I've done a little digging on this problem and found out a few things. First of all, this only affects SPARC. The AMD64 version that I built didn't have this problem.

I also tried debugging this via dbx. Here's the result:

t@1 (l@1) signal SEGV (no mapping at the fault address) in ldap_unbind_ext at line 49 in file "unbind.c"
   49           assert( LDAP_VALID( ld ) );

(dbx) where
current thread: t@1
=>[1] ldap_unbind_ext(ld = 0x9fdf00, sctrls = (nil), cctrls = (nil)), line 49 in "unbind.c"
  [2] ldap_unbind_s(ld = 0x9fdf00), line 210 in "unbind.c"
  [3] _close_ldap_link(rsrc = 0x1009f3238), line 194 in "ldap.c"
  [4] list_entry_destructor(ptr = 0x1009f3238), line 184 in "zend_list.c"
  [5] zend_hash_del_key_or_index(ht = 0x1008a6ae0, arKey = (nil), nKeyLength = 0, h = 1U, flag = 1), line 492 in "zend_hash.c"
  [6] _zend_list_delete(id = 1), line 58 in "zend_list.c"
  [7] _zval_dtor_func(zvalue = 0x1009f31f8), line 60 in "zend_variables.c"
  [8] _zval_dtor(zvalue = 0x1009f31f8), line 35 in "zend_variables.h"
  [9] _zval_ptr_dtor(zval_ptr = 0x1009fc710), line 395 in "zend_execute_API.c"
  [10] zend_hash_apply_deleter(ht = 0x1008a6988, p = 0x1009fc6f8), line 576 in "zend_hash.c"
  [11] zend_hash_graceful_reverse_destroy(ht = 0x1008a6988), line 642 in "zend_hash.c"
  [12] shutdown_executor(), line 221 in "zend_execute_API.c"
  [13] zend_deactivate(), line 854 in "zend.c"
  [14] php_request_shutdown(0x0, 0x2000, 0x130814, 0xffffffff7d5b27b8, 0xffffffff7d6e4000, 0x1009fd080), at 0x1004dccec 
  [15] main(argc = 2, argv = 0xffffffff7ffff9f8), line 1666 in "cgi_main.c"





I put a stop in list_entry_destructor (btw this wasn't arbitrary. I'm just trying to make this story as short as I can) and took a look at what was happening to ld:

(dbx) print *ld
*ld = {
    list_dtor     = (nil)
    plist_dtor    = (nil)
    list_dtor_ex  = 0x100276a80 = &`php`ldap.c`_close_ldap_link(zend_rsrc_list_entry *rsrc)
    plist_dtor_ex = (nil)
    type_name     = 0x100887b68 "ldap link"
    module_number = 15
    resource_id   = 18
    type          = '\002'
}
(dbx) step     
t@1 (l@1) stopped in list_entry_destructor at line 184 in file "zend_list.c"
  184                                           ld->list_dtor_ex(le TSRMLS_CC);
(dbx) print *ld
*ld = {
    list_dtor     = (nil)
    plist_dtor    = (nil)
    list_dtor_ex  = 0x100276a80 = &`php`ldap.c`_close_ldap_link(zend_rsrc_list_entry *rsrc)
    plist_dtor_ex = (nil)
    type_name     = 0x100887b68 "ldap link"
    module_number = 15
    resource_id   = 18
    type          = '\002'
}
(dbx) step    
t@1 (l@1) stopped in _close_ldap_link at line 192 in file "ldap.c"
  192           ldap_linkdata *ld = (ldap_linkdata *)rsrc->ptr;
(dbx) print *ld
dbx: cannot access address 0x12
(dbx) step    
t@1 (l@1) stopped in _close_ldap_link at line 194 in file "ldap.c"
  194           ldap_unbind_s(ld->link);
(dbx) print *ld
*ld = {
    link       = 0x9fdf00
    rebindproc = (nil)
}
(dbx) step    
t@1 (l@1) stopped in ldap_unbind_s at line 210 in file "unbind.c"
  210           return( ldap_unbind_ext( ld, NULL, NULL ) );
(dbx) print *ld
dbx: cannot access address 0x9fdf00
(dbx) 



My guess is that the address 0x9fdf00 is supposed to point to an LDAP structure, but it doesn't for some reason. I'll leave this to someone else to figure out why. It looks like something is getting stepped on in Zend's hash tables.

Since I only had this problem in 64-bit SPARC, I decided to play around with some of the compiler optimizations. I found out that adding -xdepend to CFLAGS and rebuilding PHP makes this problem go away.

Quick search revealed that there really has been problems with  the SunStudio. (especially with -xdepend it seems)

Interesting. I was able to make the problem go away by adding -xdepend (my previous compilation used no optimizations). The problem also went away when I used -fast (which is a macro that includes -xdepend).

I will log a bug on SunStudio and see where it goes.