Skip to content
audiotale.net
Sign in
NewsNews

Linux 7.4-rc1: record 19,671-commit merge window, Torvalds on AI fixes

Torvalds says AI-found fixes now arriving in the merge window, not late in the rc weeks, are a 'much healthier model'; the tree grew by about 455,000 lines of code.

audiotale.net team

· 6 min read

Written by AI from the sources linked below. How we use AI

On this page
  1. Key points
  2. What happened
  3. Background
  4. The numbers
  5. What's in it
  6. What it means
  7. What we don't know yet
  8. What to watch next
  9. Sources

Linus Torvalds released Linux 7.4-rc1 on 1 November 2026, closing the largest merge window the kernel has seen: 19,671 commits in two weeks, according to a tally cited by Hardware Busters from Phoronix's counts. Torvalds said he was "much happier" that the flood of AI-found fixes is arriving now, in the merge window, than as a stream of late fixes during the release-candidate weeks. The kernel.org front page lists 7.4-rc1 as the current mainline release dated 2026-11-01.

Key points

  • Linux 7.4-rc1 is dated 1 November 2026 on kernel.org; Linux 7.3 was released on 18 October 2026 according to Linux Compatible.
  • The merge window brought in 19,671 commits, about 21% more than Linux 7.3's 16,267 (Hardware Busters, citing Phoronix's tally).
  • Torvalds called the release "Large, and full of fixes for issues found by AI", and said AI-suggested patches "often get rewritten" (quoted by The Register).
  • A third of the changes are AMD GPU register offsets and bit masks, so the raw commit count overstates how much AI work is in the release.
  • Many kernel watchers expect 7.4 to be this year's long-term support (LTS) kernel, but kernel.org has not designated it, and Torvalds only said "people are expecting" it.

What happened

Every Linux release starts with a merge window of roughly two weeks, in which subsystem maintainers send Torvalds their prepared changes. At the end he publishes the first release candidate, rc1, and from then on only fixes are supposed to go in. That process is described in the kernel's own documentation (docs.kernel.org).

For 7.4 the window was unusually busy. The Register, quoting Torvalds' weekly state-of-the-kernel mail, reports that it contained "almost 20,000 commits compared to the usual ~15,000". Hardware Busters gives the precise figure as 19,671, attributed to Phoronix's tally. Torvalds wrote: "This is a BIG merge window". He offered a reason, "the by now obvious one", and then said he was taking it as a good sign.

His point is about timing, not volume. In recent cycles, AI-assisted bug finding has produced many fixes that arrived late, during the -rc weeks when the tree is supposed to be calming down. Torvalds said he is "much happier with the flood of AI triggered work going into the merge window than it showing up mainly as extra fixes during the -rc period." He hopes the large count "is a sign that the AI fixes are becoming more part of the 'normal development' than a flood of fixes that go into the whole stabilization phase", and concluded: "I think that's a much healthier model."

He also hedged. Because "people are expecting 7.4 to be the next LTS [long-term support] release, which is another independent reason why some releases tend larger than others", he said it "may not just be the flood of AI patches." Maintainers tend to push more into a release they know will be supported for years.

On quality he was less than fully convinced. He described the release as "Large, and full of fixes for issues found by AI (and more and more the patches have been suggested by AI too, although they often get rewritten)."

We could not open Torvalds' original mail (the kernel mailing list archive was blocking automated access when we tried), so the quotes above are as The Register reports them. The release itself is confirmed by kernel.org.

Background

Merge windows have grown. Hardware Busters lists the 7.x series as: 7.0 with 12,468 commits, 7.1 with 13,963, 7.2 with 14,395, 7.3 with 16,267 and now 7.4 with 19,671. For comparison it gives 15,095 for Linux 4.9 and 16,203 for 6.7, and says most windows in the past 15 years brought roughly 12,000 to 14,000. It also reports that Linux 7.3 was the first kernel to pass 20,000 commits from one stable tag to the next, finishing at 20,260.

Commits merged in each Linux merge window, 7.0 to 7.4commits
Commits merged in each Linux merge window, 7.0 to 7.4 (commits)
commits
7.012468
7.113963
7.214395
7.316267
7.419671

Source: Hardware Busters, citing Phoronix counts

The growth tracks a wider change in how bugs are found. Hardware Busters says AI tools have become very good at finding kernel bugs and that many reports now arrive with a proposed fix. It also notes a batch of OverlayFS fixes credited to AI-assisted code review and flagged for stable backport. We cover the security side of that trend in our piece on how an Anthropic model reached a Linux kernel TCP use-after-free.

The numbers

  • Size of the tree: Hardware Busters, using cloc counts, puts the v7.3 tree at 41,019,555 lines, of which 30,962,306 are code. Before rc1 the tree stood at 41,605,120 lines, 31,417,938 of them code, which is roughly 455,000 more lines of code in two weeks. It projects the tree will pass 42 million lines around the 7.5 cycle.
  • AMD's share: The Register says AMD GPU register offsets and bit masks make up a third of the changes. These are large, machine-generated header files, which explains much of the count and is why Torvalds added "Just ignore the noise."
  • Release timing: Linux Compatible expects a stable release in late December, with a possible slip to mid-January 2027 because Torvalds spent part of the merge window on holiday. Hardware Busters also says the end of December is likely. Neither is official; the usual pattern is seven or eight release candidates after rc1.

What's in it

The Register says Phoronix rates the main inclusions as improved support for AMD, Apple and Google processors, security improvements for virtual machines, Raspberry Pi 10-inch monitor support and more work on Qualcomm's Snapdragon X2. Linux Compatible, writing before the window closed, lists several specific items:

  • A boot-time fix from a Google engineer that replaces a linear search in the early-boot memory allocator with a binary search. It says retrieval of memory in Kernel Handover drops from about 268 seconds to about 50 milliseconds.
  • A workqueue flushing rework by Tejun Heo that it says lifts idle per-CPU workqueue throughput on a 192-CPU test system from 40 thousand to 4.2 million operations per second.
  • Scheduler tuning of EEVDF for short-slice tasks with up to 55% lower scheduling latency, reported from a measurement by its author.
  • Removal of the unsafe __get_user() and __put_user() helpers on x86, which Torvalds called "long over-due" according to that article.
  • AMDGPU enabling HDMI 2.1 by default, plus AMD Zen 6 preparation.

These performance figures come from patch authors' own measurements as relayed by one outlet; we have not verified them independently.

What it means

For kernel maintainers. A larger merge window means more to review before rc1, but the policy preference is clear: fixes belong early. Torvalds' remarks signal that AI-discovered fixes are being treated as ordinary development, subject to the same review, and that rewriting by humans is expected. His comment that such patches "often get rewritten" is a reminder that the maintainers, not the tools, decide what lands.

For people running Linux. Nothing changes today; rc1 is for testers. Distribution kernels and the long-term branch you may rely on (kernel.org lists 6.18 and 6.12 as longterm with projected end of life in December 2028) are unaffected. If 7.4 does become the new LTS, that will matter for the 2027 to 2030 hardware and distro cycles, but that decision is not announced.

For the AI-in-open-source debate. The kernel is an example of AI-found bugs becoming a normal input to a large project's workflow rather than an emergency. Whether the human review capacity can keep up is the open question; Hardware Busters points to a report that PostgreSQL 19 was delayed because reviewers could not clear AI-found bugs in time, which we have not checked against the PostgreSQL project.

What we don't know yet

  • How many of the 19,671 commits were suggested by AI. Nobody has published that number, and Torvalds did not give one. The commit count alone is not a count of AI contributions.
  • Whether 7.4 will be designated LTS. Kernel.org's list of longterm kernels does not include it.
  • The final release date.
  • Whether the rc weeks will in fact be quieter, which is the test of Torvalds' "healthier model" claim.

What to watch next

  • The weekly rc2 to rc4 announcements: fewer late fixes would support Torvalds' view.
  • The kernel.org longterm table for a 7.4 entry once it is released.
  • Distribution plans: which distributions say they will ship 7.4.
  • The kernel documentation's description of the -rc and stable process, linked in the sources, if you want to know what rc1 and LTS mean.

Sources

SaveStart the discussion

Written by

audiotale.net team

The official account of the people who run audiotale.net: announcements, conversation starters and the newsroom's drafts. Posts written with AI say so. Every other account here is a real person.

Read next