Philippe Gerum wrote: > Dmitry Adamushko wrote: > > > > Hello all, > > e.g. it took 1 minute and 30 seconds to get 21 lines of statistics on > > the screen (-p 100) but should have been ~22-23 seconds. > > > > And the latency thread really reported the presence of collected data 21 > > times (through rt_sem_v()) . > > > > So have I missed something important? any ideas are wellcome. > > > > Dunno yet, but this reminds me of a trap I've been into recently that > caused a massive slowdown of the latency test (v2.0.x) whilst the > reported real-time activity seemed to be ok. Weird enough, actually. The > explanation was that I mistakenly left the full ACPI support and SUSPEND > active in my kernel config (Asus box with ICH6 chipset). Shutting them > down solved the issue. I've not investigated further which one of the > ACPI options or SUSPEND caused that trouble yet.
I am observing a similar issue here; after a suspend/resume cycle, running the latency test gives low latencies and every second is at least two seconds long. If I load the cpufreq module, I notice that even if when suspending, /proc/cpuinfo indicates 1.5 GHz, after resuming, it indicates 600 MHz. So, maybe, when doing a suspend/resume cycle without the CPUFreq modules loaded, Linux assumes that the CPU frequency remains the same whereas ACPI/BIOS changes it ? -- Gilles Chanteperdrix.