Kernel Summit 2014 had a topic on performance regressions and catching them. The situation relatively recently has been good but I believe this is partially due to major distributions stabilising recently and hardware vendors working on performance for their latest platforms. On thing I fear is that we miss performance regressions because releases contains performance gains and losses, some of which balance out. After a number of mainline releases, the performance may look ok but in comparison to a distribution kernel cherry-picking the picture is not as rosy. The 3.0-longterm included a number of performance-related patches so it could be used as a good performance baseline for later releases. It is no longer maintained.
This series contains a large number of patches against 3.12-longterm that never made it to stable as the bugs were not serious enough or they were performance patches. 3.10-longterm may need other pre-requisites I did not research and 3.14-longterm should be able to apply a subset although I have not tested the result. The first 12 patches are mostly functional patches. After that is mostly but not exclusively performance-related patches. Sometimes there are patches that are clean-ups so that later patches do not have to be extensively modified for the backport. I'm aware that this series is outside the standard rules for stable. Even the stable rules for performance patches is being violated here as I do not added information to the changelog on user-visible effects or bugzilla entries beyond what is already available as the information was not recorded when these patches were identified. By the nature of many of the patches they do include performance data though. It's up to the -stable maintainers for each longterm kernel if they want to use the kernel they maintain as a performance baseline or not. Al Viro (1): callers of iov_copy_from_user_atomic() don't need pagecache_disable() Bob Liu (2): mm: thp: cleanup: mv alloc_hugepage to better place mm: thp: khugepaged: add policy for finding target node Christoph Lameter (1): vmscan: reclaim_clean_pages_from_list() must use mod_zone_page_state() Damien Ramonda (1): readahead: fix sequential read cache miss detection Dan Streetman (4): swap: change swap_info singly-linked list to list_head lib/plist: add helper functions lib/plist: add plist_requeue swap: change swap_list_head to plist, add swap_avail_head Dave Chinner (1): fs/superblock: unregister sb shrinker before ->kill_sb() David Rientjes (10): mm, thp: do not allow thp faults to avoid cpuset restrictions mm, compaction: avoid isolating pinned pages mm, compaction: determine isolation mode only once mm, compaction: ignore pageblock skip when manually invoking compaction mm, migration: add destination page freeing callback mm, compaction: return failed migration target pages back to freelist mm, compaction: add per-zone migration pfn cache for async compaction mm, compaction: embed migration mode in compact_control mm, compaction: terminate async compaction when rescheduling mm, thp: only collapse hugepages to nodes with affinity for zone_reclaim_mode Davidlohr Bueso (1): mm: per-thread vma caching Fabian Frederick (1): mm/readahead.c: inline ra_submit Han Pingtian (1): mm: prevent setting of a value less than 0 to min_free_kbytes Heesub Shin (1): mm/compaction: clean up unused code lines Hugh Dickins (4): mm: fix bad rss-counter if remap_file_pages raced migration mm: fix direct reclaim writeback regression shmem: fix init_page_accessed use to stop !PageLRU bug mm/memory.c: use entry = ACCESS_ONCE(*pte) in handle_pte_fault() Jens Axboe (1): mm/filemap.c: avoid always dirtying mapping->flags on O_DIRECT Jerome Marchand (2): mm: make copy_pte_range static again memcg, vmscan: Fix forced scan of anonymous pages Jianyu Zhan (1): mm/swap.c: clean up *lru_cache_add* functions Johannes Weiner (5): lib: radix-tree: add radix_tree_delete_item() mm: shmem: save one radix tree lookup when truncating swapped pages mm: filemap: move radix tree hole searching here mm + fs: prepare for non-page entries in page cache radix trees mm: madvise: fix MADV_WILLNEED on shmem swapouts Joonsoo Kim (7): slab: correct pfmemalloc check mm/compaction: disallow high-order page for migration target mm/compaction: do not call suitable_migration_target() on every page mm/compaction: change the timing to check to drop the spinlock mm/compaction: check pageblock suitability once per pageblock mm/compaction: clean-up code on success of ballon isolation vmalloc: use rcu list iterator to reduce vmap_area_lock contention KOSAKI Motohiro (2): mm: get rid of unnecessary overhead of trace_mm_page_alloc_extfrag() mm: __rmqueue_fallback() should respect pageblock type Linus Torvalds (1): mm: don't pointlessly use BUG_ON() for sanity check Mel Gorman (30): mm, x86: Account for TLB flushes only when debugging x86/mm: Clean up inconsistencies when flushing TLB ranges x86/mm: Eliminate redundant page table walk during TLB range flushing mm: compaction: trace compaction begin and end mm: optimize put_mems_allowed() usage mm: vmscan: use proportional scanning during direct reclaim and full scan at DEF_PRIORITY mm: page_alloc: do not update zlc unless the zlc is active mm: page_alloc: do not treat a zone that cannot be used for dirty pages as "full" include/linux/jump_label.h: expose the reference count mm: page_alloc: use jump labels to avoid checking number_of_cpusets mm: page_alloc: calculate classzone_idx once from the zonelist ref mm: page_alloc: only check the zone id check if pages are buddies mm: page_alloc: only check the alloc flags and gfp_mask for dirty once mm: page_alloc: take the ALLOC_NO_WATERMARK check out of the fast path mm: page_alloc: use unsigned int for order in more places mm: page_alloc: reduce number of times page_to_pfn is called mm: page_alloc: convert hot/cold parameter and immediate callers to bool mm: page_alloc: lookup pageblock migratetype with IRQs enabled during free mm: shmem: avoid atomic operation during shmem_getpage_gfp mm: do not use atomic operations when releasing pages mm: do not use unnecessary atomic operations when adding pages to the LRU fs: buffer: do not use unnecessary atomic operations when discarding buffers mm: non-atomically mark page accessed during page cache allocation where possible mm: avoid unnecessary atomic operations during end_page_writeback() mm: pagemap: avoid unnecessary overhead when tracepoints are deactivated mm: rearrange zone fields into read-only, page alloc, statistics and page reclaim lines mm: move zone->pages_scanned into a vmstat counter mm: vmscan: only update per-cpu thresholds for online CPU mm: page_alloc: abort fair zone allocation policy when remotes nodes are encountered mm: page_alloc: reduce cost of the fair zone allocation policy Michal Hocko (1): mm: exclude memoryless nodes from zone_reclaim Nishanth Aravamudan (1): hugetlb: ensure hugepage access is denied if hugepages are not supported Raghavendra K T (1): mm/readahead.c: fix readahead failure for memoryless NUMA nodes and limit readahead pages Sasha Levin (1): mm: remove read_cache_page_async() Shaohua Li (2): swap: add a simple detector for inappropriate swapin readahead x86/mm: In the PTE swapout page reclaim case clear the accessed bit instead of flushing the TLB Tim Chen (1): fs/superblock: avoid locking counting inodes and dentries before reclaiming them Vladimir Davydov (4): mm: vmscan: shrink all slab objects if tight on memory mm: vmscan: call NUMA-unaware shrinkers irrespective of nodemask mm: vmscan: respect NUMA policy mask when shrinking slab on direct reclaim mm: vmscan: shrink_slab: rename max_pass -> freeable Vlastimil Babka (8): mm: compaction: encapsulate defer reset logic mm: compaction: do not mark unmovable pageblocks as skipped in async compaction mm: compaction: reset scanner positions immediately when they meet mm/compaction: cleanup isolate_freepages() mm/compaction: do not count migratepages when unnecessary mm/compaction: avoid rescanning pageblocks in isolate_freepages mm, compaction: properly signal and act upon lock and need_sched() contention mm/page_alloc: prevent MIGRATE_RESERVE pages from being misplaced Yasuaki Ishimatsu (1): mm: get rid of unnecessary pageblock scanning in setup_zone_migrate_reserve arch/tile/mm/homecache.c | 2 +- arch/unicore32/include/asm/mmu_context.h | 4 +- arch/x86/include/asm/tlbflush.h | 6 +- arch/x86/kernel/cpu/mtrr/generic.c | 4 +- arch/x86/mm/pgtable.c | 21 +- arch/x86/mm/tlb.c | 52 +--- fs/btrfs/compression.c | 2 +- fs/btrfs/extent_io.c | 15 +- fs/btrfs/file.c | 10 +- fs/buffer.c | 28 +- fs/cramfs/inode.c | 3 +- fs/exec.c | 5 +- fs/ext4/mballoc.c | 14 +- fs/f2fs/checkpoint.c | 1 - fs/f2fs/node.c | 2 - fs/fuse/dev.c | 2 +- fs/fuse/file.c | 4 - fs/gfs2/aops.c | 1 - fs/gfs2/meta_io.c | 4 +- fs/hugetlbfs/inode.c | 5 + fs/jffs2/fs.c | 2 +- fs/nfs/blocklayout/blocklayout.c | 2 +- fs/ntfs/attrib.c | 1 - fs/ntfs/file.c | 1 - fs/proc/task_mmu.c | 3 +- fs/super.c | 16 +- include/linux/compaction.h | 20 +- include/linux/cpuset.h | 56 ++-- include/linux/gfp.h | 4 +- include/linux/huge_mm.h | 4 - include/linux/hugetlb.h | 10 + include/linux/jump_label.h | 20 +- include/linux/migrate.h | 11 +- include/linux/mm.h | 11 +- include/linux/mm_types.h | 4 +- include/linux/mmzone.h | 233 ++++++++------- include/linux/page-flags.h | 6 +- include/linux/pageblock-flags.h | 33 +-- include/linux/pagemap.h | 131 +++++++-- include/linux/pagevec.h | 5 + include/linux/plist.h | 45 +++ include/linux/radix-tree.h | 5 +- include/linux/sched.h | 7 + include/linux/shmem_fs.h | 1 + include/linux/swap.h | 30 +- include/linux/swapfile.h | 2 +- include/linux/vm_event_item.h | 4 +- include/linux/vmacache.h | 38 +++ include/linux/vmstat.h | 8 + include/trace/events/compaction.h | 67 ++++- include/trace/events/kmem.h | 10 +- include/trace/events/pagemap.h | 16 +- kernel/cpuset.c | 16 +- kernel/debug/debug_core.c | 14 +- kernel/fork.c | 7 +- lib/plist.c | 52 ++++ lib/radix-tree.c | 106 ++----- mm/Makefile | 2 +- mm/compaction.c | 347 +++++++++++++---------- mm/filemap.c | 470 +++++++++++++++++++++---------- mm/fremap.c | 28 +- mm/frontswap.c | 13 +- mm/huge_memory.c | 93 ++++-- mm/hugetlb.c | 17 +- mm/internal.h | 22 +- mm/madvise.c | 2 +- mm/memory-failure.c | 4 +- mm/memory.c | 4 +- mm/memory_hotplug.c | 2 +- mm/mempolicy.c | 16 +- mm/migrate.c | 56 ++-- mm/mincore.c | 20 +- mm/mmap.c | 55 ++-- mm/nommu.c | 24 +- mm/page_alloc.c | 439 ++++++++++++++++------------- mm/readahead.c | 37 +-- mm/shmem.c | 133 +++------ mm/slab.c | 12 +- mm/slub.c | 16 +- mm/swap.c | 101 ++++++- mm/swap_state.c | 65 ++++- mm/swapfile.c | 224 ++++++++------- mm/truncate.c | 74 ++++- mm/vmacache.c | 114 ++++++++ mm/vmalloc.c | 6 +- mm/vmscan.c | 144 ++++++---- mm/vmstat.c | 13 +- 87 files changed, 2374 insertions(+), 1365 deletions(-) create mode 100644 include/linux/vmacache.h create mode 100644 mm/vmacache.c -- 1.8.4.5 -- To unsubscribe from this list: send the line "unsubscribe stable" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html