[2018] in linux-scsi channel archive
Re: Scsi removable disk patch
daemon@ATHENA.MIT.EDU (Leonard N. Zubkoff)
Mon Jun 16 03:44:24 1997
Date: Mon, 16 Jun 1997 00:20:53 -0700
From: "Leonard N. Zubkoff" <lnz@dandelion.com>
To: dormouse@farsrobt.demon.co.uk
CC: linux-scsi@vger.rutgers.edu
In-reply-to: <199706160706.IAA22596@farsrobt.demon.co.uk> (message from
Richard Waltham on Mon, 16 Jun 1997 08:06:09 +0100 (BST))
From: Richard Waltham <dormouse@farsrobt.demon.co.uk>
Date: Mon, 16 Jun 1997 08:06:09 +0100 (BST)
Thank you. You just made my day:)
However there is an occasional timeout with my original patch, failing about
1 in 47 spin ups, using the following script running for about 10 hours.
while (true) do { /sbin/e2fsck /dev/sdb3 ; jazsleep 1 /dev/sdb ; /usr/bin/sleep 75; } done
e2fsck /dev/sdb3 spins up the drive but does nothing as the file system is
clean. jazsleep is a little utility I wrote to set the spin down time for
the jaz drive. Runs here in case there is a timeout as any reset puts the
spin down time back to the default 30 minutes. script then sleeps to give
the drive time to spin down. This runs about 43-44 times an hour.
I couldn't figure at first why the sd.c 15 second timeout was timing out in
9 seconds but finally traced this to a seperate timeout value specified in
scsi_ioctl.c which is only 9 seconds as opposed to the 15 second timeout in
sd.c that I expected to be used. The scsi code is certainly a bit of a
tangled web with various different timeout values in different parts of the
code:)
Changing the timeout in scsi_ioctl.c to 15 seconds as in sd.c and re-running
the script for about 12 hours over night and some of today was successful
with no timeouts.
Attached is an updated patch against 2.0.29 and an almost identical patch
against 2.1.42. It appears blank lines in sd.c in 2.0.29 have trailing
spaces, where as sd.c in 2.1.42 does not. How the patch applied to both
before is a mystery.
Actually, the START_STOP timeout really needs to be more than a standard disk
I/O timeout, since some disks can take quite a while to spin up. ioctl_command
already had a 60 second timeout for START_STOP, but that code isn't run except
for the SCSI_IOCTL_SEND_COMMAND case. I'll make the necessary corrections in
my working copy.
Leonard
