Quoting Tvrtko Ursulin (2018-02-08 08:50:58)
> On 08/02/2018 08:00, Chris Wilson wrote:
> > The other part of the puzzle was perhaps just using a timer to read the
> > registers every 1/2 wrap period, but there was no clear picture on when
> > to start and stop that timer for sysfs. However, with perf we can create
> > and destroy a background timer on PMU open/destroy.
> > Fancy hooking up a background timer?
> Could do, wasn't sure that the complication is worth it.
Just thought about 10m work to make sure it's done and dusted.
However, wait and see if Len's happy is a fair policy.
> Idea would be to schedule a timer for instance, as you say, 1/2 wrap
> period, and keep pushing it forward on manual reads.
> But then question of runtime suspend arises again. If we decide to
> estimate the counter, as in the case of PMU, then I need to move that
> code out of PMU and make intel_pm.c the one and only authoritative
> source of RC6 data.
The timer would also entail waking up the device every 30m. One more
reason to loathe adding a timer (and trying to make sure the timer is
only run as required).
Intel-gfx mailing list