Every Sunday, Linus Torvalds posts a release candidate announcement to the kernel mailing list. Most people skim them. But if you learn to read one the way Linus writes it, a boring-looking email becomes a surprisingly useful status report on the entire Linux ecosystem. Linux 7.3-rc5, tagged September 27, is a perfect example.

Start with the numbers, then ignore them

The headline figures for 7.3-rc5: 577 commits from 275 contributors, which Linus summed up as "another week, another large rc." Release candidates have been consistently big this cycle, so size alone tells you nothing alarming.

Then came the interesting part. Torvalds flagged that the diffstat looked weird: drivers made up less than a third of the changes, while selftests accounted for roughly a quarter. In a normal week, drivers dominate by a wide margin — a shape like this usually raises eyebrows.

Here's the lesson: a diffstat measures lines, not work. Linus immediately explained the quirk. Most driver patches in rc5 are small; a few selftest patches are enormous — the single biggest patch in the whole release is an arm64 KVM ITS table test. The commit history tells a much more balanced story than the raw line counts suggest.

If you ever catch yourself panicking over a strange-looking diffstat — in the kernel or in your own projects — remember rc5. Ask what the lines represent before deciding what they mean.

"The same old, same old" is good news

Torvalds' verdict on the actual content: routine. Architecture fixes (mostly KVM-related, plus some perf work), scheduler and BPF core work, networking, DRM close behind, scattered VFS and XFS fixes. One notable revert: HID++ support for the Logitech Bolt receiver got pulled back after problems.

When Linus calls a release candidate "the same old, same old," that's the best outcome a -rc can have. Release candidates aren't supposed to be exciting. Exciting in an rc means regressions, and regressions mean a delayed release. Boring is the goal.

The part worth paying attention to: AI is now a kernel developer's coworker

Buried in the coverage of rc5 is the trend that actually matters: AI and LLM tools keep uncovering bugs in a codebase approaching 41 million lines. This isn't a one-off — it's the new normal of kernel development. Automated agents find issues continuously, which is part of why every rc this cycle has been large.

This connects directly to what Canonical just did with Ubuntu's kernel schedule (see our earlier piece): the discovery rate has permanently changed, and the whole ecosystem — from Linus's tree to your distro's update cadence — is adapting to it.

Should you care about rc5?

Honest answer: probably not directly. Release candidates are for testers and developers, not production machines. But rc5 tells you what's coming:

  • Linux 7.3 is on track for October 18 (or October 25 if an eighth rc is needed). Ubuntu 26.10, shipping October 15, is built around it.
  • The 7.4 merge window opens right after, which is when queued work like the slab allocator changes and parallelized kernel builds land.
  • If you're still on 7.1, stop. The 7.1 series went end-of-life with 7.1.13. No more upstream security fixes — ever. Check with uname -r and upgrade to a supported series (7.2.x, 6.18.x, or a longterm line like 6.12).

The takeaway

Kernel RC announcements look like noise until you learn the format: numbers first, then the author's own interpretation, then the verdict. Linus gives you all three every week — most readers just stop at the numbers.

Next Sunday, when rc6 lands, try reading past the diffstat. You'll know exactly what to look for.

Sources: OSTechNix — Linux Kernel 7.3 RC5, linuxcompatible.org — 7.3-rc5 breakdown