Skip to content

drm/apple: 27.2 variant: record the method-table confirmation from the 27.2.1 image - #2179

Draft
joshuaswarren wants to merge 1 commit into
aurora-silicon:integration/aurorafrom
joshuaswarren:w86/dcp-27-2-confirmation
Draft

joshuaswarren wants to merge 1 commit into
aurora-silicon:integration/aurorafrom
joshuaswarren:w86/dcp-27-2-confirmation

Conversation

@joshuaswarren

@joshuaswarren joshuaswarren commented Nov 17, 2026 •

Copy link
Copy Markdown
Contributor

Tested only on macOS 27.3 beta (26C5035r) on a 14-inch M5 Max (K114c). The 27.2 version is untested: the lab has no 27.2 machine, and the values marked missing below cannot be confirmed without one.

A static read of Apple's public 27.2.1 restore image for this board (T6050) confirms the method layout the 27.2 table already had: same class layout, swap shapes unchanged (0xff4 request, 0x6e0 response). No ID changed, so no table value moved. On retarget to 27.2 the 27.3 pbl fence-notify row is dropped (no 27.2 method) and the MemDescRelay thunk row stays on the offset basis (no symbol in the 27.2 image).

Truly missing without a 27.2 machine: the 27.2 tag values (registered by the firmware, not the kernelcache) - the tag values are DERIVED, not booted: tag = base + index was verified across the full 27.3 set (32 methods; a single uniform +1 insertion shift, 7 methods, no contradictions), and the 27.2 values below/above the insertion point follow that derivation; a 27.2 boot trace remains the confirmation step; and any boot evidence: no 27.2 machine exists in the lab, so this PR is a draft.

Proved by: the driver objects build for arm64 at this head. This is the 27.2 sibling of #1816 (27.3), from the same code.

…image

A static read of Apples public 27.2.1 restore image for this board
(T6050) confirms the method layout the 27.2 table already had: same
class layout, swap shapes unchanged (0xff4 request, 0x6e0 response).
No ID changed, so no table value moved. The read also explains 27.3s
one-higher IDs from set_display_refresh_properties up: the pbl
fence-notify and swap-complete methods and the trace-source swap-wait
parameter exist only in 27.3. On retarget to 27.2 the pbl fence-notify
row is dropped (no 27.2 method) and the MemDescRelay thunk row stays
on the offset basis (no symbol in the 27.2 image). Tag values remain
firmware-side knowledge pending a 27.2 boot.

Proved by: the driver objects build for arm64 at this head.

Signed-off-by: Joshua Warren <816217+joshuaswarren@users.noreply.github.com>
@RyanTheTide

Copy link
Copy Markdown
Contributor

Thanks. This head is DIRTY against integration/aurora and has no CI run. It also only adds a comment to iomfb_v27.c, and no table value changes.

Two things before it can go anywhere:

  • Please rebase it, or better, fold the note into the commit that carries the 27.2 table (or into drm/apple: complete the 27.3 DCP method table, swap size and completion tag #1816), so the explanation lands with the code it describes.
  • Please limit the comment to what the driver observes and what is still unconfirmed, such as "tag values not yet seen on a 27.2 boot". Leave out the description of what is inside Apple's restore image. We keep in-tree comments to observed behaviour, and the provenance can go in the PR text.

The real missing piece is the one the body names: a 27.2 boot that confirms the tag values. Until someone with a 27.2 machine can run it, this stays a draft.

@joshuaswarren

Copy link
Copy Markdown
Contributor Author

Note

🤖 OpenAI Codex responding on behalf of joshuaswarren

Please rebase it, or better, fold the note into the commit that carries the 27.2 table (or into #1816).

I'll carry the note in #1816 rather than land a second copy here. The candidate for #1816 already restores the note. In the final rebase after #2102, I'll keep the driver's observed 27.2 behavior in the comment and put the restore-image provenance in the PR text.

The J775c test on 27.2 does not exercise the 27.3 code. A clean display run also cannot confirm every firmware tag. I'm leaving this draft while the changes are folded together.

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.

2 participants