>> I see no floppy support (for the UDB). I also see no DMA support.
>
>Right.
>
>By "DMA" i assume you mean ISA/EISA motherboard DMA controller DMA?
>(the support for which isn't there?)

Right.

>
>Basically, for ISA/EISA motherboard DMA controller DMA to be useful,
>DMA mapping and unmapping, via SGMAP entries, has to be done, since
>there is no memory whatsoever mapped in the low 16M of ISA space.
>(You can't even use a naive bounce-buffer scheme, for that reason.)
>
>
>> Anyone working on this?
>
>Not really.  Before I throw too much work at it, I'd like that much
>talked about it MI DMA mapping/unmapping scheme to appear.  I'd like
>to avoid implementing a hack, then implementing the 'real' solution
>later.
>
>That having been said, it shouldn't be too hard to make it work, if
>somebody's so inclined.
>
>
>In response to the recent messages (yours and the followup from Thomas
>Graichen), i looked at what it would take to implement simple "only
>bounce-buffered DMA" with the ISA/EISA motherboard DMACs, and to make
>the floppy driver work.
>
>Making the floppy driver compile (based on the i386 version) was
>rather easy.  (Just a matter of trimming some stuff out and working
>around its absence.)
>
>Making the DMA controller code go is a bit harder.  Basically:
>	(1) you need to allocate bounce buffer space at boot time,
>	(2) you need to set up some SGMAP entries to address that
>	    space, and
>	(3) you need to hack the isadma.c code to deal with it.
>
>Less than a few hours work, but more than I have time for, since i'm
>already busy tilting at too many flaming windmills with only a water
>pick...  (how's that for mixed metaphors? 8-)
>
>(1) and (3) are actually pretty easy, probably.  (2) requires hacking
>each bus interface chipset's code, though as long as you're simply
>trying to make it work and not make it nice, it wouldn't be too hard.
>8-)
>
>
>If anybody's actually interested in pursuing this, send me mail and
>i'd be more than happy to send you what code i've got, and give you
>advice about how to do the rest, if you need it.  I really wanted to
>hack it to work yesterday, but i just had too much else to do, and it
>requires more than the (little) bit of work that i've got time to
>throw at it right now...
>
>

Basically, I looked at #3, started thinking about #2, and realized that
if I started down this way, I'd be in for more than a few hours work
and then thought: "It has to wait until I get the AXP 8200 running",
so I'll punt until then.