Files
UniXNU/docs/devices/rothko.md
2026-11-10 14:20:18 +00:00

29 KiB

Xiaomi Redmi K70 Ultra (rothko, MediaTek Dimensity 9300)

woaphone/UniXNU-rothko

Xiaomi Redmi K70 Ultra (MT6989) UniXNU port. Device testing reached the activation page with UFS HS-G5 root access, active CMD display presentation, Goodix/HID activity and eight-core SMP; touch interaction, key wake, Wi-Fi, Bluetooth, battery and stable post-activation behavior remain open.

Source record: porting.toml, d5808a2aec900ff9b99c027e96f375a3bf120bf8.

This page is generated from the device record.

Area Status
Configuration implemented
Firmware and boot host-verified
CPU and interrupts device-verified
Memory and DMA device-verified
Storage and root device-verified
Display device-verified
Input implemented
GPU planned
Network planned
cellular planned
Audio planned
Power implemented
Image assembly host-verified
Diagnostics device-verified

Configuration

implemented

Components: ArmDTPlatform, ArmCpu, XnuPrelink.

Hardware and dependencies

  • MT6989 / Dimensity 9300 board identity, eight CPUs, 13 MHz timer, GICv3 and the iPadOS 17.7.11 / 21H461 donor profile.
  • Stock effective DT, board partition geometry and the reviewed component and patch inputs.

Configuration

  • board.toml selects the rothko board, published mt6989 common source, ipados-17.7.11-21H461 profile and device:dts/rothko.dts.
  • rothko_defconfig, soc:configs/mt6989.config and patches/xnu/series select the kernel configuration and ordered XNU changes.
  • unixnu.lock pins woaphone/xnu-mtk branch rothko, immutable patch base 53bc420e5acda41c1689f6690a85841393a347cf and final fd4edc747395525a914987bb88f063b5e549ecc7.

Procedure

  • Run unixnu defconfig rothko_defconfig -D ../UniXNU-rothko.
  • Run unixnu fetch, unixnu status --json and unixnu tree, then run scripts/porting-record.py check.

Development findings

  • The device repository consumes published MT6989 CPU, PSCI, GICv3 and timer resources.
  • The raw XNU feed names woaphone/xnu-mtk branch rothko at fd4edc7, but the OS profile currently overrides its effective revision with Apple d8b802.
  • The selected arch/SoC/OS/device graph contains 77 patches; its full application and resulting kernel have not been qualified.
  • The separate 18-patch replay from immutable 53bc420 reproduces exactly the source tree of fd4edc7.
  • Board DTS is a hardware inventory/profile input; the minimal ADT generator does not parse it.

Completion criteria

  • Effective source revision, selected patch graph, archive recipes and generated ADT must reproduce the intended K70U inputs before build qualification.

Current limits

  • The full Format 2 release is not qualified: effective XNU pin, driver recipe registration and firmware build adapter remain open.
  • The board DTS memory length currently differs from payload DRAM/owned-memory policy; a stock static memory node is not the live LK mblock map.
  • The common layer currently covers CPU/PSCI/GIC/timer; clocks, power domains, SMMU/DMA and peripheral policy remain device-specific.

Next action

  • Move genuinely reusable MT6989 bus, clock or power resources into the common repository only after a second device or independent hardware trace confirms them.

Evidence

Firmware and boot

host-verified

Components: DarwinBoot.

Hardware and dependencies

  • Separate ChallengeThePeak/Silicium-Rothko firmware repository, rothko DSC/FD paths, DarwinBoot handoff, GOP framebuffer and 56 MiB payload FAT volume.
  • UEFI device reset, PHY, refclk and BlockIO state are treated as firmware-owned and audited separately from XNU.

Configuration

  • board.toml records the UEFI checkout, rothko.dsc, SILICIUM_UEFI.fd path and firmware defines.
  • The XNU payload uses firmware loader handoff, 8 GiB owned-memory intent and PSCI SMC.

Procedure

  • Build UEFI through the official DeviceBuild.py and stuart path in the UEFI checkout.
  • Build the payload and boot image with the official UniXNU tools after the firmware FD is available.

Development findings

  • Official UEFI builds completed offline and the device reached the Darwin handoff; UFS link/DP success and later NOP/descriptor issues were observed in separate captures.
  • An earlier firmware candidate ran with a mistaken UFS reference-clock setting and is recorded as unresolved; this repository does not claim that the boot LUN was written, and it does not claim that the current firmware is safe for another hardware run.
  • The board now imports shared MT6989 CPU/GIC/timer resources from UniXNU-mt6989-common while keeping firmware paths and handoff policy board-specific.
  • The complete arch/SoC/OS/device XNU patch sequence remains in the format 2 package graph.

Completion criteria

  • A reviewed firmware revision builds, enters DarwinBoot, publishes the documented framebuffer and preserves the handoff memory contract.

Current limits

  • The UEFI source and current unx18 safety candidate live in the separate ChallengeThePeak/Silicium-Rothko repository; current firmware runtime closure and the cause of the earlier reference-clock incident remain open.

Next action

  • Qualify the official DeviceBuild/stuart adapter and complete UFS command/state boundary before any future matched runtime claim.

Evidence

CPU and interrupts

device-verified

Components: ArmCpu, MtkCpuFreq.

Hardware and dependencies

  • Eight MT6989 CPUs, GICv3, 13 MHz timer, PSCI SMC, self-IPI and MMIO-IPI paths.
  • The stock FDVFS interface and effective DT CPU vote table for the fixed boot-frequency request.

Configuration

  • payload kernel_args enables cpus=8, sheng-smp, armcpu-self-ipi and armcpu-mmio-ipi.
  • The board DTS records cpu-count=8 and timebase-frequency=13000000.

Procedure

  • Build ArmCpu and MtkCpuFreq archives, link the kernel and capture K1-K4, timer and IPI markers from pmsg.
  • Compare the fixed FDVFS request path with the stock DT and firmware table.

Development findings

  • All eight CPU startup, timer and IPI activity were observed on the device.
  • The CPU affinity, PSCI, GICv3 and 13 MHz timer description now comes from UniXNU-mt6989-common.
  • MtkCpuFreq is a fixed boot request; it is not a dynamic governor and no XNU-side frequency measurement proves that it resolves the reported system slowness.

Completion criteria

  • Every enabled CPU reaches its entry checkpoint and timer/IPI handling works; any frequency claim must include a measured runtime frequency.

Current limits

  • Dynamic cpufreq/governor behavior and the cause of perceived system slowness remain unverified.

Next action

  • Capture a same-image runtime frequency/latency trace before changing CPU policy again.

Evidence

Memory and DMA

device-verified

Components: RamBlock.

Hardware and dependencies

  • MT6989 DRAM map, 8 GiB owned-memory target, XNU 4 KiB page tables, DMA ranges and the packed seed ramdisk.
  • The 4 GiB and 8 GiB bootstrap/page-table observations from the K70U bring-up captures.

Configuration

  • payload.adt.owned_memory_size is 0x200000000 and the board DTS records the DRAM range.
  • NativeRamBlock supplies the packed Data seed without making the UEFI firmware a persistent disk writer.

Procedure

  • Generate the board ADT, link the kernel and replay/observe the native 4 KiB bootstrap table construction before payload handoff.
  • Check the seed hash and logical Data size against the recorded source artifact.

Development findings

  • The 8 GiB bootstrap table pool was enlarged after a reproducible page-table boundary failure; the corrected path reached later XNU stages.
  • The Data seed was imported in the root-matching captures; memory pressure and jetsam were still visible in earlier 2 GiB/4 GiB runs.
  • CPU topology is inherited from the shared MT6989 DTSI; RAM ownership and reservations remain board-specific.

Completion criteria

  • The declared memory and DMA regions are represented in ADT, page tables are built for the selected memory size and the packed seed imports with its recorded identity.

Current limits

  • A complete long-duration memory-pressure comparison across 4 GiB and 8 GiB is not available; this area does not claim that RAM size alone fixes the UI stalls.

Next action

  • Repeat a same-image 8 GiB run with persistent pmsg and memory-pressure counters if the device is recovered.

Evidence

Storage and root

device-verified

Components: SoftwareAES, RamBlock, AppleKeyStoreState, VerifiedRoot, NativeMtkUfs, NativeAES, NativeRamBlock.

Hardware and dependencies

  • MT6989 UFS host at 0x112b0000, UFS4 device, LU2 4096-byte user blocks, HS-G5 dual-lane XNU path, GPT unixnu window and root UUID.
  • OS/SystemOS readonly windows, Data seed and the explicit distinction between XNU reads and UEFI BlockIO ownership.

Configuration

  • board.toml records the unixnu GPT LBA range, 4096-byte blocks, root-match UUID and readonly OS/SystemOS windows.
  • NativeMtkUfs and NativeAES are included as separate XNU services; the UEFI repository is separately audited for refclk/reset/PHY writes.

Procedure

  • Build the UFS/AES/RamBlock archives, prelink the kernel and capture link, lane, capacity, GPT, root-match and read markers.
  • Keep all persistence writes inside the declared unixnu Data/Log windows and review UEFI writes independently.

Development findings

  • Device evidence showed XNU UFS HS-G5 dual-lane reads, LU2 activity, root matching and sustained reads.
  • An earlier firmware candidate that ran with a mistaken UFS reference-clock setting is not attributed to a proven boot-LUN write; the firmware-side refclk/reset/PHY audit remains separate from this XNU evidence.
  • The board firmware contract now points to the published ChallengeThePeak/Silicium-Rothko unx18 read-only UFS baseline.
  • UFS registers, driver selection, partitions and Data policy stay in the board repository after the common-resource split.
  • The complete four-layer XNU patch sequence remains selected for the storage build graph.

Completion criteria

  • Host/link/read/root/persistence boundaries match the board inventory and no write path escapes the declared unixnu windows.

Current limits

  • Long-term UFS persistence and the earlier reference-clock incident remain unresolved; this record does not claim that the current firmware is safe to run on hardware again.

Next action

  • After device recovery, run a read-first UFS capture with the reviewed UEFI safety revision and record the exact firmware and image hashes.

Evidence

Display

device-verified

Components: GopFramebuffer, NativeMobileDisplay.

Hardware and dependencies

  • 1220x2712 panel geometry, 32 bpp GOP framebuffer, CMD-mode DSI and the firmware-to-XNU framebuffer handoff.
  • The requirement to push each composed frame actively on the CMD panel.

Configuration

  • The board payload keeps the active CMD presentation flags and the DTS records 1220x2712x32 geometry.
  • NativeMobileDisplay maps the framebuffer and performs the explicit post-composition push.

Procedure

  • Build the display archive, prelink and package the payload, then observe the first frame and activation UI on the physical device.
  • Keep display and UFS logs in pmsg so a frozen screen can be separated from a reset.

Development findings

  • The device reached the activation page with the CMD active-push path.
  • The board payload contract is paired with the published ChallengeThePeak/Silicium-Rothko firmware tree that provides the GOP handoff.
  • The board DTS now imports the shared MT6989 CPU/GIC resources while retaining display geometry and panel policy locally.
  • The complete arch/SoC/OS/device XNU patch sequence remains selected for the display build graph.
  • Long-term presentation, unlock-to-desktop stability and the later black-screen behavior remain open; the evidence does not identify GPU as the cause.

Completion criteria

  • A known frame has correct geometry and colors, userland reaches the activation page and repeated frame pushes continue after handoff.

Current limits

  • Stable desktop presentation and wake-after-lock have not been demonstrated on the same final public repository revision.

Next action

  • Capture frame-push counters together with lock/unlock and wake-key events in one device run.

Evidence

Input

implemented

Components: NativeHidInput, GoodixBerlin9916K.

Hardware and dependencies

  • Goodix Berlin9916K on SPI6, reset GPIO3, IRQ GPIO4, 1220x2712 coordinate space and the corrected MT6363/GPIO key routes.
  • Touch HID report format, contact IDs and the three physical key routes in the effective DT/DTBO.

Configuration

  • The board kit binds Goodix coordinates and firmware configuration; the kernel flags enable k70u-touch and HID input.
  • The source records volume-up GPIO196 and MT6363 HOMEKEY/PWRKEY routes without treating the KPD fallback as the board's final mapping.

Procedure

  • Build NativeHidInput and the rothko kit, then capture sensor reports, HID queue movement, touch interactions and key wake.
  • Compare reported coordinates and button events with the display geometry and effective DT resources.

Development findings

  • Goodix sensor reports and HID queue activity were observed on device.
  • UI touch interaction and physical-key wake were not confirmed; the earlier three-second lock and black-screen report remains open.

Completion criteria

  • Touches map to the display, simultaneous contacts keep stable IDs, release events arrive and all three buttons produce the expected HID reports and wake behavior.

Current limits

  • Only the lower-level report/queue path is device-observed; successful UI interaction and wake-key behavior are pending.

Next action

  • Run a matched-image touch/volume-up/power/volume-down matrix and save the pmsg plus screen state for each event.

Evidence

GPU

planned

Hardware and dependencies

  • MT6989 GPU/firmware, SMMU, submission, fence and Metal client contract.

Configuration

  • The current display path uses GOP/CMD presentation and does not claim a native GPU implementation.

Procedure

  • Collect the board GPU firmware and SMMU ownership facts before selecting a driver; do not import Qualcomm GMU terminology as a MediaTek hardware fact.

Development findings

  • The observed UI and display path does not by itself prove a GPU implementation or a GPU root cause for the reported lag.
  • The MT6989 common repository intentionally exports no GPU, SMMU, firmware or Metal policy.

Completion criteria

  • GPU submissions, mappings, fences and a selected Metal client render and present a known image.

Current limits

  • No native GPU driver, firmware contract or device rendering trace is recorded in this repository.

Next action

  • Inventory the stock GPU firmware, SMMU streams and Android reference implementation.

Evidence

Collect evidence using the next action.

Network

planned

Hardware and dependencies

  • MT6639 Wi-Fi over the board's ConnV3/PCIe/AXI/WFDMA path and its firmware/calibration inputs.

Configuration

  • No Wi-Fi component is selected in board.toml; proprietary firmware is not copied into this repository.

Procedure

  • Identify the bus, firmware, interrupts, DMA and OS interface, then implement scanning and packet transfer.

Development findings

  • MT6639 Wi-Fi and UART3 Bluetooth were documented during bring-up; neither has an implemented OS transport service.
  • Wireless bus/power resources are recorded in the offline peripheral audit; the minimal board DTS does not yet describe the complete transport.

Completion criteria

  • Scanning, association, address configuration and sustained receive/transmit traffic pass with the selected firmware.

Current limits

  • The bus contract, firmware source and XNU network service are incomplete.

Next action

  • Collect the stock DT, firmware names and Android driver trace for MT6639.

Evidence

Collect evidence using the next action.

cellular

planned

Hardware and dependencies

  • Cellular/modem scope and firmware inventory for the Redmi K70 Ultra.

Configuration

  • No cellular component is selected and no modem firmware is copied into the device repository.

Procedure

  • Confirm whether this board's UniXNU scope includes modem support before selecting a protocol stack.

Development findings

  • No cellular implementation or device evidence was part of the current bring-up.
  • The common MT6989 source contains no modem or cellular policy.

Completion criteria

  • If in scope, modem firmware, control/data protocol and network registration pass; otherwise document the board exclusion.

Current limits

  • The current port has no modem implementation or verified scope decision.

Next action

  • Record a not-applicable decision or inventory the modem interface before implementation.

Evidence

Collect evidence using the next action.

Audio

planned

Hardware and dependencies

  • MT6989 audio DSP, codec, clocks, firmware services and playback/capture routes.

Configuration

  • No audio component is selected in the current board configuration.

Procedure

  • Inventory the audio DSP/codec and Android routes, then implement one playback and capture path.

Development findings

  • Audio was not part of the current device bring-up evidence.
  • The common MT6989 source contains no DSP, codec or audio-route policy.

Completion criteria

  • Declared routes carry the selected audio format through playback and capture with correct start/stop behavior.

Current limits

  • No audio source, firmware or device result is recorded.

Next action

  • Collect stock audio DT, firmware and route information.

Evidence

Collect evidence using the next action.

Power

implemented

Components: MtkCpuFreq.

Hardware and dependencies

  • MT6989 FDVFS support table, per-core vote registers, firmware-owned voltage/PLL control and the active NFG1000A gauge address.

Configuration

  • MtkCpuFreq sends the fixed boot request supported by the board's FDVFS table; it is not a runtime governor.
  • The active battery gauge is recorded as a separate pending backend and is not confused with the disabled MT6375 path.

Procedure

  • Build and link MtkCpuFreq, then compare request/readback logs with stock FDVFS metadata.
  • Implement battery/IOP power source only after the I2C11/0x55 NFG1000A path is independently verified.

Development findings

  • The fixed CPU performance request is source- and host-verified; it does not establish the cause of the reported lag.
  • Battery service is not connected to XNU and no sleep/resume transition is claimed.
  • Clock, regulator, FDVFS and battery policy remain board-specific after the shared CPU-resource split.

Completion criteria

  • Declared power controls and battery readings match observations and each claimed sleep/resume transition restores dependent peripherals.

Current limits

  • Runtime governor, battery service and sleep/resume are incomplete; only the fixed CPU request is implemented.

Next action

  • Reuse the already tested NFG1000A read-only register backend, then implement I2C11 transport and IOPMPowerSource publication; retain runtime governor and sleep/resume as separate open items.

Evidence

Image assembly

host-verified

Components: XnuPrelink, DarwinBoot.

Hardware and dependencies

  • iPadOS 17.7.11 / 21H461 kernelcache, prelink ABI, payload FAT volume, UEFI FD and stock-compatible boot image envelope.

Configuration

  • board.toml records the kernelcache hash, prelink roots/slide, payload volume, memory size and root-match UUID.
  • scripts/package.sh delegates to the official unixnu release path; package/unixnu.toml records the payload entry.

Procedure

  • Run the official kernel link, XnuPrelink, UEFI build, payload and boot_v4 packer in order.
  • Record every output hash and keep generated images outside the Git repository.

Development findings

  • The official offline build chain completed for the recorded unx15/unx16/unx17/unx18 candidates; the final public device repository intentionally contains no firmware or image binaries.
  • The package contract consumes the published ChallengeThePeak/Silicium-Rothko firmware tree through the path configured in board.toml and CI.
  • The package graph now resolves the MT6989 common commit recorded in unixnu.lock.
  • The package graph preserves arch, SoC, OS and device patch-series arrays in their intended order.
  • The packaging area owns .gitignore and the *.toml.example templates so generated images, local build overrides and local module overrides stay out of Git; the CI changed-area check no longer sees an unowned path.

Completion criteria

  • Kernel and prelink succeed, payload contains declared inputs and boot images match the firmware's format and size constraints.

Current limits

  • The earlier isolated unx15-unx18 build results do not qualify the current Format 2 release graph.
  • Current build-arm-driver.py rejects NativeMtkUfs and MtkCpuFreq; the main firmware adapter uses direct BaseTools/CLANGPDB rather than DeviceBuild/stuart.
  • Source revision override, patch graph, board input archive integration and Linux toolchain inputs need qualification before full release.

Next action

  • Run scripts/package.sh --no-deps --check from a compatible main checkout for local plan inspection, then repair and qualify the source/recipe/firmware integration before claiming full release success.

Evidence

Diagnostics

device-verified

Hardware and dependencies

  • pmsg/ramoops, UEFI console, dmesg, unixnu log ring, adb/fastboot capture and the persistent evidence reports.

Configuration

  • The payload enables serial/pmsg and the board keeps the logs needed to distinguish handoff, XNU, UFS, display and userland stages.

Procedure

  • Capture pstore/pmsg and firmware logs after each boot attempt; record image/source hashes and the last completed checkpoint.
  • Run the official porting-record checker before publishing the device revision.

Development findings

  • The record distinguishes observed XNU entry, SMP, HID, UFS/root and activation-page milestones from unresolved UI wake, peripheral and firmware-state questions.
  • The package workflow checks out the canonical UEFI repository ChallengeThePeak/Silicium-Rothko before the device release command.
  • unixnu.lock records the same MT6989 common commit pinned by feeds.toml.

Completion criteria

  • Each claimed milestone has a reproducible command, matching source/image revision and an accessible result.

Current limits

  • The public repository does not contain the large proprietary logs or generated UEFI images; references point to the compact verification record and parent bring-up reports.

Next action

  • Publish a compact per-run evidence manifest alongside the next recovered-device test.

Evidence