Subject: SCSI errors after installing 1.4.1
To: None <port-mac68k@netbsd.org>
From: Andrew J Robbie <andrew.robbie@dsto.defence.gov.au>
List: port-mac68k
Date: 02/02/2000 21:59:39
Hi people,

After a bit of a break from NetBSD I have just re-installed it on my
Mac IIvx. The install seemed to go fine, however any writes to the
disk produced this message on the console:

scsi_show: scsi_main is not running
REQ-CONNECTED: 0 0x2160000[8192] cmd[0]=a S=0 M=ff R=0 resid=8192 dr_flag=1 
Bus signals (7c/00): I/O|C/D|MSG|REQ|BSY
Dma status (30): IRQ|EPAR
phase = 0, busy:1, spl:2204
        0       scsi_main5
        1       pdma_cleanup2
        2       scsi_main1
        3       scsi_select1
        4       tpio1
        5       tpio2
        6       scsi_select10
        7       info_transf1
        8       tpio1
        9       tpio2
        10      info_transf5
        11      info_transf1
        12      info_transf3
        13      transfer_pdma0
        14      waiting for interrupt.
        15      scsi_main3
        16      scsi_main4
        17      scsi_main5
        18      drq (out)
        19      write complete
        20      read a byte to force a phase change
        21      end drq
        22      irq
        23      pdma_ready0
        24      pdma_ready1.
Unexpected phase change.

This message was printed many times, and (it seemed) proportional to the
number and size of the disk writes involved.

Initially I thought this must be because I needed to use the SBC driver rather
than the NCR, however booting with that kernel was even worse, as it just
spewed out "parity error!" messages.

I read the scsi(4) man page, which seemed to suggest that the scsi device
debugging had been turned on somehow. I wrote a small program to try turning
this off, but it didn't seem to have any effect (opening /dev/sd0c and doing
an ioctl(fd, SCIOCDEBUG, &debug_flag) where debug_flag = 0).

Has anyone else had these types of problems, or looked at the source code
enough to know what is happening? (When I did a stack trace the messages
seemed to originate in _ncr5380_irq_intr() which then called _scsi_show()).

I had release 1.3 and 1.3.1 running pretty well on this machine before, so
I guess I could run that instead, although it would hardly be optimal.


So, any ideas?

TIA,
Andrew



Machine specs:
	Macintosh IIvx

	20MB Ram

	Apple greyscale portrait display card & display.

	SCSI 0 - Quantum LPS 270S, partitioned as 30MB HFS, 192 NetBSD Root&Usr,
		 30MB swap.
	SCSI 3 - CR-ROM
	SCSI 4 - Iomega Zip (with MacOS 7.6.1 boot disk)