Linux News Digest (Sept 28, 2026): Torvalds Drops Linux 7.3-rc5, the Looming AGENTS.md Anti-AI Manifest, and a Mind-Boggling +22,900,000% Performance Metric
Linux News Digest (Sept 28, 2026): Torvalds Drops Linux 7.3-rc5, the Looming AGENTS.md Anti-AI Manifest, and a Mind-Boggling +22,900,000% Performance Metric
Текст статьи (Body):
My eyes feel like someone rubbed sandpaper over them, my keyboard has a fresh layer of dried energy drink stains, and looking toward the window confirms that the outside world is still garbage. But the open-source wheel keeps spinning. Today is September 24, 2026, and the Linux kernel community is currently dealing with unusual diffstats, existential dread over AI bots, and some absolutely ridiculous performance benchmarks.
If you spent the weekend avoiding sunlight and tuning your local system config, here is the absolute best technical breakdown of what broke and what was fixed in the open-source world over the last 24 hours.
1. Linus Torvalds Releases Linux Kernel 7.3-rc5: "Another Week, Another Large RC"
A few hours ago, Linus Torvalds officially unleashed Linux Kernel 7.3-rc5. If you've been tracking this cycle, things are getting a bit weird. Torvalds openly stated that the diffstat for this release looks highly unusual for this stage of the game. [1]
- The Breakdown: Less than a third of the entire patch consists of hardware drivers, which is almost unheard of. Instead, a massive 25% of the codebase in rc5 is entirely composed of selftests. [1]
- The Culprit: The single largest commit in this window is an ARM64 KVM ITS table test. Torvalds assured the community that while the numbers look skewed, it's just the "same old, same old" core VFS, networking, and DRM fixes underneath. This kernel is still on track to power major upcoming autumn distributions like Ubuntu 26.10. [1]
2. Kernel Developers Consider Adding "AGENTS.md" to Filter Out Raw AI Garbage
The ongoing flood of AI-generated code has finally pushed kernel maintainers to their absolute limit. A serious proposal is currently making the rounds on the Linux Kernel Mailing List (LKML) to introduce a new mandatory file to the source tree: AGENTS.md. [1]
- The Context: Ever since large language models and autonomous AI bug-hunting scripts started spamming the kernel infrastructure with automated CVE reports, human maintainers have been buried under a mountain of low-quality, unverified patches. Canonical literally had to switch Ubuntu to a weekly kernel update cycle just to keep up with the automated AI chaos. [1, 2]
- The Solution: The proposed
AGENTS.mdfile will act as a strict legal and technical guide for developers using AI assistants. It will force strict rules on code verification, meaning if you just "vibe-code" a patch using an LLM without manually auditing the memory states, your commit will be instantly dropped and blacklisted. Maintainers are tired of acting as unpaid QA testers for broken AI scripts. [1]
3. Linux 7.4 Memory Hotplugging Optimization Yields an Absurd +22,904,539% Metric
For the hardcore performance autists who live for optimization benchmarks, a patch queued up for the upcoming Linux 7.4 memory management (MM) subsystem has dropped some jaw-dropping numbers. [1]
- The Scale: An extreme optimization targeting massive multi-node enterprise enterprise servers and heavy cloud workloads has yielded a certified +22,904,539.81% performance win in a specific memory hot-unplugging micro-benchmark.
- The Reality: Before you get too excited, this is a highly specific synthetic metric testing lock contention recovery when freeing colossal pages. You won't get a 22-million-percent speed boost running an emulator or terminal matrix inside your basement setup. However, the patch series also includes more modest, real-world optimizations—such as code changes that allow the kernel to open local files for reading roughly 39% faster across the board. [1, 2]
Quick Hits from the Technical Underground:
- Linux 7.4 Steal Governor Queued: Code has been officially merged into the scheduler core branch to introduce the new "Steal Governor". This mechanism is explicitly built to reduce vCPU preemption costs in virtualized environments. Early benchmarks show a massive 44% speedup in Hackbench and up to a 1.2x performance boost for high-throughput databases like Memcached.
- Budgie 10.10.3 Desktop Environment Released: For the few surviving desktop users who still prefer a traditional floating panel layout, Budgie dropped a maintenance update. It fixes favorites tracking inside the native menu applet and improves the Labwc Wayland bridge infrastructure.
Comments
Post a Comment