Linux Hibernation RFC Cuts I/O With Contiguous Swap Runs

LG engineer Youngjun Park has posted a ten-patch RFC that changes how Linux allocates swap slots for a hibernation image. The central idea is to allocate contiguous runs of slots and batch image reads and writes around those runs instead of repeatedly handling small pieces of fragmented swap space.

Hibernation saves an image of system memory to persistent storage before power-off and reads it back during resume. Park says allocation behavior matters because fragmented image slots translate into more I/O operations. The proposed design makes those runs explicit so the kernel can issue larger batches, a pattern that is also friendlier to flash storage.

The RFC also provides useful historical context. Park reports that after Linux 6.15 removed the swap slot cache, writing a hibernation image became about ten times slower on some SSDs. A later change restored much of that performance and landed in Linux 7.1.

The new series is not simply a repeat of that fix. It tries to improve the underlying allocation and I/O shape further by giving the hibernation path contiguous runs that can be processed together. That can reduce both allocator work and the number of storage operations needed for the same image.

A fast NVMe SSD can move large sequential transfers quickly, but a hibernation path that breaks an image into many small operations still pays software and device-command overhead. Keeping image slots contiguous gives the kernel more opportunity to submit larger operations and reduces bookkeeping around individual fragments.

That does not mean raw SSD sequential bandwidth becomes the only limit. Image creation, compression where applicable, memory management, filesystem-independent swap handling, device latency and platform firmware can all influence the time between requesting hibernation and reaching power-off, as well as resume time.

Because the series was posted on September 15 as an RFC, the useful next evidence is maintainer review, later patch revisions and testing across different SSDs and swap configurations. A design that performs well on both fast NVMe storage and slower flash devices would be more meaningful than one isolated benchmark.

For Linux laptop and workstation users who rely on suspend-to-disk, the direction is still notable: the kernel developers are treating hibernation image layout and I/O batching as performance problems in their own right rather than assuming modern SSD speed hides inefficient allocation.