Date: Sun, 28 Sep 1997 11:01:42 +0200 From: Stefan Esser <se@FreeBSD.ORG> To: David Langford <langfod@dihelix.com> Cc: scsi@FreeBSD.ORG Subject: Re: Possible ncr problem? Message-ID: <19970928110142.45106@mi.uni-koeln.de> In-Reply-To: <199709280439.SAA00307@caliban.dihelix.com>; from David Langford on Sat, Sep 27, 1997 at 06:39:31PM -1000 References: <199709280439.SAA00307@caliban.dihelix.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sep 27, David Langford <langfod@dihelix.com> wrote: > Any thought on what this means? Yes, again the QUEUE_FULL problem of the DORS. This drive can't deal with a reasonable number of tagged commands at times, it seems. Since the generic SCSI layer will immediately re-issue the failed command, there will be no adverse effect, if it succeeds within 5 retries. The new generic SCSI code currently being prepared by Justin Gibbs will take care of this problem. > After reconfiguring my SCSI chain several times I think I can > rule bad termination. > > Bad drive could be. I only have seen these errors if I "dump" > the main partition on the drive or place my "obj" directory and do a > "make world". I have been wondering, whether the problem is caused by sending an unnecessary START_STOP_UNIT when the raw partition is opened. IBM drives did never like that ... You may want to remove the call of scsi_start_unit() from sd.c (there is only one occurance), and see whether the error messages are still printed ... > Fsck doesnt show problems and bad144 at least seemed to scan the drive without > any console messages showing up. Fsck does also open the raw partition, but at that time, there should not be any other activity. I'm guessing, that even a START_STOP_UNIT command that is a NOP because the drive motor had been started long ago, does collide with any other command on IBM drives. > assertion "cp" failed: file "../../pci/ncr.c", line 6228 > sd1: COMMAND FAILED (4 28) @f0497000. > assertion "cp" failed: file "../../pci/ncr.c", line 6228 > sd1: COMMAND FAILED (4 28) @f0497000. The 28 identifies QUEUE_FULL status, the 4 just is an indication, that processing of the command run to completion, as far as the NCR controller and the driver are concerned. > sd0: <IBM DORS-32160 S82C> type 0 fixed SCSI 2 > sd1: <IBM DORS-32160 S82C> type 0 fixed SCSI 2 There was roumor, that the non-wide DORS does not support as many tags as the wide version, though I never had a chance to confirm this myself. If you want to help debug the problem, then please try a kernel that does not start the SCSI drives in /sys/scsi/sd.c:sdopen(). Regards, STefan
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19970928110142.45106>