Activate Sv39 using linker-owned frames and prove kernel continuation (Batch 17)

[thanks-cohn]

Motivation

• Advance the memory-management ladder by activating a truthful RV64 Sv39 address space built from the linker-owned physical frames established in Batch 16 and prove the kernel continues executing under translation without introducing U-mode, processes, or unrelated VM machinery.
• Reuse the existing Sv39 entry/index/walker/builder/invalidation modules and the Batch 16 allocator contract so page-table pages are real owned frames rather than synthetic or identity-only mappings.

Description

• Add a small freestanding RealPageOwner adapter that allocates zeroed page-table pages from the existing PhysicalPageFrameAllocator, tracks ownership, and exposes checked read/write/release operations for the generic Sv39 builder to use. (recipes/run-hosted-morphic-runtime/src/freestanding_riscv64.zig)
• Build-time and recipe integration: import Sv39 modules into the freestanding build (build.zig), extend the freestanding linker script with __image_begin, and update the Morphic recipe to document the new active-Sv39 steps. (build.zig, recipes/run-hosted-morphic-runtime/freestanding-riscv64.ld, recipes/run-hosted-morphic-runtime/recipe.json)
• Freestanding activation path: construct real page-table pages from the owned pool, install a bounded set of 4 KiB mappings for the exact ELF continuation footprint, create one explicit non-identity virtual alias to an owned data frame, write satp with Sv39 MODE/ASID=0/root PPN, execute the existing SFENCE.VMA privileged boundary, and record a deterministic evidence frame describing pool bounds, root/intermediate frames, satp before/after, alias sentinel observations, and continuation markers. (recipes/run-hosted-morphic-runtime/src/freestanding_riscv64.zig)
• Add a focused Batch 17 verifier tools/verify-freestanding-riscv64-active-sv39.py that rejection-tests framing/field sets, decodes raw satp, reconciles ELF symbols and pool bounds, validates owned page-table frames, verifies the non-identity alias, reuses Batch 12–16 parsers, and runs two bounded system-QEMU machines for real-machine proof. (tools/verify-freestanding-riscv64-active-sv39.py)
• Update documentation and commands to describe Batch 17 self-test and real-lab commands and produce a Batch 17 report that records mapping policy, evidence fields, Snowball Yield, and explicit non-claims. (COMMANDS.md, docs/reports/AGENTIC_SNOWBALL_BATCH_17.md)
• Update generated agent/recipe indexes and recipe metadata so the new freestanding adapter and verifier are discoverable by repository tooling. (generated index updates)

Testing

• Ran the Batch 16 gate: python3 tools/verify-freestanding-riscv64-physical-memory.py --self-test and python3 tools/verify-freestanding-riscv64-physical-memory.py, which performed two real qemu-system-riscv64 runs and passed the ELF pool, Bare satp, eight owned frames, sentinel read/write, exhaustion/release/double-free/reacquisition, and preserved prior Batch 12–15 evidence. (PASS)
• Ran the Batch 17 verifier python3 tools/verify-freestanding-riscv64-active-sv39.py --self-test and python3 tools/verify-freestanding-riscv64-active-sv39.py, which executed two real qemu-system-riscv64 runs and validated that satp is Sv39 with owned root PPN, four owned page-table pages were used, the 0x80400000 -> owned non-identity alias observed the sentinel, SFENCE.VMA executed, and Morphic output remained identical across hosted/fake/two machines. (PASS)
• Ran focused module tests and recipe checks: zig build test/smoke targets for the reused Sv39 modules and zig build test-recipe-construct-and-verify-sv39-address-space plus physical allocator tests, and they completed successfully (all focused tests passed). (PASS)
• Ran repository maintenance checks: PYTHONDONTWRITEBYTECODE=1 python3 tools/build-agent-index.py, python3 tools/check-command-reference.py --check, node tools/check-port-contracts.js, python3 tools/query-reference.py agent doctor, and python3 tools/developer-command.py validate-repository to regenerate indexes and validate the repository; these verification steps passed after evidence regeneration. (PASS)

All automated validation required by the Batch 17 plan ran to completion in this environment and returned passing results; no new interactive PR creation was performed because remote push/PR tooling was not available in this execution environment.

---

Codex Task

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ede082b945

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Require exactly four page-table frames

If the builder begins consuming any count from one through seven, this verifier still returns PASS, even though the recipe's inherited_guarantees and the Batch 17 command contract specifically claim four page-table pages. A footprint or allocation regression to five pages would therefore be published as successful evidence instead of forcing the contract and resource budget to be reviewed; require count == 4 (and four rows) so this drift fails closed.

AGENTS.md reference: AGENTS.md:L92-L92

Useful? React with 👍 / 👎.