Highmem deprecation planning
summary
Proposal: deprecate highmem on 32-bit kernels over the next decade by shrinking what actually uses it, landing on a few supported paths and then removing page-cache highmem in the early 2030s. Evidence: most new 32-bit boards ship with 0.5–1 GB RAM, a small set of legacy devices keep 2–4+ GB and expect kernel updates into the 2030s. Concrete work: audit and drop GFP_HIGH allocations, land remove high PTE patches, pursue densemem for discontiguous RAM, treat VM split 4G unlikely as unfunded.
AI-written summary · about
chapters (11)
- 0:00Introduction and context
- 1:49What is highmem and why it matters
- 6:01Current highmem usage landscape
- 10:47Supporting legacy systems and updates
- 17:13Highmem on other architectures
- 19:24Proposed solutions and challenges
- 27:06Policy changes for highmem features
- 33:12Compatibility and future directions
- 36:41Timeline and deprecation plan
- 41:56Hardware longevity and support limits
- 43:25Q&A and next steps
related talks
29mHigh memory management API changes
24mConsolidating representations of the physical memory
40mInspecting and Optimizing Memory Usage in Linux
38mDeferred Memblocks Init for Boot Time Reduction
11mAdvanced Parallel Memory Virtualization
22mPreserving guest memory across kexec
41mLies Programmers Believe about Memory
22mEMD: Fair and Efficient Dynamic Memory De-bloating of Transparent Huge Pages