page_alloc.c 119 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458245924602461246224632464246524662467246824692470247124722473247424752476247724782479248024812482248324842485248624872488248924902491249224932494249524962497249824992500250125022503250425052506250725082509251025112512251325142515251625172518251925202521252225232524252525262527252825292530253125322533253425352536253725382539254025412542254325442545254625472548254925502551255225532554255525562557255825592560256125622563256425652566256725682569257025712572257325742575257625772578257925802581258225832584258525862587258825892590259125922593259425952596259725982599260026012602260326042605260626072608260926102611261226132614261526162617261826192620262126222623262426252626262726282629263026312632263326342635263626372638263926402641264226432644264526462647264826492650265126522653265426552656265726582659266026612662266326642665266626672668266926702671267226732674267526762677267826792680268126822683268426852686268726882689269026912692269326942695269626972698269927002701270227032704270527062707270827092710271127122713271427152716271727182719272027212722272327242725272627272728272927302731273227332734273527362737273827392740274127422743274427452746274727482749275027512752275327542755275627572758275927602761276227632764276527662767276827692770277127722773277427752776277727782779278027812782278327842785278627872788278927902791279227932794279527962797279827992800280128022803280428052806280728082809281028112812281328142815281628172818281928202821282228232824282528262827282828292830283128322833283428352836283728382839284028412842284328442845284628472848284928502851285228532854285528562857285828592860286128622863286428652866286728682869287028712872287328742875287628772878287928802881288228832884288528862887288828892890289128922893289428952896289728982899290029012902290329042905290629072908290929102911291229132914291529162917291829192920292129222923292429252926292729282929293029312932293329342935293629372938293929402941294229432944294529462947294829492950295129522953295429552956295729582959296029612962296329642965296629672968296929702971297229732974297529762977297829792980298129822983298429852986298729882989299029912992299329942995299629972998299930003001300230033004300530063007300830093010301130123013301430153016301730183019302030213022302330243025302630273028302930303031303230333034303530363037303830393040304130423043304430453046304730483049305030513052305330543055305630573058305930603061306230633064306530663067306830693070307130723073307430753076307730783079308030813082308330843085308630873088308930903091309230933094309530963097309830993100310131023103310431053106310731083109311031113112311331143115311631173118311931203121312231233124312531263127312831293130313131323133313431353136313731383139314031413142314331443145314631473148314931503151315231533154315531563157315831593160316131623163316431653166316731683169317031713172317331743175317631773178317931803181318231833184318531863187318831893190319131923193319431953196319731983199320032013202320332043205320632073208320932103211321232133214321532163217321832193220322132223223322432253226322732283229323032313232323332343235323632373238323932403241324232433244324532463247324832493250325132523253325432553256325732583259326032613262326332643265326632673268326932703271327232733274327532763277327832793280328132823283328432853286328732883289329032913292329332943295329632973298329933003301330233033304330533063307330833093310331133123313331433153316331733183319332033213322332333243325332633273328332933303331333233333334333533363337333833393340334133423343334433453346334733483349335033513352335333543355335633573358335933603361336233633364336533663367336833693370337133723373337433753376337733783379338033813382338333843385338633873388338933903391339233933394339533963397339833993400340134023403340434053406340734083409341034113412341334143415341634173418341934203421342234233424342534263427342834293430343134323433343434353436343734383439344034413442344334443445344634473448344934503451345234533454345534563457345834593460346134623463346434653466346734683469347034713472347334743475347634773478347934803481348234833484348534863487348834893490349134923493349434953496349734983499350035013502350335043505350635073508350935103511351235133514351535163517351835193520352135223523352435253526352735283529353035313532353335343535353635373538353935403541354235433544354535463547354835493550355135523553355435553556355735583559356035613562356335643565356635673568356935703571357235733574357535763577357835793580358135823583358435853586358735883589359035913592359335943595359635973598359936003601360236033604360536063607360836093610361136123613361436153616361736183619362036213622362336243625362636273628362936303631363236333634363536363637363836393640364136423643364436453646364736483649365036513652365336543655365636573658365936603661366236633664366536663667366836693670367136723673367436753676367736783679368036813682368336843685368636873688368936903691369236933694369536963697369836993700370137023703370437053706370737083709371037113712371337143715371637173718371937203721372237233724372537263727372837293730373137323733373437353736373737383739374037413742374337443745374637473748374937503751375237533754375537563757375837593760376137623763376437653766376737683769377037713772377337743775377637773778377937803781378237833784378537863787378837893790379137923793379437953796379737983799380038013802380338043805380638073808380938103811381238133814381538163817381838193820382138223823382438253826382738283829383038313832383338343835383638373838383938403841384238433844384538463847384838493850385138523853385438553856385738583859386038613862386338643865386638673868386938703871387238733874387538763877387838793880388138823883388438853886388738883889389038913892389338943895389638973898389939003901390239033904390539063907390839093910391139123913391439153916391739183919392039213922392339243925392639273928392939303931393239333934393539363937393839393940394139423943394439453946394739483949395039513952395339543955395639573958395939603961396239633964396539663967396839693970397139723973397439753976397739783979398039813982398339843985398639873988398939903991399239933994399539963997399839994000400140024003400440054006400740084009401040114012401340144015401640174018401940204021402240234024402540264027402840294030403140324033403440354036403740384039404040414042404340444045404640474048404940504051405240534054405540564057405840594060406140624063406440654066406740684069407040714072407340744075407640774078407940804081408240834084408540864087408840894090409140924093409440954096409740984099410041014102410341044105410641074108410941104111411241134114411541164117411841194120412141224123412441254126412741284129413041314132413341344135413641374138413941404141414241434144414541464147414841494150415141524153415441554156415741584159416041614162416341644165416641674168416941704171417241734174417541764177417841794180418141824183418441854186418741884189419041914192419341944195419641974198419942004201420242034204420542064207420842094210421142124213421442154216421742184219422042214222422342244225422642274228422942304231423242334234423542364237423842394240424142424243424442454246424742484249425042514252425342544255425642574258425942604261426242634264426542664267426842694270427142724273427442754276427742784279428042814282428342844285428642874288428942904291429242934294429542964297429842994300430143024303430443054306430743084309431043114312431343144315431643174318431943204321432243234324432543264327432843294330433143324333433443354336433743384339434043414342434343444345434643474348434943504351435243534354435543564357435843594360436143624363
  1. /*
  2. * linux/mm/page_alloc.c
  3. *
  4. * Manages the free list, the system allocates free pages here.
  5. * Note that kmalloc() lives in slab.c
  6. *
  7. * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
  8. * Swap reorganised 29.12.95, Stephen Tweedie
  9. * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
  10. * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
  11. * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
  12. * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
  13. * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
  14. * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
  15. */
  16. #include <linux/stddef.h>
  17. #include <linux/mm.h>
  18. #include <linux/swap.h>
  19. #include <linux/interrupt.h>
  20. #include <linux/pagemap.h>
  21. #include <linux/bootmem.h>
  22. #include <linux/compiler.h>
  23. #include <linux/kernel.h>
  24. #include <linux/module.h>
  25. #include <linux/suspend.h>
  26. #include <linux/pagevec.h>
  27. #include <linux/blkdev.h>
  28. #include <linux/slab.h>
  29. #include <linux/notifier.h>
  30. #include <linux/topology.h>
  31. #include <linux/sysctl.h>
  32. #include <linux/cpu.h>
  33. #include <linux/cpuset.h>
  34. #include <linux/memory_hotplug.h>
  35. #include <linux/nodemask.h>
  36. #include <linux/vmalloc.h>
  37. #include <linux/mempolicy.h>
  38. #include <linux/stop_machine.h>
  39. #include <linux/sort.h>
  40. #include <linux/pfn.h>
  41. #include <linux/backing-dev.h>
  42. #include <linux/fault-inject.h>
  43. #include <asm/tlbflush.h>
  44. #include <asm/div64.h>
  45. #include "internal.h"
  46. /*
  47. * Array of node states.
  48. */
  49. nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
  50. [N_POSSIBLE] = NODE_MASK_ALL,
  51. [N_ONLINE] = { { [0] = 1UL } },
  52. #ifndef CONFIG_NUMA
  53. [N_NORMAL_MEMORY] = { { [0] = 1UL } },
  54. #ifdef CONFIG_HIGHMEM
  55. [N_HIGH_MEMORY] = { { [0] = 1UL } },
  56. #endif
  57. [N_CPU] = { { [0] = 1UL } },
  58. #endif /* NUMA */
  59. };
  60. EXPORT_SYMBOL(node_states);
  61. unsigned long totalram_pages __read_mostly;
  62. unsigned long totalreserve_pages __read_mostly;
  63. long nr_swap_pages;
  64. int percpu_pagelist_fraction;
  65. static void __free_pages_ok(struct page *page, unsigned int order);
  66. /*
  67. * results with 256, 32 in the lowmem_reserve sysctl:
  68. * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
  69. * 1G machine -> (16M dma, 784M normal, 224M high)
  70. * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
  71. * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
  72. * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
  73. *
  74. * TBD: should special case ZONE_DMA32 machines here - in those we normally
  75. * don't need any ZONE_NORMAL reservation
  76. */
  77. int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
  78. #ifdef CONFIG_ZONE_DMA
  79. 256,
  80. #endif
  81. #ifdef CONFIG_ZONE_DMA32
  82. 256,
  83. #endif
  84. #ifdef CONFIG_HIGHMEM
  85. 32,
  86. #endif
  87. 32,
  88. };
  89. EXPORT_SYMBOL(totalram_pages);
  90. static char * const zone_names[MAX_NR_ZONES] = {
  91. #ifdef CONFIG_ZONE_DMA
  92. "DMA",
  93. #endif
  94. #ifdef CONFIG_ZONE_DMA32
  95. "DMA32",
  96. #endif
  97. "Normal",
  98. #ifdef CONFIG_HIGHMEM
  99. "HighMem",
  100. #endif
  101. "Movable",
  102. };
  103. int min_free_kbytes = 1024;
  104. unsigned long __meminitdata nr_kernel_pages;
  105. unsigned long __meminitdata nr_all_pages;
  106. static unsigned long __meminitdata dma_reserve;
  107. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  108. /*
  109. * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
  110. * ranges of memory (RAM) that may be registered with add_active_range().
  111. * Ranges passed to add_active_range() will be merged if possible
  112. * so the number of times add_active_range() can be called is
  113. * related to the number of nodes and the number of holes
  114. */
  115. #ifdef CONFIG_MAX_ACTIVE_REGIONS
  116. /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
  117. #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
  118. #else
  119. #if MAX_NUMNODES >= 32
  120. /* If there can be many nodes, allow up to 50 holes per node */
  121. #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
  122. #else
  123. /* By default, allow up to 256 distinct regions */
  124. #define MAX_ACTIVE_REGIONS 256
  125. #endif
  126. #endif
  127. static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
  128. static int __meminitdata nr_nodemap_entries;
  129. static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
  130. static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
  131. #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
  132. static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES];
  133. static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES];
  134. #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
  135. unsigned long __initdata required_kernelcore;
  136. unsigned long __initdata required_movablecore;
  137. unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
  138. /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
  139. int movable_zone;
  140. EXPORT_SYMBOL(movable_zone);
  141. #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
  142. #if MAX_NUMNODES > 1
  143. int nr_node_ids __read_mostly = MAX_NUMNODES;
  144. EXPORT_SYMBOL(nr_node_ids);
  145. #endif
  146. #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
  147. static inline int get_pageblock_migratetype(struct page *page)
  148. {
  149. return get_pageblock_flags_group(page, PB_migrate, PB_migrate_end);
  150. }
  151. static void set_pageblock_migratetype(struct page *page, int migratetype)
  152. {
  153. set_pageblock_flags_group(page, (unsigned long)migratetype,
  154. PB_migrate, PB_migrate_end);
  155. }
  156. static inline int allocflags_to_migratetype(gfp_t gfp_flags, int order)
  157. {
  158. WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
  159. /* Cluster high-order atomic allocations together */
  160. if (unlikely(order > 0) &&
  161. (!(gfp_flags & __GFP_WAIT) || in_interrupt()))
  162. return MIGRATE_HIGHATOMIC;
  163. /* Cluster based on mobility */
  164. return (((gfp_flags & __GFP_MOVABLE) != 0) << 1) |
  165. ((gfp_flags & __GFP_RECLAIMABLE) != 0);
  166. }
  167. #else
  168. static inline int get_pageblock_migratetype(struct page *page)
  169. {
  170. return MIGRATE_UNMOVABLE;
  171. }
  172. static void set_pageblock_migratetype(struct page *page, int migratetype)
  173. {
  174. }
  175. static inline int allocflags_to_migratetype(gfp_t gfp_flags, int order)
  176. {
  177. return MIGRATE_UNMOVABLE;
  178. }
  179. #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
  180. #ifdef CONFIG_DEBUG_VM
  181. static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
  182. {
  183. int ret = 0;
  184. unsigned seq;
  185. unsigned long pfn = page_to_pfn(page);
  186. do {
  187. seq = zone_span_seqbegin(zone);
  188. if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
  189. ret = 1;
  190. else if (pfn < zone->zone_start_pfn)
  191. ret = 1;
  192. } while (zone_span_seqretry(zone, seq));
  193. return ret;
  194. }
  195. static int page_is_consistent(struct zone *zone, struct page *page)
  196. {
  197. if (!pfn_valid_within(page_to_pfn(page)))
  198. return 0;
  199. if (zone != page_zone(page))
  200. return 0;
  201. return 1;
  202. }
  203. /*
  204. * Temporary debugging check for pages not lying within a given zone.
  205. */
  206. static int bad_range(struct zone *zone, struct page *page)
  207. {
  208. if (page_outside_zone_boundaries(zone, page))
  209. return 1;
  210. if (!page_is_consistent(zone, page))
  211. return 1;
  212. return 0;
  213. }
  214. #else
  215. static inline int bad_range(struct zone *zone, struct page *page)
  216. {
  217. return 0;
  218. }
  219. #endif
  220. static void bad_page(struct page *page)
  221. {
  222. printk(KERN_EMERG "Bad page state in process '%s'\n"
  223. KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
  224. KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
  225. KERN_EMERG "Backtrace:\n",
  226. current->comm, page, (int)(2*sizeof(unsigned long)),
  227. (unsigned long)page->flags, page->mapping,
  228. page_mapcount(page), page_count(page));
  229. dump_stack();
  230. page->flags &= ~(1 << PG_lru |
  231. 1 << PG_private |
  232. 1 << PG_locked |
  233. 1 << PG_active |
  234. 1 << PG_dirty |
  235. 1 << PG_reclaim |
  236. 1 << PG_slab |
  237. 1 << PG_swapcache |
  238. 1 << PG_writeback |
  239. 1 << PG_buddy );
  240. set_page_count(page, 0);
  241. reset_page_mapcount(page);
  242. page->mapping = NULL;
  243. add_taint(TAINT_BAD_PAGE);
  244. }
  245. /*
  246. * Higher-order pages are called "compound pages". They are structured thusly:
  247. *
  248. * The first PAGE_SIZE page is called the "head page".
  249. *
  250. * The remaining PAGE_SIZE pages are called "tail pages".
  251. *
  252. * All pages have PG_compound set. All pages have their ->private pointing at
  253. * the head page (even the head page has this).
  254. *
  255. * The first tail page's ->lru.next holds the address of the compound page's
  256. * put_page() function. Its ->lru.prev holds the order of allocation.
  257. * This usage means that zero-order pages may not be compound.
  258. */
  259. static void free_compound_page(struct page *page)
  260. {
  261. __free_pages_ok(page, compound_order(page));
  262. }
  263. static void prep_compound_page(struct page *page, unsigned long order)
  264. {
  265. int i;
  266. int nr_pages = 1 << order;
  267. set_compound_page_dtor(page, free_compound_page);
  268. set_compound_order(page, order);
  269. __SetPageHead(page);
  270. for (i = 1; i < nr_pages; i++) {
  271. struct page *p = page + i;
  272. __SetPageTail(p);
  273. p->first_page = page;
  274. }
  275. }
  276. static void destroy_compound_page(struct page *page, unsigned long order)
  277. {
  278. int i;
  279. int nr_pages = 1 << order;
  280. if (unlikely(compound_order(page) != order))
  281. bad_page(page);
  282. if (unlikely(!PageHead(page)))
  283. bad_page(page);
  284. __ClearPageHead(page);
  285. for (i = 1; i < nr_pages; i++) {
  286. struct page *p = page + i;
  287. if (unlikely(!PageTail(p) |
  288. (p->first_page != page)))
  289. bad_page(page);
  290. __ClearPageTail(p);
  291. }
  292. }
  293. static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
  294. {
  295. int i;
  296. VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
  297. /*
  298. * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
  299. * and __GFP_HIGHMEM from hard or soft interrupt context.
  300. */
  301. VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
  302. for (i = 0; i < (1 << order); i++)
  303. clear_highpage(page + i);
  304. }
  305. /*
  306. * function for dealing with page's order in buddy system.
  307. * zone->lock is already acquired when we use these.
  308. * So, we don't need atomic page->flags operations here.
  309. */
  310. static inline unsigned long page_order(struct page *page)
  311. {
  312. return page_private(page);
  313. }
  314. static inline void set_page_order(struct page *page, int order)
  315. {
  316. set_page_private(page, order);
  317. __SetPageBuddy(page);
  318. }
  319. static inline void rmv_page_order(struct page *page)
  320. {
  321. __ClearPageBuddy(page);
  322. set_page_private(page, 0);
  323. }
  324. /*
  325. * Locate the struct page for both the matching buddy in our
  326. * pair (buddy1) and the combined O(n+1) page they form (page).
  327. *
  328. * 1) Any buddy B1 will have an order O twin B2 which satisfies
  329. * the following equation:
  330. * B2 = B1 ^ (1 << O)
  331. * For example, if the starting buddy (buddy2) is #8 its order
  332. * 1 buddy is #10:
  333. * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
  334. *
  335. * 2) Any buddy B will have an order O+1 parent P which
  336. * satisfies the following equation:
  337. * P = B & ~(1 << O)
  338. *
  339. * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
  340. */
  341. static inline struct page *
  342. __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
  343. {
  344. unsigned long buddy_idx = page_idx ^ (1 << order);
  345. return page + (buddy_idx - page_idx);
  346. }
  347. static inline unsigned long
  348. __find_combined_index(unsigned long page_idx, unsigned int order)
  349. {
  350. return (page_idx & ~(1 << order));
  351. }
  352. /*
  353. * This function checks whether a page is free && is the buddy
  354. * we can do coalesce a page and its buddy if
  355. * (a) the buddy is not in a hole &&
  356. * (b) the buddy is in the buddy system &&
  357. * (c) a page and its buddy have the same order &&
  358. * (d) a page and its buddy are in the same zone.
  359. *
  360. * For recording whether a page is in the buddy system, we use PG_buddy.
  361. * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
  362. *
  363. * For recording page's order, we use page_private(page).
  364. */
  365. static inline int page_is_buddy(struct page *page, struct page *buddy,
  366. int order)
  367. {
  368. if (!pfn_valid_within(page_to_pfn(buddy)))
  369. return 0;
  370. if (page_zone_id(page) != page_zone_id(buddy))
  371. return 0;
  372. if (PageBuddy(buddy) && page_order(buddy) == order) {
  373. BUG_ON(page_count(buddy) != 0);
  374. return 1;
  375. }
  376. return 0;
  377. }
  378. /*
  379. * Freeing function for a buddy system allocator.
  380. *
  381. * The concept of a buddy system is to maintain direct-mapped table
  382. * (containing bit values) for memory blocks of various "orders".
  383. * The bottom level table contains the map for the smallest allocatable
  384. * units of memory (here, pages), and each level above it describes
  385. * pairs of units from the levels below, hence, "buddies".
  386. * At a high level, all that happens here is marking the table entry
  387. * at the bottom level available, and propagating the changes upward
  388. * as necessary, plus some accounting needed to play nicely with other
  389. * parts of the VM system.
  390. * At each level, we keep a list of pages, which are heads of continuous
  391. * free pages of length of (1 << order) and marked with PG_buddy. Page's
  392. * order is recorded in page_private(page) field.
  393. * So when we are allocating or freeing one, we can derive the state of the
  394. * other. That is, if we allocate a small block, and both were
  395. * free, the remainder of the region must be split into blocks.
  396. * If a block is freed, and its buddy is also free, then this
  397. * triggers coalescing into a block of larger size.
  398. *
  399. * -- wli
  400. */
  401. static inline void __free_one_page(struct page *page,
  402. struct zone *zone, unsigned int order)
  403. {
  404. unsigned long page_idx;
  405. int order_size = 1 << order;
  406. int migratetype = get_pageblock_migratetype(page);
  407. if (unlikely(PageCompound(page)))
  408. destroy_compound_page(page, order);
  409. page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
  410. VM_BUG_ON(page_idx & (order_size - 1));
  411. VM_BUG_ON(bad_range(zone, page));
  412. __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
  413. while (order < MAX_ORDER-1) {
  414. unsigned long combined_idx;
  415. struct page *buddy;
  416. buddy = __page_find_buddy(page, page_idx, order);
  417. if (!page_is_buddy(page, buddy, order))
  418. break; /* Move the buddy up one level. */
  419. list_del(&buddy->lru);
  420. zone->free_area[order].nr_free--;
  421. rmv_page_order(buddy);
  422. combined_idx = __find_combined_index(page_idx, order);
  423. page = page + (combined_idx - page_idx);
  424. page_idx = combined_idx;
  425. order++;
  426. }
  427. set_page_order(page, order);
  428. list_add(&page->lru,
  429. &zone->free_area[order].free_list[migratetype]);
  430. zone->free_area[order].nr_free++;
  431. }
  432. static inline int free_pages_check(struct page *page)
  433. {
  434. if (unlikely(page_mapcount(page) |
  435. (page->mapping != NULL) |
  436. (page_count(page) != 0) |
  437. (page->flags & (
  438. 1 << PG_lru |
  439. 1 << PG_private |
  440. 1 << PG_locked |
  441. 1 << PG_active |
  442. 1 << PG_slab |
  443. 1 << PG_swapcache |
  444. 1 << PG_writeback |
  445. 1 << PG_reserved |
  446. 1 << PG_buddy ))))
  447. bad_page(page);
  448. if (PageDirty(page))
  449. __ClearPageDirty(page);
  450. /*
  451. * For now, we report if PG_reserved was found set, but do not
  452. * clear it, and do not free the page. But we shall soon need
  453. * to do more, for when the ZERO_PAGE count wraps negative.
  454. */
  455. return PageReserved(page);
  456. }
  457. /*
  458. * Frees a list of pages.
  459. * Assumes all pages on list are in same zone, and of same order.
  460. * count is the number of pages to free.
  461. *
  462. * If the zone was previously in an "all pages pinned" state then look to
  463. * see if this freeing clears that state.
  464. *
  465. * And clear the zone's pages_scanned counter, to hold off the "all pages are
  466. * pinned" detection logic.
  467. */
  468. static void free_pages_bulk(struct zone *zone, int count,
  469. struct list_head *list, int order)
  470. {
  471. spin_lock(&zone->lock);
  472. zone->all_unreclaimable = 0;
  473. zone->pages_scanned = 0;
  474. while (count--) {
  475. struct page *page;
  476. VM_BUG_ON(list_empty(list));
  477. page = list_entry(list->prev, struct page, lru);
  478. /* have to delete it as __free_one_page list manipulates */
  479. list_del(&page->lru);
  480. __free_one_page(page, zone, order);
  481. }
  482. spin_unlock(&zone->lock);
  483. }
  484. static void free_one_page(struct zone *zone, struct page *page, int order)
  485. {
  486. spin_lock(&zone->lock);
  487. zone->all_unreclaimable = 0;
  488. zone->pages_scanned = 0;
  489. __free_one_page(page, zone, order);
  490. spin_unlock(&zone->lock);
  491. }
  492. static void __free_pages_ok(struct page *page, unsigned int order)
  493. {
  494. unsigned long flags;
  495. int i;
  496. int reserved = 0;
  497. for (i = 0 ; i < (1 << order) ; ++i)
  498. reserved += free_pages_check(page + i);
  499. if (reserved)
  500. return;
  501. if (!PageHighMem(page))
  502. debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
  503. arch_free_page(page, order);
  504. kernel_map_pages(page, 1 << order, 0);
  505. local_irq_save(flags);
  506. __count_vm_events(PGFREE, 1 << order);
  507. free_one_page(page_zone(page), page, order);
  508. local_irq_restore(flags);
  509. }
  510. /*
  511. * permit the bootmem allocator to evade page validation on high-order frees
  512. */
  513. void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
  514. {
  515. if (order == 0) {
  516. __ClearPageReserved(page);
  517. set_page_count(page, 0);
  518. set_page_refcounted(page);
  519. __free_page(page);
  520. } else {
  521. int loop;
  522. prefetchw(page);
  523. for (loop = 0; loop < BITS_PER_LONG; loop++) {
  524. struct page *p = &page[loop];
  525. if (loop + 1 < BITS_PER_LONG)
  526. prefetchw(p + 1);
  527. __ClearPageReserved(p);
  528. set_page_count(p, 0);
  529. }
  530. set_page_refcounted(page);
  531. __free_pages(page, order);
  532. }
  533. }
  534. /*
  535. * The order of subdivision here is critical for the IO subsystem.
  536. * Please do not alter this order without good reasons and regression
  537. * testing. Specifically, as large blocks of memory are subdivided,
  538. * the order in which smaller blocks are delivered depends on the order
  539. * they're subdivided in this function. This is the primary factor
  540. * influencing the order in which pages are delivered to the IO
  541. * subsystem according to empirical testing, and this is also justified
  542. * by considering the behavior of a buddy system containing a single
  543. * large block of memory acted on by a series of small allocations.
  544. * This behavior is a critical factor in sglist merging's success.
  545. *
  546. * -- wli
  547. */
  548. static inline void expand(struct zone *zone, struct page *page,
  549. int low, int high, struct free_area *area,
  550. int migratetype)
  551. {
  552. unsigned long size = 1 << high;
  553. while (high > low) {
  554. area--;
  555. high--;
  556. size >>= 1;
  557. VM_BUG_ON(bad_range(zone, &page[size]));
  558. list_add(&page[size].lru, &area->free_list[migratetype]);
  559. area->nr_free++;
  560. set_page_order(&page[size], high);
  561. }
  562. }
  563. /*
  564. * This page is about to be returned from the page allocator
  565. */
  566. static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
  567. {
  568. if (unlikely(page_mapcount(page) |
  569. (page->mapping != NULL) |
  570. (page_count(page) != 0) |
  571. (page->flags & (
  572. 1 << PG_lru |
  573. 1 << PG_private |
  574. 1 << PG_locked |
  575. 1 << PG_active |
  576. 1 << PG_dirty |
  577. 1 << PG_slab |
  578. 1 << PG_swapcache |
  579. 1 << PG_writeback |
  580. 1 << PG_reserved |
  581. 1 << PG_buddy ))))
  582. bad_page(page);
  583. /*
  584. * For now, we report if PG_reserved was found set, but do not
  585. * clear it, and do not allocate the page: as a safety net.
  586. */
  587. if (PageReserved(page))
  588. return 1;
  589. page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_readahead |
  590. 1 << PG_referenced | 1 << PG_arch_1 |
  591. 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk);
  592. set_page_private(page, 0);
  593. set_page_refcounted(page);
  594. arch_alloc_page(page, order);
  595. kernel_map_pages(page, 1 << order, 1);
  596. if (gfp_flags & __GFP_ZERO)
  597. prep_zero_page(page, order, gfp_flags);
  598. if (order && (gfp_flags & __GFP_COMP))
  599. prep_compound_page(page, order);
  600. return 0;
  601. }
  602. #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
  603. /*
  604. * This array describes the order lists are fallen back to when
  605. * the free lists for the desirable migrate type are depleted
  606. */
  607. static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
  608. [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_HIGHATOMIC },
  609. [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_HIGHATOMIC },
  610. [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE,MIGRATE_HIGHATOMIC },
  611. [MIGRATE_HIGHATOMIC] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE,MIGRATE_MOVABLE},
  612. };
  613. /*
  614. * Move the free pages in a range to the free lists of the requested type.
  615. * Note that start_page and end_pages are not aligned in a MAX_ORDER_NR_PAGES
  616. * boundary. If alignment is required, use move_freepages_block()
  617. */
  618. int move_freepages(struct zone *zone,
  619. struct page *start_page, struct page *end_page,
  620. int migratetype)
  621. {
  622. struct page *page;
  623. unsigned long order;
  624. int blocks_moved = 0;
  625. #ifndef CONFIG_HOLES_IN_ZONE
  626. /*
  627. * page_zone is not safe to call in this context when
  628. * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
  629. * anyway as we check zone boundaries in move_freepages_block().
  630. * Remove at a later date when no bug reports exist related to
  631. * CONFIG_PAGE_GROUP_BY_MOBILITY
  632. */
  633. BUG_ON(page_zone(start_page) != page_zone(end_page));
  634. #endif
  635. for (page = start_page; page <= end_page;) {
  636. if (!pfn_valid_within(page_to_pfn(page))) {
  637. page++;
  638. continue;
  639. }
  640. if (!PageBuddy(page)) {
  641. page++;
  642. continue;
  643. }
  644. order = page_order(page);
  645. list_del(&page->lru);
  646. list_add(&page->lru,
  647. &zone->free_area[order].free_list[migratetype]);
  648. page += 1 << order;
  649. blocks_moved++;
  650. }
  651. return blocks_moved;
  652. }
  653. int move_freepages_block(struct zone *zone, struct page *page, int migratetype)
  654. {
  655. unsigned long start_pfn, end_pfn;
  656. struct page *start_page, *end_page;
  657. start_pfn = page_to_pfn(page);
  658. start_pfn = start_pfn & ~(MAX_ORDER_NR_PAGES-1);
  659. start_page = pfn_to_page(start_pfn);
  660. end_page = start_page + MAX_ORDER_NR_PAGES - 1;
  661. end_pfn = start_pfn + MAX_ORDER_NR_PAGES - 1;
  662. /* Do not cross zone boundaries */
  663. if (start_pfn < zone->zone_start_pfn)
  664. start_page = page;
  665. if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
  666. return 0;
  667. return move_freepages(zone, start_page, end_page, migratetype);
  668. }
  669. /* Remove an element from the buddy allocator from the fallback list */
  670. static struct page *__rmqueue_fallback(struct zone *zone, int order,
  671. int start_migratetype)
  672. {
  673. struct free_area * area;
  674. int current_order;
  675. struct page *page;
  676. int migratetype, i;
  677. int nonatomic_fallback_atomic = 0;
  678. retry:
  679. /* Find the largest possible block of pages in the other list */
  680. for (current_order = MAX_ORDER-1; current_order >= order;
  681. --current_order) {
  682. for (i = 0; i < MIGRATE_TYPES - 1; i++) {
  683. migratetype = fallbacks[start_migratetype][i];
  684. /*
  685. * Make it hard to fallback to blocks used for
  686. * high-order atomic allocations
  687. */
  688. if (migratetype == MIGRATE_HIGHATOMIC &&
  689. start_migratetype != MIGRATE_UNMOVABLE &&
  690. !nonatomic_fallback_atomic)
  691. continue;
  692. area = &(zone->free_area[current_order]);
  693. if (list_empty(&area->free_list[migratetype]))
  694. continue;
  695. page = list_entry(area->free_list[migratetype].next,
  696. struct page, lru);
  697. area->nr_free--;
  698. /*
  699. * If breaking a large block of pages, move all free
  700. * pages to the preferred allocation list
  701. */
  702. if (unlikely(current_order >= MAX_ORDER / 2)) {
  703. migratetype = start_migratetype;
  704. move_freepages_block(zone, page, migratetype);
  705. }
  706. /* Remove the page from the freelists */
  707. list_del(&page->lru);
  708. rmv_page_order(page);
  709. __mod_zone_page_state(zone, NR_FREE_PAGES,
  710. -(1UL << order));
  711. if (current_order == MAX_ORDER - 1)
  712. set_pageblock_migratetype(page,
  713. start_migratetype);
  714. expand(zone, page, order, current_order, area, migratetype);
  715. return page;
  716. }
  717. }
  718. /* Allow fallback to high-order atomic blocks if memory is that low */
  719. if (!nonatomic_fallback_atomic) {
  720. nonatomic_fallback_atomic = 1;
  721. goto retry;
  722. }
  723. return NULL;
  724. }
  725. #else
  726. static struct page *__rmqueue_fallback(struct zone *zone, int order,
  727. int start_migratetype)
  728. {
  729. return NULL;
  730. }
  731. #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
  732. /*
  733. * Do the hard work of removing an element from the buddy allocator.
  734. * Call me with the zone->lock already held.
  735. */
  736. static struct page *__rmqueue(struct zone *zone, unsigned int order,
  737. int migratetype)
  738. {
  739. struct free_area * area;
  740. unsigned int current_order;
  741. struct page *page;
  742. /* Find a page of the appropriate size in the preferred list */
  743. for (current_order = order; current_order < MAX_ORDER; ++current_order) {
  744. area = &(zone->free_area[current_order]);
  745. if (list_empty(&area->free_list[migratetype]))
  746. continue;
  747. page = list_entry(area->free_list[migratetype].next,
  748. struct page, lru);
  749. list_del(&page->lru);
  750. rmv_page_order(page);
  751. area->nr_free--;
  752. __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
  753. expand(zone, page, order, current_order, area, migratetype);
  754. goto got_page;
  755. }
  756. page = __rmqueue_fallback(zone, order, migratetype);
  757. got_page:
  758. return page;
  759. }
  760. /*
  761. * Obtain a specified number of elements from the buddy allocator, all under
  762. * a single hold of the lock, for efficiency. Add them to the supplied list.
  763. * Returns the number of new pages which were placed at *list.
  764. */
  765. static int rmqueue_bulk(struct zone *zone, unsigned int order,
  766. unsigned long count, struct list_head *list,
  767. int migratetype)
  768. {
  769. int i;
  770. spin_lock(&zone->lock);
  771. for (i = 0; i < count; ++i) {
  772. struct page *page = __rmqueue(zone, order, migratetype);
  773. if (unlikely(page == NULL))
  774. break;
  775. list_add(&page->lru, list);
  776. set_page_private(page, migratetype);
  777. }
  778. spin_unlock(&zone->lock);
  779. return i;
  780. }
  781. #ifdef CONFIG_NUMA
  782. /*
  783. * Called from the vmstat counter updater to drain pagesets of this
  784. * currently executing processor on remote nodes after they have
  785. * expired.
  786. *
  787. * Note that this function must be called with the thread pinned to
  788. * a single processor.
  789. */
  790. void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
  791. {
  792. unsigned long flags;
  793. int to_drain;
  794. local_irq_save(flags);
  795. if (pcp->count >= pcp->batch)
  796. to_drain = pcp->batch;
  797. else
  798. to_drain = pcp->count;
  799. free_pages_bulk(zone, to_drain, &pcp->list, 0);
  800. pcp->count -= to_drain;
  801. local_irq_restore(flags);
  802. }
  803. #endif
  804. static void __drain_pages(unsigned int cpu)
  805. {
  806. unsigned long flags;
  807. struct zone *zone;
  808. int i;
  809. for_each_zone(zone) {
  810. struct per_cpu_pageset *pset;
  811. if (!populated_zone(zone))
  812. continue;
  813. pset = zone_pcp(zone, cpu);
  814. for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
  815. struct per_cpu_pages *pcp;
  816. pcp = &pset->pcp[i];
  817. local_irq_save(flags);
  818. free_pages_bulk(zone, pcp->count, &pcp->list, 0);
  819. pcp->count = 0;
  820. local_irq_restore(flags);
  821. }
  822. }
  823. }
  824. #ifdef CONFIG_HIBERNATION
  825. void mark_free_pages(struct zone *zone)
  826. {
  827. unsigned long pfn, max_zone_pfn;
  828. unsigned long flags;
  829. int order, t;
  830. struct list_head *curr;
  831. if (!zone->spanned_pages)
  832. return;
  833. spin_lock_irqsave(&zone->lock, flags);
  834. max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
  835. for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
  836. if (pfn_valid(pfn)) {
  837. struct page *page = pfn_to_page(pfn);
  838. if (!swsusp_page_is_forbidden(page))
  839. swsusp_unset_page_free(page);
  840. }
  841. for_each_migratetype_order(order, t) {
  842. list_for_each(curr, &zone->free_area[order].free_list[t]) {
  843. unsigned long i;
  844. pfn = page_to_pfn(list_entry(curr, struct page, lru));
  845. for (i = 0; i < (1UL << order); i++)
  846. swsusp_set_page_free(pfn_to_page(pfn + i));
  847. }
  848. }
  849. spin_unlock_irqrestore(&zone->lock, flags);
  850. }
  851. #endif /* CONFIG_PM */
  852. #if defined(CONFIG_HIBERNATION) || defined(CONFIG_PAGE_GROUP_BY_MOBILITY)
  853. /*
  854. * Spill all of this CPU's per-cpu pages back into the buddy allocator.
  855. */
  856. void drain_local_pages(void)
  857. {
  858. unsigned long flags;
  859. local_irq_save(flags);
  860. __drain_pages(smp_processor_id());
  861. local_irq_restore(flags);
  862. }
  863. void smp_drain_local_pages(void *arg)
  864. {
  865. drain_local_pages();
  866. }
  867. /*
  868. * Spill all the per-cpu pages from all CPUs back into the buddy allocator
  869. */
  870. void drain_all_local_pages(void)
  871. {
  872. unsigned long flags;
  873. local_irq_save(flags);
  874. __drain_pages(smp_processor_id());
  875. local_irq_restore(flags);
  876. smp_call_function(smp_drain_local_pages, NULL, 0, 1);
  877. }
  878. #else
  879. void drain_all_local_pages(void) {}
  880. #endif /* CONFIG_HIBERNATION || CONFIG_PAGE_GROUP_BY_MOBILITY */
  881. /*
  882. * Free a 0-order page
  883. */
  884. static void fastcall free_hot_cold_page(struct page *page, int cold)
  885. {
  886. struct zone *zone = page_zone(page);
  887. struct per_cpu_pages *pcp;
  888. unsigned long flags;
  889. if (PageAnon(page))
  890. page->mapping = NULL;
  891. if (free_pages_check(page))
  892. return;
  893. if (!PageHighMem(page))
  894. debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
  895. arch_free_page(page, 0);
  896. kernel_map_pages(page, 1, 0);
  897. pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
  898. local_irq_save(flags);
  899. __count_vm_event(PGFREE);
  900. list_add(&page->lru, &pcp->list);
  901. set_page_private(page, get_pageblock_migratetype(page));
  902. pcp->count++;
  903. if (pcp->count >= pcp->high) {
  904. free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
  905. pcp->count -= pcp->batch;
  906. }
  907. local_irq_restore(flags);
  908. put_cpu();
  909. }
  910. void fastcall free_hot_page(struct page *page)
  911. {
  912. free_hot_cold_page(page, 0);
  913. }
  914. void fastcall free_cold_page(struct page *page)
  915. {
  916. free_hot_cold_page(page, 1);
  917. }
  918. /*
  919. * split_page takes a non-compound higher-order page, and splits it into
  920. * n (1<<order) sub-pages: page[0..n]
  921. * Each sub-page must be freed individually.
  922. *
  923. * Note: this is probably too low level an operation for use in drivers.
  924. * Please consult with lkml before using this in your driver.
  925. */
  926. void split_page(struct page *page, unsigned int order)
  927. {
  928. int i;
  929. VM_BUG_ON(PageCompound(page));
  930. VM_BUG_ON(!page_count(page));
  931. for (i = 1; i < (1 << order); i++)
  932. set_page_refcounted(page + i);
  933. }
  934. /*
  935. * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
  936. * we cheat by calling it from here, in the order > 0 path. Saves a branch
  937. * or two.
  938. */
  939. static struct page *buffered_rmqueue(struct zonelist *zonelist,
  940. struct zone *zone, int order, gfp_t gfp_flags)
  941. {
  942. unsigned long flags;
  943. struct page *page;
  944. int cold = !!(gfp_flags & __GFP_COLD);
  945. int cpu;
  946. int migratetype = allocflags_to_migratetype(gfp_flags, order);
  947. again:
  948. cpu = get_cpu();
  949. if (likely(order == 0)) {
  950. struct per_cpu_pages *pcp;
  951. pcp = &zone_pcp(zone, cpu)->pcp[cold];
  952. local_irq_save(flags);
  953. if (!pcp->count) {
  954. pcp->count = rmqueue_bulk(zone, 0,
  955. pcp->batch, &pcp->list, migratetype);
  956. if (unlikely(!pcp->count))
  957. goto failed;
  958. }
  959. #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
  960. /* Find a page of the appropriate migrate type */
  961. list_for_each_entry(page, &pcp->list, lru)
  962. if (page_private(page) == migratetype)
  963. break;
  964. /* Allocate more to the pcp list if necessary */
  965. if (unlikely(&page->lru == &pcp->list)) {
  966. pcp->count += rmqueue_bulk(zone, 0,
  967. pcp->batch, &pcp->list, migratetype);
  968. page = list_entry(pcp->list.next, struct page, lru);
  969. }
  970. #else
  971. page = list_entry(pcp->list.next, struct page, lru);
  972. #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
  973. list_del(&page->lru);
  974. pcp->count--;
  975. } else {
  976. spin_lock_irqsave(&zone->lock, flags);
  977. page = __rmqueue(zone, order, migratetype);
  978. spin_unlock(&zone->lock);
  979. if (!page)
  980. goto failed;
  981. }
  982. __count_zone_vm_events(PGALLOC, zone, 1 << order);
  983. zone_statistics(zonelist, zone);
  984. local_irq_restore(flags);
  985. put_cpu();
  986. VM_BUG_ON(bad_range(zone, page));
  987. if (prep_new_page(page, order, gfp_flags))
  988. goto again;
  989. return page;
  990. failed:
  991. local_irq_restore(flags);
  992. put_cpu();
  993. return NULL;
  994. }
  995. #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
  996. #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
  997. #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
  998. #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
  999. #define ALLOC_HARDER 0x10 /* try to alloc harder */
  1000. #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
  1001. #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
  1002. #ifdef CONFIG_FAIL_PAGE_ALLOC
  1003. static struct fail_page_alloc_attr {
  1004. struct fault_attr attr;
  1005. u32 ignore_gfp_highmem;
  1006. u32 ignore_gfp_wait;
  1007. u32 min_order;
  1008. #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
  1009. struct dentry *ignore_gfp_highmem_file;
  1010. struct dentry *ignore_gfp_wait_file;
  1011. struct dentry *min_order_file;
  1012. #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
  1013. } fail_page_alloc = {
  1014. .attr = FAULT_ATTR_INITIALIZER,
  1015. .ignore_gfp_wait = 1,
  1016. .ignore_gfp_highmem = 1,
  1017. .min_order = 1,
  1018. };
  1019. static int __init setup_fail_page_alloc(char *str)
  1020. {
  1021. return setup_fault_attr(&fail_page_alloc.attr, str);
  1022. }
  1023. __setup("fail_page_alloc=", setup_fail_page_alloc);
  1024. static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
  1025. {
  1026. if (order < fail_page_alloc.min_order)
  1027. return 0;
  1028. if (gfp_mask & __GFP_NOFAIL)
  1029. return 0;
  1030. if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
  1031. return 0;
  1032. if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
  1033. return 0;
  1034. return should_fail(&fail_page_alloc.attr, 1 << order);
  1035. }
  1036. #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
  1037. static int __init fail_page_alloc_debugfs(void)
  1038. {
  1039. mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
  1040. struct dentry *dir;
  1041. int err;
  1042. err = init_fault_attr_dentries(&fail_page_alloc.attr,
  1043. "fail_page_alloc");
  1044. if (err)
  1045. return err;
  1046. dir = fail_page_alloc.attr.dentries.dir;
  1047. fail_page_alloc.ignore_gfp_wait_file =
  1048. debugfs_create_bool("ignore-gfp-wait", mode, dir,
  1049. &fail_page_alloc.ignore_gfp_wait);
  1050. fail_page_alloc.ignore_gfp_highmem_file =
  1051. debugfs_create_bool("ignore-gfp-highmem", mode, dir,
  1052. &fail_page_alloc.ignore_gfp_highmem);
  1053. fail_page_alloc.min_order_file =
  1054. debugfs_create_u32("min-order", mode, dir,
  1055. &fail_page_alloc.min_order);
  1056. if (!fail_page_alloc.ignore_gfp_wait_file ||
  1057. !fail_page_alloc.ignore_gfp_highmem_file ||
  1058. !fail_page_alloc.min_order_file) {
  1059. err = -ENOMEM;
  1060. debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
  1061. debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
  1062. debugfs_remove(fail_page_alloc.min_order_file);
  1063. cleanup_fault_attr_dentries(&fail_page_alloc.attr);
  1064. }
  1065. return err;
  1066. }
  1067. late_initcall(fail_page_alloc_debugfs);
  1068. #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
  1069. #else /* CONFIG_FAIL_PAGE_ALLOC */
  1070. static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
  1071. {
  1072. return 0;
  1073. }
  1074. #endif /* CONFIG_FAIL_PAGE_ALLOC */
  1075. /*
  1076. * Return 1 if free pages are above 'mark'. This takes into account the order
  1077. * of the allocation.
  1078. */
  1079. int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
  1080. int classzone_idx, int alloc_flags)
  1081. {
  1082. /* free_pages my go negative - that's OK */
  1083. long min = mark;
  1084. long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
  1085. int o;
  1086. if (alloc_flags & ALLOC_HIGH)
  1087. min -= min / 2;
  1088. if (alloc_flags & ALLOC_HARDER)
  1089. min -= min / 4;
  1090. if (free_pages <= min + z->lowmem_reserve[classzone_idx])
  1091. return 0;
  1092. for (o = 0; o < order; o++) {
  1093. /* At the next order, this order's pages become unavailable */
  1094. free_pages -= z->free_area[o].nr_free << o;
  1095. /* Require fewer higher order pages to be free */
  1096. min >>= 1;
  1097. if (free_pages <= min)
  1098. return 0;
  1099. }
  1100. return 1;
  1101. }
  1102. #ifdef CONFIG_NUMA
  1103. /*
  1104. * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
  1105. * skip over zones that are not allowed by the cpuset, or that have
  1106. * been recently (in last second) found to be nearly full. See further
  1107. * comments in mmzone.h. Reduces cache footprint of zonelist scans
  1108. * that have to skip over alot of full or unallowed zones.
  1109. *
  1110. * If the zonelist cache is present in the passed in zonelist, then
  1111. * returns a pointer to the allowed node mask (either the current
  1112. * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
  1113. *
  1114. * If the zonelist cache is not available for this zonelist, does
  1115. * nothing and returns NULL.
  1116. *
  1117. * If the fullzones BITMAP in the zonelist cache is stale (more than
  1118. * a second since last zap'd) then we zap it out (clear its bits.)
  1119. *
  1120. * We hold off even calling zlc_setup, until after we've checked the
  1121. * first zone in the zonelist, on the theory that most allocations will
  1122. * be satisfied from that first zone, so best to examine that zone as
  1123. * quickly as we can.
  1124. */
  1125. static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
  1126. {
  1127. struct zonelist_cache *zlc; /* cached zonelist speedup info */
  1128. nodemask_t *allowednodes; /* zonelist_cache approximation */
  1129. zlc = zonelist->zlcache_ptr;
  1130. if (!zlc)
  1131. return NULL;
  1132. if (jiffies - zlc->last_full_zap > 1 * HZ) {
  1133. bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
  1134. zlc->last_full_zap = jiffies;
  1135. }
  1136. allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
  1137. &cpuset_current_mems_allowed :
  1138. &node_states[N_HIGH_MEMORY];
  1139. return allowednodes;
  1140. }
  1141. /*
  1142. * Given 'z' scanning a zonelist, run a couple of quick checks to see
  1143. * if it is worth looking at further for free memory:
  1144. * 1) Check that the zone isn't thought to be full (doesn't have its
  1145. * bit set in the zonelist_cache fullzones BITMAP).
  1146. * 2) Check that the zones node (obtained from the zonelist_cache
  1147. * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
  1148. * Return true (non-zero) if zone is worth looking at further, or
  1149. * else return false (zero) if it is not.
  1150. *
  1151. * This check -ignores- the distinction between various watermarks,
  1152. * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
  1153. * found to be full for any variation of these watermarks, it will
  1154. * be considered full for up to one second by all requests, unless
  1155. * we are so low on memory on all allowed nodes that we are forced
  1156. * into the second scan of the zonelist.
  1157. *
  1158. * In the second scan we ignore this zonelist cache and exactly
  1159. * apply the watermarks to all zones, even it is slower to do so.
  1160. * We are low on memory in the second scan, and should leave no stone
  1161. * unturned looking for a free page.
  1162. */
  1163. static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
  1164. nodemask_t *allowednodes)
  1165. {
  1166. struct zonelist_cache *zlc; /* cached zonelist speedup info */
  1167. int i; /* index of *z in zonelist zones */
  1168. int n; /* node that zone *z is on */
  1169. zlc = zonelist->zlcache_ptr;
  1170. if (!zlc)
  1171. return 1;
  1172. i = z - zonelist->zones;
  1173. n = zlc->z_to_n[i];
  1174. /* This zone is worth trying if it is allowed but not full */
  1175. return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
  1176. }
  1177. /*
  1178. * Given 'z' scanning a zonelist, set the corresponding bit in
  1179. * zlc->fullzones, so that subsequent attempts to allocate a page
  1180. * from that zone don't waste time re-examining it.
  1181. */
  1182. static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
  1183. {
  1184. struct zonelist_cache *zlc; /* cached zonelist speedup info */
  1185. int i; /* index of *z in zonelist zones */
  1186. zlc = zonelist->zlcache_ptr;
  1187. if (!zlc)
  1188. return;
  1189. i = z - zonelist->zones;
  1190. set_bit(i, zlc->fullzones);
  1191. }
  1192. #else /* CONFIG_NUMA */
  1193. static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
  1194. {
  1195. return NULL;
  1196. }
  1197. static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
  1198. nodemask_t *allowednodes)
  1199. {
  1200. return 1;
  1201. }
  1202. static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
  1203. {
  1204. }
  1205. #endif /* CONFIG_NUMA */
  1206. /*
  1207. * get_page_from_freelist goes through the zonelist trying to allocate
  1208. * a page.
  1209. */
  1210. static struct page *
  1211. get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
  1212. struct zonelist *zonelist, int alloc_flags)
  1213. {
  1214. struct zone **z;
  1215. struct page *page = NULL;
  1216. int classzone_idx = zone_idx(zonelist->zones[0]);
  1217. struct zone *zone;
  1218. nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
  1219. int zlc_active = 0; /* set if using zonelist_cache */
  1220. int did_zlc_setup = 0; /* just call zlc_setup() one time */
  1221. enum zone_type highest_zoneidx = -1; /* Gets set for policy zonelists */
  1222. zonelist_scan:
  1223. /*
  1224. * Scan zonelist, looking for a zone with enough free.
  1225. * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
  1226. */
  1227. z = zonelist->zones;
  1228. do {
  1229. /*
  1230. * In NUMA, this could be a policy zonelist which contains
  1231. * zones that may not be allowed by the current gfp_mask.
  1232. * Check the zone is allowed by the current flags
  1233. */
  1234. if (unlikely(alloc_should_filter_zonelist(zonelist))) {
  1235. if (highest_zoneidx == -1)
  1236. highest_zoneidx = gfp_zone(gfp_mask);
  1237. if (zone_idx(*z) > highest_zoneidx)
  1238. continue;
  1239. }
  1240. if (NUMA_BUILD && zlc_active &&
  1241. !zlc_zone_worth_trying(zonelist, z, allowednodes))
  1242. continue;
  1243. zone = *z;
  1244. if ((alloc_flags & ALLOC_CPUSET) &&
  1245. !cpuset_zone_allowed_softwall(zone, gfp_mask))
  1246. goto try_next_zone;
  1247. if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
  1248. unsigned long mark;
  1249. if (alloc_flags & ALLOC_WMARK_MIN)
  1250. mark = zone->pages_min;
  1251. else if (alloc_flags & ALLOC_WMARK_LOW)
  1252. mark = zone->pages_low;
  1253. else
  1254. mark = zone->pages_high;
  1255. if (!zone_watermark_ok(zone, order, mark,
  1256. classzone_idx, alloc_flags)) {
  1257. if (!zone_reclaim_mode ||
  1258. !zone_reclaim(zone, gfp_mask, order))
  1259. goto this_zone_full;
  1260. }
  1261. }
  1262. page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
  1263. if (page)
  1264. break;
  1265. this_zone_full:
  1266. if (NUMA_BUILD)
  1267. zlc_mark_zone_full(zonelist, z);
  1268. try_next_zone:
  1269. if (NUMA_BUILD && !did_zlc_setup) {
  1270. /* we do zlc_setup after the first zone is tried */
  1271. allowednodes = zlc_setup(zonelist, alloc_flags);
  1272. zlc_active = 1;
  1273. did_zlc_setup = 1;
  1274. }
  1275. } while (*(++z) != NULL);
  1276. if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
  1277. /* Disable zlc cache for second zonelist scan */
  1278. zlc_active = 0;
  1279. goto zonelist_scan;
  1280. }
  1281. return page;
  1282. }
  1283. /*
  1284. * This is the 'heart' of the zoned buddy allocator.
  1285. */
  1286. struct page * fastcall
  1287. __alloc_pages(gfp_t gfp_mask, unsigned int order,
  1288. struct zonelist *zonelist)
  1289. {
  1290. const gfp_t wait = gfp_mask & __GFP_WAIT;
  1291. struct zone **z;
  1292. struct page *page;
  1293. struct reclaim_state reclaim_state;
  1294. struct task_struct *p = current;
  1295. int do_retry;
  1296. int alloc_flags;
  1297. int did_some_progress;
  1298. might_sleep_if(wait);
  1299. if (should_fail_alloc_page(gfp_mask, order))
  1300. return NULL;
  1301. restart:
  1302. z = zonelist->zones; /* the list of zones suitable for gfp_mask */
  1303. if (unlikely(*z == NULL)) {
  1304. /*
  1305. * Happens if we have an empty zonelist as a result of
  1306. * GFP_THISNODE being used on a memoryless node
  1307. */
  1308. return NULL;
  1309. }
  1310. page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
  1311. zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
  1312. if (page)
  1313. goto got_pg;
  1314. /*
  1315. * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
  1316. * __GFP_NOWARN set) should not cause reclaim since the subsystem
  1317. * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
  1318. * using a larger set of nodes after it has established that the
  1319. * allowed per node queues are empty and that nodes are
  1320. * over allocated.
  1321. */
  1322. if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
  1323. goto nopage;
  1324. for (z = zonelist->zones; *z; z++)
  1325. wakeup_kswapd(*z, order);
  1326. /*
  1327. * OK, we're below the kswapd watermark and have kicked background
  1328. * reclaim. Now things get more complex, so set up alloc_flags according
  1329. * to how we want to proceed.
  1330. *
  1331. * The caller may dip into page reserves a bit more if the caller
  1332. * cannot run direct reclaim, or if the caller has realtime scheduling
  1333. * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
  1334. * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
  1335. */
  1336. alloc_flags = ALLOC_WMARK_MIN;
  1337. if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
  1338. alloc_flags |= ALLOC_HARDER;
  1339. if (gfp_mask & __GFP_HIGH)
  1340. alloc_flags |= ALLOC_HIGH;
  1341. if (wait)
  1342. alloc_flags |= ALLOC_CPUSET;
  1343. /*
  1344. * Go through the zonelist again. Let __GFP_HIGH and allocations
  1345. * coming from realtime tasks go deeper into reserves.
  1346. *
  1347. * This is the last chance, in general, before the goto nopage.
  1348. * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
  1349. * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
  1350. */
  1351. page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
  1352. if (page)
  1353. goto got_pg;
  1354. /* This allocation should allow future memory freeing. */
  1355. rebalance:
  1356. if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
  1357. && !in_interrupt()) {
  1358. if (!(gfp_mask & __GFP_NOMEMALLOC)) {
  1359. nofail_alloc:
  1360. /* go through the zonelist yet again, ignoring mins */
  1361. page = get_page_from_freelist(gfp_mask, order,
  1362. zonelist, ALLOC_NO_WATERMARKS);
  1363. if (page)
  1364. goto got_pg;
  1365. if (gfp_mask & __GFP_NOFAIL) {
  1366. congestion_wait(WRITE, HZ/50);
  1367. goto nofail_alloc;
  1368. }
  1369. }
  1370. goto nopage;
  1371. }
  1372. /* Atomic allocations - we can't balance anything */
  1373. if (!wait)
  1374. goto nopage;
  1375. cond_resched();
  1376. /* We now go into synchronous reclaim */
  1377. cpuset_memory_pressure_bump();
  1378. p->flags |= PF_MEMALLOC;
  1379. reclaim_state.reclaimed_slab = 0;
  1380. p->reclaim_state = &reclaim_state;
  1381. did_some_progress = try_to_free_pages(zonelist->zones, order, gfp_mask);
  1382. p->reclaim_state = NULL;
  1383. p->flags &= ~PF_MEMALLOC;
  1384. cond_resched();
  1385. if (order != 0)
  1386. drain_all_local_pages();
  1387. if (likely(did_some_progress)) {
  1388. page = get_page_from_freelist(gfp_mask, order,
  1389. zonelist, alloc_flags);
  1390. if (page)
  1391. goto got_pg;
  1392. } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
  1393. /*
  1394. * Go through the zonelist yet one more time, keep
  1395. * very high watermark here, this is only to catch
  1396. * a parallel oom killing, we must fail if we're still
  1397. * under heavy pressure.
  1398. */
  1399. page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
  1400. zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
  1401. if (page)
  1402. goto got_pg;
  1403. /* The OOM killer will not help higher order allocs so fail */
  1404. if (order > PAGE_ALLOC_COSTLY_ORDER)
  1405. goto nopage;
  1406. out_of_memory(zonelist, gfp_mask, order);
  1407. goto restart;
  1408. }
  1409. /*
  1410. * Don't let big-order allocations loop unless the caller explicitly
  1411. * requests that. Wait for some write requests to complete then retry.
  1412. *
  1413. * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
  1414. * <= 3, but that may not be true in other implementations.
  1415. */
  1416. do_retry = 0;
  1417. if (!(gfp_mask & __GFP_NORETRY)) {
  1418. if ((order <= PAGE_ALLOC_COSTLY_ORDER) ||
  1419. (gfp_mask & __GFP_REPEAT))
  1420. do_retry = 1;
  1421. if (gfp_mask & __GFP_NOFAIL)
  1422. do_retry = 1;
  1423. }
  1424. if (do_retry) {
  1425. congestion_wait(WRITE, HZ/50);
  1426. goto rebalance;
  1427. }
  1428. nopage:
  1429. if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
  1430. printk(KERN_WARNING "%s: page allocation failure."
  1431. " order:%d, mode:0x%x\n",
  1432. p->comm, order, gfp_mask);
  1433. dump_stack();
  1434. show_mem();
  1435. }
  1436. got_pg:
  1437. return page;
  1438. }
  1439. EXPORT_SYMBOL(__alloc_pages);
  1440. /*
  1441. * Common helper functions.
  1442. */
  1443. fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
  1444. {
  1445. struct page * page;
  1446. page = alloc_pages(gfp_mask, order);
  1447. if (!page)
  1448. return 0;
  1449. return (unsigned long) page_address(page);
  1450. }
  1451. EXPORT_SYMBOL(__get_free_pages);
  1452. fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
  1453. {
  1454. struct page * page;
  1455. /*
  1456. * get_zeroed_page() returns a 32-bit address, which cannot represent
  1457. * a highmem page
  1458. */
  1459. VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
  1460. page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
  1461. if (page)
  1462. return (unsigned long) page_address(page);
  1463. return 0;
  1464. }
  1465. EXPORT_SYMBOL(get_zeroed_page);
  1466. void __pagevec_free(struct pagevec *pvec)
  1467. {
  1468. int i = pagevec_count(pvec);
  1469. while (--i >= 0)
  1470. free_hot_cold_page(pvec->pages[i], pvec->cold);
  1471. }
  1472. fastcall void __free_pages(struct page *page, unsigned int order)
  1473. {
  1474. if (put_page_testzero(page)) {
  1475. if (order == 0)
  1476. free_hot_page(page);
  1477. else
  1478. __free_pages_ok(page, order);
  1479. }
  1480. }
  1481. EXPORT_SYMBOL(__free_pages);
  1482. fastcall void free_pages(unsigned long addr, unsigned int order)
  1483. {
  1484. if (addr != 0) {
  1485. VM_BUG_ON(!virt_addr_valid((void *)addr));
  1486. __free_pages(virt_to_page((void *)addr), order);
  1487. }
  1488. }
  1489. EXPORT_SYMBOL(free_pages);
  1490. static unsigned int nr_free_zone_pages(int offset)
  1491. {
  1492. /* Just pick one node, since fallback list is circular */
  1493. pg_data_t *pgdat = NODE_DATA(numa_node_id());
  1494. unsigned int sum = 0;
  1495. struct zonelist *zonelist = pgdat->node_zonelists + offset;
  1496. struct zone **zonep = zonelist->zones;
  1497. struct zone *zone;
  1498. for (zone = *zonep++; zone; zone = *zonep++) {
  1499. unsigned long size = zone->present_pages;
  1500. unsigned long high = zone->pages_high;
  1501. if (size > high)
  1502. sum += size - high;
  1503. }
  1504. return sum;
  1505. }
  1506. /*
  1507. * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
  1508. */
  1509. unsigned int nr_free_buffer_pages(void)
  1510. {
  1511. return nr_free_zone_pages(gfp_zone(GFP_USER));
  1512. }
  1513. EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
  1514. /*
  1515. * Amount of free RAM allocatable within all zones
  1516. */
  1517. unsigned int nr_free_pagecache_pages(void)
  1518. {
  1519. return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
  1520. }
  1521. static inline void show_node(struct zone *zone)
  1522. {
  1523. if (NUMA_BUILD)
  1524. printk("Node %d ", zone_to_nid(zone));
  1525. }
  1526. void si_meminfo(struct sysinfo *val)
  1527. {
  1528. val->totalram = totalram_pages;
  1529. val->sharedram = 0;
  1530. val->freeram = global_page_state(NR_FREE_PAGES);
  1531. val->bufferram = nr_blockdev_pages();
  1532. val->totalhigh = totalhigh_pages;
  1533. val->freehigh = nr_free_highpages();
  1534. val->mem_unit = PAGE_SIZE;
  1535. }
  1536. EXPORT_SYMBOL(si_meminfo);
  1537. #ifdef CONFIG_NUMA
  1538. void si_meminfo_node(struct sysinfo *val, int nid)
  1539. {
  1540. pg_data_t *pgdat = NODE_DATA(nid);
  1541. val->totalram = pgdat->node_present_pages;
  1542. val->freeram = node_page_state(nid, NR_FREE_PAGES);
  1543. #ifdef CONFIG_HIGHMEM
  1544. val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
  1545. val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
  1546. NR_FREE_PAGES);
  1547. #else
  1548. val->totalhigh = 0;
  1549. val->freehigh = 0;
  1550. #endif
  1551. val->mem_unit = PAGE_SIZE;
  1552. }
  1553. #endif
  1554. #define K(x) ((x) << (PAGE_SHIFT-10))
  1555. /*
  1556. * Show free area list (used inside shift_scroll-lock stuff)
  1557. * We also calculate the percentage fragmentation. We do this by counting the
  1558. * memory on each free list with the exception of the first item on the list.
  1559. */
  1560. void show_free_areas(void)
  1561. {
  1562. int cpu;
  1563. struct zone *zone;
  1564. for_each_zone(zone) {
  1565. if (!populated_zone(zone))
  1566. continue;
  1567. show_node(zone);
  1568. printk("%s per-cpu:\n", zone->name);
  1569. for_each_online_cpu(cpu) {
  1570. struct per_cpu_pageset *pageset;
  1571. pageset = zone_pcp(zone, cpu);
  1572. printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
  1573. "Cold: hi:%5d, btch:%4d usd:%4d\n",
  1574. cpu, pageset->pcp[0].high,
  1575. pageset->pcp[0].batch, pageset->pcp[0].count,
  1576. pageset->pcp[1].high, pageset->pcp[1].batch,
  1577. pageset->pcp[1].count);
  1578. }
  1579. }
  1580. printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
  1581. " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
  1582. global_page_state(NR_ACTIVE),
  1583. global_page_state(NR_INACTIVE),
  1584. global_page_state(NR_FILE_DIRTY),
  1585. global_page_state(NR_WRITEBACK),
  1586. global_page_state(NR_UNSTABLE_NFS),
  1587. global_page_state(NR_FREE_PAGES),
  1588. global_page_state(NR_SLAB_RECLAIMABLE) +
  1589. global_page_state(NR_SLAB_UNRECLAIMABLE),
  1590. global_page_state(NR_FILE_MAPPED),
  1591. global_page_state(NR_PAGETABLE),
  1592. global_page_state(NR_BOUNCE));
  1593. for_each_zone(zone) {
  1594. int i;
  1595. if (!populated_zone(zone))
  1596. continue;
  1597. show_node(zone);
  1598. printk("%s"
  1599. " free:%lukB"
  1600. " min:%lukB"
  1601. " low:%lukB"
  1602. " high:%lukB"
  1603. " active:%lukB"
  1604. " inactive:%lukB"
  1605. " present:%lukB"
  1606. " pages_scanned:%lu"
  1607. " all_unreclaimable? %s"
  1608. "\n",
  1609. zone->name,
  1610. K(zone_page_state(zone, NR_FREE_PAGES)),
  1611. K(zone->pages_min),
  1612. K(zone->pages_low),
  1613. K(zone->pages_high),
  1614. K(zone_page_state(zone, NR_ACTIVE)),
  1615. K(zone_page_state(zone, NR_INACTIVE)),
  1616. K(zone->present_pages),
  1617. zone->pages_scanned,
  1618. (zone->all_unreclaimable ? "yes" : "no")
  1619. );
  1620. printk("lowmem_reserve[]:");
  1621. for (i = 0; i < MAX_NR_ZONES; i++)
  1622. printk(" %lu", zone->lowmem_reserve[i]);
  1623. printk("\n");
  1624. }
  1625. for_each_zone(zone) {
  1626. unsigned long nr[MAX_ORDER], flags, order, total = 0;
  1627. if (!populated_zone(zone))
  1628. continue;
  1629. show_node(zone);
  1630. printk("%s: ", zone->name);
  1631. spin_lock_irqsave(&zone->lock, flags);
  1632. for (order = 0; order < MAX_ORDER; order++) {
  1633. nr[order] = zone->free_area[order].nr_free;
  1634. total += nr[order] << order;
  1635. }
  1636. spin_unlock_irqrestore(&zone->lock, flags);
  1637. for (order = 0; order < MAX_ORDER; order++)
  1638. printk("%lu*%lukB ", nr[order], K(1UL) << order);
  1639. printk("= %lukB\n", K(total));
  1640. }
  1641. show_swap_cache_info();
  1642. }
  1643. /*
  1644. * Builds allocation fallback zone lists.
  1645. *
  1646. * Add all populated zones of a node to the zonelist.
  1647. */
  1648. static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
  1649. int nr_zones, enum zone_type zone_type)
  1650. {
  1651. struct zone *zone;
  1652. BUG_ON(zone_type >= MAX_NR_ZONES);
  1653. zone_type++;
  1654. do {
  1655. zone_type--;
  1656. zone = pgdat->node_zones + zone_type;
  1657. if (populated_zone(zone)) {
  1658. zonelist->zones[nr_zones++] = zone;
  1659. check_highest_zone(zone_type);
  1660. }
  1661. } while (zone_type);
  1662. return nr_zones;
  1663. }
  1664. /*
  1665. * zonelist_order:
  1666. * 0 = automatic detection of better ordering.
  1667. * 1 = order by ([node] distance, -zonetype)
  1668. * 2 = order by (-zonetype, [node] distance)
  1669. *
  1670. * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
  1671. * the same zonelist. So only NUMA can configure this param.
  1672. */
  1673. #define ZONELIST_ORDER_DEFAULT 0
  1674. #define ZONELIST_ORDER_NODE 1
  1675. #define ZONELIST_ORDER_ZONE 2
  1676. /* zonelist order in the kernel.
  1677. * set_zonelist_order() will set this to NODE or ZONE.
  1678. */
  1679. static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
  1680. static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
  1681. #ifdef CONFIG_NUMA
  1682. /* The value user specified ....changed by config */
  1683. static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
  1684. /* string for sysctl */
  1685. #define NUMA_ZONELIST_ORDER_LEN 16
  1686. char numa_zonelist_order[16] = "default";
  1687. /*
  1688. * interface for configure zonelist ordering.
  1689. * command line option "numa_zonelist_order"
  1690. * = "[dD]efault - default, automatic configuration.
  1691. * = "[nN]ode - order by node locality, then by zone within node
  1692. * = "[zZ]one - order by zone, then by locality within zone
  1693. */
  1694. static int __parse_numa_zonelist_order(char *s)
  1695. {
  1696. if (*s == 'd' || *s == 'D') {
  1697. user_zonelist_order = ZONELIST_ORDER_DEFAULT;
  1698. } else if (*s == 'n' || *s == 'N') {
  1699. user_zonelist_order = ZONELIST_ORDER_NODE;
  1700. } else if (*s == 'z' || *s == 'Z') {
  1701. user_zonelist_order = ZONELIST_ORDER_ZONE;
  1702. } else {
  1703. printk(KERN_WARNING
  1704. "Ignoring invalid numa_zonelist_order value: "
  1705. "%s\n", s);
  1706. return -EINVAL;
  1707. }
  1708. return 0;
  1709. }
  1710. static __init int setup_numa_zonelist_order(char *s)
  1711. {
  1712. if (s)
  1713. return __parse_numa_zonelist_order(s);
  1714. return 0;
  1715. }
  1716. early_param("numa_zonelist_order", setup_numa_zonelist_order);
  1717. /*
  1718. * sysctl handler for numa_zonelist_order
  1719. */
  1720. int numa_zonelist_order_handler(ctl_table *table, int write,
  1721. struct file *file, void __user *buffer, size_t *length,
  1722. loff_t *ppos)
  1723. {
  1724. char saved_string[NUMA_ZONELIST_ORDER_LEN];
  1725. int ret;
  1726. if (write)
  1727. strncpy(saved_string, (char*)table->data,
  1728. NUMA_ZONELIST_ORDER_LEN);
  1729. ret = proc_dostring(table, write, file, buffer, length, ppos);
  1730. if (ret)
  1731. return ret;
  1732. if (write) {
  1733. int oldval = user_zonelist_order;
  1734. if (__parse_numa_zonelist_order((char*)table->data)) {
  1735. /*
  1736. * bogus value. restore saved string
  1737. */
  1738. strncpy((char*)table->data, saved_string,
  1739. NUMA_ZONELIST_ORDER_LEN);
  1740. user_zonelist_order = oldval;
  1741. } else if (oldval != user_zonelist_order)
  1742. build_all_zonelists();
  1743. }
  1744. return 0;
  1745. }
  1746. #define MAX_NODE_LOAD (num_online_nodes())
  1747. static int node_load[MAX_NUMNODES];
  1748. /**
  1749. * find_next_best_node - find the next node that should appear in a given node's fallback list
  1750. * @node: node whose fallback list we're appending
  1751. * @used_node_mask: nodemask_t of already used nodes
  1752. *
  1753. * We use a number of factors to determine which is the next node that should
  1754. * appear on a given node's fallback list. The node should not have appeared
  1755. * already in @node's fallback list, and it should be the next closest node
  1756. * according to the distance array (which contains arbitrary distance values
  1757. * from each node to each node in the system), and should also prefer nodes
  1758. * with no CPUs, since presumably they'll have very little allocation pressure
  1759. * on them otherwise.
  1760. * It returns -1 if no node is found.
  1761. */
  1762. static int find_next_best_node(int node, nodemask_t *used_node_mask)
  1763. {
  1764. int n, val;
  1765. int min_val = INT_MAX;
  1766. int best_node = -1;
  1767. /* Use the local node if we haven't already */
  1768. if (!node_isset(node, *used_node_mask)) {
  1769. node_set(node, *used_node_mask);
  1770. return node;
  1771. }
  1772. for_each_node_state(n, N_HIGH_MEMORY) {
  1773. cpumask_t tmp;
  1774. /* Don't want a node to appear more than once */
  1775. if (node_isset(n, *used_node_mask))
  1776. continue;
  1777. /* Use the distance array to find the distance */
  1778. val = node_distance(node, n);
  1779. /* Penalize nodes under us ("prefer the next node") */
  1780. val += (n < node);
  1781. /* Give preference to headless and unused nodes */
  1782. tmp = node_to_cpumask(n);
  1783. if (!cpus_empty(tmp))
  1784. val += PENALTY_FOR_NODE_WITH_CPUS;
  1785. /* Slight preference for less loaded node */
  1786. val *= (MAX_NODE_LOAD*MAX_NUMNODES);
  1787. val += node_load[n];
  1788. if (val < min_val) {
  1789. min_val = val;
  1790. best_node = n;
  1791. }
  1792. }
  1793. if (best_node >= 0)
  1794. node_set(best_node, *used_node_mask);
  1795. return best_node;
  1796. }
  1797. /*
  1798. * Build zonelists ordered by node and zones within node.
  1799. * This results in maximum locality--normal zone overflows into local
  1800. * DMA zone, if any--but risks exhausting DMA zone.
  1801. */
  1802. static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
  1803. {
  1804. enum zone_type i;
  1805. int j;
  1806. struct zonelist *zonelist;
  1807. for (i = 0; i < MAX_NR_ZONES; i++) {
  1808. zonelist = pgdat->node_zonelists + i;
  1809. for (j = 0; zonelist->zones[j] != NULL; j++)
  1810. ;
  1811. j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
  1812. zonelist->zones[j] = NULL;
  1813. }
  1814. }
  1815. /*
  1816. * Build gfp_thisnode zonelists
  1817. */
  1818. static void build_thisnode_zonelists(pg_data_t *pgdat)
  1819. {
  1820. enum zone_type i;
  1821. int j;
  1822. struct zonelist *zonelist;
  1823. for (i = 0; i < MAX_NR_ZONES; i++) {
  1824. zonelist = pgdat->node_zonelists + MAX_NR_ZONES + i;
  1825. j = build_zonelists_node(pgdat, zonelist, 0, i);
  1826. zonelist->zones[j] = NULL;
  1827. }
  1828. }
  1829. /*
  1830. * Build zonelists ordered by zone and nodes within zones.
  1831. * This results in conserving DMA zone[s] until all Normal memory is
  1832. * exhausted, but results in overflowing to remote node while memory
  1833. * may still exist in local DMA zone.
  1834. */
  1835. static int node_order[MAX_NUMNODES];
  1836. static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
  1837. {
  1838. enum zone_type i;
  1839. int pos, j, node;
  1840. int zone_type; /* needs to be signed */
  1841. struct zone *z;
  1842. struct zonelist *zonelist;
  1843. for (i = 0; i < MAX_NR_ZONES; i++) {
  1844. zonelist = pgdat->node_zonelists + i;
  1845. pos = 0;
  1846. for (zone_type = i; zone_type >= 0; zone_type--) {
  1847. for (j = 0; j < nr_nodes; j++) {
  1848. node = node_order[j];
  1849. z = &NODE_DATA(node)->node_zones[zone_type];
  1850. if (populated_zone(z)) {
  1851. zonelist->zones[pos++] = z;
  1852. check_highest_zone(zone_type);
  1853. }
  1854. }
  1855. }
  1856. zonelist->zones[pos] = NULL;
  1857. }
  1858. }
  1859. static int default_zonelist_order(void)
  1860. {
  1861. int nid, zone_type;
  1862. unsigned long low_kmem_size,total_size;
  1863. struct zone *z;
  1864. int average_size;
  1865. /*
  1866. * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
  1867. * If they are really small and used heavily, the system can fall
  1868. * into OOM very easily.
  1869. * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
  1870. */
  1871. /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
  1872. low_kmem_size = 0;
  1873. total_size = 0;
  1874. for_each_online_node(nid) {
  1875. for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
  1876. z = &NODE_DATA(nid)->node_zones[zone_type];
  1877. if (populated_zone(z)) {
  1878. if (zone_type < ZONE_NORMAL)
  1879. low_kmem_size += z->present_pages;
  1880. total_size += z->present_pages;
  1881. }
  1882. }
  1883. }
  1884. if (!low_kmem_size || /* there are no DMA area. */
  1885. low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
  1886. return ZONELIST_ORDER_NODE;
  1887. /*
  1888. * look into each node's config.
  1889. * If there is a node whose DMA/DMA32 memory is very big area on
  1890. * local memory, NODE_ORDER may be suitable.
  1891. */
  1892. average_size = total_size /
  1893. (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
  1894. for_each_online_node(nid) {
  1895. low_kmem_size = 0;
  1896. total_size = 0;
  1897. for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
  1898. z = &NODE_DATA(nid)->node_zones[zone_type];
  1899. if (populated_zone(z)) {
  1900. if (zone_type < ZONE_NORMAL)
  1901. low_kmem_size += z->present_pages;
  1902. total_size += z->present_pages;
  1903. }
  1904. }
  1905. if (low_kmem_size &&
  1906. total_size > average_size && /* ignore small node */
  1907. low_kmem_size > total_size * 70/100)
  1908. return ZONELIST_ORDER_NODE;
  1909. }
  1910. return ZONELIST_ORDER_ZONE;
  1911. }
  1912. static void set_zonelist_order(void)
  1913. {
  1914. if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
  1915. current_zonelist_order = default_zonelist_order();
  1916. else
  1917. current_zonelist_order = user_zonelist_order;
  1918. }
  1919. static void build_zonelists(pg_data_t *pgdat)
  1920. {
  1921. int j, node, load;
  1922. enum zone_type i;
  1923. nodemask_t used_mask;
  1924. int local_node, prev_node;
  1925. struct zonelist *zonelist;
  1926. int order = current_zonelist_order;
  1927. /* initialize zonelists */
  1928. for (i = 0; i < MAX_ZONELISTS; i++) {
  1929. zonelist = pgdat->node_zonelists + i;
  1930. zonelist->zones[0] = NULL;
  1931. }
  1932. /* NUMA-aware ordering of nodes */
  1933. local_node = pgdat->node_id;
  1934. load = num_online_nodes();
  1935. prev_node = local_node;
  1936. nodes_clear(used_mask);
  1937. memset(node_load, 0, sizeof(node_load));
  1938. memset(node_order, 0, sizeof(node_order));
  1939. j = 0;
  1940. while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
  1941. int distance = node_distance(local_node, node);
  1942. /*
  1943. * If another node is sufficiently far away then it is better
  1944. * to reclaim pages in a zone before going off node.
  1945. */
  1946. if (distance > RECLAIM_DISTANCE)
  1947. zone_reclaim_mode = 1;
  1948. /*
  1949. * We don't want to pressure a particular node.
  1950. * So adding penalty to the first node in same
  1951. * distance group to make it round-robin.
  1952. */
  1953. if (distance != node_distance(local_node, prev_node))
  1954. node_load[node] = load;
  1955. prev_node = node;
  1956. load--;
  1957. if (order == ZONELIST_ORDER_NODE)
  1958. build_zonelists_in_node_order(pgdat, node);
  1959. else
  1960. node_order[j++] = node; /* remember order */
  1961. }
  1962. if (order == ZONELIST_ORDER_ZONE) {
  1963. /* calculate node order -- i.e., DMA last! */
  1964. build_zonelists_in_zone_order(pgdat, j);
  1965. }
  1966. build_thisnode_zonelists(pgdat);
  1967. }
  1968. /* Construct the zonelist performance cache - see further mmzone.h */
  1969. static void build_zonelist_cache(pg_data_t *pgdat)
  1970. {
  1971. int i;
  1972. for (i = 0; i < MAX_NR_ZONES; i++) {
  1973. struct zonelist *zonelist;
  1974. struct zonelist_cache *zlc;
  1975. struct zone **z;
  1976. zonelist = pgdat->node_zonelists + i;
  1977. zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
  1978. bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
  1979. for (z = zonelist->zones; *z; z++)
  1980. zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
  1981. }
  1982. }
  1983. #else /* CONFIG_NUMA */
  1984. static void set_zonelist_order(void)
  1985. {
  1986. current_zonelist_order = ZONELIST_ORDER_ZONE;
  1987. }
  1988. static void build_zonelists(pg_data_t *pgdat)
  1989. {
  1990. int node, local_node;
  1991. enum zone_type i,j;
  1992. local_node = pgdat->node_id;
  1993. for (i = 0; i < MAX_NR_ZONES; i++) {
  1994. struct zonelist *zonelist;
  1995. zonelist = pgdat->node_zonelists + i;
  1996. j = build_zonelists_node(pgdat, zonelist, 0, i);
  1997. /*
  1998. * Now we build the zonelist so that it contains the zones
  1999. * of all the other nodes.
  2000. * We don't want to pressure a particular node, so when
  2001. * building the zones for node N, we make sure that the
  2002. * zones coming right after the local ones are those from
  2003. * node N+1 (modulo N)
  2004. */
  2005. for (node = local_node + 1; node < MAX_NUMNODES; node++) {
  2006. if (!node_online(node))
  2007. continue;
  2008. j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
  2009. }
  2010. for (node = 0; node < local_node; node++) {
  2011. if (!node_online(node))
  2012. continue;
  2013. j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
  2014. }
  2015. zonelist->zones[j] = NULL;
  2016. }
  2017. }
  2018. /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
  2019. static void build_zonelist_cache(pg_data_t *pgdat)
  2020. {
  2021. int i;
  2022. for (i = 0; i < MAX_NR_ZONES; i++)
  2023. pgdat->node_zonelists[i].zlcache_ptr = NULL;
  2024. }
  2025. #endif /* CONFIG_NUMA */
  2026. /* return values int ....just for stop_machine_run() */
  2027. static int __build_all_zonelists(void *dummy)
  2028. {
  2029. int nid;
  2030. for_each_online_node(nid) {
  2031. pg_data_t *pgdat = NODE_DATA(nid);
  2032. build_zonelists(pgdat);
  2033. build_zonelist_cache(pgdat);
  2034. }
  2035. return 0;
  2036. }
  2037. void build_all_zonelists(void)
  2038. {
  2039. set_zonelist_order();
  2040. if (system_state == SYSTEM_BOOTING) {
  2041. __build_all_zonelists(NULL);
  2042. cpuset_init_current_mems_allowed();
  2043. } else {
  2044. /* we have to stop all cpus to guaranntee there is no user
  2045. of zonelist */
  2046. stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
  2047. /* cpuset refresh routine should be here */
  2048. }
  2049. vm_total_pages = nr_free_pagecache_pages();
  2050. printk("Built %i zonelists in %s order. Total pages: %ld\n",
  2051. num_online_nodes(),
  2052. zonelist_order_name[current_zonelist_order],
  2053. vm_total_pages);
  2054. #ifdef CONFIG_NUMA
  2055. printk("Policy zone: %s\n", zone_names[policy_zone]);
  2056. #endif
  2057. }
  2058. /*
  2059. * Helper functions to size the waitqueue hash table.
  2060. * Essentially these want to choose hash table sizes sufficiently
  2061. * large so that collisions trying to wait on pages are rare.
  2062. * But in fact, the number of active page waitqueues on typical
  2063. * systems is ridiculously low, less than 200. So this is even
  2064. * conservative, even though it seems large.
  2065. *
  2066. * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
  2067. * waitqueues, i.e. the size of the waitq table given the number of pages.
  2068. */
  2069. #define PAGES_PER_WAITQUEUE 256
  2070. #ifndef CONFIG_MEMORY_HOTPLUG
  2071. static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
  2072. {
  2073. unsigned long size = 1;
  2074. pages /= PAGES_PER_WAITQUEUE;
  2075. while (size < pages)
  2076. size <<= 1;
  2077. /*
  2078. * Once we have dozens or even hundreds of threads sleeping
  2079. * on IO we've got bigger problems than wait queue collision.
  2080. * Limit the size of the wait table to a reasonable size.
  2081. */
  2082. size = min(size, 4096UL);
  2083. return max(size, 4UL);
  2084. }
  2085. #else
  2086. /*
  2087. * A zone's size might be changed by hot-add, so it is not possible to determine
  2088. * a suitable size for its wait_table. So we use the maximum size now.
  2089. *
  2090. * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
  2091. *
  2092. * i386 (preemption config) : 4096 x 16 = 64Kbyte.
  2093. * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
  2094. * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
  2095. *
  2096. * The maximum entries are prepared when a zone's memory is (512K + 256) pages
  2097. * or more by the traditional way. (See above). It equals:
  2098. *
  2099. * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
  2100. * ia64(16K page size) : = ( 8G + 4M)byte.
  2101. * powerpc (64K page size) : = (32G +16M)byte.
  2102. */
  2103. static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
  2104. {
  2105. return 4096UL;
  2106. }
  2107. #endif
  2108. /*
  2109. * This is an integer logarithm so that shifts can be used later
  2110. * to extract the more random high bits from the multiplicative
  2111. * hash function before the remainder is taken.
  2112. */
  2113. static inline unsigned long wait_table_bits(unsigned long size)
  2114. {
  2115. return ffz(~size);
  2116. }
  2117. #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
  2118. /*
  2119. * Initially all pages are reserved - free ones are freed
  2120. * up by free_all_bootmem() once the early boot process is
  2121. * done. Non-atomic initialization, single-pass.
  2122. */
  2123. void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
  2124. unsigned long start_pfn, enum memmap_context context)
  2125. {
  2126. struct page *page;
  2127. unsigned long end_pfn = start_pfn + size;
  2128. unsigned long pfn;
  2129. for (pfn = start_pfn; pfn < end_pfn; pfn++) {
  2130. /*
  2131. * There can be holes in boot-time mem_map[]s
  2132. * handed to this function. They do not
  2133. * exist on hotplugged memory.
  2134. */
  2135. if (context == MEMMAP_EARLY) {
  2136. if (!early_pfn_valid(pfn))
  2137. continue;
  2138. if (!early_pfn_in_nid(pfn, nid))
  2139. continue;
  2140. }
  2141. page = pfn_to_page(pfn);
  2142. set_page_links(page, zone, nid, pfn);
  2143. init_page_count(page);
  2144. reset_page_mapcount(page);
  2145. SetPageReserved(page);
  2146. /*
  2147. * Mark the block movable so that blocks are reserved for
  2148. * movable at startup. This will force kernel allocations
  2149. * to reserve their blocks rather than leaking throughout
  2150. * the address space during boot when many long-lived
  2151. * kernel allocations are made
  2152. */
  2153. set_pageblock_migratetype(page, MIGRATE_MOVABLE);
  2154. INIT_LIST_HEAD(&page->lru);
  2155. #ifdef WANT_PAGE_VIRTUAL
  2156. /* The shift won't overflow because ZONE_NORMAL is below 4G. */
  2157. if (!is_highmem_idx(zone))
  2158. set_page_address(page, __va(pfn << PAGE_SHIFT));
  2159. #endif
  2160. }
  2161. }
  2162. static void __meminit zone_init_free_lists(struct pglist_data *pgdat,
  2163. struct zone *zone, unsigned long size)
  2164. {
  2165. int order, t;
  2166. for_each_migratetype_order(order, t) {
  2167. INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
  2168. zone->free_area[order].nr_free = 0;
  2169. }
  2170. }
  2171. #ifndef __HAVE_ARCH_MEMMAP_INIT
  2172. #define memmap_init(size, nid, zone, start_pfn) \
  2173. memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
  2174. #endif
  2175. static int __devinit zone_batchsize(struct zone *zone)
  2176. {
  2177. int batch;
  2178. /*
  2179. * The per-cpu-pages pools are set to around 1000th of the
  2180. * size of the zone. But no more than 1/2 of a meg.
  2181. *
  2182. * OK, so we don't know how big the cache is. So guess.
  2183. */
  2184. batch = zone->present_pages / 1024;
  2185. if (batch * PAGE_SIZE > 512 * 1024)
  2186. batch = (512 * 1024) / PAGE_SIZE;
  2187. batch /= 4; /* We effectively *= 4 below */
  2188. if (batch < 1)
  2189. batch = 1;
  2190. /*
  2191. * Clamp the batch to a 2^n - 1 value. Having a power
  2192. * of 2 value was found to be more likely to have
  2193. * suboptimal cache aliasing properties in some cases.
  2194. *
  2195. * For example if 2 tasks are alternately allocating
  2196. * batches of pages, one task can end up with a lot
  2197. * of pages of one half of the possible page colors
  2198. * and the other with pages of the other colors.
  2199. */
  2200. batch = (1 << (fls(batch + batch/2)-1)) - 1;
  2201. return batch;
  2202. }
  2203. inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
  2204. {
  2205. struct per_cpu_pages *pcp;
  2206. memset(p, 0, sizeof(*p));
  2207. pcp = &p->pcp[0]; /* hot */
  2208. pcp->count = 0;
  2209. pcp->high = 6 * batch;
  2210. pcp->batch = max(1UL, 1 * batch);
  2211. INIT_LIST_HEAD(&pcp->list);
  2212. pcp = &p->pcp[1]; /* cold*/
  2213. pcp->count = 0;
  2214. pcp->high = 2 * batch;
  2215. pcp->batch = max(1UL, batch/2);
  2216. INIT_LIST_HEAD(&pcp->list);
  2217. }
  2218. /*
  2219. * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
  2220. * to the value high for the pageset p.
  2221. */
  2222. static void setup_pagelist_highmark(struct per_cpu_pageset *p,
  2223. unsigned long high)
  2224. {
  2225. struct per_cpu_pages *pcp;
  2226. pcp = &p->pcp[0]; /* hot list */
  2227. pcp->high = high;
  2228. pcp->batch = max(1UL, high/4);
  2229. if ((high/4) > (PAGE_SHIFT * 8))
  2230. pcp->batch = PAGE_SHIFT * 8;
  2231. }
  2232. #ifdef CONFIG_NUMA
  2233. /*
  2234. * Boot pageset table. One per cpu which is going to be used for all
  2235. * zones and all nodes. The parameters will be set in such a way
  2236. * that an item put on a list will immediately be handed over to
  2237. * the buddy list. This is safe since pageset manipulation is done
  2238. * with interrupts disabled.
  2239. *
  2240. * Some NUMA counter updates may also be caught by the boot pagesets.
  2241. *
  2242. * The boot_pagesets must be kept even after bootup is complete for
  2243. * unused processors and/or zones. They do play a role for bootstrapping
  2244. * hotplugged processors.
  2245. *
  2246. * zoneinfo_show() and maybe other functions do
  2247. * not check if the processor is online before following the pageset pointer.
  2248. * Other parts of the kernel may not check if the zone is available.
  2249. */
  2250. static struct per_cpu_pageset boot_pageset[NR_CPUS];
  2251. /*
  2252. * Dynamically allocate memory for the
  2253. * per cpu pageset array in struct zone.
  2254. */
  2255. static int __cpuinit process_zones(int cpu)
  2256. {
  2257. struct zone *zone, *dzone;
  2258. int node = cpu_to_node(cpu);
  2259. node_set_state(node, N_CPU); /* this node has a cpu */
  2260. for_each_zone(zone) {
  2261. if (!populated_zone(zone))
  2262. continue;
  2263. zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
  2264. GFP_KERNEL, node);
  2265. if (!zone_pcp(zone, cpu))
  2266. goto bad;
  2267. setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
  2268. if (percpu_pagelist_fraction)
  2269. setup_pagelist_highmark(zone_pcp(zone, cpu),
  2270. (zone->present_pages / percpu_pagelist_fraction));
  2271. }
  2272. return 0;
  2273. bad:
  2274. for_each_zone(dzone) {
  2275. if (!populated_zone(dzone))
  2276. continue;
  2277. if (dzone == zone)
  2278. break;
  2279. kfree(zone_pcp(dzone, cpu));
  2280. zone_pcp(dzone, cpu) = NULL;
  2281. }
  2282. return -ENOMEM;
  2283. }
  2284. static inline void free_zone_pagesets(int cpu)
  2285. {
  2286. struct zone *zone;
  2287. for_each_zone(zone) {
  2288. struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
  2289. /* Free per_cpu_pageset if it is slab allocated */
  2290. if (pset != &boot_pageset[cpu])
  2291. kfree(pset);
  2292. zone_pcp(zone, cpu) = NULL;
  2293. }
  2294. }
  2295. static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
  2296. unsigned long action,
  2297. void *hcpu)
  2298. {
  2299. int cpu = (long)hcpu;
  2300. int ret = NOTIFY_OK;
  2301. switch (action) {
  2302. case CPU_UP_PREPARE:
  2303. case CPU_UP_PREPARE_FROZEN:
  2304. if (process_zones(cpu))
  2305. ret = NOTIFY_BAD;
  2306. break;
  2307. case CPU_UP_CANCELED:
  2308. case CPU_UP_CANCELED_FROZEN:
  2309. case CPU_DEAD:
  2310. case CPU_DEAD_FROZEN:
  2311. free_zone_pagesets(cpu);
  2312. break;
  2313. default:
  2314. break;
  2315. }
  2316. return ret;
  2317. }
  2318. static struct notifier_block __cpuinitdata pageset_notifier =
  2319. { &pageset_cpuup_callback, NULL, 0 };
  2320. void __init setup_per_cpu_pageset(void)
  2321. {
  2322. int err;
  2323. /* Initialize per_cpu_pageset for cpu 0.
  2324. * A cpuup callback will do this for every cpu
  2325. * as it comes online
  2326. */
  2327. err = process_zones(smp_processor_id());
  2328. BUG_ON(err);
  2329. register_cpu_notifier(&pageset_notifier);
  2330. }
  2331. #endif
  2332. static noinline __init_refok
  2333. int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
  2334. {
  2335. int i;
  2336. struct pglist_data *pgdat = zone->zone_pgdat;
  2337. size_t alloc_size;
  2338. /*
  2339. * The per-page waitqueue mechanism uses hashed waitqueues
  2340. * per zone.
  2341. */
  2342. zone->wait_table_hash_nr_entries =
  2343. wait_table_hash_nr_entries(zone_size_pages);
  2344. zone->wait_table_bits =
  2345. wait_table_bits(zone->wait_table_hash_nr_entries);
  2346. alloc_size = zone->wait_table_hash_nr_entries
  2347. * sizeof(wait_queue_head_t);
  2348. if (system_state == SYSTEM_BOOTING) {
  2349. zone->wait_table = (wait_queue_head_t *)
  2350. alloc_bootmem_node(pgdat, alloc_size);
  2351. } else {
  2352. /*
  2353. * This case means that a zone whose size was 0 gets new memory
  2354. * via memory hot-add.
  2355. * But it may be the case that a new node was hot-added. In
  2356. * this case vmalloc() will not be able to use this new node's
  2357. * memory - this wait_table must be initialized to use this new
  2358. * node itself as well.
  2359. * To use this new node's memory, further consideration will be
  2360. * necessary.
  2361. */
  2362. zone->wait_table = vmalloc(alloc_size);
  2363. }
  2364. if (!zone->wait_table)
  2365. return -ENOMEM;
  2366. for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
  2367. init_waitqueue_head(zone->wait_table + i);
  2368. return 0;
  2369. }
  2370. static __meminit void zone_pcp_init(struct zone *zone)
  2371. {
  2372. int cpu;
  2373. unsigned long batch = zone_batchsize(zone);
  2374. for (cpu = 0; cpu < NR_CPUS; cpu++) {
  2375. #ifdef CONFIG_NUMA
  2376. /* Early boot. Slab allocator not functional yet */
  2377. zone_pcp(zone, cpu) = &boot_pageset[cpu];
  2378. setup_pageset(&boot_pageset[cpu],0);
  2379. #else
  2380. setup_pageset(zone_pcp(zone,cpu), batch);
  2381. #endif
  2382. }
  2383. if (zone->present_pages)
  2384. printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
  2385. zone->name, zone->present_pages, batch);
  2386. }
  2387. __meminit int init_currently_empty_zone(struct zone *zone,
  2388. unsigned long zone_start_pfn,
  2389. unsigned long size,
  2390. enum memmap_context context)
  2391. {
  2392. struct pglist_data *pgdat = zone->zone_pgdat;
  2393. int ret;
  2394. ret = zone_wait_table_init(zone, size);
  2395. if (ret)
  2396. return ret;
  2397. pgdat->nr_zones = zone_idx(zone) + 1;
  2398. zone->zone_start_pfn = zone_start_pfn;
  2399. memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
  2400. zone_init_free_lists(pgdat, zone, zone->spanned_pages);
  2401. return 0;
  2402. }
  2403. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  2404. /*
  2405. * Basic iterator support. Return the first range of PFNs for a node
  2406. * Note: nid == MAX_NUMNODES returns first region regardless of node
  2407. */
  2408. static int __meminit first_active_region_index_in_nid(int nid)
  2409. {
  2410. int i;
  2411. for (i = 0; i < nr_nodemap_entries; i++)
  2412. if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
  2413. return i;
  2414. return -1;
  2415. }
  2416. /*
  2417. * Basic iterator support. Return the next active range of PFNs for a node
  2418. * Note: nid == MAX_NUMNODES returns next region regardles of node
  2419. */
  2420. static int __meminit next_active_region_index_in_nid(int index, int nid)
  2421. {
  2422. for (index = index + 1; index < nr_nodemap_entries; index++)
  2423. if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
  2424. return index;
  2425. return -1;
  2426. }
  2427. #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
  2428. /*
  2429. * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
  2430. * Architectures may implement their own version but if add_active_range()
  2431. * was used and there are no special requirements, this is a convenient
  2432. * alternative
  2433. */
  2434. int __meminit early_pfn_to_nid(unsigned long pfn)
  2435. {
  2436. int i;
  2437. for (i = 0; i < nr_nodemap_entries; i++) {
  2438. unsigned long start_pfn = early_node_map[i].start_pfn;
  2439. unsigned long end_pfn = early_node_map[i].end_pfn;
  2440. if (start_pfn <= pfn && pfn < end_pfn)
  2441. return early_node_map[i].nid;
  2442. }
  2443. return 0;
  2444. }
  2445. #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
  2446. /* Basic iterator support to walk early_node_map[] */
  2447. #define for_each_active_range_index_in_nid(i, nid) \
  2448. for (i = first_active_region_index_in_nid(nid); i != -1; \
  2449. i = next_active_region_index_in_nid(i, nid))
  2450. /**
  2451. * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
  2452. * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
  2453. * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
  2454. *
  2455. * If an architecture guarantees that all ranges registered with
  2456. * add_active_ranges() contain no holes and may be freed, this
  2457. * this function may be used instead of calling free_bootmem() manually.
  2458. */
  2459. void __init free_bootmem_with_active_regions(int nid,
  2460. unsigned long max_low_pfn)
  2461. {
  2462. int i;
  2463. for_each_active_range_index_in_nid(i, nid) {
  2464. unsigned long size_pages = 0;
  2465. unsigned long end_pfn = early_node_map[i].end_pfn;
  2466. if (early_node_map[i].start_pfn >= max_low_pfn)
  2467. continue;
  2468. if (end_pfn > max_low_pfn)
  2469. end_pfn = max_low_pfn;
  2470. size_pages = end_pfn - early_node_map[i].start_pfn;
  2471. free_bootmem_node(NODE_DATA(early_node_map[i].nid),
  2472. PFN_PHYS(early_node_map[i].start_pfn),
  2473. size_pages << PAGE_SHIFT);
  2474. }
  2475. }
  2476. /**
  2477. * sparse_memory_present_with_active_regions - Call memory_present for each active range
  2478. * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
  2479. *
  2480. * If an architecture guarantees that all ranges registered with
  2481. * add_active_ranges() contain no holes and may be freed, this
  2482. * function may be used instead of calling memory_present() manually.
  2483. */
  2484. void __init sparse_memory_present_with_active_regions(int nid)
  2485. {
  2486. int i;
  2487. for_each_active_range_index_in_nid(i, nid)
  2488. memory_present(early_node_map[i].nid,
  2489. early_node_map[i].start_pfn,
  2490. early_node_map[i].end_pfn);
  2491. }
  2492. /**
  2493. * push_node_boundaries - Push node boundaries to at least the requested boundary
  2494. * @nid: The nid of the node to push the boundary for
  2495. * @start_pfn: The start pfn of the node
  2496. * @end_pfn: The end pfn of the node
  2497. *
  2498. * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
  2499. * time. Specifically, on x86_64, SRAT will report ranges that can potentially
  2500. * be hotplugged even though no physical memory exists. This function allows
  2501. * an arch to push out the node boundaries so mem_map is allocated that can
  2502. * be used later.
  2503. */
  2504. #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
  2505. void __init push_node_boundaries(unsigned int nid,
  2506. unsigned long start_pfn, unsigned long end_pfn)
  2507. {
  2508. printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
  2509. nid, start_pfn, end_pfn);
  2510. /* Initialise the boundary for this node if necessary */
  2511. if (node_boundary_end_pfn[nid] == 0)
  2512. node_boundary_start_pfn[nid] = -1UL;
  2513. /* Update the boundaries */
  2514. if (node_boundary_start_pfn[nid] > start_pfn)
  2515. node_boundary_start_pfn[nid] = start_pfn;
  2516. if (node_boundary_end_pfn[nid] < end_pfn)
  2517. node_boundary_end_pfn[nid] = end_pfn;
  2518. }
  2519. /* If necessary, push the node boundary out for reserve hotadd */
  2520. static void __meminit account_node_boundary(unsigned int nid,
  2521. unsigned long *start_pfn, unsigned long *end_pfn)
  2522. {
  2523. printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
  2524. nid, *start_pfn, *end_pfn);
  2525. /* Return if boundary information has not been provided */
  2526. if (node_boundary_end_pfn[nid] == 0)
  2527. return;
  2528. /* Check the boundaries and update if necessary */
  2529. if (node_boundary_start_pfn[nid] < *start_pfn)
  2530. *start_pfn = node_boundary_start_pfn[nid];
  2531. if (node_boundary_end_pfn[nid] > *end_pfn)
  2532. *end_pfn = node_boundary_end_pfn[nid];
  2533. }
  2534. #else
  2535. void __init push_node_boundaries(unsigned int nid,
  2536. unsigned long start_pfn, unsigned long end_pfn) {}
  2537. static void __meminit account_node_boundary(unsigned int nid,
  2538. unsigned long *start_pfn, unsigned long *end_pfn) {}
  2539. #endif
  2540. /**
  2541. * get_pfn_range_for_nid - Return the start and end page frames for a node
  2542. * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
  2543. * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
  2544. * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
  2545. *
  2546. * It returns the start and end page frame of a node based on information
  2547. * provided by an arch calling add_active_range(). If called for a node
  2548. * with no available memory, a warning is printed and the start and end
  2549. * PFNs will be 0.
  2550. */
  2551. void __meminit get_pfn_range_for_nid(unsigned int nid,
  2552. unsigned long *start_pfn, unsigned long *end_pfn)
  2553. {
  2554. int i;
  2555. *start_pfn = -1UL;
  2556. *end_pfn = 0;
  2557. for_each_active_range_index_in_nid(i, nid) {
  2558. *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
  2559. *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
  2560. }
  2561. if (*start_pfn == -1UL)
  2562. *start_pfn = 0;
  2563. /* Push the node boundaries out if requested */
  2564. account_node_boundary(nid, start_pfn, end_pfn);
  2565. }
  2566. /*
  2567. * This finds a zone that can be used for ZONE_MOVABLE pages. The
  2568. * assumption is made that zones within a node are ordered in monotonic
  2569. * increasing memory addresses so that the "highest" populated zone is used
  2570. */
  2571. void __init find_usable_zone_for_movable(void)
  2572. {
  2573. int zone_index;
  2574. for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
  2575. if (zone_index == ZONE_MOVABLE)
  2576. continue;
  2577. if (arch_zone_highest_possible_pfn[zone_index] >
  2578. arch_zone_lowest_possible_pfn[zone_index])
  2579. break;
  2580. }
  2581. VM_BUG_ON(zone_index == -1);
  2582. movable_zone = zone_index;
  2583. }
  2584. /*
  2585. * The zone ranges provided by the architecture do not include ZONE_MOVABLE
  2586. * because it is sized independant of architecture. Unlike the other zones,
  2587. * the starting point for ZONE_MOVABLE is not fixed. It may be different
  2588. * in each node depending on the size of each node and how evenly kernelcore
  2589. * is distributed. This helper function adjusts the zone ranges
  2590. * provided by the architecture for a given node by using the end of the
  2591. * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
  2592. * zones within a node are in order of monotonic increases memory addresses
  2593. */
  2594. void __meminit adjust_zone_range_for_zone_movable(int nid,
  2595. unsigned long zone_type,
  2596. unsigned long node_start_pfn,
  2597. unsigned long node_end_pfn,
  2598. unsigned long *zone_start_pfn,
  2599. unsigned long *zone_end_pfn)
  2600. {
  2601. /* Only adjust if ZONE_MOVABLE is on this node */
  2602. if (zone_movable_pfn[nid]) {
  2603. /* Size ZONE_MOVABLE */
  2604. if (zone_type == ZONE_MOVABLE) {
  2605. *zone_start_pfn = zone_movable_pfn[nid];
  2606. *zone_end_pfn = min(node_end_pfn,
  2607. arch_zone_highest_possible_pfn[movable_zone]);
  2608. /* Adjust for ZONE_MOVABLE starting within this range */
  2609. } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
  2610. *zone_end_pfn > zone_movable_pfn[nid]) {
  2611. *zone_end_pfn = zone_movable_pfn[nid];
  2612. /* Check if this whole range is within ZONE_MOVABLE */
  2613. } else if (*zone_start_pfn >= zone_movable_pfn[nid])
  2614. *zone_start_pfn = *zone_end_pfn;
  2615. }
  2616. }
  2617. /*
  2618. * Return the number of pages a zone spans in a node, including holes
  2619. * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
  2620. */
  2621. static unsigned long __meminit zone_spanned_pages_in_node(int nid,
  2622. unsigned long zone_type,
  2623. unsigned long *ignored)
  2624. {
  2625. unsigned long node_start_pfn, node_end_pfn;
  2626. unsigned long zone_start_pfn, zone_end_pfn;
  2627. /* Get the start and end of the node and zone */
  2628. get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
  2629. zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
  2630. zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
  2631. adjust_zone_range_for_zone_movable(nid, zone_type,
  2632. node_start_pfn, node_end_pfn,
  2633. &zone_start_pfn, &zone_end_pfn);
  2634. /* Check that this node has pages within the zone's required range */
  2635. if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
  2636. return 0;
  2637. /* Move the zone boundaries inside the node if necessary */
  2638. zone_end_pfn = min(zone_end_pfn, node_end_pfn);
  2639. zone_start_pfn = max(zone_start_pfn, node_start_pfn);
  2640. /* Return the spanned pages */
  2641. return zone_end_pfn - zone_start_pfn;
  2642. }
  2643. /*
  2644. * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
  2645. * then all holes in the requested range will be accounted for.
  2646. */
  2647. unsigned long __meminit __absent_pages_in_range(int nid,
  2648. unsigned long range_start_pfn,
  2649. unsigned long range_end_pfn)
  2650. {
  2651. int i = 0;
  2652. unsigned long prev_end_pfn = 0, hole_pages = 0;
  2653. unsigned long start_pfn;
  2654. /* Find the end_pfn of the first active range of pfns in the node */
  2655. i = first_active_region_index_in_nid(nid);
  2656. if (i == -1)
  2657. return 0;
  2658. prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
  2659. /* Account for ranges before physical memory on this node */
  2660. if (early_node_map[i].start_pfn > range_start_pfn)
  2661. hole_pages = prev_end_pfn - range_start_pfn;
  2662. /* Find all holes for the zone within the node */
  2663. for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
  2664. /* No need to continue if prev_end_pfn is outside the zone */
  2665. if (prev_end_pfn >= range_end_pfn)
  2666. break;
  2667. /* Make sure the end of the zone is not within the hole */
  2668. start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
  2669. prev_end_pfn = max(prev_end_pfn, range_start_pfn);
  2670. /* Update the hole size cound and move on */
  2671. if (start_pfn > range_start_pfn) {
  2672. BUG_ON(prev_end_pfn > start_pfn);
  2673. hole_pages += start_pfn - prev_end_pfn;
  2674. }
  2675. prev_end_pfn = early_node_map[i].end_pfn;
  2676. }
  2677. /* Account for ranges past physical memory on this node */
  2678. if (range_end_pfn > prev_end_pfn)
  2679. hole_pages += range_end_pfn -
  2680. max(range_start_pfn, prev_end_pfn);
  2681. return hole_pages;
  2682. }
  2683. /**
  2684. * absent_pages_in_range - Return number of page frames in holes within a range
  2685. * @start_pfn: The start PFN to start searching for holes
  2686. * @end_pfn: The end PFN to stop searching for holes
  2687. *
  2688. * It returns the number of pages frames in memory holes within a range.
  2689. */
  2690. unsigned long __init absent_pages_in_range(unsigned long start_pfn,
  2691. unsigned long end_pfn)
  2692. {
  2693. return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
  2694. }
  2695. /* Return the number of page frames in holes in a zone on a node */
  2696. static unsigned long __meminit zone_absent_pages_in_node(int nid,
  2697. unsigned long zone_type,
  2698. unsigned long *ignored)
  2699. {
  2700. unsigned long node_start_pfn, node_end_pfn;
  2701. unsigned long zone_start_pfn, zone_end_pfn;
  2702. get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
  2703. zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
  2704. node_start_pfn);
  2705. zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
  2706. node_end_pfn);
  2707. adjust_zone_range_for_zone_movable(nid, zone_type,
  2708. node_start_pfn, node_end_pfn,
  2709. &zone_start_pfn, &zone_end_pfn);
  2710. return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
  2711. }
  2712. #else
  2713. static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
  2714. unsigned long zone_type,
  2715. unsigned long *zones_size)
  2716. {
  2717. return zones_size[zone_type];
  2718. }
  2719. static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
  2720. unsigned long zone_type,
  2721. unsigned long *zholes_size)
  2722. {
  2723. if (!zholes_size)
  2724. return 0;
  2725. return zholes_size[zone_type];
  2726. }
  2727. #endif
  2728. static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
  2729. unsigned long *zones_size, unsigned long *zholes_size)
  2730. {
  2731. unsigned long realtotalpages, totalpages = 0;
  2732. enum zone_type i;
  2733. for (i = 0; i < MAX_NR_ZONES; i++)
  2734. totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
  2735. zones_size);
  2736. pgdat->node_spanned_pages = totalpages;
  2737. realtotalpages = totalpages;
  2738. for (i = 0; i < MAX_NR_ZONES; i++)
  2739. realtotalpages -=
  2740. zone_absent_pages_in_node(pgdat->node_id, i,
  2741. zholes_size);
  2742. pgdat->node_present_pages = realtotalpages;
  2743. printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
  2744. realtotalpages);
  2745. }
  2746. #ifndef CONFIG_SPARSEMEM
  2747. /*
  2748. * Calculate the size of the zone->blockflags rounded to an unsigned long
  2749. * Start by making sure zonesize is a multiple of MAX_ORDER-1 by rounding up
  2750. * Then figure 1 NR_PAGEBLOCK_BITS worth of bits per MAX_ORDER-1, finally
  2751. * round what is now in bits to nearest long in bits, then return it in
  2752. * bytes.
  2753. */
  2754. static unsigned long __init usemap_size(unsigned long zonesize)
  2755. {
  2756. unsigned long usemapsize;
  2757. usemapsize = roundup(zonesize, MAX_ORDER_NR_PAGES);
  2758. usemapsize = usemapsize >> (MAX_ORDER-1);
  2759. usemapsize *= NR_PAGEBLOCK_BITS;
  2760. usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
  2761. return usemapsize / 8;
  2762. }
  2763. static void __init setup_usemap(struct pglist_data *pgdat,
  2764. struct zone *zone, unsigned long zonesize)
  2765. {
  2766. unsigned long usemapsize = usemap_size(zonesize);
  2767. zone->pageblock_flags = NULL;
  2768. if (usemapsize) {
  2769. zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
  2770. memset(zone->pageblock_flags, 0, usemapsize);
  2771. }
  2772. }
  2773. #else
  2774. static void inline setup_usemap(struct pglist_data *pgdat,
  2775. struct zone *zone, unsigned long zonesize) {}
  2776. #endif /* CONFIG_SPARSEMEM */
  2777. /*
  2778. * Set up the zone data structures:
  2779. * - mark all pages reserved
  2780. * - mark all memory queues empty
  2781. * - clear the memory bitmaps
  2782. */
  2783. static void __meminit free_area_init_core(struct pglist_data *pgdat,
  2784. unsigned long *zones_size, unsigned long *zholes_size)
  2785. {
  2786. enum zone_type j;
  2787. int nid = pgdat->node_id;
  2788. unsigned long zone_start_pfn = pgdat->node_start_pfn;
  2789. int ret;
  2790. pgdat_resize_init(pgdat);
  2791. pgdat->nr_zones = 0;
  2792. init_waitqueue_head(&pgdat->kswapd_wait);
  2793. pgdat->kswapd_max_order = 0;
  2794. for (j = 0; j < MAX_NR_ZONES; j++) {
  2795. struct zone *zone = pgdat->node_zones + j;
  2796. unsigned long size, realsize, memmap_pages;
  2797. size = zone_spanned_pages_in_node(nid, j, zones_size);
  2798. realsize = size - zone_absent_pages_in_node(nid, j,
  2799. zholes_size);
  2800. /*
  2801. * Adjust realsize so that it accounts for how much memory
  2802. * is used by this zone for memmap. This affects the watermark
  2803. * and per-cpu initialisations
  2804. */
  2805. memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
  2806. if (realsize >= memmap_pages) {
  2807. realsize -= memmap_pages;
  2808. printk(KERN_DEBUG
  2809. " %s zone: %lu pages used for memmap\n",
  2810. zone_names[j], memmap_pages);
  2811. } else
  2812. printk(KERN_WARNING
  2813. " %s zone: %lu pages exceeds realsize %lu\n",
  2814. zone_names[j], memmap_pages, realsize);
  2815. /* Account for reserved pages */
  2816. if (j == 0 && realsize > dma_reserve) {
  2817. realsize -= dma_reserve;
  2818. printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
  2819. zone_names[0], dma_reserve);
  2820. }
  2821. if (!is_highmem_idx(j))
  2822. nr_kernel_pages += realsize;
  2823. nr_all_pages += realsize;
  2824. zone->spanned_pages = size;
  2825. zone->present_pages = realsize;
  2826. #ifdef CONFIG_NUMA
  2827. zone->node = nid;
  2828. zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
  2829. / 100;
  2830. zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
  2831. #endif
  2832. zone->name = zone_names[j];
  2833. spin_lock_init(&zone->lock);
  2834. spin_lock_init(&zone->lru_lock);
  2835. zone_seqlock_init(zone);
  2836. zone->zone_pgdat = pgdat;
  2837. zone->prev_priority = DEF_PRIORITY;
  2838. zone_pcp_init(zone);
  2839. INIT_LIST_HEAD(&zone->active_list);
  2840. INIT_LIST_HEAD(&zone->inactive_list);
  2841. zone->nr_scan_active = 0;
  2842. zone->nr_scan_inactive = 0;
  2843. zap_zone_vm_stats(zone);
  2844. atomic_set(&zone->reclaim_in_progress, 0);
  2845. if (!size)
  2846. continue;
  2847. setup_usemap(pgdat, zone, size);
  2848. ret = init_currently_empty_zone(zone, zone_start_pfn,
  2849. size, MEMMAP_EARLY);
  2850. BUG_ON(ret);
  2851. zone_start_pfn += size;
  2852. }
  2853. }
  2854. static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
  2855. {
  2856. /* Skip empty nodes */
  2857. if (!pgdat->node_spanned_pages)
  2858. return;
  2859. #ifdef CONFIG_FLAT_NODE_MEM_MAP
  2860. /* ia64 gets its own node_mem_map, before this, without bootmem */
  2861. if (!pgdat->node_mem_map) {
  2862. unsigned long size, start, end;
  2863. struct page *map;
  2864. /*
  2865. * The zone's endpoints aren't required to be MAX_ORDER
  2866. * aligned but the node_mem_map endpoints must be in order
  2867. * for the buddy allocator to function correctly.
  2868. */
  2869. start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
  2870. end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
  2871. end = ALIGN(end, MAX_ORDER_NR_PAGES);
  2872. size = (end - start) * sizeof(struct page);
  2873. map = alloc_remap(pgdat->node_id, size);
  2874. if (!map)
  2875. map = alloc_bootmem_node(pgdat, size);
  2876. pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
  2877. }
  2878. #ifndef CONFIG_NEED_MULTIPLE_NODES
  2879. /*
  2880. * With no DISCONTIG, the global mem_map is just set as node 0's
  2881. */
  2882. if (pgdat == NODE_DATA(0)) {
  2883. mem_map = NODE_DATA(0)->node_mem_map;
  2884. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  2885. if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
  2886. mem_map -= pgdat->node_start_pfn;
  2887. #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
  2888. }
  2889. #endif
  2890. #endif /* CONFIG_FLAT_NODE_MEM_MAP */
  2891. }
  2892. void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
  2893. unsigned long *zones_size, unsigned long node_start_pfn,
  2894. unsigned long *zholes_size)
  2895. {
  2896. pgdat->node_id = nid;
  2897. pgdat->node_start_pfn = node_start_pfn;
  2898. calculate_node_totalpages(pgdat, zones_size, zholes_size);
  2899. alloc_node_mem_map(pgdat);
  2900. free_area_init_core(pgdat, zones_size, zholes_size);
  2901. }
  2902. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  2903. #if MAX_NUMNODES > 1
  2904. /*
  2905. * Figure out the number of possible node ids.
  2906. */
  2907. static void __init setup_nr_node_ids(void)
  2908. {
  2909. unsigned int node;
  2910. unsigned int highest = 0;
  2911. for_each_node_mask(node, node_possible_map)
  2912. highest = node;
  2913. nr_node_ids = highest + 1;
  2914. }
  2915. #else
  2916. static inline void setup_nr_node_ids(void)
  2917. {
  2918. }
  2919. #endif
  2920. /**
  2921. * add_active_range - Register a range of PFNs backed by physical memory
  2922. * @nid: The node ID the range resides on
  2923. * @start_pfn: The start PFN of the available physical memory
  2924. * @end_pfn: The end PFN of the available physical memory
  2925. *
  2926. * These ranges are stored in an early_node_map[] and later used by
  2927. * free_area_init_nodes() to calculate zone sizes and holes. If the
  2928. * range spans a memory hole, it is up to the architecture to ensure
  2929. * the memory is not freed by the bootmem allocator. If possible
  2930. * the range being registered will be merged with existing ranges.
  2931. */
  2932. void __init add_active_range(unsigned int nid, unsigned long start_pfn,
  2933. unsigned long end_pfn)
  2934. {
  2935. int i;
  2936. printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
  2937. "%d entries of %d used\n",
  2938. nid, start_pfn, end_pfn,
  2939. nr_nodemap_entries, MAX_ACTIVE_REGIONS);
  2940. /* Merge with existing active regions if possible */
  2941. for (i = 0; i < nr_nodemap_entries; i++) {
  2942. if (early_node_map[i].nid != nid)
  2943. continue;
  2944. /* Skip if an existing region covers this new one */
  2945. if (start_pfn >= early_node_map[i].start_pfn &&
  2946. end_pfn <= early_node_map[i].end_pfn)
  2947. return;
  2948. /* Merge forward if suitable */
  2949. if (start_pfn <= early_node_map[i].end_pfn &&
  2950. end_pfn > early_node_map[i].end_pfn) {
  2951. early_node_map[i].end_pfn = end_pfn;
  2952. return;
  2953. }
  2954. /* Merge backward if suitable */
  2955. if (start_pfn < early_node_map[i].end_pfn &&
  2956. end_pfn >= early_node_map[i].start_pfn) {
  2957. early_node_map[i].start_pfn = start_pfn;
  2958. return;
  2959. }
  2960. }
  2961. /* Check that early_node_map is large enough */
  2962. if (i >= MAX_ACTIVE_REGIONS) {
  2963. printk(KERN_CRIT "More than %d memory regions, truncating\n",
  2964. MAX_ACTIVE_REGIONS);
  2965. return;
  2966. }
  2967. early_node_map[i].nid = nid;
  2968. early_node_map[i].start_pfn = start_pfn;
  2969. early_node_map[i].end_pfn = end_pfn;
  2970. nr_nodemap_entries = i + 1;
  2971. }
  2972. /**
  2973. * shrink_active_range - Shrink an existing registered range of PFNs
  2974. * @nid: The node id the range is on that should be shrunk
  2975. * @old_end_pfn: The old end PFN of the range
  2976. * @new_end_pfn: The new PFN of the range
  2977. *
  2978. * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
  2979. * The map is kept at the end physical page range that has already been
  2980. * registered with add_active_range(). This function allows an arch to shrink
  2981. * an existing registered range.
  2982. */
  2983. void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
  2984. unsigned long new_end_pfn)
  2985. {
  2986. int i;
  2987. /* Find the old active region end and shrink */
  2988. for_each_active_range_index_in_nid(i, nid)
  2989. if (early_node_map[i].end_pfn == old_end_pfn) {
  2990. early_node_map[i].end_pfn = new_end_pfn;
  2991. break;
  2992. }
  2993. }
  2994. /**
  2995. * remove_all_active_ranges - Remove all currently registered regions
  2996. *
  2997. * During discovery, it may be found that a table like SRAT is invalid
  2998. * and an alternative discovery method must be used. This function removes
  2999. * all currently registered regions.
  3000. */
  3001. void __init remove_all_active_ranges(void)
  3002. {
  3003. memset(early_node_map, 0, sizeof(early_node_map));
  3004. nr_nodemap_entries = 0;
  3005. #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
  3006. memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
  3007. memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
  3008. #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
  3009. }
  3010. /* Compare two active node_active_regions */
  3011. static int __init cmp_node_active_region(const void *a, const void *b)
  3012. {
  3013. struct node_active_region *arange = (struct node_active_region *)a;
  3014. struct node_active_region *brange = (struct node_active_region *)b;
  3015. /* Done this way to avoid overflows */
  3016. if (arange->start_pfn > brange->start_pfn)
  3017. return 1;
  3018. if (arange->start_pfn < brange->start_pfn)
  3019. return -1;
  3020. return 0;
  3021. }
  3022. /* sort the node_map by start_pfn */
  3023. static void __init sort_node_map(void)
  3024. {
  3025. sort(early_node_map, (size_t)nr_nodemap_entries,
  3026. sizeof(struct node_active_region),
  3027. cmp_node_active_region, NULL);
  3028. }
  3029. /* Find the lowest pfn for a node */
  3030. unsigned long __init find_min_pfn_for_node(unsigned long nid)
  3031. {
  3032. int i;
  3033. unsigned long min_pfn = ULONG_MAX;
  3034. /* Assuming a sorted map, the first range found has the starting pfn */
  3035. for_each_active_range_index_in_nid(i, nid)
  3036. min_pfn = min(min_pfn, early_node_map[i].start_pfn);
  3037. if (min_pfn == ULONG_MAX) {
  3038. printk(KERN_WARNING
  3039. "Could not find start_pfn for node %lu\n", nid);
  3040. return 0;
  3041. }
  3042. return min_pfn;
  3043. }
  3044. /**
  3045. * find_min_pfn_with_active_regions - Find the minimum PFN registered
  3046. *
  3047. * It returns the minimum PFN based on information provided via
  3048. * add_active_range().
  3049. */
  3050. unsigned long __init find_min_pfn_with_active_regions(void)
  3051. {
  3052. return find_min_pfn_for_node(MAX_NUMNODES);
  3053. }
  3054. /**
  3055. * find_max_pfn_with_active_regions - Find the maximum PFN registered
  3056. *
  3057. * It returns the maximum PFN based on information provided via
  3058. * add_active_range().
  3059. */
  3060. unsigned long __init find_max_pfn_with_active_regions(void)
  3061. {
  3062. int i;
  3063. unsigned long max_pfn = 0;
  3064. for (i = 0; i < nr_nodemap_entries; i++)
  3065. max_pfn = max(max_pfn, early_node_map[i].end_pfn);
  3066. return max_pfn;
  3067. }
  3068. /*
  3069. * early_calculate_totalpages()
  3070. * Sum pages in active regions for movable zone.
  3071. * Populate N_HIGH_MEMORY for calculating usable_nodes.
  3072. */
  3073. unsigned long __init early_calculate_totalpages(void)
  3074. {
  3075. int i;
  3076. unsigned long totalpages = 0;
  3077. for (i = 0; i < nr_nodemap_entries; i++) {
  3078. unsigned long pages = early_node_map[i].end_pfn -
  3079. early_node_map[i].start_pfn;
  3080. totalpages += pages;
  3081. if (pages)
  3082. node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
  3083. }
  3084. return totalpages;
  3085. }
  3086. /*
  3087. * Find the PFN the Movable zone begins in each node. Kernel memory
  3088. * is spread evenly between nodes as long as the nodes have enough
  3089. * memory. When they don't, some nodes will have more kernelcore than
  3090. * others
  3091. */
  3092. void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
  3093. {
  3094. int i, nid;
  3095. unsigned long usable_startpfn;
  3096. unsigned long kernelcore_node, kernelcore_remaining;
  3097. unsigned long totalpages = early_calculate_totalpages();
  3098. int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
  3099. /*
  3100. * If movablecore was specified, calculate what size of
  3101. * kernelcore that corresponds so that memory usable for
  3102. * any allocation type is evenly spread. If both kernelcore
  3103. * and movablecore are specified, then the value of kernelcore
  3104. * will be used for required_kernelcore if it's greater than
  3105. * what movablecore would have allowed.
  3106. */
  3107. if (required_movablecore) {
  3108. unsigned long corepages;
  3109. /*
  3110. * Round-up so that ZONE_MOVABLE is at least as large as what
  3111. * was requested by the user
  3112. */
  3113. required_movablecore =
  3114. roundup(required_movablecore, MAX_ORDER_NR_PAGES);
  3115. corepages = totalpages - required_movablecore;
  3116. required_kernelcore = max(required_kernelcore, corepages);
  3117. }
  3118. /* If kernelcore was not specified, there is no ZONE_MOVABLE */
  3119. if (!required_kernelcore)
  3120. return;
  3121. /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
  3122. find_usable_zone_for_movable();
  3123. usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
  3124. restart:
  3125. /* Spread kernelcore memory as evenly as possible throughout nodes */
  3126. kernelcore_node = required_kernelcore / usable_nodes;
  3127. for_each_node_state(nid, N_HIGH_MEMORY) {
  3128. /*
  3129. * Recalculate kernelcore_node if the division per node
  3130. * now exceeds what is necessary to satisfy the requested
  3131. * amount of memory for the kernel
  3132. */
  3133. if (required_kernelcore < kernelcore_node)
  3134. kernelcore_node = required_kernelcore / usable_nodes;
  3135. /*
  3136. * As the map is walked, we track how much memory is usable
  3137. * by the kernel using kernelcore_remaining. When it is
  3138. * 0, the rest of the node is usable by ZONE_MOVABLE
  3139. */
  3140. kernelcore_remaining = kernelcore_node;
  3141. /* Go through each range of PFNs within this node */
  3142. for_each_active_range_index_in_nid(i, nid) {
  3143. unsigned long start_pfn, end_pfn;
  3144. unsigned long size_pages;
  3145. start_pfn = max(early_node_map[i].start_pfn,
  3146. zone_movable_pfn[nid]);
  3147. end_pfn = early_node_map[i].end_pfn;
  3148. if (start_pfn >= end_pfn)
  3149. continue;
  3150. /* Account for what is only usable for kernelcore */
  3151. if (start_pfn < usable_startpfn) {
  3152. unsigned long kernel_pages;
  3153. kernel_pages = min(end_pfn, usable_startpfn)
  3154. - start_pfn;
  3155. kernelcore_remaining -= min(kernel_pages,
  3156. kernelcore_remaining);
  3157. required_kernelcore -= min(kernel_pages,
  3158. required_kernelcore);
  3159. /* Continue if range is now fully accounted */
  3160. if (end_pfn <= usable_startpfn) {
  3161. /*
  3162. * Push zone_movable_pfn to the end so
  3163. * that if we have to rebalance
  3164. * kernelcore across nodes, we will
  3165. * not double account here
  3166. */
  3167. zone_movable_pfn[nid] = end_pfn;
  3168. continue;
  3169. }
  3170. start_pfn = usable_startpfn;
  3171. }
  3172. /*
  3173. * The usable PFN range for ZONE_MOVABLE is from
  3174. * start_pfn->end_pfn. Calculate size_pages as the
  3175. * number of pages used as kernelcore
  3176. */
  3177. size_pages = end_pfn - start_pfn;
  3178. if (size_pages > kernelcore_remaining)
  3179. size_pages = kernelcore_remaining;
  3180. zone_movable_pfn[nid] = start_pfn + size_pages;
  3181. /*
  3182. * Some kernelcore has been met, update counts and
  3183. * break if the kernelcore for this node has been
  3184. * satisified
  3185. */
  3186. required_kernelcore -= min(required_kernelcore,
  3187. size_pages);
  3188. kernelcore_remaining -= size_pages;
  3189. if (!kernelcore_remaining)
  3190. break;
  3191. }
  3192. }
  3193. /*
  3194. * If there is still required_kernelcore, we do another pass with one
  3195. * less node in the count. This will push zone_movable_pfn[nid] further
  3196. * along on the nodes that still have memory until kernelcore is
  3197. * satisified
  3198. */
  3199. usable_nodes--;
  3200. if (usable_nodes && required_kernelcore > usable_nodes)
  3201. goto restart;
  3202. /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
  3203. for (nid = 0; nid < MAX_NUMNODES; nid++)
  3204. zone_movable_pfn[nid] =
  3205. roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
  3206. }
  3207. /* Any regular memory on that node ? */
  3208. static void check_for_regular_memory(pg_data_t *pgdat)
  3209. {
  3210. #ifdef CONFIG_HIGHMEM
  3211. enum zone_type zone_type;
  3212. for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
  3213. struct zone *zone = &pgdat->node_zones[zone_type];
  3214. if (zone->present_pages)
  3215. node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
  3216. }
  3217. #endif
  3218. }
  3219. /**
  3220. * free_area_init_nodes - Initialise all pg_data_t and zone data
  3221. * @max_zone_pfn: an array of max PFNs for each zone
  3222. *
  3223. * This will call free_area_init_node() for each active node in the system.
  3224. * Using the page ranges provided by add_active_range(), the size of each
  3225. * zone in each node and their holes is calculated. If the maximum PFN
  3226. * between two adjacent zones match, it is assumed that the zone is empty.
  3227. * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
  3228. * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
  3229. * starts where the previous one ended. For example, ZONE_DMA32 starts
  3230. * at arch_max_dma_pfn.
  3231. */
  3232. void __init free_area_init_nodes(unsigned long *max_zone_pfn)
  3233. {
  3234. unsigned long nid;
  3235. enum zone_type i;
  3236. /* Sort early_node_map as initialisation assumes it is sorted */
  3237. sort_node_map();
  3238. /* Record where the zone boundaries are */
  3239. memset(arch_zone_lowest_possible_pfn, 0,
  3240. sizeof(arch_zone_lowest_possible_pfn));
  3241. memset(arch_zone_highest_possible_pfn, 0,
  3242. sizeof(arch_zone_highest_possible_pfn));
  3243. arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
  3244. arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
  3245. for (i = 1; i < MAX_NR_ZONES; i++) {
  3246. if (i == ZONE_MOVABLE)
  3247. continue;
  3248. arch_zone_lowest_possible_pfn[i] =
  3249. arch_zone_highest_possible_pfn[i-1];
  3250. arch_zone_highest_possible_pfn[i] =
  3251. max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
  3252. }
  3253. arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
  3254. arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
  3255. /* Find the PFNs that ZONE_MOVABLE begins at in each node */
  3256. memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
  3257. find_zone_movable_pfns_for_nodes(zone_movable_pfn);
  3258. /* Print out the zone ranges */
  3259. printk("Zone PFN ranges:\n");
  3260. for (i = 0; i < MAX_NR_ZONES; i++) {
  3261. if (i == ZONE_MOVABLE)
  3262. continue;
  3263. printk(" %-8s %8lu -> %8lu\n",
  3264. zone_names[i],
  3265. arch_zone_lowest_possible_pfn[i],
  3266. arch_zone_highest_possible_pfn[i]);
  3267. }
  3268. /* Print out the PFNs ZONE_MOVABLE begins at in each node */
  3269. printk("Movable zone start PFN for each node\n");
  3270. for (i = 0; i < MAX_NUMNODES; i++) {
  3271. if (zone_movable_pfn[i])
  3272. printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
  3273. }
  3274. /* Print out the early_node_map[] */
  3275. printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
  3276. for (i = 0; i < nr_nodemap_entries; i++)
  3277. printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
  3278. early_node_map[i].start_pfn,
  3279. early_node_map[i].end_pfn);
  3280. /* Initialise every node */
  3281. setup_nr_node_ids();
  3282. for_each_online_node(nid) {
  3283. pg_data_t *pgdat = NODE_DATA(nid);
  3284. free_area_init_node(nid, pgdat, NULL,
  3285. find_min_pfn_for_node(nid), NULL);
  3286. /* Any memory on that node */
  3287. if (pgdat->node_present_pages)
  3288. node_set_state(nid, N_HIGH_MEMORY);
  3289. check_for_regular_memory(pgdat);
  3290. }
  3291. }
  3292. static int __init cmdline_parse_core(char *p, unsigned long *core)
  3293. {
  3294. unsigned long long coremem;
  3295. if (!p)
  3296. return -EINVAL;
  3297. coremem = memparse(p, &p);
  3298. *core = coremem >> PAGE_SHIFT;
  3299. /* Paranoid check that UL is enough for the coremem value */
  3300. WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
  3301. return 0;
  3302. }
  3303. /*
  3304. * kernelcore=size sets the amount of memory for use for allocations that
  3305. * cannot be reclaimed or migrated.
  3306. */
  3307. static int __init cmdline_parse_kernelcore(char *p)
  3308. {
  3309. return cmdline_parse_core(p, &required_kernelcore);
  3310. }
  3311. /*
  3312. * movablecore=size sets the amount of memory for use for allocations that
  3313. * can be reclaimed or migrated.
  3314. */
  3315. static int __init cmdline_parse_movablecore(char *p)
  3316. {
  3317. return cmdline_parse_core(p, &required_movablecore);
  3318. }
  3319. early_param("kernelcore", cmdline_parse_kernelcore);
  3320. early_param("movablecore", cmdline_parse_movablecore);
  3321. #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
  3322. /**
  3323. * set_dma_reserve - set the specified number of pages reserved in the first zone
  3324. * @new_dma_reserve: The number of pages to mark reserved
  3325. *
  3326. * The per-cpu batchsize and zone watermarks are determined by present_pages.
  3327. * In the DMA zone, a significant percentage may be consumed by kernel image
  3328. * and other unfreeable allocations which can skew the watermarks badly. This
  3329. * function may optionally be used to account for unfreeable pages in the
  3330. * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
  3331. * smaller per-cpu batchsize.
  3332. */
  3333. void __init set_dma_reserve(unsigned long new_dma_reserve)
  3334. {
  3335. dma_reserve = new_dma_reserve;
  3336. }
  3337. #ifndef CONFIG_NEED_MULTIPLE_NODES
  3338. static bootmem_data_t contig_bootmem_data;
  3339. struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
  3340. EXPORT_SYMBOL(contig_page_data);
  3341. #endif
  3342. void __init free_area_init(unsigned long *zones_size)
  3343. {
  3344. free_area_init_node(0, NODE_DATA(0), zones_size,
  3345. __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
  3346. }
  3347. static int page_alloc_cpu_notify(struct notifier_block *self,
  3348. unsigned long action, void *hcpu)
  3349. {
  3350. int cpu = (unsigned long)hcpu;
  3351. if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
  3352. local_irq_disable();
  3353. __drain_pages(cpu);
  3354. vm_events_fold_cpu(cpu);
  3355. local_irq_enable();
  3356. refresh_cpu_vm_stats(cpu);
  3357. }
  3358. return NOTIFY_OK;
  3359. }
  3360. void __init page_alloc_init(void)
  3361. {
  3362. hotcpu_notifier(page_alloc_cpu_notify, 0);
  3363. }
  3364. /*
  3365. * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
  3366. * or min_free_kbytes changes.
  3367. */
  3368. static void calculate_totalreserve_pages(void)
  3369. {
  3370. struct pglist_data *pgdat;
  3371. unsigned long reserve_pages = 0;
  3372. enum zone_type i, j;
  3373. for_each_online_pgdat(pgdat) {
  3374. for (i = 0; i < MAX_NR_ZONES; i++) {
  3375. struct zone *zone = pgdat->node_zones + i;
  3376. unsigned long max = 0;
  3377. /* Find valid and maximum lowmem_reserve in the zone */
  3378. for (j = i; j < MAX_NR_ZONES; j++) {
  3379. if (zone->lowmem_reserve[j] > max)
  3380. max = zone->lowmem_reserve[j];
  3381. }
  3382. /* we treat pages_high as reserved pages. */
  3383. max += zone->pages_high;
  3384. if (max > zone->present_pages)
  3385. max = zone->present_pages;
  3386. reserve_pages += max;
  3387. }
  3388. }
  3389. totalreserve_pages = reserve_pages;
  3390. }
  3391. /*
  3392. * setup_per_zone_lowmem_reserve - called whenever
  3393. * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
  3394. * has a correct pages reserved value, so an adequate number of
  3395. * pages are left in the zone after a successful __alloc_pages().
  3396. */
  3397. static void setup_per_zone_lowmem_reserve(void)
  3398. {
  3399. struct pglist_data *pgdat;
  3400. enum zone_type j, idx;
  3401. for_each_online_pgdat(pgdat) {
  3402. for (j = 0; j < MAX_NR_ZONES; j++) {
  3403. struct zone *zone = pgdat->node_zones + j;
  3404. unsigned long present_pages = zone->present_pages;
  3405. zone->lowmem_reserve[j] = 0;
  3406. idx = j;
  3407. while (idx) {
  3408. struct zone *lower_zone;
  3409. idx--;
  3410. if (sysctl_lowmem_reserve_ratio[idx] < 1)
  3411. sysctl_lowmem_reserve_ratio[idx] = 1;
  3412. lower_zone = pgdat->node_zones + idx;
  3413. lower_zone->lowmem_reserve[j] = present_pages /
  3414. sysctl_lowmem_reserve_ratio[idx];
  3415. present_pages += lower_zone->present_pages;
  3416. }
  3417. }
  3418. }
  3419. /* update totalreserve_pages */
  3420. calculate_totalreserve_pages();
  3421. }
  3422. /**
  3423. * setup_per_zone_pages_min - called when min_free_kbytes changes.
  3424. *
  3425. * Ensures that the pages_{min,low,high} values for each zone are set correctly
  3426. * with respect to min_free_kbytes.
  3427. */
  3428. void setup_per_zone_pages_min(void)
  3429. {
  3430. unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
  3431. unsigned long lowmem_pages = 0;
  3432. struct zone *zone;
  3433. unsigned long flags;
  3434. /* Calculate total number of !ZONE_HIGHMEM pages */
  3435. for_each_zone(zone) {
  3436. if (!is_highmem(zone))
  3437. lowmem_pages += zone->present_pages;
  3438. }
  3439. for_each_zone(zone) {
  3440. u64 tmp;
  3441. spin_lock_irqsave(&zone->lru_lock, flags);
  3442. tmp = (u64)pages_min * zone->present_pages;
  3443. do_div(tmp, lowmem_pages);
  3444. if (is_highmem(zone)) {
  3445. /*
  3446. * __GFP_HIGH and PF_MEMALLOC allocations usually don't
  3447. * need highmem pages, so cap pages_min to a small
  3448. * value here.
  3449. *
  3450. * The (pages_high-pages_low) and (pages_low-pages_min)
  3451. * deltas controls asynch page reclaim, and so should
  3452. * not be capped for highmem.
  3453. */
  3454. int min_pages;
  3455. min_pages = zone->present_pages / 1024;
  3456. if (min_pages < SWAP_CLUSTER_MAX)
  3457. min_pages = SWAP_CLUSTER_MAX;
  3458. if (min_pages > 128)
  3459. min_pages = 128;
  3460. zone->pages_min = min_pages;
  3461. } else {
  3462. /*
  3463. * If it's a lowmem zone, reserve a number of pages
  3464. * proportionate to the zone's size.
  3465. */
  3466. zone->pages_min = tmp;
  3467. }
  3468. zone->pages_low = zone->pages_min + (tmp >> 2);
  3469. zone->pages_high = zone->pages_min + (tmp >> 1);
  3470. spin_unlock_irqrestore(&zone->lru_lock, flags);
  3471. }
  3472. /* update totalreserve_pages */
  3473. calculate_totalreserve_pages();
  3474. }
  3475. /*
  3476. * Initialise min_free_kbytes.
  3477. *
  3478. * For small machines we want it small (128k min). For large machines
  3479. * we want it large (64MB max). But it is not linear, because network
  3480. * bandwidth does not increase linearly with machine size. We use
  3481. *
  3482. * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
  3483. * min_free_kbytes = sqrt(lowmem_kbytes * 16)
  3484. *
  3485. * which yields
  3486. *
  3487. * 16MB: 512k
  3488. * 32MB: 724k
  3489. * 64MB: 1024k
  3490. * 128MB: 1448k
  3491. * 256MB: 2048k
  3492. * 512MB: 2896k
  3493. * 1024MB: 4096k
  3494. * 2048MB: 5792k
  3495. * 4096MB: 8192k
  3496. * 8192MB: 11584k
  3497. * 16384MB: 16384k
  3498. */
  3499. static int __init init_per_zone_pages_min(void)
  3500. {
  3501. unsigned long lowmem_kbytes;
  3502. lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
  3503. min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
  3504. if (min_free_kbytes < 128)
  3505. min_free_kbytes = 128;
  3506. if (min_free_kbytes > 65536)
  3507. min_free_kbytes = 65536;
  3508. setup_per_zone_pages_min();
  3509. setup_per_zone_lowmem_reserve();
  3510. return 0;
  3511. }
  3512. module_init(init_per_zone_pages_min)
  3513. /*
  3514. * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
  3515. * that we can call two helper functions whenever min_free_kbytes
  3516. * changes.
  3517. */
  3518. int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
  3519. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3520. {
  3521. proc_dointvec(table, write, file, buffer, length, ppos);
  3522. if (write)
  3523. setup_per_zone_pages_min();
  3524. return 0;
  3525. }
  3526. #ifdef CONFIG_NUMA
  3527. int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
  3528. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3529. {
  3530. struct zone *zone;
  3531. int rc;
  3532. rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  3533. if (rc)
  3534. return rc;
  3535. for_each_zone(zone)
  3536. zone->min_unmapped_pages = (zone->present_pages *
  3537. sysctl_min_unmapped_ratio) / 100;
  3538. return 0;
  3539. }
  3540. int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
  3541. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3542. {
  3543. struct zone *zone;
  3544. int rc;
  3545. rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  3546. if (rc)
  3547. return rc;
  3548. for_each_zone(zone)
  3549. zone->min_slab_pages = (zone->present_pages *
  3550. sysctl_min_slab_ratio) / 100;
  3551. return 0;
  3552. }
  3553. #endif
  3554. /*
  3555. * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
  3556. * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
  3557. * whenever sysctl_lowmem_reserve_ratio changes.
  3558. *
  3559. * The reserve ratio obviously has absolutely no relation with the
  3560. * pages_min watermarks. The lowmem reserve ratio can only make sense
  3561. * if in function of the boot time zone sizes.
  3562. */
  3563. int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
  3564. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3565. {
  3566. proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  3567. setup_per_zone_lowmem_reserve();
  3568. return 0;
  3569. }
  3570. /*
  3571. * percpu_pagelist_fraction - changes the pcp->high for each zone on each
  3572. * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
  3573. * can have before it gets flushed back to buddy allocator.
  3574. */
  3575. int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
  3576. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3577. {
  3578. struct zone *zone;
  3579. unsigned int cpu;
  3580. int ret;
  3581. ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  3582. if (!write || (ret == -EINVAL))
  3583. return ret;
  3584. for_each_zone(zone) {
  3585. for_each_online_cpu(cpu) {
  3586. unsigned long high;
  3587. high = zone->present_pages / percpu_pagelist_fraction;
  3588. setup_pagelist_highmark(zone_pcp(zone, cpu), high);
  3589. }
  3590. }
  3591. return 0;
  3592. }
  3593. int hashdist = HASHDIST_DEFAULT;
  3594. #ifdef CONFIG_NUMA
  3595. static int __init set_hashdist(char *str)
  3596. {
  3597. if (!str)
  3598. return 0;
  3599. hashdist = simple_strtoul(str, &str, 0);
  3600. return 1;
  3601. }
  3602. __setup("hashdist=", set_hashdist);
  3603. #endif
  3604. /*
  3605. * allocate a large system hash table from bootmem
  3606. * - it is assumed that the hash table must contain an exact power-of-2
  3607. * quantity of entries
  3608. * - limit is the number of hash buckets, not the total allocation size
  3609. */
  3610. void *__init alloc_large_system_hash(const char *tablename,
  3611. unsigned long bucketsize,
  3612. unsigned long numentries,
  3613. int scale,
  3614. int flags,
  3615. unsigned int *_hash_shift,
  3616. unsigned int *_hash_mask,
  3617. unsigned long limit)
  3618. {
  3619. unsigned long long max = limit;
  3620. unsigned long log2qty, size;
  3621. void *table = NULL;
  3622. /* allow the kernel cmdline to have a say */
  3623. if (!numentries) {
  3624. /* round applicable memory size up to nearest megabyte */
  3625. numentries = nr_kernel_pages;
  3626. numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
  3627. numentries >>= 20 - PAGE_SHIFT;
  3628. numentries <<= 20 - PAGE_SHIFT;
  3629. /* limit to 1 bucket per 2^scale bytes of low memory */
  3630. if (scale > PAGE_SHIFT)
  3631. numentries >>= (scale - PAGE_SHIFT);
  3632. else
  3633. numentries <<= (PAGE_SHIFT - scale);
  3634. /* Make sure we've got at least a 0-order allocation.. */
  3635. if (unlikely((numentries * bucketsize) < PAGE_SIZE))
  3636. numentries = PAGE_SIZE / bucketsize;
  3637. }
  3638. numentries = roundup_pow_of_two(numentries);
  3639. /* limit allocation size to 1/16 total memory by default */
  3640. if (max == 0) {
  3641. max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
  3642. do_div(max, bucketsize);
  3643. }
  3644. if (numentries > max)
  3645. numentries = max;
  3646. log2qty = ilog2(numentries);
  3647. do {
  3648. size = bucketsize << log2qty;
  3649. if (flags & HASH_EARLY)
  3650. table = alloc_bootmem(size);
  3651. else if (hashdist)
  3652. table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
  3653. else {
  3654. unsigned long order;
  3655. for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
  3656. ;
  3657. table = (void*) __get_free_pages(GFP_ATOMIC, order);
  3658. /*
  3659. * If bucketsize is not a power-of-two, we may free
  3660. * some pages at the end of hash table.
  3661. */
  3662. if (table) {
  3663. unsigned long alloc_end = (unsigned long)table +
  3664. (PAGE_SIZE << order);
  3665. unsigned long used = (unsigned long)table +
  3666. PAGE_ALIGN(size);
  3667. split_page(virt_to_page(table), order);
  3668. while (used < alloc_end) {
  3669. free_page(used);
  3670. used += PAGE_SIZE;
  3671. }
  3672. }
  3673. }
  3674. } while (!table && size > PAGE_SIZE && --log2qty);
  3675. if (!table)
  3676. panic("Failed to allocate %s hash table\n", tablename);
  3677. printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
  3678. tablename,
  3679. (1U << log2qty),
  3680. ilog2(size) - PAGE_SHIFT,
  3681. size);
  3682. if (_hash_shift)
  3683. *_hash_shift = log2qty;
  3684. if (_hash_mask)
  3685. *_hash_mask = (1 << log2qty) - 1;
  3686. return table;
  3687. }
  3688. #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
  3689. struct page *pfn_to_page(unsigned long pfn)
  3690. {
  3691. return __pfn_to_page(pfn);
  3692. }
  3693. unsigned long page_to_pfn(struct page *page)
  3694. {
  3695. return __page_to_pfn(page);
  3696. }
  3697. EXPORT_SYMBOL(pfn_to_page);
  3698. EXPORT_SYMBOL(page_to_pfn);
  3699. #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
  3700. /* Return a pointer to the bitmap storing bits affecting a block of pages */
  3701. static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
  3702. unsigned long pfn)
  3703. {
  3704. #ifdef CONFIG_SPARSEMEM
  3705. return __pfn_to_section(pfn)->pageblock_flags;
  3706. #else
  3707. return zone->pageblock_flags;
  3708. #endif /* CONFIG_SPARSEMEM */
  3709. }
  3710. static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
  3711. {
  3712. #ifdef CONFIG_SPARSEMEM
  3713. pfn &= (PAGES_PER_SECTION-1);
  3714. return (pfn >> (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS;
  3715. #else
  3716. pfn = pfn - zone->zone_start_pfn;
  3717. return (pfn >> (MAX_ORDER-1)) * NR_PAGEBLOCK_BITS;
  3718. #endif /* CONFIG_SPARSEMEM */
  3719. }
  3720. /**
  3721. * get_pageblock_flags_group - Return the requested group of flags for the MAX_ORDER_NR_PAGES block of pages
  3722. * @page: The page within the block of interest
  3723. * @start_bitidx: The first bit of interest to retrieve
  3724. * @end_bitidx: The last bit of interest
  3725. * returns pageblock_bits flags
  3726. */
  3727. unsigned long get_pageblock_flags_group(struct page *page,
  3728. int start_bitidx, int end_bitidx)
  3729. {
  3730. struct zone *zone;
  3731. unsigned long *bitmap;
  3732. unsigned long pfn, bitidx;
  3733. unsigned long flags = 0;
  3734. unsigned long value = 1;
  3735. zone = page_zone(page);
  3736. pfn = page_to_pfn(page);
  3737. bitmap = get_pageblock_bitmap(zone, pfn);
  3738. bitidx = pfn_to_bitidx(zone, pfn);
  3739. for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
  3740. if (test_bit(bitidx + start_bitidx, bitmap))
  3741. flags |= value;
  3742. return flags;
  3743. }
  3744. /**
  3745. * set_pageblock_flags_group - Set the requested group of flags for a MAX_ORDER_NR_PAGES block of pages
  3746. * @page: The page within the block of interest
  3747. * @start_bitidx: The first bit of interest
  3748. * @end_bitidx: The last bit of interest
  3749. * @flags: The flags to set
  3750. */
  3751. void set_pageblock_flags_group(struct page *page, unsigned long flags,
  3752. int start_bitidx, int end_bitidx)
  3753. {
  3754. struct zone *zone;
  3755. unsigned long *bitmap;
  3756. unsigned long pfn, bitidx;
  3757. unsigned long value = 1;
  3758. zone = page_zone(page);
  3759. pfn = page_to_pfn(page);
  3760. bitmap = get_pageblock_bitmap(zone, pfn);
  3761. bitidx = pfn_to_bitidx(zone, pfn);
  3762. for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
  3763. if (flags & value)
  3764. __set_bit(bitidx + start_bitidx, bitmap);
  3765. else
  3766. __clear_bit(bitidx + start_bitidx, bitmap);
  3767. }