[bong-water-water-bong]

docs/kernel.md opens with two present-tense claims about this box. Both are measurable, and both were
wrong; every dated 7.3 citation on this page (the A/B table, the huge-kernel rows, the historical measurements)
is left exactly as it is.

1. The box does not run 7.3 any more

- **Runs today:** 7.3.0-rc4-1bit-perfopt — Linux v7.3-rc4 plus two backports …

Measured: uname -r → 7.4.0-rc1-1bit, and the box has booted it since 2026-11-04 05:45 (six boots,
/boot/vmlinuz-7.4.0-rc1-1bit). PerfOpt is present in it, measured rather than assumed — this boot prints
amdgpu 0000:c5:00.0: AMD-Vi: PerfOpt armed on IOMMU0.

2. Neither image was built with clang — this is the more interesting one

The page asserts built with clang (LLVM=1) in the intro and CONFIG_CC_IS_CLANG=y in the config of record
bullet, and the recipe below is export LLVM=1; make LLVM=1 … bindeb-pkg. The running kernel contradicts it
three independent ways:

source	says
/boot/config-$(uname -r) — the file the page itself names as the config of record	CONFIG_CC_IS_GCC=y, CONFIG_GCC_VERSION=150200, CONFIG_CLANG_VERSION=0
/proc/config.gz (embedded in the running image)	CONFIG_CC_IS_GCC=y
the kernel's own version line	Linux version 7.4.0-rc1-1bit (bcloud@strixhalo) (gcc (Ubuntu 15.2.0-16ubuntu1) 15.2.0, GNU ld …)

The 7.3.0-rc4-1bit-perfopt+ image it replaced reports the same three ways, so this is not a regression introduced
by 7.4 — the kernels this box has booted were gcc builds while the page documents a clang build. That is
exactly the failure mode the page already warns about ("a gcc-only host silently builds the wrong compiler unless
LLVM=1 is exported"), so the warning is right and the images are the evidence for it.

I am not deciding which side should change. Rebuilding with clang and updating the recipe to record gcc are both
plausible answers, they have different consequences (the 7.3 A/B in this page was taken on a clang-built kernel),
and it is the kernel lane's call. This PR makes the page stop asserting something its own config-of-record file
contradicts, and states the measurement so that lane can act on it.

Verification

tools/copyright.py --check rc 0; site build 55 pages rc 0; rendered kernel.html carries 7.4.0-rc1-1bit,
CONFIG_CC_IS_GCC=y, GCC_VERSION=150200 and the PerfOpt line, one <h1>. Docs only, one file, no behaviour
change. Also checked and deliberately not touched: docs/hybrid.md:211 and docs/hrx.md:698 name the 7.3
kernel as the measurement context of a past verdict ("the hardware/firmware pair this verdict is for", "the
2026-10-31 A/B"), which is history and stays correct.

[bong-water-water-bong]

#1121 ready for review at c73115c — docs only, one file (+20/−7), tools/copyright.py --check rc 0, site 55 pages rc 0.

Two measurable present-tense claims on the kernel runbook's first bullets:

1. Runs today: 7.3.0-rc4-1bit-perfopt is stale — uname -r is 7.4.0-rc1-1bit, booted since 2026-11-04 05:45 (six boots), and PerfOpt is present in it: this boot prints AMD-Vi: PerfOpt armed on IOMMU0.
2. CONFIG_CC_IS_CLANG=y / "built with clang (LLVM=1)" is contradicted by the file the page names as the config of record — /boot/config-$(uname -r) says CONFIG_CC_IS_GCC=y, CONFIG_GCC_VERSION=150200, CONFIG_CLANG_VERSION=0; /proc/config.gz says the same; and so does the kernel's own version line (gcc (Ubuntu 15.2.0-16ubuntu1) 15.2.0). The 7.3 image it replaced reports the same three ways, so it is not a 7.4 regression.

That second one is the page's own warning coming true ("a gcc-only host silently builds the wrong compiler unless LLVM=1 is exported"), so the PR records the measurement and stops — whether to rebuild with clang or to correct the recipe to gcc is the kernel lane's call, and it has consequences for the 7.3 A/B measured on a clang-built kernel.

Deliberately untouched: every dated 7.3 citation (docs/hybrid.md:211, docs/hrx.md:698, the Puppy eval rows) names the kernel as the measurement context of a past verdict and is history.

BEHIND is benign (branch is origin/main + 1 commit). Note one process mistake I made and corrected: the first push of this commit went to #1112's branch by accident; that branch is force-pushed back to b4069ad and this PR now carries only c73115c (+20/−7, 1 file).

[bong-water-water-bong]

VERDICT: READY at c73115c4 — docs only, and I checked its two factual claims against the running kernel rather than reading them.

$ uname -r                                          -> 7.4.0-rc1-1bit
$ (zcat /proc/config.gz) | grep CONFIG_CC_IS_ /CONFIG_GCC_VERSION /CONFIG_CLANG_VERSION
  CONFIG_CC_IS_GCC=y
  CONFIG_GCC_VERSION=150200
  CONFIG_CLANG_VERSION=0
$ ls /proc/config.gz                                 -> present

So both claims hold: this box runs 7.4.0-rc1-1bit, and the page's own correction is right — the kernel was not built with LLVM=1/clang; it is gcc 15.2.0, with CONFIG_CLANG_VERSION=0. A page that tells the next person to rebuild with LLVM=1 when neither image was built that way is exactly the kind of documentation that costs a day, and the doc now says so in as many words ("the LLVM=1 recipe below is not what built the kernels this box has booted").

That is the right shape of correction: it keeps the 7.3 A/B measurement as the motivation for the backports (it is still evidence) and marks it as history, while stating what the box actually has.

Nothing to run — one doc file, +20/−7.

[github-actions]

Preparing review...