0. Introduction
1. Installation
2. How to use svgalib
3. Description of svgalib functions
4. Overview of supported SVGA chipsets and modes
5. Detailed comments on certain device drivers
6. Goals
7. References (location of latest version, apps etc.)
8. Known bugs
This is a low level graphics library for Linux, originally based on VGAlib 1.2 by Tommy Frandsen. VGAlib supported a number of standard VGA graphics modes, as well as Tseng ET4000 high resolution 256-color modes. As of now, support for many more chipsets has been added. See section 4 Overview of supported SVGA chipsets and modes
It supports transparent virtual console switching, that is, you can switch consoles to and from text and graphics mode consoles using alt-[function key]. Also, svgalib corrects most of VGAlib's textmode corruption behaviour by catching SIGSEGV, SIGFPE, SIGILL, and other fatal signals and ensuring that a program is running in the currently visible virtual console before setting a graphics mode.
Note right here that SIGUSR1 and SIGUSR2 are used to manage console switching internally in svgalib. You can not use them in your programs. If your program needs to use one of those signals, svgalib can be compiled to use other signals, by editing libvga.h
This version includes code to hunt for a free virtual console on its own in case you are not starting the program from one (but instead over a network or modem login, from within screen(1) or an xterm(1)). Provided there is a free console, this succeeds if you are root or if the svgalib calling user own the current console. This is to avoid people not using the console being able to fiddle with it. On graceful exit the program returns to the console from which it was started. Otherwise it remains in text mode at the VC which svgalib allocated to allow you to see any error messages. In any case, any I/O the svgalib makes in text mode (after calling vga_init(3)) will also take place at this new console.
Alas, some games misuse their suid root privilege and run as full root process. svgalib cannot detect this and allows Joe Blow User to open a new VC on the console. If this annoys you, ROOT_VC_SHORTCUT in Makefile.cfg allows you to disable allocating a new VC for root (except when he owns the current console) when you compile svgalib. This is the default.
When the library is used by a program at run-time, first the chipset is detected and the appropriate driver is used. This means that a graphics program will work on any card that is supported by svgalib, if the mode it uses is supported by the chipset driver for that card. The library is upwardly compatible with VGAlib.
The set of drawing functions provided by svgalib itself is limited (unchanged from VGAlib) and unoptimized; you can however use vga_setpage(3) and vga_getgraphmem(3) (which points to the 64K VGA framebuffer) in a program or graphics library. A fast external framebuffer graphics library for linear and banked 1, 2, 3 and 4 bytes per pixel modes is included (it also indirectly supports planar VGA modes). It is documented in vgagl(7).
One obvious application of the library is a picture viewer. Several are available, along with animation viewers. See the 7. References at the end of this document.
I have added a simple VGA textmode font restoration utility (restorefont(1)) which may help if you suffer from XFree86 textmode font corruption. It can also be used to change the textmode font. It comes with some other textmode utilities: restoretextmode(1) (which saves/restores textmode registers), restorepalette(1), and the script textmode(1). If you run the savetextmode(1) script to save textmode information to /tmp, you'll be able to restore textmode by running the textmode(1) script.
If you need information on installation see 0-INSTALL which comes with the svgalib distribution.
However, even after installation of the library you might need to configure svgalib using the file /etc/vga/libvga.config. Checkout section 4 Overview of supported SVGA chipsets and modes and libvga.config(5) for information.
For basic svgalib usage (no mouse, no raw keyboard) add #include <vga.h> at the beginning your program. Use vga_init(3) as your first svgalib call. This will give up root privileges right after initialization, making setuid-root binaries relatively safe.
The function vga_getdefaultmode(3) checks the environment variable SVGALIB_DEFAULT_MODE for a default mode, and returns the corresponding mode number. The environment string can either be a mode number or a mode name as in (G640x480x2, G640x480x16, G640x480x256 , G640x480x32K, G640x480x64K, G640x480x16M). As an example, to set the default graphics mode to 640x480, 256 colors, use:
export SVGALIB_DEFAULT_MODE=G640x480x256
on the bash(1) command line. If a program needs just a linear VGA/SVGA resolution (as required by vgagl(7)), only modes where bytesperpixel in the vga_modeinfo structure returned by vga_getmodeinfo(3) is greater or equal to 1 should be accepted (this is 0 for tweaked planar 256-color VGA modes).
Use vga_setmode(graphicsmode) to set a graphics mode. Use vga_setmode(TEXT) to restore textmode before program exit.
Programs that use svgalib must #include<vga.h>; if they also use the external graphics library vgagl(7), you must also #include<vgagl.h>. Linking must be done with -lvga (and -lvgagl before -lvga, if vgagl(7) is used). You can save binary space by removing the unused chipset drivers in Makefile.cfg if you only use specific chipsets. However this reduces the flexibility of svgalib and has a significant effect only when you use the static libraries. You should better use the shared libraries and these will load only the really used parts anyway.
Functions in the vgagl(7) library have the prefix gl_. Please see vgagl(7) for details.
There are demos with sources available which will also help to get you started, in recomended order of interest: vgatest(6), keytest(6), mousetest(6), eventtest(6), forktest(6), bg_test(6), scrolltest(6), speedtest(6), fun(6), spin(6), testlinear(6), lineart(6), testgl(6), accel(6), testaccel(6), plane(6), and wrapdemo(6).
Debugging your programs will turn out to be rather difficult, because the svgalib application can not restore textmode when it returns to the debugger.
Happy are the users with a serial terminal, X-station, or another way to log into the machine from network. These can use
textmode </dev/ttyN
on the console where the program runs and continue.
However, the vga_flip(3) function allows you to switch to textmode by entering a call to it blindly into your debugger when your program stops in graphics mode. vga_flip(3) is not very robust though. You shall not call it when svgalib is not yet initialized or in textmode.
Before continuing your program, you must then call vga_flip(3) again to return to graphics mode. If the program will not make any screen accesses or svgalib calls before it returns to the debugger, you can omit that, of course.
This will only work if your program and the debugger run in the same virtual linux console.
320x200x256, and the series of 16-color and non-standard planar 256 color modes supported by VGAlib, as well as 720x348x2.
Supports 640x480x256, 800x600x256, 1024x768x256 SVGA modes
Also known as Promotion at25. Popular as the 2D part of a voodoo rush card. As of this writing there are a few known problems with this driver. Read below.
Full support, limited RAMDAC support. Only ARK1000PV tested. Supports Clocks and Ramdac lines in config file.
This is no real driver. I do not support any new modes. However it saves additional card setup and thus allows use of the plain VGA modes even when you are using non standard text modes. It is possible to enforce use of this driver even on ATI Mach32 but not very useful.
The driver by Michael Weller supports all ATI BIOS-defined modes and more... It hits the best out of your card. Some modes may not have nice default timings but it uses the ATI's EEPROM for custom config or allows to specify modes in libvga.config(5). Some problems may occur with quite some third party cards (usually on board) Mach32 based controllers as they do not completely conform to the Mach32 data sheets. Check out svgalib.mach32(7) (and libvga.config(5)).
A driver for ATi Mach64 based cards with internal DAC.
This server was written using the SVGALIB patch from Sergio and Angelo Masci as a starting point. This version of the code resembled the XFree server code that was used up to XFree 3.1.2. As such it was incapable of programming the clocks, using linear addressing, Hi-Color, True-Color modes or the hardware acceleration. All of these features have since been added to the code. The 64200 and 64300 chips are unsupported, however these chips are very similar to the 6554x chips which are supported.
All the modes, including 256 color, 32K/64K color, 16M color (3 bytes per pixel) and 32-bit pixel 16M color modes (5434). Some bitblt functions are supported. The driver doesn't work with mode dumps, but uses a SVGA abstraction with mode timings like the X drivers.
Supported.
Is supported as an ARK2000PV
This driver was written by Matan Ziv-Av and is derived from the XFree86 driver by David J. Mckay. It lacks 24bit modes (can the card do them at all?), acceleration support and pageflipping in threeDKit is broken.
Driver by Christopher Wiles; includes 32K color modes for OTI-087.
The driver is not complete, but should work on a number of cards/RAMDACs, and 640x480x256 should work on most card. The best support is for a 801/805 with AT&T20C490-compatible RAMDAC, and S3-864 + SDAC. All 256/32K/64K/16M works for them (within the bounds of video memory & ramdac restrictions).
The supported cards include S3 Virge and S3 Trio64 cards.
None of the acceleration function is supported yet.
The chip level code should work with the 964/868/968, but most likely the card they come on would use an unsupported ramdac/clock chip. Support for these chips is slowly being added.
Clocks and Ramdac lines in libvga.config(5) supported.
The maximum pixel clock (in MHz) of the ramdac can be set using a Dacspeed line in the config file. A reasonable default is assumed if the Dacspeed line is omitted. Clocks should be the same as in XFree86. Supported ramdac IDs: Sierra32K, SC15025, SDAC, GenDAC, ATT20C490, ATT20C498, IBMRGB52x.
Example:
Clocks 25.175 28.3 40 70 50 75 36 44.9 0 118 77 31.5 110 65 72 93.5
Ramdac att20c490
DacSpeed 85
Also supported, at least in combination with the SC15025/26A ramdac, is the ICD 2061A clock chip. Since it cannot be autodetected you need to define it in the config file using a Clockchip line. As there is no way to read the current settings out of the 2061, you have the option to specify the frequency used when switching back to text mode as second argument in the Clockchip line.
This is especially required if your text mode is an 132 column mode, since these modes use a clock from the clock chip, while 80 column modes use a fixed clock of 25 MHz. The text mode frequency defaults to 40 MHz, if omitted.
Example:
ClockChip icd2061a 40.0
Derived from tvgalib by Toomas Losin. TVGA 9440 support by ARK <ark@lhq.com, root@ark.dyn.ml.or>.
Supports 640x480x256, 800x600x256, 1024x768x256 (interlaced and non-interlaced) Might be useful to add 16-color modes (for those equipped with a 512K TVGA9000) for the 8900 and 9000 cards.
320x200x{32K, 64K, 16M}, 640x480x{256, 32K, 64K, 16M}, 800x600x{256, 32K, 64K, 16M}, 1024x768x{16, 256}, 800x600x{16, 256, 32K, 64K} modes are supported for the TVGA 9440.
Autodetection can be forced with a:
line in the config file.
memory is the amount of VGA memory in KB, flags is composed of three bits:
Derived from VGAlib; not the same register values. ET4000 register values are not compatible; see svgalib.et4000(7).
Make sure the colors are right in hicolor mode; the vgatest program should draw the same color bars for 256 and hicolor modes (the DAC type is defined at compilation in et4000.regs or the dynamic registers file). ET4000/W32 based cards usually have an AT&T or Sierra 15025/6 DAC. With recent W32p based cards, you might have some luck with the AT&T DAC type. If the high resolution modes don't work, you can try dumping the registers in DOS using the program in the et4000/ directory and putting them in a file (/etc/vga/libvga.et4000 is parsed at runtime if DYNAMIC is defined in Makefile.cfg at compilation (this is default)).
Supported modes are 640x480x256, 800x600x256, 1024x768x256, 640x480x32K, 800x600x32K, 640x480x16M, etc.
Reports of ET4000/W32i/p functionality are welcome.
There may be a problem with the way the hicolor DAC register is handled; dumped registers may use one of two timing methods, with the value written to the register for a particular DAC for a hicolor mode (in vgahico.c) being correct for just one of the these methods. As a consequence some dumped resolutions may work while others don't.
Most modes of which the card is capable are supported. The 8 15 16 24 and 32 bit modes are supported.
The ET6000 has a built in DAC and there is no problem coming from that area. The ET6000 is capable of acceleration, but this as well as sprites are not yet implemented in the driver.
The driver now uses modelines in libvga.config for user defined modes. It is sometimes useful to add a modeline for a resolution which does not display well. For example, the G400x600 is too far to the right of the screen using standard modes. This is corrected by including in libvga.config the line
Modeline "400x600@72" 25.000 400 440 488 520 600 639 644 666
More examples are given below.
This driver was provided by Don Secrest.
Please read README.vesa and README.lrmi in doc subdirectory of the standard distribution.
Go figure! I turned off autodetection in the release, as a broken bios will be called too, maybe crashing the machine. Enforce VESA mode by putting a chipset VESA in the end of your libvga.config(5).
Note that it will leave protected mode and call the cards bios opening the door to many hazards.
Linear graphics is working on this card, both with and without BACKGROUND enabled, and vga_runinbackground works.
I decided it was best to quit working on more modes and try to get acceleration and sprites working.
My et6000 card is on a PCI bus. The card will run on a vesa bus, but since I don't have one on my machine I couldn't develop vesa bus handling. I quit if the bus is a vesa bus.
I check for an et6000 card, which can be unequivocally identified. The et4000 driver does not properly identify et4000 cards. It thinks the et6000 card is an et4000, but can only run it in vga modes.
I have found the following four modelines to be useful in libvga.config or in ~/.svgalibrc for proper display of some modes.
Modeline "512x384@79" 25.175 512 560 592 640 384 428 436 494
Modeline "400x300@72" 25.000 400 456 472 520 300 319 332 350 DOUBLESCAN
Modeline "512x480@71" 25.175 512 584 600 656 480 500 510 550
Modeline "400x600@72" 25.000 400 440 488 520 600 639 644 666
Don Secrest <secrest@uiuc.edu> Aug 21, 1999
As of this writing (1.2.8) fixes were made to the oak driver by Frodo Looijaard <frodol@dds.nl> to reenable OTI-067 support. It is unknown right now if they might have broken OTI-087 support. The author of the '87 support Christopher Wiles <wileyc@moscow.com> owns no longer an OTI-087 card and can thus no longer give optimal support to this driver. Thus you might be better off contacting me or Frodo for questions. If you are a knowledgable OTI-087 user and experience problems you are welcome to provide fixes. No user of a OTI-087 is currently known to me, so if you are able to fix problems with the driver please do so (and contact me) as noone else can.
Michael.
Now back to the original Oak information:
The original OTI driver, which supported the OTI-067/77 at 640x480x256, has been augmented with the following features:
The driver as it exists now is somewhat schizoid. As the '87 incorporates a completely different set of extended registers, I found it necessary to split its routines from the others. Further, I did not have access to either a '67 or a '77 for testing the new resolutions. If using them causes your monitor/video card to fry, your dog to bite you, and so forth, I warned you. The driver works on my '87, and that's all I guarantee. Period.
Heh. Now, if someone wants to try them out ... let me know if they work.
New from last release:
32K modes now work for 640x480 and 800x600. I found that the Sierra DAC information in VGADOC3.ZIP is, well, wrong. But, then again, the information for the '87 was wrong also.
64K modes do not work. I can't even get Oak's BIOS to enter those modes.
I have included a 1280x1024x16 mode, but I haven't tested it. My monitor can't handle that resolution. According to the documentation, with 2 megs the '87 should be able to do an interlaced 1280x1024x256 ... again, I couldn't get the BIOS to do the mode. I haven't 2 megs anyway, so there it sits.
I have included routines for entering and leaving linear mode. They should work, but they don't. It looks like a pointer to the frame buffer is not being passed to SVGALIB. I've been fighting with this one for a month. If anyone wants to play with this, let me know if it can be make to work. I've got exams that I need to pass.
Tidbit: I pulled the extended register info out of the video BIOS. When the information thus obtained failed to work, I procured the OTI-087 data book. It appears that Oak's video BIOS sets various modes incorrectly (e.g. setting 8-bit color as 4, wrong dot clock frequencies, etc.). Sort of makes me wonder ...
Christopher M. Wiles (a0017097@wsuaix.csc.wsu.edu)
12 September 1994
I think the ability to use a VGA/SVGA graphics resolution in one
virtual console, and being able to switch to any other virtual console
and back makes a fairly useful implementation of graphics modes in
the Linux console.
Programs that use svgalib must be setuid root. I don't know how desirable it is to have this changed; direct port access can hardly be done without. Root privileges can now be given up right after initialization. I noticed some unimplemented stuff in the kernel header files that may be useful, although doing all register I/O via the kernel would incur a significant context-switching overhead. An alternative might be to have a pseudo /dev/vga device that yields the required permissions when opened, the device being readable by programs in group vga.
It is important that textmode is restored properly and reliably; it is fairly reliable at the moment, but fast console switching back and forth between two consoles running graphics can give problems. Wild virtual console switching also sometimes corrupts the contents of the textmode screen buffer (not the textmode registers or font). Also if a program crashes it may write into the area where the saved textmode registers are stored, causing textmode not be restored correctly. It would be a good idea to somehow store this information in a 'safe' area (say a kernel buffer). Note that the vga_safety_fork(3) thing has the same idea.
Currently, programs that are in graphics mode are suspended while not in the current virtual console. Would it be a good idea to let them run in the background, virtualizing framebuffer actions (this should not be too hard for linear banked SVGA modes)? It would be nice to have, say, a raytracer with a real-time display run in the background (although just using a separate real-time viewing program is much more elegant).
Anyone wanting to rewrite it all in a cleaner way (something with loadable kernel modules shouldn't hurt performance with linear framebuffer/vgagl type applications) is encouraged.
Also, if anyone feels really strongly about a low-resource and
truecolor supporting graphical window environment with cut-and-paste,
I believe it would be surprisingly little work to come up with a
simple but very useful client-server system with shmem, the most
useful applications being fairly trivial to write (e.g. shell
window, bitmap viewer). And many X apps would port trivially.
This is old information, please be sure to read svgalib.faq(7) if you are interested in further goals.
The latest version of svgalib can be found on sunsite.unc.edu in /pub/Linux/libs/graphics or tsx-11.mit.edu in /pub/linux/sources/libs as svgalib-X.X.X.tar.gz. As of this writing the latest version is svgalib-1.4.1.tar.gz. There are countless mirrors of these ftp servers in the world. Certainly a server close to you will carry it.
The original VGAlib is on tsx-11.mit.edu, pub/linux/sources/libs/vgalib12.tar.Z. tvgalib-1.0.tar.Z is in the same directory.
SLS has long been distributing an old version of VGAlib. Slackware keeps a fairly up-to-date version of svgalib, but it may be installed in different directories from what svgalib likes to do by default. The current svgalib install tries to remove most of this. It also removes /usr/bin/setmclk and /usr/bin/convfont, which is a security risk if setuid-root. Actually the recent makefiles try to do a really good job to cleanup the mess which some distributions make.
If you want to recompile the a.out shared library, you will need the DLL 'tools' package (found on tsx-11.mit.edu, GCC dir). To make it work with recent ELF compiler's you actually need to hand patch it. You should probably not try to compile it. Compiling the ELF library is deadly simple.
And here is a list of other references which is horribly outdated. There are many more svgalib applications as well as the directories might have changed. However, these will give you a start point and names to hunt for on CD's or in ftp archives.
The XFree86 driver sources distributed with the link-kit may be helpful.
There's an alternative RAW-mode keyboard library by Russell Marks for use with svgalib on sunsite.unc.edu.
LIBGRX, the extensive framebuffer library by Csaba Biegl distributed with DJGPP, has been ported to Linux. Contact Hartmut Schirmer (phc27@rz.uni-kiel.d400.de, subject prefix "HARTMUT:"). A more up-to-date port by Daniel Jackson (djackson@icomp.intel.com) is on sunsite.unc.edu.
The vgalib ghostscript device driver sources can be found on sunsite.unc.edu, /pub/Linux/apps/graphics. Ghostscript patches from Slackware: ftp.cdrom.com, /pub/linux/misc. gnuplot patches are on sunsite.unc.edu.
Mitch D'Souza has written font functions that work in 16 color modes and can use VGA textmode (codepage format) fonts; these can be found in his g3fax package in sunsite.unc.edu. These functions may go into a later version of svgalib.
Using a 132 column textmode may cause graphics modes to fail. Try using something like 80x28.
The console switching doesn't preserve some registers that may be used to draw in planar VGA modes.
Wild console switching can cause the text screen to be corrupted, especially when switching between two graphics consoles.
On ET4000, having run XFree86 may cause high resolution modes to fail (this is more XFree86's fault).
The Trident probing routine in the XFree86 server may cause standard VGA modes to fail after exiting X on a Cirrus. Try putting a 'Chipset' line in your Xconfig to avoid the Trident probe, or use the link kit to build a server without the Trident driver. Saving and restoring the textmode registers with savetextmode/textmode (restoretextmode) should also work. [Note: svgalib now resets the particular extended register, but only if the Cirrus driver is used (i.e. the chipset is not forced to VGA)] [This is fixed in XFree86 v2.1]
Some Paradise VGA cards may not work even in standard VGA modes. Can anyone confirm this?
Piping data into a graphics program has problems. I am not sure why. A pity, since zcatting a 5Mb FLC file into flip on a 4Mb machine would be fun.
The tseng3.exe DOS program include as source in the svgalib distribution doesn't recognize any modes on some ET4000 cards. Also ET4000 cards with a Acumos/Cirrus DAC may only work correctly in 64K color mode.