Skip to content

arm64: dts: apple: t6020-j414s: enable PMP - #933

Merged
RyanTheTide merged 1 commit into
aurora-silicon:integration/aurorafrom
JJRPF:t6020-j414s-enable-pmp
Nov 16, 2026
Merged

RyanTheTide merged 1 commit into
aurora-silicon:integration/aurorafrom
JJRPF:t6020-j414s-enable-pmp

Conversation

@JJRPF

@JJRPF JJRPF commented Nov 4, 2026 •

Copy link
Copy Markdown
Contributor

Ready for review. The m1n1 side (aurora-silicon/m1n1#63) is merged; this DTB must ship together with an m1n1 that includes it. Deep S2R (SYSTEM_SUSPEND) on the J414s is not working yet, see "What is not covered".

Lets m1n1 boot the Power Management Processor on the 14-inch MacBook Pro (M2 Pro), as t6000-j314s does: #define APPLE_USE_PMP in t6020-j414s.dts (+6 lines with the comment). With the PMP up, the fabric and DCS p-states are the PMP's (pmgr-misc logs "PMP manages fabric and DCS p-states") and the DCP, AVD and ISP power domains follow the PMP's report entries (pmp-disp0, pmp-dispext0/1, pmp-isp-sys).

Result (J414s, headless, nothing attached, same kernel with and without the PMP, one boot each)

  • 16 windows of 30 s: system power 3.485 W -> 2.566 W (-0.919 W), SoC (heatpipe) 1.990 W -> 1.341 W (-0.649 W).
  • With the energy model (arm64: dts: apple: t602x: add the CPU energy model (enables EAS on M2 Pro/Max) #932) on top: 2.378 W / 1.16 W. That image held a 1 h soak (120 windows, 2.36 W / 1.13 W, 0 of 3695 pings lost, no new kernel messages).
  • Wi-Fi, Bluetooth, audio, backlight, battery, NVMe, USB and Hyprland unaffected; the only new kernel lines are PM domain pmp-isp-sys will not be powered off and deferred-probe messages that resolve.

Status against the review

  • m1n1 first: kboot: give the PMP tunables room, and leave the PMP off if they do not fit m1n1#63 merged on 2026-11-13. The PMP's tunables (81 properties, about 16K) do not fit the 96K device tree headroom of an older m1n1 on a T6020 and the boot stops in proxy mode ("failed to transfer pmp tunable"), so this DTB and an m1n1 with soc: apple: sep: keep M1 Pro enrolments across a reboot #63 have to ship together.
  • s2idle: 10 of 10 cycles of 20 s on the J414s, all returning 0, with the eDP panel and backlight, Wi-Fi and 3 USB devices back after each one and no new warnings (details in the comment below).
  • Boot of this branch's DTB: done (8f590c4, 7.3.0): the pmp node is okay with 81 tunables, "PMP manages fabric and DCS p-states", PMP firmware ready, pmp-disp0/dispext0/dispext1/isp-sys on.
  • One boot per image and headless only for the power numbers; no battery or screen-on measurement.

What is not covered

This change was developed with Claude Code (Claude Sonnet 5.5, Anthropic) at the owner's direction.

🤖 Generated with Claude Code

https://claude.ai/code/session_01RuRMQncGu3ukpyW44vHHjQ

@RyanTheTide RyanTheTide left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The change itself is fine: one APPLE_USE_PMP define, it is consumed by t602x-common.dtsi/t602x-die0.dtsi/t602x-dieX.dtsi, only t6020-j414s.dtb changes, and dt-checks and host-tests pass.

Before this leaves draft:

  1. aurora-silicon/m1n1#63 has to merge first, and this DTB can only ship together with an m1n1 that includes it. Without that, the boot stops in proxy mode ("failed to transfer pmp tunable").
  2. A J414s s2idle result with the PMP enabled: several suspend/resume cycles, and display, Wi-Fi and USB back after each one.
  3. A boot of the DTB built from this branch (the numbers above come from the 7.2.11 tree), with the same PMP lines in dmesg and no new warnings.

@JJRPF

JJRPF commented Nov 13, 2026

Copy link
Copy Markdown
Contributor Author

J414s results for this branch (8f590c4, 7.3.0), booted by m1n1 from aurora-silicon/m1n1#63 (2c324f2). With #63 merged, this covers the rest of your review: items 2 and 3 below.

Boot of the DTB built from this branch: the pmp node is okay with 81 tunables, and dmesg shows

apple-pmgr-misc 28e20c000.power-management: PMP manages fabric and DCS p-states
apple-dart 3860e8000.iommu: PM domain pmp-isp-sys will not be powered off
apple_pmp 28e700000.pmp: RTKit: Initializing (protocol version 12)
apple_pmp 28e700000.pmp: PMP firmware and endpoint ready
apple_pmp 28e700000.pmp: RTKit: syslog message: io.cpp:41: PMP started

pmp-disp0, pmp-dispext0, pmp-dispext1 and pmp-isp-sys are on, pmp-avd-sys is off. There are no PMP, RTKit, DCP or fabric warnings. The only warnings are the watchdog one (my test root keeps it armed) and the ANE firmware preflight, which 7.2.11 shows too.

s2idle: 10 of 10 cycles of 20 s, all returning 0. After each one the eDP panel came back with its backlight, Wi-Fi answered on the second ping, 3 USB devices were back, and there were no new warning or error lines. The pmp-* domains matched boot afterwards.

For the wake source I added the T602x RTC alarm cells to the DTB locally. They aren't part of this PR; the boot results above are from the DTB exactly as built from this branch.

@JJRPF
JJRPF marked this pull request as ready for review November 14, 2026 01:34
Let m1n1 boot the Power Management Processor on the 14-inch MacBook Pro
(M2 Pro), as t6000-j314s does. With the PMP up, the fabric and DCS p-states
are the PMP's (pmgr-misc reports "PMP manages fabric and DCS p-states") and
the DCP, AVD and ISP power domains follow the PMP's report entries (pmp-disp0,
pmp-dispext0/1, pmp-isp-sys).

Measured on a J414s, headless, nothing attached, the same kernel with and
without the PMP, one boot each, 16 windows of 30 s: system power 3.485 W ->
2.566 W (-0.919 W), SoC (heatpipe) 1.990 W -> 1.341 W (-0.649 W). Held for a
1 h soak with the energy model on top: 2.36 W, 0 of 3695 pings lost, no new
kernel messages.

This needs an m1n1 that makes room for the PMP's tunables (81 properties,
about 16K, which the 96K device tree headroom does not hold on a T6020);
with an older m1n1 the boot stops in m1n1 proxy mode ("failed to transfer pmp tunable", no payload). Merge the
m1n1 change first. Suspend with the PMP on the J414s is not qualified here
(it is the S2R work in progress), so only enable this once that is.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RuRMQncGu3ukpyW44vHHjQ
Signed-off-by: Justin Pfister <119864532+JJRPF@users.noreply.github.com>
@JJRPF
JJRPF force-pushed the t6020-j414s-enable-pmp branch from 8f590c4 to 8d84386 Compare November 14, 2026 01:37
@JJRPF

JJRPF commented Nov 14, 2026 •

Copy link
Copy Markdown
Contributor Author

Follow-up for the "one boot per image, headless only" gap, and a rebase.

Rebase. The branch was 3187 commits behind integration/aurora. The one commit is now on top of 5d112b1d (8d843865); the diff is still the six lines in t6020-j414s.dts, the DTB builds with the pmp node, and dt-checks and host-tests pass again on the new head.

Screen on. J414s, display on at 40% backlight, nothing attached, same kernel (7.3.0 built from this branch) and same stage 2 (m1n1 at edcd9a66, #63's head) for both boots. One boot with this branch's DTB (PMP, 81 tunables) and one with the same tree without APPLE_USE_PMP. Each: 120 s settle, then 16 windows of 30 s, SMC Total System Power and Heatpipe Power once a second, CPUs idle.

Total System Heatpipe (SoC)
without PMP 5.100 W 2.204 W
with PMP 4.349 W 1.113 W
difference -0.75 W -1.09 W

Window by window the difference stays between -0.70 and -0.83 W (total) and -1.00 and -1.28 W (heatpipe). The total rises by about 0.6 W over each run in both boots (battery and panel warming, probably) while the heatpipe stays flat, so I paired the windows rather than comparing the means.

Caveats: one boot per DTB, run in the order PMP then control, so a slow drift between the two boots would not show up here; the heatpipe difference here (-1.09 W) is larger than the headless number in the description (-0.65 W); those are different measurements (display on versus headless) and should not be compared one to one. The boot warnings were only PM domain pmp-isp-sys will not be powered off with the PMP and the test root's watchdog line without it.

What is still not covered is the same as in the description: deep S2R (SYSTEM_SUSPEND), which does not work on the J414s yet with or without the PMP.

@JJRPF
JJRPF requested a review from RyanTheTide November 15, 2026 16:56
@RyanTheTide
RyanTheTide merged commit 6e2043a into aurora-silicon:integration/aurora Nov 16, 2026
2 checks passed
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