1 # SPDX-License-Identifier: GPL-2.0-only
3 menu "Memory Management options"
5 config SELECT_MEMORY_MODEL
7 depends on ARCH_SELECT_MEMORY_MODEL
11 depends on SELECT_MEMORY_MODEL
12 default DISCONTIGMEM_MANUAL if ARCH_DISCONTIGMEM_DEFAULT
13 default SPARSEMEM_MANUAL if ARCH_SPARSEMEM_DEFAULT
14 default FLATMEM_MANUAL
16 This option allows you to change some of the ways that
17 Linux manages its memory internally. Most users will
18 only have one option here selected by the architecture
19 configuration. This is normal.
23 depends on !(ARCH_DISCONTIGMEM_ENABLE || ARCH_SPARSEMEM_ENABLE) || ARCH_FLATMEM_ENABLE
25 This option is best suited for non-NUMA systems with
26 flat address space. The FLATMEM is the most efficient
27 system in terms of performance and resource consumption
28 and it is the best option for smaller systems.
30 For systems that have holes in their physical address
31 spaces and for features like NUMA and memory hotplug,
32 choose "Sparse Memory"
34 If unsure, choose this option (Flat Memory) over any other.
36 config DISCONTIGMEM_MANUAL
37 bool "Discontiguous Memory"
38 depends on ARCH_DISCONTIGMEM_ENABLE
40 This option provides enhanced support for discontiguous
41 memory systems, over FLATMEM. These systems have holes
42 in their physical address spaces, and this option provides
43 more efficient handling of these holes.
45 Although "Discontiguous Memory" is still used by several
46 architectures, it is considered deprecated in favor of
49 If unsure, choose "Sparse Memory" over this option.
51 config SPARSEMEM_MANUAL
53 depends on ARCH_SPARSEMEM_ENABLE
55 This will be the only option for some systems, including
56 memory hot-plug systems. This is normal.
58 This option provides efficient support for systems with
59 holes is their physical address space and allows memory
60 hot-plug and hot-remove.
62 If unsure, choose "Flat Memory" over this option.
68 depends on (!SELECT_MEMORY_MODEL && ARCH_DISCONTIGMEM_ENABLE) || DISCONTIGMEM_MANUAL
72 depends on (!SELECT_MEMORY_MODEL && ARCH_SPARSEMEM_ENABLE) || SPARSEMEM_MANUAL
76 depends on (!DISCONTIGMEM && !SPARSEMEM) || FLATMEM_MANUAL
78 config FLAT_NODE_MEM_MAP
83 # Both the NUMA code and DISCONTIGMEM use arrays of pg_data_t's
84 # to represent different areas of memory. This variable allows
85 # those dependencies to exist individually.
87 config NEED_MULTIPLE_NODES
89 depends on DISCONTIGMEM || NUMA
91 config HAVE_MEMORY_PRESENT
93 depends on ARCH_HAVE_MEMORY_PRESENT || SPARSEMEM
96 # SPARSEMEM_EXTREME (which is the default) does some bootmem
97 # allocations when memory_present() is called. If this cannot
98 # be done on your architecture, select this option. However,
99 # statically allocating the mem_section[] array can potentially
100 # consume vast quantities of .bss, so be careful.
102 # This option will also potentially produce smaller runtime code
103 # with gcc 3.4 and later.
105 config SPARSEMEM_STATIC
109 # Architecture platforms which require a two level mem_section in SPARSEMEM
110 # must select this option. This is usually for architecture platforms with
111 # an extremely sparse physical address space.
113 config SPARSEMEM_EXTREME
115 depends on SPARSEMEM && !SPARSEMEM_STATIC
117 config SPARSEMEM_VMEMMAP_ENABLE
120 config SPARSEMEM_VMEMMAP
121 bool "Sparse Memory virtual memmap"
122 depends on SPARSEMEM && SPARSEMEM_VMEMMAP_ENABLE
125 SPARSEMEM_VMEMMAP uses a virtually mapped memmap to optimise
126 pfn_to_page and page_to_pfn operations. This is the most
127 efficient option when sufficient kernel resources are available.
129 config HAVE_MEMBLOCK_NODE_MAP
132 config HAVE_MEMBLOCK_PHYS_MAP
139 config ARCH_KEEP_MEMBLOCK
142 config MEMORY_ISOLATION
146 # Only be set on architectures that have completely implemented memory hotplug
147 # feature. If you are not sure, don't touch it.
149 config HAVE_BOOTMEM_INFO_NODE
152 # eventually, we can have this option just 'select SPARSEMEM'
153 config MEMORY_HOTPLUG
154 bool "Allow for memory hot-add"
155 depends on SPARSEMEM || X86_64_ACPI_NUMA
156 depends on ARCH_ENABLE_MEMORY_HOTPLUG
158 config MEMORY_HOTPLUG_SPARSE
160 depends on SPARSEMEM && MEMORY_HOTPLUG
162 config MEMORY_HOTPLUG_DEFAULT_ONLINE
163 bool "Online the newly added memory blocks by default"
164 depends on MEMORY_HOTPLUG
166 This option sets the default policy setting for memory hotplug
167 onlining policy (/sys/devices/system/memory/auto_online_blocks) which
168 determines what happens to newly added memory regions. Policy setting
169 can always be changed at runtime.
170 See Documentation/admin-guide/mm/memory-hotplug.rst for more information.
172 Say Y here if you want all hot-plugged memory blocks to appear in
173 'online' state by default.
174 Say N here if you want the default policy to keep all hot-plugged
175 memory blocks in 'offline' state.
177 config MEMORY_HOTREMOVE
178 bool "Allow for memory hot remove"
179 select MEMORY_ISOLATION
180 select HAVE_BOOTMEM_INFO_NODE if (X86_64 || PPC64)
181 depends on MEMORY_HOTPLUG && ARCH_ENABLE_MEMORY_HOTREMOVE
184 # Heavily threaded applications may benefit from splitting the mm-wide
185 # page_table_lock, so that faults on different parts of the user address
186 # space can be handled with less contention: split it at this NR_CPUS.
187 # Default to 4 for wider testing, though 8 might be more appropriate.
188 # ARM's adjust_pte (unused if VIPT) depends on mm-wide page_table_lock.
189 # PA-RISC 7xxx's spinlock_t would enlarge struct page from 32 to 44 bytes.
190 # DEBUG_SPINLOCK and DEBUG_LOCK_ALLOC spinlock_t also enlarge struct page.
192 config SPLIT_PTLOCK_CPUS
194 default "999999" if !MMU
195 default "999999" if ARM && !CPU_CACHE_VIPT
196 default "999999" if PARISC && !PA20
199 config ARCH_ENABLE_SPLIT_PMD_PTLOCK
203 # support for memory balloon
204 config MEMORY_BALLOON
208 # support for memory balloon compaction
209 config BALLOON_COMPACTION
210 bool "Allow for balloon memory compaction/migration"
212 depends on COMPACTION && MEMORY_BALLOON
214 Memory fragmentation introduced by ballooning might reduce
215 significantly the number of 2MB contiguous memory blocks that can be
216 used within a guest, thus imposing performance penalties associated
217 with the reduced number of transparent huge pages that could be used
218 by the guest workload. Allowing the compaction & migration for memory
219 pages enlisted as being part of memory balloon devices avoids the
220 scenario aforementioned and helps improving memory defragmentation.
223 # support for memory compaction
225 bool "Allow for memory compaction"
230 Compaction is the only memory management component to form
231 high order (larger physically contiguous) memory blocks
232 reliably. The page allocator relies on compaction heavily and
233 the lack of the feature can lead to unexpected OOM killer
234 invocations for high order memory requests. You shouldn't
235 disable this option unless there really is a strong reason for
236 it and then we would be really interested to hear about that at
240 # support for page migration
243 bool "Page migration"
245 depends on (NUMA || ARCH_ENABLE_MEMORY_HOTREMOVE || COMPACTION || CMA) && MMU
247 Allows the migration of the physical location of pages of processes
248 while the virtual addresses are not changed. This is useful in
249 two situations. The first is on NUMA systems to put pages nearer
250 to the processors accessing. The second is when allocating huge
251 pages as migration can relocate pages to satisfy a huge page
252 allocation instead of reclaiming.
254 config ARCH_ENABLE_HUGEPAGE_MIGRATION
257 config ARCH_ENABLE_THP_MIGRATION
261 def_bool (MEMORY_ISOLATION && COMPACTION) || CMA
263 config PHYS_ADDR_T_64BIT
267 bool "Enable bounce buffers"
269 depends on BLOCK && MMU && (ZONE_DMA || HIGHMEM)
271 Enable bounce buffers for devices that cannot access
272 the full range of memory available to the CPU. Enabled
273 by default when ZONE_DMA or HIGHMEM is selected, but you
274 may say n to override this.
284 An architecture should select this if it implements the
285 deprecated interface virt_to_bus(). All new architectures
286 should probably not select this.
294 bool "Enable KSM for page merging"
298 Enable Kernel Samepage Merging: KSM periodically scans those areas
299 of an application's address space that an app has advised may be
300 mergeable. When it finds pages of identical content, it replaces
301 the many instances by a single page with that content, so
302 saving memory until one or another app needs to modify the content.
303 Recommended for use with KVM, or with other duplicative applications.
304 See Documentation/vm/ksm.rst for more information: KSM is inactive
305 until a program has madvised that an area is MADV_MERGEABLE, and
306 root has set /sys/kernel/mm/ksm/run to 1 (if CONFIG_SYSFS is set).
308 config DEFAULT_MMAP_MIN_ADDR
309 int "Low address space to protect from user allocation"
313 This is the portion of low virtual memory which should be protected
314 from userspace allocation. Keeping a user from writing to low pages
315 can help reduce the impact of kernel NULL pointer bugs.
317 For most ia64, ppc64 and x86 users with lots of address space
318 a value of 65536 is reasonable and should cause no problems.
319 On arm and other archs it should not be higher than 32768.
320 Programs which use vm86 functionality or have some need to map
321 this low address space will need CAP_SYS_RAWIO or disable this
322 protection by setting the value to 0.
324 This value can be changed after boot using the
325 /proc/sys/vm/mmap_min_addr tunable.
327 config ARCH_SUPPORTS_MEMORY_FAILURE
330 config MEMORY_FAILURE
332 depends on ARCH_SUPPORTS_MEMORY_FAILURE
333 bool "Enable recovery from hardware memory errors"
334 select MEMORY_ISOLATION
337 Enables code to recover from some memory failures on systems
338 with MCA recovery. This allows a system to continue running
339 even when some of its memory has uncorrected errors. This requires
340 special hardware support and typically ECC memory.
342 config HWPOISON_INJECT
343 tristate "HWPoison pages injector"
344 depends on MEMORY_FAILURE && DEBUG_KERNEL && PROC_FS
345 select PROC_PAGE_MONITOR
347 config NOMMU_INITIAL_TRIM_EXCESS
348 int "Turn on mmap() excess space trimming before booting"
352 The NOMMU mmap() frequently needs to allocate large contiguous chunks
353 of memory on which to store mappings, but it can only ask the system
354 allocator for chunks in 2^N*PAGE_SIZE amounts - which is frequently
355 more than it requires. To deal with this, mmap() is able to trim off
356 the excess and return it to the allocator.
358 If trimming is enabled, the excess is trimmed off and returned to the
359 system allocator, which can cause extra fragmentation, particularly
360 if there are a lot of transient processes.
362 If trimming is disabled, the excess is kept, but not used, which for
363 long-term mappings means that the space is wasted.
365 Trimming can be dynamically controlled through a sysctl option
366 (/proc/sys/vm/nr_trim_pages) which specifies the minimum number of
367 excess pages there must be before trimming should occur, or zero if
368 no trimming is to occur.
370 This option specifies the initial value of this option. The default
371 of 1 says that all excess pages should be trimmed.
373 See Documentation/nommu-mmap.txt for more information.
375 config TRANSPARENT_HUGEPAGE
376 bool "Transparent Hugepage Support"
377 depends on HAVE_ARCH_TRANSPARENT_HUGEPAGE
381 Transparent Hugepages allows the kernel to use huge pages and
382 huge tlb transparently to the applications whenever possible.
383 This feature can improve computing performance to certain
384 applications by speeding up page faults during memory
385 allocation, by reducing the number of tlb misses and by speeding
386 up the pagetable walking.
388 If memory constrained on embedded, you may want to say N.
391 prompt "Transparent Hugepage Support sysfs defaults"
392 depends on TRANSPARENT_HUGEPAGE
393 default TRANSPARENT_HUGEPAGE_ALWAYS
395 Selects the sysfs defaults for Transparent Hugepage Support.
397 config TRANSPARENT_HUGEPAGE_ALWAYS
400 Enabling Transparent Hugepage always, can increase the
401 memory footprint of applications without a guaranteed
402 benefit but it will work automatically for all applications.
404 config TRANSPARENT_HUGEPAGE_MADVISE
407 Enabling Transparent Hugepage madvise, will only provide a
408 performance improvement benefit to the applications using
409 madvise(MADV_HUGEPAGE) but it won't risk to increase the
410 memory footprint of applications without a guaranteed
414 config ARCH_WANTS_THP_SWAP
419 depends on TRANSPARENT_HUGEPAGE && ARCH_WANTS_THP_SWAP && SWAP
421 Swap transparent huge pages in one piece, without splitting.
422 XXX: For now, swap cluster backing transparent huge page
423 will be split after swapout.
425 For selection by architectures with reasonable THP sizes.
427 config TRANSPARENT_HUGE_PAGECACHE
429 depends on TRANSPARENT_HUGEPAGE
432 # UP and nommu archs use km based percpu allocator
434 config NEED_PER_CPU_KM
440 bool "Enable cleancache driver to cache clean pages if tmem is present"
442 Cleancache can be thought of as a page-granularity victim cache
443 for clean pages that the kernel's pageframe replacement algorithm
444 (PFRA) would like to keep around, but can't since there isn't enough
445 memory. So when the PFRA "evicts" a page, it first attempts to use
446 cleancache code to put the data contained in that page into
447 "transcendent memory", memory that is not directly accessible or
448 addressable by the kernel and is of unknown and possibly
449 time-varying size. And when a cleancache-enabled
450 filesystem wishes to access a page in a file on disk, it first
451 checks cleancache to see if it already contains it; if it does,
452 the page is copied into the kernel and a disk access is avoided.
453 When a transcendent memory driver is available (such as zcache or
454 Xen transcendent memory), a significant I/O reduction
455 may be achieved. When none is available, all cleancache calls
456 are reduced to a single pointer-compare-against-NULL resulting
457 in a negligible performance hit.
459 If unsure, say Y to enable cleancache
462 bool "Enable frontswap to cache swap pages if tmem is present"
465 Frontswap is so named because it can be thought of as the opposite
466 of a "backing" store for a swap device. The data is stored into
467 "transcendent memory", memory that is not directly accessible or
468 addressable by the kernel and is of unknown and possibly
469 time-varying size. When space in transcendent memory is available,
470 a significant swap I/O reduction may be achieved. When none is
471 available, all frontswap calls are reduced to a single pointer-
472 compare-against-NULL resulting in a negligible performance hit
473 and swap data is stored as normal on the matching swap device.
475 If unsure, say Y to enable frontswap.
478 bool "Contiguous Memory Allocator"
481 select MEMORY_ISOLATION
483 This enables the Contiguous Memory Allocator which allows other
484 subsystems to allocate big physically-contiguous blocks of memory.
485 CMA reserves a region of memory and allows only movable pages to
486 be allocated from it. This way, the kernel can use the memory for
487 pagecache and when a subsystem requests for contiguous area, the
488 allocated pages are migrated away to serve the contiguous request.
493 bool "CMA debug messages (DEVELOPMENT)"
494 depends on DEBUG_KERNEL && CMA
496 Turns on debug messages in CMA. This produces KERN_DEBUG
497 messages for every CMA call as well as various messages while
498 processing calls such as dma_alloc_from_contiguous().
499 This option does not affect warning and error messages.
502 bool "CMA debugfs interface"
503 depends on CMA && DEBUG_FS
505 Turns on the DebugFS interface for CMA.
508 int "Maximum count of the CMA areas"
512 CMA allows to create CMA areas for particular purpose, mainly,
513 used as device private area. This parameter sets the maximum
514 number of CMA area in the system.
516 If unsure, leave the default value "7".
518 config MEM_SOFT_DIRTY
519 bool "Track memory changes"
520 depends on CHECKPOINT_RESTORE && HAVE_ARCH_SOFT_DIRTY && PROC_FS
521 select PROC_PAGE_MONITOR
523 This option enables memory changes tracking by introducing a
524 soft-dirty bit on pte-s. This bit it set when someone writes
525 into a page just as regular dirty bit, but unlike the latter
526 it can be cleared by hands.
528 See Documentation/admin-guide/mm/soft-dirty.rst for more details.
531 bool "Compressed cache for swap pages (EXPERIMENTAL)"
532 depends on FRONTSWAP && CRYPTO=y
536 A lightweight compressed cache for swap pages. It takes
537 pages that are in the process of being swapped out and attempts to
538 compress them into a dynamically allocated RAM-based memory pool.
539 This can result in a significant I/O reduction on swap device and,
540 in the case where decompressing from RAM is faster that swap device
541 reads, can also improve workload performance.
543 This is marked experimental because it is a new feature (as of
544 v3.11) that interacts heavily with memory reclaim. While these
545 interactions don't cause any known issues on simple memory setups,
546 they have not be fully explored on the large set of potential
547 configurations and workloads that exist.
550 tristate "Common API for compressed memory storage"
552 Compressed memory storage API. This allows using either zbud or
556 tristate "Low (Up to 2x) density storage for compressed pages"
558 A special purpose allocator for storing compressed pages.
559 It is designed to store up to two compressed pages per physical
560 page. While this design limits storage density, it has simple and
561 deterministic reclaim properties that make it preferable to a higher
562 density approach when reclaim will be used.
565 tristate "Up to 3x density storage for compressed pages"
568 A special purpose allocator for storing compressed pages.
569 It is designed to store up to three compressed pages per physical
570 page. It is a ZBUD derivative so the simplicity and determinism are
574 tristate "Memory allocator for compressed pages"
577 zsmalloc is a slab-based memory allocator designed to store
578 compressed RAM pages. zsmalloc uses virtual memory mapping
579 in order to reduce fragmentation. However, this results in a
580 non-standard allocator interface where a handle, not a pointer, is
581 returned by an alloc(). This handle must be mapped in order to
582 access the allocated space.
584 config PGTABLE_MAPPING
585 bool "Use page table mapping to access object in zsmalloc"
588 By default, zsmalloc uses a copy-based object mapping method to
589 access allocations that span two pages. However, if a particular
590 architecture (ex, ARM) performs VM mapping faster than copying,
591 then you should select this. This causes zsmalloc to use page table
592 mapping rather than copying for object mapping.
594 You can check speed with zsmalloc benchmark:
595 https://github.com/spartacus06/zsmapbench
598 bool "Export zsmalloc statistics"
602 This option enables code in the zsmalloc to collect various
603 statistics about whats happening in zsmalloc and exports that
604 information to userspace via debugfs.
607 config GENERIC_EARLY_IOREMAP
610 config MAX_STACK_SIZE_MB
611 int "Maximum user stack size for 32-bit processes (MB)"
614 depends on STACK_GROWSUP && (!64BIT || COMPAT)
616 This is the maximum stack size in Megabytes in the VM layout of 32-bit
617 user processes when the stack grows upwards (currently only on parisc
618 arch). The stack will be located at the highest memory address minus
619 the given value, unless the RLIMIT_STACK hard limit is changed to a
620 smaller value in which case that is used.
622 A sane initial value is 80 MB.
624 config DEFERRED_STRUCT_PAGE_INIT
625 bool "Defer initialisation of struct pages to kthreads"
627 depends on !NEED_PER_CPU_KM
630 Ordinarily all struct pages are initialised during early boot in a
631 single thread. On very large machines this can take a considerable
632 amount of time. If this option is set, large machines will bring up
633 a subset of memmap at boot and then initialise the rest in parallel
634 by starting one-off "pgdatinitX" kernel thread for each node X. This
635 has a potential performance impact on processes running early in the
636 lifetime of the system until these kthreads finish the
639 config IDLE_PAGE_TRACKING
640 bool "Enable idle page tracking"
641 depends on SYSFS && MMU
642 select PAGE_EXTENSION if !64BIT
644 This feature allows to estimate the amount of user pages that have
645 not been touched during a given period of time. This information can
646 be useful to tune memory cgroup limits and/or for job placement
647 within a compute cluster.
649 See Documentation/admin-guide/mm/idle_page_tracking.rst for
652 config ARCH_HAS_PTE_DEVMAP
656 bool "Device memory (pmem, HMM, etc...) hotplug support"
657 depends on MEMORY_HOTPLUG
658 depends on MEMORY_HOTREMOVE
659 depends on SPARSEMEM_VMEMMAP
660 depends on ARCH_HAS_PTE_DEVMAP
664 Device memory hotplug support allows for establishing pmem,
665 or other device driver discovered memory regions, in the
666 memmap. This allows pfn_to_page() lookups of otherwise
667 "device-physical" addresses which is needed for using a DAX
668 mapping in an O_DIRECT operation, among other things.
670 If FS_DAX is enabled, then say Y.
672 config MIGRATE_VMA_HELPER
675 config DEV_PAGEMAP_OPS
679 bool "HMM mirror CPU page table into a device page table"
680 depends on (X86_64 || PPC64)
681 depends on MMU && 64BIT
684 Select HMM_MIRROR if you want to mirror range of the CPU page table of a
685 process into a device page table. Here, mirror means "keep synchronized".
686 Prerequisites: the device must provide the ability to write-protect its
687 page tables (at PAGE_SIZE granularity), and must be able to recover from
688 the resulting potential page faults.
690 config DEVICE_PRIVATE
691 bool "Unaddressable device memory (GPU memory, ...)"
692 depends on ZONE_DEVICE
693 select DEV_PAGEMAP_OPS
696 Allows creation of struct pages to represent unaddressable device
697 memory; i.e., memory that is only accessible from the device (or
698 group of devices). You likely also want to select HMM_MIRROR.
703 config ARCH_USES_HIGH_VMA_FLAGS
705 config ARCH_HAS_PKEYS
709 bool "Collect percpu memory statistics"
711 This feature collects and exposes statistics via debugfs. The
712 information includes global and per chunk statistics, which can
713 be used to help understand percpu memory usage.
716 bool "Enable infrastructure for get_user_pages_fast() benchmarking"
718 Provides /sys/kernel/debug/gup_benchmark that helps with testing
719 performance of get_user_pages_fast().
721 See tools/testing/selftests/vm/gup_benchmark.c
723 config GUP_GET_PTE_LOW_HIGH
726 config ARCH_HAS_PTE_SPECIAL
730 # Some architectures require a special hugepage directory format that is
731 # required to support multiple hugepage sizes. For example a4fe3ce76
732 # "powerpc/mm: Allow more flexible layouts for hugepage pagetables"
733 # introduced it on powerpc. This allows for a more flexible hugepage
736 config ARCH_HAS_HUGEPD