On Fri, Aug 25, 2017 at 11:55 AM, John Stultz <john.stu...@linaro.org> wrote: > I'll look over the code again to see if I can catch anything by > review. Worse case if we can't get any traction on this in a day or so > I'll submit a revert.
I think I found the issue. In tk_update_ktime_data() I add the raw_sec and shifted down tk->tkr_raw.xtime_nsec to the base. But we already add the tk->tkr_raw.xtime_nsec to the offset and shift it all down in the timekeeping_delta_to_ns called from ktime_get_raw, so we effectively are accumulating the nsecs portion faster then we should. This only crops up for internal ktime_get_raw() users, but not getrawmonotonic64() which uses the timespec generation rather then the ktime method, which is why this wasn't seen by userspace time tests. I'll send a patch for testing shortly. thanks -john
