While experimenting with the scsi code on turbo hardware, I see a kernel
buserror while reading an esp register, just after the first part of the
first esp DMA transfer finished.

The address in question looks correct relative to intiobase,
intiobase seems unchanged, and trying to read a register in ddb gets a
similar bus error.

It seems that the mapping of the io space does not work anymore. Is there
a way to confirm/deny/debug this? PMAP_DEBUG, perhaps?

What happens is roughly the following:

esp tries to select scsi devices.  It sets up a request sense for the
first target found. The first four bytes of the response are read from
fifo, the rest via main and tail dma maps. With the next dma interrupt the
main map (16 bytes) has been read ok (verified in ddb), and in the esp
completed callback the esp dma register is flushed and nextdma_intr()
called directly (?).
The following shutdown callback tries to read NCR_FFLAG, and that bombs.

Environment:
pre-1.5 source with DISABLE_NEXT_BMAP_CHIP, DBJ's recent patches and
modified nextdma code.

Any insights?

Timm
-- 
Timm Wetzel     	<twetzel@gwdg.de>
Biomed. NMR GmbH        Tel +49 551 201-1091    FAX +49 551 201-1307
Am Fassberg 11, D-37077 Goettingen, Germany