cvs commit: src/share/man/man9 Makefile condvar.9 lock.9 mi_switch.9 mtx_pool.9 mutex.9 rwlock.9 sleep.9 sleepqueue.9 sx.9 thread_exit.9 src/sys/kern kern_synch.c src/sys/sys mutex.h rwlock.h sleepqueue.h sx.h systm.h

[John Baldwin]

On Monday 12 March 2007 19:27, Julian Elischer wrote:
> John Baldwin wrote:
> > On Monday 12 March 2007 16:03, Pawel Jakub Dawidek wrote:
> >> On Mon, Mar 12, 2007 at 03:35:21PM -0400, John Baldwin wrote:
> >>> On Monday 12 March 2007 14:56, Pawel Jakub Dawidek wrote:
> >>>> On Mon, Mar 12, 2007 at 11:16:23AM -0400, John Baldwin wrote:
> >>>>> On Saturday 10 March 2007 15:52, Pawel Jakub Dawidek wrote:
> >>>>>> What about something like this:
> >>>>>>
> >>>>>> #define	cv_wait(cv, lock)	do {
> >>>>>> 	switch (LO_CLASSINDEX((struct lock_object *)(lock))) {
> >>>>> The problem with a cast is you use type checking.  Might as well do 
> > this:
> >>>>> #define	cv_wait(cv, lock)	_cv_wait((cv), (struct lock_object *)(lock))
> >>>> This will skip type checking and my version only cast to provide type
> >>>> checking, so when you pass some random variable it will give you an
> >>>> error.
> >>> Not really, you may pass some garbage and the LO_CLASSINDEX turns out to 
> > be a 
> >>> mutex. :)  You only get a runtime error, not a compile-time one.  
> >>> Type-checking by the compiler is nice because you get compile-time 
errors.
> >> I'll get compile-time error, because cv_wait_mtx() takes
> >> 'struct condvar *' and 'struct mtx *' as arguments. So even if some
> >> garbage returns 1, which turns out to be a mutex, call to cv_wait_mtx()
> >> will generate compile-time error.
> > 
> > Err, no, actually, yours will always give compile errors actually.  Keep 
in 
> > mind that LO_CLASSINDEX() is a run-time check.  This:
> > 
> > #define cv_wait(cv, lock)       do {
> >         switch (LO_CLASSINDEX((struct lock_object *)(lock))) {
> >         case 1:
> >                 cv_wait_mtx(cv, lock);
> >                 break;
> >         case 2:
> >                 cv_wait_sx(cv, lock);
> >                 break;
> >         case 3:
> >                 cv_wait_rw(cv, lock);
> >                 break;
> >         default:
> >                 panic("Invalid lock.");
> >         }
> > } while (0)
> > 
> > Will try to pass 'lock' to three different functions, at least 2 of which 
will 
> > trigger compile errors. :)  The kernel won't choose which one to run until 
> > runtime though.  The key is that I want a compile error, not a panic(). :)
> 
> I've been asking for  awhile that for example spin and sleep mutexes should
> be different types so that we could catch those problems at compile time.

That is on my todo list actually.  Stephan and I talked at BSDCan 06 about 
various alternative strategies for spin locks.

-- 
John Baldwin