[PATCH i-g-t] tests/kms_setmode: Ignore first timestamp when computing vblank statistics

[Kahola, Mika]

> -----Original Message-----
> From: Chris Wilson <chris at chris-wilson.co.uk>
> Sent: Tuesday, May 12, 2020 1:56 PM
> To: Kahola, Mika <mika.kahola at intel.com>; intel-gfx-
> trybot at lists.freedesktop.org
> Cc: Kahola, Mika <mika.kahola at intel.com>
> Subject: Re: [PATCH i-g-t] tests/kms_setmode: Ignore first timestamp when
> computing vblank statistics
> 
> Quoting Mika Kahola (2020-05-12 11:48:42)
> > kms_setmode is flip flopping due to wondering vblank timestamps. For
> > vblank statistics the first timestamp is taken long before we enter to
> > actual loop to test vblank accurary.
> 
> They are *all* queued ahead of the loop.
> 
> > Let's ignore the first timestamp in order to ensure that it does not
> > distort the statistics.
> 
> No. There's already a vblank wait prior to the start. If you want to show the
> skew, show the histogram (ala ministats).
Ok, so there may not be a issue with the test itself. I'm wondering what could be the possible cause that we fall out of the spec on vblank intervals. Could it be hw inaccuracy?

> -Chris