Linux 7.2（v7.1 → v7.2-rc6）性能优化召回/精排清单

生成 2026-08-06 · 用日报 skills 召回精排。召回宁多勿漏，判定四问（意图/性能模式/量化证据/真实瓶颈）精排，不限行数、不设数量上限。

召回

• 范围：v7.1（2026-06-14）→ v7.2-rc6（08-02），共 17086 提交
• 方法：recall_git.py 21 子系统（日期窗口）补调度器 tag 范围（漏 merge 早提交），合并去重 1246 个非 merge 候选
• 排除统计：bugfix 500 / 非明确性能 560 / 清理 rename 113，共 1173

综合判定结果（标题 + commit message + diff）

状态	数量
待分析（优化）	28
已分析（sched/cache 系列）	25
剔除（正确性/清理）	20

待分析（优化）候选

• [内存管理] 0ef8faff490b fs: push nr_cached_objects memcg gating into individual filesystems
• [内存管理] 4edd70ee6a7d mm/sparse-vmemmap: flush_cache_vmap() after hotplugging vmemmap
• [内存管理] 1b6444331beb mm/khugepaged: enable clean pagecache folio collapse for writable files
• [内存管理] 5fea7eb1a331 mm/khugepaged: avoid unnecessary mTHP collapse attempts
• [内存管理] 90ed32d00054 mm/khugepaged: introduce mTHP collapse support
• [内存管理] 3a460f245f00 mm/khugepaged: skip collapsing mTHP to smaller orders
• [内存管理] e22f2fe72c74 mm/khugepaged: generalize collapse_huge_page for mTHP collapse
• [内存管理] 5a9c05d86683 mm/khugepaged: generalize _collapse_huge_page* for mTHP support
• [内存管理] 1a787779fe2a mm/slab: allow kmem_cache_alloc_bulk() with any gfp flags
• [io_uring] fa1ac3b9eb62 io_uring/zcrx: don’t clear master_ctx from the import path
• [io_uring] f1596ba3e6b3 io_uring/kbuf: free the replaced iovec after a successful grow
• [io_uring] ca4aa97194ae io_uring: get rid of tw_pending for !DEFER task work
• [文件系统] 0ef8faff490b fs: push nr_cached_objects memcg gating into individual filesystems
• [文件系统] 6fe4e4b8259e fscrypt: Avoid dynamic allocation in fscrypt_get_devices()
• [文件系统] 222bc257a151 f2fs: skip inode folio lookup for cached overwrite
• [文件系统] caac757a3d2d f2fs: add iostat latency tracking for direct IO
• [文件系统] 5dfb768326b9 f2fs: optimize representative type determination in GC
• [文件系统] 4dd6f6d3085a fuse: back uncached readdir buffers with pages
• [网络] c4a5f0071cc6 xfrm: cache the offload ifindex for netlink dumps
• [电源/空闲] 3a2976df778a Merge branch ‘pm-cpuidle’
• [电源/空闲] 68ff4a3ccda9 cpuidle: Allow exit latency to exceed target residency

• [架构(arm64)] ff4c5a0de1f2 arm64/mm: Optimize TLB flush in unmap_hotplug_[pmd

  pud]_range()

• [架构(arm64)] 534eb6940a89 arm64: Avoid eager DVMSync reclaim batches with C1-Pro SME erratum
• [架构(x86)] 07c60dda9c05 perf/x86/amd/core: Avoid enabling BRS from the SVM reload path
• [加密/解密] 6fe4e4b8259e fscrypt: Avoid dynamic allocation in fscrypt_get_devices()
• [调度器] 1842bf97af10 sched/deadline: Use revised wakeup rule only for running dl_server
• [调度器] abc40cca0efd sched/proxy: Optimize try_to_wake_up()
• [调度器] f13beb010e4a sched: Have try_to_wake_up() handle return-migration for PROXY_WAKING case

已分析（sched/cache 系列整体）

• 1eae219ea0e8 sched/topology: Provide arch_llc_mask for cache aware scheduling
• a26d9208c137 Merge branch ‘sched/cache’
• 91d07324c930 sched/cache: Annotate lockless accesses to mm->sc_stat.cpu
• c1e7fe5e75ed sched/cache: Add user control to adjust the aggressiveness of cache-aware scheduling
• 808915f982c2 sched/cache: Avoid cache-aware scheduling for memory-heavy processes
• 7030513a0877 sched/cache: Calculate the LLC size and store it in sched_domain
• 7b34bb1ca324 sched/cache: Skip cache-aware scheduling for single-threaded processes
• deee5e27d5b6 sched/cache: Disable cache aware scheduling for processes with high thread counts
• a2b4cf39d9d3 sched/cache: Allow only 1 thread of the process to calculate the LLC occupancy
• 067a31358143 sched/cache: Allow the user space to turn on and off cache aware scheduling
• …共 25 个补丁，作为同一系列在 sched-cache-aware-load-balancing 分析

剔除（正确性/清理，理由见标题特征）

• [内存管理] d09a8fd52147 mm/migrate: exclude hugetlb folios from MTHP_STAT_NR_ANON accounting
• [内存管理] 4bd0c3515a04 mm: decrement MTHP_STAT_NR_ANON in free_zone_device_folio()
• [块层] 1e56f30a73f3 block: Make WBT latency writes honor enable state
• [io_uring] bc0e8faf90e7 io_uring: preserve task restrictions across exec
• [io_uring] 9db03d2116ab Revert “io_uring/epoll: disallow adding an epoll file to an epoll context”
• [io_uring] 0b88f4705899 io_uring/kbuf: cap buffer selection length at MAX_RW_COUNT
• [io_uring] c77ffbc980ef io_uring/net: initialize mshot_len for send
• [io_uring] cc609376e9a4 io_uring/fs: check unused sqe fields for unlinkat
• [io_uring] f3176c8ac421 Revert “io_uring: grab RCU read lock marking task run”
• [io_uring] 1fe703cc708f io_uring/register: preserve SQ array entries on resize
• [io_uring] bdc2fc388c34 io_uring, audit: don’t log IORING_OP_RECV_ZC
• [io_uring] c554246ff4c6 io_uring/rw: preserve partial result for iopoll
• [io_uring] 5c4adb7fb46f netdev: expose io_uring rx_page_order order via netlink
• [文件系统] cf716276b0dc f2fs: avoid unnecessary sanity check on ckpt_valid_blocks
• [网络] 4787a6d2629b nexthop: avoid unlocked f6i_list walk in nh_rt_cache_flush

补挖精排（08-06 第二波）

对”待分析”清单中未深挖的 12 个候选二次精排（看真实 diff/message，判定四问），挑 4 个深挖，8 个排除。

本次深挖（4 个高价值性能点）

• [io_uring] ca4aa97194ae get rid of tw_pending（热路径） → io-uring-tw-pending-optimize
• [文件系统] 222bc257a151 f2fs skip inode folio lookup for cached overwrite（有基准） → f2fs-skip-inode-folio-lookup
• [arm64] 534eb6940a89 Avoid eager DVMSync reclaim batches（恢复 TLB batching） → arm64-dvmsync-tlb-batching
• [调度器] f13beb010e4a try_to_wake_up return-migration (PROXY_WAKING) → sched-proxy-return-migration

排除（正确性修复/诊断增强/特定硬件，非性能优化）

• 0ef8faff490b fs memcg nr_cached_objects gating（恢复 XFS memcg reclaim，正确性）
• 4edd70ee6a7d mm/sparse-vmemmap flush_cache_vmap（RISC-V 正确性）
• fa1ac3b9eb62 io_uring/zcrx master_ctx（正确性）
• f1596ba3e6b3 io_uring/kbuf iovec（正确性，dangling iovec）
• caac757a3d2d f2fs iostat DIO latency（诊断增强）
• c4a5f0071cc6 xfrm offload ifindex cache（netlink dump 竞态正确性）
• 07c60dda9c05 perf/x86 AMD BRS SVM reload（正确性，#GP）
• 1842bf97af10 sched/deadline revised wakeup（正确性，defer 规则）