diff --git a/usr.sbin/bhyve/bhyve.8 b/usr.sbin/bhyve/bhyve.8 index 1de8c516b785..6d78ad872f63 100644 --- a/usr.sbin/bhyve/bhyve.8 +++ b/usr.sbin/bhyve/bhyve.8 @@ -1,942 +1,948 @@ .\" Copyright (c) 2013 Peter Grehan .\" All rights reserved. .\" .\" Redistribution and use in source and binary forms, with or without .\" modification, are permitted provided that the following conditions .\" are met: .\" 1. Redistributions of source code must retain the above copyright .\" notice, this list of conditions and the following disclaimer. .\" 2. Redistributions in binary form must reproduce the above copyright .\" notice, this list of conditions and the following disclaimer in the .\" documentation and/or other materials provided with the distribution. .\" .\" THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND .\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE .\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE .\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE .\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL .\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS .\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) .\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT .\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY .\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF .\" SUCH DAMAGE. .\" .\" $FreeBSD$ .\" .Dd August 19, 2022 .Dt BHYVE 8 .Os .Sh NAME .Nm bhyve .Nd "run a guest operating system inside a virtual machine" .Sh SYNOPSIS .Nm .Op Fl AaCDeHhPSuWwxY .Oo .Sm off .Fl c\~ .Oo .Op Cm cpus= .Ar numcpus .Oc .Op Cm ,sockets= Ar n .Op Cm ,cores= Ar n .Op Cm ,threads= Ar n .Oc .Sm on .Op Fl G Ar port .Op Fl k Ar config_file .Oo Fl l .Sm off .Ar lpcdev Op Cm \&, Ar conf .Sm on .Oc .Oo Fl m .Sm off .Ar memsize .Oo .Cm K | Cm k | Cm M | Cm m | Cm G | Cm g | Cm T | Cm t .Oc .Sm on .Oc .Op Fl o Ar var Ns Cm = Ns Ar value .Op Fl p Ar vcpu Ns Cm \&: Ns Ar hostcpu .Op Fl r Ar file .Sm off .Oo Fl s\~ .Ar slot Cm \&, Ar emulation Op Cm \&, Ar conf .Sm on .Oc .Op Fl U Ar uuid .Ar vmname .Nm .Fl l Cm help .Nm .Fl s Cm help .Sh DESCRIPTION .Nm is a hypervisor that runs guest operating systems inside a virtual machine. .Pp Parameters such as the number of virtual CPUs, amount of guest memory, and I/O connectivity can be specified with command-line parameters. .Pp If not using a boot ROM, the guest operating system must be loaded with .Xr bhyveload 8 or a similar boot loader before running .Nm , otherwise, it is enough to run .Nm with a boot ROM of choice. .Pp .Nm runs until the guest operating system reboots or an unhandled hypervisor exit is detected. .Sh OPTIONS .Bl -tag -width 10n .It Fl A Generate ACPI tables. Required for .Fx Ns /amd64 guests. .It Fl a The guest's local APIC is configured in xAPIC mode. The xAPIC mode is the default setting so this option is redundant. It will be deprecated in a future version. .It Fl C Include guest memory in core file. .It Fl c Op Ar setting ... Number of guest virtual CPUs and/or the CPU topology. The default value for each of .Ar numcpus , .Ar sockets , .Ar cores , and .Ar threads is 1. The current maximum number of guest virtual CPUs is 16. If .Ar numcpus is not specified then it will be calculated from the other arguments. The topology must be consistent in that the .Ar numcpus must equal the product of .Ar sockets , .Ar cores , and .Ar threads . If a .Ar setting is specified more than once the last one has precedence. .It Fl D Destroy the VM on guest initiated power-off. .It Fl e Force .Nm to exit when a guest issues an access to an I/O port that is not emulated. This is intended for debug purposes. -.It Fl G Ar port +.It Fl G Xo +.Sm off +.Oo Ar w Oc +.Oo Ar bind_address: Oc +.Ar port +.Sm on +.Xc 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 +.Ar bind_address +and .Ar port to listen for debugger connections. Only a single debugger may be attached to the debug server at a time. -If -.Ar port -begins with +If the option begins with .Sq w , .Nm will pause execution at the first instruction waiting for a debugger to attach. .It Fl 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. .It Fl h Print help message and exit. .It Fl k Ar 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 .Pq Sq = , and a value. No spaces are permitted between the variable name, equals sign, or value. Blank lines and lines starting with .Sq # are ignored. See .Xr bhyve_config 5 for more details. .It Fl l Cm help Print a list of supported LPC devices. .It Fl l Ar lpcdev Ns Op Cm \&, Ns Ar conf Allow devices behind the LPC PCI-ISA bridge to be configured. The only supported devices are the TTY-class devices .Cm com1 , com2 , com3 , and .Cm com4 , the boot ROM device .Cm bootrom , and the debug/test device .Cm pc-testdev . .Pp The possible values for the .Ar conf argument are listed in the .Fl s flag description. .It Xo .Fl m Ar memsize Ns Oo .Sm off .Cm K | k | M | m | G | g | T | t .Sm on .Oc .Xc Set the guest physical memory size This must be the same size that was given to .Xr bhyveload 8 . .Pp The size argument may be suffixed with one of .Cm K , M , G or .Cm 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. .Pp The default is 256M. .It Fl o Ar var Ns Cm = Ns Ar value Set the configuration variable .Ar var to .Ar value . .It Fl P Force the guest virtual CPU to exit when a PAUSE instruction is detected. .It Fl p Ar vcpu Ns Cm \& : Ns Ar hostcpu Pin guest's virtual CPU .Em vcpu to .Em hostcpu . .It Fl r Ar file Resume a guest from a snapshot. The guest memory contents are restored from .Ar file , and the guest device and vCPU state are restored from the file .Dq Ar file Ns .kern . .Pp 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 .Fl s and .Fl l options. The count of vCPUs and memory configuration are read from the snapshot. .It Fl S Wire guest memory. .It Fl s Cm help Print a list of supported PCI devices. .It Fl s Ar slot Ns Cm \&, Ns Ar emulation Ns Op Cm \&, Ns Ar conf Configure a virtual PCI slot and function. .Pp .Nm 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. .Pp The .Ar slot can be specified in one of the following formats: .Pp .Bl -bullet -compact .It .Ar pcislot .It .Sm off .Ar pcislot Cm \&: Ar function .Sm on .It .Sm off .Ar bus Cm \&: Ar pcislot Cm \&: Ar function .Sm on .El .Pp The .Ar pcislot value is 0 to 31. The optional .Ar function value is 0 to 7. The optional .Ar bus value is 0 to 255. If not specified, the .Ar function value defaults to 0. If not specified, the .Ar bus value defaults to 0. .Pp The .Ar emulation argument can be one of the following: .Bl -tag -width "amd_hostbridge" .It Cm hostbridge A simple host bridge. This is usually configured at slot 0, and is required by most guest operating systems. .It Cm amd_hostbridge Emulation identical to .Cm hostbridge using a PCI vendor ID of AMD. .It Cm passthru PCI pass-through device. .It Cm virtio-net Virtio network interface. .It Cm virtio-blk Virtio block storage interface. .It Cm virtio-scsi Virtio SCSI interface. .It Cm virtio-9p Virtio 9p (VirtFS) interface. .It Cm virtio-rnd Virtio RNG interface. .It Cm 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. .It Cm ahci AHCI controller attached to arbitrary devices. .It Cm ahci-cd AHCI controller attached to an ATAPI CD/DVD. .It Cm ahci-hd AHCI controller attached to a SATA hard drive. .It Cm e1000 Intel e82545 network interface. .It Cm uart PCI 16550 serial device. .It Cm lpc LPC PCI-ISA bridge with COM1, COM2, COM3, and COM4 16550 serial ports, a boot ROM, and, optionally, the debug/test device. The LPC bridge emulation can only be configured on bus 0. .It Cm fbuf Raw framebuffer device attached to VNC server. .It Cm xhci eXtensible Host Controller Interface (xHCI) USB controller. .It Cm nvme NVM Express (NVMe) controller. .It Cm hda High Definition Audio Controller. .El .Pp The optional parameter .Ar conf describes the backend for device emulations. If .Ar conf is not specified, the device emulation has no backend and can be considered unconnected. .Pp Network device backends: .Sm off .Bl -bullet .It .Xo .Cm tap Ar N .Op Cm \&,mac= Ar xx:xx:xx:xx:xx:xx .Op Cm \&,mtu= Ar N .Xc .It .Xo .Cm vmnet Ar N .Op Cm \&,mac= Ar xx:xx:xx:xx:xx:xx .Op Cm \&,mtu= Ar N .Xc .It .Xo .Cm netgraph,path= Ar ADDRESS Cm \&,peerhook= Ar HOOK .Op Cm \&,socket= Ar NAME .Op Cm \&,hook= Ar HOOK .Op Cm \&,mac= Ar xx:xx:xx:xx:xx:xx .Op Cm \&,mtu= Ar N .Xc .El .Sm on .Pp If .Cm 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. .Pp The MAC address is an ASCII string in .Xr ethers 5 format. .Pp With .Cm virtio-net devices, the .Cm mtu parameter can be specified to inform the guest about the largest MTU that should be allowed, expressed in bytes. .Pp With .Cm netgraph backend, the .Cm path and .Cm peerhook parameters must be specified to set the destination node and corresponding hook. The optional parameters .Cm socket and .Cm hook may be used to set the .Xr ng_socket 4 node name and source hook. The .Ar ADDRESS , .Ar HOOK , and .Ar NAME must comply with .Xr netgraph 4 addressing rules. .Pp Block storage device backends: .Sm off .Bl -bullet .It .Ar /filename Op Cm \&, Ar block-device-options .It .Ar /dev/xxx Op Cm \&, Ar block-device-options .El .Sm on .Pp The .Ar block-device-options are: .Bl -tag -width 10n .It Cm nocache Open the file with .Dv O_DIRECT . .It Cm direct Open the file using .Dv O_SYNC . .It Cm ro Force the file to be opened read-only. .It Cm sectorsize= Ns Ar logical Ns Oo Cm \&/ Ns Ar physical Oc 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. .It Cm nodelete Disable emulation of guest trim requests via .Dv DIOCGDELETE requests. .El .Pp SCSI device backends: .Sm off .Bl -bullet .It .Pa /dev/cam/ctl Oo Ar pp Cm \&. Ar vp Oc Oo Cm \&, Ar scsi-device-options Oc .El .Sm on .Pp The .Ar scsi-device-options are: .Bl -tag -width 10n .It Cm iid= Ns Ar IID Initiator ID to use when sending requests to specified CTL port. The default value is 0. .El .Pp 9P device backends: .Sm off .Bl -bullet .It .Ar sharename Cm = Ar /path/to/share Op Cm \&, Ar 9p-device-options .El .Sm on .Pp The .Ar 9p-device-options are: .Bl -tag -width 10n .It Cm ro Expose the share in read-only mode. .El .Pp TTY device backends: .Bl -tag -width 10n .It Cm stdio Connect the serial port to the standard input and output of the .Nm process. .It Ar /dev/xxx Use the host TTY device for serial port I/O. .El .Pp Boot ROM device backends: .Bl -tag -width 10n .It Ar romfile Map .Ar romfile in the guest address space reserved for boot firmware. .El .Pp Pass-through device backends: .Sm off .Bl -bullet .It .Cm ppt Ar N Oo , Ar passthru-device-options Oc .It .Ns Ar bus Cm \&/ Ar slot Cm \&/ Ar function .Op , Ar passthru-device-options .It .Cm pci Ar bus Cm : Ar slot Cm : Ns Ar function .Op , Ar passthru-device-options .El .Sm on .Pp Connect to a PCI device on the host either named ppt .Ns Ar N or at the selector described by .Ar slot , .Ar bus , and .Ar function numbers. .Pp The .Ar passthru-device-options are: .Bl -tag -width 10n .It Cm rom= Ns Ar romfile Add .Ar romfile as option ROM to the PCI device. The ROM will be loaded by firmware and should be capable of initializing the device. .El .Pp Guest memory must be wired using the .Fl S option when a pass-through device is configured. .Pp The host device must have been reserved at boot-time using the .Va pptdevs loader variable as described in .Xr vmm 4 . .Pp Virtio console device backends: .Bl -bullet .Sm off .It .Cm port1= Ns Ar /path/to/port1.sock Ns Op Cm ,port Ns Ar N Cm \&= Ns Ar /path/to/port2.sock No \~ Ar ... .Sm on .El .Pp A maximum of 16 ports per device can be created. Every port is named and corresponds to a Unix domain socket created by .Nm . .Nm accepts at most one connection per port at a time. .Pp Limitations: .Bl -bullet .It Due to lack of destructors in .Nm , sockets on the filesystem must be cleaned up manually after .Nm exits. .It There is no way to use the .Dq console port feature, nor the console port resize at present. .It Emergency write is advertised, but no-op at present. .El .Pp Framebuffer devices backends: .Bl -bullet .Sm off .It .Op Cm rfb= Ar ip-and-port .Op Cm ,w= Ar width .Op Cm ,h= Ar height .Op Cm ,vga= Ar vgaconf .Op Cm ,wait .Op Cm ,password= Ar password .Sm on .El .Pp Configuration options are defined as follows: .Bl -tag -width 10n .It Cm rfb= Ns Ar ip-and-port Pq or Cm tcp= Ns Ar ip-and-port An IP address and a port VNC should listen on. There are two formats: .Pp .Bl -bullet -compact .It .Sm off .Op Ar IPv4 Cm \&: .Ar port .Sm on .It .Sm off .Cm \&[ Ar IPv6%zone Cm \&] Cm \&: Ar port .Sm on .El .Pp 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. .It Cm w= Ns Ar width No and Cm h= Ns Ar 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 1920x1200 pixels. .It Cm vga= Ns Ar vgaconf Possible values for this option are .Cm io (default), .Cm on , and .Cm off . PCI graphics cards have a dual personality in that they are standard PCI devices with BAR addressing, but may also implicitly decode legacy VGA I/O space .Pq Ad 0x3c0-3df and memory space .Pq 64KB at Ad 0xA0000 . The default .Cm io option should be used for guests that attempt to issue BIOS calls which result in I/O port queries, and fail to boot if I/O decode is disabled. .Pp The .Cm on option should be used along with the CSM BIOS capability in UEFI to boot traditional BIOS guests that require the legacy VGA I/O and memory regions to be available. .Pp The .Cm off option should be used for the UEFI guests that assume that VGA adapter is present if they detect the I/O ports. An example of such a guest is .Ox in UEFI mode. .Pp Please refer to the .Nm .Fx wiki page .Pq Lk https://wiki.freebsd.org/bhyve for configuration notes of particular guests. .It Cm wait Instruct .Nm to only boot upon the initiation of a VNC connection, simplifying the installation of operating systems that require immediate keyboard input. This can be removed for post-installation use. .It Cm password= Ns Ar password 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. .El .Pp xHCI USB device backends: .Bl -tag -width 10n .It Cm tablet A USB tablet device which provides precise cursor synchronization when using VNC. .El .Pp NVMe device backends: .Bl -bullet .Sm off .It .Ar devpath .Op Cm ,maxq= Ar # .Op Cm ,qsz= Ar # .Op Cm ,ioslots= Ar # .Op Cm ,sectsz= Ar # .Op Cm ,ser= Ar # .Op Cm ,eui64= Ar # .Op Cm ,dsm= Ar opt .Sm on .El .Pp Configuration options are defined as follows: .Bl -tag -width 10n .It Ar devpath Accepted device paths are: .Ar /dev/blockdev or .Ar /path/to/image or .Cm ram= Ns Ar size_in_MiB . .It Cm maxq Max number of queues. .It Cm qsz Max elements in each queue. .It Cm ioslots Max number of concurrent I/O requests. .It Cm sectsz Sector size (defaults to blockif sector size). .It Cm ser Serial number with maximum 20 characters. .It Cm eui64 IEEE Extended Unique Identifier (8 byte value). .It Cm dsm DataSet Management support. Supported values are: .Cm auto , enable , and .Cm disable . .El .Pp AHCI device backends: .Bl -bullet .It .Sm off .Op Oo Cm hd\&: | cd\&: Oc Ar path .Op Cm ,nmrr= Ar nmrr .Op Cm ,ser= Ar # .Op Cm ,rev= Ar # .Op Cm ,model= Ar # .Sm on .El .Pp Configuration options are defined as follows: .Bl -tag -width 10n .It Cm nmrr Nominal Media Rotation Rate, known as RPM. Value 1 will indicate device as Solid State Disk. Default value is 0, not report. .It Cm ser Serial Number with maximum 20 characters. .It Cm rev Revision Number with maximum 8 characters. .It Cm model Model Number with maximum 40 characters. .El .Pp HD Audio device backends: .Bl -bullet .It .Sm off .Op Cm play= Ar playback .Op Cm ,rec= Ar recording .Sm on .El .Pp Configuration options are defined as follows: .Bl -tag -width 10n .It Cm play Playback device, typically .Ar /dev/dsp0 . .It Cm rec Recording device, typically .Ar /dev/dsp0 . .El .It Fl U Ar uuid Set the universally unique identifier .Pq UUID in the guest's System Management BIOS System Information structure. By default a UUID is generated from the host's hostname and .Ar vmname . .It Fl u RTC keeps UTC time. .It Fl W Force virtio PCI device emulations to use MSI interrupts instead of MSI-X interrupts. .It Fl w Ignore accesses to unimplemented Model Specific Registers (MSRs). This is intended for debug purposes. .It Fl x The guest's local APIC is configured in x2APIC mode. .It Fl Y Disable MPtable generation. .It Ar vmname Alphanumeric name of the guest. This should be the same as that created by .Xr bhyveload 8 . .El .Sh CONFIGURATION VARIABLES .Nm uses an internal tree of configuration variables to describe global and per-device settings. When .Nm 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 .Fl 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. .Pp Once all of the command line options have been processed, the configuration values are frozen. .Nm then uses the value of configuration values to initialize device models and global settings. .Pp More details on configuration variables can be found in .Xr bhyve_config 5 . .Sh DEBUG SERVER The current debug server provides limited support for debuggers. .Ss Registers Each virtual CPU is exposed to the debugger as a thread. .Pp General purpose registers can be queried for each virtual CPU, but other registers such as floating-point and system registers cannot be queried. .Ss 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. .Ss Control The running guest can be interrupted by the debugger at any time .Pq for example, by pressing Ctrl-C in the debugger . .Pp Single stepping is only supported on Intel CPUs supporting the MTRAP VM exit. .Pp 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. .Sh SIGNAL HANDLING .Nm deals with the following signals: .Pp .Bl -tag -width SIGTERM -compact .It SIGTERM Trigger ACPI poweroff for a VM .El .Sh EXIT STATUS Exit status indicates how the VM was terminated: .Pp .Bl -tag -width indent -compact .It 0 rebooted .It 1 powered off .It 2 halted .It 3 triple fault .It 4 exited due to an error .El .Sh EXAMPLES If not using a boot ROM, the guest operating system must have been loaded with .Xr bhyveload 8 or a similar boot loader before .Xr bhyve 4 can be run. Otherwise, the boot loader is not needed. .Pp To run a virtual machine with 1GB of memory, two virtual CPUs, a virtio block device backed by the .Pa /my/image filesystem image, and a serial port for the console: .Bd -literal -offset indent bhyve -c 2 -s 0,hostbridge -s 1,lpc -s 2,virtio-blk,/my/image \\ -l com1,stdio -A -H -P -m 1G vm1 .Ed .Pp Run a 24GB single-CPU virtual machine with three network ports, one of which has a MAC address specified: .Bd -literal -offset indent 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 \\ -A -H -P -m 24G bigvm .Ed .Pp Run an 8GB 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 .Xr nmdm 4 null-modem device. .Bd -literal -offset indent 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 \\ -A -H -P -m 8G .Ed .Pp 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. .Bd -literal -offset indent 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,stdio \\ -l bootrom,/usr/local/share/uefi-firmware/BHYVE_UEFI.fd \\ uefivm .Ed .Pp Run a UEFI virtual machine with a VNC display that is bound to all IPv6 addresses on port 5900. .Bd -literal -offset indent 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,stdio \\ -l bootrom,/usr/local/share/uefi-firmware/BHYVE_UEFI.fd \\ uefivm .Ed .Sh SEE ALSO .Xr bhyve 4 , .Xr netgraph 4 , .Xr ng_socket 4 , .Xr nmdm 4 , .Xr vmm 4 , .Xr bhyve_config 5 , .Xr ethers 5 , .Xr bhyvectl 8 , .Xr bhyveload 8 .Pp .Rs .%A Intel .%B 64 and IA-32 Architectures Software Developer’s Manual .%V Volume 3 .Re .Sh HISTORY .Nm first appeared in .Fx 10.0 . .Sh AUTHORS .An Neel Natu Aq Mt neel@freebsd.org .An Peter Grehan Aq Mt grehan@freebsd.org diff --git a/usr.sbin/bhyve/bhyverun.c b/usr.sbin/bhyve/bhyverun.c index fec4ca29de80..e9fc3c98ddeb 100644 --- a/usr.sbin/bhyve/bhyverun.c +++ b/usr.sbin/bhyve/bhyverun.c @@ -1,1577 +1,1594 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2011 NetApp, Inc. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include __FBSDID("$FreeBSD$"); #include #ifndef WITHOUT_CAPSICUM #include #endif #include #ifdef BHYVE_SNAPSHOT #include #include #endif #include #ifdef BHYVE_SNAPSHOT #include #endif #include #include #include #ifndef WITHOUT_CAPSICUM #include #endif #include #include #include #include #include #ifdef BHYVE_SNAPSHOT #include #endif #include #include #include #include #include #include #include #include #ifdef BHYVE_SNAPSHOT #include #include #include #endif #include #ifndef WITHOUT_CAPSICUM #include #endif #include #include #include "bhyverun.h" #include "acpi.h" #include "atkbdc.h" #include "bootrom.h" #include "config.h" #include "inout.h" #include "debug.h" #include "fwctl.h" #include "gdb.h" #include "ioapic.h" #include "kernemu_dev.h" #include "mem.h" #include "mevent.h" #include "mptbl.h" #include "pci_emul.h" #include "pci_irq.h" #include "pci_lpc.h" #include "smbiostbl.h" #ifdef BHYVE_SNAPSHOT #include "snapshot.h" #endif #include "xmsr.h" #include "spinup_ap.h" #include "rtc.h" #include "vmgenc.h" #define GUEST_NIO_PORT 0x488 /* guest upcalls via i/o port */ #define MB (1024UL * 1024) #define GB (1024UL * MB) static const char * const vmx_exit_reason_desc[] = { [EXIT_REASON_EXCEPTION] = "Exception or non-maskable interrupt (NMI)", [EXIT_REASON_EXT_INTR] = "External interrupt", [EXIT_REASON_TRIPLE_FAULT] = "Triple fault", [EXIT_REASON_INIT] = "INIT signal", [EXIT_REASON_SIPI] = "Start-up IPI (SIPI)", [EXIT_REASON_IO_SMI] = "I/O system-management interrupt (SMI)", [EXIT_REASON_SMI] = "Other SMI", [EXIT_REASON_INTR_WINDOW] = "Interrupt window", [EXIT_REASON_NMI_WINDOW] = "NMI window", [EXIT_REASON_TASK_SWITCH] = "Task switch", [EXIT_REASON_CPUID] = "CPUID", [EXIT_REASON_GETSEC] = "GETSEC", [EXIT_REASON_HLT] = "HLT", [EXIT_REASON_INVD] = "INVD", [EXIT_REASON_INVLPG] = "INVLPG", [EXIT_REASON_RDPMC] = "RDPMC", [EXIT_REASON_RDTSC] = "RDTSC", [EXIT_REASON_RSM] = "RSM", [EXIT_REASON_VMCALL] = "VMCALL", [EXIT_REASON_VMCLEAR] = "VMCLEAR", [EXIT_REASON_VMLAUNCH] = "VMLAUNCH", [EXIT_REASON_VMPTRLD] = "VMPTRLD", [EXIT_REASON_VMPTRST] = "VMPTRST", [EXIT_REASON_VMREAD] = "VMREAD", [EXIT_REASON_VMRESUME] = "VMRESUME", [EXIT_REASON_VMWRITE] = "VMWRITE", [EXIT_REASON_VMXOFF] = "VMXOFF", [EXIT_REASON_VMXON] = "VMXON", [EXIT_REASON_CR_ACCESS] = "Control-register accesses", [EXIT_REASON_DR_ACCESS] = "MOV DR", [EXIT_REASON_INOUT] = "I/O instruction", [EXIT_REASON_RDMSR] = "RDMSR", [EXIT_REASON_WRMSR] = "WRMSR", [EXIT_REASON_INVAL_VMCS] = "VM-entry failure due to invalid guest state", [EXIT_REASON_INVAL_MSR] = "VM-entry failure due to MSR loading", [EXIT_REASON_MWAIT] = "MWAIT", [EXIT_REASON_MTF] = "Monitor trap flag", [EXIT_REASON_MONITOR] = "MONITOR", [EXIT_REASON_PAUSE] = "PAUSE", [EXIT_REASON_MCE_DURING_ENTRY] = "VM-entry failure due to machine-check event", [EXIT_REASON_TPR] = "TPR below threshold", [EXIT_REASON_APIC_ACCESS] = "APIC access", [EXIT_REASON_VIRTUALIZED_EOI] = "Virtualized EOI", [EXIT_REASON_GDTR_IDTR] = "Access to GDTR or IDTR", [EXIT_REASON_LDTR_TR] = "Access to LDTR or TR", [EXIT_REASON_EPT_FAULT] = "EPT violation", [EXIT_REASON_EPT_MISCONFIG] = "EPT misconfiguration", [EXIT_REASON_INVEPT] = "INVEPT", [EXIT_REASON_RDTSCP] = "RDTSCP", [EXIT_REASON_VMX_PREEMPT] = "VMX-preemption timer expired", [EXIT_REASON_INVVPID] = "INVVPID", [EXIT_REASON_WBINVD] = "WBINVD", [EXIT_REASON_XSETBV] = "XSETBV", [EXIT_REASON_APIC_WRITE] = "APIC write", [EXIT_REASON_RDRAND] = "RDRAND", [EXIT_REASON_INVPCID] = "INVPCID", [EXIT_REASON_VMFUNC] = "VMFUNC", [EXIT_REASON_ENCLS] = "ENCLS", [EXIT_REASON_RDSEED] = "RDSEED", [EXIT_REASON_PM_LOG_FULL] = "Page-modification log full", [EXIT_REASON_XSAVES] = "XSAVES", [EXIT_REASON_XRSTORS] = "XRSTORS" }; typedef int (*vmexit_handler_t)(struct vmctx *, struct vm_exit *, int *vcpu); extern int vmexit_task_switch(struct vmctx *, struct vm_exit *, int *vcpu); int guest_ncpus; uint16_t cpu_cores, cpu_sockets, cpu_threads; int raw_stdio = 0; static char *progname; static const int BSP = 0; static cpuset_t cpumask; static void vm_loop(struct vmctx *ctx, int vcpu, uint64_t rip); static struct vm_exit *vmexit; struct bhyvestats { uint64_t vmexit_bogus; uint64_t vmexit_reqidle; uint64_t vmexit_hlt; uint64_t vmexit_pause; uint64_t vmexit_mtrap; uint64_t vmexit_inst_emul; uint64_t cpu_switch_rotate; uint64_t cpu_switch_direct; } stats; struct mt_vmm_info { pthread_t mt_thr; struct vmctx *mt_ctx; int mt_vcpu; } *mt_vmm_info; static cpuset_t **vcpumap; static void usage(int code) { fprintf(stderr, "Usage: %s [-AaCDeHhPSuWwxY]\n" " %*s [-c [[cpus=]numcpus][,sockets=n][,cores=n][,threads=n]]\n" " %*s [-G port] [-k config_file] [-l lpc] [-m mem] [-o var=value]\n" " %*s [-p vcpu:hostcpu] [-r file] [-s pci] [-U uuid] vmname\n" " -A: create ACPI tables\n" " -a: local apic is in xAPIC mode (deprecated)\n" " -C: include guest memory in core file\n" " -c: number of CPUs and/or topology specification\n" " -D: destroy on power-off\n" " -e: exit on unhandled I/O access\n" " -G: start a debug server\n" " -H: vmexit from the guest on HLT\n" " -h: help\n" " -k: key=value flat config file\n" " -l: LPC device configuration\n" " -m: memory size\n" " -o: set config 'var' to 'value'\n" " -P: vmexit from the guest on pause\n" " -p: pin 'vcpu' to 'hostcpu'\n" #ifdef BHYVE_SNAPSHOT " -r: path to checkpoint file\n" #endif " -S: guest memory cannot be swapped\n" " -s: PCI slot config\n" " -U: UUID\n" " -u: RTC keeps UTC time\n" " -W: force virtio to use single-vector MSI\n" " -w: ignore unimplemented MSRs\n" " -x: local APIC is in x2APIC mode\n" " -Y: disable MPtable generation\n", progname, (int)strlen(progname), "", (int)strlen(progname), "", (int)strlen(progname), ""); exit(code); } /* * XXX This parser is known to have the following issues: * 1. It accepts null key=value tokens ",," as setting "cpus" to an * empty string. * * The acceptance of a null specification ('-c ""') is by design to match the * manual page syntax specification, this results in a topology of 1 vCPU. */ static int topology_parse(const char *opt) { char *cp, *str; if (*opt == '\0') { set_config_value("sockets", "1"); set_config_value("cores", "1"); set_config_value("threads", "1"); set_config_value("cpus", "1"); return (0); } str = strdup(opt); if (str == NULL) errx(4, "Failed to allocate memory"); while ((cp = strsep(&str, ",")) != NULL) { if (strncmp(cp, "cpus=", strlen("cpus=")) == 0) set_config_value("cpus", cp + strlen("cpus=")); else if (strncmp(cp, "sockets=", strlen("sockets=")) == 0) set_config_value("sockets", cp + strlen("sockets=")); else if (strncmp(cp, "cores=", strlen("cores=")) == 0) set_config_value("cores", cp + strlen("cores=")); else if (strncmp(cp, "threads=", strlen("threads=")) == 0) set_config_value("threads", cp + strlen("threads=")); #ifdef notyet /* Do not expose this until vmm.ko implements it */ else if (strncmp(cp, "maxcpus=", strlen("maxcpus=")) == 0) set_config_value("maxcpus", cp + strlen("maxcpus=")); #endif else if (strchr(cp, '=') != NULL) goto out; else set_config_value("cpus", cp); } free(str); return (0); out: free(str); return (-1); } static int parse_int_value(const char *key, const char *value, int minval, int maxval) { char *cp; long lval; errno = 0; lval = strtol(value, &cp, 0); if (errno != 0 || *cp != '\0' || cp == value || lval < minval || lval > maxval) errx(4, "Invalid value for %s: '%s'", key, value); return (lval); } /* * Set the sockets, cores, threads, and guest_cpus variables based on * the configured topology. * * The limits of UINT16_MAX are due to the types passed to * vm_set_topology(). vmm.ko may enforce tighter limits. */ static void calc_topology(void) { const char *value; bool explicit_cpus; uint64_t ncpus; value = get_config_value("cpus"); if (value != NULL) { guest_ncpus = parse_int_value("cpus", value, 1, UINT16_MAX); explicit_cpus = true; } else { guest_ncpus = 1; explicit_cpus = false; } value = get_config_value("cores"); if (value != NULL) cpu_cores = parse_int_value("cores", value, 1, UINT16_MAX); else cpu_cores = 1; value = get_config_value("threads"); if (value != NULL) cpu_threads = parse_int_value("threads", value, 1, UINT16_MAX); else cpu_threads = 1; value = get_config_value("sockets"); if (value != NULL) cpu_sockets = parse_int_value("sockets", value, 1, UINT16_MAX); else cpu_sockets = guest_ncpus; /* * Compute sockets * cores * threads avoiding overflow. The * range check above insures these are 16 bit values. */ ncpus = (uint64_t)cpu_sockets * cpu_cores * cpu_threads; if (ncpus > UINT16_MAX) errx(4, "Computed number of vCPUs too high: %ju", (uintmax_t)ncpus); if (explicit_cpus) { if (guest_ncpus != ncpus) errx(4, "Topology (%d sockets, %d cores, %d threads) " "does not match %d vCPUs", cpu_sockets, cpu_cores, cpu_threads, guest_ncpus); } else guest_ncpus = ncpus; } static int pincpu_parse(const char *opt) { const char *value; char *newval; char key[16]; int vcpu, pcpu; if (sscanf(opt, "%d:%d", &vcpu, &pcpu) != 2) { fprintf(stderr, "invalid format: %s\n", opt); return (-1); } if (vcpu < 0) { fprintf(stderr, "invalid vcpu '%d'\n", vcpu); return (-1); } if (pcpu < 0 || pcpu >= CPU_SETSIZE) { fprintf(stderr, "hostcpu '%d' outside valid range from " "0 to %d\n", pcpu, CPU_SETSIZE - 1); return (-1); } snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu); value = get_config_value(key); if (asprintf(&newval, "%s%s%d", value != NULL ? value : "", value != NULL ? "," : "", pcpu) == -1) { perror("failed to build new cpuset string"); return (-1); } set_config_value(key, newval); free(newval); return (0); } static void parse_cpuset(int vcpu, const char *list, cpuset_t *set) { char *cp, *token; int pcpu, start; CPU_ZERO(set); start = -1; token = __DECONST(char *, list); for (;;) { pcpu = strtoul(token, &cp, 0); if (cp == token) errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list); if (pcpu < 0 || pcpu >= CPU_SETSIZE) errx(4, "hostcpu '%d' outside valid range from 0 to %d", pcpu, CPU_SETSIZE - 1); switch (*cp) { case ',': case '\0': if (start >= 0) { if (start > pcpu) errx(4, "Invalid hostcpu range %d-%d", start, pcpu); while (start < pcpu) { CPU_SET(start, vcpumap[vcpu]); start++; } start = -1; } CPU_SET(pcpu, vcpumap[vcpu]); break; case '-': if (start >= 0) errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list); start = pcpu; break; default: errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list); } if (*cp == '\0') break; token = cp + 1; } } static void build_vcpumaps(void) { char key[16]; const char *value; int vcpu; vcpumap = calloc(guest_ncpus, sizeof(*vcpumap)); for (vcpu = 0; vcpu < guest_ncpus; vcpu++) { snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu); value = get_config_value(key); if (value == NULL) continue; vcpumap[vcpu] = malloc(sizeof(cpuset_t)); if (vcpumap[vcpu] == NULL) err(4, "Failed to allocate cpuset for vcpu %d", vcpu); parse_cpuset(vcpu, value, vcpumap[vcpu]); } } void vm_inject_fault(void *arg, int vcpu, int vector, int errcode_valid, int errcode) { struct vmctx *ctx; int error, restart_instruction; ctx = arg; restart_instruction = 1; error = vm_inject_exception(ctx, vcpu, vector, errcode_valid, errcode, restart_instruction); assert(error == 0); } void * paddr_guest2host(struct vmctx *ctx, uintptr_t gaddr, size_t len) { return (vm_map_gpa(ctx, gaddr, len)); } #ifdef BHYVE_SNAPSHOT uintptr_t paddr_host2guest(struct vmctx *ctx, void *addr) { return (vm_rev_map_gpa(ctx, addr)); } #endif int fbsdrun_virtio_msix(void) { return (get_config_bool_default("virtio_msix", true)); } static void * fbsdrun_start_thread(void *param) { char tname[MAXCOMLEN + 1]; struct mt_vmm_info *mtp; int vcpu; mtp = param; vcpu = mtp->mt_vcpu; snprintf(tname, sizeof(tname), "vcpu %d", vcpu); pthread_set_name_np(mtp->mt_thr, tname); #ifdef BHYVE_SNAPSHOT checkpoint_cpu_add(vcpu); #endif gdb_cpu_add(vcpu); vm_loop(mtp->mt_ctx, vcpu, vmexit[vcpu].rip); /* not reached */ exit(1); return (NULL); } void fbsdrun_addcpu(struct vmctx *ctx, int fromcpu, int newcpu, uint64_t rip) { int error; assert(fromcpu == BSP); /* * The 'newcpu' must be activated in the context of 'fromcpu'. If * vm_activate_cpu() is delayed until newcpu's pthread starts running * then vmm.ko is out-of-sync with bhyve and this can create a race * with vm_suspend(). */ error = vm_activate_cpu(ctx, newcpu); if (error != 0) err(EX_OSERR, "could not activate CPU %d", newcpu); CPU_SET_ATOMIC(newcpu, &cpumask); /* * Set up the vmexit struct to allow execution to start * at the given RIP */ vmexit[newcpu].rip = rip; vmexit[newcpu].inst_length = 0; mt_vmm_info[newcpu].mt_ctx = ctx; mt_vmm_info[newcpu].mt_vcpu = newcpu; error = pthread_create(&mt_vmm_info[newcpu].mt_thr, NULL, fbsdrun_start_thread, &mt_vmm_info[newcpu]); assert(error == 0); } static int fbsdrun_deletecpu(struct vmctx *ctx, int vcpu) { if (!CPU_ISSET(vcpu, &cpumask)) { fprintf(stderr, "Attempting to delete unknown cpu %d\n", vcpu); exit(4); } CPU_CLR_ATOMIC(vcpu, &cpumask); return (CPU_EMPTY(&cpumask)); } static int vmexit_handle_notify(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu, uint32_t eax) { #if BHYVE_DEBUG /* * put guest-driven debug here */ #endif return (VMEXIT_CONTINUE); } static int vmexit_inout(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) { int error; int bytes, port, in, out; int vcpu; vcpu = *pvcpu; port = vme->u.inout.port; bytes = vme->u.inout.bytes; in = vme->u.inout.in; out = !in; /* Extra-special case of host notifications */ if (out && port == GUEST_NIO_PORT) { error = vmexit_handle_notify(ctx, vme, pvcpu, vme->u.inout.eax); return (error); } error = emulate_inout(ctx, vcpu, vme); if (error) { fprintf(stderr, "Unhandled %s%c 0x%04x at 0x%lx\n", in ? "in" : "out", bytes == 1 ? 'b' : (bytes == 2 ? 'w' : 'l'), port, vmexit->rip); return (VMEXIT_ABORT); } else { return (VMEXIT_CONTINUE); } } static int vmexit_rdmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) { uint64_t val; uint32_t eax, edx; int error; val = 0; error = emulate_rdmsr(ctx, *pvcpu, vme->u.msr.code, &val); if (error != 0) { fprintf(stderr, "rdmsr to register %#x on vcpu %d\n", vme->u.msr.code, *pvcpu); if (get_config_bool("x86.strictmsr")) { vm_inject_gp(ctx, *pvcpu); return (VMEXIT_CONTINUE); } } eax = val; error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RAX, eax); assert(error == 0); edx = val >> 32; error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RDX, edx); assert(error == 0); return (VMEXIT_CONTINUE); } static int vmexit_wrmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) { int error; error = emulate_wrmsr(ctx, *pvcpu, vme->u.msr.code, vme->u.msr.wval); if (error != 0) { fprintf(stderr, "wrmsr to register %#x(%#lx) on vcpu %d\n", vme->u.msr.code, vme->u.msr.wval, *pvcpu); if (get_config_bool("x86.strictmsr")) { vm_inject_gp(ctx, *pvcpu); return (VMEXIT_CONTINUE); } } return (VMEXIT_CONTINUE); } static int vmexit_spinup_ap(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) { (void)spinup_ap(ctx, *pvcpu, vme->u.spinup_ap.vcpu, vme->u.spinup_ap.rip); return (VMEXIT_CONTINUE); } #define DEBUG_EPT_MISCONFIG #ifdef DEBUG_EPT_MISCONFIG #define VMCS_GUEST_PHYSICAL_ADDRESS 0x00002400 static uint64_t ept_misconfig_gpa, ept_misconfig_pte[4]; static int ept_misconfig_ptenum; #endif static const char * vmexit_vmx_desc(uint32_t exit_reason) { if (exit_reason >= nitems(vmx_exit_reason_desc) || vmx_exit_reason_desc[exit_reason] == NULL) return ("Unknown"); return (vmx_exit_reason_desc[exit_reason]); } static int vmexit_vmx(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) { fprintf(stderr, "vm exit[%d]\n", *pvcpu); fprintf(stderr, "\treason\t\tVMX\n"); fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip); fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length); fprintf(stderr, "\tstatus\t\t%d\n", vmexit->u.vmx.status); fprintf(stderr, "\texit_reason\t%u (%s)\n", vmexit->u.vmx.exit_reason, vmexit_vmx_desc(vmexit->u.vmx.exit_reason)); fprintf(stderr, "\tqualification\t0x%016lx\n", vmexit->u.vmx.exit_qualification); fprintf(stderr, "\tinst_type\t\t%d\n", vmexit->u.vmx.inst_type); fprintf(stderr, "\tinst_error\t\t%d\n", vmexit->u.vmx.inst_error); #ifdef DEBUG_EPT_MISCONFIG if (vmexit->u.vmx.exit_reason == EXIT_REASON_EPT_MISCONFIG) { vm_get_register(ctx, *pvcpu, VMCS_IDENT(VMCS_GUEST_PHYSICAL_ADDRESS), &ept_misconfig_gpa); vm_get_gpa_pmap(ctx, ept_misconfig_gpa, ept_misconfig_pte, &ept_misconfig_ptenum); fprintf(stderr, "\tEPT misconfiguration:\n"); fprintf(stderr, "\t\tGPA: %#lx\n", ept_misconfig_gpa); fprintf(stderr, "\t\tPTE(%d): %#lx %#lx %#lx %#lx\n", ept_misconfig_ptenum, ept_misconfig_pte[0], ept_misconfig_pte[1], ept_misconfig_pte[2], ept_misconfig_pte[3]); } #endif /* DEBUG_EPT_MISCONFIG */ return (VMEXIT_ABORT); } static int vmexit_svm(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) { fprintf(stderr, "vm exit[%d]\n", *pvcpu); fprintf(stderr, "\treason\t\tSVM\n"); fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip); fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length); fprintf(stderr, "\texitcode\t%#lx\n", vmexit->u.svm.exitcode); fprintf(stderr, "\texitinfo1\t%#lx\n", vmexit->u.svm.exitinfo1); fprintf(stderr, "\texitinfo2\t%#lx\n", vmexit->u.svm.exitinfo2); return (VMEXIT_ABORT); } static int vmexit_bogus(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) { assert(vmexit->inst_length == 0); stats.vmexit_bogus++; return (VMEXIT_CONTINUE); } static int vmexit_reqidle(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) { assert(vmexit->inst_length == 0); stats.vmexit_reqidle++; return (VMEXIT_CONTINUE); } static int vmexit_hlt(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) { stats.vmexit_hlt++; /* * Just continue execution with the next instruction. We use * the HLT VM exit as a way to be friendly with the host * scheduler. */ return (VMEXIT_CONTINUE); } static int vmexit_pause(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) { stats.vmexit_pause++; return (VMEXIT_CONTINUE); } static int vmexit_mtrap(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) { assert(vmexit->inst_length == 0); stats.vmexit_mtrap++; #ifdef BHYVE_SNAPSHOT checkpoint_cpu_suspend(*pvcpu); #endif gdb_cpu_mtrap(*pvcpu); #ifdef BHYVE_SNAPSHOT checkpoint_cpu_resume(*pvcpu); #endif return (VMEXIT_CONTINUE); } static int vmexit_inst_emul(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) { int err, i, cs_d; struct vie *vie; enum vm_cpu_mode mode; stats.vmexit_inst_emul++; vie = &vmexit->u.inst_emul.vie; if (!vie->decoded) { /* * Attempt to decode in userspace as a fallback. This allows * updating instruction decode in bhyve without rebooting the * kernel (rapid prototyping), albeit with much slower * emulation. */ vie_restart(vie); mode = vmexit->u.inst_emul.paging.cpu_mode; cs_d = vmexit->u.inst_emul.cs_d; if (vmm_decode_instruction(mode, cs_d, vie) != 0) goto fail; if (vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RIP, vmexit->rip + vie->num_processed) != 0) goto fail; } err = emulate_mem(ctx, *pvcpu, vmexit->u.inst_emul.gpa, vie, &vmexit->u.inst_emul.paging); if (err) { if (err == ESRCH) { EPRINTLN("Unhandled memory access to 0x%lx\n", vmexit->u.inst_emul.gpa); } goto fail; } return (VMEXIT_CONTINUE); fail: fprintf(stderr, "Failed to emulate instruction sequence [ "); for (i = 0; i < vie->num_valid; i++) fprintf(stderr, "%02x", vie->inst[i]); FPRINTLN(stderr, " ] at 0x%lx", vmexit->rip); return (VMEXIT_ABORT); } static pthread_mutex_t resetcpu_mtx = PTHREAD_MUTEX_INITIALIZER; static pthread_cond_t resetcpu_cond = PTHREAD_COND_INITIALIZER; static int vmexit_suspend(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) { enum vm_suspend_how how; how = vmexit->u.suspended.how; fbsdrun_deletecpu(ctx, *pvcpu); if (*pvcpu != BSP) { pthread_mutex_lock(&resetcpu_mtx); pthread_cond_signal(&resetcpu_cond); pthread_mutex_unlock(&resetcpu_mtx); pthread_exit(NULL); } pthread_mutex_lock(&resetcpu_mtx); while (!CPU_EMPTY(&cpumask)) { pthread_cond_wait(&resetcpu_cond, &resetcpu_mtx); } pthread_mutex_unlock(&resetcpu_mtx); switch (how) { case VM_SUSPEND_RESET: exit(0); case VM_SUSPEND_POWEROFF: if (get_config_bool_default("destroy_on_poweroff", false)) vm_destroy(ctx); exit(1); case VM_SUSPEND_HALT: exit(2); case VM_SUSPEND_TRIPLEFAULT: exit(3); default: fprintf(stderr, "vmexit_suspend: invalid reason %d\n", how); exit(100); } return (0); /* NOTREACHED */ } static int vmexit_debug(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) { #ifdef BHYVE_SNAPSHOT checkpoint_cpu_suspend(*pvcpu); #endif gdb_cpu_suspend(*pvcpu); #ifdef BHYVE_SNAPSHOT checkpoint_cpu_resume(*pvcpu); #endif return (VMEXIT_CONTINUE); } static int vmexit_breakpoint(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) { gdb_cpu_breakpoint(*pvcpu, vmexit); return (VMEXIT_CONTINUE); } static vmexit_handler_t handler[VM_EXITCODE_MAX] = { [VM_EXITCODE_INOUT] = vmexit_inout, [VM_EXITCODE_INOUT_STR] = vmexit_inout, [VM_EXITCODE_VMX] = vmexit_vmx, [VM_EXITCODE_SVM] = vmexit_svm, [VM_EXITCODE_BOGUS] = vmexit_bogus, [VM_EXITCODE_REQIDLE] = vmexit_reqidle, [VM_EXITCODE_RDMSR] = vmexit_rdmsr, [VM_EXITCODE_WRMSR] = vmexit_wrmsr, [VM_EXITCODE_MTRAP] = vmexit_mtrap, [VM_EXITCODE_INST_EMUL] = vmexit_inst_emul, [VM_EXITCODE_SPINUP_AP] = vmexit_spinup_ap, [VM_EXITCODE_SUSPENDED] = vmexit_suspend, [VM_EXITCODE_TASK_SWITCH] = vmexit_task_switch, [VM_EXITCODE_DEBUG] = vmexit_debug, [VM_EXITCODE_BPT] = vmexit_breakpoint, }; static void vm_loop(struct vmctx *ctx, int vcpu, uint64_t startrip) { int error, rc; enum vm_exitcode exitcode; cpuset_t active_cpus; if (vcpumap[vcpu] != NULL) { error = pthread_setaffinity_np(pthread_self(), sizeof(cpuset_t), vcpumap[vcpu]); assert(error == 0); } error = vm_active_cpus(ctx, &active_cpus); assert(CPU_ISSET(vcpu, &active_cpus)); error = vm_set_register(ctx, vcpu, VM_REG_GUEST_RIP, startrip); assert(error == 0); while (1) { error = vm_run(ctx, vcpu, &vmexit[vcpu]); if (error != 0) break; exitcode = vmexit[vcpu].exitcode; if (exitcode >= VM_EXITCODE_MAX || handler[exitcode] == NULL) { fprintf(stderr, "vm_loop: unexpected exitcode 0x%x\n", exitcode); exit(4); } rc = (*handler[exitcode])(ctx, &vmexit[vcpu], &vcpu); switch (rc) { case VMEXIT_CONTINUE: break; case VMEXIT_ABORT: abort(); default: exit(4); } } fprintf(stderr, "vm_run error %d, errno %d\n", error, errno); } static int num_vcpus_allowed(struct vmctx *ctx) { uint16_t sockets, cores, threads, maxcpus; int tmp, error; /* * The guest is allowed to spinup more than one processor only if the * UNRESTRICTED_GUEST capability is available. */ error = vm_get_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, &tmp); if (error != 0) return (1); error = vm_get_topology(ctx, &sockets, &cores, &threads, &maxcpus); if (error == 0) return (maxcpus); else return (1); } void fbsdrun_set_capabilities(struct vmctx *ctx, int cpu) { int err, tmp; if (get_config_bool_default("x86.vmexit_on_hlt", false)) { err = vm_get_capability(ctx, cpu, VM_CAP_HALT_EXIT, &tmp); if (err < 0) { fprintf(stderr, "VM exit on HLT not supported\n"); exit(4); } vm_set_capability(ctx, cpu, VM_CAP_HALT_EXIT, 1); if (cpu == BSP) handler[VM_EXITCODE_HLT] = vmexit_hlt; } if (get_config_bool_default("x86.vmexit_on_pause", false)) { /* * pause exit support required for this mode */ err = vm_get_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, &tmp); if (err < 0) { fprintf(stderr, "SMP mux requested, no pause support\n"); exit(4); } vm_set_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, 1); if (cpu == BSP) handler[VM_EXITCODE_PAUSE] = vmexit_pause; } if (get_config_bool_default("x86.x2apic", false)) err = vm_set_x2apic_state(ctx, cpu, X2APIC_ENABLED); else err = vm_set_x2apic_state(ctx, cpu, X2APIC_DISABLED); if (err) { fprintf(stderr, "Unable to set x2apic state (%d)\n", err); exit(4); } vm_set_capability(ctx, cpu, VM_CAP_ENABLE_INVPCID, 1); } static struct vmctx * do_open(const char *vmname) { struct vmctx *ctx; int error; bool reinit, romboot; reinit = romboot = false; if (lpc_bootrom()) romboot = true; error = vm_create(vmname); if (error) { if (errno == EEXIST) { if (romboot) { reinit = true; } else { /* * The virtual machine has been setup by the * userspace bootloader. */ } } else { perror("vm_create"); exit(4); } } else { if (!romboot) { /* * If the virtual machine was just created then a * bootrom must be configured to boot it. */ fprintf(stderr, "virtual machine cannot be booted\n"); exit(4); } } ctx = vm_open(vmname); if (ctx == NULL) { perror("vm_open"); exit(4); } #ifndef WITHOUT_CAPSICUM if (vm_limit_rights(ctx) != 0) err(EX_OSERR, "vm_limit_rights"); #endif if (reinit) { error = vm_reinit(ctx); if (error) { perror("vm_reinit"); exit(4); } } error = vm_set_topology(ctx, cpu_sockets, cpu_cores, cpu_threads, 0 /* maxcpus, unimplemented */); if (error) errx(EX_OSERR, "vm_set_topology"); return (ctx); } void spinup_vcpu(struct vmctx *ctx, int vcpu) { int error; uint64_t rip; error = vm_get_register(ctx, vcpu, VM_REG_GUEST_RIP, &rip); assert(error == 0); fbsdrun_set_capabilities(ctx, vcpu); error = vm_set_capability(ctx, vcpu, VM_CAP_UNRESTRICTED_GUEST, 1); assert(error == 0); fbsdrun_addcpu(ctx, BSP, vcpu, rip); } static bool parse_config_option(const char *option) { const char *value; char *path; value = strchr(option, '='); if (value == NULL || value[1] == '\0') return (false); path = strndup(option, value - option); if (path == NULL) err(4, "Failed to allocate memory"); set_config_value(path, value + 1); return (true); } static void parse_simple_config_file(const char *path) { FILE *fp; char *line, *cp; size_t linecap; unsigned int lineno; fp = fopen(path, "r"); if (fp == NULL) err(4, "Failed to open configuration file %s", path); line = NULL; linecap = 0; lineno = 1; for (lineno = 1; getline(&line, &linecap, fp) > 0; lineno++) { if (*line == '#' || *line == '\n') continue; cp = strchr(line, '\n'); if (cp != NULL) *cp = '\0'; if (!parse_config_option(line)) errx(4, "%s line %u: invalid config option '%s'", path, lineno, line); } free(line); fclose(fp); } +static void +parse_gdb_options(char *optarg) +{ + const char *sport; + char *colon; + + if (optarg[0] == 'w') { + set_config_bool("gdb.wait", true); + optarg++; + } + + colon = strrchr(optarg, ':'); + if (colon == NULL) { + sport = optarg; + } else { + *colon = '\0'; + colon++; + sport = colon; + set_config_value("gdb.address", optarg); + } + + set_config_value("gdb.port", sport); +} + static void set_defaults(void) { set_config_bool("acpi_tables", false); set_config_value("memory.size", "256M"); set_config_bool("x86.strictmsr", true); } int main(int argc, char *argv[]) { int c, error, err; int max_vcpus, memflags; struct vmctx *ctx; uint64_t rip; size_t memsize; const char *value, *vmname; char *optstr; #ifdef BHYVE_SNAPSHOT char *restore_file; struct restore_state rstate; int vcpu; restore_file = NULL; #endif init_config(); set_defaults(); progname = basename(argv[0]); #ifdef BHYVE_SNAPSHOT optstr = "aehuwxACDHIPSWYk:o:p:G:c:s:m:l:U:r:"; #else optstr = "aehuwxACDHIPSWYk:o:p:G:c:s:m:l:U:"; #endif while ((c = getopt(argc, argv, optstr)) != -1) { switch (c) { case 'a': set_config_bool("x86.x2apic", false); break; case 'A': set_config_bool("acpi_tables", true); break; case 'D': set_config_bool("destroy_on_poweroff", true); break; case 'p': if (pincpu_parse(optarg) != 0) { errx(EX_USAGE, "invalid vcpu pinning " "configuration '%s'", optarg); } break; case 'c': if (topology_parse(optarg) != 0) { errx(EX_USAGE, "invalid cpu topology " "'%s'", optarg); } break; case 'C': set_config_bool("memory.guest_in_core", true); break; case 'G': - if (optarg[0] == 'w') { - set_config_bool("gdb.wait", true); - optarg++; - } - set_config_value("gdb.port", optarg); + parse_gdb_options(optarg); break; case 'k': parse_simple_config_file(optarg); break; case 'l': if (strncmp(optarg, "help", strlen(optarg)) == 0) { lpc_print_supported_devices(); exit(0); } else if (lpc_device_parse(optarg) != 0) { errx(EX_USAGE, "invalid lpc device " "configuration '%s'", optarg); } break; #ifdef BHYVE_SNAPSHOT case 'r': restore_file = optarg; break; #endif case 's': if (strncmp(optarg, "help", strlen(optarg)) == 0) { pci_print_supported_devices(); exit(0); } else if (pci_parse_slot(optarg) != 0) exit(4); else break; case 'S': set_config_bool("memory.wired", true); break; case 'm': set_config_value("memory.size", optarg); break; case 'o': if (!parse_config_option(optarg)) errx(EX_USAGE, "invalid configuration option '%s'", optarg); break; case 'H': set_config_bool("x86.vmexit_on_hlt", true); break; case 'I': /* * The "-I" option was used to add an ioapic to the * virtual machine. * * An ioapic is now provided unconditionally for each * virtual machine and this option is now deprecated. */ break; case 'P': set_config_bool("x86.vmexit_on_pause", true); break; case 'e': set_config_bool("x86.strictio", true); break; case 'u': set_config_bool("rtc.use_localtime", false); break; case 'U': set_config_value("uuid", optarg); break; case 'w': set_config_bool("x86.strictmsr", false); break; case 'W': set_config_bool("virtio_msix", false); break; case 'x': set_config_bool("x86.x2apic", true); break; case 'Y': set_config_bool("x86.mptable", false); break; case 'h': usage(0); default: usage(1); } } argc -= optind; argv += optind; if (argc > 1) usage(1); #ifdef BHYVE_SNAPSHOT if (restore_file != NULL) { error = load_restore_file(restore_file, &rstate); if (error) { fprintf(stderr, "Failed to read checkpoint info from " "file: '%s'.\n", restore_file); exit(1); } vmname = lookup_vmname(&rstate); if (vmname != NULL) set_config_value("name", vmname); } #endif if (argc == 1) set_config_value("name", argv[0]); vmname = get_config_value("name"); if (vmname == NULL) usage(1); if (get_config_bool_default("config.dump", false)) { dump_config(); exit(1); } calc_topology(); build_vcpumaps(); value = get_config_value("memory.size"); error = vm_parse_memsize(value, &memsize); if (error) errx(EX_USAGE, "invalid memsize '%s'", value); ctx = do_open(vmname); #ifdef BHYVE_SNAPSHOT if (restore_file != NULL) { guest_ncpus = lookup_guest_ncpus(&rstate); memflags = lookup_memflags(&rstate); memsize = lookup_memsize(&rstate); } if (guest_ncpus < 1) { fprintf(stderr, "Invalid guest vCPUs (%d)\n", guest_ncpus); exit(1); } #endif max_vcpus = num_vcpus_allowed(ctx); if (guest_ncpus > max_vcpus) { fprintf(stderr, "%d vCPUs requested but only %d available\n", guest_ncpus, max_vcpus); exit(4); } fbsdrun_set_capabilities(ctx, BSP); memflags = 0; if (get_config_bool_default("memory.wired", false)) memflags |= VM_MEM_F_WIRED; if (get_config_bool_default("memory.guest_in_core", false)) memflags |= VM_MEM_F_INCORE; vm_set_memflags(ctx, memflags); err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL); if (err) { fprintf(stderr, "Unable to setup memory (%d)\n", errno); exit(4); } error = init_msr(); if (error) { fprintf(stderr, "init_msr error %d", error); exit(4); } init_mem(guest_ncpus); init_inout(); kernemu_dev_init(); init_bootrom(ctx); atkbdc_init(ctx); pci_irq_init(ctx); ioapic_init(ctx); rtc_init(ctx); sci_init(ctx); /* * Exit if a device emulation finds an error in its initilization */ if (init_pci(ctx) != 0) { perror("device emulation initialization error"); exit(4); } /* * Initialize after PCI, to allow a bootrom file to reserve the high * region. */ if (get_config_bool("acpi_tables")) vmgenc_init(ctx); - value = get_config_value("gdb.port"); - if (value != NULL) - init_gdb(ctx, atoi(value), get_config_bool_default("gdb.wait", - false)); + init_gdb(ctx); if (lpc_bootrom()) { if (vm_set_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, 1)) { fprintf(stderr, "ROM boot failed: unrestricted guest " "capability not available\n"); exit(4); } error = vcpu_reset(ctx, BSP); assert(error == 0); } #ifdef BHYVE_SNAPSHOT if (restore_file != NULL) { fprintf(stdout, "Pausing pci devs...\r\n"); if (vm_pause_user_devs(ctx) != 0) { fprintf(stderr, "Failed to pause PCI device state.\n"); exit(1); } fprintf(stdout, "Restoring vm mem...\r\n"); if (restore_vm_mem(ctx, &rstate) != 0) { fprintf(stderr, "Failed to restore VM memory.\n"); exit(1); } fprintf(stdout, "Restoring pci devs...\r\n"); if (vm_restore_user_devs(ctx, &rstate) != 0) { fprintf(stderr, "Failed to restore PCI device state.\n"); exit(1); } fprintf(stdout, "Restoring kernel structs...\r\n"); if (vm_restore_kern_structs(ctx, &rstate) != 0) { fprintf(stderr, "Failed to restore kernel structs.\n"); exit(1); } fprintf(stdout, "Resuming pci devs...\r\n"); if (vm_resume_user_devs(ctx) != 0) { fprintf(stderr, "Failed to resume PCI device state.\n"); exit(1); } } #endif error = vm_get_register(ctx, BSP, VM_REG_GUEST_RIP, &rip); assert(error == 0); /* * build the guest tables, MP etc. */ if (get_config_bool_default("x86.mptable", true)) { error = mptable_build(ctx, guest_ncpus); if (error) { perror("error to build the guest tables"); exit(4); } } error = smbios_build(ctx); assert(error == 0); if (get_config_bool("acpi_tables")) { error = acpi_build(ctx, guest_ncpus); assert(error == 0); } if (lpc_bootrom()) fwctl_init(); /* * Change the proc title to include the VM name. */ setproctitle("%s", vmname); #ifndef WITHOUT_CAPSICUM caph_cache_catpages(); if (caph_limit_stdout() == -1 || caph_limit_stderr() == -1) errx(EX_OSERR, "Unable to apply rights for sandbox"); if (caph_enter() == -1) errx(EX_OSERR, "cap_enter() failed"); #endif #ifdef BHYVE_SNAPSHOT if (restore_file != NULL) destroy_restore_state(&rstate); /* * checkpointing thread for communication with bhyvectl */ if (init_checkpoint_thread(ctx) < 0) printf("Failed to start checkpoint thread!\r\n"); if (restore_file != NULL) vm_restore_time(ctx); #endif /* Allocate per-VCPU resources. */ vmexit = calloc(guest_ncpus, sizeof(*vmexit)); mt_vmm_info = calloc(guest_ncpus, sizeof(*mt_vmm_info)); /* * Add CPU 0 */ fbsdrun_addcpu(ctx, BSP, BSP, rip); #ifdef BHYVE_SNAPSHOT /* * If we restore a VM, start all vCPUs now (including APs), otherwise, * let the guest OS to spin them up later via vmexits. */ if (restore_file != NULL) { for (vcpu = 0; vcpu < guest_ncpus; vcpu++) { if (vcpu == BSP) continue; fprintf(stdout, "spinning up vcpu no %d...\r\n", vcpu); spinup_vcpu(ctx, vcpu); } } #endif /* * Head off to the main event dispatch loop */ mevent_dispatch(); exit(4); } diff --git a/usr.sbin/bhyve/gdb.c b/usr.sbin/bhyve/gdb.c index dd6f19d476f0..40f226701a1a 100644 --- a/usr.sbin/bhyve/gdb.c +++ b/usr.sbin/bhyve/gdb.c @@ -1,1877 +1,1907 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2017-2018 John H. Baldwin * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #ifndef WITHOUT_CAPSICUM #include #endif #include #include #include #include #include #include #include #include #include #include #ifndef WITHOUT_CAPSICUM #include #endif #include #include #include +#include #include #include #include #include #include #include #include #include #include #include "bhyverun.h" +#include "config.h" #include "gdb.h" #include "mem.h" #include "mevent.h" /* * GDB_SIGNAL_* numbers are part of the GDB remote protocol. Most stops * use SIGTRAP. */ #define GDB_SIGNAL_TRAP 5 static void gdb_resume_vcpus(void); static void check_command(int fd); static struct mevent *read_event, *write_event; static cpuset_t vcpus_active, vcpus_suspended, vcpus_waiting; static pthread_mutex_t gdb_lock; static pthread_cond_t idle_vcpus; static bool first_stop, report_next_stop, swbreak_enabled; /* * An I/O buffer contains 'capacity' bytes of room at 'data'. For a * read buffer, 'start' is unused and 'len' contains the number of * valid bytes in the buffer. For a write buffer, 'start' is set to * the index of the next byte in 'data' to send, and 'len' contains * the remaining number of valid bytes to send. */ struct io_buffer { uint8_t *data; size_t capacity; size_t start; size_t len; }; struct breakpoint { uint64_t gpa; uint8_t shadow_inst; TAILQ_ENTRY(breakpoint) link; }; /* * When a vCPU stops to due to an event that should be reported to the * debugger, information about the event is stored in this structure. * The vCPU thread then sets 'stopped_vcpu' if it is not already set * and stops other vCPUs so the event can be reported. The * report_stop() function reports the event for the 'stopped_vcpu' * vCPU. When the debugger resumes execution via continue or step, * the event for 'stopped_vcpu' is cleared. vCPUs will loop in their * event handlers until the associated event is reported or disabled. * * An idle vCPU will have all of the boolean fields set to false. * * When a vCPU is stepped, 'stepping' is set to true when the vCPU is * released to execute the stepped instruction. When the vCPU reports * the stepping trap, 'stepped' is set. * * When a vCPU hits a breakpoint set by the debug server, * 'hit_swbreak' is set to true. */ struct vcpu_state { bool stepping; bool stepped; bool hit_swbreak; }; static struct io_buffer cur_comm, cur_resp; static uint8_t cur_csum; static struct vmctx *ctx; static int cur_fd = -1; static TAILQ_HEAD(, breakpoint) breakpoints; static struct vcpu_state *vcpu_state; static int cur_vcpu, stopped_vcpu; static bool gdb_active = false; static const int gdb_regset[] = { VM_REG_GUEST_RAX, VM_REG_GUEST_RBX, VM_REG_GUEST_RCX, VM_REG_GUEST_RDX, VM_REG_GUEST_RSI, VM_REG_GUEST_RDI, VM_REG_GUEST_RBP, VM_REG_GUEST_RSP, VM_REG_GUEST_R8, VM_REG_GUEST_R9, VM_REG_GUEST_R10, VM_REG_GUEST_R11, VM_REG_GUEST_R12, VM_REG_GUEST_R13, VM_REG_GUEST_R14, VM_REG_GUEST_R15, VM_REG_GUEST_RIP, VM_REG_GUEST_RFLAGS, VM_REG_GUEST_CS, VM_REG_GUEST_SS, VM_REG_GUEST_DS, VM_REG_GUEST_ES, VM_REG_GUEST_FS, VM_REG_GUEST_GS }; static const int gdb_regsize[] = { 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 4, 4, 4, 4, 4, 4, 4 }; #ifdef GDB_LOG #include #include static void __printflike(1, 2) debug(const char *fmt, ...) { static FILE *logfile; va_list ap; if (logfile == NULL) { logfile = fopen("/tmp/bhyve_gdb.log", "w"); if (logfile == NULL) return; #ifndef WITHOUT_CAPSICUM if (caph_limit_stream(fileno(logfile), CAPH_WRITE) == -1) { fclose(logfile); logfile = NULL; return; } #endif setlinebuf(logfile); } va_start(ap, fmt); vfprintf(logfile, fmt, ap); va_end(ap); } #else #define debug(...) #endif static void remove_all_sw_breakpoints(void); static int guest_paging_info(int vcpu, struct vm_guest_paging *paging) { uint64_t regs[4]; const int regset[4] = { VM_REG_GUEST_CR0, VM_REG_GUEST_CR3, VM_REG_GUEST_CR4, VM_REG_GUEST_EFER }; if (vm_get_register_set(ctx, vcpu, nitems(regset), regset, regs) == -1) return (-1); /* * For the debugger, always pretend to be the kernel (CPL 0), * and if long-mode is enabled, always parse addresses as if * in 64-bit mode. */ paging->cr3 = regs[1]; paging->cpl = 0; if (regs[3] & EFER_LMA) paging->cpu_mode = CPU_MODE_64BIT; else if (regs[0] & CR0_PE) paging->cpu_mode = CPU_MODE_PROTECTED; else paging->cpu_mode = CPU_MODE_REAL; if (!(regs[0] & CR0_PG)) paging->paging_mode = PAGING_MODE_FLAT; else if (!(regs[2] & CR4_PAE)) paging->paging_mode = PAGING_MODE_32; else if (regs[3] & EFER_LME) paging->paging_mode = (regs[2] & CR4_LA57) ? PAGING_MODE_64_LA57 : PAGING_MODE_64; else paging->paging_mode = PAGING_MODE_PAE; return (0); } /* * Map a guest virtual address to a physical address (for a given vcpu). * If a guest virtual address is valid, return 1. If the address is * not valid, return 0. If an error occurs obtaining the mapping, * return -1. */ static int guest_vaddr2paddr(int vcpu, uint64_t vaddr, uint64_t *paddr) { struct vm_guest_paging paging; int fault; if (guest_paging_info(vcpu, &paging) == -1) return (-1); /* * Always use PROT_READ. We really care if the VA is * accessible, not if the current vCPU can write. */ if (vm_gla2gpa_nofault(ctx, vcpu, &paging, vaddr, PROT_READ, paddr, &fault) == -1) return (-1); if (fault) return (0); return (1); } static void io_buffer_reset(struct io_buffer *io) { io->start = 0; io->len = 0; } /* Available room for adding data. */ static size_t io_buffer_avail(struct io_buffer *io) { return (io->capacity - (io->start + io->len)); } static uint8_t * io_buffer_head(struct io_buffer *io) { return (io->data + io->start); } static uint8_t * io_buffer_tail(struct io_buffer *io) { return (io->data + io->start + io->len); } static void io_buffer_advance(struct io_buffer *io, size_t amount) { assert(amount <= io->len); io->start += amount; io->len -= amount; } static void io_buffer_consume(struct io_buffer *io, size_t amount) { io_buffer_advance(io, amount); if (io->len == 0) { io->start = 0; return; } /* * XXX: Consider making this move optional and compacting on a * future read() before realloc(). */ memmove(io->data, io_buffer_head(io), io->len); io->start = 0; } static void io_buffer_grow(struct io_buffer *io, size_t newsize) { uint8_t *new_data; size_t avail, new_cap; avail = io_buffer_avail(io); if (newsize <= avail) return; new_cap = io->capacity + (newsize - avail); new_data = realloc(io->data, new_cap); if (new_data == NULL) err(1, "Failed to grow GDB I/O buffer"); io->data = new_data; io->capacity = new_cap; } static bool response_pending(void) { if (cur_resp.start == 0 && cur_resp.len == 0) return (false); if (cur_resp.start + cur_resp.len == 1 && cur_resp.data[0] == '+') return (false); return (true); } static void close_connection(void) { /* * XXX: This triggers a warning because mevent does the close * before the EV_DELETE. */ pthread_mutex_lock(&gdb_lock); mevent_delete(write_event); mevent_delete_close(read_event); write_event = NULL; read_event = NULL; io_buffer_reset(&cur_comm); io_buffer_reset(&cur_resp); cur_fd = -1; remove_all_sw_breakpoints(); /* Clear any pending events. */ memset(vcpu_state, 0, guest_ncpus * sizeof(*vcpu_state)); /* Resume any stopped vCPUs. */ gdb_resume_vcpus(); pthread_mutex_unlock(&gdb_lock); } static uint8_t hex_digit(uint8_t nibble) { if (nibble <= 9) return (nibble + '0'); else return (nibble + 'a' - 10); } static uint8_t parse_digit(uint8_t v) { if (v >= '0' && v <= '9') return (v - '0'); if (v >= 'a' && v <= 'f') return (v - 'a' + 10); if (v >= 'A' && v <= 'F') return (v - 'A' + 10); return (0xF); } /* Parses big-endian hexadecimal. */ static uintmax_t parse_integer(const uint8_t *p, size_t len) { uintmax_t v; v = 0; while (len > 0) { v <<= 4; v |= parse_digit(*p); p++; len--; } return (v); } static uint8_t parse_byte(const uint8_t *p) { return (parse_digit(p[0]) << 4 | parse_digit(p[1])); } static void send_pending_data(int fd) { ssize_t nwritten; if (cur_resp.len == 0) { mevent_disable(write_event); return; } nwritten = write(fd, io_buffer_head(&cur_resp), cur_resp.len); if (nwritten == -1) { warn("Write to GDB socket failed"); close_connection(); } else { io_buffer_advance(&cur_resp, nwritten); if (cur_resp.len == 0) mevent_disable(write_event); else mevent_enable(write_event); } } /* Append a single character to the output buffer. */ static void send_char(uint8_t data) { io_buffer_grow(&cur_resp, 1); *io_buffer_tail(&cur_resp) = data; cur_resp.len++; } /* Append an array of bytes to the output buffer. */ static void send_data(const uint8_t *data, size_t len) { io_buffer_grow(&cur_resp, len); memcpy(io_buffer_tail(&cur_resp), data, len); cur_resp.len += len; } static void format_byte(uint8_t v, uint8_t *buf) { buf[0] = hex_digit(v >> 4); buf[1] = hex_digit(v & 0xf); } /* * Append a single byte (formatted as two hex characters) to the * output buffer. */ static void send_byte(uint8_t v) { uint8_t buf[2]; format_byte(v, buf); send_data(buf, sizeof(buf)); } static void start_packet(void) { send_char('$'); cur_csum = 0; } static void finish_packet(void) { send_char('#'); send_byte(cur_csum); debug("-> %.*s\n", (int)cur_resp.len, io_buffer_head(&cur_resp)); } /* * Append a single character (for the packet payload) and update the * checksum. */ static void append_char(uint8_t v) { send_char(v); cur_csum += v; } /* * Append an array of bytes (for the packet payload) and update the * checksum. */ static void append_packet_data(const uint8_t *data, size_t len) { send_data(data, len); while (len > 0) { cur_csum += *data; data++; len--; } } static void append_string(const char *str) { append_packet_data(str, strlen(str)); } static void append_byte(uint8_t v) { uint8_t buf[2]; format_byte(v, buf); append_packet_data(buf, sizeof(buf)); } static void append_unsigned_native(uintmax_t value, size_t len) { size_t i; for (i = 0; i < len; i++) { append_byte(value); value >>= 8; } } static void append_unsigned_be(uintmax_t value, size_t len) { char buf[len * 2]; size_t i; for (i = 0; i < len; i++) { format_byte(value, buf + (len - i - 1) * 2); value >>= 8; } append_packet_data(buf, sizeof(buf)); } static void append_integer(unsigned int value) { if (value == 0) append_char('0'); else append_unsigned_be(value, (fls(value) + 7) / 8); } static void append_asciihex(const char *str) { while (*str != '\0') { append_byte(*str); str++; } } static void send_empty_response(void) { start_packet(); finish_packet(); } static void send_error(int error) { start_packet(); append_char('E'); append_byte(error); finish_packet(); } static void send_ok(void) { start_packet(); append_string("OK"); finish_packet(); } static int parse_threadid(const uint8_t *data, size_t len) { if (len == 1 && *data == '0') return (0); if (len == 2 && memcmp(data, "-1", 2) == 0) return (-1); if (len == 0) return (-2); return (parse_integer(data, len)); } /* * Report the current stop event to the debugger. If the stop is due * to an event triggered on a specific vCPU such as a breakpoint or * stepping trap, stopped_vcpu will be set to the vCPU triggering the * stop. If 'set_cur_vcpu' is true, then cur_vcpu will be updated to * the reporting vCPU for vCPU events. */ static void report_stop(bool set_cur_vcpu) { struct vcpu_state *vs; start_packet(); if (stopped_vcpu == -1) { append_char('S'); append_byte(GDB_SIGNAL_TRAP); } else { vs = &vcpu_state[stopped_vcpu]; if (set_cur_vcpu) cur_vcpu = stopped_vcpu; append_char('T'); append_byte(GDB_SIGNAL_TRAP); append_string("thread:"); append_integer(stopped_vcpu + 1); append_char(';'); if (vs->hit_swbreak) { debug("$vCPU %d reporting swbreak\n", stopped_vcpu); if (swbreak_enabled) append_string("swbreak:;"); } else if (vs->stepped) debug("$vCPU %d reporting step\n", stopped_vcpu); else debug("$vCPU %d reporting ???\n", stopped_vcpu); } finish_packet(); report_next_stop = false; } /* * If this stop is due to a vCPU event, clear that event to mark it as * acknowledged. */ static void discard_stop(void) { struct vcpu_state *vs; if (stopped_vcpu != -1) { vs = &vcpu_state[stopped_vcpu]; vs->hit_swbreak = false; vs->stepped = false; stopped_vcpu = -1; } report_next_stop = true; } static void gdb_finish_suspend_vcpus(void) { if (first_stop) { first_stop = false; stopped_vcpu = -1; } else if (report_next_stop) { assert(!response_pending()); report_stop(true); send_pending_data(cur_fd); } } /* * vCPU threads invoke this function whenever the vCPU enters the * debug server to pause or report an event. vCPU threads wait here * as long as the debug server keeps them suspended. */ static void _gdb_cpu_suspend(int vcpu, bool report_stop) { debug("$vCPU %d suspending\n", vcpu); CPU_SET(vcpu, &vcpus_waiting); if (report_stop && CPU_CMP(&vcpus_waiting, &vcpus_suspended) == 0) gdb_finish_suspend_vcpus(); while (CPU_ISSET(vcpu, &vcpus_suspended)) pthread_cond_wait(&idle_vcpus, &gdb_lock); CPU_CLR(vcpu, &vcpus_waiting); debug("$vCPU %d resuming\n", vcpu); } /* * Invoked at the start of a vCPU thread's execution to inform the * debug server about the new thread. */ void gdb_cpu_add(int vcpu) { if (!gdb_active) return; debug("$vCPU %d starting\n", vcpu); pthread_mutex_lock(&gdb_lock); assert(vcpu < guest_ncpus); CPU_SET(vcpu, &vcpus_active); if (!TAILQ_EMPTY(&breakpoints)) { vm_set_capability(ctx, vcpu, VM_CAP_BPT_EXIT, 1); debug("$vCPU %d enabled breakpoint exits\n", vcpu); } /* * If a vcpu is added while vcpus are stopped, suspend the new * vcpu so that it will pop back out with a debug exit before * executing the first instruction. */ if (!CPU_EMPTY(&vcpus_suspended)) { CPU_SET(vcpu, &vcpus_suspended); _gdb_cpu_suspend(vcpu, false); } pthread_mutex_unlock(&gdb_lock); } /* * Invoked by vCPU before resuming execution. This enables stepping * if the vCPU is marked as stepping. */ static void gdb_cpu_resume(int vcpu) { struct vcpu_state *vs; int error; vs = &vcpu_state[vcpu]; /* * Any pending event should already be reported before * resuming. */ assert(vs->hit_swbreak == false); assert(vs->stepped == false); if (vs->stepping) { error = vm_set_capability(ctx, vcpu, VM_CAP_MTRAP_EXIT, 1); assert(error == 0); } } /* * Handler for VM_EXITCODE_DEBUG used to suspend a vCPU when the guest * has been suspended due to an event on different vCPU or in response * to a guest-wide suspend such as Ctrl-C or the stop on attach. */ void gdb_cpu_suspend(int vcpu) { if (!gdb_active) return; pthread_mutex_lock(&gdb_lock); _gdb_cpu_suspend(vcpu, true); gdb_cpu_resume(vcpu); pthread_mutex_unlock(&gdb_lock); } static void gdb_suspend_vcpus(void) { assert(pthread_mutex_isowned_np(&gdb_lock)); debug("suspending all CPUs\n"); vcpus_suspended = vcpus_active; vm_suspend_cpu(ctx, -1); if (CPU_CMP(&vcpus_waiting, &vcpus_suspended) == 0) gdb_finish_suspend_vcpus(); } /* * Handler for VM_EXITCODE_MTRAP reported when a vCPU single-steps via * the VT-x-specific MTRAP exit. */ void gdb_cpu_mtrap(int vcpu) { struct vcpu_state *vs; if (!gdb_active) return; debug("$vCPU %d MTRAP\n", vcpu); pthread_mutex_lock(&gdb_lock); vs = &vcpu_state[vcpu]; if (vs->stepping) { vs->stepping = false; vs->stepped = true; vm_set_capability(ctx, vcpu, VM_CAP_MTRAP_EXIT, 0); while (vs->stepped) { if (stopped_vcpu == -1) { debug("$vCPU %d reporting step\n", vcpu); stopped_vcpu = vcpu; gdb_suspend_vcpus(); } _gdb_cpu_suspend(vcpu, true); } gdb_cpu_resume(vcpu); } pthread_mutex_unlock(&gdb_lock); } static struct breakpoint * find_breakpoint(uint64_t gpa) { struct breakpoint *bp; TAILQ_FOREACH(bp, &breakpoints, link) { if (bp->gpa == gpa) return (bp); } return (NULL); } void gdb_cpu_breakpoint(int vcpu, struct vm_exit *vmexit) { struct breakpoint *bp; struct vcpu_state *vs; uint64_t gpa; int error; if (!gdb_active) { fprintf(stderr, "vm_loop: unexpected VMEXIT_DEBUG\n"); exit(4); } pthread_mutex_lock(&gdb_lock); error = guest_vaddr2paddr(vcpu, vmexit->rip, &gpa); assert(error == 1); bp = find_breakpoint(gpa); if (bp != NULL) { vs = &vcpu_state[vcpu]; assert(vs->stepping == false); assert(vs->stepped == false); assert(vs->hit_swbreak == false); vs->hit_swbreak = true; vm_set_register(ctx, vcpu, VM_REG_GUEST_RIP, vmexit->rip); for (;;) { if (stopped_vcpu == -1) { debug("$vCPU %d reporting breakpoint at rip %#lx\n", vcpu, vmexit->rip); stopped_vcpu = vcpu; gdb_suspend_vcpus(); } _gdb_cpu_suspend(vcpu, true); if (!vs->hit_swbreak) { /* Breakpoint reported. */ break; } bp = find_breakpoint(gpa); if (bp == NULL) { /* Breakpoint was removed. */ vs->hit_swbreak = false; break; } } gdb_cpu_resume(vcpu); } else { debug("$vCPU %d injecting breakpoint at rip %#lx\n", vcpu, vmexit->rip); error = vm_set_register(ctx, vcpu, VM_REG_GUEST_ENTRY_INST_LENGTH, vmexit->u.bpt.inst_length); assert(error == 0); error = vm_inject_exception(ctx, vcpu, IDT_BP, 0, 0, 0); assert(error == 0); } pthread_mutex_unlock(&gdb_lock); } static bool gdb_step_vcpu(int vcpu) { int error, val; debug("$vCPU %d step\n", vcpu); error = vm_get_capability(ctx, vcpu, VM_CAP_MTRAP_EXIT, &val); if (error < 0) return (false); discard_stop(); vcpu_state[vcpu].stepping = true; vm_resume_cpu(ctx, vcpu); CPU_CLR(vcpu, &vcpus_suspended); pthread_cond_broadcast(&idle_vcpus); return (true); } static void gdb_resume_vcpus(void) { assert(pthread_mutex_isowned_np(&gdb_lock)); vm_resume_cpu(ctx, -1); debug("resuming all CPUs\n"); CPU_ZERO(&vcpus_suspended); pthread_cond_broadcast(&idle_vcpus); } static void gdb_read_regs(void) { uint64_t regvals[nitems(gdb_regset)]; int i; if (vm_get_register_set(ctx, cur_vcpu, nitems(gdb_regset), gdb_regset, regvals) == -1) { send_error(errno); return; } start_packet(); for (i = 0; i < nitems(regvals); i++) append_unsigned_native(regvals[i], gdb_regsize[i]); finish_packet(); } static void gdb_read_mem(const uint8_t *data, size_t len) { uint64_t gpa, gva, val; uint8_t *cp; size_t resid, todo, bytes; bool started; int error; /* Skip 'm' */ data += 1; len -= 1; /* Parse and consume address. */ cp = memchr(data, ',', len); if (cp == NULL || cp == data) { send_error(EINVAL); return; } gva = parse_integer(data, cp - data); len -= (cp - data) + 1; data += (cp - data) + 1; /* Parse length. */ resid = parse_integer(data, len); started = false; while (resid > 0) { error = guest_vaddr2paddr(cur_vcpu, gva, &gpa); if (error == -1) { if (started) finish_packet(); else send_error(errno); return; } if (error == 0) { if (started) finish_packet(); else send_error(EFAULT); return; } /* Read bytes from current page. */ todo = getpagesize() - gpa % getpagesize(); if (todo > resid) todo = resid; cp = paddr_guest2host(ctx, gpa, todo); if (cp != NULL) { /* * If this page is guest RAM, read it a byte * at a time. */ if (!started) { start_packet(); started = true; } while (todo > 0) { append_byte(*cp); cp++; gpa++; gva++; resid--; todo--; } } else { /* * If this page isn't guest RAM, try to handle * it via MMIO. For MMIO requests, use * aligned reads of words when possible. */ while (todo > 0) { if (gpa & 1 || todo == 1) bytes = 1; else if (gpa & 2 || todo == 2) bytes = 2; else bytes = 4; error = read_mem(ctx, cur_vcpu, gpa, &val, bytes); if (error == 0) { if (!started) { start_packet(); started = true; } gpa += bytes; gva += bytes; resid -= bytes; todo -= bytes; while (bytes > 0) { append_byte(val); val >>= 8; bytes--; } } else { if (started) finish_packet(); else send_error(EFAULT); return; } } } assert(resid == 0 || gpa % getpagesize() == 0); } if (!started) start_packet(); finish_packet(); } static void gdb_write_mem(const uint8_t *data, size_t len) { uint64_t gpa, gva, val; uint8_t *cp; size_t resid, todo, bytes; int error; /* Skip 'M' */ data += 1; len -= 1; /* Parse and consume address. */ cp = memchr(data, ',', len); if (cp == NULL || cp == data) { send_error(EINVAL); return; } gva = parse_integer(data, cp - data); len -= (cp - data) + 1; data += (cp - data) + 1; /* Parse and consume length. */ cp = memchr(data, ':', len); if (cp == NULL || cp == data) { send_error(EINVAL); return; } resid = parse_integer(data, cp - data); len -= (cp - data) + 1; data += (cp - data) + 1; /* Verify the available bytes match the length. */ if (len != resid * 2) { send_error(EINVAL); return; } while (resid > 0) { error = guest_vaddr2paddr(cur_vcpu, gva, &gpa); if (error == -1) { send_error(errno); return; } if (error == 0) { send_error(EFAULT); return; } /* Write bytes to current page. */ todo = getpagesize() - gpa % getpagesize(); if (todo > resid) todo = resid; cp = paddr_guest2host(ctx, gpa, todo); if (cp != NULL) { /* * If this page is guest RAM, write it a byte * at a time. */ while (todo > 0) { assert(len >= 2); *cp = parse_byte(data); data += 2; len -= 2; cp++; gpa++; gva++; resid--; todo--; } } else { /* * If this page isn't guest RAM, try to handle * it via MMIO. For MMIO requests, use * aligned writes of words when possible. */ while (todo > 0) { if (gpa & 1 || todo == 1) { bytes = 1; val = parse_byte(data); } else if (gpa & 2 || todo == 2) { bytes = 2; val = be16toh(parse_integer(data, 4)); } else { bytes = 4; val = be32toh(parse_integer(data, 8)); } error = write_mem(ctx, cur_vcpu, gpa, val, bytes); if (error == 0) { gpa += bytes; gva += bytes; resid -= bytes; todo -= bytes; data += 2 * bytes; len -= 2 * bytes; } else { send_error(EFAULT); return; } } } assert(resid == 0 || gpa % getpagesize() == 0); } assert(len == 0); send_ok(); } static bool set_breakpoint_caps(bool enable) { cpuset_t mask; int vcpu; mask = vcpus_active; while (!CPU_EMPTY(&mask)) { vcpu = CPU_FFS(&mask) - 1; CPU_CLR(vcpu, &mask); if (vm_set_capability(ctx, vcpu, VM_CAP_BPT_EXIT, enable ? 1 : 0) < 0) return (false); debug("$vCPU %d %sabled breakpoint exits\n", vcpu, enable ? "en" : "dis"); } return (true); } static void remove_all_sw_breakpoints(void) { struct breakpoint *bp, *nbp; uint8_t *cp; if (TAILQ_EMPTY(&breakpoints)) return; TAILQ_FOREACH_SAFE(bp, &breakpoints, link, nbp) { debug("remove breakpoint at %#lx\n", bp->gpa); cp = paddr_guest2host(ctx, bp->gpa, 1); *cp = bp->shadow_inst; TAILQ_REMOVE(&breakpoints, bp, link); free(bp); } TAILQ_INIT(&breakpoints); set_breakpoint_caps(false); } static void update_sw_breakpoint(uint64_t gva, int kind, bool insert) { struct breakpoint *bp; uint64_t gpa; uint8_t *cp; int error; if (kind != 1) { send_error(EINVAL); return; } error = guest_vaddr2paddr(cur_vcpu, gva, &gpa); if (error == -1) { send_error(errno); return; } if (error == 0) { send_error(EFAULT); return; } cp = paddr_guest2host(ctx, gpa, 1); /* Only permit breakpoints in guest RAM. */ if (cp == NULL) { send_error(EFAULT); return; } /* Find any existing breakpoint. */ bp = find_breakpoint(gpa); /* * Silently ignore duplicate commands since the protocol * requires these packets to be idempotent. */ if (insert) { if (bp == NULL) { if (TAILQ_EMPTY(&breakpoints) && !set_breakpoint_caps(true)) { send_empty_response(); return; } bp = malloc(sizeof(*bp)); bp->gpa = gpa; bp->shadow_inst = *cp; *cp = 0xcc; /* INT 3 */ TAILQ_INSERT_TAIL(&breakpoints, bp, link); debug("new breakpoint at %#lx\n", gpa); } } else { if (bp != NULL) { debug("remove breakpoint at %#lx\n", gpa); *cp = bp->shadow_inst; TAILQ_REMOVE(&breakpoints, bp, link); free(bp); if (TAILQ_EMPTY(&breakpoints)) set_breakpoint_caps(false); } } send_ok(); } static void parse_breakpoint(const uint8_t *data, size_t len) { uint64_t gva; uint8_t *cp; bool insert; int kind, type; insert = data[0] == 'Z'; /* Skip 'Z/z' */ data += 1; len -= 1; /* Parse and consume type. */ cp = memchr(data, ',', len); if (cp == NULL || cp == data) { send_error(EINVAL); return; } type = parse_integer(data, cp - data); len -= (cp - data) + 1; data += (cp - data) + 1; /* Parse and consume address. */ cp = memchr(data, ',', len); if (cp == NULL || cp == data) { send_error(EINVAL); return; } gva = parse_integer(data, cp - data); len -= (cp - data) + 1; data += (cp - data) + 1; /* Parse and consume kind. */ cp = memchr(data, ';', len); if (cp == data) { send_error(EINVAL); return; } if (cp != NULL) { /* * We do not advertise support for either the * ConditionalBreakpoints or BreakpointCommands * features, so we should not be getting conditions or * commands from the remote end. */ send_empty_response(); return; } kind = parse_integer(data, len); data += len; len = 0; switch (type) { case 0: update_sw_breakpoint(gva, kind, insert); break; default: send_empty_response(); break; } } static bool command_equals(const uint8_t *data, size_t len, const char *cmd) { if (strlen(cmd) > len) return (false); return (memcmp(data, cmd, strlen(cmd)) == 0); } static void check_features(const uint8_t *data, size_t len) { char *feature, *next_feature, *str, *value; bool supported; str = malloc(len + 1); memcpy(str, data, len); str[len] = '\0'; next_feature = str; while ((feature = strsep(&next_feature, ";")) != NULL) { /* * Null features shouldn't exist, but skip if they * do. */ if (strcmp(feature, "") == 0) continue; /* * Look for the value or supported / not supported * flag. */ value = strchr(feature, '='); if (value != NULL) { *value = '\0'; value++; supported = true; } else { value = feature + strlen(feature) - 1; switch (*value) { case '+': supported = true; break; case '-': supported = false; break; default: /* * This is really a protocol error, * but we just ignore malformed * features for ease of * implementation. */ continue; } value = NULL; } if (strcmp(feature, "swbreak") == 0) swbreak_enabled = supported; } free(str); start_packet(); /* This is an arbitrary limit. */ append_string("PacketSize=4096"); append_string(";swbreak+"); finish_packet(); } static void gdb_query(const uint8_t *data, size_t len) { /* * TODO: * - qSearch */ if (command_equals(data, len, "qAttached")) { start_packet(); append_char('1'); finish_packet(); } else if (command_equals(data, len, "qC")) { start_packet(); append_string("QC"); append_integer(cur_vcpu + 1); finish_packet(); } else if (command_equals(data, len, "qfThreadInfo")) { cpuset_t mask; bool first; int vcpu; if (CPU_EMPTY(&vcpus_active)) { send_error(EINVAL); return; } mask = vcpus_active; start_packet(); append_char('m'); first = true; while (!CPU_EMPTY(&mask)) { vcpu = CPU_FFS(&mask) - 1; CPU_CLR(vcpu, &mask); if (first) first = false; else append_char(','); append_integer(vcpu + 1); } finish_packet(); } else if (command_equals(data, len, "qsThreadInfo")) { start_packet(); append_char('l'); finish_packet(); } else if (command_equals(data, len, "qSupported")) { data += strlen("qSupported"); len -= strlen("qSupported"); check_features(data, len); } else if (command_equals(data, len, "qThreadExtraInfo")) { char buf[16]; int tid; data += strlen("qThreadExtraInfo"); len -= strlen("qThreadExtraInfo"); if (*data != ',') { send_error(EINVAL); return; } tid = parse_threadid(data + 1, len - 1); if (tid <= 0 || !CPU_ISSET(tid - 1, &vcpus_active)) { send_error(EINVAL); return; } snprintf(buf, sizeof(buf), "vCPU %d", tid - 1); start_packet(); append_asciihex(buf); finish_packet(); } else send_empty_response(); } static void handle_command(const uint8_t *data, size_t len) { /* Reject packets with a sequence-id. */ if (len >= 3 && data[0] >= '0' && data[0] <= '9' && data[0] >= '0' && data[0] <= '9' && data[2] == ':') { send_empty_response(); return; } switch (*data) { case 'c': if (len != 1) { send_error(EINVAL); break; } discard_stop(); gdb_resume_vcpus(); break; case 'D': send_ok(); /* TODO: Resume any stopped CPUs. */ break; case 'g': { gdb_read_regs(); break; } case 'H': { int tid; if (data[1] != 'g' && data[1] != 'c') { send_error(EINVAL); break; } tid = parse_threadid(data + 2, len - 2); if (tid == -2) { send_error(EINVAL); break; } if (CPU_EMPTY(&vcpus_active)) { send_error(EINVAL); break; } if (tid == -1 || tid == 0) cur_vcpu = CPU_FFS(&vcpus_active) - 1; else if (CPU_ISSET(tid - 1, &vcpus_active)) cur_vcpu = tid - 1; else { send_error(EINVAL); break; } send_ok(); break; } case 'm': gdb_read_mem(data, len); break; case 'M': gdb_write_mem(data, len); break; case 'T': { int tid; tid = parse_threadid(data + 1, len - 1); if (tid <= 0 || !CPU_ISSET(tid - 1, &vcpus_active)) { send_error(EINVAL); return; } send_ok(); break; } case 'q': gdb_query(data, len); break; case 's': if (len != 1) { send_error(EINVAL); break; } /* Don't send a reply until a stop occurs. */ if (!gdb_step_vcpu(cur_vcpu)) { send_error(EOPNOTSUPP); break; } break; case 'z': case 'Z': parse_breakpoint(data, len); break; case '?': report_stop(false); break; case 'G': /* TODO */ case 'v': /* Handle 'vCont' */ /* 'vCtrlC' */ case 'p': /* TODO */ case 'P': /* TODO */ case 'Q': /* TODO */ case 't': /* TODO */ case 'X': /* TODO */ default: send_empty_response(); } } /* Check for a valid packet in the command buffer. */ static void check_command(int fd) { uint8_t *head, *hash, *p, sum; size_t avail, plen; for (;;) { avail = cur_comm.len; if (avail == 0) return; head = io_buffer_head(&cur_comm); switch (*head) { case 0x03: debug("<- Ctrl-C\n"); io_buffer_consume(&cur_comm, 1); gdb_suspend_vcpus(); break; case '+': /* ACK of previous response. */ debug("<- +\n"); if (response_pending()) io_buffer_reset(&cur_resp); io_buffer_consume(&cur_comm, 1); if (stopped_vcpu != -1 && report_next_stop) { report_stop(true); send_pending_data(fd); } break; case '-': /* NACK of previous response. */ debug("<- -\n"); if (response_pending()) { cur_resp.len += cur_resp.start; cur_resp.start = 0; if (cur_resp.data[0] == '+') io_buffer_advance(&cur_resp, 1); debug("-> %.*s\n", (int)cur_resp.len, io_buffer_head(&cur_resp)); } io_buffer_consume(&cur_comm, 1); send_pending_data(fd); break; case '$': /* Packet. */ if (response_pending()) { warnx("New GDB command while response in " "progress"); io_buffer_reset(&cur_resp); } /* Is packet complete? */ hash = memchr(head, '#', avail); if (hash == NULL) return; plen = (hash - head + 1) + 2; if (avail < plen) return; debug("<- %.*s\n", (int)plen, head); /* Verify checksum. */ for (sum = 0, p = head + 1; p < hash; p++) sum += *p; if (sum != parse_byte(hash + 1)) { io_buffer_consume(&cur_comm, plen); debug("-> -\n"); send_char('-'); send_pending_data(fd); break; } send_char('+'); handle_command(head + 1, hash - (head + 1)); io_buffer_consume(&cur_comm, plen); if (!response_pending()) debug("-> +\n"); send_pending_data(fd); break; default: /* XXX: Possibly drop connection instead. */ debug("-> %02x\n", *head); io_buffer_consume(&cur_comm, 1); break; } } } static void gdb_readable(int fd, enum ev_type event, void *arg) { ssize_t nread; int pending; if (ioctl(fd, FIONREAD, &pending) == -1) { warn("FIONREAD on GDB socket"); return; } /* * 'pending' might be zero due to EOF. We need to call read * with a non-zero length to detect EOF. */ if (pending == 0) pending = 1; /* Ensure there is room in the command buffer. */ io_buffer_grow(&cur_comm, pending); assert(io_buffer_avail(&cur_comm) >= pending); nread = read(fd, io_buffer_tail(&cur_comm), io_buffer_avail(&cur_comm)); if (nread == 0) { close_connection(); } else if (nread == -1) { if (errno == EAGAIN) return; warn("Read from GDB socket"); close_connection(); } else { cur_comm.len += nread; pthread_mutex_lock(&gdb_lock); check_command(fd); pthread_mutex_unlock(&gdb_lock); } } static void gdb_writable(int fd, enum ev_type event, void *arg) { send_pending_data(fd); } static void new_connection(int fd, enum ev_type event, void *arg) { int optval, s; s = accept4(fd, NULL, NULL, SOCK_NONBLOCK); if (s == -1) { if (arg != NULL) err(1, "Failed accepting initial GDB connection"); /* Silently ignore errors post-startup. */ return; } optval = 1; if (setsockopt(s, SOL_SOCKET, SO_NOSIGPIPE, &optval, sizeof(optval)) == -1) { warn("Failed to disable SIGPIPE for GDB connection"); close(s); return; } pthread_mutex_lock(&gdb_lock); if (cur_fd != -1) { close(s); warnx("Ignoring additional GDB connection."); } read_event = mevent_add(s, EVF_READ, gdb_readable, NULL); if (read_event == NULL) { if (arg != NULL) err(1, "Failed to setup initial GDB connection"); pthread_mutex_unlock(&gdb_lock); return; } write_event = mevent_add(s, EVF_WRITE, gdb_writable, NULL); if (write_event == NULL) { if (arg != NULL) err(1, "Failed to setup initial GDB connection"); mevent_delete_close(read_event); read_event = NULL; } cur_fd = s; cur_vcpu = 0; stopped_vcpu = -1; /* Break on attach. */ first_stop = true; report_next_stop = false; gdb_suspend_vcpus(); pthread_mutex_unlock(&gdb_lock); } #ifndef WITHOUT_CAPSICUM static void limit_gdb_socket(int s) { cap_rights_t rights; unsigned long ioctls[] = { FIONREAD }; cap_rights_init(&rights, CAP_ACCEPT, CAP_EVENT, CAP_READ, CAP_WRITE, CAP_SETSOCKOPT, CAP_IOCTL); if (caph_rights_limit(s, &rights) == -1) errx(EX_OSERR, "Unable to apply rights for sandbox"); if (caph_ioctls_limit(s, ioctls, nitems(ioctls)) == -1) errx(EX_OSERR, "Unable to apply rights for sandbox"); } #endif void -init_gdb(struct vmctx *_ctx, int sport, bool wait) +init_gdb(struct vmctx *_ctx) { - struct sockaddr_in sin; int error, flags, optval, s; + struct addrinfo hints; + struct addrinfo *gdbaddr; + const char *saddr, *value; + char *sport; + bool wait; + + value = get_config_value("gdb.port"); + if (value == NULL) + return; + sport = strdup(value); + if (sport == NULL) + errx(4, "Failed to allocate memory"); + + wait = get_config_bool_default("gdb.wait", false); + + saddr = get_config_value("gdb.address"); + if (saddr == NULL) { +#if defined(INET) + saddr = "0.0.0.0"; +#elif defined(INET6) + saddr = "[::]"; +#endif + } - debug("==> starting on %d, %swaiting\n", sport, wait ? "" : "not "); + debug("==> starting on %s:%s, %swaiting\n", + saddr, sport, wait ? "" : "not "); error = pthread_mutex_init(&gdb_lock, NULL); if (error != 0) errc(1, error, "gdb mutex init"); error = pthread_cond_init(&idle_vcpus, NULL); if (error != 0) errc(1, error, "gdb cv init"); + memset(&hints, 0, sizeof(hints)); + hints.ai_family = AF_UNSPEC; + hints.ai_socktype = SOCK_STREAM; + hints.ai_flags = AI_NUMERICHOST | AI_NUMERICSERV | AI_PASSIVE; + + if (getaddrinfo(saddr, sport, &hints, &gdbaddr) != 0) + err(1, "gdb address resolve"); + ctx = _ctx; - s = socket(PF_INET, SOCK_STREAM, 0); + s = socket(gdbaddr->ai_family, gdbaddr->ai_socktype, 0); if (s < 0) err(1, "gdb socket create"); optval = 1; (void)setsockopt(s, SOL_SOCKET, SO_REUSEADDR, &optval, sizeof(optval)); - sin.sin_len = sizeof(sin); - sin.sin_family = AF_INET; - sin.sin_addr.s_addr = htonl(INADDR_ANY); - sin.sin_port = htons(sport); - - if (bind(s, (struct sockaddr *)&sin, sizeof(sin)) < 0) + if (bind(s, gdbaddr->ai_addr, gdbaddr->ai_addrlen) < 0) err(1, "gdb socket bind"); if (listen(s, 1) < 0) err(1, "gdb socket listen"); stopped_vcpu = -1; TAILQ_INIT(&breakpoints); vcpu_state = calloc(guest_ncpus, sizeof(*vcpu_state)); if (wait) { /* * Set vcpu 0 in vcpus_suspended. This will trigger the * logic in gdb_cpu_add() to suspend the first vcpu before * it starts execution. The vcpu will remain suspended * until a debugger connects. */ CPU_SET(0, &vcpus_suspended); stopped_vcpu = 0; } flags = fcntl(s, F_GETFL); if (fcntl(s, F_SETFL, flags | O_NONBLOCK) == -1) err(1, "Failed to mark gdb socket non-blocking"); #ifndef WITHOUT_CAPSICUM limit_gdb_socket(s); #endif mevent_add(s, EVF_READ, new_connection, NULL); gdb_active = true; + freeaddrinfo(gdbaddr); + free(sport); } diff --git a/usr.sbin/bhyve/gdb.h b/usr.sbin/bhyve/gdb.h index 93396c1c6705..c5fa522c63e0 100644 --- a/usr.sbin/bhyve/gdb.h +++ b/usr.sbin/bhyve/gdb.h @@ -1,39 +1,39 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2017 John H. Baldwin * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef __GDB_H__ #define __GDB_H__ void gdb_cpu_add(int vcpu); void gdb_cpu_breakpoint(int vcpu, struct vm_exit *vmexit); void gdb_cpu_mtrap(int vcpu); void gdb_cpu_suspend(int vcpu); -void init_gdb(struct vmctx *ctx, int sport, bool wait); +void init_gdb(struct vmctx *ctx); #endif /* !__GDB_H__ */