1# SPDX-License-Identifier: GPL-2.0-only 2 3menu "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# 9config ARCH_NO_SWAP 10 bool 11 12config ZPOOL 13 bool 14 15menuconfig 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 25config 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 38config CMA_REUSE 39 bool "CMA reuse feature" 40 depends on CMA 41 help 42 If enabled, it will add MIGRATE_CMA to pcp lists and movable 43 allocations with __GFP_CMA flag will use cma areas prior to 44 movable areas. 45 46 It improves the utilization ratio of cma areas. 47 48config ZSWAP_DEFAULT_ON 49 bool "Enable the compressed cache for swap pages by default" 50 depends on ZSWAP 51 help 52 If selected, the compressed cache for swap pages will be enabled 53 at boot, otherwise it will be disabled. 54 55 The selection made here can be overridden by using the kernel 56 command line 'zswap.enabled=' option. 57 58config ZSWAP_EXCLUSIVE_LOADS_DEFAULT_ON 59 bool "Invalidate zswap entries when pages are loaded" 60 depends on ZSWAP 61 help 62 If selected, exclusive loads for zswap will be enabled at boot, 63 otherwise it will be disabled. 64 65 If exclusive loads are enabled, when a page is loaded from zswap, 66 the zswap entry is invalidated at once, as opposed to leaving it 67 in zswap until the swap entry is freed. 68 69 This avoids having two copies of the same page in memory 70 (compressed and uncompressed) after faulting in a page from zswap. 71 The cost is that if the page was never dirtied and needs to be 72 swapped out again, it will be re-compressed. 73 74choice 75 prompt "Default compressor" 76 depends on ZSWAP 77 default ZSWAP_COMPRESSOR_DEFAULT_LZO 78 help 79 Selects the default compression algorithm for the compressed cache 80 for swap pages. 81 82 For an overview what kind of performance can be expected from 83 a particular compression algorithm please refer to the benchmarks 84 available at the following LWN page: 85 https://lwn.net/Articles/751795/ 86 87 If in doubt, select 'LZO'. 88 89 The selection made here can be overridden by using the kernel 90 command line 'zswap.compressor=' option. 91 92config ZSWAP_COMPRESSOR_DEFAULT_DEFLATE 93 bool "Deflate" 94 select CRYPTO_DEFLATE 95 help 96 Use the Deflate algorithm as the default compression algorithm. 97 98config ZSWAP_COMPRESSOR_DEFAULT_LZO 99 bool "LZO" 100 select CRYPTO_LZO 101 help 102 Use the LZO algorithm as the default compression algorithm. 103 104config ZSWAP_COMPRESSOR_DEFAULT_842 105 bool "842" 106 select CRYPTO_842 107 help 108 Use the 842 algorithm as the default compression algorithm. 109 110config ZSWAP_COMPRESSOR_DEFAULT_LZ4 111 bool "LZ4" 112 select CRYPTO_LZ4 113 help 114 Use the LZ4 algorithm as the default compression algorithm. 115 116config ZSWAP_COMPRESSOR_DEFAULT_LZ4HC 117 bool "LZ4HC" 118 select CRYPTO_LZ4HC 119 help 120 Use the LZ4HC algorithm as the default compression algorithm. 121 122config ZSWAP_COMPRESSOR_DEFAULT_ZSTD 123 bool "zstd" 124 select CRYPTO_ZSTD 125 help 126 Use the zstd algorithm as the default compression algorithm. 127endchoice 128 129config ZSWAP_COMPRESSOR_DEFAULT 130 string 131 depends on ZSWAP 132 default "deflate" if ZSWAP_COMPRESSOR_DEFAULT_DEFLATE 133 default "lzo" if ZSWAP_COMPRESSOR_DEFAULT_LZO 134 default "842" if ZSWAP_COMPRESSOR_DEFAULT_842 135 default "lz4" if ZSWAP_COMPRESSOR_DEFAULT_LZ4 136 default "lz4hc" if ZSWAP_COMPRESSOR_DEFAULT_LZ4HC 137 default "zstd" if ZSWAP_COMPRESSOR_DEFAULT_ZSTD 138 default "" 139 140choice 141 prompt "Default allocator" 142 depends on ZSWAP 143 default ZSWAP_ZPOOL_DEFAULT_ZBUD 144 help 145 Selects the default allocator for the compressed cache for 146 swap pages. 147 The default is 'zbud' for compatibility, however please do 148 read the description of each of the allocators below before 149 making a right choice. 150 151 The selection made here can be overridden by using the kernel 152 command line 'zswap.zpool=' option. 153 154config ZSWAP_ZPOOL_DEFAULT_ZBUD 155 bool "zbud" 156 select ZBUD 157 help 158 Use the zbud allocator as the default allocator. 159 160config ZSWAP_ZPOOL_DEFAULT_Z3FOLD 161 bool "z3fold" 162 select Z3FOLD 163 help 164 Use the z3fold allocator as the default allocator. 165 166config ZSWAP_ZPOOL_DEFAULT_ZSMALLOC 167 bool "zsmalloc" 168 select ZSMALLOC 169 help 170 Use the zsmalloc allocator as the default allocator. 171endchoice 172 173config ZSWAP_ZPOOL_DEFAULT 174 string 175 depends on ZSWAP 176 default "zbud" if ZSWAP_ZPOOL_DEFAULT_ZBUD 177 default "z3fold" if ZSWAP_ZPOOL_DEFAULT_Z3FOLD 178 default "zsmalloc" if ZSWAP_ZPOOL_DEFAULT_ZSMALLOC 179 default "" 180 181config ZBUD 182 tristate "2:1 compression allocator (zbud)" 183 depends on ZSWAP 184 help 185 A special purpose allocator for storing compressed pages. 186 It is designed to store up to two compressed pages per physical 187 page. While this design limits storage density, it has simple and 188 deterministic reclaim properties that make it preferable to a higher 189 density approach when reclaim will be used. 190 191config Z3FOLD 192 tristate "3:1 compression allocator (z3fold)" 193 depends on ZSWAP 194 help 195 A special purpose allocator for storing compressed pages. 196 It is designed to store up to three compressed pages per physical 197 page. It is a ZBUD derivative so the simplicity and determinism are 198 still there. 199 200config ZSMALLOC 201 tristate 202 prompt "N:1 compression allocator (zsmalloc)" if ZSWAP 203 depends on MMU 204 help 205 zsmalloc is a slab-based memory allocator designed to store 206 pages of various compression levels efficiently. It achieves 207 the highest storage density with the least amount of fragmentation. 208 209config ZSMALLOC_STAT 210 bool "Export zsmalloc statistics" 211 depends on ZSMALLOC 212 select DEBUG_FS 213 help 214 This option enables code in the zsmalloc to collect various 215 statistics about what's happening in zsmalloc and exports that 216 information to userspace via debugfs. 217 If unsure, say N. 218 219config ZSMALLOC_CHAIN_SIZE 220 int "Maximum number of physical pages per-zspage" 221 default 8 222 range 4 16 223 depends on ZSMALLOC 224 help 225 This option sets the upper limit on the number of physical pages 226 that a zmalloc page (zspage) can consist of. The optimal zspage 227 chain size is calculated for each size class during the 228 initialization of the pool. 229 230 Changing this option can alter the characteristics of size classes, 231 such as the number of pages per zspage and the number of objects 232 per zspage. This can also result in different configurations of 233 the pool, as zsmalloc merges size classes with similar 234 characteristics. 235 236 For more information, see zsmalloc documentation. 237 238menu "SLAB allocator options" 239 240choice 241 prompt "Choose SLAB allocator" 242 default SLUB 243 help 244 This option allows to select a slab allocator. 245 246config SLAB_DEPRECATED 247 bool "SLAB (DEPRECATED)" 248 depends on !PREEMPT_RT 249 help 250 Deprecated and scheduled for removal in a few cycles. Replaced by 251 SLUB. 252 253 If you cannot migrate to SLUB, please contact linux-mm@kvack.org 254 and the people listed in the SLAB ALLOCATOR section of MAINTAINERS 255 file, explaining why. 256 257 The regular slab allocator that is established and known to work 258 well in all environments. It organizes cache hot objects in 259 per cpu and per node queues. 260 261config SLUB 262 bool "SLUB (Unqueued Allocator)" 263 help 264 SLUB is a slab allocator that minimizes cache line usage 265 instead of managing queues of cached objects (SLAB approach). 266 Per cpu caching is realized using slabs of objects instead 267 of queues of objects. SLUB can use memory efficiently 268 and has enhanced diagnostics. SLUB is the default choice for 269 a slab allocator. 270 271endchoice 272 273config SLAB 274 bool 275 default y 276 depends on SLAB_DEPRECATED 277 278config SLUB_TINY 279 bool "Configure SLUB for minimal memory footprint" 280 depends on SLUB && EXPERT 281 select SLAB_MERGE_DEFAULT 282 help 283 Configures the SLUB allocator in a way to achieve minimal memory 284 footprint, sacrificing scalability, debugging and other features. 285 This is intended only for the smallest system that had used the 286 SLOB allocator and is not recommended for systems with more than 287 16MB RAM. 288 289 If unsure, say N. 290 291config SLAB_MERGE_DEFAULT 292 bool "Allow slab caches to be merged" 293 default y 294 depends on SLAB || SLUB 295 help 296 For reduced kernel memory fragmentation, slab caches can be 297 merged when they share the same size and other characteristics. 298 This carries a risk of kernel heap overflows being able to 299 overwrite objects from merged caches (and more easily control 300 cache layout), which makes such heap attacks easier to exploit 301 by attackers. By keeping caches unmerged, these kinds of exploits 302 can usually only damage objects in the same cache. To disable 303 merging at runtime, "slab_nomerge" can be passed on the kernel 304 command line. 305 306config SLAB_FREELIST_RANDOM 307 bool "Randomize slab freelist" 308 depends on SLAB || (SLUB && !SLUB_TINY) 309 help 310 Randomizes the freelist order used on creating new pages. This 311 security feature reduces the predictability of the kernel slab 312 allocator against heap overflows. 313 314config SLAB_FREELIST_HARDENED 315 bool "Harden slab freelist metadata" 316 depends on SLAB || (SLUB && !SLUB_TINY) 317 help 318 Many kernel heap attacks try to target slab cache metadata and 319 other infrastructure. This options makes minor performance 320 sacrifices to harden the kernel slab allocator against common 321 freelist exploit methods. Some slab implementations have more 322 sanity-checking than others. This option is most effective with 323 CONFIG_SLUB. 324 325config SLUB_STATS 326 default n 327 bool "Enable SLUB performance statistics" 328 depends on SLUB && SYSFS && !SLUB_TINY 329 help 330 SLUB statistics are useful to debug SLUBs allocation behavior in 331 order find ways to optimize the allocator. This should never be 332 enabled for production use since keeping statistics slows down 333 the allocator by a few percentage points. The slabinfo command 334 supports the determination of the most active slabs to figure 335 out which slabs are relevant to a particular load. 336 Try running: slabinfo -DA 337 338config SLUB_CPU_PARTIAL 339 default y 340 depends on SLUB && SMP && !SLUB_TINY 341 bool "SLUB per cpu partial cache" 342 help 343 Per cpu partial caches accelerate objects allocation and freeing 344 that is local to a processor at the price of more indeterminism 345 in the latency of the free. On overflow these caches will be cleared 346 which requires the taking of locks that may cause latency spikes. 347 Typically one would choose no for a realtime system. 348 349config RANDOM_KMALLOC_CACHES 350 default n 351 depends on SLUB && !SLUB_TINY 352 bool "Randomize slab caches for normal kmalloc" 353 help 354 A hardening feature that creates multiple copies of slab caches for 355 normal kmalloc allocation and makes kmalloc randomly pick one based 356 on code address, which makes the attackers more difficult to spray 357 vulnerable memory objects on the heap for the purpose of exploiting 358 memory vulnerabilities. 359 360 Currently the number of copies is set to 16, a reasonably large value 361 that effectively diverges the memory objects allocated for different 362 subsystems or modules into different caches, at the expense of a 363 limited degree of memory and CPU overhead that relates to hardware and 364 system workload. 365 366endmenu # SLAB allocator options 367 368config SHUFFLE_PAGE_ALLOCATOR 369 bool "Page allocator randomization" 370 default SLAB_FREELIST_RANDOM && ACPI_NUMA 371 help 372 Randomization of the page allocator improves the average 373 utilization of a direct-mapped memory-side-cache. See section 374 5.2.27 Heterogeneous Memory Attribute Table (HMAT) in the ACPI 375 6.2a specification for an example of how a platform advertises 376 the presence of a memory-side-cache. There are also incidental 377 security benefits as it reduces the predictability of page 378 allocations to compliment SLAB_FREELIST_RANDOM, but the 379 default granularity of shuffling on the MAX_ORDER i.e, 10th 380 order of pages is selected based on cache utilization benefits 381 on x86. 382 383 While the randomization improves cache utilization it may 384 negatively impact workloads on platforms without a cache. For 385 this reason, by default, the randomization is enabled only 386 after runtime detection of a direct-mapped memory-side-cache. 387 Otherwise, the randomization may be force enabled with the 388 'page_alloc.shuffle' kernel command line parameter. 389 390 Say Y if unsure. 391 392config COMPAT_BRK 393 bool "Disable heap randomization" 394 default y 395 help 396 Randomizing heap placement makes heap exploits harder, but it 397 also breaks ancient binaries (including anything libc5 based). 398 This option changes the bootup default to heap randomization 399 disabled, and can be overridden at runtime by setting 400 /proc/sys/kernel/randomize_va_space to 2. 401 402 On non-ancient distros (post-2000 ones) N is usually a safe choice. 403 404config MMAP_ALLOW_UNINITIALIZED 405 bool "Allow mmapped anonymous memory to be uninitialized" 406 depends on EXPERT && !MMU 407 default n 408 help 409 Normally, and according to the Linux spec, anonymous memory obtained 410 from mmap() has its contents cleared before it is passed to 411 userspace. Enabling this config option allows you to request that 412 mmap() skip that if it is given an MAP_UNINITIALIZED flag, thus 413 providing a huge performance boost. If this option is not enabled, 414 then the flag will be ignored. 415 416 This is taken advantage of by uClibc's malloc(), and also by 417 ELF-FDPIC binfmt's brk and stack allocator. 418 419 Because of the obvious security issues, this option should only be 420 enabled on embedded devices where you control what is run in 421 userspace. Since that isn't generally a problem on no-MMU systems, 422 it is normally safe to say Y here. 423 424 See Documentation/admin-guide/mm/nommu-mmap.rst for more information. 425 426config SELECT_MEMORY_MODEL 427 def_bool y 428 depends on ARCH_SELECT_MEMORY_MODEL 429 430choice 431 prompt "Memory model" 432 depends on SELECT_MEMORY_MODEL 433 default SPARSEMEM_MANUAL if ARCH_SPARSEMEM_DEFAULT 434 default FLATMEM_MANUAL 435 help 436 This option allows you to change some of the ways that 437 Linux manages its memory internally. Most users will 438 only have one option here selected by the architecture 439 configuration. This is normal. 440 441config FLATMEM_MANUAL 442 bool "Flat Memory" 443 depends on !ARCH_SPARSEMEM_ENABLE || ARCH_FLATMEM_ENABLE 444 help 445 This option is best suited for non-NUMA systems with 446 flat address space. The FLATMEM is the most efficient 447 system in terms of performance and resource consumption 448 and it is the best option for smaller systems. 449 450 For systems that have holes in their physical address 451 spaces and for features like NUMA and memory hotplug, 452 choose "Sparse Memory". 453 454 If unsure, choose this option (Flat Memory) over any other. 455 456config SPARSEMEM_MANUAL 457 bool "Sparse Memory" 458 depends on ARCH_SPARSEMEM_ENABLE 459 help 460 This will be the only option for some systems, including 461 memory hot-plug systems. This is normal. 462 463 This option provides efficient support for systems with 464 holes is their physical address space and allows memory 465 hot-plug and hot-remove. 466 467 If unsure, choose "Flat Memory" over this option. 468 469endchoice 470 471config MEMORY_MONITOR 472 bool "ENABLE MEMORY_MONITOR" 473 depends on PROC_FS 474 default n 475 help 476 MEMORY_MONITOR is a monitor of some memory reclaim method. 477 Now, kswapd wake up monitor use it. 478 479config HYPERHOLD_FILE_LRU 480 bool "Enable HyperHold FILE LRU" 481 depends on HYPERHOLD && MEMCG 482 select HYPERHOLD_MEMCG 483 default n 484 help 485 File-LRU is a mechanism that put file page in global lru list, 486 and anon page in memcg lru list(if MEMCG is enable), what's 487 more, recliam of anonymous pages and file page are separated. 488 489config HYPERHOLD_MEMCG 490 bool "Enable Memcg Management in HyperHold" 491 depends on HYPERHOLD && MEMCG 492 help 493 Add more attributes in memory cgroup, these attribute is used 494 to show information, shrink memory, swapin page and so on. 495 496config HYPERHOLD_ZSWAPD 497 bool "Enable zswapd thread to reclaim anon pages in background" 498 depends on HYPERHOLD && ZRAM 499 default n 500 help 501 zswapd is a kernel thread that reclaim anonymous pages in the 502 background. When the use of swap pages reaches the watermark 503 and the refault of anonymous pages is high, the content of 504 zram will exchanged to eswap by a certain percentage. 505 506config SPARSEMEM 507 def_bool y 508 depends on (!SELECT_MEMORY_MODEL && ARCH_SPARSEMEM_ENABLE) || SPARSEMEM_MANUAL 509 510config FLATMEM 511 def_bool y 512 depends on !SPARSEMEM || FLATMEM_MANUAL 513 514# 515# SPARSEMEM_EXTREME (which is the default) does some bootmem 516# allocations when sparse_init() is called. If this cannot 517# be done on your architecture, select this option. However, 518# statically allocating the mem_section[] array can potentially 519# consume vast quantities of .bss, so be careful. 520# 521# This option will also potentially produce smaller runtime code 522# with gcc 3.4 and later. 523# 524config SPARSEMEM_STATIC 525 bool 526 527# 528# Architecture platforms which require a two level mem_section in SPARSEMEM 529# must select this option. This is usually for architecture platforms with 530# an extremely sparse physical address space. 531# 532config SPARSEMEM_EXTREME 533 def_bool y 534 depends on SPARSEMEM && !SPARSEMEM_STATIC 535 536config SPARSEMEM_VMEMMAP_ENABLE 537 bool 538 539config SPARSEMEM_VMEMMAP 540 bool "Sparse Memory virtual memmap" 541 depends on SPARSEMEM && SPARSEMEM_VMEMMAP_ENABLE 542 default y 543 help 544 SPARSEMEM_VMEMMAP uses a virtually mapped memmap to optimise 545 pfn_to_page and page_to_pfn operations. This is the most 546 efficient option when sufficient kernel resources are available. 547# 548# Select this config option from the architecture Kconfig, if it is preferred 549# to enable the feature of HugeTLB/dev_dax vmemmap optimization. 550# 551config ARCH_WANT_OPTIMIZE_DAX_VMEMMAP 552 bool 553 554config ARCH_WANT_OPTIMIZE_HUGETLB_VMEMMAP 555 bool 556 557config HAVE_MEMBLOCK_PHYS_MAP 558 bool 559 560config HAVE_FAST_GUP 561 depends on MMU 562 bool 563 564# Don't discard allocated memory used to track "memory" and "reserved" memblocks 565# after early boot, so it can still be used to test for validity of memory. 566# Also, memblocks are updated with memory hot(un)plug. 567config ARCH_KEEP_MEMBLOCK 568 bool 569 570# Keep arch NUMA mapping infrastructure post-init. 571config NUMA_KEEP_MEMINFO 572 bool 573 574config MEMORY_ISOLATION 575 bool 576 577# IORESOURCE_SYSTEM_RAM regions in the kernel resource tree that are marked 578# IORESOURCE_EXCLUSIVE cannot be mapped to user space, for example, via 579# /dev/mem. 580config EXCLUSIVE_SYSTEM_RAM 581 def_bool y 582 depends on !DEVMEM || STRICT_DEVMEM 583 584# 585# Only be set on architectures that have completely implemented memory hotplug 586# feature. If you are not sure, don't touch it. 587# 588config HAVE_BOOTMEM_INFO_NODE 589 def_bool n 590 591config ARCH_ENABLE_MEMORY_HOTPLUG 592 bool 593 594config ARCH_ENABLE_MEMORY_HOTREMOVE 595 bool 596 597# eventually, we can have this option just 'select SPARSEMEM' 598menuconfig MEMORY_HOTPLUG 599 bool "Memory hotplug" 600 select MEMORY_ISOLATION 601 depends on SPARSEMEM 602 depends on ARCH_ENABLE_MEMORY_HOTPLUG 603 depends on 64BIT 604 select NUMA_KEEP_MEMINFO if NUMA 605 606if MEMORY_HOTPLUG 607 608config MEMORY_HOTPLUG_DEFAULT_ONLINE 609 bool "Online the newly added memory blocks by default" 610 depends on MEMORY_HOTPLUG 611 help 612 This option sets the default policy setting for memory hotplug 613 onlining policy (/sys/devices/system/memory/auto_online_blocks) which 614 determines what happens to newly added memory regions. Policy setting 615 can always be changed at runtime. 616 See Documentation/admin-guide/mm/memory-hotplug.rst for more information. 617 618 Say Y here if you want all hot-plugged memory blocks to appear in 619 'online' state by default. 620 Say N here if you want the default policy to keep all hot-plugged 621 memory blocks in 'offline' state. 622 623config MEMORY_HOTREMOVE 624 bool "Allow for memory hot remove" 625 select HAVE_BOOTMEM_INFO_NODE if (X86_64 || PPC64) 626 depends on MEMORY_HOTPLUG && ARCH_ENABLE_MEMORY_HOTREMOVE 627 depends on MIGRATION 628 629config MHP_MEMMAP_ON_MEMORY 630 def_bool y 631 depends on MEMORY_HOTPLUG && SPARSEMEM_VMEMMAP 632 depends on ARCH_MHP_MEMMAP_ON_MEMORY_ENABLE 633 634endif # MEMORY_HOTPLUG 635 636config ARCH_MHP_MEMMAP_ON_MEMORY_ENABLE 637 bool 638 639# Heavily threaded applications may benefit from splitting the mm-wide 640# page_table_lock, so that faults on different parts of the user address 641# space can be handled with less contention: split it at this NR_CPUS. 642# Default to 4 for wider testing, though 8 might be more appropriate. 643# ARM's adjust_pte (unused if VIPT) depends on mm-wide page_table_lock. 644# PA-RISC 7xxx's spinlock_t would enlarge struct page from 32 to 44 bytes. 645# SPARC32 allocates multiple pte tables within a single page, and therefore 646# a per-page lock leads to problems when multiple tables need to be locked 647# at the same time (e.g. copy_page_range()). 648# DEBUG_SPINLOCK and DEBUG_LOCK_ALLOC spinlock_t also enlarge struct page. 649# 650config SPLIT_PTLOCK_CPUS 651 int 652 default "999999" if !MMU 653 default "999999" if ARM && !CPU_CACHE_VIPT 654 default "999999" if PARISC && !PA20 655 default "999999" if SPARC32 656 default "4" 657 658config ARCH_ENABLE_SPLIT_PMD_PTLOCK 659 bool 660 661# 662# support for memory balloon 663config MEMORY_BALLOON 664 bool 665 666# 667# support for memory balloon compaction 668config BALLOON_COMPACTION 669 bool "Allow for balloon memory compaction/migration" 670 def_bool y 671 depends on COMPACTION && MEMORY_BALLOON 672 help 673 Memory fragmentation introduced by ballooning might reduce 674 significantly the number of 2MB contiguous memory blocks that can be 675 used within a guest, thus imposing performance penalties associated 676 with the reduced number of transparent huge pages that could be used 677 by the guest workload. Allowing the compaction & migration for memory 678 pages enlisted as being part of memory balloon devices avoids the 679 scenario aforementioned and helps improving memory defragmentation. 680 681# 682# support for memory compaction 683config COMPACTION 684 bool "Allow for memory compaction" 685 def_bool y 686 select MIGRATION 687 depends on MMU 688 help 689 Compaction is the only memory management component to form 690 high order (larger physically contiguous) memory blocks 691 reliably. The page allocator relies on compaction heavily and 692 the lack of the feature can lead to unexpected OOM killer 693 invocations for high order memory requests. You shouldn't 694 disable this option unless there really is a strong reason for 695 it and then we would be really interested to hear about that at 696 linux-mm@kvack.org. 697 698config COMPACT_UNEVICTABLE_DEFAULT 699 int 700 depends on COMPACTION 701 default 0 if PREEMPT_RT 702 default 1 703 704# 705# support for free page reporting 706config PAGE_REPORTING 707 bool "Free page reporting" 708 def_bool n 709 help 710 Free page reporting allows for the incremental acquisition of 711 free pages from the buddy allocator for the purpose of reporting 712 those pages to another entity, such as a hypervisor, so that the 713 memory can be freed within the host for other uses. 714 715# 716# support for page migration 717# 718config MIGRATION 719 bool "Page migration" 720 def_bool y 721 depends on (NUMA || ARCH_ENABLE_MEMORY_HOTREMOVE || COMPACTION || CMA) && MMU 722 help 723 Allows the migration of the physical location of pages of processes 724 while the virtual addresses are not changed. This is useful in 725 two situations. The first is on NUMA systems to put pages nearer 726 to the processors accessing. The second is when allocating huge 727 pages as migration can relocate pages to satisfy a huge page 728 allocation instead of reclaiming. 729 730config DEVICE_MIGRATION 731 def_bool MIGRATION && ZONE_DEVICE 732 733config ARCH_ENABLE_HUGEPAGE_MIGRATION 734 bool 735 736config ARCH_ENABLE_THP_MIGRATION 737 bool 738 739config HUGETLB_PAGE_SIZE_VARIABLE 740 def_bool n 741 help 742 Allows the pageblock_order value to be dynamic instead of just standard 743 HUGETLB_PAGE_ORDER when there are multiple HugeTLB page sizes available 744 on a platform. 745 746 Note that the pageblock_order cannot exceed MAX_ORDER and will be 747 clamped down to MAX_ORDER. 748 749config CONTIG_ALLOC 750 def_bool (MEMORY_ISOLATION && COMPACTION) || CMA 751 752config PHYS_ADDR_T_64BIT 753 def_bool 64BIT 754 755config BOUNCE 756 bool "Enable bounce buffers" 757 default y 758 depends on BLOCK && MMU && HIGHMEM 759 help 760 Enable bounce buffers for devices that cannot access the full range of 761 memory available to the CPU. Enabled by default when HIGHMEM is 762 selected, but you may say n to override this. 763 764config MMU_NOTIFIER 765 bool 766 select INTERVAL_TREE 767 768config KSM 769 bool "Enable KSM for page merging" 770 depends on MMU 771 select XXHASH 772 help 773 Enable Kernel Samepage Merging: KSM periodically scans those areas 774 of an application's address space that an app has advised may be 775 mergeable. When it finds pages of identical content, it replaces 776 the many instances by a single page with that content, so 777 saving memory until one or another app needs to modify the content. 778 Recommended for use with KVM, or with other duplicative applications. 779 See Documentation/mm/ksm.rst for more information: KSM is inactive 780 until a program has madvised that an area is MADV_MERGEABLE, and 781 root has set /sys/kernel/mm/ksm/run to 1 (if CONFIG_SYSFS is set). 782 783config DEFAULT_MMAP_MIN_ADDR 784 int "Low address space to protect from user allocation" 785 depends on MMU 786 default 4096 787 help 788 This is the portion of low virtual memory which should be protected 789 from userspace allocation. Keeping a user from writing to low pages 790 can help reduce the impact of kernel NULL pointer bugs. 791 792 For most ia64, ppc64 and x86 users with lots of address space 793 a value of 65536 is reasonable and should cause no problems. 794 On arm and other archs it should not be higher than 32768. 795 Programs which use vm86 functionality or have some need to map 796 this low address space will need CAP_SYS_RAWIO or disable this 797 protection by setting the value to 0. 798 799 This value can be changed after boot using the 800 /proc/sys/vm/mmap_min_addr tunable. 801 802config ARCH_SUPPORTS_MEMORY_FAILURE 803 bool 804 805config MEMORY_FAILURE 806 depends on MMU 807 depends on ARCH_SUPPORTS_MEMORY_FAILURE 808 bool "Enable recovery from hardware memory errors" 809 select MEMORY_ISOLATION 810 select RAS 811 help 812 Enables code to recover from some memory failures on systems 813 with MCA recovery. This allows a system to continue running 814 even when some of its memory has uncorrected errors. This requires 815 special hardware support and typically ECC memory. 816 817config HWPOISON_INJECT 818 tristate "HWPoison pages injector" 819 depends on MEMORY_FAILURE && DEBUG_KERNEL && PROC_FS 820 select PROC_PAGE_MONITOR 821 822config NOMMU_INITIAL_TRIM_EXCESS 823 int "Turn on mmap() excess space trimming before booting" 824 depends on !MMU 825 default 1 826 help 827 The NOMMU mmap() frequently needs to allocate large contiguous chunks 828 of memory on which to store mappings, but it can only ask the system 829 allocator for chunks in 2^N*PAGE_SIZE amounts - which is frequently 830 more than it requires. To deal with this, mmap() is able to trim off 831 the excess and return it to the allocator. 832 833 If trimming is enabled, the excess is trimmed off and returned to the 834 system allocator, which can cause extra fragmentation, particularly 835 if there are a lot of transient processes. 836 837 If trimming is disabled, the excess is kept, but not used, which for 838 long-term mappings means that the space is wasted. 839 840 Trimming can be dynamically controlled through a sysctl option 841 (/proc/sys/vm/nr_trim_pages) which specifies the minimum number of 842 excess pages there must be before trimming should occur, or zero if 843 no trimming is to occur. 844 845 This option specifies the initial value of this option. The default 846 of 1 says that all excess pages should be trimmed. 847 848 See Documentation/admin-guide/mm/nommu-mmap.rst for more information. 849 850config ARCH_WANT_GENERAL_HUGETLB 851 bool 852 853config ARCH_WANTS_THP_SWAP 854 def_bool n 855 856menuconfig TRANSPARENT_HUGEPAGE 857 bool "Transparent Hugepage Support" 858 depends on HAVE_ARCH_TRANSPARENT_HUGEPAGE && !PREEMPT_RT 859 select COMPACTION 860 select XARRAY_MULTI 861 help 862 Transparent Hugepages allows the kernel to use huge pages and 863 huge tlb transparently to the applications whenever possible. 864 This feature can improve computing performance to certain 865 applications by speeding up page faults during memory 866 allocation, by reducing the number of tlb misses and by speeding 867 up the pagetable walking. 868 869 If memory constrained on embedded, you may want to say N. 870 871if TRANSPARENT_HUGEPAGE 872 873choice 874 prompt "Transparent Hugepage Support sysfs defaults" 875 depends on TRANSPARENT_HUGEPAGE 876 default TRANSPARENT_HUGEPAGE_ALWAYS 877 help 878 Selects the sysfs defaults for Transparent Hugepage Support. 879 880 config TRANSPARENT_HUGEPAGE_ALWAYS 881 bool "always" 882 help 883 Enabling Transparent Hugepage always, can increase the 884 memory footprint of applications without a guaranteed 885 benefit but it will work automatically for all applications. 886 887 config TRANSPARENT_HUGEPAGE_MADVISE 888 bool "madvise" 889 help 890 Enabling Transparent Hugepage madvise, will only provide a 891 performance improvement benefit to the applications using 892 madvise(MADV_HUGEPAGE) but it won't risk to increase the 893 memory footprint of applications without a guaranteed 894 benefit. 895endchoice 896 897config THP_SWAP 898 def_bool y 899 depends on TRANSPARENT_HUGEPAGE && ARCH_WANTS_THP_SWAP && SWAP && 64BIT 900 help 901 Swap transparent huge pages in one piece, without splitting. 902 XXX: For now, swap cluster backing transparent huge page 903 will be split after swapout. 904 905 For selection by architectures with reasonable THP sizes. 906 907config READ_ONLY_THP_FOR_FS 908 bool "Read-only THP for filesystems (EXPERIMENTAL)" 909 depends on TRANSPARENT_HUGEPAGE && SHMEM 910 911 help 912 Allow khugepaged to put read-only file-backed pages in THP. 913 914 This is marked experimental because it is a new feature. Write 915 support of file THPs will be developed in the next few release 916 cycles. 917 918endif # TRANSPARENT_HUGEPAGE 919 920# 921# UP and nommu archs use km based percpu allocator 922# 923config NEED_PER_CPU_KM 924 depends on !SMP || !MMU 925 bool 926 default y 927 928config NEED_PER_CPU_EMBED_FIRST_CHUNK 929 bool 930 931config NEED_PER_CPU_PAGE_FIRST_CHUNK 932 bool 933 934config USE_PERCPU_NUMA_NODE_ID 935 bool 936 937config HAVE_SETUP_PER_CPU_AREA 938 bool 939 940config CMA 941 bool "Contiguous Memory Allocator" 942 depends on MMU 943 select MIGRATION 944 select MEMORY_ISOLATION 945 help 946 This enables the Contiguous Memory Allocator which allows other 947 subsystems to allocate big physically-contiguous blocks of memory. 948 CMA reserves a region of memory and allows only movable pages to 949 be allocated from it. This way, the kernel can use the memory for 950 pagecache and when a subsystem requests for contiguous area, the 951 allocated pages are migrated away to serve the contiguous request. 952 953 If unsure, say "n". 954 955config CMA_DEBUG 956 bool "CMA debug messages (DEVELOPMENT)" 957 depends on DEBUG_KERNEL && CMA 958 help 959 Turns on debug messages in CMA. This produces KERN_DEBUG 960 messages for every CMA call as well as various messages while 961 processing calls such as dma_alloc_from_contiguous(). 962 This option does not affect warning and error messages. 963 964config CMA_DEBUGFS 965 bool "CMA debugfs interface" 966 depends on CMA && DEBUG_FS 967 help 968 Turns on the DebugFS interface for CMA. 969 970config CMA_SYSFS 971 bool "CMA information through sysfs interface" 972 depends on CMA && SYSFS 973 help 974 This option exposes some sysfs attributes to get information 975 from CMA. 976 977config CMA_AREAS 978 int "Maximum count of the CMA areas" 979 depends on CMA 980 default 19 if NUMA 981 default 7 982 help 983 CMA allows to create CMA areas for particular purpose, mainly, 984 used as device private area. This parameter sets the maximum 985 number of CMA area in the system. 986 987 If unsure, leave the default value "7" in UMA and "19" in NUMA. 988 989config MEM_SOFT_DIRTY 990 bool "Track memory changes" 991 depends on CHECKPOINT_RESTORE && HAVE_ARCH_SOFT_DIRTY && PROC_FS 992 select PROC_PAGE_MONITOR 993 help 994 This option enables memory changes tracking by introducing a 995 soft-dirty bit on pte-s. This bit it set when someone writes 996 into a page just as regular dirty bit, but unlike the latter 997 it can be cleared by hands. 998 999 See Documentation/admin-guide/mm/soft-dirty.rst for more details. 1000 1001config GENERIC_EARLY_IOREMAP 1002 bool 1003 1004config STACK_MAX_DEFAULT_SIZE_MB 1005 int "Default maximum user stack size for 32-bit processes (MB)" 1006 default 100 1007 range 8 2048 1008 depends on STACK_GROWSUP && (!64BIT || COMPAT) 1009 help 1010 This is the maximum stack size in Megabytes in the VM layout of 32-bit 1011 user processes when the stack grows upwards (currently only on parisc 1012 arch) when the RLIMIT_STACK hard limit is unlimited. 1013 1014 A sane initial value is 100 MB. 1015 1016config DEFERRED_STRUCT_PAGE_INIT 1017 bool "Defer initialisation of struct pages to kthreads" 1018 depends on SPARSEMEM 1019 depends on !NEED_PER_CPU_KM 1020 depends on 64BIT 1021 select PADATA 1022 help 1023 Ordinarily all struct pages are initialised during early boot in a 1024 single thread. On very large machines this can take a considerable 1025 amount of time. If this option is set, large machines will bring up 1026 a subset of memmap at boot and then initialise the rest in parallel. 1027 This has a potential performance impact on tasks running early in the 1028 lifetime of the system until these kthreads finish the 1029 initialisation. 1030 1031config PAGE_IDLE_FLAG 1032 bool 1033 select PAGE_EXTENSION if !64BIT 1034 help 1035 This adds PG_idle and PG_young flags to 'struct page'. PTE Accessed 1036 bit writers can set the state of the bit in the flags so that PTE 1037 Accessed bit readers may avoid disturbance. 1038 1039config IDLE_PAGE_TRACKING 1040 bool "Enable idle page tracking" 1041 depends on SYSFS && MMU 1042 select PAGE_IDLE_FLAG 1043 help 1044 This feature allows to estimate the amount of user pages that have 1045 not been touched during a given period of time. This information can 1046 be useful to tune memory cgroup limits and/or for job placement 1047 within a compute cluster. 1048 1049 See Documentation/admin-guide/mm/idle_page_tracking.rst for 1050 more details. 1051 1052config ARCH_HAS_CACHE_LINE_SIZE 1053 bool 1054 1055config ARCH_HAS_CURRENT_STACK_POINTER 1056 bool 1057 help 1058 In support of HARDENED_USERCOPY performing stack variable lifetime 1059 checking, an architecture-agnostic way to find the stack pointer 1060 is needed. Once an architecture defines an unsigned long global 1061 register alias named "current_stack_pointer", this config can be 1062 selected. 1063 1064config ARCH_HAS_PTE_DEVMAP 1065 bool 1066 1067config ARCH_HAS_ZONE_DMA_SET 1068 bool 1069 1070config ZONE_DMA 1071 bool "Support DMA zone" if ARCH_HAS_ZONE_DMA_SET 1072 default y if ARM64 || X86 1073 1074config ZONE_DMA32 1075 bool "Support DMA32 zone" if ARCH_HAS_ZONE_DMA_SET 1076 depends on !X86_32 1077 default y if ARM64 1078 1079config ZONE_DEVICE 1080 bool "Device memory (pmem, HMM, etc...) hotplug support" 1081 depends on MEMORY_HOTPLUG 1082 depends on MEMORY_HOTREMOVE 1083 depends on SPARSEMEM_VMEMMAP 1084 depends on ARCH_HAS_PTE_DEVMAP 1085 select XARRAY_MULTI 1086 1087 help 1088 Device memory hotplug support allows for establishing pmem, 1089 or other device driver discovered memory regions, in the 1090 memmap. This allows pfn_to_page() lookups of otherwise 1091 "device-physical" addresses which is needed for using a DAX 1092 mapping in an O_DIRECT operation, among other things. 1093 1094 If FS_DAX is enabled, then say Y. 1095 1096# 1097# Helpers to mirror range of the CPU page tables of a process into device page 1098# tables. 1099# 1100config HMM_MIRROR 1101 bool 1102 depends on MMU 1103 1104config GET_FREE_REGION 1105 depends on SPARSEMEM 1106 bool 1107 1108config DEVICE_PRIVATE 1109 bool "Unaddressable device memory (GPU memory, ...)" 1110 depends on ZONE_DEVICE 1111 select GET_FREE_REGION 1112 1113 help 1114 Allows creation of struct pages to represent unaddressable device 1115 memory; i.e., memory that is only accessible from the device (or 1116 group of devices). You likely also want to select HMM_MIRROR. 1117 1118config VMAP_PFN 1119 bool 1120 1121config ARCH_USES_HIGH_VMA_FLAGS 1122 bool 1123config ARCH_HAS_PKEYS 1124 bool 1125 1126config ARCH_USES_PG_ARCH_X 1127 bool 1128 help 1129 Enable the definition of PG_arch_x page flags with x > 1. Only 1130 suitable for 64-bit architectures with CONFIG_FLATMEM or 1131 CONFIG_SPARSEMEM_VMEMMAP enabled, otherwise there may not be 1132 enough room for additional bits in page->flags. 1133 1134config VM_EVENT_COUNTERS 1135 default y 1136 bool "Enable VM event counters for /proc/vmstat" if EXPERT 1137 help 1138 VM event counters are needed for event counts to be shown. 1139 This option allows the disabling of the VM event counters 1140 on EXPERT systems. /proc/vmstat will only show page counts 1141 if VM event counters are disabled. 1142 1143config PERCPU_STATS 1144 bool "Collect percpu memory statistics" 1145 help 1146 This feature collects and exposes statistics via debugfs. The 1147 information includes global and per chunk statistics, which can 1148 be used to help understand percpu memory usage. 1149 1150config GUP_TEST 1151 bool "Enable infrastructure for get_user_pages()-related unit tests" 1152 depends on DEBUG_FS 1153 help 1154 Provides /sys/kernel/debug/gup_test, which in turn provides a way 1155 to make ioctl calls that can launch kernel-based unit tests for 1156 the get_user_pages*() and pin_user_pages*() family of API calls. 1157 1158 These tests include benchmark testing of the _fast variants of 1159 get_user_pages*() and pin_user_pages*(), as well as smoke tests of 1160 the non-_fast variants. 1161 1162 There is also a sub-test that allows running dump_page() on any 1163 of up to eight pages (selected by command line args) within the 1164 range of user-space addresses. These pages are either pinned via 1165 pin_user_pages*(), or pinned via get_user_pages*(), as specified 1166 by other command line arguments. 1167 1168 See tools/testing/selftests/mm/gup_test.c 1169 1170comment "GUP_TEST needs to have DEBUG_FS enabled" 1171 depends on !GUP_TEST && !DEBUG_FS 1172 1173config GUP_GET_PXX_LOW_HIGH 1174 bool 1175 1176config DMAPOOL_TEST 1177 tristate "Enable a module to run time tests on dma_pool" 1178 depends on HAS_DMA 1179 help 1180 Provides a test module that will allocate and free many blocks of 1181 various sizes and report how long it takes. This is intended to 1182 provide a consistent way to measure how changes to the 1183 dma_pool_alloc/free routines affect performance. 1184 1185config ARCH_HAS_PTE_SPECIAL 1186 bool 1187 1188# 1189# Some architectures require a special hugepage directory format that is 1190# required to support multiple hugepage sizes. For example a4fe3ce76 1191# "powerpc/mm: Allow more flexible layouts for hugepage pagetables" 1192# introduced it on powerpc. This allows for a more flexible hugepage 1193# pagetable layouts. 1194# 1195config ARCH_HAS_HUGEPD 1196 bool 1197 1198config MAPPING_DIRTY_HELPERS 1199 bool 1200 1201config KMAP_LOCAL 1202 bool 1203 1204config KMAP_LOCAL_NON_LINEAR_PTE_ARRAY 1205 bool 1206 1207# struct io_mapping based helper. Selected by drivers that need them 1208config IO_MAPPING 1209 bool 1210 1211config MEMFD_CREATE 1212 bool "Enable memfd_create() system call" if EXPERT 1213 1214config SECRETMEM 1215 default y 1216 bool "Enable memfd_secret() system call" if EXPERT 1217 depends on ARCH_HAS_SET_DIRECT_MAP 1218 help 1219 Enable the memfd_secret() system call with the ability to create 1220 memory areas visible only in the context of the owning process and 1221 not mapped to other processes and other kernel page tables. 1222 1223config ANON_VMA_NAME 1224 bool "Anonymous VMA name support" 1225 depends on PROC_FS && ADVISE_SYSCALLS && MMU 1226 1227 help 1228 Allow naming anonymous virtual memory areas. 1229 1230 This feature allows assigning names to virtual memory areas. Assigned 1231 names can be later retrieved from /proc/pid/maps and /proc/pid/smaps 1232 and help identifying individual anonymous memory areas. 1233 Assigning a name to anonymous virtual memory area might prevent that 1234 area from being merged with adjacent virtual memory areas due to the 1235 difference in their name. 1236 1237config USERFAULTFD 1238 bool "Enable userfaultfd() system call" 1239 depends on MMU 1240 help 1241 Enable the userfaultfd() system call that allows to intercept and 1242 handle page faults in userland. 1243 1244config HAVE_ARCH_USERFAULTFD_WP 1245 bool 1246 help 1247 Arch has userfaultfd write protection support 1248 1249config HAVE_ARCH_USERFAULTFD_MINOR 1250 bool 1251 help 1252 Arch has userfaultfd minor fault support 1253 1254config PTE_MARKER_UFFD_WP 1255 bool "Userfaultfd write protection support for shmem/hugetlbfs" 1256 default y 1257 depends on HAVE_ARCH_USERFAULTFD_WP 1258 1259 help 1260 Allows to create marker PTEs for userfaultfd write protection 1261 purposes. It is required to enable userfaultfd write protection on 1262 file-backed memory types like shmem and hugetlbfs. 1263 1264# multi-gen LRU { 1265config LRU_GEN 1266 bool "Multi-Gen LRU" 1267 depends on MMU 1268 # make sure folio->flags has enough spare bits 1269 depends on 64BIT || !SPARSEMEM || SPARSEMEM_VMEMMAP 1270 help 1271 A high performance LRU implementation to overcommit memory. See 1272 Documentation/admin-guide/mm/multigen_lru.rst for details. 1273 1274config LRU_GEN_ENABLED 1275 bool "Enable by default" 1276 depends on LRU_GEN 1277 help 1278 This option enables the multi-gen LRU by default. 1279 1280config LRU_GEN_STATS 1281 bool "Full stats for debugging" 1282 depends on LRU_GEN 1283 help 1284 Do not enable this option unless you plan to look at historical stats 1285 from evicted generations for debugging purpose. 1286 1287 This option has a per-memcg and per-node memory overhead. 1288# } 1289 1290config ARCH_SUPPORTS_PER_VMA_LOCK 1291 def_bool n 1292 1293config PER_VMA_LOCK 1294 def_bool y 1295 depends on ARCH_SUPPORTS_PER_VMA_LOCK && MMU && SMP 1296 help 1297 Allow per-vma locking during page fault handling. 1298 1299 This feature allows locking each virtual memory area separately when 1300 handling page faults instead of taking mmap_lock. 1301 1302config LOCK_MM_AND_FIND_VMA 1303 bool 1304 depends on !STACK_GROWSUP 1305 1306 1307config MEM_PURGEABLE 1308 bool "Purgeable memory feature" 1309 default n 1310 depends on 64BIT 1311 select ARCH_USES_HIGH_VMA_FLAGS 1312 help 1313 Support purgeable pages for process 1314 1315config MEM_PURGEABLE_DEBUG 1316 bool "Purgeable memory debug" 1317 default n 1318 depends on MEM_PURGEABLE 1319 help 1320 Debug info for purgeable memory 1321 1322config PURGEABLE_ASHMEM 1323 bool "Purgeable memory feature for ashmem" 1324 default n 1325 depends on MEM_PURGEABLE 1326 help 1327 Support purgeable ashmem for process 1328 1329source "mm/damon/Kconfig" 1330 1331endmenu 1332