BioDMA
From SpareMiNT Wiki
BIODMA.XIF (C) 1995-1999 by T. Lang
Contents |
[edit] Please note
Since the BIOODMA adaptors are not designed for asynchronous receiving and additionally do NOT conform to the ACSI standard they work rather bad & slow. Especially on accelerated machines (TT, PAK/3 etc.) the driver is not reliable. So I have decided to stop support for it. This is the last release of the driver. It may work for you or not. If it doesn't please note that support is stopped and that I will not give any support for this driver...
The driver for the BIONET DMA adaptor is in a very early state but seems to work quite stable. I had some problems on my 48MHz PAK/3/4ST/16TT machine but since I had connected some other equipment these problems may not be caused by the driver itself...
The driver is based on dummy-eth from Kay Roemer, the DMA code is T. Lang and is losely based on some assembler source code by Hans-Peter Jansen, LISA GmbH, Berlin.
There are some important things to know, so READ THIS BEFORE USING:
- Starting and stopping the DMA adaptor with ifconfig up or down doesn't really work since there are no possibilities to start or stop the BIODATA adaptor...
- It is important that a harddisk driver checks the FLOCK flag itself before beginning its work. At least some drivers like HUSHI don't and are NOT recommended for use together with these drivers!!!
Be careful with background DMA drivers like the floppy driver or some special harddisk drivers. The protective protocol has been changed for their use:
- when sending: Check for FLOCK, if set -> queue packet
- interrupt routine: Check FLOCK, if set -> quit, else check for queued send packets, dequeue and send them, then check for received packets, if any get them and pass them to the upper layers. So before testing any background DMA stuff you should make a backup of your harddisk!
- Floppy access causes massive delays in sending and receiving!
- Kay told me that MiNT doesn't do any task switches when FLOCK is set.
- I read in the MiNT documentation that MiNT doesn't do any task switches when the processor is in supervisor state.
- Usual harddisk drivers do DMAs in supervisor mode. The LANCEDMA driver relies on these facts. I would have written a completely interrupt driven DMA code (like I've done for my harddsikrecording system for ALL STs working on the ACSI port) but that's a lot of work due to some problems with dirty harddisk drivers like HUSHI ignoring a set FLOCK flag.
[edit] Planned features
- It should be possible that one BIODMA.XIF manages more than one DMA adaptor. At the moment you can use only one.
[edit] Still missing
- Error management is missing due to missing documentation.
- No cache management yet... ...since there are no public cache management routines in the MiNT kernel. If someone exports the internal management routines of MiNT I will use them in the next release ;-)
[edit] ATTENTION
As a replacement for cache management I've used another trick in this release of the driver: The DMA buffer is accessed only at a mirrored address. The address is ORed with $FF000000 so that the buffer is accessed in an uncached area. The only drawback with this trick is that it works DEFINITELY NOT on machines with real 32bit adressing without patching the MMU tables. This means that it works not on the Atari TT and perhaps also will cause trouble on the Eagle or the Medusa since AFAIK they don't mirror the ST RAM. If you want to use this driver on one of these machines please use an older version of the driver. Well, the driver SHOULD work on tuned STs like the MEGA ST + PAK68/3 + FRAK.
DON'T TRY THESE DRIVERS UNLESS YOU ARE ABSOLUTELY SURE WHAT YOU ARE DOING. I'm not responsible for any crashed harddisk or so.
[edit] SPECIAL OPTIONS
None
[edit] Contact
If you have any suggestions, flames, hints or so do not hesitate to contact me.
My address:
Torsten Lang
Bismarckstr. 6
35410 Hungen
EMail: Torsten@Lang.net
