On 06/20/2012 12:59 PM, Daniel Vetter wrote: > On Wed, Jun 20, 2012 at 5:53 PM, Chris Wilson <chris at chris-wilson.co.uk> wrote: >> On Wed, 20 Jun 2012 08:32:59 -0700, Jesse Barnes <jbarnes at virtuousgeek.org> wrote: >>> On Wed, 20 Jun 2012 09:38:25 +0200 >>> Daniel Vetter <daniel at ffwll.ch> wrote: >>>> And from the bikeshed departement: Can't we just print a running number? I >>>> know, substraction is bloody hard, but for anything else than total power >>>> consumption (e.g. graphing power over time) the running thing is imo >>>> simpler. We've had the same discussion for the rc6 sysfs residency timers >>>> and concluded (after Arjan yelled at us) that doing the substraction in >>>> userspace is better, least it allows multiple userspace tools to read >>>> this. >>>> >>> >>> Yeah that's a good point; this way happened to be simpler for what I >>> was doing, but just exposing the cooked register value (converted to >>> ujoules) is better. Will fix. >> >> And to keep it easy to parse, just make it return -ENODEV on older >> platforms and a simple number for gen6+. > > At that point you might also add the uJ suffix and move it to sysfs ;-) And as a final minor suggestion, could we integrate this with the /sys/kernel/debug/dri/0/i915_emon_status we have on ILK, so we could have just one file for power readings? Eugeni