Latest

Technology

Stoakes steps back from Linux kernel patch review

The kernel's memory maintainer stops reviewing general patches: half of what reaches him is AI-generated series and he can no longer keep up with the volume.

Translated from Spanish by an automated system. The Spanish original prevails. Read the Spanish original

Stoakes steps back from Linux kernel patch review
Stoakes steps back from Linux kernel patch review

Four patches were enough for Lorenzo Stoakes to shed ballast: the Arm engineer and one of the driving forces behind Linux memory development announced on November 10, 2026 that he is stepping back from reviewing what he does not maintain. The reason is neither a technical conflict nor a quiet exit: it is that half of what lands in his inbox consists of series generated by language models and he no longer has the hours to read them.

The announcement fits in a single email paragraph, but it describes reviewer saturation. The four patches Stoakes sent to the kernel mailing list touch only two files, MAINTAINERS and CREDITS: they remove his name as a reviewer for core memory, reclaim and Rust. None of that fixes a vulnerability or improves a driver. It records that the kernel's bottleneck is no longer in writing code.

The practical consequence is uncomfortable for anyone who installs distributions. The kernel receives more proposals than it can review with the current template, and every reviewer who steps down thins the filter that patches ending up in Debian, Fedora or Ubuntu must pass through. The four submissions break nothing on their own. They point to where an infrastructure that underpins much of the software we use daily is getting stuck.

Four patches and three resignations

The text accompanying the series is explicit and names no one in particular. "Given the current volume of the kernel, resulting from AI-generated series, I unfortunately do not have the bandwidth to offer reviews on a general basis, except in the areas I maintain," Stoakes wrote. The rest of the message is administrative: it splits the patches by area so that David Hildenbrand, Andrew Morton and the duo of Alice Ryhl and Danilo Krummrich take on what used to land in his inbox.

The resignation has three layers. The first is core memory review, the heart of the memory management system. The second covers reclaim, where it is decided which pages are recovered when memory runs short. The third is Rust review, a young area that depends on very few people. Stoakes keeps his four maintainerships: memory mapping, its locking, the madvise calls and reverse mapping.

The detail of the acknowledgements helps gauge what is lost. Mike Rapoport, Zi Yan, Lance Yang, David Hildenbrand, Kiryl Shutsemau and Danilo Krummrich signed off on accepting the series, and Stoakes himself recorded that his succession on transparent pages is left in the hands of Zi Yan. It is an orderly withdrawal, not an emergency exit from a specific function: that of reading and filtering what others write.

The real cost is in reading, not writing

A kernel patch is rarely a brilliant idea poorly documented. It is a proposal that someone must read in full, check against hardware behavior, discuss on the list and read again once revised. That work barely leaves a public trace when it goes well, and it consumes the time of a small group of people who also maintain their own code.

The arithmetic is easy to picture. Each merge window concentrates thousands of changes and the 7.4 cycle opened the largest merge window in its history, a record that means more email for the same reviewers. If a growing share of those series arrives generated by automated tools, the human filter becomes more expensive per useful patch.

Stoakes described that curve in the same conversation, without sugar-coating it. First he saw junk occasionally, then at least once a week, then every few days, and now "pretty much half of what I see, if not more." The phrase is not a generic complaint about artificial intelligence. It is an informal measurement of the relationship between incoming volume and review capacity, made by someone who has been reading that email since 2014.

The problem has a side effect that almost no one looks at. When reviewing costs more, it is done worse or done less, and both lead to the same place: patches that go in with fewer eyes on them. For a subsystem like memory management, where a synchronization error does not stay confined to one program, that difference shows up months later and on other people's machines.

Rust is left with fewer reviewers

Of the three resignations, the one that will make the most noise in 2027 is Rust. The language came to the kernel to reduce memory errors and its adoption depends on having people capable of evaluating new abstractions, not just translating drivers. Stoakes was one of those people, and his departure leaves the area with the same problem as the rest of the kernel: few eyes for a lot of code that promises safety but demands expert oversight.

The contrast with the official narrative is obvious. Rust rewrites are presented as a quality improvement at no cost, and the part of the cost that does not appear in the presentations is review. A language change in a desktop client can be validated with product tests and measured memory. A kernel abstraction that others will use for a decade needs reviewers who understand the ownership model and the system's memory model at the same time.

That is why Stoakes's resignation matters more than his patch volume. It is not just any contributor leaving: it is someone who could say no to a badly conceived Rust series. Young areas tolerate the loss of a reviewer worst, because there is no second row waiting.

A model that starts to fail at the filter

The kernel's response to this kind of overload has always been the same: spread it out. Co-maintainers are created, trees are split, review platforms are delegated to. It works as long as there are people willing to take on the load, and it stops working when volume grows faster than the number of people capable of evaluating it.

The numbers from the current window explain the urgency. Artificial intelligence is already used to find bugs at scale and that work also generates email: a single model detected 948 privilege-escalation bugs in Linux and each of those findings needs a patch, a report and someone to read it. The result is a circle in which tools accelerate the production of changes and the bottleneck narrows at the other end of the chain.

There is an optimistic reading and it cannot be dismissed. If review is automated in its mechanical part (compilation, testing, style checking, regression detection), the human reviewer can devote themselves to what only a person can judge: whether the change makes sense, whether it breaks an implicit contract, whether the author understood the subsystem. The question is whether that division of labor arrives before those who currently hold the filter leave.

A new development policy

Stoakes's series has already been accepted and there is no going back, but there are signals worth following. The first is who signs off on core memory reviews in December and January: if new names appear regularly, the redistribution worked; if the area goes silent, the problem will have moved from one inbox to another.

The second is the debate over long-term support, where every extra version maintained competes for the same review hours. The third directly affects anyone using Linux on the desktop and on servers: distributions that adopt young kernels depend on that filter continuing to exist. A kernel that merges faster than it reviews is not a more agile kernel, it is a kernel with less margin.

Stoakes summed it up with a phrase that is a diagnosis. He said the premise is probably false and he will end up reviewing things anyway, but that he wants to be less prone to being dragged into reviewing what he does not maintain. That "probably" is the important part. The kernel has not lost a reviewer to a technical decision; it has lost one to exhaustion of the review model, and that exhaustion cannot be fixed with a patch.

José Maldonado

Author

José Maldonado

Activist and blogger on technology, free software and blockchain.

The latest
What is happening with stablecoins in Brazil and why the Central Bank is turning off the tap

Keep reading