159e0b520SChristoph Hellwig 259e0b520SChristoph Hellwigmenu "Memory Management options" 359e0b520SChristoph Hellwig 4e1785e85SDave Hansenconfig SELECT_MEMORY_MODEL 5e1785e85SDave Hansen def_bool y 6a8826eebSKees Cook depends on ARCH_SELECT_MEMORY_MODEL 7e1785e85SDave Hansen 83a9da765SDave Hansenchoice 93a9da765SDave Hansen prompt "Memory model" 10e1785e85SDave Hansen depends on SELECT_MEMORY_MODEL 11e1785e85SDave Hansen default DISCONTIGMEM_MANUAL if ARCH_DISCONTIGMEM_DEFAULT 12d41dee36SAndy Whitcroft default SPARSEMEM_MANUAL if ARCH_SPARSEMEM_DEFAULT 13e1785e85SDave Hansen default FLATMEM_MANUAL 14*d66d109dSMike Rapoport help 15*d66d109dSMike Rapoport This option allows you to change some of the ways that 16*d66d109dSMike Rapoport Linux manages its memory internally. Most users will 17*d66d109dSMike Rapoport only have one option here selected by the architecture 18*d66d109dSMike Rapoport configuration. This is normal. 193a9da765SDave Hansen 20e1785e85SDave Hansenconfig FLATMEM_MANUAL 213a9da765SDave Hansen bool "Flat Memory" 22c898ec16SAnton Blanchard depends on !(ARCH_DISCONTIGMEM_ENABLE || ARCH_SPARSEMEM_ENABLE) || ARCH_FLATMEM_ENABLE 233a9da765SDave Hansen help 24*d66d109dSMike Rapoport This option is best suited for non-NUMA systems with 25*d66d109dSMike Rapoport flat address space. The FLATMEM is the most efficient 26*d66d109dSMike Rapoport system in terms of performance and resource consumption 27*d66d109dSMike Rapoport and it is the best option for smaller systems. 283a9da765SDave Hansen 29*d66d109dSMike Rapoport For systems that have holes in their physical address 30*d66d109dSMike Rapoport spaces and for features like NUMA and memory hotplug, 31*d66d109dSMike Rapoport choose "Sparse Memory" 32d41dee36SAndy Whitcroft 33d41dee36SAndy Whitcroft If unsure, choose this option (Flat Memory) over any other. 343a9da765SDave Hansen 35e1785e85SDave Hansenconfig DISCONTIGMEM_MANUAL 36f3519f91SDave Hansen bool "Discontiguous Memory" 373a9da765SDave Hansen depends on ARCH_DISCONTIGMEM_ENABLE 383a9da765SDave Hansen help 39785dcd44SDave Hansen This option provides enhanced support for discontiguous 40785dcd44SDave Hansen memory systems, over FLATMEM. These systems have holes 41785dcd44SDave Hansen in their physical address spaces, and this option provides 42*d66d109dSMike Rapoport more efficient handling of these holes. 43785dcd44SDave Hansen 44*d66d109dSMike Rapoport Although "Discontiguous Memory" is still used by several 45*d66d109dSMike Rapoport architectures, it is considered deprecated in favor of 46*d66d109dSMike Rapoport "Sparse Memory". 47785dcd44SDave Hansen 48*d66d109dSMike Rapoport If unsure, choose "Sparse Memory" over this option. 493a9da765SDave Hansen 50d41dee36SAndy Whitcroftconfig SPARSEMEM_MANUAL 51d41dee36SAndy Whitcroft bool "Sparse Memory" 52d41dee36SAndy Whitcroft depends on ARCH_SPARSEMEM_ENABLE 53d41dee36SAndy Whitcroft help 54d41dee36SAndy Whitcroft This will be the only option for some systems, including 55*d66d109dSMike Rapoport memory hot-plug systems. This is normal. 56d41dee36SAndy Whitcroft 57*d66d109dSMike Rapoport This option provides efficient support for systems with 58*d66d109dSMike Rapoport holes is their physical address space and allows memory 59*d66d109dSMike Rapoport hot-plug and hot-remove. 60d41dee36SAndy Whitcroft 61*d66d109dSMike Rapoport If unsure, choose "Flat Memory" over this option. 62d41dee36SAndy Whitcroft 633a9da765SDave Hansenendchoice 643a9da765SDave Hansen 65e1785e85SDave Hansenconfig DISCONTIGMEM 66e1785e85SDave Hansen def_bool y 67e1785e85SDave Hansen depends on (!SELECT_MEMORY_MODEL && ARCH_DISCONTIGMEM_ENABLE) || DISCONTIGMEM_MANUAL 68e1785e85SDave Hansen 69d41dee36SAndy Whitcroftconfig SPARSEMEM 70d41dee36SAndy Whitcroft def_bool y 711a83e175SRussell King depends on (!SELECT_MEMORY_MODEL && ARCH_SPARSEMEM_ENABLE) || SPARSEMEM_MANUAL 72d41dee36SAndy Whitcroft 73e1785e85SDave Hansenconfig FLATMEM 74e1785e85SDave Hansen def_bool y 75d41dee36SAndy Whitcroft depends on (!DISCONTIGMEM && !SPARSEMEM) || FLATMEM_MANUAL 76d41dee36SAndy Whitcroft 77d41dee36SAndy Whitcroftconfig FLAT_NODE_MEM_MAP 78d41dee36SAndy Whitcroft def_bool y 79d41dee36SAndy Whitcroft depends on !SPARSEMEM 80e1785e85SDave Hansen 8193b7504eSDave Hansen# 8293b7504eSDave Hansen# Both the NUMA code and DISCONTIGMEM use arrays of pg_data_t's 8393b7504eSDave Hansen# to represent different areas of memory. This variable allows 8493b7504eSDave Hansen# those dependencies to exist individually. 8593b7504eSDave Hansen# 8693b7504eSDave Hansenconfig NEED_MULTIPLE_NODES 8793b7504eSDave Hansen def_bool y 8893b7504eSDave Hansen depends on DISCONTIGMEM || NUMA 89af705362SAndy Whitcroft 90af705362SAndy Whitcroftconfig HAVE_MEMORY_PRESENT 91af705362SAndy Whitcroft def_bool y 92d41dee36SAndy Whitcroft depends on ARCH_HAVE_MEMORY_PRESENT || SPARSEMEM 93802f192eSBob Picco 94802f192eSBob Picco# 953e347261SBob Picco# SPARSEMEM_EXTREME (which is the default) does some bootmem 963e347261SBob Picco# allocations when memory_present() is called. If this cannot 973e347261SBob Picco# be done on your architecture, select this option. However, 983e347261SBob Picco# statically allocating the mem_section[] array can potentially 993e347261SBob Picco# consume vast quantities of .bss, so be careful. 1003e347261SBob Picco# 1013e347261SBob Picco# This option will also potentially produce smaller runtime code 1023e347261SBob Picco# with gcc 3.4 and later. 1033e347261SBob Picco# 1043e347261SBob Piccoconfig SPARSEMEM_STATIC 1059ba16087SJan Beulich bool 1063e347261SBob Picco 1073e347261SBob Picco# 10844c09201SMatt LaPlante# Architecture platforms which require a two level mem_section in SPARSEMEM 109802f192eSBob Picco# must select this option. This is usually for architecture platforms with 110802f192eSBob Picco# an extremely sparse physical address space. 111802f192eSBob Picco# 1123e347261SBob Piccoconfig SPARSEMEM_EXTREME 1133e347261SBob Picco def_bool y 1143e347261SBob Picco depends on SPARSEMEM && !SPARSEMEM_STATIC 1154c21e2f2SHugh Dickins 11629c71111SAndy Whitcroftconfig SPARSEMEM_VMEMMAP_ENABLE 1179ba16087SJan Beulich bool 11829c71111SAndy Whitcroft 11929c71111SAndy Whitcroftconfig SPARSEMEM_VMEMMAP 120a5ee6daaSGeoff Levand bool "Sparse Memory virtual memmap" 121a5ee6daaSGeoff Levand depends on SPARSEMEM && SPARSEMEM_VMEMMAP_ENABLE 122a5ee6daaSGeoff Levand default y 123a5ee6daaSGeoff Levand help 124a5ee6daaSGeoff Levand SPARSEMEM_VMEMMAP uses a virtually mapped memmap to optimise 125a5ee6daaSGeoff Levand pfn_to_page and page_to_pfn operations. This is the most 126a5ee6daaSGeoff Levand efficient option when sufficient kernel resources are available. 12729c71111SAndy Whitcroft 1287c0caeb8STejun Heoconfig HAVE_MEMBLOCK_NODE_MAP 1296341e62bSChristoph Jaeger bool 1307c0caeb8STejun Heo 13170210ed9SPhilipp Hachtmannconfig HAVE_MEMBLOCK_PHYS_MAP 1326341e62bSChristoph Jaeger bool 13370210ed9SPhilipp Hachtmann 134e585513bSKirill A. Shutemovconfig HAVE_GENERIC_GUP 1356341e62bSChristoph Jaeger bool 1362667f50eSSteve Capper 137350e88baSMike Rapoportconfig ARCH_KEEP_MEMBLOCK 1386341e62bSChristoph Jaeger bool 139c378ddd5STejun Heo 140ee6f509cSMinchan Kimconfig MEMORY_ISOLATION 1416341e62bSChristoph Jaeger bool 142ee6f509cSMinchan Kim 14346723bfaSYasuaki Ishimatsu# 14446723bfaSYasuaki Ishimatsu# Only be set on architectures that have completely implemented memory hotplug 14546723bfaSYasuaki Ishimatsu# feature. If you are not sure, don't touch it. 14646723bfaSYasuaki Ishimatsu# 14746723bfaSYasuaki Ishimatsuconfig HAVE_BOOTMEM_INFO_NODE 14846723bfaSYasuaki Ishimatsu def_bool n 14946723bfaSYasuaki Ishimatsu 1503947be19SDave Hansen# eventually, we can have this option just 'select SPARSEMEM' 1513947be19SDave Hansenconfig MEMORY_HOTPLUG 1523947be19SDave Hansen bool "Allow for memory hot-add" 153ec69acbbSKeith Mannthey depends on SPARSEMEM || X86_64_ACPI_NUMA 15440b31360SStephen Rothwell depends on ARCH_ENABLE_MEMORY_HOTPLUG 1553947be19SDave Hansen 156ec69acbbSKeith Manntheyconfig MEMORY_HOTPLUG_SPARSE 157ec69acbbSKeith Mannthey def_bool y 158ec69acbbSKeith Mannthey depends on SPARSEMEM && MEMORY_HOTPLUG 159ec69acbbSKeith Mannthey 1608604d9e5SVitaly Kuznetsovconfig MEMORY_HOTPLUG_DEFAULT_ONLINE 1618604d9e5SVitaly Kuznetsov bool "Online the newly added memory blocks by default" 1628604d9e5SVitaly Kuznetsov depends on MEMORY_HOTPLUG 1638604d9e5SVitaly Kuznetsov help 1648604d9e5SVitaly Kuznetsov This option sets the default policy setting for memory hotplug 1658604d9e5SVitaly Kuznetsov onlining policy (/sys/devices/system/memory/auto_online_blocks) which 1668604d9e5SVitaly Kuznetsov determines what happens to newly added memory regions. Policy setting 1678604d9e5SVitaly Kuznetsov can always be changed at runtime. 1688604d9e5SVitaly Kuznetsov See Documentation/memory-hotplug.txt for more information. 1698604d9e5SVitaly Kuznetsov 1708604d9e5SVitaly Kuznetsov Say Y here if you want all hot-plugged memory blocks to appear in 1718604d9e5SVitaly Kuznetsov 'online' state by default. 1728604d9e5SVitaly Kuznetsov Say N here if you want the default policy to keep all hot-plugged 1738604d9e5SVitaly Kuznetsov memory blocks in 'offline' state. 1748604d9e5SVitaly Kuznetsov 1750c0e6195SKAMEZAWA Hiroyukiconfig MEMORY_HOTREMOVE 1760c0e6195SKAMEZAWA Hiroyuki bool "Allow for memory hot remove" 17746723bfaSYasuaki Ishimatsu select MEMORY_ISOLATION 178f7e3334aSNathan Fontenot select HAVE_BOOTMEM_INFO_NODE if (X86_64 || PPC64) 1790c0e6195SKAMEZAWA Hiroyuki depends on MEMORY_HOTPLUG && ARCH_ENABLE_MEMORY_HOTREMOVE 1800c0e6195SKAMEZAWA Hiroyuki depends on MIGRATION 1810c0e6195SKAMEZAWA Hiroyuki 1824c21e2f2SHugh Dickins# Heavily threaded applications may benefit from splitting the mm-wide 1834c21e2f2SHugh Dickins# page_table_lock, so that faults on different parts of the user address 1844c21e2f2SHugh Dickins# space can be handled with less contention: split it at this NR_CPUS. 1854c21e2f2SHugh Dickins# Default to 4 for wider testing, though 8 might be more appropriate. 1864c21e2f2SHugh Dickins# ARM's adjust_pte (unused if VIPT) depends on mm-wide page_table_lock. 1877b6ac9dfSHugh Dickins# PA-RISC 7xxx's spinlock_t would enlarge struct page from 32 to 44 bytes. 188a70caa8bSHugh Dickins# DEBUG_SPINLOCK and DEBUG_LOCK_ALLOC spinlock_t also enlarge struct page. 1894c21e2f2SHugh Dickins# 1904c21e2f2SHugh Dickinsconfig SPLIT_PTLOCK_CPUS 1914c21e2f2SHugh Dickins int 1929164550eSKirill A. Shutemov default "999999" if !MMU 193a70caa8bSHugh Dickins default "999999" if ARM && !CPU_CACHE_VIPT 194a70caa8bSHugh Dickins default "999999" if PARISC && !PA20 1954c21e2f2SHugh Dickins default "4" 1967cbe34cfSChristoph Lameter 197e009bb30SKirill A. Shutemovconfig ARCH_ENABLE_SPLIT_PMD_PTLOCK 1986341e62bSChristoph Jaeger bool 199e009bb30SKirill A. Shutemov 2007cbe34cfSChristoph Lameter# 20109316c09SKonstantin Khlebnikov# support for memory balloon 20209316c09SKonstantin Khlebnikovconfig MEMORY_BALLOON 2036341e62bSChristoph Jaeger bool 20409316c09SKonstantin Khlebnikov 20509316c09SKonstantin Khlebnikov# 20618468d93SRafael Aquini# support for memory balloon compaction 20718468d93SRafael Aquiniconfig BALLOON_COMPACTION 20818468d93SRafael Aquini bool "Allow for balloon memory compaction/migration" 20918468d93SRafael Aquini def_bool y 21009316c09SKonstantin Khlebnikov depends on COMPACTION && MEMORY_BALLOON 21118468d93SRafael Aquini help 21218468d93SRafael Aquini Memory fragmentation introduced by ballooning might reduce 21318468d93SRafael Aquini significantly the number of 2MB contiguous memory blocks that can be 21418468d93SRafael Aquini used within a guest, thus imposing performance penalties associated 21518468d93SRafael Aquini with the reduced number of transparent huge pages that could be used 21618468d93SRafael Aquini by the guest workload. Allowing the compaction & migration for memory 21718468d93SRafael Aquini pages enlisted as being part of memory balloon devices avoids the 21818468d93SRafael Aquini scenario aforementioned and helps improving memory defragmentation. 21918468d93SRafael Aquini 22018468d93SRafael Aquini# 221e9e96b39SMel Gorman# support for memory compaction 222e9e96b39SMel Gormanconfig COMPACTION 223e9e96b39SMel Gorman bool "Allow for memory compaction" 22405106e6aSRik van Riel def_bool y 225e9e96b39SMel Gorman select MIGRATION 22633a93877SAndrea Arcangeli depends on MMU 227e9e96b39SMel Gorman help 228b32eaf71SMichal Hocko Compaction is the only memory management component to form 229b32eaf71SMichal Hocko high order (larger physically contiguous) memory blocks 230b32eaf71SMichal Hocko reliably. The page allocator relies on compaction heavily and 231b32eaf71SMichal Hocko the lack of the feature can lead to unexpected OOM killer 232b32eaf71SMichal Hocko invocations for high order memory requests. You shouldn't 233b32eaf71SMichal Hocko disable this option unless there really is a strong reason for 234b32eaf71SMichal Hocko it and then we would be really interested to hear about that at 235b32eaf71SMichal Hocko linux-mm@kvack.org. 236e9e96b39SMel Gorman 237e9e96b39SMel Gorman# 2387cbe34cfSChristoph Lameter# support for page migration 2397cbe34cfSChristoph Lameter# 2407cbe34cfSChristoph Lameterconfig MIGRATION 241b20a3503SChristoph Lameter bool "Page migration" 2426c5240aeSChristoph Lameter def_bool y 243de32a817SChen Gang depends on (NUMA || ARCH_ENABLE_MEMORY_HOTREMOVE || COMPACTION || CMA) && MMU 244b20a3503SChristoph Lameter help 245b20a3503SChristoph Lameter Allows the migration of the physical location of pages of processes 246e9e96b39SMel Gorman while the virtual addresses are not changed. This is useful in 247e9e96b39SMel Gorman two situations. The first is on NUMA systems to put pages nearer 248e9e96b39SMel Gorman to the processors accessing. The second is when allocating huge 249e9e96b39SMel Gorman pages as migration can relocate pages to satisfy a huge page 250e9e96b39SMel Gorman allocation instead of reclaiming. 2516550e07fSGreg Kroah-Hartman 252c177c81eSNaoya Horiguchiconfig ARCH_ENABLE_HUGEPAGE_MIGRATION 2536341e62bSChristoph Jaeger bool 254c177c81eSNaoya Horiguchi 2559c670ea3SNaoya Horiguchiconfig ARCH_ENABLE_THP_MIGRATION 2569c670ea3SNaoya Horiguchi bool 2579c670ea3SNaoya Horiguchi 2588df995f6SAlexandre Ghiticonfig CONTIG_ALLOC 2598df995f6SAlexandre Ghiti def_bool (MEMORY_ISOLATION && COMPACTION) || CMA 2608df995f6SAlexandre Ghiti 261600715dcSJeremy Fitzhardingeconfig PHYS_ADDR_T_64BIT 262d4a451d5SChristoph Hellwig def_bool 64BIT 263600715dcSJeremy Fitzhardinge 2642a7326b5SChristoph Lameterconfig BOUNCE 2659ca24e2eSVinayak Menon bool "Enable bounce buffers" 2669ca24e2eSVinayak Menon default y 2672a7326b5SChristoph Lameter depends on BLOCK && MMU && (ZONE_DMA || HIGHMEM) 2689ca24e2eSVinayak Menon help 2699ca24e2eSVinayak Menon Enable bounce buffers for devices that cannot access 2709ca24e2eSVinayak Menon the full range of memory available to the CPU. Enabled 2719ca24e2eSVinayak Menon by default when ZONE_DMA or HIGHMEM is selected, but you 2729ca24e2eSVinayak Menon may say n to override this. 2732a7326b5SChristoph Lameter 2746225e937SChristoph Lameterconfig NR_QUICK 2756225e937SChristoph Lameter int 2766225e937SChristoph Lameter depends on QUICKLIST 2776225e937SChristoph Lameter default "1" 278f057eac0SStephen Rothwell 279f057eac0SStephen Rothwellconfig VIRT_TO_BUS 2804febd95aSStephen Rothwell bool 2814febd95aSStephen Rothwell help 2824febd95aSStephen Rothwell An architecture should select this if it implements the 2834febd95aSStephen Rothwell deprecated interface virt_to_bus(). All new architectures 2844febd95aSStephen Rothwell should probably not select this. 2854febd95aSStephen Rothwell 286cddb8a5cSAndrea Arcangeli 287cddb8a5cSAndrea Arcangeliconfig MMU_NOTIFIER 288cddb8a5cSAndrea Arcangeli bool 28983fe27eaSPranith Kumar select SRCU 290fc4d5c29SDavid Howells 291f8af4da3SHugh Dickinsconfig KSM 292f8af4da3SHugh Dickins bool "Enable KSM for page merging" 293f8af4da3SHugh Dickins depends on MMU 29459e1a2f4STimofey Titovets select XXHASH 295f8af4da3SHugh Dickins help 296f8af4da3SHugh Dickins Enable Kernel Samepage Merging: KSM periodically scans those areas 297f8af4da3SHugh Dickins of an application's address space that an app has advised may be 298f8af4da3SHugh Dickins mergeable. When it finds pages of identical content, it replaces 299d0f209f6SHugh Dickins the many instances by a single page with that content, so 300f8af4da3SHugh Dickins saving memory until one or another app needs to modify the content. 301f8af4da3SHugh Dickins Recommended for use with KVM, or with other duplicative applications. 302ad56b738SMike Rapoport See Documentation/vm/ksm.rst for more information: KSM is inactive 303c73602adSHugh Dickins until a program has madvised that an area is MADV_MERGEABLE, and 304c73602adSHugh Dickins root has set /sys/kernel/mm/ksm/run to 1 (if CONFIG_SYSFS is set). 305f8af4da3SHugh Dickins 306e0a94c2aSChristoph Lameterconfig DEFAULT_MMAP_MIN_ADDR 307e0a94c2aSChristoph Lameter int "Low address space to protect from user allocation" 3086e141546SDavid Howells depends on MMU 309e0a94c2aSChristoph Lameter default 4096 310e0a94c2aSChristoph Lameter help 311e0a94c2aSChristoph Lameter This is the portion of low virtual memory which should be protected 312e0a94c2aSChristoph Lameter from userspace allocation. Keeping a user from writing to low pages 313e0a94c2aSChristoph Lameter can help reduce the impact of kernel NULL pointer bugs. 314e0a94c2aSChristoph Lameter 315e0a94c2aSChristoph Lameter For most ia64, ppc64 and x86 users with lots of address space 316e0a94c2aSChristoph Lameter a value of 65536 is reasonable and should cause no problems. 317e0a94c2aSChristoph Lameter On arm and other archs it should not be higher than 32768. 318788084abSEric Paris Programs which use vm86 functionality or have some need to map 319788084abSEric Paris this low address space will need CAP_SYS_RAWIO or disable this 320788084abSEric Paris protection by setting the value to 0. 321e0a94c2aSChristoph Lameter 322e0a94c2aSChristoph Lameter This value can be changed after boot using the 323e0a94c2aSChristoph Lameter /proc/sys/vm/mmap_min_addr tunable. 324e0a94c2aSChristoph Lameter 325d949f36fSLinus Torvaldsconfig ARCH_SUPPORTS_MEMORY_FAILURE 326d949f36fSLinus Torvalds bool 327e0a94c2aSChristoph Lameter 3286a46079cSAndi Kleenconfig MEMORY_FAILURE 3296a46079cSAndi Kleen depends on MMU 330d949f36fSLinus Torvalds depends on ARCH_SUPPORTS_MEMORY_FAILURE 3316a46079cSAndi Kleen bool "Enable recovery from hardware memory errors" 332ee6f509cSMinchan Kim select MEMORY_ISOLATION 33397f0b134SXie XiuQi select RAS 3346a46079cSAndi Kleen help 3356a46079cSAndi Kleen Enables code to recover from some memory failures on systems 3366a46079cSAndi Kleen with MCA recovery. This allows a system to continue running 3376a46079cSAndi Kleen even when some of its memory has uncorrected errors. This requires 3386a46079cSAndi Kleen special hardware support and typically ECC memory. 3396a46079cSAndi Kleen 340cae681fcSAndi Kleenconfig HWPOISON_INJECT 341413f9efbSAndi Kleen tristate "HWPoison pages injector" 34227df5068SAndi Kleen depends on MEMORY_FAILURE && DEBUG_KERNEL && PROC_FS 343478c5ffcSWu Fengguang select PROC_PAGE_MONITOR 344cae681fcSAndi Kleen 345fc4d5c29SDavid Howellsconfig NOMMU_INITIAL_TRIM_EXCESS 346fc4d5c29SDavid Howells int "Turn on mmap() excess space trimming before booting" 347fc4d5c29SDavid Howells depends on !MMU 348fc4d5c29SDavid Howells default 1 349fc4d5c29SDavid Howells help 350fc4d5c29SDavid Howells The NOMMU mmap() frequently needs to allocate large contiguous chunks 351fc4d5c29SDavid Howells of memory on which to store mappings, but it can only ask the system 352fc4d5c29SDavid Howells allocator for chunks in 2^N*PAGE_SIZE amounts - which is frequently 353fc4d5c29SDavid Howells more than it requires. To deal with this, mmap() is able to trim off 354fc4d5c29SDavid Howells the excess and return it to the allocator. 355fc4d5c29SDavid Howells 356fc4d5c29SDavid Howells If trimming is enabled, the excess is trimmed off and returned to the 357fc4d5c29SDavid Howells system allocator, which can cause extra fragmentation, particularly 358fc4d5c29SDavid Howells if there are a lot of transient processes. 359fc4d5c29SDavid Howells 360fc4d5c29SDavid Howells If trimming is disabled, the excess is kept, but not used, which for 361fc4d5c29SDavid Howells long-term mappings means that the space is wasted. 362fc4d5c29SDavid Howells 363fc4d5c29SDavid Howells Trimming can be dynamically controlled through a sysctl option 364fc4d5c29SDavid Howells (/proc/sys/vm/nr_trim_pages) which specifies the minimum number of 365fc4d5c29SDavid Howells excess pages there must be before trimming should occur, or zero if 366fc4d5c29SDavid Howells no trimming is to occur. 367fc4d5c29SDavid Howells 368fc4d5c29SDavid Howells This option specifies the initial value of this option. The default 369fc4d5c29SDavid Howells of 1 says that all excess pages should be trimmed. 370fc4d5c29SDavid Howells 371fc4d5c29SDavid Howells See Documentation/nommu-mmap.txt for more information. 372bbddff05STejun Heo 3734c76d9d1SAndrea Arcangeliconfig TRANSPARENT_HUGEPAGE 37413ece886SAndrea Arcangeli bool "Transparent Hugepage Support" 37515626062SGerald Schaefer depends on HAVE_ARCH_TRANSPARENT_HUGEPAGE 3765d689240SAndrea Arcangeli select COMPACTION 3773a08cd52SMatthew Wilcox select XARRAY_MULTI 3784c76d9d1SAndrea Arcangeli help 3794c76d9d1SAndrea Arcangeli Transparent Hugepages allows the kernel to use huge pages and 3804c76d9d1SAndrea Arcangeli huge tlb transparently to the applications whenever possible. 3814c76d9d1SAndrea Arcangeli This feature can improve computing performance to certain 3824c76d9d1SAndrea Arcangeli applications by speeding up page faults during memory 3834c76d9d1SAndrea Arcangeli allocation, by reducing the number of tlb misses and by speeding 3844c76d9d1SAndrea Arcangeli up the pagetable walking. 3854c76d9d1SAndrea Arcangeli 3864c76d9d1SAndrea Arcangeli If memory constrained on embedded, you may want to say N. 3874c76d9d1SAndrea Arcangeli 38813ece886SAndrea Arcangelichoice 38913ece886SAndrea Arcangeli prompt "Transparent Hugepage Support sysfs defaults" 39013ece886SAndrea Arcangeli depends on TRANSPARENT_HUGEPAGE 39113ece886SAndrea Arcangeli default TRANSPARENT_HUGEPAGE_ALWAYS 39213ece886SAndrea Arcangeli help 39313ece886SAndrea Arcangeli Selects the sysfs defaults for Transparent Hugepage Support. 39413ece886SAndrea Arcangeli 39513ece886SAndrea Arcangeli config TRANSPARENT_HUGEPAGE_ALWAYS 39613ece886SAndrea Arcangeli bool "always" 39713ece886SAndrea Arcangeli help 39813ece886SAndrea Arcangeli Enabling Transparent Hugepage always, can increase the 39913ece886SAndrea Arcangeli memory footprint of applications without a guaranteed 40013ece886SAndrea Arcangeli benefit but it will work automatically for all applications. 40113ece886SAndrea Arcangeli 40213ece886SAndrea Arcangeli config TRANSPARENT_HUGEPAGE_MADVISE 40313ece886SAndrea Arcangeli bool "madvise" 40413ece886SAndrea Arcangeli help 40513ece886SAndrea Arcangeli Enabling Transparent Hugepage madvise, will only provide a 40613ece886SAndrea Arcangeli performance improvement benefit to the applications using 40713ece886SAndrea Arcangeli madvise(MADV_HUGEPAGE) but it won't risk to increase the 40813ece886SAndrea Arcangeli memory footprint of applications without a guaranteed 40913ece886SAndrea Arcangeli benefit. 41013ece886SAndrea Arcangeliendchoice 41113ece886SAndrea Arcangeli 41238d8b4e6SHuang Yingconfig ARCH_WANTS_THP_SWAP 41338d8b4e6SHuang Ying def_bool n 41438d8b4e6SHuang Ying 41538d8b4e6SHuang Yingconfig THP_SWAP 41638d8b4e6SHuang Ying def_bool y 41714fef284SHuang Ying depends on TRANSPARENT_HUGEPAGE && ARCH_WANTS_THP_SWAP && SWAP 41838d8b4e6SHuang Ying help 41938d8b4e6SHuang Ying Swap transparent huge pages in one piece, without splitting. 42014fef284SHuang Ying XXX: For now, swap cluster backing transparent huge page 42114fef284SHuang Ying will be split after swapout. 42238d8b4e6SHuang Ying 42338d8b4e6SHuang Ying For selection by architectures with reasonable THP sizes. 42438d8b4e6SHuang Ying 425e496cf3dSKirill A. Shutemovconfig TRANSPARENT_HUGE_PAGECACHE 426e496cf3dSKirill A. Shutemov def_bool y 427953c66c2SAneesh Kumar K.V depends on TRANSPARENT_HUGEPAGE 428e496cf3dSKirill A. Shutemov 429e496cf3dSKirill A. Shutemov# 430bbddff05STejun Heo# UP and nommu archs use km based percpu allocator 431bbddff05STejun Heo# 432bbddff05STejun Heoconfig NEED_PER_CPU_KM 433bbddff05STejun Heo depends on !SMP 434bbddff05STejun Heo bool 435bbddff05STejun Heo default y 436077b1f83SDan Magenheimer 437077b1f83SDan Magenheimerconfig CLEANCACHE 438077b1f83SDan Magenheimer bool "Enable cleancache driver to cache clean pages if tmem is present" 439077b1f83SDan Magenheimer help 440077b1f83SDan Magenheimer Cleancache can be thought of as a page-granularity victim cache 441077b1f83SDan Magenheimer for clean pages that the kernel's pageframe replacement algorithm 442077b1f83SDan Magenheimer (PFRA) would like to keep around, but can't since there isn't enough 443077b1f83SDan Magenheimer memory. So when the PFRA "evicts" a page, it first attempts to use 444140a1ef2SMichael Witten cleancache code to put the data contained in that page into 445077b1f83SDan Magenheimer "transcendent memory", memory that is not directly accessible or 446077b1f83SDan Magenheimer addressable by the kernel and is of unknown and possibly 447077b1f83SDan Magenheimer time-varying size. And when a cleancache-enabled 448077b1f83SDan Magenheimer filesystem wishes to access a page in a file on disk, it first 449077b1f83SDan Magenheimer checks cleancache to see if it already contains it; if it does, 450077b1f83SDan Magenheimer the page is copied into the kernel and a disk access is avoided. 451077b1f83SDan Magenheimer When a transcendent memory driver is available (such as zcache or 452077b1f83SDan Magenheimer Xen transcendent memory), a significant I/O reduction 453077b1f83SDan Magenheimer may be achieved. When none is available, all cleancache calls 454077b1f83SDan Magenheimer are reduced to a single pointer-compare-against-NULL resulting 455077b1f83SDan Magenheimer in a negligible performance hit. 456077b1f83SDan Magenheimer 457077b1f83SDan Magenheimer If unsure, say Y to enable cleancache 45827c6aec2SDan Magenheimer 45927c6aec2SDan Magenheimerconfig FRONTSWAP 46027c6aec2SDan Magenheimer bool "Enable frontswap to cache swap pages if tmem is present" 46127c6aec2SDan Magenheimer depends on SWAP 46227c6aec2SDan Magenheimer help 46327c6aec2SDan Magenheimer Frontswap is so named because it can be thought of as the opposite 46427c6aec2SDan Magenheimer of a "backing" store for a swap device. The data is stored into 46527c6aec2SDan Magenheimer "transcendent memory", memory that is not directly accessible or 46627c6aec2SDan Magenheimer addressable by the kernel and is of unknown and possibly 46727c6aec2SDan Magenheimer time-varying size. When space in transcendent memory is available, 46827c6aec2SDan Magenheimer a significant swap I/O reduction may be achieved. When none is 46927c6aec2SDan Magenheimer available, all frontswap calls are reduced to a single pointer- 47027c6aec2SDan Magenheimer compare-against-NULL resulting in a negligible performance hit 47127c6aec2SDan Magenheimer and swap data is stored as normal on the matching swap device. 47227c6aec2SDan Magenheimer 47327c6aec2SDan Magenheimer If unsure, say Y to enable frontswap. 474f825c736SAneesh Kumar K.V 475f825c736SAneesh Kumar K.Vconfig CMA 476f825c736SAneesh Kumar K.V bool "Contiguous Memory Allocator" 477aca52c39SMike Rapoport depends on MMU 478f825c736SAneesh Kumar K.V select MIGRATION 479f825c736SAneesh Kumar K.V select MEMORY_ISOLATION 480f825c736SAneesh Kumar K.V help 481f825c736SAneesh Kumar K.V This enables the Contiguous Memory Allocator which allows other 482f825c736SAneesh Kumar K.V subsystems to allocate big physically-contiguous blocks of memory. 483f825c736SAneesh Kumar K.V CMA reserves a region of memory and allows only movable pages to 484f825c736SAneesh Kumar K.V be allocated from it. This way, the kernel can use the memory for 485f825c736SAneesh Kumar K.V pagecache and when a subsystem requests for contiguous area, the 486f825c736SAneesh Kumar K.V allocated pages are migrated away to serve the contiguous request. 487f825c736SAneesh Kumar K.V 488f825c736SAneesh Kumar K.V If unsure, say "n". 489f825c736SAneesh Kumar K.V 490f825c736SAneesh Kumar K.Vconfig CMA_DEBUG 491f825c736SAneesh Kumar K.V bool "CMA debug messages (DEVELOPMENT)" 492f825c736SAneesh Kumar K.V depends on DEBUG_KERNEL && CMA 493f825c736SAneesh Kumar K.V help 494f825c736SAneesh Kumar K.V Turns on debug messages in CMA. This produces KERN_DEBUG 495f825c736SAneesh Kumar K.V messages for every CMA call as well as various messages while 496f825c736SAneesh Kumar K.V processing calls such as dma_alloc_from_contiguous(). 497f825c736SAneesh Kumar K.V This option does not affect warning and error messages. 498bf550fc9SAlexander Graf 49928b24c1fSSasha Levinconfig CMA_DEBUGFS 50028b24c1fSSasha Levin bool "CMA debugfs interface" 50128b24c1fSSasha Levin depends on CMA && DEBUG_FS 50228b24c1fSSasha Levin help 50328b24c1fSSasha Levin Turns on the DebugFS interface for CMA. 50428b24c1fSSasha Levin 505a254129eSJoonsoo Kimconfig CMA_AREAS 506a254129eSJoonsoo Kim int "Maximum count of the CMA areas" 507a254129eSJoonsoo Kim depends on CMA 508a254129eSJoonsoo Kim default 7 509a254129eSJoonsoo Kim help 510a254129eSJoonsoo Kim CMA allows to create CMA areas for particular purpose, mainly, 511a254129eSJoonsoo Kim used as device private area. This parameter sets the maximum 512a254129eSJoonsoo Kim number of CMA area in the system. 513a254129eSJoonsoo Kim 514a254129eSJoonsoo Kim If unsure, leave the default value "7". 515a254129eSJoonsoo Kim 516af8d417aSDan Streetmanconfig MEM_SOFT_DIRTY 517af8d417aSDan Streetman bool "Track memory changes" 518af8d417aSDan Streetman depends on CHECKPOINT_RESTORE && HAVE_ARCH_SOFT_DIRTY && PROC_FS 519af8d417aSDan Streetman select PROC_PAGE_MONITOR 5204e2e2770SSeth Jennings help 521af8d417aSDan Streetman This option enables memory changes tracking by introducing a 522af8d417aSDan Streetman soft-dirty bit on pte-s. This bit it set when someone writes 523af8d417aSDan Streetman into a page just as regular dirty bit, but unlike the latter 524af8d417aSDan Streetman it can be cleared by hands. 525af8d417aSDan Streetman 5261ad1335dSMike Rapoport See Documentation/admin-guide/mm/soft-dirty.rst for more details. 5274e2e2770SSeth Jennings 5282b281117SSeth Jenningsconfig ZSWAP 5292b281117SSeth Jennings bool "Compressed cache for swap pages (EXPERIMENTAL)" 5302b281117SSeth Jennings depends on FRONTSWAP && CRYPTO=y 5312b281117SSeth Jennings select CRYPTO_LZO 53212d79d64SDan Streetman select ZPOOL 5332b281117SSeth Jennings help 5342b281117SSeth Jennings A lightweight compressed cache for swap pages. It takes 5352b281117SSeth Jennings pages that are in the process of being swapped out and attempts to 5362b281117SSeth Jennings compress them into a dynamically allocated RAM-based memory pool. 5372b281117SSeth Jennings This can result in a significant I/O reduction on swap device and, 5382b281117SSeth Jennings in the case where decompressing from RAM is faster that swap device 5392b281117SSeth Jennings reads, can also improve workload performance. 5402b281117SSeth Jennings 5412b281117SSeth Jennings This is marked experimental because it is a new feature (as of 5422b281117SSeth Jennings v3.11) that interacts heavily with memory reclaim. While these 5432b281117SSeth Jennings interactions don't cause any known issues on simple memory setups, 5442b281117SSeth Jennings they have not be fully explored on the large set of potential 5452b281117SSeth Jennings configurations and workloads that exist. 5462b281117SSeth Jennings 547af8d417aSDan Streetmanconfig ZPOOL 548af8d417aSDan Streetman tristate "Common API for compressed memory storage" 5490f8975ecSPavel Emelyanov help 550af8d417aSDan Streetman Compressed memory storage API. This allows using either zbud or 551af8d417aSDan Streetman zsmalloc. 5520f8975ecSPavel Emelyanov 553af8d417aSDan Streetmanconfig ZBUD 5549a001fc1SVitaly Wool tristate "Low (Up to 2x) density storage for compressed pages" 555af8d417aSDan Streetman help 556af8d417aSDan Streetman A special purpose allocator for storing compressed pages. 557af8d417aSDan Streetman It is designed to store up to two compressed pages per physical 558af8d417aSDan Streetman page. While this design limits storage density, it has simple and 559af8d417aSDan Streetman deterministic reclaim properties that make it preferable to a higher 560af8d417aSDan Streetman density approach when reclaim will be used. 561bcf1647dSMinchan Kim 5629a001fc1SVitaly Woolconfig Z3FOLD 5639a001fc1SVitaly Wool tristate "Up to 3x density storage for compressed pages" 5649a001fc1SVitaly Wool depends on ZPOOL 5659a001fc1SVitaly Wool help 5669a001fc1SVitaly Wool A special purpose allocator for storing compressed pages. 5679a001fc1SVitaly Wool It is designed to store up to three compressed pages per physical 5689a001fc1SVitaly Wool page. It is a ZBUD derivative so the simplicity and determinism are 5699a001fc1SVitaly Wool still there. 5709a001fc1SVitaly Wool 571bcf1647dSMinchan Kimconfig ZSMALLOC 572d867f203SMinchan Kim tristate "Memory allocator for compressed pages" 573bcf1647dSMinchan Kim depends on MMU 574bcf1647dSMinchan Kim help 575bcf1647dSMinchan Kim zsmalloc is a slab-based memory allocator designed to store 576bcf1647dSMinchan Kim compressed RAM pages. zsmalloc uses virtual memory mapping 577bcf1647dSMinchan Kim in order to reduce fragmentation. However, this results in a 578bcf1647dSMinchan Kim non-standard allocator interface where a handle, not a pointer, is 579bcf1647dSMinchan Kim returned by an alloc(). This handle must be mapped in order to 580bcf1647dSMinchan Kim access the allocated space. 581bcf1647dSMinchan Kim 582bcf1647dSMinchan Kimconfig PGTABLE_MAPPING 583bcf1647dSMinchan Kim bool "Use page table mapping to access object in zsmalloc" 584bcf1647dSMinchan Kim depends on ZSMALLOC 585bcf1647dSMinchan Kim help 586bcf1647dSMinchan Kim By default, zsmalloc uses a copy-based object mapping method to 587bcf1647dSMinchan Kim access allocations that span two pages. However, if a particular 588bcf1647dSMinchan Kim architecture (ex, ARM) performs VM mapping faster than copying, 589bcf1647dSMinchan Kim then you should select this. This causes zsmalloc to use page table 590bcf1647dSMinchan Kim mapping rather than copying for object mapping. 591bcf1647dSMinchan Kim 5922216ee85SBen Hutchings You can check speed with zsmalloc benchmark: 5932216ee85SBen Hutchings https://github.com/spartacus06/zsmapbench 5949e5c33d7SMark Salter 5950f050d99SGanesh Mahendranconfig ZSMALLOC_STAT 5960f050d99SGanesh Mahendran bool "Export zsmalloc statistics" 5970f050d99SGanesh Mahendran depends on ZSMALLOC 5980f050d99SGanesh Mahendran select DEBUG_FS 5990f050d99SGanesh Mahendran help 6000f050d99SGanesh Mahendran This option enables code in the zsmalloc to collect various 6010f050d99SGanesh Mahendran statistics about whats happening in zsmalloc and exports that 6020f050d99SGanesh Mahendran information to userspace via debugfs. 6030f050d99SGanesh Mahendran If unsure, say N. 6040f050d99SGanesh Mahendran 6059e5c33d7SMark Salterconfig GENERIC_EARLY_IOREMAP 6069e5c33d7SMark Salter bool 607042d27acSHelge Deller 608042d27acSHelge Dellerconfig MAX_STACK_SIZE_MB 609042d27acSHelge Deller int "Maximum user stack size for 32-bit processes (MB)" 610042d27acSHelge Deller default 80 611042d27acSHelge Deller range 8 2048 612042d27acSHelge Deller depends on STACK_GROWSUP && (!64BIT || COMPAT) 613042d27acSHelge Deller help 614042d27acSHelge Deller This is the maximum stack size in Megabytes in the VM layout of 32-bit 615042d27acSHelge Deller user processes when the stack grows upwards (currently only on parisc 6165f171577SJames Hogan arch). The stack will be located at the highest memory address minus 6175f171577SJames Hogan the given value, unless the RLIMIT_STACK hard limit is changed to a 6185f171577SJames Hogan smaller value in which case that is used. 619042d27acSHelge Deller 620042d27acSHelge Deller A sane initial value is 80 MB. 6213a80a7faSMel Gorman 6223a80a7faSMel Gormanconfig DEFERRED_STRUCT_PAGE_INIT 6231ce22103SVlastimil Babka bool "Defer initialisation of struct pages to kthreads" 624d39f8fb4SMike Rapoport depends on SPARSEMEM 625ab1e8d89SPavel Tatashin depends on !NEED_PER_CPU_KM 626889c695dSPasha Tatashin depends on 64BIT 6273a80a7faSMel Gorman help 6283a80a7faSMel Gorman Ordinarily all struct pages are initialised during early boot in a 6293a80a7faSMel Gorman single thread. On very large machines this can take a considerable 6303a80a7faSMel Gorman amount of time. If this option is set, large machines will bring up 6313a80a7faSMel Gorman a subset of memmap at boot and then initialise the rest in parallel 6321ce22103SVlastimil Babka by starting one-off "pgdatinitX" kernel thread for each node X. This 6331ce22103SVlastimil Babka has a potential performance impact on processes running early in the 6341ce22103SVlastimil Babka lifetime of the system until these kthreads finish the 6351ce22103SVlastimil Babka initialisation. 636033fbae9SDan Williams 63733c3fc71SVladimir Davydovconfig IDLE_PAGE_TRACKING 63833c3fc71SVladimir Davydov bool "Enable idle page tracking" 63933c3fc71SVladimir Davydov depends on SYSFS && MMU 64033c3fc71SVladimir Davydov select PAGE_EXTENSION if !64BIT 64133c3fc71SVladimir Davydov help 64233c3fc71SVladimir Davydov This feature allows to estimate the amount of user pages that have 64333c3fc71SVladimir Davydov not been touched during a given period of time. This information can 64433c3fc71SVladimir Davydov be useful to tune memory cgroup limits and/or for job placement 64533c3fc71SVladimir Davydov within a compute cluster. 64633c3fc71SVladimir Davydov 6471ad1335dSMike Rapoport See Documentation/admin-guide/mm/idle_page_tracking.rst for 6481ad1335dSMike Rapoport more details. 64933c3fc71SVladimir Davydov 65065f7d049SOliver O'Halloran# arch_add_memory() comprehends device memory 65165f7d049SOliver O'Halloranconfig ARCH_HAS_ZONE_DEVICE 65265f7d049SOliver O'Halloran bool 65365f7d049SOliver O'Halloran 654033fbae9SDan Williamsconfig ZONE_DEVICE 6555042db43SJérôme Glisse bool "Device memory (pmem, HMM, etc...) hotplug support" 656033fbae9SDan Williams depends on MEMORY_HOTPLUG 657033fbae9SDan Williams depends on MEMORY_HOTREMOVE 65899490f16SDan Williams depends on SPARSEMEM_VMEMMAP 65965f7d049SOliver O'Halloran depends on ARCH_HAS_ZONE_DEVICE 6603a08cd52SMatthew Wilcox select XARRAY_MULTI 661033fbae9SDan Williams 662033fbae9SDan Williams help 663033fbae9SDan Williams Device memory hotplug support allows for establishing pmem, 664033fbae9SDan Williams or other device driver discovered memory regions, in the 665033fbae9SDan Williams memmap. This allows pfn_to_page() lookups of otherwise 666033fbae9SDan Williams "device-physical" addresses which is needed for using a DAX 667033fbae9SDan Williams mapping in an O_DIRECT operation, among other things. 668033fbae9SDan Williams 669033fbae9SDan Williams If FS_DAX is enabled, then say Y. 67006a660adSLinus Torvalds 6712c8fc3dcSJérôme Glisseconfig ARCH_HAS_HMM_MIRROR 6722c8fc3dcSJérôme Glisse bool 6732c8fc3dcSJérôme Glisse default y 6742c8fc3dcSJérôme Glisse depends on (X86_64 || PPC64) 6752c8fc3dcSJérôme Glisse depends on MMU && 64BIT 6762c8fc3dcSJérôme Glisse 6772c8fc3dcSJérôme Glisseconfig ARCH_HAS_HMM_DEVICE 6782c8fc3dcSJérôme Glisse bool 6792c8fc3dcSJérôme Glisse default y 6802c8fc3dcSJérôme Glisse depends on (X86_64 || PPC64) 6812c8fc3dcSJérôme Glisse depends on MEMORY_HOTPLUG 6822c8fc3dcSJérôme Glisse depends on MEMORY_HOTREMOVE 6832c8fc3dcSJérôme Glisse depends on SPARSEMEM_VMEMMAP 6842c8fc3dcSJérôme Glisse depends on ARCH_HAS_ZONE_DEVICE 6852c8fc3dcSJérôme Glisse select XARRAY_MULTI 6862c8fc3dcSJérôme Glisse 687133ff0eaSJérôme Glisseconfig ARCH_HAS_HMM 688133ff0eaSJérôme Glisse bool 689133ff0eaSJérôme Glisse default y 690133ff0eaSJérôme Glisse depends on (X86_64 || PPC64) 691133ff0eaSJérôme Glisse depends on ZONE_DEVICE 692133ff0eaSJérôme Glisse depends on MMU && 64BIT 693133ff0eaSJérôme Glisse depends on MEMORY_HOTPLUG 694133ff0eaSJérôme Glisse depends on MEMORY_HOTREMOVE 695133ff0eaSJérôme Glisse depends on SPARSEMEM_VMEMMAP 696133ff0eaSJérôme Glisse 6976b368cd4SJérôme Glisseconfig MIGRATE_VMA_HELPER 6986b368cd4SJérôme Glisse bool 6996b368cd4SJérôme Glisse 700e7638488SDan Williamsconfig DEV_PAGEMAP_OPS 701e7638488SDan Williams bool 702e7638488SDan Williams 703133ff0eaSJérôme Glisseconfig HMM 704133ff0eaSJérôme Glisse bool 705734fb899SJérôme Glisse select MMU_NOTIFIER 7066b368cd4SJérôme Glisse select MIGRATE_VMA_HELPER 707133ff0eaSJérôme Glisse 708c0b12405SJérôme Glisseconfig HMM_MIRROR 709c0b12405SJérôme Glisse bool "HMM mirror CPU page table into a device page table" 710c0b12405SJérôme Glisse depends on ARCH_HAS_HMM 711c0b12405SJérôme Glisse select HMM 712c0b12405SJérôme Glisse help 713c0b12405SJérôme Glisse Select HMM_MIRROR if you want to mirror range of the CPU page table of a 714c0b12405SJérôme Glisse process into a device page table. Here, mirror means "keep synchronized". 715c0b12405SJérôme Glisse Prerequisites: the device must provide the ability to write-protect its 716c0b12405SJérôme Glisse page tables (at PAGE_SIZE granularity), and must be able to recover from 717c0b12405SJérôme Glisse the resulting potential page faults. 718c0b12405SJérôme Glisse 7195042db43SJérôme Glisseconfig DEVICE_PRIVATE 7205042db43SJérôme Glisse bool "Unaddressable device memory (GPU memory, ...)" 7215042db43SJérôme Glisse depends on ARCH_HAS_HMM 722df6ad698SJérôme Glisse select HMM 723e7638488SDan Williams select DEV_PAGEMAP_OPS 7245042db43SJérôme Glisse 7255042db43SJérôme Glisse help 7265042db43SJérôme Glisse Allows creation of struct pages to represent unaddressable device 7275042db43SJérôme Glisse memory; i.e., memory that is only accessible from the device (or 7285042db43SJérôme Glisse group of devices). You likely also want to select HMM_MIRROR. 7295042db43SJérôme Glisse 730df6ad698SJérôme Glisseconfig DEVICE_PUBLIC 731df6ad698SJérôme Glisse bool "Addressable device memory (like GPU memory)" 732df6ad698SJérôme Glisse depends on ARCH_HAS_HMM 733df6ad698SJérôme Glisse select HMM 734e7638488SDan Williams select DEV_PAGEMAP_OPS 735df6ad698SJérôme Glisse 736df6ad698SJérôme Glisse help 737df6ad698SJérôme Glisse Allows creation of struct pages to represent addressable device 738df6ad698SJérôme Glisse memory; i.e., memory that is accessible from both the device and 739df6ad698SJérôme Glisse the CPU 740df6ad698SJérôme Glisse 7418025e5ddSJan Karaconfig FRAME_VECTOR 7428025e5ddSJan Kara bool 74363c17fb8SDave Hansen 74463c17fb8SDave Hansenconfig ARCH_USES_HIGH_VMA_FLAGS 74563c17fb8SDave Hansen bool 74666d37570SDave Hansenconfig ARCH_HAS_PKEYS 74766d37570SDave Hansen bool 74830a5b536SDennis Zhou 74930a5b536SDennis Zhouconfig PERCPU_STATS 75030a5b536SDennis Zhou bool "Collect percpu memory statistics" 75130a5b536SDennis Zhou help 75230a5b536SDennis Zhou This feature collects and exposes statistics via debugfs. The 75330a5b536SDennis Zhou information includes global and per chunk statistics, which can 75430a5b536SDennis Zhou be used to help understand percpu memory usage. 75564c349f4SKirill A. Shutemov 75664c349f4SKirill A. Shutemovconfig GUP_BENCHMARK 75764c349f4SKirill A. Shutemov bool "Enable infrastructure for get_user_pages_fast() benchmarking" 75864c349f4SKirill A. Shutemov help 75964c349f4SKirill A. Shutemov Provides /sys/kernel/debug/gup_benchmark that helps with testing 76064c349f4SKirill A. Shutemov performance of get_user_pages_fast(). 76164c349f4SKirill A. Shutemov 76264c349f4SKirill A. Shutemov See tools/testing/selftests/vm/gup_benchmark.c 7633010a5eaSLaurent Dufour 7643010a5eaSLaurent Dufourconfig ARCH_HAS_PTE_SPECIAL 7653010a5eaSLaurent Dufour bool 76659e0b520SChristoph Hellwig 76759e0b520SChristoph Hellwigendmenu 768