Re: scsi trouble (SG_SET_TIMEOUT) on 7200-120 w/ 2.2.17: cdrecord 10a04

Does this happen only when you try to burn a dik in disk-at-once mode,
by any chance?  The symptoms you report sound eerily similar to what I
see on the mesh SCSI controller in my 7500 when I try to burn a disk in
disk-at-once mode.  But track-at-once mode works fine for me.

-Daniel

On Wed, Nov 15, 2000 at 12:19:02PM -0500, Richard Michael [off-list ref] wrote:

Hello Everyone,

This is regarding SCSI errors using 2.2.17, on a 7200 w/ a single 53c94
scsi controller, a sony cdu948s burner and cdrecord 10a04, compiled with
gcc 2.95.2 on a glibc 2.1.3 system.

I've exchanged email with J"org regarding this problem and he suggested
a possible compiler or kernel issue, and I'm not sure what to do now.
I don't think this is entirely off topic, but I'm sorry if this is
inappropriate traffic on this list.  I've read the archives regarding
drow's cdrecord difficulties, and this seems slightly different.

Given the above config, I run cdrecord (built w/ smake w/ -g opts to compiler
and linker) and receive this error:
------
<SNIP - normal cdrecord stuff up to here pretty much>
Waiting for reader process to fill input buffer ... input buffer ready.
Starting new track at sector: 0
Track 01:   0 of 417 MB written../cdrecord: Device not configured. Cannot set SG
_SET_TIMEOUT.
Currently running 'write_continue' command.
Condition not caught: SCSI ALREADY RUNNING !!.
Raisecond: not implemented.
Aborted
[root@chesslawn ppc-linux-cc]#
------

At this point, I can still move around the filesystem, but most commands
return I/O errors, including 'reboot', for example:
(note: the 'ls' below does complete normally, hence the <SNIP>)
------
[root@chesslawn /sbin]# reboot
bash: /sbin/reboot: Input/output error
[root@chesslawn /sbin]# ls -l
ls: ldconfig: Input/output error
ls: sln: Input/output error
ls: chkconfig: Input/output error
ls: install-info: Input/output error
ls: badblocks: Input/output error
ls: debugfs: Input/output error
ls: service: Input/output error
ls: pwdb_chkpwd: Input/output error
ls: poweroff: Input/output error
ls: shutdown: Input/output error
ls: sulogin: Input/output error
ls: telinit: Input/output error
ls: getty: Input/output error
ls: uugetty: Input/output error
total 3865
-rwxr-xr-x   1 root     root         4908 Apr 18  1999 accton
-rwxr-xr-x   1 root     root        35876 Jan 28  2000 arp
-rwxr-xr-x   1 root     root        58912 Jan 28  2000 cfdisk
-rwxr-xr-x   1 root     root         8024 Apr 18  1999 clock
-rwxr-xr-x   1 root     root         4280 Jan 28  2000 ctrlaltdel
<SNIP - rest of output from 'ls' normal, and back to bash prompt>
------

The machine is pretty much useless at this stage, and I can't reboot it
nicely.  The machine and the recorder must be power cycled to reset.

'cdrecord -scanbus' works properly.

I can post my kernel .config file (maybe I left something out of the config?)
and the output from strace, if they would help.

I've been using cdrecord on a variety of platforms (including x86 linux)
for quite some time, but have never seen this.  I'm quite certain that the
scsi devices are all properly terminated, etc. (but I'll triple check :).

Any comments?

Thanks everyone.
Cheers,
Richard

--
Daniel E. Eisenbud
eisenbud@cs.swarthmore.edu

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/