>-----Original Message----- >From: Mathew, JohnX >Sent: Friday, April 29, 2011 7:56 PM >To: 'Arjan van de Ven' >Cc: [email protected] >Subject: RE: [PATCH] process: report accurate wakeup numbers for ht cpus > > > >>-----Original Message----- >>From: Arjan van de Ven [mailto:[email protected]] >>Sent: Thursday, April 28, 2011 7:02 PM >>To: Mathew, JohnX >>Cc: [email protected] >>Subject: Re: [PATCH] process: report accurate wakeup numbers for ht >cpus >> >>On 4/28/2011 8:20 AM, John Mathew wrote: >>> Intel hyperthreaded cores does not exactly behave like if it was 2 >>real cores. >>> - Actual HW cstate is roughly minimal cstate of the 2 threads. >>> As most of the resources are shared, HW resources are only really >>powered >>> down when both threads are idle. >>> - Interrupts always wakes up the 2 threads. i.e. terminate the 2 >>mwait. >>> So in case of an hyperthreaded core a wakeup event is affecting power >>only when >>> both the threads are asleep. A wakeup event that occurs when one of >>> the thread is active should not be considered as a wakeup event. >>> >>> In case of hyperthreaded cores the thread ids that execute on the >same >>core >>> will be reported in the >>sys/devices/system/cpu/cpu%i/topology/thread_siblings_list. >>> A table is initially created with this thread sibling information >>corresponding to >>> a cpu. When powertop gets a power_end event on a thread the >>thread_sibling table >>> is looked up to identify the siblings of the cpu and each sibling is >>checked it >>> has a wakeup pending. If any one sibling does not have a wakeup event >>pending >>> the power end event is ignored. >>> --- >> >> >>so we already have this hierarchy in powertop (linux cpus vs cores vs >>package)... >>wonder if this should just fit into that >> >>also, Intel cpus tend to use the intel_cpus.cpp file not cpu.cpp.... > >I am using a processor which uses Atom microarchitecture , is it ok to >add the model id of >my processor to the list of ids that are being checked for Intel >processors? The current list seem to be more Nehalem based.. > >Regards, >John Mathew
I tried adding the processor model id to the list of ids being checked for Intel processors and I am not able to get the processor frequency report any more, it shows all zero percent. Seems like a separate version of intel_cpus.cpp is needed to handle Atom specific msrs. Would like to know your opinion about it. Also I don't see a reason why the patch I submitted should be moved to intel_cpus.cpp as intel_cpus.cpp file takes care of only the c-state and p-state handling and the tracepoint handling for all processors (both Intel and non Intel processors) happens in cpu.cpp. --------------------------------------------------------------------- Intel Corporation SAS (French simplified joint stock company) Registered headquarters: "Les Montalets"- 2, rue de Paris, 92196 Meudon Cedex, France Registration Number: 302 456 199 R.C.S. NANTERRE Capital: 4,572,000 Euros This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies. _______________________________________________ Power mailing list [email protected] https://bughost.org/mailman/listinfo/power
