On Mon, Oct 14, 2002 at 10:31:03PM -0400, Daniel Jacobowitz wrote: > On Tue, Oct 15, 2002 at 12:09:56PM +1000, Herbert Xu wrote: > > On Mon, Oct 14, 2002 at 07:01:46PM -0400, Daniel Jacobowitz wrote: > > > > > > > My copy of C99 says: > > > > > > > > 5 The least requirements on a conforming implementation are: > > > > - At sequence points, volatile objects are stable in the sense > > > > that previou > > > > s accesses are > > > > complete and subsequent accesses have not yet occurred. > > > > > > Only volatile objects are required to be stable. I believe that if "j" > > > > That's right. However, there is a sequence point between i++ and > > j=6, so the previous access to i should be completed at that point. > > But the value of J is not required to coordinate with any sequence > points in the implementation, only in the abstract machine... > > > I have (6.7.3 #6) > Furthermore, at every sequence point the value last stored > in the object shall agree with that prescribed by the > abstract machine, except as modified by the unknown factors > mentioned previously.114) What constitutes an access to an > object that has volatile-qualified type is implementation- > defined. > > But that paragraph explicitly applies only to volatile qualified types.
BTW, look at asm/spinlock.h: this is why there are __volatile__ markers on the asm, and a "memory" clobber. The clobber prevents the following instructions from being moved before the spinlock is taken. It's worse than that: on a machine with out-of-order memory accesses, your testcase would obviously lose unless GCC inserted explicit memory barriers, and I don't believe it's ever required to do that. -- Daniel Jacobowitz MontaVista Software Debian GNU/Linux Developer
