[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-linus-torvalds-llm-assisted-folio-patches":3,"topics-all":35,"news-related-04ac054b-bc82-4832-9064-50a2defc33b2":54},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":21,"news_slug":28,"published_at":29,"created_at":30,"modified_at":31,"is_published":32,"publish_type":33,"image_url":14,"view_count":34},"04ac054b-bc82-4832-9064-50a2defc33b2","Linus Torvalds 首批 LLM 辅助内核补丁进入审查","Linux 作者 Linus Torvalds 在维护者峰会上透露,他用大模型完成 10 个针对 folio 查找的补丁,已提交邮件列表审查。补丁识别热 folio 并设 RCU 延迟释放标志,在 RCU 保护下做 folio 数据访问,快路径无需引用计数。读密集基准实现数量级性能提升,但社区已发现竞态条件,仍需修复","在刚刚结束的 Maintainers Summit 上,Linux 内核创始人 Linus Torvalds 当众承认:他用大语言模型写出了一组针对内核长期痛点的补丁,并已经把第一版 10 个补丁递交邮件列表进入社区审查。这是 Linus 第一次正式把 LLM 参与创作的代码推上 LKML。\n\n## 这次修的是什么\n\n补丁聚焦在 folio 的查找路径。folio 是内核里把多个物理页合并管理的数据结构,在文件读、内存映射、I\u002FO 子系统里几乎无处不在;folio 的引用计数与锁竞争一直是性能优化绕不开的热点。\n\nLinus 的思路很直白:识别出「热 folio」(被反复访问的 folio),在它身上设置一个\"延迟释放直到 RCU 周期结束\"的标记。RCU(Read-Copy-Update)是 Linux 内核里一种典型读多写少场景下的同步原语,核心思想是读者几乎无锁、写者通过延迟回收来避免冲突。在 RCU 保护下做 folio 查找这件事内核早就实现了,但这次的改动更进一步——**在 RCU 保护下直接做 folio 数据访问本身**,快路径里连引用计数都不需要拿了。\n\n## 性能数据与已知问题\n\n根据 Torvalds 自述,在一组读操作密集的基准上,**性能达到了\"数量级\"的提升**——也就是 10 倍量级,而不是常见百分之几、几十。这种程度的飞跃在内核的成熟子系统里已经很久没出现过了。\n\n不过,LKML 的评审并不给面子:补丁发布没多久,社区就**已经定位出一处竞态条件**(race condition),还需要进一步修复。这其实正是 Linus 想要——他在峰会上说,用 LLM 写代码,真正的难点不在 AI 自己写得怎么样,而在社区评审能不能像审人类写的代码一样严格把它打回去。\n\n## 它为什么重要\n\n这件事的信号意义比补丁本身更大。过去两年,Linus 多次公开表达对 AI 生成代码的\"不信任\",NetBSD 等社区甚至直接把 AI 生成的代码列为污染源、禁止提交。但这一次,他亲自下场,把自己最擅长的内核子系统的代码交给 LLM 起草,然后走完整的 LKML 评审流程。\n\n换句话说,LLM 写的代码能不能进 mainline,内核社区这次是认真的。Linus 实际上把\"AI 是否能参与内核开发\"这个哲学问题,变成了一个**工程问题**:能不能过评审、过 review、过 bisect、过长期测试。能就合,不能就扔,标准只有一条,不看出处看质量。\n\n从冷启动到 LKML,这条链已经走通。下一个值得关注的节点,大概率是 -rc1 阶段这批补丁会不会被拣进 mm 或者 folio 相关维护者的分支。","https:\u002F\u002Flwn.net\u002FArticles\u002F1100449\u002F","3e3e863f-99c8-4d74-8a88-c528dbd44e19",[11,15,18],{"id":12,"name":13,"slug":13,"description":14,"color":14},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":19,"name":20,"slug":20,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[22],{"id":23,"lang":24,"title":25,"summary":26,"content":27},"873d30aa-1a78-4b98-bd28-ae39266c43dd","en","Linus posts first LLM-assisted kernel patches under review","Linus used an LLM for 10 folio-lookup kernel patches. Hot folios defer release after RCU; data accesses under RCU skip refcounts. Order-of-magnitude speedup.","At the recently concluded Maintainers Summit, Linus Torvalds openly admitted he had used a large language model to draft a patch series tackling a long-standing kernel bottleneck — and has now posted the first 10 patches to the Linux Kernel Mailing List for review. This is the first time Linus has formally submitted LLM-assisted code to LKML.\n\n## What the patches actually fix\n\nThe series targets the folio lookup path. Folios — the kernel data structure that groups multiple physical pages — are used almost everywhere a file is read, mapped, or handed to the I\u002FO subsystem; contention on folio reference counts and locks has long been one of the hottest optimization targets.\n\nThe approach is direct: detect \"hot\" folios (those that are repeatedly accessed) and set a flag that defers their release until after a grace period. RCU (Read-Copy-Update) is the kernel's classic synchronization primitive for read-mostly workloads, built on the idea that readers are nearly lock-free while writers delay reclamation. Folio lookups themselves under RCU protection are nothing new — but this series goes one step further: **the actual folio data accesses themselves happen under RCU**, eliminating the need to grab a reference count on the fast path.\n\n## Performance and the known bug\n\nAccording to Torvalds, on a read-heavy benchmark the changes deliver an **order-of-magnitude speedup** — not the usual few-percent-to-double gain you see in mature subsystems, but a 10x-class jump. That kind of leap has been rare inside the kernel for a long time.\n\nReview on LKML, however, has not been gentle. The patches were barely out the door before the community **identified a race condition** that will need further fixing. Which is, in a sense, exactly what Linus wanted: at the Summit he said the real test of LLM-written code is not how the model writes it, but whether the community's review process is rigorous enough to send it back for work.\n\n## Why it matters\n\nThe signal value here is bigger than the patches themselves. Over the past two years, Linus has been openly skeptical of AI-generated code, and communities like NetBSD have gone so far as to label such submissions as contamination and refuse to accept them. This time Linus stepped up himself, handed a kernel subsystem he understands best to an LLM to draft, and put the result through the full LKML review chain.\n\nIn other words, \"can LLM code land in mainline?\" has been turned from a philosophical question into an **engineering question**: does it pass review, does it bisect, does it survive long-term testing. The criterion is not authorship — it is quality.\n\nThe path from cold start to LKML now exists. The next checkpoint to watch is whether any of these patches get pulled into an mm or folio-related maintainer's tree during the -rc1 merge window.","linus-torvalds-llm-assisted-folio-patches","2026-10-15T02:00:00Z","2026-10-15T05:07:46.183086Z","2026-10-15T05:07:46.183094Z",true,"agent",82,[36,45],{"slug":37,"tag_slug":37,"title_zh":38,"title_en":39,"intro_zh":40,"intro_en":41,"id":42,"is_active":32,"created_at":43,"modified_at":44},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":46,"tag_slug":46,"title_zh":47,"title_en":48,"intro_zh":49,"intro_en":50,"id":51,"is_active":32,"created_at":52,"modified_at":53},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":55},[56,61,66,71,76,81],{"id":57,"title":58,"news_slug":59,"published_at":60},"50687317-89b7-4857-ad66-86e2e251ff9d","阿里开源 OpenGCUA：8 个前沿模型电脑操控平均涨 19.6 分","opengcua-open-computer-use-stack","2026-10-18T13:10:17+00:00",{"id":62,"title":63,"news_slug":64,"published_at":65},"36e4f28f-9176-485c-92b2-35cd0aac04fc","Apple MLX 联合 Meta 把 816B 模型装进 Mac Studio 集群","apple-mlx-muse-spark-mac-studio-thunderbolt-rdma","2026-10-16T03:00:00+00:00",{"id":67,"title":68,"news_slug":69,"published_at":70},"d78ea8a8-bebb-4b94-b340-127eb2874a73","Ovis 全模态嵌入 3B:综合分领先 5.19","ovis-omni-embedding-3b","2026-09-23T21:08:45+00:00",{"id":72,"title":73,"news_slug":74,"published_at":75},"12c67d52-17a2-4df5-8386-35d18ffd221a","JEPA-Anything:一套预测框架打通七个领域,湿实验也给了背书","jepa-anything-orthogonal-predictive-factorization","2026-09-19T23:10:37+00:00",{"id":77,"title":78,"news_slug":79,"published_at":80},"28c41f06-d20f-481c-b133-cd109af3aed1","答对之后停不下来:微软团队揪出在线蒸馏的 EOS 错配元凶","eos-mismatch-opd-length-inflation","2026-09-18T21:09:06+00:00",{"id":82,"title":83,"news_slug":84,"published_at":85},"cb1e799d-d6d7-4ab9-9eaf-bea0aa432b06","Mistral 模型进驻 Firefox:119B 开放权重模型驱动浏览器 AI 助手","mistral-small-4-firefox-smart-window","2026-09-16T17:07:00+00:00"]