Date: Mon, 15 Aug 2005 11:10:48 -0700 From: "Kevin Oberman" <oberman@es.net> To: Hajimu UMEMOTO <ume@freebsd.org> Cc: acpi@freebsd.org Subject: Re: Annoyances with passive thermal code (acpi_thermal) Message-ID: <20050815181048.5B2385D07@ptavv.es.net> In-Reply-To: Your message of "Mon, 15 Aug 2005 04:05:52 %2B0900." <ygezmrk2van.wl%ume@mahoroba.org>
next in thread | previous in thread | raw e-mail | index | archive | help
> Date: Mon, 15 Aug 2005 04:05:52 +0900 > From: Hajimu UMEMOTO <ume@freebsd.org> > Hi, > > >>>>> On Sat, 13 Aug 2005 19:38:42 -0700 > >>>>> "Kevin Oberman" <oberman@es.net> said: > > oberman> I've noted an unpleasant mis-behavior in acpi_thermal. > > oberman> 1. Killed powerd > oberman> 2. Set dev.cpu.0.freq to 1200 (I was on batteries and wanted to stretch > oberman> them) > oberman> 3. Started a BIG build...openoffice > > oberman> The temp went up over _PSV and the CPU was slowed to 1050. The CPU > oberman> cooled for a while and the freq was reset to 1800 which started draining > oberman> my battery way too fast. > > It's curious that even when CPU speed is slowed, the temperature go up > over _PSV. Yes, it is. I have seen that a fully loaded CPU will top out at 73C. At 1.35 GHz it reached 85C. I did this testing with a totally loaded CPU and no real I/O using 'dd if-/dev/zero count=1000 bs=1m | md5'. This loads the CPU and should push temperature to the max possible, but I was seeing the system running at 86C while building OpenOffice.org late last week and it was triggering passive thermal management which is set at 86.5C. I have no idea why building OpenOffice.org (port editors/openoffice-2.0-devel) causes the system to run hotter than the md5 test. By the way, I have noticed that dev.acpi_ibm.0.thermal returns temperatures that are more current than hw.acpi.thermal.tz0.temperature. The former seems to be immediate while the latter lags up to several seconds. Odd. > oberman> Ideally, acpi_thermal should store the frequency when it cuts speed and > oberman> restore that speed when the CPU cools, not the maximum speed. > > CPUFREQ_SET() does it, actually. However, since CPU speed is restored > by degrees, we couldn't use the facility effectively. Please try the > attached patch. I'll give it a shot as soon as I can, but I'd rather not rebuild it until I can back up my system which won't be until tomorrow. Thanks very much for looking at this. -- R. Kevin Oberman, Network Engineer Energy Sciences Network (ESnet) Ernest O. Lawrence Berkeley National Laboratory (Berkeley Lab) E-mail: oberman@es.net Phone: +1 510 486-8634
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20050815181048.5B2385D07>