Date: Mon, 4 Mar 96 0:28:46 +0000 From: Andrew.Gordon@net-tel.co.uk To: freebsd-stable@freebsd.org Subject: Re(2): panic: inconsistent empty queue Message-ID: <"359-960304003140-C1B5*/G=Andrew/S=Gordon/O=Net-Tel Computer Systems Ltd/PRMD=Net-Tel/ADMD=Gold 400/C=GB/"@MHS> In-Reply-To: <"SunOS:21708-960303190858-7A28*/DD.RFC-822=owner-stable(a)freebsd.org/O=internet/PRMD=NET-TEL/ADMD=GOLD 400/C=GB/"@MHS>
next in thread | previous in thread | raw e-mail | index | archive | help
> >I have duplicated this crash here running a 2 week old stable > >during a make world, except I don't get the scsi timeout, I drop > >right into a panic: getnewbuf: inconsistent AGE queue, qindex = 2, > >with a similiar traceback. > > > >Reverting to 2.1-RELEASE on the system eliminated the problem :-(. > > > >I do not see this problem on any of the PCI systems I have been > >building so this may be related to the EISA changes and only effecting > >EISA systems. > > I've thought this too, but I can't understand how a change in the probe > code could cause a buffer problem many hours down the line. I have a > bt747, aha1742, aha2742, and a aha2842 in my machine here, and as soon as > my disk for 2.1-stalbe returns, I'll try to track this down. Can you > repro this in a -current environment? The eisa code is very similar. I have similar symptoms on a couple of systems here (running -current as of CTM delta src-2.1.0046.gz): - This machine (486/66 VLB with 2842 SCSI) fails an hour or so into 'make world' if running -stable kernel; completes OK if running kernel.GENERIC from 2.1R - Another machine (P166 PCI, 3940 SCSI) worked OK with a corresponding -stable kernel. - I am not always seeing panics; more often partial or complete latch-ups. In the most interesting case, I was running X with a make world in one xterm that had been running overnight, but appeared to have stalled (no disc activity). The shells in the other xterms were perfectly happy, and I could ps -ax to see what was going on, 'ls' various other directories etc. However, when I tried to 'ls' in the directory where the 'make world' appeared to be working, that process hung up too. Of course, the fact that it appears to be EISA/VLB versus PCI may be a red herring, since (at least in my case) the PCI systems tend to have faster CPU and so the timing will be subsantially different...