toyos-cpuvuln holds T14 and TCG to Linux's readings of them; #556's toolchain-digest issue carried (#672)

Carries what closed pull requests #601 and #556 left worth keeping.

## Sources

- #601, `wt/toyos-s0` at 44eb3c2: the T14's and the TCG model's Linux
captures, in `toyos-t14linux/s0/{t14,tcg}/`.
- #556, `wt/toyos-release` at d30dfff: the three issue files the triage
named.

## What changed, per decision

**Only the Linux readings `toyos-cpuvuln` must agree with are committed,
and a test reads each one.**
- `toyos-cpuvuln/fixtures/t14/cpuid.txt`, `msr.txt` and `cpuinfo.txt`
are read by `the_t14s_facts_are_linuxs_reading_of_it`. It reads them the
way `Facts`' docs say Linux does:
  - a leaf past its range's highest reads as 0;
- `ARCH_CAPABILITIES` counts only where CPUID.(7,0):EDX[29] enumerates
it;
  - `MCU_OPT_CTRL` counts only where `GDS_CTRL` does.
- It also holds:
  - `Ident::new` to cpuinfo's family, model and stepping;
  - the microcode revision to both cpuinfo and MSR 0x8B;
  - `decide(&T14)`'s `vmx` to cpuinfo's `vmx` flag;
  - its `spec_ctrl` to MSR 0x48.
- Every MSR and cpuinfo field the test reads must hold one value on
every CPU.
- `toyos-cpuvuln/fixtures/tcg/console.txt` is read by
`the_tcg_models_signature_gives_its_lines`. It takes its lines from the
console's vulnerabilities lines. `fixtures/tcg.txt` was a second copy of
those lines, so it goes.
- The comparison found no disagreement.
- Deleted: `t14/`'s `cmdline`, `config-hardening`, `config-sha256`,
`dmi`, `kernel-log`, `mmap_rnd_bits`, `packages`, `uname` and `version`,
and `tcg/qemu-version`. No test reads them, and none is a reading the
classification must agree with. #601's head keeps them at 44eb3c2.
- The four kept files are #601's blobs at 44eb3c2: `git hash-object`
against `git rev-parse 44eb3c2:toyos-t14linux/s0/<path>` finds 0
mismatches. None holds a NUL, so `src/sourcegate.rs` declares none.

**The TCG fixture is held to the CPU its console names, not to the
gate.** The capture records the `-cpu` only in `tcg/SOURCE`'s command
line, and `qemu-version.txt` was QEMU's reading, not Linux's. So the
test holds the fixture to neither a QEMU pin nor `src/arch.rs`'s model,
and the renamed issue no longer calls it the PR gate's model. If the
`-cpu` or the QEMU version moves, the fixture stays Linux's verdict on
the CPU it names.

**Each `SOURCE` keeps the provenance beside its bytes.** The capture's
QEMU version and times are in 3854054's commit message.

**The Linux-readings issue is renamed** to
`issues/hardware/linuxs-readings-of-the-t14-and-the-tcg-model-lack-reads-owed-before-the-t14s-wipe.md`,
and its heading changes to match. Its four path citations move, and `git
grep HEAD` finds the old slug nowhere.
- The body names the fixtures and the two tests that hold `T14` and
`TCG` to them.
- The exit wants each owed reading committed with the test that reads
it, and `T14`'s `feat_ctl` held to 0x3A.
- No test reads a config hash any more, so the mutation is now a one-bit
change to an added reading.
- The owner is the orchestrator, which holds the T14.
`issues/filesystem/a-cached-read-from-a-file-server-is-measured-only-under-tcg.md`
names its owner the same way.

**#556's loader defect is deleted.** Stage 2 of
`issues/boot-media/the-loader-does-only-what-must-precede-the-handover.md`
already owns its fix (the bullet at :86) and its exit (:100-101). Stage
2 takes the one fact it lacked: every stick written from one image
carries the same unique GUIDs, because `src/image.rs`'s
`create_boot_image` draws them once per image (:240-241).

**`issues/build/the-toolchain-install-unpacks-an-asset-no-digest-vouches-for.md`
is carried, and still true at the merged head.**
- `release::install` (`src/release.rs:107`) checks nothing of the
tarball's bytes.
- The tag hashes `trees()`.
- The `toyos-sysroot-witness` that `check_installed_toolchain` holds an
install to is a file inside the tarball.
- `nightly.yml`'s `build` job (:54) has `contents: write` (:58) and runs
`--ci toolchain` with `GH_TOKEN` (:73-74). The bootstrap it starts
inherits the token.
- #667's `nightly.yml` change moves only the `host` job, and adds one
step to `portability-macos`.
- The repository's default workflow token is read-only: `gh api
repos/ToyOSOrg/ToyOS/actions/permissions/workflow` gives
`"default_workflow_permissions":"read"`.

**`issues/kernel/the-kernel-dies-at-boot-on-the-edk2-firmware-qemu-ships.md`
is dropped, because it is no longer true.**
- #563 (51fe14c) made the kernel boot on Homebrew QEMU 11.1.1's edk2,
and its merge message names this panic as what it fixed.
- Since dfe61e5, every harness boot takes its firmware from
`src/firmware.rs`.
- The issue's exit rests on `release_command_boots` and
`src/redlist.rs`, and both are gone from main.

#556's release-badge issue is not carried, as the triage said.

## Negative controls

For each mutation:
1. `sed` applied it, and a check confirmed the expected lines changed.
2. The tree built, and `cargo test -p toyos-cpuvuln` ran.
3. The file was restored from HEAD.

Every applied mutation built and ran red. `git status --porcelain` was
empty afterwards.

| mutation | red at | exit |
|---|---|---|
| `t14/msr.txt:17` `0a005c6b`→`0a005c6a` |
`the_t14s_facts_are_linuxs_reading_of_it`: `MSR 0x10a differs between
CPUs` | 101 |
| `t14/cpuid.txt:4` leaf 7.0 EDX `fc100710`→`fc100711` | the same test,
`cpuid_7_0_edx` | 101 |
| `t14/cpuinfo.txt` `siblings: 8`→`4` on every CPU | the same test,
`smt` | 101 |
| `t14/cpuinfo.txt` `vmx` removed from every `flags` line | the same
test, `decide`'s `vmx` | 101 |
| `t14/msr.txt` 0x48 `1`→`0` on every CPU | the same test, `spec_ctrl` |
101 |
| `t14/cpuinfo.txt` `microcode: 0xbe`→`0xbf` on every CPU | the same
test, microcode | 101 |
| `tcg/console.txt` `model: 0x6b`→`0x6c` |
`the_tcg_models_signature_gives_its_lines`, the identity | 101 |
| `tcg/console.txt` spectre_v2 `STIBP: disabled`→`conditional` | the
same test, `assert_capture`'s `spectre_v2` | 101 |

The oracle is Linux at `Ubuntu-6.8.0-142.142` on the T14 and on the TCG
model, as #601's capture read it.

## Gates at 10e2b39

This head is the branch with `origin/main` merged at 0edf266.

- `cargo run -- --ci host`: EXIT=0, `Host: 56 step(s), all green`.
- `toyos-cpuvuln`'s 53 tests ran in the host-workspace step, both
readers among them.
  - Clippy, with warnings denied, is clean.
- The licence step says `6 exception(s) stand, and nothing else is
refused`.
- Net (`git diff --shortstat origin/main...HEAD`): 16 files, +880 −61.
  - Production: 0.
  - Tests: +125 −15.
  - Captured data: +665 −17.
  - SOURCE files: +27.
  - Issues: +63 −29.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

https://claude.ai/code/session_016t9wjdQkB8SH7bmfUoiy6L

Author: Japabu

issues/boot-media/the-loader-does-only-what-must-precede-the-handover.md

@@ -86,7 +86,9 @@ Each stage lands on its own, in this order.
    - Every volume the loader opens (the slot's FAT partition, the log
      partition and the attempts file on it) is found on the boot disk, where
      exactly one match is taken. `loaderlog::volume_handle`'s machine-wide
-     first match goes.
+     first match goes. Every stick written from one image carries the same
+     unique GUIDs: `src/image.rs`'s `create_boot_image` draws them once per
+     image.
    - A pass asks firmware once: one `LoadedImage` open, one device-path walk,
      one `Disk::open` and slot-table read, and one file reader.
      `load_file_bytes`, `MAX_ESP_FILE` and the unsound `alloc_uninit` go.

issues/build/the-toolchain-install-unpacks-an-asset-no-digest-vouches-for.md

@@ -0,0 +1,25 @@
+---
+status: open
+kind: tooling
+opened: 2026-09-28
+---
+
+# The toolchain install unpacks an asset no digest vouches for
+
+`release::install` (`src/release.rs`) downloads the `toyos-toolchain.tar.zst`
+asset of the release its tree's tag names, unpacks it into `rust/build` and
+links it as rustup's `toyos`, and checks nothing about its bytes: the tag is
+the hash of the tarball's inputs (`trees`), not of the tarball, and the
+`toyos-sysroot-witness` an installed toolchain is held to comes inside it.
+Every guest job compiles and boots with what it installs.
+
+A job holding this repository's `contents: write` token can replace a
+release asset, so any code such a job runs can replace the toolchain every
+later job installs. The nightly's `build` job holds that token while it
+bootstraps the toolchain.
+
+Owner: the release module (`src/release.rs`).
+
+**Exit condition.** `install` unpacks only an asset whose SHA-256 is one that
+no job holding a write token can rewrite — committed to the tree it installs
+for — and refuses any other by name.

issues/hardware/linuxs-readings-of-the-t14-and-the-tcg-model-are-not-committed.md

@@ -0,0 +1,31 @@
+---
+status: open
+kind: tooling
+opened: 2026-09-29
+---
+
+# Linux's readings of the T14 and the TCG model lack reads owed before the T14's wipe
+
+The floor of
+`issues/kernel/the-kernel-is-at-least-as-secure-as-linux-on-every-machine-toyos-supports.md`
+is tag `Ubuntu-6.8.0-142.142` (53e5d07aac028a1523ab0b115f079d6d1bc831ef) of
+`https://git.launchpad.net/~ubuntu-kernel/ubuntu/+source/linux/+git/noble`,
+config sha256 3b8533dd9d235ca634ac58f82c5ce1ee35f12ef620693e17033184d2c9ca5890,
+and what it reads on the T14 and on the TCG model are the fixtures
+of `issues/kernel/a-pure-function-decides-a-cpus-speculation-mitigations-as-linux-does.md`,
+`toyos-cpuvuln/fixtures/`'s `t14.txt`, `t14/` and `tcg/`;
+`the_t14s_facts_are_linuxs_reading_of_it` and
+`the_tcg_models_signature_gives_its_lines` hold `T14` and `TCG` to them.
+Ubuntu leaves the T14 only after this issue and
+`issues/hardware/no-program-measures-toyos-against-linux-on-one-machine.md`
+close.
+
+**Exit**: before the wipe the T14's readings add CPUID 5, 0x19 and
+0x80000001, MSR 0xCF, MSR 0x3A (`rdmsr -a 0x3a`, IA32_FEAT_CTL), the
+split-lock line, and the config's `X86_KERNEL_IBT` and
+`X86_INTEL_MEMORY_PROTECTION_KEYS`, each committed with the test that reads
+it, and `the_t14s_facts_are_linuxs_reading_of_it` holds `T14`'s `feat_ctl`
+to 0x3A's. **Mutation**: an added reading off by one bit reds the test that
+reads it. **Oracle**: that Linux.
+
+Owner: the orchestrator, which holds the T14 the exit runs on.

issues/hardware/linuxs-readings-of-the-t14-and-the-tcg-model-lack-reads-owed-before-the-t14s-wipe.md

@@ -12,7 +12,7 @@ selects, built by the kernel and by a host test. Pull request #602 holds it.
 
 **Exit**: each committed fixture's facts give its lines: the T14's and the TCG
 model's from
-`issues/hardware/linuxs-readings-of-the-t14-and-the-tcg-model-are-not-committed.md`,
+`issues/hardware/linuxs-readings-of-the-t14-and-the-tcg-model-lack-reads-owed-before-the-t14s-wipe.md`,
 and each nightly EPYC guest's once its runner is captured. **Mutation**: `GDS`
 deleted from `cpu_vuln_blacklist`'s TIGERLAKE_L row reds the T14's fixture,
 and `SRSO` deleted from its family 0x19 row reds a family-0x19 EPYC guest's.

issues/kernel/a-pure-function-decides-a-cpus-speculation-mitigations-as-linux-does.md

@@ -9,7 +9,7 @@ opened: 2026-09-29
 **Exit**: each proving machine's ToyOS boot prints, for each vulnerabilities
 file, the line Linux at `Ubuntu-6.8.0-142.142` printed on that machine: the
 T14's from
-`issues/hardware/linuxs-readings-of-the-t14-and-the-tcg-model-are-not-committed.md`,
+`issues/hardware/linuxs-readings-of-the-t14-and-the-tcg-model-lack-reads-owed-before-the-t14s-wipe.md`,
 each EPYC guest's from its runner's capture. It omits only clauses about what
 ToyOS does not run: `spectre_v1`'s swapgs barriers while it runs no `swapgs`,
 the KVM and VM-exit clauses since it runs no guest, and `IBRS_FW` since the

issues/kernel/each-boot-prints-the-vulnerabilities-lines-linux-prints.md

@@ -14,7 +14,7 @@ probe is a `boot-actuators` arm or a `test-actuators` `SYS_DEBUG` action.
 
 **Exit**: every issue below is closed, in the order listed.
 
-- `issues/hardware/linuxs-readings-of-the-t14-and-the-tcg-model-are-not-committed.md`
+- `issues/hardware/linuxs-readings-of-the-t14-and-the-tcg-model-lack-reads-owed-before-the-t14s-wipe.md`
 - `issues/kernel/a-pure-function-decides-a-cpus-speculation-mitigations-as-linux-does.md`
 - `issues/kernel/spec-ctrl-and-gds-stay-as-firmware-left-them.md`
 - `issues/kernel/no-program-runs-with-speculative-store-bypass-disabled.md`

issues/kernel/the-kernel-is-at-least-as-secure-as-linux-on-every-machine-toyos-supports.md

@@ -21,6 +21,6 @@ T14's `CONFIG_X86_INTEL_TSX_MODE_OFF=y`, runs `tsx_init`
 `issues/kernel/a-pure-function-decides-a-cpus-speculation-mitigations-as-linux-does.md`
 carries `tsx_init`'s decision and every CPU applies it; on the T14, ToyOS's
 reads of 0x122, 0x10F and 0x123 equal the Linux reads of
-`issues/hardware/linuxs-readings-of-the-t14-and-the-tcg-model-are-not-committed.md`
+`issues/hardware/linuxs-readings-of-the-t14-and-the-tcg-model-lack-reads-owed-before-the-t14s-wipe.md`
 where each exists, and dropping the write leaves 0x122 or 0x10F at firmware's
 value and reds it.

issues/kernel/tsx-stays-as-firmware-left-it.md

@@ -0,0 +1,13 @@
+Linux's readings of the ThinkPad T14 this repository is tested on, under that
+machine's Ubuntu, read as root on the T14 by `toyos-t14linux/capture.sh` at
+commit 44eb3c2e (pull request #601), and it read
+CPUID and the MSRs through the running kernel's /dev/cpu/*/cpuid and
+/dev/cpu/*/msr, so an MSR holds what that kernel left in it.
+
+  vulnerabilities.txt   the run's `grep .` over the vulnerabilities directory,
+                        committed as ../t14.txt
+  cpuid.txt             CPU 0's leaves, one per line:
+                        leaf subleaf eax ebx ecx edx
+  msr.txt               one per line: msr cpu value; 0x122 and 0x10F are absent
+                        because ARCH_CAPABILITIES bit 7 and CPUID.(7,0):EDX
+                        bits 11 and 13 do not enumerate them

toyos-cpuvuln/fixtures/t14/SOURCE

@@ -0,0 +1,12 @@
+00000000 00000000 0000001b 756e6547 6c65746e 49656e69
+00000001 00000000 000806c1 00100800 7ffafbbf bfebfbff
+00000006 00000000 0017eff7 00000002 00000009 00000000
+00000007 00000000 00000002 f3bfa7eb 18c05fde fc100710
+00000007 00000001 00000000 00000000 00000000 00040000
+00000007 00000002 00000000 00000000 00000000 00000001
+0000000d 00000000 000002e7 00000a88 00000a88 00000000
+0000000d 00000001 0000000f 00000998 00003900 00000000
+00000014 00000000 00000001 0000004f 00000007 00000000
+80000000 00000000 80000008 00000000 00000000 00000000
+80000008 00000000 00003027 00000000 00000000 00000000
+80000021 00000000 00000000 00000000 00000000 00000000