For bisecting the regression we ran 14 test for 50 repetitions. Before the bisected regression:
commit 7b09cc5a9debc86c903c2eff8f8a1fdef773c649 Author: Pavel Tatashin <pasha.tatas...@oracle.com> Date: Wed Mar 22 16:24:17 2017 -0400 sched/clock: Fix broken stable to unstable transfer there was ~0 failing test on the Core2 machine. After regression ~350 failing tests. With your patch-set ~15 failing tests. To be honest, I must say that these test used to give unstable results on the Core2. But some time ago, the results magically stabilized at ~0 fails, by timing related fixes for other issues. Ville Syrjala now has a patch-set that we believe really solves the graphics parts of the issue. However, I believe that your patch-set still improves the situation related to the tsc instability of the Core2. /Marta > -----Original Message----- > From: Peter Zijlstra [mailto:pet...@infradead.org] > Sent: Tuesday, April 25, 2017 4:45 PM > To: Lofstedt, Marta <marta.lofst...@intel.com> > Cc: t...@linutronix.de; mi...@kernel.org; linux-kernel@vger.kernel.org; > ville.syrj...@linux.intel.com; daniel.lezc...@linaro.org; Wysocki, Rafael J > <rafael.j.wyso...@intel.com>; martin.pe...@linux.intel.com; > pasha.tatas...@oracle.com; daniel.vet...@ffwll.ch > Subject: Re: [PATCH 0/9] sched_clock fixes > > On Tue, Apr 25, 2017 at 09:31:40AM +0000, Lofstedt, Marta wrote: > > Hi Peterz, > > > > I tested your patch-set on the same Core2 machine as where we > discovered the regression. > > With the tsc=unstable boot param that passrate has improved significantly; > 350 fails -> 15 fails. > > So is that the same as before, or still worse? I don't really have a handle on > what your benchmark is here, nor what how 'good' is defined. > > If its still worse than before, I'm completely confused. Because with > "tsc=unstable" the patch you fingered is a complete no-op (__gtod_offset > == 0).