Well, Andrew Gillham was kind enough to embark on a short debugging 
session, and we managed to get DMA working on his 4/110:

He tells me that reads went from 125k/sec to 650k/sec and writes from 
100k/sec to 350k/sec.  We decided to be conservative at first and use 
polled DMA.  I think he's compiling a kernel with interrupts and reselect 
enabled as we speak :-)

As soon as I tidy up a few things, I'll send this stuff off to Paul.  
Wow!  We can really have an MI 5380 driver!  Woo hoo!

Now, if I can only track down that annoying "ie0: DMA underrun" error...

NetBSD 1.1 (ANCHOR) #6: Wed Dec  6 03:34:11 EST 1995
    root@:/usr/src/sys/arch/sparc/compile/ANCHOR
real mem = 16752640
avail mem = 14688256
using 102 buffers containing 835584 bytes of memory
bootpath: /obio0/sw0/sd@0,0
mainbus0 (root)
cpu0 at mainbus0: SUN-4/100 series (MB86910 or WTL1164/5 FPU)
obio0 at mainbus0
oclock0 at obio0 addr 3000000
eeprom0 at obio0 addr 2000000
memreg0 at obio0 addr 4000000
zs0 at obio0 addr 1000000 pri 12, softpri 6
zs0a: console i/o
ie0 at obio0 addr 6000000 pri 6 address 08:00:20:06:25:38, type onboard
sw0 at obio0 addr a000000 pri 3
sw0: options=6<NO_RESELECT,NO_DMA_INTR>
scsibus0 at sw0
sw0 targ 0 lun 0: <CDC, 94171-9, 8856> SCSI1 0/direct fixed
sd0 at scsibus0: 312MB, 1549 cyl, 9 head, 45 sec, 512 bytes/sec
vmel0 at mainbus0
vmes0 at mainbus0
root on sd0a
Enter pathname of shell or RETURN for sh:
#

------------------------------------------------------------------------------
Jason R. Thorpe                                         thorpej@Xenotropic.COM

           Just me and my collection of obsolete computer gear(s).