40 млн строк кода: как меняется ядро Linux в 2026 году

Начало 2026 года для сообщества ядра Linux получилось насыщенным. В конце января в документацию добавили раздел Linux kernel project continuity. В нем описывается, что произойдет, если ключевые участники проекта внезапно не смогут выполнять свою работу. Обсуждения шли еще на Maintainers Summit в декабре 2025-го, и теперь этот вопрос зафиксирован в документации в явном виде. Все вполне логично. Ядро Linux давно превратилось в часть огромной мировой инфраструктуры — с десятками миллионов строк кода, тысячами разработчиков из компаний по всему миру и постоянным давлением со стороны требований к производительности, безопасности и энергоэффективности. Поэтому сообщество вынуждено не только поддерживать текущий уровень стабильности, но и экспериментировать, стараться сделать систему гибче и надежнее. В 2026 году проявились несколько направлений такой работы — от Rust и планировщиков до поддержки новых архитектур. Давайте посмотрим, что тут и как.

https://habr.com/ru/companies/ru_mts/articles/992766/

#Linux #ядро_Linux #Rust #планировщик_задач #sched_ext #RISCV #LoongArch #Линус_Торвальдс #управление_проектом #open_source

A fully functional sched_ext scheduler that makes real CPU scheduling decisions based on real-time planetary positions, zodiac signs, and astrological principles. This actually loads into the Linux kernel and schedules your system tasks. Because if the universe can influence our lives, why not our CPU scheduling too?

https://github.com/zampierilucas/scx_horoscope

#linux #sched_ext

Rethinking custom kernels in my Fedora setup script.

Installing custom kernels (CachyOS) might not be the best idea after all. They can break things during system upgrades, NVIDIA drivers rebuilding, module mismatches, dependency conflicts.

Instead, considering switching to sxc_lavd scheduler. Yes, it requires a COPR, but swapping a scheduler is a much smaller trade-off than maintaining a whole custom kernel. Stock #Fedora kernel stays intact, upgrades stay clean.

#linuxgaming #sched_ext

Using machine learning to optimize/fine-tune the #Linux #kernel at runtime to the needs of the workload – this is something I guess we'll see way more often in the future.

The recent #ossna25 talk "Improve Load Balancing With Machine Learning Techniques Based on the #sched_ext Framework" from Ching-Chun ("Jim") Huang gives a glimpse into such a future.

@lwn write-up: https://lwn.net/Articles/1027096/

Recording: https://www.youtube.com/watch?v=5VXemIXAOrI

Slides: https://static.sched.com/hosted_files/ossna2025/d2/Improve-Load-Balancing-With-Machine-Learning-Techniques-based-on-sched_ext.pdf

TIL: scx_bpfland Linux scheduler can improve performance

https://kbin.melroy.org/m/til/t/717526

#sched_ext protip: when running e.g. scx_bpfland on a server, make sure to use the -k (--local-kthreads) flag, otherwise full system saturation with CPU-bound tasks (e.g. a parallel make job using all cores) will severely starve I/O (e.g. DNS, NFS).
Really should be the default.. 🚀

#linux

The mainline-wo-mergew copr of the #kernel vanilla repositories[1] for #Fedora #Linux now ship 6.12 pre-releases that have the #BPF extensible scheduler class aka #sched_ext"[2] enabled, just like Fedora Rawhide will do with the next push of a new #LinuxKernel.

Many thx to Jose Fernandez for realizing this![3]

[1] https://fedoraproject.org/wiki/Kernel_Vanilla_Repositories
[2] https://lwn.net/Articles/972710/
[3] https://gitlab.com/cki-project/kernel-ark/-/merge_requests/3397

#Linux 6.12-rc1 is out:

https://lore.kernel.org/lkml/CAHk-=wiw[email protected]/

'"Despite conference travel (both for me and several maintainers), things seemed to go mostly fairly normally. There's a couple of notable new features in here: For one thing, #PREEMPT_RT is now mainlined and enabled as a config option (you do need to enable "EXPERT" to get the question). For another, #sched_ext also got merged.

[…]

Let's get the testing and calming down period started, ok?

Linus"'

#kernel #LinuxKernel

#sched_ext, which allows scheduling policies to be implemented as #BPF programs, has been merged for #Linux 6.12:

https://git.kernel.org/torvalds/c/88264981f2082248e892a706b2c5004650faac54

See https://lwn.net/Articles/922405/ for a description of what it does and https://lwn.net/Articles/972710/ for the controversy it caused that is the reason why it took so long to land in mainline.

FWIW, the PR with the BPF extensible scheduler class aka #sched_ext for #Linux 6.11 is out for 10 days now, but Linus did not merge it yet. Peterz and @qyousef also raised a few issues in between. Remains to be seen if Linus nevertheless will merge sched_ext in the next few days.

For details, see:

https://lore.kernel.org/all/2024072316[email protected]/t/#u

https://lore.kernel.org/all/20240725011907.3f5ropfai3xoy3l3@airbuntu/t/#u

#kernel #LinuxKernel #scheduler