openldap and bdb

[neys <neys%neys.org@localhost>]

dear list,

i try to setup netbsd 4.0 with openldap on an athlon 64. first step after 
installing the basesystem was to install openldap over pkgsrc.current.

systeminfos:

AMD Athlon 64 X2 Dual Core Processor 5600+
NetBSD 4.0 amd64
OpenLDAP 2.4.11
BDB 4.6.21

the compile runs fine, no errors. after importing some data (with slapadd) 
and setting the correct file ownerships and permissions on 
/var/openldap/openldap-data, slapd can be startet without errors. two 
problems occur:

- ldapsearch over ldapi:/// does not work at all
ldapi is not accessible. ldapsearch says: ldap_sasl_bind(SIMPLE): Can't 
contact LDAP server (-1)

slapd logfile with loglevel 255 reads [snip from last lines]:
[SNIP]
Sep 11 13:39:37 [HOSTNAME] slapd[4322]: connection_get(15)
Sep 11 13:39:37 [HOSTNAME] slapd[4322]: connection_get(15): got connid=0
Sep 11 13:39:37 [HOSTNAME] slapd[4322]: connection_read(15): checking for 
input on id=0
Sep 11 13:39:37 [HOSTNAME] slapd[4322]: ber_get_next on fd 15 failed 
errno=34 (Result too large or too small)
Sep 11 13:39:37 [HOSTNAME] slapd[4322]: connection_read(15): input error=-2 
id=0, closing.
Sep 11 13:39:37 [HOSTNAME] slapd[4322]: connection_closing: readying conn=0 
sd=15 for close
Sep 11 13:39:37 [HOSTNAME] slapd[4322]: connection_close: conn=0 sd=15
Sep 11 13:39:37 [HOSTNAME] slapd[4322]: daemon: removing 15
Sep 11 13:39:37 [HOSTNAME] slapd[4322]: daemon: activity on 1 descriptor
Sep 11 13:39:37 [HOSTNAME] slapd[4322]: daemon: waked
[SNIP]


- ldapsearch over ldap://localhost/ does work, i get the whole database as 
output. slapd does not crash. but a ldapsearch with filter (e.g. 
objectClass=account) causes an slapd core dump.
ldapsearch says: ldap_result: Can't contact LDAP server (-1)

slapd logfile with loglevel 255 reads [last lines]:
[SNIP]
Sep 11 13:34:50 [HOSTNAME] slapd[5100]: => slap_access_allowed: backend 
default search access granted to "[BINDDN]"
Sep 11 13:34:50 [HOSTNAME] slapd[5100]: => access_allowed: search access 
granted by read(=rscxd)
Sep 11 13:34:50 [HOSTNAME] slapd[5100]: search_candidates: 
base="o=example.com" (0x00000001) scope=2
Sep 11 13:34:50 [HOSTNAME] slapd[5100]: => bdb_dn2idl("o=example.com")
Sep 11 13:34:50 [HOSTNAME] slapd[5100]: => bdb_filter_candidates
Sep 11 13:34:50 [HOSTNAME] slapd[5100]:     AND
Sep 11 13:34:50 [HOSTNAME] slapd[5100]: => bdb_list_candidates 0xa0
Sep 11 13:34:50 [HOSTNAME] slapd[5100]: => bdb_filter_candidates
Sep 11 13:34:50 [HOSTNAME] slapd[5100]:     OR
Sep 11 13:34:50 [HOSTNAME] slapd[5100]: => bdb_list_candidates 0xa1
Sep 11 13:34:50 [HOSTNAME] slapd[5100]: => bdb_filter_candidates
Sep 11 13:34:50 [HOSTNAME] slapd[5100]:     EQUALITY
Sep 11 13:34:50 [HOSTNAME] slapd[5100]: => bdb_equality_candidates (objectClass)
Sep 11 13:34:50 [HOSTNAME] slapd[5100]: => key_read


i have no idea how to analyse a core dump and i am not so good with ktrace, 
but tried it. ktrace output doesnt give me a hint.

i read something about db4 and mutexes, but adding
CONFIGURE_ARGS+= --with-mutex=x86/gcc-assembly
in the db46 makefile did not solve the problem.

i also tried pkgsrcQ1 from 2008 and binary packages from
ftp://ftp.netbsd.org/pub/pkgsrc/packages/NetBSD/amd64/4.0/databases/ and an 
different machine (intel core2 quad), but same issues. on the intel machine 
everything works fine with netbsd 4.0 i386 (could not try that on the athlon).


is anything known in this direction? if wanted i can send detailed logfiles 
from whatever you need, but i dont wanted to spam. please give me some hints 
on what to do.


thank you very much in advance,

neys