Thermal framework issues and limitations
summary
Kernel change in 6.10 moves mitigation decision out of the step_wise governor into the thermal core, so crossing a cooling trip stop immediately, which causes overshoot on systems with high thermal capacity. Two governors remain: stepwise (coarse, immediate changes) and P (requires per-board power coefficients, misbehaves under saturation). Proposal is to measure temperature slope to infer thermal capacity from many sensors (30–40) and to fix thermal-zone suspend ordering so mitigations don’t stop prematurely. Neural‑net/P learning is suggested but needs care.
AI-written summary · about
chapters (7)
- 0:00Introduction to thermal issues
- 2:09Stepwise governor fixes
- 6:14Governor limitations and IPA
- 10:28Measuring thermal capacity
- 13:02Tazone suspend resume problems
- 16:59Suspend coordination issues
- 18:46Real-world thermal failure cases
related talks
34mImprovisation and Demonstration of Linux Thermal Framework for Multiple Temperature Sensors
16mCombining DTPM with the thermal control framework
19msystem pressure on CPUs capacity and feedback to scheduler
11mFine grain frequency control with kernel governors
47mLinux Kernel Power Management - 101
56mPMWG: CPU Idle governor and irq prediction
16mXkernel: Principled Performance Tunability of Operating System Kernels
17mUse of Netlink for thermal kernel-user notification is problematic