On Fri, 16 Aug 2019, Joel Fernandes wrote: > On Fri, Aug 16, 2019 at 3:19 PM Alan Stern <st...@rowland.harvard.edu> wrote: > > On Fri, 16 Aug 2019, Mathieu Desnoyers wrote: > > > > > If you choose not to use READ_ONCE(), then the "load tearing" issue can > > > cause similar spurious 1 -> 0 -> 1 transitions near 16-bit counter > > > overflow as described above. The "Invented load" also becomes an issue, > > > because the compiler could use the loaded value for a branch, and re-load > > > that value between two branches which are expected to use the same value, > > > effectively generating a corrupted state. > > > > > > I think we need a statement about whether READ_ONCE/WRITE_ONCE should > > > be used in this kind of situation, or if we are fine dealing with the > > > awkward compiler side-effects when they will occur. > > > > The only real downside (apart from readability) of READ_ONCE and > > WRITE_ONCE is that they prevent the compiler from optimizing accesses > > to the location being read or written. But if you're just doing a > > single access in each place, not multiple accesses, then there's > > nothing to optimize anyway. So there's no real reason not to use > > READ_ONCE or WRITE_ONCE. > > I am also more on the side of using *_ONCE. To me, by principal, I > would be willing to convert any concurrent plain access using _ONCE, > just so we don't have to worry about it now or in the future and also > documents the access.
By that argumentation we need to plaster half of the kernel with _ONCE() and I'm so not looking forward to the insane amount of script kiddies patches to do that. Can we finally put a foot down and tell compiler and standard committee people to stop this insanity? Thanks, tglx