cachepc-qemu

Fork of AMDESE/qemu with changes for cachepc side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-qemu
Log | Files | Refs | Submodules | LICENSE | sfeed.txt

multiseat.txt (5147B)


      1
      2multiseat howto (with some multihead coverage)
      3==============================================
      4
      5host devices
      6------------
      7
      8First you must compile qemu with a user interface supporting
      9multihead/multiseat and input event routing.  Right now this
     10list includes sdl2, gtk (both 2+3) and vnc:
     11
     12  ./configure --enable-sdl
     13
     14or
     15
     16  ./configure --enable-gtk
     17
     18
     19Next put together the qemu command line (sdk/gtk):
     20
     21qemu	-accel kvm -usb $memory $disk $whatever \
     22	-display [ sdl | gtk ] \
     23	-vga std \
     24	-device usb-tablet
     25
     26That is it for the first seat, which will use the standard vga, the
     27standard ps/2 keyboard (implicitly there) and the usb-tablet.  Now the
     28additional switches for the second seat:
     29
     30	-device pci-bridge,addr=12.0,chassis_nr=2,id=head.2 \
     31	-device secondary-vga,bus=head.2,addr=02.0,id=video.2 \
     32	-device nec-usb-xhci,bus=head.2,addr=0f.0,id=usb.2 \
     33	-device usb-kbd,bus=usb.2.0,port=1,display=video.2 \
     34	-device usb-tablet,bus=usb.2.0,port=2,display=video.2
     35
     36This places a pci bridge in slot 12, connects a display adapter and
     37xhci (usb) controller to the bridge.  Then it adds a usb keyboard and
     38usb mouse, both connected to the xhci and linked to the display.
     39
     40The "display=video2" sets up the input routing.  Any input coming from
     41the window which belongs to the video.2 display adapter will be routed
     42to these input devices.
     43
     44Starting with qemu 2.4 and linux kernel 4.1 you can also use virtio
     45for the input devices, using this ...
     46
     47	-device pci-bridge,addr=12.0,chassis_nr=2,id=head.2 \
     48	-device secondary-vga,bus=head.2,addr=02.0,id=video.2 \
     49	-device virtio-keyboard-pci,bus=head.2,addr=03.0,display=video.2 \
     50	-device virtio-tablet-pci,bus=head.2,addr=03.0,display=video.2
     51
     52... instead of xhci and usb hid devices.
     53
     54host ui
     55-------
     56
     57The sdl2 ui will start up with two windows, one for each display
     58device.  The gtk ui will start with a single window and each display
     59in a separate tab.  You can either simply switch tabs to switch heads,
     60or use the "View / Detach tab" menu item to move one of the displays
     61to its own window so you can see both display devices side-by-side.
     62
     63For vnc some additional configuration on the command line is needed.
     64We'll create two vnc server instances, and bind the second one to the
     65second seat, similar to input devices:
     66
     67	-display vnc=:1,id=primary \
     68	-display vnc=:2,id=secondary,display=video.2
     69
     70Connecting to vnc display :1 gives you access to the first seat, and
     71likewise connecting to vnc display :2 shows the second seat.
     72
     73Note on spice: Spice handles multihead just fine.  But it can't do
     74multiseat.  For tablet events the event source is sent to the spice
     75agent.  But qemu can't figure it, so it can't do input routing.
     76Fixing this needs a new or extended input interface between
     77libspice-server and qemu.  For keyboard events it is even worse:  The
     78event source isn't included in the spice protocol, so the wire
     79protocol must be extended to support this.
     80
     81
     82guest side
     83----------
     84
     85You need a pretty recent linux guest.  systemd with loginctl.  kernel
     863.14+ with CONFIG_DRM_BOCHS enabled.  Fedora 20 will do.  Must be
     87fully updated for the new kernel though, i.e. the live iso doesn't cut
     88it.
     89
     90Now we'll have to configure the guest.  Boot and login.  "lspci -vt"
     91should list the pci bridge with the display adapter and usb controller:
     92
     93    [root@fedora ~]# lspci -vt
     94    -[0000:00]-+-00.0  Intel Corporation 440FX - 82441FX PMC [Natoma]
     95               [ ... ]
     96               \-12.0-[01]--+-02.0  Device 1234:1111
     97                            \-0f.0  NEC Corporation USB 3.0 Host Controller
     98
     99Good.  Now lets tell the system that the pci bridge and all devices
    100below it belong to a separate seat by dropping a file into
    101/etc/udev/rules.d:
    102
    103    [root@fedora ~]# cat /etc/udev/rules.d/70-qemu-autoseat.rules
    104    SUBSYSTEMS=="pci", DEVPATH=="*/0000:00:12.0", TAG+="seat", ENV{ID_AUTOSEAT}="1"
    105
    106Reboot.  System should come up with two seats.  With loginctl you can
    107check the configuration:
    108
    109    [root@fedora ~]# loginctl list-seats
    110    SEAT
    111    seat0
    112    seat-pci-pci-0000_00_12_0
    113
    114    2 seats listed.
    115
    116You can use "loginctl seat-status seat-pci-pci-0000_00_12_0" to list
    117the devices attached to the seat.
    118
    119Background info is here:
    120  http://www.freedesktop.org/wiki/Software/systemd/multiseat/
    121
    122
    123guest side with pci-bridge-seat
    124-------------------------------
    125
    126Qemu version 2.4 and newer has a new pci-bridge-seat device which
    127can be used instead of pci-bridge.  Just swap the device name in the
    128qemu command line above.  The only difference between the two devices
    129is the pci id.  We can match the pci id instead of the device path
    130with a nice generic rule now, which simplifies the guest
    131configuration:
    132
    133    [root@fedora ~]# cat /etc/udev/rules.d/70-qemu-pci-bridge-seat.rules
    134    SUBSYSTEM=="pci", ATTR{vendor}=="0x1b36", ATTR{device}=="0x000a", \
    135            TAG+="seat", ENV{ID_AUTOSEAT}="1"
    136
    137Patch with this rule has been submitted to upstream udev/systemd, was
    138accepted and should be included in the next systemd release (222).
    139So, if your guest has this or a newer version, multiseat will work just
    140fine without any manual guest configuration.
    141
    142Enjoy!
    143
    144--
    145Gerd Hoffmann <kraxel@redhat.com>