On Mon, Oct 21, 2019 at 03:47:59PM +0200, Peter Zijlstra wrote: > On Mon, Oct 21, 2019 at 03:19:48PM +0200, Christian Brauner wrote: > > On Mon, Oct 21, 2019 at 03:11:49PM +0200, Peter Zijlstra wrote: > > > > By popular request; here's that alternative. Completely untested :-) > > > > Am I not getting some mails? :) > > You're not on the 'right' IRC channels :-) Well, we saw that that wasn't true today. :) > > > I prefer this one as it allows us to avoid working around this in > > usercopy.c. Should especially make if this potentially helps in other > > cases as well? > > That was Josh's argument too. > > Personally I think GCC is being a moron here, because with value range > analysis it should be able to prove the shift-UB cannot happen (the < > sizeof(unsigned long) conditions on both), but alas, it emits the UBSAN > calls anyway. Ok, so I take it you route that patch somehwere through tip? I'm happy with the ubsan fix: Acked-by: Christian Brauner <christian.brauner@xxxxxxxxxx>
