Hi, I am noticing something strange in powertop reports. When i run powertop with following parameters: powertop -t 120 --html I see following in summary section Power Consumption Summary 0.1 wakeups/second, 0.0 GPU ops/second, 0.0 VFS ops/sec, 0.0 GFX wakes/sec and 5375.4% CPU use Usage Events/s Category Description 2693% 0.07 Timer tick_sched_timer 2680% 0.00 Timer ehci_watchdog 0.6% 0.00 Interrupt [7] sched(softirq) 0.2% 0.00 Interrupt [1] timer(softirq) 0.2% 0.00 Interrupt [29] twd 0.1% 0.00 kWork pm_runtime_work 0.1% 0.00 Timer led_heartbeat_function 0.1% 0.00 Interrupt [9] RCU(softirq) 0.1% 0.00 Interrupt [57] OMAP DISPC 0.1% 0.00 Process rsyslogd -c5 How is percentage over 100 %... System Info: PowerTOP Version v2.0 Kernel Version Linux version 3.4.0-01820-gcc96f22 (viresh@blr-linut-001) (gcc version 4.7.1 20120514 (prerelease) (crosstool-NG linaro-1.13.1-2012.05-20120523 - Linaro GCC 2012.05) ) #5 SMP PREEMPT Tue Aug 28 12:44:52 IST 2012 System Name CPU Information 2x OS Information Running on ARM: Panda Board. -- viresh

> Running on ARM: Panda Board.

On (09/03/12 10:37), Arjan van de Ven wrote: >>> Running on ARM: Panda Board. > > I wonder if the perf data is somehow incorrect on your platform. > powertop depends on correct timestamps for example. > > I was thinking about this. Could it be possible that we lose some perf samples? E.g. we have start, but no end timestamps.

Arjan van de Ven wrote: >shouldn't... the buffers are quite large.... By the way, I just curious, why events processing are limited to 5000? I mean the PowerTOP has f half-megabyte buffer for each type of events but wishes to process only 5000 even if data is there? void perf_event::process(void *cookie) { ... while (pc->data_tail != pc->data_head && i++ < 5000) { ... In the case of many cores sched:switch will be fired at least CONFIG_HZ * nr_cpus times per second (as I understand), and that could easily be larger than 5000. Especially when we measuring for several seconds.