Subject: Re: DPWS500/Pyxis bugs, corrupt memory etc.
To: Jason Thorpe <>
From: Olaf Seibert <>
List: port-alpha
Date: 04/11/2000 22:57:47
Sorry to get back at this so late, but I just  had the courage to re-sup
without trashing my work-in-progress on pkgsrc/cross. 

Unfortunately, the problem persists. I got a fatal kernel trap just by
repeating "disklabel -r wd0" a number of times. Some of those times the
data shown was corrupted.

Because there was already a problem *reading* the disk (which I was told
could not even be a problem) I didn't even try mounting anything or even
writing to it.

I thought a core sump would probably not be very useful, nor even a
stack backtrace, due to the nature of the corruption, but the trap
simply was within the disklabel reading code.

On Thu 24 Feb 2000 at 23:58:12 +0100, Olaf Seibert wrote:
> On Tue 22 Feb 2000 at 23:51:21 +0100, Olaf Seibert wrote:
> > 
> > I applied this patch on my 1.4.1 kernel sources. Unfortunately, as soon
> > as I did a disklabel wd0, I got a "memory management fault" or something
> > like that, so I backed it out again. Maybe it was because ithe patch did
> > not apply cleanly, or not all required files were really recompiled, or
> > something. Instead, for now I just disabled ide dme completely, and that
> > seems to work fine. At some point I'll switch to 1.4.2 anyway, and I'll
> > try -current too.
> I just tried -current, supped just a few moments ago. No go :-(.
> Mounting wd0a gives "incorrect superblock". Disklabel output seems ok.
> I tried a fsck -n, and was rewarded with a memory management fault,
> according to the trace command inside dkcksum() in readdisklabel().
> fatal kernel trap:
>     trap entry = 0x2 (memory management fault)
>     a0         = 0xfffffe0002260000
>     a1         = 0x1
>     a2         = 0x0
>     pc         = 0xfffffc000034e5a0
>     ra         = 0xfffffc00004deefc
>     curproc    = 0xfffffc0000a229b0
>         pid = 19, comm = fsck
> panic: trap
> I could arrange for a crash dump but given the nature of the pyxis bug I
> think that that may not be very helpful? Anything else I could try?
> -Olaf.
> --
> ___ Olaf 'Rhialto' Seibert - rhialto@polder.ubc.     -- If one tells the truth,
> \X/     -- one is sure, sooner or later, to be found out. (Oscar Wilde)