Hello,

I am using a Linux server (Fedora Core 2) and am experiencing problems with a SCSI library.

Due to an unknown SCSI problem, a timeout occurs on the SCSI and the Linux server tries to fix the problem performing the following steps:
1. aborts the SCSI command that has timed out
2. attempts to reset a SCSI device
3. attempts to reset a SCSI bus reset
4. attempts to reset a SCSI host bus adapter

If all these fail, the device is then set "offline" which means that it is no longer accessible.

I tried the sg_reset command to bring the device accessible but it does work. The device remains offline.
The only working "reset" is to reboot the server.

Does someone know if there is a possibility to have the device back accessible (online) without having to reboot the server.

Thanx

what hardware (controller + drive) is this happening on?

sometimes drives need a long time to actually respond to a reset so maybe raising the bus reset delay can help.

if your kernel is compiled with scsi logging support you're able to raise logging verbosity by issuing a command to the driver - have a look into your driver dokumentation/source code about this.

sl mritch.

Hi,

The hardware is a standard off-the-shelf Server with an Adaptec AHA-2944UW driving an IBM 3583 library.

The timeout error itself does not worry me too much.
What worries me is the fact that when a timeout occurs, the device goes offline and becomes unavailable. I haven’t found yet how to bring the device back online with a command/program.

Currently I must reboot the Server, which is not very funny when the problem happens over the weekend and my backups are not performed.

I designed a driver to drive the library and when a timeout error occurs, the library becomes unavailable. Any reset I tried does not re-enable the device.

After a timeout, here is what happens (the device is still there but it is unavailable):

# cat /proc/scsi/scsi
Attached devices:
Host: scsi0 Channel: 00 Id: 06 Lun: 00
Vendor: IBM Model: ULT3583-TL Rev: 2.80
Type: Medium Changer ANSI SCSI revision: 02

I have a little program that sends a TetUnitReady to the device:
testscsi <dev> <LUN> <command> where the command is a TestUnitReady

# ./testscsi /dev/sga 0 0

[E] Unable to open SCSI device /dev/sga, errno=6 (scsilib/SCSIOpenDevice).
[E] Unable to connect to device /dev/sga (testscsi/main).

I get back an error 6 which means the following:

#define ENXIO - 6 - No such device or address

According to some Linux doc I found on internet linux forums (http://www.linuxforum.com/linux-scsi/x215.html), putting the device offline may be a normal behaviour, but it does not help me.

Anyway, if someone has a “workaround” to reset the device online without rebooting, this would be great.

RX100

has this just started or have you set up the system lately? is it random? does it continue spinning?
you can try the scsi-tools to down/up the disk, but as even a host reset doesn't get the drive back online i dont think it will be any help.

check/change cables helped me a few times now solving scsi-probs.
if modern drives get too hot they are able to offline themself.
anyway iirc errorcode 9 would be timeout - don't know about 6.
i'd suggest to enable verbose scsi logging and check if you can get more info on this issue.

sl mritch.