Swap tables, flash-friendly swap, swap_ops, and more

Linux’s dusty memory closet got a glow-up — and commenters are weirdly obsessed

TLDR: Linux developers are overhauling the part of the system that juggles memory when your computer runs low, making it smaller, cleaner, and potentially kinder to SSDs. Commenters loved the nerdy ambition but roasted the celebration, joking that programmers are throwing a parade for saving a few bytes while bigger fights still loom.

Linux developers just turned one of the operating system’s oldest, messiest corners into the surprise main character of the week: swap, the feature that pushes idle stuff out of memory when your computer gets crowded. On paper, the summit news was simple: the team is slimming down swap’s overhead, cleaning up years of tangled code, making it faster, and trying to make it less rude to solid-state drives. In the room, even a tiny memory savings got actual applause — which commenters immediately treated like the nerdiest standing ovation of 2026.

That set off the usual community split. One camp was thrilled that a long-neglected subsystem is finally getting love, with fans cheering the “finally someone cleaned the basement” energy. The other camp was classic internet skeptical: are we celebrating shaving off a few bytes now? Several reactions joked that kernel developers will fight dragons for a 1-byte win, while normal people just want Chrome to stop eating their laptop alive. Others zeroed in on the real drama: developers still don’t fully agree when pages should be dropped, how aggressive reloading should be, or whether compressed RAM swap is already fast enough. Translation: yes, it’s cleaner, but the comment section heard “more arguments incoming.”

The funniest bits came from people treating swap like a chaotic storage reality show: old code called “racy,” pages getting dropped, surprise overlaps, and a possible special marker for hibernation pages. The vibe was equal parts “amazing engineering” and “this haunted attic still has goblins.”

Key Points

  • Three summit sessions in 2026 focused on the Linux kernel swap subsystem, reflecting renewed developer attention to swap performance, maintainability, and SSD friendliness.
  • Kairui Song said swap-table work merged for Linux 7.0 reduced swap metadata overhead from three to 11 bytes per page down to two to ten bytes per page.
  • Song said future work aims to eliminate static overhead, though near-term reduction is constrained by refault tracking requirements in the memory resource controller.
  • Most swap helpers are now folio-based, bypass needs for some swap-cache operations have been removed, and work continues to unify folio allocation with swap-cache handling.
  • Outstanding issues include asynchronous swap I/O overshoot, inefficient PMD-level huge-page swapping, hibernation-page readahead, dynamic swap-area resizing, and slot conflicts between anonymous pages and shmfs-backed THPs on the same device.

Hottest takes

"Kernel devs heard ‘save 1 byte’ and started clapping" — bytebandit
"This is just Marie Kondo for Linux memory" — segfault_sally
"If the fix is ‘drop the page and pray,’ the drama is not over" — cachemeoutside
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.