Re: partion code doesn't consider 2048 bytes sectors

Paul Slootman (paul@wau.mis.ah.nl)
Wed, 17 Sep 1997 16:21:17 GMT

Gerard Roudier <groudier@club-internet.fr> wrote:
>On Mon, 15 Sep 1997, Paul Slootman wrote:
>
>> I recently tried a SCSI disk formatted to 1K sectors. It looked
>> like it worked (fdisk seemed to be OK), but accessing partitions
>> after this failed with all sorts of horrifying errors. Before
>> you say, "Aha, partitions!", when I replaced the NCR810 SCSI
>> controller with an Adaptec 1542CF, it worked perfectly. I think
>> it also works with my Buslogic card.
>>
>> So, I think there is some problem with the NCR810 code.
>> You don't mention what SCSI adapter you're using?
>
>To *think* is always allowed, but did you first *do* the following for
>*each* controller and *each* possible secteur size?
>
>1 - Fill the medium with all *zeros* (cat /dev/zero >/dev/sdX (replace X
> by the driver letter).

Yup (well, dd if=/dev/zero of=/dev/sdc bs=1024k count=10; I didn't want
te spend too much time on this; this should be enough, right?)

>2 - Partition the disk using *Linux*.

You mean there's something else out there? ;-) Yes, I used Linux's
fdisk.

>3 - Check if the geometry is coherent using *fdisk*.

fdisk seemed to be satisfied with its work; a reboot also made it
clear the kernel could identify the partitions I created.

>4 - Check your SCSI bus and terminators.

I'm religious about that, I've even replaced the on-card resistor packs
on my 810 with an active terminator that plugs into the external
connector.

>5 - Run your tests.
>
>Low-level drivers are not concerned by sector size. If the problem seems
>to be due to the sector size, you should suspect another kernel module.

I find it suspicious that the exact same setup (i.e. kernel, hardware,
cabling, etc.) doesn't work with 1K sectors on a NCR810, while replacing
the controller by an Adaptec 1542CF or a BusLogic works perfectly. I
don't believe that the 810 is faulty either, as using an otherwise
identical disk but now with 512 byte sectors does not give any problems.

I'm inclined not te believe that the problem is somewhere else in the
kernel, if simply replacing the host adapter fixes the problem.
Besides, I should hope that the low-level drivers _are_ concerned with
sector size; they have to transfer the data, right? It's useful if
they know how much to transfer...

Paul Slootman

-- 
Murphy Software, Enschede, The Netherlands      | Most people would die
email work: paul@wau.mis.ah.nl / paul@murphy.nl | sooner than think --
email home: paul@wurtel.demon.nl                | in fact, they do so.
      http://www.wurtel.demon.nl                |    - Bertrand Russell