APERF counts at TDP frequency and its 64bits wide.. so unless we write some very large non-zero value you can count on it :-)
Cheers, ashok raj - Open Source Technology Center >-----Original Message----- >From: Mark.Haywood at Sun.COM [mailto:Mark.Haywood at Sun.COM] >Sent: Friday, June 20, 2008 11:02 AM >To: Raj, Ashok >Cc: Eric Saxe; Vinay Devadas; Jim.Britton at Sun.COM; tesla-dev at opensolaris.org >Subject: Re: [tesla-dev] Turbo Mode Support for Solaris > >Raj, Ashok wrote: >> You can write 0 to clear it.. actually that's part of what you do even >> after taking a sample. >> >Any idea how long it takes to wrap. I thought I remembered that it was a >verrrry long time. But maybe I made that up. ;-) > > >> >> >> Cheers, >> ashok raj >> - Open Source Technology Center >> >> >> >> >>> -----Original Message----- >>> From: tesla-dev-bounces at opensolaris.org >>> >> [mailto:tesla-dev-bounces at opensolaris.org] On Behalf Of Eric >> >>> Saxe >>> Sent: Friday, June 20, 2008 10:30 AM >>> To: Mark Haywood >>> Cc: Vinay Devadas; Jim.Britton at Sun.COM; tesla-dev at opensolaris.org >>> Subject: Re: [tesla-dev] Turbo Mode Support for Solaris >>> >>> Mark Haywood wrote: >>> >>>>> I guess we already discussed this issue before? >>>>> Turbo mode is exported as P0 = marked frequency + 1Mhz, >>>>> but allow the processors to run faster so that the performance can >>>>> be improved up to 20%. Average CPU frequency is doable by the >>>>> hardware feedback mechanism. >>>>> >>>>> >>>>> >>>> I think that Vinay was asking Jim about the RFE that he added to >>>> http://www.opensolaris.org/os/project/tesla/Work/Powertop/ >>>> >>>> As I think you and I discussed on the alias earlier, the average CPU >>>> frequency while in P0 can be determined by strategically placing >>>> APERF/MPERF value dtrace probes when transitioning to and from P0. At >>>> least that would be the first thing I'd try ... >>>> >>>> >>> We were chatting about this yesterday, and I was wondering what happens >>> if we stay at P0 long enough to where APERF/MPERF could wrap? Is that a >>> concern...or is the time it would take the counters to wrap >>> >> sufficiently >> >>> long? I haven't looked in detail yet, but I was also wondering if >>> there's a way to explicitly reset the counters at the beginning of a >>> sampling interval to maximize the time-to-wrap (if it is short). >>> >>> Thanks, >>> -Eric >>> _______________________________________________ >>> tesla-dev mailing list >>> tesla-dev at opensolaris.org >>> http://mail.opensolaris.org/mailman/listinfo/tesla-dev >>> >> _______________________________________________ >> tesla-dev mailing list >> tesla-dev at opensolaris.org >> http://mail.opensolaris.org/mailman/listinfo/tesla-dev >>
