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 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 48config 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 62choice 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 80config ZSWAP_COMPRESSOR_DEFAULT_DEFLATE 81 bool "Deflate" 82 select CRYPTO_DEFLATE 83 help 84 Use the Deflate algorithm as the default compression algorithm. 85 86config ZSWAP_COMPRESSOR_DEFAULT_LZO 87 bool "LZO" 88 select CRYPTO_LZO 89 help 90 Use the LZO algorithm as the default compression algorithm. 91 92config ZSWAP_COMPRESSOR_DEFAULT_842 93 bool "842" 94 select CRYPTO_842 95 help 96 Use the 842 algorithm as the default compression algorithm. 97 98config ZSWAP_COMPRESSOR_DEFAULT_LZ4 99 bool "LZ4" 100 select CRYPTO_LZ4 101 help 102 Use the LZ4 algorithm as the default compression algorithm. 103 104config ZSWAP_COMPRESSOR_DEFAULT_LZ4HC 105 bool "LZ4HC" 106 select CRYPTO_LZ4HC 107 help 108 Use the LZ4HC algorithm as the default compression algorithm. 109 110config ZSWAP_COMPRESSOR_DEFAULT_ZSTD 111 bool "zstd" 112 select CRYPTO_ZSTD 113 help 114 Use the zstd algorithm as the default compression algorithm. 115endchoice 116 117config 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 128choice 129 prompt "Default allocator" 130 depends on ZSWAP 131 default ZSWAP_ZPOOL_DEFAULT_ZSMALLOC if MMU 132 help 133 Selects the default allocator for the compressed cache for 134 swap pages. 135 The default is 'zbud' for compatibility, however please do 136 read the description of each of the allocators below before 137 making a right choice. 138 139 The selection made here can be overridden by using the kernel 140 command line 'zswap.zpool=' option. 141 142config ZSWAP_ZPOOL_DEFAULT_ZSMALLOC 143 bool "zsmalloc" 144 select ZSMALLOC 145 help 146 Use the zsmalloc allocator as the default allocator. 147endchoice 148 149config ZSWAP_ZPOOL_DEFAULT 150 string 151 depends on ZSWAP 152 default "zsmalloc" if ZSWAP_ZPOOL_DEFAULT_ZSMALLOC 153 default "" 154 155config ZSMALLOC 156 tristate 157 prompt "N:1 compression allocator (zsmalloc)" if (ZSWAP || ZRAM) 158 depends on MMU 159 help 160 zsmalloc is a slab-based memory allocator designed to store 161 pages of various compression levels efficiently. It achieves 162 the highest storage density with the least amount of fragmentation. 163 164config ZSMALLOC_STAT 165 bool "Export zsmalloc statistics" 166 depends on ZSMALLOC 167 select DEBUG_FS 168 help 169 This option enables code in the zsmalloc to collect various 170 statistics about what's happening in zsmalloc and exports that 171 information to userspace via debugfs. 172 If unsure, say N. 173 174config ZSMALLOC_CHAIN_SIZE 175 int "Maximum number of physical pages per-zspage" 176 default 8 177 range 4 16 178 depends on ZSMALLOC 179 help 180 This option sets the upper limit on the number of physical pages 181 that a zmalloc page (zspage) can consist of. The optimal zspage 182 chain size is calculated for each size class during the 183 initialization of the pool. 184 185 Changing this option can alter the characteristics of size classes, 186 such as the number of pages per zspage and the number of objects 187 per zspage. This can also result in different configurations of 188 the pool, as zsmalloc merges size classes with similar 189 characteristics. 190 191 For more information, see zsmalloc documentation. 192 193menu "Slab allocator options" 194 195config SLUB 196 def_bool y 197 198config KVFREE_RCU_BATCHED 199 def_bool y 200 depends on !SLUB_TINY && !TINY_RCU 201 202config SLUB_TINY 203 bool "Configure for minimal memory footprint" 204 depends on EXPERT && !COMPILE_TEST 205 select SLAB_MERGE_DEFAULT 206 help 207 Configures the slab allocator in a way to achieve minimal memory 208 footprint, sacrificing scalability, debugging and other features. 209 This is intended only for the smallest system that had used the 210 SLOB allocator and is not recommended for systems with more than 211 16MB RAM. 212 213 If unsure, say N. 214 215config SLAB_MERGE_DEFAULT 216 bool "Allow slab caches to be merged" 217 default y 218 help 219 For reduced kernel memory fragmentation, slab caches can be 220 merged when they share the same size and other characteristics. 221 This carries a risk of kernel heap overflows being able to 222 overwrite objects from merged caches (and more easily control 223 cache layout), which makes such heap attacks easier to exploit 224 by attackers. By keeping caches unmerged, these kinds of exploits 225 can usually only damage objects in the same cache. To disable 226 merging at runtime, "slab_nomerge" can be passed on the kernel 227 command line. 228 229config SLAB_FREELIST_RANDOM 230 bool "Randomize slab freelist" 231 depends on !SLUB_TINY 232 help 233 Randomizes the freelist order used on creating new pages. This 234 security feature reduces the predictability of the kernel slab 235 allocator against heap overflows. 236 237config SLAB_FREELIST_HARDENED 238 bool "Harden slab freelist metadata" 239 depends on !SLUB_TINY 240 help 241 Many kernel heap attacks try to target slab cache metadata and 242 other infrastructure. This options makes minor performance 243 sacrifices to harden the kernel slab allocator against common 244 freelist exploit methods. 245 246config SLAB_BUCKETS 247 bool "Support allocation from separate kmalloc buckets" 248 depends on !SLUB_TINY 249 default SLAB_FREELIST_HARDENED 250 help 251 Kernel heap attacks frequently depend on being able to create 252 specifically-sized allocations with user-controlled contents 253 that will be allocated into the same kmalloc bucket as a 254 target object. To avoid sharing these allocation buckets, 255 provide an explicitly separated set of buckets to be used for 256 user-controlled allocations. This may very slightly increase 257 memory fragmentation, though in practice it's only a handful 258 of extra pages since the bulk of user-controlled allocations 259 are relatively long-lived. 260 261 If unsure, say Y. 262 263config SLUB_STATS 264 default n 265 bool "Enable performance statistics" 266 depends on SYSFS && !SLUB_TINY 267 help 268 The statistics are useful to debug slab allocation behavior in 269 order find ways to optimize the allocator. This should never be 270 enabled for production use since keeping statistics slows down 271 the allocator by a few percentage points. The slabinfo command 272 supports the determination of the most active slabs to figure 273 out which slabs are relevant to a particular load. 274 Try running: slabinfo -DA 275 276config SLUB_CPU_PARTIAL 277 default y 278 depends on SMP && !SLUB_TINY 279 bool "Enable per cpu partial caches" 280 help 281 Per cpu partial caches accelerate objects allocation and freeing 282 that is local to a processor at the price of more indeterminism 283 in the latency of the free. On overflow these caches will be cleared 284 which requires the taking of locks that may cause latency spikes. 285 Typically one would choose no for a realtime system. 286 287config RANDOM_KMALLOC_CACHES 288 default n 289 depends on !SLUB_TINY 290 bool "Randomize slab caches for normal kmalloc" 291 help 292 A hardening feature that creates multiple copies of slab caches for 293 normal kmalloc allocation and makes kmalloc randomly pick one based 294 on code address, which makes the attackers more difficult to spray 295 vulnerable memory objects on the heap for the purpose of exploiting 296 memory vulnerabilities. 297 298 Currently the number of copies is set to 16, a reasonably large value 299 that effectively diverges the memory objects allocated for different 300 subsystems or modules into different caches, at the expense of a 301 limited degree of memory and CPU overhead that relates to hardware and 302 system workload. 303 304endmenu # Slab allocator options 305 306config SHUFFLE_PAGE_ALLOCATOR 307 bool "Page allocator randomization" 308 default SLAB_FREELIST_RANDOM && ACPI_NUMA 309 help 310 Randomization of the page allocator improves the average 311 utilization of a direct-mapped memory-side-cache. See section 312 5.2.27 Heterogeneous Memory Attribute Table (HMAT) in the ACPI 313 6.2a specification for an example of how a platform advertises 314 the presence of a memory-side-cache. There are also incidental 315 security benefits as it reduces the predictability of page 316 allocations to compliment SLAB_FREELIST_RANDOM, but the 317 default granularity of shuffling on the MAX_PAGE_ORDER i.e, 10th 318 order of pages is selected based on cache utilization benefits 319 on x86. 320 321 While the randomization improves cache utilization it may 322 negatively impact workloads on platforms without a cache. For 323 this reason, by default, the randomization is not enabled even 324 if SHUFFLE_PAGE_ALLOCATOR=y. The randomization may be force enabled 325 with the 'page_alloc.shuffle' kernel command line parameter. 326 327 Say Y if unsure. 328 329config COMPAT_BRK 330 bool "Disable heap randomization" 331 default y 332 help 333 Randomizing heap placement makes heap exploits harder, but it 334 also breaks ancient binaries (including anything libc5 based). 335 This option changes the bootup default to heap randomization 336 disabled, and can be overridden at runtime by setting 337 /proc/sys/kernel/randomize_va_space to 2. 338 339 On non-ancient distros (post-2000 ones) N is usually a safe choice. 340 341config MMAP_ALLOW_UNINITIALIZED 342 bool "Allow mmapped anonymous memory to be uninitialized" 343 depends on EXPERT && !MMU 344 default n 345 help 346 Normally, and according to the Linux spec, anonymous memory obtained 347 from mmap() has its contents cleared before it is passed to 348 userspace. Enabling this config option allows you to request that 349 mmap() skip that if it is given an MAP_UNINITIALIZED flag, thus 350 providing a huge performance boost. If this option is not enabled, 351 then the flag will be ignored. 352 353 This is taken advantage of by uClibc's malloc(), and also by 354 ELF-FDPIC binfmt's brk and stack allocator. 355 356 Because of the obvious security issues, this option should only be 357 enabled on embedded devices where you control what is run in 358 userspace. Since that isn't generally a problem on no-MMU systems, 359 it is normally safe to say Y here. 360 361 See Documentation/admin-guide/mm/nommu-mmap.rst for more information. 362 363config SELECT_MEMORY_MODEL 364 def_bool y 365 depends on ARCH_SELECT_MEMORY_MODEL 366 367choice 368 prompt "Memory model" 369 depends on SELECT_MEMORY_MODEL 370 default SPARSEMEM_MANUAL if ARCH_SPARSEMEM_DEFAULT 371 default FLATMEM_MANUAL 372 help 373 This option allows you to change some of the ways that 374 Linux manages its memory internally. Most users will 375 only have one option here selected by the architecture 376 configuration. This is normal. 377 378config FLATMEM_MANUAL 379 bool "Flat Memory" 380 depends on !ARCH_SPARSEMEM_ENABLE || ARCH_FLATMEM_ENABLE 381 help 382 This option is best suited for non-NUMA systems with 383 flat address space. The FLATMEM is the most efficient 384 system in terms of performance and resource consumption 385 and it is the best option for smaller systems. 386 387 For systems that have holes in their physical address 388 spaces and for features like NUMA and memory hotplug, 389 choose "Sparse Memory". 390 391 If unsure, choose this option (Flat Memory) over any other. 392 393config SPARSEMEM_MANUAL 394 bool "Sparse Memory" 395 depends on ARCH_SPARSEMEM_ENABLE 396 help 397 This will be the only option for some systems, including 398 memory hot-plug systems. This is normal. 399 400 This option provides efficient support for systems with 401 holes is their physical address space and allows memory 402 hot-plug and hot-remove. 403 404 If unsure, choose "Flat Memory" over this option. 405 406endchoice 407 408config SPARSEMEM 409 def_bool y 410 depends on (!SELECT_MEMORY_MODEL && ARCH_SPARSEMEM_ENABLE) || SPARSEMEM_MANUAL 411 412config FLATMEM 413 def_bool y 414 depends on !SPARSEMEM || FLATMEM_MANUAL 415 416# 417# SPARSEMEM_EXTREME (which is the default) does some bootmem 418# allocations when sparse_init() is called. If this cannot 419# be done on your architecture, select this option. However, 420# statically allocating the mem_section[] array can potentially 421# consume vast quantities of .bss, so be careful. 422# 423# This option will also potentially produce smaller runtime code 424# with gcc 3.4 and later. 425# 426config SPARSEMEM_STATIC 427 bool 428 429# 430# Architecture platforms which require a two level mem_section in SPARSEMEM 431# must select this option. This is usually for architecture platforms with 432# an extremely sparse physical address space. 433# 434config SPARSEMEM_EXTREME 435 def_bool y 436 depends on SPARSEMEM && !SPARSEMEM_STATIC 437 438config SPARSEMEM_VMEMMAP_ENABLE 439 bool 440 441config SPARSEMEM_VMEMMAP 442 bool "Sparse Memory virtual memmap" 443 depends on SPARSEMEM && SPARSEMEM_VMEMMAP_ENABLE 444 default y 445 help 446 SPARSEMEM_VMEMMAP uses a virtually mapped memmap to optimise 447 pfn_to_page and page_to_pfn operations. This is the most 448 efficient option when sufficient kernel resources are available. 449 450config SPARSEMEM_VMEMMAP_PREINIT 451 bool 452# 453# Select this config option from the architecture Kconfig, if it is preferred 454# to enable the feature of HugeTLB/dev_dax vmemmap optimization. 455# 456config ARCH_WANT_OPTIMIZE_DAX_VMEMMAP 457 bool 458 459config ARCH_WANT_OPTIMIZE_HUGETLB_VMEMMAP 460 bool 461 462config ARCH_WANT_HUGETLB_VMEMMAP_PREINIT 463 bool 464 465config HAVE_MEMBLOCK_PHYS_MAP 466 bool 467 468config HAVE_GUP_FAST 469 depends on MMU 470 bool 471 472# Don't discard allocated memory used to track "memory" and "reserved" memblocks 473# after early boot, so it can still be used to test for validity of memory. 474# Also, memblocks are updated with memory hot(un)plug. 475config ARCH_KEEP_MEMBLOCK 476 bool 477 478# Keep arch NUMA mapping infrastructure post-init. 479config NUMA_KEEP_MEMINFO 480 bool 481 482config MEMORY_ISOLATION 483 bool 484 485# IORESOURCE_SYSTEM_RAM regions in the kernel resource tree that are marked 486# IORESOURCE_EXCLUSIVE cannot be mapped to user space, for example, via 487# /dev/mem. 488config EXCLUSIVE_SYSTEM_RAM 489 def_bool y 490 depends on !DEVMEM || STRICT_DEVMEM 491 492# 493# Only be set on architectures that have completely implemented memory hotplug 494# feature. If you are not sure, don't touch it. 495# 496config HAVE_BOOTMEM_INFO_NODE 497 def_bool n 498 499config ARCH_ENABLE_MEMORY_HOTPLUG 500 bool 501 502config ARCH_ENABLE_MEMORY_HOTREMOVE 503 bool 504 505# eventually, we can have this option just 'select SPARSEMEM' 506menuconfig MEMORY_HOTPLUG 507 bool "Memory hotplug" 508 select MEMORY_ISOLATION 509 depends on SPARSEMEM 510 depends on ARCH_ENABLE_MEMORY_HOTPLUG 511 depends on 64BIT 512 select NUMA_KEEP_MEMINFO if NUMA 513 514if MEMORY_HOTPLUG 515 516choice 517 prompt "Memory Hotplug Default Online Type" 518 default MHP_DEFAULT_ONLINE_TYPE_OFFLINE 519 help 520 Default memory type for hotplugged memory. 521 522 This option sets the default policy setting for memory hotplug 523 onlining policy (/sys/devices/system/memory/auto_online_blocks) which 524 determines what happens to newly added memory regions. Policy setting 525 can always be changed at runtime. 526 527 The default is 'offline'. 528 529 Select offline to defer onlining to drivers and user policy. 530 Select auto to let the kernel choose what zones to utilize. 531 Select online_kernel to generally allow kernel usage of this memory. 532 Select online_movable to generally disallow kernel usage of this memory. 533 534 Example kernel usage would be page structs and page tables. 535 536 See Documentation/admin-guide/mm/memory-hotplug.rst for more information. 537 538config MHP_DEFAULT_ONLINE_TYPE_OFFLINE 539 bool "offline" 540 help 541 Hotplugged memory will not be onlined by default. 542 Choose this for systems with drivers and user policy that 543 handle onlining of hotplug memory policy. 544 545config MHP_DEFAULT_ONLINE_TYPE_ONLINE_AUTO 546 bool "auto" 547 help 548 Select this if you want the kernel to automatically online 549 hotplugged memory into the zone it thinks is reasonable. 550 This memory may be utilized for kernel data. 551 552config MHP_DEFAULT_ONLINE_TYPE_ONLINE_KERNEL 553 bool "kernel" 554 help 555 Select this if you want the kernel to automatically online 556 hotplugged memory into a zone capable of being used for kernel 557 data. This typically means ZONE_NORMAL. 558 559config MHP_DEFAULT_ONLINE_TYPE_ONLINE_MOVABLE 560 bool "movable" 561 help 562 Select this if you want the kernel to automatically online 563 hotplug memory into ZONE_MOVABLE. This memory will generally 564 not be utilized for kernel data. 565 566 This should only be used when the admin knows sufficient 567 ZONE_NORMAL memory is available to describe hotplug memory, 568 otherwise hotplug memory may fail to online. For example, 569 sufficient kernel-capable memory (ZONE_NORMAL) must be 570 available to allocate page structs to describe ZONE_MOVABLE. 571 572endchoice 573 574config 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 580config 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 585endif # MEMORY_HOTPLUG 586 587config 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# 601config SPLIT_PTE_PTLOCKS 602 def_bool y 603 depends on MMU 604 depends on SMP 605 depends on NR_CPUS >= 4 606 depends on !ARM || CPU_CACHE_VIPT 607 depends on !PARISC || PA20 608 depends on !SPARC32 609 610config ARCH_ENABLE_SPLIT_PMD_PTLOCK 611 bool 612 613config SPLIT_PMD_PTLOCKS 614 def_bool y 615 depends on SPLIT_PTE_PTLOCKS && ARCH_ENABLE_SPLIT_PMD_PTLOCK 616 617# 618# support for memory balloon 619config MEMORY_BALLOON 620 bool 621 622# 623# support for memory balloon compaction 624config BALLOON_COMPACTION 625 bool "Allow for balloon memory compaction/migration" 626 default y 627 depends on COMPACTION && MEMORY_BALLOON 628 help 629 Memory fragmentation introduced by ballooning might reduce 630 significantly the number of 2MB contiguous memory blocks that can be 631 used within a guest, thus imposing performance penalties associated 632 with the reduced number of transparent huge pages that could be used 633 by the guest workload. Allowing the compaction & migration for memory 634 pages enlisted as being part of memory balloon devices avoids the 635 scenario aforementioned and helps improving memory defragmentation. 636 637# 638# support for memory compaction 639config COMPACTION 640 bool "Allow for memory compaction" 641 default y 642 select MIGRATION 643 depends on MMU 644 help 645 Compaction is the only memory management component to form 646 high order (larger physically contiguous) memory blocks 647 reliably. The page allocator relies on compaction heavily and 648 the lack of the feature can lead to unexpected OOM killer 649 invocations for high order memory requests. You shouldn't 650 disable this option unless there really is a strong reason for 651 it and then we would be really interested to hear about that at 652 linux-mm@kvack.org. 653 654config COMPACT_UNEVICTABLE_DEFAULT 655 int 656 depends on COMPACTION 657 default 0 if PREEMPT_RT 658 default 1 659 660# 661# support for free page reporting 662config PAGE_REPORTING 663 bool "Free page reporting" 664 help 665 Free page reporting allows for the incremental acquisition of 666 free pages from the buddy allocator for the purpose of reporting 667 those pages to another entity, such as a hypervisor, so that the 668 memory can be freed within the host for other uses. 669 670# 671# support for page migration 672# 673config MIGRATION 674 bool "Page migration" 675 default y 676 depends on (NUMA || ARCH_ENABLE_MEMORY_HOTREMOVE || COMPACTION || CMA) && MMU 677 help 678 Allows the migration of the physical location of pages of processes 679 while the virtual addresses are not changed. This is useful in 680 two situations. The first is on NUMA systems to put pages nearer 681 to the processors accessing. The second is when allocating huge 682 pages as migration can relocate pages to satisfy a huge page 683 allocation instead of reclaiming. 684 685config DEVICE_MIGRATION 686 def_bool MIGRATION && ZONE_DEVICE 687 688config ARCH_ENABLE_HUGEPAGE_MIGRATION 689 bool 690 691config ARCH_ENABLE_THP_MIGRATION 692 bool 693 694config HUGETLB_PAGE_SIZE_VARIABLE 695 def_bool n 696 help 697 Allows the pageblock_order value to be dynamic instead of just standard 698 HUGETLB_PAGE_ORDER when there are multiple HugeTLB page sizes available 699 on a platform. 700 701 Note that the pageblock_order cannot exceed MAX_PAGE_ORDER and will be 702 clamped down to MAX_PAGE_ORDER. 703 704config CONTIG_ALLOC 705 def_bool (MEMORY_ISOLATION && COMPACTION) || CMA 706 707config PCP_BATCH_SCALE_MAX 708 int "Maximum scale factor of PCP (Per-CPU pageset) batch allocate/free" 709 default 5 710 range 0 6 711 help 712 In page allocator, PCP (Per-CPU pageset) is refilled and drained in 713 batches. The batch number is scaled automatically to improve page 714 allocation/free throughput. But too large scale factor may hurt 715 latency. This option sets the upper limit of scale factor to limit 716 the maximum latency. 717 718config PHYS_ADDR_T_64BIT 719 def_bool 64BIT 720 721config BOUNCE 722 bool "Enable bounce buffers" 723 default y 724 depends on BLOCK && MMU && HIGHMEM 725 help 726 Enable bounce buffers for devices that cannot access the full range of 727 memory available to the CPU. Enabled by default when HIGHMEM is 728 selected, but you may say n to override this. 729 730config MMU_NOTIFIER 731 bool 732 select INTERVAL_TREE 733 734config KSM 735 bool "Enable KSM for page merging" 736 depends on MMU 737 select XXHASH 738 help 739 Enable Kernel Samepage Merging: KSM periodically scans those areas 740 of an application's address space that an app has advised may be 741 mergeable. When it finds pages of identical content, it replaces 742 the many instances by a single page with that content, so 743 saving memory until one or another app needs to modify the content. 744 Recommended for use with KVM, or with other duplicative applications. 745 See Documentation/mm/ksm.rst for more information: KSM is inactive 746 until a program has madvised that an area is MADV_MERGEABLE, and 747 root has set /sys/kernel/mm/ksm/run to 1 (if CONFIG_SYSFS is set). 748 749config DEFAULT_MMAP_MIN_ADDR 750 int "Low address space to protect from user allocation" 751 depends on MMU 752 default 4096 753 help 754 This is the portion of low virtual memory which should be protected 755 from userspace allocation. Keeping a user from writing to low pages 756 can help reduce the impact of kernel NULL pointer bugs. 757 758 For most arm64, ppc64 and x86 users with lots of address space 759 a value of 65536 is reasonable and should cause no problems. 760 On arm and other archs it should not be higher than 32768. 761 Programs which use vm86 functionality or have some need to map 762 this low address space will need CAP_SYS_RAWIO or disable this 763 protection by setting the value to 0. 764 765 This value can be changed after boot using the 766 /proc/sys/vm/mmap_min_addr tunable. 767 768config ARCH_SUPPORTS_MEMORY_FAILURE 769 bool 770 771config MEMORY_FAILURE 772 depends on MMU 773 depends on ARCH_SUPPORTS_MEMORY_FAILURE 774 bool "Enable recovery from hardware memory errors" 775 select MEMORY_ISOLATION 776 select RAS 777 help 778 Enables code to recover from some memory failures on systems 779 with MCA recovery. This allows a system to continue running 780 even when some of its memory has uncorrected errors. This requires 781 special hardware support and typically ECC memory. 782 783config HWPOISON_INJECT 784 tristate "HWPoison pages injector" 785 depends on MEMORY_FAILURE && DEBUG_KERNEL && PROC_FS 786 select PROC_PAGE_MONITOR 787 788config NOMMU_INITIAL_TRIM_EXCESS 789 int "Turn on mmap() excess space trimming before booting" 790 depends on !MMU 791 default 1 792 help 793 The NOMMU mmap() frequently needs to allocate large contiguous chunks 794 of memory on which to store mappings, but it can only ask the system 795 allocator for chunks in 2^N*PAGE_SIZE amounts - which is frequently 796 more than it requires. To deal with this, mmap() is able to trim off 797 the excess and return it to the allocator. 798 799 If trimming is enabled, the excess is trimmed off and returned to the 800 system allocator, which can cause extra fragmentation, particularly 801 if there are a lot of transient processes. 802 803 If trimming is disabled, the excess is kept, but not used, which for 804 long-term mappings means that the space is wasted. 805 806 Trimming can be dynamically controlled through a sysctl option 807 (/proc/sys/vm/nr_trim_pages) which specifies the minimum number of 808 excess pages there must be before trimming should occur, or zero if 809 no trimming is to occur. 810 811 This option specifies the initial value of this option. The default 812 of 1 says that all excess pages should be trimmed. 813 814 See Documentation/admin-guide/mm/nommu-mmap.rst for more information. 815 816config ARCH_WANT_GENERAL_HUGETLB 817 bool 818 819config ARCH_WANTS_THP_SWAP 820 def_bool n 821 822config MM_ID 823 def_bool n 824 825menuconfig TRANSPARENT_HUGEPAGE 826 bool "Transparent Hugepage Support" 827 depends on HAVE_ARCH_TRANSPARENT_HUGEPAGE && !PREEMPT_RT 828 select COMPACTION 829 select XARRAY_MULTI 830 select MM_ID 831 help 832 Transparent Hugepages allows the kernel to use huge pages and 833 huge tlb transparently to the applications whenever possible. 834 This feature can improve computing performance to certain 835 applications by speeding up page faults during memory 836 allocation, by reducing the number of tlb misses and by speeding 837 up the pagetable walking. 838 839 If memory constrained on embedded, you may want to say N. 840 841if TRANSPARENT_HUGEPAGE 842 843choice 844 prompt "Transparent Hugepage Support sysfs defaults" 845 depends on TRANSPARENT_HUGEPAGE 846 default TRANSPARENT_HUGEPAGE_ALWAYS 847 help 848 Selects the sysfs defaults for Transparent Hugepage Support. 849 850 config TRANSPARENT_HUGEPAGE_ALWAYS 851 bool "always" 852 help 853 Enabling Transparent Hugepage always, can increase the 854 memory footprint of applications without a guaranteed 855 benefit but it will work automatically for all applications. 856 857 config TRANSPARENT_HUGEPAGE_MADVISE 858 bool "madvise" 859 help 860 Enabling Transparent Hugepage madvise, will only provide a 861 performance improvement benefit to the applications using 862 madvise(MADV_HUGEPAGE) but it won't risk to increase the 863 memory footprint of applications without a guaranteed 864 benefit. 865 866 config TRANSPARENT_HUGEPAGE_NEVER 867 bool "never" 868 help 869 Disable Transparent Hugepage by default. It can still be 870 enabled at runtime via sysfs. 871endchoice 872 873config THP_SWAP 874 def_bool y 875 depends on TRANSPARENT_HUGEPAGE && ARCH_WANTS_THP_SWAP && SWAP && 64BIT 876 help 877 Swap transparent huge pages in one piece, without splitting. 878 XXX: For now, swap cluster backing transparent huge page 879 will be split after swapout. 880 881 For selection by architectures with reasonable THP sizes. 882 883config READ_ONLY_THP_FOR_FS 884 bool "Read-only THP for filesystems (EXPERIMENTAL)" 885 depends on TRANSPARENT_HUGEPAGE && SHMEM 886 887 help 888 Allow khugepaged to put read-only file-backed pages in THP. 889 890 This is marked experimental because it is a new feature. Write 891 support of file THPs will be developed in the next few release 892 cycles. 893 894config NO_PAGE_MAPCOUNT 895 bool "No per-page mapcount (EXPERIMENTAL)" 896 help 897 Do not maintain per-page mapcounts for pages part of larger 898 allocations, such as transparent huge pages. 899 900 When this config option is enabled, some interfaces that relied on 901 this information will rely on less-precise per-allocation information 902 instead: for example, using the average per-page mapcount in such 903 a large allocation instead of the per-page mapcount. 904 905 EXPERIMENTAL because the impact of some changes is still unclear. 906 907endif # TRANSPARENT_HUGEPAGE 908 909# simple helper to make the code a bit easier to read 910config PAGE_MAPCOUNT 911 def_bool !NO_PAGE_MAPCOUNT 912 913# 914# The architecture supports pgtable leaves that is larger than PAGE_SIZE 915# 916config PGTABLE_HAS_HUGE_LEAVES 917 def_bool TRANSPARENT_HUGEPAGE || HUGETLB_PAGE 918 919# TODO: Allow to be enabled without THP 920config ARCH_SUPPORTS_HUGE_PFNMAP 921 def_bool n 922 depends on TRANSPARENT_HUGEPAGE 923 924config ARCH_SUPPORTS_PMD_PFNMAP 925 def_bool y 926 depends on ARCH_SUPPORTS_HUGE_PFNMAP && HAVE_ARCH_TRANSPARENT_HUGEPAGE 927 928config ARCH_SUPPORTS_PUD_PFNMAP 929 def_bool y 930 depends on ARCH_SUPPORTS_HUGE_PFNMAP && HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD 931 932# 933# UP and nommu archs use km based percpu allocator 934# 935config NEED_PER_CPU_KM 936 depends on !SMP || !MMU 937 bool 938 default y 939 940config NEED_PER_CPU_EMBED_FIRST_CHUNK 941 bool 942 943config NEED_PER_CPU_PAGE_FIRST_CHUNK 944 bool 945 946config USE_PERCPU_NUMA_NODE_ID 947 bool 948 949config HAVE_SETUP_PER_CPU_AREA 950 bool 951 952config CMA 953 bool "Contiguous Memory Allocator" 954 depends on MMU 955 select MIGRATION 956 select MEMORY_ISOLATION 957 help 958 This enables the Contiguous Memory Allocator which allows other 959 subsystems to allocate big physically-contiguous blocks of memory. 960 CMA reserves a region of memory and allows only movable pages to 961 be allocated from it. This way, the kernel can use the memory for 962 pagecache and when a subsystem requests for contiguous area, the 963 allocated pages are migrated away to serve the contiguous request. 964 965 If unsure, say "n". 966 967config CMA_DEBUGFS 968 bool "CMA debugfs interface" 969 depends on CMA && DEBUG_FS 970 help 971 Turns on the DebugFS interface for CMA. 972 973config CMA_SYSFS 974 bool "CMA information through sysfs interface" 975 depends on CMA && SYSFS 976 help 977 This option exposes some sysfs attributes to get information 978 from CMA. 979 980config CMA_AREAS 981 int "Maximum count of the CMA areas" 982 depends on CMA 983 default 20 if NUMA 984 default 8 985 help 986 CMA allows to create CMA areas for particular purpose, mainly, 987 used as device private area. This parameter sets the maximum 988 number of CMA area in the system. 989 990 If unsure, leave the default value "8" in UMA and "20" in NUMA. 991 992config MEM_SOFT_DIRTY 993 bool "Track memory changes" 994 depends on CHECKPOINT_RESTORE && HAVE_ARCH_SOFT_DIRTY && PROC_FS 995 select PROC_PAGE_MONITOR 996 help 997 This option enables memory changes tracking by introducing a 998 soft-dirty bit on pte-s. This bit it set when someone writes 999 into a page just as regular dirty bit, but unlike the latter 1000 it can be cleared by hands. 1001 1002 See Documentation/admin-guide/mm/soft-dirty.rst for more details. 1003 1004config GENERIC_EARLY_IOREMAP 1005 bool 1006 1007config STACK_MAX_DEFAULT_SIZE_MB 1008 int "Default maximum user stack size for 32-bit processes (MB)" 1009 default 100 1010 range 8 2048 1011 depends on STACK_GROWSUP && (!64BIT || COMPAT) 1012 help 1013 This is the maximum stack size in Megabytes in the VM layout of 32-bit 1014 user processes when the stack grows upwards (currently only on parisc 1015 arch) when the RLIMIT_STACK hard limit is unlimited. 1016 1017 A sane initial value is 100 MB. 1018 1019config DEFERRED_STRUCT_PAGE_INIT 1020 bool "Defer initialisation of struct pages to kthreads" 1021 depends on SPARSEMEM 1022 depends on !NEED_PER_CPU_KM 1023 depends on 64BIT 1024 depends on !KMSAN 1025 select PADATA 1026 help 1027 Ordinarily all struct pages are initialised during early boot in a 1028 single thread. On very large machines this can take a considerable 1029 amount of time. If this option is set, large machines will bring up 1030 a subset of memmap at boot and then initialise the rest in parallel. 1031 This has a potential performance impact on tasks running early in the 1032 lifetime of the system until these kthreads finish the 1033 initialisation. 1034 1035config PAGE_IDLE_FLAG 1036 bool 1037 select PAGE_EXTENSION if !64BIT 1038 help 1039 This adds PG_idle and PG_young flags to 'struct page'. PTE Accessed 1040 bit writers can set the state of the bit in the flags so that PTE 1041 Accessed bit readers may avoid disturbance. 1042 1043config IDLE_PAGE_TRACKING 1044 bool "Enable idle page tracking" 1045 depends on SYSFS && MMU 1046 select PAGE_IDLE_FLAG 1047 help 1048 This feature allows to estimate the amount of user pages that have 1049 not been touched during a given period of time. This information can 1050 be useful to tune memory cgroup limits and/or for job placement 1051 within a compute cluster. 1052 1053 See Documentation/admin-guide/mm/idle_page_tracking.rst for 1054 more details. 1055 1056# Architectures which implement cpu_dcache_is_aliasing() to query 1057# whether the data caches are aliased (VIVT or VIPT with dcache 1058# aliasing) need to select this. 1059config ARCH_HAS_CPU_CACHE_ALIASING 1060 bool 1061 1062config ARCH_HAS_CACHE_LINE_SIZE 1063 bool 1064 1065config ARCH_HAS_CURRENT_STACK_POINTER 1066 bool 1067 help 1068 In support of HARDENED_USERCOPY performing stack variable lifetime 1069 checking, an architecture-agnostic way to find the stack pointer 1070 is needed. Once an architecture defines an unsigned long global 1071 register alias named "current_stack_pointer", this config can be 1072 selected. 1073 1074config ARCH_HAS_PTE_DEVMAP 1075 bool 1076 1077config ARCH_HAS_ZONE_DMA_SET 1078 bool 1079 1080config ZONE_DMA 1081 bool "Support DMA zone" if ARCH_HAS_ZONE_DMA_SET 1082 default y if ARM64 || X86 1083 1084config ZONE_DMA32 1085 bool "Support DMA32 zone" if ARCH_HAS_ZONE_DMA_SET 1086 depends on !X86_32 1087 default y if ARM64 1088 1089config ZONE_DEVICE 1090 bool "Device memory (pmem, HMM, etc...) hotplug support" 1091 depends on MEMORY_HOTPLUG 1092 depends on MEMORY_HOTREMOVE 1093 depends on SPARSEMEM_VMEMMAP 1094 depends on ARCH_HAS_PTE_DEVMAP 1095 select XARRAY_MULTI 1096 1097 help 1098 Device memory hotplug support allows for establishing pmem, 1099 or other device driver discovered memory regions, in the 1100 memmap. This allows pfn_to_page() lookups of otherwise 1101 "device-physical" addresses which is needed for using a DAX 1102 mapping in an O_DIRECT operation, among other things. 1103 1104 If FS_DAX is enabled, then say Y. 1105 1106# 1107# Helpers to mirror range of the CPU page tables of a process into device page 1108# tables. 1109# 1110config HMM_MIRROR 1111 bool 1112 depends on MMU 1113 1114config GET_FREE_REGION 1115 bool 1116 1117config DEVICE_PRIVATE 1118 bool "Unaddressable device memory (GPU memory, ...)" 1119 depends on ZONE_DEVICE 1120 select GET_FREE_REGION 1121 1122 help 1123 Allows creation of struct pages to represent unaddressable device 1124 memory; i.e., memory that is only accessible from the device (or 1125 group of devices). You likely also want to select HMM_MIRROR. 1126 1127config VMAP_PFN 1128 bool 1129 1130config ARCH_USES_HIGH_VMA_FLAGS 1131 bool 1132config ARCH_HAS_PKEYS 1133 bool 1134 1135config ARCH_USES_PG_ARCH_2 1136 bool 1137config ARCH_USES_PG_ARCH_3 1138 bool 1139 1140config VM_EVENT_COUNTERS 1141 default y 1142 bool "Enable VM event counters for /proc/vmstat" if EXPERT 1143 help 1144 VM event counters are needed for event counts to be shown. 1145 This option allows the disabling of the VM event counters 1146 on EXPERT systems. /proc/vmstat will only show page counts 1147 if VM event counters are disabled. 1148 1149config PERCPU_STATS 1150 bool "Collect percpu memory statistics" 1151 help 1152 This feature collects and exposes statistics via debugfs. The 1153 information includes global and per chunk statistics, which can 1154 be used to help understand percpu memory usage. 1155 1156config GUP_TEST 1157 bool "Enable infrastructure for get_user_pages()-related unit tests" 1158 depends on DEBUG_FS 1159 help 1160 Provides /sys/kernel/debug/gup_test, which in turn provides a way 1161 to make ioctl calls that can launch kernel-based unit tests for 1162 the get_user_pages*() and pin_user_pages*() family of API calls. 1163 1164 These tests include benchmark testing of the _fast variants of 1165 get_user_pages*() and pin_user_pages*(), as well as smoke tests of 1166 the non-_fast variants. 1167 1168 There is also a sub-test that allows running dump_page() on any 1169 of up to eight pages (selected by command line args) within the 1170 range of user-space addresses. These pages are either pinned via 1171 pin_user_pages*(), or pinned via get_user_pages*(), as specified 1172 by other command line arguments. 1173 1174 See tools/testing/selftests/mm/gup_test.c 1175 1176comment "GUP_TEST needs to have DEBUG_FS enabled" 1177 depends on !GUP_TEST && !DEBUG_FS 1178 1179config GUP_GET_PXX_LOW_HIGH 1180 bool 1181 1182config DMAPOOL_TEST 1183 tristate "Enable a module to run time tests on dma_pool" 1184 depends on HAS_DMA 1185 help 1186 Provides a test module that will allocate and free many blocks of 1187 various sizes and report how long it takes. This is intended to 1188 provide a consistent way to measure how changes to the 1189 dma_pool_alloc/free routines affect performance. 1190 1191config ARCH_HAS_PTE_SPECIAL 1192 bool 1193 1194config MAPPING_DIRTY_HELPERS 1195 bool 1196 1197config KMAP_LOCAL 1198 bool 1199 1200config KMAP_LOCAL_NON_LINEAR_PTE_ARRAY 1201 bool 1202 1203# struct io_mapping based helper. Selected by drivers that need them 1204config IO_MAPPING 1205 bool 1206 1207config MEMFD_CREATE 1208 bool "Enable memfd_create() system call" if EXPERT 1209 1210config SECRETMEM 1211 default y 1212 bool "Enable memfd_secret() system call" if EXPERT 1213 depends on ARCH_HAS_SET_DIRECT_MAP 1214 help 1215 Enable the memfd_secret() system call with the ability to create 1216 memory areas visible only in the context of the owning process and 1217 not mapped to other processes and other kernel page tables. 1218 1219config ANON_VMA_NAME 1220 bool "Anonymous VMA name support" 1221 depends on PROC_FS && ADVISE_SYSCALLS && MMU 1222 1223 help 1224 Allow naming anonymous virtual memory areas. 1225 1226 This feature allows assigning names to virtual memory areas. Assigned 1227 names can be later retrieved from /proc/pid/maps and /proc/pid/smaps 1228 and help identifying individual anonymous memory areas. 1229 Assigning a name to anonymous virtual memory area might prevent that 1230 area from being merged with adjacent virtual memory areas due to the 1231 difference in their name. 1232 1233config HAVE_ARCH_USERFAULTFD_WP 1234 bool 1235 help 1236 Arch has userfaultfd write protection support 1237 1238config HAVE_ARCH_USERFAULTFD_MINOR 1239 bool 1240 help 1241 Arch has userfaultfd minor fault support 1242 1243menuconfig USERFAULTFD 1244 bool "Enable userfaultfd() system call" 1245 depends on MMU 1246 help 1247 Enable the userfaultfd() system call that allows to intercept and 1248 handle page faults in userland. 1249 1250if USERFAULTFD 1251config PTE_MARKER_UFFD_WP 1252 bool "Userfaultfd write protection support for shmem/hugetlbfs" 1253 default y 1254 depends on HAVE_ARCH_USERFAULTFD_WP 1255 1256 help 1257 Allows to create marker PTEs for userfaultfd write protection 1258 purposes. It is required to enable userfaultfd write protection on 1259 file-backed memory types like shmem and hugetlbfs. 1260endif # USERFAULTFD 1261 1262# multi-gen LRU { 1263config LRU_GEN 1264 bool "Multi-Gen LRU" 1265 depends on MMU 1266 # make sure folio->flags has enough spare bits 1267 depends on 64BIT || !SPARSEMEM || SPARSEMEM_VMEMMAP 1268 help 1269 A high performance LRU implementation to overcommit memory. See 1270 Documentation/admin-guide/mm/multigen_lru.rst for details. 1271 1272config LRU_GEN_ENABLED 1273 bool "Enable by default" 1274 depends on LRU_GEN 1275 help 1276 This option enables the multi-gen LRU by default. 1277 1278config LRU_GEN_STATS 1279 bool "Full stats for debugging" 1280 depends on LRU_GEN 1281 help 1282 Do not enable this option unless you plan to look at historical stats 1283 from evicted generations for debugging purpose. 1284 1285 This option has a per-memcg and per-node memory overhead. 1286 1287config LRU_GEN_WALKS_MMU 1288 def_bool y 1289 depends on LRU_GEN && ARCH_HAS_HW_PTE_YOUNG 1290# } 1291 1292config ARCH_SUPPORTS_PER_VMA_LOCK 1293 def_bool n 1294 1295config PER_VMA_LOCK 1296 def_bool y 1297 depends on ARCH_SUPPORTS_PER_VMA_LOCK && MMU && SMP 1298 help 1299 Allow per-vma locking during page fault handling. 1300 1301 This feature allows locking each virtual memory area separately when 1302 handling page faults instead of taking mmap_lock. 1303 1304config LOCK_MM_AND_FIND_VMA 1305 bool 1306 depends on !STACK_GROWSUP 1307 1308config IOMMU_MM_DATA 1309 bool 1310 1311config EXECMEM 1312 bool 1313 1314config NUMA_MEMBLKS 1315 bool 1316 1317config NUMA_EMU 1318 bool "NUMA emulation" 1319 depends on NUMA_MEMBLKS 1320 help 1321 Enable NUMA emulation. A flat machine will be split 1322 into virtual nodes when booted with "numa=fake=N", where N is the 1323 number of nodes. This is only useful for debugging. 1324 1325config ARCH_HAS_USER_SHADOW_STACK 1326 bool 1327 help 1328 The architecture has hardware support for userspace shadow call 1329 stacks (eg, x86 CET, arm64 GCS or RISC-V Zicfiss). 1330 1331config ARCH_SUPPORTS_PT_RECLAIM 1332 def_bool n 1333 1334config PT_RECLAIM 1335 bool "reclaim empty user page table pages" 1336 default y 1337 depends on ARCH_SUPPORTS_PT_RECLAIM && MMU && SMP 1338 select MMU_GATHER_RCU_TABLE_FREE 1339 help 1340 Try to reclaim empty user page table pages in paths other than munmap 1341 and exit_mmap path. 1342 1343 Note: now only empty user PTE page table pages will be reclaimed. 1344 1345 1346source "mm/damon/Kconfig" 1347 1348endmenu 1349