[satchlj]

Enable the preview profile by default only for the exact J613/T8122 match with both firmware compatibility and actual firmware version 26.6.2. apple_isp.preview_26=0 disables it; the read-only parameter remains mode 0400. Preview explicitly disables still MBNR and temporal history. Other profiles retain main’s capture flow and format/interval selection.

Generic IPC-failure containment has landed in #1560. This PR is now a seven-file preview-only residual at 76a7dfbcdd045f3596d2820da8952935aacc76f1, a normal merge of preceding c37f1fcb with main f1b79c4b. Original commits, authorship and signoffs remain in ancestry. No force push was used.

The resolution preserves main’s resident firmware, restartable-idle lease, heap restoration, P010 late-ACK/return ownership and containment/KUnit fixtures byte-for-byte. Only three production files change: parser/admission and parameter in isp-drv.c, the enum/flag in isp-drv.h, and two preview conditionals in isp-cam.c. Two unchanged preview scripts, two ownership lines and two runner registrations complete the diff. All 105 main test entries, their order/controls, existing 16 containment cases, failure detection and runner mode100755 are preserved; the candidate has 107 unique entries.

Validation of this integration: both existing production-extraction preview tests pass under ASan/UBSan. They cover default/opt-out and board/SoC/version/compatibility refusal, malformed firmware cells, MBNR/temporal behavior, legacy/H17 control flow and disable-error propagation. The four tested preview/parser function bodies and both scripts are byte-identical to the original feature head. This is host software coverage, not a new kernel/QEMU or physical boot result. Both tests pass again on this integration. The adjacent-helper conflict preserves the complete preview admission predicate and main's resume worker/PM notifier. Seven main sleep/resume/remove helper bodies are unchanged, and every added/removed feature line in the seven-file residual matches the preceding head. Current-head Host run 43143061939 passed on 76a7dfb, including the actual host-test step (job 130934412930).

Hardware evidence retains its original scope:

• Comment 6800184254: Satya locally viewed one recognizable daylight NV12 1920x1080 frame on J613/package 30.12 carrying f512cd7; broader image quality remains unqualified. No image left the machine.
• Comment 6804320173: f512cd7 in a combined J613 kernel passed all 65 cells (13 NV12 sizes × 15/20/24/25/30 fps), 20 further restart sessions, closed-camera and active-stream 20 s RTC sleeps with validated post-resume frames and no sequence gaps. This closes the earlier missing interval/restart/s2idle evidence for that feature head. It is not a physical boot of the current integrated 76a7dfb tree.
• Comment 6814550431: Ryan's J616s/T6040 (T6041 ISP, H17 ABI) stock-vs-test module comparison passed. The test module is fleet67/integration c6c15ac plus this head's net 42-added-line ISP delta (three production files), loaded at runtime without a reboot. Both modules passed 32/32 format/size/interval readbacks and 10/10 restart sessions across P010/NV12, with zero lost frames/sequence gaps or ISP/DART/IPC log lines; format enumeration hashes match. This closes the T6040/P010 no-regression gate. No T6040 s2idle cycle or full integrated-head kernel boot was performed.
• Firmware stop-magic/explicit-stop fallback warnings persisted. Resumed streaming does not demonstrate clean firmware suspend. The inherited command line explicitly enabled preview_26, so hardware default/opt-out isolation was not measured; host policy tests cover selection.

Software review and current-head Host CI are complete. The remaining T6040/P010 regression gate from review 5860549030 is satisfied by Ryan's stock-vs-test module report6814550431, with the scope recorded above. The J613 daylight/interval/restart/RTC evidence remains accepted at its original feature/composed-kernel scope. No new M3 or camera/media collection is requested here. Forced runtime IPC-failure hardware injection and broader image-quality/control qualification remain unclaimed.

Port of omacom/linux-aurora#344; original PR and historical evidence retained. Codex (GPT-6.1 Sol), on @DjDeveloperr’s behalf.

[satchlj]

@DjDeveloperr @RyanTheTide ready for your regression testing. It hasn't been boot-tested on J613 with 7.3 yet; I'll post the result here when I have it. It conflicts with the latest base branch right now; I'm merging that in and will push the update shortly.

[satchlj]

J613 boot result on 7.3 (rubra, M3 MacBook Air 13", macOS 26.6.2 / 25G83 firmware), 2026-11-05.

• Kernel: integration/aurora 11eccbed with all 17 J613 PRs merged (soc: apple: sep: preserve unreadable and partial identity state #962–media: apple: isp: bound exposure metadata and keep controls gated #978), plus HID: dockchannel-hid: write each frame to the FIFO under one lock #1010 (the MTP frame-lock fix). Boot chain: stage2 from the m1n1 Thunderbolt display on M1 Pro/Max (t600x), stacked on #8 #64 content, U-Boot with arm64: dts: apple: t8140/j700: keyboard, trackpad and USB-C topology (reworked, 5 patches) #30/soc: apple: sep: T6000 (M1 Pro) support, plus SPI CS timing and attach fixes #26, then Limine.
• Mesa: aurora-wip 4371cf86 + mesa cpufreq: apple: add bounded M6 T8152 CPU scaling #161–#163.
• Automatic checks passed:
  • boots to the desktop;
  • GPU: DRM driver aurora, Apple M3 (G15G) for GL and Vulkan, glmark2 1113 (1011 on 7.2.11), no GPU errors;
  • internal panel up;
  • keyboard and trackpad present with no MTP crash;
  • camera node, Wi-Fi, Bluetooth, audio cards and headphone jack;
  • Touch ID policy readback (identity_source=dt);
  • speakers stay off as intended (our stage2 passes no apple,speaker-config).
• Not checked yet, since they need me at the machine: the fingerprint match, an external display, and suspend.

[RyanTheTide]

This no longer merges into the current integration/aurora. It conflicts in MAINTAINERS and drivers/media/platform/apple/isp/{isp-drv.h,isp-fw.c,isp-v4l2.c}. Please merge the current integration/aurora into it (or rebase).

[satchlj]

Merged current integration and resolved MAINTAINERS and the ISP driver/header/firmware/V4L2 conflicts, keeping the firmware-restore and deferred-detection changes alongside IPC containment. The ISP subtree builds with KUnit enabled. All 23 ISP-filtered host runs pass, including the rollback negative control; the KUnit fragment also now enables MEDIA_PLATFORM_SUPPORT. No new hardware capture was performed on this head.

[satchlj]

GPT-6.1 Sol on @satchlj's behalf

2026-11-10 J613 camera smoke re-test with #974 head f512cd7 and #978 8f9ffa3. After returning freezer/devices/platform diagnostics, /dev/video1 (apple-isp/FaceTime HD Camera) streamed three frames to /dev/null, rc=0. The installed-default reboot repeated three-frame capture successfully. The ISP stop-magic/explicit-stop-fallback warnings also appear in the captured 30.6 baseline; this camera session repeats them. No image retained, visual image-quality assessment or exposure-control/metadata qualification. An initial attempt selected /dev/video0 (AVD decoder) and timed out; capture passed after selecting the ISP node.

J613/T8122 M3 Air, integration dc8f6f7 plus #971/#973/#974/#978/#1340/#1486/#1491, composed source 112007b70271. Hardware-tested release 7.3.0-aurora73-upstream-2-ARCH; UKI SHA256 248283ad7750793cf60f8530e3e2331f7c1a58d733d2f3e8880f8a959e87becc. Composition evidence, not an isolated PR kernel. The same binary passed a one-shot trial and the normal new-default reboot. Installed package revision 30.10 corrects only the installer's pre-extraction space estimate; all 2054 Linux-package usr regular files match the qualified 30.9 package. Stage1/Stage2 unchanged, four boot hooks masked.

[satchlj]

GPT-6.1 Sol on @satchlj's behalf

New hardware capture on the current head f512cd7e432e1d1886a36e991ffe4928e4416b73 (matches the 30.9 package source receipt).

Board: rubra, M3 MacBook Air J613 / T8122, firmware 26.6.2 / 25G83. Running 7.3.0-aurora73-upstream-2-ARCH, installed Linux/support packages 30.10 (unchanged qualified 30.9 kernel binary). No reboot or kernel/boot changes for this test.

• v4l2-ctl --all identifies apple-isp / FaceTime HD Camera at /dev/video1; media1 topology has the apple-isp capture entity. /dev/video0 is the unrelated AVD decoder.
• NV12, all 13 advertised resolutions: 1920×1080, 1080×1920, 1920×1440, 1440×1920, 1280×720, 720×1280, 1280×960, 960×1280, 640×480, 480×640, 1760×1328, 1328×1760, 1440×1440. Each streamed 90 frames at requested 30 fps. Measured timestamp rates 29.962–30.000 fps, 1170/1170 frames, zero sequence gaps.
• One additional NV12 frame saved locally per resolution; byte counts match the driver’s 64-byte-aligned stride (including portrait/non-aligned widths). No images uploaded or posted. The images are non-uniform but nearly black: e.g. 1920×1080 luma mean 15.90, variance 0.288, range 12–47; 640×480 mean 16.00, variance 0.149, range 15–43. Recognizable room-image quality is unqualified: this unattended capture cannot distinguish darkness/occlusion from an image-quality defect. Successful buffers alone are not proof of useful images.
• 20/20 start/stop sessions completed (three frames each, 640×480). Two consumers in sequence, v4l2-ctl then FFmpeg, each completed 60 frames successfully.
• Every session shutdown repeats the already-documented never received magic number from firmware / firmware did not suspend, stopping it explicit-stop fallback (48 pairs across the 26 resolution/capture sessions, 20 cycles and two consumers). No new IPC timeout, DART fault, panic or oops in the test-window delta. This is not a clean firmware-suspend result; no system suspend was attempted.

Not tested: 15/20/24/25-fps intervals, physical visual assessment, exposure-control/metadata qualification, system suspend/resume, concurrent consumers, or forced IPC failure containment. Inspection of this PR’s ISP source found no safe runtime fault-injection hook (only preview_26 module parameter); no malformed IPC or firmware disruption was injected. Measurements serialized under the alba rig lease.

[DjDeveloperr]

Codex (GPT-6.1 Sol), reviewing on @DjDeveloperr’s behalf.

Code review/evidence assessment: IPC ownership containment and preview admission are deliberately scoped, and the posted multi-resolution/repeated-session captures materially improve transport coverage. The nearly black unattended images do not yet demonstrate recognizable preview. Before merging the default-on preview, please provide local attended well-lit visual confirmation, supported frame-interval tests, stop/restart/suspend behavior and a T6040/P010 regression. Preserve the known explicit-stop-fallback warnings in the result rather than calling firmware suspend clean. Generic containment could be split from default preview if it should land earlier. No camera images/recordings need to leave the test machine.

[satchlj]

Claude Fable 5.2 on @satchlj's behalf

Daylight camera check, J613 M3 MacBook Air, kernel 7.3.0-aurora73-ppt-2-ARCH (local package 30.12, which carries this PR's current head f512cd7): 5 NV12 frames at 1920x1080 captured with v4l2-ctl from the apple-isp "FaceTime HD Camera" node after skipping 30. Luma mean 147, sampled variance about 3400, range 0-255, chroma means U 126 / V 131, so real images (last night's black frames were a dark room). @satchlj looked at one frame on the machine and confirmed a correct, recognisable picture; his words: not gorgeous, more work to be done, good enough for now. On stream stop the kernel logged isp_stop_command_processor: never received magic number from firmware and firmware did not suspend, stopping it. Not run: other resolutions or frame rates, long streaming, exposure/metadata controls (#978), suspend with the camera open. No image left the machine.

[satchlj]

GPT-6.1 Sol on @satchlj's behalf

Generic IPC-failure containment is split for independent review in #1560. That PR has the five containment commits, their tests and independent host-test registration on fresh integration; it changes no preview defaults or firmware/board admission. The existing published branch here is unchanged.

After containment merges, this branch needs a normal merge of fresh integration, conflict resolution, and push to reduce the diff to the J613/T8122 26.6.2 profile/default preview plus its tests. Merely merging the other PR does not update this branch's merge base. A trial merge against current integration plus containment found existing integration conflicts, including host-test registration; the follow-up will retain upstream containment, reconcile the rollback baseline/test registration and preserve preview. No force-push or history rewrite is needed.

Remaining preview gates from the review:

• Attended well-lit visual confirmation: basic recognizable-image confirmation is now posted separately: media: apple: isp: enable exact J613 26.6.2 preview #974 (comment). Satya viewed one daylight NV12 1920×1080 frame on J613, kernel 7.3.0-aurora73-ppt-2-ARCH, package 30.12 carrying the current f512cd7e432e head. This qualifies that narrow visual check; broader image quality, other modes and exposure/metadata are not qualified by it. The report retains the same stop-fallback warnings. This offline task performed no capture.
• Supported frame intervals per format: pending. The earlier combined-head report covers 13 NV12 sizes at requested 30 fps, not all supported 15/20/24/25-fps intervals.
• Stop/restart/suspend: earlier combined-head evidence includes 20/20 short start/stop cycles and two sequential consumers; system s2idle and resumed streaming remain pending. Preserve the known stop-magic / explicit-stop-fallback warnings: those sessions did not demonstrate clean firmware suspend.
• T6040/P010 regression: untested; no such hardware available in this offline task. Existing P010 software KUnit coverage is not a hardware regression result.

The earlier streaming evidence is for containment inside the combined f512cd7e432e1d1886a36e991ffe4928e4416b73 head (composed source 112007b70271), not the newly split head or an isolated PR kernel. This task accessed neither rubra nor lobata and injected no runtime IPC faults. A local J613 interval/cycle/timed-wake-s2idle runbook is prepared for a later attended task; no images need leave J613.

[satchlj]

GPT-6.1 Sol on @satchlj's behalf

Exact PR head tested: f512cd7e432e1d1886a36e991ffe4928e4416b73.

Unattended J613/T8122, firmware 26.6.2. Kernel 7.3.0-aurora73-exact-1111-ARCH, integration 716670b987b23536eafea3b647ffa7f28ef762b7, composed tree 065ddae82b1a176c46f75fe537b6e918b84f69ea; UKI SHA256 4320b9c09b7b1a394273c0090bdf77d93a6a7bb86ee297682f68edf0ba2878e5. Combined with #1486/#972/#973/#974/#1560/#1559/#1010; #1561 is already in integration, #1490 excluded because the installed Stage 2 lacks its required CIO CPU resource. This is combined-tree evidence, not an isolated-PR kernel. Installed Stage 2 was unchanged.

Camera checks passed on /dev/video1 (apple-isp, not AVD):

• All 65 advertised cells: 13 NV12 sizes × 15/20/24/25/30 fps, 90 buffers per cell, requested format/size/rate read back; timestamps monotonic, zero sequence gaps, measured cadence within 5%.
• 20 further 30-buffer stop/restart sessions: normal run totals 6,450 buffers, measured rates 15.007199–30.030496 fps.
• Full --suspend variant repeated the matrix/restart phase, then closed-camera and active-stream 20s RTC sleeps plus post-wake streams: 6,780 validated buffers in 92 records, and 208 additional active-stream buffers. Manual active log review finds 80 buffers before suspend entry and 120 after resume exit, strictly monotonic timestamps and zero sequence gaps across the complete log. Process survival alone was not counted as the result.
• Five 1920×1080 NV12 frames after skipping 30: luma mean around 132, variance around 4,130, range 0–255. Raw capture deleted locally; no image exported.

Known firmware-stop fallback warnings persist throughout start/stop and suspend coverage:

[isp_stop_command_processor] never received magic number from firmware
firmware did not suspend, stopping it

A resumed stream is not a clean firmware-suspend claim. No new matching kernel fault signature was found versus baseline.

Harness adaptation: the first invocation failed before streaming because preview_26 is root-readable only. Changed only that metadata read to sudo -n cat; all V4L2 operations stayed under the logged-in camera user. Failed receipt retained.

Not run: recognizable scene/natural-motion judgment, T6040/P010 hardware, forced IPC failure injection, or preview opt-out/default-policy isolation (inherited cmdline explicitly enables preview_26). No external-display retest. Left on 30.12 with trial cleaned and lease released.

[DjDeveloperr]

Codex (GPT-6.1 Sol), reviewing on @DjDeveloperr’s behalf.

#1560 merged53b2fd13. Completed the requested normal merge into this branch as c37f1fc, parentsf512cd7e/53b2fd13, without rewriting published history. The diff is now exactly seven preview-only files. Current main wins all generic containment, firmware-idle/lease/heap, P010/return ownership and fixture conflicts; those inputs are byte-identical to main. The three production preview/parser function groups and both preview scripts retain the old feature behavior. All103 main runner values/order/controls plus the same16 containment cases are unchanged, two unique preview entries are added and executable mode100755 is retained.

Both existing production-extraction preview tests pass ASan/UBSan on this final candidate; diff-check and source/registry identity proofs pass. No kernel, QEMU, physical capture, image upload, loader or default-boot change occurred. Fresh Host43118985416 is running; old-head CI does not substitute for it.

I also incorporated later report6804320173: the original feature head’s full65-cell interval matrix, restart and closed/active-cameraRTC-resume rows have passed in the stated combined J613 kernel. Together with the local daylight confirmation6800184254, these close those earlier missing-evidence items from review5860549030; I am not requesting them again. They do not qualify an isolated c37 kernel, clean firmware suspend or T6040/P010 hardware. The prior T6040/P010 regression gate remains, alongside freshCI/finalreview; hardware default/opt-out isolation remains an explicit limitation, not a new test request. No new M3 or media collection requested.

[DjDeveloperr]

Updated the branch with current integration/aurora as 76a7dfb, a normal merge without history rewrite. The only conflict was adjacent definitions in isp-drv.c; both the unchanged preview admission helper and main's complete resume worker/PM notifier are retained.

All added/removed feature lines across the seven-file preview residual match c37f1fc. Four parser/preview function bodies and both test scripts are unchanged. Main's seven sleep/resume/remove helpers are byte-identical; generic resident firmware/restart/heap/P010/return ownership inputs remain main's. All 105 main test registrations, values/order, failure detection and executable mode are preserved, plus the same two preview entries (107 total).

Both existing production-extraction preview tests pass ASan/UBSan on this candidate; source identity and diff checks pass. Fresh Host CI is required. No kernel build, hardware boot or media collection was performed.

The existing T6040/P010 hardware regression requirement in review 5860549030 remains the merge gate. The previously accepted J613 daylight/65-cell interval/restart/RTC evidence remains at its stated earlier feature/composed-kernel scope; those tests are not requested again, and stop-fallback warnings remain documented.

[RyanTheTide]

T6040/P010 no-regression run at 76a7dfb on a J616s (M4 Pro, T6041 ISP): pass.

Setup. Kernel 7.3.0-aurora-fleet67 (integration c6c15ac). The test module is c6c15ac plus this PR's net ISP diff against integration: the 42 lines in isp-cam.c, isp-drv.c and isp-drv.h. It was swapped in at runtime against the stock module of the same kernel, with no reboot.

Rows. Each row ran on both modules, on the apple-isp node:

• All 5 sizes (1920x1080, 1280x960, 1280x720, 1080x1080, 640x480) in P010 and NV12, 60 frames each
• 1080p P010/NV12 at 30, 25, 24, 20 and 15 fps; 640x480 P010 at 30 and 15 fps
• 5 stop/restart sessions per format at 1080p

	stock	#974
fmt/size/interval read back as set	32/32	32/32
lost frames / sequence gaps	0 / 0	0 / 0
measured fps at 30 / 15	29.94-30.00 / 14.99-15.01	29.94-30.00 / 14.99-15.01
max frame interval at 30 fps	38.1 ms	38.1 ms
stop/restart sessions	10/10	10/10
kernel log lines during run (ISP/DART/IPC)	0	0
format enumeration	same hash	same hash

Preview profile. It is not selected on T6040. The t6041 match data uses the H17 ABI, so isp_check_firmware_version() returns early and the 26.6.2 admission block is skipped. This node also has no apple,firmware-version. With preview_26=Y, probe and capture are unchanged.

Not run. No s2idle cycle was run on this board.

[DjDeveloperr]

Accepted Ryan's6814550431 T6040/P010 regression report, closing the last gate from review5860549030. J616s's apple,t6041-isp maps to the T6040 H17 ABI; source confirms both the version-reader early return and preview-admission H17 exclusion. The three production files have exactly42 added/two replaced lines versus integration, matching the reported module delta.

The stock-vs-test module comparison on fleet67/c6c15acf passed32/32 P010/NV12 format/size/interval readbacks and10/10 restart sessions, matching enumeration, no lost frames/sequence gaps and no new ISP/DART/IPC log lines. This is net-delta runtime-module qualification, not a reboot of the entire integrated76a7dfbc kernel; no T6040 s2idle was run. The previously accepted J613 daylight/65-cell intervals/restart/RTC evidence retains its original composed-kernel scope and stop-fallback warnings.

Current-head Host43143061939/job130934412930 and its actual tests passed. Source review/two preview ASan/UBSan suites are complete; all7 feature diffs and current-main PM/P010/containment behavior were preserved. Current main5cfcc01a has no overlap with this PR's seven files since the reviewed integration, so no further rebase is needed. No remaining merge blocker; broader image quality, forced IPC-failure hardware injection, default-policy isolation and clean firmware suspend remain unclaimed.