1e1785e85SDave Hansenconfig SELECT_MEMORY_MODEL 2e1785e85SDave Hansen def_bool y 3a8826eebSKees Cook depends on ARCH_SELECT_MEMORY_MODEL 4e1785e85SDave Hansen 53a9da765SDave Hansenchoice 63a9da765SDave Hansen prompt "Memory model" 7e1785e85SDave Hansen depends on SELECT_MEMORY_MODEL 8e1785e85SDave Hansen default DISCONTIGMEM_MANUAL if ARCH_DISCONTIGMEM_DEFAULT 9d41dee36SAndy Whitcroft default SPARSEMEM_MANUAL if ARCH_SPARSEMEM_DEFAULT 10e1785e85SDave Hansen default FLATMEM_MANUAL 113a9da765SDave Hansen 12e1785e85SDave Hansenconfig FLATMEM_MANUAL 133a9da765SDave Hansen bool "Flat Memory" 14c898ec16SAnton Blanchard depends on !(ARCH_DISCONTIGMEM_ENABLE || ARCH_SPARSEMEM_ENABLE) || ARCH_FLATMEM_ENABLE 153a9da765SDave Hansen help 163a9da765SDave Hansen This option allows you to change some of the ways that 173a9da765SDave Hansen Linux manages its memory internally. Most users will 183a9da765SDave Hansen only have one option here: FLATMEM. This is normal 193a9da765SDave Hansen and a correct option. 203a9da765SDave Hansen 21d41dee36SAndy Whitcroft Some users of more advanced features like NUMA and 22d41dee36SAndy Whitcroft memory hotplug may have different options here. 2318f65332SGeert Uytterhoeven DISCONTIGMEM is a more mature, better tested system, 24d41dee36SAndy Whitcroft but is incompatible with memory hotplug and may suffer 25d41dee36SAndy Whitcroft decreased performance over SPARSEMEM. If unsure between 26d41dee36SAndy Whitcroft "Sparse Memory" and "Discontiguous Memory", choose 27d41dee36SAndy Whitcroft "Discontiguous Memory". 28d41dee36SAndy Whitcroft 29d41dee36SAndy Whitcroft If unsure, choose this option (Flat Memory) over any other. 303a9da765SDave Hansen 31e1785e85SDave Hansenconfig DISCONTIGMEM_MANUAL 32f3519f91SDave Hansen bool "Discontiguous Memory" 333a9da765SDave Hansen depends on ARCH_DISCONTIGMEM_ENABLE 343a9da765SDave Hansen help 35785dcd44SDave Hansen This option provides enhanced support for discontiguous 36785dcd44SDave Hansen memory systems, over FLATMEM. These systems have holes 37785dcd44SDave Hansen in their physical address spaces, and this option provides 38785dcd44SDave Hansen more efficient handling of these holes. However, the vast 39785dcd44SDave Hansen majority of hardware has quite flat address spaces, and 40ad3d0a38SPhilipp Marek can have degraded performance from the extra overhead that 41785dcd44SDave Hansen this option imposes. 42785dcd44SDave Hansen 43785dcd44SDave Hansen Many NUMA configurations will have this as the only option. 44785dcd44SDave Hansen 453a9da765SDave Hansen If unsure, choose "Flat Memory" over this option. 463a9da765SDave Hansen 47d41dee36SAndy Whitcroftconfig SPARSEMEM_MANUAL 48d41dee36SAndy Whitcroft bool "Sparse Memory" 49d41dee36SAndy Whitcroft depends on ARCH_SPARSEMEM_ENABLE 50d41dee36SAndy Whitcroft help 51d41dee36SAndy Whitcroft This will be the only option for some systems, including 52d41dee36SAndy Whitcroft memory hotplug systems. This is normal. 53d41dee36SAndy Whitcroft 54d41dee36SAndy Whitcroft For many other systems, this will be an alternative to 55f3519f91SDave Hansen "Discontiguous Memory". This option provides some potential 56d41dee36SAndy Whitcroft performance benefits, along with decreased code complexity, 57d41dee36SAndy Whitcroft but it is newer, and more experimental. 58d41dee36SAndy Whitcroft 59d41dee36SAndy Whitcroft If unsure, choose "Discontiguous Memory" or "Flat Memory" 60d41dee36SAndy Whitcroft over this option. 61d41dee36SAndy Whitcroft 623a9da765SDave Hansenendchoice 633a9da765SDave Hansen 64e1785e85SDave Hansenconfig DISCONTIGMEM 65e1785e85SDave Hansen def_bool y 66e1785e85SDave Hansen depends on (!SELECT_MEMORY_MODEL && ARCH_DISCONTIGMEM_ENABLE) || DISCONTIGMEM_MANUAL 67e1785e85SDave Hansen 68d41dee36SAndy Whitcroftconfig SPARSEMEM 69d41dee36SAndy Whitcroft def_bool y 701a83e175SRussell King depends on (!SELECT_MEMORY_MODEL && ARCH_SPARSEMEM_ENABLE) || SPARSEMEM_MANUAL 71d41dee36SAndy Whitcroft 72e1785e85SDave Hansenconfig FLATMEM 73e1785e85SDave Hansen def_bool y 74d41dee36SAndy Whitcroft depends on (!DISCONTIGMEM && !SPARSEMEM) || FLATMEM_MANUAL 75d41dee36SAndy Whitcroft 76d41dee36SAndy Whitcroftconfig FLAT_NODE_MEM_MAP 77d41dee36SAndy Whitcroft def_bool y 78d41dee36SAndy Whitcroft depends on !SPARSEMEM 79e1785e85SDave Hansen 8093b7504eSDave Hansen# 8193b7504eSDave Hansen# Both the NUMA code and DISCONTIGMEM use arrays of pg_data_t's 8293b7504eSDave Hansen# to represent different areas of memory. This variable allows 8393b7504eSDave Hansen# those dependencies to exist individually. 8493b7504eSDave Hansen# 8593b7504eSDave Hansenconfig NEED_MULTIPLE_NODES 8693b7504eSDave Hansen def_bool y 8793b7504eSDave Hansen depends on DISCONTIGMEM || NUMA 88af705362SAndy Whitcroft 89af705362SAndy Whitcroftconfig HAVE_MEMORY_PRESENT 90af705362SAndy Whitcroft def_bool y 91d41dee36SAndy Whitcroft depends on ARCH_HAVE_MEMORY_PRESENT || SPARSEMEM 92802f192eSBob Picco 93802f192eSBob Picco# 943e347261SBob Picco# SPARSEMEM_EXTREME (which is the default) does some bootmem 953e347261SBob Picco# allocations when memory_present() is called. If this cannot 963e347261SBob Picco# be done on your architecture, select this option. However, 973e347261SBob Picco# statically allocating the mem_section[] array can potentially 983e347261SBob Picco# consume vast quantities of .bss, so be careful. 993e347261SBob Picco# 1003e347261SBob Picco# This option will also potentially produce smaller runtime code 1013e347261SBob Picco# with gcc 3.4 and later. 1023e347261SBob Picco# 1033e347261SBob Piccoconfig SPARSEMEM_STATIC 1049ba16087SJan Beulich bool 1053e347261SBob Picco 1063e347261SBob Picco# 10744c09201SMatt LaPlante# Architecture platforms which require a two level mem_section in SPARSEMEM 108802f192eSBob Picco# must select this option. This is usually for architecture platforms with 109802f192eSBob Picco# an extremely sparse physical address space. 110802f192eSBob Picco# 1113e347261SBob Piccoconfig SPARSEMEM_EXTREME 1123e347261SBob Picco def_bool y 1133e347261SBob Picco depends on SPARSEMEM && !SPARSEMEM_STATIC 1144c21e2f2SHugh Dickins 11529c71111SAndy Whitcroftconfig SPARSEMEM_VMEMMAP_ENABLE 1169ba16087SJan Beulich bool 11729c71111SAndy Whitcroft 1189bdac914SYinghai Luconfig SPARSEMEM_ALLOC_MEM_MAP_TOGETHER 1199bdac914SYinghai Lu def_bool y 1209bdac914SYinghai Lu depends on SPARSEMEM && X86_64 1219bdac914SYinghai Lu 12229c71111SAndy Whitcroftconfig SPARSEMEM_VMEMMAP 123a5ee6daaSGeoff Levand bool "Sparse Memory virtual memmap" 124a5ee6daaSGeoff Levand depends on SPARSEMEM && SPARSEMEM_VMEMMAP_ENABLE 125a5ee6daaSGeoff Levand default y 126a5ee6daaSGeoff Levand help 127a5ee6daaSGeoff Levand SPARSEMEM_VMEMMAP uses a virtually mapped memmap to optimise 128a5ee6daaSGeoff Levand pfn_to_page and page_to_pfn operations. This is the most 129a5ee6daaSGeoff Levand efficient option when sufficient kernel resources are available. 13029c71111SAndy Whitcroft 13195f72d1eSYinghai Luconfig HAVE_MEMBLOCK 1326341e62bSChristoph Jaeger bool 13395f72d1eSYinghai Lu 1347c0caeb8STejun Heoconfig HAVE_MEMBLOCK_NODE_MAP 1356341e62bSChristoph Jaeger bool 1367c0caeb8STejun Heo 13770210ed9SPhilipp Hachtmannconfig HAVE_MEMBLOCK_PHYS_MAP 1386341e62bSChristoph Jaeger bool 13970210ed9SPhilipp Hachtmann 140e585513bSKirill A. Shutemovconfig HAVE_GENERIC_GUP 1416341e62bSChristoph Jaeger bool 1422667f50eSSteve Capper 143c378ddd5STejun Heoconfig ARCH_DISCARD_MEMBLOCK 1446341e62bSChristoph Jaeger bool 145c378ddd5STejun Heo 14666616720SSam Ravnborgconfig NO_BOOTMEM 1476341e62bSChristoph Jaeger bool 14866616720SSam Ravnborg 149ee6f509cSMinchan Kimconfig MEMORY_ISOLATION 1506341e62bSChristoph Jaeger bool 151ee6f509cSMinchan Kim 15246723bfaSYasuaki Ishimatsu# 15346723bfaSYasuaki Ishimatsu# Only be set on architectures that have completely implemented memory hotplug 15446723bfaSYasuaki Ishimatsu# feature. If you are not sure, don't touch it. 15546723bfaSYasuaki Ishimatsu# 15646723bfaSYasuaki Ishimatsuconfig HAVE_BOOTMEM_INFO_NODE 15746723bfaSYasuaki Ishimatsu def_bool n 15846723bfaSYasuaki Ishimatsu 1593947be19SDave Hansen# eventually, we can have this option just 'select SPARSEMEM' 1603947be19SDave Hansenconfig MEMORY_HOTPLUG 1613947be19SDave Hansen bool "Allow for memory hot-add" 162ec69acbbSKeith Mannthey depends on SPARSEMEM || X86_64_ACPI_NUMA 16340b31360SStephen Rothwell depends on ARCH_ENABLE_MEMORY_HOTPLUG 1643947be19SDave Hansen 165ec69acbbSKeith Manntheyconfig MEMORY_HOTPLUG_SPARSE 166ec69acbbSKeith Mannthey def_bool y 167ec69acbbSKeith Mannthey depends on SPARSEMEM && MEMORY_HOTPLUG 168ec69acbbSKeith Mannthey 1698604d9e5SVitaly Kuznetsovconfig MEMORY_HOTPLUG_DEFAULT_ONLINE 1708604d9e5SVitaly Kuznetsov bool "Online the newly added memory blocks by default" 1718604d9e5SVitaly Kuznetsov default n 1728604d9e5SVitaly Kuznetsov depends on MEMORY_HOTPLUG 1738604d9e5SVitaly Kuznetsov help 1748604d9e5SVitaly Kuznetsov This option sets the default policy setting for memory hotplug 1758604d9e5SVitaly Kuznetsov onlining policy (/sys/devices/system/memory/auto_online_blocks) which 1768604d9e5SVitaly Kuznetsov determines what happens to newly added memory regions. Policy setting 1778604d9e5SVitaly Kuznetsov can always be changed at runtime. 1788604d9e5SVitaly Kuznetsov See Documentation/memory-hotplug.txt for more information. 1798604d9e5SVitaly Kuznetsov 1808604d9e5SVitaly Kuznetsov Say Y here if you want all hot-plugged memory blocks to appear in 1818604d9e5SVitaly Kuznetsov 'online' state by default. 1828604d9e5SVitaly Kuznetsov Say N here if you want the default policy to keep all hot-plugged 1838604d9e5SVitaly Kuznetsov memory blocks in 'offline' state. 1848604d9e5SVitaly Kuznetsov 1850c0e6195SKAMEZAWA Hiroyukiconfig MEMORY_HOTREMOVE 1860c0e6195SKAMEZAWA Hiroyuki bool "Allow for memory hot remove" 18746723bfaSYasuaki Ishimatsu select MEMORY_ISOLATION 188f7e3334aSNathan Fontenot select HAVE_BOOTMEM_INFO_NODE if (X86_64 || PPC64) 1890c0e6195SKAMEZAWA Hiroyuki depends on MEMORY_HOTPLUG && ARCH_ENABLE_MEMORY_HOTREMOVE 1900c0e6195SKAMEZAWA Hiroyuki depends on MIGRATION 1910c0e6195SKAMEZAWA Hiroyuki 1924c21e2f2SHugh Dickins# Heavily threaded applications may benefit from splitting the mm-wide 1934c21e2f2SHugh Dickins# page_table_lock, so that faults on different parts of the user address 1944c21e2f2SHugh Dickins# space can be handled with less contention: split it at this NR_CPUS. 1954c21e2f2SHugh Dickins# Default to 4 for wider testing, though 8 might be more appropriate. 1964c21e2f2SHugh Dickins# ARM's adjust_pte (unused if VIPT) depends on mm-wide page_table_lock. 1977b6ac9dfSHugh Dickins# PA-RISC 7xxx's spinlock_t would enlarge struct page from 32 to 44 bytes. 198a70caa8bSHugh Dickins# DEBUG_SPINLOCK and DEBUG_LOCK_ALLOC spinlock_t also enlarge struct page. 1994c21e2f2SHugh Dickins# 2004c21e2f2SHugh Dickinsconfig SPLIT_PTLOCK_CPUS 2014c21e2f2SHugh Dickins int 2029164550eSKirill A. Shutemov default "999999" if !MMU 203a70caa8bSHugh Dickins default "999999" if ARM && !CPU_CACHE_VIPT 204a70caa8bSHugh Dickins default "999999" if PARISC && !PA20 2054c21e2f2SHugh Dickins default "4" 2067cbe34cfSChristoph Lameter 207e009bb30SKirill A. Shutemovconfig ARCH_ENABLE_SPLIT_PMD_PTLOCK 2086341e62bSChristoph Jaeger bool 209e009bb30SKirill A. Shutemov 2107cbe34cfSChristoph Lameter# 21109316c09SKonstantin Khlebnikov# support for memory balloon 21209316c09SKonstantin Khlebnikovconfig MEMORY_BALLOON 2136341e62bSChristoph Jaeger bool 21409316c09SKonstantin Khlebnikov 21509316c09SKonstantin Khlebnikov# 21618468d93SRafael Aquini# support for memory balloon compaction 21718468d93SRafael Aquiniconfig BALLOON_COMPACTION 21818468d93SRafael Aquini bool "Allow for balloon memory compaction/migration" 21918468d93SRafael Aquini def_bool y 22009316c09SKonstantin Khlebnikov depends on COMPACTION && MEMORY_BALLOON 22118468d93SRafael Aquini help 22218468d93SRafael Aquini Memory fragmentation introduced by ballooning might reduce 22318468d93SRafael Aquini significantly the number of 2MB contiguous memory blocks that can be 22418468d93SRafael Aquini used within a guest, thus imposing performance penalties associated 22518468d93SRafael Aquini with the reduced number of transparent huge pages that could be used 22618468d93SRafael Aquini by the guest workload. Allowing the compaction & migration for memory 22718468d93SRafael Aquini pages enlisted as being part of memory balloon devices avoids the 22818468d93SRafael Aquini scenario aforementioned and helps improving memory defragmentation. 22918468d93SRafael Aquini 23018468d93SRafael Aquini# 231e9e96b39SMel Gorman# support for memory compaction 232e9e96b39SMel Gormanconfig COMPACTION 233e9e96b39SMel Gorman bool "Allow for memory compaction" 23405106e6aSRik van Riel def_bool y 235e9e96b39SMel Gorman select MIGRATION 23633a93877SAndrea Arcangeli depends on MMU 237e9e96b39SMel Gorman help 238b32eaf71SMichal Hocko Compaction is the only memory management component to form 239b32eaf71SMichal Hocko high order (larger physically contiguous) memory blocks 240b32eaf71SMichal Hocko reliably. The page allocator relies on compaction heavily and 241b32eaf71SMichal Hocko the lack of the feature can lead to unexpected OOM killer 242b32eaf71SMichal Hocko invocations for high order memory requests. You shouldn't 243b32eaf71SMichal Hocko disable this option unless there really is a strong reason for 244b32eaf71SMichal Hocko it and then we would be really interested to hear about that at 245b32eaf71SMichal Hocko linux-mm@kvack.org. 246e9e96b39SMel Gorman 247e9e96b39SMel Gorman# 2487cbe34cfSChristoph Lameter# support for page migration 2497cbe34cfSChristoph Lameter# 2507cbe34cfSChristoph Lameterconfig MIGRATION 251b20a3503SChristoph Lameter bool "Page migration" 2526c5240aeSChristoph Lameter def_bool y 253de32a817SChen Gang depends on (NUMA || ARCH_ENABLE_MEMORY_HOTREMOVE || COMPACTION || CMA) && MMU 254b20a3503SChristoph Lameter help 255b20a3503SChristoph Lameter Allows the migration of the physical location of pages of processes 256e9e96b39SMel Gorman while the virtual addresses are not changed. This is useful in 257e9e96b39SMel Gorman two situations. The first is on NUMA systems to put pages nearer 258e9e96b39SMel Gorman to the processors accessing. The second is when allocating huge 259e9e96b39SMel Gorman pages as migration can relocate pages to satisfy a huge page 260e9e96b39SMel Gorman allocation instead of reclaiming. 2616550e07fSGreg Kroah-Hartman 262c177c81eSNaoya Horiguchiconfig ARCH_ENABLE_HUGEPAGE_MIGRATION 2636341e62bSChristoph Jaeger bool 264c177c81eSNaoya Horiguchi 2659c670ea3SNaoya Horiguchiconfig ARCH_ENABLE_THP_MIGRATION 2669c670ea3SNaoya Horiguchi bool 2679c670ea3SNaoya Horiguchi 268600715dcSJeremy Fitzhardingeconfig PHYS_ADDR_T_64BIT 269600715dcSJeremy Fitzhardinge def_bool 64BIT || ARCH_PHYS_ADDR_T_64BIT 270600715dcSJeremy Fitzhardinge 2712a7326b5SChristoph Lameterconfig BOUNCE 2729ca24e2eSVinayak Menon bool "Enable bounce buffers" 2739ca24e2eSVinayak Menon default y 2742a7326b5SChristoph Lameter depends on BLOCK && MMU && (ZONE_DMA || HIGHMEM) 2759ca24e2eSVinayak Menon help 2769ca24e2eSVinayak Menon Enable bounce buffers for devices that cannot access 2779ca24e2eSVinayak Menon the full range of memory available to the CPU. Enabled 2789ca24e2eSVinayak Menon by default when ZONE_DMA or HIGHMEM is selected, but you 2799ca24e2eSVinayak Menon may say n to override this. 2802a7326b5SChristoph Lameter 281ffecfd1aSDarrick J. Wong# On the 'tile' arch, USB OHCI needs the bounce pool since tilegx will often 282ffecfd1aSDarrick J. Wong# have more than 4GB of memory, but we don't currently use the IOTLB to present 283ffecfd1aSDarrick J. Wong# a 32-bit address to OHCI. So we need to use a bounce pool instead. 284ffecfd1aSDarrick J. Wongconfig NEED_BOUNCE_POOL 285ffecfd1aSDarrick J. Wong bool 286debeb297SValentin Rothberg default y if TILE && USB_OHCI_HCD 287ffecfd1aSDarrick J. Wong 2886225e937SChristoph Lameterconfig NR_QUICK 2896225e937SChristoph Lameter int 2906225e937SChristoph Lameter depends on QUICKLIST 2916225e937SChristoph Lameter default "1" 292f057eac0SStephen Rothwell 293f057eac0SStephen Rothwellconfig VIRT_TO_BUS 2944febd95aSStephen Rothwell bool 2954febd95aSStephen Rothwell help 2964febd95aSStephen Rothwell An architecture should select this if it implements the 2974febd95aSStephen Rothwell deprecated interface virt_to_bus(). All new architectures 2984febd95aSStephen Rothwell should probably not select this. 2994febd95aSStephen Rothwell 300cddb8a5cSAndrea Arcangeli 301cddb8a5cSAndrea Arcangeliconfig MMU_NOTIFIER 302cddb8a5cSAndrea Arcangeli bool 30383fe27eaSPranith Kumar select SRCU 304fc4d5c29SDavid Howells 305f8af4da3SHugh Dickinsconfig KSM 306f8af4da3SHugh Dickins bool "Enable KSM for page merging" 307f8af4da3SHugh Dickins depends on MMU 308f8af4da3SHugh Dickins help 309f8af4da3SHugh Dickins Enable Kernel Samepage Merging: KSM periodically scans those areas 310f8af4da3SHugh Dickins of an application's address space that an app has advised may be 311f8af4da3SHugh Dickins mergeable. When it finds pages of identical content, it replaces 312d0f209f6SHugh Dickins the many instances by a single page with that content, so 313f8af4da3SHugh Dickins saving memory until one or another app needs to modify the content. 314f8af4da3SHugh Dickins Recommended for use with KVM, or with other duplicative applications. 315c73602adSHugh Dickins See Documentation/vm/ksm.txt for more information: KSM is inactive 316c73602adSHugh Dickins until a program has madvised that an area is MADV_MERGEABLE, and 317c73602adSHugh Dickins root has set /sys/kernel/mm/ksm/run to 1 (if CONFIG_SYSFS is set). 318f8af4da3SHugh Dickins 319e0a94c2aSChristoph Lameterconfig DEFAULT_MMAP_MIN_ADDR 320e0a94c2aSChristoph Lameter int "Low address space to protect from user allocation" 3216e141546SDavid Howells depends on MMU 322e0a94c2aSChristoph Lameter default 4096 323e0a94c2aSChristoph Lameter help 324e0a94c2aSChristoph Lameter This is the portion of low virtual memory which should be protected 325e0a94c2aSChristoph Lameter from userspace allocation. Keeping a user from writing to low pages 326e0a94c2aSChristoph Lameter can help reduce the impact of kernel NULL pointer bugs. 327e0a94c2aSChristoph Lameter 328e0a94c2aSChristoph Lameter For most ia64, ppc64 and x86 users with lots of address space 329e0a94c2aSChristoph Lameter a value of 65536 is reasonable and should cause no problems. 330e0a94c2aSChristoph Lameter On arm and other archs it should not be higher than 32768. 331788084abSEric Paris Programs which use vm86 functionality or have some need to map 332788084abSEric Paris this low address space will need CAP_SYS_RAWIO or disable this 333788084abSEric Paris protection by setting the value to 0. 334e0a94c2aSChristoph Lameter 335e0a94c2aSChristoph Lameter This value can be changed after boot using the 336e0a94c2aSChristoph Lameter /proc/sys/vm/mmap_min_addr tunable. 337e0a94c2aSChristoph Lameter 338d949f36fSLinus Torvaldsconfig ARCH_SUPPORTS_MEMORY_FAILURE 339d949f36fSLinus Torvalds bool 340e0a94c2aSChristoph Lameter 3416a46079cSAndi Kleenconfig MEMORY_FAILURE 3426a46079cSAndi Kleen depends on MMU 343d949f36fSLinus Torvalds depends on ARCH_SUPPORTS_MEMORY_FAILURE 3446a46079cSAndi Kleen bool "Enable recovery from hardware memory errors" 345ee6f509cSMinchan Kim select MEMORY_ISOLATION 34697f0b134SXie XiuQi select RAS 3476a46079cSAndi Kleen help 3486a46079cSAndi Kleen Enables code to recover from some memory failures on systems 3496a46079cSAndi Kleen with MCA recovery. This allows a system to continue running 3506a46079cSAndi Kleen even when some of its memory has uncorrected errors. This requires 3516a46079cSAndi Kleen special hardware support and typically ECC memory. 3526a46079cSAndi Kleen 353cae681fcSAndi Kleenconfig HWPOISON_INJECT 354413f9efbSAndi Kleen tristate "HWPoison pages injector" 35527df5068SAndi Kleen depends on MEMORY_FAILURE && DEBUG_KERNEL && PROC_FS 356478c5ffcSWu Fengguang select PROC_PAGE_MONITOR 357cae681fcSAndi Kleen 358fc4d5c29SDavid Howellsconfig NOMMU_INITIAL_TRIM_EXCESS 359fc4d5c29SDavid Howells int "Turn on mmap() excess space trimming before booting" 360fc4d5c29SDavid Howells depends on !MMU 361fc4d5c29SDavid Howells default 1 362fc4d5c29SDavid Howells help 363fc4d5c29SDavid Howells The NOMMU mmap() frequently needs to allocate large contiguous chunks 364fc4d5c29SDavid Howells of memory on which to store mappings, but it can only ask the system 365fc4d5c29SDavid Howells allocator for chunks in 2^N*PAGE_SIZE amounts - which is frequently 366fc4d5c29SDavid Howells more than it requires. To deal with this, mmap() is able to trim off 367fc4d5c29SDavid Howells the excess and return it to the allocator. 368fc4d5c29SDavid Howells 369fc4d5c29SDavid Howells If trimming is enabled, the excess is trimmed off and returned to the 370fc4d5c29SDavid Howells system allocator, which can cause extra fragmentation, particularly 371fc4d5c29SDavid Howells if there are a lot of transient processes. 372fc4d5c29SDavid Howells 373fc4d5c29SDavid Howells If trimming is disabled, the excess is kept, but not used, which for 374fc4d5c29SDavid Howells long-term mappings means that the space is wasted. 375fc4d5c29SDavid Howells 376fc4d5c29SDavid Howells Trimming can be dynamically controlled through a sysctl option 377fc4d5c29SDavid Howells (/proc/sys/vm/nr_trim_pages) which specifies the minimum number of 378fc4d5c29SDavid Howells excess pages there must be before trimming should occur, or zero if 379fc4d5c29SDavid Howells no trimming is to occur. 380fc4d5c29SDavid Howells 381fc4d5c29SDavid Howells This option specifies the initial value of this option. The default 382fc4d5c29SDavid Howells of 1 says that all excess pages should be trimmed. 383fc4d5c29SDavid Howells 384fc4d5c29SDavid Howells See Documentation/nommu-mmap.txt for more information. 385bbddff05STejun Heo 3864c76d9d1SAndrea Arcangeliconfig TRANSPARENT_HUGEPAGE 38713ece886SAndrea Arcangeli bool "Transparent Hugepage Support" 38815626062SGerald Schaefer depends on HAVE_ARCH_TRANSPARENT_HUGEPAGE 3895d689240SAndrea Arcangeli select COMPACTION 39057578c2eSMatthew Wilcox select RADIX_TREE_MULTIORDER 3914c76d9d1SAndrea Arcangeli help 3924c76d9d1SAndrea Arcangeli Transparent Hugepages allows the kernel to use huge pages and 3934c76d9d1SAndrea Arcangeli huge tlb transparently to the applications whenever possible. 3944c76d9d1SAndrea Arcangeli This feature can improve computing performance to certain 3954c76d9d1SAndrea Arcangeli applications by speeding up page faults during memory 3964c76d9d1SAndrea Arcangeli allocation, by reducing the number of tlb misses and by speeding 3974c76d9d1SAndrea Arcangeli up the pagetable walking. 3984c76d9d1SAndrea Arcangeli 3994c76d9d1SAndrea Arcangeli If memory constrained on embedded, you may want to say N. 4004c76d9d1SAndrea Arcangeli 40113ece886SAndrea Arcangelichoice 40213ece886SAndrea Arcangeli prompt "Transparent Hugepage Support sysfs defaults" 40313ece886SAndrea Arcangeli depends on TRANSPARENT_HUGEPAGE 40413ece886SAndrea Arcangeli default TRANSPARENT_HUGEPAGE_ALWAYS 40513ece886SAndrea Arcangeli help 40613ece886SAndrea Arcangeli Selects the sysfs defaults for Transparent Hugepage Support. 40713ece886SAndrea Arcangeli 40813ece886SAndrea Arcangeli config TRANSPARENT_HUGEPAGE_ALWAYS 40913ece886SAndrea Arcangeli bool "always" 41013ece886SAndrea Arcangeli help 41113ece886SAndrea Arcangeli Enabling Transparent Hugepage always, can increase the 41213ece886SAndrea Arcangeli memory footprint of applications without a guaranteed 41313ece886SAndrea Arcangeli benefit but it will work automatically for all applications. 41413ece886SAndrea Arcangeli 41513ece886SAndrea Arcangeli config TRANSPARENT_HUGEPAGE_MADVISE 41613ece886SAndrea Arcangeli bool "madvise" 41713ece886SAndrea Arcangeli help 41813ece886SAndrea Arcangeli Enabling Transparent Hugepage madvise, will only provide a 41913ece886SAndrea Arcangeli performance improvement benefit to the applications using 42013ece886SAndrea Arcangeli madvise(MADV_HUGEPAGE) but it won't risk to increase the 42113ece886SAndrea Arcangeli memory footprint of applications without a guaranteed 42213ece886SAndrea Arcangeli benefit. 42313ece886SAndrea Arcangeliendchoice 42413ece886SAndrea Arcangeli 42538d8b4e6SHuang Yingconfig ARCH_WANTS_THP_SWAP 42638d8b4e6SHuang Ying def_bool n 42738d8b4e6SHuang Ying 42838d8b4e6SHuang Yingconfig THP_SWAP 42938d8b4e6SHuang Ying def_bool y 43038d8b4e6SHuang Ying depends on TRANSPARENT_HUGEPAGE && ARCH_WANTS_THP_SWAP 43138d8b4e6SHuang Ying help 43238d8b4e6SHuang Ying Swap transparent huge pages in one piece, without splitting. 43338d8b4e6SHuang Ying XXX: For now this only does clustered swap space allocation. 43438d8b4e6SHuang Ying 43538d8b4e6SHuang Ying For selection by architectures with reasonable THP sizes. 43638d8b4e6SHuang Ying 437e496cf3dSKirill A. Shutemovconfig TRANSPARENT_HUGE_PAGECACHE 438e496cf3dSKirill A. Shutemov def_bool y 439953c66c2SAneesh Kumar K.V depends on TRANSPARENT_HUGEPAGE 440e496cf3dSKirill A. Shutemov 441e496cf3dSKirill A. Shutemov# 442bbddff05STejun Heo# UP and nommu archs use km based percpu allocator 443bbddff05STejun Heo# 444bbddff05STejun Heoconfig NEED_PER_CPU_KM 445bbddff05STejun Heo depends on !SMP 446bbddff05STejun Heo bool 447bbddff05STejun Heo default y 448077b1f83SDan Magenheimer 449077b1f83SDan Magenheimerconfig CLEANCACHE 450077b1f83SDan Magenheimer bool "Enable cleancache driver to cache clean pages if tmem is present" 451077b1f83SDan Magenheimer default n 452077b1f83SDan Magenheimer help 453077b1f83SDan Magenheimer Cleancache can be thought of as a page-granularity victim cache 454077b1f83SDan Magenheimer for clean pages that the kernel's pageframe replacement algorithm 455077b1f83SDan Magenheimer (PFRA) would like to keep around, but can't since there isn't enough 456077b1f83SDan Magenheimer memory. So when the PFRA "evicts" a page, it first attempts to use 457140a1ef2SMichael Witten cleancache code to put the data contained in that page into 458077b1f83SDan Magenheimer "transcendent memory", memory that is not directly accessible or 459077b1f83SDan Magenheimer addressable by the kernel and is of unknown and possibly 460077b1f83SDan Magenheimer time-varying size. And when a cleancache-enabled 461077b1f83SDan Magenheimer filesystem wishes to access a page in a file on disk, it first 462077b1f83SDan Magenheimer checks cleancache to see if it already contains it; if it does, 463077b1f83SDan Magenheimer the page is copied into the kernel and a disk access is avoided. 464077b1f83SDan Magenheimer When a transcendent memory driver is available (such as zcache or 465077b1f83SDan Magenheimer Xen transcendent memory), a significant I/O reduction 466077b1f83SDan Magenheimer may be achieved. When none is available, all cleancache calls 467077b1f83SDan Magenheimer are reduced to a single pointer-compare-against-NULL resulting 468077b1f83SDan Magenheimer in a negligible performance hit. 469077b1f83SDan Magenheimer 470077b1f83SDan Magenheimer If unsure, say Y to enable cleancache 47127c6aec2SDan Magenheimer 47227c6aec2SDan Magenheimerconfig FRONTSWAP 47327c6aec2SDan Magenheimer bool "Enable frontswap to cache swap pages if tmem is present" 47427c6aec2SDan Magenheimer depends on SWAP 47527c6aec2SDan Magenheimer default n 47627c6aec2SDan Magenheimer help 47727c6aec2SDan Magenheimer Frontswap is so named because it can be thought of as the opposite 47827c6aec2SDan Magenheimer of a "backing" store for a swap device. The data is stored into 47927c6aec2SDan Magenheimer "transcendent memory", memory that is not directly accessible or 48027c6aec2SDan Magenheimer addressable by the kernel and is of unknown and possibly 48127c6aec2SDan Magenheimer time-varying size. When space in transcendent memory is available, 48227c6aec2SDan Magenheimer a significant swap I/O reduction may be achieved. When none is 48327c6aec2SDan Magenheimer available, all frontswap calls are reduced to a single pointer- 48427c6aec2SDan Magenheimer compare-against-NULL resulting in a negligible performance hit 48527c6aec2SDan Magenheimer and swap data is stored as normal on the matching swap device. 48627c6aec2SDan Magenheimer 48727c6aec2SDan Magenheimer If unsure, say Y to enable frontswap. 488f825c736SAneesh Kumar K.V 489f825c736SAneesh Kumar K.Vconfig CMA 490f825c736SAneesh Kumar K.V bool "Contiguous Memory Allocator" 491de32a817SChen Gang depends on HAVE_MEMBLOCK && MMU 492f825c736SAneesh Kumar K.V select MIGRATION 493f825c736SAneesh Kumar K.V select MEMORY_ISOLATION 494f825c736SAneesh Kumar K.V help 495f825c736SAneesh Kumar K.V This enables the Contiguous Memory Allocator which allows other 496f825c736SAneesh Kumar K.V subsystems to allocate big physically-contiguous blocks of memory. 497f825c736SAneesh Kumar K.V CMA reserves a region of memory and allows only movable pages to 498f825c736SAneesh Kumar K.V be allocated from it. This way, the kernel can use the memory for 499f825c736SAneesh Kumar K.V pagecache and when a subsystem requests for contiguous area, the 500f825c736SAneesh Kumar K.V allocated pages are migrated away to serve the contiguous request. 501f825c736SAneesh Kumar K.V 502f825c736SAneesh Kumar K.V If unsure, say "n". 503f825c736SAneesh Kumar K.V 504f825c736SAneesh Kumar K.Vconfig CMA_DEBUG 505f825c736SAneesh Kumar K.V bool "CMA debug messages (DEVELOPMENT)" 506f825c736SAneesh Kumar K.V depends on DEBUG_KERNEL && CMA 507f825c736SAneesh Kumar K.V help 508f825c736SAneesh Kumar K.V Turns on debug messages in CMA. This produces KERN_DEBUG 509f825c736SAneesh Kumar K.V messages for every CMA call as well as various messages while 510f825c736SAneesh Kumar K.V processing calls such as dma_alloc_from_contiguous(). 511f825c736SAneesh Kumar K.V This option does not affect warning and error messages. 512bf550fc9SAlexander Graf 51328b24c1fSSasha Levinconfig CMA_DEBUGFS 51428b24c1fSSasha Levin bool "CMA debugfs interface" 51528b24c1fSSasha Levin depends on CMA && DEBUG_FS 51628b24c1fSSasha Levin help 51728b24c1fSSasha Levin Turns on the DebugFS interface for CMA. 51828b24c1fSSasha Levin 519a254129eSJoonsoo Kimconfig CMA_AREAS 520a254129eSJoonsoo Kim int "Maximum count of the CMA areas" 521a254129eSJoonsoo Kim depends on CMA 522a254129eSJoonsoo Kim default 7 523a254129eSJoonsoo Kim help 524a254129eSJoonsoo Kim CMA allows to create CMA areas for particular purpose, mainly, 525a254129eSJoonsoo Kim used as device private area. This parameter sets the maximum 526a254129eSJoonsoo Kim number of CMA area in the system. 527a254129eSJoonsoo Kim 528a254129eSJoonsoo Kim If unsure, leave the default value "7". 529a254129eSJoonsoo Kim 530af8d417aSDan Streetmanconfig MEM_SOFT_DIRTY 531af8d417aSDan Streetman bool "Track memory changes" 532af8d417aSDan Streetman depends on CHECKPOINT_RESTORE && HAVE_ARCH_SOFT_DIRTY && PROC_FS 533af8d417aSDan Streetman select PROC_PAGE_MONITOR 5344e2e2770SSeth Jennings help 535af8d417aSDan Streetman This option enables memory changes tracking by introducing a 536af8d417aSDan Streetman soft-dirty bit on pte-s. This bit it set when someone writes 537af8d417aSDan Streetman into a page just as regular dirty bit, but unlike the latter 538af8d417aSDan Streetman it can be cleared by hands. 539af8d417aSDan Streetman 540af8d417aSDan Streetman See Documentation/vm/soft-dirty.txt for more details. 5414e2e2770SSeth Jennings 5422b281117SSeth Jenningsconfig ZSWAP 5432b281117SSeth Jennings bool "Compressed cache for swap pages (EXPERIMENTAL)" 5442b281117SSeth Jennings depends on FRONTSWAP && CRYPTO=y 5452b281117SSeth Jennings select CRYPTO_LZO 54612d79d64SDan Streetman select ZPOOL 5472b281117SSeth Jennings default n 5482b281117SSeth Jennings help 5492b281117SSeth Jennings A lightweight compressed cache for swap pages. It takes 5502b281117SSeth Jennings pages that are in the process of being swapped out and attempts to 5512b281117SSeth Jennings compress them into a dynamically allocated RAM-based memory pool. 5522b281117SSeth Jennings This can result in a significant I/O reduction on swap device and, 5532b281117SSeth Jennings in the case where decompressing from RAM is faster that swap device 5542b281117SSeth Jennings reads, can also improve workload performance. 5552b281117SSeth Jennings 5562b281117SSeth Jennings This is marked experimental because it is a new feature (as of 5572b281117SSeth Jennings v3.11) that interacts heavily with memory reclaim. While these 5582b281117SSeth Jennings interactions don't cause any known issues on simple memory setups, 5592b281117SSeth Jennings they have not be fully explored on the large set of potential 5602b281117SSeth Jennings configurations and workloads that exist. 5612b281117SSeth Jennings 562af8d417aSDan Streetmanconfig ZPOOL 563af8d417aSDan Streetman tristate "Common API for compressed memory storage" 564af8d417aSDan Streetman default n 5650f8975ecSPavel Emelyanov help 566af8d417aSDan Streetman Compressed memory storage API. This allows using either zbud or 567af8d417aSDan Streetman zsmalloc. 5680f8975ecSPavel Emelyanov 569af8d417aSDan Streetmanconfig ZBUD 5709a001fc1SVitaly Wool tristate "Low (Up to 2x) density storage for compressed pages" 571af8d417aSDan Streetman default n 572af8d417aSDan Streetman help 573af8d417aSDan Streetman A special purpose allocator for storing compressed pages. 574af8d417aSDan Streetman It is designed to store up to two compressed pages per physical 575af8d417aSDan Streetman page. While this design limits storage density, it has simple and 576af8d417aSDan Streetman deterministic reclaim properties that make it preferable to a higher 577af8d417aSDan Streetman density approach when reclaim will be used. 578bcf1647dSMinchan Kim 5799a001fc1SVitaly Woolconfig Z3FOLD 5809a001fc1SVitaly Wool tristate "Up to 3x density storage for compressed pages" 5819a001fc1SVitaly Wool depends on ZPOOL 5829a001fc1SVitaly Wool default n 5839a001fc1SVitaly Wool help 5849a001fc1SVitaly Wool A special purpose allocator for storing compressed pages. 5859a001fc1SVitaly Wool It is designed to store up to three compressed pages per physical 5869a001fc1SVitaly Wool page. It is a ZBUD derivative so the simplicity and determinism are 5879a001fc1SVitaly Wool still there. 5889a001fc1SVitaly Wool 589bcf1647dSMinchan Kimconfig ZSMALLOC 590d867f203SMinchan Kim tristate "Memory allocator for compressed pages" 591bcf1647dSMinchan Kim depends on MMU 592bcf1647dSMinchan Kim default n 593bcf1647dSMinchan Kim help 594bcf1647dSMinchan Kim zsmalloc is a slab-based memory allocator designed to store 595bcf1647dSMinchan Kim compressed RAM pages. zsmalloc uses virtual memory mapping 596bcf1647dSMinchan Kim in order to reduce fragmentation. However, this results in a 597bcf1647dSMinchan Kim non-standard allocator interface where a handle, not a pointer, is 598bcf1647dSMinchan Kim returned by an alloc(). This handle must be mapped in order to 599bcf1647dSMinchan Kim access the allocated space. 600bcf1647dSMinchan Kim 601bcf1647dSMinchan Kimconfig PGTABLE_MAPPING 602bcf1647dSMinchan Kim bool "Use page table mapping to access object in zsmalloc" 603bcf1647dSMinchan Kim depends on ZSMALLOC 604bcf1647dSMinchan Kim help 605bcf1647dSMinchan Kim By default, zsmalloc uses a copy-based object mapping method to 606bcf1647dSMinchan Kim access allocations that span two pages. However, if a particular 607bcf1647dSMinchan Kim architecture (ex, ARM) performs VM mapping faster than copying, 608bcf1647dSMinchan Kim then you should select this. This causes zsmalloc to use page table 609bcf1647dSMinchan Kim mapping rather than copying for object mapping. 610bcf1647dSMinchan Kim 6112216ee85SBen Hutchings You can check speed with zsmalloc benchmark: 6122216ee85SBen Hutchings https://github.com/spartacus06/zsmapbench 6139e5c33d7SMark Salter 6140f050d99SGanesh Mahendranconfig ZSMALLOC_STAT 6150f050d99SGanesh Mahendran bool "Export zsmalloc statistics" 6160f050d99SGanesh Mahendran depends on ZSMALLOC 6170f050d99SGanesh Mahendran select DEBUG_FS 6180f050d99SGanesh Mahendran help 6190f050d99SGanesh Mahendran This option enables code in the zsmalloc to collect various 6200f050d99SGanesh Mahendran statistics about whats happening in zsmalloc and exports that 6210f050d99SGanesh Mahendran information to userspace via debugfs. 6220f050d99SGanesh Mahendran If unsure, say N. 6230f050d99SGanesh Mahendran 6249e5c33d7SMark Salterconfig GENERIC_EARLY_IOREMAP 6259e5c33d7SMark Salter bool 626042d27acSHelge Deller 627042d27acSHelge Dellerconfig MAX_STACK_SIZE_MB 628042d27acSHelge Deller int "Maximum user stack size for 32-bit processes (MB)" 629042d27acSHelge Deller default 80 630042d27acSHelge Deller range 8 256 if METAG 631042d27acSHelge Deller range 8 2048 632042d27acSHelge Deller depends on STACK_GROWSUP && (!64BIT || COMPAT) 633042d27acSHelge Deller help 634042d27acSHelge Deller This is the maximum stack size in Megabytes in the VM layout of 32-bit 635042d27acSHelge Deller user processes when the stack grows upwards (currently only on parisc 636042d27acSHelge Deller and metag arch). The stack will be located at the highest memory 637042d27acSHelge Deller address minus the given value, unless the RLIMIT_STACK hard limit is 638042d27acSHelge Deller changed to a smaller value in which case that is used. 639042d27acSHelge Deller 640042d27acSHelge Deller A sane initial value is 80 MB. 6413a80a7faSMel Gorman 6423a80a7faSMel Gorman# For architectures that support deferred memory initialisation 6433a80a7faSMel Gormanconfig ARCH_SUPPORTS_DEFERRED_STRUCT_PAGE_INIT 6443a80a7faSMel Gorman bool 6453a80a7faSMel Gorman 6463a80a7faSMel Gormanconfig DEFERRED_STRUCT_PAGE_INIT 6471ce22103SVlastimil Babka bool "Defer initialisation of struct pages to kthreads" 6483a80a7faSMel Gorman default n 6493a80a7faSMel Gorman depends on ARCH_SUPPORTS_DEFERRED_STRUCT_PAGE_INIT 65011e68567SGavin Shan depends on NO_BOOTMEM && MEMORY_HOTPLUG 65195794924SYang Shi depends on !FLATMEM 6523a80a7faSMel Gorman help 6533a80a7faSMel Gorman Ordinarily all struct pages are initialised during early boot in a 6543a80a7faSMel Gorman single thread. On very large machines this can take a considerable 6553a80a7faSMel Gorman amount of time. If this option is set, large machines will bring up 6563a80a7faSMel Gorman a subset of memmap at boot and then initialise the rest in parallel 6571ce22103SVlastimil Babka by starting one-off "pgdatinitX" kernel thread for each node X. This 6581ce22103SVlastimil Babka has a potential performance impact on processes running early in the 6591ce22103SVlastimil Babka lifetime of the system until these kthreads finish the 6601ce22103SVlastimil Babka initialisation. 661033fbae9SDan Williams 66233c3fc71SVladimir Davydovconfig IDLE_PAGE_TRACKING 66333c3fc71SVladimir Davydov bool "Enable idle page tracking" 66433c3fc71SVladimir Davydov depends on SYSFS && MMU 66533c3fc71SVladimir Davydov select PAGE_EXTENSION if !64BIT 66633c3fc71SVladimir Davydov help 66733c3fc71SVladimir Davydov This feature allows to estimate the amount of user pages that have 66833c3fc71SVladimir Davydov not been touched during a given period of time. This information can 66933c3fc71SVladimir Davydov be useful to tune memory cgroup limits and/or for job placement 67033c3fc71SVladimir Davydov within a compute cluster. 67133c3fc71SVladimir Davydov 67233c3fc71SVladimir Davydov See Documentation/vm/idle_page_tracking.txt for more details. 67333c3fc71SVladimir Davydov 67465f7d049SOliver O'Halloran# arch_add_memory() comprehends device memory 67565f7d049SOliver O'Halloranconfig ARCH_HAS_ZONE_DEVICE 67665f7d049SOliver O'Halloran bool 67765f7d049SOliver O'Halloran 678033fbae9SDan Williamsconfig ZONE_DEVICE 6795042db43SJérôme Glisse bool "Device memory (pmem, HMM, etc...) hotplug support" 680033fbae9SDan Williams depends on MEMORY_HOTPLUG 681033fbae9SDan Williams depends on MEMORY_HOTREMOVE 68299490f16SDan Williams depends on SPARSEMEM_VMEMMAP 68365f7d049SOliver O'Halloran depends on ARCH_HAS_ZONE_DEVICE 684ab1b597eSDan Williams select RADIX_TREE_MULTIORDER 685033fbae9SDan Williams 686033fbae9SDan Williams help 687033fbae9SDan Williams Device memory hotplug support allows for establishing pmem, 688033fbae9SDan Williams or other device driver discovered memory regions, in the 689033fbae9SDan Williams memmap. This allows pfn_to_page() lookups of otherwise 690033fbae9SDan Williams "device-physical" addresses which is needed for using a DAX 691033fbae9SDan Williams mapping in an O_DIRECT operation, among other things. 692033fbae9SDan Williams 693033fbae9SDan Williams If FS_DAX is enabled, then say Y. 69406a660adSLinus Torvalds 695133ff0eaSJérôme Glisseconfig ARCH_HAS_HMM 696133ff0eaSJérôme Glisse bool 697133ff0eaSJérôme Glisse default y 698133ff0eaSJérôme Glisse depends on (X86_64 || PPC64) 699133ff0eaSJérôme Glisse depends on ZONE_DEVICE 700133ff0eaSJérôme Glisse depends on MMU && 64BIT 701133ff0eaSJérôme Glisse depends on MEMORY_HOTPLUG 702133ff0eaSJérôme Glisse depends on MEMORY_HOTREMOVE 703133ff0eaSJérôme Glisse depends on SPARSEMEM_VMEMMAP 704133ff0eaSJérôme Glisse 7056b368cd4SJérôme Glisseconfig MIGRATE_VMA_HELPER 7066b368cd4SJérôme Glisse bool 7076b368cd4SJérôme Glisse 708133ff0eaSJérôme Glisseconfig HMM 709133ff0eaSJérôme Glisse bool 7106b368cd4SJérôme Glisse select MIGRATE_VMA_HELPER 711133ff0eaSJérôme Glisse 712c0b12405SJérôme Glisseconfig HMM_MIRROR 713c0b12405SJérôme Glisse bool "HMM mirror CPU page table into a device page table" 714c0b12405SJérôme Glisse depends on ARCH_HAS_HMM 715c0b12405SJérôme Glisse select MMU_NOTIFIER 716c0b12405SJérôme Glisse select HMM 717c0b12405SJérôme Glisse help 718c0b12405SJérôme Glisse Select HMM_MIRROR if you want to mirror range of the CPU page table of a 719c0b12405SJérôme Glisse process into a device page table. Here, mirror means "keep synchronized". 720c0b12405SJérôme Glisse Prerequisites: the device must provide the ability to write-protect its 721c0b12405SJérôme Glisse page tables (at PAGE_SIZE granularity), and must be able to recover from 722c0b12405SJérôme Glisse the resulting potential page faults. 723c0b12405SJérôme Glisse 7245042db43SJérôme Glisseconfig DEVICE_PRIVATE 7255042db43SJérôme Glisse bool "Unaddressable device memory (GPU memory, ...)" 7265042db43SJérôme Glisse depends on ARCH_HAS_HMM 727df6ad698SJérôme Glisse select HMM 7285042db43SJérôme Glisse 7295042db43SJérôme Glisse help 7305042db43SJérôme Glisse Allows creation of struct pages to represent unaddressable device 7315042db43SJérôme Glisse memory; i.e., memory that is only accessible from the device (or 7325042db43SJérôme Glisse group of devices). You likely also want to select HMM_MIRROR. 7335042db43SJérôme Glisse 734df6ad698SJérôme Glisseconfig DEVICE_PUBLIC 735df6ad698SJérôme Glisse bool "Addressable device memory (like GPU memory)" 736df6ad698SJérôme Glisse depends on ARCH_HAS_HMM 737df6ad698SJérôme Glisse select HMM 738df6ad698SJérôme Glisse 739df6ad698SJérôme Glisse help 740df6ad698SJérôme Glisse Allows creation of struct pages to represent addressable device 741df6ad698SJérôme Glisse memory; i.e., memory that is accessible from both the device and 742df6ad698SJérôme Glisse the CPU 743df6ad698SJérôme Glisse 7448025e5ddSJan Karaconfig FRAME_VECTOR 7458025e5ddSJan Kara bool 74663c17fb8SDave Hansen 74763c17fb8SDave Hansenconfig ARCH_USES_HIGH_VMA_FLAGS 74863c17fb8SDave Hansen bool 74966d37570SDave Hansenconfig ARCH_HAS_PKEYS 75066d37570SDave Hansen bool 75130a5b536SDennis Zhou 75230a5b536SDennis Zhouconfig PERCPU_STATS 75330a5b536SDennis Zhou bool "Collect percpu memory statistics" 75430a5b536SDennis Zhou default n 75530a5b536SDennis Zhou help 75630a5b536SDennis Zhou This feature collects and exposes statistics via debugfs. The 75730a5b536SDennis Zhou information includes global and per chunk statistics, which can 75830a5b536SDennis Zhou be used to help understand percpu memory usage. 759*64c349f4SKirill A. Shutemov 760*64c349f4SKirill A. Shutemovconfig GUP_BENCHMARK 761*64c349f4SKirill A. Shutemov bool "Enable infrastructure for get_user_pages_fast() benchmarking" 762*64c349f4SKirill A. Shutemov default n 763*64c349f4SKirill A. Shutemov help 764*64c349f4SKirill A. Shutemov Provides /sys/kernel/debug/gup_benchmark that helps with testing 765*64c349f4SKirill A. Shutemov performance of get_user_pages_fast(). 766*64c349f4SKirill A. Shutemov 767*64c349f4SKirill A. Shutemov See tools/testing/selftests/vm/gup_benchmark.c 768