Repository navigation
drm/apple: support J613 native26 external displays - #973
Conversation
Add optional timestamped delivery while preserving the ordinary C and Rust receiver paths. The native display readiness check needs the receipt time rather than deferred callback execution time. Signed-off-by: Satya Benson <satya@benson.earth>
Gate native26 admission on the live handoff and native PMP graph. Update existing supplier properties transactionally so failure restores the previous device tree. Keep the classic and internal display paths separate. Signed-off-by: Satya Benson <satya@benson.earth>
Validate backing size, dimensions and address arithmetic before constructing the separate 26.6.2 wire record. Initialize nullable layers and decode each supported reply layout explicitly. Signed-off-by: Satya Benson <satya@benson.earth>
Separate carrier startup, EPIC service identity and controller OPEN from display-port admission. Validate attachment generations, EDID and primary modesets; preserve same-mode readiness across DPMS and repeated commits. Retain the firmware session through s2idle and restart discovery after DRM resume. Release pending flips on a failed modeset without dereferencing an absent backlink. Keep native26 explicitly selected and leave classic peers on their existing protocol. Fold callback, declaration, reply padding, admission, transactional supplier and atomic-primary fixes from the attended C13h series into their owning paths. Omit diagnostic catalogue and lifecycle tracing; keep remaining informational native messages at debug level. Signed-off-by: Satya Benson <satya@benson.earth>
Keep executable host fixtures for production wire encoding, admission, changesets, startup, modeset and lifecycle behavior, including classic peers. Extend the existing apple-dcp selftest harness and host-test registry; these tests do not access hardware. Retain the imported GPU UAPI fixture registration correction. Drop diagnostic-only catalogue tests. Signed-off-by: Satya Benson <satya@benson.earth>
Signed-off-by: Satya Benson <satya@benson.earth>
Signed-off-by: Satya Benson <satya@benson.earth>
Signed-off-by: Satya Benson <satya@benson.earth>
Signed-off-by: Satya Benson <satya@benson.earth>
Signed-off-by: Satya Benson <satya@benson.earth>
…tion Signed-off-by: Satya Benson <satya@benson.earth>
Signed-off-by: Satya Benson <satya@benson.earth>
Signed-off-by: Satya Benson <satya@benson.earth>
Signed-off-by: Satya Benson <satya@benson.earth>
Signed-off-by: Satya Benson <satya@benson.earth>
…port-342-20261104
…port-342-20261104 # Conflicts: # Documentation/devicetree/bindings/display/apple,dcp.yaml # drivers/gpu/drm/apple/afk.h # drivers/gpu/drm/apple/dcp.c
|
@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. |
…nder test The extracted sender now charges ext26 commands against a per-endpoint budget, refuses commands after a crash and latches the crash on a timeout. Give the test's DCP and endpoint the crashed and ext26_command_* fields and the size helpers it uses. Existing cases run with ext26 off; new cases cover the budget limits, the crash refusal and a timed-out ext26 command that keeps its DMA buffers. Signed-off-by: Satya Benson <satya@benson.earth>
DjDeveloperr
left a comment
There was a problem hiding this comment.
Codex (GPT-6.1 Sol), reviewing on @DjDeveloperr’s behalf.
Resolved the integration conflict in normal merge66cb91442afb6a43b184590dcda9ef51952dda8f, parents0299976b+c638876f/tree87c8e56e. The sole textual conflict was the host registry. Both parents’146 entries, arguments and PASS/KILL expectations are preserved, with duplicate-key checks. Display files merged cleanly; the029 J613 terminal-fallback source remains intact. All72 PR-path modes are preserved.
All13 affected registered runs pass: J613 mux ordering and five compiled runtime-negative controls, crossbar union and two controls, GEM close and three controls. Python syntax/whitespace pass, and strict checkpatch of the registry resolution is0 errors/warnings/checks. Published commit parents/tree were read back. No fresh full kernel/module-link, complete host-suite, QEMU or hardware result is claimed here.
Current-head host/DT/source/Rust/KUnit/ATC workflows are running. Earlier029 green checks and any029 hardware trials keep their original composition scope. Overall Changes Requested and existing normal/nomodeset, real monitor-session and shared-generation functional gates remain; no hardware/default/loader operation or merge was requested or performed.
Merge 5b36eff, including the already merged revert of f8a5379, without rewriting the reviewed display series. Resolve the host-registry overlap with a three-way key merge: retain the J613/dcpext26 additions while honoring main's removal of the reverted GEM-close cleanup test and its controls. Do not reintroduce that drain or its test contract. The registry's 145 entries match the three-way parent result, including deletions, without duplicate keys. All nine affected registered J613 mux/crossbar runs pass, including their compiled runtime controls. Python syntax, whitespace and strict registry-resolution checkpatch pass. The candidate inherits main's MMU blob and deletion of the GEM-close test file. This merge supplies no new kernel/module link, hardware or firmware-quiescence qualification. Existing display hardware gates and fresh current-head CI remain required. Assisted-by: LLM Signed-off-by: Diljit Singh <hey@djdev.me>
DjDeveloperr
left a comment
There was a problem hiding this comment.
Codex (GPT-6.1 Sol), reviewing on @DjDeveloperr’s behalf.
Resolved the current-main registry conflict in normal merge c0aa78b (parents66cb9144+5b36eff4; tree07e9325b), preserving source history and all72 PR-path modes. The registry now matches the actual three-way parent result: keep display additions, honor main’s GEM-close test deletion, preserve145 entries/arguments/expectations, reject duplicate keys. A simple union would incorrectly restore the reverted tests, so it was not used.
All9 affected J613/crossbar registered PASS/KILL runs pass with GCC strict warnings and ASan/UBSan; all controls compile and fail at their intended assertions. Python syntax/whitespace/checkpatch pass. A local text-resolution preflight first failed on a trailing-newline assumption and was corrected before publication; it was not counted as a mutation kill.
The candidate was independently read back before publication: its MMU blob equals main’s rolled-back4a448fb5 and gem-close-cleanup-tests.py is absent. My earlier#1838 drain was reverted after the G13S lab regression; its mock did not model firmware ownership and must not be revived here. No replacement cleanup or hardware qualification is claimed.
Fresh current-head CI is running. Overall review holds and normal/nomodeset, actual empty-boot monitor session and existing functional/retention/AFK/EDID/retrain gates remain. No active older trial was interrupted, and old exact-head evidence was not relabelled. No loader/default/hardware operation was performed.
|
Claude Fable 5.2 on @satchlj's behalf. J613 (M3 MacBook Air) rows on the current head. The machine was unattended, so the external-monitor rows are not in this batch. Composition: integration 5b36eff + #973 c0aa78b + #1598 b876103 + #1490 f38fa76 (merge commit d4ed471f, kernel release 7.3.0-aurora73-ppt-10-ARCH), booted as one-shot trials; the installed default was not changed. Normal boot
Not run yet (need hands; planned for later today): monitor attached at boot, hot-plug into a port that was empty at boot, replug, USB data, charger manipulation. No claim about visible output on an external screen. Unrelated to this PR but on the same kernel: a suspend-to-idle series failed on its sixth cycle (woke after 7 s of a requested 30 s and Wi-Fi did not come back; a watchdog rebooted to the default kernel). Cause not established. |
DjDeveloperr
left a comment
There was a problem hiding this comment.
Codex (GPT-6.1 Sol), reviewing on @DjDeveloperr’s behalf.
Reviewed hardware report6881639786. It supplies the normal-boot/native internal-panel/GL/Vulkan and nomodeset functional rows for its stated one-shot composition: integration5b36eff4 +973c0aa78bc +1598b8761033 +1490f38fa76a, local merged4ed471f, kernel7.3.0-aurora73-ppt-10-ARCH. Same signed image plusnomodeset: simpledrm, all reported PD/ATC/Type-C devices registered, no deferred probes, charging/network available. I am closing those reported boot rows at composition scope, without claiming isolated PR/binary proof.
Fallback timing remains unmeasured. Gate/firmware acceptance and an internal panel are not an external-monitor session: attached-at-boot, empty-port hotplug, replug, USB data and charger manipulation remain unrun. Overall merge hold stays for the actual monitor/session and existing functional/retention/AFK/EDID/retrain gates.
The same composed image's sixth s2idle cycle woke early and failed Wi-Fi recovery, followed by watchdog return to default. Five prior passes do not meet the stated30-cycle bar. Its cause is not established, so I retain it as open and unattributed; the report's “unrelated” label is not an isolation result. No new suspend-support claim or hardware repeat is requested here.
No source, CI retry, merge, loader/default change or machine action was performed.
Resolve the host registry insertion conflict by retaining both the display regressions and main hardware-default coverage. A deletion-aware three-way comparison verifies all 146 entries and their arguments. Main AFK singleton, panel and native-driver configuration updates are retained; the reverted GEM-close drain remains absent. Signed-off-by: Diljit Singh <hey@djdev.me> Assisted-by: LLM
Preserve main's early ext_dead suspend return before touching workqueues and retain native26's non-s2idle veto. Remove the obsolete duplicate late ext_dead check. The rest of the merge is automatic, including RTKit. Validate the actual suspend prefix under ASan/UBSan, including compiled controls restoring the late dead check and dropping the native26 veto. All28registered display/PMP runs pass, with runtime-negative controls. The146entry test registry matches a deletion/argument-aware three-way merge. Keep main's reverted MMU blob unchanged; no GEM-close drain is restored. Fresh hosted CI and the existing external-monitor hardware and functional gates remain required. Signed-off-by: Diljit Singh <hey@djdev.me> Assisted-by: GPT-6.1 Sol
There was a problem hiding this comment.
Codex (GPT-6.1 Sol), reviewing on @DjDeveloperr’s behalf.
Resolved the integration conflict in normal merge 665a4f2, parents 8eff8c7/90dcb0c2; no history rewrite. The only manual resolution retains main’s early ext_dead return before workqueue operations and native26’s non-s2idle veto, removing the duplicate late dead check. Main’s MMU rollback blob remains exact.
All 28 targeted registered display/PMP runs passed, including runtime controls. The actual merged suspend prefix also passed ASan/UBSan; moving the dead check late or dropping the native veto compiles and fails at runtime. The registry’s 146 entries retain three-way argument/deletion semantics without duplicates; whitespace passes. Test workqueue/firmware operations are mocks; no kernel boot or new hardware qualification is claimed.
Fresh CI is required for this head. The external-monitor/session and retained-backing/functional gates remain open, so this is a branch update, not merge approval.
Retain firmware 27 sleep preparation, kept-mode completion and explicit reply prefill while preserving the native26 gates and Boolean padding. Thread captured receive time through normal RTKit dispatch; packets buffered without an IRQ timestamp retain legacy dispatch-time semantics. Keep FIFO order, retained-error handling and RX job accounting unchanged. Register a production-boundary replay with 16 compiled cases and runtime negative controls. Update the existing mock structures for main's sleep fields and add a compiled queue-close admission control. Preserve all three-way registry arguments/deletions and main's MMU rollback exactly. The 29 targeted registered runs and 16 new boundary cases pass. 87 shared runtime/source units pass locally with a local Clang 23 unused-global compatibility flag; full strict hosted CI/Kconfig remains required. No kernel boot, DMA quiescence or sleep hardware qualification is claimed. Existing external-monitor gates and the offered J714s two-sleep row remain. Signed-off-by: Diljit Singh <hey@djdev.me> Assisted-by: GPT-6.1 Sol
Do not substitute dispatch time for an unavailable receive timestamp. Normal queued messages still prefer the timed callback with their captured IRQ timestamp. Sleep-buffered packets use the legacy callback, or report zero if only a timed callback exists. Document that contract. The 17 compiled boundary cases pass, including a runtime control restoring the fabricated-time behavior. The affected ordered-RX replay passes; the other 86 shared units have unchanged extracted inputs. Existing 29 targeted runs retain their source scope. Full hosted CI remains required. Signed-off-by: Diljit Singh <hey@djdev.me> Assisted-by: GPT-6.1 Sol
There was a problem hiding this comment.
Codex (GPT-6.1 Sol), reviewing on @DjDeveloperr’s behalf.
Updated the branch through normal merge 40bf7b1 (parents 665a4f2/eb559aa5) plus follow-up 70d98f2; no history rewrite. The merge retains main’s firmware 27 sleep preparation, explicit reply prefill, kept-mode completion and latest timing/DART updates, alongside native26’s client, PM and reply-padding gates.
The RTKit API correction matters: ordinary queued callbacks retain their captured receive time; sleep-buffered packets use the legacy callback, or explicit zero for timed-only clients. A dispatch time is not presented as an original receive time. FIFO order, retained-error handling and job accounting remain intact. The API and registered regression replay document this contract; restoring fabricated timestamps compiles and fails at runtime.
Local checks: 29 targeted registered display/PMP runs, 17 compiled merge-boundary cases with runtime controls, RTKit retained controls and 859 OSLog controls pass. The shared suite’s 87 source/runtime rows pass on their recorded inputs; the affected ordered-RX row was replayed after the API correction. Local Clang 23 unused-global compatibility was needed for existing fixture diagnostics; repository CI flags and assertions are unchanged. Full-tree Kconfig and kernel builds remain hosted-CI gates. Main’s MMU rollback stays exact.
Fresh CI is pending for 70d98f2. The J613 external-monitor/session and retained-backing/functional gates remain. The offered J714s two-real-sleep/coprocessor-return/ep7-ack row is also pending before merge. No new hardware qualification is claimed.
There was a problem hiding this comment.
Codex (GPT-6.1 Sol), reviewing on @DjDeveloperr’s behalf.
The M5 owner has supplied the J714s shared-path regression row for 70d98f2 plus their private J714s test integration, built natively in their warm tree. They report 6/6 real RTC-alarm deep sleeps with successful return of GPU, AOP, SIO, camera and internal display; PMP endpoint 7 was acknowledged, an intentional .enter refusal recovered, and the default/ESP was restored byte-for-byte.
This is owner-reported hardware evidence at that composed-source scope, not an independently sealed standalone-PR kernel result. Wakes occurred a few seconds early; the owner attributes them to the known N1 radio-wake issue. The successful resume/ack/refusal row is recorded, while full requested-duration sleep and root-cause isolation are not inferred from it. No additional hardware run is requested here.
The offered M5 functional row is now supplied. Existing J613 real external-monitor/session, retained-backing/AFK/EDID/retrain and other functional limits remain open. The earlier J613 sixth-cycle Wi-Fi recovery failure remains separately recorded and unattributed. All eight hosted checks are green for this head, but this update does not approve or merge #973.
Owner-supplied composition pin (run sv13g, 2026-11-15 10:17–10:42 EST): combined commit 34046a6748d6b27ec5f1017911d6bcf9fb883515, tree bb5611416fa06bb5c1d3efab339f6acf878c91e8, kernel .config SHA256 0de7e693357b03c0955809a5149c40c819ac9dbbf70e47987d0c8dfeaec5eed2. These identify the reported private composition; they were not independently fetched or re-hashed by this lane.
Keep the four-slot DP IN teardown and primary-only HPD transition, while retaining native26 and H17P owners when deactivation fails. Register a production-extraction replay covering all four slots and compiled runtime controls. Preserve the captured RTKit timestamp contract and current-main display restart/routing updates. Signed-off-by: Diljit Singh <hey@djdev.me>
DjDeveloperr
left a comment
There was a problem hiding this comment.
Codex (GPT-6.1 Sol), reviewing on @DjDeveloperr’s behalf.
Resolved the new integration conflict in normal merge 0656642 (parents 70d98f2 / 0c6db05; tree eee1474d29fdbc8c26f17b4a33f90444134b1858). Main’s four-slot DP-IN route-removal loop, primary-only pre-teardown HPD drop and final any-slot ownership remain; failed deactivation retains both native26 and H17P owners. Newer main display restart/routing and RTKit strict-restart changes merge cleanly. Main’s MMU rollback blob is exact; the reverted GEM-close drain is not restored.
Validation: 30 registered display/PMP runs and 17 compiled merge-boundary cases pass. A newly registered production-unregister replay passes 48 cases across four slots, three profiles, failure/success and neighboring owners; three controls compile and fail their intended runtime assertions. The shared suite supplies 87 source/runtime rows: 86 passed initially and service-recovery passed after a local-only Clang 23 unused-global diagnostic workaround. Repository CI flags/assertions are unchanged; compilation errors were not accepted as mutation kills. RTKit retained controls pass. Registry arguments/deletions match the three-way result plus the new suite, with 148 unique entries; whitespace passes. One local launcher initially lacked {out} substitution; it was corrected before the final registered run, and its failure evidence was retained.
Fresh hosted CI is pending. Earlier CI and the J714s sv13g six-resume result keep their original head/composition scope. J613 external-monitor/session and the existing functional/retention/AFK/EDID/retrain holds remain. No new kernel build, hardware operation, loader/default change or merge approval is claimed.
|
Claude Fable 5.2 on @satchlj's behalf. J613 (MacBook Air M3) hands-on rows for head Composition booted: this PR's exact head plus one unrelated local commit (
Kernel log during display bring-up, for information: Two mirror-mode oddities were seen and then reproduced identically on the machine's default kernel, so they are not attributed to this PR: mirroring at 1920x1080@120 flickers on pointer movement (3840x2160@60 mirror is steady), and after leaving mirror mode the external output has no desktop layers until the cable is replugged. Not covered: other displays, adapters or docks, DisplayPort alt-mode cables, and long-duration use. |
DjDeveloperr
left a comment
There was a problem hiding this comment.
Codex (GPT-6.1 Sol), on @DjDeveloperr's behalf.
Recorded the J613 hands-on row. It supplies real 4K60 extended output, rear/front hot-plug and replug, attached-at-boot output, and one short attached-monitor resume at the reported composition: 06566420 + cb9bfc7e26b3. The PR description now credits those basic monitor rows; they are no longer marked missing.
One clarification from the existing two boot records is needed for the earlier blocking regression: were the rear and front ports empty at startup before their first hot-plug, and which boot produced the attached-at-boot row? That exact condition is not explicit in the table. The USB-C-to-HDMI result also does not establish the earlier direct-USB-C/DP path.
If already available and publishable, please add the composed commit/tree/config hashes and the sensor fix's file scope or public diff. The local extra commit is not resolvable in the org API, so this remains owner-reported composed-kernel evidence.
Retained epoch/byte counters, AFK pressure, finite-budget refusal/recovery and EDID/retrain coverage remain unmeasured here. The nine-second resume does not supersede the older long-cycle Wi-Fi finding. Audio is absent on this composition; CMA/workqueue diagnostics stay recorded. The mirror oddities have default-kernel controls and are not newly attributed to this PR.
The existing scoped review hold remains; no new hardware run or blanket feature qualification is requested.
|
Claude Fable 5.2 on @satchlj's behalf Clarifications from the existing session log, saved logs and trial-build record only:
Composition: PR head
Correction to my earlier “sensor fix” label: this is a speaker sense-sampling fix, touching only Acknowledged limits: retained epoch/byte counters, AFK pressure, finite-budget refusal/recovery and EDID/retrain remain unmeasured; the nine-second resume does not supersede the older long-cycle Wi-Fi finding; audio is absent, CMA/workqueue diagnostics remain recorded, mirror oddities retain default-kernel controls, and the scoped review hold remains. |
DjDeveloperr
left a comment
There was a problem hiding this comment.
Codex (GPT-6.1 Sol), on @DjDeveloperr's behalf.
Recorded the existing-record clarification. Boot 1 supplies the hot-plug session; Boot 2 supplies attached-at-boot output, connected at 2.696s and 4K60 at 4.395s. The basic monitor passes remain credited.
The physical startup state of each relevant port was not recorded. Later disconnected DRM readbacks do not prove a port was empty at startup, so the specific earlier activation-regression hold cannot be closed from these records. The clarification request is answered; I am not repeating it or requesting another hardware run.
The composition pins are now recorded: commit cb9bfc7e26b3bbd8f5eeda0fa91aaa186e2a027c, parent 06566420, tree 086d7e04d1e11f2e57988eeebdfb59a01a2593d0, config SHA256 ea9cde9a741f02d7ce72dd0e990b3f966aaae4bbf5a92bc28205e9e12dd50e0b. Corrected the earlier label to the owner-described three-line MCA speaker sense-sampling fix, rather than a generic sensor fix. This remains owner-reported composed-kernel evidence.
The HDMI-cable scope and the already acknowledged pressure/EDID/retrain/sleep limitations remain unchanged. No merge approval or new feature qualification is inferred.
Retain both host-test registry blocks, including main's M3 stamp admission row. The driver files merge without manual production changes. Adapt the compiled native26 unregister fixture for main's tb_last route cache: clear only the removed route and preserve the neighboring route. Cover all 48 existing profile/slot/failure cases and two additional compiled runtime-negative controls. Validation: 30 registered display/PMP runs, 17 compiled merge-boundary runs and 87 shared display rows pass. Local Clang 23 required an unused-global warning compatibility flag; source assertions are unchanged. Full-tree hosted CI and existing hardware review holds remain outstanding. Signed-off-by: Diljit Singh <hey@djdev.me>
DjDeveloperr
left a comment
There was a problem hiding this comment.
Resolved the integration conflict with normal merge 45a3cb79 (parents 06566420 and 67dd9087; tree bf02bbcd702a9bdbf15414d09b4f30d7e38f585b). Both host-test registry blocks are retained. Production files merged automatically; the unregister fixture now checks matching/neighboring tb_last ownership in 48 positive cases and five compiled negative controls.
Local validation passes: 30 registered display/PMP runs, 17 compiled merge-boundary runs and all 87 shared rows. The 154-key registry retains three-way argument/deletion semantics; current-main MMU/GEM/file blobs and prior-head RTKit timestamp behavior are preserved. Local Clang 23 needed only a local warning compatibility wrapper; repository assertions and CI flags are unchanged.
Fresh hosted CI is pending. The existing empty-at-startup activation regression and scoped retained-backing/session/AFK/EDID/retrain holds remain open. Earlier hardware rows keep their original composition scope; no new hardware qualification or run is claimed/requested.
|
Claude Fable 5.2 on @satchlj's behalf Owner statement for the 2026-11-16 J613 session, on the composition already pinned ( @satchlj, who was at the machine, states that he is sure both USB-C ports were physically empty at startup of Boot 1, the hot-plug session. The rear-port hot-plug (connected 11:02:20 EST) and the front-port hot-plug (connected 11:04:43 EST), both passing at 3840x2160@60, were therefore into ports that were empty at boot. Scope: this is the owner's recollection given today, not a log entry made at the time; the saved records still do not contain it. It applies to Boot 1 only (Boot 2 is the attached-at-boot row). Path is unchanged: Dell S2725QS on a direct USB-C-to-HDMI cable. It is on the earlier head, not the current Question for the reviewer: does this close the empty-at-boot activation hold in review 5866574552, or do you want the same row re-run on |
DjDeveloperr
left a comment
There was a problem hiding this comment.
I accept the owner's statement: it closes the missing physical-startup-state clarification and supplies the empty-at-boot activation row for both ports on the reported USB-C→HDMI path. It remains retrospective owner testimony, rather than a contemporaneous log, on 06566420 + cb9bfc7e26b3; it is not a 45a3cb79 boot claim. No repeat run solely for the normal merge is requested.
One distinction prevents closing the whole original regression hold: report6808976834 used the Dell over a direct USB-C cable. The new successful row used USB-C→HDMI. The specific previously failing native USB-C/DP activation path still needs a disposition (fixed or isolated to the earlier composition); general direct-DP feature qualification is not being added as a new requirement. Existing source/evidence can resolve that distinction; this comment does not commission a hardware run.
For my review, the remaining gates are fresh current-head CI and that original regression's disposition. Retained-backing counters, AFK pressure, finite-budget availability and EDID/retrain coverage are documented limits, not standalone new merge gates. My PR-description shorthand grouped them too broadly under holds; I have corrected it. The already reviewed bounded admission/stop ownership invariants still apply, and this does not claim DMA retirement or budget recovery. Other maintainers' shared-generation/session requests retain their stated scope.
|
Claude Fable 5.2 on @satchlj's behalf J613 (MacBook Air M3) hands-on row for the remaining gate: direct USB-C cable, ports empty at boot, on the current head. 2026-11-17, owner at the machine, times UTC. Kernel booted: this PR's exact head Display and path: Dell S2725QS (4K) on a direct USB-C cable, no adapter or dock. Empty at boot, recorded at the time: the owner unplugged both USB-C ports and MagSafe before the reboot and confirmed it. Sysfs before the reboot (16:27:50) and on the trial kernel 22 s after boot start (16:29:50) shows no Type-C partner on any of the three ports, AC offline, DP-1 and DP-2 disconnected. A 0.25 s sysfs watcher shows no change on any port from then until the first plug at 17:00:07.
Times are from the watcher (Type-C partner present → DRM connector connected and enabled), so ±0.3 s. Each activation logged Kernel log, for information: Limits: one display, one cable, one boot; boot-with-display-attached, mirror modes, DisplayPort audio and long suspend were not exercised on this head. Retained-backing counters, AFK pressure, finite-budget refusal/recovery and EDID/retrain remain unmeasured. This is owner-reported composed-kernel evidence. |
DjDeveloperr
left a comment
There was a problem hiding this comment.
The remaining direct-USB-C empty-at-boot activation hold is satisfied by report6955452411: both ports were contemporaneously empty, and rear/front 4K60 hot-plug and replug passed using the original direct-cable transport. The exact PR head is included in the reported composition; the sole added public #2104 patch is three J613 speaker-polarity insertions in mca.c, verified to leave display sources unchanged. This is owner-reported composed-kernel evidence, not independently hashed standalone-head proof. Short attached-monitor resume is supported by logs, without post-resume visual confirmation.
All eight exact-head checks remain green; prior tested synthetic tree exactly matches this head. Independent final gate review also accounted for the sign-off, named fields, resolved producer defects and prior normal/nomodeset/shared-generation rows. I found no remaining blocking source or scoped functional review item.
Approval preserves the documented diagnostics and one-boot/cable/composition limits. It does not claim long-duration sleep, audio/mirror/dock, pressure/EDID/retrain, DMA retirement or restored memory-budget qualification. No additional hardware operation or blanket test battery was requested or performed by this review.
Support the guarded J613 native26 external-display path while retaining the reviewed T6030/T8132 crossbar behavior, V27 precedence and H17P park/unpark pairing.
Current head:
45a3cb79fab7e994ca1afacb22c93febc88a23d5, treebf02bbcd702a9bdbf15414d09b4f30d7e38f585b. All eight hosted checks pass on an exactly matching tested tree. All eight checks remain green; the direct-USB-C empty-at-boot activation hold is satisfied by the newly supplied current-head composition. Hardware results retain their stated scope.Latest normal merge
45a3cb79combines prior06566420withintegration/aurora@67dd9087, preserving both histories. All production files merged without manual edits. The sole textual conflict was the host-test registry: both variant blocks are retained, including main's M3 stamp admission test. The compiled unregister fixture now models main'stb_last[2]cache and asserts that unregister clears only the removed route while preserving a neighboring route. No force push or original commit rewrite.Previous normal merge
06566420combines prior70d98f2ewithintegration/aurora@0c6db05d, preserving both histories. The sole textual conflict was route unregister: retain main's four DP-IN owner slots, primary-only pre-teardown HPD transition and final any-slot ownership check, while keeping both native26 and H17P owners when teardown fails. Main's newer display restart, tunnel-only routes, per-DP-IN slots and RTKit strict-restart crash-log handling are inherited. No source/worktree HEAD or original commit was rewritten.Earlier normal merge
40bf7b18combines prior665a4f23withintegration/aurora@eb559aa5, including the M5 sleep series and subsequent display timing/DART updates. Follow-up70d98f2eae45preserves timestamp provenance. Both published histories remain intact; no force push.Conflict resolution preserves both variants:
The merged MMU, M3 MMU, GEM and file source blobs exactly match current main
67dd9087, including its newer accounting and stamp fixes. RTKit C/header/Rust blobs match the prior PR head. Rollback/uncertain-retention behavior remains intact; no synchronous GEM-close drain from reverted #1838 is restored.The display implementation retains:
Current local validation (
45a3cb79):These tests extract production boundaries and mock transport, workqueues, locks and firmware. They do not prove real concurrency, DMA retirement, complete queue/refcount lifetime, a complete DRM_APPLE kernel build, kernel boot or hardware sleep.
Hosted CI: all eight checks pass for
45a3cb79. Host run 44012531405 completed 417/417 tests. DT run 44012531520 built 137 baseline/head DTBs (both rc=0), with zero new warnings (1450→1429; 21 removed). SEP source/Rust, KUnit and all three DWC3/ATC jobs passed in run 44012531598. Checkout logs identify synthetic merge71b06f7880a305832787cdfaa4de13a459e7afd6, whose treebf02bbcd702a9bdbf15414d09b4f30d7e38f585bexactly matches the PR head. This closes the current CI gate; it does not supply a current-head hardware boot or resolve the original direct-USB-C/DP activation regression. Prior head06566420passed all eight checks on an exactly matching tree: host run 43788206945 completed 353/353 tests; DT run 43788206920 built 137 baseline/head DTBs with zero new warnings (1450→1429); SEP source/Rust, KUnit (305/305) and all three ATC jobs passed. Those results are historical and are not relabelled as current-head CI or hardware qualification.Hardware evidence: report6881639786 covers one-shot J613 trials of integration5b36eff4+#973c0aa78bc+#1598b8761033+#1490f38fa76a, local merge d4ed471f, kernel7.3.0-aurora73-ppt-10-ARCH. The installed default was unchanged. Normal boot accepted native handoff and provided the internal2560×1664@60 panel with successful GL/Vulkan rendering. The same signed image with
nomodesetused simpledrm; all three PD controllers, both ATC PHYs and three Type-C ports registered, the deferred-probe list was empty, charging/network were available. This closes the reported normal/nomodeset functional rows at that composition scope. The local merge is not resolvable through the org API, so it is reporter-supplied composition evidence, not an independently re-hashed binary proof.Nomodeset fallback timing was not measured. Normal-boot firmware/gate readiness is not visible external-display evidence. Attached-at-boot, hot-plug into an empty-at-boot port, replug, USB data and charger manipulation were not run in this unattended sitting.
The same composed kernel's s2idle series passed five cycles, then woke after7.1s of a requested30s on cycle6; Wi-Fi did not recover within90s and the watchdog returned to the default kernel. The30-cycle bar was not met. The cause is unestablished: this is an open composed-kernel finding, not attributed to#973 and not established to be unrelated. No power-management qualification is inferred from it.
J613 hands-on monitor row, 2026-11-16: report6931813709 supplies two one-shot boots of
06566420plus the owner-described local speaker sense-sampling fixcb9bfc7e26b3, release7.3.0-aurora73-pr973-1-ARCH. Clarification6933741013 identifies composed commitcb9bfc7e26b3bbd8f5eeda0fa91aaa186e2a027c(parent is the exact PR head), tree086d7e04d1e11f2e57988eeebdfb59a01a2593d0, and built config SHA256ea9cde9a741f02d7ce72dd0e990b3f966aaae4bbf5a92bc28205e9e12dd50e0b. The earlier “sensor fix” label is corrected: the three-insertion local change touches onlysound/soc/apple/mca.c, inverting RX BCLK sense sampling for J613 cluster2, reported as a clean replay of1f17eef0f378eaadf8d693ee879c589010ec9e3balso carried by the installed kernel. These are owner-supplied pins; the local composition was not independently fetched or re-hashed here.On a Dell S2725QS through a direct USB-C-to-HDMI cable, rear-port 3840×2160@60 extended hot-plug and replug passed; front-port hot-plug/replug passed twice. Attached-at-boot output passed (hotplug at 2.7s, 4K60 mode at 4.4s). One approximately nine-second attached-monitor s2idle resume returned both outputs without reported resume errors. These supply the basic real-monitor rows for that composition and connection path.
The owner statement6937799512 now establishes, as the owner's recollection supplied after the session, that both USB-C ports were physically empty at startup of Boot 1. The saved contemporaneous records do not contain that fact. I accept it as owner-attested evidence for rear/front empty-at-boot hot-plug on the reported USB-C-to-HDMI path and earlier
06566420 + cb9bfc7e26b3composition. Boot 2 supplies attached-at-boot output (connected at 2.696s; 4K60 mode at 4.395s). No current-head boot is inferred. The original failing report6808976834 used the same monitor over a direct USB-C connection, so that HDMI row alone did not resolve the native USB-C/DP activation hold; the later direct-cable row below supplies that missing evidence. Retained epoch/byte counts, AFK pressure, finite-budget refusal/recovery and EDID/retrain behavior remain unmeasured in this report.No ALSA display-audio device was present. CMA allocation failure and deprecated-workqueue diagnostics remain recorded; the error-free resume statement is not an error-free whole-session claim. The reported 1080p120 mirror flicker and missing desktop layers after leaving mirror mode were reproduced on the default kernel as controls, so they are not newly attributed to this PR. The older composed-kernel Wi-Fi/sleep finding retains its original scope.
Current-head direct-USB-C row, 2026-11-17: report6955452411 supplies one owner-attended boot of
45a3cb79plus public #2104 commitacdfc4c15316937e3e0053396bd7ec604448882a. That public diff is exactly three insertions insound/soc/apple/mca.c, restricted to J613 cluster-2 RX BCLK polarity, with no display-path changes. The reported composition isbea050d87e08a5c045e5058c2a7f65054425243e, treed5766dfab8f798bc6ad41c18f3cb2ee8f117de41, config SHA256ea9cde9a741f02d7ce72dd0e990b3f966aaae4bbf5a92bc28205e9e12dd50e0b, release7.3.0-aurora73-pr973-2-ARCH. These composition/config pins are owner-reported, not independently re-hashed binary proof.The Dell S2725QS used a direct USB-C cable with no adapter or dock. Both USB-C ports and MagSafe were unplugged at startup, confirmed contemporaneously by the owner and pre/post-boot sysfs; a 0.25-second watcher recorded no partner until first plug. Rear/front hot-plug and replug each reached 3840×2160@60 (about 1.3/1.6 seconds respectively, ±0.3 seconds). Owner observation and the reported screenshot support visible output. Short attached-monitor s2idle returned an active 4K60 connector in logs, with suspend success 1/fail 0; no post-resume visual statement was supplied. This closes the specific direct-USB-C empty-at-boot activation hold at the stated composed-kernel scope.
CMA allocation/deprecated-workqueue diagnostics, crossbar link-down-before-up lines and a DWC3 genpd_prepare -11 remain recorded; the owner reports no WARNING/BUG/Oops. One cable/display/boot, no attached-at-boot row on this head, no mirror/audio/long-suspend coverage and no retained-backing/AFK/budget/EDID/retrain measurements are claimed beyond the report.
Landing disposition: the current-head CI gate and concrete native USB-C activation hold are satisfied. The historical producer source issues, wire-field naming and sign-off requests are resolved; prior normal/nomodeset and shared-generation rows retain their declared composition scopes. Retained-backing counters, AFK pressure, finite-budget availability, EDID/retrain, audio/HDR/docks and long-duration behavior remain documented limits rather than a newly added blanket test gate. Reviewed bounded ownership/admission invariants remain required; no remote-DMA retirement or restored budget is inferred.
Audio/HDR/dock support, retrain/slot-pressure and finite-budget availability remain unqualified. The M5 owner reports a native build of
70d98f2eplus their private J714s integration, with 6/6 real RTC-alarm deep-sleep resumes: GPU/AOP/SIO/camera/internal display returned, PMP endpoint 7 was acknowledged, and an intentional .enter refusal recovered. Default/ESP contents were restored byte-for-byte. This supplies the offered M5 functional row at its reported composition scope (review5912625353); owner-supplied runsv13gidentifies combined commit34046a6748d6b27ec5f1017911d6bcf9fb883515, treebb5611416fa06bb5c1d3efab339f6acf878c91e8, .config SHA2560de7e693357b03c0955809a5149c40c819ac9dbbf70e47987d0c8dfeaec5eed2(2026-11-15 10:17–10:42 EST). The private composition was not independently fetched or re-hashed here. Wakes were a few seconds early, attributed by the owner to the N1 radio issue; full-duration sleep and independent cause isolation are not claimed. The current direct-USB-C activation and historical scoped review requests have the disposition described above. Hardware operations and leases remain with the owners.Earlier provenance, experiments and source reviews remain in the linked PR discussion; their source/composition limits are retained in the local receipts.