Re: Mangled file system directory panic

[christos%zoulas.com@localhost (Christos Zoulas)]

On Aug 12,  6:51pm, jarle%uninett.no@localhost (Jarle Greipsland) wrote:
-- Subject: Re: Mangled file system directory panic

| christos%astron.com@localhost (Christos Zoulas) writes:
| > In article <20160811.153708.78150616.jarle%uninett.no@localhost>,
| > Jarle Greipsland  <jarle%uninett.no@localhost> wrote:
| >>> The panic messages reported by crash is something like:
| >>> System panicked: /mnt: bad dir ino 42369 at offset 560: NUL in name []
| >>i=0, namlen=4
| [ ... ]
| >>I have since managed to pin-point where things started to go
| >>wrong -- CVS commit-wise.  A -current kernel from CVS with tag
| >>'2016.04.03.03.00.00' works just fine, while a kernel with tag
| >>'2016.04.03.07.00.00' panics.  The only significant commit in
| >>that interval is a change of compiler for the i386 port to GCC
| >>5.3 (http://mail-index.netbsd.org/source-changes/2016/04/03/msg073782.html).
| >>
| >>Are there known bugs in our compiler for (plain) i486 systems?
| >>
| > 
| > Please try compiling ufs_lookup.c with -O0...
| I have now tried a kernel with the following lines in the config
| file:
| makeoptions     CPUFLAGS="-march=i486 -fno-tree-vrp"
| makeoptions     "CPUFLAGS.ufs_lookup.c"+="-O0"
| makeoptions     DEBUG="-g"      # compile full symbol table
| 
| It still panics.

Thanks! We are back to the drawing board on this one. The way I debugged
this the last time is by having 2 object directories compiled with the two
different compilers, and bisecting the working and the non-working object
files until I found the guilty one.

christos