[erikwb]

The J713's (MacBook Air 13" M4, T8132) display pipeline stays powered whenever the panel is on: through system sleep, while it is blanked, and while a static frame is shown. This series powers it down with the DCP's own handshake. It replaces the display half of #449, plus #501 and #503, as one drm/apple PR; the commits are unchanged.

• System sleep (commits 1–3) is what the J713's suspend-to-idle needs from the DCP; the rest of suspend-to-idle is J713 (M4 MacBook Air): suspend-to-idle #831.
• Blank and the idle cache (commits 4–5) work at runtime, with or without system sleep.

The handshake (commits 1–3)

On IOMFB 26.x the handshake is setPowerState (A473) on the out-of-band channel. The T8132 words carry that channel as stream 1 (bit 10).

• Off: with the panel blanked, setPowerState(0) makes the firmware log power on -> off, disable its M3 helpers and switch to run mode 0.
• On: setPowerState(1) brings it back (run mode 0 → 4, M3 firmware reloaded), and the next present scans out.
• Callbacks: the firmware calls back about ten times during each transition, and each one gets an answer of the exact shape it expects.

The three commits are:

1. stream-1 calls in the transport (calls nested in a stream-1 callback stay on stream 1);
2. the callback shapes and answers;
3. the suspend/resume hooks.

apple-drm is a device-link consumer of the DCP, so it has blanked before the DCP suspends and restores the CRTC after it resumes. The DCP keeps its domains on through the wakeup path, as the older DCPs do.

Power the display down while it is blanked (commit 4)

DPMS off (screen lock, idle blank, hyprctl dispatch dpms) used to only blank the display: iomfb_m4_blank() swaps to no surface, but the pipeline stays powered until the next frame.

• Blank: after a successful blank, setPowerState(0), the same call system sleep makes.
• Next frame: the first present after DPMS on sends setPowerState(1) before its swap, so the compositor's first frame brings the panel and backlight back. That's the path resume already uses.
• Suspend while powered down: suspend skips its own setPowerState(0), and the display stays down after resume until the next present.
• Refusal: a refused or failed transition fails the session, as in the sleep path. Before, that could only happen at system sleep; now it can happen at every blank and wake. The stress run below is the evidence that it holds up.
• debugfs: iomfb_m4 gains display_off, off_cycles, and off_us / on_us (last and longest).

Cache a static frame (commit 5)

With a static image, the panel keeps being scanned out of memory 60 times a second. This registers a DCP idle-cache surface once no flip has happened for appledrm.idle_cache_ms, and leaves the cache on the next present, brightness change or blank.

• The default is 1000 ms; 0 disables it, and the maximum is 60000.

Testing (J713)

On integration/aurora 41c4055 with the rest of the J713 PR stack (this series, the suspend-to-idle and USB series, #429, #432, #819/#820, #450, #531, #825):

• DPMS off: the display powers down in 129 ms; DPMS on powers it up in 55 ms; error=0.
• Idle cache: it keeps entering on a static screen.
• Real suspend-to-idle (82 s): the display comes back, with frames flowing and the cache re-entering.

Earlier, on our J713 tree:

The device tests ran on our J713 tree, which has the J713 series this branch alone lacks (the display doesn't come up on integration/aurora alone yet). The commits are identical to the ones tested. Today, on integration/aurora 41c4055 with the rest of the J713 stack, it went through several real suspend-to-idle cycles of up to 352 s, each resuming to a working desktop.

System sleep:

• Real s2idle from systemctl suspend (up to about 16 min asleep) and from the lid. Display and GPU come back (glmark2 630–642, unchanged), and so do Wi-Fi and audio.
• pm_test=devices and pm_test=platform pass.

DPMS off, power: idle on battery, 2 s samples over 120 s per phase, medians. Each column is one run on one boot, so compare within a column.

Phase	Without commit 4	With it
Display on, backlight off	3038 mW	2926 mW
DPMS off	2692 mW	1728 mW
Display on, backlight off	3017 mW	2914 mW
Saved by DPMS off	346 mW	1198 mW

DPMS stress: 200 blank/wake cycles through Omarchy's DPMS commands, checking debugfs after every step. All 200 passed, with no errors.

	Last	Longest
setPowerState(0)	128 ms	132 ms
setPowerState(1)	55 ms	56 ms

The power-down runs inside the commit that disables the CRTC, so that commit takes about 130 ms longer. Hyprland's DPMS dispatch still returns in 10–12 ms.

DPMS, other cases:

• Brightness changed while blanked: takes effect with the first frame after wake.
• Blank, then system sleep: suspend skips its own power-down, resumes with the display still down, and one key press brings it back.
• Lid suspend: the display comes back on resume; Touch ID works before and after.

Idle cache: J713, battery, backlight 0, 180 s per phase, same boot.

• Power: idle_cache_ms=0 draws 3266 mW, idle_cache_ms=1000 draws 2931 mW: −335 mW.
• Entry and exit: it cached about 93 % of a 10 min static run. The frame stays on screen, and the first swap after leaving the cache takes 5–10 ms (ordinary swaps take 12–14 ms).
• Re-entry: brightness changes, DPMS off/on and suspend-to-idle all exit the cache and re-enter it.
• fbcon: the text console, which writes the scanout buffer without flips, updates live while cached.
• Stress: a full-screen random-noise image caches without DART faults.
• Errors: fence refusals roll back cleanly and are retried, and out-of-range values are rejected.

Host tests:

• apple-dcp matches integration/aurora, with the same failures on the bare base. New cases cover stream-1 decoding and routing, and a present refused while powered off.
• m4-dcp-errors passes, including its case where a failed display domain refuses maps.
• m4-dpms-retirement, the host test of the production blank and present paths, gets stubs for the new calls in commits 4 and 5. It now also asserts that a blank powers the display down once, the next frame powers it back up, and every shown frame restarts the idle-cache countdown. It passes at both commits.

Checks (on integration/aurora 41c4055):

• all commits are signed off;
• checkpatch --strict is clean;
• full arm64 build (Image, modules, DTBs) with the J713 config, with no warnings in the touched files;
• CHECK_DTBS W=1 on all six T8132 DTBs gives the same output as the base;
• run-host-tests: 132/133. The one failing group, apple-dcp, fails the same nine cases on the bare base and nothing more.

Known limits

• A CRTC-only commit after DPMS on (no new framebuffer) leaves the display down until the next frame. Before commit 4 the same case left it blanked, so nothing that used to show a picture now doesn't.
• Unbind, shutdown or kexec while the display is down leave it down. A reboot through iBoot resets it.
• The firmware logs IOAVVideoInterface open failed after each power-up, with no effect.
• Other T8132 machines take the same path, but only the J713 device tree enables this DCP today.

This series was developed and tested by Claude (Anthropic), working with me on the J713.