Subject: Re: DMA beyond end of isa
To: None <port-i386@NetBSD.ORG>
From: Charles Hannum <mycroft@deshaw.com>
List: port-i386
Date: 01/04/1996 01:29:33
In article <199601031825.MAA02868@darkwatch.flame.org>
explorer@flame.org (Michael Graff) writes:

   >most of the code that deals with "motherboard dma," i.e. not
   >bus-mastered, should be shareable.
   >
   >the need for machine-specific stuff comes when you're mapping regions
   >for dma and unmapping them later (doing bounce-buffer handling while
   >you do that).  

   Time to dig out the source again.  :)

   Is there anyone with more clue in this area who wants to work on this
   and see what happens?

If someone presents a reasonable, coherent plan that will not break
existing devices, I will certainly look at it.  However, last time
this was mentioned to me, Chris's code was not publically available,
and he refused to give me a copy or explain how he had resolved a
particular problem, so I don't even know what it does.

One thing I would recommend is changing the `iobase' variables to a
named, machine-dependent type, such that architectures that do
memory-mapped I/O can store a partially-calculated pointer instead.
Done right, this could significantly speed up the weird mappings that
some architectures do.