MTX Lock implementation question

[John Baldwin]

On Wednesday 04 March 2009 1:45:43 pm Barney Cordoba wrote:
> 
> --- On Wed, 3/4/09, John Baldwin <jhb at freebsd.org> wrote:
> 
> > From: John Baldwin <jhb at freebsd.org>
> > Subject: Re: MTX Lock implementation question
> > To: barney_cordoba at yahoo.com
> > Cc: current at freebsd.org
> > Date: Wednesday, March 4, 2009, 1:05 PM
> > On Wednesday 04 March 2009 12:34:40 pm Barney Cordoba wrote:
> > > 
> > > --- On Wed, 3/4/09, John Baldwin
> > <jhb at freebsd.org> wrote:
> > > 
> > > > From: John Baldwin <jhb at freebsd.org>
> > > > Subject: Re: MTX Lock implementation question
> > > > To: freebsd-current at freebsd.org,
> > barney_cordoba at yahoo.com
> > > > Cc: current at freebsd.org
> > > > Date: Wednesday, March 4, 2009, 10:13 AM
> > > > On Wednesday 04 March 2009 7:54:32 am Barney
> > Cordoba wrote:
> > > > > 
> > > > > Suppose the following:
> > > > > 
> > > > > Module foo.c
> > > > > 
> > > > > foo_getlock(sc)
> > > > > {
> > > > >    FOO_LOCK(sc);
> > > > > }
> > > > > foo_unlock(sc)
> > > > > {
> > > > >    FOO_UNLOCK(sc);
> > > > > }
> > > > > foo_dosomething(sc)
> > > > > {
> > > > >    MTX_LOCK_ASSERT(sc);
> > > > >    foo_dooit();
> > > > > }
> > > > > 
> > > > > Module bar.c
> > > > > 
> > > > > bar_dofoo()
> > > > > {
> > > > >    foo_getlock(sc);
> > > > >    foo_dosomething(sc);
> > > > >    foo_unlock(sc);
> > > > > }
> > > > 
> > > > This works fine.
> > > >  
> > > > > Is this something that shouldn't work? I
> > need to
> > > > access functions
> > > > >  that require locks in a different module,
> > but this
> > > > code barfs on 
> > > > > ASSERT with witness enabled. Is this a
> > deflugalty in
> > > > WITNESS, or do 
> > > > > I need to create functions within foo that
> > do the
> > > > locking? Its been
> > > > > working ok for awhile (its not a high volume
> > function)
> > > > but when I 
> > > > > fired up witness to debug something else I
> > encountered
> > > > witness panics.
> > > > 
> > > > You probably have a real bug that WITNESS is
> > warning about
> > > > however.
> > > > 
> > > > -- 
> > > > John Baldwin
> > > 
> > > It seems that theres a problem when the mutex is
> > initialized with a type
> > > of NULL. Changing it to a non-null string eliminated
> > the issue. I'm
> > > running a 7.0 base system.
> > 
> > Hmm, mutexes are required to be named.  The
> > "type" argument is optional, but 
> > the name is not.  So you can't do:
> > 
> > 	mtx_init(&m, NULL, NULL, MTX_DEF);
> > 
> > 
> > You can, however do:
> > 
> > 	mtx_init(&m, "foo", NULL, MTX_DEF);
> 
> 
> There seems to be an issue when the type is set to NULL. I got
> panics in strcpy and strlen functions when displaying warning
> traces that cleared up when the type was set to a string.

There are literally thousands of mutexes in the kernel with a NULL type.  Are 
you only seeing panics after doing a kldlunload?  Alternatively, if you are 
not destroying your mutexes using mtx_destroy() and you are using dynamically 
created strings for your mutex name that could cause what you see.

-- 
John Baldwin