2 menu "Memory Management options"
4 config SELECT_MEMORY_MODEL
6 depends on ARCH_SELECT_MEMORY_MODEL
10 depends on SELECT_MEMORY_MODEL
11 default DISCONTIGMEM_MANUAL if ARCH_DISCONTIGMEM_DEFAULT
12 default SPARSEMEM_MANUAL if ARCH_SPARSEMEM_DEFAULT
13 default FLATMEM_MANUAL
15 This option allows you to change some of the ways that
16 Linux manages its memory internally. Most users will
17 only have one option here selected by the architecture
18 configuration. This is normal.
22 depends on !(ARCH_DISCONTIGMEM_ENABLE || ARCH_SPARSEMEM_ENABLE) || ARCH_FLATMEM_ENABLE
24 This option is best suited for non-NUMA systems with
25 flat address space. The FLATMEM is the most efficient
26 system in terms of performance and resource consumption
27 and it is the best option for smaller systems.
29 For systems that have holes in their physical address
30 spaces and for features like NUMA and memory hotplug,
31 choose "Sparse Memory"
33 If unsure, choose this option (Flat Memory) over any other.
35 config DISCONTIGMEM_MANUAL
36 bool "Discontiguous Memory"
37 depends on ARCH_DISCONTIGMEM_ENABLE
39 This option provides enhanced support for discontiguous
40 memory systems, over FLATMEM. These systems have holes
41 in their physical address spaces, and this option provides
42 more efficient handling of these holes.
44 Although "Discontiguous Memory" is still used by several
45 architectures, it is considered deprecated in favor of
48 If unsure, choose "Sparse Memory" over this option.
50 config SPARSEMEM_MANUAL
52 depends on ARCH_SPARSEMEM_ENABLE
54 This will be the only option for some systems, including
55 memory hot-plug systems. This is normal.
57 This option provides efficient support for systems with
58 holes is their physical address space and allows memory
59 hot-plug and hot-remove.
61 If unsure, choose "Flat Memory" over this option.
67 depends on (!SELECT_MEMORY_MODEL && ARCH_DISCONTIGMEM_ENABLE) || DISCONTIGMEM_MANUAL
71 depends on (!SELECT_MEMORY_MODEL && ARCH_SPARSEMEM_ENABLE) || SPARSEMEM_MANUAL
75 depends on (!DISCONTIGMEM && !SPARSEMEM) || FLATMEM_MANUAL
77 config FLAT_NODE_MEM_MAP
82 # Both the NUMA code and DISCONTIGMEM use arrays of pg_data_t's
83 # to represent different areas of memory. This variable allows
84 # those dependencies to exist individually.
86 config NEED_MULTIPLE_NODES
88 depends on DISCONTIGMEM || NUMA
90 config HAVE_MEMORY_PRESENT
92 depends on ARCH_HAVE_MEMORY_PRESENT || SPARSEMEM
95 # SPARSEMEM_EXTREME (which is the default) does some bootmem
96 # allocations when memory_present() is called. If this cannot
97 # be done on your architecture, select this option. However,
98 # statically allocating the mem_section[] array can potentially
99 # consume vast quantities of .bss, so be careful.
101 # This option will also potentially produce smaller runtime code
102 # with gcc 3.4 and later.
104 config SPARSEMEM_STATIC
108 # Architecture platforms which require a two level mem_section in SPARSEMEM
109 # must select this option. This is usually for architecture platforms with
110 # an extremely sparse physical address space.
112 config SPARSEMEM_EXTREME
114 depends on SPARSEMEM && !SPARSEMEM_STATIC
116 config SPARSEMEM_VMEMMAP_ENABLE
119 config SPARSEMEM_VMEMMAP
120 bool "Sparse Memory virtual memmap"
121 depends on SPARSEMEM && SPARSEMEM_VMEMMAP_ENABLE
124 SPARSEMEM_VMEMMAP uses a virtually mapped memmap to optimise
125 pfn_to_page and page_to_pfn operations. This is the most
126 efficient option when sufficient kernel resources are available.
128 config HAVE_MEMBLOCK_NODE_MAP
131 config HAVE_MEMBLOCK_PHYS_MAP
134 config HAVE_GENERIC_GUP
137 config ARCH_KEEP_MEMBLOCK
140 config MEMORY_ISOLATION
144 # Only be set on architectures that have completely implemented memory hotplug
145 # feature. If you are not sure, don't touch it.
147 config HAVE_BOOTMEM_INFO_NODE
150 # eventually, we can have this option just 'select SPARSEMEM'
151 config MEMORY_HOTPLUG
152 bool "Allow for memory hot-add"
153 depends on SPARSEMEM || X86_64_ACPI_NUMA
154 depends on ARCH_ENABLE_MEMORY_HOTPLUG
156 config MEMORY_HOTPLUG_SPARSE
158 depends on SPARSEMEM && MEMORY_HOTPLUG
160 config MEMORY_HOTPLUG_DEFAULT_ONLINE
161 bool "Online the newly added memory blocks by default"
162 depends on MEMORY_HOTPLUG
164 This option sets the default policy setting for memory hotplug
165 onlining policy (/sys/devices/system/memory/auto_online_blocks) which
166 determines what happens to newly added memory regions. Policy setting
167 can always be changed at runtime.
168 See Documentation/memory-hotplug.txt for more information.
170 Say Y here if you want all hot-plugged memory blocks to appear in
171 'online' state by default.
172 Say N here if you want the default policy to keep all hot-plugged
173 memory blocks in 'offline' state.
175 config MEMORY_HOTREMOVE
176 bool "Allow for memory hot remove"
177 select MEMORY_ISOLATION
178 select HAVE_BOOTMEM_INFO_NODE if (X86_64 || PPC64)
179 depends on MEMORY_HOTPLUG && ARCH_ENABLE_MEMORY_HOTREMOVE
182 # Heavily threaded applications may benefit from splitting the mm-wide
183 # page_table_lock, so that faults on different parts of the user address
184 # space can be handled with less contention: split it at this NR_CPUS.
185 # Default to 4 for wider testing, though 8 might be more appropriate.
186 # ARM's adjust_pte (unused if VIPT) depends on mm-wide page_table_lock.
187 # PA-RISC 7xxx's spinlock_t would enlarge struct page from 32 to 44 bytes.
188 # DEBUG_SPINLOCK and DEBUG_LOCK_ALLOC spinlock_t also enlarge struct page.
190 config SPLIT_PTLOCK_CPUS
192 default "999999" if !MMU
193 default "999999" if ARM && !CPU_CACHE_VIPT
194 default "999999" if PARISC && !PA20
197 config ARCH_ENABLE_SPLIT_PMD_PTLOCK
201 # support for memory balloon
202 config MEMORY_BALLOON
206 # support for memory balloon compaction
207 config BALLOON_COMPACTION
208 bool "Allow for balloon memory compaction/migration"
210 depends on COMPACTION && MEMORY_BALLOON
212 Memory fragmentation introduced by ballooning might reduce
213 significantly the number of 2MB contiguous memory blocks that can be
214 used within a guest, thus imposing performance penalties associated
215 with the reduced number of transparent huge pages that could be used
216 by the guest workload. Allowing the compaction & migration for memory
217 pages enlisted as being part of memory balloon devices avoids the
218 scenario aforementioned and helps improving memory defragmentation.
221 # support for memory compaction
223 bool "Allow for memory compaction"
228 Compaction is the only memory management component to form
229 high order (larger physically contiguous) memory blocks
230 reliably. The page allocator relies on compaction heavily and
231 the lack of the feature can lead to unexpected OOM killer
232 invocations for high order memory requests. You shouldn't
233 disable this option unless there really is a strong reason for
234 it and then we would be really interested to hear about that at
238 # support for page migration
241 bool "Page migration"
243 depends on (NUMA || ARCH_ENABLE_MEMORY_HOTREMOVE || COMPACTION || CMA) && MMU
245 Allows the migration of the physical location of pages of processes
246 while the virtual addresses are not changed. This is useful in
247 two situations. The first is on NUMA systems to put pages nearer
248 to the processors accessing. The second is when allocating huge
249 pages as migration can relocate pages to satisfy a huge page
250 allocation instead of reclaiming.
252 config ARCH_ENABLE_HUGEPAGE_MIGRATION
255 config ARCH_ENABLE_THP_MIGRATION
259 def_bool (MEMORY_ISOLATION && COMPACTION) || CMA
261 config PHYS_ADDR_T_64BIT
265 bool "Enable bounce buffers"
267 depends on BLOCK && MMU && (ZONE_DMA || HIGHMEM)
269 Enable bounce buffers for devices that cannot access
270 the full range of memory available to the CPU. Enabled
271 by default when ZONE_DMA or HIGHMEM is selected, but you
272 may say n to override this.
282 An architecture should select this if it implements the
283 deprecated interface virt_to_bus(). All new architectures
284 should probably not select this.
292 bool "Enable KSM for page merging"
296 Enable Kernel Samepage Merging: KSM periodically scans those areas
297 of an application's address space that an app has advised may be
298 mergeable. When it finds pages of identical content, it replaces
299 the many instances by a single page with that content, so
300 saving memory until one or another app needs to modify the content.
301 Recommended for use with KVM, or with other duplicative applications.
302 See Documentation/vm/ksm.rst for more information: KSM is inactive
303 until a program has madvised that an area is MADV_MERGEABLE, and
304 root has set /sys/kernel/mm/ksm/run to 1 (if CONFIG_SYSFS is set).
306 config DEFAULT_MMAP_MIN_ADDR
307 int "Low address space to protect from user allocation"
311 This is the portion of low virtual memory which should be protected
312 from userspace allocation. Keeping a user from writing to low pages
313 can help reduce the impact of kernel NULL pointer bugs.
315 For most ia64, ppc64 and x86 users with lots of address space
316 a value of 65536 is reasonable and should cause no problems.
317 On arm and other archs it should not be higher than 32768.
318 Programs which use vm86 functionality or have some need to map
319 this low address space will need CAP_SYS_RAWIO or disable this
320 protection by setting the value to 0.
322 This value can be changed after boot using the
323 /proc/sys/vm/mmap_min_addr tunable.
325 config ARCH_SUPPORTS_MEMORY_FAILURE
328 config MEMORY_FAILURE
330 depends on ARCH_SUPPORTS_MEMORY_FAILURE
331 bool "Enable recovery from hardware memory errors"
332 select MEMORY_ISOLATION
335 Enables code to recover from some memory failures on systems
336 with MCA recovery. This allows a system to continue running
337 even when some of its memory has uncorrected errors. This requires
338 special hardware support and typically ECC memory.
340 config HWPOISON_INJECT
341 tristate "HWPoison pages injector"
342 depends on MEMORY_FAILURE && DEBUG_KERNEL && PROC_FS
343 select PROC_PAGE_MONITOR
345 config NOMMU_INITIAL_TRIM_EXCESS
346 int "Turn on mmap() excess space trimming before booting"
350 The NOMMU mmap() frequently needs to allocate large contiguous chunks
351 of memory on which to store mappings, but it can only ask the system
352 allocator for chunks in 2^N*PAGE_SIZE amounts - which is frequently
353 more than it requires. To deal with this, mmap() is able to trim off
354 the excess and return it to the allocator.
356 If trimming is enabled, the excess is trimmed off and returned to the
357 system allocator, which can cause extra fragmentation, particularly
358 if there are a lot of transient processes.
360 If trimming is disabled, the excess is kept, but not used, which for
361 long-term mappings means that the space is wasted.
363 Trimming can be dynamically controlled through a sysctl option
364 (/proc/sys/vm/nr_trim_pages) which specifies the minimum number of
365 excess pages there must be before trimming should occur, or zero if
366 no trimming is to occur.
368 This option specifies the initial value of this option. The default
369 of 1 says that all excess pages should be trimmed.
371 See Documentation/nommu-mmap.txt for more information.
373 config TRANSPARENT_HUGEPAGE
374 bool "Transparent Hugepage Support"
375 depends on HAVE_ARCH_TRANSPARENT_HUGEPAGE
379 Transparent Hugepages allows the kernel to use huge pages and
380 huge tlb transparently to the applications whenever possible.
381 This feature can improve computing performance to certain
382 applications by speeding up page faults during memory
383 allocation, by reducing the number of tlb misses and by speeding
384 up the pagetable walking.
386 If memory constrained on embedded, you may want to say N.
389 prompt "Transparent Hugepage Support sysfs defaults"
390 depends on TRANSPARENT_HUGEPAGE
391 default TRANSPARENT_HUGEPAGE_ALWAYS
393 Selects the sysfs defaults for Transparent Hugepage Support.
395 config TRANSPARENT_HUGEPAGE_ALWAYS
398 Enabling Transparent Hugepage always, can increase the
399 memory footprint of applications without a guaranteed
400 benefit but it will work automatically for all applications.
402 config TRANSPARENT_HUGEPAGE_MADVISE
405 Enabling Transparent Hugepage madvise, will only provide a
406 performance improvement benefit to the applications using
407 madvise(MADV_HUGEPAGE) but it won't risk to increase the
408 memory footprint of applications without a guaranteed
412 config ARCH_WANTS_THP_SWAP
417 depends on TRANSPARENT_HUGEPAGE && ARCH_WANTS_THP_SWAP && SWAP
419 Swap transparent huge pages in one piece, without splitting.
420 XXX: For now, swap cluster backing transparent huge page
421 will be split after swapout.
423 For selection by architectures with reasonable THP sizes.
425 config TRANSPARENT_HUGE_PAGECACHE
427 depends on TRANSPARENT_HUGEPAGE
430 # UP and nommu archs use km based percpu allocator
432 config NEED_PER_CPU_KM
438 bool "Enable cleancache driver to cache clean pages if tmem is present"
440 Cleancache can be thought of as a page-granularity victim cache
441 for clean pages that the kernel's pageframe replacement algorithm
442 (PFRA) would like to keep around, but can't since there isn't enough
443 memory. So when the PFRA "evicts" a page, it first attempts to use
444 cleancache code to put the data contained in that page into
445 "transcendent memory", memory that is not directly accessible or
446 addressable by the kernel and is of unknown and possibly
447 time-varying size. And when a cleancache-enabled
448 filesystem wishes to access a page in a file on disk, it first
449 checks cleancache to see if it already contains it; if it does,
450 the page is copied into the kernel and a disk access is avoided.
451 When a transcendent memory driver is available (such as zcache or
452 Xen transcendent memory), a significant I/O reduction
453 may be achieved. When none is available, all cleancache calls
454 are reduced to a single pointer-compare-against-NULL resulting
455 in a negligible performance hit.
457 If unsure, say Y to enable cleancache
460 bool "Enable frontswap to cache swap pages if tmem is present"
463 Frontswap is so named because it can be thought of as the opposite
464 of a "backing" store for a swap device. The data is stored into
465 "transcendent memory", memory that is not directly accessible or
466 addressable by the kernel and is of unknown and possibly
467 time-varying size. When space in transcendent memory is available,
468 a significant swap I/O reduction may be achieved. When none is
469 available, all frontswap calls are reduced to a single pointer-
470 compare-against-NULL resulting in a negligible performance hit
471 and swap data is stored as normal on the matching swap device.
473 If unsure, say Y to enable frontswap.
476 bool "Contiguous Memory Allocator"
479 select MEMORY_ISOLATION
481 This enables the Contiguous Memory Allocator which allows other
482 subsystems to allocate big physically-contiguous blocks of memory.
483 CMA reserves a region of memory and allows only movable pages to
484 be allocated from it. This way, the kernel can use the memory for
485 pagecache and when a subsystem requests for contiguous area, the
486 allocated pages are migrated away to serve the contiguous request.
491 bool "CMA debug messages (DEVELOPMENT)"
492 depends on DEBUG_KERNEL && CMA
494 Turns on debug messages in CMA. This produces KERN_DEBUG
495 messages for every CMA call as well as various messages while
496 processing calls such as dma_alloc_from_contiguous().
497 This option does not affect warning and error messages.
500 bool "CMA debugfs interface"
501 depends on CMA && DEBUG_FS
503 Turns on the DebugFS interface for CMA.
506 int "Maximum count of the CMA areas"
510 CMA allows to create CMA areas for particular purpose, mainly,
511 used as device private area. This parameter sets the maximum
512 number of CMA area in the system.
514 If unsure, leave the default value "7".
516 config MEM_SOFT_DIRTY
517 bool "Track memory changes"
518 depends on CHECKPOINT_RESTORE && HAVE_ARCH_SOFT_DIRTY && PROC_FS
519 select PROC_PAGE_MONITOR
521 This option enables memory changes tracking by introducing a
522 soft-dirty bit on pte-s. This bit it set when someone writes
523 into a page just as regular dirty bit, but unlike the latter
524 it can be cleared by hands.
526 See Documentation/admin-guide/mm/soft-dirty.rst for more details.
529 bool "Compressed cache for swap pages (EXPERIMENTAL)"
530 depends on FRONTSWAP && CRYPTO=y
534 A lightweight compressed cache for swap pages. It takes
535 pages that are in the process of being swapped out and attempts to
536 compress them into a dynamically allocated RAM-based memory pool.
537 This can result in a significant I/O reduction on swap device and,
538 in the case where decompressing from RAM is faster that swap device
539 reads, can also improve workload performance.
541 This is marked experimental because it is a new feature (as of
542 v3.11) that interacts heavily with memory reclaim. While these
543 interactions don't cause any known issues on simple memory setups,
544 they have not be fully explored on the large set of potential
545 configurations and workloads that exist.
548 tristate "Common API for compressed memory storage"
550 Compressed memory storage API. This allows using either zbud or
554 tristate "Low (Up to 2x) density storage for compressed pages"
556 A special purpose allocator for storing compressed pages.
557 It is designed to store up to two compressed pages per physical
558 page. While this design limits storage density, it has simple and
559 deterministic reclaim properties that make it preferable to a higher
560 density approach when reclaim will be used.
563 tristate "Up to 3x density storage for compressed pages"
566 A special purpose allocator for storing compressed pages.
567 It is designed to store up to three compressed pages per physical
568 page. It is a ZBUD derivative so the simplicity and determinism are
572 tristate "Memory allocator for compressed pages"
575 zsmalloc is a slab-based memory allocator designed to store
576 compressed RAM pages. zsmalloc uses virtual memory mapping
577 in order to reduce fragmentation. However, this results in a
578 non-standard allocator interface where a handle, not a pointer, is
579 returned by an alloc(). This handle must be mapped in order to
580 access the allocated space.
582 config PGTABLE_MAPPING
583 bool "Use page table mapping to access object in zsmalloc"
586 By default, zsmalloc uses a copy-based object mapping method to
587 access allocations that span two pages. However, if a particular
588 architecture (ex, ARM) performs VM mapping faster than copying,
589 then you should select this. This causes zsmalloc to use page table
590 mapping rather than copying for object mapping.
592 You can check speed with zsmalloc benchmark:
593 https://github.com/spartacus06/zsmapbench
596 bool "Export zsmalloc statistics"
600 This option enables code in the zsmalloc to collect various
601 statistics about whats happening in zsmalloc and exports that
602 information to userspace via debugfs.
605 config GENERIC_EARLY_IOREMAP
608 config MAX_STACK_SIZE_MB
609 int "Maximum user stack size for 32-bit processes (MB)"
612 depends on STACK_GROWSUP && (!64BIT || COMPAT)
614 This is the maximum stack size in Megabytes in the VM layout of 32-bit
615 user processes when the stack grows upwards (currently only on parisc
616 arch). The stack will be located at the highest memory address minus
617 the given value, unless the RLIMIT_STACK hard limit is changed to a
618 smaller value in which case that is used.
620 A sane initial value is 80 MB.
622 config DEFERRED_STRUCT_PAGE_INIT
623 bool "Defer initialisation of struct pages to kthreads"
625 depends on !NEED_PER_CPU_KM
628 Ordinarily all struct pages are initialised during early boot in a
629 single thread. On very large machines this can take a considerable
630 amount of time. If this option is set, large machines will bring up
631 a subset of memmap at boot and then initialise the rest in parallel
632 by starting one-off "pgdatinitX" kernel thread for each node X. This
633 has a potential performance impact on processes running early in the
634 lifetime of the system until these kthreads finish the
637 config IDLE_PAGE_TRACKING
638 bool "Enable idle page tracking"
639 depends on SYSFS && MMU
640 select PAGE_EXTENSION if !64BIT
642 This feature allows to estimate the amount of user pages that have
643 not been touched during a given period of time. This information can
644 be useful to tune memory cgroup limits and/or for job placement
645 within a compute cluster.
647 See Documentation/admin-guide/mm/idle_page_tracking.rst for
650 # arch_add_memory() comprehends device memory
651 config ARCH_HAS_ZONE_DEVICE
655 bool "Device memory (pmem, HMM, etc...) hotplug support"
656 depends on MEMORY_HOTPLUG
657 depends on MEMORY_HOTREMOVE
658 depends on SPARSEMEM_VMEMMAP
659 depends on ARCH_HAS_ZONE_DEVICE
663 Device memory hotplug support allows for establishing pmem,
664 or other device driver discovered memory regions, in the
665 memmap. This allows pfn_to_page() lookups of otherwise
666 "device-physical" addresses which is needed for using a DAX
667 mapping in an O_DIRECT operation, among other things.
669 If FS_DAX is enabled, then say Y.
671 config ARCH_HAS_HMM_MIRROR
674 depends on (X86_64 || PPC64)
675 depends on MMU && 64BIT
677 config ARCH_HAS_HMM_DEVICE
680 depends on (X86_64 || PPC64)
681 depends on MEMORY_HOTPLUG
682 depends on MEMORY_HOTREMOVE
683 depends on SPARSEMEM_VMEMMAP
684 depends on ARCH_HAS_ZONE_DEVICE
690 depends on (X86_64 || PPC64)
691 depends on ZONE_DEVICE
692 depends on MMU && 64BIT
693 depends on MEMORY_HOTPLUG
694 depends on MEMORY_HOTREMOVE
695 depends on SPARSEMEM_VMEMMAP
697 config MIGRATE_VMA_HELPER
700 config DEV_PAGEMAP_OPS
706 select MIGRATE_VMA_HELPER
709 bool "HMM mirror CPU page table into a device page table"
710 depends on ARCH_HAS_HMM
713 Select HMM_MIRROR if you want to mirror range of the CPU page table of a
714 process into a device page table. Here, mirror means "keep synchronized".
715 Prerequisites: the device must provide the ability to write-protect its
716 page tables (at PAGE_SIZE granularity), and must be able to recover from
717 the resulting potential page faults.
719 config DEVICE_PRIVATE
720 bool "Unaddressable device memory (GPU memory, ...)"
721 depends on ARCH_HAS_HMM
723 select DEV_PAGEMAP_OPS
726 Allows creation of struct pages to represent unaddressable device
727 memory; i.e., memory that is only accessible from the device (or
728 group of devices). You likely also want to select HMM_MIRROR.
731 bool "Addressable device memory (like GPU memory)"
732 depends on ARCH_HAS_HMM
734 select DEV_PAGEMAP_OPS
737 Allows creation of struct pages to represent addressable device
738 memory; i.e., memory that is accessible from both the device and
744 config ARCH_USES_HIGH_VMA_FLAGS
746 config ARCH_HAS_PKEYS
750 bool "Collect percpu memory statistics"
752 This feature collects and exposes statistics via debugfs. The
753 information includes global and per chunk statistics, which can
754 be used to help understand percpu memory usage.
757 bool "Enable infrastructure for get_user_pages_fast() benchmarking"
759 Provides /sys/kernel/debug/gup_benchmark that helps with testing
760 performance of get_user_pages_fast().
762 See tools/testing/selftests/vm/gup_benchmark.c
764 config ARCH_HAS_PTE_SPECIAL