thermald doesn't seem to work when running with HWE-Y on Ubuntu 16.04

[James Troup]

I have a weeks old Lenovo T460s running Ubuntu 16.04. After seeing
MCEs for due to temperature, I was advised to try the upcoming HWE-Y
kernel.

I installed linux-image-4.8.0-9-generic from the
canonical-kernel-team-ubuntu-unstable-xenial PPA and rebooted.

When I returned to my machine in the morning I found it had hard
locked up over night, it was very hot and all the fans were going at
full speed.

With the 4.4 kernel, I got messages like the following:

Sep 17 22:45:13 malefic kernel: [213000.178417] CPU2: Core temperature above threshold, cpu clock throttled (total events = 9521)

With 4.8, it increased to:

Sep 20 17:14:13 malefic kernel: [24571.915198] CPU2: Core temperature above threshold, cpu clock throttled (total events = 42953)

(Although there's a fair bit of variation).

I also noticed that thermald was spewing a lot of error messages to
syslog:

  http://paste.ubuntu.com/23207813/

The thermald spew went away and the number of 'temperature above
threshold' events went down once I downgraded back to 4.4.

[Srinivas Pandruvada (srinivas-pandruvada)]

Basically sysfs writes are failing with 4.8 kernel, so all those errors in http://paste.ubuntu.com/23207813/.

[Colin Ian King (colin-king)]

Srinivas, I've asked James to test with his previous mon-4.8 kernel to see if the thermal overrun occurs; given that thermald is not supporting the /sysfs writes, is that a regression in the 4.8 kernel are is that a path change in 4.8 for the sysfs entries that need to be updated in thermald?

[Srinivas Pandruvada (srinivas-pandruvada)]

sysfs paths can't change, so need to look at the kernel. I didn't add any new stuff between 4.4 to 4.8 in kernel. I will look into this.

[Srinivas Pandruvada (srinivas-pandruvada)]

I ran on 4.8, I don't see those errors. Can you run on command line?
# sudo systemctl stop thermald
# sudo thermald --no-daemon --loglevel=debug

I want to see which sysfs write is causing issue.

[Launchpad Janitor (janitor)]

[Expired for thermald (Ubuntu) because there has been no activity for 60 days.]