changes.txt

我师兄自己写的操作系统

  > boundary.  The example that came up a few times is reading/writing 4 bytes
  > starting at 0xffffffff when the segment limit was 0xffffffff.  The
  > condition used to compare offset+length-1 with the limit, but
  > offset+length-1 was overflowing so the comparison went wrong.
- cmpxchg8b patch from Michael Hohmuth <[e-mail]>
- apply patch from Thomas Fitzsimmons <[e-mail]> to fix compile
  problems when BX_SUPPORT_PAGING and BX_USE_TLB are turned off
- fix bug introduced in 1.2.1 which caused spurious exceptions.
  See patch #439314, Exception 1 (debug) on HALT, from
  [e-mail].
- add panic in ctrl_xfer32.cc where the IRET32 implementation is broken.
  This only happens if you are NOT in vm8086 mode or protected mode.
  The intent is to warn people when they are getting bad emulation, and
  encourage people to report how they got to that point.
- apply patch from Santiago Bazerque.  See this bug report:
  [ #463018 ] retf not removing parameters sometimes
- fix bug [ #461730 ] IRETD causes problems if NT-flag is set
  reported by Peter Lammich.
- apply patch [ #455014 ] CR0 bug in 80486, described as:
  > In the register CR0, when the bit PM is enabled, the bit 4 is 0
  > when should be 1.
- apply patch from Mike Rieker <[e-mail]> associated with this bug
  report: [ #480422 ] gdt 'accessed' bit
- in task_switch when it tried to ensure that the old TSS was paged in,
  it actually used the new TSS address, fixed.
- updated the instrumentation code, and added a working example.  To try
  it, configure --enable-instrumentation=instrument/example1.  Then when
  you run bochs, you will get one line for each instruction PC and for
  each I/O access in a new file called bxevent.txt.
- set a bit in the CMOS that says the processor has an FPU.  This is
  from patch [ #455006 ] Device byte is not initialized aptly.
  Author did not leave their name.
- add logging code to the "null ethernet" which does not require host OS
  support.  All this does is print the outgoing packets from the guest OS.
- cleanup of log functions (Todd Fries)
- add BX_ERROR for every command in ATAPI-6 that bochs does not support.
  I still need to do add some commands from older specs that are obsolete
  (and not listed) in ATAPI-6.  Commands that aren't in the spec will still
  panic.
- only put 0xf into the 2nd hard disk field when the cdrom is not present.
  This is a patch from Volker Ruppert <[e-mail]>, who
  comments: "The fdisk command reports an unusable second harddisk if the cdrom
  is enabled. This patch helps, but I don't know if it is the right way."
- make hard disk code return error codes when data is not available instead
  of just panicing.  In particular, if the logical sector is out of bounds
  or the disk image cannot be read/written at the desired offset, we now
  abort the ATA command and return an error code.  Many of the old BX_PANIC
  messages are turned to BX_ERROR, so they will still appear in the
  log, but now the device model will try to communicate this fact to
  the OS instead of simply giving up.
- don't blindly reject odd length atapi commands.  There are cases when
  it's really ok according to ATA-4.
- for big endian machines, reversed the bit fields in interrupt_reason.
  This was pointed out by Nicholai Benalal.
- extended keyboard improvements by Dave Poirier
- major mouse patch from Dave Spring, that implements several missing 
  mouse modes.
- commit keyboard patch from David Haslam <[e-mail]>
  posted to mailing list, that addresses the problem of each key press
  printing ^@.  See cvs log for details.
- mouse performance fixes by Greg Alexander and Robb Main
- NE2000 fixes by Frode Vatvedt Fjeld, [e-mail], Greg Alexander,
  and [e-mail].
- fix bug [ #468340 ] pic:slave: OCW3 not implemented.  Some event handling
  code appeared in the master pic but not the slave pic.
- fix compile problems in SB16 code, related to fpos_t being treated as 
  an integer.
- patch from Volker Ruppert <[e-mail]> to fix
  midi output file so that winamp can play it.
- some cleanup of serial code by Todd Fries and Volker Ruppert, but it
  doesn't work yet.

X Windows specific:
- commit patch from David Haslam <[e-mail]>
  [ #455763 ] Cursor trail with DOS Edit/Minix vi
- error for missing fonts now points to the documentation
- new option --enable-idle-hack, by Roland Mainz, which makes Bochs more
  friendly toward other processes when its CPU is idle.  Presently, 
  this option is specific to X windows.

Win32 specific:
- now Windows 95/98/ME can read the physical cdrom (Don Becker)
- The default configuration for Win32 VC++, given in .conf.win32-vcpp,
  now enables the NE2000 and renames all .cc files to .cpp.  This keeps VC++
  happy but may make it hard to use CVS.
- The default configuration for Cygwin, given in .conf.win32-cygwin, now
  enables cdrom and SB16.
- See "new docs" on the web site for compile instructions for VC++ and Cygwin.
- The sources include a VC++ workspace, in addition to the old "nmake"
  makefile.
- ethernet support (emulated NE2000 card), coded by Don Becker.  This
  implementation requires a library called WinPCap, which you can 
  download from http://netgroup-serv.polito.it/winpcap.
- new utility called niclist.exe which lists the ID number of all your network
  cards (well probably you just have one).  The ID be used when setting up your
  .bochsrc.
- patch [ #466403 ] make text colors more accurate.  The author did not leave
  his/her name.
- fix GUI bug [ #452159 ] win32: mouse stuck if bochs win partly off screen
  Now we center the mouse periodically, whether or not the mouse has
  wandered outside of the window or not.
- event handler recognizes the extended keycode flag
- fixes for raw floppy and floppy disk images (Don Becker)

Linux specific:
- Ethernet (emulated NE2000 card) now works in Linux!  Contributed by
  [e-mail].  This has been tested using host OS kernel 2.2.14, and
  works with telnet, ftp, irc, lynx, etc.  Because it is a packet filter
  solution, you aren't able to talk to the host machine, only to other 
  machines on the network.  
- The default configuration for Linux, given in .conf.linux, now enables
  the NE2000 model.
- RPM build process configures with --prefix=/usr so that everything is
  installed in /usr/bochs instead of /usr/local/bochs.
- DLX Linux disk image is now installed so that only root can write it, to
  avoid security problems.  When you run the bochs-dlx script, it creates a
  local copy in your home directory and then runs it.
- code that determines the capacity of a cdrom now works for both ATAPI
  and SCSI drives ([e-mail])
- applied patch from [e-mail].  The comments are:
  > The Linux 2.4.5 CD-ROM driver sends a READ_DISC_INFO command which caused
  > an "unrecognized ATAPI command" panic.  Looks like READ_DISC_INFO is only
  > recognized by CD-R and CD-RW drives, so I ignore it for now.

Amiga MorphOS specific:
- Bochs now compiles and works on Amiga MorphOS.  Configure with
  --with-amigaos.  For AmigaOS only, see .bochsrc for use of fullscreeen and
  screenmode options.  The Amiga MorphsOS is written and maintained by
  Nicholai Benalal <[e-mail]>.
- raw cdrom supported if you configure with --enable-cdrom

BeOS specific:
- Bochs compiles and works on BeOS.  Configure with --with-beos.
  Bernd Thorsten Korz <[e-mail]> maintains the BeOS port.
- raw cdrom supported if you configure with --enable-cdrom

MacOS X specific:
- Bochs now compiles and works on MacOS X.  Configure with --with-carbon.
  Emmanuel Mailliard <[e-mail]> ported the Macintosh code to the
  Carbon API.
- The MacOS X application is built using (gasp) mkdir, copy, and rez.
  Surely this is not the right way, but it works.
- raw cdrom supported if you configure with --enable-cdrom

RFB mode:
- apply patch.rfb-mouse by MURANAKA Masaki ([e-mail])
  see this source forge bug [ #457968 ] Strange mouse motion on RFB
- add a retry loop in RFB code, so that if port 5900 is not available
  it can try 5901, etc.

Bochs Debugger:
- do a vga update whenever you print a debugger prompt.
- added debugger command "info fpu" that prints the FPU registers.  If you
  do "info all" you get cpu and fpu registers.
- added debugger command "info ne2k" which prints all the registers
  of the NE2000 model
- add ability to do register tracing and flag tracing (Dave Poirier).
  Try the trace-reg-on and trace-reg-off commands.
- instruction trace now includes time ticks
- fixed problems in which bochs compiled with debugger measured time 
  differently from bochs compiled without debugger.  Also when instruction
  trace was enabled, breakpoints and control-C did not work.  Also,
  breakpoints at the beginning of an interrupt handler did not work.

-------------------------------------------------------------------------
Changes in 1.2.1 (June 12, 2001):
- more work on makefile for building RPMs
- [ #432382 ] build debian packages patch
  add build/debian directory from Rob Lemley <[e-mail]>
  which allows us to make Debian packages!
- optimize for speed when simulating one processor.  Now 1-processor
  performance should be equivalent to 1.1.2.
- [ #425640 ] sb16 assumes fpos_t is long int
  This fixes compiles of iodev/sb16.cc on linux systems with newer libraries
  in which fpos_t is not an integer.
- [ #432488 ] SMP:assert "n_logfn < MAX_LOGFNS" fails
  increase MAX_LOGFNS since we ran out of them on an SMP simulation with
  4 processors
- changes to compile clean on cygwin:
  - don't use the WIN32 snprintf define for cygwin
  - add ssize_t definition for cygwin
  - only compile "struct timeval tval" if select is available
    on that platform.
- [ #432491 ] SMP: CPUID says no APIC feature
  clean up inconsistent use of BX_SUPPORT_APIC and BX_APIC_SUPPORT, which
  caused the CPUID to report no APIC was present
- [ #431025 ] --enable-external-device-models broken
  removed configure options for external-device-models and 
  external-cpu-memory.  These don't work and aren't going to be fixed.
- [ #429448 ] configure: -lreadline when not there
  Now configure allows you to choose not to use readline, even if it's found
  on your system.
- [ #428915 ] apply extended keyboard patch
  extended keyboard patch by Dave Poirier <[e-mail]>
- [ #428626 ] if no X11 found, configure&make fails
  Now configure halts if X windows is selected but no X libraries are found.
- updated rombios to version 1.13.  This fixes several problems:
  - [ #430472 ] DOS HIMEM "A20 line" error
    This problem was apparantly caused when Bryce added a function that prints
    the BIOS version, and he called it too early in the boot process.  Now the
    same function is called later, and it doesn't break the A20.
  - [ #431010 ] SMP structure overwritten in v1.2
    SMP structures were getting overwritten by BCC-generated data, 
    preventing SMP operating systems from detecting that other processors
    were available.
  - [ #431016 ] bios: SMP struct has wrong entry count
    SMP structure had the wrong entry counts
- very minor doc updates (typos, replace broken link to mtools info)
- quit when the user clicks the power button, even if they have disabled
  panics.
- win32 now defaults to having mouse capture mode turned off.  For new users,
  it would be distressing for their mouse cursor to disappear until they
  pressed F12.
- [ #428222 ] vga font not installed
  added script called "install-x11-fonts" which should help people install
  the VGA font on X windows systems, if it isn't already there.

Changes in 1.2 (June 3, 2001):
- [ #427259 ] rombios HALT calls don't print
  Fixed bios/rombios.c HALT macro so that it writes the line number of the
  panic to the PANIC_PORT (port 0x400) and then does NOT do a halt 
  instruction.  Also changed iodev/unmapped.cc so that the line number written
  to PANIC_PORT is displayed as a BX_PANIC message.  Because the HALT
  macro now triggers the normal panic behavior, it can be controlled by
  the bochsrc.
- [ #429016 ] crash if no hard drive
  rombios used to call HALT macro if no hard drive was found. Now it only
  calls HALT if a hard drive has an illegal geometry.