gwr@netbsd.org wrote:

>  > Well, I guess that ncr53c9x_dmaselect should be set
>  > *after* ncr53c9x_attach().
> 
> I did that.  The code I'm using is identical to the sparc version
> (as much as possible) but DMA seems to complete with an interrupt
> after a single byte of a seven byte transfer, and then refuses to
> send any more.  (See the debug trace sent previously).
> 
> Looking for clues...  This is an old, REV zero DMA chip, btw.

Hmm, I think the problem is only NCRCMD_SELATN with DMA.
The command was transfered into the ESP FIFO by DMA,
but ESP did not transfer it to the target device.
(I got "dmaintr: empty esp FIFO of 6" messages.)
Other DMA transfer seems working.

BTW, are you testing 1.4_ALPHA kernel? When I tried 1.4_ALPHA on
my sun3x, the kernel corrupted its root filesystem so I could not
test the esp driver by myself... Now I'm testing 1.3K sources.
---
Izumi Tsutsui
tsutsui@ceres.dti.ne.jp