Very Computer

I have a couple of Maxtor 11.5GB EIDE disks and I'm having trouble getting
them to work on 2.6. The disk has 22332 cyl, 16 hds, and 63 sec/trk. The
odd things is that I did manage to get it work on 2.5.1. Of course many
programs (format, fdisk) went beserk with overflows when displaying the
numbers, but the 2.5.1 ATA driver seems to understand the geometry I
explicitely put on the disk.

2.6 is more persistant. It keeps rounding it 16254 clyinders. No matter how
persistant I am with specifying 22332 cylinders. It looks like someone added
more masks to the driver.

Its clear Sun never tested large disks given the huge number of rounding
errors. Despite that, does anyone have experience with getting a large
EIDE disk to work on 2.6. Sun must be running into similar problems with
their new PC^H^HUltras.

Another interesting thing is that Solaris turned the disk controllers
around.  0x170 is c0d0 and 0x1f0 is c1d0. I have seen this once before
on 2.6 with network drivers. Confusing.

FYI: I installed all 2.6 recent patches and am up to DU3.

        Leendert

P.s. Of course, I'm being ridiculed around here. Linux just worked
out of the box with the disks :-(

--

Vrije Universiteit / Dept. of Math. & Comp. Sci.    +31 20 4447762
Paramecium project / De Boelelaan 1081A
1081 HV Amsterdam / The Netherlands

# I have a couple of Maxtor 11.5GB EIDE disks and I'm having trouble getting
# them to work on 2.6. The disk has 22332 cyl, 16 hds, and 63 sec/trk. The
# odd things is that I did manage to get it work on 2.5.1. Of course many
# programs (format, fdisk) went beserk with overflows when displaying the
# numbers, but the 2.5.1 ATA driver seems to understand the geometry I
# explicitely put on the disk.

# 2.6 is more persistant. It keeps rounding it 16254 clyinders. No matter how
# persistant I am with specifying 22332 cylinders. It looks like someone added
# more masks to the driver.

It turns out that setting the geometry using the ioctls keeps truncating
the amount of cylinders. Writting a disk label directly onto the disk
with the right geometry and rebooting fixed the problem. Apparently the
driver has no problem handling this large geometry.

        Leendert
--

Vrije Universiteit / Dept. of Math. & Comp. Sci.    +31 20 4447762
Paramecium project / De Boelelaan 1081A
1081 HV Amsterdam / The Netherlands

Quote:>It turns out that setting the geometry using the ioctls keeps truncating
>the amount of cylinders. Writting a disk label directly onto the disk
>with the right geometry and rebooting fixed the problem. Apparently the
>driver has no problem handling this large geometry.

Thanks for taking the time to follow-up with your resolution, BTW.

-Trevor

--
/--------------------------------=------------------------------------\


\--------------------------------|------------------------------------/

# >It turns out that setting the geometry using the ioctls keeps truncating
# >the amount of cylinders. Writting a disk label directly onto the disk
# >with the right geometry and rebooting fixed the problem. Apparently the
# >driver has no problem handling this large geometry.
#
# I'm curious... What specific method did you use to accomplish this?

Ah, wonder when someone would bring that up. I was hoping to trigger a
response from Sun :-)

Solaris 2.6 has a command fmthard that has an undocumented option
(-I geom_file). The geometry file is very similar to fdisk's one
except with two additions, partition offset and vtoc size. My geometry
file looks like:

* PCYL     NCYL     ACYL     BCYL     NHEAD NSECT SECTSIZ PARTOFF VTOCSIZ
  22330    22330    2        0        16    63    512     0       2

Now its just a matter of "fmthard -s vtoc -I geom /dev/rdsk/c1d0s2".

There are a number of catches. Fmthard complains about invalid VTOCs but
those refer to the one on the disk. Somehow, for my large disk, still some
truncating takes place. It can only access 20405952 blocks as opposed
to the 22508640. Given that I didn't have much time I didn't bother to
figure out where 1GB went. I was happy enough to see 10GB rather than the
2.5GB it saw before.

I wrote a little program to count blocks and look for aliases but everything
seemed fine.

        Leendert

BTW: fdisk -g /dev/rdsk/c1d0p0 returns the geometry as specified above.
Something I didn't get to work with fdisk -S.

--

Vrije Universiteit / Dept. of Math. & Comp. Sci.    +31 20 4447762
Paramecium project / De Boelelaan 1081A
1081 HV Amsterdam / The Netherlands