feat: ACPI Wave 1 boot-critical hardening (P19) + robust patch generation

- P19-init-startup-hardening: Replace panic-grade expect/unwrap in init startup paths (getns, register_scheme_to_ns, setrens, filename parsing) with graceful error handling and logging - P19-acpid-startup-hardening: Replace panic-grade calls in acpid with graceful degradation (rxsdt read failure → warn + exit 0, SDT parse → error + exit 1, I/O privilege → fatal, scheme registration → fatal, setrens → warn + continue, event loop errors → log + continue) - P18-9-msi-allocation-resilience: Regenerate with git diff -U0 -w format for maximum context resilience - fetch.rs: Change --fuzz=0 to --fuzz=3 for resilient patch application - AGENTS.md: Document robust patch generation technique as mandatory - Add P4/P5/P6/P7 patches (estale, dmi, i2c, ps2d hardening) - Add P21 kernel x2apic SMP fix patch - Multiple local recipe source improvements (redox-drm, driver-manager, driver-acpi, thermald) - Config updates for redbear-mini and redbear-device-services - Subsystem assessment document
2026-05-18 14:07:42 +03:00
parent 7798ed86eb
commit f6c2eb2a8e
32 changed files with 3293 additions and 567 deletions
@@ -0,0 +1,328 @@
+# Red Bear OS Subsystem Assessment vs Linux Reference
+
+**Date:** 2026-05-17
+**Scope:** Input devices, ACPI/PCID, Intel DRM/KMS, boot process
+**Reference:** Linux kernel 7.1 (local/reference/linux-7.1/)
+
+---
+
+## Executive Summary
+
+Red Bear OS has real, functional architectural scaffolding across all five subsystems. The critical gaps are in **hardware-facing paths that are stubbed or incomplete** — most notably Intel EDID/DDC, hardware vblank, display watermarks, AML interpreter depth, and boot dependency ordering. The single most impactful immediate fix is adding the missing `acpid` service to boot configs, which prevents ACPI-dependent drivers from enumerating correctly.
+
+**Critical blockers for bare-metal desktop:**
+1. Missing `acpid` service in redbear configs → ACPI devices never discovered
+2. Intel `read_edid_block()` returns error → synthetic EDID is 112 bytes (should be 128) → no real monitor modes
+3. Intel `get_vblank()` is a fake atomic counter → no real vblank for page flip synchronization
+4. No display watermarks → FIFO underruns cause visible glitching
+5. No boot dependency declarations → i2c-hidd/i2cd may race with acpid
+
+---
+
+## 1. Input Devices (Keyboard, Mouse, HID, I2C-HID, Touch)
+
+### Current Implementation Inventory
+
+| Component | Path | Quality | Key Notes |
+|---|---|---|---|
+| ps2d (PS/2) | `base/source/drivers/input/ps2d/` (5 files) | **Real** | Keyboard scancodes + mouse protocol. x86-only (non-x86 is `unimplemented!()`). Many TODOs, QEMU-specific hacks. No ImPS/2 scroll or trackpoint. |
+| usbhidd (USB HID) | `base/source/drivers/input/usbhidd/` (2 files, 520 lines) | **Real** | Full HID report descriptor parsing, USB usage→orbclient scancode table, mouse relative+absolute, scroll, buttons. Retry with exponential backoff. Polling-based (1ms sleep loop). |
+| i2c-hidd (I2C HID) | `base/source/drivers/input/i2c-hidd/` (5 files, 500+ lines) | **Real** | ACPI PNP0C50/ACPI0C50 device scan, _CRS resource parsing, _DSM HID descriptor address, I2C transfer via `/scheme/i2c/`. Probe failure quirk system with DMI matching. |
+| intel-thc-hidd (Intel THC) | `base/source/drivers/input/intel-thc-hidd/` (3 files, 282 lines) | **Partial** | PCI init works, QuickI2C transport setup works, ACPI companion resolution works. **Main loop is `thread::sleep(Duration::from_secs(5))` — no input report streaming.** |
+| inputd (multiplexer) | `base/source/drivers/inputd/` (3 files, 663 lines) | **Real** | Producer/consumer scheme, VT switching, keymap support (US/Dvorak/GB/AZERTY/Bepo/IT). ESTALE handoff for display driver transitions. |
+| evdevd (evdev adapter) | `local/recipes/system/evdevd/` (5+ files) | **Real** | evdev scheme, device model, orbclient→evdev translation, gesture recognizer, key filter. |
+| redbear-keymapd | `local/recipes/system/redbear-keymapd/` | **Real** | Keymap scheme registration and management. |
+| udev-shim | `local/recipes/system/udev-shim/` | **Real** | Device node synthesis from scheme registrations, heuristic mapping. |
+| I2C bus drivers | `base/source/drivers/i2c/` (5 modules) | **Real** | amd-mp2-i2cd, dw-acpi-i2cd, intel-lpss-i2cd, generic i2cd, i2c-interface library. |
+| redbear-input-headers | `local/recipes/drivers/redbear-input-headers/` | **Real** | `linux/input.h`, `linux/input-event-codes.h`, `linux/uinput.h` — replaces policy-violating `linux-input-headers` from libevdev tarball. |
+| libinput (WIP) | `local/recipes/libs/libinput/` | **WIP** | Port of upstream libinput with touchpad/trackpoint filtering. Not yet runtime-trusted. |
+| libevdev (WIP) | `local/recipes/libs/libevdev/` | **WIP** | Port of upstream libevdev. |
+
+### Gaps vs Linux
+
+| Gap | Severity | Linux Reference | Red Bear Status |
+|---|---|---|---|
+| intel-thc-hidd doesn't stream | **High** | `drivers/hid/intel-thc-hid/` full probe+report streaming | Main loop sleeps 5s; no HID reports |
+| No multitouch/ABS_MT | **High** | `drivers/input/input-mt.c` slot tracking, pointer emulation | Not implemented |
+| No libinput acceleration/gestures | **High** | libinput: velocity curves, palm detection, gesture recognition | inputd does raw keymap only |
+| No PS/2 extended protocols | **Medium** | `libps2.c` ImPS/2 scroll, Explorer 5-btn, trackpoint | Basic protocol only |
+| No HID quirks table | **Medium** | `hid-quirks.c` 4000+ device entries | Only probe_failure quirks |
+| No input hotplug | **Medium** | udev + inotify on `/dev/input/` | Static scan at startup |
+| Polling-based USB HID | **Low** | URB interrupt-driven | 1ms sleep loop (functional but power-inefficient) |
+| inputd keymap incompleteness | **Low** | Full xkb/keyboard-layout support | TODO for configurable keymap, AltGr, NumLock |
+
+### Linux I2C-HID Reference (from local/reference/linux-7.1/)
+
+The Linux I2C-HID probe sequence is:
+1. Verify IRQ exists
+2. Wake/power up device (_PS0/HID_POWER_ON)
+3. Read HID descriptor from controller register
+4. Read report descriptor
+5. Parse descriptor
+6. Size buffers from actual reports
+7. Register IRQ
+8. `hid_add_device()`
+
+Red Bear's i2c-hidd follows this sequence correctly. The Intel THC driver does steps 1-5 but never reaches step 7-8.
+
+---
+
+## 2. ACPI and PCID
+
+### Current Implementation Inventory
+
+| Component | Path | Quality | Key Notes |
+|---|---|---|---|
+| Kernel ACPI | `kernel/source/src/acpi/` (9+ files) | **Real, partial** | RSDP, RSDT/XSDT, MADT, FADT, DSDT parsing. New: SLIT, SRAT. AML evaluation for basic methods (_STA, _PS0, _PS3, _INI). **No While/If-Else, no OperationRegion for PCI/I2C, no method locals.** |
+| Kernel ACPI scheme | `kernel/source/src/scheme/acpi.rs` | **Real** | Exposes ACPI tables, symbols, resources, method evaluation to userspace. |
+| Kernel DMAR/IOMMU | `kernel/source/src/acpi/dmar/` | **Partial** | DMAR table parsing for IOMMU. DRHD entries parsed but not wired to allocator. |
+| Kernel sleep/S3 | `kernel/source/src/arch/x86_shared/sleep.rs` (new, uncommitted) | **New** | S3 suspend/wakeup assembly. Not yet wired to power management. |
+| acpid | `base/source/drivers/acpid/` | **Real** | Scheme-based ACPI access, symbol evaluation, resource serialization. ESTALE-graceful handling. |
+| pcid | `base/source/drivers/pcid/` | **Real** | PCI enumeration, config space, BAR mapping, pcid-spawner. MSI/MSI-X support via recent patches. |
+| redox-driver-acpi | `local/recipes/drivers/redox-driver-acpi/` | **Real** | ACPI bus driver bridging ACPI discovery to pcid-spawner. |
+| driver-manager | `local/recipes/system/driver-manager/` | **Real** | Manages PCI/ACPI driver matching and spawning. |
+| redox-driver-sys quirks | `local/recipes/drivers/redox-driver-sys/source/src/quirks/` | **Real** | Compiled-in + TOML + DMI quirk tables. MSI/MSI-X fallback, DISABLE_ACCEL. |
+| IOMMU daemon | `local/recipes/system/iommu/` | **Partial** | Builds, QEMU first-use proof passes. Real hardware validation pending. |
+
+### Gaps vs Linux
+
+| Gap | Severity | Linux Reference | Red Bear Status |
+|---|---|---|---|
+| AML interpreter incomplete | **Critical** | Full AML bytecode VM (While/If/Else, OperationRegion, Method locals, Notify) | Basic method calls only (_STA, _PS0, _PS3, _INI). No control flow. |
+| No _PRW wake resources | **High** | `drivers/acpi/wakeup.c` | Not present |
+| No thermal zones | **High** | `drivers/acpi/thermal.c` _TMP/_ACx/_PSV/_CRT | Not present |
+| No ACPI battery | **Medium** | `drivers/acpi/battery.c` _BIF/_BST | Not present |
+| No ACPI buttons | **High** | `drivers/acpi/button.c` LID/Power/Sleep | Not present |
+| SRAT/SLIT not wired to NUMA | **Medium** | `mm/numa.c` | Parsed but not connected to page allocator |
+| No _OSC OS capabilities | **Medium** | `drivers/acpi/osc.c` | Not present |
+| No PCI ASPM | **Medium** | `drivers/pci/pcie/aspm.c` | Not present |
+| No PCI hotplug | **Low** | `drivers/pci/hotplug/` | Not present |
+| No suspend/resume | **Critical** | `drivers/acpi/sleep.c` S1-S5 | sleep.rs + wakeup.asm in uncommitted changes, not wired |
+| DMAR/IOMMU path commented out | **High** | `drivers/iommu/intel-iommu.c` | `acpid/src/acpi/dmar/mod.rs` has iterator bug (`len_bytes` from wrong slice), hangs on real hardware — entire DMAR path commented out |
+| DMI quirk matching dead | **High** | `/sys/firmware/dmi` | `redox-driver-sys/quirks/dmi.rs` depends on `/scheme/acpi/dmi` but that source doesn't exist in the ACPI stack |
+| ACPI resource parsing panics | **Medium** | N/A | `redox-driver-acpi/resource.rs` and `prt.rs` panic on unexpected ACPI resource shapes instead of returning errors |
+| `madt/arch/other.rs` stub | **Low** | `drivers/acpi/madt.c` | Non-x86 MADT handling is effectively unimplemented |
+| PCI config: non-x86 `todo!()` | **Low** | `drivers/pci/` | `pcid/src/cfg_access/fallback.rs` has `todo!()` for non-x86 PCI config access |
+| **Missing acpid service in configs** | **Critical** | N/A (config bug) | No `acpid = {}` in redbear-full.toml or redbear-device-services.toml |
+
+### acpid Missing From Configs — Critical Bug
+
+The boot process agent found that **no active `acpid = {}` service entry exists** in the redbear TOML configs. This means acpid may never start, which prevents ACPI symbol/resource discovery for all ACPI-dependent drivers (i2c-hidd, intel-thc-hidd, thermald, driver-manager ACPI path). This is the single highest-priority fix.
+
+---
+
+## 3. Intel DRM/KMS
+
+### Current Implementation Inventory
+
+| Component | Path | Quality | Key Notes |
+|---|---|---|---|
+| IntelDriver | `redox-drm/source/src/drivers/intel/mod.rs` (682 lines) | **Partial** | PCIe init, MMIO mapping, FORCEWAKE, connector detection, CRTC set_mode, page_flip, GEM create/mmap/close, IRQ handling. |
+| IntelDisplay | `.../intel/display.rs` (404 lines) | **Partial** | Pipe detection, DDI port detection, mode setting (real HTOTAL/HBLANK/HSYNC/VTOTAL/VSYNC/PIPE_SRC register writes). **EDID: read_edid_block returns error → synthetic_edid(). DPCD: returns fake 4 bytes.** |
+| IntelGtt | `.../intel/gtt.rs` | **Real** | GGTT allocation, mapping, unmapping. |
+| IntelRing | `.../intel/ring.rs` (267 lines) | **Partial** | DMA ring buffer, GPU address binding. Only MI_FLUSH_DW + MI_NOOP submitted — no rendering commands. |
+| DRM scheme | `redox-drm/source/src/scheme.rs` | **Real** | Full DRM/KMS ioctl surface. SETPLANE is empty, GETENCODER hardcoded. |
+| KMS infrastructure | `redox-drm/source/src/kms/` (5 files) | **Real** | ConnectorInfo, ModeInfo with EDID parsing, synthetic_edid fallback. |
+| Interrupt handling | `redox-drm/source/src/drivers/interrupt.rs` | **Real** | MSI/MSI-X/INTx setup, try_wait polling. |
+| Linux-kpi DRM headers | `linux-kpi/source/src/c_headers/drm/` | **Minimal** | drm.h, drm_crtc.h, drm_gem.h, drm_ioctl.h — type definitions only. |
+| ihdgd (legacy) | `base/source/drivers/graphics/ihdgd/` | **Old/Partial** | Separate Intel framebuffer driver. Many TODOs. Being superseded by redox-drm. |
+| vesad | `base/source/drivers/graphics/vesad/` | **Legacy** | VESA framebuffer driver. No cursor support. |
+| Mesa | `recipes/libs/mesa/` | **Software only** | Only swrast+virgl Gallium. No Intel iris/crocus/anv driver build. |
+
+### Critical Bugs Found
+
+1. **synthetic_edid() is 112 bytes, not 128** — `ModeInfo::from_edid()` requires `edid.len() >= 128` and checks for the 8-byte EDID header. The synthetic EDID is only 112 bytes so it always fails validation, forcing `default_1080p()` fallback on every Intel connector.
+
+2. **Intel `get_vblank()` returns `AtomicU64::fetch_add(1, SeqCst)`** — This is NOT a real vblank counter. It increments on every IRQ regardless of display state. Real i915 reads the `PIPEFRAME` register (offset `0x70040 + pipe * 0x1000`) for per-pipe frame count.
+
+### Gaps vs Linux i915
+
+| Gap | Severity | Impact |
+|---|---|---|
+| EDID I2C/DDC stubbed | **Critical** | No real monitor modes — always falls back to synthetic/default |
+| Vblank counter is fake | **Critical** | Page flip has no synchronization — tearing |
+| Display watermarks absent | **Critical** | FIFO underruns → visible glitching on real hardware |
+| No panel power sequencing | **High** | eDP panels won't turn on/off properly on laptops |
+| No hardware cursor | **High** | No visible cursor in DRM mode |
+| No DP AUX channel | **High** | No DisplayPort monitor support |
+ synthetic_edid too short (bug) | **Critical** | EDID validation always fails |
+| No DMC firmware loading | **Medium** | No DC5/DC6 power state for Gen9+ |
+| No HPD pulse detection | **Medium** | Monitor hotplug is crude |
+| No render commands | **Medium** | Ring only does flush — no 2D/3D acceleration |
+| No GGTT PTE invalidation | **Medium** | Stale TLB entries after GGTT updates |
+| Mesa has no Intel driver | **High** | No hardware-accelerated OpenGL/Vulkan |
+
+---
+
+## 4. Boot Process
+
+### Boot Sequence (as configured)
+
+```
+UEFI bootloader
+  → kernel (startup, ACPI, scheme registration)
+  → init (PID 1)
+    → logd
+    → scheme registration (memory, irq, event, pipe, debug, etc.)
+    → numbered services from init.d/:
+      00_*  : base daemons (ipcd, ptyd, randd)
+      02_*  : driver-manager (or legacy pcid-spawner)
+      04_*  : device drivers
+      06_*  : D-Bus, sessiond, seatd
+      08_*  : console/greeter
+```
+
+### Dependency Analysis
+
+The `init` system supports `requires_weak` for service dependencies, but **most services don't declare dependencies**. The boot agent found:
+
+- **`requires_weak`** means "if the dependency exists, wait for it; if not, proceed anyway." This is good for optional services but inadequate for strict ordering.
+- **No explicit `acpid = {}` service** in redbear-full.toml or redbear-device-services.toml — ACPI-dependent drivers may never discover their devices.
+- **`driver-manager`** retries deferred probes, but missing schemes (especially `acpi`) can leave drivers permanently skipped.
+- **Greeter/session path works only** if dbus, seatd, redox-drm, and authd are all present. `redbear-greeterd` waits for Wayland socket, not a stronger compositor readiness signal.
+
+### Gaps
+
+| Gap | Severity | Notes |
+|---|---|---|
+| **Missing acpid service in configs** | **Critical** | No ACPI symbol discovery for i2c-hidd, thermald, driver-manager ACPI path |
+| No dependency declarations | **High** | Services use number-based ordering only |
+| No service readiness signaling | **High** | No sd_notify equivalent; init doesn't gate on daemon.ready() |
+| No filesystem check | **Medium** | No fsck on boot; dirty filesystem mounts anyway |
+| initfs→rootfs transition | **Medium** | No re-evaluation of service readiness after root switch |
+
+---
+
+## 5. Phased Implementation Plan
+
+### Phase 1: Boot Dependency Fix (1–2 weeks)
+
+**Priority: Unblocks everything downstream.**
+
+| # | Task | Files | Complexity |
+|---|------|-------|------------|
+| 1.1 | Add `acpid = {}` to redbear-device-services.toml and redbear-full.toml | `config/redbear-device-services.toml`, `config/redbear-full.toml` | Low |
+| 1.2 | Add `requires=` / `wants=` declarations to init service format | `recipes/core/base/source/init/src/` | Medium |
+| 1.3 | Implement dependency-aware startup: wait for `scheme:<dep>` before starting dependents | `recipes/core/base/source/init/src/` | Medium |
+| 1.4 | Add `provides= scheme:acpi` / `requires= scheme:acpi` to ACPI-dependent services | Service TOML files | Low |
+| 1.5 | Wire ESTALE-retry into i2c-hidd/intel-thc-hidd as fallback (already partial) | `drivers/input/i2c-hidd/`, `intel-thc-hidd/` | Low |
+
+### Phase 2: Intel Display Critical Fixes (3–5 weeks)
+
+**Priority: Highest impact for bare-metal desktop.**
+
+| # | Task | Complexity | Risk | Blocks |
+|---|------|------------|------|--------|
+| 2.1 | Implement I2C master-mode in i2cd | High | Medium | 2.2, 2.7, 3.1 |
+| 2.2 | Implement real EDID via DDC (I2C at 0xA0). Fix synthetic_edid to 128 bytes as fallback | High | Medium | — |
+| 2.3 | Implement hardware vblank (read PIPEFRAME register) | Medium | Low | — |
+| 2.4 | Implement display watermarks (WM_LINETIME, WM levels per pipe) | High | Medium | — |
+| 2.5 | Implement eDP panel power sequencing (PP_ON/OFF/CYCLE) | Medium | Medium | — |
+| 2.6 | Implement hardware cursor (CUR_CTL/CUR_BASE/CUR_POS) | Medium | Low | — |
+| 2.7 | Implement DP AUX channel (I2C-over-AUX for DisplayPort) | High | Medium | Depends on 2.1 |
+
+**Ordering:** EDID (2.2) → vblank (2.3) → watermarks (2.4) → panel (2.5) → cursor (2.6) → DP AUX (2.7)
+
+### Phase 3: Input Stack Completion (2–4 weeks)
+
+**Can parallel with Phase 2 once I2C master-mode (2.1) is done.**
+
+| # | Task | Complexity |
+|---|------|------------|
+| 3.1 | Complete intel-thc-hidd input streaming (replace sleep loop with HID report read) | Medium |
+| 3.2 | Add PS/2 extended protocols (ImPS/2 scroll, Explorer 5-btn, trackpoint) | Medium |
+| 3.3 | Add input device hotplug (dynamic producer registration in inputd) | Medium |
+| 3.4 | Add multitouch protocol (ABS_MT slots, touch report parsing) | Medium |
+| 3.5 | Add pointer acceleration to inputd | Low |
+| 3.6 | Port bounded subset of Linux hid-quirks for device workarounds | Medium |
+
+### Phase 4: AML Interpreter Depth (4–8 weeks)
+
+**Risk gate: scope strictly to _PS0/_PS3/_PRW/_BIF/_BST opcodes first.**
+
+| # | Task | Complexity |
+|---|------|------------|
+| 4.1 | AML While/If-Else/Method-with-locals (bounded, not full spec) | Very High |
+| 4.2 | OperationRegion handlers for PCI config and I2C | High |
+| 4.3 | _PRW (power resources for wake) | Medium |
+| 4.4 | ACPI battery (_BIF/_BST) | Medium |
+| 4.5 | ACPI buttons (LID, power, sleep) | Low |
+| 4.6 | Thermal zone evaluation (_TMP, _ACx, _PSV, _CRT) | Medium |
+
+### Phase 5: Advanced Features (4–8 weeks)
+
+After Phases 2–4 are stable.
+
+| # | Task |
+|---|------|
+| 5.1 | PCI ASPM power management (_OSC, L0s/L1) |
+| 5.2 | PCI hotplug (acpiphp/pciehp) |
+| 5.3 | SRAT/SLIT → NUMA allocator wiring |
+| 5.4 | Display FIFO underrun recovery |
+| 5.5 | HPD pulse detection |
+| 5.6 | I2C bus error recovery (SMBus timeout, multi-controller) |
+
+### Dependency Graph
+
+```
+Phase 1 (boot deps)
+    │
+    ├──→ Phase 2 (Intel display) ──→ Phase 5.4, 5.5
+    │         │
+    │         └──→ 2.1 (I2C master) blocks 2.2, 2.7, 3.1
+    │
+    ├──→ Phase 3 (input) ──→ 3.1 needs I2C (shared with 2.1)
+    │
+    ├──→ Phase 4 (AML) ──→ Phase 5.1, 5.2
+    │         │
+    │         └──→ 4.1 gates 4.3–4.6
+    │
+    └──→ Phase 5 (advanced) ──→ depends on Phases 2, 3, 4
+```
+
+### Effort Estimate (2 developers)
+
+| Phase | Duration | Parallelizable? |
+|-------|----------|-----------------|
+| Phase 1 | 1–2 weeks | Yes (with Phase 2 start) |
+| Phase 2 | 3–5 weeks | Partially (2.1 blocks 2.2–2.7) |
+| Phase 3 | 2–4 weeks | Yes (parallel with Phase 2) |
+| Phase 4 | 4–8 weeks | Partially (4.1 gates rest) |
+| Phase 5 | 4–8 weeks | After Phases 2–4 |
+| **Total** | **14–27 weeks** | ~8–14 months |
+
+### Key Risks
+
+1. **I2C master-mode** is a shared dependency between EDID (2.2), THC input (3.1), and DDC (2.2). Implement it first in i2cd.
+2. **AML interpreter scope creep** — the full AML spec is enormous. Strictly bound the first pass to opcodes needed for _PS0/_PS3/_PRW/_BIF/_BST. Fallback: bounded userspace AML evaluator in acpid.
+3. **Intel watermark programming varies by generation** — start with Gen9 Skylake, then generalize.
+4. **synthetic_edid 112-byte bug** must be fixed IMMEDIATELY — it affects every Intel display attempt.
+5. **Missing acpid service** in configs must be fixed IMMEDIATELY — it blocks all ACPI-dependent device discovery.
+6. **DMAR/IOMMU iterator bug** in `acpid/src/acpi/dmar/mod.rs` causes hangs on real hardware; entire DMAR path is commented out.
+7. **DMI quirk matching is dead** — `redox-driver-sys/quirks/dmi.rs` reads `/scheme/acpi/dmi` but no code provides that scheme.
+
+---
+
+## Appendix A: Linux Reference File Map
+
+From `local/reference/linux-7.1/`:
+
+| Subsystem | Key Files |
+|---|---|
+| HID core | `drivers/hid/hid-core.c`, `hid-input.c`, `hid-quirks.c` |
+| I2C-HID | `drivers/hid/i2c-hid/i2c-hid-core.c`, `i2c-hid-acpi.c` |
+| USB HID | `drivers/hid/usbhid/hid-core.c`, `usbkbd.c`, `usbmouse.c` |
+| Input core | `drivers/input/input.c`, `input-mt.c`, `evdev.c` |
+| PS/2 | `drivers/input/serio/i8042.c`, `libps2.c`, `atkbd.c`, `psmouse-base.c` |
+| I2C core | `drivers/i2c/i2c-core-acpi.c`, `i2c-core-base.c` |
+| i915 | `drivers/gpu/drm/i915/` (6M+ lines) |
+| ACPI | `drivers/acpi/` (full AML interpreter, 15k+ lines) |
+
+## Appendix B: Uncommitted Changes (as of 2026-05-17)
+
+The `bootprocess` branch has 63 changed files including:
+- Kernel ACPI: `slit.rs`, `srat.rs` (NUMA), `msi.rs`, `vector.rs` (MSI/MSI-X), `sleep.rs` + `s3_wakeup.asm` (S3)
+- Kernel: `numa.rs`, `mcs.rs` (MCS lock), context/percpu/event/sync improvements
+- Base patches: `P4-acpi-estale-graceful.patch`, `P4-hwd-estale-graceful.patch`, `P4-ucsid-estale-graceful.patch`
+- Kernel patch: `P21-x2apic-smp-fix.patch`
+- Modified: pcid, driver-manager, thermald, redox-drm, redox-driver-acpi source files