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

Phase 3: the GMU gets a driver, the crypto node goes (2026-11-03)

Attempt 2026-11-03-26 (kernel #17, commits a6e3987 and 63bddc5, image 963fb3004304e941), cold start. Owner pick from the Omarchy review (2026-11-03, “1 2 3”). Sources: the netlog dumps/firstboot/2026-11-03-netlog.txt and the AP console log (gitignored), and Omarchy’s linux-omarchy-aarch64 package (https://github.com/omacom/omarchy-pkgs/tree/master/pkgbuilds/linux-omarchy-aarch64, read 2026-11-03).

Why

Every boot left unbound=3d6a000.gmu and unbound=1dfa000.crypto (the sync_state() pending due to … lines that /init reports since the retro), handoff lead 3 and the gap list’s crypto lead.

  • GMU: nothing binds the GPU’s management unit, so gcc and gpucc never reach sync_state(). Akhil P Oommen’s (Qualcomm) “drm/msm: Attach a driver to the GMU” series binds it as a component of the GPU master; Omarchy carries it as 2170-2174 with its own fix-up 2178.
  • Crypto: the pin makes QCE depend on BROKEN (drivers/crypto/Kconfig:531), so nothing binds the node, and aggre2_noc and mc_virt never reach sync_state(); Omarchy (2210) disables the node.

Changes

  • kernel/patches/0002-0007 (ADR 0007): the six patches unchanged, each with a provenance header. They apply in order to the pin; check 20 now applies the series cumulatively to the pin’s files (0005-0007 build on 0002-0004).
  • dts/x1e80100-google-mica.dts: &crypto { status = "disabled"; }, marked not from the stock tree; qcom,qce left kernel/dt-gaps.txt.

Result: pass

  • adreno 3d00000.gpu: bound 3d6a000.gmu (ops a6xx_gmu_bind_ops) at 0.555 s.
  • unbound= went from 8 to 6: 3d6a000.gmu and 1dfa000.crypto are gone. Left: the four CPU bandwidth-monitor PMUs (24091000, 240b3400, 240b5400, 240b6400) and the video and camera clock controllers (aaf0000, ade0000).
  • The GPU is unchanged on the msm.separate_gpu_kms=1 path that patch 0005 reworks: gpu-node=renderD128 driver=adreno, gpu-open=opened, gpu-freq=300000000 max=1250000000, SQE and GMU firmware loaded, zap off.
  • errors=0 (from 1). SCMI full (0x20000, 3 of 3, 78 replies, max 3.9 ms, 0 timeouts).
  • Not measured yet: whether suspend now reaches CX collapse and DDR self-refresh (the reason Omarchy carries 2210); there is no suspend test on mica yet.