Messages in this thread | | | Date | Wed, 29 Jul 2015 14:35:55 +0200 | From | Peter Zijlstra <> | Subject | Re: [PATCH 1/5] Add functions producing system time given a backing counter value |
| |
On Wed, Jul 29, 2015 at 01:48:41PM +0200, Richard Cochran wrote: > On Wed, Jul 29, 2015 at 12:49:44PM +0200, Peter Zijlstra wrote: > > This is still fuzzy, right? The hardware ART timestamp could be from > > _before_ the NTP adjust; or the NTP adjust could happen while we do this > > conversion and we'll take the retry loop. > > In the original series, yes. > > > In both cases, the resulting value is computed using a different slope > > than was in effect at the time of the stamp. With the end result being > > slightly off from what it should be. > > In Thomas' patch the get_ts() is meant to read fresh pairs of time > stamps from within the loop.
Ah, indeed, I missed that. Yes if that's possible we get it right.
| |