Re: Easy Guide / new tarball

[John Tobey <spam@john-edwin-tobey.org>]

Roland McGrath <frob@debian.org> wrote:
> > Well, it scrolled by pretty fast with the kernel messages appearing,
> > but I believe it was libdiskfs/disk-pager.c:93 `env && "unexpected
> > fault on disk image"'.  I saw this with the last hurd (19990923?) too.
> 
> Hmm.  I don't recall anyone hitting that one before.  That is certainly odd.
> Here is a hack to hopefully make it give more informative messages when it
> crashes there.  I've put up some binaries I just built using this in:
> 	http://www.ai.mit.edu/people/roland/hurd/
> if you want to give it a try without rebuilding from source.  The
> (big) non-".stripped" files there have full debugging symbols; but
> since these are my binaries, I can figure some things out even
> from backtraces with raw addresses and no symbols.

After first starting ext2fs for /jtobey, I saw some console messages
that scrolled off the screen too fast for me to read.  I was getting

    ext2fs: /dev/hd0s5: warning: set_hyper(0,0) 1 => 0

    ext2fs: /dev/hd0s5: warning: finished set_hyper(0,0) 1 => 0

    ext2fs: /dev/hd0s6: warning: set_hyper(0,0) 1 => 0

    ext2fs: /dev/hd0s6: warning: finished set_hyper(0,0) 1 => 0

every couple of seconds, but the first messages were something else.

I restarted my perl build and ran `vmstat 5'.  After 4-5 minutes,
vmstat stopped, but the console warnings about set_hyper continued.
The vmstat output was:

    (pagesize: 4K, size: 20.2M, swap size: 98.9M)
     free  actv inact wired zeroed  react  pgins pgouts  pfaults\
      cowpfs hrat swfree
[cut]
     492K 4.75M 9.42M 5.54M    +8K +1.05M +1.99M  +600K     +531\
          +0  60%  67.4M

I had two xterm processes running, one with make/gcc, the other a root
shell running vmstat.  Both were able to redisplay, but neither
responded to ^C or ^Z.  I dismissed the build window cleanly with the
window manager's "delete" command, but then the other window became
unable to redisplay itself and had to be zapped.

I was able to open another rsh/xterm and type

    # fsysopts /jtobey --readonly

whereupon the console gave your new assertion about EKERN_MEMORY_ERROR
.  I could not read the sig, sigcode, etc. before they scrolled past.

Then I did

    # fsysopts /install --readonly

with no such assertion failure, and the stream of set_hyper messages
ended with

    ext2fs: /dev/hd0s5: warning: finished set_hyper(0,0) 1 => 0x1

(note 0x1).  Next, I did

    # kill `cat /var/run/syslog.pid`
    # fsysopts / --readonly

to prepare for reboot.  fsysopts returned to the prompt.  I selected
the above two lines with the mouse to paste them into this message,
but after highlighting the lines, xterm froze, and I had to type them
in.  Now the Hurd machine is unresponsive to ping and to its own
keyboard.