On Thu, Jul 05, 2018 at 10:09:49PM +0200, Imre Palik wrote: > Hi All, > > first of all, thanks for the great book. On behalf of all of its contributors, glad you like it! > I have some doubts about some of the counting examples. Namely, those > that implement inc_count() as a simple increment operation. If there > is nothing to restrain the compiler/linker, they can possibly merge > several increments together, or optionally keep the counter value in a > register. > > Of course, this won't happen with counttorture, because of the > barrier() in the loop. But I wonder if that barrier() should be part > of inc_count() instead. What do you think? I actually put this on my todo list as a results of the C++ discussion, but given that I am in the process of (almost) rewriting Linux-kernel RCU, it will take time to get to it. So if you are interested, I would welcome a patch that added READ_ONCE() and WRITE_ONCE as appropriate. If you are curious about the historical perspective, some of the counter code predates ACCESS_ONCE() -- let alone READ_ONCE() and WRITE_ONCE() -- in the Linux kernel. Compilers have gotten more aggressive over the decades. ;-) Thanx, Paul -- To unsubscribe from this list: send the line "unsubscribe perfbook" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
