Skip to content

arm64: dts: apple: t8132-j773g: wire pmc_dispext0/1 interfaces (4K@120Hz bandwidth) - #2761

Open
DjDeveloperr wants to merge 1 commit into
djdev/j773g-272-dcpext0from
djdev/j773g-pmc-dispext-bw
Open

DjDeveloperr wants to merge 1 commit into
djdev/j773g-272-dcpext0from
djdev/j773g-pmc-dispext-bw

Conversation

@DjDeveloperr

Copy link
Copy Markdown
Contributor

Summary

Stacked on #2541 (djdev/j773g-272-dcpext0).

On T8132, the PMC ignores an external display pipe's bandwidth request (BWR) and display-clock boost votes while its interface enable bits in 0x3828ac020 (bwr_enable) and 0x3828ac028 (boost_enable) remain clear (drivers/pmdomain/apple/pmc.c). t8132-j713-j715.dtsi wires pmc_dispext0 and pmc_dispext1 into ps_dispext0_fe and ps_dispext1_fe, whereas t8132-j773g.dts left them unconnected — leaving dispext0 BWR/boost off at 4K@120Hz (perf_state=5) and halving available DRAM bandwidth across both GPU and CPU.

  • Wire <&pmc_dispext0> and <&pmc_dispext1> as the second power-domain parent of &ps_dispext0_fe and &ps_dispext1_fe, keeping &ps_dispext0_sys at index 0 (matching t8132-j713-j715.dtsi).
  • Point &framebuffer0 at <&ps_dispext0_sys> instead of <&ps_dispext0_fe> so simpledrm still binds early when CONFIG_APPLE_PMP=m (apple-pmc defers until apple-pmp binds); ps_dispext0_fe is apple,always-on and never powered off by genpd.

Hardware Verification (M4 Mac mini J773g, macOS 27.2 / 26B123)

  • Boot-default verification (7.3.0-ARCH-m4prod1, dcp61-m4prod1-full, zero /dev/mem writes):
    • /sys/kernel/debug/apple-pmc/382800000.power-management/status comes up automatically at boot with bwr enable: 0x00000018 (saved 0x00000000) and boost enable: 0x00000007 (saved 0x00000001) (0 deferred devices).
  • 4K@120Hz bandwidth measurement (dcp55, J773g 26B123):
    • Before (bwr_enable=0x00): GPU DRAM BW 51.54 GB/s, CPU 4P BW 45.59 GB/s.
    • After (bwr_enable=0x18, boost_enable=0x07): GPU DRAM BW 100.39 GB/s in-place and 102.54 GB/s after re-modeset (+9.8% over native macOS 27.2 93.42 GB/s), CPU 4P BW 97.70 GB/s.
  • scripts/checkpatch.pl --strict: 0 errors, 0 warnings, 0 checks.
  • t8132-j773g.dtb builds cleanly.

Assisted-by: LLM (Gemini:4 Neon)

@RyanTheTide

Copy link
Copy Markdown
Contributor

Reviewed the top commit (9bc2bc9) only; the rest is #2541.

The fix is right and the numbers make the case: with the PMC interfaces as second parents of the dispext FE domains, the BWR/boost enables come up at boot (0x18 / 0x07), and DRAM bandwidth under a 4K@120 pipe goes from ~51 to ~100 GB/s. Moving framebuffer0 to ps_dispext0_sys keeps simpledrm from waiting on apple-pmp, and since the FE domain is always-on, the boot framebuffer loses nothing.

One request: this is now the second copy of the same wiring (t8132-j713-j715.dtsi:64 has the dispext0 half). The PMC interfaces are SoC blocks (t8132-memory.dtsi), and every T8132 board with these pipes needs the same parents, including the 14-inch MacBook Pro and the iMacs, which would otherwise ship with half the bandwidth under an external display. Please move &ps_dispext{0,1}_fe { power-domains = <&ps_dispext{0,1}_sys>, <&pmc_dispext{0,1}>; } into a shared T8132 dtsi and drop the per-board copies, J713/J715 included. That needs a J713 boot to show nothing changed there (the PMC status shows the same enables, external display still lights).

Assisted-by: LLM (Claude:Opus 5.5)

@DjDeveloperr
DjDeveloperr force-pushed the djdev/j773g-pmc-dispext-bw branch from 9bc2bc9 to d86ac07 Compare November 22, 2026 02:25
@DjDeveloperr

Copy link
Copy Markdown
Contributor Author

Updated top commit to d86ac07ca41fd (arm64: dts: apple: t8132: wire pmc_dispext0/1 interfaces in shared t8132.dtsi):

  • Moved &ps_dispext0_fe / &ps_dispext1_fe PMC parent wiring (<&ps_dispext{0,1}_sys>, <&pmc_dispext{0,1}>) into shared arch/arm64/boot/dts/apple/t8132.dtsi (right after #include "t8132-memory.dtsi", so drivers/soc/apple/t8132-platform-data.dts — which includes only t8132-memory.dtsi and deletes pmc_dispext{0,1} — is untouched) and dropped the per-board copies from t8132-j713-j715.dtsi and t8132-j773g.dts.
  • Kept the &framebuffer0 -> <&ps_dispext0_sys>, <&ps_dptx_phy> fix on t8132-j773g.dts.
  • Verified scripts/checkpatch.pl --strict (0/0/0) and all t8132-*.dts compile cleanly with dtc.

Assisted-by: LLM (Gemini:4 Neon)

@DjDeveloperr
DjDeveloperr force-pushed the djdev/j773g-272-dcpext0 branch from 9952e46 to bc0067c Compare November 22, 2026 04:39
…132.dtsi

On T8132, the PMC ignores an external display pipe's bandwidth request
(BWR) and display-clock boost votes while its interface enable bits in
0x3828ac020 (bwr_enable) and 0x3828ac028 (boost_enable) remain clear
(drivers/pmdomain/apple/pmc.c). Previously only t8132-j713-j715.dtsi wired
pmc_dispext0 and pmc_dispext1 into ps_dispext0_fe and ps_dispext1_fe,
leaving J773g (as well as J604 and the M4 iMacs) unconnected and cutting
available DRAM bandwidth in half under a 4K@120Hz pipe (51.54 GB/s vs
102.54 GB/s).

Move the shared <&pmc_dispext0> and <&pmc_dispext1> second power-domain
parent wiring for &ps_dispext0_fe and &ps_dispext1_fe into t8132.dtsi
(keeping &ps_dispext0_sys at index 0 so apple_pmgr_ps_hold_parents()
enables pmc_dispext0 cleanly even when ps_dispext0_fe starts active from
iBoot, while leaving t8132-platform-data.dts unaffected because it
includes only t8132-memory.dtsi), and drop the per-board copy from
t8132-j713-j715.dtsi.

On t8132-j773g.dts, point &framebuffer0 at <&ps_dispext0_sys> instead of
<&ps_dispext0_fe> so simpledrm still binds early when CONFIG_APPLE_PMP=m
(apple-pmc defers until apple-pmp binds); ps_dispext0_fe is
apple,always-on and never powered off by genpd.

Signed-off-by: Diljit Singh <hey@djdev.me>
@DjDeveloperr
DjDeveloperr force-pushed the djdev/j773g-pmc-dispext-bw branch from d86ac07 to b76381c Compare November 22, 2026 05:10
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