* there is a filedescriptor leak! * move the 2 data files somewhere sane * check various tunables (see powertop 1.x) into a tunable checklist * pretty print estimate to only do 2 decimal places instead of 3 * inode dirty trace point * have a "show me my process" page * fix memory leaks * find out why CPU %age and C state C0 %age don't mesh sometimes * device events/sec into the process view * USB urbs as events/sec counter * investigate tick_sched_timer and why it wakes up * find if we can do more detailed reporting about tasklet(softirq) * deal with tasklets -- needs kernel patch * audio calibration? Need appropriate sample * "report" mode * CSV-output mode * delta mode (Tony) done ---- * on NHM, cpu P state stats accumulate and don't reset at measurment time * rescan for USB devices each measurement * brightness calibration * rfkill calibration * wakeup calibration * disk calibration * in learning, find the biggest delta and only adjust that parameter, not all of them as they go * in learning, try slightly lower base power * display device power in power top 20 * find out why CPU power and sum of components doesn't add up * exclude xorg/etc from wakeup blame backcounting * GPU ops in interrupt context -> do not account special ? * try to see if we can account graphics ops done by X to the guy who woke up X * don't save measurements < 5 seconds * auto save every 10 or so measurements * between 500 - 1000 measurements, start replacing (at random) existing ones with a hard cap at 1000 * network interfaces as device * network interfaces -- packet counters as events from /proc/net/dev * identify kernel threads in the process list for visual differentiation * deal with the "no measurements" case better with default values * add device runtime PM stats for pci, spi and i2c * looks like events/sec keep adding up for processes * Readable PCI names for runtime PM pci devices * debug why wakeups sometimes go to all 0's -looks like I fixed it, keep an eye on it * pretty print estimates to only show 2 significant places cleanly * pretty print various kernel pieces * deal with debugfs not being in /sys/kernel/debug * if you plug in the AC during a run, the battery code is not detecting that * status bar at the bottom * health report (reporting mode, not interactive) needs to include devices that don't support runtime PM * for unknown USB devices, list VID/PID