1ec8f24b7SThomas Gleixner# SPDX-License-Identifier: GPL-2.0-only 259e0b520SChristoph Hellwig 359e0b520SChristoph Hellwigmenu "Memory Management options" 459e0b520SChristoph Hellwig 57b42f104SJohannes Weiner# 67b42f104SJohannes Weiner# For some reason microblaze and nios2 hard code SWAP=n. Hopefully we can 77b42f104SJohannes Weiner# add proper SWAP support to them, in which case this can be remove. 87b42f104SJohannes Weiner# 97b42f104SJohannes Weinerconfig ARCH_NO_SWAP 107b42f104SJohannes Weiner bool 117b42f104SJohannes Weiner 12b3fbd58fSJohannes Weinerconfig ZPOOL 13b3fbd58fSJohannes Weiner bool 14b3fbd58fSJohannes Weiner 15519bcb79SJohannes Weinermenuconfig SWAP 167b42f104SJohannes Weiner bool "Support for paging of anonymous memory (swap)" 177b42f104SJohannes Weiner depends on MMU && BLOCK && !ARCH_NO_SWAP 187b42f104SJohannes Weiner default y 197b42f104SJohannes Weiner help 207b42f104SJohannes Weiner This option allows you to choose whether you want to have support 217b42f104SJohannes Weiner for so called swap devices or swap files in your kernel that are 227b42f104SJohannes Weiner used to provide more virtual memory than the actual RAM present 237b42f104SJohannes Weiner in your computer. If unsure say Y. 247b42f104SJohannes Weiner 25519bcb79SJohannes Weinerconfig ZSWAP 26fcab9b44SDavid Heidelberg bool "Compressed cache for swap pages" 27b3fbd58fSJohannes Weiner depends on SWAP 28b3fbd58fSJohannes Weiner select CRYPTO 29519bcb79SJohannes Weiner select ZPOOL 30519bcb79SJohannes Weiner help 31519bcb79SJohannes Weiner A lightweight compressed cache for swap pages. It takes 32519bcb79SJohannes Weiner pages that are in the process of being swapped out and attempts to 33519bcb79SJohannes Weiner compress them into a dynamically allocated RAM-based memory pool. 34519bcb79SJohannes Weiner This can result in a significant I/O reduction on swap device and, 351a44131dSSophia Gabriella in the case where decompressing from RAM is faster than swap device 36519bcb79SJohannes Weiner reads, can also improve workload performance. 37519bcb79SJohannes Weiner 38b3fbd58fSJohannes Weinerconfig ZSWAP_DEFAULT_ON 39b3fbd58fSJohannes Weiner bool "Enable the compressed cache for swap pages by default" 40b3fbd58fSJohannes Weiner depends on ZSWAP 41b3fbd58fSJohannes Weiner help 42b3fbd58fSJohannes Weiner If selected, the compressed cache for swap pages will be enabled 43b3fbd58fSJohannes Weiner at boot, otherwise it will be disabled. 44b3fbd58fSJohannes Weiner 45b3fbd58fSJohannes Weiner The selection made here can be overridden by using the kernel 46b3fbd58fSJohannes Weiner command line 'zswap.enabled=' option. 47b3fbd58fSJohannes Weiner 48b9c91c43SYosry Ahmedconfig ZSWAP_EXCLUSIVE_LOADS_DEFAULT_ON 49b9c91c43SYosry Ahmed bool "Invalidate zswap entries when pages are loaded" 50b9c91c43SYosry Ahmed depends on ZSWAP 51b9c91c43SYosry Ahmed help 52b9c91c43SYosry Ahmed If selected, exclusive loads for zswap will be enabled at boot, 53b9c91c43SYosry Ahmed otherwise it will be disabled. 54b9c91c43SYosry Ahmed 55b9c91c43SYosry Ahmed If exclusive loads are enabled, when a page is loaded from zswap, 56b9c91c43SYosry Ahmed the zswap entry is invalidated at once, as opposed to leaving it 57b9c91c43SYosry Ahmed in zswap until the swap entry is freed. 58b9c91c43SYosry Ahmed 59b9c91c43SYosry Ahmed This avoids having two copies of the same page in memory 60b9c91c43SYosry Ahmed (compressed and uncompressed) after faulting in a page from zswap. 61b9c91c43SYosry Ahmed The cost is that if the page was never dirtied and needs to be 62b9c91c43SYosry Ahmed swapped out again, it will be re-compressed. 63b9c91c43SYosry Ahmed 64519bcb79SJohannes Weinerchoice 65b3fbd58fSJohannes Weiner prompt "Default compressor" 66519bcb79SJohannes Weiner depends on ZSWAP 67519bcb79SJohannes Weiner default ZSWAP_COMPRESSOR_DEFAULT_LZO 68519bcb79SJohannes Weiner help 69519bcb79SJohannes Weiner Selects the default compression algorithm for the compressed cache 70519bcb79SJohannes Weiner for swap pages. 71519bcb79SJohannes Weiner 72519bcb79SJohannes Weiner For an overview what kind of performance can be expected from 73519bcb79SJohannes Weiner a particular compression algorithm please refer to the benchmarks 74519bcb79SJohannes Weiner available at the following LWN page: 75519bcb79SJohannes Weiner https://lwn.net/Articles/751795/ 76519bcb79SJohannes Weiner 77519bcb79SJohannes Weiner If in doubt, select 'LZO'. 78519bcb79SJohannes Weiner 79519bcb79SJohannes Weiner The selection made here can be overridden by using the kernel 80519bcb79SJohannes Weiner command line 'zswap.compressor=' option. 81519bcb79SJohannes Weiner 82519bcb79SJohannes Weinerconfig ZSWAP_COMPRESSOR_DEFAULT_DEFLATE 83519bcb79SJohannes Weiner bool "Deflate" 84519bcb79SJohannes Weiner select CRYPTO_DEFLATE 85519bcb79SJohannes Weiner help 86519bcb79SJohannes Weiner Use the Deflate algorithm as the default compression algorithm. 87519bcb79SJohannes Weiner 88519bcb79SJohannes Weinerconfig ZSWAP_COMPRESSOR_DEFAULT_LZO 89519bcb79SJohannes Weiner bool "LZO" 90519bcb79SJohannes Weiner select CRYPTO_LZO 91519bcb79SJohannes Weiner help 92519bcb79SJohannes Weiner Use the LZO algorithm as the default compression algorithm. 93519bcb79SJohannes Weiner 94519bcb79SJohannes Weinerconfig ZSWAP_COMPRESSOR_DEFAULT_842 95519bcb79SJohannes Weiner bool "842" 96519bcb79SJohannes Weiner select CRYPTO_842 97519bcb79SJohannes Weiner help 98519bcb79SJohannes Weiner Use the 842 algorithm as the default compression algorithm. 99519bcb79SJohannes Weiner 100519bcb79SJohannes Weinerconfig ZSWAP_COMPRESSOR_DEFAULT_LZ4 101519bcb79SJohannes Weiner bool "LZ4" 102519bcb79SJohannes Weiner select CRYPTO_LZ4 103519bcb79SJohannes Weiner help 104519bcb79SJohannes Weiner Use the LZ4 algorithm as the default compression algorithm. 105519bcb79SJohannes Weiner 106519bcb79SJohannes Weinerconfig ZSWAP_COMPRESSOR_DEFAULT_LZ4HC 107519bcb79SJohannes Weiner bool "LZ4HC" 108519bcb79SJohannes Weiner select CRYPTO_LZ4HC 109519bcb79SJohannes Weiner help 110519bcb79SJohannes Weiner Use the LZ4HC algorithm as the default compression algorithm. 111519bcb79SJohannes Weiner 112519bcb79SJohannes Weinerconfig ZSWAP_COMPRESSOR_DEFAULT_ZSTD 113519bcb79SJohannes Weiner bool "zstd" 114519bcb79SJohannes Weiner select CRYPTO_ZSTD 115519bcb79SJohannes Weiner help 116519bcb79SJohannes Weiner Use the zstd algorithm as the default compression algorithm. 117519bcb79SJohannes Weinerendchoice 118519bcb79SJohannes Weiner 119519bcb79SJohannes Weinerconfig ZSWAP_COMPRESSOR_DEFAULT 120519bcb79SJohannes Weiner string 121519bcb79SJohannes Weiner depends on ZSWAP 122519bcb79SJohannes Weiner default "deflate" if ZSWAP_COMPRESSOR_DEFAULT_DEFLATE 123519bcb79SJohannes Weiner default "lzo" if ZSWAP_COMPRESSOR_DEFAULT_LZO 124519bcb79SJohannes Weiner default "842" if ZSWAP_COMPRESSOR_DEFAULT_842 125519bcb79SJohannes Weiner default "lz4" if ZSWAP_COMPRESSOR_DEFAULT_LZ4 126519bcb79SJohannes Weiner default "lz4hc" if ZSWAP_COMPRESSOR_DEFAULT_LZ4HC 127519bcb79SJohannes Weiner default "zstd" if ZSWAP_COMPRESSOR_DEFAULT_ZSTD 128519bcb79SJohannes Weiner default "" 129519bcb79SJohannes Weiner 130519bcb79SJohannes Weinerchoice 131b3fbd58fSJohannes Weiner prompt "Default allocator" 132519bcb79SJohannes Weiner depends on ZSWAP 133519bcb79SJohannes Weiner default ZSWAP_ZPOOL_DEFAULT_ZBUD 134519bcb79SJohannes Weiner help 135519bcb79SJohannes Weiner Selects the default allocator for the compressed cache for 136519bcb79SJohannes Weiner swap pages. 137519bcb79SJohannes Weiner The default is 'zbud' for compatibility, however please do 138519bcb79SJohannes Weiner read the description of each of the allocators below before 139519bcb79SJohannes Weiner making a right choice. 140519bcb79SJohannes Weiner 141519bcb79SJohannes Weiner The selection made here can be overridden by using the kernel 142519bcb79SJohannes Weiner command line 'zswap.zpool=' option. 143519bcb79SJohannes Weiner 144519bcb79SJohannes Weinerconfig ZSWAP_ZPOOL_DEFAULT_ZBUD 145519bcb79SJohannes Weiner bool "zbud" 146519bcb79SJohannes Weiner select ZBUD 147519bcb79SJohannes Weiner help 148519bcb79SJohannes Weiner Use the zbud allocator as the default allocator. 149519bcb79SJohannes Weiner 150519bcb79SJohannes Weinerconfig ZSWAP_ZPOOL_DEFAULT_Z3FOLD 151519bcb79SJohannes Weiner bool "z3fold" 152519bcb79SJohannes Weiner select Z3FOLD 153519bcb79SJohannes Weiner help 154519bcb79SJohannes Weiner Use the z3fold allocator as the default allocator. 155519bcb79SJohannes Weiner 156519bcb79SJohannes Weinerconfig ZSWAP_ZPOOL_DEFAULT_ZSMALLOC 157519bcb79SJohannes Weiner bool "zsmalloc" 158519bcb79SJohannes Weiner select ZSMALLOC 159519bcb79SJohannes Weiner help 160519bcb79SJohannes Weiner Use the zsmalloc allocator as the default allocator. 161519bcb79SJohannes Weinerendchoice 162519bcb79SJohannes Weiner 163519bcb79SJohannes Weinerconfig ZSWAP_ZPOOL_DEFAULT 164519bcb79SJohannes Weiner string 165519bcb79SJohannes Weiner depends on ZSWAP 166519bcb79SJohannes Weiner default "zbud" if ZSWAP_ZPOOL_DEFAULT_ZBUD 167519bcb79SJohannes Weiner default "z3fold" if ZSWAP_ZPOOL_DEFAULT_Z3FOLD 168519bcb79SJohannes Weiner default "zsmalloc" if ZSWAP_ZPOOL_DEFAULT_ZSMALLOC 169519bcb79SJohannes Weiner default "" 170519bcb79SJohannes Weiner 171519bcb79SJohannes Weinerconfig ZBUD 172b3fbd58fSJohannes Weiner tristate "2:1 compression allocator (zbud)" 173b3fbd58fSJohannes Weiner depends on ZSWAP 174519bcb79SJohannes Weiner help 175519bcb79SJohannes Weiner A special purpose allocator for storing compressed pages. 176519bcb79SJohannes Weiner It is designed to store up to two compressed pages per physical 177519bcb79SJohannes Weiner page. While this design limits storage density, it has simple and 178519bcb79SJohannes Weiner deterministic reclaim properties that make it preferable to a higher 179519bcb79SJohannes Weiner density approach when reclaim will be used. 180519bcb79SJohannes Weiner 181519bcb79SJohannes Weinerconfig Z3FOLD 182b3fbd58fSJohannes Weiner tristate "3:1 compression allocator (z3fold)" 183b3fbd58fSJohannes Weiner depends on ZSWAP 184519bcb79SJohannes Weiner help 185519bcb79SJohannes Weiner A special purpose allocator for storing compressed pages. 186519bcb79SJohannes Weiner It is designed to store up to three compressed pages per physical 187519bcb79SJohannes Weiner page. It is a ZBUD derivative so the simplicity and determinism are 188519bcb79SJohannes Weiner still there. 189519bcb79SJohannes Weiner 190519bcb79SJohannes Weinerconfig ZSMALLOC 191b3fbd58fSJohannes Weiner tristate 192b3fbd58fSJohannes Weiner prompt "N:1 compression allocator (zsmalloc)" if ZSWAP 193519bcb79SJohannes Weiner depends on MMU 194519bcb79SJohannes Weiner help 195519bcb79SJohannes Weiner zsmalloc is a slab-based memory allocator designed to store 196b3fbd58fSJohannes Weiner pages of various compression levels efficiently. It achieves 197b3fbd58fSJohannes Weiner the highest storage density with the least amount of fragmentation. 198519bcb79SJohannes Weiner 199519bcb79SJohannes Weinerconfig ZSMALLOC_STAT 200519bcb79SJohannes Weiner bool "Export zsmalloc statistics" 201519bcb79SJohannes Weiner depends on ZSMALLOC 202519bcb79SJohannes Weiner select DEBUG_FS 203519bcb79SJohannes Weiner help 204519bcb79SJohannes Weiner This option enables code in the zsmalloc to collect various 205519bcb79SJohannes Weiner statistics about what's happening in zsmalloc and exports that 206519bcb79SJohannes Weiner information to userspace via debugfs. 207519bcb79SJohannes Weiner If unsure, say N. 208519bcb79SJohannes Weiner 2094ff93b29SSergey Senozhatskyconfig ZSMALLOC_CHAIN_SIZE 2104ff93b29SSergey Senozhatsky int "Maximum number of physical pages per-zspage" 211b46402faSSergey Senozhatsky default 8 2124ff93b29SSergey Senozhatsky range 4 16 2134ff93b29SSergey Senozhatsky depends on ZSMALLOC 2144ff93b29SSergey Senozhatsky help 2154ff93b29SSergey Senozhatsky This option sets the upper limit on the number of physical pages 2164ff93b29SSergey Senozhatsky that a zmalloc page (zspage) can consist of. The optimal zspage 2174ff93b29SSergey Senozhatsky chain size is calculated for each size class during the 2184ff93b29SSergey Senozhatsky initialization of the pool. 2194ff93b29SSergey Senozhatsky 2204ff93b29SSergey Senozhatsky Changing this option can alter the characteristics of size classes, 2214ff93b29SSergey Senozhatsky such as the number of pages per zspage and the number of objects 2224ff93b29SSergey Senozhatsky per zspage. This can also result in different configurations of 2234ff93b29SSergey Senozhatsky the pool, as zsmalloc merges size classes with similar 2244ff93b29SSergey Senozhatsky characteristics. 2254ff93b29SSergey Senozhatsky 2264ff93b29SSergey Senozhatsky For more information, see zsmalloc documentation. 2274ff93b29SSergey Senozhatsky 228519bcb79SJohannes Weinermenu "SLAB allocator options" 229519bcb79SJohannes Weiner 2307b42f104SJohannes Weinerchoice 2317b42f104SJohannes Weiner prompt "Choose SLAB allocator" 2327b42f104SJohannes Weiner default SLUB 2337b42f104SJohannes Weiner help 2347b42f104SJohannes Weiner This option allows to select a slab allocator. 2357b42f104SJohannes Weiner 236eb07c4f3SVlastimil Babkaconfig SLAB_DEPRECATED 237eb07c4f3SVlastimil Babka bool "SLAB (DEPRECATED)" 2387b42f104SJohannes Weiner depends on !PREEMPT_RT 2397b42f104SJohannes Weiner help 240eb07c4f3SVlastimil Babka Deprecated and scheduled for removal in a few cycles. Replaced by 241eb07c4f3SVlastimil Babka SLUB. 242eb07c4f3SVlastimil Babka 243eb07c4f3SVlastimil Babka If you cannot migrate to SLUB, please contact linux-mm@kvack.org 244eb07c4f3SVlastimil Babka and the people listed in the SLAB ALLOCATOR section of MAINTAINERS 245eb07c4f3SVlastimil Babka file, explaining why. 246eb07c4f3SVlastimil Babka 2477b42f104SJohannes Weiner The regular slab allocator that is established and known to work 2487b42f104SJohannes Weiner well in all environments. It organizes cache hot objects in 2497b42f104SJohannes Weiner per cpu and per node queues. 2507b42f104SJohannes Weiner 2517b42f104SJohannes Weinerconfig SLUB 2527b42f104SJohannes Weiner bool "SLUB (Unqueued Allocator)" 2537b42f104SJohannes Weiner help 2547b42f104SJohannes Weiner SLUB is a slab allocator that minimizes cache line usage 2557b42f104SJohannes Weiner instead of managing queues of cached objects (SLAB approach). 2567b42f104SJohannes Weiner Per cpu caching is realized using slabs of objects instead 2577b42f104SJohannes Weiner of queues of objects. SLUB can use memory efficiently 2587b42f104SJohannes Weiner and has enhanced diagnostics. SLUB is the default choice for 2597b42f104SJohannes Weiner a slab allocator. 2607b42f104SJohannes Weiner 2617b42f104SJohannes Weinerendchoice 2627b42f104SJohannes Weiner 263eb07c4f3SVlastimil Babkaconfig SLAB 264eb07c4f3SVlastimil Babka bool 265eb07c4f3SVlastimil Babka default y 266eb07c4f3SVlastimil Babka depends on SLAB_DEPRECATED 267eb07c4f3SVlastimil Babka 268e240e53aSVlastimil Babkaconfig SLUB_TINY 269e240e53aSVlastimil Babka bool "Configure SLUB for minimal memory footprint" 270e240e53aSVlastimil Babka depends on SLUB && EXPERT 271e240e53aSVlastimil Babka select SLAB_MERGE_DEFAULT 272e240e53aSVlastimil Babka help 273e240e53aSVlastimil Babka Configures the SLUB allocator in a way to achieve minimal memory 274e240e53aSVlastimil Babka footprint, sacrificing scalability, debugging and other features. 275e240e53aSVlastimil Babka This is intended only for the smallest system that had used the 276e240e53aSVlastimil Babka SLOB allocator and is not recommended for systems with more than 277e240e53aSVlastimil Babka 16MB RAM. 278e240e53aSVlastimil Babka 279e240e53aSVlastimil Babka If unsure, say N. 280e240e53aSVlastimil Babka 2817b42f104SJohannes Weinerconfig SLAB_MERGE_DEFAULT 2827b42f104SJohannes Weiner bool "Allow slab caches to be merged" 2837b42f104SJohannes Weiner default y 2847b42f104SJohannes Weiner depends on SLAB || SLUB 2857b42f104SJohannes Weiner help 2867b42f104SJohannes Weiner For reduced kernel memory fragmentation, slab caches can be 2877b42f104SJohannes Weiner merged when they share the same size and other characteristics. 2887b42f104SJohannes Weiner This carries a risk of kernel heap overflows being able to 2897b42f104SJohannes Weiner overwrite objects from merged caches (and more easily control 2907b42f104SJohannes Weiner cache layout), which makes such heap attacks easier to exploit 2917b42f104SJohannes Weiner by attackers. By keeping caches unmerged, these kinds of exploits 2927b42f104SJohannes Weiner can usually only damage objects in the same cache. To disable 2937b42f104SJohannes Weiner merging at runtime, "slab_nomerge" can be passed on the kernel 2947b42f104SJohannes Weiner command line. 2957b42f104SJohannes Weiner 2967b42f104SJohannes Weinerconfig SLAB_FREELIST_RANDOM 2977b42f104SJohannes Weiner bool "Randomize slab freelist" 298e240e53aSVlastimil Babka depends on SLAB || (SLUB && !SLUB_TINY) 2997b42f104SJohannes Weiner help 3007b42f104SJohannes Weiner Randomizes the freelist order used on creating new pages. This 3017b42f104SJohannes Weiner security feature reduces the predictability of the kernel slab 3027b42f104SJohannes Weiner allocator against heap overflows. 3037b42f104SJohannes Weiner 3047b42f104SJohannes Weinerconfig SLAB_FREELIST_HARDENED 3057b42f104SJohannes Weiner bool "Harden slab freelist metadata" 306e240e53aSVlastimil Babka depends on SLAB || (SLUB && !SLUB_TINY) 3077b42f104SJohannes Weiner help 3087b42f104SJohannes Weiner Many kernel heap attacks try to target slab cache metadata and 3097b42f104SJohannes Weiner other infrastructure. This options makes minor performance 3107b42f104SJohannes Weiner sacrifices to harden the kernel slab allocator against common 3117b42f104SJohannes Weiner freelist exploit methods. Some slab implementations have more 3127b42f104SJohannes Weiner sanity-checking than others. This option is most effective with 3137b42f104SJohannes Weiner CONFIG_SLUB. 3147b42f104SJohannes Weiner 3150710d012SVlastimil Babkaconfig SLUB_STATS 3160710d012SVlastimil Babka default n 3170710d012SVlastimil Babka bool "Enable SLUB performance statistics" 318e240e53aSVlastimil Babka depends on SLUB && SYSFS && !SLUB_TINY 3190710d012SVlastimil Babka help 3200710d012SVlastimil Babka SLUB statistics are useful to debug SLUBs allocation behavior in 3210710d012SVlastimil Babka order find ways to optimize the allocator. This should never be 3220710d012SVlastimil Babka enabled for production use since keeping statistics slows down 3230710d012SVlastimil Babka the allocator by a few percentage points. The slabinfo command 3240710d012SVlastimil Babka supports the determination of the most active slabs to figure 3250710d012SVlastimil Babka out which slabs are relevant to a particular load. 3260710d012SVlastimil Babka Try running: slabinfo -DA 3270710d012SVlastimil Babka 328519bcb79SJohannes Weinerconfig SLUB_CPU_PARTIAL 329519bcb79SJohannes Weiner default y 330e240e53aSVlastimil Babka depends on SLUB && SMP && !SLUB_TINY 331519bcb79SJohannes Weiner bool "SLUB per cpu partial cache" 332519bcb79SJohannes Weiner help 333519bcb79SJohannes Weiner Per cpu partial caches accelerate objects allocation and freeing 334519bcb79SJohannes Weiner that is local to a processor at the price of more indeterminism 335519bcb79SJohannes Weiner in the latency of the free. On overflow these caches will be cleared 336519bcb79SJohannes Weiner which requires the taking of locks that may cause latency spikes. 337519bcb79SJohannes Weiner Typically one would choose no for a realtime system. 338519bcb79SJohannes Weiner 339519bcb79SJohannes Weinerconfig RANDOM_KMALLOC_CACHES 340519bcb79SJohannes Weiner default n 3417b42f104SJohannes Weiner depends on SLUB && !SLUB_TINY 3427b42f104SJohannes Weiner bool "Randomize slab caches for normal kmalloc" 3437b42f104SJohannes Weiner help 3447b42f104SJohannes Weiner A hardening feature that creates multiple copies of slab caches for 3457b42f104SJohannes Weiner normal kmalloc allocation and makes kmalloc randomly pick one based 3467b42f104SJohannes Weiner on code address, which makes the attackers more difficult to spray 3477b42f104SJohannes Weiner vulnerable memory objects on the heap for the purpose of exploiting 3487b42f104SJohannes Weiner memory vulnerabilities. 3497b42f104SJohannes Weiner 3507b42f104SJohannes Weiner Currently the number of copies is set to 16, a reasonably large value 3517b42f104SJohannes Weiner that effectively diverges the memory objects allocated for different 35223baf831SKirill A. Shutemov subsystems or modules into different caches, at the expense of a 35323baf831SKirill A. Shutemov limited degree of memory and CPU overhead that relates to hardware and 35423baf831SKirill A. Shutemov system workload. 3557b42f104SJohannes Weiner 3567b42f104SJohannes Weinerendmenu # SLAB allocator options 3577b42f104SJohannes Weiner 3587b42f104SJohannes Weinerconfig SHUFFLE_PAGE_ALLOCATOR 3597b42f104SJohannes Weiner bool "Page allocator randomization" 3607b42f104SJohannes Weiner default SLAB_FREELIST_RANDOM && ACPI_NUMA 3617b42f104SJohannes Weiner help 3627b42f104SJohannes Weiner Randomization of the page allocator improves the average 3637b42f104SJohannes Weiner utilization of a direct-mapped memory-side-cache. See section 3647b42f104SJohannes Weiner 5.2.27 Heterogeneous Memory Attribute Table (HMAT) in the ACPI 3650710d012SVlastimil Babka 6.2a specification for an example of how a platform advertises 3660710d012SVlastimil Babka the presence of a memory-side-cache. There are also incidental 3670710d012SVlastimil Babka security benefits as it reduces the predictability of page 3680710d012SVlastimil Babka allocations to compliment SLAB_FREELIST_RANDOM, but the 3690710d012SVlastimil Babka default granularity of shuffling on the MAX_ORDER i.e, 10th 3700710d012SVlastimil Babka order of pages is selected based on cache utilization benefits 3710710d012SVlastimil Babka on x86. 3720710d012SVlastimil Babka 3730710d012SVlastimil Babka While the randomization improves cache utilization it may 3740710d012SVlastimil Babka negatively impact workloads on platforms without a cache. For 3750710d012SVlastimil Babka this reason, by default, the randomization is enabled only 3760710d012SVlastimil Babka after runtime detection of a direct-mapped memory-side-cache. 3770710d012SVlastimil Babka Otherwise, the randomization may be force enabled with the 3780710d012SVlastimil Babka 'page_alloc.shuffle' kernel command line parameter. 3790710d012SVlastimil Babka 3800710d012SVlastimil Babka Say Y if unsure. 3810710d012SVlastimil Babka 3820710d012SVlastimil Babkaconfig COMPAT_BRK 3830710d012SVlastimil Babka bool "Disable heap randomization" 3840710d012SVlastimil Babka default y 3850710d012SVlastimil Babka help 3860710d012SVlastimil Babka Randomizing heap placement makes heap exploits harder, but it 3870710d012SVlastimil Babka also breaks ancient binaries (including anything libc5 based). 3880710d012SVlastimil Babka This option changes the bootup default to heap randomization 3890710d012SVlastimil Babka disabled, and can be overridden at runtime by setting 3900710d012SVlastimil Babka /proc/sys/kernel/randomize_va_space to 2. 3910710d012SVlastimil Babka 3920710d012SVlastimil Babka On non-ancient distros (post-2000 ones) N is usually a safe choice. 3930710d012SVlastimil Babka 3940710d012SVlastimil Babkaconfig MMAP_ALLOW_UNINITIALIZED 3950710d012SVlastimil Babka bool "Allow mmapped anonymous memory to be uninitialized" 3960710d012SVlastimil Babka depends on EXPERT && !MMU 3970710d012SVlastimil Babka default n 3980710d012SVlastimil Babka help 399e1785e85SDave Hansen Normally, and according to the Linux spec, anonymous memory obtained 400e1785e85SDave Hansen from mmap() has its contents cleared before it is passed to 401a8826eebSKees Cook userspace. Enabling this config option allows you to request that 402e1785e85SDave Hansen mmap() skip that if it is given an MAP_UNINITIALIZED flag, thus 4033a9da765SDave Hansen providing a huge performance boost. If this option is not enabled, 4043a9da765SDave Hansen then the flag will be ignored. 405e1785e85SDave Hansen 406d41dee36SAndy Whitcroft This is taken advantage of by uClibc's malloc(), and also by 407e1785e85SDave Hansen ELF-FDPIC binfmt's brk and stack allocator. 408d66d109dSMike Rapoport 409d66d109dSMike Rapoport Because of the obvious security issues, this option should only be 410d66d109dSMike Rapoport enabled on embedded devices where you control what is run in 411d66d109dSMike Rapoport userspace. Since that isn't generally a problem on no-MMU systems, 412d66d109dSMike Rapoport it is normally safe to say Y here. 4133a9da765SDave Hansen 414e1785e85SDave Hansen See Documentation/admin-guide/mm/nommu-mmap.rst for more information. 4153a9da765SDave Hansen 416bb1c50d3SMike Rapoportconfig SELECT_MEMORY_MODEL 4173a9da765SDave Hansen def_bool y 418d66d109dSMike Rapoport depends on ARCH_SELECT_MEMORY_MODEL 419d66d109dSMike Rapoport 420d66d109dSMike Rapoportchoice 421d66d109dSMike Rapoport prompt "Memory model" 4223a9da765SDave Hansen depends on SELECT_MEMORY_MODEL 423d66d109dSMike Rapoport default SPARSEMEM_MANUAL if ARCH_SPARSEMEM_DEFAULT 424d66d109dSMike Rapoport default FLATMEM_MANUAL 425dd33d29aSRandy Dunlap help 426d41dee36SAndy Whitcroft This option allows you to change some of the ways that 427d41dee36SAndy Whitcroft Linux manages its memory internally. Most users will 4283a9da765SDave Hansen only have one option here selected by the architecture 429d41dee36SAndy Whitcroft configuration. This is normal. 430d41dee36SAndy Whitcroft 431d41dee36SAndy Whitcroftconfig FLATMEM_MANUAL 432d41dee36SAndy Whitcroft bool "Flat Memory" 433d41dee36SAndy Whitcroft depends on !ARCH_SPARSEMEM_ENABLE || ARCH_FLATMEM_ENABLE 434d66d109dSMike Rapoport help 435d41dee36SAndy Whitcroft This option is best suited for non-NUMA systems with 436d66d109dSMike Rapoport flat address space. The FLATMEM is the most efficient 437d66d109dSMike Rapoport system in terms of performance and resource consumption 438d66d109dSMike Rapoport and it is the best option for smaller systems. 439d41dee36SAndy Whitcroft 440d66d109dSMike Rapoport For systems that have holes in their physical address 441d41dee36SAndy Whitcroft spaces and for features like NUMA and memory hotplug, 4423a9da765SDave Hansen choose "Sparse Memory". 4433a9da765SDave Hansen 444d41dee36SAndy Whitcroft If unsure, choose this option (Flat Memory) over any other. 445d41dee36SAndy Whitcroft 4461a83e175SRussell Kingconfig SPARSEMEM_MANUAL 447d41dee36SAndy Whitcroft bool "Sparse Memory" 448e1785e85SDave Hansen depends on ARCH_SPARSEMEM_ENABLE 449e1785e85SDave Hansen help 450bb1c50d3SMike Rapoport This will be the only option for some systems, including 451d41dee36SAndy Whitcroft memory hot-plug systems. This is normal. 45293b7504eSDave Hansen 4533e347261SBob Picco This option provides efficient support for systems with 454c89ab04fSMike Rapoport holes is their physical address space and allows memory 4553e347261SBob Picco hot-plug and hot-remove. 4563e347261SBob Picco 4573e347261SBob Picco If unsure, choose "Flat Memory" over this option. 4583e347261SBob Picco 4593e347261SBob Piccoendchoice 4603e347261SBob Picco 4613e347261SBob Piccoconfig SPARSEMEM 4623e347261SBob Picco def_bool y 4639ba16087SJan Beulich depends on (!SELECT_MEMORY_MODEL && ARCH_SPARSEMEM_ENABLE) || SPARSEMEM_MANUAL 4643e347261SBob Picco 4653e347261SBob Piccoconfig FLATMEM 46644c09201SMatt LaPlante def_bool y 467802f192eSBob Picco depends on !SPARSEMEM || FLATMEM_MANUAL 468802f192eSBob Picco 469802f192eSBob Picco# 4703e347261SBob Picco# SPARSEMEM_EXTREME (which is the default) does some bootmem 4713e347261SBob Picco# allocations when sparse_init() is called. If this cannot 4723e347261SBob Picco# be done on your architecture, select this option. However, 4734c21e2f2SHugh Dickins# statically allocating the mem_section[] array can potentially 47429c71111SAndy Whitcroft# consume vast quantities of .bss, so be careful. 4759ba16087SJan Beulich# 47629c71111SAndy Whitcroft# This option will also potentially produce smaller runtime code 47729c71111SAndy Whitcroft# with gcc 3.4 and later. 478a5ee6daaSGeoff Levand# 479a5ee6daaSGeoff Levandconfig SPARSEMEM_STATIC 480a5ee6daaSGeoff Levand bool 481a5ee6daaSGeoff Levand 482a5ee6daaSGeoff Levand# 483a5ee6daaSGeoff Levand# Architecture platforms which require a two level mem_section in SPARSEMEM 484a5ee6daaSGeoff Levand# must select this option. This is usually for architecture platforms with 4850b376f1eSAneesh Kumar K.V# an extremely sparse physical address space. 4860b376f1eSAneesh Kumar K.V# 4870b376f1eSAneesh Kumar K.Vconfig SPARSEMEM_EXTREME 4880b376f1eSAneesh Kumar K.V def_bool y 4890b6f1582SAneesh Kumar K.V depends on SPARSEMEM && !SPARSEMEM_STATIC 4900b6f1582SAneesh Kumar K.V 4910b6f1582SAneesh Kumar K.Vconfig SPARSEMEM_VMEMMAP_ENABLE 4920b6f1582SAneesh Kumar K.V bool 4930b376f1eSAneesh Kumar K.V 49429c71111SAndy Whitcroftconfig SPARSEMEM_VMEMMAP 49570210ed9SPhilipp Hachtmann bool "Sparse Memory virtual memmap" 4966341e62bSChristoph Jaeger depends on SPARSEMEM && SPARSEMEM_VMEMMAP_ENABLE 49770210ed9SPhilipp Hachtmann default y 49867a929e0SChristoph Hellwig help 499050a9adcSChristoph Hellwig SPARSEMEM_VMEMMAP uses a virtually mapped memmap to optimise 5006341e62bSChristoph Jaeger pfn_to_page and page_to_pfn operations. This is the most 5012667f50eSSteve Capper efficient option when sufficient kernel resources are available. 50252219aeaSDavid Hildenbrand# 50352219aeaSDavid Hildenbrand# Select this config option from the architecture Kconfig, if it is preferred 50452219aeaSDavid Hildenbrand# to enable the feature of HugeTLB/dev_dax vmemmap optimization. 505350e88baSMike Rapoport# 5066341e62bSChristoph Jaegerconfig ARCH_WANT_OPTIMIZE_DAX_VMEMMAP 507c378ddd5STejun Heo bool 5081e5d8e1eSDan Williams 5091e5d8e1eSDan Williamsconfig ARCH_WANT_OPTIMIZE_HUGETLB_VMEMMAP 5101e5d8e1eSDan Williams bool 5111e5d8e1eSDan Williams 512ee6f509cSMinchan Kimconfig HAVE_MEMBLOCK_PHYS_MAP 5136341e62bSChristoph Jaeger bool 514ee6f509cSMinchan Kim 515a9e7b8d4SDavid Hildenbrandconfig HAVE_FAST_GUP 516a9e7b8d4SDavid Hildenbrand depends on MMU 517a9e7b8d4SDavid Hildenbrand bool 518a9e7b8d4SDavid Hildenbrand 519a9e7b8d4SDavid Hildenbrand# Don't discard allocated memory used to track "memory" and "reserved" memblocks 520a9e7b8d4SDavid Hildenbrand# after early boot, so it can still be used to test for validity of memory. 521a9e7b8d4SDavid Hildenbrand# Also, memblocks are updated with memory hot(un)plug. 52246723bfaSYasuaki Ishimatsuconfig ARCH_KEEP_MEMBLOCK 52346723bfaSYasuaki Ishimatsu bool 52446723bfaSYasuaki Ishimatsu 52546723bfaSYasuaki Ishimatsu# Keep arch NUMA mapping infrastructure post-init. 52646723bfaSYasuaki Ishimatsuconfig NUMA_KEEP_MEMINFO 52746723bfaSYasuaki Ishimatsu bool 52846723bfaSYasuaki Ishimatsu 52991024b3cSAnshuman Khandualconfig MEMORY_ISOLATION 53091024b3cSAnshuman Khandual bool 53191024b3cSAnshuman Khandual 532519bcb79SJohannes Weiner# IORESOURCE_SYSTEM_RAM regions in the kernel resource tree that are marked 533519bcb79SJohannes Weiner# IORESOURCE_EXCLUSIVE cannot be mapped to user space, for example, via 534519bcb79SJohannes Weiner# /dev/mem. 5353947be19SDave Hansenconfig EXCLUSIVE_SYSTEM_RAM 536519bcb79SJohannes Weiner def_bool y 537519bcb79SJohannes Weiner depends on !DEVMEM || STRICT_DEVMEM 538b30c5927SDavid Hildenbrand 53971b6f2ddSDavid Hildenbrand# 54040b31360SStephen Rothwell# Only be set on architectures that have completely implemented memory hotplug 5417ec58a2bSDavid Hildenbrand# feature. If you are not sure, don't touch it. 5421e5d8e1eSDan Williams# 5433947be19SDave Hansenconfig HAVE_BOOTMEM_INFO_NODE 544519bcb79SJohannes Weiner def_bool n 545519bcb79SJohannes Weiner 5468604d9e5SVitaly Kuznetsovconfig ARCH_ENABLE_MEMORY_HOTPLUG 5478604d9e5SVitaly Kuznetsov bool 5488604d9e5SVitaly Kuznetsov 5498604d9e5SVitaly Kuznetsovconfig ARCH_ENABLE_MEMORY_HOTREMOVE 5508604d9e5SVitaly Kuznetsov bool 5518604d9e5SVitaly Kuznetsov 5528604d9e5SVitaly Kuznetsov# eventually, we can have this option just 'select SPARSEMEM' 5538604d9e5SVitaly Kuznetsovmenuconfig MEMORY_HOTPLUG 554cb1aaebeSMauro Carvalho Chehab bool "Memory hotplug" 5558604d9e5SVitaly Kuznetsov select MEMORY_ISOLATION 5568604d9e5SVitaly Kuznetsov depends on SPARSEMEM 5578604d9e5SVitaly Kuznetsov depends on ARCH_ENABLE_MEMORY_HOTPLUG 5588604d9e5SVitaly Kuznetsov depends on 64BIT 5598604d9e5SVitaly Kuznetsov select NUMA_KEEP_MEMINFO if NUMA 5608604d9e5SVitaly Kuznetsov 5610c0e6195SKAMEZAWA Hiroyukiif MEMORY_HOTPLUG 5620c0e6195SKAMEZAWA Hiroyuki 563f7e3334aSNathan Fontenotconfig MEMORY_HOTPLUG_DEFAULT_ONLINE 5640c0e6195SKAMEZAWA Hiroyuki bool "Online the newly added memory blocks by default" 5650c0e6195SKAMEZAWA Hiroyuki depends on MEMORY_HOTPLUG 5660c0e6195SKAMEZAWA Hiroyuki help 567a08a2ae3SOscar Salvador This option sets the default policy setting for memory hotplug 568a08a2ae3SOscar Salvador onlining policy (/sys/devices/system/memory/auto_online_blocks) which 569a08a2ae3SOscar Salvador determines what happens to newly added memory regions. Policy setting 570a08a2ae3SOscar Salvador can always be changed at runtime. 571a08a2ae3SOscar Salvador See Documentation/admin-guide/mm/memory-hotplug.rst for more information. 572519bcb79SJohannes Weiner 573519bcb79SJohannes Weiner Say Y here if you want all hot-plugged memory blocks to appear in 574*04d5ea46SAneesh Kumar K.V 'online' state by default. 575*04d5ea46SAneesh Kumar K.V Say N here if you want the default policy to keep all hot-plugged 576*04d5ea46SAneesh Kumar K.V memory blocks in 'offline' state. 5774c21e2f2SHugh Dickins 5784c21e2f2SHugh Dickinsconfig MEMORY_HOTREMOVE 5794c21e2f2SHugh Dickins bool "Allow for memory hot remove" 5804c21e2f2SHugh Dickins select HAVE_BOOTMEM_INFO_NODE if (X86_64 || PPC64) 5814c21e2f2SHugh Dickins depends on MEMORY_HOTPLUG && ARCH_ENABLE_MEMORY_HOTREMOVE 5827b6ac9dfSHugh Dickins depends on MIGRATION 58360bccaa6SWill Deacon 58460bccaa6SWill Deaconconfig MHP_MEMMAP_ON_MEMORY 58560bccaa6SWill Deacon def_bool y 586a70caa8bSHugh Dickins depends on MEMORY_HOTPLUG && SPARSEMEM_VMEMMAP 5874c21e2f2SHugh Dickins depends on ARCH_MHP_MEMMAP_ON_MEMORY_ENABLE 5884c21e2f2SHugh Dickins 5894c21e2f2SHugh Dickinsendif # MEMORY_HOTPLUG 5909164550eSKirill A. Shutemov 591a70caa8bSHugh Dickinsconfig ARCH_MHP_MEMMAP_ON_MEMORY_ENABLE 592a70caa8bSHugh Dickins bool 59360bccaa6SWill Deacon 5944c21e2f2SHugh Dickins# Heavily threaded applications may benefit from splitting the mm-wide 5957cbe34cfSChristoph Lameter# page_table_lock, so that faults on different parts of the user address 596e009bb30SKirill A. Shutemov# space can be handled with less contention: split it at this NR_CPUS. 5976341e62bSChristoph Jaeger# Default to 4 for wider testing, though 8 might be more appropriate. 598e009bb30SKirill A. Shutemov# ARM's adjust_pte (unused if VIPT) depends on mm-wide page_table_lock. 5997cbe34cfSChristoph Lameter# PA-RISC 7xxx's spinlock_t would enlarge struct page from 32 to 44 bytes. 60009316c09SKonstantin Khlebnikov# SPARC32 allocates multiple pte tables within a single page, and therefore 60109316c09SKonstantin Khlebnikov# a per-page lock leads to problems when multiple tables need to be locked 6026341e62bSChristoph Jaeger# at the same time (e.g. copy_page_range()). 60309316c09SKonstantin Khlebnikov# DEBUG_SPINLOCK and DEBUG_LOCK_ALLOC spinlock_t also enlarge struct page. 60409316c09SKonstantin Khlebnikov# 60518468d93SRafael Aquiniconfig SPLIT_PTLOCK_CPUS 60618468d93SRafael Aquini int 60718468d93SRafael Aquini default "999999" if !MMU 60818468d93SRafael Aquini default "999999" if ARM && !CPU_CACHE_VIPT 60909316c09SKonstantin Khlebnikov default "999999" if PARISC && !PA20 61018468d93SRafael Aquini default "999999" if SPARC32 61118468d93SRafael Aquini default "4" 61218468d93SRafael Aquini 61318468d93SRafael Aquiniconfig ARCH_ENABLE_SPLIT_PMD_PTLOCK 61418468d93SRafael Aquini bool 61518468d93SRafael Aquini 61618468d93SRafael Aquini# 61718468d93SRafael Aquini# support for memory balloon 61818468d93SRafael Aquiniconfig MEMORY_BALLOON 61918468d93SRafael Aquini bool 620e9e96b39SMel Gorman 621e9e96b39SMel Gorman# 622e9e96b39SMel Gorman# support for memory balloon compaction 62305106e6aSRik van Rielconfig BALLOON_COMPACTION 624e9e96b39SMel Gorman bool "Allow for balloon memory compaction/migration" 62533a93877SAndrea Arcangeli def_bool y 626e9e96b39SMel Gorman depends on COMPACTION && MEMORY_BALLOON 627b32eaf71SMichal Hocko help 628b32eaf71SMichal Hocko Memory fragmentation introduced by ballooning might reduce 629b32eaf71SMichal Hocko significantly the number of 2MB contiguous memory blocks that can be 630b32eaf71SMichal Hocko used within a guest, thus imposing performance penalties associated 631b32eaf71SMichal Hocko with the reduced number of transparent huge pages that could be used 632b32eaf71SMichal Hocko by the guest workload. Allowing the compaction & migration for memory 633b32eaf71SMichal Hocko pages enlisted as being part of memory balloon devices avoids the 634b32eaf71SMichal Hocko scenario aforementioned and helps improving memory defragmentation. 635e9e96b39SMel Gorman 636c7e0b3d0SThomas Gleixner# 637c7e0b3d0SThomas Gleixner# support for memory compaction 638c7e0b3d0SThomas Gleixnerconfig COMPACTION 639c7e0b3d0SThomas Gleixner bool "Allow for memory compaction" 640c7e0b3d0SThomas Gleixner def_bool y 641c7e0b3d0SThomas Gleixner select MIGRATION 642e9e96b39SMel Gorman depends on MMU 64336e66c55SAlexander Duyck help 64436e66c55SAlexander Duyck Compaction is the only memory management component to form 64536e66c55SAlexander Duyck high order (larger physically contiguous) memory blocks 64636e66c55SAlexander Duyck reliably. The page allocator relies on compaction heavily and 64736e66c55SAlexander Duyck the lack of the feature can lead to unexpected OOM killer 64836e66c55SAlexander Duyck invocations for high order memory requests. You shouldn't 64936e66c55SAlexander Duyck disable this option unless there really is a strong reason for 65036e66c55SAlexander Duyck it and then we would be really interested to hear about that at 65136e66c55SAlexander Duyck linux-mm@kvack.org. 65236e66c55SAlexander Duyck 65336e66c55SAlexander Duyckconfig COMPACT_UNEVICTABLE_DEFAULT 6547cbe34cfSChristoph Lameter int 6557cbe34cfSChristoph Lameter depends on COMPACTION 6567cbe34cfSChristoph Lameter default 0 if PREEMPT_RT 657b20a3503SChristoph Lameter default 1 6586c5240aeSChristoph Lameter 659de32a817SChen Gang# 660b20a3503SChristoph Lameter# support for free page reporting 661b20a3503SChristoph Lameterconfig PAGE_REPORTING 662e9e96b39SMel Gorman bool "Free page reporting" 663e9e96b39SMel Gorman def_bool n 664e9e96b39SMel Gorman help 665e9e96b39SMel Gorman Free page reporting allows for the incremental acquisition of 666e9e96b39SMel Gorman free pages from the buddy allocator for the purpose of reporting 6676550e07fSGreg Kroah-Hartman those pages to another entity, such as a hypervisor, so that the 66876cbbeadSChristoph Hellwig memory can be freed within the host for other uses. 669d90a25f8SChristoph Hellwig 67076cbbeadSChristoph Hellwig# 671c177c81eSNaoya Horiguchi# support for page migration 6726341e62bSChristoph Jaeger# 673c177c81eSNaoya Horiguchiconfig MIGRATION 6749c670ea3SNaoya Horiguchi bool "Page migration" 6759c670ea3SNaoya Horiguchi def_bool y 6769c670ea3SNaoya Horiguchi depends on (NUMA || ARCH_ENABLE_MEMORY_HOTREMOVE || COMPACTION || CMA) && MMU 6774bfb68a0SAnshuman Khandual help 6784bfb68a0SAnshuman Khandual Allows the migration of the physical location of pages of processes 6794bfb68a0SAnshuman Khandual while the virtual addresses are not changed. This is useful in 6804bfb68a0SAnshuman Khandual two situations. The first is on NUMA systems to put pages nearer 6814bfb68a0SAnshuman Khandual to the processors accessing. The second is when allocating huge 6824bfb68a0SAnshuman Khandual pages as migration can relocate pages to satisfy a huge page 6834bfb68a0SAnshuman Khandual allocation instead of reclaiming. 68423baf831SKirill A. Shutemov 68523baf831SKirill A. Shutemovconfig DEVICE_MIGRATION 686b3d40a2bSDavid Hildenbrand def_bool MIGRATION && ZONE_DEVICE 6878df995f6SAlexandre Ghiti 6888df995f6SAlexandre Ghiticonfig ARCH_ENABLE_HUGEPAGE_MIGRATION 6898df995f6SAlexandre Ghiti bool 690600715dcSJeremy Fitzhardinge 691d4a451d5SChristoph Hellwigconfig ARCH_ENABLE_THP_MIGRATION 692600715dcSJeremy Fitzhardinge bool 6932a7326b5SChristoph Lameter 6949ca24e2eSVinayak Menonconfig HUGETLB_PAGE_SIZE_VARIABLE 6959ca24e2eSVinayak Menon def_bool n 696ce288e05SChristoph Hellwig help 6979ca24e2eSVinayak Menon Allows the pageblock_order value to be dynamic instead of just standard 698ce288e05SChristoph Hellwig HUGETLB_PAGE_ORDER when there are multiple HugeTLB page sizes available 699ce288e05SChristoph Hellwig on a platform. 700ce288e05SChristoph Hellwig 7012a7326b5SChristoph Lameter Note that the pageblock_order cannot exceed MAX_ORDER and will be 702cddb8a5cSAndrea Arcangeli clamped down to MAX_ORDER. 703cddb8a5cSAndrea Arcangeli 70499cb252fSJason Gunthorpeconfig CONTIG_ALLOC 705fc4d5c29SDavid Howells def_bool (MEMORY_ISOLATION && COMPACTION) || CMA 706f8af4da3SHugh Dickins 707f8af4da3SHugh Dickinsconfig PHYS_ADDR_T_64BIT 708f8af4da3SHugh Dickins def_bool 64BIT 70959e1a2f4STimofey Titovets 710f8af4da3SHugh Dickinsconfig BOUNCE 711f8af4da3SHugh Dickins bool "Enable bounce buffers" 712f8af4da3SHugh Dickins default y 713f8af4da3SHugh Dickins depends on BLOCK && MMU && HIGHMEM 714d0f209f6SHugh Dickins help 715f8af4da3SHugh Dickins Enable bounce buffers for devices that cannot access the full range of 716f8af4da3SHugh Dickins memory available to the CPU. Enabled by default when HIGHMEM is 717ee65728eSMike Rapoport selected, but you may say n to override this. 718c73602adSHugh Dickins 719c73602adSHugh Dickinsconfig MMU_NOTIFIER 720f8af4da3SHugh Dickins bool 721e0a94c2aSChristoph Lameter select INTERVAL_TREE 722e0a94c2aSChristoph Lameter 7236e141546SDavid Howellsconfig KSM 724e0a94c2aSChristoph Lameter bool "Enable KSM for page merging" 725e0a94c2aSChristoph Lameter depends on MMU 726e0a94c2aSChristoph Lameter select XXHASH 727e0a94c2aSChristoph Lameter help 728e0a94c2aSChristoph Lameter Enable Kernel Samepage Merging: KSM periodically scans those areas 729e0a94c2aSChristoph Lameter of an application's address space that an app has advised may be 730e0a94c2aSChristoph Lameter mergeable. When it finds pages of identical content, it replaces 731e0a94c2aSChristoph Lameter the many instances by a single page with that content, so 732e0a94c2aSChristoph Lameter saving memory until one or another app needs to modify the content. 733788084abSEric Paris Recommended for use with KVM, or with other duplicative applications. 734788084abSEric Paris See Documentation/mm/ksm.rst for more information: KSM is inactive 735788084abSEric Paris until a program has madvised that an area is MADV_MERGEABLE, and 736e0a94c2aSChristoph Lameter root has set /sys/kernel/mm/ksm/run to 1 (if CONFIG_SYSFS is set). 737e0a94c2aSChristoph Lameter 738e0a94c2aSChristoph Lameterconfig DEFAULT_MMAP_MIN_ADDR 739e0a94c2aSChristoph Lameter int "Low address space to protect from user allocation" 740d949f36fSLinus Torvalds depends on MMU 741d949f36fSLinus Torvalds default 4096 742e0a94c2aSChristoph Lameter help 7436a46079cSAndi Kleen This is the portion of low virtual memory which should be protected 7446a46079cSAndi Kleen from userspace allocation. Keeping a user from writing to low pages 745d949f36fSLinus Torvalds can help reduce the impact of kernel NULL pointer bugs. 7466a46079cSAndi Kleen 747ee6f509cSMinchan Kim For most ia64, ppc64 and x86 users with lots of address space 74897f0b134SXie XiuQi a value of 65536 is reasonable and should cause no problems. 7496a46079cSAndi Kleen On arm and other archs it should not be higher than 32768. 7506a46079cSAndi Kleen Programs which use vm86 functionality or have some need to map 7516a46079cSAndi Kleen this low address space will need CAP_SYS_RAWIO or disable this 7526a46079cSAndi Kleen protection by setting the value to 0. 7536a46079cSAndi Kleen 7546a46079cSAndi Kleen This value can be changed after boot using the 755cae681fcSAndi Kleen /proc/sys/vm/mmap_min_addr tunable. 756413f9efbSAndi Kleen 75727df5068SAndi Kleenconfig ARCH_SUPPORTS_MEMORY_FAILURE 758478c5ffcSWu Fengguang bool 759cae681fcSAndi Kleen 760fc4d5c29SDavid Howellsconfig MEMORY_FAILURE 761fc4d5c29SDavid Howells depends on MMU 762fc4d5c29SDavid Howells depends on ARCH_SUPPORTS_MEMORY_FAILURE 763fc4d5c29SDavid Howells bool "Enable recovery from hardware memory errors" 764fc4d5c29SDavid Howells select MEMORY_ISOLATION 765fc4d5c29SDavid Howells select RAS 766fc4d5c29SDavid Howells help 767fc4d5c29SDavid Howells Enables code to recover from some memory failures on systems 768fc4d5c29SDavid Howells with MCA recovery. This allows a system to continue running 769fc4d5c29SDavid Howells even when some of its memory has uncorrected errors. This requires 770fc4d5c29SDavid Howells special hardware support and typically ECC memory. 771fc4d5c29SDavid Howells 772fc4d5c29SDavid Howellsconfig HWPOISON_INJECT 773fc4d5c29SDavid Howells tristate "HWPoison pages injector" 774fc4d5c29SDavid Howells depends on MEMORY_FAILURE && DEBUG_KERNEL && PROC_FS 775fc4d5c29SDavid Howells select PROC_PAGE_MONITOR 776fc4d5c29SDavid Howells 777fc4d5c29SDavid Howellsconfig NOMMU_INITIAL_TRIM_EXCESS 778fc4d5c29SDavid Howells int "Turn on mmap() excess space trimming before booting" 779fc4d5c29SDavid Howells depends on !MMU 780fc4d5c29SDavid Howells default 1 781fc4d5c29SDavid Howells help 782fc4d5c29SDavid Howells The NOMMU mmap() frequently needs to allocate large contiguous chunks 783fc4d5c29SDavid Howells of memory on which to store mappings, but it can only ask the system 784fc4d5c29SDavid Howells allocator for chunks in 2^N*PAGE_SIZE amounts - which is frequently 785fc4d5c29SDavid Howells more than it requires. To deal with this, mmap() is able to trim off 786dd19d293SStephen Kitt the excess and return it to the allocator. 787bbddff05STejun Heo 788519bcb79SJohannes Weiner If trimming is enabled, the excess is trimmed off and returned to the 789519bcb79SJohannes Weiner system allocator, which can cause extra fragmentation, particularly 790519bcb79SJohannes Weiner if there are a lot of transient processes. 791519bcb79SJohannes Weiner 792519bcb79SJohannes Weiner If trimming is disabled, the excess is kept, but not used, which for 793519bcb79SJohannes Weiner long-term mappings means that the space is wasted. 794519bcb79SJohannes Weiner 79513ece886SAndrea Arcangeli Trimming can be dynamically controlled through a sysctl option 796554b0f3cSSebastian Andrzej Siewior (/proc/sys/vm/nr_trim_pages) which specifies the minimum number of 7975d689240SAndrea Arcangeli excess pages there must be before trimming should occur, or zero if 7983a08cd52SMatthew Wilcox no trimming is to occur. 7994c76d9d1SAndrea Arcangeli 8004c76d9d1SAndrea Arcangeli This option specifies the initial value of this option. The default 8014c76d9d1SAndrea Arcangeli of 1 says that all excess pages should be trimmed. 8024c76d9d1SAndrea Arcangeli 8034c76d9d1SAndrea Arcangeli See Documentation/admin-guide/mm/nommu-mmap.rst for more information. 8044c76d9d1SAndrea Arcangeli 8054c76d9d1SAndrea Arcangeliconfig ARCH_WANT_GENERAL_HUGETLB 8064c76d9d1SAndrea Arcangeli bool 8074c76d9d1SAndrea Arcangeli 8084c76d9d1SAndrea Arcangeliconfig ARCH_WANTS_THP_SWAP 809519bcb79SJohannes Weiner def_bool n 810519bcb79SJohannes Weiner 81113ece886SAndrea Arcangelimenuconfig TRANSPARENT_HUGEPAGE 81213ece886SAndrea Arcangeli bool "Transparent Hugepage Support" 81313ece886SAndrea Arcangeli depends on HAVE_ARCH_TRANSPARENT_HUGEPAGE && !PREEMPT_RT 81413ece886SAndrea Arcangeli select COMPACTION 81513ece886SAndrea Arcangeli select XARRAY_MULTI 81613ece886SAndrea Arcangeli help 81713ece886SAndrea Arcangeli Transparent Hugepages allows the kernel to use huge pages and 81813ece886SAndrea Arcangeli huge tlb transparently to the applications whenever possible. 81913ece886SAndrea Arcangeli This feature can improve computing performance to certain 82013ece886SAndrea Arcangeli applications by speeding up page faults during memory 82113ece886SAndrea Arcangeli allocation, by reducing the number of tlb misses and by speeding 82213ece886SAndrea Arcangeli up the pagetable walking. 82313ece886SAndrea Arcangeli 82413ece886SAndrea Arcangeli If memory constrained on embedded, you may want to say N. 82513ece886SAndrea Arcangeli 82613ece886SAndrea Arcangeliif TRANSPARENT_HUGEPAGE 82713ece886SAndrea Arcangeli 82813ece886SAndrea Arcangelichoice 82913ece886SAndrea Arcangeli prompt "Transparent Hugepage Support sysfs defaults" 83013ece886SAndrea Arcangeli depends on TRANSPARENT_HUGEPAGE 83113ece886SAndrea Arcangeli default TRANSPARENT_HUGEPAGE_ALWAYS 83213ece886SAndrea Arcangeli help 83313ece886SAndrea Arcangeli Selects the sysfs defaults for Transparent Hugepage Support. 83413ece886SAndrea Arcangeli 83538d8b4e6SHuang Ying config TRANSPARENT_HUGEPAGE_ALWAYS 83638d8b4e6SHuang Ying bool "always" 837dad6a5ebSHugh Dickins help 83838d8b4e6SHuang Ying Enabling Transparent Hugepage always, can increase the 83938d8b4e6SHuang Ying memory footprint of applications without a guaranteed 84014fef284SHuang Ying benefit but it will work automatically for all applications. 84114fef284SHuang Ying 84238d8b4e6SHuang Ying config TRANSPARENT_HUGEPAGE_MADVISE 84338d8b4e6SHuang Ying bool "madvise" 84438d8b4e6SHuang Ying help 845519bcb79SJohannes Weiner Enabling Transparent Hugepage madvise, will only provide a 846519bcb79SJohannes Weiner performance improvement benefit to the applications using 847519bcb79SJohannes Weiner madvise(MADV_HUGEPAGE) but it won't risk to increase the 848519bcb79SJohannes Weiner memory footprint of applications without a guaranteed 849519bcb79SJohannes Weiner benefit. 850519bcb79SJohannes Weinerendchoice 851519bcb79SJohannes Weiner 852519bcb79SJohannes Weinerconfig THP_SWAP 853519bcb79SJohannes Weiner def_bool y 854519bcb79SJohannes Weiner depends on TRANSPARENT_HUGEPAGE && ARCH_WANTS_THP_SWAP && SWAP && 64BIT 855519bcb79SJohannes Weiner help 856519bcb79SJohannes Weiner Swap transparent huge pages in one piece, without splitting. 857519bcb79SJohannes Weiner XXX: For now, swap cluster backing transparent huge page 858e496cf3dSKirill A. Shutemov will be split after swapout. 859bbddff05STejun Heo 860bbddff05STejun Heo For selection by architectures with reasonable THP sizes. 861bbddff05STejun Heo 8623583521aSVladimir Murzinconfig READ_ONLY_THP_FOR_FS 863bbddff05STejun Heo bool "Read-only THP for filesystems (EXPERIMENTAL)" 864bbddff05STejun Heo depends on TRANSPARENT_HUGEPAGE && SHMEM 865077b1f83SDan Magenheimer 8667ecd19cfSKefeng Wang help 8677ecd19cfSKefeng Wang Allow khugepaged to put read-only file-backed pages in THP. 8687ecd19cfSKefeng Wang 8697ecd19cfSKefeng Wang This is marked experimental because it is a new feature. Write 8707ecd19cfSKefeng Wang support of file THPs will be developed in the next few release 8717ecd19cfSKefeng Wang cycles. 8727ecd19cfSKefeng Wang 8737ecd19cfSKefeng Wangendif # TRANSPARENT_HUGEPAGE 8747ecd19cfSKefeng Wang 8757ecd19cfSKefeng Wang# 8767ecd19cfSKefeng Wang# UP and nommu archs use km based percpu allocator 8777ecd19cfSKefeng Wang# 878f825c736SAneesh Kumar K.Vconfig NEED_PER_CPU_KM 879f825c736SAneesh Kumar K.V depends on !SMP || !MMU 880aca52c39SMike Rapoport bool 881f825c736SAneesh Kumar K.V default y 882f825c736SAneesh Kumar K.V 883f825c736SAneesh Kumar K.Vconfig NEED_PER_CPU_EMBED_FIRST_CHUNK 884f825c736SAneesh Kumar K.V bool 885f825c736SAneesh Kumar K.V 886f825c736SAneesh Kumar K.Vconfig NEED_PER_CPU_PAGE_FIRST_CHUNK 887f825c736SAneesh Kumar K.V bool 888f825c736SAneesh Kumar K.V 889f825c736SAneesh Kumar K.Vconfig USE_PERCPU_NUMA_NODE_ID 890f825c736SAneesh Kumar K.V bool 891f825c736SAneesh Kumar K.V 892f825c736SAneesh Kumar K.Vconfig HAVE_SETUP_PER_CPU_AREA 893f825c736SAneesh Kumar K.V bool 894f825c736SAneesh Kumar K.V 895f825c736SAneesh Kumar K.Vconfig CMA 896f825c736SAneesh Kumar K.V bool "Contiguous Memory Allocator" 897f825c736SAneesh Kumar K.V depends on MMU 898f825c736SAneesh Kumar K.V select MIGRATION 899f825c736SAneesh Kumar K.V select MEMORY_ISOLATION 900f825c736SAneesh Kumar K.V help 901bf550fc9SAlexander Graf This enables the Contiguous Memory Allocator which allows other 90228b24c1fSSasha Levin subsystems to allocate big physically-contiguous blocks of memory. 90328b24c1fSSasha Levin CMA reserves a region of memory and allows only movable pages to 90428b24c1fSSasha Levin be allocated from it. This way, the kernel can use the memory for 90528b24c1fSSasha Levin pagecache and when a subsystem requests for contiguous area, the 90628b24c1fSSasha Levin allocated pages are migrated away to serve the contiguous request. 90728b24c1fSSasha Levin 90843ca106fSMinchan Kim If unsure, say "n". 90943ca106fSMinchan Kim 91043ca106fSMinchan Kimconfig CMA_DEBUG 91143ca106fSMinchan Kim bool "CMA debug messages (DEVELOPMENT)" 91243ca106fSMinchan Kim depends on DEBUG_KERNEL && CMA 91343ca106fSMinchan Kim help 91443ca106fSMinchan Kim Turns on debug messages in CMA. This produces KERN_DEBUG 915a254129eSJoonsoo Kim messages for every CMA call as well as various messages while 916a254129eSJoonsoo Kim processing calls such as dma_alloc_from_contiguous(). 917a254129eSJoonsoo Kim This option does not affect warning and error messages. 918b7176c26SBarry Song 919a254129eSJoonsoo Kimconfig CMA_DEBUGFS 920a254129eSJoonsoo Kim bool "CMA debugfs interface" 921a254129eSJoonsoo Kim depends on CMA && DEBUG_FS 922a254129eSJoonsoo Kim help 923a254129eSJoonsoo Kim Turns on the DebugFS interface for CMA. 924a254129eSJoonsoo Kim 925b7176c26SBarry Songconfig CMA_SYSFS 926a254129eSJoonsoo Kim bool "CMA information through sysfs interface" 927af8d417aSDan Streetman depends on CMA && SYSFS 928af8d417aSDan Streetman help 929af8d417aSDan Streetman This option exposes some sysfs attributes to get information 930af8d417aSDan Streetman from CMA. 9314e2e2770SSeth Jennings 932af8d417aSDan Streetmanconfig CMA_AREAS 933af8d417aSDan Streetman int "Maximum count of the CMA areas" 934af8d417aSDan Streetman depends on CMA 935af8d417aSDan Streetman default 19 if NUMA 936af8d417aSDan Streetman default 7 9371ad1335dSMike Rapoport help 9384e2e2770SSeth Jennings CMA allows to create CMA areas for particular purpose, mainly, 9399e5c33d7SMark Salter used as device private area. This parameter sets the maximum 9409e5c33d7SMark Salter number of CMA area in the system. 941042d27acSHelge Deller 94222ee3ea5SHelge Deller If unsure, leave the default value "7" in UMA and "19" in NUMA. 94322ee3ea5SHelge Deller 94422ee3ea5SHelge Dellerconfig MEM_SOFT_DIRTY 945042d27acSHelge Deller bool "Track memory changes" 946042d27acSHelge Deller depends on CHECKPOINT_RESTORE && HAVE_ARCH_SOFT_DIRTY && PROC_FS 947042d27acSHelge Deller select PROC_PAGE_MONITOR 948042d27acSHelge Deller help 949042d27acSHelge Deller This option enables memory changes tracking by introducing a 95022ee3ea5SHelge Deller soft-dirty bit on pte-s. This bit it set when someone writes 951042d27acSHelge Deller into a page just as regular dirty bit, but unlike the latter 95222ee3ea5SHelge Deller it can be cleared by hands. 9533a80a7faSMel Gorman 9543a80a7faSMel Gorman See Documentation/admin-guide/mm/soft-dirty.rst for more details. 9551ce22103SVlastimil Babka 956d39f8fb4SMike Rapoportconfig GENERIC_EARLY_IOREMAP 957ab1e8d89SPavel Tatashin bool 958889c695dSPasha Tatashin 959e4443149SDaniel Jordanconfig STACK_MAX_DEFAULT_SIZE_MB 9603a80a7faSMel Gorman int "Default maximum user stack size for 32-bit processes (MB)" 9613a80a7faSMel Gorman default 100 9623a80a7faSMel Gorman range 8 2048 9633a80a7faSMel Gorman depends on STACK_GROWSUP && (!64BIT || COMPAT) 964e4443149SDaniel Jordan help 965e4443149SDaniel Jordan This is the maximum stack size in Megabytes in the VM layout of 32-bit 9661ce22103SVlastimil Babka user processes when the stack grows upwards (currently only on parisc 9671ce22103SVlastimil Babka arch) when the RLIMIT_STACK hard limit is unlimited. 968033fbae9SDan Williams 9691c676e0dSSeongJae Park A sane initial value is 100 MB. 9701c676e0dSSeongJae Park 9711c676e0dSSeongJae Parkconfig DEFERRED_STRUCT_PAGE_INIT 9721c676e0dSSeongJae Park bool "Defer initialisation of struct pages to kthreads" 9731c676e0dSSeongJae Park depends on SPARSEMEM 9741c676e0dSSeongJae Park depends on !NEED_PER_CPU_KM 9751c676e0dSSeongJae Park depends on 64BIT 9761c676e0dSSeongJae Park select PADATA 97733c3fc71SVladimir Davydov help 97833c3fc71SVladimir Davydov Ordinarily all struct pages are initialised during early boot in a 97933c3fc71SVladimir Davydov single thread. On very large machines this can take a considerable 9801c676e0dSSeongJae Park amount of time. If this option is set, large machines will bring up 98133c3fc71SVladimir Davydov a subset of memmap at boot and then initialise the rest in parallel. 98233c3fc71SVladimir Davydov This has a potential performance impact on tasks running early in the 98333c3fc71SVladimir Davydov lifetime of the system until these kthreads finish the 98433c3fc71SVladimir Davydov initialisation. 98533c3fc71SVladimir Davydov 98633c3fc71SVladimir Davydovconfig PAGE_IDLE_FLAG 9871ad1335dSMike Rapoport bool 9881ad1335dSMike Rapoport select PAGE_EXTENSION if !64BIT 98933c3fc71SVladimir Davydov help 990c2280be8SAnshuman Khandual This adds PG_idle and PG_young flags to 'struct page'. PTE Accessed 991c2280be8SAnshuman Khandual bit writers can set the state of the bit in the flags so that PTE 992c2280be8SAnshuman Khandual Accessed bit readers may avoid disturbance. 9932792d84eSKees Cook 9942792d84eSKees Cookconfig IDLE_PAGE_TRACKING 9952792d84eSKees Cook bool "Enable idle page tracking" 9962792d84eSKees Cook depends on SYSFS && MMU 9972792d84eSKees Cook select PAGE_IDLE_FLAG 9982792d84eSKees Cook help 9992792d84eSKees Cook This feature allows to estimate the amount of user pages that have 10002792d84eSKees Cook not been touched during a given period of time. This information can 10012792d84eSKees Cook be useful to tune memory cgroup limits and/or for job placement 100217596731SRobin Murphy within a compute cluster. 100365f7d049SOliver O'Halloran 100465f7d049SOliver O'Halloran See Documentation/admin-guide/mm/idle_page_tracking.rst for 100563703f37SKefeng Wang more details. 100663703f37SKefeng Wang 100763703f37SKefeng Wangconfig ARCH_HAS_CACHE_LINE_SIZE 100863703f37SKefeng Wang bool 100963703f37SKefeng Wang 101063703f37SKefeng Wangconfig ARCH_HAS_CURRENT_STACK_POINTER 101163703f37SKefeng Wang bool 101263703f37SKefeng Wang help 101363703f37SKefeng Wang In support of HARDENED_USERCOPY performing stack variable lifetime 101463703f37SKefeng Wang checking, an architecture-agnostic way to find the stack pointer 101563703f37SKefeng Wang is needed. Once an architecture defines an unsigned long global 101663703f37SKefeng Wang register alias named "current_stack_pointer", this config can be 1017033fbae9SDan Williams selected. 10185042db43SJérôme Glisse 1019033fbae9SDan Williamsconfig ARCH_HAS_PTE_DEVMAP 1020033fbae9SDan Williams bool 102199490f16SDan Williams 102217596731SRobin Murphyconfig ARCH_HAS_ZONE_DMA_SET 10233a08cd52SMatthew Wilcox bool 1024033fbae9SDan Williams 1025033fbae9SDan Williamsconfig ZONE_DMA 1026033fbae9SDan Williams bool "Support DMA zone" if ARCH_HAS_ZONE_DMA_SET 1027033fbae9SDan Williams default y if ARM64 || X86 1028033fbae9SDan Williams 1029033fbae9SDan Williamsconfig ZONE_DMA32 1030033fbae9SDan Williams bool "Support DMA32 zone" if ARCH_HAS_ZONE_DMA_SET 1031033fbae9SDan Williams depends on !X86_32 1032033fbae9SDan Williams default y if ARM64 103306a660adSLinus Torvalds 10349c240a7bSChristoph Hellwigconfig ZONE_DEVICE 10359c240a7bSChristoph Hellwig bool "Device memory (pmem, HMM, etc...) hotplug support" 10369c240a7bSChristoph Hellwig depends on MEMORY_HOTPLUG 10379c240a7bSChristoph Hellwig depends on MEMORY_HOTREMOVE 1038c0b12405SJérôme Glisse depends on SPARSEMEM_VMEMMAP 10399c240a7bSChristoph Hellwig depends on ARCH_HAS_PTE_DEVMAP 1040f442c283SChristoph Hellwig select XARRAY_MULTI 1041c0b12405SJérôme Glisse 104214b80582SDan Williams help 104314b80582SDan Williams Device memory hotplug support allows for establishing pmem, 104414b80582SDan Williams or other device driver discovered memory regions, in the 104514b80582SDan Williams memmap. This allows pfn_to_page() lookups of otherwise 10465042db43SJérôme Glisse "device-physical" addresses which is needed for using a DAX 10475042db43SJérôme Glisse mapping in an O_DIRECT operation, among other things. 10487328d9ccSChristoph Hellwig 104914b80582SDan Williams If FS_DAX is enabled, then say Y. 10505042db43SJérôme Glisse 10515042db43SJérôme Glisse# 10525042db43SJérôme Glisse# Helpers to mirror range of the CPU page tables of a process into device page 10535042db43SJérôme Glisse# tables. 10545042db43SJérôme Glisse# 10555042db43SJérôme Glisseconfig HMM_MIRROR 10563e9a9e25SChristoph Hellwig bool 10573e9a9e25SChristoph Hellwig depends on MMU 10583e9a9e25SChristoph Hellwig 105963c17fb8SDave Hansenconfig GET_FREE_REGION 106063c17fb8SDave Hansen depends on SPARSEMEM 106166d37570SDave Hansen bool 106266d37570SDave Hansen 106330a5b536SDennis Zhouconfig DEVICE_PRIVATE 1064b0284cd2SCatalin Marinas bool "Unaddressable device memory (GPU memory, ...)" 1065b0284cd2SCatalin Marinas depends on ZONE_DEVICE 1066b0284cd2SCatalin Marinas select GET_FREE_REGION 1067b0284cd2SCatalin Marinas 1068b0284cd2SCatalin Marinas help 1069b0284cd2SCatalin Marinas Allows creation of struct pages to represent unaddressable device 1070b0284cd2SCatalin Marinas memory; i.e., memory that is only accessible from the device (or 1071b0284cd2SCatalin Marinas group of devices). You likely also want to select HMM_MIRROR. 10720710d012SVlastimil Babka 10730710d012SVlastimil Babkaconfig VMAP_PFN 10740710d012SVlastimil Babka bool 10750710d012SVlastimil Babka 10760710d012SVlastimil Babkaconfig ARCH_USES_HIGH_VMA_FLAGS 10770710d012SVlastimil Babka bool 10780710d012SVlastimil Babkaconfig ARCH_HAS_PKEYS 10790710d012SVlastimil Babka bool 10800710d012SVlastimil Babka 108130a5b536SDennis Zhouconfig ARCH_USES_PG_ARCH_X 108230a5b536SDennis Zhou bool 108330a5b536SDennis Zhou help 108430a5b536SDennis Zhou Enable the definition of PG_arch_x page flags with x > 1. Only 108530a5b536SDennis Zhou suitable for 64-bit architectures with CONFIG_FLATMEM or 108630a5b536SDennis Zhou CONFIG_SPARSEMEM_VMEMMAP enabled, otherwise there may not be 108764c349f4SKirill A. Shutemov enough room for additional bits in page->flags. 10889c84f229SJohn Hubbard 10899c84f229SJohn Hubbardconfig VM_EVENT_COUNTERS 1090d0de8241SBarry Song default y 109164c349f4SKirill A. Shutemov bool "Enable VM event counters for /proc/vmstat" if EXPERT 10929c84f229SJohn Hubbard help 10939c84f229SJohn Hubbard VM event counters are needed for event counts to be shown. 10949c84f229SJohn Hubbard This option allows the disabling of the VM event counters 109564c349f4SKirill A. Shutemov on EXPERT systems. /proc/vmstat will only show page counts 10969c84f229SJohn Hubbard if VM event counters are disabled. 10979c84f229SJohn Hubbard 10989c84f229SJohn Hubbardconfig PERCPU_STATS 10999c84f229SJohn Hubbard bool "Collect percpu memory statistics" 1100f4f9bda4SJohn Hubbard help 1101f4f9bda4SJohn Hubbard This feature collects and exposes statistics via debugfs. The 1102f4f9bda4SJohn Hubbard information includes global and per chunk statistics, which can 1103f4f9bda4SJohn Hubbard be used to help understand percpu memory usage. 1104f4f9bda4SJohn Hubbard 1105f4f9bda4SJohn Hubbardconfig GUP_TEST 1106baa489faSSeongJae Park bool "Enable infrastructure for get_user_pages()-related unit tests" 11073010a5eaSLaurent Dufour depends on DEBUG_FS 1108d0de8241SBarry Song help 1109d0de8241SBarry Song Provides /sys/kernel/debug/gup_test, which in turn provides a way 11103010a5eaSLaurent Dufour to make ioctl calls that can launch kernel-based unit tests for 11116ca297d4SPeter Zijlstra the get_user_pages*() and pin_user_pages*() family of API calls. 111239656e83SChristoph Hellwig 111339656e83SChristoph Hellwig These tests include benchmark testing of the _fast variants of 1114def85743SKeith Busch get_user_pages*() and pin_user_pages*(), as well as smoke tests of 1115def85743SKeith Busch the non-_fast variants. 1116def85743SKeith Busch 1117def85743SKeith Busch There is also a sub-test that allows running dump_page() on any 1118def85743SKeith Busch of up to eight pages (selected by command line args) within the 1119def85743SKeith Busch range of user-space addresses. These pages are either pinned via 1120def85743SKeith Busch pin_user_pages*(), or pinned via get_user_pages*(), as specified 1121def85743SKeith Busch by other command line arguments. 1122def85743SKeith Busch 11233010a5eaSLaurent Dufour See tools/testing/selftests/mm/gup_test.c 11243010a5eaSLaurent Dufour 112559e0b520SChristoph Hellwigcomment "GUP_TEST needs to have DEBUG_FS enabled" 1126cbd34da7SChristoph Hellwig depends on !GUP_TEST && !DEBUG_FS 1127cbd34da7SChristoph Hellwig 1128cbd34da7SChristoph Hellwigconfig GUP_GET_PXX_LOW_HIGH 1129cbd34da7SChristoph Hellwig bool 1130cbd34da7SChristoph Hellwig 1131cbd34da7SChristoph Hellwigconfig DMAPOOL_TEST 1132cbd34da7SChristoph Hellwig tristate "Enable a module to run time tests on dma_pool" 1133cbd34da7SChristoph Hellwig depends on HAS_DMA 1134cbd34da7SChristoph Hellwig help 1135cbd34da7SChristoph Hellwig Provides a test module that will allocate and free many blocks of 1136c5acad84SThomas Hellstrom various sizes and report how long it takes. This is intended to 1137c5acad84SThomas Hellstrom provide a consistent way to measure how changes to the 1138c5acad84SThomas Hellstrom dma_pool_alloc/free routines affect performance. 1139298fa1adSThomas Gleixner 1140298fa1adSThomas Gleixnerconfig ARCH_HAS_PTE_SPECIAL 1141298fa1adSThomas Gleixner bool 1142825c43f5SArd Biesheuvel 1143825c43f5SArd Biesheuvel# 1144825c43f5SArd Biesheuvel# Some architectures require a special hugepage directory format that is 11451fbaf8fcSChristoph Hellwig# required to support multiple hugepage sizes. For example a4fe3ce76 11461fbaf8fcSChristoph Hellwig# "powerpc/mm: Allow more flexible layouts for hugepage pagetables" 11471fbaf8fcSChristoph Hellwig# introduced it on powerpc. This allows for a more flexible hugepage 11481507f512SMike Rapoport# pagetable layouts. 1149626e98cbSThomas Weißschuh# 1150626e98cbSThomas Weißschuhconfig ARCH_HAS_HUGEPD 1151626e98cbSThomas Weißschuh bool 11521507f512SMike Rapoport 115374947724SLukas Bulwahnconfig MAPPING_DIRTY_HELPERS 115474947724SLukas Bulwahn bool 115574947724SLukas Bulwahn 115674947724SLukas Bulwahnconfig KMAP_LOCAL 115774947724SLukas Bulwahn bool 115874947724SLukas Bulwahn 115974947724SLukas Bulwahnconfig KMAP_LOCAL_NON_LINEAR_PTE_ARRAY 11601507f512SMike Rapoport bool 11619a10064fSColin Cross 11629a10064fSColin Cross# struct io_mapping based helper. Selected by drivers that need them 11639a10064fSColin Crossconfig IO_MAPPING 11649a10064fSColin Cross bool 11659a10064fSColin Cross 11669a10064fSColin Crossconfig MEMFD_CREATE 11679a10064fSColin Cross bool "Enable memfd_create() system call" if EXPERT 11689a10064fSColin Cross 11699a10064fSColin Crossconfig SECRETMEM 11709a10064fSColin Cross default y 11719a10064fSColin Cross bool "Enable memfd_secret() system call" if EXPERT 11729a10064fSColin Cross depends on ARCH_HAS_SET_DIRECT_MAP 11739a10064fSColin Cross help 11749a10064fSColin Cross Enable the memfd_secret() system call with the ability to create 1175430529b5SPeter Xu memory areas visible only in the context of the owning process and 1176430529b5SPeter Xu not mapped to other processes and other kernel page tables. 1177430529b5SPeter Xu 1178430529b5SPeter Xuconfig ANON_VMA_NAME 1179430529b5SPeter Xu bool "Anonymous VMA name support" 1180430529b5SPeter Xu depends on PROC_FS && ADVISE_SYSCALLS && MMU 1181430529b5SPeter Xu 1182430529b5SPeter Xu help 1183430529b5SPeter Xu Allow naming anonymous virtual memory areas. 1184430529b5SPeter Xu 1185430529b5SPeter Xu This feature allows assigning names to virtual memory areas. Assigned 1186430529b5SPeter Xu names can be later retrieved from /proc/pid/maps and /proc/pid/smaps 1187430529b5SPeter Xu and help identifying individual anonymous memory areas. 1188430529b5SPeter Xu Assigning a name to anonymous virtual memory area might prevent that 1189430529b5SPeter Xu area from being merged with adjacent virtual memory areas due to the 1190430529b5SPeter Xu difference in their name. 1191430529b5SPeter Xu 11921db9dbc2SPeter Xuconfig USERFAULTFD 119381e0f15fSPeter Xu bool "Enable userfaultfd() system call" 119481e0f15fSPeter Xu depends on MMU 119581e0f15fSPeter Xu help 11961db9dbc2SPeter Xu Enable the userfaultfd() system call that allows to intercept and 11971db9dbc2SPeter Xu handle page faults in userland. 11981db9dbc2SPeter Xu 11991db9dbc2SPeter Xuconfig HAVE_ARCH_USERFAULTFD_WP 12001db9dbc2SPeter Xu bool 12011db9dbc2SPeter Xu help 1202ac35a490SYu Zhao Arch has userfaultfd write protection support 1203ec1c86b2SYu Zhao 1204ec1c86b2SYu Zhaoconfig HAVE_ARCH_USERFAULTFD_MINOR 1205ec1c86b2SYu Zhao bool 1206ec1c86b2SYu Zhao help 1207ec1c86b2SYu Zhao Arch has userfaultfd minor fault support 1208ec1c86b2SYu Zhao 120907017acbSYu Zhaoconfig PTE_MARKER_UFFD_WP 121007017acbSYu Zhao bool "Userfaultfd write protection support for shmem/hugetlbfs" 1211ec1c86b2SYu Zhao default y 1212354ed597SYu Zhao depends on HAVE_ARCH_USERFAULTFD_WP 1213354ed597SYu Zhao 1214354ed597SYu Zhao help 1215354ed597SYu Zhao Allows to create marker PTEs for userfaultfd write protection 1216354ed597SYu Zhao purposes. It is required to enable userfaultfd write protection on 1217354ed597SYu Zhao file-backed memory types like shmem and hugetlbfs. 1218ac35a490SYu Zhao 1219ac35a490SYu Zhao# multi-gen LRU { 1220ac35a490SYu Zhaoconfig LRU_GEN 1221ac35a490SYu Zhao bool "Multi-Gen LRU" 1222ac35a490SYu Zhao depends on MMU 1223ac35a490SYu Zhao # make sure folio->flags has enough spare bits 1224ac35a490SYu Zhao depends on 64BIT || !SPARSEMEM || SPARSEMEM_VMEMMAP 1225ac35a490SYu Zhao help 1226ac35a490SYu Zhao A high performance LRU implementation to overcommit memory. See 1227ac35a490SYu Zhao Documentation/admin-guide/mm/multigen_lru.rst for details. 12280b6cc04fSSuren Baghdasaryan 12290b6cc04fSSuren Baghdasaryanconfig LRU_GEN_ENABLED 12300b6cc04fSSuren Baghdasaryan bool "Enable by default" 12310b6cc04fSSuren Baghdasaryan depends on LRU_GEN 12320b6cc04fSSuren Baghdasaryan help 12330b6cc04fSSuren Baghdasaryan This option enables the multi-gen LRU by default. 12340b6cc04fSSuren Baghdasaryan 12350b6cc04fSSuren Baghdasaryanconfig LRU_GEN_STATS 12360b6cc04fSSuren Baghdasaryan bool "Full stats for debugging" 12370b6cc04fSSuren Baghdasaryan depends on LRU_GEN 12380b6cc04fSSuren Baghdasaryan help 12390b6cc04fSSuren Baghdasaryan Do not enable this option unless you plan to look at historical stats 1240c2508ec5SLinus Torvalds from evicted generations for debugging purpose. 1241c2508ec5SLinus Torvalds 1242c2508ec5SLinus Torvalds This option has a per-memcg and per-node memory overhead. 1243c2508ec5SLinus Torvalds# } 12442224d848SSeongJae Park 12452224d848SSeongJae Parkconfig ARCH_SUPPORTS_PER_VMA_LOCK 124659e0b520SChristoph Hellwig def_bool n 1247 1248config PER_VMA_LOCK 1249 def_bool y 1250 depends on ARCH_SUPPORTS_PER_VMA_LOCK && MMU && SMP 1251 help 1252 Allow per-vma locking during page fault handling. 1253 1254 This feature allows locking each virtual memory area separately when 1255 handling page faults instead of taking mmap_lock. 1256 1257config LOCK_MM_AND_FIND_VMA 1258 bool 1259 depends on !STACK_GROWSUP 1260 1261source "mm/damon/Kconfig" 1262 1263endmenu 1264