Uncached Buffered I/O & Some Other Nice Memory Management Optimizations With Linux 6.14

There is a ton of exciting MM changes this cycle for the Linux 6.14 kernel release due out in March. Some of the most exciting changes I found from this Linux 6.14 MM pull include:
- The Uncached buffered I/O support from Jens Axboe is included! With the new RWF_DONTCACHE flag, this uncached buffered I/O support is a nice performance win.
- A rework of the swap allocator locks that also simplifies swap allocator locking was found to provide a 400% speedup for one workload while a 35% reduction for kernel build times when using swap on zRAM. The 400% win was in a VM scalability test with pmem as SWAP. Details within this patch series.
- Google's Yu Zhao contributed a few MGLRU fixes as well as some new MGLRU performance optimizations to that important code.
- Support for large folios with TMPFS rather than being only limited to PMD-sized folios.
- Removing the global swap cgroup lock was found to provide a 10% speed-up for TMPFS-based kernel builds.
- Improved memory accounting accuracy by accounting for page tables at all levels.
- Various DAMON improvements like page level properties based monitoring.
See this pull request for the full list of MM feature changes submitted for Linux 6.14.
8 Comments