The page allocator tracks the number of zones that have unaccepted memory
using static_branch_enc/dec() and uses that static branch in hot paths to
determine if it needs to deal with unaccepted memory.
Borislav and Thomas pointed out that the tracking is racy: operations on
static_branch are not serialized against adding/removing unaccepted pages
to/from the zone.
Sanity checks inside static_branch machinery detects it:
WARNING: CPU: 0 PID: 10 at kernel/jump_label.c:276 __static_key_slow_dec_cpuslocked+0x8e/0xa0
The comment around the WARN() explains the problem:
/*
* Warn about the '-1' case though; since that means a
* decrement is concurrent with a first (0->1) increment. IOW
* people are trying to disable something that wasn't yet fully
* enabled. This suggests an ordering problem on the user side.
*/
The effect of this static_branch optimization is only visible on
microbenchmark.
Instead of adding more complexity around it, remove it altogether.
Link: https://lkml.kernel.org/r/20250506133207.1009676-1-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Fixes: dcdfdd40fa82 ("mm: Add support for unaccepted memory")
Link: https://lore.kernel.org/all/20250506092445.GBaBnVXXyvnazly6iF@fat_crate.local
Reported-by: Borislav Petkov <bp@alien8.de>
Tested-by: Borislav Petkov (AMD) <bp@alien8.de>
Reported-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Brendan Jackman <jackmanb@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: <stable@vger.kernel.org> [6.5+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
On XEN PV, folio_pte_batch() can incorrectly batch beyond the end of a
folio due to a corner case in pte_advance_pfn(). Specifically, when the
PFN following the folio maps to an invalidated MFN,
expected_pte = pte_advance_pfn(expected_pte, nr);
produces a pte_none(). If the actual next PTE in memory is also
pte_none(), the pte_same() succeeds,
if (!pte_same(pte, expected_pte))
break;
the loop is not broken, and batching continues into unrelated memory.
For example, with a 4-page folio, the PTE layout might look like this:
[ 53.465673] [ T2552] folio_pte_batch: printing PTE values at addr=0x7f1ac9dc5000
[ 53.465674] [ T2552] PTE[453] = 000000010085c125
[ 53.465679] [ T2552] PTE[454] = 000000010085d125
[ 53.465682] [ T2552] PTE[455] = 000000010085e125
[ 53.465684] [ T2552] PTE[456] = 000000010085f125
[ 53.465686] [ T2552] PTE[457] = 0000000000000000 <-- not present
[ 53.465689] [ T2552] PTE[458] = 0000000101da7125
pte_advance_pfn(PTE[456]) returns a pte_none() due to invalid PFN->MFN
mapping. The next actual PTE (PTE[457]) is also pte_none(), so the loop
continues and includes PTE[457] in the batch, resulting in 5 batched
entries for a 4-page folio. This triggers the following warning:
[ 53.465751] [ T2552] page: refcount:85 mapcount:20 mapping:ffff88813ff4f6a8 index:0x110 pfn:0x10085c
[ 53.465754] [ T2552] head: order:2 mapcount:80 entire_mapcount:0 nr_pages_mapped:4 pincount:0
[ 53.465756] [ T2552] memcg:ffff888003573000
[ 53.465758] [ T2552] aops:0xffffffff8226fd20 ino:82467c dentry name(?):"libc.so.6"
[ 53.465761] [ T2552] flags: 0x2000000000416c(referenced|uptodate|lru|active|private|head|node=0|zone=2)
[ 53.465764] [ T2552] raw: 002000000000416c ffffea0004021f08 ffffea0004021908 ffff88813ff4f6a8
[ 53.465767] [ T2552] raw: 0000000000000110 ffff888133d8bd40 0000005500000013 ffff888003573000
[ 53.465768] [ T2552] head: 002000000000416c ffffea0004021f08 ffffea0004021908 ffff88813ff4f6a8
[ 53.465770] [ T2552] head: 0000000000000110 ffff888133d8bd40 0000005500000013 ffff888003573000
[ 53.465772] [ T2552] head: 0020000000000202 ffffea0004021701 000000040000004f 00000000ffffffff
[ 53.465774] [ T2552] head: 0000000300000003 8000000300000002 0000000000000013 0000000000000004
[ 53.465775] [ T2552] page dumped because: VM_WARN_ON_FOLIO((_Generic((page + nr_pages - 1), const struct page *: (const struct folio *)_compound_head(page + nr_pages - 1), struct page *: (struct folio *)_compound_head(page + nr_pages - 1))) != folio)
Original code works as expected everywhere, except on XEN PV, where
pte_advance_pfn() can yield a pte_none() after balloon inflation due to
MFNs invalidation. In XEN, pte_advance_pfn() ends up calling
__pte()->xen_make_pte()->pte_pfn_to_mfn(), which returns pte_none() when
mfn == INVALID_P2M_ENTRY.
The pte_pfn_to_mfn() documents that nastiness:
If there's no mfn for the pfn, then just create an
empty non-present pte. Unfortunately this loses
information about the original pfn, so
pte_mfn_to_pfn is asymmetric.
While such hacks should certainly be removed, we can do better in
folio_pte_batch() and simply check ahead of time how many PTEs we can
possibly batch in our folio.
This way, we can not only fix the issue but cleanup the code: removing the
pte_pfn() check inside the loop body and avoiding end_ptr comparison +
arithmetic.
Link: https://lkml.kernel.org/r/20250502215019.822-2-arkamar@atlas.cz
Fixes: f8d937761d65 ("mm/memory: optimize fork() with PTE-mapped THP")
Co-developed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Petr Vaněk <arkamar@atlas.cz>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
When the last page in the zone is accepted, __accept_page() calls
static_branch_dec(). This function takes cpu_hotplug_lock, which can lead
to a deadlock if the allocation occurs during CPU bringup path as
_cpu_up() also takes the lock.
To prevent this deadlock, defer static_branch_dec() to a workqueue.
Call static_branch_dec() only when the workqueue is not yet initialized.
Workqueues are initialized before CPU bring up, so this will not conflict
with the first scenario.
Link: https://lkml.kernel.org/r/20250329171030.3942298-1-kirill.shutemov@linux.intel.com
Fixes: 55ad43e8ba0f ("mm: add a helper to accept page")
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reported-by: Srikanth Aithal <sraithal@amd.com>
Tested-by: Srikanth Aithal <sraithal@amd.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Ashish Kalra <ashish.kalra@amd.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: "Edgecombe, Rick P" <rick.p.edgecombe@intel.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: "Mike Rapoport (IBM)" <rppt@kernel.org>
Cc: Thomas Lendacky <thomas.lendacky@amd.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Pull MM updates from Andrew Morton: - The series "Enable strict percpu address space checks" from Uros Bizjak uses x86 named address space qualifiers to provide compile-time checking of percpu area accesses. This has caused a small amount of fallout - two or three issues were reported. In all cases the calling code was found to be incorrect. - The series "Some cleanup for memcg" from Chen Ridong implements some relatively monir cleanups for the memcontrol code. - The series "mm: fixes for device-exclusive entries (hmm)" from David Hildenbrand fixes a boatload of issues which David found then using device-exclusive PTE entries when THP is enabled. More work is needed, but this makes thins better - our own HMM selftests now succeed. - The series "mm: zswap: remove z3fold and zbud" from Yosry Ahmed remove the z3fold and zbud implementations. They have been deprecated for half a year and nobody has complained. - The series "mm: further simplify VMA merge operation" from Lorenzo Stoakes implements numerous simplifications in this area. No runtime effects are anticipated. - The series "mm/madvise: remove redundant mmap_lock operations from process_madvise()" from SeongJae Park rationalizes the locking in the madvise() implementation. Performance gains of 20-25% were observed in one MADV_DONTNEED microbenchmark. - The series "Tiny cleanup and improvements about SWAP code" from Baoquan He contains a number of touchups to issues which Baoquan noticed when working on the swap code. - The series "mm: kmemleak: Usability improvements" from Catalin Marinas implements a couple of improvements to the kmemleak user-visible output. - The series "mm/damon/paddr: fix large folios access and schemes handling" from Usama Arif provides a couple of fixes for DAMON's handling of large folios. - The series "mm/damon/core: fix wrong and/or useless damos_walk() behaviors" from SeongJae Park fixes a few issues with the accuracy of kdamond's walking of DAMON regions. - The series "expose mapping wrprotect, fix fb_defio use" from Lorenzo Stoakes changes the interaction between framebuffer deferred-io and core MM. No functional changes are anticipated - this is preparatory work for the future removal of page structure fields. - The series "mm/damon: add support for hugepage_size DAMOS filter" from Usama Arif adds a DAMOS filter which permits the filtering by huge page sizes. - The series "mm: permit guard regions for file-backed/shmem mappings" from Lorenzo Stoakes extends the guard region feature from its present "anon mappings only" state. The feature now covers shmem and file-backed mappings. - The series "mm: batched unmap lazyfree large folios during reclamation" from Barry Song cleans up and speeds up the unmapping for pte-mapped large folios. - The series "reimplement per-vma lock as a refcount" from Suren Baghdasaryan puts the vm_lock back into the vma. Our reasons for pulling it out were largely bogus and that change made the code more messy. This patchset provides small (0-10%) improvements on one microbenchmark. - The series "Docs/mm/damon: misc DAMOS filters documentation fixes and improves" from SeongJae Park does some maintenance work on the DAMON docs. - The series "hugetlb/CMA improvements for large systems" from Frank van der Linden addresses a pile of issues which have been observed when using CMA on large machines. - The series "mm/damon: introduce DAMOS filter type for unmapped pages" from SeongJae Park enables users of DMAON/DAMOS to filter my the page's mapped/unmapped status. - The series "zsmalloc/zram: there be preemption" from Sergey Senozhatsky teaches zram to run its compression and decompression operations preemptibly. - The series "selftests/mm: Some cleanups from trying to run them" from Brendan Jackman fixes a pile of unrelated issues which Brendan encountered while runnimg our selftests. - The series "fs/proc/task_mmu: add guard region bit to pagemap" from Lorenzo Stoakes permits userspace to use /proc/pid/pagemap to determine whether a particular page is a guard page. - The series "mm, swap: remove swap slot cache" from Kairui Song removes the swap slot cache from the allocation path - it simply wasn't being effective. - The series "mm: cleanups for device-exclusive entries (hmm)" from David Hildenbrand implements a number of unrelated cleanups in this code. - The series "mm: Rework generic PTDUMP configs" from Anshuman Khandual implements a number of preparatoty cleanups to the GENERIC_PTDUMP Kconfig logic. - The series "mm/damon: auto-tune aggregation interval" from SeongJae Park implements a feedback-driven automatic tuning feature for DAMON's aggregation interval tuning. - The series "Fix lazy mmu mode" from Ryan Roberts fixes some issues in powerpc, sparc and x86 lazy MMU implementations. Ryan did this in preparation for implementing lazy mmu mode for arm64 to optimize vmalloc. - The series "mm/page_alloc: Some clarifications for migratetype fallback" from Brendan Jackman reworks some commentary to make the code easier to follow. - The series "page_counter cleanup and size reduction" from Shakeel Butt cleans up the page_counter code and fixes a size increase which we accidentally added late last year. - The series "Add a command line option that enables control of how many threads should be used to allocate huge pages" from Thomas Prescher does that. It allows the careful operator to significantly reduce boot time by tuning the parallalization of huge page initialization. - The series "Fix calculations in trace_balance_dirty_pages() for cgwb" from Tang Yizhou fixes the tracing output from the dirty page balancing code. - The series "mm/damon: make allow filters after reject filters useful and intuitive" from SeongJae Park improves the handling of allow and reject filters. Behaviour is made more consistent and the documention is updated accordingly. - The series "Switch zswap to object read/write APIs" from Yosry Ahmed updates zswap to the new object read/write APIs and thus permits the removal of some legacy code from zpool and zsmalloc. - The series "Some trivial cleanups for shmem" from Baolin Wang does as it claims. - The series "fs/dax: Fix ZONE_DEVICE page reference counts" from Alistair Popple regularizes the weird ZONE_DEVICE page refcount handling in DAX, permittig the removal of a number of special-case checks. - The series "refactor mremap and fix bug" from Lorenzo Stoakes is a preparatoty refactoring and cleanup of the mremap() code. - The series "mm: MM owner tracking for large folios (!hugetlb) + CONFIG_NO_PAGE_MAPCOUNT" from David Hildenbrand reworks the manner in which we determine whether a large folio is known to be mapped exclusively into a single MM. - The series "mm/damon: add sysfs dirs for managing DAMOS filters based on handling layers" from SeongJae Park adds a couple of new sysfs directories to ease the management of DAMON/DAMOS filters. - The series "arch, mm: reduce code duplication in mem_init()" from Mike Rapoport consolidates many per-arch implementations of mem_init() into code generic code, where that is practical. - The series "mm/damon/sysfs: commit parameters online via damon_call()" from SeongJae Park continues the cleaning up of sysfs access to DAMON internal data. - The series "mm: page_ext: Introduce new iteration API" from Luiz Capitulino reworks the page_ext initialization to fix a boot-time crash which was observed with an unusual combination of compile and cmdline options. - The series "Buddy allocator like (or non-uniform) folio split" from Zi Yan reworks the code to split a folio into smaller folios. The main benefit is lessened memory consumption: fewer post-split folios are generated. - The series "Minimize xa_node allocation during xarry split" from Zi Yan reduces the number of xarray xa_nodes which are generated during an xarray split. - The series "drivers/base/memory: Two cleanups" from Gavin Shan performs some maintenance work on the drivers/base/memory code. - The series "Add tracepoints for lowmem reserves, watermarks and totalreserve_pages" from Martin Liu adds some more tracepoints to the page allocator code. - The series "mm/madvise: cleanup requests validations and classifications" from SeongJae Park cleans up some warts which SeongJae observed during his earlier madvise work. - The series "mm/hwpoison: Fix regressions in memory failure handling" from Shuai Xue addresses two quite serious regressions which Shuai has observed in the memory-failure implementation. - The series "mm: reliable huge page allocator" from Johannes Weiner makes huge page allocations cheaper and more reliable by reducing fragmentation. - The series "Minor memcg cleanups & prep for memdescs" from Matthew Wilcox is preparatory work for the future implementation of memdescs. - The series "track memory used by balloon drivers" from Nico Pache introduces a way to track memory used by our various balloon drivers. - The series "mm/damon: introduce DAMOS filter type for active pages" from Nhat Pham permits users to filter for active/inactive pages, separately for file and anon pages. - The series "Adding Proactive Memory Reclaim Statistics" from Hao Jia separates the proactive reclaim statistics from the direct reclaim statistics. - The series "mm/vmscan: don't try to reclaim hwpoison folio" from Jinjiang Tu fixes our handling of hwpoisoned pages within the reclaim code. * tag 'mm-stable-2025-03-30-16-52' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm: (431 commits) mm/page_alloc: remove unnecessary __maybe_unused in order_to_pindex() x86/mm: restore early initialization of high_memory for 32-bits mm/vmscan: don't try to reclaim hwpoison folio mm/hwpoison: introduce folio_contain_hwpoisoned_page() helper cgroup: docs: add pswpin and pswpout items in cgroup v2 doc mm: vmscan: split proactive reclaim statistics from direct reclaim statistics selftests/mm: speed up split_huge_page_test selftests/mm: uffd-unit-tests support for hugepages > 2M docs/mm/damon/design: document active DAMOS filter type mm/damon: implement a new DAMOS filter type for active pages fs/dax: don't disassociate zero page entries MM documentation: add "Unaccepted" meminfo entry selftests/mm: add commentary about 9pfs bugs fork: use __vmalloc_node() for stack allocation docs/mm: Physical Memory: Populate the "Zones" section xen: balloon: update the NR_BALLOON_PAGES state hv_balloon: update the NR_BALLOON_PAGES state balloon_compaction: update the NR_BALLOON_PAGES state meminfo: add a per node counter for balloon drivers mm: remove references to folio in __memcg_kmem_uncharge_page() ...
Pull bpf try_alloc_pages() support from Alexei Starovoitov:
"The pull includes work from Sebastian, Vlastimil and myself with a lot
of help from Michal and Shakeel.
This is a first step towards making kmalloc reentrant to get rid of
slab wrappers: bpf_mem_alloc, kretprobe's objpool, etc. These patches
make page allocator safe from any context.
Vlastimil kicked off this effort at LSFMM 2024:
https://lwn.net/Articles/974138/
and we continued at LSFMM 2025:
https://lore.kernel.org/all/CAADnVQKfkGxudNUkcPJgwe3nTZ=xohnRshx9kLZBTmR_E1DFEg@mail.gmail.com/
Why:
SLAB wrappers bind memory to a particular subsystem making it
unavailable to the rest of the kernel. Some BPF maps in production
consume Gbytes of preallocated memory. Top 5 in Meta: 1.5G, 1.2G,
1.1G, 300M, 200M. Once we have kmalloc that works in any context BPF
map preallocation won't be necessary.
How:
Synchronous kmalloc/page alloc stack has multiple stages going from
fast to slow: cmpxchg16 -> slab_alloc -> new_slab -> alloc_pages ->
rmqueue_pcplist -> __rmqueue, where rmqueue_pcplist was already
relying on trylock.
This set changes rmqueue_bulk/rmqueue_buddy to attempt a trylock and
return ENOMEM if alloc_flags & ALLOC_TRYLOCK. It then wraps this
functionality into try_alloc_pages() helper. We make sure that the
logic is sane in PREEMPT_RT.
End result: try_alloc_pages()/free_pages_nolock() are safe to call
from any context.
try_kmalloc() for any context with similar trylock approach will
follow. It will use try_alloc_pages() when slab needs a new page.
Though such try_kmalloc/page_alloc() is an opportunistic allocator,
this design ensures that the probability of successful allocation of
small objects (up to one page in size) is high.
Even before we have try_kmalloc(), we already use try_alloc_pages() in
BPF arena implementation and it's going to be used more extensively in
BPF"
* tag 'bpf_try_alloc_pages' of git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next:
mm: Fix the flipped condition in gfpflags_allow_spinning()
bpf: Use try_alloc_pages() to allocate pages for bpf needs.
mm, bpf: Use memcg in try_alloc_pages().
memcg: Use trylock to access memcg stock_lock.
mm, bpf: Introduce free_pages_nolock()
mm, bpf: Introduce try_alloc_pages() for opportunistic page allocation
locking/local_lock: Introduce localtry_lock_t
Pull sysctl updates from Joel Granados:
- Move vm_table members out of kernel/sysctl.c
All vm_table array members have moved to their respective subsystems
leading to the removal of vm_table from kernel/sysctl.c. This
increases modularity by placing the ctl_tables closer to where they
are actually used and at the same time reducing the chances of merge
conflicts in kernel/sysctl.c.
- ctl_table range fixes
Replace the proc_handler function that checks variable ranges in
coredump_sysctls and vdso_table with the one that actually uses the
extra{1,2} pointers as min/max values. This tightens the range of the
values that users can pass into the kernel effectively preventing
{under,over}flows.
- Misc fixes
Correct grammar errors and typos in test messages. Update sysctl
files in MAINTAINERS. Constified and removed array size in
declaration for alignment_tbl
* tag 'sysctl-6.15-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/sysctl/sysctl: (22 commits)
selftests/sysctl: fix wording of help messages
selftests: fix spelling/grammar errors in sysctl/sysctl.sh
MAINTAINERS: Update sysctl file list in MAINTAINERS
sysctl: Fix underflow value setting risk in vm_table
coredump: Fixes core_pipe_limit sysctl proc_handler
sysctl: remove unneeded include
sysctl: remove the vm_table
sh: vdso: move the sysctl to arch/sh/kernel/vsyscall/vsyscall.c
x86: vdso: move the sysctl to arch/x86/entry/vdso/vdso32-setup.c
fs: dcache: move the sysctl to fs/dcache.c
sunrpc: simplify rpcauth_cache_shrink_count()
fs: drop_caches: move sysctl to fs/drop_caches.c
fs: fs-writeback: move sysctl to fs/fs-writeback.c
mm: nommu: move sysctl to mm/nommu.c
security: min_addr: move sysctl to security/min_addr.c
mm: mmap: move sysctl to mm/mmap.c
mm: util: move sysctls to mm/util.c
mm: vmscan: move vmscan sysctls to mm/vmscan.c
mm: swap: move sysctl to mm/swap.c
mm: filemap: move sysctl to mm/filemap.c
...
__init_reserved_page_zone() function finds the zone for pfn and nid and performs initialization of a struct page with that zone and nid. There is nothing in that function about reserved pages and it is misnamed. Rename it to __init_page_from_nid() to better reflect what the function does. Link: https://lkml.kernel.org/r/20250225083017.567649-2-rppt@kernel.org Signed-off-by: Mike Rapoport (Microsoft) <rppt@kernel.org> Reviewed-by: Wei Yang <richard.weiyang@gmail.com> Cc: Frank van der Linden <fvdl@google.com> Cc: Muchun Song <muchun.song@linux.dev> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
The previous patch added pageblock_order reclaim to kswapd/kcompactd,
which helps, but produces only one block at a time. Allocation stalls and
THP failure rates are still higher than they could be.
To adequately reflect ALLOC_NOFRAGMENT demand for pageblocks, change the
watermarking for kswapd & kcompactd: instead of targeting the high
watermark in order-0 pages and checking for one suitable block, simply
require that the high watermark is entirely met in pageblocks.
To this end, track the number of free pages within contiguous pageblocks,
then change pgdat_balanced() and compact_finished() to check watermarks
against this new value.
This further reduces THP latencies and allocation stalls, and improves THP
success rates against the previous patch:
DEFRAGMODE-ASYNC DEFRAGMODE-ASYNC-WMARKS
Hugealloc Time mean 34300.36 ( +0.00%) 28904.00 ( -15.73%)
Hugealloc Time stddev 36390.42 ( +0.00%) 33464.37 ( -8.04%)
Kbuild Real time 196.13 ( +0.00%) 196.59 ( +0.23%)
Kbuild User time 1234.74 ( +0.00%) 1231.67 ( -0.25%)
Kbuild System time 62.62 ( +0.00%) 59.10 ( -5.54%)
THP fault alloc 57054.53 ( +0.00%) 63223.67 ( +10.81%)
THP fault fallback 11581.40 ( +0.00%) 5412.47 ( -53.26%)
Direct compact fail 107.80 ( +0.00%) 59.07 ( -44.79%)
Direct compact success 4.53 ( +0.00%) 2.80 ( -31.33%)
Direct compact success rate % 3.20 ( +0.00%) 3.99 ( +18.66%)
Compact daemon scanned migrate 5461033.93 ( +0.00%) 2267500.33 ( -58.48%)
Compact daemon scanned free 5824897.93 ( +0.00%) 2339773.00 ( -59.83%)
Compact direct scanned migrate 58336.93 ( +0.00%) 47659.93 ( -18.30%)
Compact direct scanned free 32791.87 ( +0.00%) 40729.67 ( +24.21%)
Compact total migrate scanned 5519370.87 ( +0.00%) 2315160.27 ( -58.05%)
Compact total free scanned 5857689.80 ( +0.00%) 2380502.67 ( -59.36%)
Alloc stall 2424.60 ( +0.00%) 638.87 ( -73.62%)
Pages kswapd scanned 2657018.33 ( +0.00%) 4002186.33 ( +50.63%)
Pages kswapd reclaimed 559583.07 ( +0.00%) 718577.80 ( +28.41%)
Pages direct scanned 722094.07 ( +0.00%) 355172.73 ( -50.81%)
Pages direct reclaimed 107257.80 ( +0.00%) 31162.80 ( -70.95%)
Pages total scanned 3379112.40 ( +0.00%) 4357359.07 ( +28.95%)
Pages total reclaimed 666840.87 ( +0.00%) 749740.60 ( +12.43%)
Swap out 77238.20 ( +0.00%) 110084.33 ( +42.53%)
Swap in 11712.80 ( +0.00%) 24457.00 ( +108.80%)
File refaults 143438.80 ( +0.00%) 188226.93 ( +31.22%)
Also of note is that compaction work overall is reduced. The reason for
this is that when free pageblocks are more readily available, allocations
are also much more likely to get physically placed in LRU order, instead
of being forced to scavenge free space here and there. This means that
reclaim by itself has better chances of freeing up whole blocks, and the
system relies less on compaction.
Comparing all changes to the vanilla kernel:
VANILLA DEFRAGMODE-ASYNC-WMARKS
Hugealloc Time mean 52739.45 ( +0.00%) 28904.00 ( -45.19%)
Hugealloc Time stddev 56541.26 ( +0.00%) 33464.37 ( -40.81%)
Kbuild Real time 197.47 ( +0.00%) 196.59 ( -0.44%)
Kbuild User time 1240.49 ( +0.00%) 1231.67 ( -0.71%)
Kbuild System time 70.08 ( +0.00%) 59.10 ( -15.45%)
THP fault alloc 46727.07 ( +0.00%) 63223.67 ( +35.30%)
THP fault fallback 21910.60 ( +0.00%) 5412.47 ( -75.29%)
Direct compact fail 195.80 ( +0.00%) 59.07 ( -69.48%)
Direct compact success 7.93 ( +0.00%) 2.80 ( -57.46%)
Direct compact success rate % 3.51 ( +0.00%) 3.99 ( +10.49%)
Compact daemon scanned migrate 3369601.27 ( +0.00%) 2267500.33 ( -32.71%)
Compact daemon scanned free 5075474.47 ( +0.00%) 2339773.00 ( -53.90%)
Compact direct scanned migrate 161787.27 ( +0.00%) 47659.93 ( -70.54%)
Compact direct scanned free 163467.53 ( +0.00%) 40729.67 ( -75.08%)
Compact total migrate scanned 3531388.53 ( +0.00%) 2315160.27 ( -34.44%)
Compact total free scanned 5238942.00 ( +0.00%) 2380502.67 ( -54.56%)
Alloc stall 2371.07 ( +0.00%) 638.87 ( -73.02%)
Pages kswapd scanned 2160926.73 ( +0.00%) 4002186.33 ( +85.21%)
Pages kswapd reclaimed 533191.07 ( +0.00%) 718577.80 ( +34.77%)
Pages direct scanned 400450.33 ( +0.00%) 355172.73 ( -11.31%)
Pages direct reclaimed 94441.73 ( +0.00%) 31162.80 ( -67.00%)
Pages total scanned 2561377.07 ( +0.00%) 4357359.07 ( +70.12%)
Pages total reclaimed 627632.80 ( +0.00%) 749740.60 ( +19.46%)
Swap out 47959.53 ( +0.00%) 110084.33 ( +129.53%)
Swap in 7276.00 ( +0.00%) 24457.00 ( +236.10%)
File refaults 138043.00 ( +0.00%) 188226.93 ( +36.35%)
THP allocation latencies and %sys time are down dramatically.
THP allocation failures are down from nearly 50% to 8.5%. And to recall
previous data points, the success rates are steady and reliable without
the cumulative deterioration of fragmentation events.
Compaction work is down overall. Direct compaction work especially is
drastically reduced. As an aside, its success rate of 4% indicates there
is room for improvement. For now it's good to rely on it less.
Reclaim work is up overall, however direct reclaim work is down. Part of
the increase can be attributed to a higher use of THPs, which due to
internal fragmentation increase the memory footprint. This is not
necessarily an unexpected side-effect for users of THP.
However, taken both points together, there may well be some opportunities
for fine tuning in the reclaim/compaction coordination.
[hannes@cmpxchg.org: fix squawks from rebasing]
Link: https://lkml.kernel.org/r/20250314210558.GD1316033@cmpxchg.org
Link: https://lkml.kernel.org/r/20250313210647.1314586-6-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
The point where the memory is released from memblock to the buddy allocator is hidden inside arch-specific mem_init()s and the call to memblock_free_all() is needlessly duplicated in every artiste cure and after introduction of arch_mm_preinit() hook, mem_init() implementation on many architecture only contains the call to memblock_free_all(). Pull memblock_free_all() call into mm_core_init() and drop mem_init() on relevant architectures to make it more explicit where the free memory is released from memblock to the buddy allocator and to reduce code duplication in architecture specific code. Link: https://lkml.kernel.org/r/20250313135003.836600-14-rppt@kernel.org Signed-off-by: Mike Rapoport (Microsoft) <rppt@kernel.org> Acked-by: Dave Hansen <dave.hansen@linux.intel.com> [x86] Acked-by: Geert Uytterhoeven <geert@linux-m68k.org> [m68k] Tested-by: Mark Brown <broonie@kernel.org> Cc: Alexander Gordeev <agordeev@linux.ibm.com> Cc: Andreas Larsson <andreas@gaisler.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: Ard Biesheuvel <ardb@kernel.org> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Borislav Betkov <bp@alien8.de> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: David S. Miller <davem@davemloft.net> Cc: Dinh Nguyen <dinguyen@kernel.org> Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com> Cc: Guo Ren (csky) <guoren@kernel.org> Cc: Heiko Carstens <hca@linux.ibm.com> Cc: Helge Deller <deller@gmx.de> Cc: Huacai Chen <chenhuacai@kernel.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiaxun Yang <jiaxun.yang@flygoat.com> Cc: Johannes Berg <johannes@sipsolutions.net> Cc: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de> Cc: Madhavan Srinivasan <maddy@linux.ibm.com> Cc: Matt Turner <mattst88@gmail.com> Cc: Max Filippov <jcmvbkbc@gmail.com> Cc: Michael Ellerman <mpe@ellerman.id.au> Cc: Michal Simek <monstr@monstr.eu> Cc: Palmer Dabbelt <palmer@dabbelt.com> Cc: Richard Weinberger <richard@nod.at> Cc: Russel King <linux@armlinux.org.uk> Cc: Stafford Horne <shorne@gmail.com> Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de> Cc: Thomas Gleinxer <tglx@linutronix.de> Cc: Vasily Gorbik <gor@linux.ibm.com> Cc: Vineet Gupta <vgupta@kernel.org> Cc: Will Deacon <will@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Everything is in place to stop using the per-page mapcounts in large
folios: the mapcount of tail pages will always be logically 0 (-1 value),
just like it currently is for hugetlb folios already, and the page
mapcount of the head page is either 0 (-1 value) or contains a page type
(e.g., hugetlb).
Maintaining _nr_pages_mapped without per-page mapcounts is impossible, so
that one also has to go with CONFIG_NO_PAGE_MAPCOUNT.
There are two remaining implications:
(1) Per-node, per-cgroup and per-lruvec stats of "NR_ANON_MAPPED"
("mapped anonymous memory") and "NR_FILE_MAPPED"
("mapped file memory"):
As soon as any page of the folio is mapped -- folio_mapped() -- we
now account the complete folio as mapped. Once the last page is
unmapped -- !folio_mapped() -- we account the complete folio as
unmapped.
This implies that ...
* "AnonPages" and "Mapped" in /proc/meminfo and
/sys/devices/system/node/*/meminfo
* cgroup v2: "anon" and "file_mapped" in "memory.stat" and
"memory.numa_stat"
* cgroup v1: "rss" and "mapped_file" in "memory.stat" and
"memory.numa_stat
... can now appear higher than before. But note that these folios do
consume that memory, simply not all pages are actually currently
mapped.
It's worth nothing that other accounting in the kernel (esp. cgroup
charging on allocation) is not affected by this change.
[why oh why is "anon" called "rss" in cgroup v1]
(2) Detecting partial mappings
Detecting whether anon THPs are partially mapped gets a bit more
unreliable. As long as a single MM maps such a large folio
("exclusively mapped"), we can reliably detect it. Especially before
fork() / after a short-lived child process quit, we will detect
partial mappings reliably, which is the common case.
In essence, if the average per-page mapcount in an anon THP is < 1,
we know for sure that we have a partial mapping.
However, as soon as multiple MMs are involved, we might miss detecting
partial mappings: this might be relevant with long-lived child
processes. If we have a fully-mapped anon folio before fork(), once
our child processes and our parent all unmap (zap/COW) the same pages
(but not the complete folio), we might not detect the partial mapping.
However, once the child processes quit we would detect the partial
mapping.
How relevant this case is in practice remains to be seen.
Swapout/migration will likely mitigate this.
In the future, RMAP walkers could check for that for that case
(e.g., when collecting access bits during reclaim) and simply flag
them for deferred-splitting.
Link: https://lkml.kernel.org/r/20250303163014.1128035-21-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Andy Lutomirks^H^Hski <luto@kernel.org>
Cc: Borislav Betkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Matthew Wilcow (Oracle) <willy@infradead.org>
Cc: Michal Koutn <mkoutny@suse.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: tejun heo <tj@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Zefan Li <lizefan.x@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>