open() and close() from multithread application

Hello,

I have application with three threads, second and third threads are 
created with pthread_create().
In kernel module my d_open and d_close handlers contains reference count 
incrementing in d_open and decrementing in d_close handler.
All is right in several first sequences open(), ioctl(), close() .
These sequences can be called from all three threads simultaneously.
After some time module comes to state that d_close is never reached when 
called close(), I checked close() in user-mode debugger and it returns 
always 0.
I am sure that d_close is not called I don't see my printf() output,
but I see all another d_open calls, and when checked my reference count
it is on every d_open incremented and never decremented.

What to investigate to find why d_close handler is not called ?
I use kernel with enabled DIAGNOSTIC, DEBUG, LOCKDEBUG.