[satchlj]

GPT-6.1 Sol on @satchlj's behalf

Port the J775c (Mac Studio M5 Max) board description from the incrementally booted stack, preserving the installed loader's working compatibility layout. Covers CPU/cache topology, PCIe/N1/NVRAM/fans, USB/PD/router suppliers, desktop display and audio wiring, and disabled inventory. No Thunderbolt recovery or other driver series is included.

Dependencies and review order:

• Depends on arm64: dts: apple: t6050: add disabled acio0/1/3 routers to the shared files #1579, exact shared-router head 953cbcc02ebeced384136e416daa9cc4f27977e3, still open when prepared. That fork branch is retained in ancestry, merged with fresh integration de6a8d46f755237a012435a735251a5efb325d34. GitHub cannot use a fork-only branch as the base of an Aurora-targeted PR, so this targets integration and includes the dependency prefix until arm64: dts: apple: t6050: add disabled acio0/1/3 routers to the shared files #1579 merges. Review board commits after merge a2d763268; no duplicate shared-router implementation is proposed.
• Requires the isolated PCIe Clock PM fix in PCI: apple: mask Clock PM before enabling adopted T6050 endpoints #1618 for the tested J775c kernel path. That kernel change is deliberately absent from this DT PR.
• SD reader remains loader-dependent: adopted port1/GL9755 description is disabled until power/train by aura #163 (1f10b75f3bc315e745550c846ff85871ba261957, draft/unmerged). It is retained to match the booted board; SD functionality is not standalone or merge-qualified. No MMC config/driver series or SD media claim here.

Reviewable commits consolidate bisect-only steps while separating: desktop internal-panel DCP omission; inventory alignment; installed GPU paths/fallback compatible; MTP DART disablement; three ANE supplier references; final OPP/capture cleanup; and J775c-only bus-layout scoping. Original incremental commits all have author Satya Benson satya@benson.earth, preserved. Original copyright notices remain. The board is derived from Jeff Queisser's J775D and Erik Bourget's one-die J716c; Jeff's and Dj's shared foundations remain in ancestry, not recast as Satya's work.

Compatibility decisions backed by the retained J775c trials:

• Adding 102 inventory nodes did not return; changing only MTP DART iommu@294800000 to disabled passed. Active power/handoff remains unqualified.
• With MTP DART disabled, migrating existing properties did not return; restoring only three ANE_SYS supplier properties passed. Keep these when the -ane wrapper enables the DARTs; this is not an ANE workload result.
• Generic internal-panel dcp@48ae00000 booted but lost HDMI. Omitting only that node restored HDMI and display audio on this desktop.
• Installed Aura b6e6faf4 needs old GPU paths under /soc@2300000000 and apple,agx-t6050 fallback beside modern compatibles. Retaining both restored GPU metadata and renderD128. Modern-only paths/compatible need a separately qualified loader update.
• Scope the identity-mapped compatibility base to J775c; other boards retain integration's shared bus paths.

Hardware identity: final whole-stack 9c038a05a1aaf2c2722ca1e354b24125011c0f7c, with integration 716670b9 + Clock PM kernel b88dec9637cf3e4ce56d7b146af907ef4d58ff5f, build #6, installed Aura b6e6faf4. Source DTB SHA256 bffee06f90270ddb50d933082bdff86f01447a979e470521d5209ea492b0bb84. Final inc-final-whole health: CPUs0-17,62GiB,root NVMe,wired/Wi-Fi,USB2/3 hubs and microphone,HDMI connected,USB audio plus two display audio cards,GL9755,all four router DT statuses okay,alba TB20Gbps in both directions,renderD128. SDDM/NetworkManager/sshd active; zero failed units after the stable-path ac-power-mode service fix. Analog AOP card absent on this integration kernel.

Current boot identity: tonight's hardware receipt supersedes the earlier unbooted statement. Head 32149e08e8b191a500b40435eb0820cc78ec22fe booted with #1618 and integration 050f6a517, local combined merge fe75e6f6f5574140685ce268f675c9594c6b2294. ANE-wrapper DTB SHA256 dc88aa139f329bb568e48ba7ba80417e7d2c01ec6be49514a95ca24285dfd8d9 (127787 bytes). CPUs0–17,62GiB,NVMe,Ethernet/Wi-Fi,USB,HDMI,renderD128,GL9755 and four ACIO nodes passed boot/health; audio devices listed only, no playback. Known AOP/DCP AV firmware limitations and untested workloads are recorded in that receipt.

This offline update's first-parent commits and J775c DTB byte effects versus tonight's tree:

• ab72eb0b9f68527638e7627046e84e1b9da44648: documentation only, zero changed bytes.
• 3acbd5e3f1a8258267002005f9dae5f669a8136c: merge integration 050f6a517; zero changed bytes in both J775c DTBs versus the original PR tree.
• 9b4806242babd9df64c88b0d490917d9071f889c: remove duplicate GPU calibration YAML keys introduced by the merge, retaining upstream definitions; zero changed DTB bytes.
• 7759346bc4a0235d62d5c588f8ca6dc75228ab55: documentation warning-count update; zero changed bytes.
• fa191334fdaae077a50e09d1eb53a899a2334537: remove inherited duplicate G17 schema enum member; zero changed DTB bytes.

Final ANE-wrapper DTB is byte-identical to tonight’s booted dc88aa139f329bb568e48ba7ba80417e7d2c01ec6be49514a95ca24285dfd8d9; zero differing bytes. Normal-wrapper SHA256 bc057c1970d5f885fb37dff94497fc0a6ae2450eb87b5042219c93a99a42b077 is unchanged versus 32149e08; the normal wrapper has not itself been boot-qualified by that receipt. No loader paths, compatibles or driver matches were changed by the fixes. The new kernel/commit combination itself is not yet booted; the DTB byte identity is the specific boot-evidence claim. Earlier 659db5b5e removed four duplicate router port/endpoint pairs and added the ANE DART reg-names pair relative to bffee06f; those bytes are now covered by tonight's boot, not awaiting one.
Offline warning audit (full line-by-line classification retained in work/j775c-dt-gate-evidence/warning-classification.json):

• (a) 0 of the 59 can be eliminated solely by ordering, labels or formatting. Separate readable-router override commit changes zero DTB bytes.
• (b) 13 lines fixed by safe DTB changes: eight router port schema failures plus four nonreciprocal graph warnings, and one enabled ANE DART0 missing-reg-names warning. Current PD endpoints already point to each router's ports/port@1/endpoint; drivers/thunderbolt/apple.c obtains remote port 1, not the deleted duplicates. Installed Aura b6e6faf4:src/kboot.c uses router aliases and DART/NHI child paths, not those duplicate nodes. DART0's second existing bank is named auxiliary, deliberately avoiding dapf/gapf names that enable extra accesses in drivers/iommu/apple-dart.c. No compatibles or match-table selection changed.
• (c) 36 lines fixed in bindings: eight audio compatible/machine warnings; fourteen ANE type, firmware-window/full disabled inventory and supplier warnings; eight AVE fallback/five-bank inventory warnings; two GPIO hog warnings; four GPU calibration/SRAM property typing warnings. Preserve existing standalone AVE compatible and earlier-generation audio power requirements. H17 type 0x220 is confirmed by the production ANE platform descriptor; the disabled full seven-window inventory is distinct from the firmware backend's named windows. No new runtime ANE claim. Two pre-existing invalid ANE if/else constructs are corrected for edited-binding validation.
• (d) 10 GPU lines remained at the original CI merge (eight remain after fresh integration), five for each of t6050-j775c.dtb and t6050-j775c-ane.dtb: unmatched three-string compatible, compatible oneOf failure, reg too long, reg-names too long, and reg-names[0] expecting sgx. Installed Aura b6e6faf4:src/kboot_gpu.c requires /soc@2300000000/gpu@2302600000, apple,agx-t6050, and its ASC/SGX handoff shape. The modern apple,agx-g17s binding expects one SGX bank; deleting the fallback or changing the resource contract can lose GPU handoff. Leave these visible rather than weakening the modern binding. Follow-up: separately qualify a loader that supports the modern GPU path/compatible/resources, migrate the DT, and re-boot with renderD128/HDMI health before claiming closure.

MTP DART remains disabled; all three ANE_SYS suppliers and the desktop internal-panel DCP omission remain. Neither MTP nor ANE supplier requirements are deferred to a loader update.

Draft / current offline validation: Original failed run 43142214806 used synthetic merge acbf5b33932a515fcf0fb2fe295d90ad5f45b1dc, first parent f1b79c4beaef1a4a48738f1c820b0bbc16e84bb3. Reproduced its exact script invocation and all-board builds:137/138 DTBs,399→409 warning lines,10 new/0 removed,exit1. All six edited-binding checks and ADT coverage/selftest pass. This is not a merge-base error or coverage failure.

Fresh-base gate (scripts/dtc/apple-dt-check.sh 050f6a517 HEAD WORK): all137/138 DTBs build;402→410 normalized warning lines,8 new/0 removed,exit1 solely for the listed GPU loader contract. All six edited bindings pass, including GPU; ADT selftest, six required manifests and base/head coverage pass. The final offline gate tree is byte-for-byte the published Git tree c3d4ae07293cd835c2c2fef59c781c1f7dff0b04. The full host run used identical driver/tool/script/include/DTS inputs; subsequent changes only update documentation and remove a duplicate schema enum member. Fresh integration's overlapping GPU additions introduced duplicate calibration mapping keys on automatic merge; the binding-only fix keeps upstream definitions. An explicit GPU binding check also exposed an inherited duplicate G17 enum member (command exit0 but schema error in the log); a second binding-only commit removes it. Both defects are fixed without touching DT source.

There is no schema-warning allowlist, baseline file or documented board exception in this repository's DT gate. Existing warnings are tolerated only by base/head set comparison. ADT coverage waivers apply to coverage, not CHECK_DTBS. No warning filter, new waiver mechanism or weakened legacy GPU schema was added. One remaining DT gate item: qualify a loader supporting the modern GPU path/compatible/resource contract, then migrate and boot-test GPU handoff/HDMI. The eight remaining GPU diagnostics remain visible; the compatibility layout is preserved. MTP disablement, ANE suppliers and desktop internal-panel omission remain as required by boot evidence.

Checkpatch: both binding fixes and final documentation update0 errors/0 warnings; documentation0 errors/1 new-file ownership prompt. Full PR diff0 errors/383 warnings/28 checks (inherited generated tuples/prose retained). Whitespace check passes. Documentation at Documentation/devicetree/apple-j775c-compatibility.rst records boot identity and the CI/loader constraint.

Full CI host runner: original merge254/284 in374s; fresh-base production inputs256/286 in348s. Both runs have the same30 environment/fixture failures:8 historical-ref/path controls absent from the retained shallow history, and22 SEP runs failing at ARM64 Rust/native linking (__stack_chk_guard, DSO missing from command line). All other runs pass, including the two tests added by integration. No unrelated test or driver changes were made. Local toolchains: Ubuntu ARM64,Clang18.1.3,GCC13.3,Rust1.96.0,Python3.12.3,dtschema2026.9. DT workflow uses Python3.12; GitHub host CI uses x86-64/Python3.14, so local host results are not a CI-equivalent green claim. Original head's GitHub host-tests PASS. New head fa191334fdaae077a50e09d1eb53a899a2334537 was fast-forward pushed; fresh GitHub checks are pending at this update (not asserted green). No reviews; draft retained because DT CI fails.

Receipts and warning lists: preparation workspace work/j775c-ci-audit-evidence-20261112/; DT archive SHA256 c52de51916129a26a76c38708f109b472b9da473b75a7ea474590c04358500d6, host archive 020ceb537cc38d9f7652a51eedba20d0069be0ae28260770b4173689588ac8c3. No hardware access, subagents or contact outside this PR in this pass.
Not qualified on this stack: HDMI replug/each Type-C display route, audio playback/speaker safety, SD media/physical hotplug/discard, RTC wake and other-board boots (J714/J716 not booted by us). The exact #1579 shared resource definitions did boot inside this compatibility layout; #1579 itself leaves new routers disabled. This is not universal board-layout clearance. Thunderbolt recovery drivers remain separate: R3 one-shot boot passed, soak failed stop/start3 after two passes; not25/25 and freezer0/5. No hardware access was made while preparing these PRs.

🤖 Generated with Codex (GPT-6.1 Sol)

[satchlj]

GPT-6.1 Sol on @satchlj's behalf

Boot-tested current head 32149e08e8b191a500b40435eb0820cc78ec22fe on J775c/T6050 (lobata), combined with PCIe Clock PM #1618 bb464b0c13c22c1dcf82ad9edd92897639ac26b9 and fresh integration 050f6a517fe103c362c63091d5111a9f91fb8691. Local combined merge fe75e6f6f5574140685ce268f675c9594c6b2294; PCIe conflict resolution retains both Neo helpers and Clock-PM host-init wrapper. Native GCC16.1.1/Rust1.93.1 build; one-shot Aura b6e6faf4 trial, installed loader unchanged, command line includes arm64.nomte. Final kernel 7.3.0-lobata-night1619+ #2. Image SHA256 33309f0d7cd27a7004db5576123fb8f74364416d07f51aa1cf9d5a40f403ebbd; DTB dc88aa139f329bb568e48ba7ba80417e7d2c01ec6be49514a95ca24285dfd8d9.

Health: CPUs0–17, ~62GiB RAM, root NVMe mounted rw, wired Ethernet and Wi-Fi connected, USB2/USB3 hubs and USB microphone present, HDMI connected, GPU driver probed with renderD128, GL9755 SD controller present. All four ACIO DT nodes status okay; cabled Mac mini peer Mac16,10 at20Gb/s each direction, bidirectional 3/3 DF1472 pings. SDDM/NetworkManager/sshd active, zero failed systemd units. Audio devices listed only: USB microphone plus two DisplayPort cards. No playback.

Compared with previously booted incremental 9c038a05 source DTB bffee06f…: four duplicate ACIO /port/endpoint remote links removed, along with their eight port/endpoint nodes; disabled /soc/iommu@509800000 adds reg-names = "dart", "auxiliary". All other normalized properties match (excluding phandle renumbering). The changed DTB now boots; no DT source fix needed.

Config migration detail: the initial boot lacked Wi-Fi because new CONFIG_APPLE_CENTAURI_T6050 was off. Enabled it (and default-y _WIFI) locally, rebuilt, and re-booted successfully with the same DTB. MTE is disabled in both build configurations; arm64.nomte also verified live.

This is a boot/health pass with known integration limitations, not warning-free parity with private candidate44: AOP probe returns -61 and analog audio card is absent, as in earlier incremental integration tests; DCP AV endpoints report unsupported firmware/-19 despite HDMI and display audio cards being present. Existing lane-count/provider-clock, DROM, regulatory.db and firmware AT-read diagnostics remain; no panic/oops observed. Current integration also emits zram crypto-zstd modprobe warning. No claim that these are caused or fixed by this DT delta.

Not tested: audio playback/speaker safety, SD media/hotplug/discard, HDMI replug or other Type-C display routes, suspend/RTC, USB4 recovery/throughput soak, GPU rendering workload or other boards. No clean poweroff, promotion or installed Stage1/Stage2 change; default44 restored after testing. Zero Shelly power cycles.

Leaving draft: gh pr checks has host-tests PASS and dt-checks FAIL. The documented ten GPU loader-compatibility schema warnings remain outstanding, so the green-CI requirement for ready-for-review is unmet.

[satchlj]

GPT-6.1 Sol on @satchlj's behalf

Fast-forward refreshed this draft to 0075a007a50701db8cb2ac8aab5f6910377ed414. Merged current integration/aurora 87476b97637ad6a66405a06664bed18685b55bb5 with merge commit 42b8da3861125a4087c7dc4bd983dae26a3c5a1d (no conflicts, no rebase). Reconciled the loader-compatible J775c copies with merged #1676: RSM DART now references DEBUG_GATED and AUSS0 DART references AUSS. Both remain disabled, so this verifies the properties reach the live tree, not active power operation of those two DARTs. Shared modern includes cannot replace these copies without changing loader-sensitive bus paths.

Audited other T6050 SoC changes since the previous integration merge and carried over ProRes reg-names/reset and DART DMA aperture, both JPEG register lengths (16 KiB) and shared DART DMA aperture, CLPC's SMC supplier, and the CPU PMU node. The automatic merge also combined identical GPIO-hog binding patterns; removed one duplicate without changing DTB bytes. GPU paths, compatibles and ASC/SGX resources remain unchanged.

Native CHECK_DTBS base/head: all 137/138 DTBs build, 402→410 normalized warning lines, exactly 8 new / 0 removed. These are the same four loader-dependent GPU diagnostics for each J775c wrapper (compatible oneOf, reg length, reg-names length and first name). No warning suppression or binding weakening. All five edited-binding checks and ADT selftest/comparison pass. Direct dt-validate exits 0 but prints the retained GPU diagnostics (18 lines per wrapper), so it is not diagnostic-free. Checkpatch: reconciliation 0 errors/1 unwrapped commit-message warning; duplicate-binding fix 0/0; integration-merge net diff 0 errors/23 inherited warnings. Whitespace passes. Draft retained.

Boot PASS on lobata/J775c. Native GCC 16.1.1 / Rust 1.93.1, candidate44-derived config with Centauri enabled; no driver shim. Exact inputs: integration 87476b97637ad6a66405a06664bed18685b55bb5, #1618 bb464b0c13c22c1dcf82ad9edd92897639ac26b9, updated #1619 0075a007a50701db8cb2ac8aab5f6910377ed414; local combined boot head 12992dd42a849501c5eff93d97e1f54793e6c9e9. Test-only PCIe conflict resolution retains upstream Neo helpers and #1618's Clock PM host-init wrapper. Image.gz SHA256 751f1181c567fb250827fd673a244ae4d34881d39c9abff91414b79674e9930d; ANE-wrapper DTB SHA256 0a4ec6fdb639e9eedb4e0736419ba2a847879b3fe2c1a10af54cdfaa74befa28, byte-identical to the checked PR DTB.

One-shot via lobata-try-kernel, installed loader unchanged, live arm64.nomte, kernel 7.3.0-lobata-refresh1619+ #2. SSH returned; all 23 previous DART bindings remain; no SError, DART/IOMMU fault, kernel WARN/BUG/Oops signatures in final log. GPU binds and primary/secondary initdata are acknowledged. HDMI-A-1 connected on apple-drm, appledrmfb; explicit modetest 1920x1080@60 returns 0 with active/enabled CRTC readback. USB hubs/microphone, root NVMe rw, wired/Wi-Fi connected, cabled Mac16,10 Thunderbolt peer with two 20-Gb/s lanes in each direction and carrier up; zero failed units. Existing AOP -61, DCP AV -19 and regulatory/module diagnostics remain.

Compared with the previous compatibility boot's live FDT: only the ten intended source-level node/property changes above, phandle renumbering, random KASLR seed, initrd/CPU release-address relocation, and relocated same-size ADT/OS-log reserved regions. No unexplained resource change. Full property comparison is retained in work/upstream-1619-refresh-20261112/live-delta.json alongside the report work/UPSTREAM-1619-REFRESH-20261112.md.

Not tested: rendering/compute workloads, active RSM/AUSS0 power behavior, ProRes/JPEG/PMU workloads, audio playback, HDMI replug/Type-C display routes, SD media/hotplug/discard, suspend/RTC, Thunderbolt recovery/throughput/soak, or other boards. No newer loader, default promotion, Stage1/installed Stage2 change, clean poweroff or recovery power cycle. Default44 #25 restored healthy, protected hashes match.