Futureweb

Software

A Linux 7.4 scheduler pull targets short-slice latency. It is not a tarball.

Ingo Molnar's October 18 sched/core pull for Linux 7.4 carries Vincent Guittot's short-slice changes. His timings stay his. The pull is not a 7.4 tarball.

The short version

Ingo Molnar's October 18 pull asks Linus to take sched/core for Linux 7.4, including Vincent Guittot's changes for short-slice tasks. Guittot's earlier measurements are his, not a merged-result table. The pull is not a 7.4 tarball. It is a different document from the close-path pull already filed.

The pull is the document

On October 18, 2026 Ingo Molnar sent sched/core for the Linux 7.4 merge window. He points at the tip tag sched-core-2026-10-18 and at tip commit b99b7f60c5b473d374c68b53753153b4aa9e0702. Inside that pull, Vincent Guittot has three changes: keep a protected slice from rising above a minimum, align the update path with the set path, and handle more cases where a short-slice task wakes. That is the latency record. It is not a claim that 7.4 has shipped. Asked directly, Is this the file-descriptor close pull? No. That pull was filed separately. This one is sched/core. Same merge window, different dependency.the 7.4 close pull.

His cyclictest numbers stay his

Guittot's cover series is dated October 2, 2026. The measurements in it are his description of the series, not a table someone published after Linus merged the result. This page will not promote those figures into a before-and-after for the whole 7.4 window. A reader who wants the latency story wants his message. A reader who wants a released kernel wants a tarball that does not exist yet. Phoronix's news item reports the window. It does not replace the pull. An October 26 brief files a third change in that window, the EFI stub publishing a loader disk identifier when firmware boots the kernel directly, and that variable is not this scheduler pull.the EFI variable is a different pull.

The Vera line is Phoronix's headline

Phoronix's October 22 headline says Linux 7.4 improves scheduling latency for short-slice tasks and also names an NVIDIA Vera SMT preference. The short-slice half matches the pull this brief is built on. The SMT half is in the headline. This page did not open a separate Vera scheduler commit and will not invent the mechanism from the headline alone. If that preference matters, its own message is the document. It is not this pull's three Guittot changes. Asked directly, What tag does the pull name? Molnar points at the tip tag sched-core-2026-10-18. The tip commit he names is b99b7f60c5b473d374c68b53753153b4aa9e0702.HPE's Vera rack is an announcement.

A different pull in the same window

This paper already filed Christian Brauner's close-path pull for the same 7.4 window. That document is about putting file closes inline. This document is about scheduler slices. They share a version number and they do not share a dependency. A reader who followed the close story here is not reading it again. kernel.org listing 7.3 as mainline, which the earlier brief recorded on October 21, is still the right picture of a merge window: requests go in before a tarball exists.Linux 7.3 without the rc.

What the three changes are for

A short-slice task is one the scheduler has already decided should not run long. Guittot's note is that the protection around that slice could drift, and that some wakeups were missing the short-slice path. The pull says the protection should stay at the minimum and that more of those wakeups should take the short path. That is a policy correction inside the scheduler. It is not a userspace tool, and it is not a tuning guide this page will write. Operators do not flip it on with a command this brief is going to print. An October 24 brief files a different 7.4 change, a memblock search, and that search is not this scheduler pull.the memblock search is a different pull.

What this page will not paste

It will not paste the scheduler source. An earlier fair.c series that circulated in June is not this October pull, and copying it would attach the wrong diff to the wrong message. The identifier is the October 18 pull and the tip commit it names. People who review kernel mail want that link. People who want a distribution package are early. An October 25 brief files Takashi Iwai's sound pull for the same window, and that pull is not this scheduler change. Asked directly, What does Phoronix add about Vera? Phoronix's October 22 headline also names an NVIDIA Vera SMT preference in the 7.4 window. This page does not invent how that preference works.the sound pull names an AI flood.

Who is in scope

Kernel readers tracking the 7.4 merge window, and anyone who hit the close-path brief and needs to know this is the other pull. Latency-sensitive workloads are the case Guittot is measuring. A desktop user waiting on a distribution kernel is not yet holding this change. No distribution changelog is in the sources. Asked directly, Is Linux 7.4 a released kernel on October 22? No. The October 18 message is a pull into the 7.4 merge window. A pull is not a tarball a distribution can point at.what an unsafe block has to name.

What stays on the record

October 18, 2026 sched/core pull from Ingo Molnar. Tip tag sched-core-2026-10-18. Tip commit b99b7f60c5b473d374c68b53753153b4aa9e0702. Three Guittot changes on short-slice protection and wakeups. His October 2 measurements stay his. Phoronix names a Vera SMT preference in the same window without a mechanism this page will invent. Linux 7.4 is not a tarball. Distinct from the close-path pull. Asked directly, Whose latency numbers are in the series? Vincent Guittot's, from his October 2 cover message. They describe the series he measured. They are not a table of the merged result.four stable kernels.

What would move the row

A Linux 7.4 tarball, or a stable patch that carries this scheduler change, would be the next document. A Vera SMT commit message, read on its own, would be a separate row. Until then the record is Molnar's pull, Guittot's series, and Phoronix's account of the window. The short answer this page will stand on is this: Ingo Molnar's October 18 pull asks Linus to take sched/core for Linux 7.4, including Vincent Guittot's changes for short-slice tasks. Guittot's earlier measurements are his, not a merged-result table. The pull is not a 7.4 tarball. It is a different document from the close-path pull already filed.the software desk.

Sources

The reports this brief is filing. Futureweb did not republish them.

  1. LKML
  2. Scheduler series
  3. Phoronix

Questions

Is Linux 7.4 a released kernel on October 22?

No. The October 18 message is a pull into the 7.4 merge window. A pull is not a tarball a distribution can point at.

Whose latency numbers are in the series?

Vincent Guittot's, from his October 2 cover message. They describe the series he measured. They are not a table of the merged result.

What does Phoronix add about Vera?

Phoronix's October 22 headline also names an NVIDIA Vera SMT preference in the 7.4 window. This page does not invent how that preference works.

Is this the file-descriptor close pull?

No. That pull was filed separately. This one is sched/core. Same merge window, different dependency.

What tag does the pull name?

Molnar points at the tip tag sched-core-2026-10-18. The tip commit he names is b99b7f60c5b473d374c68b53753153b4aa9e0702.