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"
32 A lightweight compressed cache for swap pages. It takes
33 pages that are in the process of being swapped out and attempts to
34 compress them into a dynamically allocated RAM-based memory pool.
35 This can result in a significant I/O reduction on swap device and,
36 in the case where decompressing from RAM is faster than swap device
37 reads, can also improve workload performance.
39 config ZSWAP_DEFAULT_ON
40 bool "Enable the compressed cache for swap pages by default"
43 If selected, the compressed cache for swap pages will be enabled
44 at boot, otherwise it will be disabled.
46 The selection made here can be overridden by using the kernel
47 command line 'zswap.enabled=' option.
50 prompt "Default compressor"
52 default ZSWAP_COMPRESSOR_DEFAULT_LZO
54 Selects the default compression algorithm for the compressed cache
57 For an overview what kind of performance can be expected from
58 a particular compression algorithm please refer to the benchmarks
59 available at the following LWN page:
60 https://lwn.net/Articles/751795/
62 If in doubt, select 'LZO'.
64 The selection made here can be overridden by using the kernel
65 command line 'zswap.compressor=' option.
67 config ZSWAP_COMPRESSOR_DEFAULT_DEFLATE
71 Use the Deflate algorithm as the default compression algorithm.
73 config ZSWAP_COMPRESSOR_DEFAULT_LZO
77 Use the LZO algorithm as the default compression algorithm.
79 config ZSWAP_COMPRESSOR_DEFAULT_842
83 Use the 842 algorithm as the default compression algorithm.
85 config ZSWAP_COMPRESSOR_DEFAULT_LZ4
89 Use the LZ4 algorithm as the default compression algorithm.
91 config ZSWAP_COMPRESSOR_DEFAULT_LZ4HC
95 Use the LZ4HC algorithm as the default compression algorithm.
97 config ZSWAP_COMPRESSOR_DEFAULT_ZSTD
101 Use the zstd algorithm as the default compression algorithm.
104 config ZSWAP_COMPRESSOR_DEFAULT
107 default "deflate" if ZSWAP_COMPRESSOR_DEFAULT_DEFLATE
108 default "lzo" if ZSWAP_COMPRESSOR_DEFAULT_LZO
109 default "842" if ZSWAP_COMPRESSOR_DEFAULT_842
110 default "lz4" if ZSWAP_COMPRESSOR_DEFAULT_LZ4
111 default "lz4hc" if ZSWAP_COMPRESSOR_DEFAULT_LZ4HC
112 default "zstd" if ZSWAP_COMPRESSOR_DEFAULT_ZSTD
116 prompt "Default allocator"
118 default ZSWAP_ZPOOL_DEFAULT_ZBUD
120 Selects the default allocator for the compressed cache for
122 The default is 'zbud' for compatibility, however please do
123 read the description of each of the allocators below before
124 making a right choice.
126 The selection made here can be overridden by using the kernel
127 command line 'zswap.zpool=' option.
129 config ZSWAP_ZPOOL_DEFAULT_ZBUD
133 Use the zbud allocator as the default allocator.
135 config ZSWAP_ZPOOL_DEFAULT_Z3FOLD
139 Use the z3fold allocator as the default allocator.
141 config ZSWAP_ZPOOL_DEFAULT_ZSMALLOC
145 Use the zsmalloc allocator as the default allocator.
148 config ZSWAP_ZPOOL_DEFAULT
151 default "zbud" if ZSWAP_ZPOOL_DEFAULT_ZBUD
152 default "z3fold" if ZSWAP_ZPOOL_DEFAULT_Z3FOLD
153 default "zsmalloc" if ZSWAP_ZPOOL_DEFAULT_ZSMALLOC
157 tristate "2:1 compression allocator (zbud)"
160 A special purpose allocator for storing compressed pages.
161 It is designed to store up to two compressed pages per physical
162 page. While this design limits storage density, it has simple and
163 deterministic reclaim properties that make it preferable to a higher
164 density approach when reclaim will be used.
167 tristate "3:1 compression allocator (z3fold)"
170 A special purpose allocator for storing compressed pages.
171 It is designed to store up to three compressed pages per physical
172 page. It is a ZBUD derivative so the simplicity and determinism are
177 prompt "N:1 compression allocator (zsmalloc)" if ZSWAP
180 zsmalloc is a slab-based memory allocator designed to store
181 pages of various compression levels efficiently. It achieves
182 the highest storage density with the least amount of fragmentation.
185 bool "Export zsmalloc statistics"
189 This option enables code in the zsmalloc to collect various
190 statistics about what's happening in zsmalloc and exports that
191 information to userspace via debugfs.
194 config ZSMALLOC_CHAIN_SIZE
195 int "Maximum number of physical pages per-zspage"
200 This option sets the upper limit on the number of physical pages
201 that a zmalloc page (zspage) can consist of. The optimal zspage
202 chain size is calculated for each size class during the
203 initialization of the pool.
205 Changing this option can alter the characteristics of size classes,
206 such as the number of pages per zspage and the number of objects
207 per zspage. This can also result in different configurations of
208 the pool, as zsmalloc merges size classes with similar
211 For more information, see zsmalloc documentation.
213 menu "SLAB allocator options"
216 prompt "Choose SLAB allocator"
219 This option allows to select a slab allocator.
223 depends on !PREEMPT_RT
224 select HAVE_HARDENED_USERCOPY_ALLOCATOR
226 The regular slab allocator that is established and known to work
227 well in all environments. It organizes cache hot objects in
228 per cpu and per node queues.
231 bool "SLUB (Unqueued Allocator)"
232 select HAVE_HARDENED_USERCOPY_ALLOCATOR
234 SLUB is a slab allocator that minimizes cache line usage
235 instead of managing queues of cached objects (SLAB approach).
236 Per cpu caching is realized using slabs of objects instead
237 of queues of objects. SLUB can use memory efficiently
238 and has enhanced diagnostics. SLUB is the default choice for
244 bool "Configure SLUB for minimal memory footprint"
245 depends on SLUB && EXPERT
246 select SLAB_MERGE_DEFAULT
248 Configures the SLUB allocator in a way to achieve minimal memory
249 footprint, sacrificing scalability, debugging and other features.
250 This is intended only for the smallest system that had used the
251 SLOB allocator and is not recommended for systems with more than
256 config SLAB_MERGE_DEFAULT
257 bool "Allow slab caches to be merged"
259 depends on SLAB || SLUB
261 For reduced kernel memory fragmentation, slab caches can be
262 merged when they share the same size and other characteristics.
263 This carries a risk of kernel heap overflows being able to
264 overwrite objects from merged caches (and more easily control
265 cache layout), which makes such heap attacks easier to exploit
266 by attackers. By keeping caches unmerged, these kinds of exploits
267 can usually only damage objects in the same cache. To disable
268 merging at runtime, "slab_nomerge" can be passed on the kernel
271 config SLAB_FREELIST_RANDOM
272 bool "Randomize slab freelist"
273 depends on SLAB || (SLUB && !SLUB_TINY)
275 Randomizes the freelist order used on creating new pages. This
276 security feature reduces the predictability of the kernel slab
277 allocator against heap overflows.
279 config SLAB_FREELIST_HARDENED
280 bool "Harden slab freelist metadata"
281 depends on SLAB || (SLUB && !SLUB_TINY)
283 Many kernel heap attacks try to target slab cache metadata and
284 other infrastructure. This options makes minor performance
285 sacrifices to harden the kernel slab allocator against common
286 freelist exploit methods. Some slab implementations have more
287 sanity-checking than others. This option is most effective with
292 bool "Enable SLUB performance statistics"
293 depends on SLUB && SYSFS && !SLUB_TINY
295 SLUB statistics are useful to debug SLUBs allocation behavior in
296 order find ways to optimize the allocator. This should never be
297 enabled for production use since keeping statistics slows down
298 the allocator by a few percentage points. The slabinfo command
299 supports the determination of the most active slabs to figure
300 out which slabs are relevant to a particular load.
301 Try running: slabinfo -DA
303 config SLUB_CPU_PARTIAL
305 depends on SLUB && SMP && !SLUB_TINY
306 bool "SLUB per cpu partial cache"
308 Per cpu partial caches accelerate objects allocation and freeing
309 that is local to a processor at the price of more indeterminism
310 in the latency of the free. On overflow these caches will be cleared
311 which requires the taking of locks that may cause latency spikes.
312 Typically one would choose no for a realtime system.
314 endmenu # SLAB allocator options
316 config SHUFFLE_PAGE_ALLOCATOR
317 bool "Page allocator randomization"
318 default SLAB_FREELIST_RANDOM && ACPI_NUMA
320 Randomization of the page allocator improves the average
321 utilization of a direct-mapped memory-side-cache. See section
322 5.2.27 Heterogeneous Memory Attribute Table (HMAT) in the ACPI
323 6.2a specification for an example of how a platform advertises
324 the presence of a memory-side-cache. There are also incidental
325 security benefits as it reduces the predictability of page
326 allocations to compliment SLAB_FREELIST_RANDOM, but the
327 default granularity of shuffling on the MAX_ORDER i.e, 10th
328 order of pages is selected based on cache utilization benefits
331 While the randomization improves cache utilization it may
332 negatively impact workloads on platforms without a cache. For
333 this reason, by default, the randomization is enabled only
334 after runtime detection of a direct-mapped memory-side-cache.
335 Otherwise, the randomization may be force enabled with the
336 'page_alloc.shuffle' kernel command line parameter.
341 bool "Disable heap randomization"
344 Randomizing heap placement makes heap exploits harder, but it
345 also breaks ancient binaries (including anything libc5 based).
346 This option changes the bootup default to heap randomization
347 disabled, and can be overridden at runtime by setting
348 /proc/sys/kernel/randomize_va_space to 2.
350 On non-ancient distros (post-2000 ones) N is usually a safe choice.
352 config MMAP_ALLOW_UNINITIALIZED
353 bool "Allow mmapped anonymous memory to be uninitialized"
354 depends on EXPERT && !MMU
357 Normally, and according to the Linux spec, anonymous memory obtained
358 from mmap() has its contents cleared before it is passed to
359 userspace. Enabling this config option allows you to request that
360 mmap() skip that if it is given an MAP_UNINITIALIZED flag, thus
361 providing a huge performance boost. If this option is not enabled,
362 then the flag will be ignored.
364 This is taken advantage of by uClibc's malloc(), and also by
365 ELF-FDPIC binfmt's brk and stack allocator.
367 Because of the obvious security issues, this option should only be
368 enabled on embedded devices where you control what is run in
369 userspace. Since that isn't generally a problem on no-MMU systems,
370 it is normally safe to say Y here.
372 See Documentation/admin-guide/mm/nommu-mmap.rst for more information.
374 config SELECT_MEMORY_MODEL
376 depends on ARCH_SELECT_MEMORY_MODEL
379 prompt "Memory model"
380 depends on SELECT_MEMORY_MODEL
381 default SPARSEMEM_MANUAL if ARCH_SPARSEMEM_DEFAULT
382 default FLATMEM_MANUAL
384 This option allows you to change some of the ways that
385 Linux manages its memory internally. Most users will
386 only have one option here selected by the architecture
387 configuration. This is normal.
389 config FLATMEM_MANUAL
391 depends on !ARCH_SPARSEMEM_ENABLE || ARCH_FLATMEM_ENABLE
393 This option is best suited for non-NUMA systems with
394 flat address space. The FLATMEM is the most efficient
395 system in terms of performance and resource consumption
396 and it is the best option for smaller systems.
398 For systems that have holes in their physical address
399 spaces and for features like NUMA and memory hotplug,
400 choose "Sparse Memory".
402 If unsure, choose this option (Flat Memory) over any other.
404 config SPARSEMEM_MANUAL
406 depends on ARCH_SPARSEMEM_ENABLE
408 This will be the only option for some systems, including
409 memory hot-plug systems. This is normal.
411 This option provides efficient support for systems with
412 holes is their physical address space and allows memory
413 hot-plug and hot-remove.
415 If unsure, choose "Flat Memory" over this option.
421 depends on (!SELECT_MEMORY_MODEL && ARCH_SPARSEMEM_ENABLE) || SPARSEMEM_MANUAL
425 depends on !SPARSEMEM || FLATMEM_MANUAL
428 # SPARSEMEM_EXTREME (which is the default) does some bootmem
429 # allocations when sparse_init() is called. If this cannot
430 # be done on your architecture, select this option. However,
431 # statically allocating the mem_section[] array can potentially
432 # consume vast quantities of .bss, so be careful.
434 # This option will also potentially produce smaller runtime code
435 # with gcc 3.4 and later.
437 config SPARSEMEM_STATIC
441 # Architecture platforms which require a two level mem_section in SPARSEMEM
442 # must select this option. This is usually for architecture platforms with
443 # an extremely sparse physical address space.
445 config SPARSEMEM_EXTREME
447 depends on SPARSEMEM && !SPARSEMEM_STATIC
449 config SPARSEMEM_VMEMMAP_ENABLE
452 config SPARSEMEM_VMEMMAP
453 bool "Sparse Memory virtual memmap"
454 depends on SPARSEMEM && SPARSEMEM_VMEMMAP_ENABLE
457 SPARSEMEM_VMEMMAP uses a virtually mapped memmap to optimise
458 pfn_to_page and page_to_pfn operations. This is the most
459 efficient option when sufficient kernel resources are available.
461 # Select this config option from the architecture Kconfig, if it is preferred
462 # to enable the feature of HugeTLB/dev_dax vmemmap optimization.
464 config ARCH_WANT_OPTIMIZE_VMEMMAP
467 config HAVE_MEMBLOCK_PHYS_MAP
474 # Don't discard allocated memory used to track "memory" and "reserved" memblocks
475 # after early boot, so it can still be used to test for validity of memory.
476 # Also, memblocks are updated with memory hot(un)plug.
477 config ARCH_KEEP_MEMBLOCK
480 # Keep arch NUMA mapping infrastructure post-init.
481 config NUMA_KEEP_MEMINFO
484 config MEMORY_ISOLATION
487 # IORESOURCE_SYSTEM_RAM regions in the kernel resource tree that are marked
488 # IORESOURCE_EXCLUSIVE cannot be mapped to user space, for example, via
490 config EXCLUSIVE_SYSTEM_RAM
492 depends on !DEVMEM || STRICT_DEVMEM
495 # Only be set on architectures that have completely implemented memory hotplug
496 # feature. If you are not sure, don't touch it.
498 config HAVE_BOOTMEM_INFO_NODE
501 config ARCH_ENABLE_MEMORY_HOTPLUG
504 config ARCH_ENABLE_MEMORY_HOTREMOVE
507 # eventually, we can have this option just 'select SPARSEMEM'
508 menuconfig MEMORY_HOTPLUG
509 bool "Memory hotplug"
510 select MEMORY_ISOLATION
512 depends on ARCH_ENABLE_MEMORY_HOTPLUG
514 select NUMA_KEEP_MEMINFO if NUMA
518 config MEMORY_HOTPLUG_DEFAULT_ONLINE
519 bool "Online the newly added memory blocks by default"
520 depends on MEMORY_HOTPLUG
522 This option sets the default policy setting for memory hotplug
523 onlining policy (/sys/devices/system/memory/auto_online_blocks) which
524 determines what happens to newly added memory regions. Policy setting
525 can always be changed at runtime.
526 See Documentation/admin-guide/mm/memory-hotplug.rst for more information.
528 Say Y here if you want all hot-plugged memory blocks to appear in
529 'online' state by default.
530 Say N here if you want the default policy to keep all hot-plugged
531 memory blocks in 'offline' state.
533 config MEMORY_HOTREMOVE
534 bool "Allow for memory hot remove"
535 select HAVE_BOOTMEM_INFO_NODE if (X86_64 || PPC64)
536 depends on MEMORY_HOTPLUG && ARCH_ENABLE_MEMORY_HOTREMOVE
539 config MHP_MEMMAP_ON_MEMORY
541 depends on MEMORY_HOTPLUG && SPARSEMEM_VMEMMAP
542 depends on ARCH_MHP_MEMMAP_ON_MEMORY_ENABLE
544 endif # MEMORY_HOTPLUG
546 # Heavily threaded applications may benefit from splitting the mm-wide
547 # page_table_lock, so that faults on different parts of the user address
548 # space can be handled with less contention: split it at this NR_CPUS.
549 # Default to 4 for wider testing, though 8 might be more appropriate.
550 # ARM's adjust_pte (unused if VIPT) depends on mm-wide page_table_lock.
551 # PA-RISC 7xxx's spinlock_t would enlarge struct page from 32 to 44 bytes.
552 # SPARC32 allocates multiple pte tables within a single page, and therefore
553 # a per-page lock leads to problems when multiple tables need to be locked
554 # at the same time (e.g. copy_page_range()).
555 # DEBUG_SPINLOCK and DEBUG_LOCK_ALLOC spinlock_t also enlarge struct page.
557 config SPLIT_PTLOCK_CPUS
559 default "999999" if !MMU
560 default "999999" if ARM && !CPU_CACHE_VIPT
561 default "999999" if PARISC && !PA20
562 default "999999" if SPARC32
565 config ARCH_ENABLE_SPLIT_PMD_PTLOCK
569 # support for memory balloon
570 config MEMORY_BALLOON
574 # support for memory balloon compaction
575 config BALLOON_COMPACTION
576 bool "Allow for balloon memory compaction/migration"
578 depends on COMPACTION && MEMORY_BALLOON
580 Memory fragmentation introduced by ballooning might reduce
581 significantly the number of 2MB contiguous memory blocks that can be
582 used within a guest, thus imposing performance penalties associated
583 with the reduced number of transparent huge pages that could be used
584 by the guest workload. Allowing the compaction & migration for memory
585 pages enlisted as being part of memory balloon devices avoids the
586 scenario aforementioned and helps improving memory defragmentation.
589 # support for memory compaction
591 bool "Allow for memory compaction"
596 Compaction is the only memory management component to form
597 high order (larger physically contiguous) memory blocks
598 reliably. The page allocator relies on compaction heavily and
599 the lack of the feature can lead to unexpected OOM killer
600 invocations for high order memory requests. You shouldn't
601 disable this option unless there really is a strong reason for
602 it and then we would be really interested to hear about that at
605 config COMPACT_UNEVICTABLE_DEFAULT
607 depends on COMPACTION
608 default 0 if PREEMPT_RT
612 # support for free page reporting
613 config PAGE_REPORTING
614 bool "Free page reporting"
617 Free page reporting allows for the incremental acquisition of
618 free pages from the buddy allocator for the purpose of reporting
619 those pages to another entity, such as a hypervisor, so that the
620 memory can be freed within the host for other uses.
623 # support for page migration
626 bool "Page migration"
628 depends on (NUMA || ARCH_ENABLE_MEMORY_HOTREMOVE || COMPACTION || CMA) && MMU
630 Allows the migration of the physical location of pages of processes
631 while the virtual addresses are not changed. This is useful in
632 two situations. The first is on NUMA systems to put pages nearer
633 to the processors accessing. The second is when allocating huge
634 pages as migration can relocate pages to satisfy a huge page
635 allocation instead of reclaiming.
637 config DEVICE_MIGRATION
638 def_bool MIGRATION && ZONE_DEVICE
640 config ARCH_ENABLE_HUGEPAGE_MIGRATION
643 config ARCH_ENABLE_THP_MIGRATION
646 config HUGETLB_PAGE_SIZE_VARIABLE
649 Allows the pageblock_order value to be dynamic instead of just standard
650 HUGETLB_PAGE_ORDER when there are multiple HugeTLB page sizes available
653 Note that the pageblock_order cannot exceed MAX_ORDER and will be
654 clamped down to MAX_ORDER.
657 def_bool (MEMORY_ISOLATION && COMPACTION) || CMA
659 config PHYS_ADDR_T_64BIT
663 bool "Enable bounce buffers"
665 depends on BLOCK && MMU && HIGHMEM
667 Enable bounce buffers for devices that cannot access the full range of
668 memory available to the CPU. Enabled by default when HIGHMEM is
669 selected, but you may say n to override this.
676 bool "Enable KSM for page merging"
680 Enable Kernel Samepage Merging: KSM periodically scans those areas
681 of an application's address space that an app has advised may be
682 mergeable. When it finds pages of identical content, it replaces
683 the many instances by a single page with that content, so
684 saving memory until one or another app needs to modify the content.
685 Recommended for use with KVM, or with other duplicative applications.
686 See Documentation/mm/ksm.rst for more information: KSM is inactive
687 until a program has madvised that an area is MADV_MERGEABLE, and
688 root has set /sys/kernel/mm/ksm/run to 1 (if CONFIG_SYSFS is set).
690 config DEFAULT_MMAP_MIN_ADDR
691 int "Low address space to protect from user allocation"
695 This is the portion of low virtual memory which should be protected
696 from userspace allocation. Keeping a user from writing to low pages
697 can help reduce the impact of kernel NULL pointer bugs.
699 For most ia64, ppc64 and x86 users with lots of address space
700 a value of 65536 is reasonable and should cause no problems.
701 On arm and other archs it should not be higher than 32768.
702 Programs which use vm86 functionality or have some need to map
703 this low address space will need CAP_SYS_RAWIO or disable this
704 protection by setting the value to 0.
706 This value can be changed after boot using the
707 /proc/sys/vm/mmap_min_addr tunable.
709 config ARCH_SUPPORTS_MEMORY_FAILURE
712 config MEMORY_FAILURE
714 depends on ARCH_SUPPORTS_MEMORY_FAILURE
715 bool "Enable recovery from hardware memory errors"
716 select MEMORY_ISOLATION
719 Enables code to recover from some memory failures on systems
720 with MCA recovery. This allows a system to continue running
721 even when some of its memory has uncorrected errors. This requires
722 special hardware support and typically ECC memory.
724 config HWPOISON_INJECT
725 tristate "HWPoison pages injector"
726 depends on MEMORY_FAILURE && DEBUG_KERNEL && PROC_FS
727 select PROC_PAGE_MONITOR
729 config NOMMU_INITIAL_TRIM_EXCESS
730 int "Turn on mmap() excess space trimming before booting"
734 The NOMMU mmap() frequently needs to allocate large contiguous chunks
735 of memory on which to store mappings, but it can only ask the system
736 allocator for chunks in 2^N*PAGE_SIZE amounts - which is frequently
737 more than it requires. To deal with this, mmap() is able to trim off
738 the excess and return it to the allocator.
740 If trimming is enabled, the excess is trimmed off and returned to the
741 system allocator, which can cause extra fragmentation, particularly
742 if there are a lot of transient processes.
744 If trimming is disabled, the excess is kept, but not used, which for
745 long-term mappings means that the space is wasted.
747 Trimming can be dynamically controlled through a sysctl option
748 (/proc/sys/vm/nr_trim_pages) which specifies the minimum number of
749 excess pages there must be before trimming should occur, or zero if
750 no trimming is to occur.
752 This option specifies the initial value of this option. The default
753 of 1 says that all excess pages should be trimmed.
755 See Documentation/admin-guide/mm/nommu-mmap.rst for more information.
757 config ARCH_WANT_GENERAL_HUGETLB
760 config ARCH_WANTS_THP_SWAP
763 menuconfig TRANSPARENT_HUGEPAGE
764 bool "Transparent Hugepage Support"
765 depends on HAVE_ARCH_TRANSPARENT_HUGEPAGE && !PREEMPT_RT
769 Transparent Hugepages allows the kernel to use huge pages and
770 huge tlb transparently to the applications whenever possible.
771 This feature can improve computing performance to certain
772 applications by speeding up page faults during memory
773 allocation, by reducing the number of tlb misses and by speeding
774 up the pagetable walking.
776 If memory constrained on embedded, you may want to say N.
778 if TRANSPARENT_HUGEPAGE
781 prompt "Transparent Hugepage Support sysfs defaults"
782 depends on TRANSPARENT_HUGEPAGE
783 default TRANSPARENT_HUGEPAGE_ALWAYS
785 Selects the sysfs defaults for Transparent Hugepage Support.
787 config TRANSPARENT_HUGEPAGE_ALWAYS
790 Enabling Transparent Hugepage always, can increase the
791 memory footprint of applications without a guaranteed
792 benefit but it will work automatically for all applications.
794 config TRANSPARENT_HUGEPAGE_MADVISE
797 Enabling Transparent Hugepage madvise, will only provide a
798 performance improvement benefit to the applications using
799 madvise(MADV_HUGEPAGE) but it won't risk to increase the
800 memory footprint of applications without a guaranteed
806 depends on TRANSPARENT_HUGEPAGE && ARCH_WANTS_THP_SWAP && SWAP && 64BIT
808 Swap transparent huge pages in one piece, without splitting.
809 XXX: For now, swap cluster backing transparent huge page
810 will be split after swapout.
812 For selection by architectures with reasonable THP sizes.
814 config READ_ONLY_THP_FOR_FS
815 bool "Read-only THP for filesystems (EXPERIMENTAL)"
816 depends on TRANSPARENT_HUGEPAGE && SHMEM
819 Allow khugepaged to put read-only file-backed pages in THP.
821 This is marked experimental because it is a new feature. Write
822 support of file THPs will be developed in the next few release
825 endif # TRANSPARENT_HUGEPAGE
828 # UP and nommu archs use km based percpu allocator
830 config NEED_PER_CPU_KM
831 depends on !SMP || !MMU
835 config NEED_PER_CPU_EMBED_FIRST_CHUNK
838 config NEED_PER_CPU_PAGE_FIRST_CHUNK
841 config USE_PERCPU_NUMA_NODE_ID
844 config HAVE_SETUP_PER_CPU_AREA
851 bool "Contiguous Memory Allocator"
854 select MEMORY_ISOLATION
856 This enables the Contiguous Memory Allocator which allows other
857 subsystems to allocate big physically-contiguous blocks of memory.
858 CMA reserves a region of memory and allows only movable pages to
859 be allocated from it. This way, the kernel can use the memory for
860 pagecache and when a subsystem requests for contiguous area, the
861 allocated pages are migrated away to serve the contiguous request.
866 bool "CMA debug messages (DEVELOPMENT)"
867 depends on DEBUG_KERNEL && CMA
869 Turns on debug messages in CMA. This produces KERN_DEBUG
870 messages for every CMA call as well as various messages while
871 processing calls such as dma_alloc_from_contiguous().
872 This option does not affect warning and error messages.
875 bool "CMA debugfs interface"
876 depends on CMA && DEBUG_FS
878 Turns on the DebugFS interface for CMA.
881 bool "CMA information through sysfs interface"
882 depends on CMA && SYSFS
884 This option exposes some sysfs attributes to get information
888 int "Maximum count of the CMA areas"
893 CMA allows to create CMA areas for particular purpose, mainly,
894 used as device private area. This parameter sets the maximum
895 number of CMA area in the system.
897 If unsure, leave the default value "7" in UMA and "19" in NUMA.
899 config MEM_SOFT_DIRTY
900 bool "Track memory changes"
901 depends on CHECKPOINT_RESTORE && HAVE_ARCH_SOFT_DIRTY && PROC_FS
902 select PROC_PAGE_MONITOR
904 This option enables memory changes tracking by introducing a
905 soft-dirty bit on pte-s. This bit it set when someone writes
906 into a page just as regular dirty bit, but unlike the latter
907 it can be cleared by hands.
909 See Documentation/admin-guide/mm/soft-dirty.rst for more details.
911 config GENERIC_EARLY_IOREMAP
914 config STACK_MAX_DEFAULT_SIZE_MB
915 int "Default maximum user stack size for 32-bit processes (MB)"
918 depends on STACK_GROWSUP && (!64BIT || COMPAT)
920 This is the maximum stack size in Megabytes in the VM layout of 32-bit
921 user processes when the stack grows upwards (currently only on parisc
922 arch) when the RLIMIT_STACK hard limit is unlimited.
924 A sane initial value is 100 MB.
926 config DEFERRED_STRUCT_PAGE_INIT
927 bool "Defer initialisation of struct pages to kthreads"
929 depends on !NEED_PER_CPU_KM
933 Ordinarily all struct pages are initialised during early boot in a
934 single thread. On very large machines this can take a considerable
935 amount of time. If this option is set, large machines will bring up
936 a subset of memmap at boot and then initialise the rest in parallel.
937 This has a potential performance impact on tasks running early in the
938 lifetime of the system until these kthreads finish the
941 config PAGE_IDLE_FLAG
943 select PAGE_EXTENSION if !64BIT
945 This adds PG_idle and PG_young flags to 'struct page'. PTE Accessed
946 bit writers can set the state of the bit in the flags so that PTE
947 Accessed bit readers may avoid disturbance.
949 config IDLE_PAGE_TRACKING
950 bool "Enable idle page tracking"
951 depends on SYSFS && MMU
952 select PAGE_IDLE_FLAG
954 This feature allows to estimate the amount of user pages that have
955 not been touched during a given period of time. This information can
956 be useful to tune memory cgroup limits and/or for job placement
957 within a compute cluster.
959 See Documentation/admin-guide/mm/idle_page_tracking.rst for
962 config ARCH_HAS_CACHE_LINE_SIZE
965 config ARCH_HAS_CURRENT_STACK_POINTER
968 In support of HARDENED_USERCOPY performing stack variable lifetime
969 checking, an architecture-agnostic way to find the stack pointer
970 is needed. Once an architecture defines an unsigned long global
971 register alias named "current_stack_pointer", this config can be
974 config ARCH_HAS_PTE_DEVMAP
977 config ARCH_HAS_ZONE_DMA_SET
981 bool "Support DMA zone" if ARCH_HAS_ZONE_DMA_SET
982 default y if ARM64 || X86
985 bool "Support DMA32 zone" if ARCH_HAS_ZONE_DMA_SET
990 bool "Device memory (pmem, HMM, etc...) hotplug support"
991 depends on MEMORY_HOTPLUG
992 depends on MEMORY_HOTREMOVE
993 depends on SPARSEMEM_VMEMMAP
994 depends on ARCH_HAS_PTE_DEVMAP
998 Device memory hotplug support allows for establishing pmem,
999 or other device driver discovered memory regions, in the
1000 memmap. This allows pfn_to_page() lookups of otherwise
1001 "device-physical" addresses which is needed for using a DAX
1002 mapping in an O_DIRECT operation, among other things.
1004 If FS_DAX is enabled, then say Y.
1007 # Helpers to mirror range of the CPU page tables of a process into device page
1014 config GET_FREE_REGION
1015 depends on SPARSEMEM
1018 config DEVICE_PRIVATE
1019 bool "Unaddressable device memory (GPU memory, ...)"
1020 depends on ZONE_DEVICE
1021 select GET_FREE_REGION
1024 Allows creation of struct pages to represent unaddressable device
1025 memory; i.e., memory that is only accessible from the device (or
1026 group of devices). You likely also want to select HMM_MIRROR.
1031 config ARCH_USES_HIGH_VMA_FLAGS
1033 config ARCH_HAS_PKEYS
1036 config ARCH_USES_PG_ARCH_X
1039 Enable the definition of PG_arch_x page flags with x > 1. Only
1040 suitable for 64-bit architectures with CONFIG_FLATMEM or
1041 CONFIG_SPARSEMEM_VMEMMAP enabled, otherwise there may not be
1042 enough room for additional bits in page->flags.
1044 config VM_EVENT_COUNTERS
1046 bool "Enable VM event counters for /proc/vmstat" if EXPERT
1048 VM event counters are needed for event counts to be shown.
1049 This option allows the disabling of the VM event counters
1050 on EXPERT systems. /proc/vmstat will only show page counts
1051 if VM event counters are disabled.
1054 bool "Collect percpu memory statistics"
1056 This feature collects and exposes statistics via debugfs. The
1057 information includes global and per chunk statistics, which can
1058 be used to help understand percpu memory usage.
1061 bool "Enable infrastructure for get_user_pages()-related unit tests"
1064 Provides /sys/kernel/debug/gup_test, which in turn provides a way
1065 to make ioctl calls that can launch kernel-based unit tests for
1066 the get_user_pages*() and pin_user_pages*() family of API calls.
1068 These tests include benchmark testing of the _fast variants of
1069 get_user_pages*() and pin_user_pages*(), as well as smoke tests of
1070 the non-_fast variants.
1072 There is also a sub-test that allows running dump_page() on any
1073 of up to eight pages (selected by command line args) within the
1074 range of user-space addresses. These pages are either pinned via
1075 pin_user_pages*(), or pinned via get_user_pages*(), as specified
1076 by other command line arguments.
1078 See tools/testing/selftests/mm/gup_test.c
1080 comment "GUP_TEST needs to have DEBUG_FS enabled"
1081 depends on !GUP_TEST && !DEBUG_FS
1083 config GUP_GET_PXX_LOW_HIGH
1087 tristate "Enable a module to run time tests on dma_pool"
1090 Provides a test module that will allocate and free many blocks of
1091 various sizes and report how long it takes. This is intended to
1092 provide a consistent way to measure how changes to the
1093 dma_pool_alloc/free routines affect performance.
1095 config ARCH_HAS_PTE_SPECIAL
1099 # Some architectures require a special hugepage directory format that is
1100 # required to support multiple hugepage sizes. For example a4fe3ce76
1101 # "powerpc/mm: Allow more flexible layouts for hugepage pagetables"
1102 # introduced it on powerpc. This allows for a more flexible hugepage
1103 # pagetable layouts.
1105 config ARCH_HAS_HUGEPD
1108 config MAPPING_DIRTY_HELPERS
1114 config KMAP_LOCAL_NON_LINEAR_PTE_ARRAY
1117 # struct io_mapping based helper. Selected by drivers that need them
1123 bool "Enable memfd_secret() system call" if EXPERT
1124 depends on ARCH_HAS_SET_DIRECT_MAP
1126 Enable the memfd_secret() system call with the ability to create
1127 memory areas visible only in the context of the owning process and
1128 not mapped to other processes and other kernel page tables.
1130 config ANON_VMA_NAME
1131 bool "Anonymous VMA name support"
1132 depends on PROC_FS && ADVISE_SYSCALLS && MMU
1135 Allow naming anonymous virtual memory areas.
1137 This feature allows assigning names to virtual memory areas. Assigned
1138 names can be later retrieved from /proc/pid/maps and /proc/pid/smaps
1139 and help identifying individual anonymous memory areas.
1140 Assigning a name to anonymous virtual memory area might prevent that
1141 area from being merged with adjacent virtual memory areas due to the
1142 difference in their name.
1145 bool "Enable userfaultfd() system call"
1148 Enable the userfaultfd() system call that allows to intercept and
1149 handle page faults in userland.
1151 config HAVE_ARCH_USERFAULTFD_WP
1154 Arch has userfaultfd write protection support
1156 config HAVE_ARCH_USERFAULTFD_MINOR
1159 Arch has userfaultfd minor fault support
1161 config PTE_MARKER_UFFD_WP
1162 bool "Userfaultfd write protection support for shmem/hugetlbfs"
1164 depends on HAVE_ARCH_USERFAULTFD_WP
1167 Allows to create marker PTEs for userfaultfd write protection
1168 purposes. It is required to enable userfaultfd write protection on
1169 file-backed memory types like shmem and hugetlbfs.
1173 bool "Multi-Gen LRU"
1175 # make sure folio->flags has enough spare bits
1176 depends on 64BIT || !SPARSEMEM || SPARSEMEM_VMEMMAP
1178 A high performance LRU implementation to overcommit memory. See
1179 Documentation/admin-guide/mm/multigen_lru.rst for details.
1181 config LRU_GEN_ENABLED
1182 bool "Enable by default"
1185 This option enables the multi-gen LRU by default.
1187 config LRU_GEN_STATS
1188 bool "Full stats for debugging"
1191 Do not enable this option unless you plan to look at historical stats
1192 from evicted generations for debugging purpose.
1194 This option has a per-memcg and per-node memory overhead.
1197 config ARCH_SUPPORTS_PER_VMA_LOCK
1202 depends on ARCH_SUPPORTS_PER_VMA_LOCK && MMU && SMP
1204 Allow per-vma locking during page fault handling.
1206 This feature allows locking each virtual memory area separately when
1207 handling page faults instead of taking mmap_lock.
1209 source "mm/damon/Kconfig"