VirtualBox 4.3.14 is a maintenance release. The following items were fixed and/or added since VirtualBox 4.3.12: VMM: more fixes for MSR emulation on certain hardware (bugs #12784, #12949, #13034) VMM: improve MSI handling under rare circumstances (only relevant for the ICH9 chipset) VMM: fixed #UD exception for 64-bit guests with the EFER.SCE bit and the SYSCALL instruction (VT-x only; 4.3 regression; bug #13008) VMM: fixed timekeeping after resuming SMP guests VMM: properly wake up a halted VCPU on NMI/SMI GUI: fixed a potential crash GUI: fixed stuck AltGr key on Windows hosts (bug #2537) GUI: fixed a potential error during the version check GUI: shortcut change should not require Enter/Return (or other trigger) to confirm (bugs #12828, #12847, #12937, #13087) GUI: fixed update check which was broken due to changing the location of the root certificates (bug #13096) VBoxManage: fixed typo in showvminfo --machinereadable (bug #13176) NAT: fixed inbound half-close (bug #13116) NAT: fixed slow upload speed under certain conditions (bug #10034) NAT Network: fixed potential loss of inbound TCP data NAT Network: fixed potential infinite stalls of TCP connections over IPv6 NAT Network: fixed resets of TCP connections on Windows hosts NAT Network: fixed inbound half-close on Mac OS X hosts NAT Network: fixed socket leak on Solaris hosts NAT Network: fixed ping of mapped host loopback on Mac OS X and Solaris hosts, fixed proxying of IMCP errors on Mac OS X Host-Only Network: fixed SNMP ifConnectorPresent value on Windows (bug #13143) Storage: fixed a possible crash with CD/DVD passthrough under certain circumstances Storage: fixed a crash when trying to open an inaccessible QED or QCOW image (bug #12613) Storage: fixed data corruption or read errors under rare circumstances AHCI: fixed a crash under rare circumstances USB: performance fixes ICH9: properly reset MSI capability on reset Keyboard: active modifier keys during suspend were stuck after resuming the host 3D: fixed misbehavior with huge guests (i.e. guest more than 4GB guest memory 3D: several fixes API: properly detect the Windows 8.1 guest OS type (bug #13090) ExtPack: cleanup of dangling uninstallation directories Linux hosts / guests: compile fix for EL7 (bug #12638) Linux Additions: made 3D pass-through work with recent versions of Mesa in the guest (bug #12941) Linux Additions: Linux 3.16 fixes (bug #13123) Mac OS X hosts: when scanning for host CD/DVD devices also consider BlueRay devices OS/2 Additions: fixed gengradd.dll library name (bug #12785) Solaris Additions: fixed permissions of files and directories located on shared folders Windows host installer: fixed the need for rebooting Windows after installation or upgrade, extended logging for NetFlt/NetAdp (un)installation

- Debugger TR command didn't handle linked resident lists correctly (inverted mask..). - TO debugger command also lists FileSysStartupMsg and DosEnvec of mounted devices. - Slow down 68EC020 RTE instruction, fast clock rate multiplier modes could have executed MOVE to INTREQ + RTE combination too quickly (before Paula had cleared the interrupt) causing interrupt to re-trigger immediately after RTE. - Added SCSI controller selection to GUI and config file. ("SCSI (Auto)" does same as old plain SCSI selection = use first enabled SCSI controller). - More than 1 SCSI controller can be now active simultaneously. (For example A4000T + A4091) - Added support for two A590/A2091 and two A4091 boards, both (or even all 4) can be active at the same time. - Secondary A590/A2091 and A4091 share primary board's boot ROM by default. a2091_2_rom_file and a4091_2_rom_file config file entry can be used to select other ROM version. - 280b7 sprite update missed AGA bordersprite condition, causing clipped sprites in some situations (Silicon Graphics 2 / WFMH) - Rewritten directory filesystem Examine Next logic, now returned fib_DiskKey always have file/directory entry specific unique value. Some programs check fib_DiskKey and (incorrectly) assume that it is unique value and may also assume file must be same if both file's have same fib_DiskKey. Now also C:List KEY output looks "correct". This needs testing because original (Tripos?) directory enumeration packet design is not that good.. - AGA border sprite bit set and if same scanline had at least 1 sprite that was outside of playfield and at least 1 bitplane was active and BPLCON4 palette XOR value was non-zero: wrong background color was visible in some situations. (Nexus 7 / Andromeda "shade cluster" part) - Some BEAMCON0 programmed mode emulation updates. VBSTRT/VBSTOP register values only affect screen if VARVBEN bit is set. - Use also value stored in VBSTRT (vertical blank start) register when setting up programmed mode vertical size, some weird modes can have (much) smaller VBSTRT than VSSTRT (vertical sync start). (Jtxrules by Illusion) - HBSTRT/HBSTOP register values are only used if BPLCON3 EXTBLKEN bit is set. (Demos Contactro and Jtxrules by Illusion) - BEAMCON0 HARDDIS bit also disables DDFSTRT hardware limit. (Weird stuff, VARBEAMEN disables it too, even superhires mode disables it..) - Fixed "CPU trace blahblah" error after 68020+ state file was loaded and loaded program used bitfield instructions later. - Split ROM images didn't load (again..) correctly if relative path mode was active. - Added support for missing Z2 RAM board sizes (64k/128k/256k/512k). Note that small boards (256k and less) can be mapped at IO autoconfig base (0xe90000+) which makes predicting final address difficult: JIT (if enabled) can't use direct mode to access this RAM board. - Added support for second Z2 fast RAM board. Size must be smaller or same as first board. (Makes JIT board address calculation simpler) - Above two changes mean it is now possible to have total 6M of Z2 RAM in 24 bit addressing config with 2M gfx board config (4M + 2M and 2M gfx) or less useful 8M fast + extra 256k of fast config... - Z2 gfx board JIT direct calculation didn't correctly align the board base address if Z2 RAM size was less than Z2 gfx board size. - Implemented no-hack Z3 memory mapping, lets OS handle Z3 autoconfig without patching base addressess from >=0x40000000 to >=0x10000000. This mode is not JIT Direct compatible. Option in Memory panel. Do not enable unless you know what are you doing and you really need 100% matching real hardware Z3 address mapping. - Use no-hack Z3 allocation automatically if all configured boards fit in official Z3 space, fully JIT Direct compatible. Requires 64-bit Windows and available space will be about 300M to 500M. RAM panel text string shows free space. 32-bit Chip RAM does not count, it is not Z3 board and it is always located at 0x10000000. - Added 384M and 768M 32-bit Chip RAM size options. (768M fits perfectly at 0x10000000-0x3fffffff) - Added Z3 autoconfig advanced chipset option. If ticked (or A3000/A3000T/A4000/A4000T chipset extra selected), Z33 boards will now appear at Z3 autoconfig space (0xFF000000), previously Z2 space was always used. Z2 space is supported by AOS but most real, but not all, Z3 boards use Z3 space. Optional because not all KS ROMs (for example A1200) support Z3 autoconfig space but do support Z3 boards. - Try to keep number of cylinders under 65536 when generating default geometry even if partition hardfile is very large (>100G) for better old AOS compatibility. (Helps my big partition PFS3 testing..) - Hardware (Cirrus Logic) RTG board emulation incorrectly required UAE boot ROM. - 68030 MMU update from Previous, fixes some supervisor/write protect edge cases. - Added "history" menu to filesystem, hardfile and tape path selection text boxes. - Added portable mode (.ini) on/off state to Paths panel. Read-only, enabling it on the fly would cause all kinds of side-effects. Unsupported warning removed

* Ported Stella to SDL2, which brings many new features. Among the largest improvements is native hardware acceleration support for Windows (Direct3D) and Linux/OSX (OpenGL). It is also now possible to port Stella to iOS and Android devices using OpenGLES. Hardware acceleration is now required, which means up-to-date drivers are needed. Software rendering is still present, but is somewhat unoptimized and unsupported going forward. * Fullscreen video modes now use the desktop resolution. Switching to fullscreen and back to windowed mode no longer rearranges icons on your desktop. * TIA TV effects are now available in all video modes, since hardware acceleration is a requirement. * Added a much more detailed view of cart extended RAM to a new debugger tab. Special thanks to SpiceWare for this implementation. * Added preliminary support for 'DASH' bankswitching scheme by A. Davie. * The AtariVox and SaveKey controllers now have the ability in the debugger to completely erase the virtual EEPROM data. * Added 'savesnap' debugger prompt command, and also associated context menu item to the debugger TIA output area. This saves the current TIA image to a PNG file. * Added 'hidecursor' commandline option, which allows to completely disable showing the mouse cursor (useful on systems that don't have a mouse). * Removed 'uipalette' option, as the original palette is no longer supported. * Updated included PNG library to latest stable version.

2.8.0 bugs fixed: - JIT on/off on the fly switching was unreliable. - A500 reading write-only/non-existing registers method introduced in 2.8.0 was incorrectly also used in AGA modes. Updates: - Fast CPU mode audio DMA wait hack compatibility improved (workaround for programs that use CPU delay loops in audio code) - Some chipset emulation updates, Critical Mass / Parallax finally works. - AROS ROM updated. New features: - Directory filesystem MorphOS compatible >4G file size DOS packet support. - Paths panel relative path mode now supports relative paths that point outside of winuae root directory. Bug fixes: - Filter modes in DirectDraw crashed in some situations. - Disk insert using GUI after emulation was started inserted disk in write protected state. - Quickstart disk eject button was unreliable. - Video recording with "Capture before filtering" unticked used wrong video size. - "Toggle between mouse grabbed and un-grabbed" input event fixed.