Decoupling Large Folios from Transparent Huge Pages
summary
Goal is to let file-backed large folios work without depending on CONFIG_TRANSPARENT_HUGEPAGE by removing the huge_memory.c splitting dependency. One quick hack sets min_order == max_order so folios never invoke splitting; this runs xfstests on XFS and passes. A more involved path allocates PMD-size folios but maps them with page-table entries instead of PMD mappings. RFC patches are posted and the remaining questions are which use cases disable THP (real-time, tiny kernels, arches lacking PMD).
AI-written summary · about
chapters (7)
- 0:00Introduction and background
- 2:06Why decouple large folios from THP
- 4:12Current issues with THP dependency
- 6:00First solution: min and max order alignment
- 7:50Second solution: separate allocation and mapping
- 9:55Long-term goal: remove config THP
- 12:42Use cases for disabling THP
related talks
43mHybrid THP Mechanism. Selective Use of Huge Pages by Hot Applications
47mLinux's Leap Beyond Fixed Pages
44mLarge Pages in Linux
22mEMD: Fair and Efficient Dynamic Memory De-bloating of Transparent Huge Pages
12mEvery Mapping Counts in Large Amounts: Folio Accounting
29mHigh memory management API changes
12mRethinking File Mapping for Persistent Memory
30mFlexible Paging in Linux: Per-process Page Size