Skip to content

ASB-V37

Choose a tag to compare

View all tags
@krilikd krilikd released this 15 Apr 20:33
· 3531 commits to main since this release
37
9fcdcbb
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

AutoSystemBoost -- Changelog

V37 -- Honest Thermal Telemetry + Release Discipline

V36 rebuilt the profiles around sustainable gaming and deeper battery saving.
V37 makes the thermal telemetry honest, isolates broken sensors, and aligns versioning across every file in the module.

The V37 release is not a profile rebalance. The profiles that landed in V36 (battery, balanced, performance) tested well across multi-hour sessions on real hardware and are kept untouched. V37's job is to make the module's diagnostics trustworthy enough for a public GitHub release.

🌡 Honest Thermal Telemetry

V37 dev iterations added separate skin_temp and surface_hotspot channels, but the fallback logic collapsed them back into a single value when the literal shell sensors went silent. V37 fixes that and adds a sanity gate around the SoC die hotspot sensor that was returning garbage on some firmwares.

skin_temp is now strictly literal shell

| | V36 / V37 release |
|:--|:--|:--|
| skin_temp_c source | shell_front/frame/back, falling back to sys-therm-6 if all zero | shell_front/frame/back only, returns 0 if all silent |
| surface_hotspot_c source | sys-therm-6 / board_temp | sys-therm-6 / board_temp (unchanged) |
| Result on OP15 (shell sensors silent) | skin and surface both report 40C from sys-therm-6 — taxonomy meaningless | skin reports 0 (with thermal_skin_zone=-1), surface reports 40C — clean separation |

The two channels were always meant to answer different questions: "what does the user's hand feel" vs "where's the body-adjacent hot spot". V37 stops conflating them.

socd sanity gate

The SoC die-hotspot sensor socd is the highest-priority CPU thermal source on Snapdragon 8 Elite Gen 5, but on at least one OP15 firmware revision it reports raw=5 (effectively garbage) while real per-core sensors read 90–100C.

V37 adds an explicit sanity gate at two layers:

  1. Discovery time: when thermal_discover() validates socd, it does an immediate live read. If the value is <=10C, the zone is rejected and discovery falls through to cpu-1-1-* (priority 2) and cpu-0-5-* (priority 3). The reason is logged at startup.
  2. Read time: even if socd was selected, every per-tick read re-checks c <= 10. If true, the read is treated as invalid (temp_valid=0) with temp_invalid_reason="raw_too_low" while keeping the cached value. This prevents one broken read from cascading into FSM decisions.

temp_invalid_reason field

Status JSON and session_history.jsonl now carry a temp_invalid_reason string with these possible values:

Value Meaning
ok Sensor read succeeded, temperature is fresh
init First read hasn't completed yet
read_fail sysfs read returned 0 or error
raw_too_low Sensor returned an implausibly low value (likely broken socd)
no_zone No thermal CPU zone bound at all

This replaces the old "is temp_valid 0 because the sensor is dead, or because it just hasn't refreshed yet?" guessing game.

Session-level sensor health

session_history.jsonl schema bumped from v:8 to v:9. New per-session fields:

"temp_invalid_n": 12,
"temp_last_reason": "raw_too_low"

These let tools/asb_session_report.py and any future analysis tooling see at a glance which sessions ran with degraded thermal telemetry vs which had a healthy sensor stack.

thermal_summary: startup log line

The startup tz_dump: block prints all 96+ thermal zones for postmortem use, but reading 96 lines to learn which sensor was actually selected is annoying. V37 adds a single summary line right after tz_dump::

thermal_summary: cpu=socd(z67,val=5,valid=0) skin=none(z-1,val=0) surface=sys-therm-6(z74,val=40)

One line tells you: which CPU sensor won, which skin sensor won, which surface sensor won, current readings, validity. Everything else is just for context.

🩺 Type-Resolved Thermal Diagnostics in asb_doctor

tools/asb_doctor.sh previously didn't dump thermal state at all — users had to chase sensors manually with cat /sys/class/thermal/thermal_zone*/type. V37 adds a 🌡 Thermal Sensors (type-resolved) section that:

  1. Walks every /sys/class/thermal/thermal_zone* directory
  2. Reads each zone's type attribute
  3. Looks for a fixed set of recognised types (socd, cpu-1-1-0, shell_front, sys-therm-6, board_temp, etc.)
  4. Resolves each one to its current zone ID dynamically and prints name → zone → temperature

Critically: zone IDs are never hardcoded. Zone IDs renumber across kernel boots; any tool that hardcodes thermal_zone57 is wrong on the very next reboot. Doctor now does the right thing.

Assets 3
Loading