lubich@ethz.UUCP (Hannes Lubich) (08/22/90)
Hi,
is anybody using Fujitsu M2351A-Eagle disks with Emulex controllers on a Vax?
We've got such a beast (474.2 MB unformatted) on an Emulex QD32 controller
for our uVax II running Ultrix V2.0-1. We are running an Emulex QD21 with
two smaller Fujitsu disks on the same machine already without problems.
I can format and verify the Eagle (45 bad blocks) by talking to the diagnostic
software on the controller NOVRAM and I have verified that the strapping is
correctly set for 48 sectors. I've added
controller uda2 at uba0
controller uq2 at uda2 csr 0160344 vector uqintr
disk ra8 at uq2 drive 0
to the kernel config file. When booting, uq2 and ra8 are shown as being
present on the bus - no problems so far.
I'm using the standard eagle disktab entry:
eagle|Eagle|Fujitsu Eagle (48 sectors):\
:ty=winchester:ns#48:nt#20:nc#842:rm#3900:\
:pa#15884:ba#8192:fa#1024:\
:pb#66880:bb#4096:fb#512:\
:pc#808320:bc#8192:fc#1024:\
:pd#15884:bd#8192:fd#1024:\
:pe#307200:be#4096:fe#512:\
:pf#109248:bf#4096:ff#1024:\
:pg#432768:bg#4096:fg#1024:\
:ph#291346:bh#4096:fh#512:
So I've tried to run "newfs -v /dev/rra8a eagle" which produces the following
output:
/etc/mkfs /dev/rra8a 15884 48 20 8192 1024 16 10 65 2048
Warning: inode blocks/cyl group (36) >= data blocks (32) in last
cylinder group. This implies 512 sector(s) cannot be allocated.
/dev/rra8a: 15360 sectors in 16 cylinders of 20 tracks, 48 sectors
7.9Mb in 1 cyl groups (16 c/g, 7.86Mb/g, 1856 i/g)
super-block backups (for fsck -b#) at:
32,
installing boot code
sector 0 boot + 1st level boot = /usr/mdec/vaxboot
Then I try to fsck it with "fsck /dev/rra8a" which results in:
** /dev/rra8a
BAD SUPER BLOCK: TRASHED VALUES IN SUPER BLOCK
USE -b OPTION TO FSCK TO SPECIFY LOCATION OF AN ALTERNATE
SUPER-BLOCK TO SUPPLY NEEDED INFORMATION; SEE fsck(8).
I then use "fsck -b32 /dev/rra8a" which finds lots and lots of partially
allocated inodes, beginning with inode 5. When trying to clear them
using "fsck -b32 -y /dev/rra8a" they get cleared in ther first phase,
but then I get the following:
** Phase 2 - Check Pathnames
DIRECTORY CORRUPTED I=3 OWNER=root MODE=40755
SIZE=8192 MTIME=Aug 22 03:28 1990
DIR=/lost+found
SALVAGE? yes
It seems to me as if the newfs works ok (sort of) but I can't figure out
why the fsck doesn't get along with the superblock.
The problem is there for all partitions, not only for a. I've tried a smaller
number for the cylinders/group with newfs, "newfs -v -c 8 /dev/rra8a eagle"
The result looks different:
/etc/mkfs /dev/rra8a 15884 48 20 8192 1024 8 10 65 2048
Warning: 448 sector(s) in last cylinder unallocated
/dev/rra8a: 15872 sectors in 17 cylinders of 20 tracks, 48 sectors
8.1Mb in 3 cyl groups (8 c/g, 3.93Mb/g, 1216 i/g)
super-block backups (for fsck -b#) at:
32, 7760, 15488,
installing boot code
sector 0 boot + 1st level boot = /usr/mdec/vaxboot
but fsck still gives the same error "BAD SUPER BLOCK" and specifying any
of the alternate boot blocks again results in this enormous amount of
error messages about partially allocated inodes. The next phases of fsck,
however, are different:
** Phase 2 - Check Pathnames
** Phase 3 - Check Connectivity
** Phase 4 - Check Reference Counts
** Phase 5 - Check Cyl groups
CG 1: BAD MAGIC NUMBER
EXCESSIVE BAD BLKS IN BIT MAPS.
CONTINUE? yes
BAD CYLINDER GROUPS
SALVAGE? yes
** Phase 6 - Salvage Cylinder Groups
**** FILE SYSTEM MODIFIED, VERIFYING
** /dev/rra8a
BAD SUPER BLOCK: TRASHED VALUES IN SUPER BLOCK
USE -b OPTION TO FSCK TO SPECIFY LOCATION OF AN ALTERNATE
SUPER-BLOCK TO SUPPLY NEEDED INFORMATION; SEE fsck(8).
Following the advice of our disk supplier, I changed the disktab entry
from
:ty=winchester:ns#48:nt#20:nc#842:rm#3900:\
to
:ty=winchester:ns#46:nt#20:nc#842:rm#3961:\
which didn't help much.
I'd appreciate any help or pointers.
Thanks a lot in advance
--HaL
--
~ UUCP/Usenet : {known world}!mcvax!cernvax!ethz!lubich
~ or : lubich@ethz.uucp
~ CSNET/ARPA/BITNET : lubich@inf.ethz.ch / lubich%inf.ethz.ch@relay.cs.net
~ The usual disclaimer : No, it wasn't me, somebody must have used my account.hans@umd5.umd.edu (Hans Breitenlohner) (08/23/90)
We are running a setup such as yours quite successfully. (M2351A disk,
QD32 controller on a Microvax 3200, used to be a Microvax II).
First, the disktab entry for the eagle is wrong. ns should be 47 (one
spare sector per track), and nc should be 840 (2 spare cylinders).
This would give a total of 789600 sectors per disk, but somehow
only 787156 sectors are available. (2444 sectors missing, perhaps used
for block replacement tables?)
Having said this, I must stress that, as far as I know, the only penalty
for giving newfs bad information about the disk geometry would be bad
performance. Our disks were newfs'd as RA81s, and have worked fine for
years.
Have you run the data verify or the data reliability section of the
diagnostics?
My first guess would be that the disk drive is not strapped properly.
One Emulex manual states that the sector size should be 594, but the
correct number is 587.
You can determine if this is a problem with the following little
exercise:
dd if=/vmunix of=xxx bs=10k count=10
dd if=xxx of=/dev/rra8a bs=10k count=10
dd if=/dev/rra8a of=yyy bs=10k count=10
cmp xxx yyy
I would guess that you will find xxx and yyy to be different, with
the differences confined to certain sectors.
If you need more information, please send mail, and specify which
version of Ultrix, and which revision of the QD32 ROMs you have.
Hans Breitenlohner
hans@umd5.umd.edu