[Intel-gfx] [PATCH 1/2] tests/pm_rps: Round requested freq correctly

Ville Syrjälä ville.syrjala at linux.intel.com
Fri Feb 7 15:52:30 CET 2014


On Fri, Feb 07, 2014 at 08:44:14AM -0600, Jeff McGee wrote:
> On Fri, Feb 07, 2014 at 11:15:15AM +0100, Daniel Vetter wrote:
> > On Fri, Feb 7, 2014 at 10:33 AM, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> > > On Fri, Feb 07, 2014 at 10:03:33AM +0100, Daniel Vetter wrote:
> > >> The kernel will round it, so if we don't we'll have a spurious
> > >> mismatch. Happens on my machine here with 650-1300MHz range, where the
> > >> midpoint is 975.
> > >>
> > >> Cc: Jeff McGee <jeff.mcgee at intel.com>
> > >> Signed-off-by: Daniel Vetter <daniel.vetter at ffwll.ch>
> > >> ---
> > >>  tests/pm_rps.c | 3 +++
> > >>  1 file changed, 3 insertions(+)
> > >>
> > >> diff --git a/tests/pm_rps.c b/tests/pm_rps.c
> > >> index 467038104ec6..27e758755e3f 100644
> > >> --- a/tests/pm_rps.c
> > >> +++ b/tests/pm_rps.c
> > >> @@ -350,6 +350,9 @@ static void min_max_config(void (*check)(void))
> > >>  {
> > >>       int fmid = (origfreqs[RPn] + origfreqs[RP0]) / 2;
> > >>
> > >> +     /* hw (and so kernel) currently rounds to 50 MHz ... */
> > >
> > > s/rounds/truncates/ or if it really does round, you need to adjust the
> > > calculation.
> > 
> > We just need to use something divisible by 50 so that the value we
> > write and the one we get match up. Whether it's truncating or rounding
> > doesn't matter really.
> > -Daniel
> 
> Darn, I considered this possibility but forgot to account for it in the test.
> I think what I was going to do was to create another writeval variant
> which doesn't do read back matching check. This was because I didn't want to
> assume that all systems use a 50 Mhz frequency increment (do they all?).

VLV sure doesn't.

-- 
Ville Syrjälä
Intel OTC



More information about the Intel-gfx mailing list