Hi Umesh,
On Fri, Oct 10, 2025 at 10:45:39AM -0700, Umesh Nerlige Ramappa wrote:
> On Thu, Oct 09, 2025 at 02:24:52PM +0200, Andi Shyti wrote:
> > > @@ -171,7 +173,12 @@ static u32 read_clock_frequency(struct intel_uncore
> > > *uncore)
> > >
> > > void intel_gt_init_clock_frequency(struct intel_gt *gt)
> > > {
> > > + unsigned long clock_period_scale;
> > > +
> > > gt->clock_frequency = read_clock_frequency(gt->uncore);
> >
> > Shall we add a paranoid check here to raise a warning if
> > clock_frequency is '0'?
> >
> > It should never happen, but for robustness I would suggest:
>
> I did post that change, but BAT fails for PNV since clock_frequency is 0 on
> PNV (gen3). I will use this version without the check.
I saw the failure, haven't checked the all the paths, but isn't
then PNV affected by division by 0?
Andi
> >
> > GEM_WARN_ON(!gt->clock_frequency);
> >
> > (or even BUG_ON, though that might be overkill).
> >
> > Andi
> >
> > > + clock_period_scale = gcd(NSEC_PER_SEC, gt->clock_frequency);
> > > + gt->clock_nsec_scaled = NSEC_PER_SEC / clock_period_scale;
> > > + gt->clock_freq_scaled = gt->clock_frequency / clock_period_scale;
> > >
> > > /* Icelake appears to use another fixed frequency for CTX_TIMESTAMP */
> > > if (GRAPHICS_VER(gt->i915) == 11)
