| From | Vitaly Kuznetsov <> | Subject | Re: [patch 10/11] x86/vdso: Move cycle_last handling into the caller | Date | Fri, 14 Sep 2018 17:26:01 +0200 |
| |
Thomas Gleixner <tglx@linutronix.de> writes:
> Dereferencing gtod->cycle_last all over the place and foing the cycles < > last comparison in the vclock read functions generates horrible code. Doing > it at the call site is much better and gains a few cycles both for TSC and > pvclock. > > Caveat: This adds the comparison to the hyperv vclock as well, but I have > no way to test that.
While the change looks obviously correct for Hyper-V vclock too, by saying that you encouraged me to give the whole series a shot. I did and turns out Hyper-V vclock is broken: simple clock_gettime(CLOCK_REALTIME, ) test goes from 70 cpu cycles to 1000 (meaning that we're falling back to syscall I guess). Bisecting now, stay tuned!
-- Vitaly
|