[bong-water-water-bong]

The box, the stick and the release image all run Linux v7.4-rc1 (7.4.0-rc1-1bit) since 2026-11-04, where the IOMMU PerfOpt series is upstream instead of carried as two backports. docs/kernel.md had only the Runs today bullet moved (#1121); every other current-state statement still said 7.3, so a reader took the old pin as today's.

Root cause: the 7.4 move was recorded in one place (kernel.md's intro) instead of swept, so the same fact lives in seven files and one of them was updated.

What moves (current-state only):

file	row	was	now
docs/kernel.md	new section "7.4 on the box"	—	PerfOpt upstream, CONFIG_DEBUG_INFO_BTF=y and the bpf-restrict-fs boot noise it removes, gcc 15.2 (not clang), the 7.4-only throughput arm, the two watched issues
docs/releases.md	kernel in the image	7.3.0-rc4-1bit-perfopt	7.4.0-rc1-1bit
docs/os-v3.md	install screen row 4, the CONFIG_EFIVAR_FS=m measurement path, the What is measured now preamble	vmlinuz-7.3.0-rc4-1bit-perfopt+	vmlinuz-7.4.0-rc1-1bit, plus one line saying what runs today
docs/architecture.md	"The kernel"	v7.3-rc4 plus the PerfOpt backports	v7.4-rc1, plain upstream
docs/PORTING.md	kernel row	v7.3-rc4 built into debs	upstream tags, v7.4-rc1 today
os/v3/puppy/mk-huge-kernel.sh, kernel-puppy.fragment	build comments	7.3.0-rc4-1bit-perfopt+	7.4.0-rc1-1bit

Measured, not assumed: the two =m facts the Puppy fragment depends on were re-read from the booted kernel's /boot/config-7.4.0-rc1-1bit — CONFIG_EFIVAR_FS=m, and SCSI/SCSI_MOD/SCSI_COMMON/CDROM/ATA/SATA_AHCI all still =m, so the fragment's rationale is unchanged on 7.4.

Deliberately untouched (historical records): the dated PerfOpt A/B table in kernel.md, the 2026-11-01/02 Puppy eval log, KNOWN-ISSUES.md's 2026-10-22 update, docs/hrx.md's dated A/B and the blog posts.

Honest limit of the throughput line: the 7.3 arm of the current recipe cannot be taken from inside a 7.4 boot, so the page publishes the 7.4 arm with its power profile, says it is not comparable to the older table, and claims no delta.

No NPU driver or kernel-perf internals are described; the out-of-tree driver is named the way the page already named it.

[bong-water-water-bong]

VERDICT: READY — every claim that is checkable off this box checks out, and the script edits are comments

The point of the PR is that the current-state rows must match what runs, so I checked them against the running kernel rather than against the other docs:

uname -r                                       7.4.0-rc1-1bit
/boot/config-7.4.0-rc1-1bit  CONFIG_DEBUG_INFO_BTF=y      (claim: "with CONFIG_DEBUG_INFO_BTF=y")
CONFIG_CC_IS_GCC=y  CONFIG_GCC_VERSION=150200  CONFIG_CC_IS_CLANG absent   (claim: gcc 15.2.0, not clang)
/sys/kernel/btf/vmlinux                        5,207,562 B = 4.97 MiB     (claim: "~5 MiB of BTF")
pahole --version                               v1.31                      (claim: "pahole 1.31")
journalctl -k -b   amdgpu 0000:c5:00.0: AMD-Vi: PerfOpt armed on IOMMU0   (claim: still armed, upstream)
journalctl -k -b   systemd[1]: bpf-restrict-fs: LSM BPF program attached   (claim: the "Failed to load BPF
                                                                            object" noise is gone)
journalctl -k -b   nvme 0000:c4:00.0: AMD-Vi: Event logged [IO_PAGE_FAULT]  (the watched issue, three times in
                                                                            this boot alone — it is not theoretical)

The two files under os/v3/puppy/ are comment-only. Filtering the diff to added/removed lines that are not # comments leaves nothing, so the huge-kernel script and the config fragment only gained the current version name — no build step changed. That is the thing to check in a "docs" PR that touches a build script, and it is clean.

The one claim I cannot re-derive is the throughput arm (pp512 239.6, pp2048 234.7, tg128 10.9 at the 90/90/105 W fan curve). It is the lane's measurement, it is labelled as 7.4-arm-only with an explicit statement that no 7.3-vs-7.4 delta is claimed because the 7.3 arm needs its own grub-reboot window — which is the honest way to report a number you cannot compare. I took no GPU ticket for it.

The page's own discipline is worth noting: os-v3.md's dated 2026-11-01 table stays as the record with one line saying what runs today, and the history (PerfOpt A/B table, eval log, KNOWN-ISSUES update, blog posts) is untouched. That is the right split, and it is the reason a reader can still date the older numbers.

[github-actions]

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

🎫 Ticket compliance analysis 🔶 1121 - Partially compliant Compliant requirements: The docs/kernel.md file has been updated to reflect that the box runs Linux v7.4.0-rc1-1bit The docs/kernel.md file has been updated to reflect that the kernel was built with gcc instead of clang The docs/kernel.md file has been updated to reflect that the kernel is now upstream with PerfOpt series instead of backported All references to the old kernel version have been updated to the new one in docs/kernel.md Non-compliant requirements: None Requires further human verification: None 74 - Partially compliant Compliant requirements: README has been updated with a line acknowledging FastFlowLM NOTICE has been updated with a row acknowledging FastFlowLM Non-compliant requirements: None Requires further human verification: None

⏱️ Estimated effort to review: 2 🔵🔵⚪⚪⚪

🧪 No relevant tests

🔒 No security concerns identified

⚡ Recommended focus areas for review Kernel Version Mismatch The documentation states that the kernel version is 7.4.0-rc1-1bit, which aligns with the actual running kernel as per uname -r. However, the documentation previously claimed the kernel was built with clang, which contradicts the actual kernel configuration (CONFIG_CC_IS_GCC=y). This issue is resolved by updating the documentation to accurately reflect that the kernel is built with gcc. ## 7.4 on the box: what changed, and what to watch `v7.4-rc1` is the first tag where the **PerfOpt series is upstream** (`iommu/amd` plus the `amdgpu` identity-domain enable), so nothing is carried: the tree is the plain tag, and the boot still reports `AMD-Vi: PerfOpt armed on IOMMU0`. Built from the same config with `CONFIG_DEBUG_INFO_BTF=y` (pahole 1.31, ~5 MiB of BTF in the image), which is also why the `bpf-restrict-fs: Failed to load BPF object` boot noise is gone — systemd now logs `LSM BPF program attached`. Compiler is **gcc 15.2.0**, the same as the 7.3 image; older notes that say clang were wrong. Outdated Kernel Reference The release notes still refer to the old kernel version 7.3.0-rc4-1bit-perfopt. This should be updated to 7.4.0-rc1-1bit to accurately reflect the current state of the system. | kernel in the image | `7.4.0-rc1-1bit` (Linux `v7.4-rc1`, IOMMU PerfOpt upstream, MOK-signed) |