Unfortunately I haven't documented this yet. However, the current code has been completely rewritten from this.
DOSEMU DMA code Copyright (C) 1995 Joel N. Weber II This dma code is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
This is intended to be a reasonably complete implementation of dma. However, the following has been omitted:
There is no command register. Half the bits control hardware details, two bits are related to memory to memory transfer, and two bits have misc. functions that might be useful execpt that they are global, so probably only the bios messes with them. And dosemu's psuedo-bios certainly doesn't need them.
16 bit DMA isn't supported, because I'm not quite sure how it's supposed to work. DMA channels 4-7 are treated the same way as 0-3 execpt that they have different address. Also, my currnet goal is a sb pro emulator, and the sb pro doesn't use 16 bit DMA. So it's really a combination of lack of information and lack of need.
Verify is not supported. It doesn't actually move any data, so how can it possibly be simulated?
The mode selection that determines whether to transfer one byte at a time or several is ignored because we want to feed the data to the kernel efficiently and let it control the speed.
Cascade mode really isn't supported; however all the channels are available so I don't consider it nessisary.
Autoinitialization may be broken. From the docs I have, the current and reload registers might be written at the same time??? I can't tell.
The docs I have refer to a "temporary register". I can't figure out what it is, so guess what: It ain't implemented. (It's only a read register).
The following is known to be broken. It should be fixed. Any volunteers? :-)
dma_ff1 is not used yet
Address decrement is not supported. As far as I know, there's no simple, effecient flag to pass to the kernel.
I should have used much more bitwise logic and conditional expressions.
This is my second real C program, and I had a lot of experience in Pascal before that.
Read include/dma.h. In the dma_ch struct, you'll find some fields that don't exist on the real DMA controller itself. Those are for you to fill in. I trust that they are self-explainatory.
One trick that you should know: if you know you're writing to a device which will fill up and you want the transfer to occur in the background, open the file with O_NONBLOCK.
PC Game Programers Encyclopedia ftp://teeri.oulu.fi/pub/msdos/programming/gpe/
The Intel Microprocessors: 8086/8088, 80186, 80286, 80386, 80486, Barry B. Brey,ISBN 0-02-314250-2,1994,Macmillan
(The only reason I use this book so extensively is it's the only one like it that I can find in the Hawaii State Library System :-)