On Fri 29-07-16 16:14:10, Michael S. Tsirkin wrote: > On Fri, Jul 29, 2016 at 08:04:22AM +0200, Michal Hocko wrote: > > On Thu 28-07-16 23:41:53, Michael S. Tsirkin wrote: > > > On Thu, Jul 28, 2016 at 09:42:33PM +0200, Michal Hocko wrote: [...] > > > > and the reader would hit a page fault > > > > + * if it stumbled over a reaped memory. > > > > > > This last point I don't get. flag read could bypass data read > > > if that happens data read could happen after unmap > > > yes it might get a PF but you handle that, correct? > > > > The point I've tried to make is that if the reader really page faults > > then get_user will imply the full barrier already. If get_user didn't > > page fault then the state of the flag is not really important because > > the reaper shouldn't have touched it. Does it make more sense now or > > I've missed your question? > > Can task flag read happen before the get_user pagefault? Do you mean? get_user_mm() temp = false <- test_bit(MMF_UNSTABLE, &mm->flags) ret = __get_user(x, ptr) #PF if (!ret && temp) # misses the flag The code is basically doing if (!__get_user() && test_bit(MMF_UNSTABLE, &mm->flags)) so test_bit part of the conditional cannot be evaluated before __get_user() part is done. Compiler cannot reorder two depending subconditions AFAIK. -- Michal Hocko SUSE Labs -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>