On Thu, 6 Mar 1997, Bill Sommerfeld wrote:

> > #define mutex_init(m,s)		(m).spl=(s)
> > 
> > #define mutex_enter(m)		(m).oldspl = splraise((m).spl)
> > #define mutex_exit(m)		splx((m).oldspl)
> > 
> > You will need to replace all the spl stuff with this.  Then you can
> > start worrying about isolating critical sections that are not currently
> > protected due to our single-threaded kernel design.  
> 
> The problem with this implementation is that `oldspl' needs to be
> stuffed in a temporary variable with a scope covering the lock/unlock,
> not in a global variable..
> 
> i.e., it's currently legal to do:
> 
> 	x = splfoo();
> 	y = splfoo();
> 
> 	splx(y);
> 	splx(x);
> 
> Nesting your mutex_enter/mutex_exit, on the other hand, leave things
> locked on exit...

While you are correct, in most mutex implementations I am familliar with
it is illegal to attempt to acquire the same mutex more than once, so this
is not really an issue.  So we make the following changes:

#define mutex_enter(m) \
             do { \
		if ((m).oldspl) \
			panic("Recursive mutex enter"); \
		(m).oldspl = splraise((m).spl); \
	     } while(0)
#define mutex_exit(m)  \
	      do { \
		int s = (m).oldspl; \
		(m).oldspl = 0; \
		splx(s); \
	      } while (0)

Then it'll behave just like Solaris 8^)


=========================================================================
Eduardo Horvath				eeh@btr.com
"Cliffs are for climbing.  That's why God invented grappling hooks."
					- Benton Frasier