1 # SPDX-License-Identifier: GPL-2.0-only
3 menu "Memory Management options"
6 # For some reason microblaze and nios2 hard code SWAP=n. Hopefully we can
7 # add proper SWAP support to them, in which case this can be remove.
16 bool "Support for paging of anonymous memory (swap)"
17 depends on MMU && BLOCK && !ARCH_NO_SWAP
20 This option allows you to choose whether you want to have support
21 for so called swap devices or swap files in your kernel that are
22 used to provide more virtual memory than the actual RAM present
23 in your computer. If unsure say Y.
26 bool "Compressed cache for swap pages"
31 A lightweight compressed cache for swap pages. It takes
32 pages that are in the process of being swapped out and attempts to
33 compress them into a dynamically allocated RAM-based memory pool.
34 This can result in a significant I/O reduction on swap device and,
35 in the case where decompressing from RAM is faster than swap device
36 reads, can also improve workload performance.
38 config ZSWAP_DEFAULT_ON
39 bool "Enable the compressed cache for swap pages by default"
42 If selected, the compressed cache for swap pages will be enabled
43 at boot, otherwise it will be disabled.
45 The selection made here can be overridden by using the kernel
46 command line 'zswap.enabled=' option.
48 config ZSWAP_EXCLUSIVE_LOADS_DEFAULT_ON
49 bool "Invalidate zswap entries when pages are loaded"
52 If selected, exclusive loads for zswap will be enabled at boot,
53 otherwise it will be disabled.
55 If exclusive loads are enabled, when a page is loaded from zswap,
56 the zswap entry is invalidated at once, as opposed to leaving it
57 in zswap until the swap entry is freed.
59 This avoids having two copies of the same page in memory
60 (compressed and uncompressed) after faulting in a page from zswap.
61 The cost is that if the page was never dirtied and needs to be
62 swapped out again, it will be re-compressed.
65 prompt "Default compressor"
67 default ZSWAP_COMPRESSOR_DEFAULT_LZO
69 Selects the default compression algorithm for the compressed cache
72 For an overview what kind of performance can be expected from
73 a particular compression algorithm please refer to the benchmarks
74 available at the following LWN page:
75 https://lwn.net/Articles/751795/
77 If in doubt, select 'LZO'.
79 The selection made here can be overridden by using the kernel
80 command line 'zswap.compressor=' option.
82 config ZSWAP_COMPRESSOR_DEFAULT_DEFLATE
86 Use the Deflate algorithm as the default compression algorithm.
88 config ZSWAP_COMPRESSOR_DEFAULT_LZO
92 Use the LZO algorithm as the default compression algorithm.
94 config ZSWAP_COMPRESSOR_DEFAULT_842
98 Use the 842 algorithm as the default compression algorithm.
100 config ZSWAP_COMPRESSOR_DEFAULT_LZ4
104 Use the LZ4 algorithm as the default compression algorithm.
106 config ZSWAP_COMPRESSOR_DEFAULT_LZ4HC
110 Use the LZ4HC algorithm as the default compression algorithm.
112 config ZSWAP_COMPRESSOR_DEFAULT_ZSTD
116 Use the zstd algorithm as the default compression algorithm.
119 config ZSWAP_COMPRESSOR_DEFAULT
122 default "deflate" if ZSWAP_COMPRESSOR_DEFAULT_DEFLATE
123 default "lzo" if ZSWAP_COMPRESSOR_DEFAULT_LZO
124 default "842" if ZSWAP_COMPRESSOR_DEFAULT_842
125 default "lz4" if ZSWAP_COMPRESSOR_DEFAULT_LZ4
126 default "lz4hc" if ZSWAP_COMPRESSOR_DEFAULT_LZ4HC
127 default "zstd" if ZSWAP_COMPRESSOR_DEFAULT_ZSTD
131 prompt "Default allocator"
133 default ZSWAP_ZPOOL_DEFAULT_ZSMALLOC if MMU
134 default ZSWAP_ZPOOL_DEFAULT_ZBUD
136 Selects the default allocator for the compressed cache for
138 The default is 'zbud' for compatibility, however please do
139 read the description of each of the allocators below before
140 making a right choice.
142 The selection made here can be overridden by using the kernel
143 command line 'zswap.zpool=' option.
145 config ZSWAP_ZPOOL_DEFAULT_ZBUD
149 Use the zbud allocator as the default allocator.
151 config ZSWAP_ZPOOL_DEFAULT_Z3FOLD
155 Use the z3fold allocator as the default allocator.
157 config ZSWAP_ZPOOL_DEFAULT_ZSMALLOC
161 Use the zsmalloc allocator as the default allocator.
164 config ZSWAP_ZPOOL_DEFAULT
167 default "zbud" if ZSWAP_ZPOOL_DEFAULT_ZBUD
168 default "z3fold" if ZSWAP_ZPOOL_DEFAULT_Z3FOLD
169 default "zsmalloc" if ZSWAP_ZPOOL_DEFAULT_ZSMALLOC
173 tristate "2:1 compression allocator (zbud)"
176 A special purpose allocator for storing compressed pages.
177 It is designed to store up to two compressed pages per physical
178 page. While this design limits storage density, it has simple and
179 deterministic reclaim properties that make it preferable to a higher
180 density approach when reclaim will be used.
183 tristate "3:1 compression allocator (z3fold)"
186 A special purpose allocator for storing compressed pages.
187 It is designed to store up to three compressed pages per physical
188 page. It is a ZBUD derivative so the simplicity and determinism are
193 prompt "N:1 compression allocator (zsmalloc)" if ZSWAP
196 zsmalloc is a slab-based memory allocator designed to store
197 pages of various compression levels efficiently. It achieves
198 the highest storage density with the least amount of fragmentation.
201 bool "Export zsmalloc statistics"
205 This option enables code in the zsmalloc to collect various
206 statistics about what's happening in zsmalloc and exports that
207 information to userspace via debugfs.
210 config ZSMALLOC_CHAIN_SIZE
211 int "Maximum number of physical pages per-zspage"
216 This option sets the upper limit on the number of physical pages
217 that a zmalloc page (zspage) can consist of. The optimal zspage
218 chain size is calculated for each size class during the
219 initialization of the pool.
221 Changing this option can alter the characteristics of size classes,
222 such as the number of pages per zspage and the number of objects
223 per zspage. This can also result in different configurations of
224 the pool, as zsmalloc merges size classes with similar
227 For more information, see zsmalloc documentation.
229 menu "SLAB allocator options"
232 prompt "Choose SLAB allocator"
235 This option allows to select a slab allocator.
237 config SLAB_DEPRECATED
238 bool "SLAB (DEPRECATED)"
239 depends on !PREEMPT_RT
241 Deprecated and scheduled for removal in a few cycles. Replaced by
245 and the people listed in the SLAB ALLOCATOR section of MAINTAINERS
246 file, explaining why.
248 The regular slab allocator that is established and known to work
249 well in all environments. It organizes cache hot objects in
250 per cpu and per node queues.
253 bool "SLUB (Unqueued Allocator)"
255 SLUB is a slab allocator that minimizes cache line usage
256 instead of managing queues of cached objects (SLAB approach).
257 Per cpu caching is realized using slabs of objects instead
258 of queues of objects. SLUB can use memory efficiently
259 and has enhanced diagnostics. SLUB is the default choice for
267 depends on SLAB_DEPRECATED
270 bool "Configure SLUB for minimal memory footprint"
271 depends on SLUB && EXPERT
272 select SLAB_MERGE_DEFAULT
274 Configures the SLUB allocator in a way to achieve minimal memory
275 footprint, sacrificing scalability, debugging and other features.
276 This is intended only for the smallest system that had used the
277 SLOB allocator and is not recommended for systems with more than
282 config SLAB_MERGE_DEFAULT
283 bool "Allow slab caches to be merged"
285 depends on SLAB || SLUB
287 For reduced kernel memory fragmentation, slab caches can be
288 merged when they share the same size and other characteristics.
289 This carries a risk of kernel heap overflows being able to
290 overwrite objects from merged caches (and more easily control
291 cache layout), which makes such heap attacks easier to exploit
292 by attackers. By keeping caches unmerged, these kinds of exploits
293 can usually only damage objects in the same cache. To disable
294 merging at runtime, "slab_nomerge" can be passed on the kernel
297 config SLAB_FREELIST_RANDOM
298 bool "Randomize slab freelist"
299 depends on SLAB || (SLUB && !SLUB_TINY)
301 Randomizes the freelist order used on creating new pages. This
302 security feature reduces the predictability of the kernel slab
303 allocator against heap overflows.
305 config SLAB_FREELIST_HARDENED
306 bool "Harden slab freelist metadata"
307 depends on SLAB || (SLUB && !SLUB_TINY)
309 Many kernel heap attacks try to target slab cache metadata and
310 other infrastructure. This options makes minor performance
311 sacrifices to harden the kernel slab allocator against common
312 freelist exploit methods. Some slab implementations have more
313 sanity-checking than others. This option is most effective with
318 bool "Enable SLUB performance statistics"
319 depends on SLUB && SYSFS && !SLUB_TINY
321 SLUB statistics are useful to debug SLUBs allocation behavior in
322 order find ways to optimize the allocator. This should never be
323 enabled for production use since keeping statistics slows down
324 the allocator by a few percentage points. The slabinfo command
325 supports the determination of the most active slabs to figure
326 out which slabs are relevant to a particular load.
327 Try running: slabinfo -DA
329 config SLUB_CPU_PARTIAL
331 depends on SLUB && SMP && !SLUB_TINY
332 bool "SLUB per cpu partial cache"
334 Per cpu partial caches accelerate objects allocation and freeing
335 that is local to a processor at the price of more indeterminism
336 in the latency of the free. On overflow these caches will be cleared
337 which requires the taking of locks that may cause latency spikes.
338 Typically one would choose no for a realtime system.
340 config RANDOM_KMALLOC_CACHES
342 depends on SLUB && !SLUB_TINY
343 bool "Randomize slab caches for normal kmalloc"
345 A hardening feature that creates multiple copies of slab caches for
346 normal kmalloc allocation and makes kmalloc randomly pick one based
347 on code address, which makes the attackers more difficult to spray
348 vulnerable memory objects on the heap for the purpose of exploiting
349 memory vulnerabilities.
351 Currently the number of copies is set to 16, a reasonably large value
352 that effectively diverges the memory objects allocated for different
353 subsystems or modules into different caches, at the expense of a
354 limited degree of memory and CPU overhead that relates to hardware and
357 endmenu # SLAB allocator options
359 config SHUFFLE_PAGE_ALLOCATOR
360 bool "Page allocator randomization"
361 default SLAB_FREELIST_RANDOM && ACPI_NUMA
363 Randomization of the page allocator improves the average
364 utilization of a direct-mapped memory-side-cache. See section
365 5.2.27 Heterogeneous Memory Attribute Table (HMAT) in the ACPI
366 6.2a specification for an example of how a platform advertises
367 the presence of a memory-side-cache. There are also incidental
368 security benefits as it reduces the predictability of page
369 allocations to compliment SLAB_FREELIST_RANDOM, but the
370 default granularity of shuffling on the MAX_ORDER i.e, 10th
371 order of pages is selected based on cache utilization benefits
374 While the randomization improves cache utilization it may
375 negatively impact workloads on platforms without a cache. For
376 this reason, by default, the randomization is enabled only
377 after runtime detection of a direct-mapped memory-side-cache.
378 Otherwise, the randomization may be force enabled with the
379 'page_alloc.shuffle' kernel command line parameter.
384 bool "Disable heap randomization"
387 Randomizing heap placement makes heap exploits harder, but it
388 also breaks ancient binaries (including anything libc5 based).
389 This option changes the bootup default to heap randomization
390 disabled, and can be overridden at runtime by setting
391 /proc/sys/kernel/randomize_va_space to 2.
393 On non-ancient distros (post-2000 ones) N is usually a safe choice.
395 config MMAP_ALLOW_UNINITIALIZED
396 bool "Allow mmapped anonymous memory to be uninitialized"
397 depends on EXPERT && !MMU
400 Normally, and according to the Linux spec, anonymous memory obtained
401 from mmap() has its contents cleared before it is passed to
402 userspace. Enabling this config option allows you to request that
403 mmap() skip that if it is given an MAP_UNINITIALIZED flag, thus
404 providing a huge performance boost. If this option is not enabled,
405 then the flag will be ignored.
407 This is taken advantage of by uClibc's malloc(), and also by
408 ELF-FDPIC binfmt's brk and stack allocator.
410 Because of the obvious security issues, this option should only be
411 enabled on embedded devices where you control what is run in
412 userspace. Since that isn't generally a problem on no-MMU systems,
413 it is normally safe to say Y here.
415 See Documentation/admin-guide/mm/nommu-mmap.rst for more information.
417 config SELECT_MEMORY_MODEL
419 depends on ARCH_SELECT_MEMORY_MODEL
422 prompt "Memory model"
423 depends on SELECT_MEMORY_MODEL
424 default SPARSEMEM_MANUAL if ARCH_SPARSEMEM_DEFAULT
425 default FLATMEM_MANUAL
427 This option allows you to change some of the ways that
428 Linux manages its memory internally. Most users will
429 only have one option here selected by the architecture
430 configuration. This is normal.
432 config FLATMEM_MANUAL
434 depends on !ARCH_SPARSEMEM_ENABLE || ARCH_FLATMEM_ENABLE
436 This option is best suited for non-NUMA systems with
437 flat address space. The FLATMEM is the most efficient
438 system in terms of performance and resource consumption
439 and it is the best option for smaller systems.
441 For systems that have holes in their physical address
442 spaces and for features like NUMA and memory hotplug,
443 choose "Sparse Memory".
445 If unsure, choose this option (Flat Memory) over any other.
447 config SPARSEMEM_MANUAL
449 depends on ARCH_SPARSEMEM_ENABLE
451 This will be the only option for some systems, including
452 memory hot-plug systems. This is normal.
454 This option provides efficient support for systems with
455 holes is their physical address space and allows memory
456 hot-plug and hot-remove.
458 If unsure, choose "Flat Memory" over this option.
464 depends on (!SELECT_MEMORY_MODEL && ARCH_SPARSEMEM_ENABLE) || SPARSEMEM_MANUAL
468 depends on !SPARSEMEM || FLATMEM_MANUAL
471 # SPARSEMEM_EXTREME (which is the default) does some bootmem
472 # allocations when sparse_init() is called. If this cannot
473 # be done on your architecture, select this option. However,
474 # statically allocating the mem_section[] array can potentially
475 # consume vast quantities of .bss, so be careful.
477 # This option will also potentially produce smaller runtime code
478 # with gcc 3.4 and later.
480 config SPARSEMEM_STATIC
484 # Architecture platforms which require a two level mem_section in SPARSEMEM
485 # must select this option. This is usually for architecture platforms with
486 # an extremely sparse physical address space.
488 config SPARSEMEM_EXTREME
490 depends on SPARSEMEM && !SPARSEMEM_STATIC
492 config SPARSEMEM_VMEMMAP_ENABLE
495 config SPARSEMEM_VMEMMAP
496 bool "Sparse Memory virtual memmap"
497 depends on SPARSEMEM && SPARSEMEM_VMEMMAP_ENABLE
500 SPARSEMEM_VMEMMAP uses a virtually mapped memmap to optimise
501 pfn_to_page and page_to_pfn operations. This is the most
502 efficient option when sufficient kernel resources are available.
504 # Select this config option from the architecture Kconfig, if it is preferred
505 # to enable the feature of HugeTLB/dev_dax vmemmap optimization.
507 config ARCH_WANT_OPTIMIZE_DAX_VMEMMAP
510 config ARCH_WANT_OPTIMIZE_HUGETLB_VMEMMAP
513 config HAVE_MEMBLOCK_PHYS_MAP
520 # Don't discard allocated memory used to track "memory" and "reserved" memblocks
521 # after early boot, so it can still be used to test for validity of memory.
522 # Also, memblocks are updated with memory hot(un)plug.
523 config ARCH_KEEP_MEMBLOCK
526 # Keep arch NUMA mapping infrastructure post-init.
527 config NUMA_KEEP_MEMINFO
530 config MEMORY_ISOLATION
533 # IORESOURCE_SYSTEM_RAM regions in the kernel resource tree that are marked
534 # IORESOURCE_EXCLUSIVE cannot be mapped to user space, for example, via
536 config EXCLUSIVE_SYSTEM_RAM
538 depends on !DEVMEM || STRICT_DEVMEM
541 # Only be set on architectures that have completely implemented memory hotplug
542 # feature. If you are not sure, don't touch it.
544 config HAVE_BOOTMEM_INFO_NODE
547 config ARCH_ENABLE_MEMORY_HOTPLUG
550 config ARCH_ENABLE_MEMORY_HOTREMOVE
553 # eventually, we can have this option just 'select SPARSEMEM'
554 menuconfig MEMORY_HOTPLUG
555 bool "Memory hotplug"
556 select MEMORY_ISOLATION
558 depends on ARCH_ENABLE_MEMORY_HOTPLUG
560 select NUMA_KEEP_MEMINFO if NUMA
564 config MEMORY_HOTPLUG_DEFAULT_ONLINE
565 bool "Online the newly added memory blocks by default"
566 depends on MEMORY_HOTPLUG
568 This option sets the default policy setting for memory hotplug
569 onlining policy (/sys/devices/system/memory/auto_online_blocks) which
570 determines what happens to newly added memory regions. Policy setting
571 can always be changed at runtime.
572 See Documentation/admin-guide/mm/memory-hotplug.rst for more information.
574 Say Y here if you want all hot-plugged memory blocks to appear in
575 'online' state by default.
576 Say N here if you want the default policy to keep all hot-plugged
577 memory blocks in 'offline' state.
579 config MEMORY_HOTREMOVE
580 bool "Allow for memory hot remove"
581 select HAVE_BOOTMEM_INFO_NODE if (X86_64 || PPC64)
582 depends on MEMORY_HOTPLUG && ARCH_ENABLE_MEMORY_HOTREMOVE
585 config MHP_MEMMAP_ON_MEMORY
587 depends on MEMORY_HOTPLUG && SPARSEMEM_VMEMMAP
588 depends on ARCH_MHP_MEMMAP_ON_MEMORY_ENABLE
590 endif # MEMORY_HOTPLUG
592 config ARCH_MHP_MEMMAP_ON_MEMORY_ENABLE
595 # Heavily threaded applications may benefit from splitting the mm-wide
596 # page_table_lock, so that faults on different parts of the user address
597 # space can be handled with less contention: split it at this NR_CPUS.
598 # Default to 4 for wider testing, though 8 might be more appropriate.
599 # ARM's adjust_pte (unused if VIPT) depends on mm-wide page_table_lock.
600 # PA-RISC 7xxx's spinlock_t would enlarge struct page from 32 to 44 bytes.
601 # SPARC32 allocates multiple pte tables within a single page, and therefore
602 # a per-page lock leads to problems when multiple tables need to be locked
603 # at the same time (e.g. copy_page_range()).
604 # DEBUG_SPINLOCK and DEBUG_LOCK_ALLOC spinlock_t also enlarge struct page.
606 config SPLIT_PTLOCK_CPUS
608 default "999999" if !MMU
609 default "999999" if ARM && !CPU_CACHE_VIPT
610 default "999999" if PARISC && !PA20
611 default "999999" if SPARC32
614 config ARCH_ENABLE_SPLIT_PMD_PTLOCK
618 # support for memory balloon
619 config MEMORY_BALLOON
623 # support for memory balloon compaction
624 config BALLOON_COMPACTION
625 bool "Allow for balloon memory compaction/migration"
627 depends on COMPACTION && MEMORY_BALLOON
629 Memory fragmentation introduced by ballooning might reduce
630 significantly the number of 2MB contiguous memory blocks that can be
631 used within a guest, thus imposing performance penalties associated
632 with the reduced number of transparent huge pages that could be used
633 by the guest workload. Allowing the compaction & migration for memory
634 pages enlisted as being part of memory balloon devices avoids the
635 scenario aforementioned and helps improving memory defragmentation.
638 # support for memory compaction
640 bool "Allow for memory compaction"
645 Compaction is the only memory management component to form
646 high order (larger physically contiguous) memory blocks
647 reliably. The page allocator relies on compaction heavily and
648 the lack of the feature can lead to unexpected OOM killer
649 invocations for high order memory requests. You shouldn't
650 disable this option unless there really is a strong reason for
651 it and then we would be really interested to hear about that at
654 config COMPACT_UNEVICTABLE_DEFAULT
656 depends on COMPACTION
657 default 0 if PREEMPT_RT
661 # support for free page reporting
662 config PAGE_REPORTING
663 bool "Free page reporting"
666 Free page reporting allows for the incremental acquisition of
667 free pages from the buddy allocator for the purpose of reporting
668 those pages to another entity, such as a hypervisor, so that the
669 memory can be freed within the host for other uses.
672 # support for page migration
675 bool "Page migration"
677 depends on (NUMA || ARCH_ENABLE_MEMORY_HOTREMOVE || COMPACTION || CMA) && MMU
679 Allows the migration of the physical location of pages of processes
680 while the virtual addresses are not changed. This is useful in
681 two situations. The first is on NUMA systems to put pages nearer
682 to the processors accessing. The second is when allocating huge
683 pages as migration can relocate pages to satisfy a huge page
684 allocation instead of reclaiming.
686 config DEVICE_MIGRATION
687 def_bool MIGRATION && ZONE_DEVICE
689 config ARCH_ENABLE_HUGEPAGE_MIGRATION
692 config ARCH_ENABLE_THP_MIGRATION
695 config HUGETLB_PAGE_SIZE_VARIABLE
698 Allows the pageblock_order value to be dynamic instead of just standard
699 HUGETLB_PAGE_ORDER when there are multiple HugeTLB page sizes available
702 Note that the pageblock_order cannot exceed MAX_ORDER and will be
703 clamped down to MAX_ORDER.
706 def_bool (MEMORY_ISOLATION && COMPACTION) || CMA
708 config PCP_BATCH_SCALE_MAX
709 int "Maximum scale factor of PCP (Per-CPU pageset) batch allocate/free"
713 In page allocator, PCP (Per-CPU pageset) is refilled and drained in
714 batches. The batch number is scaled automatically to improve page
715 allocation/free throughput. But too large scale factor may hurt
716 latency. This option sets the upper limit of scale factor to limit
719 config PHYS_ADDR_T_64BIT
723 bool "Enable bounce buffers"
725 depends on BLOCK && MMU && HIGHMEM
727 Enable bounce buffers for devices that cannot access the full range of
728 memory available to the CPU. Enabled by default when HIGHMEM is
729 selected, but you may say n to override this.
736 bool "Enable KSM for page merging"
740 Enable Kernel Samepage Merging: KSM periodically scans those areas
741 of an application's address space that an app has advised may be
742 mergeable. When it finds pages of identical content, it replaces
743 the many instances by a single page with that content, so
744 saving memory until one or another app needs to modify the content.
745 Recommended for use with KVM, or with other duplicative applications.
746 See Documentation/mm/ksm.rst for more information: KSM is inactive
747 until a program has madvised that an area is MADV_MERGEABLE, and
748 root has set /sys/kernel/mm/ksm/run to 1 (if CONFIG_SYSFS is set).
750 config DEFAULT_MMAP_MIN_ADDR
751 int "Low address space to protect from user allocation"
755 This is the portion of low virtual memory which should be protected
756 from userspace allocation. Keeping a user from writing to low pages
757 can help reduce the impact of kernel NULL pointer bugs.
759 For most ia64, ppc64 and x86 users with lots of address space
760 a value of 65536 is reasonable and should cause no problems.
761 On arm and other archs it should not be higher than 32768.
762 Programs which use vm86 functionality or have some need to map
763 this low address space will need CAP_SYS_RAWIO or disable this
764 protection by setting the value to 0.
766 This value can be changed after boot using the
767 /proc/sys/vm/mmap_min_addr tunable.
769 config ARCH_SUPPORTS_MEMORY_FAILURE
772 config MEMORY_FAILURE
774 depends on ARCH_SUPPORTS_MEMORY_FAILURE
775 bool "Enable recovery from hardware memory errors"
776 select MEMORY_ISOLATION
779 Enables code to recover from some memory failures on systems
780 with MCA recovery. This allows a system to continue running
781 even when some of its memory has uncorrected errors. This requires
782 special hardware support and typically ECC memory.
784 config HWPOISON_INJECT
785 tristate "HWPoison pages injector"
786 depends on MEMORY_FAILURE && DEBUG_KERNEL && PROC_FS
787 select PROC_PAGE_MONITOR
789 config NOMMU_INITIAL_TRIM_EXCESS
790 int "Turn on mmap() excess space trimming before booting"
794 The NOMMU mmap() frequently needs to allocate large contiguous chunks
795 of memory on which to store mappings, but it can only ask the system
796 allocator for chunks in 2^N*PAGE_SIZE amounts - which is frequently
797 more than it requires. To deal with this, mmap() is able to trim off
798 the excess and return it to the allocator.
800 If trimming is enabled, the excess is trimmed off and returned to the
801 system allocator, which can cause extra fragmentation, particularly
802 if there are a lot of transient processes.
804 If trimming is disabled, the excess is kept, but not used, which for
805 long-term mappings means that the space is wasted.
807 Trimming can be dynamically controlled through a sysctl option
808 (/proc/sys/vm/nr_trim_pages) which specifies the minimum number of
809 excess pages there must be before trimming should occur, or zero if
810 no trimming is to occur.
812 This option specifies the initial value of this option. The default
813 of 1 says that all excess pages should be trimmed.
815 See Documentation/admin-guide/mm/nommu-mmap.rst for more information.
817 config ARCH_WANT_GENERAL_HUGETLB
820 config ARCH_WANTS_THP_SWAP
823 menuconfig TRANSPARENT_HUGEPAGE
824 bool "Transparent Hugepage Support"
825 depends on HAVE_ARCH_TRANSPARENT_HUGEPAGE && !PREEMPT_RT
829 Transparent Hugepages allows the kernel to use huge pages and
830 huge tlb transparently to the applications whenever possible.
831 This feature can improve computing performance to certain
832 applications by speeding up page faults during memory
833 allocation, by reducing the number of tlb misses and by speeding
834 up the pagetable walking.
836 If memory constrained on embedded, you may want to say N.
838 if TRANSPARENT_HUGEPAGE
841 prompt "Transparent Hugepage Support sysfs defaults"
842 depends on TRANSPARENT_HUGEPAGE
843 default TRANSPARENT_HUGEPAGE_ALWAYS
845 Selects the sysfs defaults for Transparent Hugepage Support.
847 config TRANSPARENT_HUGEPAGE_ALWAYS
850 Enabling Transparent Hugepage always, can increase the
851 memory footprint of applications without a guaranteed
852 benefit but it will work automatically for all applications.
854 config TRANSPARENT_HUGEPAGE_MADVISE
857 Enabling Transparent Hugepage madvise, will only provide a
858 performance improvement benefit to the applications using
859 madvise(MADV_HUGEPAGE) but it won't risk to increase the
860 memory footprint of applications without a guaranteed
866 depends on TRANSPARENT_HUGEPAGE && ARCH_WANTS_THP_SWAP && SWAP && 64BIT
868 Swap transparent huge pages in one piece, without splitting.
869 XXX: For now, swap cluster backing transparent huge page
870 will be split after swapout.
872 For selection by architectures with reasonable THP sizes.
874 config READ_ONLY_THP_FOR_FS
875 bool "Read-only THP for filesystems (EXPERIMENTAL)"
876 depends on TRANSPARENT_HUGEPAGE && SHMEM
879 Allow khugepaged to put read-only file-backed pages in THP.
881 This is marked experimental because it is a new feature. Write
882 support of file THPs will be developed in the next few release
885 endif # TRANSPARENT_HUGEPAGE
888 # UP and nommu archs use km based percpu allocator
890 config NEED_PER_CPU_KM
891 depends on !SMP || !MMU
895 config NEED_PER_CPU_EMBED_FIRST_CHUNK
898 config NEED_PER_CPU_PAGE_FIRST_CHUNK
901 config USE_PERCPU_NUMA_NODE_ID
904 config HAVE_SETUP_PER_CPU_AREA
908 bool "Contiguous Memory Allocator"
911 select MEMORY_ISOLATION
913 This enables the Contiguous Memory Allocator which allows other
914 subsystems to allocate big physically-contiguous blocks of memory.
915 CMA reserves a region of memory and allows only movable pages to
916 be allocated from it. This way, the kernel can use the memory for
917 pagecache and when a subsystem requests for contiguous area, the
918 allocated pages are migrated away to serve the contiguous request.
923 bool "CMA debug messages (DEVELOPMENT)"
924 depends on DEBUG_KERNEL && CMA
926 Turns on debug messages in CMA. This produces KERN_DEBUG
927 messages for every CMA call as well as various messages while
928 processing calls such as dma_alloc_from_contiguous().
929 This option does not affect warning and error messages.
932 bool "CMA debugfs interface"
933 depends on CMA && DEBUG_FS
935 Turns on the DebugFS interface for CMA.
938 bool "CMA information through sysfs interface"
939 depends on CMA && SYSFS
941 This option exposes some sysfs attributes to get information
945 int "Maximum count of the CMA areas"
950 CMA allows to create CMA areas for particular purpose, mainly,
951 used as device private area. This parameter sets the maximum
952 number of CMA area in the system.
954 If unsure, leave the default value "7" in UMA and "19" in NUMA.
956 config MEM_SOFT_DIRTY
957 bool "Track memory changes"
958 depends on CHECKPOINT_RESTORE && HAVE_ARCH_SOFT_DIRTY && PROC_FS
959 select PROC_PAGE_MONITOR
961 This option enables memory changes tracking by introducing a
962 soft-dirty bit on pte-s. This bit it set when someone writes
963 into a page just as regular dirty bit, but unlike the latter
964 it can be cleared by hands.
966 See Documentation/admin-guide/mm/soft-dirty.rst for more details.
968 config GENERIC_EARLY_IOREMAP
971 config STACK_MAX_DEFAULT_SIZE_MB
972 int "Default maximum user stack size for 32-bit processes (MB)"
975 depends on STACK_GROWSUP && (!64BIT || COMPAT)
977 This is the maximum stack size in Megabytes in the VM layout of 32-bit
978 user processes when the stack grows upwards (currently only on parisc
979 arch) when the RLIMIT_STACK hard limit is unlimited.
981 A sane initial value is 100 MB.
983 config DEFERRED_STRUCT_PAGE_INIT
984 bool "Defer initialisation of struct pages to kthreads"
986 depends on !NEED_PER_CPU_KM
990 Ordinarily all struct pages are initialised during early boot in a
991 single thread. On very large machines this can take a considerable
992 amount of time. If this option is set, large machines will bring up
993 a subset of memmap at boot and then initialise the rest in parallel.
994 This has a potential performance impact on tasks running early in the
995 lifetime of the system until these kthreads finish the
998 config PAGE_IDLE_FLAG
1000 select PAGE_EXTENSION if !64BIT
1002 This adds PG_idle and PG_young flags to 'struct page'. PTE Accessed
1003 bit writers can set the state of the bit in the flags so that PTE
1004 Accessed bit readers may avoid disturbance.
1006 config IDLE_PAGE_TRACKING
1007 bool "Enable idle page tracking"
1008 depends on SYSFS && MMU
1009 select PAGE_IDLE_FLAG
1011 This feature allows to estimate the amount of user pages that have
1012 not been touched during a given period of time. This information can
1013 be useful to tune memory cgroup limits and/or for job placement
1014 within a compute cluster.
1016 See Documentation/admin-guide/mm/idle_page_tracking.rst for
1019 config ARCH_HAS_CACHE_LINE_SIZE
1022 config ARCH_HAS_CURRENT_STACK_POINTER
1025 In support of HARDENED_USERCOPY performing stack variable lifetime
1026 checking, an architecture-agnostic way to find the stack pointer
1027 is needed. Once an architecture defines an unsigned long global
1028 register alias named "current_stack_pointer", this config can be
1031 config ARCH_HAS_PTE_DEVMAP
1034 config ARCH_HAS_ZONE_DMA_SET
1038 bool "Support DMA zone" if ARCH_HAS_ZONE_DMA_SET
1039 default y if ARM64 || X86
1042 bool "Support DMA32 zone" if ARCH_HAS_ZONE_DMA_SET
1047 bool "Device memory (pmem, HMM, etc...) hotplug support"
1048 depends on MEMORY_HOTPLUG
1049 depends on MEMORY_HOTREMOVE
1050 depends on SPARSEMEM_VMEMMAP
1051 depends on ARCH_HAS_PTE_DEVMAP
1055 Device memory hotplug support allows for establishing pmem,
1056 or other device driver discovered memory regions, in the
1057 memmap. This allows pfn_to_page() lookups of otherwise
1058 "device-physical" addresses which is needed for using a DAX
1059 mapping in an O_DIRECT operation, among other things.
1061 If FS_DAX is enabled, then say Y.
1064 # Helpers to mirror range of the CPU page tables of a process into device page
1071 config GET_FREE_REGION
1072 depends on SPARSEMEM
1075 config DEVICE_PRIVATE
1076 bool "Unaddressable device memory (GPU memory, ...)"
1077 depends on ZONE_DEVICE
1078 select GET_FREE_REGION
1081 Allows creation of struct pages to represent unaddressable device
1082 memory; i.e., memory that is only accessible from the device (or
1083 group of devices). You likely also want to select HMM_MIRROR.
1088 config ARCH_USES_HIGH_VMA_FLAGS
1090 config ARCH_HAS_PKEYS
1093 config ARCH_USES_PG_ARCH_X
1096 Enable the definition of PG_arch_x page flags with x > 1. Only
1097 suitable for 64-bit architectures with CONFIG_FLATMEM or
1098 CONFIG_SPARSEMEM_VMEMMAP enabled, otherwise there may not be
1099 enough room for additional bits in page->flags.
1101 config VM_EVENT_COUNTERS
1103 bool "Enable VM event counters for /proc/vmstat" if EXPERT
1105 VM event counters are needed for event counts to be shown.
1106 This option allows the disabling of the VM event counters
1107 on EXPERT systems. /proc/vmstat will only show page counts
1108 if VM event counters are disabled.
1111 bool "Collect percpu memory statistics"
1113 This feature collects and exposes statistics via debugfs. The
1114 information includes global and per chunk statistics, which can
1115 be used to help understand percpu memory usage.
1118 bool "Enable infrastructure for get_user_pages()-related unit tests"
1121 Provides /sys/kernel/debug/gup_test, which in turn provides a way
1122 to make ioctl calls that can launch kernel-based unit tests for
1123 the get_user_pages*() and pin_user_pages*() family of API calls.
1125 These tests include benchmark testing of the _fast variants of
1126 get_user_pages*() and pin_user_pages*(), as well as smoke tests of
1127 the non-_fast variants.
1129 There is also a sub-test that allows running dump_page() on any
1130 of up to eight pages (selected by command line args) within the
1131 range of user-space addresses. These pages are either pinned via
1132 pin_user_pages*(), or pinned via get_user_pages*(), as specified
1133 by other command line arguments.
1135 See tools/testing/selftests/mm/gup_test.c
1137 comment "GUP_TEST needs to have DEBUG_FS enabled"
1138 depends on !GUP_TEST && !DEBUG_FS
1140 config GUP_GET_PXX_LOW_HIGH
1144 tristate "Enable a module to run time tests on dma_pool"
1147 Provides a test module that will allocate and free many blocks of
1148 various sizes and report how long it takes. This is intended to
1149 provide a consistent way to measure how changes to the
1150 dma_pool_alloc/free routines affect performance.
1152 config ARCH_HAS_PTE_SPECIAL
1156 # Some architectures require a special hugepage directory format that is
1157 # required to support multiple hugepage sizes. For example a4fe3ce76
1158 # "powerpc/mm: Allow more flexible layouts for hugepage pagetables"
1159 # introduced it on powerpc. This allows for a more flexible hugepage
1160 # pagetable layouts.
1162 config ARCH_HAS_HUGEPD
1165 config MAPPING_DIRTY_HELPERS
1171 config KMAP_LOCAL_NON_LINEAR_PTE_ARRAY
1174 # struct io_mapping based helper. Selected by drivers that need them
1179 bool "Enable memfd_create() system call" if EXPERT
1183 bool "Enable memfd_secret() system call" if EXPERT
1184 depends on ARCH_HAS_SET_DIRECT_MAP
1186 Enable the memfd_secret() system call with the ability to create
1187 memory areas visible only in the context of the owning process and
1188 not mapped to other processes and other kernel page tables.
1190 config ANON_VMA_NAME
1191 bool "Anonymous VMA name support"
1192 depends on PROC_FS && ADVISE_SYSCALLS && MMU
1195 Allow naming anonymous virtual memory areas.
1197 This feature allows assigning names to virtual memory areas. Assigned
1198 names can be later retrieved from /proc/pid/maps and /proc/pid/smaps
1199 and help identifying individual anonymous memory areas.
1200 Assigning a name to anonymous virtual memory area might prevent that
1201 area from being merged with adjacent virtual memory areas due to the
1202 difference in their name.
1204 config HAVE_ARCH_USERFAULTFD_WP
1207 Arch has userfaultfd write protection support
1209 config HAVE_ARCH_USERFAULTFD_MINOR
1212 Arch has userfaultfd minor fault support
1214 menuconfig USERFAULTFD
1215 bool "Enable userfaultfd() system call"
1218 Enable the userfaultfd() system call that allows to intercept and
1219 handle page faults in userland.
1222 config PTE_MARKER_UFFD_WP
1223 bool "Userfaultfd write protection support for shmem/hugetlbfs"
1225 depends on HAVE_ARCH_USERFAULTFD_WP
1228 Allows to create marker PTEs for userfaultfd write protection
1229 purposes. It is required to enable userfaultfd write protection on
1230 file-backed memory types like shmem and hugetlbfs.
1235 bool "Multi-Gen LRU"
1237 # make sure folio->flags has enough spare bits
1238 depends on 64BIT || !SPARSEMEM || SPARSEMEM_VMEMMAP
1240 A high performance LRU implementation to overcommit memory. See
1241 Documentation/admin-guide/mm/multigen_lru.rst for details.
1243 config LRU_GEN_ENABLED
1244 bool "Enable by default"
1247 This option enables the multi-gen LRU by default.
1249 config LRU_GEN_STATS
1250 bool "Full stats for debugging"
1253 Do not enable this option unless you plan to look at historical stats
1254 from evicted generations for debugging purpose.
1256 This option has a per-memcg and per-node memory overhead.
1259 config ARCH_SUPPORTS_PER_VMA_LOCK
1264 depends on ARCH_SUPPORTS_PER_VMA_LOCK && MMU && SMP
1266 Allow per-vma locking during page fault handling.
1268 This feature allows locking each virtual memory area separately when
1269 handling page faults instead of taking mmap_lock.
1271 config LOCK_MM_AND_FIND_VMA
1273 depends on !STACK_GROWSUP
1275 source "mm/damon/Kconfig"