Either my laptop is having hot flashes, or ACPI is playing games with me.

I was ripping and encoding some music earlier this evening, when my laptop suddenly shut down.  As I was out of the room at the time, I figured it was yet another kernel or hardware bug, and didn't look through /var/log until it did it again not even ten minutes later:

Nov 24 19:41:23 kernel: ACPI: Critical trip point
Nov 24 19:41:23 kernel: Critical temperature reached (93 C), shutting down.
Nov 24 19:41:23 acpid: received event "thermal_zone THM0 000000f0 00000001"
Nov 24 19:41:23 acpid: notifying client 6182[0:0]
Nov 24 19:41:23 acpid: executing action "/etc/acpi/default.sh thermal_zone THM0 000000f0 00000001"
Nov 24 19:41:23 logger: ACPI event unhandled: thermal_zone THM0 000000f0 00000001
Nov 24 19:41:23 acpid: action exited with status 0
Nov 24 19:41:23 acpid: completed event "thermal_zone THM0 000000f0 00000001"
Nov 24 19:41:24 acpid: received event "processor CPU 00000080 00000006"
Nov 24 19:41:24 acpid: notifying client 6182[0:0]
Nov 24 19:41:24 acpid: executing action "/etc/acpi/default.sh processor CPU 00000080 00000006"
Nov 24 19:41:24 shutdown[7868]: shutting down for system halt

First off, I'm not even sure if the thermal readings are accurate.  Sure the unit was warm to the touch, but 93 celsius in the core?  That is very hot.

I rebooted and attempted to rip and encode for the third time, this time disabling acpid, but running a watch on /proc/acpi/thermal_zone/THRM/*.  Under normal conditions w/the ondemand cpufreq governer, ACPI reports the following:

<setting not supported>
<polling disabled>
state:                   ok
temperature:             49 C
critical (S5):           93 C
passive:                 89 C: tc1=5 tc2=4 tsp=600 devices= CPU

Note the "passive" line at the bottom.  The ACPI documentation seems to indicate that this is the point when OSPM modulates the CPU clock in order to cool it down.  In actuality, when watching the statistics under high load, CPU temperature continually oscillates between 81 and 90 celsius, with the period being somewhere between 10 and 15 seconds.

This still leaves me with a few unanswered questions:

1. How long does the clock modulation last, or does it wait for a safe temperature (say.. 81?) before it returns the clock to its original frequency?
2. What is the clock modulation delay, when a critical threshold is reached?

With regard to the second question, I'm thinking there may be no delay in the OSPM reaction.  When doing a watch -n0, which actually runs the command every 0.1 sec, on the ACPI statistics, it looked like the temperature was being read every two or three seconds.  So, it's possible that the CPU temperature may have risen at a rate larger than 4-5 degrees/sec, and so the OS saw the temperature go from 88 to 93 instantaneously.  I suppose the shutdown (S5) reaction took precedence over any type of clock modulation.

Anyway, I'll add my laptop to MRTG when I get home, and plot CPU temperature.  I still think 93 celsius is bogus, to begin with.

New comments are currently disabled for this entry.