- "Kevin P. Lawton" <kevin@mandrakesoft.com>: Tue Sep 11 13:53:39 CEST 2001
    Fixed compilation problem with curses.cc.  Still some kind of problem
      (perhaps with screen updates), so only the VGA BIOS init messages
      are displaying so far.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Mon Sep 10 20:44:42 CEST 2001
    Cleaned up duplicate include directives ("-I") in Makefiles.  They
      were a result of mods to allow for compilation in a separate directory,
      but the diffs left duplicates of some include dirs.
    Fixed the 'bitmaps' directory problem with CVS.  You shouldn't need
      a symlink anymore.  Make sure to "cvs update -APd" before trying.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Wed Aug 15 18:32:28 CEST 2001
    Many more moularization changes.  Now, all the IO devices which are
      compiled by default, use a plugin interface and are loaded as
      individual plugins.  I haven't cleaned up some of the devices which
      are not compiled in by default.
    Only tested and fixed the conf/ files for freedos and win95.
      If you try with your own conf files, you must modify them
      like 'freedos' or 'win95'.  This interface will be in flux for
      a while, as I'm making several passes through the code to
      clean it up, while keeping things in working condition.
    Currently, options are set through the old bx_options structure,
      but next I plan to fix this, passing options directly to each
      plugin via a more modular/intelligent get/set option interface.
    Here is a list of the devices which are now plugins (an
      excerpt from a conf file):

      plugin = ./plugins/bochs/iodev/cmos.so
      plugin = ./plugins/bochs/iodev/dma.so
      plugin = ./plugins/bochs/iodev/system.so
      plugin = ./plugins/bochs/iodev/vga.so
      plugin = ./plugins/bochs/iodev/serial.so
      plugin = ./plugins/bochs/iodev/parallel.so
      plugin = ./plugins/bochs/iodev/keyboard.so
      plugin = ./plugins/bochs/iodev/floppy.so
      plugin = ./plugins/bochs/iodev/harddrv.so

    Also as I make passes through the code, I'll clean up the plugin
      interface too.  It's a little hacky in parts.  But if I redesign
      it all in one shot, I'm sure I'm going to introduce some big errors
      which will make testing/debugging really difficult.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Fri Aug 10 18:53:17 CEST 2001
    Yet another wad of modularization changes to IO devices and glue logic.
      CMOS and DMA interaction now go through a plugin interface.  Getting
      there...  Most of the inter-device interaction is now modularized to
      use plugin interfaces.  iodev/devices.cc is trimmed way down.  Soon
      I can work on compiling/calling the devices as individual plugins.
      And some .plex86rc directives to go with this.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Wed Aug  8 14:03:17 CEST 2001
    More cleanup of IO devices glue logic.  Removed pc_devices.{cc,h}
      completely.  Moved logic up to the plugin interface, or down to
      the DMA controller emulation.
    I'll need to make a few more logical passes through the code to effect
      the splitup to IO device plugins.
    Ultimately, will put in a requirments flag/mechanism so, for example,
      you can only load a floppy plugin if you have already loaded a 
      DMA plugin and a PIC plugin.  (or they are intrinsically offered
      by the VM core)

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Tue Aug  7 16:44:54 CEST 2001
    Cleanup of plugin glue code in user/plugins/bochs/.  I'm working toward
      splitting the IO devices into individual plugins.  This will give
      us more flexibility for sharing devices with bochs, and distributing
      updated device models without updating the entire core.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Mon Aug  6 11:37:57 CEST 2001
    Integrated small diff for Linux header space conflict of timer_t
      from Arjan van de Ven <arjanv@redhat.com>

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Mon Jun 11 00:19:20 EDT 2001
    Oops, there was an include problem - when the default linux source
      path was used, an extra /include component was appended.  The path
      should point to the top linux directory.  The Makefile adds
      in the /include component.  When I added in that `uname -r` idea
      I didn't notice it broke things until I passed '-v' to GCC.  Fixed.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sun Jun 10 12:34:56 EDT 2001
    ACK!  I had a goto jumping around the native CS:EIP fetch accelerations,
      in emulation/emulation.c.  Forgot and left that in there.  Man,
      do things work faster now!!!  Check this out:

               Yesterday    Today
      Pragma   5:50         2:54
      Win95    3:23         1:59

      And we still have many performance tweaks to do, running app code
      natively, better video (still getting wacked with 1E+06 redirects for
      VGA 640x480x16color latches), drivers, better page-on-demand logic,
      etc.  Pragma is getting hit with a lot of redirects to the CMOS
      in user space.  That can be fixed also, by a move to VM space, like
      I did with the PICs/PITs already.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sat Jun  9 23:04:50 EDT 2001
    Added code to handle OUTS in chunks.  (Already had INS support). This
      helps with Windows a little - less redirects to user space.
    Designated FS descriptor for virtualized guest CS to avoid reloading
      a descriptor many times for each instruction fetch.
    Optimized access to PDE/PTE fields in function to map a guest
      page into the monitor.
    Moved the port 0x80 handler to the VM.  Pragma was nailing it with
      millions of accesses.  Eliminated that waste.
    Win95 boots in 3:23, Pragma in 5:50.  Win95 is getting hit with
      about a million memory mapped IO redirects during a boot/shutdown.
      Perhaps I'll move the latches into the VM space.  Pragma is hitting
      the CMOS IO ports a few million times.  Yuk.  Gotta fix that too.
    Added some instrumentation macros, turn on by editing end of config.h
      after you run ./configure.  Will add more.  Some are not compatible
      with multiple VMs when enabled (because I was lazy and used global
      variables).
    My guess is that the current 1-page-at-a-time demand paging scheme
      is not so good for performance either.  Should really map in
      neighboring guest pages.  Could handle virtualized page tables
      much better.  I have a feeling if this is done right, the above
      execution times will go down signficantly.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sat Jun  9 03:13:32 EDT 2001
    More optimizations, performance is definitely better.
      Added native seg:offset accesses for emulation of read-modify-write
        instructions.  Previously I only implemented for strict read or
        write instructions.
      Modified assembly handlers for native accesses, to take advantage
        of GCC doing caller-saves on certain registers.  So no need to
        push/pop state info when using them.  Also biased the C part
        of this stuff for 32bit accesses in the if-then-else constructs.
      Restructured the keyboard emulation a little, to eliminate the
        constant delivery of time events.  Only one event per request
        is now delivered.  That eliminated a lot of wasteful round-trips
        to user space!

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Fri Jun  8 15:39:42 EDT 2001
    Added idea from Arjan van de Ven <arjanv@redhat.com>, to default
      Linux kernel include path to /lib/modules/`uname -r`/build/include.
    Added mods from Roland McGrath <roland@frob.com> to build plex86
      in a separate directory.
    Added mods from Eyal Lotem for physical memory access outside of
      bounds (for OS memory sizing).

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Thu Jun  7 11:26:47 EDT 2001
    Moved PIC/PIT device emulation into the VM monitor.  Definitely
      a performance win.  A lot less round trips to user space!
      Other components could be moved to the VM monitor space also.
      Some more instrumentation measurements; the number of requests
      from monitor to user space:

        Win95:

        874354  MemMapIOWriteRequest
        811118  TimeElapsed
        299860  IOOutRequest
        285533  MemMapIOReadRequest
        123386  IOInRequest
         10450  IOInBatchRequest
           105  PrintBuf

        Pragma Linux:
        
        3029470 IOOutRequest
        1448653 TimeElapsed
        1032759 IOInRequest
          27931 MemMapIOWriteRequest
             38 PrintBuf
              1 IOInBatchRequest

      Win95 is taking a lot of MemMapIO hits (VGA framebuffer).  That
      can be remedied with smarter video handling (emulation of a better
      card, special drivers, or moving part of the VGA emulation to the
      monitor also).  Both OSes are being hit with TimeElapsed request,
      which comes from the timer framework in the VM monitor.  This is
      due to the ::periodic function in the keyboard emulation which
      is receiving timer events every 100useconds, whether it needs it
      or not.  I'll squash that soon!  Anyways, it's a lot easier to
      move devices to the monitor, since the PIC/PIT are there now.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Mon Jun  4 22:58:26 EDT 2001
    Fixed the bug that prevented 1st execution of plex86 to work.  Linux
      appears to be demand allocating/mapping the pages of memory returned
      by a vmalloc.  No problem, I just touch each page with one read
      to force the lazy allocation to map the page in, so that we can
      get the physical page address from the page tables.  Just added a
      little code in host_alloc().  Don't know if other hostOSes will need
      this also.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Mon Jun  4 17:03:44 EDT 2001
    Changes to host-linux.c, thanks to suggestions from Jeff Garzik.
      Moved plex86 proc entry to /proc/driver, where it should be.
      Moved assignment of file_operations to declaration statement
        and used field labels (GCC specific).
      Elminated use of MOD_{DEC,INC}_USE_COUNT for kernel 2.4, and set
        file_operations::owner to THIS_MODULE instead.  This eliminates
        a race between the time dev->release is called and module
        unload is initiated on another CPU.
      Used {>>,<<} instead of {/,*} where possible for page size oriented
        calculations, in case the compiler doesn't optimize this for us.
    Manage a set of virtualized descriptors in the monitor GDT, one for
      each guest descriptor, and use those directly for native memory
      accesses in the emulation, rather than always loading a new
      GDT entry for each access.  Things can be optimized further.
    TODO next:
      Fix error on first invocation of plex86.
      Move some critical IO device emulation to VM monitor space for
        performance.  Mostly, the PIT/PIC and the hardware component
        of the VGA emulation.  JFYI, here's some instrumentation data
        I gathered from a Win95 boot in plex86.  I quote the number of
        VM to user space requests necessary.  The MemMapIOW is due to
        writes to the VGA memory (latched planar mode sucks).  TimeElapsed
        is mostly due to time delivery events to the PIT.  IOOut in
        large part due to PIC EOI and etc.  If I move PIT/PIC/VGA
        into the VM, these will disappear.  These redirects are expensive,
        so I expect to see performance increase nicely without them!
        The GUI component of the VGA emulation will stay in user space,
        of course.

          872068  MemMapIOWriteRequest
          800406  TimeElapsed
          324176  IOOutRequest
          283248  MemMapIOReadRequest
          153986  IOInRequest
           10449  IOInBatchRequest
            4029  IACRequest
              78  PrintBuf

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Tue May 29 01:22:49 EDT 2001
    The emulation code attempts to use native segment:offset accesses from
      special assembly routines.  A monitor fault handler catches when
      these don't work and fixes things so that code flows through to
      normal emulation.  That really helped emulation speed!
    Adjusted the cycles-per-instruction factor, since emulation is faster.
      That helped significantly too.
    I've got a bunch more accelerations to do.  Will add some of these
      before adding ability to run application code natively.
    Some IO accelerations will be necessary *soon*.  The faster the VM
      gets, the slower IO is relatively speaking.  Video/disk/floppy are
      prime candidates.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sat May 26 00:55:39 EDT 2001
    Run everything in pure emulation mode (in VM monitor space).
      I'll optimize the cpuEmulate() loop tomorrow.  Currently, it
      does the map_guest_laddr() call for every iteration.  Only
      need to do this when branching out of the current page, or paging
      change occurs.  Fortunately, I've done this trick before in bochs. :^)
    Rebuilt the timer delivery system.  I gutted out the framework that
      was in plugins/bochs/pc_system.cc and built timer delivery services
      into the VM monitor.  The video/GUI updates are still based on
      host time though.
    I tested this stuff with FreeDOS/Pragma Linux.
    I'm going to tune the emulation and some other things that are long
      overdue next.
    You may find some things run faster already, even with the inefficient
      cpu emulation loop.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Thu May 24 15:01:00 EDT 2001
    Created 'PATCHES' files at top-level.
    Added static branch translations from DT prototype.  Tested on
      FreeDOS/Pragma - both work.
    OK, most of the pieces are here now.  There's still some more
      instructions to deal with translations for, but it's more of the
      same stuff.  Plex86 is running the guest in 100% DT mode currently,
      (though some instructions are translated as INT3 for now)
      which is great for testing (abusing) the DT logic.  But not
      good for performance.  Only really active ("hot") code should be DT'd.
      Anyways, with enough DT working, on to the next steps:
         1) Make plex86 default to running everything in emulation mode.
            (should be quick to implement as it can already do this)
         2) Let protected mode user code run natively.  (Should
            be reasonable - I did some of the work previously)
         3) Release code for 1) & 2) to CVS.  This should be a reasonable
            step forward with performance.
         4) Gather instruction frequency use and other data from 1),
            and DT only active and well behaved code.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Tue May 15 12:33:05 EDT 2001
    Integrated small NetBSD diffs from Frank van der Linden.
    I moved a small amount of code out of the host-*.c files, into
      monitor-host.c, beginning the process of migrating as much generic
      code out of the kernel-specific modules as possible.
    Uploading all new code (including DT stuff) to CVS.
    BTW, I moved the names of the space specific files in kernel/, so
      they made more sense.  For example, 'mon-mode.c' became 'mode-mon.c'.
      This way I can see all the related files for mode switching grouped
      together with one 'ls' command.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Mon May 14 14:03:20 EDT 2001
    Integrated a lot of new DT architecture (from dt-testbed/proto4) into
      plex86.  Eliminated the previous Software Instruction Virtualization
      layer (SBE).  Parts that are now integrated/added:
        - Tcode (translated code) storage/allocation.
        - Guest instruction address to tcode address sparse tables.  These
          are on a per-page basis.
        - Reverse tcode address to guest instruction address sparse tables
          for exception handling.
        - Maintenance of two hash tables for high speed lookups.  One
          for guest offset (EIP) to tcode address, and one for linear
          page address to page meta info index.  These will be more important
          when I integrate the branch translations.
        - Invalidation of tcode based on writes to managed tcode pages.
        - Revalidation of tcode pages across guest OS context switches.
        - Arbitrary code expansion.  We're not locked into maintain the
          same page offsets anymore.
        - Updated the 'PERFORMANCE' file with more related ideas/todos.
    Parts not integrated yet.  The only thing that wasn't integrated is
      the branch translations.  To make debugging easier, for this phase,
      _all_ branches (even static intra-page branches) are translated as
      INT3 and emulated by the monitor.  This means performance will suck
      until I add them.  The code is prototyped already in dt-testbed/proto4,
      and now the hash table management logic is tested and works.  So this
      should be reasonable to add in short time.
    FreeDOS / Pragma Linux (kernel 2.0.33) / Win95 all work again.
      I used 'conf/freedos', 'conf/pragma', and 'conf/win95'.
    This release is for testing only.  To decrease noise on the developers
      list, test it only with host = {single processor Linux kernel}, and
      guest = {FreeDOS, Pragma Linux} and guest = {Win95} if you already
      have an installed disk image.
    Performance will be very slow for this phase.  This is OK and expected,
      as every guest branch instruction is monitored/emulated.  First,
      I need your help to test this code.  After a small amount of time
      to shake out bugs, I will integrate the branch translation logic
      from the prototype (phase#2), and things should get fun!

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sat Apr 28 15:30:00 EDT 2001
    Changed xml FAQ doc to real xml QandA type.
    More cleanup of proto4.
    Enough work on the prototype - time to integrate this puppy!

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Fri Apr 27 14:17:02 EDT 2001
    Added top-level 'FAQ' which points users to docs dir.  Better notes
      in docs/README on how to ftp or generate formatted docs.
    Entry in FAQ letting users know to wait for new OSes to run to
      reduce list noise/redundancy.
    Some more testing of proto4.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Thu Apr 26 18:00:12 EDT 2001
    More work on dt-testbed/proto4.  Implemented reverse address lookup
      table (tcode address to instruction address) capability.  This is
      needed for proper exception handling when the code is integrated
      into the real VM environment.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Tue Apr 24 15:04:33 EDT 2001
    More work on dt-testbed/proto4.  Implemented real sparse lookup table
      and storage mechanisms for tcode (translated code).  Each code page
      has it's own lookup table and a set of storage chunks, in which tcode
      and lookup table information is stored.  This plays nicely with the
      dynamic page invalidation/revalidation scheme and guest instruction
      address to tcode address hash table centricies.
    Next, to implement a reverse lookup capability; mapping tcode addresses
      to the corresponding guest instruction address, for handling
      guest exceptions.  This will use the same mechanisms as already
      implemented so it should be fairly straighforward.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Fri Apr 20 16:58:11 EDT 2001
    Created dt-testbed/proto4.  Added some framework in the header files
      for tcode storage and forward/reverse instruction addr <--> tcode addr
      lookup mechanisms.  This is the last major component of this
      testbed code to be coded, before integrating the new architecture
      into the main code.  Hopefully the last prototype revision.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Fri Apr 13 12:43:45 EDT 2001
    Wrote some more text in the Plex86 Users Guide.
    A little docs/ clean-up.
    Uploaded a formatted docs (docs/output/) tarball to
      ftp://plex86.org/docs/

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Wed Apr 11 23:59:02 EDT 2001
    Restrutured docs/xml.  Put Plex86 Users Guide (PUG) and the
      Plex86 Internals Guide (PIG) in separate directories.  Makefiles
      now use stock DocBook/XML tools from Mandrake 8.0 Beta 3.
    Added one more graphic to the PIG (DT code cache storage).
      More explanation text to come for this.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Tue Jan 23 19:31:58 EST 2001
    Rerun of performance tests for dynamic branch handling, better
      results now.
    More notes in dt-testbed/proto3/README.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: 
    Small mod to get NetBSD to boot from Krister Walfridsson in vga.cc.
    Integrated dt-testbed/proto3 from Ramon.  Ramon restructured some
      things so he can research more advanced DT techniques in parallel
      to the lightweight DT development.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Fri Jan 19 12:11:14 EST 2001
    More enhancements to dt-testbed/proto2, and more notes in the README.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sun Jan 14 21:36:48 EST 2001
    More enhancements to dt-testbed/proto2, and more notes in the README.
      Modeled computed branches.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Fri Jan 12 22:57:11 EST 2001
    More enhancements to dt-testbed/proto2, and more notes in the README.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Fri Jan 12 15:54:22 EST 2001
    More enhancements to dt-testbed/proto2, and more notes in the README.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Fri Jan 12 01:13:29 EST 2001
    Created new dir dt-testbed/proto2.  Check out README in that dir.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Wed Jan 10 17:24:18 EST 2001
    Moved previous files in dt-testbed to dt-testbed/proto0.
    Created new dir dt-testbed/proto1.  Check out README in that dir.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sat Jan  6 23:39:58 EST 2001
    Integrated updates to write-cache from Tom Vijlbrief.
    Reorganized docs/ a little.  Split the xml docs into 2 docbooks.
      There is now a PUG (Plex86 User's Guide) and a
      PIG (Plex86 Internals Guide).  I removed the tarball out of
      the CVS tree, and will upload pre-generated docs to the FTP
      server periodically.  Look for docs-yyyy_hhmm.tar.gz.
      The tarball now contains both HTML and PS formats.
      There's also some notes on what packages are needed to generate
      and develop your own docs.  (docs/xml/README)  This release should
      make it easier for people to help out.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Mon Jan  1 23:26:44 EST 2001
    Added another docs chapter (17) to the user's manual, relating to
      dynamic translation (DT) ideas for maintaining linear to translated
      code address mappings and some other stuff.  You can update from
      CVS and just untar the docs/ tarball if you want.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Mon Jan  1 11:04:34 EST 2001
    Integrated write-cache plugin from Tom Vijlbrief, and his related
      mods to the hard drive code, to handle > 32-bit file IO.
    Integrated small FreeBSD Makefile mods from Alexander Langer.
    Regenerated configure.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sat Dec 30 10:17:23 EST 2000
    Integrated FreeBSD host port by Alexander Langer <alex@big.endian.de>
      Note: this port does not work yet, but was uploaded so others
      can work on it.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Thu Dec 28 13:51:50 EST 2000
    Integrated LBA patches from Tom Vijlbrief, to rombios.c and harddrv.cc.
      I modified the patches slightly, and have not tested them much.
      Am uploading changes to let Tom test them as integrated.
      Tom was able to access up to 2GB partitions with his changes.
    Generated new BIOS.  Changed all conf files to use new BIOS.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Thu Dec 28 10:32:57 EST 2000
    Added dt-testbed dir.  This is just a small separate testbed program
      for developing thoughts about generated code for a quasi-dynamic
      translation scheme.  If you're interested in such developments,
      look at the README file in that dir.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Tue Dec 26 16:34:49 EST 2000
    Added framework for write-cache buffering (for "persistent disks").
      Created new plugin, in directory user/plugins/write-cache/.  It
      does nothing currently, other than pass disk accesses to libc.
      This is to make it easier for others to implement the actual logic,
      without having to know much about plex86 internals.  There are some
      notes in the plugin C code.
    I also created a new button (really lame) on the GUI.  This triggers
      a call to the plugin, so the user can select when to flush (commit)
      the writes in the write-cache to disk and clear the cache.
      Eventually, a nice menu system might do this with a "Are you sure?"
      dialog or something similar.
    It is possible that a large number of disk writes may exceed the
      size of the write-cache (which should be configured by an option
      to the plugin eventually).  In that case, it would be good to present
      the user with 2 options: 1) Quit altogether and ignore writes,
      and 2) commit writes and continue.  This should eventually be
      GUI oriented, by stdin would be OK for now.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sun Dec 24 19:07:34 EST 2000
    Integrated mods from Warwick Duncan for some remaining C++ -> C
      comment conversions in plugin directories.
    Added dynamic translation idea chapter to docs.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Thu Dec 14 11:00:07 EST 2000
    Guest NetBSD 1.5 works, small patches from:
      "Frank van der Linden" <fvdl@wasabisystems.com>
    Added loopback mount instructions to docs/ tree.
      New docs/output.tar.gz file.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Thu Dec 14 08:11:40 EST 2000
    Integrated small diffs from Frank van der Linden:
      Small patches to host-netbsd.c to fix minor interface changes,
      and make plex86 compile on upcoming 1.5.x releases, as well as
      the development NetBSD sources.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Thu Dec 14 01:04:17 EST 2000
    Integrated NetBSD port from
      "Frank van der Linden" <fvdl@wasabisystems.com>
    Added framework for running user code (ring3) without prescanning.
      This isn't complete yet, so don't turn off prescanning.
    IsNullSelector() takes type selector_t now.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sun Dec 10 17:27:17 EST 2000
    Windows NT 4.0 Build 1381 Service Pack 1 boots. (added conf/winnt)
    Enabled CDROM emulation.  By default, CDROM emulation is compiled
      into the bochs plugin now.  Pass '--enable-cdrom=no' to configure
      if you don't want it.
    Added OUTSB_DXXb instruction emulation.
    Replaced some monpanic calls with monprint.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sat Dec  9 15:14:55 EST 2000
    Added 'docs/xml' directory.
    Moved most existing documentation to XML/DocBook format.
    There is a file 'docs/output.tar.gz' which is a pre-generated
      set of HTML docs files, since most people won't want to generate
      their own from the XML/DocBook source.  Whenever you update
      from CVS, go into 'docs' and type:

        user:docs-> make clean
        user:docs-> tar xvfz output.tar.gz

      This will put files in 'docs/output/html/', as if you generated them
      yourself.  Then point your browser at 'docs/output/html/book1.htm'.
    These docs need to be filled out and beautified.  I just copied some
      of the text files into <literallayout> sections.
    Ultimately, I'm hoping some other people take over maintaining the
      documentation tree.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Thu Dec  7 21:38:38 EST 2000
    Fixed handling of 16-bit stack segments in guest when monitored in PM.
    QNX v4 demo boot floppy works now. (added conf/qnx)

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sun Nov 26 17:52:27 EST 2000
    Urgent fix to nexus-mode.c.  Was using pointers from wrong space!
      If you ran a version recently, make sure to do the following:

        - Reboot you system.  You may have stepped on random memory addresses.
        - 'make clean; make' from the top.

    I'm going to give QNX a spin again after this fix.
    If the recent mods broke something for you, give it a try again.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sun Nov 26 00:16:11 EST 2000
    Optimization: prescan() returns -1 if the very first instruction
      has already been prescanned, and is virtualized.  Since this
      instruction must be emulated, flagging this condition allows
      us to eliminate unnecessary exception processing.  Pretty cool,
      just a few lines of code eliminated about 5% of execution time
      in plex86.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sat Nov 25 23:22:53 EST 2000
    Implemented IO instructions which were bombing Debian rescue floppy.
    Fixed mode handling of monitor running guest code in v86 mode.
    I booted the Debian rescue floppy image (2.2.17 kernel).  The
      root floppy seemed to mount OK, but I didn't have the patience
      to continue the boot process.  Let me know if you test this.
    Added 'createdisk.sh' & 'createdisk.README' to the misc/ dir,
      from Eric Laberge.  These are scripts to create empty hard disk
      images.  I haven't tried them.
    Also reran Prama Linux (2.0.33 kernel), FreeDOS, and Win95.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Fri Nov 24 16:29:41 EST 2000
    Integrated small bug fix from Robert Fitzsimons for plex86.c for
      processing of command args.
    Created REQUIREMENTS file, which lists CPUs that plex86 will work on.
      This list was generated some time ago from the TLB test program.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Fri Nov 24 14:18:59 EST 2000
    Integrated small floppy.cc patch from Hanish Menon.  If you specify
      a floppy A or B as the boot device, but don't provide a valid
      file/device, a panic will be generated.  The BIOS still needs to
      be fixed to detect if you do provide a floppy but it is not bootable.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Fri Nov 24 10:38:25 EST 2000
    Fixed problem with FreeDOS mini boot floppy image EIP > limit.
      I forgot that (floppy) DMA transfers are writing to memory mapped
      guest memory, so the prescan cache was not being invalidated.
      Added an ioctl to invalidate the cache, and works fine again.
      This may fix problems with other boot floppies not working.
      BTW, the FreeDOS beta5 mini.bin image is not good.  I got it from:
      
        http://www.ibiblio.org/pub/micro/pc-stuff/freedos/files/distributions/beta5/mini/mini.bin

      Execution gets "sent into the weeds".  After some time debugging
      it, I popped it in a real PC, and it hangs on that too.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Thu Nov 23 21:28:30 EST 2000
    Replaced C structure copy with a call to our helper function
      copy_memory() in nexus-prescan.c.  User reported undef'd memcpy
      without this.
    Rehashed plex86.c options handling.  You can pass all the config
      file options on the command line now.  Just use -o "option line".
      Make sure to quote the text of the option if it has spaces or
      magic characters, as each argument is passed to the config
      option parser as it where an entire line in the config file.
      You can also use multiple config files, or no config files at all,
      though some options have to be set to do anything useful.  Options
      and config files are processed in the order in which they are
      passed.  You can now do things like:

        ./plex86 -o "memory=4" -f ../conf/freedos -o "db_syntax = at&t"
        ./plex86 -f /tmp/conf1 -f /tmp/conf2 -o "..."
        ./plex86 -o "..." -o "..." -o "..."
        ./plex86 -f ../conf/freedos

    Added intelligent error reporting for options from command line
      and/or config files.  Exact line and character of errors are
      reported to user, along with a message.
    Config file option "load-rom" now has a colon following it, and
      a comma delimiting options, for example:
      "load-rom: file=../bios/BIOS-plex86-2000_0817a, address=0xf0000"
    Config file option "prescan_depth" was replaced by:
      "prescan: depth = N, ring3 = {on,off,auto}"
      The "ring3" clause is not used yet, but will ultimately modify
      the behaviour of SBE on ring3 code, according to your preference:
        on   = always control execution of ring3 code with SBE. (current)
        auto = turn off SBE intervention when conditions look reasonable.
        off  = always turn off SBE intervention of ring3 code.
    Some other cosmetics.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Wed Nov 22 14:55:57 EST 2000
    Integrated small curses GUI patch from Josh Wilmes.
    Integrated small load_module.sh patch from Mark Zealey.
    Implemented INSB_YbDX instruction.
    Small hacks to dma.cc to allow word read.
    Use different virtualization opcode map when running guest in
      v86 mode.  This allows natural execution of instructions which
      read selectors or reload segment registers.  I notice this shaved
      about 15% off the boot time for FreeDOS.  Protected mode code
      won't benefit from this mod.
    Commented panic which was causing Windows to panic when you fired
      up the Settings or Programs menu.  Should handle cross dword boundary
      memory mapped IO better in case it matters for a device in the
      future.  Anyways, I was able to start the Control Panel, and
      shutdown Windows all the way to the final "OK to shutdown" screen.
    Added docs/txt/DOS622.HOWTO from Michael Madore's email.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Tue Nov 21 14:33:14 EST 2000
    Integrated patches for color text handling in curses GUI from
      Josh Wilmes.
    Converted all C++ comments to C for files in kernel space.
      There's only a couple more straight C files in user space left to go.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Tue Nov 21 00:49:47 EST 2000
    Added a bogus RDTSC instruction, implemented RETfar32_Iw and
      defined bits in CR4 to get QNX to boot quite a ways.  Panics
      on use of PCD/PWT bits in PTE.  Probably harmless.  Will
      try to update handling and re-run later.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Mon Nov 20 22:25:53 EST 2000
    Bug fixes and recoding of mon-paging.c.  Some of the comments
      for the mods need to be changed.  I'll do that next.
    Added an acceleration for INSW, transfer is done in user space
      potentially in blocks of words direct to guest memory.  I notice
      FreeDOS boots from the disk image file in only a few seconds
      after this mod.  :^)  This only helps if you do a REP INSW, which
      the BIOS does.  Could extend this method to other instructions.
      Gives us an idea what special data transfer can do for performance.
      This should model the kinds of gains we can get with special
        guest OS device drivers.
    VCode cache entries are invalidated individually.  Before I was
      dumping the whole set when one got stepped on.


  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sun Nov 19 16:04:40 EST 2000
    Implemented WBINVD instruction.  Ingnored for now.  This was
      tripping up some people's attempts to install/boot OSes.
    Moved part of mon-prescan.c to host-prescan.c.  This will eventually
      contain multiple maps, one for each of various CPU modes.  The
      current map is used for all modes, which is very inefficient, since
      in v86 mode, there is no need to monitor segment loads or selector
      reads etc.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sun Nov 19 12:51:38 EST 2000
    Integrated patches from Josh Wilmes for gui/curses.cc.
    Integrated/completed code cache patches from Ramon van Handel.
      There are now N code cache pages, rather than 1.  We'll need
      to look at various page algorithms to figure out which pages
      to kick out of the cache when a new one is needed.  For now,
      ICACHE_PAGES pages is hardwired in kernel/include/monitor.h
      Keep in mind there is one meta page and one vcode page for each
      virtualized code page, thus we consume 2*ICACHE_PAGES.  I have
      this set to 64 pages which is 2*64*4096 = 512KB total.
    Reworked mon-paging.c a little.
    Bug fixed of macro LOG_BUFF_SIZE:
      from (LOG_BUFF_PAGES*512) to ((LOG_BUFF_PAGES)*4096)
    Single word disk IO is still dragging down performance.  Need to
      handle disk IO in blocks.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Thu Nov 16 02:27:24 EST 2000
    Changed use of macro current_got_fatal_signal() to
      signal_pending(current), for Linux2.4.0-test10, as
      per suggestion from Eric Warmenhoven.  That macro was in
      wrapper.h and went away with test10.
    Fixes to emulation/fpu.c.  Had '&&' instead of '||'.
    Renamed files in kernel/.  Files starting with 'host-' contain
      code existing in host space.  Files starting with 'mon-' contain
      code existing in monitor space.  And files starting with 'nexus-'
      contain code existing in nexus space (access to both spaces).
      This way routines can be coded to access memory in a particular
      space more efficiently, without extra pointer dereferencing.
    Added support of the A20 Enable line.
    Many changes regarding handling of various guest CPU modes, and
      the modes they are 'monitored' in.
    Implemented more protected mode instructions and control transfer
      logic.
    Support guest v86 mode.
    Handle legacy protected mode values in descriptor caches, after
      transitioning to real mode, using virtualized segments and running
      guest code in protected mode.
    There are many #warning directives.  Excuse the mess for now,
      there is much to clean up.
      

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Thu Nov  2 16:56:09 EST 2000
    Redirect memory mapped IO addresses in the VGA framebuffer,
      to the VGA emulation.  Took out hack to periodically copy
      physical guest memory to VGA framebuffer.  VGA graphics
      modes should be enabled now.
    Added an abort_code to the vm_t so we can tell where a panic
      happened, if it is not safe to print a message.  (for example
      in the exception handler)

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Tue Oct 31 17:45:07 EST 2000
    Filled in a little more emulation of task switches
      and interrupt handling.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sun Oct 29 21:28:04 EST 2000
    Linux kernel 2.4 works as a host now.  (I used linux-2.4.0-test9).
      There is a bug as of that version in the Linux kernel header
      file 'include/linux/wrapper.h'.  Macros mem_map_reserve()
      and mem_map_unreserve() do not use the proper parenthesis.
      They are listed as:

        #define mem_map_reserve(p) set_bit(PG_reserved, &p->flags)
        #define mem_map_unreserve(p) clear_bit(PG_reserved, &p->flags)

      ...but should be:

        #define mem_map_reserve(p) set_bit(PG_reserved, &((p)->flags))
        #define mem_map_unreserve(p) clear_bit(PG_reserved, &((p)->flags))

      I just used set_bit() and clear_bit() directly to get around using
      these macros.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Fri Oct 27 14:58:38 EDT 2000
    Added toplevel files 'TODO' and 'PERFORMANCE'.  These describe
      general stuff to-do and performance to-dos respectively.
    Integrated patches from Jeff Dubrule:
      user/user.c: also try opening /dev/misc/plex86.
      kernel/monprint.c: get rid of varargs warning.
    Couple small hacks to avoid panics in harddrv.cc for Tom's root-boot.
      Search in that file for tomsrtbt, and uncomment the hacks.  I
      don't want to have these hacks in the code normally yet.


  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Thu Oct 26 23:17:55 EDT 2000
    Integrated DevFS support patches from Jeff Dubrule


  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Thu Oct 26 12:55:59 EDT 2000
    Linux 2.0.31 boots!  (Used a disk image file from bochs with
      Linux already installed)
    Added RTEMS to GUESTS file
    Moved guest_context_t to kernel space.  User code now interfaces
      with monitor via guest_cpu_t.  Got rid of set_guest_context()
      and get_guest_context().  This broke the ICE plugin, which
      I have commented out temporarily in the Makefile.
    Monitor can now handle running guest in 3 modes:
      1) Execution as normal (SBE controlled)
      2) Execution by breakpointing (setting TF and run guest for
         duration of one instruction at a time.
      3) Emulation.  Monitor keeps in an emulation loop.
    Monitor print output is now channeled to user space program,
      rather than system log.  Mapped the 4k monitor print buffer
      to user space for efficiency.  Still have to clean up the
      print code.
    Fixed bug Ramon found in handle_fault, where 'from =' code was
      evaluated too soon, possibly before a guest mode transition
      resulting from emulation of a guest system instruction.
    Lot's of #warning directives in C files to remind myself things
      that need to be cleaned up.  :^)
    Fixed bug in kernel/prescan.c.  If a branch instruction was found
      at the maximum level of prescan recursion, then it would not
      be virtualized.
    Updated the paging code, both in kernel/vpaging.c and
      emulation/paging.c.  It now accurately handles error codes
      and A&D bit updates in the page dir/table entries.
    Changed the delivery of guest execution times and user space
      periodic callbacks to the user code.  Some stuff is hacky
      and needs to be reworked.
    Turned SBE on for real mode code (run in monitor as v86 code).
      This will slow FreeDOS down until I optimize things.
    SBE still only is dealing with one page at a time.  This I
      will change very soon.  This rev will be very slow because
      of this.  Picture that every out of page branch dumps the
      virtualized code cache.  Ouch!
    I think tomorrow, I can put a preinstalled Linux disk image file
      on the 'net for developer to try.  It's 126Meg file generated
      some time ago with bochs.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Fri Sep  8 09:05:49 EDT 2000
    Added FWAIT to list of instructions to virtualize
    Coded POPAD32.
    Filled out a little more of jump_protected
    Created a suite of tests to examine updates of A&D bits in
      PDE/PTE, test various paging permission handling, and
      error reporting on real processor.  Updated paging.c accordingly.
    The virtualized paging in the monitor (vpaging.c) needs to be
      synchronized with the emulation paging facility.  Also, A&D
      bits from the actual monitor page tables and the guest's page
      tables need to be synchronized.  (Same deal for the A bit of
      data segment descriptors)
    Need to detect case where guest code is modifying the page directory,
      or the page table of the currently executing code.  On a processor,
      the TLB cache will be used until reloaded.  Since plex86 is
      dumping the page tables and rebuilding upon a write, this can
      cause problems, and send guest code "into the weeds".
    I'm also occasionally preparing some tests to throw into a test
      suite for plex86, to validate the virtualization/emulation.
      Think I'll make a separate validation module for the plex86
      CVS tree.  This well let people interested in being "testers"
      help out more.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Mon Sep  4 16:53:14 EDT 2000
    In special cosimulation mode, plex86 boots Linux!
    Fixes to virtualized paging; CR2 was not being set
    More emulation of instructions which require virtualization.
    Mods to paging code.
    Created kernel/vsegment_nexus.c.  Moved stuff here that needs to
      be accessed from either host or monitor space.
    Fixed some handling of segment/virtualized-segment synchronicity.
    Fixed error handling init_module().  Must unregister services if
      an error is returned, otherwise kernel goes unstable.  Also
      return negative error now.
    Added emulation of hardware tasking.
    Increased PLEX86_MAX_MONITOR_PAGES so bigger monitor will fit.
    Fixed a bunch of inline assembly statements, so they explicitly
      request 8-bit registers when necessary.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Thu Aug 31 20:33:04 EDT 2000
    Implemented OUTS instruction.
    Fixed inline asms for some 8-bit instructions - added directives
      to force compiler to use 8-bit capable registers, where
      appropriate.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Thu Aug 31 15:37:01 EDT 2000
    Finished emulation of user-level integer instructions.  The rest of
      the instructions can be added as they are encountered.  That
      was a lot of typing.  Anyways, I tried to implement emulation in
      terms of inlined assembly use of the native instruction, whenever
      possible for speed.
    Now back to virtualization framework hacking, while booting
      Linux/Windows in plex86...

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Wed Aug 30 17:03:40 EDT 2000
    Added emulation of more user level integer-oriented instructions.
      Only a few more to go...  Then on to running Linux/Windows in the VM.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Mon Aug 28 14:39:23 EDT 2000
    Added emulation of a bunch more instructions.  I'm working towards
      being able to emulate all integer unit instructions.  There's
      another batch to do.  Soon, I'll move on to getting Linux and
      Windows running in plex86.  Note that a lot of the emulation
      isn't regularly used, but it will eliminate future panics
      when a virtualization condition hits a random instruction.
      It is also very useful for cosimulating against bochs and
      debugging.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Mon Aug 21 13:17:05 EDT 2000
    Changes from Ramon:
      BOCHS plugin now recognises the vga_update_interval
        setting correctly.
      Reduced the user executable size from approx 1.4MB
        to 0.1MB.
      Removed the superfluous --with-sdl command from
        configure.in.  {I generated a new configure (KPL)}

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Thu Aug 17 00:55:17 EDT 2000
    Added SDL code from Martin Garton (untested)
    System BIOS now scans for additional ROMs, and calls the initialization
      routine in them, if they exist.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Mon Aug 14 14:04:16 EDT 2000
    Bugfix to kernel/emulation/fetchdecode.c.  Not all prefixes would
      be handled properly with respect to page boundaries.
    Moved plex86.conf.* files to a newly created top-level 'conf' dir.
    Added VGABIOS-elpin-LICENSE file.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sun Aug 13 17:31:04 EDT 2000
    Tagged last version with 'plex86-0_1_0'

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>:
    Plex86 now boots FreeDOS!  Some real base instructions in
      README.DOS
    VGA BIOS from Elpin Systems, Inc, is now licensed for use with
      plex86, courtesy of MandrakeSoft.  It's in bios/VGABIOS-elpin-2.40.
    Implemented emulation of a bunch of 16-bit instructions in
      case virtualization conditions necessitate them.
    Add ioctl calls to set/get complete guest CPU state.
    Debug information now displays correct disassembly information for
      real-mode, and protected-mode with segment bases not 0.
    Converted C++ to C comments for files in the kernel/ tree
      and top level of the user/ directory.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Thu Aug  3 13:14:16 EDT 2000
    Integrated patches from Ramon.  He coded a new and more efficient
      IO redirection service for routing IO requests from the VM
      to the proper user space handler.
    Integrated patches from Josh Wilmes.  Minor error reporting on
      failed open()/fstat() calls, and a wait option before terminating
      program execution.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Wed Aug  2 15:30:46 EDT 2000
    DOS 6.22 boots from floppy and disk image file.  Anything other
      than booting and typing 'DIR' will likely resulting in a panic,
      since I only implemented enough instruction to get there.
    Added running real-mode guest code capability to the VM.  Code
      is run in v86 mode in the VM.  Modified the guest_context
      stack frame conventions a little to accommodate the extra
      data seg selector pushes in v86 mode.
    Added more emulation of instructions.
    Rearranged the user/plugin code to add more flexibility to
      the plex86.conf initialization.
    Plex86.conf can now pass bochs options to the bochs plugins.
    Moved the BIOS over from bochs.  A precompiled BIOS is available
      in bios/BIOS-plex86-*.  Added instructions to the toplevel README
      on how to compile your own.
    You can now load a system BIOS and VGA BIOS into memory.  If
      you load a VGA BIOS, then you don't need to use the replay_io
      plugin.
    Hacks:
      Int 0x15, AH=0x87 is hacked in kernel/emulation/soft_int.c to
        return CF=1 (error).  This is because I don't support a
        transition to PM yet from RM, necessary for extended memory
        BIOS operations.
      SBE is effectively disabled for running RM guest code.  In
        kernel/fault.c, there is a goto hack.  I will take this
        out at some point, after I get more instructions emulated,
        so that SBE works.
      In user/plugin.c, there are hacks to inport and output routines
        for port 0x01f0 (hard disk).

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sat Jul 22 12:26:15 EDT 2000
    prescan() is passed a linear address to read from instead of opening
      a guest phy page.  This is more efficient as the guest addr has
      to be mapped in anyways.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Sat Jul 22 12:26:15 EDT 2000
    Added all the remaining devices from bochs to plex86: CDROM,
      CMOS, floppy, ...
    Moved x86 instruction emulation code to kernel/emulation.
      Corresponding header info was moved to kernel/include/emulation.h.
      Emulation files are compile with -DEMULATION; emulation.h
      exports only a few features to the monitor without this defined.
    Linux as a guest partially boots.  You will need 'initrd.img' and
      'linux.img' if you want to play with this.  These are the initial
      ramdisk and kernel image files.  Should panic with a HLT with IF==0.
      The VGA text framebuffer at the time of panic is dumped to the
      system log.  Grab these files from the ftp site.
    Fixed some bugs.
    Added extra options to plex86.conf and code to be able to load
      Linux into memory and set up the environment as it were loaded by
      a real boot-loader.
    Rehashed the way time events are delivered.  This is unfinished.
      Eventually there will be both a wall-clock time reference for
      things like GUI updates, and a VM time reference.  Will fix this.
      VGA updates will be slow for now.  Don't worry about this for now.
    Added a macro ANAL_CHECKS to config.h.in.  By default it's on.  Some
      extra safety checks are made or debugging purposes.
    There is now a bios directory.  This will ultimately contain the sytem
      ROM BIOS.  For now, it's a null BIOS.  I changed the way the kernel
      image loading was hooked with the BIOS.  It is independent now.
    VGA IO event replaying behaves like it did previously.  Events are
      read from a file.
    Implemented a setjmp/longjmp for use with guest excepion handling.
      In the file kernel/include/vmsetjmp.h, you can also use the gcc
      builtin ones as well.  These calls are bracketed with CLI/STI
      instructions because I was debugging stuff, and I just realized
      now that I left them in there.
    Need to move 'user/plugins/bios' to something more suitable like
      'user/plugins/load-kernel'.
    Other stuff that I forgot about.  Also, there might be some legacy
      debug stuff in there that I forgot about.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: Thu Jun 29 01:25:34 EDT 2000
    Bios plugin no longer loads the kernel image.  This is done
      separately now, by 'load-kernel.so'.
    Added 'replay_io.so' in lieu of the vga.S file.
      It replays the IO trace file 'misc/vga_io.log' to the vga
      emulation.  This is more efficient, and allowed for separation
      of the BIOS functionality.  This is in a new plugin directory 'misc'.
    Added 'load-rom.so' which does nothing effectively, since we
      don't have a ROM BIOS yet, but will load one when we do.
    I changed plugin_load() so it adds modules to the end of the list.
      This way, they are added in the order in which they are
      used in the 'plex86.conf' file.
    Created a new toplevel dir 'bios' which will ultimately hold
      our ROM BIOS.
    Stack_address is now a parameter to the kernel loader plugin.
    Added line continuation character to plex86.conf for readability.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>:
    Incoporated fix from Ulrich for broken logic in fetchdecode.c.
    Added support for monitor interruptibility.  Interrupts
      disabled dwell time is really short.  This version should
      be a lot nicer to your host OS.
    Fixed bug in sbe(); for guest CR0.PG==1, getting address of
      monitor page table was wrong.
    Fixed bug in get_cpu_capability(); added ecx and edx to list
      of registers 'touched' by cpuid instruction for eax==1 case.
    Split vm->state into vm->mon_state and vm->mon_request, to
      keep things sane for an interruptible monitor.  Because even
      a HW interrupt needs to call sbe() after it's done, even the
      end of the HW interrupt handler has to be interruptible
      (the sbe() function can be long).  Thus, HW int redirection
      does not step on the vm->mon_state.
    I added some extra push/pops to the transition code.  Not sure
      if all of it was needed.  Will look into eliminating code
      not needed.  But, I wanted to quickly get a release out
      that had the above fixes in int.
    Please try all 3 guests: cooperative, preemptible, paging on
      your machine and report.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>:
    Added prescan_depth option to plex86.conf, and corresponding
      ioctl kernel code so this parameter is now configurable.
    Filled out some support for a guest with paging enabled.
    The 'paging' guest now works!
    Added a function to grok the CPUID instruction for feature
      support.
    Changed use of KERNEL_VERSION() to VERSION_CODE() in host-linux.
      That wasn't defined on a lesser kernel.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Minor simplicfication of SBE logic.
    Various interface cleanups.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: (plex86-2000_0618g)
    Removed user/{emulation.c,emulation.h}; no longer needed.  This
      stuff is all done in the monitor now.
    Abstracted a lot of the host-linux.c code to host-all.c.  This
      is a first step in making porting to other platforms easier.
    Reworked some framework so that supporting a guest that uses
      paging will be easier.
    Renabled passing interrupts from user code to monitor.
    The 'preemptive' guest works again.
    Page tables are built dynamically.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>: (plex86-2000_0605b)
    Added monitor log facility to use (printk like) to print
      debug messages in either host or monitor space.  Text
      is redirected to the host kernel print facility.
    Changed nexus functions to switch from host<-->monitor,
      rather than host<-->guest.  This gives us flexibility to
      return to monitor code anywhere, including in the middle
      of emulating an instruction.
    Moved some fields out of nexus.  Only data needed to
      transition between host<-->monitor/guest context should be
      in nexus.  All other fields are easily accessible from either
      space and should be in VM structure proper.
    kernel/emulation.c now uses same fetchdecode function as
      prescan code.
    Totally recoded emulation.c, and split out emulation of
      many functions into *.c files, which strangely mirror
      the names of files in bochs. :^)  Sorry, this helped
      me keep things straight.
    Added segmentation protection checks to data accesses,
      control transfers, etc, in instruction emulation.
    Added much protection model framework to instruction
      emulation.
    Paged guests are not supported yet, though much of the
      new framework has been thought out with that in mind.
    Likely only the cooperative guest works so far, as that's
      the only one I've been working with, and I implemented
      just enough to get it going.
    Changed the user<-->host<-->monitor interface.  The user
      code does not participate in the emulation of instructions
      any more, except to serve IO and other hardware requests.
    SBE (Scan Before Execute = Prescan) logic is now an integral
      part of plex86.  In fact, both ring0 and ring3 code are
      currently always controlled by SBE.  Much thought went into
      structuring the code, so that we can make a dynamic decision
      to switch SBE off when running ring3 code, yet the emulation
      functions will still work.
    Lots of other doings...
    NOTE: Things should and do run a _lot_ slower for the moment,
      as everything is being controlled by SBE, and I'm not doing
      some things very optimally.  Things will speed up greatly
      in the future.

  - "Edouard G. Parmelan" <Edouard.Parmelan@quadratec.fr>:
    Add module option to choose static device major number.

  - "Jens Nerche" <jn4@os.inf.tu-dresden.de>:
    Support for multiboot guest kernels.
    Add shadow descriptors after LGDT.  
    Add ljmp emulation.

  - "Ramon van Handel" <vhandel@chem.vu.nl>:
    Added BIOS plugin; moved guest loading and VGA init there.
    Port I/O dispatching fixes.
    Don't call X routines from within signal handler.
    Some coding style clean-up.
   
  - "Bryan Meredith" <bryan@bear-mountain.demon.co.uk>:
    Use dynamically allocated /dev/plex86 major number.

  - "Edouard G. Parmelan" <Edouard.Parmelan@quadratec.fr>:
    Fix incorrect plex86_proc_entry.namelen.
    Fix OOPS in host2guest when compiled without SBE.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Automatically choose appropriate warning level when using gcc.
    configure.linux removed.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Removed some more references to FreeMWare or FMW.

  - "Edouard G. Parmelan" <Edouard.Parmelan@quadratec.fr>:
    Fixed Makefiles to allow build in separate directory.
    Added --with-linux-source= configure option.

  - "C Hanish Menon" <hanish@innomedia.soft.net>:
    Don't load guest code beyond available guest memory.
    Removed data / bss offsets.

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>:
    Mapped VM (vm_t) structure into monitor space.
    Rehashed the vm_t structure and other framework so that
      we can access VM fields from either host or guest address
      spaces.  If you stick to accessing VM fields using only
      pointers in the 'common' substructure, code will work in
      either space.  The 'addr' field is actually just a pointer
      which is changed during a space transition, to point to
      set of pointers for that space.  This way, general code
      doesn't need to know or care what space it executes in.
    Functions in kernel/fault.c now pass the 'vm' pointer on
      to functions in kernel/emulate.c.  Also, all functions in
      kernel/emulate.c pass the vm pointer.  These functions should
      execute in either space now.
    Moved vm_nexus() from kernel/include/monitor.c to kernel/fault.c
      so it won't be used outside of these functions.  It is space
      dependent, and shouldn't be used outside of this file anymore.
      It is used to grab the vm pointer from the nexus upon fault,
      and is not needed thereafter during the fault handling.

  - "Janek Hiis" <janekh@math.ut.ee>
    Added #ifdef in kernel/host-linux.c for move of i_mmap field,
      in newer Linux kernels.  2.3.99?

  - "Kevin P. Lawton" <kevin@mandrakesoft.com>:
    Integrated scan-before-execute (prescan) code into main
      source tree.  (It was previously developed independently)
    Integrated split I&D TLB code into main source tree.  (It
      was previously implemented only in a small test program)
    The above components are not fully operational yet.  They
      are not used by default.  To try them, after typing
      configure.linux, modify the kernel/Makefile CFLAGS
      variable as per the commented line, then compile.
      Otherwise they are not used, and things should function
      as per previous revs.
    Changed references to FreeMWare to plex86.
    Added checks to return value of retrieve_vm_pages() calls
      in kernel/host-linux.c.  Also fixed one minor flaw there.
    Created a docs/ directory structure.  Moved the public paper
      and memory address space graphic there.

  - "Nick Bastin" <nbastin@mil3.com>:
    Partially converted public paper to sgml/DocBook format.
    Added new output targets to docs Makefile system.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Initial import of freemware-20000213 into plex86.
    FreeMWare ChangeLog file is appended.

 ==========================================================================

  - "Ramon van Handel" <vhandel@chem.vu.nl>:
    New guest vs. monitor eflags handling.
    IOPL dependent behaviour of STI/CLI/IRET.
    Bugfix: correct limit handling for big segments.
    Added new 'paging' guest kernel.

  - "Ramon van Handel" <vhandel@chem.vu.nl>:
    Added interrupt interception to kernel module and plugin interface.
    Moved IF handling to kernel-side, using a VIF/VIP mechanism.
    Various fixes/cleanups for the ICE plugin.
    Use US keyboard map in virtcode.

  - "Ramon van Handel" <vhandel@chem.vu.nl>:
    Added VGA font and updated README accordingly.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Simplified loading of guest kernel.
    Store page numbers instead of addresses in VM data structure.
    Removed all dependecies on host.h from non-OS specific files.
    Moved freemware.h to root, don't include kernel headers from user.
    Removed unnecessary #include directives.
    Removed Bochs macintosh GUI.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Removed the 'NT-flag hack'; instead set all but current code segment
    descriptors to type 'data' to make iret trap.  This allows to run the
    *unmodified* cooperative and preemptive kernels.

  - "Ramon van Handel" <vhandel@chem.vu.nl>:
    Clean up: split user.c into multiple files.
    Main executable renamed to 'freemware'.
    Added emulation for ins/outs instructions (including rep variants).
      [partially based on a patch due to Jens Nerche] 
    Added emulation of int3 instruction.
    Always use 'bx_printf' for Bochs debug output.

  - "Josh Wilmes"  <josh@hitchhiker.org>:
    Added curses UI for bochs plugin.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Added support for raising hardware interrupts to kernel module.
    Added INTR interface between user and plugins.
    Added PIC/PIT hardware models from Bochs.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Cleaned up user/kernel interface a bit.
    General clean up: CFLAGS simplified, compiler warnings fixed.

  - "Jens Nerche" <jn4@os.inf.tu-dresden.de>:
    Added emulation for more instructions.
    Updated 'virtcode' app to include minishell.

  - "Ramon van Handel" <vhandel@chem.vu.nl>:
    Miscellaneous bugfixes.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Simplified user mode instruction emulation interface.
    Increased Bochs timer update frequency.

  - "Ramon van Handel" <vhandel@chem.vu.nl>:
    Resource allocation and event handling for plugins.
    user.c reorganization / cleanup.
    Added I/O emulation plugin interface to Bochs.
    Updated 'virtcode' app to test keyboard input.

  - "Jens Nerche" <jn4@os.inf.tu-dresden.de>:
    Added support for in/out instruction emulation.
    Fixed and improved VM dump.

  - "Ramon van Handel" <vhandel@chem.vu.nl>:
    Fixed AT&T syntax suffixes.
    Added gdb remote protocol documentation.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Allow to change 'safe' guest flags from user mode.

  - "Ramon van Handel" <vhandel@chem.vu.nl>:
    Bugfix to 'mywait'.  Make VGA timer handler restartable.
    Fixed 'Power' button.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Removed .bin guest kernels (always use ELF).
    Compile shared libraries with -fPIC.
    Compile everything except kernel with debug symbols.
    Fixed memory overwrite bug in plugin handling.

  - "Oleg Drokin" <green@crimea.edu>:
    Disable VGA timer handler before unmapping guest memory.

  - "Ramon van Handel" <vhandel@chem.vu.nl>:
    Added support for plugins.  
    Make Bochs device emulation framework a plugin.
    Added ICE plugin for remote GDB debugger support.
    Added AT&T syntax instruction decoding to decode.c.
    Removed timeout handling; general clean-up of user.c.

  - "Kevin P. Lawton" <kevin@bochs.com>
    Integrated VGA, keyboard and certain IO device emulation framework
    from bochs into FreeMWare.  Writes to the text video memory
    are now displayed in an X window via the VGA emulation.  The
    cooperative kernel was tested and displays fine.
    Added periodic timer signal handler to user code, so it can
    drive updates in the IO device emulation.
    Expanded copyrights in user/ from 1999 to 1999-2000.

  - "Ramon van Handel" <vhandel@chem.vu.nl>:
    Updated README file.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Bugfix: return modified guest_context on EINTR.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Bugfix: monitor_pages should contain virtual, not physical addresses.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Removed linker script; retrieve module address range from kernel data.
    Use standard 'current_got_fatal_signal()' test.
    Return -EINTR instead of -ERESTARTSYS.
    Added some missing .globl definitions in nexus.S.
    Added NT flag hack to preemptive kernel as well.
    Bugfix: do *not* allow direct I/O access to guest!

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Added emulation for lgdt/lidt/lldt/ltr/int/iret instructions.
    Activated nexus selector migration (after lgdt).
    Hack: set NT flag during guest execution so that iret traps.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Improved/fixed kernel emulation operand decoding.
    Added decoding of some more instructions.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Separated monitor from guest initialization code.
    Prepared for monitor/nexus linear address / selector migration.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Made transition code more self-contained; don't rely on host 
    parameters not changing over time.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Map module pages into guest at arbitrary linear adresses
    by shifting the monitor segment bases accordingly. 
    Re-enabled monitor-side fault handlers.

  - "Kevin P. Lawton" <kevin@bochs.com>:
    Simplified nexus access via C structure.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Support for multiple VMs on the host (Linux) side.

  - "Kevin P. Lawton" <kevin@bochs.com>:
    Implemented transition into guest space using 'nexus' page.
    Prepared for managing multiple VMs.

  - <beef@niggard.org>:
    Fixed some warnings.

  - "Ben" <ben@comp.uark.edu>:
    Portability fixes for Linux versions >= 2.3.25.

  - "Todd T. Fries" <toddf@acm.org>:
    Portability fixes.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Fixed incorrect decoding of some instructions (user).
    Added effective address decoding to kernel emulator.

  - "Jens Nerche" <jn4@os.inf.tu-dresden.de>:
    Added beginnings of emulation framework.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Added Intel instruction decoder to 'user'.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Bug fix: run guest at IOPL 0, not IOPL 3.

  - "Oleg Drokin" <green@crimea.edu>:
    Fix signal handling on glibc systems.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Pass guest context from/to user space on FMWRUNGUEST ioctl.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Removed 'quanta' argument to FMWRUNGUEST ioctl.
    Replaced 'quanta' handling in 'user' by SIGALRM-based timeout mechanism.

  - "Jens Nerche" <jn4@os.inf.tu-dresden.de>:
    here comes the patch for loading ELF binaries. I enhanced
    the config file: text address, data address, stack address
    and bss address are new options, and user.c undertands -d
    for debugging messages now.
    If the file is in ELF format, the text address is taken from
    file, if it is a pure binary it's taken from config file.
    data address and bss address are still unused.

  - "Love" <lha@stacken.kth.se>:
    Adapted Makefiles to allow build outside the source-tree.

  - "Jens Nerche" <jn4@os.inf.tu-dresden.de>:
    Command line parameters and config file for 'user'.

  - "Ulrich Weigand" <weigand@informatik.uni-erlangen.de>:
    Created 'guest' directory to contain various test guest kernels.
    Moved dummy virtcode to guest/virtcode.
    Added Ramon's nullkernel as guest/cooperative and guest/preemptive.
    Made guest kernel to load a command line parameter to 'user'.
    Reorganized interrupt/exception handling in kernel module.

  - "Kevin P. Lawton" <kevin@bochs.com>:
    Fixed soft_int macro.


fmw-19990826a:

  - (UW) [host-linux.c] [host-linux.h]
    Added check to prevent the guest physical memory from being
    freed while still mapped to user space.
    Added check to prevent duplicate allocation (memory leak!).
    Use 'mem_map_(un)reserve' macros instead of manipulating page
    flags directly.

  - (UW) [user.c]
    Unmap guest physical memory before freeing it.


fmw-19990825a:

  [patch1 from UW]
  - Use some linker/ELF magic to retrieve the linear address space
    occupied by the kernel module at runtime.

  - Walk the host page tables to retrieve a list of all physical pages
    underlying the kernel module's code and data sections.

  - Allocate memory for the monitor page tables, and set them up as follows:
    * An identity mapping of the complete guest physical memory (at address 0)
    * A mapping of the kernel module's code/data pages at the same
      linear address as in the host space.

  - Remove the dynamic allocation of interrupt reflection stubs
    (sorry Ramon :-/).
    These stubs are called in the guest VM, so using kmalloc() -- which
    returns some arbitrary address inside host physical memory -- to
    allocate them is not a real good idea ;-)   I've moved them to memory
    inside the module's data pages.

  - Add the actual PDBR reloading.

  [patch2 from UW]
  - The do_nothing task is removed from the kernel module.
    Instead, it is built as standalone executable 'virtcode.bin',
    which gets loaded by the user app into the guest memory.

  - The kernel module, on initial switch to the monitor, starts
    executing the guest code in ring-3.

  - A nasty problem with the mmap() implementation appeared:
    It seems that Linux doesn't like if we map normal memory
    pages using remap_page_range; apparently you are only allowed
    to map 'hardware' pages.  While it did seem to work, what
    actually happened was that remap_page_range recognized the
    page as normal memory page, and decided to map in an anonymous
    page instead.  Thus, the guest could access the mapped memory
    perfectly well --- the only problem is that is accessed completely
    different pages, which were mapped in by the standard nopage
    handler :-/

    This patch circumvent the problem by a very ugly hack:  it
    simply sets the PG_reserved bit of all those pages used for
    guest memory, thereby declaring them as 'hardware range' ;-)
    This works fine on my system (2.0.36), but I'm not at all sure
    that it will work on other versions ...

    Anyway, this needs to be fixed correctly, of course.  I'll be
    looking into this problem.  For now, we need the hack, because
    otherwise the user app is unable to load the guest code in ...

fmw-19990822b:

  - (KPL) Converted to using GNU autoconf for easier configuration
    on multiple host OS platforms.  You only have to run autoconf
    if you change the configure.in file.  Otherwise, re-run
    configure.  To make things simple, and so that it's easy
    to run configure with known flags etc, run ./configure.hostos,
    where hostos is your host OS.  I made one for Linux and
    BeOS, though the BeOS one may be broken.

  - (KPL) Integrated patch from Ulrich regarding looking for
    signals in kernel module before looping on running guest OS.

  - (KPL) Moved sti() call in host-linux.c into soft_int macro.
    The IF change will not be recognised until after the INT
    instruction, in this case, so there is no chance of an
    interrupt intervening.

  - (KPL) Created a NEED_RESCHED macro, so we can compile
    for various Linux kernel versions.  I don't know when the
    change was made to put need_resched in the "current" structure.
    I guessed 2.2.0.  If this is not right, please let us know
    when the change occurred.

  - (KPL) added inline function pending_signals() to host-linux.c.
    Please also fix the version number code that controls which
    method to use to access the signals field in the "current"
    structure.  This was a small fix to get Ramon's changes
    to compile on my 2.2.5 kernel.  I'm not sure if the signals
    > 31 are interesting to us.  If not, we don't need the loop
    code I put in the inline function.
    

fmw-19990821a:

  - (RvH) I cleaned up all of the code, giving it bigger indents
    and converting to C-style comments whenever things got really
    ugly.  It's still not the clearest code I've ever seen but it'll
    do for now.  I also moved the virtualised test code into a
    separate file, virtcode.c.

  - (RvH) I changed host-linux.c so that it takes the amount of
    quanta to run as an argument to ioctl().  Now it doesn't trap
    back to the user application on ever quantum.  Question:
    on 2.2 we use current->need_resched, but is this true for
    2.0 too, or does it have need_resched as a global variable ?

  - (RvH) I integrated Ulrich's mmap() code into the module, and
    fixed it so that it compiles on 2.2 kernels.

  - (RvH) I changed the way interrupts are allocated.  Now, the
    host code needs to call init_reflect() (in monitor.c) after
    init_monitor() for every interrupt that needs to be reflected
    back to the host OS.  The monitor.c code then dynamically
    generates and relocates a correct interrupt stub.  The host
    code is responsible for allocating/freeing memory for the
    stubs.

  - (RvH) I fixed up the interrupts so that the code works on my
    SMP machine.  I still have 2.0 code in there too, but I can't
    test that --- can anybody test this on a 2.0 SMP box ?
    2.2 SMP is rather annoying because it allocates lots of IPIs,
    and most of them aren't fixed (2.2 SMP does not use the 0x20-
    0x2f mapping for the PIC, but has everything go through the
    I/O APIC with rather weird interrupt mappings).  In order to
    allocate the correct interrupts I need access to a kernel
    array called irq_vector[], which unfortunately is not exported
    to modules.  Thus you need to patch the linux kernel if you
    want to run this code on a 2.2 SMP box.  The patch is simple:
    put the following code in linux/arch/i386/kernel/i386_ksyms.c:

        #ifdef __SMP__
        #include "irq.h"
        EXPORT_SYMBOL(irq_vector);
        #endif

    and everything will work just beautifully (or anyway, that's
    what it looks like.)

fmw-990817c:

  - (KPL) I added irq16 vector to the init_monitor() call.
    If you pass a 0, no irq16 redirect handler is installed.
    Irq16 is the APIC IPI interrupt, which is used on multiple
    processor compiles of Linux (SMP).  The code is untested.
    In kernel/include/host-linux.h, I do this:

    #ifdef __SMP__
    #  define IRQ16_BASE_VECTOR 0x30  // IPI vector
    #else
    #  define IRQ16_BASE_VECTOR 0     // IPI not used on non-SMP
    #endif

    Let me know if this makes this code work on your SMP
    Linux machine.

  - (KPL) Fixed the previous interrupt bug, thanks to a find
    by Ulrich and Ramon.

fmw-990817b:
  - (KPL) Initial release.  Has interrupt bug.