1 # SPDX-License-Identifier: GPL-2.0-only
2
3 menu "Memory Management options"
4
5 #
6 # For some reason microblaze and nios2 hard code SWAP=n. Hopefully we can
7 # add proper SWAP support to them, in which case this can be remove.
8 #
9 config ARCH_NO_SWAP
10 bool
11
12 config ZPOOL
13 bool
14
15 menuconfig SWAP
16 bool "Support for paging of anonymous memory (swap)"
17 depends on MMU && BLOCK && !ARCH_NO_SWAP
18 default y
19 help
20 This option allows you to choose whether you want to have support
21 for so called swap devices or swap files in your kernel that are
22 used to provide more virtual memory than the actual RAM present
23 in your computer. If unsure say Y.
24
25 config ZSWAP
26 bool "Compressed cache for swap pages"
27 depends on SWAP
28 select CRYPTO
29 select ZPOOL
30 help
31 A lightweight compressed cache for swap pages. It takes
32 pages that are in the process of being swapped out and attempts to
33 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 device
36 reads, can also improve workload performance.
37
38 config ZSWAP_DEFAULT_ON
39 bool "Enable the compressed cache for swap pages by default"
40 depends on ZSWAP
41 help
42 If selected, the compressed cache for swap pages will be enabled
43 at boot, otherwise it will be disabled.
44
45 The selection made here can be overridden by using the kernel
46 command line 'zswap.enabled=' option.
47
48 config ZSWAP_SHRINKER_DEFAULT_ON
49 bool "Shrink the zswap pool on memory pressure"
50 depends on ZSWAP
51 default n
52 help
53 If selected, the zswap shrinker will be enabled, and the pages
54 stored in the zswap pool will become available for reclaim (i.e
55 written back to the backing swap device) on memory pressure.
56
57 This means that zswap writeback could happen even if the pool is
58 not yet full, or the cgroup zswap limit has not been reached,
59 reducing the chance that cold pages will reside in the zswap pool
60 and consume memory indefinitely.
61
62 choice
63 prompt "Default compressor"
64 depends on ZSWAP
65 default ZSWAP_COMPRESSOR_DEFAULT_LZO
66 help
67 Selects the default compression algorithm for the compressed cache
68 for swap pages.
69
70 For an overview what kind of performance can be expected from
71 a particular compression algorithm please refer to the benchmarks
72 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 kernel
78 command line 'zswap.compressor=' option.
79
80 config ZSWAP_COMPRESSOR_DEFAULT_DEFLATE
81 bool "Deflate"
82 select CRYPTO_DEFLATE
83 help
84 Use the Deflate algorithm as the default compression algorithm.
85
86 config ZSWAP_COMPRESSOR_DEFAULT_LZO
87 bool "LZO"
88 select CRYPTO_LZO
89 help
90 Use the LZO algorithm as the default compression algorithm.
91
92 config ZSWAP_COMPRESSOR_DEFAULT_842
93 bool "842"
94 select CRYPTO_842
95 help
96 Use the 842 algorithm as the default compression algorithm.
97
98 config ZSWAP_COMPRESSOR_DEFAULT_LZ4
99 bool "LZ4"
100 select CRYPTO_LZ4
101 help
102 Use the LZ4 algorithm as the default compression algorithm.
103
104 config ZSWAP_COMPRESSOR_DEFAULT_LZ4HC
105 bool "LZ4HC"
106 select CRYPTO_LZ4HC
107 help
108 Use the LZ4HC algorithm as the default compression algorithm.
109
110 config ZSWAP_COMPRESSOR_DEFAULT_ZSTD
111 bool "zstd"
112 select CRYPTO_ZSTD
113 help
114 Use the zstd algorithm as the default compression algorithm.
115 endchoice
116
117 config ZSWAP_COMPRESSOR_DEFAULT
118 string
119 depends on ZSWAP
120 default "deflate" if ZSWAP_COMPRESSOR_DEFAULT_DEFLATE
121 default "lzo" if ZSWAP_COMPRESSOR_DEFAULT_LZO
122 default "842" if ZSWAP_COMPRESSOR_DEFAULT_842
123 default "lz4" if ZSWAP_COMPRESSOR_DEFAULT_LZ4
124 default "lz4hc" if ZSWAP_COMPRESSOR_DEFAULT_LZ4HC
125 default "zstd" if ZSWAP_COMPRESSOR_DEFAULT_ZSTD
126 default ""
127
128 choice
129 prompt "Default allocator"
130 depends on ZSWAP
131 default ZSWAP_ZPOOL_DEFAULT_ZSMALLOC if HAVE_ZSMALLOC
132 default ZSWAP_ZPOOL_DEFAULT_ZBUD
133 help
134 Selects the default allocator for the compressed cache for
135 swap pages.
136 The default is 'zbud' for compatibility, however please do
137 read the description of each of the allocators below before
138 making a right choice.
139
140 The selection made here can be overridden by using the kernel
141 command line 'zswap.zpool=' option.
142
143 config ZSWAP_ZPOOL_DEFAULT_ZBUD
144 bool "zbud"
145 select ZBUD
146 help
147 Use the zbud allocator as the default allocator.
148
149 config ZSWAP_ZPOOL_DEFAULT_Z3FOLD_DEPRECATED
150 bool "z3foldi (DEPRECATED)"
151 select Z3FOLD_DEPRECATED
152 help
153 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
158 config ZSWAP_ZPOOL_DEFAULT_ZSMALLOC
159 bool "zsmalloc"
160 depends on HAVE_ZSMALLOC
161 select ZSMALLOC
162 help
163 Use the zsmalloc allocator as the default allocator.
164 endchoice
165
166 config ZSWAP_ZPOOL_DEFAULT
167 string
168 depends on ZSWAP
169 default "zbud" if ZSWAP_ZPOOL_DEFAULT_ZBUD
170 default "z3fold" if ZSWAP_ZPOOL_DEFAULT_Z3FOLD_DEPRECATED
171 default "zsmalloc" if ZSWAP_ZPOOL_DEFAULT_ZSMALLOC
172 default ""
173
174 config ZBUD
175 tristate "2:1 compression allocator (zbud)"
176 depends on ZSWAP
177 help
178 A special purpose allocator for storing compressed pages.
179 It is designed to store up to two compressed pages per physical
180 page. While this design limits storage density, it has simple and
181 deterministic reclaim properties that make it preferable to a higher
182 density approach when reclaim will be used.
183
184 config Z3FOLD_DEPRECATED
185 tristate "3:1 compression allocator (z3fold) (DEPRECATED)"
186 depends on ZSWAP
187 help
188 Deprecated and scheduled for removal in a few cycles. If you have
189 a good reason for using Z3FOLD over ZSMALLOC, please contact
190 linux-mm@kvack.org and the zswap maintainers.
191
192 A special purpose allocator for storing compressed pages.
193 It is designed to store up to three compressed pages per physical
194 page. It is a ZBUD derivative so the simplicity and determinism are
195 still there.
196
197 config Z3FOLD
198 tristate
199 default y if Z3FOLD_DEPRECATED=y
200 default m if Z3FOLD_DEPRECATED=m
201 depends on Z3FOLD_DEPRECATED
202
203 config HAVE_ZSMALLOC
204 def_bool y
205 depends on MMU
206 depends on PAGE_SIZE_LESS_THAN_256KB # we want <= 64 KiB
207
208 config ZSMALLOC
209 tristate
210 prompt "N:1 compression allocator (zsmalloc)" if ZSWAP
211 depends on HAVE_ZSMALLOC
212 help
213 zsmalloc is a slab-based memory allocator designed to store
214 pages of various compression levels efficiently. It achieves
215 the highest storage density with the least amount of fragmentation.
216
217 config ZSMALLOC_STAT
218 bool "Export zsmalloc statistics"
219 depends on ZSMALLOC
220 select DEBUG_FS
221 help
222 This option enables code in the zsmalloc to collect various
223 statistics about what's happening in zsmalloc and exports that
224 information to userspace via debugfs.
225 If unsure, say N.
226
227 config ZSMALLOC_CHAIN_SIZE
228 int "Maximum number of physical pages per-zspage"
229 default 8
230 range 4 16
231 depends on ZSMALLOC
232 help
233 This option sets the upper limit on the number of physical pages
234 that a zmalloc page (zspage) can consist of. The optimal zspage
235 chain size is calculated for each size class during the
236 initialization of the pool.
237
238 Changing this option can alter the characteristics of size classes,
239 such as the number of pages per zspage and the number of objects
240 per zspage. This can also result in different configurations of
241 the pool, as zsmalloc merges size classes with similar
242 characteristics.
243
244 For more information, see zsmalloc documentation.
245
246 menu "Slab allocator options"
247
248 config SLUB
249 def_bool y
250
251 config SLUB_TINY
252 bool "Configure for minimal memory footprint"
253 depends on EXPERT
254 select SLAB_MERGE_DEFAULT
255 help
256 Configures the slab allocator in a way to achieve minimal memory
257 footprint, sacrificing scalability, debugging and other features.
258 This is intended only for the smallest system that had used the
259 SLOB allocator and is not recommended for systems with more than
260 16MB RAM.
261
262 If unsure, say N.
263
264 config SLAB_MERGE_DEFAULT
265 bool "Allow slab caches to be merged"
266 default y
267 help
268 For reduced kernel memory fragmentation, slab caches can be
269 merged when they share the same size and other characteristics.
270 This carries a risk of kernel heap overflows being able to
271 overwrite objects from merged caches (and more easily control
272 cache layout), which makes such heap attacks easier to exploit
273 by attackers. By keeping caches unmerged, these kinds of exploits
274 can usually only damage objects in the same cache. To disable
275 merging at runtime, "slab_nomerge" can be passed on the kernel
276 command line.
277
278 config SLAB_FREELIST_RANDOM
279 bool "Randomize slab freelist"
280 depends on !SLUB_TINY
281 help
282 Randomizes the freelist order used on creating new pages. This
283 security feature reduces the predictability of the kernel slab
284 allocator against heap overflows.
285
286 config SLAB_FREELIST_HARDENED
287 bool "Harden slab freelist metadata"
288 depends on !SLUB_TINY
289 help
290 Many kernel heap attacks try to target slab cache metadata and
291 other infrastructure. This options makes minor performance
292 sacrifices to harden the kernel slab allocator against common
293 freelist exploit methods.
294
295 config SLAB_BUCKETS
296 bool "Support allocation from separate kmalloc buckets"
297 depends on !SLUB_TINY
298 default SLAB_FREELIST_HARDENED
299 help
300 Kernel heap attacks frequently depend on being able to create
301 specifically-sized allocations with user-controlled contents
302 that will be allocated into the same kmalloc bucket as a
303 target object. To avoid sharing these allocation buckets,
304 provide an explicitly separated set of buckets to be used for
305 user-controlled allocations. This may very slightly increase
306 memory fragmentation, though in practice it's only a handful
307 of extra pages since the bulk of user-controlled allocations
308 are relatively long-lived.
309
310 If unsure, say Y.
311
312 config SLUB_STATS
313 default n
314 bool "Enable performance statistics"
315 depends on SYSFS && !SLUB_TINY
316 help
317 The statistics are useful to debug slab allocation behavior in
318 order find ways to optimize the allocator. This should never be
319 enabled for production use since keeping statistics slows down
320 the allocator by a few percentage points. The slabinfo command
321 supports the determination of the most active slabs to figure
322 out which slabs are relevant to a particular load.
323 Try running: slabinfo -DA
324
325 config SLUB_CPU_PARTIAL
326 default y
327 depends on SMP && !SLUB_TINY
328 bool "Enable per cpu partial caches"
329 help
330 Per cpu partial caches accelerate objects allocation and freeing
331 that is local to a processor at the price of more indeterminism
332 in the latency of the free. On overflow these caches will be cleared
333 which requires the taking of locks that may cause latency spikes.
334 Typically one would choose no for a realtime system.
335
336 config RANDOM_KMALLOC_CACHES
337 default n
338 depends on !SLUB_TINY
339 bool "Randomize slab caches for normal kmalloc"
340 help
341 A hardening feature that creates multiple copies of slab caches for
342 normal kmalloc allocation and makes kmalloc randomly pick one based
343 on code address, which makes the attackers more difficult to spray
344 vulnerable memory objects on the heap for the purpose of exploiting
345 memory vulnerabilities.
346
347 Currently the number of copies is set to 16, a reasonably large value
348 that effectively diverges the memory objects allocated for different
349 subsystems or modules into different caches, at the expense of a
350 limited degree of memory and CPU overhead that relates to hardware and
351 system workload.
352
353 endmenu # Slab allocator options
354
355 config SHUFFLE_PAGE_ALLOCATOR
356 bool "Page allocator randomization"
357 default SLAB_FREELIST_RANDOM && ACPI_NUMA
358 help
359 Randomization of the page allocator improves the average
360 utilization of a direct-mapped memory-side-cache. See section
361 5.2.27 Heterogeneous Memory Attribute Table (HMAT) in the ACPI
362 6.2a specification for an example of how a platform advertises
363 the presence of a memory-side-cache. There are also incidental
364 security benefits as it reduces the predictability of page
365 allocations to compliment SLAB_FREELIST_RANDOM, but the
366 default granularity of shuffling on the MAX_PAGE_ORDER i.e, 10th
367 order of pages is selected based on cache utilization benefits
368 on x86.
369
370 While the randomization improves cache utilization it may
371 negatively impact workloads on platforms without a cache. For
372 this reason, by default, the randomization is not enabled even
373 if SHUFFLE_PAGE_ALLOCATOR=y. The randomization may be force enabled
374 with the 'page_alloc.shuffle' kernel command line parameter.
375
376 Say Y if unsure.
377
378 config COMPAT_BRK
379 bool "Disable heap randomization"
380 default y
381 help
382 Randomizing heap placement makes heap exploits harder, but it
383 also breaks ancient binaries (including anything libc5 based).
384 This option changes the bootup default to heap randomization
385 disabled, and can be overridden at runtime by setting
386 /proc/sys/kernel/randomize_va_space to 2.
387
388 On non-ancient distros (post-2000 ones) N is usually a safe choice.
389
390 config MMAP_ALLOW_UNINITIALIZED
391 bool "Allow mmapped anonymous memory to be uninitialized"
392 depends on EXPERT && !MMU
393 default n
394 help
395 Normally, and according to the Linux spec, anonymous memory obtained
396 from mmap() has its contents cleared before it is passed to
397 userspace. Enabling this config option allows you to request that
398 mmap() skip that if it is given an MAP_UNINITIALIZED flag, thus
399 providing a huge performance boost. If this option is not enabled,
400 then the flag will be ignored.
401
402 This is taken advantage of by uClibc's malloc(), and also by
403 ELF-FDPIC binfmt's brk and stack allocator.
404
405 Because of the obvious security issues, this option should only be
406 enabled on embedded devices where you control what is run in
407 userspace. Since that isn't generally a problem on no-MMU systems,
408 it is normally safe to say Y here.
409
410 See Documentation/admin-guide/mm/nommu-mmap.rst for more information.
411
412 config SELECT_MEMORY_MODEL
413 def_bool y
414 depends on ARCH_SELECT_MEMORY_MODEL
415
416 choice
417 prompt "Memory model"
418 depends on SELECT_MEMORY_MODEL
419 default SPARSEMEM_MANUAL if ARCH_SPARSEMEM_DEFAULT
420 default FLATMEM_MANUAL
421 help
422 This option allows you to change some of the ways that
423 Linux manages its memory internally. Most users will
424 only have one option here selected by the architecture
425 configuration. This is normal.
426
427 config FLATMEM_MANUAL
428 bool "Flat Memory"
429 depends on !ARCH_SPARSEMEM_ENABLE || ARCH_FLATMEM_ENABLE
430 help
431 This option is best suited for non-NUMA systems with
432 flat address space. The FLATMEM is the most efficient
433 system in terms of performance and resource consumption
434 and it is the best option for smaller systems.
435
436 For systems that have holes in their physical address
437 spaces and for features like NUMA and memory hotplug,
438 choose "Sparse Memory".
439
440 If unsure, choose this option (Flat Memory) over any other.
441
442 config SPARSEMEM_MANUAL
443 bool "Sparse Memory"
444 depends on ARCH_SPARSEMEM_ENABLE
445 help
446 This will be the only option for some systems, including
447 memory hot-plug systems. This is normal.
448
449 This option provides efficient support for systems with
450 holes is their physical address space and allows memory
451 hot-plug and hot-remove.
452
453 If unsure, choose "Flat Memory" over this option.
454
455 endchoice
456
457 config SPARSEMEM
458 def_bool y
459 depends on (!SELECT_MEMORY_MODEL && ARCH_SPARSEMEM_ENABLE) || SPARSEMEM_MANUAL
460
461 config FLATMEM
462 def_bool y
463 depends on !SPARSEMEM || FLATMEM_MANUAL
464
465 #
466 # SPARSEMEM_EXTREME (which is the default) does some bootmem
467 # allocations when sparse_init() is called. If this cannot
468 # be done on your architecture, select this option. However,
469 # statically allocating the mem_section[] array can potentially
470 # consume vast quantities of .bss, so be careful.
471 #
472 # This option will also potentially produce smaller runtime code
473 # with gcc 3.4 and later.
474 #
475 config SPARSEMEM_STATIC
476 bool
477
478 #
479 # Architecture platforms which require a two level mem_section in SPARSEMEM
480 # must select this option. This is usually for architecture platforms with
481 # an extremely sparse physical address space.
482 #
483 config SPARSEMEM_EXTREME
484 def_bool y
485 depends on SPARSEMEM && !SPARSEMEM_STATIC
486
487 config SPARSEMEM_VMEMMAP_ENABLE
488 bool
489
490 config SPARSEMEM_VMEMMAP
491 bool "Sparse Memory virtual memmap"
492 depends on SPARSEMEM && SPARSEMEM_VMEMMAP_ENABLE
493 default y
494 help
495 SPARSEMEM_VMEMMAP uses a virtually mapped memmap to optimise
496 pfn_to_page and page_to_pfn operations. This is the most
497 efficient option when sufficient kernel resources are available.
498 #
499 # Select this config option from the architecture Kconfig, if it is preferred
500 # to enable the feature of HugeTLB/dev_dax vmemmap optimization.
501 #
502 config ARCH_WANT_OPTIMIZE_DAX_VMEMMAP
503 bool
504
505 config ARCH_WANT_OPTIMIZE_HUGETLB_VMEMMAP
506 bool
507
508 config HAVE_MEMBLOCK_PHYS_MAP
509 bool
510
511 config HAVE_GUP_FAST
512 depends on MMU
513 bool
514
515 # Don't discard allocated memory used to track "memory" and "reserved" memblocks
516 # after early boot, so it can still be used to test for validity of memory.
517 # Also, memblocks are updated with memory hot(un)plug.
518 config ARCH_KEEP_MEMBLOCK
519 bool
520
521 # Keep arch NUMA mapping infrastructure post-init.
522 config NUMA_KEEP_MEMINFO
523 bool
524
525 config MEMORY_ISOLATION
526 bool
527
528 # IORESOURCE_SYSTEM_RAM regions in the kernel resource tree that are marked
529 # IORESOURCE_EXCLUSIVE cannot be mapped to user space, for example, via
530 # /dev/mem.
531 config EXCLUSIVE_SYSTEM_RAM
532 def_bool y
533 depends on !DEVMEM || STRICT_DEVMEM
534
535 #
536 # Only be set on architectures that have completely implemented memory hotplug
537 # feature. If you are not sure, don't touch it.
538 #
539 config HAVE_BOOTMEM_INFO_NODE
540 def_bool n
541
542 config ARCH_ENABLE_MEMORY_HOTPLUG
543 bool
544
545 config ARCH_ENABLE_MEMORY_HOTREMOVE
546 bool
547
548 # eventually, we can have this option just 'select SPARSEMEM'
549 menuconfig MEMORY_HOTPLUG
550 bool "Memory hotplug"
551 select MEMORY_ISOLATION
552 depends on SPARSEMEM
553 depends on ARCH_ENABLE_MEMORY_HOTPLUG
554 depends on 64BIT
555 select NUMA_KEEP_MEMINFO if NUMA
556
557 if MEMORY_HOTPLUG
558
559 config MEMORY_HOTPLUG_DEFAULT_ONLINE
560 bool "Online the newly added memory blocks by default"
561 depends on MEMORY_HOTPLUG
562 help
563 This option sets the default policy setting for memory hotplug
564 onlining policy (/sys/devices/system/memory/auto_online_blocks) which
565 determines what happens to newly added memory regions. Policy setting
566 can always be changed at runtime.
567 See Documentation/admin-guide/mm/memory-hotplug.rst for more information.
568
569 Say Y here if you want all hot-plugged memory blocks to appear in
570 'online' state by default.
571 Say N here if you want the default policy to keep all hot-plugged
572 memory blocks in 'offline' state.
573
574 config MEMORY_HOTREMOVE
575 bool "Allow for memory hot remove"
576 select HAVE_BOOTMEM_INFO_NODE if (X86_64 || PPC64)
577 depends on MEMORY_HOTPLUG && ARCH_ENABLE_MEMORY_HOTREMOVE
578 depends on MIGRATION
579
580 config MHP_MEMMAP_ON_MEMORY
581 def_bool y
582 depends on MEMORY_HOTPLUG && SPARSEMEM_VMEMMAP
583 depends on ARCH_MHP_MEMMAP_ON_MEMORY_ENABLE
584
585 endif # MEMORY_HOTPLUG
586
587 config ARCH_MHP_MEMMAP_ON_MEMORY_ENABLE
588 bool
589
590 # Heavily threaded applications may benefit from splitting the mm-wide
591 # page_table_lock, so that faults on different parts of the user address
592 # space can be handled with less contention: split it at this NR_CPUS.
593 # Default to 4 for wider testing, though 8 might be more appropriate.
594 # ARM's adjust_pte (unused if VIPT) depends on mm-wide page_table_lock.
595 # PA-RISC 7xxx's spinlock_t would enlarge struct page from 32 to 44 bytes.
596 # SPARC32 allocates multiple pte tables within a single page, and therefore
597 # a per-page lock leads to problems when multiple tables need to be locked
598 # at the same time (e.g. copy_page_range()).
599 # DEBUG_SPINLOCK and DEBUG_LOCK_ALLOC spinlock_t also enlarge struct page.
600 #
601 config SPLIT_PTLOCK_CPUS
602 int
603 default "999999" if !MMU
604 default "999999" if ARM && !CPU_CACHE_VIPT
605 default "999999" if PARISC && !PA20
606 default "999999" if SPARC32
607 default "4"
608
609 config ARCH_ENABLE_SPLIT_PMD_PTLOCK
610 bool
611
612 #
613 # support for memory balloon
614 config MEMORY_BALLOON
615 bool
616
617 #
618 # support for memory balloon compaction
619 config BALLOON_COMPACTION
620 bool "Allow for balloon memory compaction/migration"
621 default y
622 depends on COMPACTION && MEMORY_BALLOON
623 help
624 Memory fragmentation introduced by ballooning might reduce
625 significantly the number of 2MB contiguous memory blocks that can be
626 used within a guest, thus imposing performance penalties associated
627 with the reduced number of transparent huge pages that could be used
628 by the guest workload. Allowing the compaction & migration for memory
629 pages enlisted as being part of memory balloon devices avoids the
630 scenario aforementioned and helps improving memory defragmentation.
631
632 #
633 # support for memory compaction
634 config COMPACTION
635 bool "Allow for memory compaction"
636 default y
637 select MIGRATION
638 depends on MMU
639 help
640 Compaction is the only memory management component to form
641 high order (larger physically contiguous) memory blocks
642 reliably. The page allocator relies on compaction heavily and
643 the lack of the feature can lead to unexpected OOM killer
644 invocations for high order memory requests. You shouldn't
645 disable this option unless there really is a strong reason for
646 it and then we would be really interested to hear about that at
647 linux-mm@kvack.org.
648
649 config COMPACT_UNEVICTABLE_DEFAULT
650 int
651 depends on COMPACTION
652 default 0 if PREEMPT_RT
653 default 1
654
655 #
656 # support for free page reporting
657 config PAGE_REPORTING
658 bool "Free page reporting"
659 help
660 Free page reporting allows for the incremental acquisition of
661 free pages from the buddy allocator for the purpose of reporting
662 those pages to another entity, such as a hypervisor, so that the
663 memory can be freed within the host for other uses.
664
665 #
666 # support for page migration
667 #
668 config MIGRATION
669 bool "Page migration"
670 default y
671 depends on (NUMA || ARCH_ENABLE_MEMORY_HOTREMOVE || COMPACTION || CMA) && MMU
672 help
673 Allows the migration of the physical location of pages of processes
674 while the virtual addresses are not changed. This is useful in
675 two situations. The first is on NUMA systems to put pages nearer
676 to the processors accessing. The second is when allocating huge
677 pages as migration can relocate pages to satisfy a huge page
678 allocation instead of reclaiming.
679
680 config DEVICE_MIGRATION
681 def_bool MIGRATION && ZONE_DEVICE
682
683 config ARCH_ENABLE_HUGEPAGE_MIGRATION
684 bool
685
686 config ARCH_ENABLE_THP_MIGRATION
687 bool
688
689 config HUGETLB_PAGE_SIZE_VARIABLE
690 def_bool n
691 help
692 Allows the pageblock_order value to be dynamic instead of just standard
693 HUGETLB_PAGE_ORDER when there are multiple HugeTLB page sizes available
694 on a platform.
695
696 Note that the pageblock_order cannot exceed MAX_PAGE_ORDER and will be
697 clamped down to MAX_PAGE_ORDER.
698
699 config CONTIG_ALLOC
700 def_bool (MEMORY_ISOLATION && COMPACTION) || CMA
701
702 config PCP_BATCH_SCALE_MAX
703 int "Maximum scale factor of PCP (Per-CPU pageset) batch allocate/free"
704 default 5
705 range 0 6
706 help
707 In page allocator, PCP (Per-CPU pageset) is refilled and drained in
708 batches. The batch number is scaled automatically to improve page
709 allocation/free throughput. But too large scale factor may hurt
710 latency. This option sets the upper limit of scale factor to limit
711 the maximum latency.
712
713 config PHYS_ADDR_T_64BIT
714 def_bool 64BIT
715
716 config BOUNCE
717 bool "Enable bounce buffers"
718 default y
719 depends on BLOCK && MMU && HIGHMEM
720 help
721 Enable bounce buffers for devices that cannot access the full range of
722 memory available to the CPU. Enabled by default when HIGHMEM is
723 selected, but you may say n to override this.
724
725 config MMU_NOTIFIER
726 bool
727 select INTERVAL_TREE
728
729 config KSM
730 bool "Enable KSM for page merging"
731 depends on MMU
732 select XXHASH
733 help
734 Enable Kernel Samepage Merging: KSM periodically scans those areas
735 of an application's address space that an app has advised may be
736 mergeable. When it finds pages of identical content, it replaces
737 the many instances by a single page with that content, so
738 saving memory until one or another app needs to modify the content.
739 Recommended for use with KVM, or with other duplicative applications.
740 See Documentation/mm/ksm.rst for more information: KSM is inactive
741 until a program has madvised that an area is MADV_MERGEABLE, and
742 root has set /sys/kernel/mm/ksm/run to 1 (if CONFIG_SYSFS is set).
743
744 config DEFAULT_MMAP_MIN_ADDR
745 int "Low address space to protect from user allocation"
746 depends on MMU
747 default 4096
748 help
749 This is the portion of low virtual memory which should be protected
750 from userspace allocation. Keeping a user from writing to low pages
751 can help reduce the impact of kernel NULL pointer bugs.
752
753 For most arm64, ppc64 and x86 users with lots of address space
754 a value of 65536 is reasonable and should cause no problems.
755 On arm and other archs it should not be higher than 32768.
756 Programs which use vm86 functionality or have some need to map
757 this low address space will need CAP_SYS_RAWIO or disable this
758 protection by setting the value to 0.
759
760 This value can be changed after boot using the
761 /proc/sys/vm/mmap_min_addr tunable.
762
763 config ARCH_SUPPORTS_MEMORY_FAILURE
764 bool
765
766 config MEMORY_FAILURE
767 depends on MMU
768 depends on ARCH_SUPPORTS_MEMORY_FAILURE
769 bool "Enable recovery from hardware memory errors"
770 select MEMORY_ISOLATION
771 select RAS
772 help
773 Enables code to recover from some memory failures on systems
774 with MCA recovery. This allows a system to continue running
775 even when some of its memory has uncorrected errors. This requires
776 special hardware support and typically ECC memory.
777
778 config HWPOISON_INJECT
779 tristate "HWPoison pages injector"
780 depends on MEMORY_FAILURE && DEBUG_KERNEL && PROC_FS
781 select PROC_PAGE_MONITOR
782
783 config NOMMU_INITIAL_TRIM_EXCESS
784 int "Turn on mmap() excess space trimming before booting"
785 depends on !MMU
786 default 1
787 help
788 The NOMMU mmap() frequently needs to allocate large contiguous chunks
789 of memory on which to store mappings, but it can only ask the system
790 allocator for chunks in 2^N*PAGE_SIZE amounts - which is frequently
791 more than it requires. To deal with this, mmap() is able to trim off
792 the excess and return it to the allocator.
793
794 If trimming is enabled, the excess is trimmed off and returned to the
795 system allocator, which can cause extra fragmentation, particularly
796 if there are a lot of transient processes.
797
798 If trimming is disabled, the excess is kept, but not used, which for
799 long-term mappings means that the space is wasted.
800
801 Trimming can be dynamically controlled through a sysctl option
802 (/proc/sys/vm/nr_trim_pages) which specifies the minimum number of
803 excess pages there must be before trimming should occur, or zero if
804 no trimming is to occur.
805
806 This option specifies the initial value of this option. The default
807 of 1 says that all excess pages should be trimmed.
808
809 See Documentation/admin-guide/mm/nommu-mmap.rst for more information.
810
811 config ARCH_WANT_GENERAL_HUGETLB
812 bool
813
814 config ARCH_WANTS_THP_SWAP
815 def_bool n
816
817 menuconfig TRANSPARENT_HUGEPAGE
818 bool "Transparent Hugepage Support"
819 depends on HAVE_ARCH_TRANSPARENT_HUGEPAGE && !PREEMPT_RT
820 select COMPACTION
821 select XARRAY_MULTI
822 help
823 Transparent Hugepages allows the kernel to use huge pages and
824 huge tlb transparently to the applications whenever possible.
825 This feature can improve computing performance to certain
826 applications by speeding up page faults during memory
827 allocation, by reducing the number of tlb misses and by speeding
828 up the pagetable walking.
829
830 If memory constrained on embedded, you may want to say N.
831
832 if TRANSPARENT_HUGEPAGE
833
834 choice
835 prompt "Transparent Hugepage Support sysfs defaults"
836 depends on TRANSPARENT_HUGEPAGE
837 default TRANSPARENT_HUGEPAGE_ALWAYS
838 help
839 Selects the sysfs defaults for Transparent Hugepage Support.
840
841 config TRANSPARENT_HUGEPAGE_ALWAYS
842 bool "always"
843 help
844 Enabling Transparent Hugepage always, can increase the
845 memory footprint of applications without a guaranteed
846 benefit but it will work automatically for all applications.
847
848 config TRANSPARENT_HUGEPAGE_MADVISE
849 bool "madvise"
850 help
851 Enabling Transparent Hugepage madvise, will only provide a
852 performance improvement benefit to the applications using
853 madvise(MADV_HUGEPAGE) but it won't risk to increase the
854 memory footprint of applications without a guaranteed
855 benefit.
856
857 config TRANSPARENT_HUGEPAGE_NEVER
858 bool "never"
859 help
860 Disable Transparent Hugepage by default. It can still be
861 enabled at runtime via sysfs.
862 endchoice
863
864 config THP_SWAP
865 def_bool y
866 depends on TRANSPARENT_HUGEPAGE && ARCH_WANTS_THP_SWAP && SWAP && 64BIT
867 help
868 Swap transparent huge pages in one piece, without splitting.
869 XXX: For now, swap cluster backing transparent huge page
870 will be split after swapout.
871
872 For selection by architectures with reasonable THP sizes.
873
874 config READ_ONLY_THP_FOR_FS
875 bool "Read-only THP for filesystems (EXPERIMENTAL)"
876 depends on TRANSPARENT_HUGEPAGE && SHMEM
877
878 help
879 Allow khugepaged to put read-only file-backed pages in THP.
880
881 This is marked experimental because it is a new feature. Write
882 support of file THPs will be developed in the next few release
883 cycles.
884
885 endif # TRANSPARENT_HUGEPAGE
886
887 #
888 # The architecture supports pgtable leaves that is larger than PAGE_SIZE
889 #
890 config PGTABLE_HAS_HUGE_LEAVES
891 def_bool TRANSPARENT_HUGEPAGE || HUGETLB_PAGE
892
893 #
894 # UP and nommu archs use km based percpu allocator
895 #
896 config NEED_PER_CPU_KM
897 depends on !SMP || !MMU
898 bool
899 default y
900
901 config NEED_PER_CPU_EMBED_FIRST_CHUNK
902 bool
903
904 config NEED_PER_CPU_PAGE_FIRST_CHUNK
905 bool
906
907 config USE_PERCPU_NUMA_NODE_ID
908 bool
909
910 config HAVE_SETUP_PER_CPU_AREA
911 bool
912
913 config CMA
914 bool "Contiguous Memory Allocator"
915 depends on MMU
916 select MIGRATION
917 select MEMORY_ISOLATION
918 help
919 This enables the Contiguous Memory Allocator which allows other
920 subsystems to allocate big physically-contiguous blocks of memory.
921 CMA reserves a region of memory and allows only movable pages to
922 be allocated from it. This way, the kernel can use the memory for
923 pagecache and when a subsystem requests for contiguous area, the
924 allocated pages are migrated away to serve the contiguous request.
925
926 If unsure, say "n".
927
928 config CMA_DEBUGFS
929 bool "CMA debugfs interface"
930 depends on CMA && DEBUG_FS
931 help
932 Turns on the DebugFS interface for CMA.
933
934 config CMA_SYSFS
935 bool "CMA information through sysfs interface"
936 depends on CMA && SYSFS
937 help
938 This option exposes some sysfs attributes to get information
939 from CMA.
940
941 config CMA_AREAS
942 int "Maximum count of the CMA areas"
943 depends on CMA
944 default 20 if NUMA
945 default 8
946 help
947 CMA allows to create CMA areas for particular purpose, mainly,
948 used as device private area. This parameter sets the maximum
949 number of CMA area in the system.
950
951 If unsure, leave the default value "8" in UMA and "20" in NUMA.
952
953 config MEM_SOFT_DIRTY
954 bool "Track memory changes"
955 depends on CHECKPOINT_RESTORE && HAVE_ARCH_SOFT_DIRTY && PROC_FS
956 select PROC_PAGE_MONITOR
957 help
958 This option enables memory changes tracking by introducing a
959 soft-dirty bit on pte-s. This bit it set when someone writes
960 into a page just as regular dirty bit, but unlike the latter
961 it can be cleared by hands.
962
963 See Documentation/admin-guide/mm/soft-dirty.rst for more details.
964
965 config GENERIC_EARLY_IOREMAP
966 bool
967
968 config STACK_MAX_DEFAULT_SIZE_MB
969 int "Default maximum user stack size for 32-bit processes (MB)"
970 default 100
971 range 8 2048
972 depends on STACK_GROWSUP && (!64BIT || COMPAT)
973 help
974 This is the maximum stack size in Megabytes in the VM layout of 32-bit
975 user processes when the stack grows upwards (currently only on parisc
976 arch) when the RLIMIT_STACK hard limit is unlimited.
977
978 A sane initial value is 100 MB.
979
980 config DEFERRED_STRUCT_PAGE_INIT
981 bool "Defer initialisation of struct pages to kthreads"
982 depends on SPARSEMEM
983 depends on !NEED_PER_CPU_KM
984 depends on 64BIT
985 depends on !KMSAN
986 select PADATA
987 help
988 Ordinarily all struct pages are initialised during early boot in a
989 single thread. On very large machines this can take a considerable
990 amount of time. If this option is set, large machines will bring up
991 a subset of memmap at boot and then initialise the rest in parallel.
992 This has a potential performance impact on tasks running early in the
993 lifetime of the system until these kthreads finish the
994 initialisation.
995
996 config PAGE_IDLE_FLAG
997 bool
998 select PAGE_EXTENSION if !64BIT
999 help
1000 This adds PG_idle and PG_young flags to 'struct page'. PTE Accessed
1001 bit writers can set the state of the bit in the flags so that PTE
1002 Accessed bit readers may avoid disturbance.
1003
1004 config IDLE_PAGE_TRACKING
1005 bool "Enable idle page tracking"
1006 depends on SYSFS && MMU
1007 select PAGE_IDLE_FLAG
1008 help
1009 This feature allows to estimate the amount of user pages that have
1010 not been touched during a given period of time. This information can
1011 be useful to tune memory cgroup limits and/or for job placement
1012 within a compute cluster.
1013
1014 See Documentation/admin-guide/mm/idle_page_tracking.rst for
1015 more details.
1016
1017 # Architectures which implement cpu_dcache_is_aliasing() to query
1018 # whether the data caches are aliased (VIVT or VIPT with dcache
1019 # aliasing) need to select this.
1020 config ARCH_HAS_CPU_CACHE_ALIASING
1021 bool
1022
1023 config ARCH_HAS_CACHE_LINE_SIZE
1024 bool
1025
1026 config ARCH_HAS_CURRENT_STACK_POINTER
1027 bool
1028 help
1029 In support of HARDENED_USERCOPY performing stack variable lifetime
1030 checking, an architecture-agnostic way to find the stack pointer
1031 is needed. Once an architecture defines an unsigned long global
1032 register alias named "current_stack_pointer", this config can be
1033 selected.
1034
1035 config ARCH_HAS_PTE_DEVMAP
1036 bool
1037
1038 config ARCH_HAS_ZONE_DMA_SET
1039 bool
1040
1041 config ZONE_DMA
1042 bool "Support DMA zone" if ARCH_HAS_ZONE_DMA_SET
1043 default y if ARM64 || X86
1044
1045 config ZONE_DMA32
1046 bool "Support DMA32 zone" if ARCH_HAS_ZONE_DMA_SET
1047 depends on !X86_32
1048 default y if ARM64
1049
1050 config ZONE_DEVICE
1051 bool "Device memory (pmem, HMM, etc...) hotplug support"
1052 depends on MEMORY_HOTPLUG
1053 depends on MEMORY_HOTREMOVE
1054 depends on SPARSEMEM_VMEMMAP
1055 depends on ARCH_HAS_PTE_DEVMAP
1056 select XARRAY_MULTI
1057
1058 help
1059 Device memory hotplug support allows for establishing pmem,
1060 or other device driver discovered memory regions, in the
1061 memmap. This allows pfn_to_page() lookups of otherwise
1062 "device-physical" addresses which is needed for using a DAX
1063 mapping in an O_DIRECT operation, among other things.
1064
1065 If FS_DAX is enabled, then say Y.
1066
1067 #
1068 # Helpers to mirror range of the CPU page tables of a process into device page
1069 # tables.
1070 #
1071 config HMM_MIRROR
1072 bool
1073 depends on MMU
1074
1075 config GET_FREE_REGION
1076 depends on SPARSEMEM
1077 bool
1078
1079 config DEVICE_PRIVATE
1080 bool "Unaddressable device memory (GPU memory, ...)"
1081 depends on ZONE_DEVICE
1082 select GET_FREE_REGION
1083
1084 help
1085 Allows creation of struct pages to represent unaddressable device
1086 memory; i.e., memory that is only accessible from the device (or
1087 group of devices). You likely also want to select HMM_MIRROR.
1088
1089 config VMAP_PFN
1090 bool
1091
1092 config ARCH_USES_HIGH_VMA_FLAGS
1093 bool
1094 config ARCH_HAS_PKEYS
1095 bool
1096
1097 config ARCH_USES_PG_ARCH_X
1098 bool
1099 help
1100 Enable the definition of PG_arch_x page flags with x > 1. Only
1101 suitable for 64-bit architectures with CONFIG_FLATMEM or
1102 CONFIG_SPARSEMEM_VMEMMAP enabled, otherwise there may not be
1103 enough room for additional bits in page->flags.
1104
1105 config VM_EVENT_COUNTERS
1106 default y
1107 bool "Enable VM event counters for /proc/vmstat" if EXPERT
1108 help
1109 VM event counters are needed for event counts to be shown.
1110 This option allows the disabling of the VM event counters
1111 on EXPERT systems. /proc/vmstat will only show page counts
1112 if VM event counters are disabled.
1113
1114 config PERCPU_STATS
1115 bool "Collect percpu memory statistics"
1116 help
1117 This feature collects and exposes statistics via debugfs. The
1118 information includes global and per chunk statistics, which can
1119 be used to help understand percpu memory usage.
1120
1121 config GUP_TEST
1122 bool "Enable infrastructure for get_user_pages()-related unit tests"
1123 depends on DEBUG_FS
1124 help
1125 Provides /sys/kernel/debug/gup_test, which in turn provides a way
1126 to make ioctl calls that can launch kernel-based unit tests for
1127 the get_user_pages*() and pin_user_pages*() family of API calls.
1128
1129 These tests include benchmark testing of the _fast variants of
1130 get_user_pages*() and pin_user_pages*(), as well as smoke tests of
1131 the non-_fast variants.
1132
1133 There is also a sub-test that allows running dump_page() on any
1134 of up to eight pages (selected by command line args) within the
1135 range of user-space addresses. These pages are either pinned via
1136 pin_user_pages*(), or pinned via get_user_pages*(), as specified
1137 by other command line arguments.
1138
1139 See tools/testing/selftests/mm/gup_test.c
1140
1141 comment "GUP_TEST needs to have DEBUG_FS enabled"
1142 depends on !GUP_TEST && !DEBUG_FS
1143
1144 config GUP_GET_PXX_LOW_HIGH
1145 bool
1146
1147 config DMAPOOL_TEST
1148 tristate "Enable a module to run time tests on dma_pool"
1149 depends on HAS_DMA
1150 help
1151 Provides a test module that will allocate and free many blocks of
1152 various sizes and report how long it takes. This is intended to
1153 provide a consistent way to measure how changes to the
1154 dma_pool_alloc/free routines affect performance.
1155
1156 config ARCH_HAS_PTE_SPECIAL
1157 bool
1158
1159 config MAPPING_DIRTY_HELPERS
1160 bool
1161
1162 config KMAP_LOCAL
1163 bool
1164
1165 config KMAP_LOCAL_NON_LINEAR_PTE_ARRAY
1166 bool
1167
1168 # struct io_mapping based helper. Selected by drivers that need them
1169 config IO_MAPPING
1170 bool
1171
1172 config MEMFD_CREATE
1173 bool "Enable memfd_create() system call" if EXPERT
1174
1175 config SECRETMEM
1176 default y
1177 bool "Enable memfd_secret() system call" if EXPERT
1178 depends on ARCH_HAS_SET_DIRECT_MAP
1179 help
1180 Enable the memfd_secret() system call with the ability to create
1181 memory areas visible only in the context of the owning process and
1182 not mapped to other processes and other kernel page tables.
1183
1184 config ANON_VMA_NAME
1185 bool "Anonymous VMA name support"
1186 depends on PROC_FS && ADVISE_SYSCALLS && MMU
1187
1188 help
1189 Allow naming anonymous virtual memory areas.
1190
1191 This feature allows assigning names to virtual memory areas. Assigned
1192 names can be later retrieved from /proc/pid/maps and /proc/pid/smaps
1193 and help identifying individual anonymous memory areas.
1194 Assigning a name to anonymous virtual memory area might prevent that
1195 area from being merged with adjacent virtual memory areas due to the
1196 difference in their name.
1197
1198 config HAVE_ARCH_USERFAULTFD_WP
1199 bool
1200 help
1201 Arch has userfaultfd write protection support
1202
1203 config HAVE_ARCH_USERFAULTFD_MINOR
1204 bool
1205 help
1206 Arch has userfaultfd minor fault support
1207
1208 menuconfig USERFAULTFD
1209 bool "Enable userfaultfd() system call"
1210 depends on MMU
1211 help
1212 Enable the userfaultfd() system call that allows to intercept and
1213 handle page faults in userland.
1214
1215 if USERFAULTFD
1216 config PTE_MARKER_UFFD_WP
1217 bool "Userfaultfd write protection support for shmem/hugetlbfs"
1218 default y
1219 depends on HAVE_ARCH_USERFAULTFD_WP
1220
1221 help
1222 Allows to create marker PTEs for userfaultfd write protection
1223 purposes. It is required to enable userfaultfd write protection on
1224 file-backed memory types like shmem and hugetlbfs.
1225 endif # USERFAULTFD
1226
1227 # multi-gen LRU {
1228 config LRU_GEN
1229 bool "Multi-Gen LRU"
1230 depends on MMU
1231 # make sure folio->flags has enough spare bits
1232 depends on 64BIT || !SPARSEMEM || SPARSEMEM_VMEMMAP
1233 help
1234 A high performance LRU implementation to overcommit memory. See
1235 Documentation/admin-guide/mm/multigen_lru.rst for details.
1236
1237 config LRU_GEN_ENABLED
1238 bool "Enable by default"
1239 depends on LRU_GEN
1240 help
1241 This option enables the multi-gen LRU by default.
1242
1243 config LRU_GEN_STATS
1244 bool "Full stats for debugging"
1245 depends on LRU_GEN
1246 help
1247 Do not enable this option unless you plan to look at historical stats
1248 from evicted generations for debugging purpose.
1249
1250 This option has a per-memcg and per-node memory overhead.
1251
1252 config LRU_GEN_WALKS_MMU
1253 def_bool y
1254 depends on LRU_GEN && ARCH_HAS_HW_PTE_YOUNG
1255 # }
1256
1257 config ARCH_SUPPORTS_PER_VMA_LOCK
1258 def_bool n
1259
1260 config PER_VMA_LOCK
1261 def_bool y
1262 depends on ARCH_SUPPORTS_PER_VMA_LOCK && MMU && SMP
1263 help
1264 Allow per-vma locking during page fault handling.
1265
1266 This feature allows locking each virtual memory area separately when
1267 handling page faults instead of taking mmap_lock.
1268
1269 config LOCK_MM_AND_FIND_VMA
1270 bool
1271 depends on !STACK_GROWSUP
1272
1273 config IOMMU_MM_DATA
1274 bool
1275
1276 config EXECMEM
1277 bool
1278
1279 source "mm/damon/Kconfig"
1280
1281 endmenu
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.