On Sat, Oct 07, 2006 at 09:36:09PM -0700, Jason Thorpe wrote:
> 
> On Oct 4, 2006, at 11:39 PM, Jason Thorpe wrote:
> 
> >Ok, here is a generalized (and currently untested) "specificdata"  
> >implementation, based on Yamamoto-san's.  proc, lwp, and session  
> >specific-data could be built on top of this, a'la (note this is  
> >incomplete):
> 
> I have checked this in, added the wrappers for proc and lwp, and  
> converted kern/uipc_sem.c to use the new proc_{get,set}specific() as  
> an example of how it should be used.
> 
> I will probably make a few more tweaks over the next few days, but  
> nothing too drastic.  Jury is still out as to whether or not  
> getspecific_unlocked is really a good idea (even though lwps are  
> currently using it).
> 
> -- thorpej

As I asked before:

As my upcoming file system transactions should use this interface to store
a list in lwp-context I'm missing two things:

- Specificdata_setspecific() and the functions using it should also work
  (and possibly fail) if called from a non-sleepable context.  Either a
  flag (WAIT/NOWAIT) or a special function xxx_nowait().

- It would be nice to have a static __link_set_xxx based creator so a subsystem
   may create a key without the need of an explicit init() function.  

-- 
Juergen Hannken-Illjes - hannken@eis.cs.tu-bs.de - TU Braunschweig (Germany)