On Sat, 11 Oct 1997, Miloslaw Smyk wrote:
> I've been testing new 1.2G kernel (phase5 SCSI support) with my CyberSCSI
> MK-II attached to CyberStorm MK-II 060/50MHz. I booted it from my IDE
> drive and:
> 
> It recognized SCSI interface and attached devices correctly: (...)
> FIREBALL_TM3200S, 300X> SCSI2 0/direct fixed
> Oct 11 19:52:23 amiga /netbsd: sd1: 3067MB, 6810 cyl, 5 head, 184 sec, 512
> bytes/sect x 6281856 sectors

Maybe my results can shed some more light on this subject.
My system is similar to yours. I have a CyberSCSI mkII with a CSmk2
060/50, and I boot NetBSD 1.2 from my IDE drive. I've been testing the new
generic kernels, both a 19971007 (built by Ignatios?) and the -1.2G-phase5
kernel. Both report similar behaviour with my disk:
sd1 at scsibus1 targ 6 lun 0 <QUANTUM FIREBALL_TM2110S, 300X> SCSI2
0/direct fixed
sd1: 2014 MB, 6810 cyl, 4 head, 151 sec 512 bytes/sect x 4124736 sectors
> 
> Then, following lines were displayed:
> 
> Oct 11 19:52:23 amiga /netbsd: warning found rdb->secpercyl(794) !=
> rdb->nsectors(159) * rdb->nheads(5)
> Oct 11 19:52:23 amiga /netbsd: warning lp->d_sparespercyl(1) not multiple of
> lp->d_ntracks(5)
> Oct 11 19:52:24 amiga /netbsd: Partition 'hd0' geometry 5/159 differs from
> RDB 5/159
> Oct 11 19:52:24 amiga /netbsd: Partition 'hd1' geometry 5/159 differs from
> RDB 5/159
> Oct 11 19:52:24 amiga /netbsd: Partition 'hd3' geometry 5/159 differs from
> RDB 5/159

and I get something like this..
warning found rdb->secpercyl(158) != rdb->nsectors(159) * rdb->nheads(1)
Partition HD0 geometry 1/159 differs from RDB 1/159
...all partitions
Partition HD5 geometry 1/159 differs from RDB 1/159

> What's more, instead of listing unrecognized partitions once at boot time,
> the kernel insisted on displaying them every 30 seconds or so.

I only got this listing once, on bootup. However, disklabel sd1 reports
these partitions: (sorry about crappy CnP, I only wrote this down)
sd1c: cyl     2-26105
sd1d: cyl     2-262
sd1e: cyl   262-6745
sd1f: cyl  6745-12190
sd1g: cyl 12190-15510
sd1h: cyl 15510-20824
sd1i: cyl 20824-26105
partition i: partition extends past end of unit.
 
Indeed, mounting partitions d-h went ok, with only the same warning that
partition geometry differs from RDB. Mounting partition i didn't work at
all, illegal parameter or something.

> I then prepared one of SCSI drive's partitions with
> 
> newfs -b4096 <partition>
> So far so good.

I really couldn't spare any of my scsi partitions for trying this, since
they're all in use now.

> I mounted the partition under /mnt and copied kernel file there. Then, I
> unmounted the partition and ran `fsck -f` on it. It reported no problems. So
> I mounted the partition again and tried comparing (cmp) the original kernel
> file with the one on /mnt. At this point I got the following msg:
> 
> panic: pmap_alloc_pv: pgi_nfree inconsistent
> Stopped at 0xb1a26: unlk a6

I don't know whether this might be in any way related, but I accidentally
tried to mount the c partition (entire drive?) and soon I was in the
debugger too..
trap type 3, code=0, v=0
panic: zero divide
stopped at 0xb1a26: unlk a6

Coincidence or no?
> Oh, one more thing: under AmigaOS FFS makes no problem of it, but AFS
> does - it miscalculates the size of the partition.

This drive works with no problem in FFS, I've never used AFS, but I've
tested it with 68k-linux, and had no problems there. There are alot of
Quantum FB's out there, seems to me something could be done with the
kernel to fix this (minor?) problem.

big Thanks to <is> and all who made this p5 scsi support, now netbsd is
finally usable, I even installed X that runs on aga, and this is quite a
bit on the side, but the signal from grf0 only appeared on my 1942 in the
amiga rgb output, and not in the scandoubled output on my cv3d. My 17"
just went blank when I tried startx..

Who's working on a grfx mode for CV64/3D? I know, a FAQ, but I really need
this now :)

cheers,
-- 
  //            Trygve Lunheim               | trygvelu@stud.ntnu.no 
\X//A4000/060mk2,50MB,CV64/3D,2.5G HD,f SCSII|         @pvv.org