Grabbing a standard VGA graphics mode (from svgalib or dosemu, or even from DOS itself, using grabmode.exe ) for inclusion in your own TextConfig (or XF86Config) file is probably the most attractive one. "vgaset", included in most X-distributions, can give you that, but it does NOT tell you the pixel clock currently used, neither does it tell you about doublescan modes, sync polarities, etc. The output of grabmode can be pasted right into your XF86Config file (if it's correct...).

Of course, you could also grab the default Linux (and/or DOS) text modes to include them in your TextConfig file. So you can show your friends how the old modes looked, and how the SVGATextMode-"enhanced" modes look in contrast ;-)

Debugging XFree86 or SVGATextMode setups is another one. E.g. if SVGATextMode or XFree86 fails to show you the kind of screen you'd expect, and/or your monitor doesn't seem to like what you send it, then the most common reason for that is that the pixel clock is not programmed correctly, or not at all. Clockprobe or grabmode (even when typed blindly, redirected to a file for later viewing) should be able to tell you what clock you are ACTUALLY using. It's not because your "clocks" line says "50 MHz" for some clock, that that clock index actually IS 50 MHz. You could be using the wrong clocks line, or there could be a bug in the clock setting code. In either case, the clockprobe can tell you the ACTUAL clock.

clockprobe: Estimated vertical scanrate = 69.754 Hz.
clockprobe: Estimated horizontal scanrate = 55.803 kHz.
clockprobe: Estimated pixel clock = 74.83 MHz

Note that the clockprobe (both in clockprobe and grabmode) only measures the vertical refresh rate. The rest is calculated from the VGA register contents and the vertical refresh. Extended and special modes which could cause grabmode to mis-detect a mode, will also cause the horizontal frequency and/or the pixel clock to be wrong.

(the output comes on one line)

"116x48" 74.832 928 968 1104 1192 768 775 776 800 +Hsync +Vsync font 9x16 # 55.803kHz/69.75Hz

adds one more line to that:

# Active special VGA register flags: WORDMODE

This is a rather trivial feature, and is (for as far as I know) reliable enough to go without warning. Doublescan is a way to display modes with very few lines on a screen that was made to display many more. The most common application is the CGA-emulation mode 320x200. Displaying only 200 lines on a VGA screen would look awful (you'd see large black lines in between the 200 video lines).

Doublescan displays each line twice, so there are actually 400 lines on screen, but only 200 different ones. In this mode, the VGA is programmed for 400 lines, but with Doublescan enabled, the ACTUAL line count is half that. Grabmode will thus divide all vertical timings by 2 before displaying them.

The standard VGA register set only allows for a maximum of 2048 pixels horizontally and 1024 lines vertically (this is the FULL size, including sync and blanking, so the actual visible size is even smaller). Modern SVGA cards can do modes with 1024 active lines and more, requiring more than 1024 lines in total. A 1280x1024 mode for example needs 1024 active lines. Add sync and blanking time to that, and the grand total is over 1024 lines.

Although standard VGA has a "vertical double" mode bit, which multiplies all vertical values by 2, most SVGA drivers prefer to use extended (chipset dependent) registers to accomplish that. Grabmode cannot read these extended registers (we want to keep it chipset-independent, remember?), so it will show wrapped-around timings for those modes with around 1024 lines.

example:

Depending on how exactly a mode is implemented on your specific VGA card, a 1280x1024 mode at 60 Hz, which should be detected as:

"1280x1024"  110.34  1280 1328 1512 1712  1024 1025 1028 1054

will be detected (in "raw" mode) as:

"1280x1024"  3.131  1280 1328 1512 1712  1024 1 4 30

Grabmode will see that the sequence of vertical timings is wrong, and will assume that this was an extended mode, using extended registers. It will correct the sequence to make it look like the correct one.

This feature only works when the vertical timings happen to cross over the 10-bit standard VGA limit, which means it will only work for 1280x1024 modes. Far from perfect, but better than not at all...

The "golden ratio" rule suggests that most displays have a physical aspect ratio of 4x3 (width/height = 4/3). Most applications assume pixels are square (same width and height), so the normal case would be to use the same ratio in programming the VGA card.

640x480 modes have an aspect ratio of 640/480 = 1.333 (=4/3). 800x600, 1024x768, 1280x1024, 1600x1200: all those modes use an aspect ratio of 4/3 (or a close match) (anyone have a 16/9 computer display -- they exist) ?

If grabmode detects a mode that looks like double or triple that ratio, it assumes this is a special mode (see below), and divides the horizontal timings by 2 or 3. In most common cases, this is the correct result. An appropriate warning is given.

The golden ratio assumption is then used by the CGA emulation detector, the Hi-TrueColor mode detector and the interlace detector to adjust the mode timings.

Grabmode (and clockprobe) report the _actual_ pixel clock used at which data is being transferred from the VGA chip to the RAMDAC. This means that any special data transfer scheme will mislead the clock probe, causing a faulty report.

HighColor and TrueColor modes on most cheap cards are one such example: the VGA registers for e.g. a 640x480x16M mode on a normal VGA card with a standard 8-bit RAMDAC (like most cheap VGA cards, e.g. using the S3 GenDAC or ATT20C490/1/2) are set up as if the mode was actually 2x (16 bpp) or 3x (24 bpp) or 4x (32 bpp) the actual horizontal width, and the pixel clock is multiplied by the same value. So in that case a 24-bit 640x480 mode at 25 MHz pixel clock is programmed in the VGA chip as a 1920x480 mode at 75 MHz! For this mode, the RAMDAC is programmed to assemble three consecutive bytes into one giant 24-bit color pixel.

This ramdac programming is very RAMDAC-dependent, so it is impossible to detect what is going on without knowing the RAMDAC type. RAMDAC's with 16-bit interfaces (like the S3 SDAC) use this scheme for 24/32-bit modes: they transfer one 24-bit pixel in 2 clock periods, or, worse even, for 24-bit packed pixel modes, in 3/2 clock periods...

To make things even worse, grabmode will not even correctly report simple 256-color (8-bit) modes on some cards. Especially high-end cards (like #9, Diamond, ...) using special VGA DAC's (Bt485,...) with pixel multiplexing. They will report completely wrong clocks and refresh frequencies things on higher (>67.5 MHz, for example) pixel clocks. Such RAMDAC's can transfer 2 or more entire pixels in one clock period, the opposite of simple RAMDAC's in 16/24/32-bit modes.

On many (if not all) VGA cards, interlaced modes can be used to allow a high resolution mode to be displayed on a low-end display (= with a low maximum horizontal scan rate) without having to resort to ridiculously low vertical refresh rates.

In order to do this, they program the VGA chip with HALF the amount of actual lines on the screen, and set a special bit that tells the chip this is an interlaced mode.

This "interlace-bit" is, again, in the extended (non-standard) VGA registers, and can thus not be detected by the mode grabber. What _can_ be detected is that the amount of lines programmed in the VGA registers is only half of the height of the screen. This would be easy if there was a way to know how much lines the screen actually has...

But since there isn't, grabmode must resort to some assumptions: assuming that most graphics modes use one the standard screen sizes below:

640x480
800x600
1024x768
1152x900
1152x910
1280x1024
1600x1200

the mode grabber can then check if the current mode has exactly HALF the amount of lines of one of these modes: 240, 300, ..., 600 lines. If so, it ASSUMES (!) that this is an interlaced mode, and changes the timings accoringly before displaying them.

This immediately shows that the mode grabber will not be able to detect non-standard interlaced modes.

A very popular graphics mode is the 320x200 mode with 256 colors. This is CGA resolution, but with 256 colors (BIOS mode 13h): the only standard VGA 256 color mode. In this mode, the VGA reads 4 bits per pixel clock, assembles each 2 groups of 4 bits into one 8-bit pixel and sends this to the VGA. The actual VGA registers are thus programmed for a 640x400 mode: 400 lines with DoubleScan is 200 real lines, and 640 pixels with each 2 pixels output as one is 320 real pixels. Why make it easy?

This pixel joining scheme is enabled by a certain bit in the VGA registers, which is detected by grabmode to trigger it to transform the mode into a "CGA" emulated mode. Easy.

BUT. Extended VGA modes use extended VGA registers, and those have priority over the standard VGA registers. Hence, even if this special CGA-emulation bit is set ON, it doesn't necessarily mean pixels are being combined.

All in all, even if this special mode is set in the standard VGA registers, the extended registers could override that. Detecting this mode is thus not trivial. This is why the "golden ratio" detection is used for this: after doublescan has been accounted for, the mode timings come out as 640x200, which triggers the golden ratio detector to transform this into 320x200, with an appropriate warning.

The clock probe measures the average time between two vertical retraces (busy-waiting. Yuck!). This means it will wait forever when there is no vertical retrace signal. This can be caused either by a program actually wrecking the current video mode (SVGATextMode has been known to do that :-) or when the display system is in power-saving mode.

When there is no vertical retrace detected within a few seconds, the probe assumes it will never get one, and bail out with a message to that extent. This avoids that the probe hangs forever.

The clock probe (used by clockprobe and grabmode) works best when the system is not severely loaded. Especially disk access seems to have a bad influence. DOSEMU is even worse. Very slow machines (i386) might run out of steam while measuring. On 486 machines, the probe works fine.

However, even busy machines should give a reasonable result, unlike the X clock probe. Even heavily loaded (swapping!) machines should give a rather close estimate (maybe a few MHz off, but that should be all). System loads to up to 3 or 4 will probably still yield results which are only a few MHz wrong.

The clockprobe first performs 100 measurements of the time between two vertical retraces, then calculates a histogram of those measurements. If the histogram does not have a nice peak (mostly due to heavy system load, causing many measurements to be wrong), the measurement is done all over again. After three failures, the clock probe will stop retrying, using the last measurement, but also warning you about the unreliability of the result.

In most cases, the clock probe will measure a value slightly higher or slightly lower than you expected. It might be for example that you programmed a mode line that specified a 70.000 MHz pixel clock, but the clock probe insists that it is 75.175 MHz. This is due to the fact that any clock generator cannot produce just _any_ clock you ask for, but must pick the closest available one, either from a list (the `Clocks' line in SVGATextMode or XFree86), or from the possible settings on a programmable clock chip.

If the clockprobe insists on the same value over and over, it means that that is the actual clock. If it insists on something far from the one you expected, then it's either a bug in the probe, or a misconception on your side.

Some special modes like CGA emulation, HiCilor, TrueColor, pixel multiplexing, ..., will cause grabmode to fail to detect the correct number of pixels per line. And because the horizontal sync and clock speed are all calculated from the H and V timings, they too are wrong. The vertical sync speed is ALWAYS correct, since it is MEASURED, independantly from ANY timing register.

The DOS grabmode/clockprobe programs will report heavy system load when they are ran from a Windows DOS box. Even when you are not running any other program. This is typical Windows behaviour: wasting time while doing nothing...

When executed in an MSWindows DOS box, you will ALWAYS probe the VGA parameters from the full-screen DOS version, even when you are running the DOS box in a window and not full-screen. This is because MSWindows hides its VGA registers from normal programs (probably through the MMU address mapping), so the mode grabber gets fooled into believing this is still a text mode box.

For the same reason, doing a grabmode in a DOS window will even lie about the font size. Windows will give you an "appropriate" font size, which doesn't have to be the real one.

A Full-screen DOS box WILL give a correct probe, although there could be a warning about serious system-load and the resulting timing-inaccuracy.

Clock probing is a DOS window is a hoax. You will get extremely high clock rates, or even floating point overflows. Use the "-c" option (don't probe clock rate) to avoid this.

Conclusion: Probing a DOS window is TOTALLY USELESS. You get the wrong VGA register contents (those from the text mode, not those from the Windows mode), and the wrong pixel clock.

Too many to count. Or even more...

Most bugs not yet mentionned above result from the various assumptions the mode grabber has to make along the way, which in turn is caused by the wealth of different methods VGA chip makers come up with to make special mode programming non-standard.

As mentionned above, the following special modes could cause grabmode and/or clockprobe to get it all wrong:

Interlaced modes
Hicolor modes
TrueColor modes
Pixel multiplexed modes
Extended VGA modes ( > 1280x1024)

The reasons for this are explained in one of the sections above.

If you don't trust the result, use the "-r" (raw probing), and add some common sense. Then determine what the real mode parameters should be. In most cases, you know what the actual mode is (e.g. 1024x768), so you can use that knowledge to interpret the raw timing report.

Some IBM 8514 compatible cards (like ATI Mach) use a completely different set of registers than the normal VGA set when they are in graphics mode. Because of the chipset-independance of grabmode, grabmode will not notice this, and report the values which are in the now unused VGA register set. In most cases, this is the text mode from which the graphics program (XFree86, for example) was started.

The main reason for all these problems are the author's stubbornness (combined with a fair amount of lazyness) to keep grabmode chipset-independent.

Grabmode is only reliable when used to grab a pure text mode, or when the mode sizes are given on the command line. I just HATE "intelligent" software, don't you?