From owner-freebsd-hardware Tue Dec 19 15:38:38 1995 Return-Path: owner-hardware Received: (from root@localhost) by freefall.freebsd.org (8.7.3/8.7.3) id PAA15554 for hardware-outgoing; Tue, 19 Dec 1995 15:38:38 -0800 (PST) Received: from Sysiphos (Sysiphos.MI.Uni-Koeln.DE [134.95.212.10]) by freefall.freebsd.org (8.7.3/8.7.3) with SMTP id PAA15532 for ; Tue, 19 Dec 1995 15:38:10 -0800 (PST) Received: by Sysiphos id AA04096 (5.67b/IDA-1.5 for hardware@freebsd.org); Wed, 20 Dec 1995 00:26:48 +0100 Message-Id: <199512192326.AA04096@Sysiphos> From: se@zpr.uni-koeln.de (Stefan Esser) Date: Wed, 20 Dec 1995 00:26:47 +0100 In-Reply-To: Dmitry Kohmanyuk "PCI NCR assertion failed on disk access" (Dec 13, 0:54) X-Mailer: Mail User's Shell (7.2.6 alpha(2) 7/9/95) To: Dmitry Kohmanyuk Subject: Re: PCI NCR assertion failed on disk access Cc: hardware@freebsd.org Sender: owner-hardware@freebsd.org Precedence: bulk On Dec 13, 0:54, Dmitry Kohmanyuk wrote: } Subject: PCI NCR assertion failed on disk access } } hi folks, } } I have a SiS PCI motherboard w/builtin IDE and NCR PCI (53c810). } There are 2 IDE HDs installed, one on each controller (for performance } reasons). (they are Quantum Trailblazer and Quantum Fireball if this matters). } } I have a newly bought Conner 1G SCSI-2. Tested in other machine (PCI UMC) under } 2.1-RELEASE, same card (the system ran just from that drive, no IDE was } installed). } } When I install it, and boot my 2.0.5-RELEASE (haven't upgraded yet, would } ASAP), it recognizes all the drives. } } I've disklabeled and newfs'ed my new sd0 just fine. } } When I start to `cd / ; pax -rw -X -pe -t . /mnt/root' (just for example), } in plain single-user, nothing but a shell running before, } it starts, then, after some short time copying files, I got: } } assertion "cp == np->header.cp" failed, line 5235 } assertion "cp" failed, line 5236 } ERROR (80:100) (7-a3-0) (8/13) @ b5c:720071d Hmmm, this looks a little strange ... The NCR asserts control signals corresponding to a MESSAGE IN phase, but the drive wants to deliver the last command's status. The NCR's registers don't indicate any special problem: dstat: 80 = DMA FIFO empty sist: 100 = handshake timeout This is not a curent driver, so I've got a little problem to locate the exact place of failure. The NCR command 0x07...... is an ADD with Carry, but this command does not a occur a single time in the driver ... I'd suppose some memory coherency problem, but this is hard to verify. } this is from my paper notes (the machine is at work, I write this from } home), but I hope I haven't made any mistake in numbers. Well, if the command was an 0x70......, then there are exactly two places in the NCR code, that could have been active at the time of the fault. } Here are the code lines in question: } } static void ncr_int_ma (ncb_p np) } { } [...] } /* } ** verify cp } */ } dsa = INL (nc_dsa); } cp = &np->ccb; } while (cp && (vtophys(&cp->phys) != dsa)) } cp = cp->link_ccb; } } assert (cp == np->header.cp); } assert (cp); } if (!cp) } return; Well, it doesn't help to know where the host CPU detected the premature termination of the command (which was marked as complete by the timeout code, since there was no progress for 1.6 seconds), but which code the NCR was executing at the time of the SCSI bus freeze ... } my simple question is: } } is this motherboard, RAM, PCI card, hard drive, } or (I hope) the driver? Well, can't really tell. Please try a more recent driver. There are some changes in 2.1R, which should help the NCR recover from such SCSI bus timeouts in more cases (it sometimes failed to reset the SCSI bus, before). } It seems that all works fine _until_ I access both IDE and SCSI. } It that the problem? That really points at some hardware problem, IMHO. There could be PCI bus arbitration problems, or cache coherency problems. There is some EIDE chip (RZ-1000) that is known to cause problems (it is found on most current motherboards, AFAIK). } I have read some guys' report in relcom newsgroup telling just the same: } NCR works fine but with IDE. Well, never heard such a thing before. But then, I don't actually know anybody running a mixed IDE/SCSI system. Regards, STefan -- Stefan Esser, Zentrum fuer Paralleles Rechnen Tel: +49 221 4706021 Universitaet zu Koeln, Weyertal 80, 50931 Koeln FAX: +49 221 4705160 ============================================================================== http://www.zpr.uni-koeln.de/~se