Re: [RFC PATCH 4/4] x86/TSC: Use RDTSCP
From: Andy Lutomirski
Date: Wed Dec 12 2018 - 15:09:18 EST
On Wed, Dec 12, 2018 at 12:00 PM Borislav Petkov <bp@xxxxxxxxx> wrote:
> On Wed, Dec 12, 2018 at 10:50:30AM -0800, Andy Lutomirski wrote:
> > As far as I know, RDTSCP gets the job done, as does LFENCE, RDTSC on
> > Intel.
> Same on AMD when LFENCE has been made dispatch-serializing.
> > There was a big discussion a few years ago where we changed it
> > from LFENCE;RDTSC;LFENCE to just LFENCE;RDTSC after everyone was
> > reasonably convinced that the uarch would not dispatch two RDTSCs
> > backwards if the first one was immediately preceeded by LFENCE.
> Yeah, the second one won't pass the LFENCE so you won't see time going
> backwards, sure.
Also that the uarch doesn't turn:
load some memory
into a load followed a cycle later by RDTSC. If that happened, then
some cooperating CPUs could observe time going backwards.
But RDTSC has no dependencies, so, barring odd register renaming
effects or similar, this isn't going to be a problem in practice.
Also, the actual clock reading code is complicated enough that we have
several cycles of wiggle room :)