Skip to content

arm64: dts: exynos: bluejay: reserve retained firmware buffers - #226

Merged
derekross merged 1 commit into
derekross:omamobilefrom
latentoperator:omamobile-bluejay-memory
Nov 13, 2026
Merged

derekross merged 1 commit into
derekross:omamobilefrom
latentoperator:omamobile-bluejay-memory

Conversation

@latentoperator

@latentoperator latentoperator commented Nov 9, 2026 •

Copy link
Copy Markdown
Contributor

The commit identifies every excluded range by source node, address and size, with Google's pinned GS101 GPL sources and the reconstructed stock MP1.0 tree as provenance. The saved native FDT/iomem evidence explains the independent ABL tail exclusion.

Keep the shared dashed debug-info node. The earlier capture shows ABL inserting an underscore-named duplicate, but does not establish that the source must use that spelling. A fresh live run now confirms both spellings at the same page, as recorded below; no source rename or duplicate-resolution claim is made. Consumers remain unbound and the extents are no-map exclusions.

Runtime update — November 12–13, 2026

November 12–13 runtime follow-up:

Tested signed integration: 720a495233734e4328a09c33c095d72a8291e1c4, based on fc7086a68fcaaed8960d09f1bdf6c51c9926c759; uname -r = 7.3.0-rc6-omamobile-bluejay-review-20261111, with its matching Image, stage DTs and modules. This is an integration result containing the submitted changes, rather than a boot of each PR in isolation.

The revised kernel passed both the RAM-rescue baseline (userdata unmounted, no modules loaded, all eight experimental nodes disabled) and a fresh persistent Arch-root baseline (all 49 partition guards, actual-host-SSH ready-marker acknowledgement, no failed system units or fatal kernel log lines). The later isolated touch and pool/PCIe/Wi-Fi stages also booted this retained-memory layout.

The fresh persistent-root log confirms that ABL still introduces debug_kinfo_reserved@fd3fe000 alongside the source's debug-kinfo-reserved@fd3fe000: both describe the same 4 KiB no-map page at 0xfd3fe000. This is the requested fresh observation, not a claim that the duplicate is resolved or that the shared source node should be renamed. The log also reports the ABL exclusion at 0xf8800000..0xf97fffff. Reserved consumers were left unbound in the baseline; their hardware functionality is not qualified by these boots.

These were attended headless stages with storage guards, a PID-1-owned 20-second watchdog, host logging and fresh SSH postchecks. They do not qualify the revised desktop/GPU path, native panel modesetting, suspend, endurance or a release installer. The historical graphical desktop has been restored. No kernel commit or DCO metadata changed; AUTO_WRITE=no and masked automatic boot-success remain in force.

Prior source/build validation

Validation and limits

  • Independent scoped review found no confirmed runtime safety blocker after corrections.
  • Earlier source-checkpoint integration based on 1a882ca builds eight changed C objects with W=1, ten Google DTs with dtbs_check and no new warnings, and passes four binding checks. Strict checkpatch has no errors; new-file coverage reminders are disclosed.
  • The revised signed integration has now run in RAM-rescue, persistent-root and isolated hardware stages (runtime section below). The exact old tested c934e31 source and original signed refs remain preserved.
  • Mobile development evidence and the split are originally tracked by the now-closed omarchy-mobile #543, with documentation #561, initramfs #562, installer #563 and bring-up #565.
  • No firmware, private captures, credentials, device addresses or calibration data are submitted.

Human DCO certification completed on November 9. Matching author sign-offs, unchanged source trees and stable patch IDs were independently verified before publication. The agent added no human tags. Signed-head policy/style and clean-merge checks pass against current upstream; the common build/schema evidence remains valid because signing changed metadata only.

@derekross

Copy link
Copy Markdown
Owner

Thanks, Chris — reserving these before anything else touches memory is the right order. Could you cite where each range comes from (Google's bluejay device tree or your stock iomem capture, described, not attached), and either show that ABL needs the underscore node name or keep the dashed one, which builds without a W=1 warning?

Preserve retained stock MP1.0 extents before Linux's CMA and SWIOTLB
allocators can use them. These are no-map exclusions, without driver or
pool bindings. The ranges are from the stock B0 base plus MP1.0 dtbo,
reconstructed offline from the owner's verified slot A images. Google's
GPL gs101-rmem.dtsi corroborates the ranges below except xhci, which is
in gs101-dma-heap.dtsi. Snapshot constants are in debug-snapshot-def.h.

Source node               Start       Size
  ect_binary              90000000    0003a000
  xhci_dma                97000000    00400000
  seclog_mem              c3000000    00080000
  header                  d8100000    00010000
  log_kevents             d8110000    00700000
  log_s2d                 d8810000    01800000
  log_array_reset         da010000    00a00000
  log_bcm                 dd610000    00400000
  log_itmon               dda10000    00010000
  cp_rmem_1               e8000000    01c00000
  cp_rmem_2               ee000000    00400000
  cp_rmem                 f0000000    00800000
  cp_msi_rmem             f6200000    00001000
  ramoops_mem             fd3ff000    00400000
  suspend_diag_log        fd90c000    00001000

The saved native live FDT and iomem show ABL reducing abl@f8800000 from
32 MiB to 16 MiB. Preserve its f9800000/01000000 tail independently.
Keep the shared dashed debug-info node: the capture shows ABL inserting
an underscore-named duplicate, but does not prove ABL needs that spelling
in our source. A future live run must inspect the resulting nodes again.
No capture or firmware payload is submitted. No new phone boot is claimed.

Link: https://android.googlesource.com/kernel/devices/google/gs101/+/refs/tags/android-16.0.0_r0.9/dts/gs101-rmem.dtsi
Link: https://android.googlesource.com/kernel/devices/google/gs101/+/refs/tags/android-16.0.0_r0.9/dts/gs101-dma-heap.dtsi
Assisted-by: LLM
Signed-off-by: Chris Gray <chrisgray70@gmail.com>
@latentoperator
latentoperator force-pushed the omamobile-bluejay-memory branch from 47788f9 to 3fb092d Compare November 9, 2026 18:35
@derekross

Copy link
Copy Markdown
Owner

Thanks, Chris. The per-range citations are exactly what I wanted, and keeping the dashed node name avoids the warning. Happy to take this once it's out of draft and CI has built it.

@latentoperator
latentoperator marked this pull request as ready for review November 10, 2026 18:17
@latentoperator

Copy link
Copy Markdown
Contributor Author

November 12–13 runtime follow-up:

Tested signed integration: 720a495233734e4328a09c33c095d72a8291e1c4, based on fc7086a68fcaaed8960d09f1bdf6c51c9926c759; uname -r = 7.3.0-rc6-omamobile-bluejay-review-20261111, with its matching Image, stage DTs and modules. This is an integration result containing the submitted changes, rather than a boot of each PR in isolation.

The revised kernel passed both the RAM-rescue baseline (userdata unmounted, no modules loaded, all eight experimental nodes disabled) and a fresh persistent Arch-root baseline (all 49 partition guards, actual-host-SSH ready-marker acknowledgement, no failed system units or fatal kernel log lines). The later isolated touch and pool/PCIe/Wi-Fi stages also booted this retained-memory layout.

The fresh persistent-root log confirms that ABL still introduces debug_kinfo_reserved@fd3fe000 alongside the source's debug-kinfo-reserved@fd3fe000: both describe the same 4 KiB no-map page at 0xfd3fe000. This is the requested fresh observation, not a claim that the duplicate is resolved or that the shared source node should be renamed. The log also reports the ABL exclusion at 0xf8800000..0xf97fffff. Reserved consumers were left unbound in the baseline; their hardware functionality is not qualified by these boots.

These were attended headless stages with storage guards, a PID-1-owned 20-second watchdog, host logging and fresh SSH postchecks. They do not qualify the revised desktop/GPU path, native panel modesetting, suspend, endurance or a release installer. The historical graphical desktop has been restored. No kernel commit or DCO metadata changed; AUTO_WRITE=no and masked automatic boot-success remain in force.

@derekross

Copy link
Copy Markdown
Owner

Thank you, Chris. This is ready, and I'll merge it as it is.

I re-checked it on today's omamobile (2db2bc3): a clean merge, your sign-off, checkpatch --strict with nothing to read, and the ten device trees with no new warnings. dt-validate on gs101-bluejay.dtb prints nothing, before or after. CI on this head is green. I also read each range against Google's gs101-rmem.dtsi and gs101-dma-heap.dtsi at the tag your links name. ECT, the four modem regions, ramoops, the suspend log, ABL's region and the xHCI pool all match, and each has a fixed reg there, so these are real addresses and not pools that happened to land there. Thanks also for the fresh look at the duplicate debug-kinfo node, and for saying exactly what it does and doesn't show.

Two questions for later, neither holding this up:

  • Google's file also has log_array_panic, log_slcdump and log_preslcdump, listed between log_array_reset and log_bcm, and none of them is reserved here. If they sit where the order suggests, that's System RAM in this tree. Did you leave them out on purpose (does the MP1.0 overlay drop them?), or should they be reserved too?
  • Your 11-13 report is from 720a495233734e4328a09c33c095d72a8291e1c4, which isn't on GitHub. Could you push it to a branch on your fork? Then I can match its files to each PR's head, here and on the other Bluejay PRs.

@derekross
derekross merged commit 8f5d0cb into derekross:omamobile Nov 13, 2026
1 check passed
@latentoperator

Copy link
Copy Markdown
Contributor Author

Following up on the three dump regions: they are present in the saved MP1.0 stock tree, but all three are disabled. The reason is the shared GS101 user debug policy, rather than an MP1.0 overlay deleting the nodes. gs101-dpm-user.dtsi disables log_arrdumppanic, log_slcdump and log_preslcdump; gs101-dpm.dtsi selects the policy.

The reserved-memory scanner in the tested Linux tree skips unavailable nodes, and Google's GPL debug-snapshot consumer independently checks each memory-region target's availability before using it. The omitted regions total 44 MiB. This evidence supports leaving them inactive under the stock user policy; it does not prove every firmware/debug mode leaves the ranges untouched. Enabling a different dump policy or its consumers needs a fresh reservation audit. No additional reservation or hardware change was made for this follow-up.

The exact tested integration is now published unchanged: bringup/bluejay-reviewed-tested-20261111, commit 720a495233734e4328a09c33c095d72a8291e1c4. Its source tree and existing human sign-offs are unchanged.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants