Kernel: MGLRU swappiness, pseudo-fs backing, and shadow stacks

Memory reclaim, filesystem backing, and protection bypasses led the day's kernel traffic. Library and driver fixes covered keyrings, pagecache refcounts, PCS attachment, and an ASoC regression.

MGLRU v9 makes swappiness match classic LRU

A nine-revision patch series for multi-generational LRU replaces get_type_to_scan and its fallbacks with refault-driven proportional scanning of anonymous and file pages. The change is meant to honor swappiness the way the classic LRU does, so reclaim mix tracks the administrator's setting instead of drifting under fallbacks.

RFC blocks pseudo-fs files as backing storage

An RFC of 42 patches adds file_check_backing and the SB_I_PSEUDO flag so loop, erofs, and fuse cannot use files from pseudo filesystems as backing storage. The series targets capability bypasses that arise when those files are treated as ordinary storage.

Lib fixes reopen kvmalloc GFP_ATOMIC debate

A four-patch set labeled for 7.3 drew discussion among Linus and mm maintainers on whether kvmalloc should support GFP_ATOMIC and vmalloc after an rhashtable deadlock and maple_tree RCU fixes. The thread is still a design debate rather than a settled interface change.

Keyrings drop assoc_array for rhashtable

Version 11 of a KEYS series replaces the only remaining user of assoc_array, the keyring code, with rhashtable. It removes about 1,700 lines plus documentation and includes s390 keyring fixes.

HMM rejects shadow-stack mappings

A patch stops HMM from returning HMM_PFN_WRITE, or a read mapping, for VMAs marked VM_SHADOW_STACK. The intent is to stop device DMA from bypassing x86 shadow-stack protection.

Pagecache folio refcounts get overflow checks

Version 4 of a memory-management series adds refcount limits and checks for pagecache folios. The checks are aimed at overflows that large transparent huge pages and file mappings can provoke.

Fwnode PCS API reaches v20

A net-next series posts a fwnode-based provider and consumer API for physical coding sublayers, with phylink changes for late attach and release. It is infrastructure for describing PCS devices through firmware nodes rather than a board-specific hook.

ASoC runtime component count reset

A one-line class of fix resets rtd->num_components after allocation. The bug came from a devm_kzalloc_flex change combined with __counted_by, which left ASoC runtimes with no components.