Skip to content

kernel: Adopt the iPad 7 boot framebuffer in the A9/A10 display driver - #166

Merged
henrikklee merged 2 commits into
OrchardLinux:mainfrom
jerricky26:j171-adbe-panel-size
Nov 18, 2026
Merged

henrikklee merged 2 commits into
OrchardLinux:mainfrom
jerricky26:j171-adbe-panel-size

Conversation

@jerricky26

Copy link
Copy Markdown
Contributor

Follow-up to #158. The display pipe driver only accepted the 1536x2048 boot framebuffer of the 9.7-inch iPad 5 and 6. On the iPad 7 (J171) the 10.2-inch panel runs at 1620x2160, with iBoot padding each row to 6528 bytes, so probe failed with no 1536x2048 boot framebuffer (-EINVAL) and the display stayed on simpledrm. #158's verified DART scanout never ran there.

The driver never programs timings, so it now takes the size and stride from the simple-framebuffer iBoot describes and accepts the two known panels, each with its own fixed mode and physical size. The adopted-plane check, park, plane limits and display DART verification use that size and stride. J71b/J72b behaviour is unchanged. J172 shares the 1620x2160 panel.

Tested on J171: adbe probes and replaces simpledrm, the display DART self-check passes and the panel scans out through the DART (scanout through the display DART, boot framebuffer at 0xbb5e0000), the connector reports 1620x2160, and the desktop and touch work normally. Not tested on J172, J71b or J72b.

jerricky26 and others added 2 commits November 17, 2026 19:11
The display pipe driver only accepted the 1536x2048 boot framebuffer of
the 9.7-inch iPad 5 and 6, so on the iPad 7 (1620x2160, rows padded to
6528 bytes) it failed to probe and the display stayed on simpledrm.
Take the size and stride from the boot simple-framebuffer and accept
both known panels.

Assisted-by: LLM
Build the display pipe's fixed mode from the boot simple-framebuffer's
width and height instead of matching a table of two panels, so every
A9/A10 board's panel is adopted, the iPad 7 being the first beyond the
iPad 5/6. Only the physical size of the known panels stays in a table;
an unknown panel reports its size as unknown instead of failing probe.

Assisted-by: LLM
@henrikklee
henrikklee merged commit 5b12117 into OrchardLinux:main Nov 18, 2026
12 checks passed
@henrikklee

Copy link
Copy Markdown
Contributor

Thanks @jerricky26, merged. I generalized it on top: the driver now takes its mode from whatever boot framebuffer iBoot describes, so every A9/A10 board's panel is adopted, not just the two known sizes; only the panel's physical size still comes from a small table.

Could you check on your iPad 7, with the next nightly (or a build of main), that the desktop is GPU-accelerated end to end? For example cat /sys/kernel/debug/dri/*/name or the compositor log showing the PowerVR renderer rather than llvmpipe/pixman, and glmark2-es2-wayland (or any GL app) reporting the PowerVR device. A short note with what you saw would be great.

henrikklee added a commit to brkosebay/hardware that referenced this pull request Nov 18, 2026
… switches

Rework of OrchardLinux#170 on current main:

- Renumber after main: drivers/drm 0010 (binding) and 0011 (driver), whose
  0009 collided with OrchardLinux#166's; devices/s800x 0003.
- drm/adbe: the transmitter's AUX channel is a struct drm_dp_aux with a
  transfer hook; DPCD access, retries, the capability and link status
  reads, the lane checks and DP_SET_POWER come from the shared DP helpers.
  The no-handshake TPS1/TPS2 sequence stays in the driver: it drives the
  transmitter's own training pattern register and the helpers have none.
- Panel-off sleep is unconditional wherever the device tree gives the
  panel supply: the sleep_panel_off parameter, the adbe_panel debugfs
  file, the post-power-off witness read and the info-level logging are
  gone (gate reasons are dev_dbg).
- The binding no longer forbids the panel supply on the A10 and allows
  the "dp" window there too; the T8010 device trees are unchanged until
  their supply line is known.
- The driver selects DRM_DISPLAY_DP_HELPER (configs updated).

Assisted-by: LLM
@jerricky26

Copy link
Copy Markdown
Contributor Author

Tested on iPad 7 (J171), but on my local build (7.3.0-rc1-nhpce, our version of this change plus the BT/charger patches), not yet on main with your generalization; I'll retest on the next nightly.

  • Scanout goes through apple-adbe (card1); labwc runs WLR_RENDERER=gles2 on renderD128 (powervr), with libvulkan_powervr_mesa.so loaded via zink; no lavapipe installed.
  • glmark2-es2-wayland (Mesa 26.2.7): GL_RENDERER: zink Vulkan 1.2(PowerVR Apple A10 (IMAGINATION_OPEN_SOURCE_MESA)), GLES 3.1. 32 of 33 default scenes ran, typically 5–10× llvmpipe on the same device (e.g. build 1269 vs 214 fps).
  • [shadow] hangs a fragment job reproducibly: Apple GPU job timed out / hang: fragment job …, GPU reset, zink VK_ERROR_DEVICE_LOST. The GPU recovers and the following scenes run. Happy to open a separate issue for that.
  • Note: GL apps launched outside the session need PVR_I_WANT_A_BROKEN_VULKAN_DRIVER=1 or Mesa silently falls back to llvmpipe.

@jerricky26
jerricky26 deleted the j171-adbe-panel-size branch November 19, 2026 02:34
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