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_SHRINKER_DEFAULT_ON
49 bool "Shrink the zswap pool on memory pressure"
53 If selected, the zswap shrinker will be enabled, and the pages
54 stored in the zswap pool will become available for reclaim (i.e
55 written back to the backing swap device) on memory pressure.
57 This means that zswap writeback could happen even if the pool is
58 not yet full, or the cgroup zswap limit has not been reached,
59 reducing the chance that cold pages will reside in the zswap pool
60 and consume memory indefinitely.
63 prompt "Default compressor"
65 default ZSWAP_COMPRESSOR_DEFAULT_LZO
67 Selects the default compression algorithm for the compressed cache
70 For an overview what kind of performance can be expected from
71 a particular compression algorithm please refer to the benchmarks
72 available at the following LWN page:
73 https://lwn.net/Articles/751795/
75 If in doubt, select 'LZO'.
77 The selection made here can be overridden by using the kernel
78 command line 'zswap.compressor=' option.
80 config ZSWAP_COMPRESSOR_DEFAULT_DEFLATE
84 Use the Deflate algorithm as the default compression algorithm.
86 config ZSWAP_COMPRESSOR_DEFAULT_LZO
90 Use the LZO algorithm as the default compression algorithm.
92 config ZSWAP_COMPRESSOR_DEFAULT_842
96 Use the 842 algorithm as the default compression algorithm.
98 config ZSWAP_COMPRESSOR_DEFAULT_LZ4
102 Use the LZ4 algorithm as the default compression algorithm.
104 config ZSWAP_COMPRESSOR_DEFAULT_LZ4HC
108 Use the LZ4HC algorithm as the default compression algorithm.
110 config ZSWAP_COMPRESSOR_DEFAULT_ZSTD
114 Use the zstd algorithm as the default compression algorithm.
117 config ZSWAP_COMPRESSOR_DEFAULT
120 default "deflate" if ZSWAP_COMPRESSOR_DEFAULT_DEFLATE
121 default "lzo" if ZSWAP_COMPRESSOR_DEFAULT_LZO
122 default "842" if ZSWAP_COMPRESSOR_DEFAULT_842
123 default "lz4" if ZSWAP_COMPRESSOR_DEFAULT_LZ4
124 default "lz4hc" if ZSWAP_COMPRESSOR_DEFAULT_LZ4HC
125 default "zstd" if ZSWAP_COMPRESSOR_DEFAULT_ZSTD
129 prompt "Default allocator"
131 default ZSWAP_ZPOOL_DEFAULT_ZSMALLOC if HAVE_ZSMALLOC
132 default ZSWAP_ZPOOL_DEFAULT_ZBUD
134 Selects the default allocator for the compressed cache for
136 The default is 'zbud' for compatibility, however please do
137 read the description of each of the allocators below before
138 making a right choice.
140 The selection made here can be overridden by using the kernel
141 command line 'zswap.zpool=' option.
143 config ZSWAP_ZPOOL_DEFAULT_ZBUD
147 Use the zbud allocator as the default allocator.
149 config ZSWAP_ZPOOL_DEFAULT_Z3FOLD
153 Use the z3fold allocator as the default allocator.
155 config ZSWAP_ZPOOL_DEFAULT_ZSMALLOC
157 depends on HAVE_ZSMALLOC
160 Use the zsmalloc allocator as the default allocator.
163 config ZSWAP_ZPOOL_DEFAULT
166 default "zbud" if ZSWAP_ZPOOL_DEFAULT_ZBUD
167 default "z3fold" if ZSWAP_ZPOOL_DEFAULT_Z3FOLD
168 default "zsmalloc" if ZSWAP_ZPOOL_DEFAULT_ZSMALLOC
172 tristate "2:1 compression allocator (zbud)"
175 A special purpose allocator for storing compressed pages.
176 It is designed to store up to two compressed pages per physical
177 page. While this design limits storage density, it has simple and
178 deterministic reclaim properties that make it preferable to a higher
179 density approach when reclaim will be used.
182 tristate "3:1 compression allocator (z3fold)"
185 A special purpose allocator for storing compressed pages.
186 It is designed to store up to three compressed pages per physical
187 page. It is a ZBUD derivative so the simplicity and determinism are
193 depends on PAGE_SIZE_LESS_THAN_256KB # we want <= 64 KiB
197 prompt "N:1 compression allocator (zsmalloc)" if ZSWAP
198 depends on HAVE_ZSMALLOC
200 zsmalloc is a slab-based memory allocator designed to store
201 pages of various compression levels efficiently. It achieves
202 the highest storage density with the least amount of fragmentation.
205 bool "Export zsmalloc statistics"
209 This option enables code in the zsmalloc to collect various
210 statistics about what's happening in zsmalloc and exports that
211 information to userspace via debugfs.
214 config ZSMALLOC_CHAIN_SIZE
215 int "Maximum number of physical pages per-zspage"
220 This option sets the upper limit on the number of physical pages
221 that a zmalloc page (zspage) can consist of. The optimal zspage
222 chain size is calculated for each size class during the
223 initialization of the pool.
225 Changing this option can alter the characteristics of size classes,
226 such as the number of pages per zspage and the number of objects
227 per zspage. This can also result in different configurations of
228 the pool, as zsmalloc merges size classes with similar
231 For more information, see zsmalloc documentation.
233 menu "Slab allocator options"
239 bool "Configure for minimal memory footprint"
241 select SLAB_MERGE_DEFAULT
243 Configures the slab allocator in a way to achieve minimal memory
244 footprint, sacrificing scalability, debugging and other features.
245 This is intended only for the smallest system that had used the
246 SLOB allocator and is not recommended for systems with more than
251 config SLAB_MERGE_DEFAULT
252 bool "Allow slab caches to be merged"
255 For reduced kernel memory fragmentation, slab caches can be
256 merged when they share the same size and other characteristics.
257 This carries a risk of kernel heap overflows being able to
258 overwrite objects from merged caches (and more easily control
259 cache layout), which makes such heap attacks easier to exploit
260 by attackers. By keeping caches unmerged, these kinds of exploits
261 can usually only damage objects in the same cache. To disable
262 merging at runtime, "slab_nomerge" can be passed on the kernel
265 config SLAB_FREELIST_RANDOM
266 bool "Randomize slab freelist"
267 depends on !SLUB_TINY
269 Randomizes the freelist order used on creating new pages. This
270 security feature reduces the predictability of the kernel slab
271 allocator against heap overflows.
273 config SLAB_FREELIST_HARDENED
274 bool "Harden slab freelist metadata"
275 depends on !SLUB_TINY
277 Many kernel heap attacks try to target slab cache metadata and
278 other infrastructure. This options makes minor performance
279 sacrifices to harden the kernel slab allocator against common
280 freelist exploit methods.
283 bool "Support allocation from separate kmalloc buckets"
284 depends on !SLUB_TINY
285 default SLAB_FREELIST_HARDENED
287 Kernel heap attacks frequently depend on being able to create
288 specifically-sized allocations with user-controlled contents
289 that will be allocated into the same kmalloc bucket as a
290 target object. To avoid sharing these allocation buckets,
291 provide an explicitly separated set of buckets to be used for
292 user-controlled allocations. This may very slightly increase
293 memory fragmentation, though in practice it's only a handful
294 of extra pages since the bulk of user-controlled allocations
295 are relatively long-lived.
301 bool "Enable performance statistics"
302 depends on SYSFS && !SLUB_TINY
304 The statistics are useful to debug slab allocation behavior in
305 order find ways to optimize the allocator. This should never be
306 enabled for production use since keeping statistics slows down
307 the allocator by a few percentage points. The slabinfo command
308 supports the determination of the most active slabs to figure
309 out which slabs are relevant to a particular load.
310 Try running: slabinfo -DA
312 config SLUB_CPU_PARTIAL
314 depends on SMP && !SLUB_TINY
315 bool "Enable per cpu partial caches"
317 Per cpu partial caches accelerate objects allocation and freeing
318 that is local to a processor at the price of more indeterminism
319 in the latency of the free. On overflow these caches will be cleared
320 which requires the taking of locks that may cause latency spikes.
321 Typically one would choose no for a realtime system.
323 config RANDOM_KMALLOC_CACHES
325 depends on !SLUB_TINY
326 bool "Randomize slab caches for normal kmalloc"
328 A hardening feature that creates multiple copies of slab caches for
329 normal kmalloc allocation and makes kmalloc randomly pick one based
330 on code address, which makes the attackers more difficult to spray
331 vulnerable memory objects on the heap for the purpose of exploiting
332 memory vulnerabilities.
334 Currently the number of copies is set to 16, a reasonably large value
335 that effectively diverges the memory objects allocated for different
336 subsystems or modules into different caches, at the expense of a
337 limited degree of memory and CPU overhead that relates to hardware and
340 endmenu # Slab allocator options
342 config SHUFFLE_PAGE_ALLOCATOR
343 bool "Page allocator randomization"
344 default SLAB_FREELIST_RANDOM && ACPI_NUMA
346 Randomization of the page allocator improves the average
347 utilization of a direct-mapped memory-side-cache. See section
348 5.2.27 Heterogeneous Memory Attribute Table (HMAT) in the ACPI
349 6.2a specification for an example of how a platform advertises
350 the presence of a memory-side-cache. There are also incidental
351 security benefits as it reduces the predictability of page
352 allocations to compliment SLAB_FREELIST_RANDOM, but the
353 default granularity of shuffling on the MAX_PAGE_ORDER i.e, 10th
354 order of pages is selected based on cache utilization benefits
357 While the randomization improves cache utilization it may
358 negatively impact workloads on platforms without a cache. For
359 this reason, by default, the randomization is not enabled even
360 if SHUFFLE_PAGE_ALLOCATOR=y. The randomization may be force enabled
361 with the 'page_alloc.shuffle' kernel command line parameter.
366 bool "Disable heap randomization"
369 Randomizing heap placement makes heap exploits harder, but it
370 also breaks ancient binaries (including anything libc5 based).
371 This option changes the bootup default to heap randomization
372 disabled, and can be overridden at runtime by setting
373 /proc/sys/kernel/randomize_va_space to 2.
375 On non-ancient distros (post-2000 ones) N is usually a safe choice.
377 config MMAP_ALLOW_UNINITIALIZED
378 bool "Allow mmapped anonymous memory to be uninitialized"
379 depends on EXPERT && !MMU
382 Normally, and according to the Linux spec, anonymous memory obtained
383 from mmap() has its contents cleared before it is passed to
384 userspace. Enabling this config option allows you to request that
385 mmap() skip that if it is given an MAP_UNINITIALIZED flag, thus
386 providing a huge performance boost. If this option is not enabled,
387 then the flag will be ignored.
389 This is taken advantage of by uClibc's malloc(), and also by
390 ELF-FDPIC binfmt's brk and stack allocator.
392 Because of the obvious security issues, this option should only be
393 enabled on embedded devices where you control what is run in
394 userspace. Since that isn't generally a problem on no-MMU systems,
395 it is normally safe to say Y here.
397 See Documentation/admin-guide/mm/nommu-mmap.rst for more information.
399 config SELECT_MEMORY_MODEL
401 depends on ARCH_SELECT_MEMORY_MODEL
404 prompt "Memory model"
405 depends on SELECT_MEMORY_MODEL
406 default SPARSEMEM_MANUAL if ARCH_SPARSEMEM_DEFAULT
407 default FLATMEM_MANUAL
409 This option allows you to change some of the ways that
410 Linux manages its memory internally. Most users will
411 only have one option here selected by the architecture
412 configuration. This is normal.
414 config FLATMEM_MANUAL
416 depends on !ARCH_SPARSEMEM_ENABLE || ARCH_FLATMEM_ENABLE
418 This option is best suited for non-NUMA systems with
419 flat address space. The FLATMEM is the most efficient
420 system in terms of performance and resource consumption
421 and it is the best option for smaller systems.
423 For systems that have holes in their physical address
424 spaces and for features like NUMA and memory hotplug,
425 choose "Sparse Memory".
427 If unsure, choose this option (Flat Memory) over any other.
429 config SPARSEMEM_MANUAL
431 depends on ARCH_SPARSEMEM_ENABLE
433 This will be the only option for some systems, including
434 memory hot-plug systems. This is normal.
436 This option provides efficient support for systems with
437 holes is their physical address space and allows memory
438 hot-plug and hot-remove.
440 If unsure, choose "Flat Memory" over this option.
446 depends on (!SELECT_MEMORY_MODEL && ARCH_SPARSEMEM_ENABLE) || SPARSEMEM_MANUAL
450 depends on !SPARSEMEM || FLATMEM_MANUAL
453 # SPARSEMEM_EXTREME (which is the default) does some bootmem
454 # allocations when sparse_init() is called. If this cannot
455 # be done on your architecture, select this option. However,
456 # statically allocating the mem_section[] array can potentially
457 # consume vast quantities of .bss, so be careful.
459 # This option will also potentially produce smaller runtime code
460 # with gcc 3.4 and later.
462 config SPARSEMEM_STATIC
466 # Architecture platforms which require a two level mem_section in SPARSEMEM
467 # must select this option. This is usually for architecture platforms with
468 # an extremely sparse physical address space.
470 config SPARSEMEM_EXTREME
472 depends on SPARSEMEM && !SPARSEMEM_STATIC
474 config SPARSEMEM_VMEMMAP_ENABLE
477 config SPARSEMEM_VMEMMAP
478 bool "Sparse Memory virtual memmap"
479 depends on SPARSEMEM && SPARSEMEM_VMEMMAP_ENABLE
482 SPARSEMEM_VMEMMAP uses a virtually mapped memmap to optimise
483 pfn_to_page and page_to_pfn operations. This is the most
484 efficient option when sufficient kernel resources are available.
486 # Select this config option from the architecture Kconfig, if it is preferred
487 # to enable the feature of HugeTLB/dev_dax vmemmap optimization.
489 config ARCH_WANT_OPTIMIZE_DAX_VMEMMAP
492 config ARCH_WANT_OPTIMIZE_HUGETLB_VMEMMAP
495 config HAVE_MEMBLOCK_PHYS_MAP
502 # Don't discard allocated memory used to track "memory" and "reserved" memblocks
503 # after early boot, so it can still be used to test for validity of memory.
504 # Also, memblocks are updated with memory hot(un)plug.
505 config ARCH_KEEP_MEMBLOCK
508 # Keep arch NUMA mapping infrastructure post-init.
509 config NUMA_KEEP_MEMINFO
512 config MEMORY_ISOLATION
515 # IORESOURCE_SYSTEM_RAM regions in the kernel resource tree that are marked
516 # IORESOURCE_EXCLUSIVE cannot be mapped to user space, for example, via
518 config EXCLUSIVE_SYSTEM_RAM
520 depends on !DEVMEM || STRICT_DEVMEM
523 # Only be set on architectures that have completely implemented memory hotplug
524 # feature. If you are not sure, don't touch it.
526 config HAVE_BOOTMEM_INFO_NODE
529 config ARCH_ENABLE_MEMORY_HOTPLUG
532 config ARCH_ENABLE_MEMORY_HOTREMOVE
535 # eventually, we can have this option just 'select SPARSEMEM'
536 menuconfig MEMORY_HOTPLUG
537 bool "Memory hotplug"
538 select MEMORY_ISOLATION
540 depends on ARCH_ENABLE_MEMORY_HOTPLUG
542 select NUMA_KEEP_MEMINFO if NUMA
546 config MEMORY_HOTPLUG_DEFAULT_ONLINE
547 bool "Online the newly added memory blocks by default"
548 depends on MEMORY_HOTPLUG
550 This option sets the default policy setting for memory hotplug
551 onlining policy (/sys/devices/system/memory/auto_online_blocks) which
552 determines what happens to newly added memory regions. Policy setting
553 can always be changed at runtime.
554 See Documentation/admin-guide/mm/memory-hotplug.rst for more information.
556 Say Y here if you want all hot-plugged memory blocks to appear in
557 'online' state by default.
558 Say N here if you want the default policy to keep all hot-plugged
559 memory blocks in 'offline' state.
561 config MEMORY_HOTREMOVE
562 bool "Allow for memory hot remove"
563 select HAVE_BOOTMEM_INFO_NODE if (X86_64 || PPC64)
564 depends on MEMORY_HOTPLUG && ARCH_ENABLE_MEMORY_HOTREMOVE
567 config MHP_MEMMAP_ON_MEMORY
569 depends on MEMORY_HOTPLUG && SPARSEMEM_VMEMMAP
570 depends on ARCH_MHP_MEMMAP_ON_MEMORY_ENABLE
572 endif # MEMORY_HOTPLUG
574 config ARCH_MHP_MEMMAP_ON_MEMORY_ENABLE
577 # Heavily threaded applications may benefit from splitting the mm-wide
578 # page_table_lock, so that faults on different parts of the user address
579 # space can be handled with less contention: split it at this NR_CPUS.
580 # Default to 4 for wider testing, though 8 might be more appropriate.
581 # ARM's adjust_pte (unused if VIPT) depends on mm-wide page_table_lock.
582 # PA-RISC 7xxx's spinlock_t would enlarge struct page from 32 to 44 bytes.
583 # SPARC32 allocates multiple pte tables within a single page, and therefore
584 # a per-page lock leads to problems when multiple tables need to be locked
585 # at the same time (e.g. copy_page_range()).
586 # DEBUG_SPINLOCK and DEBUG_LOCK_ALLOC spinlock_t also enlarge struct page.
588 config SPLIT_PTLOCK_CPUS
590 default "999999" if !MMU
591 default "999999" if ARM && !CPU_CACHE_VIPT
592 default "999999" if PARISC && !PA20
593 default "999999" if SPARC32
596 config ARCH_ENABLE_SPLIT_PMD_PTLOCK
600 # support for memory balloon
601 config MEMORY_BALLOON
605 # support for memory balloon compaction
606 config BALLOON_COMPACTION
607 bool "Allow for balloon memory compaction/migration"
609 depends on COMPACTION && MEMORY_BALLOON
611 Memory fragmentation introduced by ballooning might reduce
612 significantly the number of 2MB contiguous memory blocks that can be
613 used within a guest, thus imposing performance penalties associated
614 with the reduced number of transparent huge pages that could be used
615 by the guest workload. Allowing the compaction & migration for memory
616 pages enlisted as being part of memory balloon devices avoids the
617 scenario aforementioned and helps improving memory defragmentation.
620 # support for memory compaction
622 bool "Allow for memory compaction"
627 Compaction is the only memory management component to form
628 high order (larger physically contiguous) memory blocks
629 reliably. The page allocator relies on compaction heavily and
630 the lack of the feature can lead to unexpected OOM killer
631 invocations for high order memory requests. You shouldn't
632 disable this option unless there really is a strong reason for
633 it and then we would be really interested to hear about that at
636 config COMPACT_UNEVICTABLE_DEFAULT
638 depends on COMPACTION
639 default 0 if PREEMPT_RT
643 # support for free page reporting
644 config PAGE_REPORTING
645 bool "Free page reporting"
647 Free page reporting allows for the incremental acquisition of
648 free pages from the buddy allocator for the purpose of reporting
649 those pages to another entity, such as a hypervisor, so that the
650 memory can be freed within the host for other uses.
653 # support for page migration
656 bool "Page migration"
658 depends on (NUMA || ARCH_ENABLE_MEMORY_HOTREMOVE || COMPACTION || CMA) && MMU
660 Allows the migration of the physical location of pages of processes
661 while the virtual addresses are not changed. This is useful in
662 two situations. The first is on NUMA systems to put pages nearer
663 to the processors accessing. The second is when allocating huge
664 pages as migration can relocate pages to satisfy a huge page
665 allocation instead of reclaiming.
667 config DEVICE_MIGRATION
668 def_bool MIGRATION && ZONE_DEVICE
670 config ARCH_ENABLE_HUGEPAGE_MIGRATION
673 config ARCH_ENABLE_THP_MIGRATION
676 config HUGETLB_PAGE_SIZE_VARIABLE
679 Allows the pageblock_order value to be dynamic instead of just standard
680 HUGETLB_PAGE_ORDER when there are multiple HugeTLB page sizes available
683 Note that the pageblock_order cannot exceed MAX_PAGE_ORDER and will be
684 clamped down to MAX_PAGE_ORDER.
687 def_bool (MEMORY_ISOLATION && COMPACTION) || CMA
689 config PCP_BATCH_SCALE_MAX
690 int "Maximum scale factor of PCP (Per-CPU pageset) batch allocate/free"
694 In page allocator, PCP (Per-CPU pageset) is refilled and drained in
695 batches. The batch number is scaled automatically to improve page
696 allocation/free throughput. But too large scale factor may hurt
697 latency. This option sets the upper limit of scale factor to limit
700 config PHYS_ADDR_T_64BIT
704 bool "Enable bounce buffers"
706 depends on BLOCK && MMU && HIGHMEM
708 Enable bounce buffers for devices that cannot access the full range of
709 memory available to the CPU. Enabled by default when HIGHMEM is
710 selected, but you may say n to override this.
717 bool "Enable KSM for page merging"
721 Enable Kernel Samepage Merging: KSM periodically scans those areas
722 of an application's address space that an app has advised may be
723 mergeable. When it finds pages of identical content, it replaces
724 the many instances by a single page with that content, so
725 saving memory until one or another app needs to modify the content.
726 Recommended for use with KVM, or with other duplicative applications.
727 See Documentation/mm/ksm.rst for more information: KSM is inactive
728 until a program has madvised that an area is MADV_MERGEABLE, and
729 root has set /sys/kernel/mm/ksm/run to 1 (if CONFIG_SYSFS is set).
731 config DEFAULT_MMAP_MIN_ADDR
732 int "Low address space to protect from user allocation"
736 This is the portion of low virtual memory which should be protected
737 from userspace allocation. Keeping a user from writing to low pages
738 can help reduce the impact of kernel NULL pointer bugs.
740 For most arm64, ppc64 and x86 users with lots of address space
741 a value of 65536 is reasonable and should cause no problems.
742 On arm and other archs it should not be higher than 32768.
743 Programs which use vm86 functionality or have some need to map
744 this low address space will need CAP_SYS_RAWIO or disable this
745 protection by setting the value to 0.
747 This value can be changed after boot using the
748 /proc/sys/vm/mmap_min_addr tunable.
750 config ARCH_SUPPORTS_MEMORY_FAILURE
753 config MEMORY_FAILURE
755 depends on ARCH_SUPPORTS_MEMORY_FAILURE
756 bool "Enable recovery from hardware memory errors"
757 select MEMORY_ISOLATION
760 Enables code to recover from some memory failures on systems
761 with MCA recovery. This allows a system to continue running
762 even when some of its memory has uncorrected errors. This requires
763 special hardware support and typically ECC memory.
765 config HWPOISON_INJECT
766 tristate "HWPoison pages injector"
767 depends on MEMORY_FAILURE && DEBUG_KERNEL && PROC_FS
768 select PROC_PAGE_MONITOR
770 config NOMMU_INITIAL_TRIM_EXCESS
771 int "Turn on mmap() excess space trimming before booting"
775 The NOMMU mmap() frequently needs to allocate large contiguous chunks
776 of memory on which to store mappings, but it can only ask the system
777 allocator for chunks in 2^N*PAGE_SIZE amounts - which is frequently
778 more than it requires. To deal with this, mmap() is able to trim off
779 the excess and return it to the allocator.
781 If trimming is enabled, the excess is trimmed off and returned to the
782 system allocator, which can cause extra fragmentation, particularly
783 if there are a lot of transient processes.
785 If trimming is disabled, the excess is kept, but not used, which for
786 long-term mappings means that the space is wasted.
788 Trimming can be dynamically controlled through a sysctl option
789 (/proc/sys/vm/nr_trim_pages) which specifies the minimum number of
790 excess pages there must be before trimming should occur, or zero if
791 no trimming is to occur.
793 This option specifies the initial value of this option. The default
794 of 1 says that all excess pages should be trimmed.
796 See Documentation/admin-guide/mm/nommu-mmap.rst for more information.
798 config ARCH_WANT_GENERAL_HUGETLB
801 config ARCH_WANTS_THP_SWAP
804 menuconfig TRANSPARENT_HUGEPAGE
805 bool "Transparent Hugepage Support"
806 depends on HAVE_ARCH_TRANSPARENT_HUGEPAGE && !PREEMPT_RT
810 Transparent Hugepages allows the kernel to use huge pages and
811 huge tlb transparently to the applications whenever possible.
812 This feature can improve computing performance to certain
813 applications by speeding up page faults during memory
814 allocation, by reducing the number of tlb misses and by speeding
815 up the pagetable walking.
817 If memory constrained on embedded, you may want to say N.
819 if TRANSPARENT_HUGEPAGE
822 prompt "Transparent Hugepage Support sysfs defaults"
823 depends on TRANSPARENT_HUGEPAGE
824 default TRANSPARENT_HUGEPAGE_ALWAYS
826 Selects the sysfs defaults for Transparent Hugepage Support.
828 config TRANSPARENT_HUGEPAGE_ALWAYS
831 Enabling Transparent Hugepage always, can increase the
832 memory footprint of applications without a guaranteed
833 benefit but it will work automatically for all applications.
835 config TRANSPARENT_HUGEPAGE_MADVISE
838 Enabling Transparent Hugepage madvise, will only provide a
839 performance improvement benefit to the applications using
840 madvise(MADV_HUGEPAGE) but it won't risk to increase the
841 memory footprint of applications without a guaranteed
844 config TRANSPARENT_HUGEPAGE_NEVER
847 Disable Transparent Hugepage by default. It can still be
848 enabled at runtime via sysfs.
853 depends on TRANSPARENT_HUGEPAGE && ARCH_WANTS_THP_SWAP && SWAP && 64BIT
855 Swap transparent huge pages in one piece, without splitting.
856 XXX: For now, swap cluster backing transparent huge page
857 will be split after swapout.
859 For selection by architectures with reasonable THP sizes.
861 config READ_ONLY_THP_FOR_FS
862 bool "Read-only THP for filesystems (EXPERIMENTAL)"
863 depends on TRANSPARENT_HUGEPAGE && SHMEM
866 Allow khugepaged to put read-only file-backed pages in THP.
868 This is marked experimental because it is a new feature. Write
869 support of file THPs will be developed in the next few release
872 endif # TRANSPARENT_HUGEPAGE
875 # The architecture supports pgtable leaves that is larger than PAGE_SIZE
877 config PGTABLE_HAS_HUGE_LEAVES
878 def_bool TRANSPARENT_HUGEPAGE || HUGETLB_PAGE
881 # UP and nommu archs use km based percpu allocator
883 config NEED_PER_CPU_KM
884 depends on !SMP || !MMU
888 config NEED_PER_CPU_EMBED_FIRST_CHUNK
891 config NEED_PER_CPU_PAGE_FIRST_CHUNK
894 config USE_PERCPU_NUMA_NODE_ID
897 config HAVE_SETUP_PER_CPU_AREA
901 bool "Contiguous Memory Allocator"
904 select MEMORY_ISOLATION
906 This enables the Contiguous Memory Allocator which allows other
907 subsystems to allocate big physically-contiguous blocks of memory.
908 CMA reserves a region of memory and allows only movable pages to
909 be allocated from it. This way, the kernel can use the memory for
910 pagecache and when a subsystem requests for contiguous area, the
911 allocated pages are migrated away to serve the contiguous request.
916 bool "CMA debugfs interface"
917 depends on CMA && DEBUG_FS
919 Turns on the DebugFS interface for CMA.
922 bool "CMA information through sysfs interface"
923 depends on CMA && SYSFS
925 This option exposes some sysfs attributes to get information
929 int "Maximum count of the CMA areas"
934 CMA allows to create CMA areas for particular purpose, mainly,
935 used as device private area. This parameter sets the maximum
936 number of CMA area in the system.
938 If unsure, leave the default value "8" in UMA and "20" in NUMA.
940 config MEM_SOFT_DIRTY
941 bool "Track memory changes"
942 depends on CHECKPOINT_RESTORE && HAVE_ARCH_SOFT_DIRTY && PROC_FS
943 select PROC_PAGE_MONITOR
945 This option enables memory changes tracking by introducing a
946 soft-dirty bit on pte-s. This bit it set when someone writes
947 into a page just as regular dirty bit, but unlike the latter
948 it can be cleared by hands.
950 See Documentation/admin-guide/mm/soft-dirty.rst for more details.
952 config GENERIC_EARLY_IOREMAP
955 config STACK_MAX_DEFAULT_SIZE_MB
956 int "Default maximum user stack size for 32-bit processes (MB)"
959 depends on STACK_GROWSUP && (!64BIT || COMPAT)
961 This is the maximum stack size in Megabytes in the VM layout of 32-bit
962 user processes when the stack grows upwards (currently only on parisc
963 arch) when the RLIMIT_STACK hard limit is unlimited.
965 A sane initial value is 100 MB.
967 config DEFERRED_STRUCT_PAGE_INIT
968 bool "Defer initialisation of struct pages to kthreads"
970 depends on !NEED_PER_CPU_KM
975 Ordinarily all struct pages are initialised during early boot in a
976 single thread. On very large machines this can take a considerable
977 amount of time. If this option is set, large machines will bring up
978 a subset of memmap at boot and then initialise the rest in parallel.
979 This has a potential performance impact on tasks running early in the
980 lifetime of the system until these kthreads finish the
983 config PAGE_IDLE_FLAG
985 select PAGE_EXTENSION if !64BIT
987 This adds PG_idle and PG_young flags to 'struct page'. PTE Accessed
988 bit writers can set the state of the bit in the flags so that PTE
989 Accessed bit readers may avoid disturbance.
991 config IDLE_PAGE_TRACKING
992 bool "Enable idle page tracking"
993 depends on SYSFS && MMU
994 select PAGE_IDLE_FLAG
996 This feature allows to estimate the amount of user pages that have
997 not been touched during a given period of time. This information can
998 be useful to tune memory cgroup limits and/or for job placement
999 within a compute cluster.
1001 See Documentation/admin-guide/mm/idle_page_tracking.rst for
1004 # Architectures which implement cpu_dcache_is_aliasing() to query
1005 # whether the data caches are aliased (VIVT or VIPT with dcache
1006 # aliasing) need to select this.
1007 config ARCH_HAS_CPU_CACHE_ALIASING
1010 config ARCH_HAS_CACHE_LINE_SIZE
1013 config ARCH_HAS_CURRENT_STACK_POINTER
1016 In support of HARDENED_USERCOPY performing stack variable lifetime
1017 checking, an architecture-agnostic way to find the stack pointer
1018 is needed. Once an architecture defines an unsigned long global
1019 register alias named "current_stack_pointer", this config can be
1022 config ARCH_HAS_PTE_DEVMAP
1025 config ARCH_HAS_ZONE_DMA_SET
1029 bool "Support DMA zone" if ARCH_HAS_ZONE_DMA_SET
1030 default y if ARM64 || X86
1033 bool "Support DMA32 zone" if ARCH_HAS_ZONE_DMA_SET
1038 bool "Device memory (pmem, HMM, etc...) hotplug support"
1039 depends on MEMORY_HOTPLUG
1040 depends on MEMORY_HOTREMOVE
1041 depends on SPARSEMEM_VMEMMAP
1042 depends on ARCH_HAS_PTE_DEVMAP
1046 Device memory hotplug support allows for establishing pmem,
1047 or other device driver discovered memory regions, in the
1048 memmap. This allows pfn_to_page() lookups of otherwise
1049 "device-physical" addresses which is needed for using a DAX
1050 mapping in an O_DIRECT operation, among other things.
1052 If FS_DAX is enabled, then say Y.
1055 # Helpers to mirror range of the CPU page tables of a process into device page
1062 config GET_FREE_REGION
1063 depends on SPARSEMEM
1066 config DEVICE_PRIVATE
1067 bool "Unaddressable device memory (GPU memory, ...)"
1068 depends on ZONE_DEVICE
1069 select GET_FREE_REGION
1072 Allows creation of struct pages to represent unaddressable device
1073 memory; i.e., memory that is only accessible from the device (or
1074 group of devices). You likely also want to select HMM_MIRROR.
1079 config ARCH_USES_HIGH_VMA_FLAGS
1081 config ARCH_HAS_PKEYS
1084 config ARCH_USES_PG_ARCH_X
1087 Enable the definition of PG_arch_x page flags with x > 1. Only
1088 suitable for 64-bit architectures with CONFIG_FLATMEM or
1089 CONFIG_SPARSEMEM_VMEMMAP enabled, otherwise there may not be
1090 enough room for additional bits in page->flags.
1092 config VM_EVENT_COUNTERS
1094 bool "Enable VM event counters for /proc/vmstat" if EXPERT
1096 VM event counters are needed for event counts to be shown.
1097 This option allows the disabling of the VM event counters
1098 on EXPERT systems. /proc/vmstat will only show page counts
1099 if VM event counters are disabled.
1102 bool "Collect percpu memory statistics"
1104 This feature collects and exposes statistics via debugfs. The
1105 information includes global and per chunk statistics, which can
1106 be used to help understand percpu memory usage.
1109 bool "Enable infrastructure for get_user_pages()-related unit tests"
1112 Provides /sys/kernel/debug/gup_test, which in turn provides a way
1113 to make ioctl calls that can launch kernel-based unit tests for
1114 the get_user_pages*() and pin_user_pages*() family of API calls.
1116 These tests include benchmark testing of the _fast variants of
1117 get_user_pages*() and pin_user_pages*(), as well as smoke tests of
1118 the non-_fast variants.
1120 There is also a sub-test that allows running dump_page() on any
1121 of up to eight pages (selected by command line args) within the
1122 range of user-space addresses. These pages are either pinned via
1123 pin_user_pages*(), or pinned via get_user_pages*(), as specified
1124 by other command line arguments.
1126 See tools/testing/selftests/mm/gup_test.c
1128 comment "GUP_TEST needs to have DEBUG_FS enabled"
1129 depends on !GUP_TEST && !DEBUG_FS
1131 config GUP_GET_PXX_LOW_HIGH
1135 tristate "Enable a module to run time tests on dma_pool"
1138 Provides a test module that will allocate and free many blocks of
1139 various sizes and report how long it takes. This is intended to
1140 provide a consistent way to measure how changes to the
1141 dma_pool_alloc/free routines affect performance.
1143 config ARCH_HAS_PTE_SPECIAL
1146 config MAPPING_DIRTY_HELPERS
1152 config KMAP_LOCAL_NON_LINEAR_PTE_ARRAY
1155 # struct io_mapping based helper. Selected by drivers that need them
1160 bool "Enable memfd_create() system call" if EXPERT
1164 bool "Enable memfd_secret() system call" if EXPERT
1165 depends on ARCH_HAS_SET_DIRECT_MAP
1167 Enable the memfd_secret() system call with the ability to create
1168 memory areas visible only in the context of the owning process and
1169 not mapped to other processes and other kernel page tables.
1171 config ANON_VMA_NAME
1172 bool "Anonymous VMA name support"
1173 depends on PROC_FS && ADVISE_SYSCALLS && MMU
1176 Allow naming anonymous virtual memory areas.
1178 This feature allows assigning names to virtual memory areas. Assigned
1179 names can be later retrieved from /proc/pid/maps and /proc/pid/smaps
1180 and help identifying individual anonymous memory areas.
1181 Assigning a name to anonymous virtual memory area might prevent that
1182 area from being merged with adjacent virtual memory areas due to the
1183 difference in their name.
1185 config HAVE_ARCH_USERFAULTFD_WP
1188 Arch has userfaultfd write protection support
1190 config HAVE_ARCH_USERFAULTFD_MINOR
1193 Arch has userfaultfd minor fault support
1195 menuconfig USERFAULTFD
1196 bool "Enable userfaultfd() system call"
1199 Enable the userfaultfd() system call that allows to intercept and
1200 handle page faults in userland.
1203 config PTE_MARKER_UFFD_WP
1204 bool "Userfaultfd write protection support for shmem/hugetlbfs"
1206 depends on HAVE_ARCH_USERFAULTFD_WP
1209 Allows to create marker PTEs for userfaultfd write protection
1210 purposes. It is required to enable userfaultfd write protection on
1211 file-backed memory types like shmem and hugetlbfs.
1216 bool "Multi-Gen LRU"
1218 # make sure folio->flags has enough spare bits
1219 depends on 64BIT || !SPARSEMEM || SPARSEMEM_VMEMMAP
1221 A high performance LRU implementation to overcommit memory. See
1222 Documentation/admin-guide/mm/multigen_lru.rst for details.
1224 config LRU_GEN_ENABLED
1225 bool "Enable by default"
1228 This option enables the multi-gen LRU by default.
1230 config LRU_GEN_STATS
1231 bool "Full stats for debugging"
1234 Do not enable this option unless you plan to look at historical stats
1235 from evicted generations for debugging purpose.
1237 This option has a per-memcg and per-node memory overhead.
1239 config LRU_GEN_WALKS_MMU
1241 depends on LRU_GEN && ARCH_HAS_HW_PTE_YOUNG
1244 config ARCH_SUPPORTS_PER_VMA_LOCK
1249 depends on ARCH_SUPPORTS_PER_VMA_LOCK && MMU && SMP
1251 Allow per-vma locking during page fault handling.
1253 This feature allows locking each virtual memory area separately when
1254 handling page faults instead of taking mmap_lock.
1256 config LOCK_MM_AND_FIND_VMA
1258 depends on !STACK_GROWSUP
1260 config IOMMU_MM_DATA
1266 source "mm/damon/Kconfig"