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 12menuconfig SWAP 13 bool "Support for paging of anonymous memory (swap)" 14 depends on MMU && BLOCK && !ARCH_NO_SWAP 15 default y 16 help 17 This option allows you to choose whether you want to have support 18 for so called swap devices or swap files in your kernel that are 19 used to provide more virtual memory than the actual RAM present 20 in your computer. If unsure say Y. 21 22config ZSWAP 23 bool "Compressed cache for swap pages" 24 depends on SWAP 25 select CRYPTO 26 select ZSMALLOC 27 help 28 A lightweight compressed cache for swap pages. It takes 29 pages that are in the process of being swapped out and attempts to 30 compress them into a dynamically allocated RAM-based memory pool. 31 This can result in a significant I/O reduction on swap device and, 32 in the case where decompressing from RAM is faster than swap device 33 reads, can also improve workload performance. 34 35config ZSWAP_DEFAULT_ON 36 bool "Enable the compressed cache for swap pages by default" 37 depends on ZSWAP 38 help 39 If selected, the compressed cache for swap pages will be enabled 40 at boot, otherwise it will be disabled. 41 42 The selection made here can be overridden by using the kernel 43 command line 'zswap.enabled=' option. 44 45config ZSWAP_SHRINKER_DEFAULT_ON 46 bool "Shrink the zswap pool on memory pressure" 47 depends on ZSWAP 48 default n 49 help 50 If selected, the zswap shrinker will be enabled, and the pages 51 stored in the zswap pool will become available for reclaim (i.e 52 written back to the backing swap device) on memory pressure. 53 54 This means that zswap writeback could happen even if the pool is 55 not yet full, or the cgroup zswap limit has not been reached, 56 reducing the chance that cold pages will reside in the zswap pool 57 and consume memory indefinitely. 58 59choice 60 prompt "Default compressor" 61 depends on ZSWAP 62 default ZSWAP_COMPRESSOR_DEFAULT_LZO 63 help 64 Selects the default compression algorithm for the compressed cache 65 for swap pages. 66 67 For an overview what kind of performance can be expected from 68 a particular compression algorithm please refer to the benchmarks 69 available at the following LWN page: 70 https://lwn.net/Articles/751795/ 71 72 If in doubt, select 'LZO'. 73 74 The selection made here can be overridden by using the kernel 75 command line 'zswap.compressor=' option. 76 77config ZSWAP_COMPRESSOR_DEFAULT_DEFLATE 78 bool "Deflate" 79 select CRYPTO_DEFLATE 80 help 81 Use the Deflate algorithm as the default compression algorithm. 82 83config ZSWAP_COMPRESSOR_DEFAULT_LZO 84 bool "LZO" 85 select CRYPTO_LZO 86 help 87 Use the LZO algorithm as the default compression algorithm. 88 89config ZSWAP_COMPRESSOR_DEFAULT_842 90 bool "842" 91 select CRYPTO_842 92 help 93 Use the 842 algorithm as the default compression algorithm. 94 95config ZSWAP_COMPRESSOR_DEFAULT_LZ4 96 bool "LZ4" 97 select CRYPTO_LZ4 98 help 99 Use the LZ4 algorithm as the default compression algorithm. 100 101config ZSWAP_COMPRESSOR_DEFAULT_LZ4HC 102 bool "LZ4HC" 103 select CRYPTO_LZ4HC 104 help 105 Use the LZ4HC algorithm as the default compression algorithm. 106 107config ZSWAP_COMPRESSOR_DEFAULT_ZSTD 108 bool "zstd" 109 select CRYPTO_ZSTD 110 help 111 Use the zstd algorithm as the default compression algorithm. 112endchoice 113 114config ZSWAP_COMPRESSOR_DEFAULT 115 string 116 depends on ZSWAP 117 default "deflate" if ZSWAP_COMPRESSOR_DEFAULT_DEFLATE 118 default "lzo" if ZSWAP_COMPRESSOR_DEFAULT_LZO 119 default "842" if ZSWAP_COMPRESSOR_DEFAULT_842 120 default "lz4" if ZSWAP_COMPRESSOR_DEFAULT_LZ4 121 default "lz4hc" if ZSWAP_COMPRESSOR_DEFAULT_LZ4HC 122 default "zstd" if ZSWAP_COMPRESSOR_DEFAULT_ZSTD 123 default "" 124 125config ZSMALLOC 126 tristate 127 128if ZSMALLOC 129 130menu "Zsmalloc allocator options" 131 depends on ZSMALLOC 132 133comment "Zsmalloc is a common backend allocator for zswap & zram" 134 135config ZSMALLOC_STAT 136 bool "Export zsmalloc statistics" 137 select DEBUG_FS 138 help 139 This option enables code in the zsmalloc to collect various 140 statistics about what's happening in zsmalloc and exports that 141 information to userspace via debugfs. 142 If unsure, say N. 143 144config ZSMALLOC_CHAIN_SIZE 145 int "Maximum number of physical pages per-zspage" 146 default 8 147 range 4 16 148 help 149 This option sets the upper limit on the number of physical pages 150 that a zmalloc page (zspage) can consist of. The optimal zspage 151 chain size is calculated for each size class during the 152 initialization of the pool. 153 154 Changing this option can alter the characteristics of size classes, 155 such as the number of pages per zspage and the number of objects 156 per zspage. This can also result in different configurations of 157 the pool, as zsmalloc merges size classes with similar 158 characteristics. 159 160 For more information, see zsmalloc documentation. 161 162endmenu 163 164endif 165 166menu "Slab allocator options" 167 168config SLUB 169 def_bool y 170 select IRQ_WORK 171 172config KVFREE_RCU_BATCHED 173 def_bool y 174 depends on !SLUB_TINY && !TINY_RCU 175 depends on !RCU_STRICT_GRACE_PERIOD 176 177config SLUB_TINY 178 bool "Configure for minimal memory footprint" 179 depends on EXPERT && !COMPILE_TEST 180 select SLAB_MERGE_DEFAULT 181 help 182 Configures the slab allocator in a way to achieve minimal memory 183 footprint, sacrificing scalability, debugging and other features. 184 This is intended only for the smallest system that had used the 185 SLOB allocator and is not recommended for systems with more than 186 16MB RAM. 187 188 If unsure, say N. 189 190config SLAB_MERGE_DEFAULT 191 bool "Allow slab caches to be merged" 192 default y 193 help 194 For reduced kernel memory fragmentation, slab caches can be 195 merged when they share the same size and other characteristics. 196 This carries a risk of kernel heap overflows being able to 197 overwrite objects from merged caches (and more easily control 198 cache layout), which makes such heap attacks easier to exploit 199 by attackers. By keeping caches unmerged, these kinds of exploits 200 can usually only damage objects in the same cache. To disable 201 merging at runtime, "slab_nomerge" can be passed on the kernel 202 command line. 203 204config SLAB_FREELIST_RANDOM 205 bool "Randomize slab freelist" 206 depends on !SLUB_TINY 207 help 208 Randomizes the freelist order used on creating new pages. This 209 security feature reduces the predictability of the kernel slab 210 allocator against heap overflows. 211 212config SLAB_FREELIST_HARDENED 213 bool "Harden slab freelist metadata" 214 depends on !SLUB_TINY 215 help 216 Many kernel heap attacks try to target slab cache metadata and 217 other infrastructure. This options makes minor performance 218 sacrifices to harden the kernel slab allocator against common 219 freelist exploit methods. 220 221config SLAB_BUCKETS 222 bool "Support allocation from separate kmalloc buckets" 223 depends on !SLUB_TINY 224 default SLAB_FREELIST_HARDENED 225 help 226 Kernel heap attacks frequently depend on being able to create 227 specifically-sized allocations with user-controlled contents 228 that will be allocated into the same kmalloc bucket as a 229 target object. To avoid sharing these allocation buckets, 230 provide an explicitly separated set of buckets to be used for 231 user-controlled allocations. This may very slightly increase 232 memory fragmentation, though in practice it's only a handful 233 of extra pages since the bulk of user-controlled allocations 234 are relatively long-lived. 235 236 If unsure, say Y. 237 238config SLUB_STATS 239 default n 240 bool "Enable performance statistics" 241 depends on SYSFS && !SLUB_TINY 242 help 243 The statistics are useful to debug slab allocation behavior in 244 order find ways to optimize the allocator. This should never be 245 enabled for production use since keeping statistics slows down 246 the allocator by a few percentage points. The slabinfo command 247 supports the determination of the most active slabs to figure 248 out which slabs are relevant to a particular load. 249 Try running: slabinfo -DA 250 251config KMALLOC_PARTITION_CACHES 252 depends on !SLUB_TINY 253 bool "Partitioned slab caches for normal kmalloc" 254 default RANDOM_KMALLOC_CACHES 255 help 256 A hardening feature that creates multiple isolated copies of slab 257 caches for normal kmalloc allocations. This makes it more difficult 258 to exploit memory-safety vulnerabilities by attacking vulnerable 259 co-located memory objects. Several modes are provided. 260 261 Currently the number of copies is set to 16, a reasonably large value 262 that effectively diverges the memory objects allocated for different 263 subsystems or modules into different caches, at the expense of a 264 limited degree of memory and CPU overhead that relates to hardware 265 and system workload. 266 267choice 268 prompt "Partitioned slab cache mode" 269 depends on KMALLOC_PARTITION_CACHES 270 default KMALLOC_PARTITION_TYPED if CC_HAS_ALLOC_TOKEN 271 default KMALLOC_PARTITION_RANDOM 272 help 273 Selects the slab cache partitioning mode. 274 275config KMALLOC_PARTITION_RANDOM 276 bool "Randomize slab caches for normal kmalloc" 277 help 278 Randomly pick a slab cache based on code address and a per-boot 279 random seed. 280 281 This makes it harder for attackers to predict object co-location. 282 The placement is random: while attackers don't know which kmalloc 283 cache an object will be allocated from, they might circumvent 284 the randomization by retrying attacks across multiple machines until 285 the target objects are co-located. 286 287config KMALLOC_PARTITION_TYPED 288 bool "Type based slab cache selection for normal kmalloc" 289 depends on CC_HAS_ALLOC_TOKEN 290 help 291 Rely on Clang's allocation tokens to choose a slab cache, where token 292 IDs are derived from the allocated type. 293 294 Unlike KMALLOC_PARTITION_RANDOM, cache assignment is deterministic based 295 on type, which guarantees that objects of certain types are not 296 placed in the same cache. This effectively mitigates certain classes 297 of exploits that probabilistic defenses like KMALLOC_PARTITION_RANDOM 298 only make harder but not impossible. However, this also means the 299 cache assignment is predictable. 300 301 Clang's default token ID calculation returns a bounded hash with 302 disjoint ranges for pointer-containing and pointerless objects: when 303 used as the slab cache index, this prevents buffer overflows on 304 primitive buffers from directly corrupting pointer-containing 305 objects. 306 307 The current effectiveness of Clang's type inference can be judged by 308 -Rpass=alloc-token, which provides diagnostics where (after dead-code 309 elimination) type inference failed. 310 311 Requires Clang 22 or later. 312 313endchoice 314 315config RANDOM_KMALLOC_CACHES 316 bool 317 transitional 318 help 319 Transitional config for migration to KMALLOC_PARTITION_CACHES. 320 321endmenu # Slab allocator options 322 323config SHUFFLE_PAGE_ALLOCATOR 324 bool "Page allocator randomization" 325 default SLAB_FREELIST_RANDOM && ACPI_NUMA 326 help 327 Randomization of the page allocator improves the average 328 utilization of a direct-mapped memory-side-cache. See section 329 5.2.27 Heterogeneous Memory Attribute Table (HMAT) in the ACPI 330 6.2a specification for an example of how a platform advertises 331 the presence of a memory-side-cache. There are also incidental 332 security benefits as it reduces the predictability of page 333 allocations to compliment SLAB_FREELIST_RANDOM, but the 334 default granularity of shuffling on the MAX_PAGE_ORDER i.e, 10th 335 order of pages is selected based on cache utilization benefits 336 on x86. 337 338 While the randomization improves cache utilization it may 339 negatively impact workloads on platforms without a cache. For 340 this reason, by default, the randomization is not enabled even 341 if SHUFFLE_PAGE_ALLOCATOR=y. The randomization may be force enabled 342 with the 'page_alloc.shuffle' kernel command line parameter. 343 344 Say Y if unsure. 345 346config COMPAT_BRK 347 bool "Disable heap randomization" 348 default y 349 help 350 Randomizing heap placement makes heap exploits harder, but it 351 also breaks ancient binaries (including anything libc5 based). 352 This option changes the bootup default to heap randomization 353 disabled, and can be overridden at runtime by setting 354 /proc/sys/kernel/randomize_va_space to 2. 355 356 On non-ancient distros (post-2000 ones) N is usually a safe choice. 357 358config MMAP_ALLOW_UNINITIALIZED 359 bool "Allow mmapped anonymous memory to be uninitialized" 360 depends on EXPERT && !MMU 361 default n 362 help 363 Normally, and according to the Linux spec, anonymous memory obtained 364 from mmap() has its contents cleared before it is passed to 365 userspace. Enabling this config option allows you to request that 366 mmap() skip that if it is given an MAP_UNINITIALIZED flag, thus 367 providing a huge performance boost. If this option is not enabled, 368 then the flag will be ignored. 369 370 This is taken advantage of by uClibc's malloc(), and also by 371 ELF-FDPIC binfmt's brk and stack allocator. 372 373 Because of the obvious security issues, this option should only be 374 enabled on embedded devices where you control what is run in 375 userspace. Since that isn't generally a problem on no-MMU systems, 376 it is normally safe to say Y here. 377 378 See Documentation/admin-guide/mm/nommu-mmap.rst for more information. 379 380config SELECT_MEMORY_MODEL 381 def_bool y 382 depends on ARCH_SELECT_MEMORY_MODEL 383 384choice 385 prompt "Memory model" 386 depends on SELECT_MEMORY_MODEL 387 default SPARSEMEM_MANUAL if ARCH_SPARSEMEM_DEFAULT 388 default FLATMEM_MANUAL 389 help 390 This option allows you to change some of the ways that 391 Linux manages its memory internally. Most users will 392 only have one option here selected by the architecture 393 configuration. This is normal. 394 395config FLATMEM_MANUAL 396 bool "Flat Memory" 397 depends on !ARCH_SPARSEMEM_ENABLE || ARCH_FLATMEM_ENABLE 398 help 399 This option is best suited for non-NUMA systems with 400 flat address space. The FLATMEM is the most efficient 401 system in terms of performance and resource consumption 402 and it is the best option for smaller systems. 403 404 For systems that have holes in their physical address 405 spaces and for features like NUMA and memory hotplug, 406 choose "Sparse Memory". 407 408 If unsure, choose this option (Flat Memory) over any other. 409 410config SPARSEMEM_MANUAL 411 bool "Sparse Memory" 412 depends on ARCH_SPARSEMEM_ENABLE 413 help 414 This will be the only option for some systems, including 415 memory hot-plug systems. This is normal. 416 417 This option provides efficient support for systems with 418 holes is their physical address space and allows memory 419 hot-plug and hot-remove. 420 421 If unsure, choose "Flat Memory" over this option. 422 423endchoice 424 425config SPARSEMEM 426 def_bool y 427 depends on (!SELECT_MEMORY_MODEL && ARCH_SPARSEMEM_ENABLE) || SPARSEMEM_MANUAL 428 429config FLATMEM 430 def_bool y 431 depends on !SPARSEMEM || FLATMEM_MANUAL 432 433# 434# SPARSEMEM_EXTREME (which is the default) does some bootmem 435# allocations when sparse_init() is called. If this cannot 436# be done on your architecture, select this option. However, 437# statically allocating the mem_section[] array can potentially 438# consume vast quantities of .bss, so be careful. 439# 440# This option will also potentially produce smaller runtime code 441# with gcc 3.4 and later. 442# 443config SPARSEMEM_STATIC 444 bool 445 446# 447# Architecture platforms which require a two level mem_section in SPARSEMEM 448# must select this option. This is usually for architecture platforms with 449# an extremely sparse physical address space. 450# 451config SPARSEMEM_EXTREME 452 def_bool y 453 depends on SPARSEMEM && !SPARSEMEM_STATIC 454 455config SPARSEMEM_VMEMMAP_ENABLE 456 bool 457 458config SPARSEMEM_VMEMMAP 459 def_bool y 460 depends on SPARSEMEM && SPARSEMEM_VMEMMAP_ENABLE 461 help 462 SPARSEMEM_VMEMMAP uses a virtually mapped memmap to optimise 463 pfn_to_page and page_to_pfn operations. This is the most 464 efficient option when sufficient kernel resources are available. 465 466config SPARSEMEM_VMEMMAP_PREINIT 467 bool 468# 469# Select this config option from the architecture Kconfig, if it is preferred 470# to enable the feature of HugeTLB/dev_dax vmemmap optimization. 471# 472config ARCH_WANT_OPTIMIZE_DAX_VMEMMAP 473 bool 474 475config ARCH_WANT_OPTIMIZE_HUGETLB_VMEMMAP 476 bool 477 478config ARCH_WANT_HUGETLB_VMEMMAP_PREINIT 479 bool 480 481config HAVE_MEMBLOCK_PHYS_MAP 482 bool 483 484config HAVE_GUP_FAST 485 depends on MMU 486 bool 487 488# Enable memblock support for scratch memory which is needed for kexec handover 489config MEMBLOCK_KHO_SCRATCH 490 bool 491 492# Don't discard allocated memory used to track "memory" and "reserved" memblocks 493# after early boot, so it can still be used to test for validity of memory. 494# Also, memblocks are updated with memory hot(un)plug. 495config ARCH_KEEP_MEMBLOCK 496 bool 497 498# Keep arch NUMA mapping infrastructure post-init. 499config NUMA_KEEP_MEMINFO 500 bool 501 502config MEMORY_ISOLATION 503 bool 504 505# IORESOURCE_SYSTEM_RAM regions in the kernel resource tree that are marked 506# IORESOURCE_EXCLUSIVE cannot be mapped to user space, for example, via 507# /dev/mem. 508config EXCLUSIVE_SYSTEM_RAM 509 def_bool y 510 depends on !DEVMEM || STRICT_DEVMEM 511 512config ARCH_ENABLE_MEMORY_HOTPLUG 513 bool 514 515# eventually, we can have this option just 'select SPARSEMEM' 516menuconfig MEMORY_HOTPLUG 517 bool "Memory hotplug" 518 select MEMORY_ISOLATION 519 depends on SPARSEMEM_VMEMMAP 520 depends on ARCH_ENABLE_MEMORY_HOTPLUG 521 depends on 64BIT 522 select NUMA_KEEP_MEMINFO if NUMA 523 524if MEMORY_HOTPLUG 525 526choice 527 prompt "Memory Hotplug Default Online Type" 528 default MHP_DEFAULT_ONLINE_TYPE_OFFLINE 529 help 530 Default memory type for hotplugged memory. 531 532 This option sets the default policy setting for memory hotplug 533 onlining policy (/sys/devices/system/memory/auto_online_blocks) which 534 determines what happens to newly added memory regions. Policy setting 535 can always be changed at runtime. 536 537 The default is 'offline'. 538 539 Select offline to defer onlining to drivers and user policy. 540 Select auto to let the kernel choose what zones to utilize. 541 Select online_kernel to generally allow kernel usage of this memory. 542 Select online_movable to generally disallow kernel usage of this memory. 543 544 Example kernel usage would be page structs and page tables. 545 546 See Documentation/admin-guide/mm/memory-hotplug.rst for more information. 547 548config MHP_DEFAULT_ONLINE_TYPE_OFFLINE 549 bool "offline" 550 help 551 Hotplugged memory will not be onlined by default. 552 Choose this for systems with drivers and user policy that 553 handle onlining of hotplug memory policy. 554 555config MHP_DEFAULT_ONLINE_TYPE_ONLINE_AUTO 556 bool "auto" 557 help 558 Select this if you want the kernel to automatically online 559 hotplugged memory into the zone it thinks is reasonable. 560 This memory may be utilized for kernel data. 561 562config MHP_DEFAULT_ONLINE_TYPE_ONLINE_KERNEL 563 bool "kernel" 564 help 565 Select this if you want the kernel to automatically online 566 hotplugged memory into a zone capable of being used for kernel 567 data. This typically means ZONE_NORMAL. 568 569config MHP_DEFAULT_ONLINE_TYPE_ONLINE_MOVABLE 570 bool "movable" 571 help 572 Select this if you want the kernel to automatically online 573 hotplug memory into ZONE_MOVABLE. This memory will generally 574 not be utilized for kernel data. 575 576 This should only be used when the admin knows sufficient 577 ZONE_NORMAL memory is available to describe hotplug memory, 578 otherwise hotplug memory may fail to online. For example, 579 sufficient kernel-capable memory (ZONE_NORMAL) must be 580 available to allocate page structs to describe ZONE_MOVABLE. 581 582endchoice 583 584config MEMORY_HOTREMOVE 585 bool "Allow for memory hot remove" 586 depends on MEMORY_HOTPLUG 587 select MIGRATION 588 589config MHP_MEMMAP_ON_MEMORY 590 def_bool y 591 depends on MEMORY_HOTPLUG && SPARSEMEM_VMEMMAP 592 depends on ARCH_MHP_MEMMAP_ON_MEMORY_ENABLE 593 594endif # MEMORY_HOTPLUG 595 596config ARCH_MHP_MEMMAP_ON_MEMORY_ENABLE 597 bool 598 599# Heavily threaded applications may benefit from splitting the mm-wide 600# page_table_lock, so that faults on different parts of the user address 601# space can be handled with less contention: split it at this NR_CPUS. 602# Default to 4 for wider testing, though 8 might be more appropriate. 603# ARM's adjust_pte (unused if VIPT) depends on mm-wide page_table_lock. 604# PA-RISC 7xxx's spinlock_t would enlarge struct page from 32 to 44 bytes. 605# SPARC32 allocates multiple pte tables within a single page, and therefore 606# a per-page lock leads to problems when multiple tables need to be locked 607# at the same time (e.g. copy_page_range()). 608# DEBUG_SPINLOCK and DEBUG_LOCK_ALLOC spinlock_t also enlarge struct page. 609# 610config SPLIT_PTE_PTLOCKS 611 def_bool y 612 depends on MMU 613 depends on SMP 614 depends on NR_CPUS >= 4 615 depends on !ARM || CPU_CACHE_VIPT 616 depends on !PARISC || PA20 617 depends on !SPARC32 618 depends on !UML 619 620config ARCH_ENABLE_SPLIT_PMD_PTLOCK 621 bool 622 623config SPLIT_PMD_PTLOCKS 624 def_bool y 625 depends on SPLIT_PTE_PTLOCKS && ARCH_ENABLE_SPLIT_PMD_PTLOCK 626 627# 628# support for memory balloon 629config BALLOON 630 bool 631 632# 633# support for memory balloon page migration 634config BALLOON_MIGRATION 635 bool "Allow for balloon memory migration" 636 default y 637 depends on MIGRATION && BALLOON 638 help 639 Allow for migration of pages inflated in a memory balloon such that 640 they can be allocated from memory areas only available for movable 641 allocations (e.g., ZONE_MOVABLE, CMA) and such that they can be 642 migrated for memory defragmentation purposes by memory compaction. 643 644# 645# support for memory compaction 646config COMPACTION 647 bool "Allow for memory compaction" 648 default y 649 select MIGRATION 650 depends on MMU 651 help 652 Compaction is the only memory management component to form 653 high order (larger physically contiguous) memory blocks 654 reliably. The page allocator relies on compaction heavily and 655 the lack of the feature can lead to unexpected OOM killer 656 invocations for high order memory requests. You shouldn't 657 disable this option unless there really is a strong reason for 658 it and then we would be really interested to hear about that at 659 linux-mm@kvack.org. 660 661config COMPACT_UNEVICTABLE_DEFAULT 662 int 663 depends on COMPACTION 664 default 0 if PREEMPT_RT 665 default 1 666 667# 668# support for free page reporting 669config PAGE_REPORTING 670 bool "Free page reporting" 671 help 672 Free page reporting allows for the incremental acquisition of 673 free pages from the buddy allocator for the purpose of reporting 674 those pages to another entity, such as a hypervisor, so that the 675 memory can be freed within the host for other uses. 676 677config NUMA_MIGRATION 678 bool "NUMA page migration" 679 default y 680 depends on NUMA && MMU 681 select MIGRATION 682 help 683 Support the migration of pages to other NUMA nodes, available to 684 user space through interfaces like migrate_pages(), move_pages(), 685 and mbind(). Selecting this option also enables support for page 686 demotion for memory tiering. 687 688config MIGRATION 689 bool 690 depends on MMU 691 692config DEVICE_MIGRATION 693 def_bool MIGRATION && ZONE_DEVICE 694 695config ARCH_ENABLE_HUGEPAGE_MIGRATION 696 bool 697 698config ARCH_HAS_PMD_SOFTLEAVES 699 bool 700 701config HUGETLB_PAGE_SIZE_VARIABLE 702 def_bool n 703 help 704 Allows the pageblock_order value to be dynamic instead of just standard 705 HUGETLB_PAGE_ORDER when there are multiple HugeTLB page sizes available 706 on a platform. 707 708 Note that the pageblock_order cannot exceed MAX_PAGE_ORDER and will be 709 clamped down to MAX_PAGE_ORDER. 710 711config CONTIG_ALLOC 712 def_bool (MEMORY_ISOLATION && COMPACTION) || CMA 713 714config PCP_BATCH_SCALE_MAX 715 int "Maximum scale factor of PCP (Per-CPU pageset) batch allocate/free" 716 default 5 717 range 0 6 718 help 719 In page allocator, PCP (Per-CPU pageset) is refilled and drained in 720 batches. The batch number is scaled automatically to improve page 721 allocation/free throughput. But too large scale factor may hurt 722 latency. This option sets the upper limit of scale factor to limit 723 the maximum latency. 724 725config PHYS_ADDR_T_64BIT 726 def_bool 64BIT 727 728config MMU_NOTIFIER 729 bool 730 select INTERVAL_TREE 731 732config KSM 733 bool "Enable KSM for page merging" 734 depends on MMU 735 select XXHASH 736 help 737 Enable Kernel Samepage Merging: KSM periodically scans those areas 738 of an application's address space that an app has advised may be 739 mergeable. When it finds pages of identical content, it replaces 740 the many instances by a single page with that content, so 741 saving memory until one or another app needs to modify the content. 742 Recommended for use with KVM, or with other duplicative applications. 743 See Documentation/mm/ksm.rst for more information: KSM is inactive 744 until a program has madvised that an area is MADV_MERGEABLE, and 745 root has set /sys/kernel/mm/ksm/run to 1 (if CONFIG_SYSFS is set). 746 747config DEFAULT_MMAP_MIN_ADDR 748 int "Low address space to protect from user allocation" 749 depends on MMU 750 default 4096 751 help 752 This is the portion of low virtual memory which should be protected 753 from userspace allocation. Keeping a user from writing to low pages 754 can help reduce the impact of kernel NULL pointer bugs. 755 756 For most arm64, ppc64 and x86 users with lots of address space 757 a value of 65536 is reasonable and should cause no problems. 758 On arm and other archs it should not be higher than 32768. 759 Programs which use vm86 functionality or have some need to map 760 this low address space will need CAP_SYS_RAWIO or disable this 761 protection by setting the value to 0. 762 763 This value can be changed after boot using the 764 /proc/sys/vm/mmap_min_addr tunable. 765 766config ARCH_SUPPORTS_MEMORY_FAILURE 767 bool 768 769config MEMORY_FAILURE 770 depends on MMU 771 depends on ARCH_SUPPORTS_MEMORY_FAILURE 772 bool "Enable recovery from hardware memory errors" 773 select INTERVAL_TREE 774 help 775 Enables code to recover from some memory failures on systems 776 with MCA recovery. This allows a system to continue running 777 even when some of its memory has uncorrected errors. This requires 778 special hardware support and typically ECC memory. 779 780config HWPOISON_INJECT 781 tristate "HWPoison pages injector" 782 depends on MEMORY_FAILURE && DEBUG_KERNEL && PROC_FS 783 select PROC_PAGE_MONITOR 784 785config NOMMU_INITIAL_TRIM_EXCESS 786 int "Turn on mmap() excess space trimming before booting" 787 depends on !MMU 788 default 1 789 help 790 The NOMMU mmap() frequently needs to allocate large contiguous chunks 791 of memory on which to store mappings, but it can only ask the system 792 allocator for chunks in 2^N*PAGE_SIZE amounts - which is frequently 793 more than it requires. To deal with this, mmap() is able to trim off 794 the excess and return it to the allocator. 795 796 If trimming is enabled, the excess is trimmed off and returned to the 797 system allocator, which can cause extra fragmentation, particularly 798 if there are a lot of transient processes. 799 800 If trimming is disabled, the excess is kept, but not used, which for 801 long-term mappings means that the space is wasted. 802 803 Trimming can be dynamically controlled through a sysctl option 804 (/proc/sys/vm/nr_trim_pages) which specifies the minimum number of 805 excess pages there must be before trimming should occur, or zero if 806 no trimming is to occur. 807 808 This option specifies the initial value of this option. The default 809 of 1 says that all excess pages should be trimmed. 810 811 See Documentation/admin-guide/mm/nommu-mmap.rst for more information. 812 813config ARCH_WANT_GENERAL_HUGETLB 814 bool 815 816config ARCH_WANTS_THP_SWAP 817 def_bool n 818 819config PERSISTENT_HUGE_ZERO_FOLIO 820 bool "Allocate a PMD sized folio for zeroing" 821 depends on TRANSPARENT_HUGEPAGE 822 help 823 Enable this option to reduce the runtime refcounting overhead 824 of the huge zero folio and expand the places in the kernel 825 that can use huge zero folios. For instance, block I/O benefits 826 from access to large folios for zeroing memory. 827 828 With this option enabled, the huge zero folio is allocated 829 once and never freed. One full huge page's worth of memory shall 830 be used. 831 832 Say Y if your system has lots of memory. Say N if you are 833 memory constrained. 834 835config MM_ID 836 def_bool n 837 838menuconfig TRANSPARENT_HUGEPAGE 839 bool "Transparent Hugepage Support" 840 depends on HAVE_ARCH_TRANSPARENT_HUGEPAGE && !PREEMPT_RT 841 select COMPACTION 842 select XARRAY_MULTI 843 select MM_ID 844 help 845 Transparent Hugepages allows the kernel to use huge pages and 846 huge tlb transparently to the applications whenever possible. 847 This feature can improve computing performance to certain 848 applications by speeding up page faults during memory 849 allocation, by reducing the number of tlb misses and by speeding 850 up the pagetable walking. 851 852 If memory constrained on embedded, you may want to say N. 853 854if TRANSPARENT_HUGEPAGE 855 856choice 857 prompt "Transparent Hugepage Support sysfs defaults" 858 default TRANSPARENT_HUGEPAGE_ALWAYS 859 help 860 Selects the sysfs defaults for Transparent Hugepage Support. 861 862 config TRANSPARENT_HUGEPAGE_ALWAYS 863 bool "always" 864 help 865 Enabling Transparent Hugepage always, can increase the 866 memory footprint of applications without a guaranteed 867 benefit but it will work automatically for all applications. 868 869 config TRANSPARENT_HUGEPAGE_MADVISE 870 bool "madvise" 871 help 872 Enabling Transparent Hugepage madvise, will only provide a 873 performance improvement benefit to the applications using 874 madvise(MADV_HUGEPAGE) but it won't risk to increase the 875 memory footprint of applications without a guaranteed 876 benefit. 877 878 config TRANSPARENT_HUGEPAGE_NEVER 879 bool "never" 880 help 881 Disable Transparent Hugepage by default. It can still be 882 enabled at runtime via sysfs. 883endchoice 884 885choice 886 prompt "Shmem hugepage allocation defaults" 887 default TRANSPARENT_HUGEPAGE_SHMEM_HUGE_NEVER 888 help 889 Selects the hugepage allocation policy defaults for 890 the internal shmem mount. 891 892 The selection made here can be overridden by using the kernel 893 command line 'transparent_hugepage_shmem=' option. 894 895 config TRANSPARENT_HUGEPAGE_SHMEM_HUGE_NEVER 896 bool "never" 897 help 898 Disable hugepage allocation for shmem mount by default. It can 899 still be enabled with the kernel command line 900 'transparent_hugepage_shmem=' option or at runtime via sysfs 901 knob. Note that madvise(MADV_COLLAPSE) can still cause 902 transparent huge pages to be obtained even if this mode is 903 specified. 904 905 config TRANSPARENT_HUGEPAGE_SHMEM_HUGE_ALWAYS 906 bool "always" 907 help 908 Always attempt to allocate hugepage for shmem mount, can 909 increase the memory footprint of applications without a 910 guaranteed benefit but it will work automatically for all 911 applications. 912 913 config TRANSPARENT_HUGEPAGE_SHMEM_HUGE_WITHIN_SIZE 914 bool "within_size" 915 help 916 Enable hugepage allocation for shmem mount if the allocation 917 will be fully within the i_size. This configuration also takes 918 into account any madvise(MADV_HUGEPAGE) hints that may be 919 provided by the applications. 920 921 config TRANSPARENT_HUGEPAGE_SHMEM_HUGE_ADVISE 922 bool "advise" 923 help 924 Enable hugepage allocation for the shmem mount exclusively when 925 applications supply the madvise(MADV_HUGEPAGE) hint. 926 This ensures that hugepages are used only in response to explicit 927 requests from applications. 928endchoice 929 930choice 931 prompt "Tmpfs hugepage allocation defaults" 932 default TRANSPARENT_HUGEPAGE_TMPFS_HUGE_NEVER 933 help 934 Selects the hugepage allocation policy defaults for 935 the tmpfs mount. 936 937 The selection made here can be overridden by using the kernel 938 command line 'transparent_hugepage_tmpfs=' option. 939 940 config TRANSPARENT_HUGEPAGE_TMPFS_HUGE_NEVER 941 bool "never" 942 help 943 Disable hugepage allocation for tmpfs mount by default. It can 944 still be enabled with the kernel command line 945 'transparent_hugepage_tmpfs=' option. Note that 946 madvise(MADV_COLLAPSE) can still cause transparent huge pages 947 to be obtained even if this mode is specified. 948 949 config TRANSPARENT_HUGEPAGE_TMPFS_HUGE_ALWAYS 950 bool "always" 951 help 952 Always attempt to allocate hugepage for tmpfs mount, can 953 increase the memory footprint of applications without a 954 guaranteed benefit but it will work automatically for all 955 applications. 956 957 config TRANSPARENT_HUGEPAGE_TMPFS_HUGE_WITHIN_SIZE 958 bool "within_size" 959 help 960 Enable hugepage allocation for tmpfs mount if the allocation 961 will be fully within the i_size. This configuration also takes 962 into account any madvise(MADV_HUGEPAGE) hints that may be 963 provided by the applications. 964 965 config TRANSPARENT_HUGEPAGE_TMPFS_HUGE_ADVISE 966 bool "advise" 967 help 968 Enable hugepage allocation for the tmpfs mount exclusively when 969 applications supply the madvise(MADV_HUGEPAGE) hint. 970 This ensures that hugepages are used only in response to explicit 971 requests from applications. 972endchoice 973 974config THP_SWAP 975 def_bool y 976 depends on ARCH_WANTS_THP_SWAP && SWAP && 64BIT 977 help 978 Swap transparent huge pages in one piece, without splitting. 979 XXX: For now, swap cluster backing transparent huge page 980 will be split after swapout. 981 982 For selection by architectures with reasonable THP sizes. 983 984config NO_PAGE_MAPCOUNT 985 bool "No per-page mapcount (EXPERIMENTAL)" 986 help 987 Do not maintain per-page mapcounts for pages part of larger 988 allocations, such as transparent huge pages. 989 990 When this config option is enabled, some interfaces that relied on 991 this information will rely on less-precise per-allocation information 992 instead: for example, using the average per-page mapcount in such 993 a large allocation instead of the per-page mapcount. 994 995 EXPERIMENTAL because the impact of some changes is still unclear. 996 997endif # TRANSPARENT_HUGEPAGE 998 999# simple helper to make the code a bit easier to read 1000config PAGE_MAPCOUNT 1001 def_bool !NO_PAGE_MAPCOUNT 1002 1003# 1004# The architecture supports pgtable leaves that is larger than PAGE_SIZE 1005# 1006config PGTABLE_HAS_HUGE_LEAVES 1007 def_bool TRANSPARENT_HUGEPAGE || HUGETLB_PAGE 1008 1009# 1010# We can end up creating gigantic folio. 1011# 1012config HAVE_GIGANTIC_FOLIOS 1013 def_bool (HUGETLB_PAGE && ARCH_HAS_GIGANTIC_PAGE) || \ 1014 (ZONE_DEVICE && HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD) 1015 1016config ASYNC_KERNEL_PGTABLE_FREE 1017 def_bool n 1018 1019# TODO: Allow to be enabled without THP 1020config ARCH_SUPPORTS_HUGE_PFNMAP 1021 def_bool n 1022 depends on TRANSPARENT_HUGEPAGE 1023 1024config ARCH_SUPPORTS_PMD_PFNMAP 1025 def_bool y 1026 depends on ARCH_SUPPORTS_HUGE_PFNMAP && HAVE_ARCH_TRANSPARENT_HUGEPAGE 1027 1028config ARCH_SUPPORTS_PUD_PFNMAP 1029 def_bool y 1030 depends on ARCH_SUPPORTS_HUGE_PFNMAP && HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD 1031 1032# 1033# Architectures that always use weak definitions for percpu 1034# variables in modules should set this. 1035# 1036config ARCH_MODULE_NEEDS_WEAK_PER_CPU 1037 bool 1038 1039# 1040# UP and nommu archs use km based percpu allocator 1041# 1042config NEED_PER_CPU_KM 1043 depends on !SMP || !MMU 1044 bool 1045 default y 1046 1047config NEED_PER_CPU_EMBED_FIRST_CHUNK 1048 bool 1049 1050config NEED_PER_CPU_PAGE_FIRST_CHUNK 1051 bool 1052 1053config USE_PERCPU_NUMA_NODE_ID 1054 bool 1055 1056config HAVE_SETUP_PER_CPU_AREA 1057 bool 1058 1059config CMA 1060 bool "Contiguous Memory Allocator" 1061 depends on MMU 1062 select MIGRATION 1063 select MEMORY_ISOLATION 1064 help 1065 This enables the Contiguous Memory Allocator which allows other 1066 subsystems to allocate big physically-contiguous blocks of memory. 1067 CMA reserves a region of memory and allows only movable pages to 1068 be allocated from it. This way, the kernel can use the memory for 1069 pagecache and when a subsystem requests for contiguous area, the 1070 allocated pages are migrated away to serve the contiguous request. 1071 1072 If unsure, say "n". 1073 1074config CMA_DEBUGFS 1075 bool "CMA debugfs interface" 1076 depends on CMA && DEBUG_FS 1077 help 1078 Turns on the DebugFS interface for CMA. 1079 1080config CMA_SYSFS 1081 bool "CMA information through sysfs interface" 1082 depends on CMA && SYSFS 1083 help 1084 This option exposes some sysfs attributes to get information 1085 from CMA. 1086 1087config CMA_AREAS 1088 int "Maximum count of the CMA areas" 1089 depends on CMA 1090 default 20 if NUMA 1091 default 8 1092 help 1093 CMA allows to create CMA areas for particular purpose, mainly, 1094 used as device private area. This parameter sets the maximum 1095 number of CMA area in the system. 1096 1097 If unsure, leave the default value "8" in UMA and "20" in NUMA. 1098 1099# 1100# Select this config option from the architecture Kconfig, if available, to set 1101# the max page order for physically contiguous allocations. 1102# 1103config ARCH_FORCE_MAX_ORDER 1104 int 1105 1106# 1107# When ARCH_FORCE_MAX_ORDER is not defined, 1108# the default page block order is MAX_PAGE_ORDER (10) as per 1109# include/linux/mmzone.h. 1110# 1111config PAGE_BLOCK_MAX_ORDER 1112 int "Page Block Order Upper Limit" 1113 range 1 10 if ARCH_FORCE_MAX_ORDER = 0 1114 default 10 if ARCH_FORCE_MAX_ORDER = 0 1115 range 1 ARCH_FORCE_MAX_ORDER if ARCH_FORCE_MAX_ORDER != 0 1116 default ARCH_FORCE_MAX_ORDER if ARCH_FORCE_MAX_ORDER != 0 1117 help 1118 The page block order refers to the power of two number of pages that 1119 are physically contiguous and can have a migrate type associated to 1120 them. The maximum size of the page block order is at least limited by 1121 ARCH_FORCE_MAX_ORDER/MAX_PAGE_ORDER. 1122 1123 This config adds a new upper limit of default page block 1124 order when the page block order is required to be smaller than 1125 ARCH_FORCE_MAX_ORDER/MAX_PAGE_ORDER or other limits 1126 (see include/linux/pageblock-flags.h for details). 1127 1128 Reducing pageblock order can negatively impact THP generation 1129 success rate. If your workloads use THP heavily, please use this 1130 option with caution. 1131 1132 Don't change if unsure. 1133 1134config MEM_SOFT_DIRTY 1135 bool "Track memory changes" 1136 depends on CHECKPOINT_RESTORE && HAVE_ARCH_SOFT_DIRTY && PROC_FS 1137 select PROC_PAGE_MONITOR 1138 help 1139 This option enables memory changes tracking by introducing a 1140 soft-dirty bit on pte-s. This bit it set when someone writes 1141 into a page just as regular dirty bit, but unlike the latter 1142 it can be cleared by hands. 1143 1144 See Documentation/admin-guide/mm/soft-dirty.rst for more details. 1145 1146config GENERIC_EARLY_IOREMAP 1147 bool 1148 1149config STACK_MAX_DEFAULT_SIZE_MB 1150 int "Default maximum user stack size for 32-bit processes (MB)" 1151 default 100 1152 range 8 2048 1153 depends on STACK_GROWSUP && (!64BIT || COMPAT) 1154 help 1155 This is the maximum stack size in Megabytes in the VM layout of 32-bit 1156 user processes when the stack grows upwards (currently only on parisc 1157 arch) when the RLIMIT_STACK hard limit is unlimited. 1158 1159 A sane initial value is 100 MB. 1160 1161config DEFERRED_STRUCT_PAGE_INIT 1162 bool "Defer initialisation of struct pages to kthreads" 1163 depends on SPARSEMEM 1164 depends on !NEED_PER_CPU_KM 1165 depends on 64BIT 1166 depends on !KMSAN 1167 select PADATA 1168 help 1169 Ordinarily all struct pages are initialised during early boot in a 1170 single thread. On very large machines this can take a considerable 1171 amount of time. If this option is set, large machines will bring up 1172 a subset of memmap at boot and then initialise the rest in parallel. 1173 This has a potential performance impact on tasks running early in the 1174 lifetime of the system until these kthreads finish the 1175 initialisation. 1176 1177config PAGE_IDLE_FLAG 1178 bool 1179 select PAGE_EXTENSION if !64BIT 1180 help 1181 This adds PG_idle and PG_young flags to 'struct page'. PTE Accessed 1182 bit writers can set the state of the bit in the flags so that PTE 1183 Accessed bit readers may avoid disturbance. 1184 1185config IDLE_PAGE_TRACKING 1186 bool "Enable idle page tracking" 1187 depends on SYSFS && MMU 1188 select PAGE_IDLE_FLAG 1189 help 1190 This feature allows to estimate the amount of user pages that have 1191 not been touched during a given period of time. This information can 1192 be useful to tune memory cgroup limits and/or for job placement 1193 within a compute cluster. 1194 1195 See Documentation/admin-guide/mm/idle_page_tracking.rst for 1196 more details. 1197 1198# Architectures which implement cpu_dcache_is_aliasing() to query 1199# whether the data caches are aliased (VIVT or VIPT with dcache 1200# aliasing) need to select this. 1201config ARCH_HAS_CPU_CACHE_ALIASING 1202 bool 1203 1204config ARCH_HAS_CACHE_LINE_SIZE 1205 bool 1206 1207config ARCH_HAS_CURRENT_STACK_POINTER 1208 bool 1209 help 1210 In support of HARDENED_USERCOPY performing stack variable lifetime 1211 checking, an architecture-agnostic way to find the stack pointer 1212 is needed. Once an architecture defines an unsigned long global 1213 register alias named "current_stack_pointer", this config can be 1214 selected. 1215 1216config ARCH_HAS_ZONE_DMA_SET 1217 bool 1218 1219config ZONE_DMA 1220 bool "Support DMA zone" if ARCH_HAS_ZONE_DMA_SET 1221 default y if ARM64 || X86 1222 1223config ZONE_DMA32 1224 bool "Support DMA32 zone" if ARCH_HAS_ZONE_DMA_SET 1225 depends on !X86_32 1226 default y if ARM64 1227 1228config ZONE_DEVICE 1229 bool "Device memory (pmem, HMM, etc...) hotplug support" 1230 depends on MEMORY_HOTPLUG 1231 depends on MEMORY_HOTREMOVE 1232 depends on SPARSEMEM_VMEMMAP 1233 select XARRAY_MULTI 1234 1235 help 1236 Device memory hotplug support allows for establishing pmem, 1237 or other device driver discovered memory regions, in the 1238 memmap. This allows pfn_to_page() lookups of otherwise 1239 "device-physical" addresses which is needed for DAX, PCI_P2PDMA, and 1240 DEVICE_PRIVATE features among others. 1241 1242 Enabling this option will reduce the entropy of x86 KASLR memory 1243 regions. For example - on a 46 bit system, the entropy goes down 1244 from 16 bits to 15 bits. The actual reduction in entropy depends 1245 on the physical address bits, on processor features, kernel config 1246 (5 level page table) and physical memory present on the system. 1247 1248# 1249# Helpers to mirror range of the CPU page tables of a process into device page 1250# tables. 1251# 1252config HMM_MIRROR 1253 bool 1254 depends on MMU 1255 select MMU_NOTIFIER 1256 1257config GET_FREE_REGION 1258 bool 1259 1260config DEVICE_PRIVATE 1261 bool "Unaddressable device memory (GPU memory, ...)" 1262 depends on ZONE_DEVICE 1263 select GET_FREE_REGION 1264 1265 help 1266 Allows creation of struct pages to represent unaddressable device 1267 memory; i.e., memory that is only accessible from the device (or 1268 group of devices). You likely also want to select HMM_MIRROR. 1269 1270config VMAP_PFN 1271 bool 1272 1273config ARCH_USES_HIGH_VMA_FLAGS 1274 bool 1275config ARCH_HAS_PKEYS 1276 bool 1277 1278config ARCH_USES_PG_ARCH_2 1279 bool 1280config ARCH_USES_PG_ARCH_3 1281 bool 1282 1283config VM_EVENT_COUNTERS 1284 default y 1285 bool "Enable VM event counters for /proc/vmstat" if EXPERT 1286 help 1287 VM event counters are needed for event counts to be shown. 1288 This option allows the disabling of the VM event counters 1289 on EXPERT systems. /proc/vmstat will only show page counts 1290 if VM event counters are disabled. 1291 1292config PERCPU_STATS 1293 bool "Collect percpu memory statistics" 1294 help 1295 This feature collects and exposes statistics via debugfs. The 1296 information includes global and per chunk statistics, which can 1297 be used to help understand percpu memory usage. 1298 1299config GUP_TEST 1300 bool "Enable infrastructure for get_user_pages()-related unit tests" 1301 depends on DEBUG_FS 1302 help 1303 Provides /sys/kernel/debug/gup_test, which in turn provides a way 1304 to make ioctl calls that can launch kernel-based unit tests for 1305 the get_user_pages*() and pin_user_pages*() family of API calls. 1306 1307 These tests include benchmark testing of the _fast variants of 1308 get_user_pages*() and pin_user_pages*(), as well as smoke tests of 1309 the non-_fast variants. 1310 1311 There is also a sub-test that allows running dump_page() on any 1312 of up to eight pages (selected by command line args) within the 1313 range of user-space addresses. These pages are either pinned via 1314 pin_user_pages*(), or pinned via get_user_pages*(), as specified 1315 by other command line arguments. 1316 1317 See tools/testing/selftests/mm/gup_test.c 1318 1319comment "GUP_TEST needs to have DEBUG_FS enabled" 1320 depends on !GUP_TEST && !DEBUG_FS 1321 1322config GUP_GET_PXX_LOW_HIGH 1323 bool 1324 1325config DMAPOOL_TEST 1326 tristate "Enable a module to run time tests on dma_pool" 1327 depends on HAS_DMA 1328 help 1329 Provides a test module that will allocate and free many blocks of 1330 various sizes and report how long it takes. This is intended to 1331 provide a consistent way to measure how changes to the 1332 dma_pool_alloc/free routines affect performance. 1333 1334config ARCH_HAS_PTE_SPECIAL 1335 bool 1336 1337config MAPPING_DIRTY_HELPERS 1338 bool 1339 1340config KMAP_LOCAL 1341 bool 1342 1343config KMAP_LOCAL_NON_LINEAR_PTE_ARRAY 1344 bool 1345 1346config MEMFD_CREATE 1347 bool "Enable memfd_create() system call" if EXPERT 1348 1349config SECRETMEM 1350 default y 1351 bool "Enable memfd_secret() system call" if EXPERT 1352 depends on ARCH_HAS_SET_DIRECT_MAP 1353 help 1354 Enable the memfd_secret() system call with the ability to create 1355 memory areas visible only in the context of the owning process and 1356 not mapped to other processes and other kernel page tables. 1357 1358config ANON_VMA_NAME 1359 bool "Anonymous VMA name support" 1360 depends on PROC_FS && ADVISE_SYSCALLS && MMU 1361 1362 help 1363 Allow naming anonymous virtual memory areas. 1364 1365 This feature allows assigning names to virtual memory areas. Assigned 1366 names can be later retrieved from /proc/pid/maps and /proc/pid/smaps 1367 and help identifying individual anonymous memory areas. 1368 Assigning a name to anonymous virtual memory area might prevent that 1369 area from being merged with adjacent virtual memory areas due to the 1370 difference in their name. 1371 1372config HAVE_ARCH_USERFAULTFD_WP 1373 bool 1374 help 1375 Arch has userfaultfd write protection support 1376 1377config HAVE_ARCH_USERFAULTFD_MINOR 1378 bool 1379 help 1380 Arch has userfaultfd minor fault support 1381 1382config USERFAULTFD_RWP 1383 def_bool y 1384 depends on 64BIT && ARCH_HAS_PTE_PROTNONE && HAVE_ARCH_USERFAULTFD_WP 1385 help 1386 Userfaultfd read-write protection (UFFDIO_RWPROTECT) delivers a 1387 userfaultfd notification on every access -- read or write -- to a 1388 protected range, letting userspace observe the working set of a 1389 process. 1390 1391menuconfig USERFAULTFD 1392 bool "Enable userfaultfd() system call" 1393 depends on MMU 1394 help 1395 Enable the userfaultfd() system call that allows to intercept and 1396 handle page faults in userland. 1397 1398if USERFAULTFD 1399config PTE_MARKER_UFFD_WP 1400 bool "Userfaultfd write protection support for shmem/hugetlbfs" 1401 default y 1402 depends on HAVE_ARCH_USERFAULTFD_WP 1403 1404 help 1405 Allows to create marker PTEs for userfaultfd write protection 1406 purposes. It is required to enable userfaultfd write protection on 1407 file-backed memory types like shmem and hugetlbfs. 1408endif # USERFAULTFD 1409 1410# multi-gen LRU { 1411config LRU_GEN 1412 bool "Multi-Gen LRU" 1413 depends on MMU 1414 # make sure folio->flags has enough spare bits 1415 depends on 64BIT || !SPARSEMEM || SPARSEMEM_VMEMMAP 1416 help 1417 A high performance LRU implementation to overcommit memory. See 1418 Documentation/admin-guide/mm/multigen_lru.rst for details. 1419 1420config LRU_GEN_ENABLED 1421 bool "Enable by default" 1422 depends on LRU_GEN 1423 help 1424 This option enables the multi-gen LRU by default. 1425 1426config LRU_GEN_STATS 1427 bool "Full stats for debugging" 1428 depends on LRU_GEN 1429 help 1430 Do not enable this option unless you plan to look at historical stats 1431 from evicted generations for debugging purpose. 1432 1433 This option has a per-memcg and per-node memory overhead. 1434 1435config LRU_GEN_WALKS_MMU 1436 def_bool y 1437 depends on LRU_GEN && ARCH_HAS_HW_PTE_YOUNG 1438# } 1439 1440config ARCH_SUPPORTS_PER_VMA_LOCK 1441 def_bool n 1442 1443config PER_VMA_LOCK 1444 def_bool y 1445 depends on ARCH_SUPPORTS_PER_VMA_LOCK && MMU && SMP 1446 help 1447 Allow per-vma locking during page fault handling. 1448 1449 This feature allows locking each virtual memory area separately when 1450 handling page faults instead of taking mmap_lock. 1451 1452config LOCK_MM_AND_FIND_VMA 1453 bool 1454 depends on !STACK_GROWSUP 1455 1456config IOMMU_MM_DATA 1457 bool 1458 1459config EXECMEM 1460 bool 1461 1462config NUMA_MEMBLKS 1463 bool 1464 1465config NUMA_EMU 1466 bool "NUMA emulation" 1467 depends on NUMA_MEMBLKS 1468 depends on X86 || GENERIC_ARCH_NUMA 1469 help 1470 Enable NUMA emulation. A flat machine will be split 1471 into virtual nodes when booted with "numa=fake=N", where N is the 1472 number of nodes. This is only useful for debugging. 1473 1474config ARCH_HAS_USER_SHADOW_STACK 1475 bool 1476 help 1477 The architecture has hardware support for userspace shadow call 1478 stacks (eg, x86 CET, arm64 GCS or RISC-V Zicfiss). 1479 1480config HAVE_ARCH_TLB_REMOVE_TABLE 1481 def_bool n 1482 1483config PT_RECLAIM 1484 def_bool y 1485 depends on MMU_GATHER_RCU_TABLE_FREE && !HAVE_ARCH_TLB_REMOVE_TABLE 1486 help 1487 Try to reclaim empty user page table pages in paths other than munmap 1488 and exit_mmap path. 1489 1490 Note: now only empty user PTE page table pages will be reclaimed. 1491 1492config FIND_NORMAL_PAGE 1493 def_bool n 1494 1495config ARCH_HAS_LAZY_MMU_MODE 1496 bool 1497 help 1498 The architecture uses the lazy MMU mode. This allows changes to 1499 MMU-related architectural state to be deferred until the mode is 1500 exited. See <linux/pgtable.h> for details. 1501 1502config LAZY_MMU_MODE_KUNIT_TEST 1503 tristate "KUnit tests for the lazy MMU mode" if !KUNIT_ALL_TESTS 1504 depends on ARCH_HAS_LAZY_MMU_MODE 1505 depends on KUNIT 1506 default KUNIT_ALL_TESTS 1507 help 1508 Enable this option to check that the lazy MMU mode interface behaves 1509 as expected. Only tests for the generic interface are included (not 1510 architecture-specific behaviours). 1511 1512 If unsure, say N. 1513 1514source "mm/damon/Kconfig" 1515 1516endmenu 1517