These changes are based on the SF patch #742670. It did not work with
plugins disabled.
Here is the descriiption of patch #742670:
the library dependencies (see objdump -p file) in the
main bochs binary and the gui plugins are wrong:
- the gui plugins don't depend on their respective gui
libraries
- the main bochs binary depends on all the gui libraries
this doesn't seem much of a problem, since the needed
libraries are loaded anyway, but it's problematic when
building packages for binary distribution of bochs.
in Debian, the package building tools will set package
dependencies according to the library dependencies of
each file, so the result is:
- the main bochs package depends on all libraries. this
includes libwx, libsdl, libncurses and xlibs.
- the indivudual plugin packages don't depend on any
gui library
so when a user wants to run bochs with the curses gui
and installs the bochs-curses package, he/she will be
forced to install Xfree86, WxWindows and GTK+ on
his/her system.
the attached patch fixes the problem by moving the
linker flags related to GUI libraries to the right
makefile (gui/Makefile.in)
in text mode. The new variables 'text_rows' and 'text_cols' are used in the
text_update() function.
- optimizations in function UpdateScreen()
- desriptions of the functions text_update() and dimension_update() updated
It's not clear what the correct behavior is in that case, (we
clearly don't currently handle it correctly) so
simply avoiding it is the easiest thing to do. As such, this
option is ON by default.
* renamed variables textres_x and textres_y to text_cols and text_rows
* use text_rows in function text_update() instead of the function argument
(unused for now - should be removed)
* fixed a warning (bochs_icon_bits)
cursor enable/disable support implemented
- function dimension_update() now calculates the number of visible rows and
columns in text mode
- draw a border if the physical screen is larger than the bochs text screen
- initialize color pairs only when color support is present
- german umlaut characters added in function get_term_char()
- function get_term_char(): unsupported characters are now replaced by spaces
and implemented in SDL
* sequencer controller register 0x01 bit 0 controls the width of the characters.
This value is used to calculate the screen width.
* attribute controller register 0x10 bit 2 controls the appearance of graphics
characters (ASCII 0xC0 - 0cDF). A change of this value forces a charmap update
to rebuild the font bitmaps.
* the SDL display library uses the new feature described above
* the other display libraries recalculate the screen width, since they are using
a fixed font width of 8 for now.
- VGA: attribute controller register 0x10 bit 2 (enable_line_graphics) does not
switch the palatte in CGA mode
check.
Commented out a number of instances of invalidate_prefetch_q(),
for branches which do not change CS since the EIP window mechanism
takes care of validating that EIP lands in the current page or not
in the main cpu loop anyways.
Fixed a couple cases (v8086 mode and real mode) of loading CS where
the EIP page window was not invalidated in segment_ctrl_pro.cc.
That may fix some aliasing problems reported before (OS2).
- line offset handling simplified
- calculate textres_x and textres_y after changing the font dimensions
- unnecessary size check in function dimension_update() removed
ATTENTION: the Elpin VGA BIOS uses a delay of 280 vertical retraces after
displaying the copyright text
- changes to the horizontal or vertical pel panning registers force a redraw of
the screen
- screen update problems in text mode fixed
- sort order of the CRT register write cases changed
- fixed calculation of the byte offset for standard VGA modes if line compare
is active
- force a resize of the screen after a change of the display mode (text/graphics)
- mem_write(): skip the rotation of the cpu byte if 'data_rotate' is 0
You can now choose for each disk on the ata interfaces the
disk mode, between :
- flat : one file flat layout
- concat : multiple files layout
- external : developer's specific, through a C++ class
- dll : developer's specific, through a DLL
- sparse : stackable, commitable, rollbackable
- vmware3 : vmware3 disk support
- undoable : flat file with commitable redolog
- growable : growing file
- volatile : flat file with volatile redolog
- z-undoable : gziped flat file with commitable redolog
- z-volatile : gziped flat file with volatile redolog
A new "bxcommit" utility can merge commitable redologs to
flat images.
Notes from the author:
Here is another one of my speed up patches. Unlike my previous speedups
this one will help more platforms than just X86. It cleans up the Data
Xfer instructions. Since the Data Xfer instructions are the most often
executed instructions it gives a noticable boost in speed. The basic
optimization technique was to eliminate intermediate variables and pass
a pointer to the final destination or original source to the
read_virtual_whatever and the write_virtual_whatever functions.
If set, the screen size will be horizontally doubled in CGA and standard
VGA modes.
- doublescan feature for CGA mode 320x200x4 added
- removed trailing spaces in VBE update code
the address offset between two lines on the screen in all VGA modes.
It depends on the CRTC offset register value and the address mode (byte, word,
dword)
- moved screen update code to the end of the write handler. If a register change
should force a redraw of the screen, the variable 'needs_update' must be set
to 1.
- changes to the attribute controller palette index register now force a redraw
of the screen
- modeX: calculation of the tile numbers fixed
* check memory mapping before everything else
* read mode 1 optimized using a part of patch.vga-mode2-speed-dohzono
- function mem_write(): check memory mapping before everything else
- writes to the CRT registers are handled only if the new value differs
- a CRT start address change forces a redraw of the screen
with 100 rows and a char height of 4
- raster operations AND, OR and XOR in write mode 2 implemented (part of patch #707931)
- use the vga_tile_updated array in modeX like other modes do
- small optimizations in the graphics update code
I've got a game that uses OPL2 functions. It's still not
working yet, but this patch brings it closer. Using Jeffrey S.
Lee's "Programming the AdLib/Sound Blaster FM Music Chips (Ver
2.0)"[1] as a reference, I found an inversion of the OPL timer
masks in the sb16 code. Without this patch, the emulator uses
the wrong bits for masking, and completely botches the flags
register (note the bitwise AND vs OR).
double width)
- cleanup in modeX emulation similar to mode 13h
- consider start address in function mem_write for VGA modes
- register 0x03c8 is readable, too
- clear tile array when switching to graphics mode, clear text snapshot when
switching to text mode, do the same when changing the palette or disable video
- simplified the function determine_screen_dimensions()
- fixed the code for the CGA mode 640x200x2 in update() and mem_write()
According to the Intel manuals:
The LOCK prefix can be prepended only to the following instructions
and only to those forms of the instructions where the destination
operand is a memory operand: ADD, ADC, AND, BTC, BTR, BTS, CMPXCHG,
CMPXCH8B, DEC, INC, NEG, NOT, OR, SBB, SUB, XOR, XADD, and XCHG. If
the LOCK prefix is used with one of these instructions and the source
operand is a memory operand, an undefined opcode exception (#UD) will
be generated. An undefined opcode exception will also be generated if
the LOCK prefix is used with any instruction not in the above list.
Checking of the LOCK prefix done in fetchDecode state and not overloads
Bochs's execution.
- it works only on x86 with gcc2.95+
- uses the GCC function atribute "regparm(n)" to declare that certain
functions use the register calling convention
- performance improvement is about 6%
1) fixed the type of "hostPageAddr" and associated typecasts.
2) fixed the type of "pages" and associated typecasts (overloaded variable)
3) patch to cpu.cc to calculate "eipPageBias" correctly in 64 bit mode
1) fixed some errors running 32 bit compat mode. IMPORTANT FIX.
2) added IST processing (uses IST1-IST7 in 64 bit TSS)
3) cosmetic - debugging stuff to console.
a minor optimization. Also in transition from compat mode to 64 bit mode (e.g. interrupt to inner
privelege with mode change), SS may not be properly defined - this avoids other messiness.
DEV_kbd_gen_scancode() is called instead of DEV_kbd_put_scancode(). This
change enables the support for different scancode sets. The win32 scancodes
are now converted to bx_key values.
* renamed CPU_ID to BX_CPU_ID.
with this new name there is no possibility for name contentions and BX_CPU_ID
definition could be moved out to NEED_CPU_REG_SHORTCUTS block
* returned back `unsigned BX_CPU::which_cpu(void)` function
* added BX_CPU_ID parameter for
BX_INSTR_PHY_READ(a20addr, len);
BX_INSTR_PHY_WRITE(a20addr, len);
now it will be
BX_INSTR_PHY_READ(cpu_id, a20addr, len);
BX_INSTR_PHY_WRITE(cpu_id, a20addr, len);
It uses special gcc attribute regparm, available on gcc3.2 on x86.
I get about 7% increase when booting windows 95.
The patch still need some integration work before being included
in the main code.
- function userbutton_handler(): stop parsing the shortcut string if the
key name is invalid
- function userbutton_handler(): incrementing variable 'len' simplified
- function snapshot_checker(): fixed a warning
> CPU_ID is defined as
> #define CPU_ID (BX_CPU_THIS_PTR local_apic.get_id())
> This is not true when the APIC name is changed (true in Linux). Please
> change this to:
> #define CPU_ID (BX_CPU_THIS - BX_CPU(0))
and improved by Christophe Bothamy and Volker Ruppert)
Comment from the author:
Attached is a "patch" file detailing what you need to do
to add USB support (UHCI only for now) to your existing
Bochs (2.0.xx) source code.
I use Win32 and VC++ but the source and modifications
should be platform and compiler independant.
Please let me know if this patch some how breaks the
build process of your compilation (Bochs 2.0.0 or above
only).
I would also like any feedback on how this code works
(or doesn't work) on your platform and within your
images. If you explain in as much detail as you can on
how it did or did not work, I will try to add to the source
to help it work on all platforms and images.
- added "ioaddr" and "irq" to the "usb1" bochsrc option
- number of ports is limited to 2 for a USB UHCI root hub
- the PCI configuration space is now a part of the usb hub structure
- new symbol BX_USE_PCIVGA_SMF for the experimental PCI VGA device
- pcivga.cc: removed unnecessary include statement
- pcivga.cc: fixed two warnings
- pcivga.h: removed unnecessary symbol BX_IODEV_PCIVGA_H
From the author:
I've implemented functions ALOb(), ObAL(), YbAL(),
ALXb() and eAXXv() from dis_groups.cc.
I've also changed dis_decode.cc to print cmpsd if code
is 32 bits and cmpsw if code is 16 bits. The same with
stosd, lodsd and scasd.
