Current Status and Future of HWPOISON Subsystem
summary
Explains concrete kernel work: error handling now keeps poisoned pages in pagecache (so data loss on writeback is reduced), accounting changes land since 5.18, patches enabling 1GB hugepage support merged around 6.0 and fixes for hotplug interactions land in 6.2; memfd/ramfs got the pagecache fix, and a safe memory copy path is implemented to avoid panics on action‑required MCEs. Good if wanting patch numbers, regressions and current status of each fix.
AI-written summary · about
chapters (24)
- 0:00Intro
- 1:05Memory error background
- 2:23Hardware Error
- 4:10level handling for memory error
- 5:02Terminology Confusion
- 6:44HWPOISON basic concept
- 8:30Action Required/Optional Memory Error
- 10:04Example of hard page offline
- 11:33Example of soft page offline
- 13:07HWPOISON Internal: hard-offline
- 15:32Notification to Userspace
- 17:42Recent development topics
- 18:07Soft offline rework
- 21:092. More topics on pagecache handling
- 22:31Folio related issues
- 24:27Huge) Zero page
- 25:44Persistent memory
- 26:56Hugetlb pinning race
- 28:39Subpage hwpoison tracking
- 30:001GB hugetib page
- 30:34Memory hotplug integration
- 32:58Machine Check safe memory copy
- 34:41Management interface
- 35:52Summary
related talks
37mHardware Error Handling Improvement for Reliable KVM Hypervisor
47mLinux's Leap Beyond Fixed Pages
38mKernel Hardening: Protecting the Protection Mechanisms
30miommu page table consolidation and iommufd
23mCPU Isolation state of the art
47mLinux on POWER Driver Porting, Optimization, and Test
43mHybrid THP Mechanism. Selective Use of Huge Pages by Hot Applications
22mPreserving guest memory across kexec