1301 lines · plain
1# SPDX-License-Identifier: GPL-2.0-only2 3menu "Memory Management options"4 5#6# For some reason microblaze and nios2 hard code SWAP=n. Hopefully we can7# add proper SWAP support to them, in which case this can be remove.8#9config ARCH_NO_SWAP10 bool11 12config ZPOOL13 bool14 15menuconfig SWAP16 bool "Support for paging of anonymous memory (swap)"17 depends on MMU && BLOCK && !ARCH_NO_SWAP18 default y19 help20 This option allows you to choose whether you want to have support21 for so called swap devices or swap files in your kernel that are22 used to provide more virtual memory than the actual RAM present23 in your computer. If unsure say Y.24 25config ZSWAP26 bool "Compressed cache for swap pages"27 depends on SWAP28 select CRYPTO29 select ZPOOL30 help31 A lightweight compressed cache for swap pages. It takes32 pages that are in the process of being swapped out and attempts to33 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 device36 reads, can also improve workload performance.37 38config ZSWAP_DEFAULT_ON39 bool "Enable the compressed cache for swap pages by default"40 depends on ZSWAP41 help42 If selected, the compressed cache for swap pages will be enabled43 at boot, otherwise it will be disabled.44 45 The selection made here can be overridden by using the kernel46 command line 'zswap.enabled=' option.47 48config ZSWAP_SHRINKER_DEFAULT_ON49 bool "Shrink the zswap pool on memory pressure"50 depends on ZSWAP51 default n52 help53 If selected, the zswap shrinker will be enabled, and the pages54 stored in the zswap pool will become available for reclaim (i.e55 written back to the backing swap device) on memory pressure.56 57 This means that zswap writeback could happen even if the pool is58 not yet full, or the cgroup zswap limit has not been reached,59 reducing the chance that cold pages will reside in the zswap pool60 and consume memory indefinitely.61 62choice63 prompt "Default compressor"64 depends on ZSWAP65 default ZSWAP_COMPRESSOR_DEFAULT_LZO66 help67 Selects the default compression algorithm for the compressed cache68 for swap pages.69 70 For an overview what kind of performance can be expected from71 a particular compression algorithm please refer to the benchmarks72 available at the following LWN page:73 https://lwn.net/Articles/751795/74 75 If in doubt, select 'LZO'.76 77 The selection made here can be overridden by using the kernel78 command line 'zswap.compressor=' option.79 80config ZSWAP_COMPRESSOR_DEFAULT_DEFLATE81 bool "Deflate"82 select CRYPTO_DEFLATE83 help84 Use the Deflate algorithm as the default compression algorithm.85 86config ZSWAP_COMPRESSOR_DEFAULT_LZO87 bool "LZO"88 select CRYPTO_LZO89 help90 Use the LZO algorithm as the default compression algorithm.91 92config ZSWAP_COMPRESSOR_DEFAULT_84293 bool "842"94 select CRYPTO_84295 help96 Use the 842 algorithm as the default compression algorithm.97 98config ZSWAP_COMPRESSOR_DEFAULT_LZ499 bool "LZ4"100 select CRYPTO_LZ4101 help102 Use the LZ4 algorithm as the default compression algorithm.103 104config ZSWAP_COMPRESSOR_DEFAULT_LZ4HC105 bool "LZ4HC"106 select CRYPTO_LZ4HC107 help108 Use the LZ4HC algorithm as the default compression algorithm.109 110config ZSWAP_COMPRESSOR_DEFAULT_ZSTD111 bool "zstd"112 select CRYPTO_ZSTD113 help114 Use the zstd algorithm as the default compression algorithm.115endchoice116 117config ZSWAP_COMPRESSOR_DEFAULT118 string119 depends on ZSWAP120 default "deflate" if ZSWAP_COMPRESSOR_DEFAULT_DEFLATE121 default "lzo" if ZSWAP_COMPRESSOR_DEFAULT_LZO122 default "842" if ZSWAP_COMPRESSOR_DEFAULT_842123 default "lz4" if ZSWAP_COMPRESSOR_DEFAULT_LZ4124 default "lz4hc" if ZSWAP_COMPRESSOR_DEFAULT_LZ4HC125 default "zstd" if ZSWAP_COMPRESSOR_DEFAULT_ZSTD126 default ""127 128choice129 prompt "Default allocator"130 depends on ZSWAP131 default ZSWAP_ZPOOL_DEFAULT_ZSMALLOC if MMU132 default ZSWAP_ZPOOL_DEFAULT_ZBUD133 help134 Selects the default allocator for the compressed cache for135 swap pages.136 The default is 'zbud' for compatibility, however please do137 read the description of each of the allocators below before138 making a right choice.139 140 The selection made here can be overridden by using the kernel141 command line 'zswap.zpool=' option.142 143config ZSWAP_ZPOOL_DEFAULT_ZBUD144 bool "zbud"145 select ZBUD146 help147 Use the zbud allocator as the default allocator.148 149config ZSWAP_ZPOOL_DEFAULT_Z3FOLD_DEPRECATED150 bool "z3foldi (DEPRECATED)"151 select Z3FOLD_DEPRECATED152 help153 Use the z3fold allocator as the default allocator.154 155 Deprecated and scheduled for removal in a few cycles,156 see CONFIG_Z3FOLD_DEPRECATED.157 158config ZSWAP_ZPOOL_DEFAULT_ZSMALLOC159 bool "zsmalloc"160 select ZSMALLOC161 help162 Use the zsmalloc allocator as the default allocator.163endchoice164 165config ZSWAP_ZPOOL_DEFAULT166 string167 depends on ZSWAP168 default "zbud" if ZSWAP_ZPOOL_DEFAULT_ZBUD169 default "z3fold" if ZSWAP_ZPOOL_DEFAULT_Z3FOLD_DEPRECATED170 default "zsmalloc" if ZSWAP_ZPOOL_DEFAULT_ZSMALLOC171 default ""172 173config ZBUD174 tristate "2:1 compression allocator (zbud)"175 depends on ZSWAP176 help177 A special purpose allocator for storing compressed pages.178 It is designed to store up to two compressed pages per physical179 page. While this design limits storage density, it has simple and180 deterministic reclaim properties that make it preferable to a higher181 density approach when reclaim will be used.182 183config Z3FOLD_DEPRECATED184 tristate "3:1 compression allocator (z3fold) (DEPRECATED)"185 depends on ZSWAP186 help187 Deprecated and scheduled for removal in a few cycles. If you have188 a good reason for using Z3FOLD over ZSMALLOC, please contact189 linux-mm@kvack.org and the zswap maintainers.190 191 A special purpose allocator for storing compressed pages.192 It is designed to store up to three compressed pages per physical193 page. It is a ZBUD derivative so the simplicity and determinism are194 still there.195 196config Z3FOLD197 tristate198 default y if Z3FOLD_DEPRECATED=y199 default m if Z3FOLD_DEPRECATED=m200 depends on Z3FOLD_DEPRECATED201 202config ZSMALLOC203 tristate204 prompt "N:1 compression allocator (zsmalloc)" if (ZSWAP || ZRAM)205 depends on MMU206 help207 zsmalloc is a slab-based memory allocator designed to store208 pages of various compression levels efficiently. It achieves209 the highest storage density with the least amount of fragmentation.210 211config ZSMALLOC_STAT212 bool "Export zsmalloc statistics"213 depends on ZSMALLOC214 select DEBUG_FS215 help216 This option enables code in the zsmalloc to collect various217 statistics about what's happening in zsmalloc and exports that218 information to userspace via debugfs.219 If unsure, say N.220 221config ZSMALLOC_CHAIN_SIZE222 int "Maximum number of physical pages per-zspage"223 default 8224 range 4 16225 depends on ZSMALLOC226 help227 This option sets the upper limit on the number of physical pages228 that a zmalloc page (zspage) can consist of. The optimal zspage229 chain size is calculated for each size class during the230 initialization of the pool.231 232 Changing this option can alter the characteristics of size classes,233 such as the number of pages per zspage and the number of objects234 per zspage. This can also result in different configurations of235 the pool, as zsmalloc merges size classes with similar236 characteristics.237 238 For more information, see zsmalloc documentation.239 240menu "Slab allocator options"241 242config SLUB243 def_bool y244 245config SLUB_TINY246 bool "Configure for minimal memory footprint"247 depends on EXPERT248 select SLAB_MERGE_DEFAULT249 help250 Configures the slab allocator in a way to achieve minimal memory251 footprint, sacrificing scalability, debugging and other features.252 This is intended only for the smallest system that had used the253 SLOB allocator and is not recommended for systems with more than254 16MB RAM.255 256 If unsure, say N.257 258config SLAB_MERGE_DEFAULT259 bool "Allow slab caches to be merged"260 default y261 help262 For reduced kernel memory fragmentation, slab caches can be263 merged when they share the same size and other characteristics.264 This carries a risk of kernel heap overflows being able to265 overwrite objects from merged caches (and more easily control266 cache layout), which makes such heap attacks easier to exploit267 by attackers. By keeping caches unmerged, these kinds of exploits268 can usually only damage objects in the same cache. To disable269 merging at runtime, "slab_nomerge" can be passed on the kernel270 command line.271 272config SLAB_FREELIST_RANDOM273 bool "Randomize slab freelist"274 depends on !SLUB_TINY275 help276 Randomizes the freelist order used on creating new pages. This277 security feature reduces the predictability of the kernel slab278 allocator against heap overflows.279 280config SLAB_FREELIST_HARDENED281 bool "Harden slab freelist metadata"282 depends on !SLUB_TINY283 help284 Many kernel heap attacks try to target slab cache metadata and285 other infrastructure. This options makes minor performance286 sacrifices to harden the kernel slab allocator against common287 freelist exploit methods.288 289config SLAB_BUCKETS290 bool "Support allocation from separate kmalloc buckets"291 depends on !SLUB_TINY292 default SLAB_FREELIST_HARDENED293 help294 Kernel heap attacks frequently depend on being able to create295 specifically-sized allocations with user-controlled contents296 that will be allocated into the same kmalloc bucket as a297 target object. To avoid sharing these allocation buckets,298 provide an explicitly separated set of buckets to be used for299 user-controlled allocations. This may very slightly increase300 memory fragmentation, though in practice it's only a handful301 of extra pages since the bulk of user-controlled allocations302 are relatively long-lived.303 304 If unsure, say Y.305 306config SLUB_STATS307 default n308 bool "Enable performance statistics"309 depends on SYSFS && !SLUB_TINY310 help311 The statistics are useful to debug slab allocation behavior in312 order find ways to optimize the allocator. This should never be313 enabled for production use since keeping statistics slows down314 the allocator by a few percentage points. The slabinfo command315 supports the determination of the most active slabs to figure316 out which slabs are relevant to a particular load.317 Try running: slabinfo -DA318 319config SLUB_CPU_PARTIAL320 default y321 depends on SMP && !SLUB_TINY322 bool "Enable per cpu partial caches"323 help324 Per cpu partial caches accelerate objects allocation and freeing325 that is local to a processor at the price of more indeterminism326 in the latency of the free. On overflow these caches will be cleared327 which requires the taking of locks that may cause latency spikes.328 Typically one would choose no for a realtime system.329 330config RANDOM_KMALLOC_CACHES331 default n332 depends on !SLUB_TINY333 bool "Randomize slab caches for normal kmalloc"334 help335 A hardening feature that creates multiple copies of slab caches for336 normal kmalloc allocation and makes kmalloc randomly pick one based337 on code address, which makes the attackers more difficult to spray338 vulnerable memory objects on the heap for the purpose of exploiting339 memory vulnerabilities.340 341 Currently the number of copies is set to 16, a reasonably large value342 that effectively diverges the memory objects allocated for different343 subsystems or modules into different caches, at the expense of a344 limited degree of memory and CPU overhead that relates to hardware and345 system workload.346 347endmenu # Slab allocator options348 349config SHUFFLE_PAGE_ALLOCATOR350 bool "Page allocator randomization"351 default SLAB_FREELIST_RANDOM && ACPI_NUMA352 help353 Randomization of the page allocator improves the average354 utilization of a direct-mapped memory-side-cache. See section355 5.2.27 Heterogeneous Memory Attribute Table (HMAT) in the ACPI356 6.2a specification for an example of how a platform advertises357 the presence of a memory-side-cache. There are also incidental358 security benefits as it reduces the predictability of page359 allocations to compliment SLAB_FREELIST_RANDOM, but the360 default granularity of shuffling on the MAX_PAGE_ORDER i.e, 10th361 order of pages is selected based on cache utilization benefits362 on x86.363 364 While the randomization improves cache utilization it may365 negatively impact workloads on platforms without a cache. For366 this reason, by default, the randomization is not enabled even367 if SHUFFLE_PAGE_ALLOCATOR=y. The randomization may be force enabled368 with the 'page_alloc.shuffle' kernel command line parameter.369 370 Say Y if unsure.371 372config COMPAT_BRK373 bool "Disable heap randomization"374 default y375 help376 Randomizing heap placement makes heap exploits harder, but it377 also breaks ancient binaries (including anything libc5 based).378 This option changes the bootup default to heap randomization379 disabled, and can be overridden at runtime by setting380 /proc/sys/kernel/randomize_va_space to 2.381 382 On non-ancient distros (post-2000 ones) N is usually a safe choice.383 384config MMAP_ALLOW_UNINITIALIZED385 bool "Allow mmapped anonymous memory to be uninitialized"386 depends on EXPERT && !MMU387 default n388 help389 Normally, and according to the Linux spec, anonymous memory obtained390 from mmap() has its contents cleared before it is passed to391 userspace. Enabling this config option allows you to request that392 mmap() skip that if it is given an MAP_UNINITIALIZED flag, thus393 providing a huge performance boost. If this option is not enabled,394 then the flag will be ignored.395 396 This is taken advantage of by uClibc's malloc(), and also by397 ELF-FDPIC binfmt's brk and stack allocator.398 399 Because of the obvious security issues, this option should only be400 enabled on embedded devices where you control what is run in401 userspace. Since that isn't generally a problem on no-MMU systems,402 it is normally safe to say Y here.403 404 See Documentation/admin-guide/mm/nommu-mmap.rst for more information.405 406config SELECT_MEMORY_MODEL407 def_bool y408 depends on ARCH_SELECT_MEMORY_MODEL409 410choice411 prompt "Memory model"412 depends on SELECT_MEMORY_MODEL413 default SPARSEMEM_MANUAL if ARCH_SPARSEMEM_DEFAULT414 default FLATMEM_MANUAL415 help416 This option allows you to change some of the ways that417 Linux manages its memory internally. Most users will418 only have one option here selected by the architecture419 configuration. This is normal.420 421config FLATMEM_MANUAL422 bool "Flat Memory"423 depends on !ARCH_SPARSEMEM_ENABLE || ARCH_FLATMEM_ENABLE424 help425 This option is best suited for non-NUMA systems with426 flat address space. The FLATMEM is the most efficient427 system in terms of performance and resource consumption428 and it is the best option for smaller systems.429 430 For systems that have holes in their physical address431 spaces and for features like NUMA and memory hotplug,432 choose "Sparse Memory".433 434 If unsure, choose this option (Flat Memory) over any other.435 436config SPARSEMEM_MANUAL437 bool "Sparse Memory"438 depends on ARCH_SPARSEMEM_ENABLE439 help440 This will be the only option for some systems, including441 memory hot-plug systems. This is normal.442 443 This option provides efficient support for systems with444 holes is their physical address space and allows memory445 hot-plug and hot-remove.446 447 If unsure, choose "Flat Memory" over this option.448 449endchoice450 451config SPARSEMEM452 def_bool y453 depends on (!SELECT_MEMORY_MODEL && ARCH_SPARSEMEM_ENABLE) || SPARSEMEM_MANUAL454 455config FLATMEM456 def_bool y457 depends on !SPARSEMEM || FLATMEM_MANUAL458 459#460# SPARSEMEM_EXTREME (which is the default) does some bootmem461# allocations when sparse_init() is called. If this cannot462# be done on your architecture, select this option. However,463# statically allocating the mem_section[] array can potentially464# consume vast quantities of .bss, so be careful.465#466# This option will also potentially produce smaller runtime code467# with gcc 3.4 and later.468#469config SPARSEMEM_STATIC470 bool471 472#473# Architecture platforms which require a two level mem_section in SPARSEMEM474# must select this option. This is usually for architecture platforms with475# an extremely sparse physical address space.476#477config SPARSEMEM_EXTREME478 def_bool y479 depends on SPARSEMEM && !SPARSEMEM_STATIC480 481config SPARSEMEM_VMEMMAP_ENABLE482 bool483 484config SPARSEMEM_VMEMMAP485 bool "Sparse Memory virtual memmap"486 depends on SPARSEMEM && SPARSEMEM_VMEMMAP_ENABLE487 default y488 help489 SPARSEMEM_VMEMMAP uses a virtually mapped memmap to optimise490 pfn_to_page and page_to_pfn operations. This is the most491 efficient option when sufficient kernel resources are available.492#493# Select this config option from the architecture Kconfig, if it is preferred494# to enable the feature of HugeTLB/dev_dax vmemmap optimization.495#496config ARCH_WANT_OPTIMIZE_DAX_VMEMMAP497 bool498 499config ARCH_WANT_OPTIMIZE_HUGETLB_VMEMMAP500 bool501 502config HAVE_MEMBLOCK_PHYS_MAP503 bool504 505config HAVE_GUP_FAST506 depends on MMU507 bool508 509# Don't discard allocated memory used to track "memory" and "reserved" memblocks510# after early boot, so it can still be used to test for validity of memory.511# Also, memblocks are updated with memory hot(un)plug.512config ARCH_KEEP_MEMBLOCK513 bool514 515# Keep arch NUMA mapping infrastructure post-init.516config NUMA_KEEP_MEMINFO517 bool518 519config MEMORY_ISOLATION520 bool521 522# IORESOURCE_SYSTEM_RAM regions in the kernel resource tree that are marked523# IORESOURCE_EXCLUSIVE cannot be mapped to user space, for example, via524# /dev/mem.525config EXCLUSIVE_SYSTEM_RAM526 def_bool y527 depends on !DEVMEM || STRICT_DEVMEM528 529#530# Only be set on architectures that have completely implemented memory hotplug531# feature. If you are not sure, don't touch it.532#533config HAVE_BOOTMEM_INFO_NODE534 def_bool n535 536config ARCH_ENABLE_MEMORY_HOTPLUG537 bool538 539config ARCH_ENABLE_MEMORY_HOTREMOVE540 bool541 542# eventually, we can have this option just 'select SPARSEMEM'543menuconfig MEMORY_HOTPLUG544 bool "Memory hotplug"545 select MEMORY_ISOLATION546 depends on SPARSEMEM547 depends on ARCH_ENABLE_MEMORY_HOTPLUG548 depends on 64BIT549 select NUMA_KEEP_MEMINFO if NUMA550 551if MEMORY_HOTPLUG552 553config MEMORY_HOTPLUG_DEFAULT_ONLINE554 bool "Online the newly added memory blocks by default"555 depends on MEMORY_HOTPLUG556 help557 This option sets the default policy setting for memory hotplug558 onlining policy (/sys/devices/system/memory/auto_online_blocks) which559 determines what happens to newly added memory regions. Policy setting560 can always be changed at runtime.561 See Documentation/admin-guide/mm/memory-hotplug.rst for more information.562 563 Say Y here if you want all hot-plugged memory blocks to appear in564 'online' state by default.565 Say N here if you want the default policy to keep all hot-plugged566 memory blocks in 'offline' state.567 568config MEMORY_HOTREMOVE569 bool "Allow for memory hot remove"570 select HAVE_BOOTMEM_INFO_NODE if (X86_64 || PPC64)571 depends on MEMORY_HOTPLUG && ARCH_ENABLE_MEMORY_HOTREMOVE572 depends on MIGRATION573 574config MHP_MEMMAP_ON_MEMORY575 def_bool y576 depends on MEMORY_HOTPLUG && SPARSEMEM_VMEMMAP577 depends on ARCH_MHP_MEMMAP_ON_MEMORY_ENABLE578 579endif # MEMORY_HOTPLUG580 581config ARCH_MHP_MEMMAP_ON_MEMORY_ENABLE582 bool583 584# Heavily threaded applications may benefit from splitting the mm-wide585# page_table_lock, so that faults on different parts of the user address586# space can be handled with less contention: split it at this NR_CPUS.587# Default to 4 for wider testing, though 8 might be more appropriate.588# ARM's adjust_pte (unused if VIPT) depends on mm-wide page_table_lock.589# PA-RISC 7xxx's spinlock_t would enlarge struct page from 32 to 44 bytes.590# SPARC32 allocates multiple pte tables within a single page, and therefore591# a per-page lock leads to problems when multiple tables need to be locked592# at the same time (e.g. copy_page_range()).593# DEBUG_SPINLOCK and DEBUG_LOCK_ALLOC spinlock_t also enlarge struct page.594#595config SPLIT_PTE_PTLOCKS596 def_bool y597 depends on MMU598 depends on SMP599 depends on NR_CPUS >= 4600 depends on !ARM || CPU_CACHE_VIPT601 depends on !PARISC || PA20602 depends on !SPARC32603 604config ARCH_ENABLE_SPLIT_PMD_PTLOCK605 bool606 607config SPLIT_PMD_PTLOCKS608 def_bool y609 depends on SPLIT_PTE_PTLOCKS && ARCH_ENABLE_SPLIT_PMD_PTLOCK610 611#612# support for memory balloon613config MEMORY_BALLOON614 bool615 616#617# support for memory balloon compaction618config BALLOON_COMPACTION619 bool "Allow for balloon memory compaction/migration"620 default y621 depends on COMPACTION && MEMORY_BALLOON622 help623 Memory fragmentation introduced by ballooning might reduce624 significantly the number of 2MB contiguous memory blocks that can be625 used within a guest, thus imposing performance penalties associated626 with the reduced number of transparent huge pages that could be used627 by the guest workload. Allowing the compaction & migration for memory628 pages enlisted as being part of memory balloon devices avoids the629 scenario aforementioned and helps improving memory defragmentation.630 631#632# support for memory compaction633config COMPACTION634 bool "Allow for memory compaction"635 default y636 select MIGRATION637 depends on MMU638 help639 Compaction is the only memory management component to form640 high order (larger physically contiguous) memory blocks641 reliably. The page allocator relies on compaction heavily and642 the lack of the feature can lead to unexpected OOM killer643 invocations for high order memory requests. You shouldn't644 disable this option unless there really is a strong reason for645 it and then we would be really interested to hear about that at646 linux-mm@kvack.org.647 648config COMPACT_UNEVICTABLE_DEFAULT649 int650 depends on COMPACTION651 default 0 if PREEMPT_RT652 default 1653 654#655# support for free page reporting656config PAGE_REPORTING657 bool "Free page reporting"658 help659 Free page reporting allows for the incremental acquisition of660 free pages from the buddy allocator for the purpose of reporting661 those pages to another entity, such as a hypervisor, so that the662 memory can be freed within the host for other uses.663 664#665# support for page migration666#667config MIGRATION668 bool "Page migration"669 default y670 depends on (NUMA || ARCH_ENABLE_MEMORY_HOTREMOVE || COMPACTION || CMA) && MMU671 help672 Allows the migration of the physical location of pages of processes673 while the virtual addresses are not changed. This is useful in674 two situations. The first is on NUMA systems to put pages nearer675 to the processors accessing. The second is when allocating huge676 pages as migration can relocate pages to satisfy a huge page677 allocation instead of reclaiming.678 679config DEVICE_MIGRATION680 def_bool MIGRATION && ZONE_DEVICE681 682config ARCH_ENABLE_HUGEPAGE_MIGRATION683 bool684 685config ARCH_ENABLE_THP_MIGRATION686 bool687 688config HUGETLB_PAGE_SIZE_VARIABLE689 def_bool n690 help691 Allows the pageblock_order value to be dynamic instead of just standard692 HUGETLB_PAGE_ORDER when there are multiple HugeTLB page sizes available693 on a platform.694 695 Note that the pageblock_order cannot exceed MAX_PAGE_ORDER and will be696 clamped down to MAX_PAGE_ORDER.697 698config CONTIG_ALLOC699 def_bool (MEMORY_ISOLATION && COMPACTION) || CMA700 701config PCP_BATCH_SCALE_MAX702 int "Maximum scale factor of PCP (Per-CPU pageset) batch allocate/free"703 default 5704 range 0 6705 help706 In page allocator, PCP (Per-CPU pageset) is refilled and drained in707 batches. The batch number is scaled automatically to improve page708 allocation/free throughput. But too large scale factor may hurt709 latency. This option sets the upper limit of scale factor to limit710 the maximum latency.711 712config PHYS_ADDR_T_64BIT713 def_bool 64BIT714 715config BOUNCE716 bool "Enable bounce buffers"717 default y718 depends on BLOCK && MMU && HIGHMEM719 help720 Enable bounce buffers for devices that cannot access the full range of721 memory available to the CPU. Enabled by default when HIGHMEM is722 selected, but you may say n to override this.723 724config MMU_NOTIFIER725 bool726 select INTERVAL_TREE727 728config KSM729 bool "Enable KSM for page merging"730 depends on MMU731 select XXHASH732 help733 Enable Kernel Samepage Merging: KSM periodically scans those areas734 of an application's address space that an app has advised may be735 mergeable. When it finds pages of identical content, it replaces736 the many instances by a single page with that content, so737 saving memory until one or another app needs to modify the content.738 Recommended for use with KVM, or with other duplicative applications.739 See Documentation/mm/ksm.rst for more information: KSM is inactive740 until a program has madvised that an area is MADV_MERGEABLE, and741 root has set /sys/kernel/mm/ksm/run to 1 (if CONFIG_SYSFS is set).742 743config DEFAULT_MMAP_MIN_ADDR744 int "Low address space to protect from user allocation"745 depends on MMU746 default 4096747 help748 This is the portion of low virtual memory which should be protected749 from userspace allocation. Keeping a user from writing to low pages750 can help reduce the impact of kernel NULL pointer bugs.751 752 For most arm64, ppc64 and x86 users with lots of address space753 a value of 65536 is reasonable and should cause no problems.754 On arm and other archs it should not be higher than 32768.755 Programs which use vm86 functionality or have some need to map756 this low address space will need CAP_SYS_RAWIO or disable this757 protection by setting the value to 0.758 759 This value can be changed after boot using the760 /proc/sys/vm/mmap_min_addr tunable.761 762config ARCH_SUPPORTS_MEMORY_FAILURE763 bool764 765config MEMORY_FAILURE766 depends on MMU767 depends on ARCH_SUPPORTS_MEMORY_FAILURE768 bool "Enable recovery from hardware memory errors"769 select MEMORY_ISOLATION770 select RAS771 help772 Enables code to recover from some memory failures on systems773 with MCA recovery. This allows a system to continue running774 even when some of its memory has uncorrected errors. This requires775 special hardware support and typically ECC memory.776 777config HWPOISON_INJECT778 tristate "HWPoison pages injector"779 depends on MEMORY_FAILURE && DEBUG_KERNEL && PROC_FS780 select PROC_PAGE_MONITOR781 782config NOMMU_INITIAL_TRIM_EXCESS783 int "Turn on mmap() excess space trimming before booting"784 depends on !MMU785 default 1786 help787 The NOMMU mmap() frequently needs to allocate large contiguous chunks788 of memory on which to store mappings, but it can only ask the system789 allocator for chunks in 2^N*PAGE_SIZE amounts - which is frequently790 more than it requires. To deal with this, mmap() is able to trim off791 the excess and return it to the allocator.792 793 If trimming is enabled, the excess is trimmed off and returned to the794 system allocator, which can cause extra fragmentation, particularly795 if there are a lot of transient processes.796 797 If trimming is disabled, the excess is kept, but not used, which for798 long-term mappings means that the space is wasted.799 800 Trimming can be dynamically controlled through a sysctl option801 (/proc/sys/vm/nr_trim_pages) which specifies the minimum number of802 excess pages there must be before trimming should occur, or zero if803 no trimming is to occur.804 805 This option specifies the initial value of this option. The default806 of 1 says that all excess pages should be trimmed.807 808 See Documentation/admin-guide/mm/nommu-mmap.rst for more information.809 810config ARCH_WANT_GENERAL_HUGETLB811 bool812 813config ARCH_WANTS_THP_SWAP814 def_bool n815 816menuconfig TRANSPARENT_HUGEPAGE817 bool "Transparent Hugepage Support"818 depends on HAVE_ARCH_TRANSPARENT_HUGEPAGE && !PREEMPT_RT819 select COMPACTION820 select XARRAY_MULTI821 help822 Transparent Hugepages allows the kernel to use huge pages and823 huge tlb transparently to the applications whenever possible.824 This feature can improve computing performance to certain825 applications by speeding up page faults during memory826 allocation, by reducing the number of tlb misses and by speeding827 up the pagetable walking.828 829 If memory constrained on embedded, you may want to say N.830 831if TRANSPARENT_HUGEPAGE832 833choice834 prompt "Transparent Hugepage Support sysfs defaults"835 depends on TRANSPARENT_HUGEPAGE836 default TRANSPARENT_HUGEPAGE_ALWAYS837 help838 Selects the sysfs defaults for Transparent Hugepage Support.839 840 config TRANSPARENT_HUGEPAGE_ALWAYS841 bool "always"842 help843 Enabling Transparent Hugepage always, can increase the844 memory footprint of applications without a guaranteed845 benefit but it will work automatically for all applications.846 847 config TRANSPARENT_HUGEPAGE_MADVISE848 bool "madvise"849 help850 Enabling Transparent Hugepage madvise, will only provide a851 performance improvement benefit to the applications using852 madvise(MADV_HUGEPAGE) but it won't risk to increase the853 memory footprint of applications without a guaranteed854 benefit.855 856 config TRANSPARENT_HUGEPAGE_NEVER857 bool "never"858 help859 Disable Transparent Hugepage by default. It can still be860 enabled at runtime via sysfs.861endchoice862 863config THP_SWAP864 def_bool y865 depends on TRANSPARENT_HUGEPAGE && ARCH_WANTS_THP_SWAP && SWAP && 64BIT866 help867 Swap transparent huge pages in one piece, without splitting.868 XXX: For now, swap cluster backing transparent huge page869 will be split after swapout.870 871 For selection by architectures with reasonable THP sizes.872 873config READ_ONLY_THP_FOR_FS874 bool "Read-only THP for filesystems (EXPERIMENTAL)"875 depends on TRANSPARENT_HUGEPAGE && SHMEM876 877 help878 Allow khugepaged to put read-only file-backed pages in THP.879 880 This is marked experimental because it is a new feature. Write881 support of file THPs will be developed in the next few release882 cycles.883 884endif # TRANSPARENT_HUGEPAGE885 886#887# The architecture supports pgtable leaves that is larger than PAGE_SIZE888#889config PGTABLE_HAS_HUGE_LEAVES890 def_bool TRANSPARENT_HUGEPAGE || HUGETLB_PAGE891 892# TODO: Allow to be enabled without THP893config ARCH_SUPPORTS_HUGE_PFNMAP894 def_bool n895 depends on TRANSPARENT_HUGEPAGE896 897config ARCH_SUPPORTS_PMD_PFNMAP898 def_bool y899 depends on ARCH_SUPPORTS_HUGE_PFNMAP && HAVE_ARCH_TRANSPARENT_HUGEPAGE900 901config ARCH_SUPPORTS_PUD_PFNMAP902 def_bool y903 depends on ARCH_SUPPORTS_HUGE_PFNMAP && HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD904 905#906# UP and nommu archs use km based percpu allocator907#908config NEED_PER_CPU_KM909 depends on !SMP || !MMU910 bool911 default y912 913config NEED_PER_CPU_EMBED_FIRST_CHUNK914 bool915 916config NEED_PER_CPU_PAGE_FIRST_CHUNK917 bool918 919config USE_PERCPU_NUMA_NODE_ID920 bool921 922config HAVE_SETUP_PER_CPU_AREA923 bool924 925config CMA926 bool "Contiguous Memory Allocator"927 depends on MMU928 select MIGRATION929 select MEMORY_ISOLATION930 help931 This enables the Contiguous Memory Allocator which allows other932 subsystems to allocate big physically-contiguous blocks of memory.933 CMA reserves a region of memory and allows only movable pages to934 be allocated from it. This way, the kernel can use the memory for935 pagecache and when a subsystem requests for contiguous area, the936 allocated pages are migrated away to serve the contiguous request.937 938 If unsure, say "n".939 940config CMA_DEBUGFS941 bool "CMA debugfs interface"942 depends on CMA && DEBUG_FS943 help944 Turns on the DebugFS interface for CMA.945 946config CMA_SYSFS947 bool "CMA information through sysfs interface"948 depends on CMA && SYSFS949 help950 This option exposes some sysfs attributes to get information951 from CMA.952 953config CMA_AREAS954 int "Maximum count of the CMA areas"955 depends on CMA956 default 20 if NUMA957 default 8958 help959 CMA allows to create CMA areas for particular purpose, mainly,960 used as device private area. This parameter sets the maximum961 number of CMA area in the system.962 963 If unsure, leave the default value "8" in UMA and "20" in NUMA.964 965config MEM_SOFT_DIRTY966 bool "Track memory changes"967 depends on CHECKPOINT_RESTORE && HAVE_ARCH_SOFT_DIRTY && PROC_FS968 select PROC_PAGE_MONITOR969 help970 This option enables memory changes tracking by introducing a971 soft-dirty bit on pte-s. This bit it set when someone writes972 into a page just as regular dirty bit, but unlike the latter973 it can be cleared by hands.974 975 See Documentation/admin-guide/mm/soft-dirty.rst for more details.976 977config GENERIC_EARLY_IOREMAP978 bool979 980config STACK_MAX_DEFAULT_SIZE_MB981 int "Default maximum user stack size for 32-bit processes (MB)"982 default 100983 range 8 2048984 depends on STACK_GROWSUP && (!64BIT || COMPAT)985 help986 This is the maximum stack size in Megabytes in the VM layout of 32-bit987 user processes when the stack grows upwards (currently only on parisc988 arch) when the RLIMIT_STACK hard limit is unlimited.989 990 A sane initial value is 100 MB.991 992config DEFERRED_STRUCT_PAGE_INIT993 bool "Defer initialisation of struct pages to kthreads"994 depends on SPARSEMEM995 depends on !NEED_PER_CPU_KM996 depends on 64BIT997 depends on !KMSAN998 select PADATA999 help1000 Ordinarily all struct pages are initialised during early boot in a1001 single thread. On very large machines this can take a considerable1002 amount of time. If this option is set, large machines will bring up1003 a subset of memmap at boot and then initialise the rest in parallel.1004 This has a potential performance impact on tasks running early in the1005 lifetime of the system until these kthreads finish the1006 initialisation.1007 1008config PAGE_IDLE_FLAG1009 bool1010 select PAGE_EXTENSION if !64BIT1011 help1012 This adds PG_idle and PG_young flags to 'struct page'. PTE Accessed1013 bit writers can set the state of the bit in the flags so that PTE1014 Accessed bit readers may avoid disturbance.1015 1016config IDLE_PAGE_TRACKING1017 bool "Enable idle page tracking"1018 depends on SYSFS && MMU1019 select PAGE_IDLE_FLAG1020 help1021 This feature allows to estimate the amount of user pages that have1022 not been touched during a given period of time. This information can1023 be useful to tune memory cgroup limits and/or for job placement1024 within a compute cluster.1025 1026 See Documentation/admin-guide/mm/idle_page_tracking.rst for1027 more details.1028 1029# Architectures which implement cpu_dcache_is_aliasing() to query1030# whether the data caches are aliased (VIVT or VIPT with dcache1031# aliasing) need to select this.1032config ARCH_HAS_CPU_CACHE_ALIASING1033 bool1034 1035config ARCH_HAS_CACHE_LINE_SIZE1036 bool1037 1038config ARCH_HAS_CURRENT_STACK_POINTER1039 bool1040 help1041 In support of HARDENED_USERCOPY performing stack variable lifetime1042 checking, an architecture-agnostic way to find the stack pointer1043 is needed. Once an architecture defines an unsigned long global1044 register alias named "current_stack_pointer", this config can be1045 selected.1046 1047config ARCH_HAS_PTE_DEVMAP1048 bool1049 1050config ARCH_HAS_ZONE_DMA_SET1051 bool1052 1053config ZONE_DMA1054 bool "Support DMA zone" if ARCH_HAS_ZONE_DMA_SET1055 default y if ARM64 || X861056 1057config ZONE_DMA321058 bool "Support DMA32 zone" if ARCH_HAS_ZONE_DMA_SET1059 depends on !X86_321060 default y if ARM641061 1062config ZONE_DEVICE1063 bool "Device memory (pmem, HMM, etc...) hotplug support"1064 depends on MEMORY_HOTPLUG1065 depends on MEMORY_HOTREMOVE1066 depends on SPARSEMEM_VMEMMAP1067 depends on ARCH_HAS_PTE_DEVMAP1068 select XARRAY_MULTI1069 1070 help1071 Device memory hotplug support allows for establishing pmem,1072 or other device driver discovered memory regions, in the1073 memmap. This allows pfn_to_page() lookups of otherwise1074 "device-physical" addresses which is needed for using a DAX1075 mapping in an O_DIRECT operation, among other things.1076 1077 If FS_DAX is enabled, then say Y.1078 1079#1080# Helpers to mirror range of the CPU page tables of a process into device page1081# tables.1082#1083config HMM_MIRROR1084 bool1085 depends on MMU1086 1087config GET_FREE_REGION1088 bool1089 1090config DEVICE_PRIVATE1091 bool "Unaddressable device memory (GPU memory, ...)"1092 depends on ZONE_DEVICE1093 select GET_FREE_REGION1094 1095 help1096 Allows creation of struct pages to represent unaddressable device1097 memory; i.e., memory that is only accessible from the device (or1098 group of devices). You likely also want to select HMM_MIRROR.1099 1100config VMAP_PFN1101 bool1102 1103config ARCH_USES_HIGH_VMA_FLAGS1104 bool1105config ARCH_HAS_PKEYS1106 bool1107 1108config ARCH_USES_PG_ARCH_21109 bool1110config ARCH_USES_PG_ARCH_31111 bool1112 1113config VM_EVENT_COUNTERS1114 default y1115 bool "Enable VM event counters for /proc/vmstat" if EXPERT1116 help1117 VM event counters are needed for event counts to be shown.1118 This option allows the disabling of the VM event counters1119 on EXPERT systems. /proc/vmstat will only show page counts1120 if VM event counters are disabled.1121 1122config PERCPU_STATS1123 bool "Collect percpu memory statistics"1124 help1125 This feature collects and exposes statistics via debugfs. The1126 information includes global and per chunk statistics, which can1127 be used to help understand percpu memory usage.1128 1129config GUP_TEST1130 bool "Enable infrastructure for get_user_pages()-related unit tests"1131 depends on DEBUG_FS1132 help1133 Provides /sys/kernel/debug/gup_test, which in turn provides a way1134 to make ioctl calls that can launch kernel-based unit tests for1135 the get_user_pages*() and pin_user_pages*() family of API calls.1136 1137 These tests include benchmark testing of the _fast variants of1138 get_user_pages*() and pin_user_pages*(), as well as smoke tests of1139 the non-_fast variants.1140 1141 There is also a sub-test that allows running dump_page() on any1142 of up to eight pages (selected by command line args) within the1143 range of user-space addresses. These pages are either pinned via1144 pin_user_pages*(), or pinned via get_user_pages*(), as specified1145 by other command line arguments.1146 1147 See tools/testing/selftests/mm/gup_test.c1148 1149comment "GUP_TEST needs to have DEBUG_FS enabled"1150 depends on !GUP_TEST && !DEBUG_FS1151 1152config GUP_GET_PXX_LOW_HIGH1153 bool1154 1155config DMAPOOL_TEST1156 tristate "Enable a module to run time tests on dma_pool"1157 depends on HAS_DMA1158 help1159 Provides a test module that will allocate and free many blocks of1160 various sizes and report how long it takes. This is intended to1161 provide a consistent way to measure how changes to the1162 dma_pool_alloc/free routines affect performance.1163 1164config ARCH_HAS_PTE_SPECIAL1165 bool1166 1167config MAPPING_DIRTY_HELPERS1168 bool1169 1170config KMAP_LOCAL1171 bool1172 1173config KMAP_LOCAL_NON_LINEAR_PTE_ARRAY1174 bool1175 1176# struct io_mapping based helper. Selected by drivers that need them1177config IO_MAPPING1178 bool1179 1180config MEMFD_CREATE1181 bool "Enable memfd_create() system call" if EXPERT1182 1183config SECRETMEM1184 default y1185 bool "Enable memfd_secret() system call" if EXPERT1186 depends on ARCH_HAS_SET_DIRECT_MAP1187 help1188 Enable the memfd_secret() system call with the ability to create1189 memory areas visible only in the context of the owning process and1190 not mapped to other processes and other kernel page tables.1191 1192config ANON_VMA_NAME1193 bool "Anonymous VMA name support"1194 depends on PROC_FS && ADVISE_SYSCALLS && MMU1195 1196 help1197 Allow naming anonymous virtual memory areas.1198 1199 This feature allows assigning names to virtual memory areas. Assigned1200 names can be later retrieved from /proc/pid/maps and /proc/pid/smaps1201 and help identifying individual anonymous memory areas.1202 Assigning a name to anonymous virtual memory area might prevent that1203 area from being merged with adjacent virtual memory areas due to the1204 difference in their name.1205 1206config HAVE_ARCH_USERFAULTFD_WP1207 bool1208 help1209 Arch has userfaultfd write protection support1210 1211config HAVE_ARCH_USERFAULTFD_MINOR1212 bool1213 help1214 Arch has userfaultfd minor fault support1215 1216menuconfig USERFAULTFD1217 bool "Enable userfaultfd() system call"1218 depends on MMU1219 help1220 Enable the userfaultfd() system call that allows to intercept and1221 handle page faults in userland.1222 1223if USERFAULTFD1224config PTE_MARKER_UFFD_WP1225 bool "Userfaultfd write protection support for shmem/hugetlbfs"1226 default y1227 depends on HAVE_ARCH_USERFAULTFD_WP1228 1229 help1230 Allows to create marker PTEs for userfaultfd write protection1231 purposes. It is required to enable userfaultfd write protection on1232 file-backed memory types like shmem and hugetlbfs.1233endif # USERFAULTFD1234 1235# multi-gen LRU {1236config LRU_GEN1237 bool "Multi-Gen LRU"1238 depends on MMU1239 # make sure folio->flags has enough spare bits1240 depends on 64BIT || !SPARSEMEM || SPARSEMEM_VMEMMAP1241 help1242 A high performance LRU implementation to overcommit memory. See1243 Documentation/admin-guide/mm/multigen_lru.rst for details.1244 1245config LRU_GEN_ENABLED1246 bool "Enable by default"1247 depends on LRU_GEN1248 help1249 This option enables the multi-gen LRU by default.1250 1251config LRU_GEN_STATS1252 bool "Full stats for debugging"1253 depends on LRU_GEN1254 help1255 Do not enable this option unless you plan to look at historical stats1256 from evicted generations for debugging purpose.1257 1258 This option has a per-memcg and per-node memory overhead.1259 1260config LRU_GEN_WALKS_MMU1261 def_bool y1262 depends on LRU_GEN && ARCH_HAS_HW_PTE_YOUNG1263# }1264 1265config ARCH_SUPPORTS_PER_VMA_LOCK1266 def_bool n1267 1268config PER_VMA_LOCK1269 def_bool y1270 depends on ARCH_SUPPORTS_PER_VMA_LOCK && MMU && SMP1271 help1272 Allow per-vma locking during page fault handling.1273 1274 This feature allows locking each virtual memory area separately when1275 handling page faults instead of taking mmap_lock.1276 1277config LOCK_MM_AND_FIND_VMA1278 bool1279 depends on !STACK_GROWSUP1280 1281config IOMMU_MM_DATA1282 bool1283 1284config EXECMEM1285 bool1286 1287config NUMA_MEMBLKS1288 bool1289 1290config NUMA_EMU1291 bool "NUMA emulation"1292 depends on NUMA_MEMBLKS1293 help1294 Enable NUMA emulation. A flat machine will be split1295 into virtual nodes when booted with "numa=fake=N", where N is the1296 number of nodes. This is only useful for debugging.1297 1298source "mm/damon/Kconfig"1299 1300endmenu1301