https://bugzilla.kernel.org/show_bug.cgi?id=23551
--- Comment #34 from Guenter Roeck <li...@roeck-us.net> 2013-02-25 14:45:10 --- Graphics(In reply to comment #33) > (In reply to comment #31) > > Unfortunately the CPU only reports one value for all cores combined, not per > > CPU. You can see that clearly when the load is high. > > This is odd. But then shouldn't "Core power" be changed to "Cores power"? > Ok, no problem. > Also, not sure how "Pkg power" can be so much higher than "Core power". What > else in the CPU package is drawing significant power? > Graphics, most likely, in my case. On servers with no graphics it might be the DRAM controller. Note that I don't report those since it would require restructuring the entire driver. The code is already a bit kludgy, as the core cpower is reported with CPU 0. If someone takes CPU 0 offline, the core power/attributes would disappear. So all power/energy should really be reported through the package instance, but if I do that I would have more than one power attribute associated with it, which would make things a bit difficult. Question is if I should rearrange the code and report all power domains. Thoughts on that ? > > The initial energy reading starts with the current value in the registers. > > The > > counter can wrap every 60 seconds, and the pkg and core counters wrap at > > different times. This can result in the initially higher value for the core > > energy. > > I see. Then I question the value of these measurements for user-space. I > assume > that the driver uses these values to compute the power figures, so the energy > values are not only confusing but also redundant. Don't you think exposing the > power values would be sufficient and less confusing? No, because power is instantaneous and energy is cumulative. Both values are important: With one the user can see how much energy the chip consumed since the driver was loaded, and with the other how much power it consumes now. We have a couple of options: I can start the count from the time the driver is loaded (which would miss some energy, but be a one-liner), I could add a wraparound value to the pkg energy if the core energy reading is higher (a bit complicated), or I could leave it is as. -- Configure bugmail: https://bugzilla.kernel.org/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug. ------------------------------------------------------------------------------ Everyone hates slow websites. So do we. Make your web apps faster with AppDynamics Download AppDynamics Lite for free today: http://p.sf.net/sfu/appdyn_d2d_feb _______________________________________________ acpi-bugzilla mailing list acpi-bugzilla@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/acpi-bugzilla
