Subject: Re: 1.6_BETA3/amiga 200206260000 crashed
To: None <current-users@netbsd.org>
From: Klaus Heinz <k.heinz.jun.zwei@onlinehome.de>
List: current-users
Date: 06/30/2002 23:57:30
Michael L. Hitch wrote:

>   Hmmm.  I'm running 1.6_BETA2 on my A2000/Zeus and do not see any
> problems when running an X build [other than errors in the source code -
> which I've just committed the fixes].  I'm just about ready to upgrade to
> a newer build (which will be BETA3).  It's also possible that it may be a
> 60060 specific problem (my Zeus is 68040).

After getting your changes to the X sources I started the build again.
Some time later, I saw many error messages regarding missing include files
(which existed before) and stopped the 'make' program. Now the
following happens in my shell:

  $ ls
  Segmentation fault (core dumped)
  $ su -
  Floating point exception
  $ /bin/ls
  Segmentation fault (core dumped)

On the console:
  login: root
  Shared object "nfo" not found

As I cannot do anything else, I went into ddb:

  db> sync
  syncing disks...call_sicallback: 11 more dynamic structures 62 total

and nothing else happens for at least 20 minutes. Again no dump possible :-/

Yesterday the system ran for hours and the build of X completed as far as
the assembler syntax errors let it go.

To my untrained eye this looks like problems with memory/buffer management.
Shared/buffered objects seem to get somehow mangled and the whole system
becomes unstable. Something similar also occurred to me in March with the
miniroot of Ignatios' latest snapshot for 1.5ZA, for details you can look
at PR16021.

After a reset of the machine I see fsck problems:

ffs: /dev/rsd1n (/altroot/usr): EXITED WITH SIGNAL 11
ffs: /dev/rsd1m (/altroot/var): EXITED WITH SIGNAL 11
ffs: /dev/rsd1l (/var/tmp2): UNEXPECTED INCONSISTENCY; RUN fsck_ffs MANUALLY

The last one is the slice (shared between 1.5.2 and 1.6) where the source
tree (X and 1.6) resided.
Manual fsck shows that rsd1n, rsd1m and rsd1l have many 'partially allocated
inode' errors and are in fact hosed :-/
No wonder the compiler complained about missing files...

The strange thing is: The 1.5.2 slices rsd1n and rsd1m were mounted but
not used _at all_. The 1.6_BETA3 slices seem to be ok.


ciao
     Klaus