[joske]

#1727 gave the M4 Max macOS's Low Power mode, with the CPU ceiling from the J616c's ADT (cpu-low-power-frequency-max = 2808 MHz) for every apple,t6041 machine. The 14" J614c's ADT has 2616 MHz there. The macOS 26.6.2 ADTs show the same 14"/16" split on the M4 Pro (J614s 2616, J616s 2808). So in power-saver the J614c's P clusters ran one table step higher than macOS lets them (2868 instead of 2616 MHz).

Change

• The single low_power_machine compatible becomes a per-machine table, first match wins: apple,j614c 2616 MHz, then apple,t6041 2808 MHz. So the J616c, and the J575c whose ADT I don't have, keep 2808. The M4 Pro still gets no profile, and T6050 keeps its ceiling on any machine.
• KUnit: a new case checks the J614c, J616c, J575c, J614s and table-less results. With the two table rows swapped, it fails.

Testing

J614c (MacBook Pro 14" M4 Max), os-fw 26.6.2, kernel integration/aurora fdef45b + this commit + #1793, with CONFIG_ACPI=y (see below).

• apple-clpc offers low-power balanced performance; powerprofilesctl lists power-saver.

• scaling_max_freq:

  Cluster	balanced	power-saver
  P (policy0, policy10)	4104 MHz	2616 MHz
  E (policy1)	2592 MHz	2592 MHz

  Back to balanced restores 4104 MHz.

• An all-core spin (16 threads) for 14 s, on battery, SMC "Total System Power" after 7 and 10 s:

  Profile	P clusters	System power
  balanced	4104 MHz	91.2 / 91.8 W
  power-saver	2616 MHz	43.2 / 43.3 W

  Idle is 4.7 W in both. I didn't measure the old 2868 MHz step.

• KUnit clpc on arm64 QEMU: 13 passed, 1 skipped (apple_g17_rx_util_min_test, needs CONFIG_UCLAMP_TASK).

Note: the profile only registers with CONFIG_ACPI_PLATFORM_PROFILE, which lives under drivers/acpi/ and is only built with CONFIG_ACPI=y. The aurora package config has # CONFIG_ACPI is not set, so with the shipped kernel no T6041 or T6050 machine gets a platform profile. For this test I built with CONFIG_ACPI=y; with no ACPI tables, the kernel logs "ACPI: Interpreter disabled" and boots from the DT as before.

🤖 Generated with Claude Code