version dosemu-1.4.0
This document is the preliminary draft of a manual to help people understand the inner workings of dosemu. It is the goal of this document to create new dosemu hackers. This concept was inspired by the linux kernel hackers guide.
This Guide was concieved and originally written by "Corey Sweeney" <corey@interaccess.com>. It has been completely revised. It is now generated automatically directly from the source code. Special thanks to "James B. MacLean" <macleajb@ednet.ns.ca> for supplying the original information. (It was mostly ripped out of a mail message.) "Jochen Hein" has made many useful comments & suggestions.
At the end if this document is a section detailing how this guide is put together. This may help you when trying to locate the relevant pieces of code. If you add new code, it would be useful if the relevant markers are added where appropriate.
This file is a collective effort. If you don't like one of the explanations, or want to add anything, please send me something!
These files are used to start DOSEMU as well as hold globally called functions and global vars.
These are the functions defined in dos.c.
Arguments are:
argc - Count of argumnents.
argc - Actual arguments.
These are the functions defined in emu.c.
call the emulate function by way of the dll headers. Always make sure that this line is the first of emu.c and link emu.o as the first object file to the lib
Arguments are:
argc - Argument count.
argv - Arguments.
DOSEMU must not work within the 1 meg DOS limit, so start of code is loaded at a higher address, at some time this could conflict with other shared libs. If DOSEMU is compiled statically (without shared libs), and org instruction is used to provide the jump above 1 meg.
The `vm86_struct` is used to pass all the necessary status/registers to DOSEMU when running in vm86 mode.
-----
DOSEMU keeps system wide configuration status in a structure called config.
-----
The var `fatalerr` can be given a true value at any time to have DOSEMU exit on the next return from vm86 mode.
-----
The var 'running_DosC' is set by the DosC kernel and is used to handle some things differently, e.g. the redirector. It interfaces via INTe6,0xDC (DOS_HELPER_DOSC), but only if running_DosC is !=0. At the very startup DosC issues a INTe6,0xdcDC to set running_DosC with the contents of BX (which is the internal DosC version).
These files are used for initialization and runtime configuration of DOSEMU
These are the functions defined in base/init/init.c.
Setup the dosemu amazing custom BIOS, quietly overwriting anything was copied there before. Do not overwrite graphic fonts!
Set up all memory areas as would be present on a typical i86 during the boot phase.
Calls all initialization routines for devices (keyboard, video, serial, disks, etc.)
These are the functions defined in base/init/config.c.
Process user CPU override from the config file ('cpu xxx') or from the command line. Returns the selected CPU identifier or -1 on error.
register a function Enforce consistency upon the `config` structure after all values have been set to remove silly option combinations
Enforce consistency upon the `config` structure after all values have been set to remove silly option combinations
For simpler support of X, DOSEMU can be started by a symbolic link called `xdos` which DOSEMU will use to switch into X-mode.
DPMI is Lutz's Baby. It's a really important part of the Emulator as far as we are concerned, since it will allow us to run so many more programs and, most importantly, bcc. This is the one thing that the WINE developers want that we haven't been able to give them.
If you think you can help .... "Away you Go!" (Sorry to those non-UK folks ... Thats a reference to a UK kids sports programme from my youth ... anyway ... enough of this banter. You'll be wanting to know that this is all about DPMI ...)
These are the functions defined in dosext/dpmi/dpmi.c.
This function is similar to the vm86() syscall in the kernel and switches to dpmi code.
This routine is used for running protected mode hardware interrupts. run_pm_int() switches to the locked protected mode stack and calls the handler. If no handler is installed the real mode interrupt routine is called.
This routine is used for reflecting the software interrupts 0x1c, 0x23 and 0x24 to protected mode.
This is the default CPU exception handler. Exceptions 0, 1, 2, 3, 4, 5 and 7 are reflected to real mode. All other exceptions are terminating the client (and may be dosemu too :-)).
This routine switches to the locked protected mode stack, disables interrupts and calls the DPMI client exception handler. If no handler is installed the default handler is called.
This is the brain of DPMI. All CPU exceptions are first reflected (from the signal handlers) to this code.
Exception from nonprivileged instructions INT XX, STI, CLI, HLT and from WINDOWS 3.1 are handled here.
All here unhandled exceptions are reflected to do_cpu_exception()
Note for cpu-emu: exceptions generated from the emulator are handled here. 'Real' system exceptions (e.g. from an emulator fault) are redirected to emu_dpmi_fault() in fullemu mode
We are caching ldt here for speed reasons and for Windows 3.1. I would love to have an readonly ldt-alias (located in the first 16MByte for use with 16-Bit descriptors (WIN-LDT)). This is on my wish list for the kernel hackers (Linus mainly) :-))))))).
-----
DPMI is designed such that the stack change needs a task switch. We are doing it via an SIGSEGV - instead of one task switch we have now four :-(. Arrgh this is the point where I should start to include DPMI stuff in the kernel, but then we could include the rest of dosemu too. Would Linus love this? I don't :-((((. Anyway I would love to see first a working DPMI port, maybe we will later (with version 0.9 or similar :-)) start with it to get a really fast dos emulator...............
NOTE: Using DIRECT_DPMI_CONTEXT_SWITCH we avoid these 4 taskswitches actually doing 0. We don't need a 'physical' taskswitch (not having different TSS for us and DPMI), we only need a complete register (context) replacement. For back-switching, however, we need the sigcontext technique, so we build a proper sigcontext structure even for 'hand made taskswitch'. (Hans Lermen, June 1996)
-- the whole emu_stack_frame could be eliminated except for eip/rip and esp/rsp. For the most part GCC can worry about clobbered registers (Bart Oldeman, October 2006)
dpmi_control is called only from dpmi_run when in_dpmi_dos_int==0
-----
Hopefully the below LAR can serve as a replacement for the KERNEL_LDT, which we are abandoning now. Especially the 'accessed-bit' will get updated in the ldt-cache with the code below. Most DPMI-clients fortunately _are_ using LAR also to get this info, however, some do not. Some of those which do _not_, at least use the DPMI-GetDescriptor function, so this may solve the problem. (Hans Lermen, July 1996)
-----
Here we handle all prefixes prior switching to the appropriate routines The exception CS:EIP will point to the first prefix that effects the the faulting instruction, hence, 0x65 0x66 is same as 0x66 0x65. So we collect all prefixes and remember them. - Hans Lermen
We shouldn't return to dosemu code if IF=0, but it helps - WHY? */
All of the Video handling code is in the "video" subdirectory.
There is one file for each video card or chipset and the master file. To Add a new card, it needs a set of save & restore routines putting in a file here.
These are the functions defined in env/video/video.c.
Set pointer to correct structure of functions to initialize, close, etc... video routines.
Here the sleeping lion will be awoken and eat much of CPU time !!!
The result of setting VM86_SCREEN_BITMAP (at state of Linux 1.1.56): Each vm86 call will set 32 pages of video mem RD-only (there may be 1000000 per second) Write access to RD-only page results in page-fault (mm/memory.c), which will set a bit in current->screen_bitmap and calls do_wp_page() which does __get_free_page(GFP_KERNEL) but frees it immediatly, because copy-on-write is not neccessary and sets RD/WR for the page. (this could happen 32000000 per second, if the CPU were fast enough) It would be better to get the DIRTY-bit directly from the page table, isn't it? A special syscall in emumodule could do this.
-----
reserve_video_memory()
This procedure is trying to eke out all the UMB blocks possible to maximize your memory under DOSEMU. If you know about dual monitor setups, you can contribute by putting in the correct graphics page address values.
These are the functions defined in plugin/X/X.c.
called by mouse.c to hide/display the mouse and set it's position. This is currently the only callback from mouse.c to X.
Allocate a colormap for graphics modes and initialize it. Do mostly nothing on true color displays. Otherwise, do: - if colormaps have less than 256 entries (notably 16 or 2 colors), don't use a private colormap - if a shared map is requested and there are less than 36 colors (3/4/3) available, use a private colormap
Note: Text modes always use the screen's default colormap.
This is the interface function called by the video subsystem to set a video mode.
NOTE: The actual mode is taken from the global variable "video_mode".
Set the video mode. If mode_class is -1, this will only reinitialize the current mode. The other arguments are ignored in this case.
DO NOT REMOVE THIS TEST!!! It is magic, without it EMS fails on my machine under X. Perhaps someday when we don't use a buggy /proc/self/mem.. -- EB 18 May 1998 A slightly further look says it's not the test so much as suppressing noop resize events... -- EB 1 June 1998
These are the functions defined in env/video/vgaemu.c.
Emulates writes to VGA ports. This is a hardware emulation function.
Arguments are:
port - The port being written to.
value - The value written,
Emulates reads from VGA ports. This is a hardware emulation function.
Arguments are:
port - The port being read from.
vga_emu_fault() is used to catch video access, and handle it. This function is called from arch/linux/async/sigsegv.c::dosemu_fault1(). The sigcontext_struct is defined in include/cpu.h. Now it catches only changes in a 4K page, but maybe it is useful to catch each video access. The problem when you do that is, you have to simulate each instruction which could write to the video memory. It is easy to get the place where the exception happens (scp->cr2), but what are those changes? An other problem is, it could eat a lot of time, but it does now also.
MODIFICATION: VGA mode 12h under X is supported in exactly the way that was suggested above. Not every instruction needs to be simulated in order to make this feature useful, just the ones used to access video RAM by key applications (Borland BGI, Protel, etc.).
MODIFICATION: all VGA modes now work and almost all instructions are simulated.
Arguments are:
scp - A pointer to a struct sigcontext_struct holding some relevant data.
vga_emu_init() must be called before using the VGAEmu functions. It is only called from env/video/X.c::X_init() at the moment. This function basically initializes the global variable `vga' and allocates the VGA memory.
It does in particular *not* map any memory into the range 0xa0000 - 0xc0000, this is done as part of a VGA mode switch.
There should be an accompanying vga_emu_done().
Arguments are:
vedt - Pointer to struct describing the type of display we are actually
attached to.
vga_emu_update() scans the VGA memory for dirty (= written to since last update) pages and returns the changed area in *veut. See the definition of vga_emu_update_type in env/video/vgaemu_inside.h for details.
You will need to call this function repeatedly until it returns 0 to grab all changes. You can specify an upper limit for the size of the area that will be returned using `veut->max_max_len' and `veut->max_len'. See the example in env/video/X.c how this works.
If the return value of vga_emu_update() is >= 0, it is the number of changed pages, -1 means there are still changed pages but the maximum update chunk size (`veut->max_max_len') was exceeded.
This function does in its current form not work for Hercules modes; it does, however work for text modes, although this feature is currently not used.
Arguments are:
veut - A pointer to a vga_emu_update_type object holding all relevant info.
vgaemu_switch_plane() maps the specified plane.
This function returns True on success and False on error, usually indicating an invalid bank number.
Arguments are:
plane (0..3) - The plane to map.
vga_emu_switch_bank() is used to emulate video-bankswitching.
This function returns True on success and False on error, usually indicating an invalid bank number.
Arguments are:
bank - The bank to switch to.
Searches a video mode with the requested mode number.
The search starts with the mode *after* the mode `vmi' points to. If `vmi' == NULL, starts at the beginning of the internal mode table. `mode' may be a standard VGA mode number (0 ... 0x7f) or a VESA mode number (>= 0x100). The mode number may have its don't-clear-bit (bit 7 or bit 15) or its use-lfb-bit (bit 14) set. The special mode number -1 will match any mode and may be used to scan through the whole table.
Returns NULL if no mode was found and a pointer into the mode table otherwise. The returned pointer is a suitable argument for subsequent calls to this function.
You should (and can) access the mode table only through this function.
Arguments are:
mode - video mode.
vmi - pointer into internal mode list
Set a video mode.
Switches to `mode' with text sizes `width' and `height' or (if no such mode was found) at least `width' and `height'.
Arguments are:
mode - The new video mode.
width - Number of text columns.
height - Number of text rows.
Sets the text mode resolution. Typically called after a font change.
Arguments are:
width - Number of text columns.
height - Number of text rows.
Checks DAC and Attribute Controller to find all colors with changed RGB-values. Returns number of changed colors. Note: the list _must_ be large enough, that is, have at least min(256, (1 << vga.pixel_size)) entries!
Arguments are:
de - list of DAC entries to store changed colors in
These are the functions defined in env/video/vesa.c.
Initializes the VGA/VBE BIOS and the VBE support.
Arguments are:
vedt - Pointer to struct describing the type of display we are actually
attached to.
This is the VESA interrupt handler.
It is called from base/bios/int10.c::int10(). The VESA interrupt is called with 0x4f in AH and the function number (0x00 ... 0x10) in AL.
These are the functions defined in env/video/attremu.c.
Directly reads the Attribute Controller's registers. This is an interface function.
Directly sets the Attribute Controller's registers. This is an interface function.
Emulates reads from the attribute controller. This is a hardware emulation function.
Emulates writes to attribute controller combined index and data register. Read VGADOC for details. This is a hardware emulation function.
Returns the current index of the attribute controller. This is a hardware emulation function, though in fact this function is undefined in a real attribute controller. Well, it is exactly what my VGA board (S3) does. -- sw This is a hardware emulation function.
These are the functions defined in env/video/dacemu.c.
Initializes the DAC. It depends on a correct value in vga.pixel_size. This function should be called during VGA mode initialization. This is an interface function.
Sets the DAC width. Typical values are 6 or 8 bits. In theory, we support other values as well (untested). This is an interface function.
Returns a complete DAC entry (r, g, b). Don't forget to set DAC_entry.index first! This is an interface function.
Converts a DAC register's RGB values to gray scale. This is an interface function.
Specifies which palette entry is read. This is a hardware emulation function.
Specifies which palette entry is written. This is a hardware emulation function.
Read a value from the DAC. Each read will cycle through the registers for red, green and blue. After a ``blue read'' the read index will be incremented. Read VGADOC4 if you want to know more about the DAC. This is a hardware emulation function.
Write a value to the DAC. Each write will cycle through the registers for red, green and blue. After a ``blue write'' the write index will be incremented. This is a hardware emulation function.
Returns the current PEL mask. Note that changed_vga_colors() already applies the PEL mask; so applications should not worry too much about it. This is a hardware emulation function.
Sets the PEL mask and marks all DAC entries as dirty. This is a hardware emulation function.
These are the functions defined in env/video/dualmon.c.
Initializes the monochrome card. First detects which monochrome card is used, because the Hercules RamFont and the Hercules InColor need one more register to be initialized. If there is no monochrome card at all, we just think there is one and poke an peek in the void. After the detection the card is initialized.
returns: nothing
Arguments are:
none
After MDA_init() the VGA is configured, something in video.c or console.c "reprograms" the monochrome card again in such a way that I always have to run hgc.com before I can use any program that uses the monochrome card. I've spent a day trying to find it, but I can't figure out. Something is writing to one of the following ports: 0x3b4, 0x3b5, 0x3b8, 0x3b9, 0x3ba, 0x3bb, 0x3bf. The problem occurs at (at least) the following 2 systems:
- AMD 386DX40, Trident 9000/512Kb ISA, Hercules Graphics Card Plus - Intel 486DX2/66, Cirrus Logic 5426/1Mb VLB, Hercules clone
The problem doesn't occur when I start dosemu from a telnet connection or from a VT100 terminal. (Erik Mouw, jakmouw@et.tudelft.nl)
These are the functions defined in env/video/vgaemu.c.
Emulates writes to VGA ports. This is a hardware emulation function.
Arguments are:
port - The port being written to.
value - The value written,
Emulates reads from VGA ports. This is a hardware emulation function.
Arguments are:
port - The port being read from.
vga_emu_fault() is used to catch video access, and handle it. This function is called from arch/linux/async/sigsegv.c::dosemu_fault1(). The sigcontext_struct is defined in include/cpu.h. Now it catches only changes in a 4K page, but maybe it is useful to catch each video access. The problem when you do that is, you have to simulate each instruction which could write to the video memory. It is easy to get the place where the exception happens (scp->cr2), but what are those changes? An other problem is, it could eat a lot of time, but it does now also.
MODIFICATION: VGA mode 12h under X is supported in exactly the way that was suggested above. Not every instruction needs to be simulated in order to make this feature useful, just the ones used to access video RAM by key applications (Borland BGI, Protel, etc.).
MODIFICATION: all VGA modes now work and almost all instructions are simulated.
Arguments are:
scp - A pointer to a struct sigcontext_struct holding some relevant data.
vga_emu_init() must be called before using the VGAEmu functions. It is only called from env/video/X.c::X_init() at the moment. This function basically initializes the global variable `vga' and allocates the VGA memory.
It does in particular *not* map any memory into the range 0xa0000 - 0xc0000, this is done as part of a VGA mode switch.
There should be an accompanying vga_emu_done().
Arguments are:
vedt - Pointer to struct describing the type of display we are actually
attached to.
vga_emu_update() scans the VGA memory for dirty (= written to since last update) pages and returns the changed area in *veut. See the definition of vga_emu_update_type in env/video/vgaemu_inside.h for details.
You will need to call this function repeatedly until it returns 0 to grab all changes. You can specify an upper limit for the size of the area that will be returned using `veut->max_max_len' and `veut->max_len'. See the example in env/video/X.c how this works.
If the return value of vga_emu_update() is >= 0, it is the number of changed pages, -1 means there are still changed pages but the maximum update chunk size (`veut->max_max_len') was exceeded.
This function does in its current form not work for Hercules modes; it does, however work for text modes, although this feature is currently not used.
Arguments are:
veut - A pointer to a vga_emu_update_type object holding all relevant info.
vgaemu_switch_plane() maps the specified plane.
This function returns True on success and False on error, usually indicating an invalid bank number.
Arguments are:
plane (0..3) - The plane to map.
vga_emu_switch_bank() is used to emulate video-bankswitching.
This function returns True on success and False on error, usually indicating an invalid bank number.
Arguments are:
bank - The bank to switch to.
Searches a video mode with the requested mode number.
The search starts with the mode *after* the mode `vmi' points to. If `vmi' == NULL, starts at the beginning of the internal mode table. `mode' may be a standard VGA mode number (0 ... 0x7f) or a VESA mode number (>= 0x100). The mode number may have its don't-clear-bit (bit 7 or bit 15) or its use-lfb-bit (bit 14) set. The special mode number -1 will match any mode and may be used to scan through the whole table.
Returns NULL if no mode was found and a pointer into the mode table otherwise. The returned pointer is a suitable argument for subsequent calls to this function.
You should (and can) access the mode table only through this function.
Arguments are:
mode - video mode.
vmi - pointer into internal mode list
Set a video mode.
Switches to `mode' with text sizes `width' and `height' or (if no such mode was found) at least `width' and `height'.
Arguments are:
mode - The new video mode.
width - Number of text columns.
height - Number of text rows.
Sets the text mode resolution. Typically called after a font change.
Arguments are:
width - Number of text columns.
height - Number of text rows.
Checks DAC and Attribute Controller to find all colors with changed RGB-values. Returns number of changed colors. Note: the list _must_ be large enough, that is, have at least min(256, (1 << vga.pixel_size)) entries!
Arguments are:
de - list of DAC entries to store changed colors in
These are the functions defined in env/video/instremu.c.
instr_sim is used to simulate instructions that access the VGA video memory in planar modes when using X as the video output device.
It is necessary to do this in order to simulate the effects of the hardware VGA controller in X mode.
If the return value is 0, it means the instruction was not one that for which a simulation is provided. The return value is 1 for success, but the function exits because we need to go back to the DOSEMU's main loop or count runs out.
Arguments are:
x86: the structure holding everything about the cpu-state we need.
instr_emu is the main interface to instr_sim. It puts the processor state in the x86 structure.
Arguments are:
scp - A pointer to a struct sigcontext_struct holding some relevant data.
pmode - flags protected mode
cnt - number of instructions to be simulated
Most of the New Keyboard handling code is in the "plugin/kbd_unicode" subdirectory.
These are the functions defined in plugin/kbd_unicode/serv_xlat.c.
The task of compute_keynum() is to 'collect' keyboard bytes (e.g. 0xe0 prefixes) until it thinks it has assembled an entire keyboard event. The entire keyboard event is then returned, otherwise NUM_VOID is returned.
translate_key takes a keysym event and calculates the appropriate bios translation.
As a side effect translate_key updates the apropriate pieces of state to reflect the current keyboard state.
Calling translate_key twice on the same data is likely to be hazardous.
This function sends a raw keycode byte, e.g. read directly from the hardware, to DOS. It is both queued for the port60h emulation and processed for the BIOS keyboard buffer, using the national translation tables etc.
For DOS applications using int16h we will therefore not have to load KEYB.EXE, others (e.g. games) need their own drivers anyway.
This function is used if we are at the console and config.rawkeyboard=on.
This does all the work of sending a key event to DOS. Either pressing a key releasing one. The key to move is the key specified by keynum.
keynum - the keynum from keynum.h indicating a physical key make - TRUE for key press, FALSE for release
Applications using int16h will always see the appropriate ASCII code for the given keyboard key and the current keyboard state. All the chracter translation is done for you to keep from reporting inconsistent key events.
An emulated hardware scancode is also sent to port60h.
Note that you have to send both MAKE (press) and BREAK (release) events. If no BREAK codes are available (e.g. terminal mode), send them immediately after the MAKE codes.
Allows peeking into the keytables. This returns the keynum a given keysym sits on.
This does all the work of sending a key event to DOS. Either pressing a key releasing one. The key to move is the key that is labeled with the specified keysym.
key - the keysym, one of the KEY_ constants from new-kbd.h make - TRUE for key press, FALSE for release
Applications using int16h will always see the appropriate ASCII code for the given keyboard key and the current keyboard state. All the chracter translation is done for you to keep from reporting inconsistent key events.
An emulated hardware scancode is also sent to port60h.
Note that you have to send both MAKE (press) and BREAK (release) events. If no BREAK codes are available (e.g. terminal mode), send them immediately after the MAKE codes.
This does all the work of sending a key event to DOS. sym -- The unicode value of the symbol you want to send
Applications using int16h will always see the symbol passed here, if it is representable in the current dos character set. The appropriate scancodes are generated automatically to keep the keyboard code consistent.
An emulated hardware scancode is also sent to port60h.
Note that you have to send both MAKE (press) and BREAK (release) events. If no BREAK codes are available (e.g. terminal mode), send them immediately after the MAKE codes.
This does all the work of sending a key event to DOS. sym -- The unicode value of the symbol you want to send modifiers -- modifiers like alt etc you what to change your symbol with.
This function is a concession to the reality, in which key events are a composed of active modifiers, and a key label.
This function behaves as put_symbol does, except before pressing the key it adds the specified modifiers to the modifiers it would normally use.
For cases where the symbol can only be created by an alt# combination or by pressing a dead key (Basically any case where more than one key is requried, after setting the shiftstate) it gives up and just sends the symbol.
Note that you have to send both MAKE (press) and BREAK (release) events. If no BREAK codes are available (e.g. terminal mode), send them immediately after the MAKE codes.
This simply reads the keyboard server's shift state.
This is intended to be used in conjunction with set_shiftstate to sync up a shiftstate with a source of key events.
With the addition of this function the keyboard inteface is clean enough so if needed a completly different translation engine can be dropped in to support a totally different environment (windows or whatever).
This simply sets the keyboard server's shift state.
If there are shiftstate bits you want to keep fixed simply grab them with get_shiftstate, before calling this function.
This changes the keyboard flags by generating the appropriate shift key make/break codes that normally come along with such changes. So this function should be safe in any context.
Note also that you can't simply write to the shiftstate variable instead of using this function.
These are the functions defined in plugin/kbd_unicode/keyb_clients.c.
Figures out which keyboard client to use and initialises it.
First it calls the probe method to see if it should use the client, Then it call init to set that client up.
If probe or init fails it trys another client.
Eventually falling back to Keyboard_none a dummy client, which does nothing.
These are the functions defined in plugin/kbd_unicode/keyb_none.c.
Succeed if we can run the dummy keyboard client, (we always can). but first try the other fall-back (slang keyboard)
These are the functions defined in plugin/kbd_unicode/keyb_raw.c.
These are the functions defined in plugin/term/keyb_slang.c.
Initialize the keyboard in pc scancode mode. This functionality is ideal but rarely supported on a terminal.
Set the terminal back to a keyboard mode other programs can understand.
These are the remaining important files, that do not really fit into another group. These should not be dismissed as unimportant - rather, they are often amongst the most important.
These are the functions defined in base/async/int.c.
int 0x1A call
This has (among other things) the calls that DOS makes to get/set its sense of time. On booting, DOS gets the RTC time and date with AH=2 and AH=4, after that it should use AH=0 calls to read the 'tick' counter from BIOS memory. Each time this crosses midnight, a flag is set that DOS uses to increment its date.
Here we can now change the 'view' of time so the calls either return BIOS tick (most DOS like), read the PIT counter (avoids INT-8 changes) or gets LINUX time (most accurate for long term NTP-adjusted time keeping).
int0x21 call
we trap this for two functions: simulating the EMMXXXX0 device and fudging the CONFIG.XXX and AUTOEXEC.XXX bootup files.
note that the emulation herein may cause problems with programs that like to take control of certain int 21h functions, or that change functions that the true int 21h functions use. An example of the latter is ANSI.SYS, which changes int 10h, and int 21h uses int 10h. for the moment, ANSI.SYS won't work anyway, so it's no problem.
This function runs the specified caller function in response to an int instruction. Where i is the interrupt function to execute.
revect specifies whether we call a non-revectored leaf interrupt function or a "watcher" that sits in between: the leaf interrupt function is called if cs:ip is at f000:i*10 or if (the int vector points there and the int is labelled non-revectored) otherwise the non-leaf interrupt function is called, which may chain through to the real interrupt function (if it returns 0)
This function runs the instruction with the following model _CS:_IP is the address to start executing at after the caller function terminates, and _EFLAGS are the flags to use after termination. For the simple case of an int instruction this is easy. _CS:_IP = retCS:retIP and _FLAGS = retFLAGS as well equally the current values (retIP = curIP +2 technically).
However if the function is called (from dos) by simulating an int instruction (something that is common with chained interrupt vectors) _CS:_IP = BIOS_SEG:HLT_OFF(i) and _FLAGS = curFLAGS while retCS, retIP, and retFlags are on the stack. These I pop and place in the appropriate registers.
This functions actions certainly correct for functions executing an int/iret discipline. And almost certianly correct for functions executing an int/retf#2 discipline (with flag changes), as an iret is always possilbe. However functions like dos int 0x25 and 0x26 executing with a int/retf will not be handled correctlty by this function and if you need to handle them inside dosemu use a halt handler instead.
Finally there is a possible trouble spot lurking in this code. Interrupts are only implicitly disabled when it calls the caller function, so if for some reason the main loop would be entered before the caller function returns wrong code may execute if the retFLAGS have interrupts enabled!
This is only a real handicap for sequences of dosemu code execute for long periods of time as we try to improve timer response and prevent signal queue overflows! -- EB 10 March 1997
Grumble do to code that executes before interrupts, and the semantics of default_interupt, I can't implement this function as I would like. In the tricky case of being called from dos by simulating an int instruction, I must leave retCS, retIP, on the stack. But I can safely read retFlags so I do. I pop retCS, and retIP just before returning to dos, as well as dropping the stack slot that held retFlags.
This improves consistency of interrupt handling, but not quite as much as if I could handle it the way I would like. -- EB 30 Nov 1997
Trying to get it right now -- BO 25 Jan 2003
This function returns 1 if it's completely finished (no need to run real_run_int()), otherwise 0.
SETUP_INTERRUPTS is used to initialize the interrupt_function array which directs handling of interrupts in protected mode and also initializes the base vector for interrupts in real mode.
Setup initial interrupts which can be revectored so that the kernel does not need to return to DOSEMU if such an interrupt occurs.
Many video BIOSes use hi interrupt vector locations as scratchpad area - this is because they come before DOS and feel safe to do it. But we are initializing vectors before video, so this only causes trouble. I assume no video BIOS will ever: - change vectors < 0xe0 (0:380-0:3ff area) - change anything in the vector area _after_ installation - AV
These are the functions defined in arch/linux/async/sigsegv.c.
All CPU exceptions (except 13=general_protection from V86 mode, which is directly scanned by the kernel) are handled here.
Prints information about an exception: exception number, error code, address, reason, etc.
These are the functions defined in arch/linux/async/signal.c.
Arguments are:
sig - the signal to have a handler installed to.
fun - the signal handler function to install
The IRQ numbers to monitor are taken from config.sillyint, each bit corresponding to one IRQ. The higher 16 bit are defining the use of SIGIO
Initialize the signals to have NONE being blocked. Currently this is NOT of much use to DOSEMU.
Due to signals happening at any time, the actual work to be done because a signal occurs is done here in a serial fashion.
The concept, should this eventualy work, is that a signal should only flag that it has occurred and let DOSEMU deal with it in an orderly fashion as it executes the rest of it's code.
Arguments are:
context - signal context to save.
signal_call - signal handling routine to be called.
Whenever I/O occurs on devices allowing SIGIO to occur, DOSEMU will be flagged to run this call which inturn checks which fd(s) was set and execute the proper routine to get the I/O from that device.
We assume system call restarting... under linux 0.99pl8 and earlier, this was the default. SA_RESTART was defined in 0.99pl8 to explicitly request restarting (and thus does nothing). However, if this ever changes, I want to be safe
-----
Check for keyboard coming from client For now, first byte is interrupt requests from Client
These are the functions defined in base/dev/misc/timers.c.
Every time we get a TIMER signal from Linux, this procedure is called. It checks to see if we should queue a timer interrupt based on the current values.
do_sound handles the _emulated_ mode pc-speaker emulation.
As far as I can determine all cases of the pc-speaker are now emulated. But I am not sure where Rainer Zimmerman got his (pit[2].mode == 2) || (pit[2].mode == 3) test in the original implementation, it doesn't seem to cause problems though.
The implementation of speaker_on & speaker_off can be found in src/base/speaker.c
Major Changes from version written by Rainter Zimmerman.
o Added support for programs that control the directly through bit 1 of port61.
o Added a generic interface to allow multiple speaker backends.
o Implemented X speaker code so the emulated speaker now works in X.
--EB 21 September 1997
This is experimental TIMER-IRQ CHAIN code! This is a function to determine whether it is time to invoke a new timer irq 0 event. Normally it is 18 times a second, but many video games set it to 100 times per second or more. Since the kernel cannot keep an accurate timer interrupt, the job of this routine is to perform a chained timer irq 0 right after the previous timer irq 0. This routine should, ideally, be called right after the end of a timer irq, if possible.
This would speed up high frequency timer interrupts if this code can be converted into an assembly macro equivalent!
PLEASE NOTE
This code has been replaced by interrupt scheduling code in pic. The result is that we simply call pic_sched and run the dos interrupt. If the new code causes no problems, I'll revise this section permanently.
These are the functions defined in base/misc/dos2linux.c.
Runs a command and prints the (stdout and stderr) output on the dosemu screen.
Return values mean:
Arguments are:
buffer - string with command to execute
These are the functions defined in base/misc/ioctl.c.
Arguments are:
fd - File handle to add to select statment
want_sigio - want SIGIO (1) if it's available, or not (0).
Arguments are:
fd - File handle to remove from select statment.
used_sigio - used SIGIO (1) if it's available, or not (0).
These are the functions defined in base/dev/misc/pci.c.
Use standard 32-bit (type 1) access method to read PCI configuration space data
These are the functions defined in base/dev/misc/joystick.c.
Tells DOSEMU whether or not it is time to update its internal status of the joystick (for nonblocking reads only).
DOS programs read/poll from the joystick port hundreds of thousands of times per second so the idea is that we really don't need to read from Linux for every such query (increasing performance by about 40%) because:
1. humans are incapable of changing the status of the joystick (moving, pressing buttons) more than about 10 times per second
2. no one will not notice a delay in DOS registering the joystick status (if it is in the order of a few milliseconds)
Of course, this means that you should not set joy_latency in dosemu.conf to more than 1000/(#times I can press a button/move joy per second * 2), unless you want DOSEMU to miss quick axis/button changes and want to wait a ridiculous amount of time before DOSEMU registers any changes at all.
Update the button status for each joystick.
We must perform "button mapping" if only the first joystick is enabled i.e. we are required to map the "excessive" buttons (>2) from the first joystick onto the second:
a) 3rd button of 1st joy --> 1st button of 2nd joy
b) 4th button of 1st joy --> 2nd button of 2nd joy
Update the axis status for each joystick.
We must perform "axis mapping" if only the first joystick is enabled i.e. we are required to map the "excessive" axes (>2) from the first joystick onto the second:
a) 3rd axis of 1st joy --> 2st axis of 2nd joy
b) 4th axis of 1st joy --> 1st axis of 2nd joy (yes, these are reversed deliberately because it's what happens in DOS)
Convert a Linux joystick axis reading to a DOS one by making use of the differences in the allowable range of axis values.
NOTE: I don't know whether or not Linux returns exponential values for the joystick but (apparently) DOS programs expect the values to be exponential and so if this is to be fixed, it should probably be done in this function.
Update global joystick status variables given a Linux joystick event.
Process the event queue for _both_ linux joysticks using nonblocking reads with the new joystick API (joy_driver_new).
This should be done before (well, actually, not before _every_ single read -- see joy_latency_over()) the joystick status is returned to DOS as all Linux joystick events are queued until they are processed and we want to return a reasonably current state of the joystick -- not what it was a long time ago. _Both_ joysticks are processed here because of axis/button mapping affecting the status of both emulated joysticks (what DOS sees).
Read both the current position and current button status of the joystick from Linux (joy_driver_old).
Eventually called from DOS to get the button status of the joysticks. The threaded version will simply get the status from global variables. The unthreaded versions will perform non-blocking reads.
Eventually called from DOS to get the axis status of the joysticks. The threaded version will simply get the values from global variables. The unthreaded versions will perform non-blocking reads.
Arguments are:
invalid_val - value to return to signify a non-existent axis
update - whether DOSEMU should update its internal axis values from Linux
(for each read of the joystick position, set this flag _only_
on the first of the 4 calls to this function unless you want
the axis positions to be from different points in time :)
This is the int15 function 0x84 handler (i.e. BIOS joystick emulation), called from src/base/async/int.c.
The real BIOS actually reads its values straight from the joystick port (0x201) but we don't bother because we can do it faster :)
Because of this, it returns the joystick axis values with the same range as port 0x201 BUT the range for a real BIOS varies between computers as it is dependant on how it reads from the port (hopefully this won't cause any problems).
This function emulates reads from the joystick port (0x201) -- this is the most common way of detecting and reading the joystick.
The real implementation of this port sets a bit for each axis for a certain period of time, corresponding to analog measurements of the position of each axis (so "if you count the analog values in software, a faster machine yields different values from a slow machine [unless you use a timer]" - DOS 6: A Developer's Guide).
In contrast, this implementation sets the bits high for a certain number of port reads, corresponding to the position of each axis (independent of time). This means that, for most programs, the axis range will be that specified in dosemu.conf (which is rather convenient) and avoids the issue of super-fast computers causing DOS program axis counters to overflow (e.g. in a real system, if the program used an 8-bit variable for storing the position of an axis and the system was fast enough to read from the port more than 127 or 255 times, there would be trouble).
We make a runtime decision based on the detected joystick API version and #ifdef USE_PTHREADS, on the way in which we obtain the joystick status from Linux (a "driver"):
1. joy_driver_nojoy: simply tells DOS programs that you have no joystick(s)
2. joy_driver_old: uses old, non-blocking joystick API (<1.0.0); limited to 2 axes; supported because DOSEMU supports old kernels
3. joy_driver_new: uses new, non-blocking joystick API (>=1.0.0); a (little) slower than joy_driver_new_threaded
4. joy_driver_new_threaded: uses new, BLOCKING joystick API (>=1.0.0); efficient but requires pthreads (which is known to make DOSEMU unstable!)
The same driver is used for both joysticks.
-----
if the 2nd joystick is enabled, we ignore any button >= 2 regardless of which joystick it is (if it's the 1st, the 2nd joystick would overwrite its buttons; if it's the 2nd, it would be out of range)
-----
if the 2nd joystick is enabled, we ignore any axis >= 2 regardless of which joystick it is (if it's the 1st, the 2nd joystick would overwrite its axes; if it's the 2nd, it would be out of range)
-----
Apparently, the Carry Flag is cleared if int15 function 0x84 exists and it is set if it doesn't exist.
But what does this mean? Does the existence of such a BIOS function depend on the existence of a Game Card/SoundBlaster, or does it just mean that there is such an implemented BIOS function, regardless of whether or not you have a joystick?
I have never seen a real BIOS set the Carry Flag on this call, even on a computer without a joystick -- so to mimick what happens in the real world, I just clear the Carry Flag regardless of whether the user has a joystick or not. This could be incorrect behaviour so it may have to be changed in the future.
-----
Here we set bits based on joystick axis counters. The code here is particularly tricky and if you try to change it, you will probably break it :)
-----
Here we read the button status from Linux (programs can read the button status from the port, _without_ making a dummy write to the port first so the Linux read must be done _here_) and return it.
does this code work for ports other than 0x201?
-----
joy_reset() is called immediately after joy_init(), which is rather inconvenient (anyone heard of a port_unregister_handler()?) so we don't bother resetting at all but in the future this could cause problems
-----
perhaps we should not have been called to start with?
The Helper Interrupt uses the following groups:
0x00 - Check for DOSEMU 0x01-0x11 - Initialisation functions & Debugging 0x12 - Set hogthreshold (aka garrot?) 0x20 - MFS functions 0x21-0x22 - EMS functions 0x28 - Garrot Functions for use with the mouse 0x29 - Serial functions 0x30 - Whether to use the BOOTDISK predicate 0x33 - Mouse Functions 0x40 - CD-ROM functions 0x50-0x5f - DOSEMU/Linux communications 50 -- run unix command in ES:DX 51,52? 53 -- do system(ES:DX) 54 -- get CPU speed 55 -- get terminal type 0x60-0x6f - reserved for plug-ins 0x7a - IPX functions 0x8x -- utility functions 0x80 -- getcwd(ES:DX, size AX) 0x81 -- chdir(ES:DX) 0xdc - helper for DosC kernel 0xfe - called from our MBR, emulate MBR-code. 0xff - Terminate DOSEMU
There are (as yet) no guidelines on choosing areas for new functions.
These files all relate to Intel-x86 specific code.
These are the functions defined in emu-i386/cpu.c.
process opcodes 0F xx xx trapped by GP_fault returns 1 if handled, 0 otherwise Main difference with previous version: bits in our pseudo-control regs can now be written. This should make CPU detection pgms happy.
Setup initial interrupts which can be revectored so that the kernel does not need to return to DOSEMU if such an interrupt occurs.
These are the functions defined in emu-i386/ports.c.
Handles/simulates an inw() port IO read. Usually this invokes port_inb() twice, but it may be necessary to do full word i/o for some video boards.
I don't know what to do of this stuff... it was added incrementally to port.c and has mainly to do with video code. This is not the right place for it... Anyway, this implements some HGC stuff for X and the emuretrace port access for 0x3c0/0x3da
Resets all the port port_handler information. This must be called before parsing the config file - This must NOT be called again when warm booting! Can't use debug logging, it is called too early.
Catch all the special cases previously defined in ports.c mainly video stuff that should be moved away from here This must be called at the end of initialization phase
NOTE: the order in which these inits are done could be significant! I tried to keep it the same it was in ports.c but this code surely can still have bugs
Assigns a handle in the port table to a range of ports with or without a device, and registers the ports
The following port_{in|out}{bwd} functions are the main entry points to the port code. They look into the port_handle_table and call the appropriate code, usually the std_port_ functions, but each device is free to register its own functions which in turn will call std_port or directly access I/O (like video code does), or emulate it - AV
-----
optimized versions for rep - basically we avoid changing privileges and iopl on and off lots of times. We are safe letting iopl=3 here since we don't exit from this code until finished. This code is shared between VM86 and DPMI.
-----
This is the core of the new emuretrace algorithm: If a read of port 0x3da is performed we just set it as pending and set ioperm OFF for port 0x3c0 When a write to port 0x3c0 is then trapped, we perform any pending read to 0x3da and reset the ioperm for 0x3c0 in the default ON state. This way we avoid extra port accesses when the program is only looking for the sync bits, and we don't miss the case where the read to 0x3da is used to reset the index/data flipflop for port 0x3c0. Futher accesses to port 0x3c0 are handled at full speed.
-----
find out whether the port address request is available; this way, try to deny uncoordinated access
If it is not listed in /proc/ioports, register them (we need some syscall to do so bo 960609)... (we have a module to do so AV 970813) if it is registered, we need the name of a device to open if we can't open it, we disallow access to that port
-----
We need to check if our required port range is in use by some device. So we look into proc/ioports to check the addresses. Fine, but at this point we must supply a device name ourselves, and we can't check from here if it's the right one. The device is then open and left open until dosemu ends; for the rest, in the original code the device wasn't used, just locked, and only then port access was granted.
This stuff should be moved to video code!!
-----
we should free the name but we are going to exit anyway
These are the functions defined in emu-i386/do_vm86.c.
All from the kernel unhandled general protection faults from V86 mode are handled here. This are mainly port IO and the HLT instruction.
Here is where DOSEMU runs VM86 mode with the vm86() call which also has the registers that it will be called with. It will stop vm86 mode for many reasons, like trying to execute an interrupt, doing port I/O to ports not opened for I/O, etc ...
Here we collect all stuff, that has to be executed within _one_ pass (step) of a loop containing run_vm86().
Here we handle all prefixes prior switching to the appropriate routines The exception CS:EIP will point to the first prefix that effects the faulting instruction, hence, 0x65 0x66 is same as 0x66 0x65. So we collect all prefixes and remember them. - Hans Lermen
These are the functions defined in emu-i386/cputime.c.
GETcpuTIME is a pointer to a function which returns the relative CPU time. Different methods of getting the time can then be implemented, currently there are two using gettimeofday() for 486 and TSC for pentium
GETusTIME returns the DOS time with 1-usec resolution using GETcpuTIME to get the implementation-dependent CPU time. The 'sc' parameter is unused.
GETtickTIME returns the DOS time with 838ns resolution using GETcpuTIME to get the implementation-dependent CPU time. The 'sc' parameter is unused.
At the heart of the timing system in dosemu >= 0.67.11 is the availability of the system time as a 64-bit [type hitimer_t] monoton value. (a 64-bit timer on a 200MHz CPU increments by 2^48 a day).
Dosemu needs this time under two resolutions:
- a MICROSECOND resolution for general timing purposes
- a TICK(838ns) resolution for PIT |
On non-pentium machines, only the first one is available via the kernel call gettimeofday(). On the pentium and up, the situation is better since we have a cheap hi-res timer on-chip, and worse since this timer runs at a speed depending from the CPU clock (which we need to know/measure, and could be not 100% accurate esp. if the speed is a non-integer multiple of 33.3333).
dosemu >= 0.67.11 can use both timing methods (call them 486 and pentium), and switch between them in a dynamic way when configuring.
At the first level (local to the file cputime.c) there are the RAW timer functions, addressed by RAWcpuTIME(). These get the actual absolute CPU time in usecs.
At the second level, GETcpuTIME() returns the relative, zero-based system time. This is where the 486/pentium switch happens.
The third level is the actual timer interface for dosemu and is made of two functions:
- GETusTIME(s) gives the time in usecs
- GETtickTIME(s) gives the time in ticks |
The 's' parameter can be used to control secondary time functions like 'time stretching' (see the READMEs). The function GETusSYSTIME() never activates this stretching, and is used only by the realtime thread-based 1-sec timer in rtc.c.
All timing are RELATIVE to a base. The use of a based time allows us to play more freely with time, e.g. stop and restart it during debugging, stretch it, make it go at different speeds between real-time and CPU emulation, etc. The base has been chosen to be zero, because it will avoid overflows in calculations, produce more readable and more easily comparable debug log files, and also because only int0x1a and BIOS timer require knowledge of the actual time, PIT and PIC are not sensitive.
These are the functions defined in emu-i386/simx86/sigsegv.c.
All CPU exceptions (except 13=general_protection from V86 mode, which is directly scanned by the kernel) are handled here.
This is the code that works our serial emulation. This needs to be very fast if we are to convince DOS that we have a very fast serial port.
Extensions to serial debugging.
SER_DEBUG_MAIN (0 or 1) - extra debug output on the most critical information.
SER_DEBUG_HEAVY (0 or 1) - super-heavy extra debug output, including all ports reads and writes, and every character received and transmitted!
SER_DEBUG_INTERRUPT (0 or 1) - additional debug output related to serial interrupt code, including flagging serial interrupts, or PIC-driven code.
SER_DEBUG_FOSSIL_RW (0 or 1) - heavy FOSSIL debug output, including all reads and writes.
SER_DEBUG_FOSSIL_STATUS (0 or 1) - super-heavy FOSSIL debug output, including all status checks.
You must recompile dosemu everytime one of these constants are modified. Just type 'make' in the dosemu dir and it will recompile the changes only.
-----
IMPORTANT INFO about com[] variable array structure used in serial.c
Most of the serial variables are stored in the com[] array. The com[] array is a structure in itself. Take a look at the 'serial_t' struct declaration in the serial.h module for more info about this. Only the most commonly referenced global variables are listed here:
config.num_ser Number of serial ports active. com[x].base_port The base port address of emulated serial port. com[x].real_comport The COM port number. com[x].interrupt The PIC interrupt level (based on IRQ number) com[x].mouse Flag mouse (to enable extended features) com[x].fd File descriptor for port device com[x].dev[] Filename of port port device com[x].dev_locked Flag whether device has been locked
The arbritary example variable 'x' in com[x] can have a minimum value of 0 and a maximum value of (config.numser - 1). There can be no gaps for the value 'x', even though gaps between actual COM ports are permitted. It is strongly noted that the 'x' does not equal the COM port number. This example code illustrates the fact, and how the com[] array works:
for (i = 0; i < config.numser; i++) s_printf("COM port number %d has a base address of %x", com[i].real_comport, com[i].base_port);
These are the functions defined in base/serial/ser_init.c.
This is the master serial initialization function that is called upon startup of DOSEMU to initialize ALL the emulated UARTs for all configured serial ports. The UART is initialized via the initialize_uart function, which opens the serial ports and defines variables for the specific UART.
If the port is a mouse, the port is only initialized when i
This needs more work before it is implemented into /etc/dosemu.conf as an 'rtscts' option.
These are the functions defined in base/serial/ser_ports.c.
The following function returns a value from an I/O port. The port is an I/O address such as 0x3F8 (the base port address of COM1). There are 8 I/O addresses for each serial port which ranges from the base port (ie 0x3F8) to the base port plus seven (ie 0x3FF). [num = abritary port number for serial line, address = I/O port address]
The following function writes a value to an I/O port. The port is an I/O address such as 0x3F8 (the base port address of COM1). [num = abritary port number for serial line, address = I/O port address, val = value to write to I/O port address]
Should clearing UART cause THRE int if it's enabled? */
-----
Fix the calculation assumption
-----
Is this safe to put this here? */
-----
Is this safe to put this here? */
These are the functions defined in base/serial/ser_irq.c.
This function is the serial interrupts scheduler. Its purpose is to update interrupt status and/or invoke a requested serial interrupt. If interrupts are not enabled, the Interrupt Identification Register is still updated and the function returns. See pic_serial_run() below it is executed right at the instant the interrupt is actually invoked.
Since it is not possible to run the interrupt on the spot, it triggers the interrupt via the pic_request() function (which is in pic.c) and sets a flag that an interrupt is going to be occur soon.
Please read pic_serial_run() for more information about interrupts. [num = port, int_requested = the requested serial interrupt]
This function is called by the priority iunterrupt controller when a serial interrupt occurs. It executes the highest priority serial interrupt for that port. (Priority order is: RLSI, RDI, THRI, MSI)
Because it is theoretically possible for things to change between the interrupt trigger and the actual interrupt, some checks must be repeated.
This is the main housekeeping function, which should be called about 20 to 100 times per second. The more frequent, the better, up to a certain point. However, it should be self-compensating if it executes 10 times or even 1000 times per second. Serial performance increases with frequency of execution of serial_run.
Serial mouse performance becomes more smooth if the time between calls to serial_run are smaller.
Linux code hackers: How do I detect a break signal without having to rely on Linux signals? Can I peek a 'break state bit'? Also, how do I 'turn on' and 'turn off' the break state, via an ioctl() or tcsetattr(), rather than using POSIX tcsendbrk()?
how do we cancel a PIC interrupt, when we have come this far?
These are the functions defined in base/serial/int14.c.
The following function executes a BIOS interrupt 0x14 function. This code by Mark Rejhon replaced some very buggy, old int14 interface a while back. These routines are not flawless since it does not wait for a character during receive, and this may confuse some programs.
If any of you coders are ambitious, try thinking of the following: - Converting this into inline assembler and use direct port access
Why does a RX_BUFFER_SIZE of 256 cause slower performance than a size of 128?
All of the Mouse handling code is in the "mouse" subdirectory.
There are only 2 main files, mouse.c and mouseint.c.
I have not properly tested this INT74 - JES 96/10/20 I have removed it. INT74 is irq 12. Which I suppose is the proper irq for a ps2 mouse. It appears initial support was planned to support irq 12 and at Mouse_ROUTINE_OFF is a routine that acknowledges an irq. That routine is probably what should be acknowledging irq12, and what int 0x74 should point to. I have disabled int0x74 support for now. --EB 29 Oct 1997 Got it working --BO 4 Nov 2004
-----
Whoever wrote the dos mouse driver spec was brain dead... For some video modes the mouse driver appears to randomly pick a shift factor, possibly to keep at least a 640x200 resolution.
The general programming documentation doesn't make this clear. And says that in text modes it is safe to divide the resolution by 8 to get the coordinates in characters.
The only safe way to handle the mouse driver is to call function 0x26 Get max x & max y coordinates and scale whatever the driver returns yourself.
To handle programs written by programmers who weren't so cautious a doctrine of least suprise has been implemented.
As much as possible do the same as a standard dos mouse driver in the original vga modes 0,1,2,3,4,5,6,7,13,14,15,16,17,18,19.
For other text modes allow the divide by 8 technique to work. For other graphics modes return x & y in screen coordinates. Except when those modes are either 40x?? or 320x??? and then handle the x resolution as in 40x25 and 320x200 modes.
320x200 modes are slightly controversial as I have indications that not all mouse drivers do the same thing. So I have taken the simplest, and most common route, which is also long standing dosemu practice of always shifting the xaxis by 1. When I researched this I could find no examples that did otherwise.
-- Eric Biederman 19 August 1998
This code has now been updated so it defaults as above but allows work arounds if necessary. Because tweaking dosemu is easier than fixing applications without source.
-- Eric Biederman 29 May 2000
All of the Bios code is in the "bios" subdirectory.
DOSEMU requires certain code to be coded in assembler and also code to be located in the F000 segment. This is where all such code should be put.
All of the PIC handling code is in the "PIC" subdirectory.
These are the functions defined in base/dev/pic/pic.c.
This is the pic debug message printer. It writes out some basic information, followed by an informative message. The basic information consists of: interrupt nesting counter change flag (+, -, or blank) interrupt nesting count (pic_icount) interrupt level change flag (+, -, or blank) current interrupt level interrupt in-service register interrupt mask register interrupt request register message part one decimal data value message part two
If the message part 2 pointer is a null pointer, then only message part one (without the data value) is printed.
The change flags are there to facilitate grepping for changes in pic_ilevel and pic_icount
To avoid line wrap, the first seven values are printed without labels. Instead, a header line is printed every 15 messages.
write_pic_0() and write_pic1() implement dos writes to the pic ports. They are called by the code that emulates inb and outb instructions. Each function implements both ports for the pic: pic0 is on ports 0x20 and 0x21; pic1 is on ports 0xa0 and 0xa1. These functions take
Arguments are:
read_pic0 and read_pic1 return the values for the interrupt mask register (port 1), or either the in service register or interrupt request register, as determined by the last OCW3 command (port 0). These functions take a single parameter, which is a port number (0 or 1). They are called by code that emulates the inb instruction.
pic_seti is used to initialize an interrupt for dosemu. It requires four parameters. The first parameter is the interrupt level, which may select the NMI, any of the IRQs, or any of the 16 extra levels (16 - 31). The second parameter is the dosemu function to be called when the interrupt is activated. This function should call do_irq() if the DOS interrupt is really to be activated. If there is no special dosemu code to call, the second parameter can specify do_irq(), but see that description for some special considerations. The third parameter is a number of an interrupt to activate if there is no default interrupt for this ilevel. The fourth parameter is the dosemu function to be called from do_irq(). Required by some internal dosemu drivers that needs some additional code before calling an actual handler. This function MUST provide a EOI at the end of a callback.
run_irqs, which is initiated via the macro pic_run, is the "brains" of the pic. It is called from the vm86() loop, checks for the highest priority interrupt requested, and executes it. This function is written in assembly language in order to take advantage of atomic (indivisible) instructions, so that it should be safe for a two process model, even in a multiple CPU machine. A c language version was started, but it became impossible, even with in-line assembly macros, because such macros can only return a single result. If I find a way to do it in c, I will, but don't hold your breath.
I found a way to write it in C --EB 15 Jan 97
do_irq() calls the correct do_int(). It then executes a vm86 loop until an outb( end-of-interrupt) is found. For priority levels 0 and >15 (not real IRQs), vm86 executes once, then returns, since no outb20 will come. Returns: 0 = complete, 1 = interrupt not run because it directly calls our "bios" See run_timer_tick() in timer.c for an example To assure notification when the irq completes, we push flags, ip, and cs here and fake cs:ip to PIC_[SEG,OFF], where there is a hlt. This makes the irq generate a sigsegv, which calls pic_iret when it completes. pic_iret then pops the real cs:ip from the stack. This routine is RE-ENTRANT - it calls run_irqs, which may call an interrupt routine, which may call do_irq(). Be Careful! !!!!!!!!!!!!!!!!!! No single interrupt is ever re-entered.
Callers: base/misc/ioctl.c base/keyboard/serv_8042.c base/keyboard/keyboard-server.c base/serial/ser_irq.c dosext/sound/sound.c dosext/net/net/pktnew.c
pic_resched decrements a count of interrupts on the stack (set by do_irq()). If the count is then less or equal to some pre-defined value (normally 1, pic_icount_od), pic_resched moves all queued interrupts to the interrupt request register.
Normally it is called from pic_iret(), but it can also be called directly if dosemu was fooled by the program and failed to catch iret.
pic_request triggers an interrupt. There is presently no way to "un-trigger" an interrupt. The interrupt will be initiated the next time pic_run is called, unless masked or superceded by a higher priority interrupt. pic_request takes one argument, an interrupt level, which specifies the interrupt to be triggered. If that interrupt is already active, the request will be queued until all active interrupts have been completed. The queue is only one request deep for each interrupt, so it is the responsibility of the interrupt code to retrigger itself if more interrupts are needed.
pic_iret is used to sense that all active interrupts are really complete, so that interrupts queued by pic_request can be triggered. Interrupts end when they issue an outb 0x20 to the pic, however it is not yet safe at that time to retrigger interrupts, since the stack has not been restored to its initial state by an iret. pic_iret is called whenever interrupts have been enabled by a popf, sti, or iret. It determines if an iret was the cause by comparing stack contents with cs and ip. If so, it calls pic_resched() and does the actual iret by pop'ing ip and cs from stack. It is possible for pic_iret to be fooled by dos code; for this reason active interrupts are checked, any queued interrupts that are also active will remain queued. Also, some programs fake an iret, so that it is possible for pic_iret to fail. See pic_watch for the watchdog timer that catches and fixes this event.
pic_watch is a watchdog timer for pending interrupts. If pic_iret somehow fails to activate a pending interrupt request for 2 consecutive timer ticks, pic_watch will activate them anyway. pic_watch is called ONLY by timer_tick, the interval timer signal handler, so the two functions will probably be merged.
This function returns a non-zero value if the designated interrupt has been requested and is not masked. In these circumstances, it is important for a hardware emulation to return a status which does *not* reflect the event(s) which caused the request, until the interrupt actually gets processed. This, in turn, hides the interrupt latency of pic from the dos software.
The single parameter ilevel is the interrupt level (see pic.h) of the interrupt of interest.
If the requested interrupt level is currently active, the returned status will depend upon whether the interrupt code has re-requested itself. If no re-request has occurred, a value of false (zero) will be returned.
pic_activate requests any interrupts whose scheduled time has arrived. anything after pic_dos_time and before pic_sys_time is activated. pic_dos_time is advanced to the earliest time scheduled.
pic_sched schedules an interrupt for activation after a designated time interval. The time measurement is in unis of 1193047/second, the same rate as the pit counters. This is convenient for timer emulation, but can also be used for pacing other functions, such as serial emulation, incoming keystrokes, or video updates. Some sample intervals:
rate/sec: 5 7.5 11 13.45 15 30 60 interval: 238608 159072 108459 88702 79536 39768 19884
rate/sec: 120 180 200 240 360 480 720 interval: 9942 6628 5965 4971 3314 2485 1657
rate/sec: 960 1440 1920 2880 3840 5760 11520 interval: 1243 829 621 414 311 207 103
pic_sched expects two parameters: an interrupt level and an interval. To assure proper repeat scheduling, pic_sched should be called from within the interrupt handler for the same interrupt. The maximum interval is 15 minutes (0x3fffffff).
pic_push,pic_pop
Pic maintains two stacks of the current interrupt level. an internal one is maintained by run_irqs, and is valid whenever the emulator code for an interrupt is active. These functions maintain an external stack, which is valid from the time the dos interrupt code is called until the code has issued all necessary EOIs. Because pic will not necessarily get control immediately after an EOI, another EOI (for another interrupt) could occur. This external stack is kept strictly synchronized with the actions of the dos code to avoid any problems. pic_push and pic_pop maintain the external stack.
The sound code provides emulation of the SB. The actual emulation provided depends upon the support available from the kernel sound driver. Because this is very OS dependant the driver code itself is kept in architecture specifc files under src/arch/osname/dosext/sound/. Communication is via a set of interface functions and the device independant structures.
These are the functions defined in dosext/sound/sound.c.
Arguments are:
port - The I/O port being read from.
Arguments are:
port - The I/O port being read from.
Arguments are:
port - The I/O port being read from.
Arguments are:
port - The I/O port being written to.
value - The value being output.
Arguments are:
value - The value being written to the DSP.
Write silence could probably be implemented by setting up a "DMA" transfer from /dev/null - AM
The file header needs tidying up a _LOT_ ! */
-----
Adlib status reads are unimplemented */
-----
Advanced adlib reads are unimplemented */
-----
CMS Writes are unimplemented.
-----
DSP Status is unimplemented
-----
Adlib Waveform tests are unimplemented */
-----
Advanced Adlib register writes are unimplemented */
-----
Advanced Adlib data writes are unimplemented */
-----
SB Midi is Unimplemented
-----
Sine Generation is unimplemented
-----
AUX Status is Unimplemented
**** WARNING **** This Code _HAS_ changed.
-----
The Emulated DMA channels are provided by using files and writes. This means that they are easy to track. It might cause problems when attempting to interface to the REAL DMA controller. (Necessary to talk to hardware which uses DMA.)
Note that DMA controller 2 uses word granular addressing and controller 1 uses byte granular address ... this simplifies the code !
: Cascade Mode Reads are not supported
-----
: The Verify Mode is not supported
-----
: The Invalid Mode is not supported (!)
This hold all kind of accessing files on a Unix filesysten from DOS.
The msdos_dir_ent structure has much more than 28 bytes. Is this significant?
-----
Added compares to device so that newer versions of Foxpro which test directories using xx\yy\device perform closer to whats DOS does.
We probably should use llseek here for file > 2 GBytes
-----
returned size of struct dir_ent seems wrong at 28 bytes. */
The Following items are used to delimit the text used to create this file. Whilst it is not necessary to know this, they are included because they may be useful for searching, as they are (at least at the moment) reasonably unique.
DANG_BEGIN_MODULE / DANG_END_MODULE This will bracket a description of the file (normally at the start). Within this you may have the keyword 'Maintainer:' followed by a list (one line each) of maintainers for this packet. These will be turned into URLs.
DANG_BEGIN_FUNCTION / DANG_END_FUNCTION This brackets a description of functions (good this, isn't it!) Not every function needs to be described in this way - just the major ones. Within this you may have the keywords: `arguments:', `return:' and `description:', which will sort out the information following it to build proper lists.
DANG_BEGIN_STRUCT / DANG_END_STRUCT This brackets a description of structures and data definitions Not every structure needs to be described in this way - just the major ones. Within this you may have the keywords: `elements:', and `description:', which will sort out the information following it to build proper lists. Also, you may bracket the structur definition of real C-code, given you have one element per line. In this case comments (/*...*/) behind the element will be inserted properly into the formatted list while the C-code itself is still compilable.
DANG_BEGIN_REMARK / DANG_END_REMARK This brackets descriptions of obscure items, like data structures and architecture.
DANG_FIXTHIS This is a one line item, indicating a an area requiring a fix, or redesign.
DANG_BEGIN_NEWIDEA / DANG_END_NEWIDEA New Ideas Start Here! As Ideas are proposed, that get added with their description, so that future generations can laugh at or code the ideas ..... These bracket the idea description.
DANG_BEGIN_CHANGELOG / DANG_END_CHANGELOG Changelogs - very useful for bug fixing, and avvailable for use with DPR (or that's the theory)
In addition there are some keywords, that are recognized within a bracket.
VERB ... /VERB This formats the enclosed text verbatim. This is valid within *_MODULE, *_REMARK, *_STRUCT, *_FUNCTION
REMARK ... /REMARK This is only valid within *_MODULE and also can contain VERB brackets. Its useful to when you want to have a global modul description
PROTO ... /PROTO This is only valid within *_FUNCTION and takes a C-function prototype as `verbatim' until either a `{' or a /PROTO is seen. After this all input is `skipped' until the next PROTO or a /SKIP.
SKIP ... /SKIP This is only valid within *_FUNCTION and skips formatting until either PROTO or /SKIP is seen.