Hi Andy, On Mon, Jan 08, 2018 at 03:05:48PM -0800, Andy Lutomirski wrote: > On 01/08/2018 09:03 AM, Dave Hansen wrote: > > On 01/08/2018 08:12 AM, Willy Tarreau wrote: > I vote per-thread.
The per-mm approach that Linus suggested doesn't look bad either and makes quite some sense. > Anyway, we can easily sync the NX-clearing: just catch the spurious page > fault and clear the bit. Avoiding infinite loops will need a bit of > thought, but it's surely doable. That's an excellent idea, eventhough I have no idea how to implement it :-) > Or we set a per-mm flag saying "no NX", then do synchronize_sched() or > similar if we were the first to set it (or take the pagetable lock), then > clear all the NX bits. Again, needs some care, but doable. > > FWIW, the NX trick quite nicely emulates SMEP on non-SMEP hardware, which is > fantastic for Spectre resistance and general hardening. Yes I figured exactly this when I faced this protection! > Turning it off totally defeats that, which hurts a bit. I agree, that's why I'd like it to be conditional. Probably that with your idea of catching the page fault and the per-mm flag it would work quite well, but before being able to do this I still have a lot to explore :-/ > Also, Kees should be CC'd here. Yes I've added him and you (and a few others) in CC of all forthcoming patches. Sorry for not adding you initially, I simply wanted to share a quick experiment and initiate a discussion. Willy
