The Many Memory Management Improvements In Linux 6.18
The many memory management "MM" changes were recently merged into the Linux 6.18 kernel, consisting of a number of interesting patch series.
Andrew Morton submitted the big set of MM changes for Linux 6.18. The highlights this cycle from the long list of patches include:
- Improvements to the cluster scan strategy for the swapping code. This improves the large allocation performance. A Tencent engineer found that for a kernel build test with 96 jobs and 10G of zRAM with 64KkB mTHP enabled, the system time was nearly halved and also saw a lower swap failure rate.
- Improvements to the Kexec Handover functionality introduced back in Linux 6.16.
- Kernel file-mapped folios introduce the notion of "kernel file pages".
- Extending PR_SET_THP_DISABLE to allow individual processes to opt-out of THP always behavior into "madvise" THP mode without affecting other running workloads on the system.
- Tiny optimizations for large read operations to enhance the page cache read path.
- Adding persistent huge zero folio support.
- The Zpool indirection layer has been removed.
- Enhancements around Rust allocator support. There is also now Rust abstractions for maple trees, driven by Nouveau and Nova driver needs.
- The preliminary code intrdoucing swap tables as swap cache.
- A number of improvements to the out-of-memory (OOM) killer.
This complements the other non-MM pull request to also already having been merged for Linux 6.18.
More details on the many patches making up the MM feature pull for Linux 6.18 via this pull.
Andrew Morton submitted the big set of MM changes for Linux 6.18. The highlights this cycle from the long list of patches include:
- Improvements to the cluster scan strategy for the swapping code. This improves the large allocation performance. A Tencent engineer found that for a kernel build test with 96 jobs and 10G of zRAM with 64KkB mTHP enabled, the system time was nearly halved and also saw a lower swap failure rate.
- Improvements to the Kexec Handover functionality introduced back in Linux 6.16.
- Kernel file-mapped folios introduce the notion of "kernel file pages".
- Extending PR_SET_THP_DISABLE to allow individual processes to opt-out of THP always behavior into "madvise" THP mode without affecting other running workloads on the system.
- Tiny optimizations for large read operations to enhance the page cache read path.
- Adding persistent huge zero folio support.
- The Zpool indirection layer has been removed.
- Enhancements around Rust allocator support. There is also now Rust abstractions for maple trees, driven by Nouveau and Nova driver needs.
- The preliminary code intrdoucing swap tables as swap cache.
- A number of improvements to the out-of-memory (OOM) killer.
This complements the other non-MM pull request to also already having been merged for Linux 6.18.
More details on the many patches making up the MM feature pull for Linux 6.18 via this pull.