Thermal panic for systems with brain damaged ECs
Hi all.
I have an HP TC1100 (tablet PC) that I've been pretty lucky with under Linux in terms of component support. I have, however, been noticing a potentially dangerous situation occurring, possibly due to a brain damaged EC, and possibly due to my lack of acpid. Namely, when the system gets fatally hot, it does not shut down.
I have previously had one tc1100 replaced under warranty for this (system sitting in carrying case, unintentionally powered on for excessive periods of time; system locked up and would not POST on powerup). Regardless of the state in which software lies, the system should be able to automatically shut down before it kills itself, although the EC doesn't seem to be doing this for me.
So, I'm wondering if it's a good idea to have the thermal module provide an option to automatically umount ro/sync/power down above a certain temperature. I argue that this is not suitable for user space as it could potentially damage the system, and therefore needs to act as reliably as possible.
I notice that in thermal.c, poweroff is called, but this might not be sufficient if the system is already starting to misbehave due to thermal stress. It also seems that my DSDT has incorrect trip points (or the thermal sensor does not properly read): S5 is listed as 80C, but the system was only reading at 65C when the plastic case was too hot to touch. Does it make sense to add a user-defineable hard poweroff point?
joshua