Skip to content

arm64: dts: apple: t8122, t603x: complete the M3 ProRes node descriptions - #2428

Open
RyanTheTide wants to merge 2 commits into
integration/aurorafrom
m3-media/prores
Open

RyanTheTide wants to merge 2 commits into
integration/aurorafrom
m3-media/prores

Conversation

@RyanTheTide

Copy link
Copy Markdown
Contributor

Problem

Every M3 SoC (T8122, T6030, T6031/T6034, T6032) already has disabled ProRes and DART nodes from the M3 inventory work, and the binding already lists their compatibles. The nodes were never finished: none has reg-names, so the driver, which maps its windows by name, could not map them even with match data; none has resets; and the DARTs on T8122, T6030 and T6032 have no power domain. The T6032 branch of the binding also rejects reg-names and resets.

Change

  • Binding: the T6032 branch accepts the same core/wrapper/aux reg-names and one reset as the T6030/T6031/T8122 branch.
  • DT: each M3 ProRes node gets reg-names and a reset through its own power domain (SCODEC for apr/apr0/apr2, PRORES for apr1/apr3). The T8122, T6030 and T6032 apr DARTs get the engine's domain, as T6031 already has. Everything stays disabled.

No driver match data. The M3 engines are their own ADT revisions (apr,f0 on T8122, apr,h0 on T6030, apr,h1 on T6031/T6034, apr,h2 on the second T6032 die). None of them is a revision the driver already supports (a0, j0/j1, T6050, T8140), and the window layout is different too: the M3 aux window sits at core+0x78000, not core+0xc0000 as on T6041 and T8132, and the T8122/T6030 wrapper window is 32 KiB, not 16 MiB. So reusing T6041's data isn't justified. What the engine needs before its first job, its power-up initialization rows, is not in the ADT. We have no capture of it for any M3 yet, so this PR doesn't guess it. See "Still to run".

Testing

  • All 14 M3 DTBs (t8122-j433/j434/j504/j613/j613-25g83/j615/j615-25g83, t6030-j514s/j516s, t6031-j514c/j516c, t6034-j514m/j516m, t6032-j575d) build with W=1. CHECK_DTBS gives no findings for the ProRes or DART nodes. The 205 findings it does report are all in other nodes and were already there.
  • dt_binding_check for media/apple,prores.yaml is clean.
  • The driver isn't touched by this PR, so its module build and KUnit suite are not affected. The W=1 module rebuild didn't run here: modules_prepare on the buildbox config fails in rust/bindings.o, which has nothing to do with this change. It's listed under "Still to run".
  • checkpatch --strict: clean, apart from one long commit <sha> ("title") reference line.
  • Values: reg, interrupts, DART SIDs, power domains and the reset target were checked against the ADT of each board and our own IODeviceTree capture of a J615 under macOS 27.2. The reset target is the ADT device_reset PMGR device, and it is the same device as the power domain.

Still to run

Per-SoC ADT facts used
SoC ADT node compatible windows IRQs DART SID domain / reset
T8122 apr apr,f0 core 16K, wrapper 32K, aux 16K 2 1 (two DART instances) SCODEC
T6030 apr apr,h0 core 16K, wrapper 32K, aux 16K 2 1 (two DART instances) SCODEC
T6031/T6034 apr0 apr,h1 core 16K, wrapper 16M, aux 16K 1 1 (two DART instances) SCODEC
T6031/T6034 apr1 apr,h1 core 16K, wrapper 16M 1 0 PRORES
T6032 die0 apr0/apr1 apr,h1 as T6031 1 each as T6031 SCODEC / PRORES
T6032 die1 apr2/apr3 apr,h2 as T6031 1 each as T6031 SCODEC / PRORES (die 1)

All boards of each SoC have the same ProRes description, so the nodes stay in the SoC dtsi.

Related

Assisted-by: LLM (Claude:Opus 5.5)

The T6032 branch of the schema lists reg, interrupts, power-domains and
iommus only, so a T6032 ProRes node that names its windows or points at
its reset fails validation under unevaluatedProperties. Every other
variant already allows both.

Allow the same core/wrapper/aux reg-names and a single reset as the
T6030/T6031/T8122 branch. The second engine on each die (apr1/apr3) has
only the core and wrapper windows, so reg-names keeps minItems 2.

Signed-off-by: Ryan Murray <ryan@aurorasilicon.org>
The disabled ProRes nodes and their DARTs on T8122, T6030, T6031 and
T6032 came with the M3 resource inventory, both by Ryan Murray:
commit 982f330 ("arm64: dts: apple: describe disabled M3 peripheral resources")
commit 4c28191 ("arm64: dts: apple: describe disabled T6032 package resources")

Checked against the ADT of every M3 board (J613, J615, J433, J434,
J504, J514s/c/m, J516s/c/m, J575d) and our own IODeviceTree capture of
a J615 under macOS 27.2, their reg, interrupts, DART SIDs and power
domains are right, but three things are missing:

- reg-names. The driver maps its windows by name (core, wrapper, aux),
  so without them no variant could ever map these nodes. apr0 on each
  die has all three windows; apr1 on T6031/T6032 has core and wrapper
  only, as in the ADT.
- resets. The ADT's device_reset function for each engine points at the
  same PMGR device as its power domain (SCODEC for apr/apr0/apr2,
  PRORES for apr1/apr3), so the engine is reset through that domain.
- power-domains on the T8122, T6030 and T6032 DARTs. The DART sits in
  the engine's domain; T6031 already says so, the others did not.

All nodes stay disabled: the driver has no match data for these engines
yet (see the PR for why).

Signed-off-by: Ryan Murray <ryan@aurorasilicon.org>
@RyanTheTide

Copy link
Copy Markdown
Contributor Author

No blockers.

I checked every hunk at 3b5e96d against the tree. The reset phandles are valid: ps_scodec and ps_prores (and their _die1 copies from DIE_NODE() in t6031-pmgr.dtsi:1617/1635) have #reset-cells = <0>, and so do t8122-pmgr.dtsi:768 and t6030-pmgr.dtsi:845. The new DART domains follow the IRQ grouping that T6031 already uses (t6031-peripherals.dtsi:40/50/60): IRQ 1737 maps to SCODEC and IRQ 1782 maps to PRORES, and the T6032 die0 and die1 DARTs now match. Each reg-names entry matches its node's reg count: three on apr/apr0/apr2, two on apr1/apr3. The T6032 binding branch now has the same reg-names/resets shape as the generic branch (apple,prores.yaml:126-141).

Runtime effect is none. Every node is still status = "disabled", and apr_of_match (apple-prores-drv.c:1910) only matches t6000, t6041, t6050 and t8140, with no generic apple,prores entry. Even an enabled node can't bind with guessed data. dt-checks and host-tests are green, and the PR merges cleanly. Not adding M3 match data until there is a traced init sequence is the right call.

This can leave draft. The hardware row belongs to the later M3 match-data PR, not this one.

HW row: none needed for this DT-only change (nodes stay disabled). When the M3 match data lands, run on J615 at that head: probe, engine ID readout, Proxy/LT/422/HQ at 1080p and 4K with every frame decoded by ffmpeg, 0 DART faults, s2idle with the module loaded, 0 AVC, 0 failed units, no new kernel errors.

Assisted-by: LLM (Claude:Opus 5.5)

@RyanTheTide
RyanTheTide marked this pull request as ready for review November 22, 2026 10:11
@RyanTheTide

Copy link
Copy Markdown
Contributor Author

Adversarial pass at 3b5e96d: no blockers.

  • Built every Apple DTS with cpp + dtc at integration/aurora and with this PR merged in (clean merge). Exactly the 14 M3 trees change (seven T8122, two each of T6030/T6031/T6034, and the T6032 J575d), no other board changes, and no dtc warning counts move.
  • Every new resets/power-domains phandle resolves, including the _die1 copies on T6032. Each node's reg-names count matches its reg (three windows on apr/apr0/apr2, two on apr1/apr3), and the T6032 schema branch now takes that shape.
  • Every ProRes and ProRes DART node is still disabled, and the driver has no generic apple,prores match, so there is no runtime change on any M3.

No hardware row needed for this PR; the J615 row comes with the M3 match data.

Assisted-by: LLM (Claude:Opus 5.5)

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.

1 participant