From mboxrd@z Thu Jan 1 00:00:00 1970 From: Daniel Lezcano Subject: Re: [RFC PATCH 1/2] cpuidle: Add idle enter/exit time stamp for notifying current idle state. Date: Tue, 02 Apr 2013 13:18:27 +0200 Message-ID: <515ABE83.3070200@linaro.org> References: <1364804657-16590-1-git-send-email-jonghwa3.lee@samsung.com> <1364804657-16590-2-git-send-email-jonghwa3.lee@samsung.com> <515A65DB.8070803@linaro.org> <515A77FC.70008@samsung.com> <515A8A21.6070509@linaro.org> <515AA6C3.3060408@samsung.com> <515AAE2C.3040407@linaro.org> <515ABC0C.1090102@samsung.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Return-path: In-Reply-To: <515ABC0C.1090102@samsung.com> Sender: cpufreq-owner@vger.kernel.org To: jonghwa3.lee@samsung.com Cc: "Rafael J. Wysocki" , linux-kernel@vger.kernel.org, linux-pm@vger.kernel.org, cpufreq@vger.kernel.org, MyungJoo Ham , Lukasz Majewski , Kyungmin Park , Chanwoo Choi , sw0312.kim@samsung.com, m.szyprowski@samsung.com List-Id: linux-pm@vger.kernel.org On 04/02/2013 01:07 PM, jonghwa3.lee@samsung.com wrote: > On 2013=EB=85=84 04=EC=9B=94 02=EC=9D=BC 19:08, Daniel Lezcano wrote: >=20 >> On 04/02/2013 11:37 AM, jonghwa3.lee@samsung.com wrote: >>> On 2013=EB=85=84 04=EC=9B=94 02=EC=9D=BC 16:34, Daniel Lezcano wrot= e: >>> >>>> On 04/02/2013 08:17 AM, jonghwa3.lee@samsung.com wrote: >>>>> On 2013=EB=85=84 04=EC=9B=94 02=EC=9D=BC 14:00, Daniel Lezcano wr= ote: >>>>> >>>>>> On 04/01/2013 10:24 AM, Jonghwa Lee wrote: >>>>>>> This patch adds idle state time stamp to cpuidle device structu= re to >>>>>>> notify its current idle state. If last enter time is newer than= last >>>>>>> exit time, then it means that the core is in idle now. >>>>>>> >>>>>>> Signed-off-by: Jonghwa Lee >>>>>>> --- >>>>>> >>>>>> The patch description does not explain what problem you want to = solve, >>>>>> how to solve it and the patch itself shows nothing. >>>>>> >>>>>> Could you elaborate ? >>>>> >>>>> >>>>> I'm sorry for lacking description. I supplement more. >>>>> >>>>> This patch does add time-stamp for idle enter/exit only nothing m= ore. >>>>> The reason why I needed them is that I wanted to know current cpu= idle >>>>> state. It is hard to know whether cpu is in idle or not now. >>>> >>>> Did you looked at: >>>> >>>> include/linux/sched.h:extern int idle_cpu(int cpu); >>>> >>> >>> >>> Yes, I did. >>> >>>> ? >>>> >>>>> When I check the cpuidle state usage, sometimes the information i= s wrong. >>>>> Because it is updated only when the cpu exits the idle state. So = while the >>>>> cpu is idling, the cpuidle state usage holds past one. Therefore = I put >>>>> the time-stamp for cpuidle enter/exit for checking current idling= and >>>>> calculating idle state usage correctly. >>>>> >>>>> I just make this patch temporary for my cpufreq governor work. So= , it just >>>>> use time-stamp for all idle state together. After RFC working, I = have a plan >>>>> to update this patch to use timestamp for each idle state. >>>> >>>> I suggest you look at the enter_idle / exit_idle function and make= your >>>> governor to subscribe to the IDLE_START/EXIT notifiers. >>>> >>>> arch/x86/kernel/process.c >>>> >>>> These are defined for the x86 architecture, maybe worth to add it = to >>>> another architecture. >>>> >>> >>> >>> Thanks for your opinion. >>> >>> Actually, I work on ARM architecture and I knew that the attempt of= applying >>> idle notifier was failed. You probably knew it, because the link yo= u gave me >>> before is that attempt. (https://lkml.org/lkml/2012/2/7/504) :) >> >> Yeah, now I recall this thread. >> >=20 >=20 > Oh my, I thought you gave the link but you didn't. It was Viresh Kuma= r from > other patch of the patchset. Sorry. >=20 >>> Currently, there >>> is only notifying call which is for led in arch/arm/kernel/process.= c and I think >>> it isn't for me to use. Anyway, Do you really think it is better wa= y to use >>> notifier than my way? Because I think it is too heavy for me. On my= board, >>> sometimes entering idle happened hundreds times during the 100ms. I= don't want >>> to call notifier that much time. IMO, just moving local variable to= per-cpu >>> variable for showing the enter/exit time looks better although it r= equires code >>> modification on cpudile side. What do you think? >> >> Sorry, but it is hard to figure out what you are trying to achieve w= ith >> a single patch. >> >> IIUC, you want to know how long the cpu is idle including the curren= t >> state, right ? So you need to know if the cpu is idle and when it >> entered the idle state, correct ? >> >=20 >=20 > Exactly. >=20 > =20 >> If the cpu is idle and the information is per cpu, how will you read >> this value from another cpu without introducing a locking mechanism = ? >> >=20 >=20 > I think it might be tolerated for incoherency of that data. Governor = reads the > data only, and if recoded start time or end time are different in few= usec with > real one then it doesn't effect to the result much. >=20 >=20 >> Does it mean the cpufreq governor needs cpuidle ? I am wondering if >> these informations shouldn't be retrieved from the scheduler, not fr= om >> cpuidle. >> >=20 >=20 > Yes, tick_sched per-cpu variable has all information that I need. But= it isn't > global variable. And I'm afraid to change static variable to global o= ne as my > pleases. It is a global variable but there is a function to get access: extern struct tick_sched *tick_get_tick_sched(int cpu); Does it fit better for what you want to achieve ? Thanks -- Daniel --=20 Linaro.org =E2=94=82 Open source software for= ARM SoCs =46ollow Linaro: Facebook | Twitter | Blog