffers.

-audiodev sndio,id=id[,prop=value[,...]] — creates a backend using SNDIO. This backend is available on OpenBSD and most other Unix-like systems. 🐡

Sndio specific options are:

in|out.dev=device — specify the sndio device to use for input and/or output. Default is default.
in|out.latency=usecs — sets the desired period length in microseconds.

-audiodev spice,id=id[,prop=value[,...]] — creates a backend that sends audio through SPICE. This backend requires -spice and is automatically selected in that case, so usually you can ignore this option. This backend has no backend specific properties. 🌶️
-audiodev wav,id=id[,prop=value[,...]] — creates a backend that writes audio to a WAV file. 🎀

Backend specific options are:

path=path — write recorded audio into the specified file. Default is qemu.wav.

-device driver[,prop=value[,...]] — drops in a driver, bestie. prop=value sets that driver's properties, and which properties are even valid totally depends on which driver you picked. Want the receipts? Hit -device help for the driver list, and -device driver,help to see what a specific driver will let you set. Iconic little self-docs moment. 💅

Some of the drivers you can flex are:

🧚 `-device ipmi-bmc-sim,id=id[,prop=value[,...]]`

Adds an IPMI BMC — basically it's a whole simulation of that hardware management interface processor that normally just sits there on a system, doing its thing. It gives you a watchdog, plus the power to reset and power-control the system. But heads up: it's giving "useless on its own" energy until you connect it to an IPMI interface. 💀

Okay so the IPMI slave address used for the BMC — default is 0x20. That's the BMC's address on the I2C network of management controllers. And if that sentence made your brain do the dial-up noise, it's genuinely safe to just ignore it. 🙈

id=id — the BMC id, aka how interfaces refer to this device.
slave_addr=val — sets the slave address the BMC uses. Default is 0x20.
sdrfile=file — a file holding raw Sensor Data Records (SDR) data. Default is none.
fruareasize=val — the size of a Field Replaceable Unit (FRU) area. Default is 1024.
frudatafile=file — a file holding raw Field Replaceable Unit (FRU) inventory data. Default is none.
guid=uuid — the GUID value for the BMC, in standard UUID format. Set it, and a "Get GUID" command to the BMC returns it. Don't set it, and "Get GUID" returns an error. no cap.
lan.channel=val — channel number of the fake LAN channel. It can be 0 or any integer from 0x1 to 0xb inclusive. Set it to 0 and the LAN channel is disabled; anything else and the LAN channel is enabled at that channel number. Default is 0. 🫶
lan.ipaddr=addr — the ipv4 address of the BMC LAN channel.
lan.ipsrc=val — the ipv4 address source of the BMC LAN channel. Options are 0 (unspecified source), 1 (static address), 2 (DHCP), 3 (BIOS or system software), or 4 (other source). Default is 0.
lan.macaddr=mac — the mac address of the BMC LAN channel.
lan.netmask=mask — the ipv4 subnet mask for the BMC LAN channel.
lan.defgw_ipaddr=addr — the default gateway ipv4 address for the BMC LAN channel.
lan.defgw_macaddr=mac — the default gateway mac address for the BMC LAN channel.

🦋 `-device ipmi-bmc-extern,id=id,chardev=id[,slave_addr=val]`

Okay so instead of emulating the BMC locally like the thing above, this one makes a connection to an external entity that provides the IPMI services. Externally-sourced BMC, very independent, main character energy. ✨

The tea: you're connecting to an external BMC simulator. And if you do this, it is strongly recommended that you use the "reconnect-ms=" chardev option so it reconnects to the simulator if the connection gets lost. 🫶

💀 Careful with this one, bestie — if it's not used carefully it can be a security issue, because that interface has the ability to send resets, NMIs, and power off the VM. Like, full chaos potential. The move is to have QEMU connect to an external simulator running on a secure port on localhost, so neither the simulator nor QEMU is ever exposed to any outside network. periodt.

Wanna know more about the external interface? Peep the "lanserv/README.vm" file in the OpenIPMI library. 📚

🎀 `-device isa-ipmi-kcs,bmc=id[,ioport=val][,irq=val]`

Adds a KCS IPMI interface on the ISA bus. It also adds the corresponding ACPI and SMBIOS entries when appropriate, so it's a whole package deal. 💅

bmc=id — the BMC to connect to, one of ipmi-bmc-sim or ipmi-bmc-extern from above.
ioport=val — sets the I/O address of the interface. Default is 0xca0 for KCS.
irq=val — sets the interrupt to use. Default is 5. Wanna disable interrupts? Set it to 0. 🚫

🍓 `-device isa-ipmi-bt,bmc=id[,ioport=val][,irq=val]`

Same vibe as the KCS interface, but it defines a BT interface. Default port is 0xe4 and the default interrupt is 5. ⭐

🌸 `-device pci-ipmi-kcs,bmc=id`

Adds a KCS IPMI interface on the PCI bus. 🚌

bmc=id — the BMC to connect to, one of ipmi-bmc-sim or ipmi-bmc-extern above.

🌈 `-device pci-ipmi-bt,bmc=id`

Like the KCS interface, but this one defines a BT interface on the PCI bus. Same energy, different bus. 💫

👑 `-device intel-iommu[,option=...]`

Lowkey important: this is only supported by -machine q35, which enables Intel VT-d emulation inside the guest. It supports these options:

intremap=on|off (default: auto) — enables the interrupt remapping feature. It's required to enable a complete x2apic. Right now it only supports kvm kernel-irqchip modes off or split — full kernel-irqchip is not yet supported. 🙈 The default is "auto", which gets decided by the mode of kernel-irqchip.
caching-mode=on|off (default: off) — enables caching mode for the VT-d emulated device. When it's on, every guest DMA buffer mapping generates an IOTLB invalidation going from the guest IOMMU driver to the vIOMMU device, and it does it synchronously. It's required for -device vfio-pci to work with the VT-d device, because host assigned devices need the DMA mapping set up on the host before guest DMA starts. fr fr. 🫶
device-iotlb=on|off (default: off) — enables the device-iotlb capability for the emulated VT-d device. So far virtio/vhost should be the only real user of this parameter, paired with ats=on configured for the device. 💅
aw-bits=39|48 (default: 39) — decides the address width of the IOVA address space. That address space is 39 bits wide for 3-level IOMMU page tables, and 48 bits for 4-level IOMMU page tables.

Also, please do refer to the wiki page for the general scenarios of VT-d emulation in QEMU: <https://wiki.qemu.org/Features/VT-d>. 📖

🧚 `-device virtio-iommu-pci[,option=...]`

Only supported by -machine q35 (x86_64) and -machine virt (ARM). It supports these options:

granule=val (possible values are 4k, 8k, 16k, 64k and host; default: host) — decides the default granule that the virtio-iommu exposes. If it's host, the granule matches the host page size. 🎀
aw-bits=val (val between 32 and 64, default depends on machine) — decides the address width of the IOVA address space.

💜 `-device arm-smmuv3,primary-bus=id[,option=...]`

Only supported by -machine virt (ARM). 🌸

primary-bus=id — accepts either the default root complex (pcie.0) or a pxb-pcie based root complex.
accel=on|off (default: off) — lets the guest leverage host SMMUv3 features for acceleration. Turning on accel configures the host SMMUv3 in nested mode to support vfio-pci passthrough. 🔥

Okay so the following options get set to auto by default if you don't set them manually. When accel=on and these properties are set to auto, the value is derived from the host SMMUv3 capabilities via IOMMU_GET_HW_INFO. And with accel=on, this requires at least one cold-plugged vfio-pci device — if there isn't one present at machine init, QEMU will abort. 😭 No cap, that's real.

If accel=off, the auto values resolve to the non-accel defaults given below. 🫶

ril=on|off|auto (default: auto) — support for Range Invalidation, which lets the SMMUv3 driver invalidate TLB entries for a range of IOVAs all at once instead of firing off separate commands to invalidate each page. Has to match with host SMMUv3 Range Invalidation support.
- With accel=off, auto is resolved to 'on'.
ats=on|off|auto (default: auto) — support for Address Translation Services, which lets PCIe devices cache address translations in their local TLB and reduce latency. Host SMMUv3 must support ATS in order to enable this feature for the vIOMMU.
- With accel=off, auto is resolved to 'off'.
oas=val|auto (supported values are 44 and 48. default: auto) — sets the Output Address Size in bits. Whatever value you set here must be less than or equal to the host SMMUv3's supported OAS, so that the intermediate physical addresses (IPA) consumed by host SMMU for stage-2 translation don't exceed the host's max supported IPA size. ⭐
- With accel=off, auto is resolved to 44.
ssidsize=val|auto (val between 0 and 20. default: auto) — sets the Substream ID size in bits. When you set it to a non-zero value, PASID capability is advertised to the vIOMMU and accelerated use cases like Shared Virtual Addressing (SVA) are supported. 💫
- With accel=off, auto is resolved to 0.
cmdqv=on|off|auto (default: auto) — enables hardware Command Queue Virtualization (CMDQV) for the SMMUv3 command queue. Currently only the NVIDIA Tegra241 CMDQV implementation is supported. 👑
- With accel=on, auto means the value is automatically derived from the host SMMU.
- With accel=off, auto is resolved to 'off'.

🦋 `-device amd-iommu[,option=...]`

Enables emulation of an AMD-Vi I/O Memory Management Unit (IOMMU). Only available with -machine q35, and it supports this option:

dma-remap=on|off (default: off) — support for DMA address translation and access permission checking for guests that attach passthrough devices to paging domains, using the AMD v1 I/O Page Table format. This is what makes -device vfio-pci,... work correctly with a guest that's using the DMA remapping feature of the vIOMMU. 🫶
intremap=on|off (default: auto) — this is the generic x86 IOMMU functionality, brought to you by the amd-iommu device ✨. It switches on the interrupt remapping feature inside guests — and yes, you need it to enable x2apic support too. Right now it only plays nice with kernel-irqchip=off|split: it turns itself on automatically when either of those modes is in use, and stays off with kernel-irqchip=on. No cap, that's just how it is 💅
xtsup=on|off (default: off) — lets the interrupt remapping table support x2apic mode, which means the guest can use the 128-bit IRTE format with a 32-bit destination field 🎀. You need this to route interrupts to vCPUs whose APIC IDs are bigger than 0xff. Big-ID girlies, this one's for you 🫶
-name name — sets the name of the guest 💖. This name shows up in the SDL window caption, and it's also what the VNC server uses. Optionally it also sets the top visible process name on Linux, and on Linux you can even enable naming of individual threads to help with debugging 🧚
-uuid uuid — sets the system UUID. Simple as that, bestie ⭐

💾 Block device options 💾

Okay so this is where it gets deep — the QEMU block device handling options have a long history and have gone through several iterations as the feature set and the complexity of the block layer grew 🌱. Real talk: lots of online guides to QEMU reference older, deprecated options, and that's exactly how you end up confused 😭.

The most explicit way to describe disks is to pair -device (which specifies the hardware device) with -blockdev (which describes the backend). The device defines what the guest sees, and the backend describes how QEMU handles the data. It's the only guaranteed stable interface for describing block devices, so it's the recommended one for management tools and scripting 👑

The -drive option squishes the device and the backend into a single command line option, which is more human friendly 🥺. But there's no interface stability guarantee — although some older board models still need updating to work with the modern blockdev forms.

Older options like -hda are essentially macros that expand into -drive options for various drive interfaces. The original forms bake in a whole lot of assumptions from back when QEMU was emulating a legacy PC, so they are not recommended for modern configurations 🙈.

Legacy energy is cute in a photo dump, not in your disk config. Periodt 💅

-fda file, -fdb file — use file as floppy disk 0/1 image 🌸 (see the Disk Images <#disk-images> chapter in the System Emulation Users Guide).
-hda file, -hdb file, -hdc file, -hdd file — use file as hard disk 0, 1, 2 or 3 image on the default bus of the emulated machine 💖 (that's, for example, the IDE bus on most x86 machines, but it can also be SCSI, virtio or something else on other target architectures). See also the Disk Images <#disk-images> chapter in the System Emulation Users Guide.
-cdrom file — use file as CD-ROM image on the default bus of the emulated machine 🎀 (which is IDE1 master on x86, so you cannot use -hdc and -cdrom at the same time there — pick a lane bestie). On systems that support it, you can use the host CD-ROM by using /dev/cdrom as the filename 🌈.
-blockdev option[,option[,option[,...]]] — define a new block driver node ✨. Some of the options apply to all block drivers, other options are only accepted for a specific block driver — see below for a list of generic options and the options for the most common block drivers.

Options that expect a reference to another node (e.g. file) can be given in two ways 🫶. Either you specify the node name of an already existing node (file=node-name), or you define a new node inline by adding options for the referenced node after a dot (file.filename=path,file.aio=native).

A block driver node created with -blockdev can be used for a guest device by specifying its node name for the drive property in a -device argument that defines a block device 👑.

Valid options for any block driver node:

driver — specifies the block driver to use for the given node.
node-name — this defines the name of the block driver node by which it will be referenced later 🔖. The name must be unique, i.e. it must not match the name of a different block driver node, or (if you use -drive as well) the ID of a drive.

If no node name is specified, it is automatically generated. The generated node name is not intended to be predictable and changes between QEMU invocations 🙈. For the top level, an explicit node name must be specified.

read-only — open the node read-only. Guest write attempts will fail 💀.

Note that some block drivers support only read-only access, either generally or in certain configurations. In this case, the default value read-only=off does not work and the option must be specified explicitly.

auto-read-only — if auto-read-only=on is set, QEMU may fall back to read-only usage even when read-only=off is requested, or even switch between modes as needed 🦋 — like, depending on whether the image file is writable or whether a writing user is attached to the node.
force-share — override the image locking system of QEMU by forcing the node to utilize weaker shared access for permissions where it would normally request exclusive access 🔓. When there's the potential for multiple instances to have the same file open (whether this invocation of QEMU is the first or the second instance), both instances must permit shared access for the second instance to succeed at opening the file.

Enabling force-share=on requires read-only=on ⚠️.

cache.direct — the host page cache can be avoided with cache.direct=on. This will attempt to do disk IO directly to the guest's memory 💨. QEMU may still perform an internal copy of the data.
cache.no-flush — in case you don't care about data integrity over host failures, you can use cache.no-flush=on 😬. This option tells QEMU that it never needs to write any data to the disk but can instead keep things in cache. If anything goes wrong — like your host losing power, the disk storage getting disconnected accidentally, etc. — your image will most probably be rendered unusable 💀🔥.

Fr fr, only reach for this if you genuinely do not care about that image. It said "most probably unusable" and it meant it 😭

discard=discard — discard is one of ignore (or off) or unmap (or on) and controls whether discard (also known as trim or unmap) requests are ignored or passed to the filesystem 🧹. Some machine types may not support discard requests.
detect-zeroes=detect-zeroes — detect-zeroes is off, on or unmap and enables the automatic conversion of plain zero writes by the OS to driver specific optimized zero write commands ✨. You may even choose unmap if discard is set to unmap, to allow a zero write to be converted to an unmap operation.

🗂️ Driver-specific options for `file` 🗂️

This is the protocol-level block driver for accessing regular files 📄.

filename — the path to the image file in the local filesystem.
aio — specifies the AIO backend (threads/native/io_uring, default: threads) ⚡
locking — specifies whether the image file is protected with Linux OFD / POSIX locks 🔐. The default is to use the Linux Open File Descriptor API if available, otherwise no lock is applied. (auto/on/off, default: auto)

Example:

-blockdev driver=file,node-name=disk,filename=disk.img

🧊 Driver-specific options for `raw` 🧊

This is the image format block driver for raw images. It's usually stacked on top of a protocol level block driver such as file 🥞.

file — reference to or definition of the data source block driver node (e.g. a file driver node).

Example 1:

-blockdev driver=file,node-name=disk_file,filename=disk.img
-blockdev driver=raw,node-name=disk,file=disk_file

Example 2:

-blockdev driver=raw,node-name=disk,file.driver=file,file.filename=disk.img

🦋 Driver-specific options for `qcow2` 🦋

This is the image format block driver for qcow2 images. It's usually stacked on top of a protocol level block driver such as file 🥞.

file — reference to or definition of the data source block driver node (e.g. a file driver node).
backing — reference to or definition of the backing file block device (the default is taken from the image file) 🎀. It is allowed to pass null here in order to disable the default backing file.
lazy-refcounts — whether to enable the lazy refcounts feature (on/off; the default is taken from the image file) 😴
cache-size — the maximum total size of the L2 table and refcount block caches in bytes (default: the sum of l2-cache-size and refcount-cache-size) 💾
l2-cache-size — the biggest the L2 table cache can get, in bytes. Default is lowkey conditional: if cache-size isn't given, it's 32M on Linux platforms and 8M on non-Linux platforms; but if cache-size is given, it's as large as possible inside that cache-size while still leaving room for the requested (or minimal) refcount cache size. 💅
refcount-cache-size — the biggest the refcount block cache can get, in bytes. Default: 4 times the cluster size; or, if cache-size is set, the chunk of it that isn't being used for the L2 cache. 🎀
cache-clean-interval — sweeps unused entries out of the L2 and refcount caches, on an interval measured in seconds. Default is 600 on supporting platforms and 0 on the others. Setting it to 0 disables the whole feature, periodt. ✨
pass-discard-request — should discard requests aimed at the qcow2 device get forwarded to the data source? (on/off; default: on if discard=unmap is specified, off otherwise 🌸)
pass-discard-snapshot — should discard requests for the data source be issued when a snapshot operation (think: deleting a snapshot) frees clusters in the qcow2 file? (on/off; default: on 💖)
pass-discard-other — should discard requests for the data source be issued on other occasions where a cluster gets freed? (on/off; default: off 🦋)
discard-no-unref — when you turn this on, data clusters stay preallocated even once they're no longer used — like, because they got discarded or converted to zero clusters. Same as always, whether the old data actually gets discarded or kept at the protocol level (i.e. inside the image file) depends on how pass-discard-request is set. Keeping the clusters preallocated stops the qcow2 fragmentation that freeing and re-allocating them later would otherwise cause. Ngl though: besides the potential performance hit, that kind of fragmentation can make clusters get allocated past the end of the image file, which gives you image files whose file length grows way bigger than their guest disk size would ever suggest. 👀 So if image file length is something you care about (say, when you're storing qcow2 images straight onto block devices), you should fr consider enabling this option.
overlap-check — which overlap checks to run for writes to the image (none/constant/cached/all; default: cached). For the deets, or if you want finer granularity control, go look at the QAPI documentation of blockdev-add. 🧚

Example 1:

-blockdev driver=file,node-name=my_file,filename=/tmp/disk.qcow2
-blockdev driver=qcow2,node-name=hda,file=my_file,overlap-check=none,cache-size=16777216

Example 2:

-blockdev driver=qcow2,node-name=disk,file.driver=http,file.filename=http://example.com/image.qcow2

🌈 Driver-specific options for other drivers

Please refer to the QAPI documentation of the blockdev-add QMP command. That's it, that's the whole vibe here. 🫶

✨ `-drive` `option[,option[,option[,...]]]` ✨

Okay bestie, this one spins up a whole new drive: it creates a block driver node (the backend) and a guest device, all in one go. It's basically a shortcut so you don't have to hand-write the matching -blockdev and -device options yourself. Main character convenience energy. 👑

-drive accepts every option that -blockdev accepts. On top of those, it also knows the following:

file=file — says which disk image to use with this drive (see the Disk Images <#disk-images> chapter in the System Emulation Users Guide). Lowkey important: if the filename contains a comma, you have to double it — so file=my,,file is how you'd point at the file my,file. 💀

Special files like iSCSI devices can be given using protocol specific URLs. See the section for "Device URL Syntax" for more info.

if=interface — picks which type of interface the drive connects to. The available types are: ide, scsi, sd, mtd, floppy, pflash, virtio, none. 🌸
bus=bus,unit=unit — these two set where the drive is connected by naming the bus number and the unit id.
index=index — sets where the drive connects by using an index into the list of available connectors for a given interface type.
media=media — the type of the media: disk or cdrom. 💿
snapshot=snapshot — snapshot is on or off and controls snapshot mode for the given drive (see -snapshot).
cache=cache — cache is none, writeback, unsafe, directsync or writethrough, and it controls how the host cache is used to access block data. It's a shortcut that sets the cache.direct and cache.no-flush options (as in -blockdev), plus cache.writeback, which provides a default for the write-cache option of block guest devices (as in -device). Here's how the modes line up:
- writeback — cache.writeback on, cache.direct off, cache.no-flush off
- none — cache.writeback on, cache.direct on, cache.no-flush off
- writethrough — cache.writeback off, cache.direct off, cache.no-flush off
- directsync — cache.writeback off, cache.direct on, cache.no-flush off
- unsafe — cache.writeback on, cache.direct off, cache.no-flush on

The default mode is cache=writeback. ✨

aio=aio — aio is threads, native, or io_uring, and it picks between pthread based disk I/O, native Linux AIO, or the Linux io_uring API. 🦋
format=format — tell it which disk format to use instead of detecting it. You can use format=raw to avoid interpreting an untrusted format header — yes, that's a whole security thing, do it. 🔥
werror=action,rerror=action — which action to take on write and read errors. Valid actions are: ignore (ignore the error and try to continue), stop (pause QEMU), report (report the error to the guest), enospc (pause QEMU only if the host disk is full; report the error to the guest otherwise). The default setting is werror=enospc and rerror=report.
copy-on-read=copy-on-read — copy-on-read is on or off, and it enables whether to copy read backing file sectors into the image file. 🙈
bps=b,bps_rd=r,bps_wr=w — bandwidth throttling limits in bytes per second, either for all request types or for reads or writes only. Small values can lead to timeouts or hangs inside the guest, so heads up: a safe minimum for disks is 2 MiB/s. ⭐
bps_max=bm,bps_rd_max=rm,bps_wr_max=wm — bursts in bytes per second, either for all request types or for reads or writes only. Bursts let the guest I/O spike above the limit temporarily.
iops=i,iops_rd=r,iops_wr=w — request rate limits in requests per second, either for all request types or for reads or writes only.
iops_max=bm,iops_rd_max=rm,iops_wr_max=wm — bursts in requests per second, either for all request types or for reads or writes only. Bursts let the guest I/O spike above the limit temporarily. 💫
iops_size=is — let every is bytes of a request count as a new request for iops throttling purposes. Use this so guests can't dodge the iops limits by sending fewer but bigger requests. 💅
group=g — join a throttling quota group with the given name g. Every drive that's a member of the same group gets accounted for together. Use this to stop guests from circumventing throttling limits by using a bunch of tiny disks instead of one bigger disk. 👑

Like, okay, bestie, back to the disk cache discourse because this part is important — cache.writeback is literally giving "trust fall with your data" energy. 💾✨

By default, the cache.writeback=on mode is used. It reports data writes as finished the second the data hits the host page cache. That's totally safe as long as your guest OS is a responsible queen who actually flushes disk caches when she's supposed to. But if your guest OS does not handle volatile disk write caches correctly and your host crashes or loses power? Yeah. Guest may experience data corruption. 💀 Not the vibe.

For those chaotic guests, you should lowkey consider cache.writeback=off. This means the host page cache still gets used to read and write data, but the write notification only gets sent to the guest after QEMU has made sure to flush each write to the disk. Heads up though: this has a major impact on performance. Like, it ate, but it ate your throughput. 🐌

When you're using the -snapshot option, unsafe caching is always used. No exceptions, no negotiations. 🫶

Copy-on-read avoids accessing the same backing file sectors over and over — iconic behavior, honestly — and it's super useful when the backing file lives over a slow network. By default copy-on-read is off.

Instead of -cdrom you can use:

qemu-system-x86_64 -drive file=file,index=2,media=cdrom

Instead of -hda, -hdb, -hdc, -hdd, you can use:

qemu-system-x86_64 -drive file=file,index=0,media=disk
qemu-system-x86_64 -drive file=file,index=1,media=disk
qemu-system-x86_64 -drive file=file,index=2,media=disk
qemu-system-x86_64 -drive file=file,index=3,media=disk

You can also open an image using pre-opened file descriptors from an fd set — yes, that's as main character as it sounds:

qemu-system-x86_64 \
 -add-fd fd=3,set=2,opaque="rdwr:/path/to/file" \
 -add-fd fd=4,set=2,opaque="rdonly:/path/to/file" \
 -drive file=/dev/fdset/2,index=0,media=disk

You can connect a CDROM to the slave of ide0:

qemu-system-x86_64 -drive file=file,if=ide,index=1,media=cdrom

And if you don't specify the file= argument, congrats, you've defined an empty drive:

qemu-system-x86_64 -drive if=ide,index=1,media=cdrom

Instead of -fda, -fdb, you can use:

qemu-system-x86_64 -drive file=file,index=0,if=floppy
qemu-system-x86_64 -drive file=file,index=1,if=floppy

By default, interface is "ide" and index is automatically incremented. So this:

qemu-system-x86_64 -drive file=a -drive file=b

is interpreted like:

qemu-system-x86_64 -hda a -hdb b

🌸 `-mtdblock` file

Use file as on-board Flash memory image. Cute, tiny, nostalgic. ✨

🎀 `-sd` file

Use file as SecureDigital card image. Yes, like an SD card. She's versatile!

🦋 `-snapshot`

Write to temporary files instead of disk image files. In this case, the raw disk image you use is not written back. You can, however, force the write back by pressing C-a s (see the Disk Images <#disk-images> chapter in the System Emulation Users Guide).

⚠️ Warning: snapshot is incompatible with -blockdev (instead use qemu-img to manually create snapshot images to attach to your blockdev). If you have mixed -blockdev and -drive declarations you can use the 'snapshot' property on your drive declarations instead of this global option.

💜 `-fsdev`

-fsdev local,id=id,path=path,security_model=security_model
[,writeout=writeout][,readonly=on][,fmode=fmode][,dmode=dmode][,max_xattr=max]
[,throttling.option=value[,throttling.option=value[,...]]]

-fsdev synth,id=id[,readonly=on][,max_xattr=max]

Define a new file system device. Valid options are:

local — Accesses to the filesystem are done by QEMU.
synth — Synthetic filesystem, only used by QTests.
id=id — Specifies identifier for this device.
path=path — Specifies the export path for the file system device. Files under this path will be available to the 9p client on the guest.
security_model=security_model — Specifies the security model to be used for this export path. Supported security models are "passthrough", "mapped-xattr", "mapped-file" and "none". In "passthrough" security model, files are stored using the same credentials as they are created on the guest. This requires QEMU to run as root. In "mapped-xattr" security model, some of the file attributes like uid, gid, mode bits and link target are stored as file attributes. For "mapped-file" these attributes are stored in the hidden .virtfs_metadata directory. Directories exported by this security model cannot interact with other unix tools. "none" security model is same as passthrough except the sever won't report failures if it fails to set file attributes like ownership. Security model is mandatory only for local fsdriver. 👑
writeout=writeout — This is an optional argument. The only supported value is "immediate". This means that host page cache will be used to read and write data but write notification will be sent to the guest only when the data has been reported as written by the storage subsystem.
readonly=on — Enables exporting 9p share as a readonly mount for guests. By default read-write access is given.
fmode=fmode — Specifies the default mode for newly created files on the host. Works only with security models "mapped-xattr" and "mapped-file".
dmode=dmode — Specifies the default mode for newly created directories on the host. Works only with security models "mapped-xattr" and "mapped-file".
throttling.bps-total=b, throttling.bps-read=r, throttling.bps-write=w — Specify bandwidth throttling limits in bytes per second, either for all request types or for reads or writes only.
throttling.bps-total-max=bm, bps-read-max=rm, bps-write-max=wm — Specify bursts in bytes per second, either for all request types or for reads or writes only. Bursts allow the guest I/O to spike above the limit temporarily.
throttling.iops-total=i, throttling.iops-read=r, throttling.iops-write=w — Specify request rate limits in requests per second, either for all request types or for reads or writes only.
throttling.iops-total-max=im, throttling.iops-read-max=irm, throttling.iops-write-max=iwm — Specify bursts in requests per second, either for all request types or for reads or writes only. Bursts allow the guest I/O to spike above the limit temporarily.
throttling.iops-size=is — Let every is bytes of a request count as a new request for iops throttling purposes.
max_xattr=max — Specifies the maximum number of concurrent xattr FIDs allowed for this export. The default is 1024. Set to 0 for allowing an infinite number of xattr FIDs. This limit prevents host memory exhaustion attacks by capping the number of simultaneous xattr FIDs. 🛡️

The -fsdev option is used along with -device driver "virtio-9p-...".

🧚 `-device virtio-9p-type,fsdev=id,mount_tag=mount_tag`

Options for virtio-9p-... driver are:

type — Specifies the variant to be used. Supported values are "pci", "ccw" or "device", depending on the machine type.
fsdev=id — Specifies the id value specified along with -fsdev option.
mount_tag=mount_tag — Specifies the tag name to be used by the guest to mount this export point.

🌈 `-virtfs`

-virtfs local,path=path,mount_tag=mount_tag
,security_model=security_model[,writeout=writeout][,readonly=on]
[,fmode=fmode][,dmode=dmode][,multidevs=multidevs][,max_xattr=max]

-virtfs synth,mount_tag=mount_tag[,max_xattr=max]

Define a new virtual filesystem device and expose it to the guest using a virtio-9p-device (a.k.a. 9pfs) — which essentially means that a certain directory on host is made directly accessible by guest as a pass-through file system by using the 9P network protocol for communication between host and guests, if desired even accessible, shared by several guests simultaneously. It's giving "sharing is caring" but make it filesystem. 🫶✨

Note that -virtfs is actually just a convenience shortcut for its generalized form -fsdev -device virtio-9p-pci.

The general form of pass-through file system options are:

local — Accesses to the filesystem are done by QEMU.
synth — Synthetic filesystem, only used by QTests.
id=id — Specifies identifier for the filesystem device.
path=path — Specifies the export path for the file system device. Files under this path will be available to the 9p client on the guest.
security_model=security_model — Specifies the security model to be used for this export path. Supported security models are "passthrough", "mapped-xattr", "mapped-file" and "none". In "passthrough" security model, files are stored using the same credentials as they are created on the guest. This requires QEMU to run as root. In "mapped-xattr" security model, some of the file attributes like uid, gid, mode bits and link target are stored as file attributes. For "mapped-file" these attributes are stored in the hidden .virtfs_metadata directory. Directories exported by this security model cannot interact with other unix tools. "none" security model is same as passthrough except the sever won't report failures if it fails to set file attributes like ownership. Security model is mandatory only for local fsdriver.
writeout=writeout — This is an optional argument. The only supported value is "immediate". This means that host page cache will be used to read and write data but write notification will be sent to the guest only when the data has been reported as written by the storage subsystem.
readonly=on — Enables exporting 9p share as a readonly mount for guests. By default read-write access is given.
fmode=fmode — sets the default mode for brand-new files on the host. Ngl this one only works with the mapped-xattr and mapped-file security models, so don't come for me if it does nothing elsewhere 💅
dmode=dmode — same vibe but for brand-new directories on the host. Also mapped-xattr / mapped-file only, bestie.
mount_tag=mount_tag — the tag name the guest uses to mount this export point. That's literally it. That's the whole job 🎀
max_xattr=max — caps how many concurrent xattr FIDs this export is allowed. Default is 1024. Set it to 0 and you get an infinite number of xattr FIDs — yes, infinite, she's feral 😭. Why does this exist? It stops host memory exhaustion attacks by capping how many simultaneous xattr FIDs can be open at once. Iconic defensive queen behavior fr.
multidevs=remap|forbid|warn — tells QEMU how to handle multiple devices being shared through the same 9p export so you don't get file ID collisions on the guest. The supported behaviours are "remap" (the default), "forbid", or "warn".
- remap : assumes it's possible that more than one device is shared with the same 9p export. So inode numbers from the host get remapped for the guest in a way that prevents file ID collisions on the guest. Why is remapping required? Because the original device IDs from the host are never passed through and exposed on the guest — instead, all files of an export shared with virtfs always share the same device ID on the guest. So two files with identical inode numbers but from actually different devices on the host would otherwise cause a file ID collision, and therefore potential severe misbehaviours on the guest. Not the vibe. 🚫
- warn : virtfs 9p expects only one device to be shared with the same export. But if more than one device does get shared and accessed via the same 9p export, then all that happens is a warning message gets logged (once) by QEMU on the host side. No further action is performed in this case that would prevent file ID collisions on the guest. So this could lead to severe misbehaviours — like wrong files being accessed and data corruption on the exported tree. Yeah. Corruption. Bestie, be careful. 💀
- forbid : assumes, like "warn", that only one device is shared by the same 9p export — except it won't just log a warning, it will also deny access to additional devices on the guest. Cute, right? Note though that "forbid" does currently not block all possible file access operations (e.g. readdir() would still return entries from other devices). So it's giving security theatre a little bit 🙈.
-iscsi — configures iSCSI session parameters. That's the whole sentence, she's brief ✨

🔌 USB convenience options 🔌

-usb — enables USB emulation on machine types with an on-board USB host controller (if it's not already enabled by default). Heads up though: on-board USB host controllers may not support USB 3.0. In that case you can use -device qemu-xhci instead, on machines with PCI. Slay 💅
-usbdevice devname — adds the USB device devname, and enables an on-board USB controller if that's possible and necessary (just like it can be done via -machine usb=on). Note that this option is mainly intended for the user's convenience only — more fine-grained control can be achieved by selecting a USB host controller (if necessary) and the desired USB device via the -device option instead.

For example, instead of using -usbdevice mouse it's possible to use -device qemu-xhci -device usb-mouse to connect the USB mouse to a USB 3.0 controller instead (at least on machines that support PCI and do not have a USB controller enabled by default yet).

For more details, see the chapter about Connecting USB devices <#connecting-usb-devices> in the System Emulation Users Guide. Possible devices for devname are:

braille — Braille device. This will use BrlAPI to display the braille output on a real or fake device (i.e. it also creates a corresponding braille chardev automatically beside the usb-braille USB device). Accessibility queen 🫶
keyboard — Standard USB keyboard. Will override the PS/2 keyboard (if present). Out with the old, in with the new ✨
mouse — Virtual Mouse. This will override the PS/2 mouse emulation when activated.
tablet — Pointer device that uses absolute coordinates (like a touchscreen). This means QEMU is able to report the mouse position without having to grab the mouse. Also overrides the PS/2 mouse emulation when activated. No grabbing required, it's giving touchscreen main character energy 👑
wacom-tablet — Wacom PenPartner USB tablet.

🖥️ Display options 🖥️

-display type — selects the type of display to use. Side note: this may also influence the default audio backend selection. Lowkey connected, iykyk.

Use -display help to list the available display types. Valid values for type are:

spice-app[,gl=on|off] — starts QEMU as a Spice server and launches the default Spice client application. The Spice server will redirect the serial consoles and QEMU monitors. (Since 4.0) 🎀
dbus — exports the display over D-Bus interfaces. (Since 7.0)

If no audio backend is specified and the dbus display backend is specified, the dbus audio backend is used by default. Match made in heaven fr.

The connection is registered with the "org.qemu" name (and queued when already owned).

addr=<dbusaddr> : D-Bus bus address to connect to.
p2p=yes|no : use peer-to-peer connection, accepted via QMP add_client.
gl=on|off|core|es : use OpenGL for rendering (the D-Bus interface will share framebuffers with DMABUF file descriptors).

sdl — display video output via SDL (usually in a separate graphics window; see the SDL documentation for other possibilities). Valid parameters are:
- grab-mod=<mods> : used to select the modifier keys for toggling the mouse grabbing in conjunction with the "g" key. <mods> can be either lshift-lctrl-lalt or rctrl.
- gl=on|off|core|es : use OpenGL for displaying
- show-cursor=on|off : force showing the mouse cursor
- window-close=on|off : allow to quit qemu with window close button
gtk — display video output in a GTK window. This interface provides drop-down menus and other UI elements to configure and control the VM during runtime. Valid parameters are:
- clipboard=on|off (Enable host-guest clipboard sharing,) — defaults to "off"
- full-screen=on|off : start in fullscreen mode
- gl=on|off : use OpenGL for displaying
- grab-on-hover=on|off : grab keyboard input on mouse hover
- show-tabs=on|off (Display the tab bar for switching between the) — various graphical interfaces (e.g. VGA and virtual console character devices) by default.
- show-cursor=on|off : force showing the mouse cursor
- window-close=on|off : allow to quit qemu with window close button
- show-menubar=on|off : display the main window menubar, defaults to "on"
- zoom-to-fit=on|off (Expand video output to the window size,) — defaults to "off"
curses[,charset=<encoding>] — display video output via curses. For graphics device models which support a text mode, QEMU can display this output using a curses/ncurses interface. Nothing is displayed when the graphics device is in graphical mode or if the graphics device does not support a text mode. Generally only the VGA device models support text mode. The font charset used by the guest can be specified with the charset option, for example charset=CP850 for IBM CP850 encoding. The default is CP437. Retro aesthetic, honestly obsessed 🌸
cocoa — display video output in a Cocoa window. Mac only. This interface provides drop-down menus and other UI elements to configure and control the VM during runtime. Valid parameters are:
- full-grab=on|off (Capture all key presses, including system combos.) — this requires accessibility permissions, since it performs a global grab on key events. (default: off) See <https://support.apple.com/en-in/guide/mac-help/mh32356/mac>
- swap-opt-cmd=on|off (Swap the Option and Command keys so that their) — key codes match their position on non-Mac keyboards and you can use Meta/Super and Alt where you expect them. (default: off)
- show-cursor=on|off : force showing the mouse cursor
- left-command-key=on|off : disable forwarding left command key to host
- full-screen=on|off : start in fullscreen mode
- zoom-to-fit=on|off (Expand video output to the window size,) — defaults to "off"
egl-headless[,rendernode=<file>] — offloads all OpenGL operations to a local DRI device. For any graphical display, this display needs to be paired with either VNC or SPICE displays. Teamwork makes the dream work 🦋
vnc=<display> — start a VNC server on display <display>
none — do not display video output. The guest will still see an emulated graphics card, but its output will not be displayed to the QEMU user. This option differs from the -nographic option in that it only affects what is done with video output; -nographic also changes the destination of the serial and parallel port data. Two very different girls, don't confuse them 💅

-nographic — okay so normally, if QEMU was compiled with graphical window support, it pops your guest graphics, guest console, and the QEMU monitor into a little window 🪟. With this option you can totally kill the graphical output so QEMU is just a plain little command-line application, no window energy at all. The emulated serial port gets redirected onto the console and muxed with the monitor (unless you explicitly redirected it somewhere else, obviously 💅). Which means — slay — you can still use QEMU to debug a Linux kernel with a serial console. Hit Ctrl+a then h for help on switching between the console and the monitor. iykyk 🫶
-spice option[,option[,...]] — turns on the spice remote desktop protocol ✨. Valid options are:
- port=<nr> — set the TCP port spice is listening on for plaintext channels.
- addr=<addr> — set the IP address spice is listening on. Default is any address.
- ipv4=on|off; ipv6=on|off; unix=on|off — force using the specified IP version.
- password-secret=<secret-id> — set the ID of the secret object containing the password you need to authenticate 🔐.
- sasl=on|off — require that the client use SASL to authenticate with the spice. The exact choice of authentication method used is controlled from the system / user's SASL configuration file for the 'qemu' service. This is typically found in /etc/sasl2/qemu.conf. If running QEMU as an unprivileged user, an environment variable SASL_CONF_PATH can be used to make it search alternate locations for the service config 🗂️. While some SASL auth methods can also provide data encryption (eg GSSAPI), it is recommended that SASL always be combined with the 'tls' and 'x509' settings to enable use of SSL and server certificates. This ensures a data encryption preventing compromise of authentication credentials — bestie, do not skip that part, fr fr.
- disable-ticketing=on|off — allow client connects without authentication.
- disable-copy-paste=on|off — disable copy paste between the client and the guest.
- disable-agent-file-xfer=on|off — disable spice-vdagent based file-xfer between the client and the guest.
- tls-port=<nr> — set the TCP port spice is listening on for encrypted channels.
- x509-dir=<dir> — set the x509 file directory. Expects same filenames as -vnc $display,x509=$dir.
- x509-key-file=<file>; x509-key-password=<file>; x509-cert-file=<file>; x509-cacert-file=<file>; x509-dh-key-file=<file> — the x509 file names can also be configured individually ✨.
- tls-ciphers=<list> — specify which ciphers to use.
- tls-channel=main|display|cursor|inputs|record|playback; plaintext-channel=main|display|cursor|inputs|record|playback — force a specific channel to be used with or without TLS encryption. The options can be specified multiple times to configure multiple channels. The special name "default" can be used to set the default mode. For channels which are not explicitly forced into one mode the spice client is allowed to pick tls/plaintext as he pleases 🎀.
- image-compression=auto_glz|auto_lz|quic|glz|lz|off — configure image compression (lossless). Default is auto_glz.
- jpeg-wan-compression=auto|never|always; zlib-glz-wan-compression=auto|never|always — configure wan image compression (lossy for slow links). Default is auto.
- streaming-video=off|all|filter — configure video stream detection. Default is off.
- agent-mouse=on|off — enable/disable passing mouse events via vdagent. Default is on.
- playback-compression=on|off — enable/disable audio stream compression (using celt 0.5.1). Default is on.
- seamless-migration=on|off — enable/disable spice seamless migration. Default is off.
- video-codec=<codec> — provide the preferred codec the Spice server should use with the Gstreamer encoder 🎬. This option is only relevant when gl=on is specified. If no codec is provided, then the codec gstreamer:h264 would be used as default. And, for the case where gl=off, the default codec to be used is determined by the Spice server.
- max-refresh-rate=rate — provide the maximum refresh rate (or FPS) at which the encoding requests should be sent to the Spice server. Default would be 30.
- gl=on|off — enable/disable OpenGL context. Default is off.
- rendernode=<file> — DRM render node for OpenGL rendering. If not specified, it will pick the first available. (Since 2.9) ⭐
-vga type — pick the type of VGA card to emulate 🎨. Valid values for type are:
- cirrus — Cirrus Logic GD5446 Video card. All Windows versions starting from Windows 95 should recognize and use this graphic card. For optimal performances, use 16 bit color depth in the guest and the host OS. (This card was the default before QEMU 2.2)
- std — Standard VGA card with Bochs VBE extensions. If your guest OS supports the VESA 2.0 VBE extensions (e.g. Windows XP) and if you want to use high resolution modes (>= 1280x1024x16) then you should use this option. (This card is the default since QEMU 2.2) 👑
- vmware — VMWare SVGA-II compatible adapter. Use it if you have sufficiently recent XFree86/XOrg server or Windows guest with a driver for this card.
- qxl — QXL paravirtual graphic card. It is VGA compatible (including VESA 2.0 VBE support). Works best with qxl guest drivers installed though. Recommended choice when using the spice protocol 💅.
- tcx — (sun4m only) Sun TCX framebuffer. This is the default framebuffer for sun4m machines and offers both 8-bit and 24-bit colour depths at a fixed resolution of 1024x768.
- cg3 — (sun4m only) Sun cgthree framebuffer. This is a simple 8-bit framebuffer for sun4m machines available in both 1024x768 (OpenBIOS) and 1152x900 (OBP) resolutions aimed at people wishing to run older Solaris versions 🌸.
- virtio — Virtio VGA card.
- none — disable VGA card. (yes, that's a thing, and yes it's a mood 💀)
-full-screen — start in full screen 🖥️✨.
-g widthxheight[xdepth] — set the initial graphical resolution and depth (PPC, SPARC only).

For PPC the default is 800x600x32.

For SPARC with the TCX graphics device, the default is 1024x768x8 with the option of 1024x768x24. For cgthree, the default is 1024x768x8 with the option of 1152x900x8 for people who wish to use OBP 🦋.

-vnc display[,option[,option[,...]]] — so normally, if QEMU is compiled with graphical window support, it displays output such as guest graphics, guest console, and the QEMU monitor in a window. With this option, you can have QEMU listen on VNC display display and redirect the VGA display over the VNC session 🖼️. It is very useful to enable the usb tablet device when using this option (option -device usb-tablet). When using the VNC display, you must use the -k parameter to set the keyboard layout if you are not using en-us. Valid syntax for the display is:
- to=L — with this option, QEMU will try next available VNC displays, until the number L, if the originally defined -vnc display is not available, e.g. port 5900+display is already used by another application. By default, to=0.
- host:d — TCP connections will only be allowed from host on display d. By convention the TCP port is 5900+d. Optionally, host can be omitted in which case the server will accept connections from any host.
- unix:path — connections will be allowed over UNIX domain sockets where path is the location of a unix socket to listen for connections on.
- none — VNC is initialized but not started. The monitor change command can be used to later start the VNC server.

Following the display value there may be one or more option flags separated by commas. Valid options are:

reverse=on|off — connect to a listening VNC client via a "reverse" connection. The client is specified by the display. For reverse network connections (host:d,reverse), the d argument is a TCP port number, not a display number 🔄.
websocket=on|off — opens an additional TCP listening port dedicated to VNC Websocket connections. If a bare websocket option is given, the Websocket port is 5700+display. An alternative port can be specified with the syntax websocket=port.

If host is specified connections will only be allowed from this host. It is possible to control the websocket listen address independently, using the syntax websocket=host:port.

Websocket could be allowed over UNIX domain socket, using the syntax websocket=unix:path, where path is the location of a unix socket to listen for connections on 🌐.

If no TLS credentials are provided, the websocket connection runs in unencrypted mode. If TLS credentials are provided, the websocket connection requires encrypted client connections 🔒.

password=on|off — require that password based authentication is used for client connections.

The password must be set separately using the set_password command in the QEMU Monitor <#qemu-monitor>. The syntax to change your password is: set_password <protocol> <password> where <protocol> could be either "vnc" or "spice" 🔑.

If you wanna change the <protocol> password expiration, hit up expire_password <protocol> <expiration-time> — and that expiration time can be one of these: now, never, +seconds, or a straight-up UNIX timestamp. Like +60 makes the password expire in 60 seconds, or 1335196800 makes it expire on "Mon Apr 23 12:00:00 EDT 2012" (that's the UNIX time for that date, bestie).

You can also just throw in the keywords now or never for the expiration time if you want the <protocol> password to die immediately or literally never expire. Your call 💅

password-secret=<secret-id> — makes password-based auth mandatory for client connections, pulling the password from the secret object with that secret-id. Locked down 🔒
tls-creds=ID — hands over the ID of a set of TLS credentials that secure the VNC server. These apply to both the normal VNC server socket and the websocket socket (if you've got that enabled). Setting TLS creds makes the VNC server socket flip on the VeNCrypt auth mechanism. Oh, and those credentials should've already been made earlier with the -object tls-creds argument — no winging it ✨
tls-authz=ID — gives the ID of the QAuthZ authorization object that the client's x509 distinguished name gets validated against. This object only gets resolved at time of use, so you can literally delete and recreate it on the fly while the VNC server is running. If it's missing, default is "deny access" — no object, no entry 🙈
sasl=on|off — forces the client to authenticate with the VNC server via SASL. The exact auth method is controlled from the system / user's SASL config file for the qemu service, which normally lives at /etc/sasl2/qemu.conf. If you're running QEMU as an unprivileged user, the SASL_CONF_PATH environment variable lets you point it at alternate locations for the service config. Some SASL auth methods can also do data encryption (like GSSAPI), but the recommendation is to always pair SASL with the tls and x509 settings so SSL and server certs are in play — that guarantees data encryption, which prevents your auth credentials from getting compromised. Full details live in the VNC security <#vnc-security> section in the System Emulation Users Guide 🧚
sasl-authz=ID — gives the ID of the QAuthZ authorization object that the client's SASL username gets validated against. Same deal as above: only resolved at time of use, so deleting and recreating it live while the VNC server is active is totally fine. Missing = deny by default 🫶
acl=on|off — the legacy way to enable client authorization against the x509 distinguished name and SASL username. It creates two authz-list objects with the IDs vnc.username and vnc.x509dname, and you gotta configure the rules for those objects with the HMP ACL commands.

⚠️ This option is deprecated and should no longer be used. The new sasl-authz and tls-authz options are the replacement, bestie. Move on 💀

lossy=on|off — turns on lossy compression methods (gradient, JPEG, ...). With this set, the VNC client may get lossy framebuffer updates depending on its encoding settings. Big bandwidth savings, but at the expense of quality — pick your fighter 🔥
non-adaptive=on|off — disables adaptive encodings, which are on by default. An adaptive encoding tries to spot frequently-updated screen regions and send updates there with a lossy encoding (like JPEG) — genuinely so helpful for saving bandwidth when you're playing videos. Disabling them restores the original static behavior of encodings like Tight. No cap, it's giving "vibes vs. efficiency" 🌸
share=allow-exclusive|force-shared|ignore — sets the display sharing policy, three flavors:
- allow-exclusive lets clients ask for exclusive access. Per the rfb spec, that's implemented by dropping other connections. If you want multiple clients connected in parallel, all of them have to ask for a shared session (vncviewer: -shared switch). This is the default.
- force-shared disables exclusive client access entirely. Perfect for shared desktop sessions where you don't want somebody forgetting -shared and disconnecting literally everybody else.
- ignore completely ignores the shared flag and lets everybody connect unconditionally. Doesn't conform to the rfb spec, but it's the traditional QEMU behavior 😌
key-delay-ms — sets keyboard delay for key down and key up events, in milliseconds. Default is 10. Keyboards are low-bandwidth devices, so this slowdown helps the device and the guest keep up and not lose events when they're arriving in bulk. Possible causes of bulk arrivals: flaky network connections, or scripts for automated testing 👑
audiodev=audiodev — uses the specified audiodev when the VNC client requests audio transmission. If you're not using an -audiodev argument, this option must be omitted; otherwise it must be present and specify a valid audiodev 🎀
power-control=on|off — lets the remote client issue shutdown, reboot or reset power control requests. Giving them the aux cord AND the power button, bold ⭐

💻 i386 target only 💻

-win2k-hack — use this when you're installing Windows 2000 to dodge that disk full bug. Once Windows 2000 is installed you don't need it anymore (and heads up: this option slows down the IDE transfers). It's a synonym of -global ide-device.win2k-install-hack=on 🦋
-no-fd-bootchk — disables boot signature checking for floppy disks in BIOS. You might need this to boot from old floppy disks. Synonym of -m fd-bootchk=off — vintage tech rights 💾
-acpitable [sig=str][,rev=n][,oem_id=str][,oem_table_id=str][,oem_rev=n][,asl_compiler_id=str][,asl_compiler_rev=n][,data=file1[:file2]...] — adds an ACPI table with the specified header fields and context from the specified files. For file=, it takes the whole ACPI table from the specified files, including all the ACPI headers (which may get overridden by other options). For data=, only the data portion of the table is used — all the header info comes from the command line. And if a SLIC table is supplied to QEMU, then the SLIC's oem_id and oem_table_id fields override the same ones in the RSDT and the FADT (a.k.a. FACP), so the field matches what the Microsoft SLIC spec and the ACPI spec require 🧠
-smbios file=binary — loads the SMBIOS entry from a binary file 📁
-smbios type=0[,vendor=str][,version=str][,date=str][,release=%d.%d][,uefi=on|off] — specifies SMBIOS type 0 fields ✨
-smbios type=1[,manufacturer=str][,product=str][,version=str][,serial=str][,uuid=uuid][,sku=str][,family=str] — specifies SMBIOS type 1 fields 🎀
-smbios type=2[,manufacturer=str][,product=str][,version=str][,serial=str][,asset=str][,location=str] — specifies SMBIOS type 2 fields 🌸
-smbios type=3[,manufacturer=str][,version=str][,serial=str][,asset=str][,sku=str] — specifies SMBIOS type 3 fields 🦋
-smbios type=4[,sock_pfx=str][,manufacturer=str][,version=str][,serial=str][,asset=str][,part=str][,processor-family=%d][,processor-id=%d] — specifies SMBIOS type 4 fields 💜
-smbios type=9[,slot_designation=str][,slot_type=%d][,slot_data_bus_width=%d][,current_usage=%d][,slot_length=%d][,slot_id=%d][,slot_characteristics1=%d][,slot_characteristics12=%d][,pci_device=str] — specifies SMBIOS type 9 fields 🌈
-smbios type=11[,value=str][,path=filename] — specifies SMBIOS type 11 fields 👑

This argument can be repeated multiple times, and values get added in the order they're parsed. If your app is planning to use OEM strings data, you're encouraged to prefix the value string with your application's name — that makes it way easier to pass info for multiple applications at the same time, fr fr.

The value=str syntax provides the string data inline, while the path=filename syntax loads data from a file on disk. Note that the file is not permitted to contain any NUL bytes 🚫

Both the value and path options can be repeated multiple times and will be added to the SMBIOS table in the order in which they appear.

⚠️ Important: on the x86 architecture, the total size of all SMBIOS tables is limited to 65535 bytes. So OEM strings data is not suitable for passing large amounts of data into the guest. Instead, use it as an indicator telling the guest where to find the real data set — for example, by specifying the serial ID of a block device 😭

Here's an example passing three strings:

-smbios type=11,value=cloud-init:ds=nocloud-net;s=http://10.10.0.1:8000/,\
                value=anaconda:method=http://dl.fedoraproject.org/pub/fedora/linux/releases/25/x86_64/os,\
                path=/some/file/with/oemstringsdata.txt

In the guest OS this is visible with the dmidecode command:

$ dmidecode -t 11
Handle 0x0E00, DMI type 11, 5 bytes
OEM Strings
     String 1: cloud-init:ds=nocloud-net;s=http://10.10.0.1:8000/
     String 2: anaconda:method=http://dl.fedoraproject.org/pub/fedora/linux/releases/25/x86_64/os
     String 3: myapp:some extra data

-smbios type=17[,loc_pfx=str][,bank=str][,manufacturer=str][,serial=str][,asset=str][,part=str][,speed=%d] — specifies SMBIOS type 17 fields 🍓
-smbios type=41[,designation=str][,kind=str][,instance=%d][,pcidev=str] — specifies SMBIOS type 41 fields 💖

This argument can be repeated multiple times. Its main use is to allow network interfaces be created as enoX on Linux, with X being the instance number, instead of the name depending on the interface position on the PCI bus.

Here's what that looks like in action, bestie:

-netdev user,id=internet \
-device virtio-net-pci,mac=50:54:00:00:00:42,netdev=internet,id=internet-dev \
-smbios type=41,designation='Onboard LAN',instance=1,kind=ethernet,pcidev=internet-dev

Then in the guest OS, the device should pop up as eno1, like:

$ ip -brief l
lo               UNKNOWN        00:00:00:00:00:00 <LOOPBACK,UP,LOWER_UP>
eno1             UP             50:54:00:00:00:42 <BROADCAST,MULTICAST,UP,LOWER_UP>

Heads up 🫶 — currently, the PCI device has to be attached to the root bus.

🌐 Network options 🌐

`-nic` `[tap|passt|bridge|user|l2tpv3|vde|netmap|af-xdp|vhost-user|socket][,...][,mac=macaddr][,model=mn]`

Okay so this one is the shortcut queen 👑 — it configures both the on-board (default) guest NIC hardware and the host network backend in one single go. The host backend options are the same as with the corresponding -netdev options below. The guest NIC model gets set with model=modelname — and yes, model=help lists out all the available device types for you ✨ The hardware MAC address goes in mac=macaddr.

These two examples do exactly the same thing — they're just here to show how -nic can shorten your command line length:

qemu-system-x86_64 -netdev user,id=n1,ipv6=off -device e1000,netdev=n1,mac=52:54:98:76:54:32
qemu-system-x86_64 -nic user,ipv6=off,model=e1000,mac=52:54:98:76:54:32

`-nic none`

Says nope, no network devices should be configured at all 💅 It's used to override the default configuration (default NIC with "user" host network backend) — which is what gets activated if you don't provide any other networking options.

`-netdev passt,id=str[,option][,...]`

Configures a passt network backend, which is iconic because it requires no administrator privilege to run 🔥 Valid options are:

id=id — assign a symbolic name for use in monitor commands.
path=file — filename of the passt program to run. If it's not provided, the passt command gets started with a little help from the PATH environment variable.
quiet=on|off — by default, quiet=on, which disables informational messages from passt. quiet=on is passed as --quiet to passt.
vhost-user=on|off — by default, vhost-user=off and QEMU uses the stream network backend to communicate with passt. If vhost-user=on, passt is started with --vhost-user and QEMU uses the vhost-user network backend to communicate with passt.
@mtu — assign MTU via DHCP/NDP
address — IPv4 or IPv6 address
netmask — IPv4 mask
mac — source MAC address
gateway — IPv4 or IPv6 address as gateway
interface — interface for addresses and routes
outbound — bind to address as outbound source
outbound-if4 — bind to outbound interface for IPv4
outbound-if6 — bind to outbound interface for IPv6
dns — IPv4 or IPv6 address as DNS
search — search domains
fqdn — FQDN to configure client with
dhcp-dns — enable/disable DNS list in DHCP/DHCPv6/NDP
dhcp-search — enable/disable list in DHCP/DHCPv6/NDP
map-host-loopback — address to refer to host
map-guest-addr — addr to translate to guest's address
dns-forward — forward DNS queries sent to
dns-host — host nameserver to direct queries to
tcp — enable/disable TCP
udp — enable/disable UDP
icmp — enable/disable ICMP
dhcp — enable/disable DHCP
ndp — enable/disable NDP
dhcpv6 — enable/disable DHCPv6
ra — enable/disable route advertisements
freebind — bind to any address for forwarding
ipv4 — enable/disable IPv4
ipv6 — enable/disable IPv6
tcp-ports — TCP ports to forward
udp-ports — UDP ports to forward
param=string — string will be passed to passt as a command line parameter. You can have multiple occurrences of the param parameter to pass multiple parameters to passt. 🎀

So for instance, to pass --trace --log=trace.log:

qemu-system-x86_64 -nic passt,param=--trace,param=--log=trace.log

`-netdev user,id=id[,option][,option][,...]`

Configures the user mode host network backend — the one that needs no administrator privilege to run, we love that for us 💖 Valid options are:

id=id — assign a symbolic name for use in monitor commands.
ipv4=on|off and ipv6=on|off — specify that either IPv4 or IPv6 must be enabled. If neither is specified, both protocols are enabled.
net=addr[/mask] — set the IP network address the guest will see. Optionally specify the netmask, either in the form a.b.c.d or as a number of valid top-most bits. Default is 10.0.2.0/24.
host=addr — specify the guest-visible address of the host. Default is the 2nd IP in the guest network, i.e. x.x.x.2.
ipv6-net=addr[/int] — set the IPv6 network address the guest will see (default is fec0::/64). The network prefix is given in the usual hexadecimal IPv6 address notation. The prefix size is optional, and is given as the number of valid top-most bits (default is 64).
ipv6-host=addr — specify the guest-visible IPv6 address of the host. Default is the 2nd IPv6 in the guest network, i.e. xxxx::2.
restrict=on|off — if this option is enabled, the guest gets isolated 😭 i.e. it will not be able to contact the host and no guest IP packets will be routed over the host to the outside. This option does not affect any explicitly set forwarding rules.
hostname=name — specifies the client hostname reported by the built-in DHCP server.
dhcpstart=addr — specify the first of the 16 IPs the built-in DHCP server can assign. Default is the 15th to 31st IP in the guest network, i.e. x.x.x.15 to x.x.x.31.
dns=addr — specify the guest-visible address of the virtual nameserver. The address must be different from the host address. Default is the 3rd IP in the guest network, i.e. x.x.x.3.
ipv6-dns=addr — specify the guest-visible address of the IPv6 virtual nameserver. The address must be different from the host address. Default is the 3rd IP in the guest network, i.e. xxxx::3.
dnssearch=domain — provides an entry for the domain-search list sent by the built-in DHCP server. More than one domain suffix can be transmitted by specifying this option multiple times. If supported, this will cause the guest to automatically try to append the given domain suffix(es) in case a domain name can not be resolved. 🦋

Like, for example:

qemu-system-x86_64 -nic user,dnssearch=mgmt.example.org,dnssearch=example.org

domainname=domain — specifies the client domain name reported by the built-in DHCP server.
tftp=dir — when using the user mode network stack, activate a built-in TFTP server. The files in dir will be exposed as the root of a TFTP server. The TFTP client on the guest must be configured in binary mode (use the command bin of the Unix TFTP client). The built-in TFTP server is read-only 👀 it does not implement any command for writing files. QEMU will not write to this directory.
tftp-server-name=name — in the BOOTP reply, broadcast name as the "TFTP server name" (RFC2132 option 66). This can be used to advise the guest to load boot files or configurations from a different server than the host address.
bootfile=file — when using the user mode network stack, broadcast file as the BOOTP filename. In conjunction with tftp, this can be used to network boot a guest from a local directory.

Example (using pxelinux):

qemu-system-x86_64 -hda linux.img -boot n -device e1000,netdev=n1 \
    -netdev user,id=n1,tftp=/path/to/tftp/files,bootfile=/pxelinux.0

smb=dir[,smbserver=addr] — when using the user mode network stack, activate a built-in SMB server so that Windows OSes can access the host files in dir transparently ✨ The IP address of the SMB server can be set to addr. By default the 4th IP in the guest network is used, i.e. x.x.x.4.

In the guest Windows OS, this line:

10.0.2.4 smbserver #PRE #NOFNR

must be added in the LMHOSTS file. In this line, #PRE requests pre-caching of the mapping, which speeds up the initial connection on some Windows versions. #NOFNR is necessary for Windows NT 3.1 to tell it not to send NBNS query packets that QEMU does not handle, and is harmlessly ignored on other versions. 🎀

The LMHOSTS file may be in different locations depending on the Windows version:

C:\WINDOWS\LMHOSTS for Windows 3x/9x/Me
C:\WINNT\SYSTEM32\DRIVERS\ETC\LMHOSTS for Windows NT/2000
C:\WINDOWS\SYSTEM32\DRIVERS\ETC\LMHOSTS for Windows XP and newer

Then dir can be accessed in \\smbserver\qemu.

Note that a SAMBA server must be installed on the host OS. 💅

🌈 `hostfwd=[tcp|udp|unix]:[[hostaddr]:hostport|hostpath]-[guestaddr]:guestport`

Okay so this one is literally the ✨magic portal✨ option — it takes incoming TCP, UDP or UNIX connections that show up on the host port hostport and yeets them straight to the guest IP guestaddr on guest port guestport 💫

If you don't say what guestaddr is, it just chills at x.x.x.15 — which is the default first address the built-in DHCP server hands out. Periodt.
Wanna lock the rule to one specific host interface? Say hostaddr out loud and it's done 🫶
Don't set a connection type? TCP. That's the default, no cap.
This option can be given multiple times — stack 'em up, queen 👑

💅 For example, to redirect a host X11 connection from screen 1 over to guest screen 0, you do this:

# on the host
qemu-system-x86_64 -nic user,hostfwd=tcp:127.0.0.1:6001-:6000
# this host xterm should open in the guest X11 server
xterm -display :1

Want to redirect telnet connections from host port 5555 to the telnet port on the guest? Say less 🌸

# on the host
qemu-system-x86_64 -nic user,hostfwd=tcp::5555-:23
telnet localhost 5555

Then when you run telnet localhost 5555 on the host, boom — you're connected to the guest telnet server 🔥

And to redirect host unix socket /tmp/vm to guest tcp socket 23, use the following:

# on the host
qemu-system-x86_64 -nic user,hostfwd=unix:/tmp/vm-:23

🌀 `guestfwd=[tcp]:server:port-dev`<br>`guestfwd=[tcp]:server:port-cmd:command`

Now we're going the other way — guest TCP connections get forwarded to IP address server on port port, either to the character device dev or to a program run by cmd:command which gets spawned for each connection 😤 This option can be given multiple times too.

Two vibes, two ways to do it:

Option A — use a chardev directly and have that one hang around for QEMU's whole lifetime, like in the following example:

# open 10.10.1.1:4321 on bootup, connect 10.0.2.100:1234 to it whenever
# the guest accesses it
qemu-system-x86_64 -nic user,guestfwd=tcp:10.0.2.100:1234-<tcp:10.10.1.1:4321>

Option B — execute a command on every single TCP connection the guest establishes, so QEMU basically behaves like an inetd process for that virtual server 🤖

# call "netcat 10.10.1.1 4321" on every TCP connection to 10.0.2.100:1234
# and connect the TCP stream to its stdin/stdout
qemu-system-x86_64 -nic  'user,id=n1,guestfwd=tcp:10.0.2.100:1234-cmd:netcat 10.10.1.1 4321'

🧚 `-netdev tap,id=id[,fd=h][,ifname=name][,script=file][,downscript=dfile][,br=bridge][,helper=helper]`

Configure a host TAP network backend with ID id. Slay, networking time 💖

Use the network script file to configure it, and the network script dfile to deconfigure it. If name isn't provided, the OS just automatically gives you one — generous king 👑 The default network configure script is /etc/qemu-ifup and the default network deconfigure script is /etc/qemu-ifdown. Wanna skip script execution entirely? Use script=no or downscript=no 💅

If you're running QEMU as an unprivileged user, use the network helper to configure the TAP interface and attach it to the bridge. The default network helper executable is /path/to/qemu-bridge-helper and the default bridge device is br0 🫶

fd=h can be used to specify the handle of an already opened host TAP interface.

Examples:

#launch a QEMU instance with the default network script
qemu-system-x86_64 linux.img -nic tap

#launch a QEMU instance with two NICs, each one connected
#to a TAP device
qemu-system-x86_64 linux.img \
        -netdev tap,id=nd0,ifname=tap0 -device e1000,netdev=nd0 \
        -netdev tap,id=nd1,ifname=tap1 -device rtl8139,netdev=nd1

#launch a QEMU instance with the default network helper to
#connect a TAP device to bridge br0
qemu-system-x86_64 linux.img -device virtio-net-pci,netdev=n1 \
        -netdev tap,id=n1,"helper=/path/to/qemu-bridge-helper"

🌉 `-netdev bridge,id=id[,br=bridge][,helper=helper]`

Connect a host TAP network interface to a host bridge device. Bridging the vibes, fr fr 🌈

Use the network helper helper to configure the TAP interface and attach it to the bridge. The default network helper executable is /path/to/qemu-bridge-helper and the default bridge device is br0.

Examples:

#launch a QEMU instance with the default network helper to
#connect a TAP device to bridge br0
qemu-system-x86_64 linux.img -netdev bridge,id=n1 -device virtio-net,netdev=n1

#launch a QEMU instance with the default network helper to
#connect a TAP device to bridge qemubr0
qemu-system-x86_64 linux.img -netdev bridge,br=qemubr0,id=n1 -device virtio-net,netdev=n1

🔌 `-netdev socket,id=id[,fd=h][,listen=[host]:port][,connect=host:port]`

This host network backend lets you connect the guest's network to another QEMU virtual machine using a TCP socket connection 🦋 Two VMs, one brain cell, one socket.

If listen is specified, QEMU waits for incoming connections on port (host is optional).
connect is how you connect to another QEMU instance that's using the listen option.
fd=h specifies an already opened TCP socket.

Example:

# launch a first QEMU instance
qemu-system-x86_64 linux.img \
                 -device e1000,netdev=n1,mac=52:54:00:12:34:56 \
                 -netdev socket,id=n1,listen=:1234
# connect the network of this instance to the network of the first instance
qemu-system-x86_64 linux.img \
                 -device e1000,netdev=n2,mac=52:54:00:12:34:57 \
                 -netdev socket,id=n2,connect=127.0.0.1:1234

📡 `-netdev socket,id=id[,fd=h][,mcast=maddr:port[,localaddr=addr]]`

Configure a socket host network backend to share the guest's network traffic with other QEMU virtual machines using a UDP multicast socket — effectively making a bus for every QEMU with the same multicast address maddr and port 🚌✨

NOTES, bestie, read these 👀:

Several QEMU instances can be running on different hosts and share the same bus (assuming correct multicast setup for these hosts).
mcast support is compatible with User Mode Linux (argument ethN=mcast), see <http://user-mode-linux.sf.net>.
Use fd=h to specify an already opened UDP multicast socket.

Example:

# launch one QEMU instance
qemu-system-x86_64 linux.img \
                 -device e1000,netdev=n1,mac=52:54:00:12:34:56 \
                 -netdev socket,id=n1,mcast=230.0.0.1:1234
# launch another QEMU instance on same "bus"
qemu-system-x86_64 linux.img \
                 -device e1000,netdev=n2,mac=52:54:00:12:34:57 \
                 -netdev socket,id=n2,mcast=230.0.0.1:1234
# launch yet another QEMU instance on same "bus"
qemu-system-x86_64 linux.img \
                 -device e1000,netdev=n3,mac=52:54:00:12:34:58 \
                 -netdev socket,id=n3,mcast=230.0.0.1:1234

Example (User Mode Linux compat.):

# launch QEMU instance (note mcast address selected is UML's default)
qemu-system-x86_64 linux.img \
                 -device e1000,netdev=n1,mac=52:54:00:12:34:56 \
                 -netdev socket,id=n1,mcast=239.192.168.1:1102
# launch UML
/path/to/linux ubd0=/path/to/root_fs eth0=mcast

Example (send packets from host's 1.2.3.4):

qemu-system-x86_64 linux.img \
                 -device e1000,netdev=n1,mac=52:54:00:12:34:56 \
                 -netdev socket,id=n1,mcast=239.192.168.1:1102,localaddr=1.2.3.4

🌊 `-netdev stream,id=str[,server=on|off],addr.type=inet,addr.host=host,addr.port=port[,to=maxport][,numeric=on|off][,keep-alive=on|off][,mptcp=on|off][,addr.ipv4=on|off][,addr.ipv6=on|off][,reconnect-ms=milliseconds]`

Configure a network backend to connect to another QEMU virtual machine or a proxy using a TCP/IP socket 🔥

server=on|off — if on, create a server socket.
addr.host=host, addr.port=port — the socket address to listen on (server=on) or connect to (server=off).
to=maxport — if present, this is a range of possible addresses, with port between port and maxport.
numeric=on|off — if on, host and port are guaranteed to be numeric; otherwise a name resolution should be attempted (default: off).
keep-alive=on|off — enable keep-alive when connecting to this socket. Not supported for passive sockets (sorry bestie 😭).
mptcp=on|off — enable multipath TCP.
ipv4=on|off — whether to accept IPv4 addresses; default is to try both IPv4 and IPv6.
ipv6=on|off — whether to accept IPv6 addresses; default is to try both IPv4 and IPv6.
reconnect-ms=milliseconds — for a client socket, if the socket gets disconnected, then attempt a reconnect after the given number of milliseconds. Setting this to zero disables this function. (default: 0)

Example (two guests connected using a TCP/IP socket):

# first VM
qemu-system-x86_64 linux.img \
              -device virtio-net,netdev=net0,mac=52:54:00:12:34:56 \
              -netdev stream,id=net0,server=on,addr.type=inet,addr.host=localhost,addr.port=1234
# second VM
qemu-system-x86_64 linux.img \
              -device virtio-net,netdev=net0,mac=52:54:00:12:34:57 \
              -netdev stream,id=net0,server=off,addr.type=inet,addr.host=localhost,addr.port=1234,reconnect-ms=5000

🐚 `-netdev stream,id=str[,server=on|off],addr.type=unix,addr.path=path[,abstract=on|off][,tight=on|off][,reconnect-ms=milliseconds]`

Configure a network backend to connect to another QEMU virtual machine or a proxy using a stream oriented unix domain socket 🍓 Same energy, different address family.

server=on|off — if on, create a server socket.
addr.path=path — the filesystem path to use.
abstract=on|off — if on, this is a Linux abstract socket address.

and that's where this part cuts off — the rest of the unix socket options are coming right up 💜

💅 `tight=on|off`

ok so if this is false, QEMU pads an abstract socket address with enough null bytes so it fills the struct sockaddr_un member sun_path. that's it, that's the whole vibe.

🔁 `reconnect-ms=milliseconds`

for a client socket — if the socket gets disconnected, it'll try to reconnect after however many milliseconds you said. set it to zero and this whole function is off. (default: 0)

🫶 example time, this one's using passt as a glow-up replacement for -netdev user:

# start passt server as a non privileged user
passt
UNIX domain socket bound at /tmp/passt_1.socket
# start QEMU to connect to passt
qemu-system-x86_64 linux.img \
              -device virtio-net,netdev=net0 \
              -netdev stream,id=net0,server=off,addr.type=unix,addr.path=/tmp/passt_1.socket

🦋 and here's two guests being besties over a stream oriented unix domain socket:

# first VM
qemu-system-x86_64 linux.img \
              -device virtio-net,netdev=net0,mac=52:54:00:12:34:56 \
              netdev stream,id=net0,server=on,addr.type=unix,addr.path=/tmp/qemu0
# second VM
qemu-system-x86_64 linux.img \
              -device virtio-net,netdev=net0,mac=52:54:00:12:34:57 \
              -netdev stream,id=net0,server=off,addr.type=unix,addr.path=/tmp/qemu0,reconnect-ms=5000

🌈 `-netdev stream,id=str[,server=on|off],addr.type=fd,addr.str=file-descriptor[,reconnect-ms=milliseconds]`

sets up a network backend that connects to another QEMU virtual machine or a proxy using a stream oriented socket file descriptor. like, a socket but make it a filedesc, ykwim?

server=on|off — on means create a server socket ✨
addr.str=file-descriptor — the file descriptor number to use as a socket
reconnect-ms=milliseconds — client socket behaviour: if it gets disconnected, attempt a reconnect after this many milliseconds. zero disables it. (default: 0)

📡 `-netdev dgram,id=str,remote.type=inet,remote.host=maddr,remote.port=port[,local.type=inet,local.host=addr]`

network backend that connects to a multicast address. very "everyone's invited to the group chat" of it.

remote.host=maddr,remote.port=port — the multicast address 💖
local.host=addr — specify the host address to send packets from

example, bestie:

# launch one QEMU instance
qemu-system-x86_64 linux.img \
              -device virtio-net,netdev=net0,mac=52:54:00:12:34:56 \
              -netdev dgram,id=net0,remote.type=inet,remote.host=224.0.0.1,remote.port=1234
# launch another QEMU instance on same "bus"
qemu-system-x86_64 linux.img \
              -device virtio-net,netdev=net0,mac=52:54:00:12:34:57 \
              -netdev dgram,id=net0,remote.type=inet,remote.host=224.0.0.1,remote.port=1234
# launch yet another QEMU instance on same "bus"
qemu-system-x86_64 linux.img \
              -device virtio-net,netdev=net0,mac=52:54:00:12:34:58 \
              -netdev dgram,id=net0,remote.type=inet,remote.host=224.0.0.1,remote.port=1234

🧚 `-netdev dgram,id=str,remote.type=inet,remote.host=maddr,remote.port=port[,local.type=fd,local.str=file-descriptor]`

same multicast address energy, but connecting using a UDP socket file descriptor.

remote.host=maddr,remote.port=port — the multicast address
local.str=file-descriptor — file descriptor to use to send packets

🍓 `-netdev dgram,id=str,local.type=inet,local.host=addr,local.port=port[,remote.type=inet,remote.host=addr,remote.port=port]`

ok this one connects to another QEMU virtual machine or a proxy using a datagram oriented unix domain socket.

local.host=addr,local.port=port — IP address to use to send the packets from
remote.host=addr,remote.port=port — destination IP address

example — two guests, UDP/IP socket, iconic:

# first VM
qemu-system-x86_64 linux.img \
              -device virtio-net,netdev=net0,mac=52:54:00:12:34:56 \
              -netdev dgram,id=net0,local.type=inet,local.host=localhost,local.port=1234,remote.type=inet,remote.host=localhost,remote.port=1235
# second VM
qemu-system-x86_64 linux.img \
              -device virtio-net,netdev=net0,mac=52:54:00:12:34:56 \
              -netdev dgram,id=net0,local.type=inet,local.host=localhost,local.port=1235,remote.type=inet,remote.host=localhost,remote.port=1234

(and yes, both MACs really are 52:54:00:12:34:56 in that example — i'm not fixing it, i'm just reporting it 💀)

🎀 `-netdev dgram,id=str,local.type=unix,local.path=path[,remote.type=unix,remote.path=path]`

another datagram oriented unix socket setup, connecting two QEMU virtual machines or a proxy. same deal, different address family.

local.path=path — filesystem path to use to bind the socket
remote.path=path — filesystem path to use as a destination (see sendto(2))

example — UDP/UNIX socket, two guests:

# first VM
qemu-system-x86_64 linux.img \
              -device virtio-net,netdev=net0,mac=52:54:00:12:34:56 \
              -netdev dgram,id=net0,local.type=unix,local.path=/tmp/qemu0,remote.type=unix,remote.path=/tmp/qemu1
# second VM
qemu-system-x86_64 linux.img \
              -device virtio-net,netdev=net0,mac=52:54:00:12:34:57 \
              -netdev dgram,id=net0,local.type=unix,local.path=/tmp/qemu1,remote.type=unix,remote.path=/tmp/qemu0

🦋 `-netdev dgram,id=str,local.type=fd,local.str=file-descriptor`

datagram oriented socket file descriptor version — connects to another QEMU virtual machine or a proxy.

local.str=file-descriptor — file descriptor to use to send packets

👑 `-netdev l2tpv3,id=id,src=srcaddr,dst=dstaddr[,srcport=srcport][,dstport=dstport],txsession=txsession[,rxsession=rxsession][,ipv6=on|off][,udp=on|off][,cookie64=on|off][,counter=on|off][,pincounter=on|off][,txcookie=txcookie][,rxcookie=rxcookie][,offset=offset]`

okay this one is kinda the main character ngl. it configures an L2TPv3 pseudowire host network backend. L2TPv3 (RFC3931) is a super popular protocol for transporting Ethernet (and other Layer 2) data frames between two systems. it lives in routers, firewalls, and the Linux kernel (from version 3.3 onwards). so like, it's everywhere already.

this transport lets a VM talk to another VM, router, or firewall directly. no middleman fr fr.

src=srcaddr — source address (mandatory ⚠️)
dst=dstaddr — destination address (mandatory ⚠️)
udp=on — select udp encapsulation (default is ip).
srcport=srcport — source udp port.
dstport=dstport — destination udp port.
ipv6=on — force v6, otherwise defaults to v4.
rxcookie=rxcookie; txcookie=txcookie — cookies are a weak form of security in the l2tpv3 spec. their whole job is mostly to prevent misconfiguration. by default they're 32 bit.
cookie64=on — set cookie size to 64 bit instead of the default 32
counter=off — force a 'cut-down' L2TPv3 with no counter, as in draft-mkonstan-l2tpext-keyed-ipv6-tunnel-00
pincounter=on — work around broken counter handling in the peer. this may also help on networks which have packet reorder.
offset=offset — add an extra offset between header and data

for example, to attach a VM running on host 4.3.2.1 via L2TPv3 to the bridge br-lan on the remote Linux host 1.2.3.4:

# Setup tunnel on linux host using raw ip as encapsulation
# on 1.2.3.4
ip l2tp add tunnel remote 4.3.2.1 local 1.2.3.4 tunnel_id 1 peer_tunnel_id 1 \
    encap udp udp_sport 16384 udp_dport 16384
ip l2tp add session tunnel_id 1 name vmtunnel0 session_id \
    0xFFFFFFFF peer_session_id 0xFFFFFFFF
ifconfig vmtunnel0 mtu 1500
ifconfig vmtunnel0 up
brctl addif br-lan vmtunnel0

# on 4.3.2.1
# launch QEMU instance - if your network has reorder or is very lossy add ,pincounter

qemu-system-x86_64 linux.img -device e1000,netdev=n1 \
    -netdev l2tpv3,id=n1,src=4.2.3.1,dst=1.2.3.4,udp=on,srcport=16384,dstport=16384,rxsession=0xffffffff,txsession=0xffffffff,counter=on

🌸 `-netdev vde,id=id[,sock=socketpath][,port=n][,group=groupname][,mode=octalmode]`

sets up a VDE backend to connect to PORT n of a vde switch running on the host and listening for incoming connections on socketpath. use GROUP groupname and MODE octalmode to change the default ownership and permissions for the communication port.

⚠️ this option is only available if QEMU has been compiled with vde support enabled. no vde compile, no vde vibes, sorry bestie 😭

example:

# launch vde switch
vde_switch -F -sock /tmp/myswitch
# launch QEMU instance
qemu-system-x86_64 linux.img -nic vde,sock=/tmp/myswitch

⭐ `-netdev af-xdp,id=str,ifname=name[,mode=native|skb][,force-copy=on|off][,queues=n][,start-queue=m][,inhibit=on|off][,sock-fds=x:y:...:z][,map-path=/path/to/socket/map][,map-start-index=i]`

configures an AF_XDP backend to connect to a network interface name using an AF_XDP socket. a specific program attach mode for a default XDP program can be forced with mode, and it defaults to best-effort, where the likely most performant mode will be in use. the number of queues n should generally match the number of queues in the interface, and defaults to 1.

💀 heads up: traffic arriving on non-configured device queues will not be delivered to the network backend. so configure accordingly, okay?

# set number of queues to 4
ethtool -L eth0 combined 4
# launch QEMU instance
qemu-system-x86_64 linux.img -device virtio-net-pci,netdev=n1 \
    -netdev af-xdp,id=n1,ifname=eth0,queues=4

...and honestly the start-queue option is right there for you if you need a specific range of queues [m, m + n] in use. Like, this is lowkey necessary sometimes to use certain NICs in native mode, no cap. The kernel lets the driver create a separate set of XDP queues on top of the regular ones, and only those queues can be used for AF_XDP sockets. NICs that work this way may also require an additional traffic redirection with ethtool to these special queues. It's giving extra homework but bestie we do it:

# set number of queues to 1
ethtool -L eth0 combined 1
# redirect all the traffic to the second queue (id: 1)
# note: drivers may require non-empty key/mask pair.
ethtool -N eth0 flow-type ether \
    dst 00:00:00:00:00:00 m FF:FF:FF:FF:FF:FE action 1
ethtool -N eth0 flow-type ether \
    dst 00:00:00:00:00:01 m FF:FF:FF:FF:FF:FE action 1
# launch QEMU instance
qemu-system-x86_64 linux.img -device virtio-net-pci,netdev=n1 \
    -netdev af-xdp,id=n1,ifname=eth0,queues=1,start-queue=1

Alsooo the XDP program can be loaded externally too 🫶. In that case the inhibit option should be set to on. Either sock-fds or map-path can be used with inhibit enabled. sock-fds can be provided with file descriptors for already open but not bound XDP sockets already added to a socket map for corresponding queues. One socket per queue, periodt:

qemu-system-x86_64 linux.img -device virtio-net-pci,netdev=n1 \
    -netdev af-xdp,id=n1,ifname=eth0,queues=3,inhibit=on,sock-fds=15:16:17

For the inhibit option set to on used together with map-path, it is expected that the XDP program with the socket map is already loaded on the networking device and the map pinned into BPF file system. The path to the pinned map is then passed to QEMU, which then creates the file descriptors and inserts them into the existing socket map. Iconic behaviour, honestly:

qemu-system-x86_64 linux.img -device virtio-net-pci,netdev=n1 \
    -netdev af-xdp,id=n1,ifname=eth0,queues=2,inhibit=on,map-path=/sys/fs/bpf/xsks_map

Additionally, map-start-index can be used to specify the start offset for insertion into the socket map. And hear me out: the combination of map-path and sock-fds together is not supported 🙅‍♀️. Pick one, babe.

-netdev vhost-user,chardev=id[,vhostforce=on|off][,queues=n] — establish a vhost-user netdev, backed by a chardev id. The chardev should be a unix domain socket backed one. The vhost-user uses a specifically defined protocol to pass vhost ioctl replacement messages to an application on the other end of the socket. On non-MSIX guests, the feature can be forced with vhostforce. Use queues=n to specify the number of queues to be created for multiqueue vhost-user. Slay.

Example:

qemu -m 512 -object memory-backend-file,id=mem,size=512M,mem-path=/hugetlbfs,share=on \
     -numa node,memdev=mem \
     -chardev socket,id=chr0,path=/path/to/socket \
     -netdev type=vhost-user,id=net0,chardev=chr0 \
     -device virtio-net-pci,netdev=net0

-netdev vhost-vdpa[,vhostdev=/path/to/dev][,vhostfd=h] — establish a vhost-vdpa netdev. A vDPA device is a device that uses a datapath which complies with the virtio specifications with a vendor specific control path. vDPA devices can be both physically located on the hardware or emulated by software. Either way, main character energy 💅.
-netdev hubport,id=id,hubid=hubid[,netdev=nd] — create a hub port on the emulated hub with ID hubid. The hubport netdev lets you connect a NIC to a QEMU emulated hub instead of a single netdev. Alternatively, you can also connect the hubport to another netdev with ID nd by using the netdev=nd option. Networking but make it a group chat 🫶.
-net nic[,netdev=nd][,macaddr=mac][,model=type][,name=name][,addr=addr][,vectors=v] — the legacy option to configure or create an on-board (or machine default) Network Interface Card (NIC) and connect it either to the emulated hub with ID 0 (i.e. the default hub), or to the netdev nd. If model is omitted, then the default NIC model associated with the machine type is used. Note that the default NIC model may change in future QEMU releases, so it is highly recommended to always specify a model (fr fr, don't leave it up to fate 🙈). Optionally, the MAC address can be changed to mac, the device address set to addr (PCI cards only), and a name can be assigned for use in monitor commands. Optionally, for PCI cards, you can specify the number v of MSI-X vectors that the card should have; this option currently only affects virtio cards; set v = 0 to disable MSI-X. If no -net option is specified, a single NIC is created. QEMU can emulate several different models of network card. Use -net nic,model=help for a list of available devices for your target. Chef's kiss.
-net user|passt|tap|bridge|socket|l2tpv3|vde[,...][,name=name] — configure a host network backend (with the options corresponding to the same -netdev option) and connect it to the emulated hub 0 (the default hub). Use name to specify the name of the hub port. Choose your fighter 💖.

🎀 Character device options 🎀

Okay so the general form of a character device option is:

-chardev backend,id=id[,mux=on|off][,options]

backend is one of: null, socket, udp, msmouse, hub, vc, ringbuf, file, pipe, console, serial, pty, stdio, braille, parallel, spicevmc, spiceport, dbus. The specific backend will determine the applicable options. That's right, pick your fave from the roster 🌈.

Use -chardev help to print all available chardev backend types. Handy, use it, love it.

All devices must have an id, which can be any string up to 127 characters long. It is used to uniquely identify this device in other command line directives. No anonymous chardevs, bestie, everyone gets a name 🎀.

A character device may be used in multiplexing mode by multiple front-ends. Specify mux=on to enable this mode. A multiplexer is a "1:N" device, and here the "1" end is your specified chardev backend, and the "N" end is the various parts of QEMU that can talk to a chardev. If you create a chardev with id=myid and mux=on, QEMU will create a multiplexer with your specified ID, and you can then configure multiple front ends to use that chardev ID for their input/output. Up to four different front ends can be connected to a single multiplexed chardev. (Without multiplexing enabled, a chardev can only be used by a single front end — no sharing, no slay.) For instance you could use this to allow a single stdio chardev to be used by two serial ports and the QEMU monitor:

-chardev stdio,mux=on,id=char0 \
-object monitor-hmp,id=hmp0,chardev=char0 \
-serial chardev:char0 \
-serial chardev:char0

You can have more than one multiplexer in a system configuration, because apparently we're extra like that ✨. For instance you could have a TCP port multiplexed between UART 0 and UART 1, and stdio multiplexed between the QEMU monitor and a parallel port:

-chardev stdio,mux=on,id=char0 \
-object monitor-hmp,id=hmp0,chardev=char0 \
-parallel chardev:char0 \
-chardev tcp,...,mux=on,id=char1 \
-serial chardev:char1 \
-serial chardev:char1

When you're using a multiplexed character device, some escape sequences are interpreted in the input. See the chapter about Keys in the character backend multiplexer <#keys-in-the-character-backend-multiplexer> in the System Emulation Users Guide for more details. Go read it, iykyk 📖.

Note that some other command line options may implicitly create multiplexed character backends; for instance -serial mon:stdio creates a multiplexed stdio backend connected to the serial port and the QEMU monitor, and -nographic also multiplexes the console and the monitor to stdio. Sneaky, but valid 💀.

If you need to aggregate data in the opposite direction (where one QEMU frontend interface receives input and output from multiple backend chardev devices), please refer to the paragraph below regarding chardev hub aggregator device configuration. Different vibe, different paragraph 🦋.

Every backend supports the logfile option, which supplies the path to a file to record all data transmitted via the backend. The logappend option controls whether the log file will be truncated or appended to when opened. Receipts for everything, we love to see it 📝.

💜 The available backends are:

-chardev null,id=id — a void device. This device will not emit any data, and will drop any data it receives. The null backend does not take any options. Truly the "seen, no reply" of chardevs 🙈.
-chardev socket,id=id[,TCP options or unix options][,server=on|off][,wait=on|off][,telnet=on|off][,websocket=on|off][,reconnect-ms=milliseconds][,tls-creds=id][,tls-authz=id] — create a two-way stream socket, which can be either a TCP or a unix socket. A unix socket will be created if path is specified. Behaviour is undefined if TCP options are specified for a unix socket ⚠️ (so like, don't).
- server=on|off specifies that the socket shall be a listening socket.
- wait=on|off specifies that QEMU should not block waiting for a client to connect to a listening socket.
- telnet=on|off specifies that traffic on the socket should interpret telnet escape sequences.
- websocket=on|off specifies that the socket uses WebSocket protocol for communication.
- reconnect-ms sets the timeout for reconnecting on non-server sockets when the remote end goes away. qemu will delay this many milliseconds and then attempt to reconnect. Zero disables reconnecting, and is the default.
- tls-creds requests enablement of the TLS protocol for encryption, and specifies the id of the TLS credentials to use for the handshake. The credentials must be previously created with the -object tls-creds argument.
- tls-auth provides the ID of the QAuthZ authorization object against which the client's x509 distinguished name will be validated. This object is only resolved at time of use, so can be deleted and recreated on the fly while the chardev server is active. If missing, it will default to denying access. Locked door, no key, we respect it 💅.

TCP and unix socket options are given below:

-chardev msmouse,id=id — forwards QEMU's emulated msmouse events straight to the guest 🐭. msmouse takes, like, zero options. Simple. Iconic.
-chardev hub,id=id,chardevs.0=id[,chardevs.N=id] — makes an explicit chardev backend hub device ✨. And bestie, this one's actually so cool: it aggregates input from multiple backend devices and forwards it all to a single frontend device. But wait, there's more — the hub device also takes the output from the frontend device and sends it back out to all the connected backend devices. So everything talks to everything. Seamless interaction between different backend devices and one single frontend interface 🫶. Aggregation is supported for up to 4 chardev devices. (Since 10.0)

For example, here's a use case with 2 backend devices: a virtual console vc0 and a pseudo TTY pty0, connected to a single virtio hvc console frontend device with a hub hub0 help. The virtual console renders text to an image, which can be shared over the VNC protocol. In turn, the pty backend provides bidirectional communication to the virtio hvc console over the pseudo TTY file. The example configuration can be as follows:

-chardev pty,path=/tmp/pty,id=pty0 \
-chardev vc,id=vc0 \
-chardev hub,id=hub0,chardevs.0=pty0,chardevs.1=vc0 \
-device virtconsole,chardev=hub0 \
-vnc 0.0.0.0:0

Once QEMU starts, a VNC client and any TTY emulator can be used to control a single hvc console:

# Start TTY emulator
tio /tmp/pty

# Start VNC client and switch to virtual console Ctrl-Alt-2
vncviewer :0

Heads up 🚨: several frontend devices is not supported. Stacking of multiplexers and hub devices is not supported either. Sad but true.

-chardev vc,id=id[[,width=width][,height=height]][[,cols=cols][,rows=rows]][,encoding=ENCODING] — connects to a QEMU text console 💻. The implementation and supported feature set depend on the selected display backend, bestie:
- The GTK backend uses libvte for the emulation and display (when available).
- The D-Bus backend exports the character device as a Chardev object.
- The spice-app backend exports it as a Spice port.

In other cases, QEMU uses its own emulated VT100, and vc may optionally be given a specific size.

width and height specify the width and height respectively of the console, in pixels.

cols and rows specify that the console be sized to fit a text console with the given dimensions.

encoding specifies the character set expected from the guest: utf8 or cp437 (8-bit Extended ASCII / VGA).

-chardev ringbuf,id=id[,size=size] — creates a ring buffer with fixed size size 🎀. size must be a power of two and defaults to 64K.
-chardev file,id=id,path=path[,input-path=input-path] — logs all traffic received from the guest to a file 📝.

path specifies the path of the file to be opened. This file will be created if it does not already exist, and overwritten if it does. path is required.

If input-path is specified, this is the path of a second file which will be used for input. If input-path is not specified, no input will be available from the chardev.

Note that input-path is not supported on Windows hosts 🙈.

-chardev pipe,id=id,path=path — creates a two-way connection to the guest 🥤. The behaviour differs slightly between Windows hosts and other hosts:

On Windows, a single duplex pipe will be created at \\.pipe\path.

On other hosts, 2 pipes will be created called path.in and path.out. Data written to path.in will be received by the guest. Data written by the guest can be read from path.out. QEMU will not create these fifos, and requires them to be present.

path forms part of the pipe path as described above. path is required.

-chardev console,id=id — sends traffic from the guest to QEMU's standard output 📤. console takes no options at all.

console is only available on Windows hosts.

-chardev serial,id=id,path=path — sends traffic from the guest to a serial device on the host 🔌.

On Unix hosts, serial will actually accept any tty device, not only serial lines. Slay.

path specifies the name of the serial device to open.

-chardev pty,id=id[,path=path] — creates a new pseudo-terminal on the host and connects to it 🖥️.

pty is not available on Windows hosts.

If path is specified, QEMU will create a symbolic link at that location which points to the new PTY device.

This avoids having to make QMP or HMP monitor queries to find out what the new PTY device path is. Genuinely so convenient.

Okay this part matters: QEMU will remove the symlink when it exits gracefully, but it will not do so in case of crashes or on certain startup errors 💀. It is recommended that the user checks and removes the symlink after QEMU terminates to account for this.

-chardev stdio,id=id[,signal=on|off] — connects to standard input and standard output of the QEMU process ⌨️.

signal controls if signals are enabled on the terminal, and that includes exiting QEMU with the key sequence Control+c. This option is enabled by default — use signal=off to disable it.

-chardev braille,id=id — connects to a local BrlAPI server 🫶. braille takes no options.
-chardev parallel,id=id,path=path — parallel is only available on Linux, FreeBSD and DragonFlyBSD hosts. Connects to a local parallel port.

path specifies the path to the parallel port device. path is required.

-chardev spicevmc,id=id,debug=debug,name=name — spicevmc is only available when spice support is built in 🌶️.

debug — debug level for spicevmc.

name — name of spice channel to connect to.

Connects to a spice virtual machine channel, such as vdiport.

-chardev spiceport,id=id,debug=debug,name=name — spiceport is only available when spice support is built in 🌶️.

debug — debug level for spicevmc.

name — name of spice port to connect to.

Connects to a spice port, allowing a Spice client to handle the traffic identified by a name (preferably a fqdn).

-chardev dbus,id=id,name=name — dbus is only available when D-Bus display support is built in 🚌.

name — name of the chardev as exported on the D-Bus display interface.

Exports the character device on the D-Bus display interface, so that a D-Bus client can connect to it.

🎀 TPM device options 🎀

The general form of a TPM device option is:

-tpmdev backend,id=id[,options]

-tpmdev backend,id=id[,options] — the specific backend type will determine the applicable options 🔐. The -tpmdev option creates the TPM backend and requires a -device option that specifies the TPM frontend interface model.

Use -tpmdev help to print all available TPM backend types. iykyk 💅

The available backends are:

-tpmdev passthrough,id=id,path=path,cancel-path=cancel-path — (Linux-host only) enable access to the host's TPM using the passthrough driver 🔓.

path specifies the path to the host's TPM device, i.e. on a Linux host this would be /dev/tpm0. path is optional and by default /dev/tpm0 is used.

cancel-path specifies the path to the host TPM device's sysfs entry allowing for cancellation of an ongoing TPM command. cancel-path is optional and by default QEMU will search for the sysfs entry to use.

Some notes about using the host's TPM with the passthrough driver:

The TPM device accessed by the passthrough driver must not be used by any other application on the host. No sharing, bestie 🚫.

Like, real talk bestie — because your host's firmware (BIOS/UEFI) already went ahead and initialized the TPM, the VM's firmware (BIOS/UEFI) literally can't initialize it again, so it might not even show you a TPM-specific menu that would otherwise let you configure the thing, e.g. enable/disable or activate/deactivate the TPM. 😭 And it gets worse: if TPM ownership is released from inside a VM, then your host's TPM gets disabled and deactivated. To enable and activate the TPM again afterwards, the host has to be rebooted and you have to go into the firmware's menu to enable and activate the TPM. If the TPM is left disabled and/or deactivated, most TPM commands will fail. 💀

To create a passthrough TPM, use these two options:

-tpmdev passthrough,id=tpm0 -device tpm-tis,tpmdev=tpm0

Note that the -tpmdev id is tpm0 and is referenced by tpmdev=tpm0 in the device option. iykyk 🎀

-tpmdev emulator,id=id,chardev=dev — (Linux-host only, bestie 💅) Enable access to a TPM emulator using a Unix domain socket based chardev backend.

chardev specifies the unique ID of a character device backend that provides the connection to the software TPM server.

To create a TPM emulator backend device with the chardev socket backend:

-chardev socket,id=chrtpm,path=/tmp/swtpm-sock -tpmdev emulator,id=tpm0,chardev=chrtpm -device tpm-tis,tpmdev=tpm0

🥾 Boot Image or Kernel specific 🥾

So there are broadly 4 ways you can boot a system with QEMU. Pick your fighter ✨

specify a firmware and let it control finding a kernel
specify a firmware and pass a hint to the kernel to boot
direct kernel image boot
manually load files into the guest's address space

The third method is lowkey useful for quickly testing kernels, but — and this is the catch — since there's no firmware to pass configuration information to the kernel, the hardware must either be probeable, the kernel built for the exact configuration, or passed some configuration data (e.g. a DTB blob) which tells the kernel what drivers it needs. These exact details are often hardware specific. 🙈

The final method is the most generic way of loading images into the guest address space, and it's used mostly for bare metal type development where the reset vectors of the processor are taken into account.

For x86 machines and some other architectures, -bios will generally do the right thing with whatever it is given. For other machines, the more strict -pflash option needs an image that is sized for the flash device for the given machine type. ⭐

Please see the QEMU System Emulator Targets <#system-targets-ref> section of the manual for more detailed documentation.

-bios file — Set the filename for the BIOS.
-pflash file — Use file as a parallel flash image.

The kernel options were designed to work with Linux kernels, although other things (like hypervisors) can be packaged up as a kernel executable image. The exact format of an executable image is usually architecture specific. 🦋

The way in which the kernel is started (what address it is loaded at, what if any information is passed to it via CPU registers, the state of the hardware when it is started, and so on) is also architecture specific. Typically it follows the specification laid down by the Linux kernel for how kernels for that architecture must be started.

-kernel bzImage — Use bzImage as kernel image. The kernel can be either a Linux kernel or in multiboot format.
-shim shim.efi — Use shim.efi to boot the kernel.
-append cmdline — Use cmdline as kernel command line.
-initrd file — Use file as initial ram disk.
-initrd "file1 arg=foo,file2" — This syntax is only available with multiboot.

Use file1 and file2 as modules and pass arg=foo as parameter to the first module. Commas can be provided in module parameters by doubling them on the command line to escape them:

-initrd "bzImage earlyprintk=xen,,keep root=/dev/xvda1,initrd.img" — Multiboot only. Use bzImage as the first module with "earlyprintk=xen,keep root=/dev/xvda1" as its command line, and initrd.img as the second module. 🔥
-dtb file — Use file as a device tree binary (dtb) image and pass it to the kernel on boot.

Finally, you can also manually load images directly into the address space of the guest. This is most useful for developers who already know the layout of their guest and take care to ensure something sane will happen when the reset vector executes. 👑

The generic loader can be invoked by using the loader device:

-device loader,addr=<addr>,data=<data>,data-len=<data-len>[,data-be=<data-be>][,cpu-num=<cpu-num>]

There's also the guest loader, which operates in a similar way but tweaks the DTB so a hypervisor loaded via -kernel can find where the guest image is:

-device guest-loader,addr=<addr>[,kernel=<path>,[bootargs=<arguments>]][,initrd=<path>]

🌈 Debug/Expert options 🌈

This is the part of the page where things get experimental and a lil spicy — some of these are literally marked "(experimental)", so don't come for me if the vibes shift. 💀

-compat [deprecated-input=<input-policy>][,deprecated-output=<output-policy>] — Set policy for handling deprecated management interfaces (experimental):
- deprecated-input=accept (default) — Accept deprecated commands and arguments
- deprecated-input=reject — Reject deprecated commands and arguments
- deprecated-input=crash — Crash on deprecated commands and arguments
- deprecated-output=accept (default) — Emit deprecated command results and events
- deprecated-output=hide — Suppress deprecated command results and events

Limitation: covers only syntactic aspects of QMP. 🫶

-compat [unstable-input=<input-policy>][,unstable-output=<output-policy>] — Set policy for handling unstable management interfaces (experimental):
- unstable-input=accept (default) — Accept unstable commands and arguments
- unstable-input=reject — Reject unstable commands and arguments
- unstable-input=crash — Crash on unstable commands and arguments
- unstable-output=accept (default) — Emit unstable command results and events
- unstable-output=hide — Suppress unstable command results and events

Limitation: covers only syntactic aspects of QMP. ✨

-fw_cfg [name=]name,file=file — Add named fw_cfg entry with contents from file file. If the filename contains a comma, you must double it (for instance, "file=my,,file" to use file "my,file"). 🍓
-fw_cfg [name=]name,string=str — Add named fw_cfg entry with contents from string str. If the string contains a comma, you must double it (for instance, "string=my,,string" to use file "my,string").

The terminating NUL character of the contents of str will not be included as part of the fw_cfg item data. To insert contents with embedded NUL characters, you have to use the file parameter. Ngl, that's the kinda detail that bites you later.

The fw_cfg entries are passed by QEMU through to the guest.

Example:

-fw_cfg name=opt/com.mycompany/blob,file=./my_blob.bin

This creates an fw_cfg entry named opt/com.mycompany/blob with contents from ./my_blob.bin. Slay. 💅

-serial dev — Redirect the virtual serial port to host character device dev. The default device is vc in graphical mode and stdio in non graphical mode.

This option can be used several times to simulate multiple serial ports. 👑

You can use -serial none to suppress the creation of default serial devices.

Available character devices are:

vc[:WxH] — Virtual console. Optionally, a width and height can be given in pixel with

vc:800x600

It is also possible to specify width or height in characters:

vc:80Cx24C

pty[:path] — [Linux only] Pseudo TTY (a new PTY is automatically allocated).

If path is specified, QEMU will create a symbolic link at that location which points to the new PTY device. 🎀

This avoids having to make QMP or HMP monitor queries to find out what the new PTY device path is.

Note that while QEMU will remove the symlink when it exits gracefully, it will not do so in case of crashes or on certain startup errors. It is recommended that the user checks and removes the symlink after QEMU terminates to account for this. Don't get caught lacking by a stale symlink bestie 😭

none — No device is allocated. Note that for machine types which emulate systems where a serial device is always present in real hardware, this may be equivalent to the null option, in that the serial device is still present but all output is discarded. For boards where the number of serial ports is truly variable, this suppresses the creation of the device.
null — A guest will see the UART or serial device as present in the machine, but all output is discarded, and there is no input. Conceptually equivalent to redirecting the output to /dev/null. 💀
chardev:id — Use a named character device defined with the -chardev option.
/dev/XXX — [Linux only] Use host tty, e.g. /dev/ttyS0. The host serial port parameters are set according to the emulated ones.
/dev/parportN — [Linux only, parallel port only] Use host parallel port N. Currently SPP and EPP parallel port features can be used.
file:filename — Write output to filename. No character can be read.
stdio — [Unix only] standard input/output
pipe:filename — name pipe filename
COMn — [Windows only] Use host serial port n
udp:[remote_host]:remote_port[@[src_ip]:src_port] — This implements UDP Net Console. When remote_host or src_ip are not specified they default to 0.0.0.0. When not using a specified src_port a random port is automatically chosen.

If you just want a simple readonly console you can use netcat or nc, by starting QEMU with: -serial udp::4555 and nc as: nc -u -l -p 4555. Any time QEMU writes something to that port it will appear in the netconsole session. 🌸

ok so if you're planning to send characters back through netconsole, or you're the type who starts and stops QEMU a million times a day (relatable 💅), you lowkey want QEMU to use the same source port every single time — do that with something like -serial udp::4555@:4556 passed to QEMU. Another whole approach? Use a patched version of netcat that can listen on a TCP port and send and receive characters over udp. And if you've got a patched netcat that activates telnet remote echo and single char transfer, then bestie you can use the following options to set up a netcat redirector so telnet on port 5555 can reach the QEMU port. 🎀

QEMU Options:

-serial udp::4555@:4556

netcat options:

-u -P 4555 -L 0.0.0.0:4556 -t -p 5555 -I -T

telnet options:

localhost 5555

tcp:[host]:port[,server=on|off][,wait=on|off][,nodelay=on|off][,reconnect-ms=milliseconds] — the TCP Net Console, and it's giving two modes of operation, bestie. It can either send the serial I/O somewhere or wait for a connection to come to it. By default the TCP Net Console is sent to host at the port. If you slap on server=on, QEMU will wait for a client socket application to connect to the port before continuing — unless you also specified the wait=on|off option. The nodelay=on|off option disables the Nagle buffering algorithm. The reconnect-ms option only applies if server=no is set, and if the connection goes down it'll attempt to reconnect at the given interval. If host is omitted, 0.0.0.0 is assumed. Only one TCP connection at a time is accepted. You can use telnet=on to connect to the corresponding character device.
- Example to send tcp console to 192.168.0.2 port 4444:

-serial <tcp:192.168.0.2:4444>

Example to listen and wait on port 4444 for connection:

-serial <tcp::4444,server=on>

Example to not wait and listen on ip 192.168.0.100 port 4444:

-serial <tcp:192.168.0.100:4444,server=on,wait=off>

telnet:host:port[,server=on|off][,wait=on|off][,nodelay=on|off] — same energy, except the telnet protocol is used instead of raw tcp sockets. The options work exactly the same as if you'd specified -serial tcp. The difference (and this is the ✨iconic✨ part) is that the port acts like a telnet server or client using telnet option negotiation. This also lets you send the MAGIC_SYSRQ sequence if you use a telnet that supports sending the break sequence. Typically in unix telnet you do it with Control+] and then type send break followed by pressing the enter key.
websocket:host:port,server=on[,wait=on|off][,nodelay=on|off] — the WebSocket protocol instead of a raw tcp socket. The port acts as a WebSocket server. Client mode is not supported, so don't even try it 🙈.
unix:path[,server=on|off][,wait=on|off][,reconnect-ms=milliseconds] — a unix domain socket instead of a tcp socket. Works the same as if you'd specified -serial tcp, except the unix domain socket path is used for connections.
mon:dev_string — this one's special 🫶, it lets the monitor be multiplexed onto another serial port. You access the monitor with the key sequence Control+a and then pressing c. dev_string should be any one of the serial devices specified above. Example of multiplexing the monitor onto a telnet server listening on port 4444:

-serial mon:telnet::4444,server=on,wait=off

Heads up bestie: when the monitor is multiplexed to stdio this way, Ctrl+C will not terminate QEMU anymore — it gets passed to the guest instead. 😭

braille — Braille device. Uses BrlAPI to display the braille output on a real or fake device.
msmouse — three button serial mouse. Configure the guest to use Microsoft protocol.

-parallel dev — redirect the virtual parallel port to host device dev (same devices as the serial port, iykyk). On Linux hosts, /dev/parportN can be used to use hardware devices connected on the corresponding host parallel port.

This option can be used several times to simulate up to 3 parallel ports.

Use -parallel none to disable all parallel ports.

-monitor dev — redirect the monitor to host device dev (again, same devices as the serial port). The default device is vc in graphical mode and stdio in non graphical mode. Use -monitor none to disable the default monitor.

The use of -monitor dev is syntactic sugar for creating a character device from dev and pairing it with -object monitor-hmp. Both the character device and monitor object will be given an ID compat_monitorNNN where NNN is a counter starting from 0.

-qmp dev — like -monitor but opens in 'control' mode. For example, to make QMP available on localhost port 4444:

-qmp tcp:localhost:4444,server=on,wait=off

The use of -qmp dev is syntactic sugar for creating a character device from dev and pairing it with -object monitor-qmp. Both the character device and monitor object will be given an ID compat_monitorNNN where NNN is a counter starting from 0.

-qmp-pretty dev — like -qmp but uses pretty JSON formatting. Aesthetic and functional 💖.
-mon [chardev=]name[,mode=readline|control][,pretty=on|off] — set up a monitor connected to the chardev name. QEMU supports two monitors: the Human Monitor Protocol (HMP; for human interaction, obviously) and the QEMU Monitor Protocol (QMP; a JSON RPC-style protocol). The default is HMP; mode=control selects QMP instead. pretty is only valid when mode=control, turning on JSON pretty printing to ease human reading and debugging.

For example:

-chardev socket,id=mon1,host=localhost,port=4444,server=on,wait=off \
-mon chardev=mon1,mode=control,pretty=on

enables the QMP monitor on localhost port 4444 with pretty-printing.

The use of -mon mode=readline is deprecated syntactic sugar for the new -object monitor-hmp option, each use of which creates an object with the ID compat_monitorNNN where NNN is a counter starting from 0.

The use of -mon mode=control is deprecated syntactic sugar for the new -object monitor-qmp option, each use of which creates an object with the ID compat_monitorNNN where NNN is a counter starting from 0.

-debugcon dev — redirect the debug console to host device dev (same devices as the serial port, per usual 💅). The debug console is an I/O port which is typically port 0xe9; writing to that I/O port sends output to this device. The default device is vc in graphical mode and stdio in non graphical mode.
-pidfile file — store the QEMU process PID in file. Super useful if you launch QEMU from a script 🫶.
--preconfig — pause QEMU for interactive configuration before the machine is created, which allows querying and configuring properties that will affect machine initialization. Use the QMP command x-exit-preconfig to exit the preconfig state and move to the next state (i.e. run guest if -S isn't used, or pause the second time if -S is used). This option is experimental, so proceed accordingly 🧚.
-S — do not start CPU at startup (you must type c in the monitor). That's it, that's the whole vibe.
-overcommit mem-lock=on|off|on-fault
-overcommit cpu-pm=on|off — run qemu with hints about host resource overcommit. The default is to assume that the host overcommits all resources.

Locking qemu and guest memory can be enabled via mem-lock=on or mem-lock=on-fault (disabled by default). This works when host memory is not overcommitted and reduces the worst-case latency for the guest. The on-fault option is better for reducing the memory footprint since it makes allocations lazy — but the pages still get locked in place once faulted by the guest or QEMU. Note that the two options are mutually exclusive, so pick one, bestie. ✨

Guest ability to manage power state of host cpus (increasing latency for other processes on the same host cpu, but decreasing latency for the guest) can be enabled via cpu-pm=on (disabled by default). This works best when the host CPU is not overcommitted. When used, host estimates of CPU cycle and power utilization will be incorrect, not taking into account guest idle time. 💀

-gdb dev — accept a gdb connection on device dev (see the GDB usage chapter in the System Emulation Users Guide). Note that this option does not pause QEMU execution — if you want QEMU to not start the guest until you connect with gdb and issue a continue command, you'll need to also pass the -S option to QEMU. Two-step slay.

The most usual configuration is to listen on a local TCP socket:

-gdb tcp::3117

but you can specify other backends; UDP, pseudo TTY, or even stdio are all reasonable use cases. For example, a stdio connection allows you to start QEMU from within gdb and establish the connection via a pipe:

(gdb) target remote | exec qemu-system-x86_64 -gdb stdio ...

-s — shorthand for -gdb <tcp::1234>, i.e. open a gdbserver on TCP port 1234 (see the GDB usage chapter in the System Emulation Users Guide). Tiny flag, big energy ⭐
-d item1[,...] — enable logging of specified items. Use -d help for a list of log items.
-D logfile — output log in logfile instead of to stderr.
-dfilter range1[,...] — filter debug output to that relevant to a range of target addresses. The filter spec can be either start+size, start-size or start..end where start, end and size are the addresses and sizes required. For example:

-dfilter 0x8000..0x8fff,0xffffffc000080000+0x200,0xffffffc000060000-0x1000

three whole ranges in one line, she ate 🔥

Will dump the output for literally any code in the 0x1000-sized block starting at 0x8000, plus the 0x200-sized block starting at 0xffffffc000080000, and another 0x1000-sized block starting at 0xffffffc00005f000. bestie that's three blocks, we love a triple threat 💅

-seed number — forces the guest to use a deterministic pseudo-random number generator, seeded with number. Lowkey important: this does not affect crypto routines within the host. nope, host crypto stays doing its own thing 🔥
-L path — sets the directory for the BIOS, VGA BIOS and keymaps.

Wanna see all the data directories? Just hit -L help and it'll list them for you. 🎀

-enable-kvm — turns on KVM full virtualization support. Heads up: this option only exists if KVM support was enabled when compiling. like, it's not a vibe you can just summon ✨
-xen-domid id — specify the xen guest domain id (XEN only).
-xen-attach — attach to an existing xen domain. libxl will use this when starting QEMU (XEN only). Also restricts the set of available xen operations to the specified domain id (XEN only). three-for-three on the XEN-only energy 💀
-no-reboot — exit instead of rebooting. that's it, that's the whole thing.
-no-shutdown — don't exit QEMU when the guest shuts down, just stop the emulation instead. this is iconic because it lets you, for instance, switch to the monitor and commit changes to the disk image. 🫶
-action event=action — this one modifies QEMU's default behavior when certain guest events occur. it's the generic way to specify the same behaviors that -no-reboot and -no-shutdown modify. very "one option to rule them all" of it 👑

Examples:

-action panic=none -action reboot=shutdown,shutdown=pause -device i6300esb -action watchdog=pause

-loadvm file — start right away with a saved state (that's loadvm in the monitor, iykyk).
-daemonize — daemonizes the QEMU process after initialization. QEMU will not detach from standard IO until it's ready to receive connections on any of its devices. this option is genuinely so useful for external programs launching QEMU without having to cope with initialization race conditions. no cap, it's giving "responsible adult" 🌸
-option-rom file — loads the contents of file as an option ROM. super handy for loading things like EtherBoot. 🦋
-rtc [base=utc|localtime|datetime][,clock=host|rt|vm][,driftfix=none|slew] — specify base as utc or localtime so the RTC starts at the current UTC or local time respectively. localtime is required for the correct date in MS-DOS or Windows. wanna start at a specific point in time? provide datetime in the format 2006-06-17T16:01:21 or 2006-06-17. the default base is UTC. ⭐

By default the RTC is driven by the host system time. this lets you use the RTC as an accurate reference clock inside the guest, especially if the host time is smoothly following an accurate external reference clock, e.g. via NTP. If you want to isolate the guest time from the host, set clock to rt instead — that gives you a host monotonic clock if the host supports it. To even prevent the RTC from progressing during suspension, set clock to vm (virtual clock). 'clock=vm' is recommended especially in icount mode in order to preserve determinism; however, note that in icount mode the speed of the virtual clock is variable and can in general differ from the host clock. 🌀

Enable driftfix (i386 targets only) if you're experiencing time drift problems, specifically with Windows' ACPI HAL. This option will try to figure out how many timer interrupts were not processed by the Windows guest and will re-inject them. very "we found your lost packages and redelivered them" 📦

-icount [shift=N|auto][,align=on|off][,sleep=on|off][,rr=record|replay,rrfile=filename[,rrsnapshot=snapshot]] — enables the virtual instruction counter. the virtual cpu will execute one instruction every 2^N ns of virtual time. if auto is specified then the virtual cpu speed will be automatically adjusted to keep virtual time within a few seconds of real time. ⏱️

Note that while this option can give deterministic behavior, it does not provide cycle accurate emulation. Modern CPUs contain superscalar out of order cores with complex cache hierarchies. The number of instructions executed often has little or no correlation with actual performance. so like, deterministic ≠ accurate, don't come for me 💀

When the virtual cpu is sleeping, the virtual time will advance at default speed unless sleep=off is specified. With sleep=off, the virtual time will jump to the next timer deadline instantly whenever the virtual cpu goes to sleep mode and will not advance if no timer is enabled. This behavior gives deterministic execution times from the guest point of view. The default if icount is enabled is sleep=on. sleep=off cannot be used together with either shift=auto or align=on. ⚠️

align=on will activate the delay algorithm which will try to synchronise the host clock and the virtual clock. The goal is to have a guest running at the real frequency imposed by the shift option. Whenever the guest clock is behind the host clock and if align=on is specified then we print a message to the user to inform about the delay. Currently this option does not work when shift is auto. Note: The sync algorithm will work for those shift values for which the guest clock runs ahead of the host clock. Typically this happens when the shift value is high (how high depends on the host machine). The default if icount is enabled is align=off. 🌈

When the rr option is specified deterministic record/replay is enabled. The rrfile= option must also be provided to specify the path to the replay log. In record mode data is written to this file, and in replay mode it is read back. If the rrsnapshot option is given then it specifies a VM snapshot name. In record mode, a new VM snapshot with the given name is created at the start of execution recording. In replay mode this option specifies the snapshot name used to load the initial VM state. 🎥

-watchdog-action action — the action controls what QEMU will do when the watchdog timer expires. The default is reset (forcefully reset the guest). Other possible actions are: shutdown (attempt to gracefully shutdown the guest), poweroff (forcefully poweroff the guest), inject-nmi (inject a NMI into the guest), pause (pause the guest), debug (print a debug message and continue), or none (do nothing). 🐕

Note that the shutdown action requires that the guest responds to ACPI signals, which it may not be able to do in the sort of situations where the watchdog would have expired, and thus -watchdog-action shutdown is not recommended for production use. fr fr, don't ship that 🙈

Examples:

-device i6300esb -watchdog-action pause

-echr numeric_ascii_value — changes the escape character used for switching to the monitor when using monitor and serial sharing. The default is 0x01 when using the -nographic option. 0x01 is equal to pressing Control+a. You can select a different character from the ascii control keys where 1 through 26 map to Control+a through Control+z. For instance you could use either of the following to change the escape character to Control+t. 🎀

-echr 0x14; -echr 20

The -incoming option specifies the migration channel for an incoming migration. It may be used multiple times to specify multiple migration channel types. The channel type is specified in <channel>, or is 'main' for all other forms of -incoming. If multiple -incoming options are specified for a channel type, the last one takes precedence. 👑

-incoming tcp:[host]:port[,to=maxport][,ipv4=on|off][,ipv6=on|off]
-incoming rdma:host:port[,ipv4=on|off][,ipv6=on|off]

Prepare for incoming migration, listen on a given tcp port. 🛰️

-incoming unix:socketpath — prepare for incoming migration, listen on a given unix socket.
-incoming fd:fd — accept incoming migration from a given file descriptor.
-incoming file:filename[,offset=offset] — accept incoming migration from a given file starting at offset. offset allows the common size suffixes, or a 0x prefix, but not both. pick a lane bestie 💅
-incoming exec:cmdline — accept incoming migration as an output from specified external command.
-incoming <channel> — accept incoming migration on the migration channel. For the syntax of <channel>, see the QAPI documentation of MigrationChannel. Examples:

-incoming '{"channel-type": "main",
            "addr": { "transport": "socket",
                      "type": "unix",
                      "path": "my.sock" }}'

-incoming main,addr.transport=socket,addr.type=unix,addr.path=my.sock

-incoming defer — wait for the URI to be specified via migrate_incoming. The monitor can be used to change settings (such as migration parameters) prior to issuing the migrate_incoming to allow the migration to begin. ⏳
-only-migratable — only allow migratable devices. Devices will not be allowed to enter an unmigratable state. 🔒
-nodefaults — don't create default devices. Normally, QEMU sets the default devices like serial port, parallel port, virtual console, monitor device, VGA adapter, floppy and CD-ROM drive and others. The -nodefaults option will disable all those default devices. clean slate energy ✨
-prom-env variable=value — sets the OpenBIOS nvram variable to the given value (PPC, SPARC only).

qemu-system-sparc -prom-env 'auto-boot?=false' \
 -prom-env 'boot-device=sd(0,2,0):d' -prom-env 'boot-args=linux single'

qemu-system-ppc -prom-env 'auto-boot?=false' \
 -prom-env 'boot-device=hd:2,\yaboot' \
 -prom-env 'boot-args=conf=hd:2,\yaboot.conf'

-semihosting — enables Semihosting mode (ARM, M68K, Xtensa, MIPS, RISC-V only). 🧚

Warning: Note that this allows guest direct access to the host filesystem, so should only be used with a trusted guest OS. 🚨 like, genuinely, trusted guest only, no exceptions.

See the -semihosting-config option documentation for further information about the facilities this enables.

-semihosting-config [enable=on|off][,target=native|gdb|auto][,chardev=id][,userspace=on|off][,arg=str[,...]] — enable and configure Semihosting (ARM, M68K, Xtensa, MIPS, RISC-V only).

Warning: Note that this allows guest direct access to the host filesystem, so should only be used with a trusted guest OS. 🔥 same warning, same seriousness.

target=native|gdb|auto — this is the "who's catching the semihosting calls" knob 👀 — calls get addressed either to QEMU (native) or to GDB (gdb). Default is auto, which is the ✨reads-the-room✨ option: it means gdb during debug sessions and native otherwise.
chardev=str1 — send the output to a chardev backend output, for native — or for auto output when you're not in gdb. That's where the semihosting output actually goes, bestie 🎀
userspace=on|off — lets code running in guest userspace access the semihosting interface. Default vibe is strict: only privileged guest code can make semihosting calls. Note that setting userspace=on should only be used if all guest code is trusted (for example, in bare-metal test case code).

🫶 fr fr, that trust warning is not decorative — it's the whole point of the default.

arg=str1,arg=str2,... — lets you pass input arguments, and you can use it multiple times to build up a whole list 🧾. The old-style -kernel/-append method of passing a command line is still supported for backward compatibility. And if both the --semihosting-config arg and the -kernel/-append are specified, the former is passed to semihosting, because it always takes precedence 👑 periodt.

-sandbox arg[,obsolete=string][,elevateprivileges=string][,spawn=string][,resourcecontrol=string] — turns on the Seccomp mode 2 system call filter 🛡️. 'on' will enable syscall filtering and 'off' will disable it. The default is 'off'.
- obsolete=string — enable Obsolete system calls.
- elevateprivileges=string — disable set*uid|gid system calls.
- spawn=string — disable *fork and execve.
- resourcecontrol=string — disable process affinity and scheduler priority.
-readconfig file — read device configuration from file 📄. This approach is a lifesaver when you want to spawn a QEMU process with so many command line options but you do not want to exceed the command line character limit. Very "my options outgrew the terminal" energy.
-no-user-config — the -no-user-config option makes QEMU not load any of the user-provided config files on sysconfdir. Simple. Iconic. No user config, no problem 💅
-trace [[enable=]pattern][,events=file][,file=file] — specify tracing options 🔍.
- [enable=]PATTERN — immediately enable events matching PATTERN (either event name or a globbing pattern). This option is only available if QEMU has been compiled with the simple, log or ftrace tracing backend. To specify multiple events or patterns, just specify the -trace option multiple times. Use -trace help to print a list of names of trace points — extremely handy, use it 🧚
- events=FILE — immediately enable events listed in FILE. The file must contain one event name (as listed in the trace-events-all file) per line; globbing patterns are accepted too. This option is only available if QEMU has been compiled with the simple, log or ftrace tracing backend.
- file=FILE — log output traces to FILE. This option is only available if QEMU has been compiled with the simple tracing backend.
-plugin file=file[,argname=argvalue] — load a plugin 🔌.
- file=file — load the given plugin from a shared library file.
- argname=argvalue — argument passed to the plugin. (Can be given multiple times.)
-run-with [async-teardown=on|off][,chroot=dir][,exit-with-parent=on|off][,user=username|uid:gid] — set QEMU process lifecycle options 🌱.
- async-teardown=on enables asynchronous teardown. A new process called "cleanup/<QEMU_PID>" will be created at startup sharing the address space with the main QEMU process, using clone. It will wait for the main QEMU process to terminate completely, and then exit. This allows QEMU to terminate very quickly even if the guest was huge, leaving the teardown of the address space to the cleanup process. Since the cleanup process shares the same cgroups as the main QEMU process, accounting is performed correctly.

💀 Important caveat: this only works if the cleanup process is not forcefully killed with SIGKILL before the main QEMU process has terminated completely. Kill it early and the whole trick falls apart.

chroot=dir can be used for doing a chroot to the specified directory immediately before starting the guest execution. This is especially useful in combination with user=....
exit-with-parent=on causes QEMU to exit if the parent process of QEMU exits. This can be used when QEMU runs a captive appliance, where the lifetime of the appliance is scoped to the parent process. In case the parent process crashes, QEMU is still cleaned up. This only works on Linux, FreeBSD and macOS platforms 🐧🦋.
user=username or user=uid:gid can be used to drop root privileges before starting guest execution. QEMU will use the setuid and setgid system calls to switch to the specified identity. Note that the user=username syntax will also apply the full set of supplementary groups for the user, whereas the user=uid:gid will use only the gid group.

-msg [timestamp=on|off][,guest-name=on|off] — control error message format 💬.
- timestamp=on|off — prefix messages with a timestamp. Default is off.
- guest-name=on|off — prefix messages with guest name but only if -name guest option is set, otherwise the option is ignored. Default is off.
-dump-vmstate file — dump json-encoded vmstate information for the current machine type to file in file 📦.
-enable-sync-profile — enable synchronization profiling.
-perfmap — generate a map file for Linux perf tools that will allow basic profiling information to be broken down into basic blocks 🔥.
-jitdump — generate a dump file for Linux perf tools that maps basic blocks to symbol names, line numbers and JITted code.

🎀 Generic object creation 🎀

-object typename[,prop1=value1,...] — create a new object of type typename, setting properties in the order they are specified 💅. Note that the 'id' property must be set. These objects are placed in the '/objects' path.

🌸 `-object monitor-hmp,id=id,chardev=chardev_id,readline=on|off`

Set up a monitor running the Human Monitor Protocol, connected to the chardev chardev_id.

The id parameter is a unique ID that can be used to dynamically delete the monitor at runtime. Note that monitors created using the historical syntax will be allocated IDs following the pattern compat_monmitorNNN. Mixing -object with -monitor syntax is discouraged.

The readline parameter, which defaults to on, controls whether the monitor provides line editing ✍️.

💜 `-object monitor-qmp,id=id,chardev=chardev_id,pretty=on|off,close-action=none|delete`

Set up a monitor running the QEMU Monitor Protocol, connected to the chardev chardev_id.

The id parameter is a unique ID that can be used to dynamically delete the monitor at runtime. Note that monitors created using the historical syntax will be allocated IDs following the pattern compat_monitorNNN. Mixing -object with -qmp and -qmp-pretty syntax is discouraged.

The pretty parameter, which defaults to off, controls whether the monitor responses are pretty printed as multi-line indented JSON, as opposed to constrained to a single line without extraneous whitespace ✨.

The close-action parameter, which defaults to none, controls what happens when the connection to the monitor is terminated by the user. If set to delete, then the monitor-qmp object and its associated character device are both immediately deleted. This can be useful if an extra monitor was hotplugged for a specific task and should be unplugged when completed 👋.

🦋 `-object memory-backend-file,id=id,size=size,mem-path=dir,share=on|off,discard-data=on|off,merge=on|off,dump=on|off,prealloc=on|off,host-nodes=host-nodes,policy=default|preferred|bind|interleave,align=align,offset=offset,readonly=on|off,rom=on|off|auto`

Creates a memory file backend object, which can be used to back the guest RAM with huge pages 🐘.

The id parameter is a unique ID that will be used to reference this memory region in other parameters, e.g. -numa, -device nvdimm, etc.

The size option provides the size of the memory region, and accepts common suffixes, e.g. 500M.

The mem-path provides the path to either a shared memory or huge page filesystem mount.

The share boolean option determines whether the memory region is marked as private to QEMU, or shared. The latter allows a co-operating external process to access the QEMU memory region.

⚠️ Setting share=on might affect the ability to configure NUMA bindings for the memory backend under some circumstances, see Documentation/vm/numa_memory_policy.txt on the Linux kernel source tree for additional details.

Setting the discard-data boolean option to on indicates that file contents can be destroyed when QEMU exits, to avoid unnecessarily flushing data to the backing file. Note that discard-data is only an optimization, and QEMU might not discard file contents if it aborts unexpectedly or is terminated using SIGKILL 😭 — so don't rely on it like it's a guarantee.

The merge boolean option enables memory merge, also known as MADV_MERGEABLE, so that Kernel Samepage Merging will consider the pages for memory deduplication.

Setting the dump boolean option to off excludes the memory from core dumps. This feature is also known as MADV_DONTDUMP.

The prealloc boolean option enables memory preallocation.

The host-nodes option binds the memory range to a list of NUMA host nodes.

The policy option sets the NUMA policy to one of the following values:

default — default host policy.

preferred — prefer the given host node list for allocation ✨
bind — restrict memory allocation to the given host node list 🔒
interleave — interleave memory allocations across the given host node list 🌈

The align option sets the base address alignment that QEMU uses when it does an mmap(2) on mem-path, and it takes the usual suffixes, like 2M. Some backends named by mem-path need a different alignment than QEMU's default, bestie — for example the device DAX /dev/dax0.0 wants 2M alignment instead of 4K. In those cases you just pass the alignment you need through this option, easy 💅

The offset option says where inside the target file the region starts. You can use it to back multiple regions with just one file, which is honestly such a slay 🎀

The pmem option says whether the backing file named by mem-path lives in host persistent memory that can be accessed through the SNIA NVM programming model (e.g. Intel NVDIMM). If pmem is set to 'on', QEMU does the necessary operations to guarantee its own writes to mem-path persist (e.g. in vNVDIMM label emulation and live migration). It also maps the backend file with the MAP_SYNC flag, which keeps the file metadata in sync for mem-path even if the host crashes or the power dies 😭 MAP_SYNC needs support from both the host kernel (since Linux kernel 4.15) and the filesystem of mem-path being mounted with the DAX option. Both. No cap.

The readonly option says whether the backing file opens read-only or read-write (read-write is the default).

The rom option says whether to create Read Only Memory (ROM) that the VM can't modify. Any write attempts to that ROM get denied, periodt. Most use cases actually want proper RAM instead of ROM. But some selected use cases, like R/O NVDIMMs, really do benefit from ROM. So: set it to on → create ROM; set it to off → create writable RAM; set it to auto (the default) → whatever the readonly option says gets used. This option is mainly helpful when you want writable RAM in configs that would traditionally create ROM before the rom option even existed — think VM templating, where we want to open a file readonly (readonly=on) and mark the memory private for QEMU (share=off). For that use case we need writable RAM instead of ROM, and we also want to set rom=off 🫶

-object memory-backend-ram,id=id,merge=on|off,dump=on|off,share=on|off,prealloc=on|off,size=size,host-nodes=host-nodes,policy=default|preferred|bind|interleave — creates a memory backend object that can back the guest RAM. Memory backend objects give you way more control than the -m option everyone traditionally uses to define guest RAM. For a rundown of the options, go peek at memory-backend-file 👀
-object memory-backend-memfd,id=id,merge=on|off,dump=on|off,share=on|off,prealloc=on|off,size=size,host-nodes=host-nodes,policy=default|preferred|bind|interleave,seal=on|off,hugetlb=on|off,hugetlbsize=size — creates an anonymous memory file backend object, which lets QEMU share the memory with an external process (e.g. when you're using vhost-user). The memory is allocated with memfd and optional sealing. (Linux only, bestie 🐧)

The seal option creates a sealed-file, which blocks any further resizing of the memory ('on' by default).

The hugetlb option says the file being created lives in the hugetlbfs filesystem (since Linux 4.14). Used together with the hugetlb option, the hugetlbsize option sets the hugetlb page size on systems that support multiple hugetlb page sizes (it has to be a power of 2 value that the system supports).

On some versions of Linux, the hugetlb option is incompatible with the seal option (needs at least Linux 4.16) — so like, watch out for that combo 💀

For a description of the other options, refer to memory-backend-file.

The share boolean option is on by default with memfd ✨

-object memory-backend-shm,id=id,merge=on|off,dump=on|off,share=on|off,prealloc=on|off,size=size,host-nodes=host-nodes,policy=default|preferred|bind|interleave — creates a POSIX shared memory backend object, which lets QEMU share the memory with an external process (e.g. when using vhost-user).

memory-backend-shm is basically a more portable but less featureful version of memory-backend-memfd. So you can use it on any POSIX system, especially when memfd isn't supported 🦋

For a description of the options, refer to memory-backend-file.

The share boolean option is on by default with shm. Setting it to off will cause a failure during allocation, because this backend just doesn't support that — not me finding out the hard way 🙈

-object iommufd,id=id[,fd=fd] — creates an iommufd backend that lets you control DMA mapping through the /dev/iommu device.

The id parameter is a unique ID that frontends (like vfio-pci or vdpa) use to connect to the iommufd backend.

The fd parameter is an optional pre-opened file descriptor that comes from opening /dev/iommu. Usually iommufd is shared across all subsystems, which brings the benefit of centralized reference counting 👑

-object rng-builtin,id=id — creates a random number generator backend that grabs entropy from QEMU's builtin functions. The id parameter is a unique ID used to reference this entropy backend from the virtio-rng device. By default, the virtio-rng device uses this RNG backend ⭐
-object rng-random,id=id,filename=/dev/random — creates a random number generator backend that grabs entropy from a device on the host. The id parameter is a unique ID used to reference this entropy backend from the virtio-rng device. The filename parameter says which file to grab entropy from, and if you leave it out it defaults to /dev/urandom 🍓
-object rng-egd,id=id,chardev=chardevid — creates a random number generator backend that grabs entropy from an external daemon running on the host. The id parameter is a unique ID used to reference this entropy backend from the virtio-rng device. The chardev parameter is the unique ID of a character device backend that provides the connection to the RNG daemon 🔥
-object tls-creds-anon,id=id,endpoint=endpoint,dir=/path/to/cred/dir,verify-peer=on|off — creates a TLS anonymous credentials object, which can provide TLS support on network backends. The id parameter is a unique ID that network backends use to access the credentials. The endpoint is either server or client, depending on whether the QEMU network backend using the credentials will be acting as a client or as a server. If verify-peer is enabled (it's the default), then once the handshake finishes, the peer credentials get verified — though honestly that's a no-op for anonymous credentials 🤷‍♀️

The dir parameter tells QEMU where to find the credential files. For server endpoints, this directory may contain a file dh-params.pem providing diffie-hellman parameters to use for the TLS server. If that file is missing, QEMU will generate a set of DH parameters at startup. That's a computationally expensive operation that eats random pool entropy, so it's recommended that you generate a persistent set of parameters up front and save them. Future-you will thank you 💜

-object tls-creds-psk,id=id,endpoint=endpoint,dir=/path/to/keys/dir[,username=username] — creates a TLS Pre-Shared Keys (PSK) credentials object, which can provide TLS support on network backends. The id parameter is a unique ID that network backends use to access the credentials. The endpoint is either server or client, depending on whether the QEMU network backend using the credentials will be acting as a client or as a server. For clients only, username is the username that gets sent to the server. If you omit it, it defaults to "qemu" ✨

The dir parameter tells QEMU where to find the keys file. It's called "dir/keys.psk" and contains "username:key" pairs. Easiest way to create this file? The GnuTLS psktool program, bestie 🎀

For server endpoints, dir may also contain a file dh-params.pem providing diffie-hellman parameters to use for the TLS server. If that file is missing, QEMU will generate a set of DH parameters at startup. Again, that's computationally expensive and consumes random pool entropy, so it's recommended that you generate a persistent set of parameters up front and save them 🧚

-object tls-creds-x509,id=id,endpoint=endpoint,dir=/path/to/cred/dir,priority=priority,verify-peer=on|off,passwordid=id — okay so this one makes a TLS anonymous credentials object, and it's what gives your network backends their TLS glow-up. The id is a unique ID that the network backends use to grab the credentials. endpoint is either server or client, depending on whether the QEMU network backend using these creds is acting as a client or as a server. If verify-peer is enabled (it's the default, bestie) then once the handshake finishes, the peer credentials get verified. With x509 certs that means clients have to be given valid client certificates too — no cert, no entry 💅

The dir parameter tells QEMU where the credential files live. For server endpoints, this directory may contain a file dh-params.pem giving diffie-hellman parameters for the TLS server. If that file is missing, QEMU will generate a set of DH parameters at startup — and that is a computationally expensive operation that eats random pool entropy, so the recommended move is to generate a persistent set of parameters upfront and save them. Learn from the manpage's mistakes, girlies 🫶

For x509 certificate credentials, the directory holds further files with the x509 certificates. Certs must be stored in PEM format, in these filenames: ca-cert.pem, ca-crl.pem (optional), server-cert.pem (only servers), server-key.pem (only servers), client-cert.pem (only clients), and client-key.pem (only clients).

For server-key.pem and client-key.pem — the ones holding sensitive private keys, so like, handle with care 😭 — you can use an encrypted version by providing the passwordid parameter. That gives the ID of a previously created secret object containing the password for decryption.

The priority parameter lets you override the global default priority used by gnutls. Super useful if the sysadmin needs a weaker set of crypto priorities for QEMU without forcing that weakness onto every other application. Or, flip side: if you want a stronger default for QEMU than for all other apps, you can do that here too. Its format is a gnutls priority string, as described at https://gnutls.org/manual/html_node/Priority-Strings.html.

-object tls-cipher-suites,id=id,priority=priority — makes a TLS cipher suites object, which controls the TLS cipher/protocol algorithms applications are allowed to use. Very "you can't sit with us" of it.

The id is a unique ID that frontends use to access the ordered list of permitted TLS cipher suites from the host.

The priority parameter overrides the global default priority used by gnutls. Same deal as above: handy if a system administrator needs weaker crypto priorities for QEMU without potentially forcing the weakness onto all applications — or conversely, if someone wants a stronger default for QEMU than for all other applications, this parameter is how. Format is a gnutls priority string as described at https://gnutls.org/manual/html_node/Priority-Strings.html.

A real serve for this object: controlling UEFI HTTPS Boot. The tls-cipher-suites object exposes the ordered list of permitted TLS cipher suites from the host side to the guest firmware via fw_cfg. The list is represented as an array of IANA_TLS_CIPHER objects, and the firmware uses that IANA_TLS_CIPHER array for configuring guest-side TLS. Obsessed, honestly.

In the following example, the priority at which the host-side policy is retrieved is given by the priority property. Given that QEMU uses GNUTLS, priority=@SYSTEM may be used to refer to /etc/crypto-policies/back-ends/gnutls.config. Iconic ✨

# qemu-system-x86_64 \
    -object tls-cipher-suites,id=mysuite0,priority=@SYSTEM \
    -fw_cfg name=etc/edk2/https/ciphers,gen_id=mysuite0

-object filter-buffer,id=id,netdev=netdevid,interval=t[,queue=all|rx|tx][,status=on|off][,position=head|tail|id=<id>][,insert=behind|before] — interval t can't be 0 (fr fr, do not), and this filter batches packet delivery: all packets arriving in a given interval on netdev netdevid get delayed until the end of the interval. The interval is in microseconds. status is optional and says whether the netfilter is on (enabled) or off (disabled) — the default status for a netfilter is 'on'.

queue all|rx|tx is an option you can slap on any netfilter:

all: the filter is attached both to the receive and the transmit queue of the netdev (default).
rx: the filter is attached to the receive queue of the netdev, where it receives packets sent to the netdev.
tx: the filter is attached to the transmit queue of the netdev, where it receives packets sent by the netdev.

position head|tail|id=<id> specifies where the filter should be inserted in the filter list. Also applies to any netfilter:

head: the filter is inserted at the head of the filter list, before any existing filters.
tail: the filter is inserted at the tail of the filter list, behind any existing filters (default).
id=<id>: the filter is inserted before or behind the filter specified by <id> — see the insert option below.

insert behind|before specifies where to insert the new filter relative to the one given with position=id=<id>. Applies to any netfilter too:

before: insert before the specified filter.
behind: insert behind the specified filter (default).

-object filter-mirror,id=id,netdev=netdevid,outdev=chardevid,queue=all|rx|tx[,vnet_hdr_support][,position=head|tail|id=<id>][,insert=behind|before] — filter-mirror on netdev netdevid mirrors net packets to chardevid. If it has the vnet_hdr_support flag, filter-mirror will mirror the packet with vnet_hdr_len. Copy-paste but for packets, lowkey genius 🪞
-object filter-redirector,id=id,netdev=netdevid,indev=chardevid,outdev=chardevid,queue=all|rx|tx[,vnet_hdr_support][,position=head|tail|id=<id>][,insert=behind|before] — filter-redirector on netdev netdevid redirects the filter's net packet to chardev chardevid, and redirects indev's packet to the filter. If it has the vnet_hdr_support flag, filter-redirector will redirect the packet with vnet_hdr_len.

When you create a filter-redirector you need the outdev id to differ from the indev id — id can not be the same. You can just use indev or outdev, but at least one of indev or outdev needs to be specified.

-object filter-rewriter,id=id,netdev=netdevid,queue=all|rx|tx,[vnet_hdr_support][,position=head|tail|id=<id>][,insert=behind|before] — filter-rewriter is a part of the COLO project. It rewrites tcp packets going to the secondary from the primary to keep the secondary tcp connection alive, and rewrites tcp packets going to the primary from the secondary so the client can handle them. If it has the vnet_hdr_support flag, we can parse packets with the vnet header.

usage: colo secondary: -object filter-redirector,id=f1,netdev=hn0,queue=tx,indev=red0 -object filter-redirector,id=f2,netdev=hn0,queue=rx,outdev=red1 -object filter-rewriter,id=rew0,netdev=hn0,queue=all

-object filter-dump,id=id,netdev=dev[,file=filename][,maxlen=len][,position=head|tail|id=<id>][,insert=behind|before] — dumps the network traffic on netdev dev to the file given by filename. At most len bytes per packet are stored (64k by default). The file format is libpcap, so you can analyze it with tools like tcpdump or Wireshark. Snitch filter, honestly 💀
-object colo-compare,id=<id>,primary_in=<chardevid>,secondary_in=<chardevid>,outdev=<chardevid>,iothread=<id>[,vnet_hdr_support][,notify_dev=<id>][,compare_timeout=<time_ms>][,expired_scan_cycle=<time_ms>][,max_queue_size=<maxsize>] — colo-compare grabs packets from the chardev backends specified by primary_in and secondary_in, and compares whether the payloads of the primary packet and the secondary packet are the same. If they match, it outputs the primary packet to the chardev backend specified by outdev; otherwise it notifies the COLO-framework to do a checkpoint and sends the primary packet to outdev.

To improve efficiency, the comparison task needs to live in another iothread — the iothread option specifies that iothread object (which your commandline should create, don't forget bestie 👀).

The vnet_hdr_support flag tells colo compare to pass the vnet header length when it sends and receives packets.

The compare_timeout option sets the maximum time that colo-compare will hold the packet for, in ms.

The expired_scan_cycle option sets the period of scanning expired primary node network packets, in ms.

The max_queue_size option sets the max compare queue size.

If you want to use Xen COLO, you need to specify notify_dev to tell colo-compare how to notify Xen colo-frame to do a checkpoint.

COLO-compare must be used with the help of filter-mirror, filter-redirector and filter-rewriter. It does not solo, ever.

KVM COLO

...and the whole COLO setup is literally right here, bestie — both sides, full send 🫶

primary:
-netdev tap,id=hn0,vhost=off
-device e1000,id=e0,netdev=hn0,mac=52:a4:00:12:78:66
-chardev socket,id=mirror0,host=3.3.3.3,port=9003,server=on,wait=off
-chardev socket,id=compare1,host=3.3.3.3,port=9004,server=on,wait=off
-chardev socket,id=compare0,host=3.3.3.3,port=9001,server=on,wait=off
-chardev socket,id=compare0-0,host=3.3.3.3,port=9001
-chardev socket,id=compare_out,host=3.3.3.3,port=9005,server=on,wait=off
-chardev socket,id=compare_out0,host=3.3.3.3,port=9005
-object iothread,id=iothread1
-object filter-mirror,id=m0,netdev=hn0,queue=tx,outdev=mirror0
-object filter-redirector,netdev=hn0,id=redire0,queue=rx,indev=compare_out
-object filter-redirector,netdev=hn0,id=redire1,queue=rx,outdev=compare0
-object colo-compare,id=comp0,primary_in=compare0-0,secondary_in=compare1,outdev=compare_out0,iothread=iothread1

and the backup bestie:

secondary:
-netdev tap,id=hn0,vhost=off
-device e1000,netdev=hn0,mac=52:a4:00:12:78:66
-chardev socket,id=red0,host=3.3.3.3,port=9003
-chardev socket,id=red1,host=3.3.3.3,port=9004
-object filter-redirector,id=f1,netdev=hn0,queue=tx,indev=red0
-object filter-redirector,id=f2,netdev=hn0,queue=rx,outdev=red1

🌈 Xen COLO — same energy, except the primary also brings a notify_way chardev and hands notify_dev=notify_way to the colo-compare 🌸

primary:
-netdev tap,id=hn0,vhost=off
-device e1000,id=e0,netdev=hn0,mac=52:a4:00:12:78:66
-chardev socket,id=mirror0,host=3.3.3.3,port=9003,server=on,wait=off
-chardev socket,id=compare1,host=3.3.3.3,port=9004,server=on,wait=off
-chardev socket,id=compare0,host=3.3.3.3,port=9001,server=on,wait=off
-chardev socket,id=compare0-0,host=3.3.3.3,port=9001
-chardev socket,id=compare_out,host=3.3.3.3,port=9005,server=on,wait=off
-chardev socket,id=compare_out0,host=3.3.3.3,port=9005
-chardev socket,id=notify_way,host=3.3.3.3,port=9009,server=on,wait=off
-object filter-mirror,id=m0,netdev=hn0,queue=tx,outdev=mirror0
-object filter-redirector,netdev=hn0,id=redire0,queue=rx,indev=compare_out
-object filter-redirector,netdev=hn0,id=redire1,queue=rx,outdev=compare0
-object iothread,id=iothread1
-object colo-compare,id=comp0,primary_in=compare0-0,secondary_in=compare1,outdev=compare_out0,notify_dev=notify_way,iothread=iothread1

secondary:
-netdev tap,id=hn0,vhost=off
-device e1000,netdev=hn0,mac=52:a4:00:12:78:66
-chardev socket,id=red0,host=3.3.3.3,port=9003
-chardev socket,id=red1,host=3.3.3.3,port=9004
-object filter-redirector,id=f1,netdev=hn0,queue=tx,indev=red0
-object filter-redirector,id=f2,netdev=hn0,queue=rx,outdev=red1

If you want the tea on those command lines, read the colo-compare git log — it's all in there, iykyk 📜✨

-object cryptodev-backend-builtin,id=id[,queues=queues] — creates a cryptodev backend that runs crypto operations straight from the QEMU cipher APIs 💜. The id param is a unique ID that you'll use to reference this cryptodev backend from the virtio-crypto device. The queues param is optional and sets the queue number of the cryptodev backend — the default of queues is 1.

# qemu-system-x86_64 \
  [...] \
      -object cryptodev-backend-builtin,id=cryptodev0 \
      -device virtio-crypto-pci,id=crypto0,cryptodev=cryptodev0 \
  [...]

-object cryptodev-vhost-user,id=id,chardev=chardevid[,queues=queues] — makes a vhost-user cryptodev backend, backed by a chardev chardevid 🦋. The id param is a unique ID that will be used to reference this cryptodev backend from the virtio-crypto device. That chardev should be a unix domain socket backed one. vhost-user uses a specifically defined protocol to pass vhost ioctl replacement messages to an application on the other end of the socket. The queues param is optional and specifies the queue number of cryptodev backend for multiqueue vhost-user — the default of queues is 1.

# qemu-system-x86_64 \
  [...] \
      -chardev socket,id=chardev0,path=/path/to/socket \
      -object cryptodev-vhost-user,id=cryptodev0,chardev=chardev0 \
      -device virtio-crypto-pci,id=crypto0,cryptodev=cryptodev0 \
  [...]

-object secret,id=id,data=string,format=raw|base64[,keyid=secretid,iv=string]
-object secret,id=id,file=filename,format=raw|base64[,keyid=secretid,iv=string] — defines a secret to stash a password, encryption key, or some other sensitive data 🔐. The sensitive data can either be passed directly via the data parameter, or indirectly via the file parameter. Heads up: using the data parameter is insecure unless the sensitive data is encrypted.

The sensitive data can be provided in raw format (which is the default), or base64. When it's encoded as JSON, the raw format only supports valid UTF-8 characters, so base64 is recommended for sending binary data. QEMU converts from whichever format you provide to the format it needs internally — eg, an RBD password can be provided in raw format, even though it will be base64 encoded when passed onto the RBD server.

For added protection, you can encrypt the data attached to a secret using the AES-256-CBC cipher, no cap 🎀. You indicate encryption by providing the keyid and iv parameters. The keyid parameter gives the ID of a previously defined secret that contains the AES-256 decryption key — this key should be 32-bytes long and be base64 encoded. The iv parameter provides the random initialization vector used for encryption of this particular secret, and should be a base64 encrypted string of the 16-byte IV.

The simplest (insecure) usage is to provide the secret inline:

# qemu-system-x86_64 -object secret,id=sec0,data=letmein,format=raw

The simplest secure usage is to provide the secret via a file:

# printf "letmein" > mypasswd.txt # QEMU_SYSTEM_MACRO -object secret,id=sec0,file=mypasswd.txt,format=raw

For greater security, AES-256-CBC should be used 💅. To illustrate, consider the openssl command line tool, which can encrypt the data. Note that when encrypting, the plaintext must be padded to the cipher block size (32 bytes) using the standard PKCS#5/6 compatible padding algorithm.

First a master key needs to be created in base64 encoding:

# openssl rand -base64 32 > key.b64
# KEY=$(base64 -d key.b64 | hexdump  -v -e '/1 "%02X"')

Each secret to be encrypted needs to have a random initialization vector generated — these do not need to be kept secret:

# openssl rand -base64 16 > iv.b64
# IV=$(base64 -d iv.b64 | hexdump  -v -e '/1 "%02X"')

The secret to be defined can now be encrypted — in this case we're telling openssl to base64 encode the result, but it could be left as raw bytes if you wanted:

# SECRET=$(printf "letmein" |
           openssl enc -aes-256-cbc -a -K $KEY -iv $IV)

When launching QEMU, create a master secret pointing to key.b64 and specify that to be used to decrypt the user password. Pass the contents of iv.b64 to the second secret:

# qemu-system-x86_64 \
    -object secret,id=secmaster0,format=base64,file=key.b64 \
    -object secret,id=sec0,keyid=secmaster0,format=base64,\
        data=$SECRET,iv=$(<iv.b64)

-object sev-guest,id=id,cbitpos=cbitpos,reduced-phys-bits=val,[sev-device=string,policy=policy,handle=handle,dh-cert-file=file,session-file=file,kernel-hashes=on|off] — creates a Secure Encrypted Virtualization (SEV) guest object, which can be used to provide the guest memory encryption support on AMD processors 👑.

When memory encryption is enabled, one of the physical address bits (aka the C-bit) is utilized to mark if a memory page is protected. The cbitpos is used to provide the C-bit position — and because the C-bit position is Host family dependent, you (yes, you!) must provide this value. On EPYC, the value should be 47.

When memory encryption is enabled, we lose certain bits in physical address space. The reduced-phys-bits is used to provide the number of bits we lose in physical address space. Similar to C-bit, the value is Host family dependent. On EPYC, a guest will lose a maximum of 1 bit, so the value should be 1.

The sev-device provides the device file to use for communicating with the SEV firmware running inside the AMD Secure Processor. The default device is /dev/sev. If the hardware supports memory encryption then /dev/sev devices are created by the CCP driver.

The policy provides the guest policy to be enforced by the SEV firmware and restricts what configuration and operational commands can be performed on this guest by the hypervisor 🔥. The policy should be provided by the guest owner and is bound to the guest and cannot be changed throughout the lifetime of the guest. The default is 0.

okay so if guest policy allows sharing the key with another SEV guest, then handle can be used to provide the handle of the guest from which to share the key. 🫶

The dh-cert-file and session-file provide the guest owner's Public Diffie-Hillman key defined in SEV spec. The PDH and session parameters are used for establishing a cryptographic session with the guest owner to negotiate keys used for attestation. The file must be encoded in base64. no cap, base64 is the vibe here.

The kernel-hashes adds the hashes of given kernel/initrd/cmdline to a designated guest firmware page for measured Linux boot with -kernel. The default is off. (Since 6.2) ✨

e.g. to launch a SEV guest:

# qemu-system-x86_64 \
    ...... \
    -object sev-guest,id=sev0,cbitpos=47,reduced-phys-bits=1 \
    -machine ...,confidential-guest-support=sev0 \
    .....

-object tdx-guest,id=id,[attributes=attrs,sept-ve-disable=on|off,mrconfigid=sha384_digest,mrowner=sha384_digest,mrownerconfig=sha384_digest,quote-generation-socket=socketaddr] — creates an Intel Trusted Domain eXtensions (TDX) guest object, which is the type of confidential-guest-support object. When you pass the object ID to the machine's confidential-guest-support property, it can create a TDX guest. 👑
- attributes — a 64-bit integer, which specifies the TD attributes of the TD.
- sept-ve-disable — controls the bit 28 of TD attributes specifically. When it's on, the EPT violation conversion to #VE on guest access of PENDING pages is disabled. Some guest OS (e.g., Linux TD guest) may require this to be set, otherwise they refuse to boot. The default value is on. literally do not skip this if your guest is picky.
- mrconfigid — base64 encoded SHA384 digest, which provides the ID for non-owner-defined configuration of the guest TD, e.g., run-time or OS configuration. The default value is all zeros.
- mrowner — base64 encoded SHA384 digest, which provides the ID for guest TD's owner. The default value is all zeros.
- mrownerconfig — base64 encoded SHA384 digest, which provides the ID for owner-defined configuration of the guest TD, e.g., the configuration specific to the workload rather than the run-time of OS. The default value is all zeros.
- quote-generation-socket — specifies the socket address of the Quote Generation Service (QGS). QGS is a daemon running on the host. QEMU forwards the <GetQuote> request from TD guest to QGS and sends the reply (which contains generated QUOTE on success) from QGS to guest TD.

# qemu-system-x86_64 \
    ...... \
    -object '{"qom-type":"tdx-guest","id":"tdx","quote-generation-socket":{"type":"unix","path":"/var/run/qgs.socket"}}' \
    -machine ...,confidential-guest-support=tdx \
    ......

-object igvm-cfg,file=file — creates an IGVM configuration object that defines the initial state of the guest using a file that conforms to the Independent Guest Virtual Machine (IGVM) file format. 🦋

This is currently only supported by -machine q35 and -machine pc.

The file parameter is used to specify the IGVM file to load. When provided, the IGVM file is used to populate the initial memory of the virtual machine and, depending on the platform, can define the initial processor state, memory map and parameters.

The IGVM file is expected to contain the firmware for the virtual machine, therefore an igvm-cfg object cannot be provided along with other ways of specifying firmware, such as the -bios parameter on x86 machines. that's a hard no on mixing firmware sources, bestie.

e.g. to launch a machine providing the firmware in an IGVM file:

# qemu-system-x86_64 \
    ...... \
    -object igvm-cfg,id=igvm0,file=bios.igvm \
    -machine ...,igvm-cfg=igvm0 \
    .....

-object authz-simple,id=id,identity=string — creates an authorization object that will control access to network services. 🎀

The identity parameter identifies the user, and its format depends on the network service that authorization object is associated with. For authorizing based on TLS x509 certificates, the identity must be the x509 distinguished name.

⚠️ Note that care must be taken to escape any commas in the distinguished name. fr fr, those commas are not decorative.

An example authorization object to validate a x509 distinguished name would look like:

# qemu-system-x86_64 \
    ... \
    -object 'authz-simple,id=auth0,identity=CN=laptop.example.com,,O=Example Org,,L=London,,ST=London,,C=GB' \
    ...

Note the use of quotes due to the x509 distinguished name containing whitespace, and escaping of ','. iykyk.

-object authz-listfile,id=id,filename=path,refresh=on|off — creates an authorization object that will control access to network services. 💜

The filename parameter is the fully qualified path to a file containing the access control list rules in JSON format.

An example set of rules that match against SASL usernames might look like:

{
  "rules": [
     { "match": "fred", "policy": "allow", "format": "exact" },
     { "match": "bob", "policy": "allow", "format": "exact" },
     { "match": "danb", "policy": "deny", "format": "glob" },
     { "match": "dan*", "policy": "allow", "format": "exact" },
  ],
  "policy": "deny"
}

When checking access the object will iterate over all the rules and the first rule to match will have its policy value returned as the result. If no rules match, then the default policy value is returned. first match wins, periodt.

The rules can either be an exact string match, or they can use the simple UNIX glob pattern matching to allow wildcards to be used.

If refresh is set to true the file will be monitored and automatically reloaded whenever its content changes.

As with the authz-simple object, the format of the identity strings being matched depends on the network service, but is usually a TLS x509 distinguished name, or a SASL username.

An example authorization object to validate a SASL username would look like:

# qemu-system-x86_64 \
    ... \
    -object authz-simple,id=auth0,filename=/etc/qemu/vnc-sasl.acl,refresh=on \
    ...

-object authz-pam,id=id,service=string — creates an authorization object that will control access to network services. 🌸

The service parameter provides the name of a PAM service to use for authorization. It requires that a file /etc/pam.d/service exist to provide the configuration for the account subsystem.

An example authorization object to validate a TLS x509 distinguished name would look like:

# qemu-system-x86_64 \
    ... \
    -object authz-pam,id=auth0,service=qemu-vnc \
    ...

There would then be a corresponding config file for PAM at /etc/pam.d/qemu-vnc that contains:

account requisite  pam_listfile.so item=user sense=allow \
           file=/etc/qemu/vnc.allow

Finally the /etc/qemu/vnc.allow file would contain the list of x509 distinguished names that are permitted access:

CN=laptop.example.com,O=Example Home,L=London,ST=London,C=GB

-object iothread,id=id,poll-max-ns=poll-max-ns,poll-grow=poll-grow,poll-shrink=poll-shrink,poll-weight=poll-weight,aio-max-batch=aio-max-batch — creates a dedicated event loop thread that devices can be assigned to. This is known as an IOThread. By default device emulation happens in vCPU threads or the main event loop thread. This can become a scalability bottleneck. IOThreads allow device emulation and I/O to run on other host CPUs. 🔥

The id parameter is a unique ID that will be used to reference this IOThread from -device ...,iothread=id. Multiple devices can be assigned to an IOThread. Note that not all devices support an iothread parameter. vibe check your device first.

The query-iothreads QMP command lists IOThreads and reports their thread IDs so that the user can configure host CPU pinning/affinity.

IOThreads use an adaptive polling algorithm to reduce event loop latency. Instead of entering a blocking system call to monitor file descriptors and then pay the cost of being woken up when an event occurs, the polling algorithm spins waiting for events for a short time. The algorithm's default parameters are suitable for many cases but can be adjusted based on knowledge of the workload and/or host device latency.

The poll-max-ns parameter is the maximum number of nanoseconds to busy wait for events. Polling can be disabled by setting this value to 0. ⭐

poll-grow is the multiplier that bumps up the polling time when the algorithm catches itself missing events because it wasn’t polling long enough. 💅

poll-shrink is the divisor that decreases the polling time when the algorithm realizes it’s spending way too long polling without hitting any events. 🫶

poll-weight is the weight factor for adaptive polling. It decides how much the most recent event interval affects the next polling duration calculation. If you set it to 0, the system default value of 3 is used. Typical values: 1 (high weight on recent interval), 2-4 (moderate weight on recent interval). ✨

aio-max-batch is the maximum number of requests in a batch for the AIO engine. 0 means the engine will use its default. 👑

The IOThread parameters can be modified at run-time using the qom-set command (where iothread1 is the IOThread’s id):

(qemu) qom-set /objects/iothread1 poll-max-ns 100000

During graphical emulation, you can use special key combinations from the following table to change modes. By default the modifier is Ctrl+Alt (used in the table below) which can be changed with -display suboption mod= where appropriate. For example, -display sdl,grab-mod=lshift-lctrl-lalt changes the modifier key to Ctrl+Alt+Shift, while -display sdl,grab-mod=rctrl changes it to the right Ctrl key. 🎀

🎮 Multiplexer Keys

Ctrl+Alt+f — Toggle full screen
Ctrl+Alt++ — Enlarge the screen
Ctrl+Alt+- — Shrink the screen
Ctrl+Alt+0 — Restore the screen’s un-scaled dimensions
Ctrl+Alt+n — Switch to virtual console 'n'. Standard console mappings are:
- 1: Target system display
- 2: Monitor
- 3: Serial port
Ctrl+Alt+g — Toggle mouse and keyboard grab.

In the virtual consoles, you can use Ctrl+Up, Ctrl+Down, Ctrl+PageUp and Ctrl+PageDown to move in the back log. 🦋

During emulation, if you are using a character backend multiplexer (which is the default if you are using -nographic) then several commands are available via an escape sequence. These key sequences all start with an escape character, which is Ctrl+a by default, but can be changed with -echr. The list below assumes you’re using the default. 🌸

🕹️ Multiplexer Keys

Ctrl+a h — Print this help
Ctrl+a x — Exit emulator
Ctrl+a s — Save disk data back to file (if -snapshot)
Ctrl+a t — Toggle console timestamps
Ctrl+a b — Send break (magic sysrq in Linux)
Ctrl+a c — Rotate between the frontends connected to the multiplexer (usually this switches between the monitor and the console)
Ctrl+a Ctrl+a — Send the escape character to the frontend

🌐 NOTES 🌐

okay so real talk bestie — you're not limited to boring local image files for your emulated storage devices 💾✨ QEMU can also pull from networked resources like iSCSI devices, and those get specified using a special URL syntax. iykyk 🫶

🧚 iSCSI

iSCSI support means QEMU can reach iSCSI resources directly and use them as images for the guest storage — and yes, both disk</> and cdrom images are supported 👑

Syntax for specifying iSCSI LUNs is:

"iscsi://<target-ip>[:<port>]/<target-iqn>/<lun>"

By default qemu will use the iSCSI initiator-name 'iqn.2008-11.org.linux-kvm[:<name>]' — but this can also be set from the command line or a configuration file, so you're not stuck with the default 💅

Since version QEMU 2.4 it's possible to specify an iSCSI request timeout to detect stalled requests and force a reestablishment of the session ⏰ The timeout is specified in seconds. The default is 0, which means no timeout. Libiscsi 1.15.0 or greater is required for this feature, no cap 📌

Example (without authentication):

qemu-system-x86_64 -iscsi initiator-name=iqn.2001-04.com.example:my-initiator \
                 -cdrom iscsi://192.0.2.1/iqn.2001-04.com.example/2 \
                 -drive file=iscsi://192.0.2.1/iqn.2001-04.com.example/1

Example (CHAP username/password via URL):

qemu-system-x86_64 -drive file=iscsi://user%password@192.0.2.1/iqn.2001-04.com.example/1

Example (CHAP username/password via environment variables):

LIBISCSI_CHAP_USERNAME="user" \
LIBISCSI_CHAP_PASSWORD="password" \
qemu-system-x86_64 -drive file=iscsi://192.0.2.1/iqn.2001-04.com.example/1

🦋 NBD

QEMU supports NBD (Network Block Devices) both using the TCP protocol as well as Unix Domain Sockets. With TCP, the default port is 10809 📞

Syntax for specifying a NBD device using TCP, in the preferred URI form:

"nbd://<server-ip>[:<port>]/[<export>]"

Syntax for specifying a NBD device using Unix Domain Sockets — and bestie, remember that '?' is a shell glob character so it may need quoting 🙈:

"nbd+unix:///[<export>]?socket=<domain-socket>"

Older syntax that is also recognized:

"nbd:<server-ip>:<port>[:exportname=<export>]"

Syntax for specifying a NBD device using Unix Domain Sockets:

"nbd:unix:<domain-socket>[:exportname=<export>]"

Example for TCP

qemu-system-x86_64 --drive file=nbd:192.0.2.1:30000

Example for Unix Domain Sockets

qemu-system-x86_64 --drive file=nbd:unix:/tmp/nbd-socket

🎀 SSH

QEMU supports SSH (Secure Shell) access to remote disks, periodt 🔐

Examples:

qemu-system-x86_64 -drive file=ssh://user@host/path/to/disk.img
qemu-system-x86_64 -drive file.driver=ssh,file.user=user,file.host=host,file.port=22,file.path=/path/to/disk.img

Currently authentication must be done using ssh-agent. Other authentication methods may be supported in future — so watch this space 👀

🌈 HTTP/HTTPS/FTP/FTPS

QEMU supports read-only access to files accessed over http(s) and ftp(s) 🔥

Syntax using a single filename:

<protocol>://[<username>[:<password>]@]<host>/<path>

where:

protocol — 'http', 'https', 'ftp', or 'ftps' 💅
username — optional username for authentication to the remote server 🙋‍♀️
password — optional password for authentication to the remote server 🤫
host — address of the remote server 🖥️
path — path on the remote server, including any query string 🛤️

The following options are also supported:

url — the full URL when passing options to the driver explicitly 🔗
readahead — the amount of data to read ahead with each range request to the remote server. This value may optionally have the suffix 'T', 'G', 'M', 'K', 'k' or 'b'. If it does not have a suffix, it will be assumed to be in bytes. The value must be a multiple of 512 bytes. It defaults to 256k ⚡
sslverify — whether to verify the remote server's certificate when connecting over SSL. It can have the value 'on' or 'off'. It defaults to 'on' ✅
cookie — send this cookie (it can also be a list of cookies separated by ';') with each outgoing request. Only supported when using protocols such as HTTP which support cookies, otherwise ignored 🍪
timeout — set the timeout in seconds of the CURL connection. This timeout is the time that CURL waits for a response from the remote server to get the size of the image to be downloaded. If not set, the default timeout of 5 seconds is used ⏳
force-range — don't issue a HEAD HTTP request to discover if the http server supports range requests and rely only on GET requests. This is especially useful for S3 presigned URLs where HEAD requests are unauthorized. It defaults to 'false' 🚫

Note that when passing options to qemu explicitly, driver is the value of <protocol> — don't forget it bestie 💖

Example: boot from a remote Fedora 20 live ISO image

qemu-system-x86_64 --drive media=cdrom,file=https://archives.fedoraproject.org/pub/archive/fedora/linux/releases/20/Live/x86_64/Fedora-Live-Desktop-x86_64-20-1.iso,readonly

qemu-system-x86_64 --drive media=cdrom,file.driver=http,file.url=http://archives.fedoraproject.org/pub/fedora/linux/releases/20/Live/x86_64/Fedora-Live-Desktop-x86_64-20-1.iso,readonly

Example: boot from a remote Fedora 20 cloud image using a local overlay for writes, copy-on-read, and a readahead of 64k

qemu-img create -f qcow2 -o backing_file='json:{"file.driver":"http",, "file.url":"<http://archives.fedoraproject.org/pub/archive/fedora/linux/releases/20/Images/x86_64/Fedora-x86_64-20-20131211.1-sda.qcow2>",, "file.readahead":"64k"}' /tmp/Fedora-x86_64-20-20131211.1-sda.qcow2

qemu-system-x86_64 -drive file=/tmp/Fedora-x86_64-20-20131211.1-sda.qcow2,copy-on-read=on

Example: boot from an image stored on a VMware vSphere server with a self-signed certificate using a local overlay for writes, a readahead of 64k and a timeout of 10 seconds

qemu-img create -f qcow2 -o backing_file='json:{"file.driver":"https",, "file.url":"<https://user:password@vsphere.example.com/folder/test/test-flat.vmdk?dcPath=Datacenter&dsName=datastore1>",, "file.sslverify":"off",, "file.readahead":"64k",, "file.timeout":10}' /tmp/test.qcow2

qemu-system-x86_64 -drive file=/tmp/test.qcow2

👀 SEE ALSO 👀

The HTML documentation of QEMU for more precise information and Linux user mode emulator invocation — go read it, it's giving ✨ extra ✨ detail 🫶

👑 Author 👑

Fabrice Bellard — the icon, the legend 💜

📜 Copyright 📜

and that's a wrap bestie!! 🎉💖 this page was brought to you by QEMU version 11.1.3, dated Oct 29, 2026 — go forth and virtualize, you absolute main character 😭🔥💅

📦 where this page comes from

arch rolling: qemu-common 11.1.3-1 · BSD-2-Clause AND BSD-2-Clause-Patent AND BSD-3-Clause AND CC0-1.0 AND FSFAP AND GPL-1.0-or-later AND GPL-2.0-only AND GPL-2.0-only WITH Linux-syscall-note AND GPL-2.0-only WITH Linux-syscall-note OR BSD-2-Clause AND GPL-2.0-only WITH Linux-syscall-note OR BSD-3-Clause AND GPL-2.0-or-later AND GPL-2.0-or-later OR BSD-3-Clause AND GPL-2.0-or-later WITH Linux-syscall-note AND ISC AND LGPL-2.1-only OR BSD-3-Clause AND LGPL-2.0-or-later AND LGPL-2.1-or-later AND LicenseRef-Public-Domain AND MIT
💖 girlified by deepseek-ai/DeepSeek-V4.1-Flash-1017 on 2026-11-04: a language model's rewrite of the original, under its terms

QEMU(1)

✨ QEMU ✨

✨ NAME ✨

✨ SYNOPSIS ✨

✨ DESCRIPTION ✨

🎀 Machine-specific options 🎀

✨ OPTIONS ✨

🌈 Standard options 🌈

🦋 cpu — the new-and-improved way to place CPUs

🍓 memdev — the one you should actually be using

🧚 the legacy mem situation

🫶 initiator — the bestie node

📏 distances — the math part, stay with me

🌈 hmat-lb — latency and bandwidth, but fancy

🎀 hmat-cache — cache info, but make it structured

💅 -add-fd fd=fd,set=set[,opaque=opaque]

⭐ -set group.id.arg=value

🔥 -global driver.prop=value / -global driver=driver,property=property,value=value

🧚 -boot

🧚 -device ipmi-bmc-sim,id=id[,prop=value[,...]]

🦋 -device ipmi-bmc-extern,id=id,chardev=id[,slave_addr=val]

🎀 -device isa-ipmi-kcs,bmc=id[,ioport=val][,irq=val]

🍓 -device isa-ipmi-bt,bmc=id[,ioport=val][,irq=val]

🌸 -device pci-ipmi-kcs,bmc=id

🌈 -device pci-ipmi-bt,bmc=id

👑 -device intel-iommu[,option=...]

🧚 -device virtio-iommu-pci[,option=...]

💜 -device arm-smmuv3,primary-bus=id[,option=...]

🦋 -device amd-iommu[,option=...]

💾 Block device options 💾

🗂️ Driver-specific options for file 🗂️

🧊 Driver-specific options for raw 🧊

🦋 Driver-specific options for qcow2 🦋

🌈 Driver-specific options for other drivers

✨ -drive option[,option[,option[,...]]] ✨

🌸 -mtdblock file

🎀 -sd file

🦋 -snapshot

💜 -fsdev

🧚 -device virtio-9p-type,fsdev=id,mount_tag=mount_tag

🌈 -virtfs

🔌 USB convenience options 🔌

🖥️ Display options 🖥️

💻 i386 target only 💻

🌐 Network options 🌐

-nic [tap|passt|bridge|user|l2tpv3|vde|netmap|af-xdp|vhost-user|socket][,...][,mac=macaddr][,model=mn]

-nic none

-netdev passt,id=str[,option][,...]

-netdev user,id=id[,option][,option][,...]

🌈 hostfwd=[tcp|udp|unix]:[[hostaddr]:hostport|hostpath]-[guestaddr]:guestport

🌀 guestfwd=[tcp]:server:port-dev<br>guestfwd=[tcp]:server:port-cmd:command

🧚 -netdev tap,id=id[,fd=h][,ifname=name][,script=file][,downscript=dfile][,br=bridge][,helper=helper]

🌉 -netdev bridge,id=id[,br=bridge][,helper=helper]

🔌 -netdev socket,id=id[,fd=h][,listen=[host]:port][,connect=host:port]

📡 -netdev socket,id=id[,fd=h][,mcast=maddr:port[,localaddr=addr]]

🌊 -netdev stream,id=str[,server=on|off],addr.type=inet,addr.host=host,addr.port=port[,to=maxport][,numeric=on|off][,keep-alive=on|off][,mptcp=on|off][,addr.ipv4=on|off][,addr.ipv6=on|off][,reconnect-ms=milliseconds]

🐚 -netdev stream,id=str[,server=on|off],addr.type=unix,addr.path=path[,abstract=on|off][,tight=on|off][,reconnect-ms=milliseconds]

💅 tight=on|off

🔁 reconnect-ms=milliseconds

🌈 -netdev stream,id=str[,server=on|off],addr.type=fd,addr.str=file-descriptor[,reconnect-ms=milliseconds]

📡 -netdev dgram,id=str,remote.type=inet,remote.host=maddr,remote.port=port[,local.type=inet,local.host=addr]

🧚 -netdev dgram,id=str,remote.type=inet,remote.host=maddr,remote.port=port[,local.type=fd,local.str=file-descriptor]

🍓 -netdev dgram,id=str,local.type=inet,local.host=addr,local.port=port[,remote.type=inet,remote.host=addr,remote.port=port]

🎀 -netdev dgram,id=str,local.type=unix,local.path=path[,remote.type=unix,remote.path=path]

🦋 -netdev dgram,id=str,local.type=fd,local.str=file-descriptor

👑 -netdev l2tpv3,id=id,src=srcaddr,dst=dstaddr[,srcport=srcport][,dstport=dstport],txsession=txsession[,rxsession=rxsession][,ipv6=on|off][,udp=on|off][,cookie64=on|off][,counter=on|off][,pincounter=on|off][,txcookie=txcookie][,rxcookie=rxcookie][,offset=offset]

🌸 -netdev vde,id=id[,sock=socketpath][,port=n][,group=groupname][,mode=octalmode]

⭐ -netdev af-xdp,id=str,ifname=name[,mode=native|skb][,force-copy=on|off][,queues=n][,start-queue=m][,inhibit=on|off][,sock-fds=x:y:...:z][,map-path=/path/to/socket/map][,map-start-index=i]

🎀 Character device options 🎀

💜 The available backends are:

🎀 TPM device options 🎀

🥾 Boot Image or Kernel specific 🥾

🌈 Debug/Expert options 🌈

🎀 Generic object creation 🎀

🌸 -object monitor-hmp,id=id,chardev=chardev_id,readline=on|off

💜 -object monitor-qmp,id=id,chardev=chardev_id,pretty=on|off,close-action=none|delete

🦋 -object memory-backend-file,id=id,size=size,mem-path=dir,share=on|off,discard-data=on|off,merge=on|off,dump=on|off,prealloc=on|off,host-nodes=host-nodes,policy=default|preferred|bind|interleave,align=align,offset=offset,readonly=on|off,rom=on|off|auto

🎮 Multiplexer Keys

🕹️ Multiplexer Keys

🌐 NOTES 🌐

🦋 `cpu` — the new-and-improved way to place CPUs

🍓 `memdev` — the one you should actually be using

🧚 the legacy `mem` situation

🫶 `initiator` — the bestie node

🌈 `hmat-lb` — latency and bandwidth, but fancy

🎀 `hmat-cache` — cache info, but make it structured

💅 `-add-fd fd=fd,set=set[,opaque=opaque]`

⭐ `-set group.id.arg=value`

🔥 `-global driver.prop=value` / `-global driver=driver,property=property,value=value`

🧚 `-boot`

🧚 `-device ipmi-bmc-sim,id=id[,prop=value[,...]]`

🦋 `-device ipmi-bmc-extern,id=id,chardev=id[,slave_addr=val]`

🎀 `-device isa-ipmi-kcs,bmc=id[,ioport=val][,irq=val]`

🍓 `-device isa-ipmi-bt,bmc=id[,ioport=val][,irq=val]`

🌸 `-device pci-ipmi-kcs,bmc=id`

🌈 `-device pci-ipmi-bt,bmc=id`

👑 `-device intel-iommu[,option=...]`

🧚 `-device virtio-iommu-pci[,option=...]`

💜 `-device arm-smmuv3,primary-bus=id[,option=...]`

🦋 `-device amd-iommu[,option=...]`

🗂️ Driver-specific options for `file` 🗂️

🧊 Driver-specific options for `raw` 🧊

🦋 Driver-specific options for `qcow2` 🦋

✨ `-drive` `option[,option[,option[,...]]]` ✨

🌸 `-mtdblock` file

🎀 `-sd` file

🦋 `-snapshot`

💜 `-fsdev`

🧚 `-device virtio-9p-type,fsdev=id,mount_tag=mount_tag`

🌈 `-virtfs`

`-nic` `[tap|passt|bridge|user|l2tpv3|vde|netmap|af-xdp|vhost-user|socket][,...][,mac=macaddr][,model=mn]`

`-nic none`

`-netdev passt,id=str[,option][,...]`

`-netdev user,id=id[,option][,option][,...]`

🌈 `hostfwd=[tcp|udp|unix]:[[hostaddr]:hostport|hostpath]-[guestaddr]:guestport`

🌀 `guestfwd=[tcp]:server:port-dev`<br>`guestfwd=[tcp]:server:port-cmd:command`

🧚 `-netdev tap,id=id[,fd=h][,ifname=name][,script=file][,downscript=dfile][,br=bridge][,helper=helper]`

🌉 `-netdev bridge,id=id[,br=bridge][,helper=helper]`

🔌 `-netdev socket,id=id[,fd=h][,listen=[host]:port][,connect=host:port]`

📡 `-netdev socket,id=id[,fd=h][,mcast=maddr:port[,localaddr=addr]]`

🌊 `-netdev stream,id=str[,server=on|off],addr.type=inet,addr.host=host,addr.port=port[,to=maxport][,numeric=on|off][,keep-alive=on|off][,mptcp=on|off][,addr.ipv4=on|off][,addr.ipv6=on|off][,reconnect-ms=milliseconds]`

🐚 `-netdev stream,id=str[,server=on|off],addr.type=unix,addr.path=path[,abstract=on|off][,tight=on|off][,reconnect-ms=milliseconds]`

💅 `tight=on|off`

🔁 `reconnect-ms=milliseconds`

🌈 `-netdev stream,id=str[,server=on|off],addr.type=fd,addr.str=file-descriptor[,reconnect-ms=milliseconds]`

📡 `-netdev dgram,id=str,remote.type=inet,remote.host=maddr,remote.port=port[,local.type=inet,local.host=addr]`

🧚 `-netdev dgram,id=str,remote.type=inet,remote.host=maddr,remote.port=port[,local.type=fd,local.str=file-descriptor]`

🍓 `-netdev dgram,id=str,local.type=inet,local.host=addr,local.port=port[,remote.type=inet,remote.host=addr,remote.port=port]`

🎀 `-netdev dgram,id=str,local.type=unix,local.path=path[,remote.type=unix,remote.path=path]`

🦋 `-netdev dgram,id=str,local.type=fd,local.str=file-descriptor`

👑 `-netdev l2tpv3,id=id,src=srcaddr,dst=dstaddr[,srcport=srcport][,dstport=dstport],txsession=txsession[,rxsession=rxsession][,ipv6=on|off][,udp=on|off][,cookie64=on|off][,counter=on|off][,pincounter=on|off][,txcookie=txcookie][,rxcookie=rxcookie][,offset=offset]`

🌸 `-netdev vde,id=id[,sock=socketpath][,port=n][,group=groupname][,mode=octalmode]`

⭐ `-netdev af-xdp,id=str,ifname=name[,mode=native|skb][,force-copy=on|off][,queues=n][,start-queue=m][,inhibit=on|off][,sock-fds=x:y:...:z][,map-path=/path/to/socket/map][,map-start-index=i]`

🌸 `-object monitor-hmp,id=id,chardev=chardev_id,readline=on|off`

💜 `-object monitor-qmp,id=id,chardev=chardev_id,pretty=on|off,close-action=none|delete`

🦋 `-object memory-backend-file,id=id,size=size,mem-path=dir,share=on|off,discard-data=on|off,merge=on|off,dump=on|off,prealloc=on|off,host-nodes=host-nodes,policy=default|preferred|bind|interleave,align=align,offset=offset,readonly=on|off,rom=on|off|auto`