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
135 config HAVE_GENERIC_GUP
138 config ARCH_KEEP_MEMBLOCK
141 config MEMORY_ISOLATION
145 # Only be set on architectures that have completely implemented memory hotplug
146 # feature. If you are not sure, don't touch it.
148 config HAVE_BOOTMEM_INFO_NODE
151 # eventually, we can have this option just 'select SPARSEMEM'
152 config MEMORY_HOTPLUG
153 bool "Allow for memory hot-add"
154 depends on SPARSEMEM || X86_64_ACPI_NUMA
155 depends on ARCH_ENABLE_MEMORY_HOTPLUG
157 config MEMORY_HOTPLUG_SPARSE
159 depends on SPARSEMEM && MEMORY_HOTPLUG
161 config MEMORY_HOTPLUG_DEFAULT_ONLINE
162 bool "Online the newly added memory blocks by default"
163 depends on MEMORY_HOTPLUG
165 This option sets the default policy setting for memory hotplug
166 onlining policy (/sys/devices/system/memory/auto_online_blocks) which
167 determines what happens to newly added memory regions. Policy setting
168 can always be changed at runtime.
169 See Documentation/memory-hotplug.txt for more information.
171 Say Y here if you want all hot-plugged memory blocks to appear in
172 'online' state by default.
173 Say N here if you want the default policy to keep all hot-plugged
174 memory blocks in 'offline' state.
176 config MEMORY_HOTREMOVE
177 bool "Allow for memory hot remove"
178 select MEMORY_ISOLATION
179 select HAVE_BOOTMEM_INFO_NODE if (X86_64 || PPC64)
180 depends on MEMORY_HOTPLUG && ARCH_ENABLE_MEMORY_HOTREMOVE
183 # Heavily threaded applications may benefit from splitting the mm-wide
184 # page_table_lock, so that faults on different parts of the user address
185 # space can be handled with less contention: split it at this NR_CPUS.
186 # Default to 4 for wider testing, though 8 might be more appropriate.
187 # ARM's adjust_pte (unused if VIPT) depends on mm-wide page_table_lock.
188 # PA-RISC 7xxx's spinlock_t would enlarge struct page from 32 to 44 bytes.
189 # DEBUG_SPINLOCK and DEBUG_LOCK_ALLOC spinlock_t also enlarge struct page.
191 config SPLIT_PTLOCK_CPUS
193 default "999999" if !MMU
194 default "999999" if ARM && !CPU_CACHE_VIPT
195 default "999999" if PARISC && !PA20
198 config ARCH_ENABLE_SPLIT_PMD_PTLOCK
202 # support for memory balloon
203 config MEMORY_BALLOON
207 # support for memory balloon compaction
208 config BALLOON_COMPACTION
209 bool "Allow for balloon memory compaction/migration"
211 depends on COMPACTION && MEMORY_BALLOON
213 Memory fragmentation introduced by ballooning might reduce
214 significantly the number of 2MB contiguous memory blocks that can be
215 used within a guest, thus imposing performance penalties associated
216 with the reduced number of transparent huge pages that could be used
217 by the guest workload. Allowing the compaction & migration for memory
218 pages enlisted as being part of memory balloon devices avoids the
219 scenario aforementioned and helps improving memory defragmentation.
222 # support for memory compaction
224 bool "Allow for memory compaction"
229 Compaction is the only memory management component to form
230 high order (larger physically contiguous) memory blocks
231 reliably. The page allocator relies on compaction heavily and
232 the lack of the feature can lead to unexpected OOM killer
233 invocations for high order memory requests. You shouldn't
234 disable this option unless there really is a strong reason for
235 it and then we would be really interested to hear about that at
239 # support for page migration
242 bool "Page migration"
244 depends on (NUMA || ARCH_ENABLE_MEMORY_HOTREMOVE || COMPACTION || CMA) && MMU
246 Allows the migration of the physical location of pages of processes
247 while the virtual addresses are not changed. This is useful in
248 two situations. The first is on NUMA systems to put pages nearer
249 to the processors accessing. The second is when allocating huge
250 pages as migration can relocate pages to satisfy a huge page
251 allocation instead of reclaiming.
253 config ARCH_ENABLE_HUGEPAGE_MIGRATION
256 config ARCH_ENABLE_THP_MIGRATION
260 def_bool (MEMORY_ISOLATION && COMPACTION) || CMA
262 config PHYS_ADDR_T_64BIT
266 bool "Enable bounce buffers"
268 depends on BLOCK && MMU && (ZONE_DMA || HIGHMEM)
270 Enable bounce buffers for devices that cannot access
271 the full range of memory available to the CPU. Enabled
272 by default when ZONE_DMA or HIGHMEM is selected, but you
273 may say n to override this.
283 An architecture should select this if it implements the
284 deprecated interface virt_to_bus(). All new architectures
285 should probably not select this.
293 bool "Enable KSM for page merging"
297 Enable Kernel Samepage Merging: KSM periodically scans those areas
298 of an application's address space that an app has advised may be
299 mergeable. When it finds pages of identical content, it replaces
300 the many instances by a single page with that content, so
301 saving memory until one or another app needs to modify the content.
302 Recommended for use with KVM, or with other duplicative applications.
303 See Documentation/vm/ksm.rst for more information: KSM is inactive
304 until a program has madvised that an area is MADV_MERGEABLE, and
305 root has set /sys/kernel/mm/ksm/run to 1 (if CONFIG_SYSFS is set).
307 config DEFAULT_MMAP_MIN_ADDR
308 int "Low address space to protect from user allocation"
312 This is the portion of low virtual memory which should be protected
313 from userspace allocation. Keeping a user from writing to low pages
314 can help reduce the impact of kernel NULL pointer bugs.
316 For most ia64, ppc64 and x86 users with lots of address space
317 a value of 65536 is reasonable and should cause no problems.
318 On arm and other archs it should not be higher than 32768.
319 Programs which use vm86 functionality or have some need to map
320 this low address space will need CAP_SYS_RAWIO or disable this
321 protection by setting the value to 0.
323 This value can be changed after boot using the
324 /proc/sys/vm/mmap_min_addr tunable.
326 config ARCH_SUPPORTS_MEMORY_FAILURE
329 config MEMORY_FAILURE
331 depends on ARCH_SUPPORTS_MEMORY_FAILURE
332 bool "Enable recovery from hardware memory errors"
333 select MEMORY_ISOLATION
336 Enables code to recover from some memory failures on systems
337 with MCA recovery. This allows a system to continue running
338 even when some of its memory has uncorrected errors. This requires
339 special hardware support and typically ECC memory.
341 config HWPOISON_INJECT
342 tristate "HWPoison pages injector"
343 depends on MEMORY_FAILURE && DEBUG_KERNEL && PROC_FS
344 select PROC_PAGE_MONITOR
346 config NOMMU_INITIAL_TRIM_EXCESS
347 int "Turn on mmap() excess space trimming before booting"
351 The NOMMU mmap() frequently needs to allocate large contiguous chunks
352 of memory on which to store mappings, but it can only ask the system
353 allocator for chunks in 2^N*PAGE_SIZE amounts - which is frequently
354 more than it requires. To deal with this, mmap() is able to trim off
355 the excess and return it to the allocator.
357 If trimming is enabled, the excess is trimmed off and returned to the
358 system allocator, which can cause extra fragmentation, particularly
359 if there are a lot of transient processes.
361 If trimming is disabled, the excess is kept, but not used, which for
362 long-term mappings means that the space is wasted.
364 Trimming can be dynamically controlled through a sysctl option
365 (/proc/sys/vm/nr_trim_pages) which specifies the minimum number of
366 excess pages there must be before trimming should occur, or zero if
367 no trimming is to occur.
369 This option specifies the initial value of this option. The default
370 of 1 says that all excess pages should be trimmed.
372 See Documentation/nommu-mmap.txt for more information.
374 config TRANSPARENT_HUGEPAGE
375 bool "Transparent Hugepage Support"
376 depends on HAVE_ARCH_TRANSPARENT_HUGEPAGE
380 Transparent Hugepages allows the kernel to use huge pages and
381 huge tlb transparently to the applications whenever possible.
382 This feature can improve computing performance to certain
383 applications by speeding up page faults during memory
384 allocation, by reducing the number of tlb misses and by speeding
385 up the pagetable walking.
387 If memory constrained on embedded, you may want to say N.
390 prompt "Transparent Hugepage Support sysfs defaults"
391 depends on TRANSPARENT_HUGEPAGE
392 default TRANSPARENT_HUGEPAGE_ALWAYS
394 Selects the sysfs defaults for Transparent Hugepage Support.
396 config TRANSPARENT_HUGEPAGE_ALWAYS
399 Enabling Transparent Hugepage always, can increase the
400 memory footprint of applications without a guaranteed
401 benefit but it will work automatically for all applications.
403 config TRANSPARENT_HUGEPAGE_MADVISE
406 Enabling Transparent Hugepage madvise, will only provide a
407 performance improvement benefit to the applications using
408 madvise(MADV_HUGEPAGE) but it won't risk to increase the
409 memory footprint of applications without a guaranteed
413 config ARCH_WANTS_THP_SWAP
418 depends on TRANSPARENT_HUGEPAGE && ARCH_WANTS_THP_SWAP && SWAP
420 Swap transparent huge pages in one piece, without splitting.
421 XXX: For now, swap cluster backing transparent huge page
422 will be split after swapout.
424 For selection by architectures with reasonable THP sizes.
426 config TRANSPARENT_HUGE_PAGECACHE
428 depends on TRANSPARENT_HUGEPAGE
431 # UP and nommu archs use km based percpu allocator
433 config NEED_PER_CPU_KM
439 bool "Enable cleancache driver to cache clean pages if tmem is present"
441 Cleancache can be thought of as a page-granularity victim cache
442 for clean pages that the kernel's pageframe replacement algorithm
443 (PFRA) would like to keep around, but can't since there isn't enough
444 memory. So when the PFRA "evicts" a page, it first attempts to use
445 cleancache code to put the data contained in that page into
446 "transcendent memory", memory that is not directly accessible or
447 addressable by the kernel and is of unknown and possibly
448 time-varying size. And when a cleancache-enabled
449 filesystem wishes to access a page in a file on disk, it first
450 checks cleancache to see if it already contains it; if it does,
451 the page is copied into the kernel and a disk access is avoided.
452 When a transcendent memory driver is available (such as zcache or
453 Xen transcendent memory), a significant I/O reduction
454 may be achieved. When none is available, all cleancache calls
455 are reduced to a single pointer-compare-against-NULL resulting
456 in a negligible performance hit.
458 If unsure, say Y to enable cleancache
461 bool "Enable frontswap to cache swap pages if tmem is present"
464 Frontswap is so named because it can be thought of as the opposite
465 of a "backing" store for a swap device. The data is stored into
466 "transcendent memory", memory that is not directly accessible or
467 addressable by the kernel and is of unknown and possibly
468 time-varying size. When space in transcendent memory is available,
469 a significant swap I/O reduction may be achieved. When none is
470 available, all frontswap calls are reduced to a single pointer-
471 compare-against-NULL resulting in a negligible performance hit
472 and swap data is stored as normal on the matching swap device.
474 If unsure, say Y to enable frontswap.
477 bool "Contiguous Memory Allocator"
480 select MEMORY_ISOLATION
482 This enables the Contiguous Memory Allocator which allows other
483 subsystems to allocate big physically-contiguous blocks of memory.
484 CMA reserves a region of memory and allows only movable pages to
485 be allocated from it. This way, the kernel can use the memory for
486 pagecache and when a subsystem requests for contiguous area, the
487 allocated pages are migrated away to serve the contiguous request.
492 bool "CMA debug messages (DEVELOPMENT)"
493 depends on DEBUG_KERNEL && CMA
495 Turns on debug messages in CMA. This produces KERN_DEBUG
496 messages for every CMA call as well as various messages while
497 processing calls such as dma_alloc_from_contiguous().
498 This option does not affect warning and error messages.
501 bool "CMA debugfs interface"
502 depends on CMA && DEBUG_FS
504 Turns on the DebugFS interface for CMA.
507 int "Maximum count of the CMA areas"
511 CMA allows to create CMA areas for particular purpose, mainly,
512 used as device private area. This parameter sets the maximum
513 number of CMA area in the system.
515 If unsure, leave the default value "7".
517 config MEM_SOFT_DIRTY
518 bool "Track memory changes"
519 depends on CHECKPOINT_RESTORE && HAVE_ARCH_SOFT_DIRTY && PROC_FS
520 select PROC_PAGE_MONITOR
522 This option enables memory changes tracking by introducing a
523 soft-dirty bit on pte-s. This bit it set when someone writes
524 into a page just as regular dirty bit, but unlike the latter
525 it can be cleared by hands.
527 See Documentation/admin-guide/mm/soft-dirty.rst for more details.
530 bool "Compressed cache for swap pages (EXPERIMENTAL)"
531 depends on FRONTSWAP && CRYPTO=y
535 A lightweight compressed cache for swap pages. It takes
536 pages that are in the process of being swapped out and attempts to
537 compress them into a dynamically allocated RAM-based memory pool.
538 This can result in a significant I/O reduction on swap device and,
539 in the case where decompressing from RAM is faster that swap device
540 reads, can also improve workload performance.
542 This is marked experimental because it is a new feature (as of
543 v3.11) that interacts heavily with memory reclaim. While these
544 interactions don't cause any known issues on simple memory setups,
545 they have not be fully explored on the large set of potential
546 configurations and workloads that exist.
549 tristate "Common API for compressed memory storage"
551 Compressed memory storage API. This allows using either zbud or
555 tristate "Low (Up to 2x) density storage for compressed pages"
557 A special purpose allocator for storing compressed pages.
558 It is designed to store up to two compressed pages per physical
559 page. While this design limits storage density, it has simple and
560 deterministic reclaim properties that make it preferable to a higher
561 density approach when reclaim will be used.
564 tristate "Up to 3x density storage for compressed pages"
567 A special purpose allocator for storing compressed pages.
568 It is designed to store up to three compressed pages per physical
569 page. It is a ZBUD derivative so the simplicity and determinism are
573 tristate "Memory allocator for compressed pages"
576 zsmalloc is a slab-based memory allocator designed to store
577 compressed RAM pages. zsmalloc uses virtual memory mapping
578 in order to reduce fragmentation. However, this results in a
579 non-standard allocator interface where a handle, not a pointer, is
580 returned by an alloc(). This handle must be mapped in order to
581 access the allocated space.
583 config PGTABLE_MAPPING
584 bool "Use page table mapping to access object in zsmalloc"
587 By default, zsmalloc uses a copy-based object mapping method to
588 access allocations that span two pages. However, if a particular
589 architecture (ex, ARM) performs VM mapping faster than copying,
590 then you should select this. This causes zsmalloc to use page table
591 mapping rather than copying for object mapping.
593 You can check speed with zsmalloc benchmark:
594 https://github.com/spartacus06/zsmapbench
597 bool "Export zsmalloc statistics"
601 This option enables code in the zsmalloc to collect various
602 statistics about whats happening in zsmalloc and exports that
603 information to userspace via debugfs.
606 config GENERIC_EARLY_IOREMAP
609 config MAX_STACK_SIZE_MB
610 int "Maximum user stack size for 32-bit processes (MB)"
613 depends on STACK_GROWSUP && (!64BIT || COMPAT)
615 This is the maximum stack size in Megabytes in the VM layout of 32-bit
616 user processes when the stack grows upwards (currently only on parisc
617 arch). The stack will be located at the highest memory address minus
618 the given value, unless the RLIMIT_STACK hard limit is changed to a
619 smaller value in which case that is used.
621 A sane initial value is 80 MB.
623 config DEFERRED_STRUCT_PAGE_INIT
624 bool "Defer initialisation of struct pages to kthreads"
626 depends on !NEED_PER_CPU_KM
629 Ordinarily all struct pages are initialised during early boot in a
630 single thread. On very large machines this can take a considerable
631 amount of time. If this option is set, large machines will bring up
632 a subset of memmap at boot and then initialise the rest in parallel
633 by starting one-off "pgdatinitX" kernel thread for each node X. This
634 has a potential performance impact on processes running early in the
635 lifetime of the system until these kthreads finish the
638 config IDLE_PAGE_TRACKING
639 bool "Enable idle page tracking"
640 depends on SYSFS && MMU
641 select PAGE_EXTENSION if !64BIT
643 This feature allows to estimate the amount of user pages that have
644 not been touched during a given period of time. This information can
645 be useful to tune memory cgroup limits and/or for job placement
646 within a compute cluster.
648 See Documentation/admin-guide/mm/idle_page_tracking.rst for
651 # arch_add_memory() comprehends device memory
652 config ARCH_HAS_ZONE_DEVICE
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_ZONE_DEVICE
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 ARCH_HAS_HMM_MIRROR
675 depends on (X86_64 || PPC64)
676 depends on MMU && 64BIT
678 config ARCH_HAS_HMM_DEVICE
681 depends on (X86_64 || PPC64)
682 depends on MEMORY_HOTPLUG
683 depends on MEMORY_HOTREMOVE
684 depends on SPARSEMEM_VMEMMAP
685 depends on ARCH_HAS_ZONE_DEVICE
691 depends on (X86_64 || PPC64)
692 depends on ZONE_DEVICE
693 depends on MMU && 64BIT
694 depends on MEMORY_HOTPLUG
695 depends on MEMORY_HOTREMOVE
696 depends on SPARSEMEM_VMEMMAP
698 config MIGRATE_VMA_HELPER
701 config DEV_PAGEMAP_OPS
707 select MIGRATE_VMA_HELPER
710 bool "HMM mirror CPU page table into a device page table"
711 depends on ARCH_HAS_HMM
714 Select HMM_MIRROR if you want to mirror range of the CPU page table of a
715 process into a device page table. Here, mirror means "keep synchronized".
716 Prerequisites: the device must provide the ability to write-protect its
717 page tables (at PAGE_SIZE granularity), and must be able to recover from
718 the resulting potential page faults.
720 config DEVICE_PRIVATE
721 bool "Unaddressable device memory (GPU memory, ...)"
722 depends on ARCH_HAS_HMM
724 select DEV_PAGEMAP_OPS
727 Allows creation of struct pages to represent unaddressable device
728 memory; i.e., memory that is only accessible from the device (or
729 group of devices). You likely also want to select HMM_MIRROR.
732 bool "Addressable device memory (like GPU memory)"
733 depends on ARCH_HAS_HMM
735 select DEV_PAGEMAP_OPS
738 Allows creation of struct pages to represent addressable device
739 memory; i.e., memory that is accessible from both the device and
745 config ARCH_USES_HIGH_VMA_FLAGS
747 config ARCH_HAS_PKEYS
751 bool "Collect percpu memory statistics"
753 This feature collects and exposes statistics via debugfs. The
754 information includes global and per chunk statistics, which can
755 be used to help understand percpu memory usage.
758 bool "Enable infrastructure for get_user_pages_fast() benchmarking"
760 Provides /sys/kernel/debug/gup_benchmark that helps with testing
761 performance of get_user_pages_fast().
763 See tools/testing/selftests/vm/gup_benchmark.c
765 config ARCH_HAS_PTE_SPECIAL