Index of /OpenVMS/QEMU_TCG/

NameLast ModifiedSizeType
../ -  Directory
01-lapic-timer-fix.patch2026-Nov-15 16:40:222.0Ktext/x-diff; charset=utf-8
02-max-insns-per-tb.patch2026-Nov-15 16:40:333.2Ktext/x-diff; charset=utf-8
README.txt2026-Nov-15 23:03:3110.9Ktext/plain; charset=utf-8
benchmark.c2026-Nov-15 22:49:295.8Ktext/x-csrc; charset=utf-8
disable-services.com2026-Nov-15 22:10:512.0Kapplication/x-msdos-program
 _____                  _   ____  ___ _____    _____ ________  ____   _ 
|  _  |                | | | |  \/  |/  ___|  |  _  |  ___|  \/  | | | |
| | | |_ __   ___ _ __ | | | | .  . |\ `--.   | | | | |__ | .  . | | | |
| | | | '_ \ / _ \ '_ \| | | | |\/| | `--. \  | | | |  __|| |\/| | | | |
\ \_/ / |_) |  __/ | | \ \_/ / |  | |/\__/ /  \ \/' / |___| |  | | |_| |
 \___/| .__/ \___|_| |_|\___/\_|  |_/\____/    \_/\_\____/\_|  |_/\___/ 
      | |                                                               
      |_|                                                               

        _________         .    .
       (..       \_    ,  |\  /|
        \       0  \  /|  \ \/ /
         \______    \/ |   \  /
            vvvv\    \ |   /  |
            \^^^^  ==   \_/   |
             `\_   ===    \.  |
             / /\_   \ /      |
             |/   \_  \|      /
                    \________/



Running OpenVMS 9.2-3 (x86_64) on macOS arm64 with QEMU TCG
=============================================================

Complete guide: from zero to a running OpenVMS system on Apple Silicon.

Patches included in this directory fix two issues that prevent VMS
from running correctly under QEMU's TCG (software emulation) on arm64:

  01-lapic-timer-fix.patch
    Fixes LAPIC timer handling for VMS's mask-set-unmask pattern.
    Without this, VMS hangs during boot (timer interrupt never fires).

  02-max-insns-per-tb.patch
    Adds a "max-insns-per-tb" TCG property that limits translation
    block size via a soft translator clamp (keeps TB chaining active).
    Without this, VMS crashes with %RMS-F-BUG during any heavy file
    I/O (compilers, shutdown, TCP/IP startup). The root cause is a
    TCG codegen issue on arm64 with multi-instruction translation
    blocks. Setting max-insns-per-tb=1 avoids the bug while keeping
    TB chaining for performance.

Tested against QEMU commit e9fa1f6 (staging branch, Nov 2026).
Patches should apply cleanly to any QEMU 11.x release.


STEP 1: BUILD SDL2
-------------------

QEMU needs SDL2 for the graphical display window.

  curl -LO https://github.com/libsdl-org/SDL/releases/download/release-2.30.10/SDL2-2.30.10.tar.gz
  tar xf SDL2-2.30.10.tar.gz
  cd SDL2-2.30.10
  ./configure --prefix=/usr/local
  make -j$(sysctl -n hw.ncpu)
  sudo make install
  cd ..


STEP 2: BUILD SOCAT (optional, for QEMU monitor access)
---------------------------------------------------------

  curl -LO http://www.dest-unreach.org/socat/download/socat-1.8.0.2.tar.gz
  tar xf socat-1.8.0.2.tar.gz
  cd socat-1.8.0.2
  ./configure --prefix=/usr/local
  make -j$(sysctl -n hw.ncpu)
  sudo make install
  cd ..


STEP 3: CLONE AND PATCH QEMU
------------------------------

  git clone https://gitlab.com/qemu-project/qemu.git
  cd qemu
  git submodule update --init --recursive

  git apply /path/to/QEMU-PATCH/01-lapic-timer-fix.patch
  git apply /path/to/QEMU-PATCH/02-max-insns-per-tb.patch


STEP 4: CONFIGURE AND BUILD QEMU
----------------------------------

  mkdir -p build && cd build
  ../configure \
    --target-list=x86_64-softmmu \
    --enable-tcg \
    --enable-sdl \
    --disable-werror \
    --disable-docs \
    --disable-gtk \
    --disable-cocoa \
    --disable-vnc \
    --disable-spice-protocol \
    --disable-usb-redir \
    --disable-libusb \
    --disable-brlapi \
    --prefix=/usr/local
  make -j$(sysctl -n hw.ncpu)

The build takes about 2-3 minutes on Apple Silicon.


STEP 5: SET UP THE WORKING DIRECTORY
--------------------------------------

Create a directory for the VM and copy the binaries:

  mkdir -p ~/openvms && cd ~/openvms

  cp /path/to/qemu/build/qemu-system-x86_64 .
  cp /path/to/qemu/build/qemu-img .

macOS requires clearing extended attributes and code-signing:

  xattr -cr qemu-system-x86_64 qemu-img
  codesign -f -s - qemu-system-x86_64
  codesign -f -s - qemu-img

Without this step, macOS kills the binary immediately (exit code 137).


STEP 6: EXTRACT AND CONVERT THE OPENVMS IMAGE
-----------------------------------------------

The VSI community image is distributed as a RAR archive containing
VMDK files (VMware format). Download it from VSI's community program.
The archive is named something like community_2026.rar (~1.3 GB).

Extract the VMDK files:

  unrar x community_2026.rar

This creates a directory (e.g. community_2026/) with VMDK files:

  community_2026/
    X86_V923-comm-2026.vmdk        (descriptor)
    X86_V923-comm-2026-s001.vmdk   (data extent 1)
    X86_V923-comm-2026-s002.vmdk   (data extent 2)
    ...

Convert the VMDK to qcow2 (point at the descriptor, not the extents):

  ./qemu-img convert -f vmdk -O qcow2 \
    community_2026/X86_V923-comm-2026.vmdk openvms-vanilla.qcow2

This produces a ~4 GB qcow2 file. The conversion takes about 1 minute.

Create an overlay so the original image stays untouched:

  ./qemu-img create -f qcow2 \
    -b openvms-vanilla.qcow2 -F qcow2 openvms-run.qcow2

Create a writable UEFI variable store (required for snapshots):

  ./qemu-img convert -f raw -O qcow2 \
    /usr/local/share/qemu/edk2-i386-vars.fd OVMF_VARS.qcow2

You can now delete the RAR and extracted VMDK directory to save space.


STEP 7: CREATE THE BOOT SCRIPT
--------------------------------

Create boot-vms.sh:

  #!/bin/bash
  BASEDIR="$(cd "$(dirname "$0")" && pwd)"
  QEMU="$BASEDIR/qemu-system-x86_64"
  FIRMWARE=/usr/local/share/qemu
  export DYLD_LIBRARY_PATH=/usr/local/lib

  exec "$QEMU" \
    -machine q35,hpet=off \
    -accel tcg,thread=single,max-insns-per-tb=1 \
    -cpu max \
    -m 4096 \
    -smp 1 \
    -icount shift=auto,align=off \
    -rtc base=utc,clock=vm \
    -nodefaults \
    -device VGA,vgamem_mb=8 \
    -drive if=pflash,format=raw,readonly=on,file="$FIRMWARE/edk2-x86_64-code.fd" \
    -drive if=pflash,format=qcow2,file="$BASEDIR/OVMF_VARS.qcow2" \
    -drive file="$BASEDIR/openvms-run.qcow2",format=qcow2,if=none,id=disk0 \
    -device ahci,id=ahci0 \
    -device ide-hd,drive=disk0,bus=ahci0.0 \
    -monitor unix:"$BASEDIR/monitor.sock",server,nowait \
    -display sdl \
    -no-reboot \
    -netdev user,id=net0,net=10.0.2.0/24,dhcpstart=10.0.2.15,hostfwd=tcp::2222-:22,hostfwd=tcp::2223-:23,hostfwd=tcp::2221-:21 \
    -device e1000e,netdev=net0

Make it executable:

  chmod +x boot-vms.sh


STEP 8: FIRST BOOT
--------------------

  ./boot-vms.sh

Boot sequence:

  1. UEFI loads (~1 minute). You see the boot manager.
  2. type OPA0 at the UEFI shell.
     type BOOT at the UEFI shell.  
  3. VMS boots through SYSINIT. Watch the SDL window.
  4. Login prompt appears after ~5-10 minutes. (be patient, default VSI image is heavy) 
  5. Default login:  SYSTEM / your-password

First boot is the slowest because VMS initializes databases and
starts all services. Subsequent boots from snapshot are instant.
We will also cleanup some unecessary daemons to speedup the boot time.



STEP 9: POST-INSTALL TUNING
------------------------------

After first login, disable unnecessary services to speed up boot
and reduce resource usage. Upload and run disable-services.com
(included alongside this README), or do it manually:

  $ @SYS$STARTUP:SAMBA$SHUTDOWN.COM
  $ RENAME SYS$STARTUP:APACHE$STARTUP.COM SYS$STARTUP:APACHE$STARTUP.COM_DISABLED
  $ RENAME SYS$STARTUP:SAMBA$STARTUP.COM SYS$STARTUP:SAMBA$STARTUP.COM_DISABLED
  $ RENAME SYS$COMMON:[SYSMGR]TOMCAT$DEFINE_LOGICALS.COM -
           SYS$COMMON:[SYSMGR]TOMCAT$DEFINE_LOGICALS.COM_DISABLED

If you want the SYSTEM password to never expire:

  $ MC AUTHORIZE MODIFY SYSTEM /PWDLIFETIME=NONE

STEP 10: SAVE A SNAPSHOT
--------------------------

Once the system is configured the way you want it, save a snapshot
so you never have to cold-boot again:

  echo "savevm vms-clean" | socat - UNIX-CONNECT:monitor.sock

Restore instantly next time by adding -loadvm to the boot command:

  exec "$QEMU" ... -loadvm vms-clean


STEP 11: SSH ACCESS
---------------------

The boot script forwards ports via QEMU's NAT networking:

  Host port 2222 -> VMS port 22 (SSH)
  Host port 2223 -> VMS port 23 (Telnet)
  Host port 2221 -> VMS port 21 (FTP)

Connect from your Mac:

  ssh -p 2222 system@127.0.0.1

VMS SSH is slow to handshake. Use a long timeout:

  ssh -p 2222 -o ConnectTimeout=120 -o ServerAliveInterval=30 system@127.0.0.1

For scripted access (sshpass must be installed):

  sshpass -p 'your-password' ssh -p 2222 \
    -o StrictHostKeyChecking=no \
    -o UserKnownHostsFile=/dev/null \
    -o ConnectTimeout=120 \
    system@127.0.0.1 'SHOW SYSTEM'


PERFORMANCE (MacBook Air M2, arm64)
--------------------------------------

Host: Apple MacBook Air M2 (8-core, 8 GB RAM)
Guest: OpenVMS V9.2-3 x86_64, single vCPU, 4 GB RAM
QEMU: TCG with max-insns-per-tb=1, icount shift=auto

Boot and general:

  Cold boot to login prompt     2:40        (services disabled)
  Snapshot restore to login     instant     (< 1 second)
  CC compile + LINK             ~5 seconds
  System shutdown               ~30 seconds
  Dhrystone 2.1 VUPs            95.1 VUPs   (167,084 Dhrystones/sec)

Benchmark results (benchmark.c):

  Test                    Time         Notes
  ---------------------   ----------   ---------------------------
  Integer arithmetic        0.79 sec   10M iters, add/mul/div/mod/bit
  Floating point           40.18 sec   10M iters, sin/cos/sqrt/log/exp
  Memory bandwidth          0.52 sec   160 MB sequential r/w (4 MB buf)
  String operations       253.36 sec   1M iters, memcpy/memset/strcmp 8KB
  Sieve of Eratosthenes     1.10 sec   10 passes, primes < 2M
  File I/O                  1.07 sec   write+read 4 MB, 4 KB blocks
  ---------------------   ----------
  TOTAL                   297.04 sec

  Integer, memory, sieve, and file I/O are fast. Floating point and
  string operations are slow because TCG must emulate x86 SSE and
  REP-prefixed string instructions individually on arm64.

The system is fully usable for development: compiling, linking,
running programs, and network access all work at interactive speed.


KEY FLAGS EXPLAINED
--------------------

  -machine q35,hpet=off
    Use Q35 chipset. Disable HPET (VMS doesn't use it, avoids issues).

  -accel tcg,thread=single,max-insns-per-tb=1
    Software emulation. Single thread (required for icount). Limit
    each translation block to 1 guest instruction with TB chaining
    enabled. This is the fix that makes VMS stable.

  -cpu max
    Expose all CPU features the host can emulate. VMS requires at
    least XSAVE and SSE4.1 (qemu64 model is not enough).

  -icount shift=auto,align=off
    Tie virtual time to instruction count. Without this, UEFI timed
    delays take ~1000x longer (hours instead of seconds). The
    shift=auto lets QEMU calibrate automatically. align=off avoids
    unnecessary synchronization overhead.

  -device VGA,vgamem_mb=8
    VGA display. VMS sends all output to VGA after early boot.

  -device e1000e,netdev=net0
    Intel e1000e NIC. VMS recognizes this as interface IE0.
    Connected to NAT with DHCP + port forwarding.

  -netdev user,...,hostfwd=tcp::2222-:22,...
    QEMU user-mode (NAT) networking. VMS gets IP via DHCP from
    QEMU (10.0.2.15). Port forwards provide host access to SSH,
    Telnet, and FTP.
Webserver 1.0