Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Every enabled DT node binds: bandwidth monitors, videocc, camcc (2026-11-04)

Attempt 2026-11-03-35 (written before midnight, booted 2026-11-04 about 12:00 AM ET): kernel #18 (commit d41ab28, kernel/mica-sync.config), the board DTS with ADR 0008 (no PSCI domain idle states), image 4c8699959d256a77. One cold start (the EC reset itself at power-on: Reset cause: sysjump hard efs). Sources: the netlog dumps/firstboot/2026-11-04-netlog.txt and the AP console log (both gitignored), attempt.sh compare against -34 boot 3.

Change

kernel/mica-sync.config builds in three drivers that arm64 defconfig has as modules (the initramfs loads none):

  • QCOM_ICC_BWMON: the four bandwidth-monitor PMUs pmu@24091000 (qcom,sc7280-llcc-bwmon) and pmu@240b3400, @240b5400, @240b6400 (qcom,sdm845-bwmon).
  • SM_VIDEOCC_8550: clock-controller@aaf0000 (qcom,x1e80100-videocc).
  • CLK_X1E80100_CAMCC: clock-controller@ade0000.

The symbols come from scripts/dt-kconfig-map.py at a804c57; their four lines left kernel/dt-gaps.txt in the same commit.

Result

  • unbound= went from 6 to 0. No line in dmesg says sync_state() pending due to, so every interconnect and clock provider reached sync_state(). deferred= is empty.
  • No probe error from the new drivers (dmesg has no failure line naming bwmon, videocc or camcc).
  • SCMI: 78 replies, max 3.9 ms, 0 timeouts. That is the fourth stall-free cold boot with no domain idle states. cpufreq is on all three clusters (710 MHz to 3.42 GHz).
  • The touchscreen still fails (hid-generic ... -22). Its rail and enable pin read on this time (ts-vreg=enabled users=1, gpio88 out high, against disabled/low on -34 boot 3), which is the same failure with a different probe order (inference).
  • compare labels this boot “warm”: it took the AP reset at EC 179 s, which was /init’s reboot after the run, while the EC’s own reset at power-on restarted its clock. The boot was cold.

Open

  • What the providers’ sync_state() changes in power (bandwidth and clock votes dropped to what consumers ask for) has not been measured. The suspend test covers it.