media: qcom: restore SP11 camera parity gaps

[ooaklee]

Why this repair exists

The final-tree parity audit after packaging 7.2.2-sp11beta1 found that the
camera re-lift in #35 had omitted active behavior from the hardware-qualified
sp11/integration-7.2.x source. The affected candidate and package were
withdrawn before hardware testing.

This topic restores those behaviors as seven reviewable commits. The source
author dates for Leon Silcott are retained, while his author, committer, and
sign-off identity is consolidated to the preferred leon@boasi.io address. It
must remain draft until the exact stack passes the camera gate below.

Restored behavior

1. exact stable replay of b910f68a07fe: disable unconfigured CSID680 RDI
   horizontal/vertical crop and drop engines;
2. exact stable replay of 651e711719df: select MIPI RAW write-client mode
   for VFE680 RDI, consistent with the v17x/v480 RDI implementations;
3. restore the exact one-trio C-PHY RAW10 3844-to-3840 receiver crop and
   source-pad propagation from 73960c772d24, including the final standalone
   IMX681 format signature from 6621d73e732c;
4. restore explicit keep-all values for every CSID680 drop engine from
   1592ec377418;
5. restore the CAMSS partial-stream-start unwind from 0097c12b0fec; and
6. restore the exact X1E80100/IMX681-scoped read-only CSID and VFE stop
   diagnostics from 4d190bc96139; and
7. contain the shared CSID680 and VFE680 register changes to the exact
   X1E80100 C-PHY RAW10 IMX681 signature, restoring prior behavior on every
   other route.

The complete camss-video.c and camss-csid.h results match the qualified
source tree. VFE and CSID preserve the qualified SP11 behavior but deliberately
differ elsewhere: C-PHY RX selection is limited to X1E80100, the X1E CSID clock
override is limited to C-PHY, and the restored register programming is selected
only by the qualified IMX681 transport and geometry signature.

Deliberate exclusions

• The old CCS IMX681 experiment is not restored. In the qualified final tree,
  the Denali camera node binds sony,imx681 to the standalone driver; no live
  device-tree node binds the CCS IMX681 path. Restoring both implementations
  would retain unbound superseded code and risk a future ownership conflict.
• The CCS-oriented camss_get_link_freq() upstream walk is not restored. The
  active standalone sensor is directly connected and exposes its link-frequency
  control on the sensor source pad. This remains documented as a latent CCS-only
  difference rather than being silently treated as parity.
• Sensor MODE_SELECT diagnostics from the superseded CCS path are not copied
  into the standalone driver because they are absent from the qualified final
  standalone implementation. Active CSID/VFE diagnostics are retained.

Exact-current-head software evidence

• seven per-commit strict checkpatch.pl runs: 0 errors, 0 warnings, 0 checks;
• integration validator: pass, with only the three already recorded aggregate
  warnings and three pre-existing DWC3 alignment checks;
• clean case-sensitive ARM64 ubuntu_x1e_defconfig build with W=1: all
  changed camera objects and complete qcom-camss.o link pass;
• X1E/OLED, X1P/LCD, and Dell XPS 13 9345 DTBs: pass and byte-identical to the
  beta checkpoint;
• independent read-only OpenCode final-tree audit corroborated the active
  omissions and the dead/replaced CCS classification.

Current re-lift verification

The seven-commit repair was re-lifted onto current beta bf631f9, producing head f1ee66a. Ordered stable patch IDs, Leon author identity, and subjects match the pre-sync series. git range-diff pairs every commit in order; only the first two commit messages add explicit Jeroen Karsies source-commit links. GitHub CI and a local detached-head rerun of the exact-current-head integration check pass. A fresh case-sensitive ARM64 W=1 build at this exact head and tree 4659d5af7dc6 passes for every changed camera object, the complete qcom-camss.o link, and all three representative DTBs. The X1E OLED, X1P LCD, and Dell DTB hashes are respectively 0448aa1d7060, 81e289dfbd49, and 7baa7f970464, matching the current beta evidence.

Required hardware gates

• X1E/OLED: probe and topology, repeated and continuous 3840x2640 RAW10
  capture, correct stride/layout, controls, and privacy LED;
• verify CSID packet/error counters, VFE completion, and no crop/drop or buffer
  corruption;
• force or reproduce a partial stream-start failure, then confirm a clean retry
  without runtime-PM imbalance;
• libcamera, PipeWire, browser preview, repeated stop/start, and suspend/resume;
• retain the complete stop diagnostics for any failure report; and
• non-SP11 VFE680 RDI regression testing before upstream submission.

This PR does not authorize merging this topic or the held aggregate stack.

Fixes: #35
Context: jglathe#79 (comment)

[ooaklee]

Aggregate hardware qualification is tracked in #41 at exact candidate
08249986a3f4efd7e0826c29e830601bfde928ed (tree
da7479fbffe997cdc980a147b4cb929d71253185).

This topic remains draft. A green aggregate build or CI result does not satisfy
its topic-specific hardware, dependency, provenance, or non-SP11 gates. Record
full-system evidence in #41 and link the relevant result back here before this
topic leaves draft state.

PR #41 is a test-only aggregate and must close without merge.

[ooaklee]

Camera provenance rewrite

The branch head is now 5dba023bc2ab7077b66be23b3d7080fc16542523.

The rewrite changes commit messages only. The previous head b26aac2cfb29facb8a814ba4f4b10daa0426d801 and current head have the same Git tree (098e0347e48a2f18dee04162237f965d6e85728c), and git diff --exit-code reports no content change. CI is green on the new head.

The first two guarded CAMSS adaptations now explicitly state that they are based on Jeroen Karsies' SP11 camera snapshot and link the exact source commit/blob. Leon remains the author and sole Signed-off-by identity (Leon Silcott <leon@boasi.io>). We intentionally did not add Co-developed-by for Jeroen because the source snapshot does not provide the corresponding developer sign-off required by kernel DCO conventions.

The rebuilt aggregate qualification bundle at 25b65b5ece94ab051e537bd85884a3bb1efcea31 includes this rewritten camera series and completed successfully with the same source tree/content as the prior aggregate head. Physical camera qualification is still required; this PR remains draft.

[ooaklee]

Provenance audit: every relift/parity commit has a matching Leon Silcott <leon@boasi.io> sign-off and retains the public turbineBMW, Jeroen Karsies, and golden-integration source links. The inspected public turbineBMW patch series and Jeroen import do not carry reusable kernel DCO sign-offs for these rewritten commits. Preserve the current explicit source credit, do not fabricate contributor trailers, and obtain contributor confirmation before choosing any From:/Co-developed-by attribution for upstream submission. This is an upstream provenance gate; it does not turn software or future hardware qualification into permission to merge prematurely.

[ooaklee]

Final-tree semantic audit against qualified v19: PASS

Independent OpenCode review and direct Git-object checks compared this branch (5dba023bc2ab) with the hardware-qualified v19 final tree (2cbd1ec3e2da):

• camss-video.c is byte-identical in both trees (blob 7ff03ebb5450).
• camss-csid.h is byte-identical in both trees (blob 7a91ff319ae9).
• The remaining CSID/VFE differences only add containment: X1E80100, C-PHY, RAW10, 3840/3844×2640, and packed 3840×2640 stride 4800 as applicable.
• The shipped x1e80100-microsoft-denali-oled.dts is byte-identical to v19 and supplies C-PHY trio 0, 1,203,000,000 Hz link frequency, 19.2 MHz MCLK4, sony,imx681, and the 3840×2640 RAW10 mode. Every containment predicate therefore evaluates true on the qualified route.
• Active v19 RX selection, 3844→3840 crop, explicit keep-all drop state, MIPI RAW VFE mode, partial-start unwind, and stop diagnostics remain present.
• Excluding the old unbound CCS experiment removes no active v19 DT path.

The standalone sensor now rejects an unsupported multi-trio/extra-frequency/non-19.2-MHz description at probe. That is an intentional guardrail, not a parity gap; the shipped Denali graph passes it.

No source-semantic blocker was found. Hardware camera gates in the PR body remain mandatory before merge.

[ooaklee]

Message-only submission cleanup

The bare external SHA in prose was replaced with “the first linked commit” so full-patch checkpatch no longer treats it as an unavailable local commit. The authoritative external Link: remains unchanged. Final tree, all author identities/dates/subjects, and source are unchanged. Current head: 9e90f8aaf0f6287cdefe7582dd7d2df68d460ca3; all seven full patches and source checks pass with zero findings.