BHYVE(8)

📦
version 15.1 · 2 versions
, in freebsd 15.1

👑 bhyve — System Manager's Manual 👑

okay bestie, welcome to bhyve(8), straight out of the System Manager's Manual. This is the hypervisor one. Buckle up 💅

🌸 NAME 🌸

bhyve — run a guest operating system inside a virtual machine ✨

🦋 SYNOPSIS 🦋

bhyve [-aCDeHhPSuWwxY]
      [-c [[cpus=]numcpus][,sockets=n][,cores=n][,threads=n]] [-f
      name,[string|file]=data] [-G [w][bind_address:]port]
      [-k config_file] [-K layout] [-l lpcdev[,conf]]
      [-m memsize[K|k|M|m|G|g|T|t]] [-o var=value] [-p vcpu:hostcpu]
      [-r file] [-s slot,emulation[,conf]] [-U uuid] vmname
bhyve -l help
bhyve -s help

💖 DESCRIPTION 💖

so bhyve is a hypervisor — it runs guest operating systems inside a virtual machine. It can run guests on amd64 and arm64 platforms, as long as you've got the suitable hardware support. no cap 🫶

Stuff like the number of virtual CPUs, how much guest memory you get, and I/O connectivity? All of that is set with command-line parameters. iconic.

bhyve is typically used with a boot ROM that can load the guest operating system. On arm64 platforms, this is currently required. If you're not using a boot ROM, the guest operating system must be loaded with bhyveload(8) or a similar boot loader before running bhyve, otherwise. On amd64, the edk2-bhyve package provides a UEFI firmware that can be used to boot the guest; on arm64 the u-boot-bhyve-arm64 package provides a U-Boot image that can be used to boot the guest. 🧚

bhyve runs until the guest operating system reboots or an unhandled hypervisor exit is detected. that's the whole lifecycle, periodt.

✨ OPTIONS ✨

  • -a — the guest's local APIC is configured in xAPIC mode. This option only applies to the amd64 platform. Heads up though: xAPIC mode is the default setting, so this option is redundant 🙈 and it will be deprecated in a future version.
  • -C — include guest memory in core files.
  • -c [[cpus=]numcpus][,sockets=n][,cores=n][,threads=n] — number of guest virtual CPUs and/or the CPU topology. The default value for each of numcpus, sockets, cores, and threads is 1. If numcpus is not specified then it gets calculated from the other arguments. The topology has to be consistent — meaning numcpus must equal the product of sockets, cores, and threads. And if a setting is specified more than once, the last one has precedence (last one wins, always 💅).

The maximum number of virtual CPUs defaults to the number of active physical CPUs in the system, available via the hw.vmm.maxcpu sysctl(8) variable. That limit can be adjusted via the hw.vmm.maxcpu loader tunable.

  • -D — destroy the VM on guest initiated power-off. bye bye 👋
  • -e — force bhyve to exit when a guest issues an access to an I/O port that is not emulated. This is intended for debug purposes and only applies to the amd64 platform.
  • -f name,[string|file]=data — add a fw_cfg file name to the fw_cfg interface. If a string is specified, the fw_cfg file contains the string as data. If a file is specified, bhyve reads the file and adds the file content as fw_cfg data.
  • -G [w][bind_address:]port — start a debug server that uses the GDB protocol to export guest state to a debugger. An IPv4 TCP socket will be bound to the supplied bind_address and port to listen for debugger connections. Only a single debugger may be attached to the debug server at a time. And if the option begins with w, bhyve will pause execution at the first instruction, waiting for a debugger to attach 🫶
  • -H — yield the virtual CPU thread when a HLT instruction is detected. If this option is not specified, virtual CPUs will use 100% of a host CPU 😭. This option applies only to the amd64 platform.
  • -h — print help message and exit.
  • -k config_file — set configuration variables from a simple, key-value config file. Each line of the config file is expected to consist of a config variable name, an equals sign (=), and a value. No spaces are permitted between the variable name, equals sign, or value. Blank lines and lines starting with # are ignored. See bhyve_config(5) for more details.
  • -K layout — specify the keyboard layout. The value you can specify sets the file name in /usr/share/bhyve/kbdlayout. This specification only works when loaded with UEFI mode for VNC. When using a VNC client that supports QEMU Extended Key Event Message (e.g. TigerVNC), this option isn't needed. When using a VNC client that doesn't support QEMU Extended Key Event Message (e.g. TightVNC), the layout defaults to the US keyboard unless specified otherwise. ty keyboard girlies 🎀
  • -l help — print a list of supported LPC devices.
  • -l lpcdev[,conf] — allow devices behind the LPC PCI-ISA bridge to be configured. The only supported devices are the TTY-class devices com1, com2, com3, and com4, the TPM module tpm, the boot ROM device bootrom, the fwcfg type and the debug/test device pc-testdev.

The possible values for the conf argument are listed in the -s flag description.

This option applies only to the amd64 platform. On arm64, the console and boot ROM devices are configured using the more generic -o option.

  • -m memsize[K|k|M|m|G|g|T|t] — set the guest physical memory size. This must be the same size that was given to bhyveload(8).

The size argument may be suffixed with one of K, M, G or T (either upper or lower case) to indicate a multiple of kilobytes, megabytes, gigabytes, or terabytes. If no suffix is given, the value is assumed to be in megabytes. The default is 256M.

  • -n id,size,cpus[,domain_policy] — configure guest NUMA domains. This option applies only to the amd64 platform.

The -n option allows the guest physical address space to be partitioned into domains. The layout of each domain is encoded in an ACPI table visible to the guest operating system. The -n option also allows the specification of a domainset(9) memory allocation policy for the host memory backing a given NUMA domain. A guest can have up to 8 NUMA domains. This feature requires that the guest use a boot ROM, and in particular cannot be used if the guest was initialized using bhyveload(8). 💀

Each domain is identified by a numerical id. The domain memory size is specified using the same format as the -m flag. The sum of all size parameters overrides the total VM memory size specified by the -m flag. However, if at least one domain memory size parameter is missing, the total VM memory size will be equally distributed across all emulated domains. The cpuset parameter specifies the set of CPUs that are part of the domain. The domain_policy parameter may be optionally used to configure the domainset(9) host NUMA memory allocation policy for an emulated domain. See the -n flag in cpuset(1) for a list of valid NUMA memory allocation policies and their formats.

  • -o var=value — set the configuration variable var to value. See bhyve_config(5) for configuration options.
  • -P — force the guest virtual CPU to exit when a PAUSE instruction is detected. This option applies only to the amd64 platform.
  • -p vcpu:hostcpu — pin guest's virtual CPU vcpu to hostcpu. Host CPUs and guest virtual CPUs are numbered starting from 0. A -p option is required for every guest vCPU to be pinned. To map a 4 vCPU guest to host CPUs 12-15:
-p 0:12 -p 1:13 -p 2:14 -p 3:15

match made in heaven, honestly 🥰

  • -r file — resume a guest from a snapshot. The guest memory contents are restored from file, and the guest device and vCPU state are restored from the file “file.kern”.
Note that the current snapshot file format requires that the configuration of devices in the new VM match the VM from which the snapshot was taken by specifying the same -s and -l options. The count of vCPUs and memory configuration are read from the snapshot.
  • -S — wire guest memory.
  • -s help — print a list of supported PCI devices.
  • -s slot,emulation[,conf] — configure a virtual PCI slot and function.

bhyve provides PCI bus emulation and virtual devices that can be attached to slots on the bus. There are 32 available slots, with the option of providing up to 8 functions per slot.

The slot can be specified in one of the following formats:

  • pcislot
  • pcislot:function
  • bus:pcislot:function

The pcislot value is 0 to 31. The optional function value is 0 to 7. The optional bus value is 0 to 255. If not specified, the function value defaults to 0. If not specified, the bus value defaults to 0.

See “PCI EMULATION” for available options for the emulation argument. 👀

  • -U uuid — set the universally unique identifier (UUID) in the guest's System Management BIOS System Information structure. By default a UUID is generated from the host's hostname and vmname.
  • -u — RTC keeps UTC time. 🕰️
  • -W — force virtio PCI device emulations to use MSI interrupts instead of MSI-X interrupts.
  • -w — ignore accesses to unimplemented Model Specific Registers (MSRs). This is intended for debug purposes.
  • -x — the guest's local APIC is configured in x2APIC mode. This option applies only to the amd64 platform.
  • -Y — disable MPtable generation. This option applies only to the amd64 platform.
  • vmname — alphanumeric name of the guest. This should be the same as that created by bhyveload(8).

🌈 PCI EMULATION 🌈

bhyve provides emulation for various PCI devices. They are specified by the -s slot,emulation,conf configuration's emulation argument, which can be one of the following:

  • hostbridge — a simple host bridge. This is usually configured at slot 0, and is required by most guest operating systems. lowkey the main character of PCI slots 👑
  • amd_hostbridge — literally the same emulation as hostbridge, except the PCI vendor ID says AMD instead. bestie branding, that's it 💅
  • passthru — a PCI pass-through device. yes, the real one.
  • virtio-net — virtio network interface 🌐
  • virtio-blk — virtio block storage interface 💾
  • virtio-scsi — virtio SCSI interface
  • virtio-9p — virtio 9p (VirtFS) interface
  • virtio-rnd — virtio RNG interface 🎲
  • virtio-console — virtio console interface, which exposes multiple ports to the guest in the form of simple char devices for simple IO between the guest and host userspaces
  • virtio-input — virtio input interface
  • ahci — an AHCI controller attached to arbitrary devices
  • ahci-cd — an AHCI controller attached to an ATAPI CD/DVD 💿
  • ahci-hd — an AHCI controller attached to a SATA hard drive
  • e1000 — Intel e82545 network interface
  • uart — PCI 16550 serial device
  • lpc — LPC PCI-ISA bridge with COM1, COM2, COM3, and COM4 16550 serial ports, a boot ROM, and, optionally, a TPM module, a fw_cfg type, and the debug/test device. ngl the LPC bridge emulation can only be configured on bus 0 — that's the rule, no cap 🚫
  • fbuf — raw framebuffer device attached to a VNC server 🖼️
  • xhci — eXtensible Host Controller Interface (xHCI) USB controller 🔌
  • nvme — NVM Express (NVMe) controller ⚡
  • hda — High Definition Audio Controller 🔊

okay so the optional parameter conf describes the backend for device emulations. If conf is not specified, the device emulation has no backend and can be considered unconnected. it's giving "ghosted, but in a hardware way" 🙈

🔌 Network device backends 🔌

  • tapN[,mac=xx:xx:xx:xx:xx:xx][,mtu=N]
  • vmnetN[,mac=xx:xx:xx:xx:xx:xx][,mtu=N]
  • ngdN
  • netgraph,path=ADDRESS,peerhook=HOOK[,socket=NAME][,hook=HOOK][,mac=xx:xx:xx:xx:xx:xx][,mtu=N]
  • slirp[,open][,hostfwd=proto:hostaddr:hostport-guestaddr:guestport][,mac=xx:xx:xx:xx:xx:xx][,mtu=N]

If mac is not specified, the MAC address is derived from a fixed OUI, and the remaining bytes from an MD5 hash of the slot and function numbers and the device name. If specified, it must be a unicast MAC address. (no multicast shenanigans, bestie ✨)

The MAC address is an ASCII string in ethers(5) format.

A ngd device can be used to connect a guest to a netgraph(4) through a ng_device(4) node. This can be used to run bhyve in a VNET(9) jail, and give it access to the host's netgraph, that cannot be reached directly, by exposing the ng_device through devfs(8). lowkey genius 💖

With virtio-net devices, the mtu parameter can be specified to inform the guest about the largest MTU that should be allowed, expressed in bytes.

With the netgraph backend, the path and peerhook parameters must be specified to set the destination node and corresponding hook. The optional parameters socket and hook may be used to set the ng_socket(4) node name and source hook. The ADDRESS, HOOK, and NAME must comply with netgraph(4) addressing rules 📐

The slirp backend can be used to provide a NATed network to the guest. This backend has poor performance but does not require any network configuration on the host system 😭 it depends on the net/libslirp port. If the open keyword is set, the guest will be able to make outbound network connections, and bhyve will transparently handle the necessary address translation. The hostfwd option takes a 5-tuple describing how connections from the host are to be forwarded to the guest. Multiple rules can be specified, separated by semicolons. Note that semicolons must be escaped or quoted to prevent the shell from interpreting them 🐚 The backend will provide DHCP and DNS service to the guest.

💾 Block storage device backends 💾

  • /filename[,block-device-options]
  • /dev/xxx[,block-device-options]

The block-device-options are:

  • nocache — open the file with O_DIRECT.
  • direct — open the file using O_SYNC.
  • ro — force the file to be opened read-only 🔒
  • sectorsize=logical[/physical] — specify the logical and physical sector sizes of the emulated disk. The physical sector size is optional and is equal to the logical sector size if not explicitly specified. math is mathing 📏
  • nodelete — disable emulation of guest trim requests via DIOCGDELETE requests.
  • bootindex=index — add the device to the boot order at index. A fw_cfg file is used to specify the boot order. The guest firmware may ignore or doesn't support this fw_cfg file. In that case, this feature doesn't work as expected 🫠 (fr fr, don't be shocked when it does nothing)

🧵 SCSI device backends 🧵

  • /dev/cam/ctl[pp.vp][,scsi-device-options]

The scsi-device-options are:

  • iid=IID — initiator ID to use when sending requests to specified CTL port. The default value is 0.
  • bootindex=index — add the device to the boot order at index. A fw_cfg file is used to specify the boot order. The guest firmware may ignore or not support this fw_cfg file. In that case, this feature doesn't work as expected 😭 (yes, same disclaimer, it's a whole vibe)

🍓 9P device backends 🍓

  • sharename=/path/to/share[,9p-device-options]

The 9p-device-options are:

  • ro — expose the share in read-only mode 🔐

🖥️ TTY device backends 🖥️

  • stdio — connect the serial port to the standard input and output of the bhyve process.
  • /dev/xxx — use the host TTY device for serial port I/O.
  • tcp=ip:port — use the TCP server for serial port I/O. Configuring this option will start a TCP server that waits for connections. Only one connection is allowed at any time. The TCP server will immediately close new connections while an existing connection is active. Note that this feature allows unprivileged users to access the guest console, so ensure that access is appropriately restricted.
⚠️ bestie read that again — unprivileged users + guest console = lock it down. this is not a drill 🚨

🔐 TPM device backends 🔐

  • type,path[,tpm-device-options]

Emulate a TPM device. Supported options for type:

  • passthru — use a physical TPM device. The argument path needs to point to a valid TPM device path, i.e. /dev/tpm0.
  • swtpm — connect to a running swtpm instance. The argument path needs to point to a UNIX domain socket that a swtpm process is listening on.

The tpm-device-options are:

  • version=version — version of the TPM device according to the TCG specification. Defaults to 2.0, which is the only version currently supported. (so... yeah, 2.0. that's the whole menu 🎀)

👑 Boot ROM device backends 👑

  • romfile[,varfile]

Map romfile in the guest address space reserved for boot firmware.

If varfile is provided, that file is also mapped in the boot firmware guest address space, and any modifications the guest makes will be saved to that file. it's giving persistent memory, love that for you 💖

fw_cfg types:

  • fwcfg — the fw_cfg interface is used to pass information such as the CPU count or ACPI tables to the guest firmware. Supported values are bhyve and qemu. Due to backward compatibility reasons, bhyve is the default option. When bhyve is used, bhyve's fwctl interface is used. It currently reports only the CPU count to the guest firmware. The qemu option uses QEMU's fw_cfg interface. This interface is widely used and allows user-defined information to be passed to the guest. It is used for passing the CPU count, ACPI tables, a boot order and many other things to the guest. Some operating systems such as Fedora CoreOS can be configured by QEMU's fw_cfg interface as well 🔥

🦋 Pass-through device backends 🦋

  • pptN[,passthru-device-options]
  • bus/slot/function[,passthru-device-options]
  • pcibus:slot:function[,passthru-device-options]

Connect to a PCI device on the host either named pptN or at the selector described by slot, bus, and function numbers.

The passthru-device-options are:

  • rom=romfile — add romfile as option ROM to the PCI device. The ROM will be loaded by firmware and should be capable of initializing the device.
  • bootindex=index — add the device to the boot order at index. A fw_cfg file is used to specify the boot order. The guest firmware may ignore or doesn't support this fw_cfg file. In that case, this feature doesn't work as expected 💀
📌 Guest memory must be wired using the -S option when a pass-through device is configured.
📌 The host device must have been reserved at boot time using the pptdevs loader variable as described in vmm(4).

🎀 Virtio console device backends 🎀

  • port1=/path/to/port1.sock[,portN=/path/to/port2.sock ...]

A maximum of 16 ports per device can be created. Every port is named and corresponds to a Unix domain socket created by bhyve. bhyve accepts at most one connection per port at a time 🫶

Limitations:

  • Due to the lack of destructors in bhyve, sockets on the filesystem must be cleaned up manually after bhyve exits. (yes you, personally. go clean them up 🧹)
  • There is no way to use the "console port" feature, nor the console port resize at present.
  • Emergency write is advertised, but no-op at present. iconic behavior honestly 💅

🌸 Virtio input device backends 🌸

  • /dev/input/eventX

Send input events of /dev/input/eventX to guest by VirtIO Input Interface.

🌈 Framebuffer device backends 🌈

  • [rfb=address][,w=width][,h=height][,vga=vgaconf][,wait][,password=password]

Configuration options are defined as follows:

  • rfb=address (or tcp=address) — a UNIX domain socket or IP address and a port VNC should listen on. There are three possible formats:
    • [IPv4:]port
    • [IPv6%zone]:port
    • unix:my/unix.sock

The default is to listen on localhost IPv4 address and default VNC port 5900. An IPv6 address must be enclosed in square brackets and may contain an optional zone identifier.

  • w=width and h=height — a display resolution, width and height, respectively. If not specified, a default resolution of 1024x768 pixels will be used. Minimal supported resolution is 640x480 pixels, and maximum is 3840x2160 pixels 📺
  • vga=vgaconf — the VGA personality dial 🎛️. Possible values are io (that's the default, bestie), on, and off. So here's the lore: PCI graphics cards are lowkey dual-personality — they're standard PCI devices with BAR addressing, but they can also implicitly decode legacy VGA I/O space (0x3c0-0xdf) and memory space (64 KiB at 0xA0000). Wild, right?
    • The default io option is the one you want for guests that try to issue BIOS calls which result in I/O port queries, and then straight up fail to boot if I/O decode is disabled. Like, they literally cannot function 💀.
    • The on option goes hand-in-hand with the CSM BIOS capability in UEFI, so you can boot traditional BIOS guests that require the legacy VGA I/O and memory regions to be available. Old school, but make it work ✨.
    • The off option is for UEFI guests that assume a VGA adapter is present if they detect the I/O ports. The iconic example of such a guest is OpenBSD in UEFI mode. Noted 📝.

Please refer to the bhyve FreeBSD wiki page (https://wiki.freebsd.org/bhyve) for configuration notes of particular guests. iykyk 🫶

  • wait — tells bhyve to only boot upon the initiation of a VNC connection. Why? Because it simplifies the installation of operating systems that require immediate keyboard input — no more missing the boot prompt while you're still alt-tabbing, queen 💅. This can be removed for post-installation use.
  • password=password — okay so real talk: this type of authentication is known to be cryptographically weak and is not intended for use on untrusted networks. Many implementations will want to use stronger security, such as running the session over an encrypted channel provided by IPsec or SSH. Protect yourself, bestie 🔒.

🔌 xHCI USB device backends

  • tablet

A USB tablet device that provides precise cursor synchronization when using VNC. Cursor lines up perfectly, no cap 🖱️✨.

💾 NVMe device backends

  • devpath[,maxq=#][,qsz=#][,ioslots=#][,sectsz=#][,ser=#][,eui64=#][,dsm=opt]

Configuration options are defined as follows:

  • devpath — accepted device paths are: /dev/blockdev or /path/to/image or ram=size_in_MiB.
  • maxq — max number of queues.
  • qsz — max elements in each queue.
  • ioslots — max number of concurrent I/O requests.
  • sectsz — sector size (defaults to blockif sector size).
  • ser — serial number with maximum 20 characters.
  • eui64 — IEEE Extended Unique Identifier (8 byte value).
  • dsm — Dataset Management support. Supported values are: auto, enable, and disable.

🗄️ AHCI device backends

  • [[hd:|cd:]path][,nmrr=nmrr][,ser=#][,rev=#][,model=#]

Configuration options are defined as follows:

  • nmrr — Nominal Media Rotation Rate, known as RPM. A value of 1 indicates that the device is a solid state drive, i.e., non-rotational. Default value is 0.
  • ser — serial number with maximum 20 characters.
  • rev — revision number with maximum 8 characters.
  • model — model number with maximum 40 characters.

🔊 HD Audio device backends

  • [play=playback][,rec=recording]

Configuration options are defined as follows:

  • play — playback device, typically /dev/dsp0.
  • rec — recording device, typically /dev/dsp0.

🌳 CONFIGURATION VARIABLES

So bhyve uses an internal tree of configuration variables to describe global and per-device settings, and honestly? It's giving file system but make it settings 🌲.

When bhyve starts, it parses command line options (including config files) in the order given on the command line. Each command line option sets one or more configuration variables. For example, the -s option creates a new tree node for a PCI device and sets one or more variables under that node — including the device model and device model-specific variables. Variables may be set multiple times during this parsing stage, with the final value overriding previous values. Last one wins, periodt 💅.

Once all of the command line options have been processed, the configuration values are frozen. bhyve then uses the value of configuration values to initialize device models and global settings. No take-backs, the vibes are locked in ❄️.

More details on configuration variables can be found in bhyve_config(5).

📝 CONFIGURATION FILE CREATION

The -k flag lets you provide a path to a configuration file holding all the settings, which otherwise would need to be defined by providing a long list of program arguments to bhyve. Like, imagine typing all of that out. No thanks 😭.

There is a very simple way to translate a complex set of program arguments to an equivalent configuration file in bhyve_config(5) format. Iconic behaviour, honestly 👑.

Use -o config.dump=1 to make bhyve dump a configuration file representing the used flags and arguments to stdout. You can pipe the output into a file to persist the generated settings. Chef's kiss 🍓.

⚠️ Make sure to remove the config.dump line from the resulting configuration file before using it to start bhyve. Do not skip this step, bestie, I'm begging 🙈.

🐛 DEBUG SERVER

The current debug server provides limited support for debuggers. It's giving "we have a debugger at home" energy, but it still works 💀.

🧮 Registers

Each virtual CPU is exposed to the debugger as a thread.

General purpose registers can be queried for each virtual CPU, but other registers such as floating-point and system registers cannot be queried. Sad but true 😔.

🧠 Memory

Memory (including memory mapped I/O regions) can be read and written by the debugger. Memory operations use virtual addresses that are resolved to physical addresses via the current virtual CPU's active address translation. So it's virtual → physical, the classic glow-up ✨.

🎮 Control

The running guest can be interrupted by the debugger at any time (for example, by pressing Ctrl+C in the debugger). Gotcha, you're paused now 💅.

Single stepping is only supported on Intel CPUs supporting the MTRAP VM exit. Intel girls only for that one, sorry 🌸.

Breakpoints are supported on Intel CPUs that support single stepping. Note that continuing from a breakpoint while interrupts are enabled in the guest may not work as expected due to timer interrupts firing while single stepping over the breakpoint. Lowkey chaotic, but now you know 🫠.

🚨 SIGNAL HANDLING

bhyve deals with the following signals:

  • SIGTERM — trigger ACPI poweroff for a VM. Politely asking the VM to go night-night 😴.

🚪 EXIT STATUS

Exit status indicates how the VM was terminated:

  1. 0 — rebooted
  2. 1 — powered off
  3. 2 — halted
  4. 3 — triple fault (amd64 only)
  5. 4 — exited due to an error
  6. 5 — suspended

🌈 EXAMPLES

If not using a boot ROM, the guest operating system must have been loaded with bhyveload(8) or a similar boot loader before bhyve(4) can be run. Otherwise, the boot loader is not needed. Choose your fighter 🥊.

To run a virtual machine with 1 GiB of memory, two virtual CPUs, a virtio block device backed by the /my/image filesystem image, and a serial port for the console:

bhyve -c 2 -s 0,hostbridge -s 1,lpc -s 2,virtio-blk,/my/image \
  -l com1,stdio -H -P -m 1G vm1

To do the same on arm64:

bhyve -c 2 -s 0,hostbridge -s 1,virtio-blk,/my/image -o console=stdio \
  -o bootrom=/usr/local/share/u-boot/u-boot-bhyve-arm64/u-boot.bin -m 1G vm1

Run a 24 GiB single-CPU virtual machine with three network ports, one of which has a MAC address specified:

bhyve -s 0,hostbridge -s 1,lpc -s 2:0,virtio-net,tap0 \
  -s 2:1,virtio-net,tap1 \
  -s 2:2,virtio-net,tap2,mac=00:be:fa:76:45:00 \
  -s 3,virtio-blk,/my/image -l com1,stdio \
  -H -P -m 24G bigvm

Run an 8 GiB quad-CPU virtual machine with 8 AHCI SATA disks, an AHCI ATAPI CD-ROM, a single virtio network port, an AMD hostbridge, and the console port connected to an nmdm(4) null-modem device.

bhyve -c 4 \
  -s 0,amd_hostbridge -s 1,lpc \
  -s 1:0,ahci,hd:/images/disk.1,hd:/images/disk.2,\
hd:/images/disk.3,hd:/images/disk.4,\
hd:/images/disk.5,hd:/images/disk.6,\
hd:/images/disk.7,hd:/images/disk.8,\
cd:/images/install.iso \
  -s 3,virtio-net,tap0 \
  -l com1,/dev/nmdm0A \
  -H -P -m 8G

Run a UEFI virtual machine with a display resolution of 800 by 600 pixels that can be accessed via VNC at: 0.0.0.0:5900 or via serial console over TCP at: 127.0.0.1:1234 (unsafe if you expose serial console without protection).

bhyve -c 2 -m 4G -w -H \
  -s 0,hostbridge \
  -s 3,ahci-cd,/path/to/uefi-OS-install.iso \
  -s 4,ahci-hd,disk.img \
  -s 5,virtio-net,tap0 \
  -s 29,fbuf,tcp=0.0.0.0:5900,w=800,h=600,wait \
  -s 30,xhci,tablet \
  -s 31,lpc -l com1,tcp=127.0.0.1:1234 \
  -l bootrom,/usr/local/share/uefi-firmware/BHYVE_UEFI.fd \
   uefivm

Run a UEFI virtual machine with a VNC display that is bound to all IPv6 addresses on port 5900 and a serial I/O port bound to TCP port 1234 of loopback address (unsafe if you expose serial console without protection).

bhyve -c 2 -m 4G -w -H \
  -s 0,hostbridge \
  -s 4,ahci-hd,disk.img \
  -s 5,virtio-net,tap0 \
  -s 29,fbuf,tcp=[::]:5900,w=800,h=600 \
  -s 30,xhci,tablet \
  -s 31,lpc -l com1,tcp=[::1]:1234 \
  -l bootrom,/usr/local/share/uefi-firmware/BHYVE_UEFI.fd \
   uefivm

Run a UEFI virtual machine with a VARS file to save EFI variables. Note that bhyve will write guest modifications to the given VARS file. Be sure to create a per-guest copy of the template VARS file from /usr. Like, don't share your VARS file, that's not giving boundaries 🚩.

bhyve -c 2 -m 4g -w -H \
  -s 0,hostbridge \
  -s 31,lpc -l com1,stdio \
  -l bootrom,/usr/local/share/uefi-firmware/BHYVE_UEFI_CODE.fd,BHYVE_UEFI_VARS.fd
   uefivm

To create a configuration file configfile for a virtual machine, use -o config.dump=1:

/usr/sbin/bhyve -c 2 -m 256 -H -P \
  -s 0:0,hostbridge -s 1:0,virtio-net,tap0 \
  -s 2:0,ahci-hd,./vm0.img \
  -s 31,lpc -l com1,stdio \
  -o config.dump=1 vm0 > configfile

Then use an editor of your choice to remove the line "config.dump=1" from the newly generated configfile. Yes, actually remove it — I said what I said 💅.

To start bhyve using this configuration file, use flag -k:

/usr/sbin/bhyve -k configfile vm0

Run a UEFI virtual machine with four CPUs and two emulated NUMA domains:

bhyve -c 4 -w -H \
  -s 0,hostbridge \
  -s 4,ahci-hd,disk.img \
  -s 31,lpc -l com1,stdio \
  -l bootrom,/usr/local/share/uefi-firmware/BHYVE_UEFI.fd \
  -n id=0,size=4G,cpus=0-1 \
  -n id=1,size=4G,cpus=2-3 \
   numavm

Assuming a host machine with two NUMA domains, run a UEFI virtual machine with four CPUs using a prefer domainset(9) policy to allocate guest memory from the first host NUMA domain only.

bhyve -c 2 -w -H \
  -s 0,hostbridge \
  -s 4,ahci-hd,disk.img \
  -s 31,lpc -l com1,stdio \
  -l bootrom,/usr/local/share/uefi-firmware/BHYVE_UEFI.fd \
  -n id=0,size=4G,cpus=0-1,domain_policy=prefer:0 \
   numavm

✨ SEE ALSO ✨

okay bestie, here's the squad — all the other pages that live in the same friend group, go read them if you want the full lore 💖

also! Intel, 64 and IA-32 Architectures Software Developer's Manual, Volume 3. — literally the sacred text, the big book, the one you keep open in another window at 2am 🫶

🌸 HISTORY 🌸

bhyve first appeared in FreeBSD 10.0 — so she's been on the scene for a minute, and she's still main character energy 👑

💜 AUTHORS 💜

the icons who actually made this happen, give them their flowers fr fr 🌷

  • Neel Natu <neel@freebsd.org>
  • Peter Grehan <grehan@freebsd.org>
so yeah, that's the whole page bestie — go spin up a VM, be careful with -H because that's genuinely no going back, and text me if it doesn't boot 💀✨

<nmdm>ich</nmdm>

Debian January 5, 2026 BHYVE(8) — that's the source and the date, mwah, bye! 💅🎀🦋

📦 where this page comes from