On Sun, 23 Aug 1998, Thor Lancelot Simon wrote:

> On Mon, Aug 24, 1998 at 09:52:22AM +1000, Simon Burge wrote:
> > On Wed, 19 Aug 1998 16:24:48 -0700 (PDT)  "Aaron J. Grier" wrote:
> > 
> > > I already have four rz23s I'm trying to get into a ccd, if I can keep
> > > 1.3.2 from spontaneously rebooting under heavy disk load.

After supping down -current, changing trap.c, and upgrading gcc to
whatever ships in 1.3.2 comp.tgz, I was able to compile up -current.  The
good news is that my 5000/240 doesn't spontaneously reboot anymore when
put under heavy SCSI IO load.  :)  The bad news is, mirroring under ccd is
acting funny.

goldberry# ccdconfig -g
ccd0            32      6       /dev/rz1g /dev/rz2g /dev/rz3g /dev/rz4g
goldberry# disklabel ccd0
# /dev/rccd0c:
[SNIP]
bytes/sector: 512
sectors/track: 2048
tracks/cylinder: 1
sectors/cylinder: 2048
cylinders: 199
total sectors: 409152
[SNIP]
1 partitions:
#        size   offset    fstype   [fsize bsize   cpg]
  a:   409152        0      4.2BSD     1024  8192    16   # (Cyl.    0 - 199*)
goldberry# newfs /dev/rccd0a
Warning: 448 sector(s) in last cylinder unallocated
/dev/rccd0a:    409152 sectors in 200 cylinders of 1 tracks, 2048 sectors
        199.8MB in 13 cyl groups (16 c/g, 16.00MB/g, 3840 i/g)
kvtophys: pte not valid for c3f31f30
kvtophys: pte not valid for c3f32000
kvtophys: pte not valid for c3f33000
sukvtophys: pte not valid for c3f31fb0
kvtophys: pte not valid for c3f32000
per-block backups (for fsck -b #) at:
 32,Aug 25 12:44:38 goldberry /netbsd: kvtophys: pte not valid for c3f31f30

etc...  any disk IO on the mirrored partition spits out more kvtophys
messages from the kernel.  (I have DEBUG enabled.)

> Changing NKMEMCLUSTERS may help.  It would probably help more to use the
> pool allocator for CCD buffers instead of allocating them with malloc();
> I'm not at a location where I have a NetBSD source tree, so I can't
> easily check if that's been done in -current.

I'll try changing NKMEMCLUSTERS and see what happens...

----
  Aaron J. Grier  | "Not your ordinary poofy goof." | agrier@poofy.goof.com
  "No one's jumped off the top of the [Windows] building here, so I guess
   that's a pretty good indicator that it can't be all that bad."
    -- Rob Bennett, Microsoft group product manager, on Windows 98