[hishamMounir]

This series brings the bootloader up on T8142 (Apple M5), as seen on the 14"
MacBook Pro (Mac17,2 / J704). With it the tree reaches a usable proxy on that
machine and chainloads Linux from the internal SSD.

What is here

• pcie: support t8142 — needed for the internal NVMe and WLAN.
• kboot: raise WLAN_REG_ON (SMC gP13) before PCIe init — without it the
  WLAN device never appears on J704. Includes the small SMC write plumbing.
• dapf: find the DAPF reg by its ADT instance name — on t8142 the reg
  index used on earlier SoCs lands on a page shared with PMGR and SErrors on
  access. The index path is kept for the SoCs that use it.
• kboot: reserve the CTRR ranges on t8142 — adds t8142 to the existing
  CTRR reservation, and reserves the second CTRR instance that covers the
  DeviceTree/TrustCache area below the bootloader. See the porting note below.
• kboot: reserve the DCP carveouts and split DCP mapping runs, reserve unlisted iBoot DCP mappings, skip iommu-addresses without phandle — iBoot
  maps some regions (t8142 dcp_data) in several IOVA runs, and Linux replaces
  a locked DART's tables with exactly what iommu-addresses lists, so each run
  has to be described. A run with no mapping now fails rather than silently
  handing Linux a truncated list, once at least one run has been described.
• kboot: match DT CPU nodes to running CPUs by MPIDR — index-based
  matching does not hold on this SoC.
• kboot_atc: copy the atc-phy tunables on t8142 — same tunable layout as
  t8122/t8132.
• pmgr_adt2dt: resolve no_ps parents to their nearest real ancestors.
• smp: log once when a call target hasn't taken a call after 1 s — a
  secondary that never takes a cross-CPU call otherwise looks exactly like a
  dead board. Logs once, then waits as before.

Porting notes

These were developed against AsahiLinux/m1n1 and rebased onto this tree, so
three of them were reworked rather than applied:

• CTRR: this tree already reserves the CTRR range its own way, keyed on
  chip_id, and has no names for the second instance's registers. Rather than
  import Asahi's renamed CTRR_C_*/CTRR_D_* set, I added only the two
  D-instance encodings (3, 0, 11, 0, 2/3) next to the existing pair and
  extended the existing block. The D reservation is gated on t8142 and skipped
  when the instance reads back zero, so the M4 targets are untouched.
• kboot_atc: only the t8142 compatible is added. Asahi's t8130 entry and
  its CIO-lane tunable block do not exist here and were deliberately not
  imported.
• smp: include fixup only (iodev.h for the console flush).

Testing

J704 (Mac17,2, T8142, 16 GB) with the bootloader as the custom boot object of a
second macOS install (26.6.2 25G83, later 27.2 26B123; system firmware
mBoot-20457.41.3). Linux boots from the internal SSD and the proxy runs over
USB-C from an M4 Mac mini. AIC3, pmgr (386 devices), framebuffer, dockchannel
UART, USB (3 DARTs, t8110), PCIe and NVMe all come up.

Caveats I would rather state than have you discover: that testing was done with
the AsahiLinux-based build, not with this tree — the three reworked commits
above are compile-tested here but have not been booted. Every C file the series
touches compiles warning-free with aarch64-linux-gnu-gcc; I could not link a
full image because neither machine I have here has a bare-metal Rust target for
librust.a, and clang-format is not installed either, so no formatting pass
was run. The new DCP failure path has not been hit on hardware: it needs a
region iBoot maps in several runs with a gap, which this machine has not
produced.

Provenance

• Clean room: every commit here is platform bring-up derived from the ADT,
  hypervisor traces and driver experiments. No Apple binary was disassembled for
  any of them. My GPU work on this machine is not clean-room and is staying
  local — none of it is in this series, and I will not send it here.
• These patches were briefly open as AsahiLinux/m1n1#747, which I closed in
  favour of carrying them in this tree.
• sh1ftmaker's m5-linux-research was a reference during this bring-up. Their
  T8142 GPU firmware memory handoff patch is not included — it is their work to
  send.
• The hypervisor side of this machine's bring-up (12 commits: t8142 timer
  trapping via IMASK + ECV, impdef sysreg handling, a watchdog mode,
  run_guest --disable-node) rests on the SPTM/SPRR emulation by Sven Peter and
  Lukas Radermacher. I have it ready but am holding it, since janmarek98's
  M4 Pro/Max SPTM PR is still open here and one of my commits reverts an
  upstream change I cannot yet explain. Happy to send it whenever you want it.
• I also have t8142 kernel-side DT and an irq-apple-aic fix, which overlap
  Lostosword's open PR HV fixups for macOS 26.6.2 #4 in GravityLinux/linux. I would rather rebase the
  non-overlapping parts onto that branch than open a competing PR — tell me
  which you prefer.

🤖 Generated with Claude Code