On Thu, 2013-03-21 at 18:08 +0100, Oleg Nesterov wrote: > OK... since nobody volunteered to make a patch, what do you think about > the change below? > > It should "fix" atomic_add_unless() (only on x86) and optimize > atomic_inc/dec_unless. > > With this change atomic_*_unless() can do the unnecessary mb() after > cmpxchg() fails, but I think this case is very unlikely. > > And, in the likely case atomic_inc/dec_unless avoids the 1st cmpxchg() > which in most cases just reads the memory for the next cmpxchg(). > > Oleg.
Hmm, cmpxchg() has different effect on MESI transaction, than a plain read. Given there is a single user of atomic_inc_unless_negative(), (get_write_access(struct inode *inode) ) maybe the 'hint' idea used in atomic_inc_not_zero_hint() could be used. (The caller might know what is the expected current value, instead of using atomic_read() and possibly not use appropriate transaction) -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
