Re: problem: are floppies from outer space?

["ali (Anders Lindgren)" <dat94ali%ludat.lth.se@localhost>]

On Mon, 8 Feb 1999, ali (Anders Lindgren), which would be me, wrote:

[snip]

More hard-resets, more info:

> simple data corruption - randomly, the kernel will hang indeterminately
> until the box is hard-reset(!) when trying to mount an 1.44MB MS-DOS
> floppy (mount /dev/fd0b /floppy), seemingly with a 15-20% probability.
          ^^^^^^^^^^^^^^^^^^^^^^^
This in fact says "mount_msdos /dev/fd0b /floppy", everything else is
just a figment of your imagination. ;-)

Although yesterday I peaked out at a 50% chance of failure.

>   The only other clue I can give about it is that I suspect (but can't
> confirm) that the failures only happened after I tried to rm a (seemingly)
> corrupt file on one of the floppies (a zero-sized tarball, corrupt fs).

I can now confirm that this is *not* the case. It happens anyway, no
matter the state of the fs, and no matter what I do to the floppy.

>   The nature of the failures are as follows; "mount" will simply hang,
> doing nothing and not responding to signals (blocked in syscall I
> suppose). All other processes will continue to run, until any IO occur
> when they will instead also be deadlocked. Needless to say, it's hard to

Updated clue: most (all?) processes will continue to run fine, until
something IOish happens that involves the _floppy-device_. This can be
sync, fsck, umount, ls /floppy or just anything that causes any attempt
to retrieve information about the floppy-device or its file-system. If I
stay away from the floppy device I'm fine. ps aux reports mount_msdos
as in state D+ (until it gets swapped out => DW+), but the ``... or other
short-term uniterruptible wait.'' is permament.

When anything involving the floppy-device happens, the entire system
deadlocks. 

> do anything without causing IO to occur. :-) On a few occasions I was able
> to su to root and halt the system, but usually any IO will deadlock the

su to root works fine (in an other xterm -- naturally I was already root
in the shell used to mount the floppy, but that deadlocks...), but halting
the system does not, since it involves syncing and unmounting the
not-yet-finished-mounting floppy, and the kernel therefore deadlocks
during shutdown as well.

-- 
/ali: Computer Science Major and aspiring cartoonist. :-) 
(dept) dat94ali%ludat.lth.se@localhost - http:// www.ludat.lth.se/~dat94ali
* a4ooo/o4o/18Mb/1,5Gb/OS3.o/Ariadne/Picasso2+ - cogito, amiga sum *