[215] in linux-scsi channel archive
Re: SCSI disk SPIN-UP on pwr-up ?
daemon@ATHENA.MIT.EDU (Pascal Haible)
Fri May 26 14:13:16 1995
To: submit-linux-dev-scsi@ratatosk.yggdrasil.com
From: haible@izfm.uni-stuttgart.de (Pascal Haible)
Date: 26 May 1995 16:51:43 GMT
In article <9505191357.ZM25227@aib.com>, Eric Youngdale <eric@aib.com> wrote:
>On May 19, 1:07pm, Rob Janssen wrote:
>> Subject: Re: SCSI disk SPIN-UP on pwr-up ?
> Most scsi drives have a jumper that indicates whether the drive
>should spin up at power up. The main point of doing this is that
>on large systems with lots of disks, the power supply is loaded fairly
>heavily if all of the disks attempt to spin up simultanously.
>
[...]
> Yes. After we have detected the disks themselves, send the
>TEST_UNIT_READY command to see whether the disk is up or not. If not,
>we send the START_STOP command which instructs the drive to spin up.
>Again, we do this one drive at a time to minimize power supply loading.
[...]
> I added it, but I never had a way of testing it. Some people
>reported that it worked - my guess is that if the sense code is
>UNIT_ATTENTION, we just need to retry the TEST_UNIT_READY.
It works, but I recently had problems with it. As my new Quantum
hd produces a nasty vibration, I wanted it to spin up only when I really
need it, i.e. when I boot linux. But the 50 seconds timeout is too small,
it actually needs 70 seconds(!) to report ready. Unfortunately, the
error recovery in the case of the timeout is - hmm - unsufficient,
the kernel dies a strange death.
There are 2 ways to solve this - let it spin up by the SCSI controller,
or patch the kernel to > 70s timeout; I did the latter, and I suggest
making this change permanent.
On a side note, I had other kernel boot crashes when I played around
partitioning the hd. The code that checks the partition tables takes
the 15(16?) devices per disk serious, but doesn't check it:
for testing I had created 20 logical partitions inside an extended
partition (fdisk didn't object), and the kernel crashed after
printing 'sda15'.
With drives getting Gigs large, this limitation is becoming serious.
Couldn't the PC partitioning scheme (including logical partitions)
go further?
Pascal
--
Dipl.-Ing. Pascal Haible haible@IZFM.Uni-Stuttgart.DE
