page_alloc.c 135 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019202020212022202320242025202620272028202920302031203220332034203520362037203820392040204120422043204420452046204720482049205020512052205320542055205620572058205920602061206220632064206520662067206820692070207120722073207420752076207720782079208020812082208320842085208620872088208920902091209220932094209520962097209820992100210121022103210421052106210721082109211021112112211321142115211621172118211921202121212221232124212521262127212821292130213121322133213421352136213721382139214021412142214321442145214621472148214921502151215221532154215521562157215821592160216121622163216421652166216721682169217021712172217321742175217621772178217921802181218221832184218521862187218821892190219121922193219421952196219721982199220022012202220322042205220622072208220922102211221222132214221522162217221822192220222122222223222422252226222722282229223022312232223322342235223622372238223922402241224222432244224522462247224822492250225122522253225422552256225722582259226022612262226322642265226622672268226922702271227222732274227522762277227822792280228122822283228422852286228722882289229022912292229322942295229622972298229923002301230223032304230523062307230823092310231123122313231423152316231723182319232023212322232323242325232623272328232923302331233223332334233523362337233823392340234123422343234423452346234723482349235023512352235323542355235623572358235923602361236223632364236523662367236823692370237123722373237423752376237723782379238023812382238323842385238623872388238923902391239223932394239523962397239823992400240124022403240424052406240724082409241024112412241324142415241624172418241924202421242224232424242524262427242824292430243124322433243424352436243724382439244024412442244324442445244624472448244924502451245224532454245524562457245824592460246124622463246424652466246724682469247024712472247324742475247624772478247924802481248224832484248524862487248824892490249124922493249424952496249724982499250025012502250325042505250625072508250925102511251225132514251525162517251825192520252125222523252425252526252725282529253025312532253325342535253625372538253925402541254225432544254525462547254825492550255125522553255425552556255725582559256025612562256325642565256625672568256925702571257225732574257525762577257825792580258125822583258425852586258725882589259025912592259325942595259625972598259926002601260226032604260526062607260826092610261126122613261426152616261726182619262026212622262326242625262626272628262926302631263226332634263526362637263826392640264126422643264426452646264726482649265026512652265326542655265626572658265926602661266226632664266526662667266826692670267126722673267426752676267726782679268026812682268326842685268626872688268926902691269226932694269526962697269826992700270127022703270427052706270727082709271027112712271327142715271627172718271927202721272227232724272527262727272827292730273127322733273427352736273727382739274027412742274327442745274627472748274927502751275227532754275527562757275827592760276127622763276427652766276727682769277027712772277327742775277627772778277927802781278227832784278527862787278827892790279127922793279427952796279727982799280028012802280328042805280628072808280928102811281228132814281528162817281828192820282128222823282428252826282728282829283028312832283328342835283628372838283928402841284228432844284528462847284828492850285128522853285428552856285728582859286028612862286328642865286628672868286928702871287228732874287528762877287828792880288128822883288428852886288728882889289028912892289328942895289628972898289929002901290229032904290529062907290829092910291129122913291429152916291729182919292029212922292329242925292629272928292929302931293229332934293529362937293829392940294129422943294429452946294729482949295029512952295329542955295629572958295929602961296229632964296529662967296829692970297129722973297429752976297729782979298029812982298329842985298629872988298929902991299229932994299529962997299829993000300130023003300430053006300730083009301030113012301330143015301630173018301930203021302230233024302530263027302830293030303130323033303430353036303730383039304030413042304330443045304630473048304930503051305230533054305530563057305830593060306130623063306430653066306730683069307030713072307330743075307630773078307930803081308230833084308530863087308830893090309130923093309430953096309730983099310031013102310331043105310631073108310931103111311231133114311531163117311831193120312131223123312431253126312731283129313031313132313331343135313631373138313931403141314231433144314531463147314831493150315131523153315431553156315731583159316031613162316331643165316631673168316931703171317231733174317531763177317831793180318131823183318431853186318731883189319031913192319331943195319631973198319932003201320232033204320532063207320832093210321132123213321432153216321732183219322032213222322332243225322632273228322932303231323232333234323532363237323832393240324132423243324432453246324732483249325032513252325332543255325632573258325932603261326232633264326532663267326832693270327132723273327432753276327732783279328032813282328332843285328632873288328932903291329232933294329532963297329832993300330133023303330433053306330733083309331033113312331333143315331633173318331933203321332233233324332533263327332833293330333133323333333433353336333733383339334033413342334333443345334633473348334933503351335233533354335533563357335833593360336133623363336433653366336733683369337033713372337333743375337633773378337933803381338233833384338533863387338833893390339133923393339433953396339733983399340034013402340334043405340634073408340934103411341234133414341534163417341834193420342134223423342434253426342734283429343034313432343334343435343634373438343934403441344234433444344534463447344834493450345134523453345434553456345734583459346034613462346334643465346634673468346934703471347234733474347534763477347834793480348134823483348434853486348734883489349034913492349334943495349634973498349935003501350235033504350535063507350835093510351135123513351435153516351735183519352035213522352335243525352635273528352935303531353235333534353535363537353835393540354135423543354435453546354735483549355035513552355335543555355635573558355935603561356235633564356535663567356835693570357135723573357435753576357735783579358035813582358335843585358635873588358935903591359235933594359535963597359835993600360136023603360436053606360736083609361036113612361336143615361636173618361936203621362236233624362536263627362836293630363136323633363436353636363736383639364036413642364336443645364636473648364936503651365236533654365536563657365836593660366136623663366436653666366736683669367036713672367336743675367636773678367936803681368236833684368536863687368836893690369136923693369436953696369736983699370037013702370337043705370637073708370937103711371237133714371537163717371837193720372137223723372437253726372737283729373037313732373337343735373637373738373937403741374237433744374537463747374837493750375137523753375437553756375737583759376037613762376337643765376637673768376937703771377237733774377537763777377837793780378137823783378437853786378737883789379037913792379337943795379637973798379938003801380238033804380538063807380838093810381138123813381438153816381738183819382038213822382338243825382638273828382938303831383238333834383538363837383838393840384138423843384438453846384738483849385038513852385338543855385638573858385938603861386238633864386538663867386838693870387138723873387438753876387738783879388038813882388338843885388638873888388938903891389238933894389538963897389838993900390139023903390439053906390739083909391039113912391339143915391639173918391939203921392239233924392539263927392839293930393139323933393439353936393739383939394039413942394339443945394639473948394939503951395239533954395539563957395839593960396139623963396439653966396739683969397039713972397339743975397639773978397939803981398239833984398539863987398839893990399139923993399439953996399739983999400040014002400340044005400640074008400940104011401240134014401540164017401840194020402140224023402440254026402740284029403040314032403340344035403640374038403940404041404240434044404540464047404840494050405140524053405440554056405740584059406040614062406340644065406640674068406940704071407240734074407540764077407840794080408140824083408440854086408740884089409040914092409340944095409640974098409941004101410241034104410541064107410841094110411141124113411441154116411741184119412041214122412341244125412641274128412941304131413241334134413541364137413841394140414141424143414441454146414741484149415041514152415341544155415641574158415941604161416241634164416541664167416841694170417141724173417441754176417741784179418041814182418341844185418641874188418941904191419241934194419541964197419841994200420142024203420442054206420742084209421042114212421342144215421642174218421942204221422242234224422542264227422842294230423142324233423442354236423742384239424042414242424342444245424642474248424942504251425242534254425542564257425842594260426142624263426442654266426742684269427042714272427342744275427642774278427942804281428242834284428542864287428842894290429142924293429442954296429742984299430043014302430343044305430643074308430943104311431243134314431543164317431843194320432143224323432443254326432743284329433043314332433343344335433643374338433943404341434243434344434543464347434843494350435143524353435443554356435743584359436043614362436343644365436643674368436943704371437243734374437543764377437843794380438143824383438443854386438743884389439043914392439343944395439643974398439944004401440244034404440544064407440844094410441144124413441444154416441744184419442044214422442344244425442644274428442944304431443244334434443544364437443844394440444144424443444444454446444744484449445044514452445344544455445644574458445944604461446244634464446544664467446844694470447144724473447444754476447744784479448044814482448344844485448644874488448944904491449244934494449544964497449844994500450145024503450445054506450745084509451045114512451345144515451645174518451945204521452245234524452545264527452845294530453145324533453445354536453745384539454045414542454345444545454645474548454945504551455245534554455545564557455845594560456145624563456445654566456745684569457045714572457345744575457645774578457945804581458245834584458545864587458845894590459145924593459445954596459745984599460046014602460346044605460646074608460946104611461246134614461546164617461846194620462146224623462446254626462746284629463046314632463346344635463646374638463946404641464246434644464546464647464846494650465146524653465446554656465746584659466046614662466346644665466646674668466946704671467246734674467546764677467846794680468146824683468446854686468746884689469046914692469346944695469646974698469947004701470247034704470547064707470847094710471147124713471447154716471747184719472047214722472347244725472647274728472947304731473247334734473547364737473847394740474147424743474447454746474747484749475047514752475347544755475647574758475947604761476247634764476547664767476847694770477147724773477447754776477747784779478047814782478347844785478647874788478947904791479247934794479547964797479847994800480148024803480448054806480748084809481048114812481348144815481648174818481948204821482248234824482548264827482848294830483148324833483448354836483748384839484048414842484348444845484648474848484948504851485248534854485548564857485848594860486148624863486448654866486748684869487048714872487348744875487648774878487948804881488248834884488548864887488848894890489148924893489448954896489748984899490049014902490349044905490649074908490949104911491249134914491549164917491849194920492149224923492449254926492749284929493049314932493349344935493649374938493949404941494249434944
  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/jiffies.h>
  22. #include <linux/bootmem.h>
  23. #include <linux/compiler.h>
  24. #include <linux/kernel.h>
  25. #include <linux/kmemcheck.h>
  26. #include <linux/module.h>
  27. #include <linux/suspend.h>
  28. #include <linux/pagevec.h>
  29. #include <linux/blkdev.h>
  30. #include <linux/slab.h>
  31. #include <linux/oom.h>
  32. #include <linux/notifier.h>
  33. #include <linux/topology.h>
  34. #include <linux/sysctl.h>
  35. #include <linux/cpu.h>
  36. #include <linux/cpuset.h>
  37. #include <linux/memory_hotplug.h>
  38. #include <linux/nodemask.h>
  39. #include <linux/vmalloc.h>
  40. #include <linux/mempolicy.h>
  41. #include <linux/stop_machine.h>
  42. #include <linux/sort.h>
  43. #include <linux/pfn.h>
  44. #include <linux/backing-dev.h>
  45. #include <linux/fault-inject.h>
  46. #include <linux/page-isolation.h>
  47. #include <linux/page_cgroup.h>
  48. #include <linux/debugobjects.h>
  49. #include <linux/kmemleak.h>
  50. #include <asm/tlbflush.h>
  51. #include <asm/div64.h>
  52. #include "internal.h"
  53. /*
  54. * Array of node states.
  55. */
  56. nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
  57. [N_POSSIBLE] = NODE_MASK_ALL,
  58. [N_ONLINE] = { { [0] = 1UL } },
  59. #ifndef CONFIG_NUMA
  60. [N_NORMAL_MEMORY] = { { [0] = 1UL } },
  61. #ifdef CONFIG_HIGHMEM
  62. [N_HIGH_MEMORY] = { { [0] = 1UL } },
  63. #endif
  64. [N_CPU] = { { [0] = 1UL } },
  65. #endif /* NUMA */
  66. };
  67. EXPORT_SYMBOL(node_states);
  68. unsigned long totalram_pages __read_mostly;
  69. unsigned long totalreserve_pages __read_mostly;
  70. unsigned long highest_memmap_pfn __read_mostly;
  71. int percpu_pagelist_fraction;
  72. gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
  73. #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
  74. int pageblock_order __read_mostly;
  75. #endif
  76. static void __free_pages_ok(struct page *page, unsigned int order);
  77. /*
  78. * results with 256, 32 in the lowmem_reserve sysctl:
  79. * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
  80. * 1G machine -> (16M dma, 784M normal, 224M high)
  81. * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
  82. * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
  83. * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
  84. *
  85. * TBD: should special case ZONE_DMA32 machines here - in those we normally
  86. * don't need any ZONE_NORMAL reservation
  87. */
  88. int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
  89. #ifdef CONFIG_ZONE_DMA
  90. 256,
  91. #endif
  92. #ifdef CONFIG_ZONE_DMA32
  93. 256,
  94. #endif
  95. #ifdef CONFIG_HIGHMEM
  96. 32,
  97. #endif
  98. 32,
  99. };
  100. EXPORT_SYMBOL(totalram_pages);
  101. static char * const zone_names[MAX_NR_ZONES] = {
  102. #ifdef CONFIG_ZONE_DMA
  103. "DMA",
  104. #endif
  105. #ifdef CONFIG_ZONE_DMA32
  106. "DMA32",
  107. #endif
  108. "Normal",
  109. #ifdef CONFIG_HIGHMEM
  110. "HighMem",
  111. #endif
  112. "Movable",
  113. };
  114. int min_free_kbytes = 1024;
  115. unsigned long __meminitdata nr_kernel_pages;
  116. unsigned long __meminitdata nr_all_pages;
  117. static unsigned long __meminitdata dma_reserve;
  118. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  119. /*
  120. * MAX_ACTIVE_REGIONS determines the maximum number of distinct
  121. * ranges of memory (RAM) that may be registered with add_active_range().
  122. * Ranges passed to add_active_range() will be merged if possible
  123. * so the number of times add_active_range() can be called is
  124. * related to the number of nodes and the number of holes
  125. */
  126. #ifdef CONFIG_MAX_ACTIVE_REGIONS
  127. /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
  128. #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
  129. #else
  130. #if MAX_NUMNODES >= 32
  131. /* If there can be many nodes, allow up to 50 holes per node */
  132. #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
  133. #else
  134. /* By default, allow up to 256 distinct regions */
  135. #define MAX_ACTIVE_REGIONS 256
  136. #endif
  137. #endif
  138. static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
  139. static int __meminitdata nr_nodemap_entries;
  140. static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
  141. static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
  142. static unsigned long __initdata required_kernelcore;
  143. static unsigned long __initdata required_movablecore;
  144. static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
  145. /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
  146. int movable_zone;
  147. EXPORT_SYMBOL(movable_zone);
  148. #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
  149. #if MAX_NUMNODES > 1
  150. int nr_node_ids __read_mostly = MAX_NUMNODES;
  151. int nr_online_nodes __read_mostly = 1;
  152. EXPORT_SYMBOL(nr_node_ids);
  153. EXPORT_SYMBOL(nr_online_nodes);
  154. #endif
  155. int page_group_by_mobility_disabled __read_mostly;
  156. static void set_pageblock_migratetype(struct page *page, int migratetype)
  157. {
  158. if (unlikely(page_group_by_mobility_disabled))
  159. migratetype = MIGRATE_UNMOVABLE;
  160. set_pageblock_flags_group(page, (unsigned long)migratetype,
  161. PB_migrate, PB_migrate_end);
  162. }
  163. bool oom_killer_disabled __read_mostly;
  164. #ifdef CONFIG_DEBUG_VM
  165. static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
  166. {
  167. int ret = 0;
  168. unsigned seq;
  169. unsigned long pfn = page_to_pfn(page);
  170. do {
  171. seq = zone_span_seqbegin(zone);
  172. if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
  173. ret = 1;
  174. else if (pfn < zone->zone_start_pfn)
  175. ret = 1;
  176. } while (zone_span_seqretry(zone, seq));
  177. return ret;
  178. }
  179. static int page_is_consistent(struct zone *zone, struct page *page)
  180. {
  181. if (!pfn_valid_within(page_to_pfn(page)))
  182. return 0;
  183. if (zone != page_zone(page))
  184. return 0;
  185. return 1;
  186. }
  187. /*
  188. * Temporary debugging check for pages not lying within a given zone.
  189. */
  190. static int bad_range(struct zone *zone, struct page *page)
  191. {
  192. if (page_outside_zone_boundaries(zone, page))
  193. return 1;
  194. if (!page_is_consistent(zone, page))
  195. return 1;
  196. return 0;
  197. }
  198. #else
  199. static inline int bad_range(struct zone *zone, struct page *page)
  200. {
  201. return 0;
  202. }
  203. #endif
  204. static void bad_page(struct page *page)
  205. {
  206. static unsigned long resume;
  207. static unsigned long nr_shown;
  208. static unsigned long nr_unshown;
  209. /*
  210. * Allow a burst of 60 reports, then keep quiet for that minute;
  211. * or allow a steady drip of one report per second.
  212. */
  213. if (nr_shown == 60) {
  214. if (time_before(jiffies, resume)) {
  215. nr_unshown++;
  216. goto out;
  217. }
  218. if (nr_unshown) {
  219. printk(KERN_ALERT
  220. "BUG: Bad page state: %lu messages suppressed\n",
  221. nr_unshown);
  222. nr_unshown = 0;
  223. }
  224. nr_shown = 0;
  225. }
  226. if (nr_shown++ == 0)
  227. resume = jiffies + 60 * HZ;
  228. printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
  229. current->comm, page_to_pfn(page));
  230. printk(KERN_ALERT
  231. "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
  232. page, (void *)page->flags, page_count(page),
  233. page_mapcount(page), page->mapping, page->index);
  234. dump_stack();
  235. out:
  236. /* Leave bad fields for debug, except PageBuddy could make trouble */
  237. __ClearPageBuddy(page);
  238. add_taint(TAINT_BAD_PAGE);
  239. }
  240. /*
  241. * Higher-order pages are called "compound pages". They are structured thusly:
  242. *
  243. * The first PAGE_SIZE page is called the "head page".
  244. *
  245. * The remaining PAGE_SIZE pages are called "tail pages".
  246. *
  247. * All pages have PG_compound set. All pages have their ->private pointing at
  248. * the head page (even the head page has this).
  249. *
  250. * The first tail page's ->lru.next holds the address of the compound page's
  251. * put_page() function. Its ->lru.prev holds the order of allocation.
  252. * This usage means that zero-order pages may not be compound.
  253. */
  254. static void free_compound_page(struct page *page)
  255. {
  256. __free_pages_ok(page, compound_order(page));
  257. }
  258. void prep_compound_page(struct page *page, unsigned long order)
  259. {
  260. int i;
  261. int nr_pages = 1 << order;
  262. set_compound_page_dtor(page, free_compound_page);
  263. set_compound_order(page, order);
  264. __SetPageHead(page);
  265. for (i = 1; i < nr_pages; i++) {
  266. struct page *p = page + i;
  267. __SetPageTail(p);
  268. p->first_page = page;
  269. }
  270. }
  271. static int destroy_compound_page(struct page *page, unsigned long order)
  272. {
  273. int i;
  274. int nr_pages = 1 << order;
  275. int bad = 0;
  276. if (unlikely(compound_order(page) != order) ||
  277. unlikely(!PageHead(page))) {
  278. bad_page(page);
  279. bad++;
  280. }
  281. __ClearPageHead(page);
  282. for (i = 1; i < nr_pages; i++) {
  283. struct page *p = page + i;
  284. if (unlikely(!PageTail(p) || (p->first_page != page))) {
  285. bad_page(page);
  286. bad++;
  287. }
  288. __ClearPageTail(p);
  289. }
  290. return bad;
  291. }
  292. static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
  293. {
  294. int i;
  295. /*
  296. * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
  297. * and __GFP_HIGHMEM from hard or soft interrupt context.
  298. */
  299. VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
  300. for (i = 0; i < (1 << order); i++)
  301. clear_highpage(page + i);
  302. }
  303. static inline void set_page_order(struct page *page, int order)
  304. {
  305. set_page_private(page, order);
  306. __SetPageBuddy(page);
  307. }
  308. static inline void rmv_page_order(struct page *page)
  309. {
  310. __ClearPageBuddy(page);
  311. set_page_private(page, 0);
  312. }
  313. /*
  314. * Locate the struct page for both the matching buddy in our
  315. * pair (buddy1) and the combined O(n+1) page they form (page).
  316. *
  317. * 1) Any buddy B1 will have an order O twin B2 which satisfies
  318. * the following equation:
  319. * B2 = B1 ^ (1 << O)
  320. * For example, if the starting buddy (buddy2) is #8 its order
  321. * 1 buddy is #10:
  322. * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
  323. *
  324. * 2) Any buddy B will have an order O+1 parent P which
  325. * satisfies the following equation:
  326. * P = B & ~(1 << O)
  327. *
  328. * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
  329. */
  330. static inline struct page *
  331. __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
  332. {
  333. unsigned long buddy_idx = page_idx ^ (1 << order);
  334. return page + (buddy_idx - page_idx);
  335. }
  336. static inline unsigned long
  337. __find_combined_index(unsigned long page_idx, unsigned int order)
  338. {
  339. return (page_idx & ~(1 << order));
  340. }
  341. /*
  342. * This function checks whether a page is free && is the buddy
  343. * we can do coalesce a page and its buddy if
  344. * (a) the buddy is not in a hole &&
  345. * (b) the buddy is in the buddy system &&
  346. * (c) a page and its buddy have the same order &&
  347. * (d) a page and its buddy are in the same zone.
  348. *
  349. * For recording whether a page is in the buddy system, we use PG_buddy.
  350. * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
  351. *
  352. * For recording page's order, we use page_private(page).
  353. */
  354. static inline int page_is_buddy(struct page *page, struct page *buddy,
  355. int order)
  356. {
  357. if (!pfn_valid_within(page_to_pfn(buddy)))
  358. return 0;
  359. if (page_zone_id(page) != page_zone_id(buddy))
  360. return 0;
  361. if (PageBuddy(buddy) && page_order(buddy) == order) {
  362. VM_BUG_ON(page_count(buddy) != 0);
  363. return 1;
  364. }
  365. return 0;
  366. }
  367. /*
  368. * Freeing function for a buddy system allocator.
  369. *
  370. * The concept of a buddy system is to maintain direct-mapped table
  371. * (containing bit values) for memory blocks of various "orders".
  372. * The bottom level table contains the map for the smallest allocatable
  373. * units of memory (here, pages), and each level above it describes
  374. * pairs of units from the levels below, hence, "buddies".
  375. * At a high level, all that happens here is marking the table entry
  376. * at the bottom level available, and propagating the changes upward
  377. * as necessary, plus some accounting needed to play nicely with other
  378. * parts of the VM system.
  379. * At each level, we keep a list of pages, which are heads of continuous
  380. * free pages of length of (1 << order) and marked with PG_buddy. Page's
  381. * order is recorded in page_private(page) field.
  382. * So when we are allocating or freeing one, we can derive the state of the
  383. * other. That is, if we allocate a small block, and both were
  384. * free, the remainder of the region must be split into blocks.
  385. * If a block is freed, and its buddy is also free, then this
  386. * triggers coalescing into a block of larger size.
  387. *
  388. * -- wli
  389. */
  390. static inline void __free_one_page(struct page *page,
  391. struct zone *zone, unsigned int order,
  392. int migratetype)
  393. {
  394. unsigned long page_idx;
  395. if (unlikely(PageCompound(page)))
  396. if (unlikely(destroy_compound_page(page, order)))
  397. return;
  398. VM_BUG_ON(migratetype == -1);
  399. page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
  400. VM_BUG_ON(page_idx & ((1 << order) - 1));
  401. VM_BUG_ON(bad_range(zone, page));
  402. while (order < MAX_ORDER-1) {
  403. unsigned long combined_idx;
  404. struct page *buddy;
  405. buddy = __page_find_buddy(page, page_idx, order);
  406. if (!page_is_buddy(page, buddy, order))
  407. break;
  408. /* Our buddy is free, merge with it and move up one order. */
  409. list_del(&buddy->lru);
  410. zone->free_area[order].nr_free--;
  411. rmv_page_order(buddy);
  412. combined_idx = __find_combined_index(page_idx, order);
  413. page = page + (combined_idx - page_idx);
  414. page_idx = combined_idx;
  415. order++;
  416. }
  417. set_page_order(page, order);
  418. list_add(&page->lru,
  419. &zone->free_area[order].free_list[migratetype]);
  420. zone->free_area[order].nr_free++;
  421. }
  422. #ifdef CONFIG_HAVE_MLOCKED_PAGE_BIT
  423. /*
  424. * free_page_mlock() -- clean up attempts to free and mlocked() page.
  425. * Page should not be on lru, so no need to fix that up.
  426. * free_pages_check() will verify...
  427. */
  428. static inline void free_page_mlock(struct page *page)
  429. {
  430. __dec_zone_page_state(page, NR_MLOCK);
  431. __count_vm_event(UNEVICTABLE_MLOCKFREED);
  432. }
  433. #else
  434. static void free_page_mlock(struct page *page) { }
  435. #endif
  436. static inline int free_pages_check(struct page *page)
  437. {
  438. if (unlikely(page_mapcount(page) |
  439. (page->mapping != NULL) |
  440. (atomic_read(&page->_count) != 0) |
  441. (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
  442. bad_page(page);
  443. return 1;
  444. }
  445. if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
  446. page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
  447. return 0;
  448. }
  449. /*
  450. * Frees a list of pages.
  451. * Assumes all pages on list are in same zone, and of same order.
  452. * count is the number of pages to free.
  453. *
  454. * If the zone was previously in an "all pages pinned" state then look to
  455. * see if this freeing clears that state.
  456. *
  457. * And clear the zone's pages_scanned counter, to hold off the "all pages are
  458. * pinned" detection logic.
  459. */
  460. static void free_pages_bulk(struct zone *zone, int count,
  461. struct list_head *list, int order)
  462. {
  463. spin_lock(&zone->lock);
  464. zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
  465. zone->pages_scanned = 0;
  466. __mod_zone_page_state(zone, NR_FREE_PAGES, count << order);
  467. while (count--) {
  468. struct page *page;
  469. VM_BUG_ON(list_empty(list));
  470. page = list_entry(list->prev, struct page, lru);
  471. /* have to delete it as __free_one_page list manipulates */
  472. list_del(&page->lru);
  473. __free_one_page(page, zone, order, page_private(page));
  474. }
  475. spin_unlock(&zone->lock);
  476. }
  477. static void free_one_page(struct zone *zone, struct page *page, int order,
  478. int migratetype)
  479. {
  480. spin_lock(&zone->lock);
  481. zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
  482. zone->pages_scanned = 0;
  483. __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
  484. __free_one_page(page, zone, order, migratetype);
  485. spin_unlock(&zone->lock);
  486. }
  487. static void __free_pages_ok(struct page *page, unsigned int order)
  488. {
  489. unsigned long flags;
  490. int i;
  491. int bad = 0;
  492. int wasMlocked = TestClearPageMlocked(page);
  493. kmemcheck_free_shadow(page, order);
  494. for (i = 0 ; i < (1 << order) ; ++i)
  495. bad += free_pages_check(page + i);
  496. if (bad)
  497. return;
  498. if (!PageHighMem(page)) {
  499. debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
  500. debug_check_no_obj_freed(page_address(page),
  501. PAGE_SIZE << order);
  502. }
  503. arch_free_page(page, order);
  504. kernel_map_pages(page, 1 << order, 0);
  505. local_irq_save(flags);
  506. if (unlikely(wasMlocked))
  507. free_page_mlock(page);
  508. __count_vm_events(PGFREE, 1 << order);
  509. free_one_page(page_zone(page), page, order,
  510. get_pageblock_migratetype(page));
  511. local_irq_restore(flags);
  512. }
  513. /*
  514. * permit the bootmem allocator to evade page validation on high-order frees
  515. */
  516. void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
  517. {
  518. if (order == 0) {
  519. __ClearPageReserved(page);
  520. set_page_count(page, 0);
  521. set_page_refcounted(page);
  522. __free_page(page);
  523. } else {
  524. int loop;
  525. prefetchw(page);
  526. for (loop = 0; loop < BITS_PER_LONG; loop++) {
  527. struct page *p = &page[loop];
  528. if (loop + 1 < BITS_PER_LONG)
  529. prefetchw(p + 1);
  530. __ClearPageReserved(p);
  531. set_page_count(p, 0);
  532. }
  533. set_page_refcounted(page);
  534. __free_pages(page, order);
  535. }
  536. }
  537. /*
  538. * The order of subdivision here is critical for the IO subsystem.
  539. * Please do not alter this order without good reasons and regression
  540. * testing. Specifically, as large blocks of memory are subdivided,
  541. * the order in which smaller blocks are delivered depends on the order
  542. * they're subdivided in this function. This is the primary factor
  543. * influencing the order in which pages are delivered to the IO
  544. * subsystem according to empirical testing, and this is also justified
  545. * by considering the behavior of a buddy system containing a single
  546. * large block of memory acted on by a series of small allocations.
  547. * This behavior is a critical factor in sglist merging's success.
  548. *
  549. * -- wli
  550. */
  551. static inline void expand(struct zone *zone, struct page *page,
  552. int low, int high, struct free_area *area,
  553. int migratetype)
  554. {
  555. unsigned long size = 1 << high;
  556. while (high > low) {
  557. area--;
  558. high--;
  559. size >>= 1;
  560. VM_BUG_ON(bad_range(zone, &page[size]));
  561. list_add(&page[size].lru, &area->free_list[migratetype]);
  562. area->nr_free++;
  563. set_page_order(&page[size], high);
  564. }
  565. }
  566. /*
  567. * This page is about to be returned from the page allocator
  568. */
  569. static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
  570. {
  571. if (unlikely(page_mapcount(page) |
  572. (page->mapping != NULL) |
  573. (atomic_read(&page->_count) != 0) |
  574. (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
  575. bad_page(page);
  576. return 1;
  577. }
  578. set_page_private(page, 0);
  579. set_page_refcounted(page);
  580. arch_alloc_page(page, order);
  581. kernel_map_pages(page, 1 << order, 1);
  582. if (gfp_flags & __GFP_ZERO)
  583. prep_zero_page(page, order, gfp_flags);
  584. if (order && (gfp_flags & __GFP_COMP))
  585. prep_compound_page(page, order);
  586. return 0;
  587. }
  588. /*
  589. * Go through the free lists for the given migratetype and remove
  590. * the smallest available page from the freelists
  591. */
  592. static inline
  593. struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
  594. int migratetype)
  595. {
  596. unsigned int current_order;
  597. struct free_area * area;
  598. struct page *page;
  599. /* Find a page of the appropriate size in the preferred list */
  600. for (current_order = order; current_order < MAX_ORDER; ++current_order) {
  601. area = &(zone->free_area[current_order]);
  602. if (list_empty(&area->free_list[migratetype]))
  603. continue;
  604. page = list_entry(area->free_list[migratetype].next,
  605. struct page, lru);
  606. list_del(&page->lru);
  607. rmv_page_order(page);
  608. area->nr_free--;
  609. expand(zone, page, order, current_order, area, migratetype);
  610. return page;
  611. }
  612. return NULL;
  613. }
  614. /*
  615. * This array describes the order lists are fallen back to when
  616. * the free lists for the desirable migrate type are depleted
  617. */
  618. static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
  619. [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
  620. [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
  621. [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
  622. [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
  623. };
  624. /*
  625. * Move the free pages in a range to the free lists of the requested type.
  626. * Note that start_page and end_pages are not aligned on a pageblock
  627. * boundary. If alignment is required, use move_freepages_block()
  628. */
  629. static int move_freepages(struct zone *zone,
  630. struct page *start_page, struct page *end_page,
  631. int migratetype)
  632. {
  633. struct page *page;
  634. unsigned long order;
  635. int pages_moved = 0;
  636. #ifndef CONFIG_HOLES_IN_ZONE
  637. /*
  638. * page_zone is not safe to call in this context when
  639. * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
  640. * anyway as we check zone boundaries in move_freepages_block().
  641. * Remove at a later date when no bug reports exist related to
  642. * grouping pages by mobility
  643. */
  644. BUG_ON(page_zone(start_page) != page_zone(end_page));
  645. #endif
  646. for (page = start_page; page <= end_page;) {
  647. /* Make sure we are not inadvertently changing nodes */
  648. VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
  649. if (!pfn_valid_within(page_to_pfn(page))) {
  650. page++;
  651. continue;
  652. }
  653. if (!PageBuddy(page)) {
  654. page++;
  655. continue;
  656. }
  657. order = page_order(page);
  658. list_del(&page->lru);
  659. list_add(&page->lru,
  660. &zone->free_area[order].free_list[migratetype]);
  661. page += 1 << order;
  662. pages_moved += 1 << order;
  663. }
  664. return pages_moved;
  665. }
  666. static int move_freepages_block(struct zone *zone, struct page *page,
  667. int migratetype)
  668. {
  669. unsigned long start_pfn, end_pfn;
  670. struct page *start_page, *end_page;
  671. start_pfn = page_to_pfn(page);
  672. start_pfn = start_pfn & ~(pageblock_nr_pages-1);
  673. start_page = pfn_to_page(start_pfn);
  674. end_page = start_page + pageblock_nr_pages - 1;
  675. end_pfn = start_pfn + pageblock_nr_pages - 1;
  676. /* Do not cross zone boundaries */
  677. if (start_pfn < zone->zone_start_pfn)
  678. start_page = page;
  679. if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
  680. return 0;
  681. return move_freepages(zone, start_page, end_page, migratetype);
  682. }
  683. /* Remove an element from the buddy allocator from the fallback list */
  684. static inline struct page *
  685. __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
  686. {
  687. struct free_area * area;
  688. int current_order;
  689. struct page *page;
  690. int migratetype, i;
  691. /* Find the largest possible block of pages in the other list */
  692. for (current_order = MAX_ORDER-1; current_order >= order;
  693. --current_order) {
  694. for (i = 0; i < MIGRATE_TYPES - 1; i++) {
  695. migratetype = fallbacks[start_migratetype][i];
  696. /* MIGRATE_RESERVE handled later if necessary */
  697. if (migratetype == MIGRATE_RESERVE)
  698. continue;
  699. area = &(zone->free_area[current_order]);
  700. if (list_empty(&area->free_list[migratetype]))
  701. continue;
  702. page = list_entry(area->free_list[migratetype].next,
  703. struct page, lru);
  704. area->nr_free--;
  705. /*
  706. * If breaking a large block of pages, move all free
  707. * pages to the preferred allocation list. If falling
  708. * back for a reclaimable kernel allocation, be more
  709. * agressive about taking ownership of free pages
  710. */
  711. if (unlikely(current_order >= (pageblock_order >> 1)) ||
  712. start_migratetype == MIGRATE_RECLAIMABLE) {
  713. unsigned long pages;
  714. pages = move_freepages_block(zone, page,
  715. start_migratetype);
  716. /* Claim the whole block if over half of it is free */
  717. if (pages >= (1 << (pageblock_order-1)))
  718. set_pageblock_migratetype(page,
  719. start_migratetype);
  720. migratetype = start_migratetype;
  721. }
  722. /* Remove the page from the freelists */
  723. list_del(&page->lru);
  724. rmv_page_order(page);
  725. if (current_order == pageblock_order)
  726. set_pageblock_migratetype(page,
  727. start_migratetype);
  728. expand(zone, page, order, current_order, area, migratetype);
  729. return page;
  730. }
  731. }
  732. return NULL;
  733. }
  734. /*
  735. * Do the hard work of removing an element from the buddy allocator.
  736. * Call me with the zone->lock already held.
  737. */
  738. static struct page *__rmqueue(struct zone *zone, unsigned int order,
  739. int migratetype)
  740. {
  741. struct page *page;
  742. retry_reserve:
  743. page = __rmqueue_smallest(zone, order, migratetype);
  744. if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
  745. page = __rmqueue_fallback(zone, order, migratetype);
  746. /*
  747. * Use MIGRATE_RESERVE rather than fail an allocation. goto
  748. * is used because __rmqueue_smallest is an inline function
  749. * and we want just one call site
  750. */
  751. if (!page) {
  752. migratetype = MIGRATE_RESERVE;
  753. goto retry_reserve;
  754. }
  755. }
  756. return page;
  757. }
  758. /*
  759. * Obtain a specified number of elements from the buddy allocator, all under
  760. * a single hold of the lock, for efficiency. Add them to the supplied list.
  761. * Returns the number of new pages which were placed at *list.
  762. */
  763. static int rmqueue_bulk(struct zone *zone, unsigned int order,
  764. unsigned long count, struct list_head *list,
  765. int migratetype, int cold)
  766. {
  767. int i;
  768. spin_lock(&zone->lock);
  769. for (i = 0; i < count; ++i) {
  770. struct page *page = __rmqueue(zone, order, migratetype);
  771. if (unlikely(page == NULL))
  772. break;
  773. /*
  774. * Split buddy pages returned by expand() are received here
  775. * in physical page order. The page is added to the callers and
  776. * list and the list head then moves forward. From the callers
  777. * perspective, the linked list is ordered by page number in
  778. * some conditions. This is useful for IO devices that can
  779. * merge IO requests if the physical pages are ordered
  780. * properly.
  781. */
  782. if (likely(cold == 0))
  783. list_add(&page->lru, list);
  784. else
  785. list_add_tail(&page->lru, list);
  786. set_page_private(page, migratetype);
  787. list = &page->lru;
  788. }
  789. __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
  790. spin_unlock(&zone->lock);
  791. return i;
  792. }
  793. #ifdef CONFIG_NUMA
  794. /*
  795. * Called from the vmstat counter updater to drain pagesets of this
  796. * currently executing processor on remote nodes after they have
  797. * expired.
  798. *
  799. * Note that this function must be called with the thread pinned to
  800. * a single processor.
  801. */
  802. void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
  803. {
  804. unsigned long flags;
  805. int to_drain;
  806. local_irq_save(flags);
  807. if (pcp->count >= pcp->batch)
  808. to_drain = pcp->batch;
  809. else
  810. to_drain = pcp->count;
  811. free_pages_bulk(zone, to_drain, &pcp->list, 0);
  812. pcp->count -= to_drain;
  813. local_irq_restore(flags);
  814. }
  815. #endif
  816. /*
  817. * Drain pages of the indicated processor.
  818. *
  819. * The processor must either be the current processor and the
  820. * thread pinned to the current processor or a processor that
  821. * is not online.
  822. */
  823. static void drain_pages(unsigned int cpu)
  824. {
  825. unsigned long flags;
  826. struct zone *zone;
  827. for_each_populated_zone(zone) {
  828. struct per_cpu_pageset *pset;
  829. struct per_cpu_pages *pcp;
  830. pset = zone_pcp(zone, cpu);
  831. pcp = &pset->pcp;
  832. local_irq_save(flags);
  833. free_pages_bulk(zone, pcp->count, &pcp->list, 0);
  834. pcp->count = 0;
  835. local_irq_restore(flags);
  836. }
  837. }
  838. /*
  839. * Spill all of this CPU's per-cpu pages back into the buddy allocator.
  840. */
  841. void drain_local_pages(void *arg)
  842. {
  843. drain_pages(smp_processor_id());
  844. }
  845. /*
  846. * Spill all the per-cpu pages from all CPUs back into the buddy allocator
  847. */
  848. void drain_all_pages(void)
  849. {
  850. on_each_cpu(drain_local_pages, NULL, 1);
  851. }
  852. #ifdef CONFIG_HIBERNATION
  853. void mark_free_pages(struct zone *zone)
  854. {
  855. unsigned long pfn, max_zone_pfn;
  856. unsigned long flags;
  857. int order, t;
  858. struct list_head *curr;
  859. if (!zone->spanned_pages)
  860. return;
  861. spin_lock_irqsave(&zone->lock, flags);
  862. max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
  863. for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
  864. if (pfn_valid(pfn)) {
  865. struct page *page = pfn_to_page(pfn);
  866. if (!swsusp_page_is_forbidden(page))
  867. swsusp_unset_page_free(page);
  868. }
  869. for_each_migratetype_order(order, t) {
  870. list_for_each(curr, &zone->free_area[order].free_list[t]) {
  871. unsigned long i;
  872. pfn = page_to_pfn(list_entry(curr, struct page, lru));
  873. for (i = 0; i < (1UL << order); i++)
  874. swsusp_set_page_free(pfn_to_page(pfn + i));
  875. }
  876. }
  877. spin_unlock_irqrestore(&zone->lock, flags);
  878. }
  879. #endif /* CONFIG_PM */
  880. /*
  881. * Free a 0-order page
  882. */
  883. static void free_hot_cold_page(struct page *page, int cold)
  884. {
  885. struct zone *zone = page_zone(page);
  886. struct per_cpu_pages *pcp;
  887. unsigned long flags;
  888. int wasMlocked = TestClearPageMlocked(page);
  889. kmemcheck_free_shadow(page, 0);
  890. if (PageAnon(page))
  891. page->mapping = NULL;
  892. if (free_pages_check(page))
  893. return;
  894. if (!PageHighMem(page)) {
  895. debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
  896. debug_check_no_obj_freed(page_address(page), PAGE_SIZE);
  897. }
  898. arch_free_page(page, 0);
  899. kernel_map_pages(page, 1, 0);
  900. pcp = &zone_pcp(zone, get_cpu())->pcp;
  901. set_page_private(page, get_pageblock_migratetype(page));
  902. local_irq_save(flags);
  903. if (unlikely(wasMlocked))
  904. free_page_mlock(page);
  905. __count_vm_event(PGFREE);
  906. if (cold)
  907. list_add_tail(&page->lru, &pcp->list);
  908. else
  909. list_add(&page->lru, &pcp->list);
  910. pcp->count++;
  911. if (pcp->count >= pcp->high) {
  912. free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
  913. pcp->count -= pcp->batch;
  914. }
  915. local_irq_restore(flags);
  916. put_cpu();
  917. }
  918. void free_hot_page(struct page *page)
  919. {
  920. free_hot_cold_page(page, 0);
  921. }
  922. void free_cold_page(struct page *page)
  923. {
  924. free_hot_cold_page(page, 1);
  925. }
  926. /*
  927. * split_page takes a non-compound higher-order page, and splits it into
  928. * n (1<<order) sub-pages: page[0..n]
  929. * Each sub-page must be freed individually.
  930. *
  931. * Note: this is probably too low level an operation for use in drivers.
  932. * Please consult with lkml before using this in your driver.
  933. */
  934. void split_page(struct page *page, unsigned int order)
  935. {
  936. int i;
  937. VM_BUG_ON(PageCompound(page));
  938. VM_BUG_ON(!page_count(page));
  939. #ifdef CONFIG_KMEMCHECK
  940. /*
  941. * Split shadow pages too, because free(page[0]) would
  942. * otherwise free the whole shadow.
  943. */
  944. if (kmemcheck_page_is_tracked(page))
  945. split_page(virt_to_page(page[0].shadow), order);
  946. #endif
  947. for (i = 1; i < (1 << order); i++)
  948. set_page_refcounted(page + i);
  949. }
  950. /*
  951. * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
  952. * we cheat by calling it from here, in the order > 0 path. Saves a branch
  953. * or two.
  954. */
  955. static inline
  956. struct page *buffered_rmqueue(struct zone *preferred_zone,
  957. struct zone *zone, int order, gfp_t gfp_flags,
  958. int migratetype)
  959. {
  960. unsigned long flags;
  961. struct page *page;
  962. int cold = !!(gfp_flags & __GFP_COLD);
  963. int cpu;
  964. again:
  965. cpu = get_cpu();
  966. if (likely(order == 0)) {
  967. struct per_cpu_pages *pcp;
  968. pcp = &zone_pcp(zone, cpu)->pcp;
  969. local_irq_save(flags);
  970. if (!pcp->count) {
  971. pcp->count = rmqueue_bulk(zone, 0,
  972. pcp->batch, &pcp->list,
  973. migratetype, cold);
  974. if (unlikely(!pcp->count))
  975. goto failed;
  976. }
  977. /* Find a page of the appropriate migrate type */
  978. if (cold) {
  979. list_for_each_entry_reverse(page, &pcp->list, lru)
  980. if (page_private(page) == migratetype)
  981. break;
  982. } else {
  983. list_for_each_entry(page, &pcp->list, lru)
  984. if (page_private(page) == migratetype)
  985. break;
  986. }
  987. /* Allocate more to the pcp list if necessary */
  988. if (unlikely(&page->lru == &pcp->list)) {
  989. pcp->count += rmqueue_bulk(zone, 0,
  990. pcp->batch, &pcp->list,
  991. migratetype, cold);
  992. page = list_entry(pcp->list.next, struct page, lru);
  993. }
  994. list_del(&page->lru);
  995. pcp->count--;
  996. } else {
  997. if (unlikely(gfp_flags & __GFP_NOFAIL)) {
  998. /*
  999. * __GFP_NOFAIL is not to be used in new code.
  1000. *
  1001. * All __GFP_NOFAIL callers should be fixed so that they
  1002. * properly detect and handle allocation failures.
  1003. *
  1004. * We most definitely don't want callers attempting to
  1005. * allocate greater than order-1 page units with
  1006. * __GFP_NOFAIL.
  1007. */
  1008. WARN_ON_ONCE(order > 1);
  1009. }
  1010. spin_lock_irqsave(&zone->lock, flags);
  1011. page = __rmqueue(zone, order, migratetype);
  1012. __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
  1013. spin_unlock(&zone->lock);
  1014. if (!page)
  1015. goto failed;
  1016. }
  1017. __count_zone_vm_events(PGALLOC, zone, 1 << order);
  1018. zone_statistics(preferred_zone, zone);
  1019. local_irq_restore(flags);
  1020. put_cpu();
  1021. VM_BUG_ON(bad_range(zone, page));
  1022. if (prep_new_page(page, order, gfp_flags))
  1023. goto again;
  1024. return page;
  1025. failed:
  1026. local_irq_restore(flags);
  1027. put_cpu();
  1028. return NULL;
  1029. }
  1030. /* The ALLOC_WMARK bits are used as an index to zone->watermark */
  1031. #define ALLOC_WMARK_MIN WMARK_MIN
  1032. #define ALLOC_WMARK_LOW WMARK_LOW
  1033. #define ALLOC_WMARK_HIGH WMARK_HIGH
  1034. #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
  1035. /* Mask to get the watermark bits */
  1036. #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
  1037. #define ALLOC_HARDER 0x10 /* try to alloc harder */
  1038. #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
  1039. #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
  1040. #ifdef CONFIG_FAIL_PAGE_ALLOC
  1041. static struct fail_page_alloc_attr {
  1042. struct fault_attr attr;
  1043. u32 ignore_gfp_highmem;
  1044. u32 ignore_gfp_wait;
  1045. u32 min_order;
  1046. #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
  1047. struct dentry *ignore_gfp_highmem_file;
  1048. struct dentry *ignore_gfp_wait_file;
  1049. struct dentry *min_order_file;
  1050. #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
  1051. } fail_page_alloc = {
  1052. .attr = FAULT_ATTR_INITIALIZER,
  1053. .ignore_gfp_wait = 1,
  1054. .ignore_gfp_highmem = 1,
  1055. .min_order = 1,
  1056. };
  1057. static int __init setup_fail_page_alloc(char *str)
  1058. {
  1059. return setup_fault_attr(&fail_page_alloc.attr, str);
  1060. }
  1061. __setup("fail_page_alloc=", setup_fail_page_alloc);
  1062. static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
  1063. {
  1064. if (order < fail_page_alloc.min_order)
  1065. return 0;
  1066. if (gfp_mask & __GFP_NOFAIL)
  1067. return 0;
  1068. if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
  1069. return 0;
  1070. if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
  1071. return 0;
  1072. return should_fail(&fail_page_alloc.attr, 1 << order);
  1073. }
  1074. #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
  1075. static int __init fail_page_alloc_debugfs(void)
  1076. {
  1077. mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
  1078. struct dentry *dir;
  1079. int err;
  1080. err = init_fault_attr_dentries(&fail_page_alloc.attr,
  1081. "fail_page_alloc");
  1082. if (err)
  1083. return err;
  1084. dir = fail_page_alloc.attr.dentries.dir;
  1085. fail_page_alloc.ignore_gfp_wait_file =
  1086. debugfs_create_bool("ignore-gfp-wait", mode, dir,
  1087. &fail_page_alloc.ignore_gfp_wait);
  1088. fail_page_alloc.ignore_gfp_highmem_file =
  1089. debugfs_create_bool("ignore-gfp-highmem", mode, dir,
  1090. &fail_page_alloc.ignore_gfp_highmem);
  1091. fail_page_alloc.min_order_file =
  1092. debugfs_create_u32("min-order", mode, dir,
  1093. &fail_page_alloc.min_order);
  1094. if (!fail_page_alloc.ignore_gfp_wait_file ||
  1095. !fail_page_alloc.ignore_gfp_highmem_file ||
  1096. !fail_page_alloc.min_order_file) {
  1097. err = -ENOMEM;
  1098. debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
  1099. debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
  1100. debugfs_remove(fail_page_alloc.min_order_file);
  1101. cleanup_fault_attr_dentries(&fail_page_alloc.attr);
  1102. }
  1103. return err;
  1104. }
  1105. late_initcall(fail_page_alloc_debugfs);
  1106. #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
  1107. #else /* CONFIG_FAIL_PAGE_ALLOC */
  1108. static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
  1109. {
  1110. return 0;
  1111. }
  1112. #endif /* CONFIG_FAIL_PAGE_ALLOC */
  1113. /*
  1114. * Return 1 if free pages are above 'mark'. This takes into account the order
  1115. * of the allocation.
  1116. */
  1117. int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
  1118. int classzone_idx, int alloc_flags)
  1119. {
  1120. /* free_pages my go negative - that's OK */
  1121. long min = mark;
  1122. long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
  1123. int o;
  1124. if (alloc_flags & ALLOC_HIGH)
  1125. min -= min / 2;
  1126. if (alloc_flags & ALLOC_HARDER)
  1127. min -= min / 4;
  1128. if (free_pages <= min + z->lowmem_reserve[classzone_idx])
  1129. return 0;
  1130. for (o = 0; o < order; o++) {
  1131. /* At the next order, this order's pages become unavailable */
  1132. free_pages -= z->free_area[o].nr_free << o;
  1133. /* Require fewer higher order pages to be free */
  1134. min >>= 1;
  1135. if (free_pages <= min)
  1136. return 0;
  1137. }
  1138. return 1;
  1139. }
  1140. #ifdef CONFIG_NUMA
  1141. /*
  1142. * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
  1143. * skip over zones that are not allowed by the cpuset, or that have
  1144. * been recently (in last second) found to be nearly full. See further
  1145. * comments in mmzone.h. Reduces cache footprint of zonelist scans
  1146. * that have to skip over a lot of full or unallowed zones.
  1147. *
  1148. * If the zonelist cache is present in the passed in zonelist, then
  1149. * returns a pointer to the allowed node mask (either the current
  1150. * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
  1151. *
  1152. * If the zonelist cache is not available for this zonelist, does
  1153. * nothing and returns NULL.
  1154. *
  1155. * If the fullzones BITMAP in the zonelist cache is stale (more than
  1156. * a second since last zap'd) then we zap it out (clear its bits.)
  1157. *
  1158. * We hold off even calling zlc_setup, until after we've checked the
  1159. * first zone in the zonelist, on the theory that most allocations will
  1160. * be satisfied from that first zone, so best to examine that zone as
  1161. * quickly as we can.
  1162. */
  1163. static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
  1164. {
  1165. struct zonelist_cache *zlc; /* cached zonelist speedup info */
  1166. nodemask_t *allowednodes; /* zonelist_cache approximation */
  1167. zlc = zonelist->zlcache_ptr;
  1168. if (!zlc)
  1169. return NULL;
  1170. if (time_after(jiffies, zlc->last_full_zap + HZ)) {
  1171. bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
  1172. zlc->last_full_zap = jiffies;
  1173. }
  1174. allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
  1175. &cpuset_current_mems_allowed :
  1176. &node_states[N_HIGH_MEMORY];
  1177. return allowednodes;
  1178. }
  1179. /*
  1180. * Given 'z' scanning a zonelist, run a couple of quick checks to see
  1181. * if it is worth looking at further for free memory:
  1182. * 1) Check that the zone isn't thought to be full (doesn't have its
  1183. * bit set in the zonelist_cache fullzones BITMAP).
  1184. * 2) Check that the zones node (obtained from the zonelist_cache
  1185. * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
  1186. * Return true (non-zero) if zone is worth looking at further, or
  1187. * else return false (zero) if it is not.
  1188. *
  1189. * This check -ignores- the distinction between various watermarks,
  1190. * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
  1191. * found to be full for any variation of these watermarks, it will
  1192. * be considered full for up to one second by all requests, unless
  1193. * we are so low on memory on all allowed nodes that we are forced
  1194. * into the second scan of the zonelist.
  1195. *
  1196. * In the second scan we ignore this zonelist cache and exactly
  1197. * apply the watermarks to all zones, even it is slower to do so.
  1198. * We are low on memory in the second scan, and should leave no stone
  1199. * unturned looking for a free page.
  1200. */
  1201. static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
  1202. nodemask_t *allowednodes)
  1203. {
  1204. struct zonelist_cache *zlc; /* cached zonelist speedup info */
  1205. int i; /* index of *z in zonelist zones */
  1206. int n; /* node that zone *z is on */
  1207. zlc = zonelist->zlcache_ptr;
  1208. if (!zlc)
  1209. return 1;
  1210. i = z - zonelist->_zonerefs;
  1211. n = zlc->z_to_n[i];
  1212. /* This zone is worth trying if it is allowed but not full */
  1213. return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
  1214. }
  1215. /*
  1216. * Given 'z' scanning a zonelist, set the corresponding bit in
  1217. * zlc->fullzones, so that subsequent attempts to allocate a page
  1218. * from that zone don't waste time re-examining it.
  1219. */
  1220. static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
  1221. {
  1222. struct zonelist_cache *zlc; /* cached zonelist speedup info */
  1223. int i; /* index of *z in zonelist zones */
  1224. zlc = zonelist->zlcache_ptr;
  1225. if (!zlc)
  1226. return;
  1227. i = z - zonelist->_zonerefs;
  1228. set_bit(i, zlc->fullzones);
  1229. }
  1230. #else /* CONFIG_NUMA */
  1231. static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
  1232. {
  1233. return NULL;
  1234. }
  1235. static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
  1236. nodemask_t *allowednodes)
  1237. {
  1238. return 1;
  1239. }
  1240. static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
  1241. {
  1242. }
  1243. #endif /* CONFIG_NUMA */
  1244. /*
  1245. * get_page_from_freelist goes through the zonelist trying to allocate
  1246. * a page.
  1247. */
  1248. static struct page *
  1249. get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
  1250. struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
  1251. struct zone *preferred_zone, int migratetype)
  1252. {
  1253. struct zoneref *z;
  1254. struct page *page = NULL;
  1255. int classzone_idx;
  1256. struct zone *zone;
  1257. nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
  1258. int zlc_active = 0; /* set if using zonelist_cache */
  1259. int did_zlc_setup = 0; /* just call zlc_setup() one time */
  1260. classzone_idx = zone_idx(preferred_zone);
  1261. zonelist_scan:
  1262. /*
  1263. * Scan zonelist, looking for a zone with enough free.
  1264. * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
  1265. */
  1266. for_each_zone_zonelist_nodemask(zone, z, zonelist,
  1267. high_zoneidx, nodemask) {
  1268. if (NUMA_BUILD && zlc_active &&
  1269. !zlc_zone_worth_trying(zonelist, z, allowednodes))
  1270. continue;
  1271. if ((alloc_flags & ALLOC_CPUSET) &&
  1272. !cpuset_zone_allowed_softwall(zone, gfp_mask))
  1273. goto try_next_zone;
  1274. BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
  1275. if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
  1276. unsigned long mark;
  1277. int ret;
  1278. mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
  1279. if (zone_watermark_ok(zone, order, mark,
  1280. classzone_idx, alloc_flags))
  1281. goto try_this_zone;
  1282. if (zone_reclaim_mode == 0)
  1283. goto this_zone_full;
  1284. ret = zone_reclaim(zone, gfp_mask, order);
  1285. switch (ret) {
  1286. case ZONE_RECLAIM_NOSCAN:
  1287. /* did not scan */
  1288. goto try_next_zone;
  1289. case ZONE_RECLAIM_FULL:
  1290. /* scanned but unreclaimable */
  1291. goto this_zone_full;
  1292. default:
  1293. /* did we reclaim enough */
  1294. if (!zone_watermark_ok(zone, order, mark,
  1295. classzone_idx, alloc_flags))
  1296. goto this_zone_full;
  1297. }
  1298. }
  1299. try_this_zone:
  1300. page = buffered_rmqueue(preferred_zone, zone, order,
  1301. gfp_mask, migratetype);
  1302. if (page)
  1303. break;
  1304. this_zone_full:
  1305. if (NUMA_BUILD)
  1306. zlc_mark_zone_full(zonelist, z);
  1307. try_next_zone:
  1308. if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
  1309. /*
  1310. * we do zlc_setup after the first zone is tried but only
  1311. * if there are multiple nodes make it worthwhile
  1312. */
  1313. allowednodes = zlc_setup(zonelist, alloc_flags);
  1314. zlc_active = 1;
  1315. did_zlc_setup = 1;
  1316. }
  1317. }
  1318. if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
  1319. /* Disable zlc cache for second zonelist scan */
  1320. zlc_active = 0;
  1321. goto zonelist_scan;
  1322. }
  1323. return page;
  1324. }
  1325. static inline int
  1326. should_alloc_retry(gfp_t gfp_mask, unsigned int order,
  1327. unsigned long pages_reclaimed)
  1328. {
  1329. /* Do not loop if specifically requested */
  1330. if (gfp_mask & __GFP_NORETRY)
  1331. return 0;
  1332. /*
  1333. * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
  1334. * means __GFP_NOFAIL, but that may not be true in other
  1335. * implementations.
  1336. */
  1337. if (order <= PAGE_ALLOC_COSTLY_ORDER)
  1338. return 1;
  1339. /*
  1340. * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
  1341. * specified, then we retry until we no longer reclaim any pages
  1342. * (above), or we've reclaimed an order of pages at least as
  1343. * large as the allocation's order. In both cases, if the
  1344. * allocation still fails, we stop retrying.
  1345. */
  1346. if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
  1347. return 1;
  1348. /*
  1349. * Don't let big-order allocations loop unless the caller
  1350. * explicitly requests that.
  1351. */
  1352. if (gfp_mask & __GFP_NOFAIL)
  1353. return 1;
  1354. return 0;
  1355. }
  1356. static inline struct page *
  1357. __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
  1358. struct zonelist *zonelist, enum zone_type high_zoneidx,
  1359. nodemask_t *nodemask, struct zone *preferred_zone,
  1360. int migratetype)
  1361. {
  1362. struct page *page;
  1363. /* Acquire the OOM killer lock for the zones in zonelist */
  1364. if (!try_set_zone_oom(zonelist, gfp_mask)) {
  1365. schedule_timeout_uninterruptible(1);
  1366. return NULL;
  1367. }
  1368. /*
  1369. * Go through the zonelist yet one more time, keep very high watermark
  1370. * here, this is only to catch a parallel oom killing, we must fail if
  1371. * we're still under heavy pressure.
  1372. */
  1373. page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
  1374. order, zonelist, high_zoneidx,
  1375. ALLOC_WMARK_HIGH|ALLOC_CPUSET,
  1376. preferred_zone, migratetype);
  1377. if (page)
  1378. goto out;
  1379. /* The OOM killer will not help higher order allocs */
  1380. if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_NOFAIL))
  1381. goto out;
  1382. /* Exhausted what can be done so it's blamo time */
  1383. out_of_memory(zonelist, gfp_mask, order);
  1384. out:
  1385. clear_zonelist_oom(zonelist, gfp_mask);
  1386. return page;
  1387. }
  1388. /* The really slow allocator path where we enter direct reclaim */
  1389. static inline struct page *
  1390. __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
  1391. struct zonelist *zonelist, enum zone_type high_zoneidx,
  1392. nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
  1393. int migratetype, unsigned long *did_some_progress)
  1394. {
  1395. struct page *page = NULL;
  1396. struct reclaim_state reclaim_state;
  1397. struct task_struct *p = current;
  1398. cond_resched();
  1399. /* We now go into synchronous reclaim */
  1400. cpuset_memory_pressure_bump();
  1401. /*
  1402. * The task's cpuset might have expanded its set of allowable nodes
  1403. */
  1404. p->flags |= PF_MEMALLOC;
  1405. lockdep_set_current_reclaim_state(gfp_mask);
  1406. reclaim_state.reclaimed_slab = 0;
  1407. p->reclaim_state = &reclaim_state;
  1408. *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
  1409. p->reclaim_state = NULL;
  1410. lockdep_clear_current_reclaim_state();
  1411. p->flags &= ~PF_MEMALLOC;
  1412. cond_resched();
  1413. if (order != 0)
  1414. drain_all_pages();
  1415. if (likely(*did_some_progress))
  1416. page = get_page_from_freelist(gfp_mask, nodemask, order,
  1417. zonelist, high_zoneidx,
  1418. alloc_flags, preferred_zone,
  1419. migratetype);
  1420. return page;
  1421. }
  1422. /*
  1423. * This is called in the allocator slow-path if the allocation request is of
  1424. * sufficient urgency to ignore watermarks and take other desperate measures
  1425. */
  1426. static inline struct page *
  1427. __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
  1428. struct zonelist *zonelist, enum zone_type high_zoneidx,
  1429. nodemask_t *nodemask, struct zone *preferred_zone,
  1430. int migratetype)
  1431. {
  1432. struct page *page;
  1433. do {
  1434. page = get_page_from_freelist(gfp_mask, nodemask, order,
  1435. zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
  1436. preferred_zone, migratetype);
  1437. if (!page && gfp_mask & __GFP_NOFAIL)
  1438. congestion_wait(BLK_RW_ASYNC, HZ/50);
  1439. } while (!page && (gfp_mask & __GFP_NOFAIL));
  1440. return page;
  1441. }
  1442. static inline
  1443. void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
  1444. enum zone_type high_zoneidx)
  1445. {
  1446. struct zoneref *z;
  1447. struct zone *zone;
  1448. for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
  1449. wakeup_kswapd(zone, order);
  1450. }
  1451. static inline int
  1452. gfp_to_alloc_flags(gfp_t gfp_mask)
  1453. {
  1454. struct task_struct *p = current;
  1455. int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
  1456. const gfp_t wait = gfp_mask & __GFP_WAIT;
  1457. /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
  1458. BUILD_BUG_ON(__GFP_HIGH != ALLOC_HIGH);
  1459. /*
  1460. * The caller may dip into page reserves a bit more if the caller
  1461. * cannot run direct reclaim, or if the caller has realtime scheduling
  1462. * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
  1463. * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
  1464. */
  1465. alloc_flags |= (gfp_mask & __GFP_HIGH);
  1466. if (!wait) {
  1467. alloc_flags |= ALLOC_HARDER;
  1468. /*
  1469. * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
  1470. * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
  1471. */
  1472. alloc_flags &= ~ALLOC_CPUSET;
  1473. } else if (unlikely(rt_task(p)))
  1474. alloc_flags |= ALLOC_HARDER;
  1475. if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
  1476. if (!in_interrupt() &&
  1477. ((p->flags & PF_MEMALLOC) ||
  1478. unlikely(test_thread_flag(TIF_MEMDIE))))
  1479. alloc_flags |= ALLOC_NO_WATERMARKS;
  1480. }
  1481. return alloc_flags;
  1482. }
  1483. static inline struct page *
  1484. __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
  1485. struct zonelist *zonelist, enum zone_type high_zoneidx,
  1486. nodemask_t *nodemask, struct zone *preferred_zone,
  1487. int migratetype)
  1488. {
  1489. const gfp_t wait = gfp_mask & __GFP_WAIT;
  1490. struct page *page = NULL;
  1491. int alloc_flags;
  1492. unsigned long pages_reclaimed = 0;
  1493. unsigned long did_some_progress;
  1494. struct task_struct *p = current;
  1495. /*
  1496. * In the slowpath, we sanity check order to avoid ever trying to
  1497. * reclaim >= MAX_ORDER areas which will never succeed. Callers may
  1498. * be using allocators in order of preference for an area that is
  1499. * too large.
  1500. */
  1501. if (WARN_ON_ONCE(order >= MAX_ORDER))
  1502. return NULL;
  1503. /*
  1504. * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
  1505. * __GFP_NOWARN set) should not cause reclaim since the subsystem
  1506. * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
  1507. * using a larger set of nodes after it has established that the
  1508. * allowed per node queues are empty and that nodes are
  1509. * over allocated.
  1510. */
  1511. if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
  1512. goto nopage;
  1513. wake_all_kswapd(order, zonelist, high_zoneidx);
  1514. /*
  1515. * OK, we're below the kswapd watermark and have kicked background
  1516. * reclaim. Now things get more complex, so set up alloc_flags according
  1517. * to how we want to proceed.
  1518. */
  1519. alloc_flags = gfp_to_alloc_flags(gfp_mask);
  1520. restart:
  1521. /* This is the last chance, in general, before the goto nopage. */
  1522. page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
  1523. high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
  1524. preferred_zone, migratetype);
  1525. if (page)
  1526. goto got_pg;
  1527. rebalance:
  1528. /* Allocate without watermarks if the context allows */
  1529. if (alloc_flags & ALLOC_NO_WATERMARKS) {
  1530. page = __alloc_pages_high_priority(gfp_mask, order,
  1531. zonelist, high_zoneidx, nodemask,
  1532. preferred_zone, migratetype);
  1533. if (page)
  1534. goto got_pg;
  1535. }
  1536. /* Atomic allocations - we can't balance anything */
  1537. if (!wait)
  1538. goto nopage;
  1539. /* Avoid recursion of direct reclaim */
  1540. if (p->flags & PF_MEMALLOC)
  1541. goto nopage;
  1542. /* Try direct reclaim and then allocating */
  1543. page = __alloc_pages_direct_reclaim(gfp_mask, order,
  1544. zonelist, high_zoneidx,
  1545. nodemask,
  1546. alloc_flags, preferred_zone,
  1547. migratetype, &did_some_progress);
  1548. if (page)
  1549. goto got_pg;
  1550. /*
  1551. * If we failed to make any progress reclaiming, then we are
  1552. * running out of options and have to consider going OOM
  1553. */
  1554. if (!did_some_progress) {
  1555. if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
  1556. if (oom_killer_disabled)
  1557. goto nopage;
  1558. page = __alloc_pages_may_oom(gfp_mask, order,
  1559. zonelist, high_zoneidx,
  1560. nodemask, preferred_zone,
  1561. migratetype);
  1562. if (page)
  1563. goto got_pg;
  1564. /*
  1565. * The OOM killer does not trigger for high-order
  1566. * ~__GFP_NOFAIL allocations so if no progress is being
  1567. * made, there are no other options and retrying is
  1568. * unlikely to help.
  1569. */
  1570. if (order > PAGE_ALLOC_COSTLY_ORDER &&
  1571. !(gfp_mask & __GFP_NOFAIL))
  1572. goto nopage;
  1573. goto restart;
  1574. }
  1575. }
  1576. /* Check if we should retry the allocation */
  1577. pages_reclaimed += did_some_progress;
  1578. if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
  1579. /* Wait for some write requests to complete then retry */
  1580. congestion_wait(BLK_RW_ASYNC, HZ/50);
  1581. goto rebalance;
  1582. }
  1583. nopage:
  1584. if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
  1585. printk(KERN_WARNING "%s: page allocation failure."
  1586. " order:%d, mode:0x%x\n",
  1587. p->comm, order, gfp_mask);
  1588. dump_stack();
  1589. show_mem();
  1590. }
  1591. return page;
  1592. got_pg:
  1593. if (kmemcheck_enabled)
  1594. kmemcheck_pagealloc_alloc(page, order, gfp_mask);
  1595. return page;
  1596. }
  1597. /*
  1598. * This is the 'heart' of the zoned buddy allocator.
  1599. */
  1600. struct page *
  1601. __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
  1602. struct zonelist *zonelist, nodemask_t *nodemask)
  1603. {
  1604. enum zone_type high_zoneidx = gfp_zone(gfp_mask);
  1605. struct zone *preferred_zone;
  1606. struct page *page;
  1607. int migratetype = allocflags_to_migratetype(gfp_mask);
  1608. gfp_mask &= gfp_allowed_mask;
  1609. lockdep_trace_alloc(gfp_mask);
  1610. might_sleep_if(gfp_mask & __GFP_WAIT);
  1611. if (should_fail_alloc_page(gfp_mask, order))
  1612. return NULL;
  1613. /*
  1614. * Check the zones suitable for the gfp_mask contain at least one
  1615. * valid zone. It's possible to have an empty zonelist as a result
  1616. * of GFP_THISNODE and a memoryless node
  1617. */
  1618. if (unlikely(!zonelist->_zonerefs->zone))
  1619. return NULL;
  1620. /* The preferred zone is used for statistics later */
  1621. first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone);
  1622. if (!preferred_zone)
  1623. return NULL;
  1624. /* First allocation attempt */
  1625. page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
  1626. zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
  1627. preferred_zone, migratetype);
  1628. if (unlikely(!page))
  1629. page = __alloc_pages_slowpath(gfp_mask, order,
  1630. zonelist, high_zoneidx, nodemask,
  1631. preferred_zone, migratetype);
  1632. return page;
  1633. }
  1634. EXPORT_SYMBOL(__alloc_pages_nodemask);
  1635. /*
  1636. * Common helper functions.
  1637. */
  1638. unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
  1639. {
  1640. struct page * page;
  1641. page = alloc_pages(gfp_mask, order);
  1642. if (!page)
  1643. return 0;
  1644. return (unsigned long) page_address(page);
  1645. }
  1646. EXPORT_SYMBOL(__get_free_pages);
  1647. unsigned long get_zeroed_page(gfp_t gfp_mask)
  1648. {
  1649. struct page * page;
  1650. /*
  1651. * get_zeroed_page() returns a 32-bit address, which cannot represent
  1652. * a highmem page
  1653. */
  1654. VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
  1655. page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
  1656. if (page)
  1657. return (unsigned long) page_address(page);
  1658. return 0;
  1659. }
  1660. EXPORT_SYMBOL(get_zeroed_page);
  1661. void __pagevec_free(struct pagevec *pvec)
  1662. {
  1663. int i = pagevec_count(pvec);
  1664. while (--i >= 0)
  1665. free_hot_cold_page(pvec->pages[i], pvec->cold);
  1666. }
  1667. void __free_pages(struct page *page, unsigned int order)
  1668. {
  1669. if (put_page_testzero(page)) {
  1670. if (order == 0)
  1671. free_hot_page(page);
  1672. else
  1673. __free_pages_ok(page, order);
  1674. }
  1675. }
  1676. EXPORT_SYMBOL(__free_pages);
  1677. void free_pages(unsigned long addr, unsigned int order)
  1678. {
  1679. if (addr != 0) {
  1680. VM_BUG_ON(!virt_addr_valid((void *)addr));
  1681. __free_pages(virt_to_page((void *)addr), order);
  1682. }
  1683. }
  1684. EXPORT_SYMBOL(free_pages);
  1685. /**
  1686. * alloc_pages_exact - allocate an exact number physically-contiguous pages.
  1687. * @size: the number of bytes to allocate
  1688. * @gfp_mask: GFP flags for the allocation
  1689. *
  1690. * This function is similar to alloc_pages(), except that it allocates the
  1691. * minimum number of pages to satisfy the request. alloc_pages() can only
  1692. * allocate memory in power-of-two pages.
  1693. *
  1694. * This function is also limited by MAX_ORDER.
  1695. *
  1696. * Memory allocated by this function must be released by free_pages_exact().
  1697. */
  1698. void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
  1699. {
  1700. unsigned int order = get_order(size);
  1701. unsigned long addr;
  1702. addr = __get_free_pages(gfp_mask, order);
  1703. if (addr) {
  1704. unsigned long alloc_end = addr + (PAGE_SIZE << order);
  1705. unsigned long used = addr + PAGE_ALIGN(size);
  1706. split_page(virt_to_page((void *)addr), order);
  1707. while (used < alloc_end) {
  1708. free_page(used);
  1709. used += PAGE_SIZE;
  1710. }
  1711. }
  1712. return (void *)addr;
  1713. }
  1714. EXPORT_SYMBOL(alloc_pages_exact);
  1715. /**
  1716. * free_pages_exact - release memory allocated via alloc_pages_exact()
  1717. * @virt: the value returned by alloc_pages_exact.
  1718. * @size: size of allocation, same value as passed to alloc_pages_exact().
  1719. *
  1720. * Release the memory allocated by a previous call to alloc_pages_exact.
  1721. */
  1722. void free_pages_exact(void *virt, size_t size)
  1723. {
  1724. unsigned long addr = (unsigned long)virt;
  1725. unsigned long end = addr + PAGE_ALIGN(size);
  1726. while (addr < end) {
  1727. free_page(addr);
  1728. addr += PAGE_SIZE;
  1729. }
  1730. }
  1731. EXPORT_SYMBOL(free_pages_exact);
  1732. static unsigned int nr_free_zone_pages(int offset)
  1733. {
  1734. struct zoneref *z;
  1735. struct zone *zone;
  1736. /* Just pick one node, since fallback list is circular */
  1737. unsigned int sum = 0;
  1738. struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
  1739. for_each_zone_zonelist(zone, z, zonelist, offset) {
  1740. unsigned long size = zone->present_pages;
  1741. unsigned long high = high_wmark_pages(zone);
  1742. if (size > high)
  1743. sum += size - high;
  1744. }
  1745. return sum;
  1746. }
  1747. /*
  1748. * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
  1749. */
  1750. unsigned int nr_free_buffer_pages(void)
  1751. {
  1752. return nr_free_zone_pages(gfp_zone(GFP_USER));
  1753. }
  1754. EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
  1755. /*
  1756. * Amount of free RAM allocatable within all zones
  1757. */
  1758. unsigned int nr_free_pagecache_pages(void)
  1759. {
  1760. return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
  1761. }
  1762. static inline void show_node(struct zone *zone)
  1763. {
  1764. if (NUMA_BUILD)
  1765. printk("Node %d ", zone_to_nid(zone));
  1766. }
  1767. void si_meminfo(struct sysinfo *val)
  1768. {
  1769. val->totalram = totalram_pages;
  1770. val->sharedram = 0;
  1771. val->freeram = global_page_state(NR_FREE_PAGES);
  1772. val->bufferram = nr_blockdev_pages();
  1773. val->totalhigh = totalhigh_pages;
  1774. val->freehigh = nr_free_highpages();
  1775. val->mem_unit = PAGE_SIZE;
  1776. }
  1777. EXPORT_SYMBOL(si_meminfo);
  1778. #ifdef CONFIG_NUMA
  1779. void si_meminfo_node(struct sysinfo *val, int nid)
  1780. {
  1781. pg_data_t *pgdat = NODE_DATA(nid);
  1782. val->totalram = pgdat->node_present_pages;
  1783. val->freeram = node_page_state(nid, NR_FREE_PAGES);
  1784. #ifdef CONFIG_HIGHMEM
  1785. val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
  1786. val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
  1787. NR_FREE_PAGES);
  1788. #else
  1789. val->totalhigh = 0;
  1790. val->freehigh = 0;
  1791. #endif
  1792. val->mem_unit = PAGE_SIZE;
  1793. }
  1794. #endif
  1795. #define K(x) ((x) << (PAGE_SHIFT-10))
  1796. /*
  1797. * Show free area list (used inside shift_scroll-lock stuff)
  1798. * We also calculate the percentage fragmentation. We do this by counting the
  1799. * memory on each free list with the exception of the first item on the list.
  1800. */
  1801. void show_free_areas(void)
  1802. {
  1803. int cpu;
  1804. struct zone *zone;
  1805. for_each_populated_zone(zone) {
  1806. show_node(zone);
  1807. printk("%s per-cpu:\n", zone->name);
  1808. for_each_online_cpu(cpu) {
  1809. struct per_cpu_pageset *pageset;
  1810. pageset = zone_pcp(zone, cpu);
  1811. printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
  1812. cpu, pageset->pcp.high,
  1813. pageset->pcp.batch, pageset->pcp.count);
  1814. }
  1815. }
  1816. printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
  1817. " inactive_file:%lu"
  1818. " unevictable:%lu"
  1819. " dirty:%lu writeback:%lu unstable:%lu\n"
  1820. " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
  1821. global_page_state(NR_ACTIVE_ANON),
  1822. global_page_state(NR_ACTIVE_FILE),
  1823. global_page_state(NR_INACTIVE_ANON),
  1824. global_page_state(NR_INACTIVE_FILE),
  1825. global_page_state(NR_UNEVICTABLE),
  1826. global_page_state(NR_FILE_DIRTY),
  1827. global_page_state(NR_WRITEBACK),
  1828. global_page_state(NR_UNSTABLE_NFS),
  1829. global_page_state(NR_FREE_PAGES),
  1830. global_page_state(NR_SLAB_RECLAIMABLE) +
  1831. global_page_state(NR_SLAB_UNRECLAIMABLE),
  1832. global_page_state(NR_FILE_MAPPED),
  1833. global_page_state(NR_PAGETABLE),
  1834. global_page_state(NR_BOUNCE));
  1835. for_each_populated_zone(zone) {
  1836. int i;
  1837. show_node(zone);
  1838. printk("%s"
  1839. " free:%lukB"
  1840. " min:%lukB"
  1841. " low:%lukB"
  1842. " high:%lukB"
  1843. " active_anon:%lukB"
  1844. " inactive_anon:%lukB"
  1845. " active_file:%lukB"
  1846. " inactive_file:%lukB"
  1847. " unevictable:%lukB"
  1848. " present:%lukB"
  1849. " pages_scanned:%lu"
  1850. " all_unreclaimable? %s"
  1851. "\n",
  1852. zone->name,
  1853. K(zone_page_state(zone, NR_FREE_PAGES)),
  1854. K(min_wmark_pages(zone)),
  1855. K(low_wmark_pages(zone)),
  1856. K(high_wmark_pages(zone)),
  1857. K(zone_page_state(zone, NR_ACTIVE_ANON)),
  1858. K(zone_page_state(zone, NR_INACTIVE_ANON)),
  1859. K(zone_page_state(zone, NR_ACTIVE_FILE)),
  1860. K(zone_page_state(zone, NR_INACTIVE_FILE)),
  1861. K(zone_page_state(zone, NR_UNEVICTABLE)),
  1862. K(zone->present_pages),
  1863. zone->pages_scanned,
  1864. (zone_is_all_unreclaimable(zone) ? "yes" : "no")
  1865. );
  1866. printk("lowmem_reserve[]:");
  1867. for (i = 0; i < MAX_NR_ZONES; i++)
  1868. printk(" %lu", zone->lowmem_reserve[i]);
  1869. printk("\n");
  1870. }
  1871. for_each_populated_zone(zone) {
  1872. unsigned long nr[MAX_ORDER], flags, order, total = 0;
  1873. show_node(zone);
  1874. printk("%s: ", zone->name);
  1875. spin_lock_irqsave(&zone->lock, flags);
  1876. for (order = 0; order < MAX_ORDER; order++) {
  1877. nr[order] = zone->free_area[order].nr_free;
  1878. total += nr[order] << order;
  1879. }
  1880. spin_unlock_irqrestore(&zone->lock, flags);
  1881. for (order = 0; order < MAX_ORDER; order++)
  1882. printk("%lu*%lukB ", nr[order], K(1UL) << order);
  1883. printk("= %lukB\n", K(total));
  1884. }
  1885. printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
  1886. show_swap_cache_info();
  1887. }
  1888. static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
  1889. {
  1890. zoneref->zone = zone;
  1891. zoneref->zone_idx = zone_idx(zone);
  1892. }
  1893. /*
  1894. * Builds allocation fallback zone lists.
  1895. *
  1896. * Add all populated zones of a node to the zonelist.
  1897. */
  1898. static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
  1899. int nr_zones, enum zone_type zone_type)
  1900. {
  1901. struct zone *zone;
  1902. BUG_ON(zone_type >= MAX_NR_ZONES);
  1903. zone_type++;
  1904. do {
  1905. zone_type--;
  1906. zone = pgdat->node_zones + zone_type;
  1907. if (populated_zone(zone)) {
  1908. zoneref_set_zone(zone,
  1909. &zonelist->_zonerefs[nr_zones++]);
  1910. check_highest_zone(zone_type);
  1911. }
  1912. } while (zone_type);
  1913. return nr_zones;
  1914. }
  1915. /*
  1916. * zonelist_order:
  1917. * 0 = automatic detection of better ordering.
  1918. * 1 = order by ([node] distance, -zonetype)
  1919. * 2 = order by (-zonetype, [node] distance)
  1920. *
  1921. * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
  1922. * the same zonelist. So only NUMA can configure this param.
  1923. */
  1924. #define ZONELIST_ORDER_DEFAULT 0
  1925. #define ZONELIST_ORDER_NODE 1
  1926. #define ZONELIST_ORDER_ZONE 2
  1927. /* zonelist order in the kernel.
  1928. * set_zonelist_order() will set this to NODE or ZONE.
  1929. */
  1930. static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
  1931. static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
  1932. #ifdef CONFIG_NUMA
  1933. /* The value user specified ....changed by config */
  1934. static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
  1935. /* string for sysctl */
  1936. #define NUMA_ZONELIST_ORDER_LEN 16
  1937. char numa_zonelist_order[16] = "default";
  1938. /*
  1939. * interface for configure zonelist ordering.
  1940. * command line option "numa_zonelist_order"
  1941. * = "[dD]efault - default, automatic configuration.
  1942. * = "[nN]ode - order by node locality, then by zone within node
  1943. * = "[zZ]one - order by zone, then by locality within zone
  1944. */
  1945. static int __parse_numa_zonelist_order(char *s)
  1946. {
  1947. if (*s == 'd' || *s == 'D') {
  1948. user_zonelist_order = ZONELIST_ORDER_DEFAULT;
  1949. } else if (*s == 'n' || *s == 'N') {
  1950. user_zonelist_order = ZONELIST_ORDER_NODE;
  1951. } else if (*s == 'z' || *s == 'Z') {
  1952. user_zonelist_order = ZONELIST_ORDER_ZONE;
  1953. } else {
  1954. printk(KERN_WARNING
  1955. "Ignoring invalid numa_zonelist_order value: "
  1956. "%s\n", s);
  1957. return -EINVAL;
  1958. }
  1959. return 0;
  1960. }
  1961. static __init int setup_numa_zonelist_order(char *s)
  1962. {
  1963. if (s)
  1964. return __parse_numa_zonelist_order(s);
  1965. return 0;
  1966. }
  1967. early_param("numa_zonelist_order", setup_numa_zonelist_order);
  1968. /*
  1969. * sysctl handler for numa_zonelist_order
  1970. */
  1971. int numa_zonelist_order_handler(ctl_table *table, int write,
  1972. struct file *file, void __user *buffer, size_t *length,
  1973. loff_t *ppos)
  1974. {
  1975. char saved_string[NUMA_ZONELIST_ORDER_LEN];
  1976. int ret;
  1977. if (write)
  1978. strncpy(saved_string, (char*)table->data,
  1979. NUMA_ZONELIST_ORDER_LEN);
  1980. ret = proc_dostring(table, write, file, buffer, length, ppos);
  1981. if (ret)
  1982. return ret;
  1983. if (write) {
  1984. int oldval = user_zonelist_order;
  1985. if (__parse_numa_zonelist_order((char*)table->data)) {
  1986. /*
  1987. * bogus value. restore saved string
  1988. */
  1989. strncpy((char*)table->data, saved_string,
  1990. NUMA_ZONELIST_ORDER_LEN);
  1991. user_zonelist_order = oldval;
  1992. } else if (oldval != user_zonelist_order)
  1993. build_all_zonelists();
  1994. }
  1995. return 0;
  1996. }
  1997. #define MAX_NODE_LOAD (nr_online_nodes)
  1998. static int node_load[MAX_NUMNODES];
  1999. /**
  2000. * find_next_best_node - find the next node that should appear in a given node's fallback list
  2001. * @node: node whose fallback list we're appending
  2002. * @used_node_mask: nodemask_t of already used nodes
  2003. *
  2004. * We use a number of factors to determine which is the next node that should
  2005. * appear on a given node's fallback list. The node should not have appeared
  2006. * already in @node's fallback list, and it should be the next closest node
  2007. * according to the distance array (which contains arbitrary distance values
  2008. * from each node to each node in the system), and should also prefer nodes
  2009. * with no CPUs, since presumably they'll have very little allocation pressure
  2010. * on them otherwise.
  2011. * It returns -1 if no node is found.
  2012. */
  2013. static int find_next_best_node(int node, nodemask_t *used_node_mask)
  2014. {
  2015. int n, val;
  2016. int min_val = INT_MAX;
  2017. int best_node = -1;
  2018. const struct cpumask *tmp = cpumask_of_node(0);
  2019. /* Use the local node if we haven't already */
  2020. if (!node_isset(node, *used_node_mask)) {
  2021. node_set(node, *used_node_mask);
  2022. return node;
  2023. }
  2024. for_each_node_state(n, N_HIGH_MEMORY) {
  2025. /* Don't want a node to appear more than once */
  2026. if (node_isset(n, *used_node_mask))
  2027. continue;
  2028. /* Use the distance array to find the distance */
  2029. val = node_distance(node, n);
  2030. /* Penalize nodes under us ("prefer the next node") */
  2031. val += (n < node);
  2032. /* Give preference to headless and unused nodes */
  2033. tmp = cpumask_of_node(n);
  2034. if (!cpumask_empty(tmp))
  2035. val += PENALTY_FOR_NODE_WITH_CPUS;
  2036. /* Slight preference for less loaded node */
  2037. val *= (MAX_NODE_LOAD*MAX_NUMNODES);
  2038. val += node_load[n];
  2039. if (val < min_val) {
  2040. min_val = val;
  2041. best_node = n;
  2042. }
  2043. }
  2044. if (best_node >= 0)
  2045. node_set(best_node, *used_node_mask);
  2046. return best_node;
  2047. }
  2048. /*
  2049. * Build zonelists ordered by node and zones within node.
  2050. * This results in maximum locality--normal zone overflows into local
  2051. * DMA zone, if any--but risks exhausting DMA zone.
  2052. */
  2053. static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
  2054. {
  2055. int j;
  2056. struct zonelist *zonelist;
  2057. zonelist = &pgdat->node_zonelists[0];
  2058. for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
  2059. ;
  2060. j = build_zonelists_node(NODE_DATA(node), zonelist, j,
  2061. MAX_NR_ZONES - 1);
  2062. zonelist->_zonerefs[j].zone = NULL;
  2063. zonelist->_zonerefs[j].zone_idx = 0;
  2064. }
  2065. /*
  2066. * Build gfp_thisnode zonelists
  2067. */
  2068. static void build_thisnode_zonelists(pg_data_t *pgdat)
  2069. {
  2070. int j;
  2071. struct zonelist *zonelist;
  2072. zonelist = &pgdat->node_zonelists[1];
  2073. j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
  2074. zonelist->_zonerefs[j].zone = NULL;
  2075. zonelist->_zonerefs[j].zone_idx = 0;
  2076. }
  2077. /*
  2078. * Build zonelists ordered by zone and nodes within zones.
  2079. * This results in conserving DMA zone[s] until all Normal memory is
  2080. * exhausted, but results in overflowing to remote node while memory
  2081. * may still exist in local DMA zone.
  2082. */
  2083. static int node_order[MAX_NUMNODES];
  2084. static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
  2085. {
  2086. int pos, j, node;
  2087. int zone_type; /* needs to be signed */
  2088. struct zone *z;
  2089. struct zonelist *zonelist;
  2090. zonelist = &pgdat->node_zonelists[0];
  2091. pos = 0;
  2092. for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
  2093. for (j = 0; j < nr_nodes; j++) {
  2094. node = node_order[j];
  2095. z = &NODE_DATA(node)->node_zones[zone_type];
  2096. if (populated_zone(z)) {
  2097. zoneref_set_zone(z,
  2098. &zonelist->_zonerefs[pos++]);
  2099. check_highest_zone(zone_type);
  2100. }
  2101. }
  2102. }
  2103. zonelist->_zonerefs[pos].zone = NULL;
  2104. zonelist->_zonerefs[pos].zone_idx = 0;
  2105. }
  2106. static int default_zonelist_order(void)
  2107. {
  2108. int nid, zone_type;
  2109. unsigned long low_kmem_size,total_size;
  2110. struct zone *z;
  2111. int average_size;
  2112. /*
  2113. * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
  2114. * If they are really small and used heavily, the system can fall
  2115. * into OOM very easily.
  2116. * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
  2117. */
  2118. /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
  2119. low_kmem_size = 0;
  2120. total_size = 0;
  2121. for_each_online_node(nid) {
  2122. for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
  2123. z = &NODE_DATA(nid)->node_zones[zone_type];
  2124. if (populated_zone(z)) {
  2125. if (zone_type < ZONE_NORMAL)
  2126. low_kmem_size += z->present_pages;
  2127. total_size += z->present_pages;
  2128. }
  2129. }
  2130. }
  2131. if (!low_kmem_size || /* there are no DMA area. */
  2132. low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
  2133. return ZONELIST_ORDER_NODE;
  2134. /*
  2135. * look into each node's config.
  2136. * If there is a node whose DMA/DMA32 memory is very big area on
  2137. * local memory, NODE_ORDER may be suitable.
  2138. */
  2139. average_size = total_size /
  2140. (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
  2141. for_each_online_node(nid) {
  2142. low_kmem_size = 0;
  2143. total_size = 0;
  2144. for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
  2145. z = &NODE_DATA(nid)->node_zones[zone_type];
  2146. if (populated_zone(z)) {
  2147. if (zone_type < ZONE_NORMAL)
  2148. low_kmem_size += z->present_pages;
  2149. total_size += z->present_pages;
  2150. }
  2151. }
  2152. if (low_kmem_size &&
  2153. total_size > average_size && /* ignore small node */
  2154. low_kmem_size > total_size * 70/100)
  2155. return ZONELIST_ORDER_NODE;
  2156. }
  2157. return ZONELIST_ORDER_ZONE;
  2158. }
  2159. static void set_zonelist_order(void)
  2160. {
  2161. if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
  2162. current_zonelist_order = default_zonelist_order();
  2163. else
  2164. current_zonelist_order = user_zonelist_order;
  2165. }
  2166. static void build_zonelists(pg_data_t *pgdat)
  2167. {
  2168. int j, node, load;
  2169. enum zone_type i;
  2170. nodemask_t used_mask;
  2171. int local_node, prev_node;
  2172. struct zonelist *zonelist;
  2173. int order = current_zonelist_order;
  2174. /* initialize zonelists */
  2175. for (i = 0; i < MAX_ZONELISTS; i++) {
  2176. zonelist = pgdat->node_zonelists + i;
  2177. zonelist->_zonerefs[0].zone = NULL;
  2178. zonelist->_zonerefs[0].zone_idx = 0;
  2179. }
  2180. /* NUMA-aware ordering of nodes */
  2181. local_node = pgdat->node_id;
  2182. load = nr_online_nodes;
  2183. prev_node = local_node;
  2184. nodes_clear(used_mask);
  2185. memset(node_load, 0, sizeof(node_load));
  2186. memset(node_order, 0, sizeof(node_order));
  2187. j = 0;
  2188. while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
  2189. int distance = node_distance(local_node, node);
  2190. /*
  2191. * If another node is sufficiently far away then it is better
  2192. * to reclaim pages in a zone before going off node.
  2193. */
  2194. if (distance > RECLAIM_DISTANCE)
  2195. zone_reclaim_mode = 1;
  2196. /*
  2197. * We don't want to pressure a particular node.
  2198. * So adding penalty to the first node in same
  2199. * distance group to make it round-robin.
  2200. */
  2201. if (distance != node_distance(local_node, prev_node))
  2202. node_load[node] = load;
  2203. prev_node = node;
  2204. load--;
  2205. if (order == ZONELIST_ORDER_NODE)
  2206. build_zonelists_in_node_order(pgdat, node);
  2207. else
  2208. node_order[j++] = node; /* remember order */
  2209. }
  2210. if (order == ZONELIST_ORDER_ZONE) {
  2211. /* calculate node order -- i.e., DMA last! */
  2212. build_zonelists_in_zone_order(pgdat, j);
  2213. }
  2214. build_thisnode_zonelists(pgdat);
  2215. }
  2216. /* Construct the zonelist performance cache - see further mmzone.h */
  2217. static void build_zonelist_cache(pg_data_t *pgdat)
  2218. {
  2219. struct zonelist *zonelist;
  2220. struct zonelist_cache *zlc;
  2221. struct zoneref *z;
  2222. zonelist = &pgdat->node_zonelists[0];
  2223. zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
  2224. bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
  2225. for (z = zonelist->_zonerefs; z->zone; z++)
  2226. zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
  2227. }
  2228. #else /* CONFIG_NUMA */
  2229. static void set_zonelist_order(void)
  2230. {
  2231. current_zonelist_order = ZONELIST_ORDER_ZONE;
  2232. }
  2233. static void build_zonelists(pg_data_t *pgdat)
  2234. {
  2235. int node, local_node;
  2236. enum zone_type j;
  2237. struct zonelist *zonelist;
  2238. local_node = pgdat->node_id;
  2239. zonelist = &pgdat->node_zonelists[0];
  2240. j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
  2241. /*
  2242. * Now we build the zonelist so that it contains the zones
  2243. * of all the other nodes.
  2244. * We don't want to pressure a particular node, so when
  2245. * building the zones for node N, we make sure that the
  2246. * zones coming right after the local ones are those from
  2247. * node N+1 (modulo N)
  2248. */
  2249. for (node = local_node + 1; node < MAX_NUMNODES; node++) {
  2250. if (!node_online(node))
  2251. continue;
  2252. j = build_zonelists_node(NODE_DATA(node), zonelist, j,
  2253. MAX_NR_ZONES - 1);
  2254. }
  2255. for (node = 0; node < local_node; node++) {
  2256. if (!node_online(node))
  2257. continue;
  2258. j = build_zonelists_node(NODE_DATA(node), zonelist, j,
  2259. MAX_NR_ZONES - 1);
  2260. }
  2261. zonelist->_zonerefs[j].zone = NULL;
  2262. zonelist->_zonerefs[j].zone_idx = 0;
  2263. }
  2264. /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
  2265. static void build_zonelist_cache(pg_data_t *pgdat)
  2266. {
  2267. pgdat->node_zonelists[0].zlcache_ptr = NULL;
  2268. }
  2269. #endif /* CONFIG_NUMA */
  2270. /* return values int ....just for stop_machine() */
  2271. static int __build_all_zonelists(void *dummy)
  2272. {
  2273. int nid;
  2274. for_each_online_node(nid) {
  2275. pg_data_t *pgdat = NODE_DATA(nid);
  2276. build_zonelists(pgdat);
  2277. build_zonelist_cache(pgdat);
  2278. }
  2279. return 0;
  2280. }
  2281. void build_all_zonelists(void)
  2282. {
  2283. set_zonelist_order();
  2284. if (system_state == SYSTEM_BOOTING) {
  2285. __build_all_zonelists(NULL);
  2286. mminit_verify_zonelist();
  2287. cpuset_init_current_mems_allowed();
  2288. } else {
  2289. /* we have to stop all cpus to guarantee there is no user
  2290. of zonelist */
  2291. stop_machine(__build_all_zonelists, NULL, NULL);
  2292. /* cpuset refresh routine should be here */
  2293. }
  2294. vm_total_pages = nr_free_pagecache_pages();
  2295. /*
  2296. * Disable grouping by mobility if the number of pages in the
  2297. * system is too low to allow the mechanism to work. It would be
  2298. * more accurate, but expensive to check per-zone. This check is
  2299. * made on memory-hotadd so a system can start with mobility
  2300. * disabled and enable it later
  2301. */
  2302. if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
  2303. page_group_by_mobility_disabled = 1;
  2304. else
  2305. page_group_by_mobility_disabled = 0;
  2306. printk("Built %i zonelists in %s order, mobility grouping %s. "
  2307. "Total pages: %ld\n",
  2308. nr_online_nodes,
  2309. zonelist_order_name[current_zonelist_order],
  2310. page_group_by_mobility_disabled ? "off" : "on",
  2311. vm_total_pages);
  2312. #ifdef CONFIG_NUMA
  2313. printk("Policy zone: %s\n", zone_names[policy_zone]);
  2314. #endif
  2315. }
  2316. /*
  2317. * Helper functions to size the waitqueue hash table.
  2318. * Essentially these want to choose hash table sizes sufficiently
  2319. * large so that collisions trying to wait on pages are rare.
  2320. * But in fact, the number of active page waitqueues on typical
  2321. * systems is ridiculously low, less than 200. So this is even
  2322. * conservative, even though it seems large.
  2323. *
  2324. * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
  2325. * waitqueues, i.e. the size of the waitq table given the number of pages.
  2326. */
  2327. #define PAGES_PER_WAITQUEUE 256
  2328. #ifndef CONFIG_MEMORY_HOTPLUG
  2329. static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
  2330. {
  2331. unsigned long size = 1;
  2332. pages /= PAGES_PER_WAITQUEUE;
  2333. while (size < pages)
  2334. size <<= 1;
  2335. /*
  2336. * Once we have dozens or even hundreds of threads sleeping
  2337. * on IO we've got bigger problems than wait queue collision.
  2338. * Limit the size of the wait table to a reasonable size.
  2339. */
  2340. size = min(size, 4096UL);
  2341. return max(size, 4UL);
  2342. }
  2343. #else
  2344. /*
  2345. * A zone's size might be changed by hot-add, so it is not possible to determine
  2346. * a suitable size for its wait_table. So we use the maximum size now.
  2347. *
  2348. * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
  2349. *
  2350. * i386 (preemption config) : 4096 x 16 = 64Kbyte.
  2351. * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
  2352. * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
  2353. *
  2354. * The maximum entries are prepared when a zone's memory is (512K + 256) pages
  2355. * or more by the traditional way. (See above). It equals:
  2356. *
  2357. * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
  2358. * ia64(16K page size) : = ( 8G + 4M)byte.
  2359. * powerpc (64K page size) : = (32G +16M)byte.
  2360. */
  2361. static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
  2362. {
  2363. return 4096UL;
  2364. }
  2365. #endif
  2366. /*
  2367. * This is an integer logarithm so that shifts can be used later
  2368. * to extract the more random high bits from the multiplicative
  2369. * hash function before the remainder is taken.
  2370. */
  2371. static inline unsigned long wait_table_bits(unsigned long size)
  2372. {
  2373. return ffz(~size);
  2374. }
  2375. #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
  2376. /*
  2377. * Mark a number of pageblocks as MIGRATE_RESERVE. The number
  2378. * of blocks reserved is based on min_wmark_pages(zone). The memory within
  2379. * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
  2380. * higher will lead to a bigger reserve which will get freed as contiguous
  2381. * blocks as reclaim kicks in
  2382. */
  2383. static void setup_zone_migrate_reserve(struct zone *zone)
  2384. {
  2385. unsigned long start_pfn, pfn, end_pfn;
  2386. struct page *page;
  2387. unsigned long reserve, block_migratetype;
  2388. /* Get the start pfn, end pfn and the number of blocks to reserve */
  2389. start_pfn = zone->zone_start_pfn;
  2390. end_pfn = start_pfn + zone->spanned_pages;
  2391. reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
  2392. pageblock_order;
  2393. for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
  2394. if (!pfn_valid(pfn))
  2395. continue;
  2396. page = pfn_to_page(pfn);
  2397. /* Watch out for overlapping nodes */
  2398. if (page_to_nid(page) != zone_to_nid(zone))
  2399. continue;
  2400. /* Blocks with reserved pages will never free, skip them. */
  2401. if (PageReserved(page))
  2402. continue;
  2403. block_migratetype = get_pageblock_migratetype(page);
  2404. /* If this block is reserved, account for it */
  2405. if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
  2406. reserve--;
  2407. continue;
  2408. }
  2409. /* Suitable for reserving if this block is movable */
  2410. if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
  2411. set_pageblock_migratetype(page, MIGRATE_RESERVE);
  2412. move_freepages_block(zone, page, MIGRATE_RESERVE);
  2413. reserve--;
  2414. continue;
  2415. }
  2416. /*
  2417. * If the reserve is met and this is a previous reserved block,
  2418. * take it back
  2419. */
  2420. if (block_migratetype == MIGRATE_RESERVE) {
  2421. set_pageblock_migratetype(page, MIGRATE_MOVABLE);
  2422. move_freepages_block(zone, page, MIGRATE_MOVABLE);
  2423. }
  2424. }
  2425. }
  2426. /*
  2427. * Initially all pages are reserved - free ones are freed
  2428. * up by free_all_bootmem() once the early boot process is
  2429. * done. Non-atomic initialization, single-pass.
  2430. */
  2431. void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
  2432. unsigned long start_pfn, enum memmap_context context)
  2433. {
  2434. struct page *page;
  2435. unsigned long end_pfn = start_pfn + size;
  2436. unsigned long pfn;
  2437. struct zone *z;
  2438. if (highest_memmap_pfn < end_pfn - 1)
  2439. highest_memmap_pfn = end_pfn - 1;
  2440. z = &NODE_DATA(nid)->node_zones[zone];
  2441. for (pfn = start_pfn; pfn < end_pfn; pfn++) {
  2442. /*
  2443. * There can be holes in boot-time mem_map[]s
  2444. * handed to this function. They do not
  2445. * exist on hotplugged memory.
  2446. */
  2447. if (context == MEMMAP_EARLY) {
  2448. if (!early_pfn_valid(pfn))
  2449. continue;
  2450. if (!early_pfn_in_nid(pfn, nid))
  2451. continue;
  2452. }
  2453. page = pfn_to_page(pfn);
  2454. set_page_links(page, zone, nid, pfn);
  2455. mminit_verify_page_links(page, zone, nid, pfn);
  2456. init_page_count(page);
  2457. reset_page_mapcount(page);
  2458. SetPageReserved(page);
  2459. /*
  2460. * Mark the block movable so that blocks are reserved for
  2461. * movable at startup. This will force kernel allocations
  2462. * to reserve their blocks rather than leaking throughout
  2463. * the address space during boot when many long-lived
  2464. * kernel allocations are made. Later some blocks near
  2465. * the start are marked MIGRATE_RESERVE by
  2466. * setup_zone_migrate_reserve()
  2467. *
  2468. * bitmap is created for zone's valid pfn range. but memmap
  2469. * can be created for invalid pages (for alignment)
  2470. * check here not to call set_pageblock_migratetype() against
  2471. * pfn out of zone.
  2472. */
  2473. if ((z->zone_start_pfn <= pfn)
  2474. && (pfn < z->zone_start_pfn + z->spanned_pages)
  2475. && !(pfn & (pageblock_nr_pages - 1)))
  2476. set_pageblock_migratetype(page, MIGRATE_MOVABLE);
  2477. INIT_LIST_HEAD(&page->lru);
  2478. #ifdef WANT_PAGE_VIRTUAL
  2479. /* The shift won't overflow because ZONE_NORMAL is below 4G. */
  2480. if (!is_highmem_idx(zone))
  2481. set_page_address(page, __va(pfn << PAGE_SHIFT));
  2482. #endif
  2483. }
  2484. }
  2485. static void __meminit zone_init_free_lists(struct zone *zone)
  2486. {
  2487. int order, t;
  2488. for_each_migratetype_order(order, t) {
  2489. INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
  2490. zone->free_area[order].nr_free = 0;
  2491. }
  2492. }
  2493. #ifndef __HAVE_ARCH_MEMMAP_INIT
  2494. #define memmap_init(size, nid, zone, start_pfn) \
  2495. memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
  2496. #endif
  2497. static int zone_batchsize(struct zone *zone)
  2498. {
  2499. #ifdef CONFIG_MMU
  2500. int batch;
  2501. /*
  2502. * The per-cpu-pages pools are set to around 1000th of the
  2503. * size of the zone. But no more than 1/2 of a meg.
  2504. *
  2505. * OK, so we don't know how big the cache is. So guess.
  2506. */
  2507. batch = zone->present_pages / 1024;
  2508. if (batch * PAGE_SIZE > 512 * 1024)
  2509. batch = (512 * 1024) / PAGE_SIZE;
  2510. batch /= 4; /* We effectively *= 4 below */
  2511. if (batch < 1)
  2512. batch = 1;
  2513. /*
  2514. * Clamp the batch to a 2^n - 1 value. Having a power
  2515. * of 2 value was found to be more likely to have
  2516. * suboptimal cache aliasing properties in some cases.
  2517. *
  2518. * For example if 2 tasks are alternately allocating
  2519. * batches of pages, one task can end up with a lot
  2520. * of pages of one half of the possible page colors
  2521. * and the other with pages of the other colors.
  2522. */
  2523. batch = rounddown_pow_of_two(batch + batch/2) - 1;
  2524. return batch;
  2525. #else
  2526. /* The deferral and batching of frees should be suppressed under NOMMU
  2527. * conditions.
  2528. *
  2529. * The problem is that NOMMU needs to be able to allocate large chunks
  2530. * of contiguous memory as there's no hardware page translation to
  2531. * assemble apparent contiguous memory from discontiguous pages.
  2532. *
  2533. * Queueing large contiguous runs of pages for batching, however,
  2534. * causes the pages to actually be freed in smaller chunks. As there
  2535. * can be a significant delay between the individual batches being
  2536. * recycled, this leads to the once large chunks of space being
  2537. * fragmented and becoming unavailable for high-order allocations.
  2538. */
  2539. return 0;
  2540. #endif
  2541. }
  2542. static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
  2543. {
  2544. struct per_cpu_pages *pcp;
  2545. memset(p, 0, sizeof(*p));
  2546. pcp = &p->pcp;
  2547. pcp->count = 0;
  2548. pcp->high = 6 * batch;
  2549. pcp->batch = max(1UL, 1 * batch);
  2550. INIT_LIST_HEAD(&pcp->list);
  2551. }
  2552. /*
  2553. * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
  2554. * to the value high for the pageset p.
  2555. */
  2556. static void setup_pagelist_highmark(struct per_cpu_pageset *p,
  2557. unsigned long high)
  2558. {
  2559. struct per_cpu_pages *pcp;
  2560. pcp = &p->pcp;
  2561. pcp->high = high;
  2562. pcp->batch = max(1UL, high/4);
  2563. if ((high/4) > (PAGE_SHIFT * 8))
  2564. pcp->batch = PAGE_SHIFT * 8;
  2565. }
  2566. #ifdef CONFIG_NUMA
  2567. /*
  2568. * Boot pageset table. One per cpu which is going to be used for all
  2569. * zones and all nodes. The parameters will be set in such a way
  2570. * that an item put on a list will immediately be handed over to
  2571. * the buddy list. This is safe since pageset manipulation is done
  2572. * with interrupts disabled.
  2573. *
  2574. * Some NUMA counter updates may also be caught by the boot pagesets.
  2575. *
  2576. * The boot_pagesets must be kept even after bootup is complete for
  2577. * unused processors and/or zones. They do play a role for bootstrapping
  2578. * hotplugged processors.
  2579. *
  2580. * zoneinfo_show() and maybe other functions do
  2581. * not check if the processor is online before following the pageset pointer.
  2582. * Other parts of the kernel may not check if the zone is available.
  2583. */
  2584. static struct per_cpu_pageset boot_pageset[NR_CPUS];
  2585. /*
  2586. * Dynamically allocate memory for the
  2587. * per cpu pageset array in struct zone.
  2588. */
  2589. static int __cpuinit process_zones(int cpu)
  2590. {
  2591. struct zone *zone, *dzone;
  2592. int node = cpu_to_node(cpu);
  2593. node_set_state(node, N_CPU); /* this node has a cpu */
  2594. for_each_populated_zone(zone) {
  2595. zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
  2596. GFP_KERNEL, node);
  2597. if (!zone_pcp(zone, cpu))
  2598. goto bad;
  2599. setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
  2600. if (percpu_pagelist_fraction)
  2601. setup_pagelist_highmark(zone_pcp(zone, cpu),
  2602. (zone->present_pages / percpu_pagelist_fraction));
  2603. }
  2604. return 0;
  2605. bad:
  2606. for_each_zone(dzone) {
  2607. if (!populated_zone(dzone))
  2608. continue;
  2609. if (dzone == zone)
  2610. break;
  2611. kfree(zone_pcp(dzone, cpu));
  2612. zone_pcp(dzone, cpu) = &boot_pageset[cpu];
  2613. }
  2614. return -ENOMEM;
  2615. }
  2616. static inline void free_zone_pagesets(int cpu)
  2617. {
  2618. struct zone *zone;
  2619. for_each_zone(zone) {
  2620. struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
  2621. /* Free per_cpu_pageset if it is slab allocated */
  2622. if (pset != &boot_pageset[cpu])
  2623. kfree(pset);
  2624. zone_pcp(zone, cpu) = &boot_pageset[cpu];
  2625. }
  2626. }
  2627. static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
  2628. unsigned long action,
  2629. void *hcpu)
  2630. {
  2631. int cpu = (long)hcpu;
  2632. int ret = NOTIFY_OK;
  2633. switch (action) {
  2634. case CPU_UP_PREPARE:
  2635. case CPU_UP_PREPARE_FROZEN:
  2636. if (process_zones(cpu))
  2637. ret = NOTIFY_BAD;
  2638. break;
  2639. case CPU_UP_CANCELED:
  2640. case CPU_UP_CANCELED_FROZEN:
  2641. case CPU_DEAD:
  2642. case CPU_DEAD_FROZEN:
  2643. free_zone_pagesets(cpu);
  2644. break;
  2645. default:
  2646. break;
  2647. }
  2648. return ret;
  2649. }
  2650. static struct notifier_block __cpuinitdata pageset_notifier =
  2651. { &pageset_cpuup_callback, NULL, 0 };
  2652. void __init setup_per_cpu_pageset(void)
  2653. {
  2654. int err;
  2655. /* Initialize per_cpu_pageset for cpu 0.
  2656. * A cpuup callback will do this for every cpu
  2657. * as it comes online
  2658. */
  2659. err = process_zones(smp_processor_id());
  2660. BUG_ON(err);
  2661. register_cpu_notifier(&pageset_notifier);
  2662. }
  2663. #endif
  2664. static noinline __init_refok
  2665. int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
  2666. {
  2667. int i;
  2668. struct pglist_data *pgdat = zone->zone_pgdat;
  2669. size_t alloc_size;
  2670. /*
  2671. * The per-page waitqueue mechanism uses hashed waitqueues
  2672. * per zone.
  2673. */
  2674. zone->wait_table_hash_nr_entries =
  2675. wait_table_hash_nr_entries(zone_size_pages);
  2676. zone->wait_table_bits =
  2677. wait_table_bits(zone->wait_table_hash_nr_entries);
  2678. alloc_size = zone->wait_table_hash_nr_entries
  2679. * sizeof(wait_queue_head_t);
  2680. if (!slab_is_available()) {
  2681. zone->wait_table = (wait_queue_head_t *)
  2682. alloc_bootmem_node(pgdat, alloc_size);
  2683. } else {
  2684. /*
  2685. * This case means that a zone whose size was 0 gets new memory
  2686. * via memory hot-add.
  2687. * But it may be the case that a new node was hot-added. In
  2688. * this case vmalloc() will not be able to use this new node's
  2689. * memory - this wait_table must be initialized to use this new
  2690. * node itself as well.
  2691. * To use this new node's memory, further consideration will be
  2692. * necessary.
  2693. */
  2694. zone->wait_table = vmalloc(alloc_size);
  2695. }
  2696. if (!zone->wait_table)
  2697. return -ENOMEM;
  2698. for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
  2699. init_waitqueue_head(zone->wait_table + i);
  2700. return 0;
  2701. }
  2702. static __meminit void zone_pcp_init(struct zone *zone)
  2703. {
  2704. int cpu;
  2705. unsigned long batch = zone_batchsize(zone);
  2706. for (cpu = 0; cpu < NR_CPUS; cpu++) {
  2707. #ifdef CONFIG_NUMA
  2708. /* Early boot. Slab allocator not functional yet */
  2709. zone_pcp(zone, cpu) = &boot_pageset[cpu];
  2710. setup_pageset(&boot_pageset[cpu],0);
  2711. #else
  2712. setup_pageset(zone_pcp(zone,cpu), batch);
  2713. #endif
  2714. }
  2715. if (zone->present_pages)
  2716. printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
  2717. zone->name, zone->present_pages, batch);
  2718. }
  2719. __meminit int init_currently_empty_zone(struct zone *zone,
  2720. unsigned long zone_start_pfn,
  2721. unsigned long size,
  2722. enum memmap_context context)
  2723. {
  2724. struct pglist_data *pgdat = zone->zone_pgdat;
  2725. int ret;
  2726. ret = zone_wait_table_init(zone, size);
  2727. if (ret)
  2728. return ret;
  2729. pgdat->nr_zones = zone_idx(zone) + 1;
  2730. zone->zone_start_pfn = zone_start_pfn;
  2731. mminit_dprintk(MMINIT_TRACE, "memmap_init",
  2732. "Initialising map node %d zone %lu pfns %lu -> %lu\n",
  2733. pgdat->node_id,
  2734. (unsigned long)zone_idx(zone),
  2735. zone_start_pfn, (zone_start_pfn + size));
  2736. zone_init_free_lists(zone);
  2737. return 0;
  2738. }
  2739. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  2740. /*
  2741. * Basic iterator support. Return the first range of PFNs for a node
  2742. * Note: nid == MAX_NUMNODES returns first region regardless of node
  2743. */
  2744. static int __meminit first_active_region_index_in_nid(int nid)
  2745. {
  2746. int i;
  2747. for (i = 0; i < nr_nodemap_entries; i++)
  2748. if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
  2749. return i;
  2750. return -1;
  2751. }
  2752. /*
  2753. * Basic iterator support. Return the next active range of PFNs for a node
  2754. * Note: nid == MAX_NUMNODES returns next region regardless of node
  2755. */
  2756. static int __meminit next_active_region_index_in_nid(int index, int nid)
  2757. {
  2758. for (index = index + 1; index < nr_nodemap_entries; index++)
  2759. if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
  2760. return index;
  2761. return -1;
  2762. }
  2763. #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
  2764. /*
  2765. * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
  2766. * Architectures may implement their own version but if add_active_range()
  2767. * was used and there are no special requirements, this is a convenient
  2768. * alternative
  2769. */
  2770. int __meminit __early_pfn_to_nid(unsigned long pfn)
  2771. {
  2772. int i;
  2773. for (i = 0; i < nr_nodemap_entries; i++) {
  2774. unsigned long start_pfn = early_node_map[i].start_pfn;
  2775. unsigned long end_pfn = early_node_map[i].end_pfn;
  2776. if (start_pfn <= pfn && pfn < end_pfn)
  2777. return early_node_map[i].nid;
  2778. }
  2779. /* This is a memory hole */
  2780. return -1;
  2781. }
  2782. #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
  2783. int __meminit early_pfn_to_nid(unsigned long pfn)
  2784. {
  2785. int nid;
  2786. nid = __early_pfn_to_nid(pfn);
  2787. if (nid >= 0)
  2788. return nid;
  2789. /* just returns 0 */
  2790. return 0;
  2791. }
  2792. #ifdef CONFIG_NODES_SPAN_OTHER_NODES
  2793. bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
  2794. {
  2795. int nid;
  2796. nid = __early_pfn_to_nid(pfn);
  2797. if (nid >= 0 && nid != node)
  2798. return false;
  2799. return true;
  2800. }
  2801. #endif
  2802. /* Basic iterator support to walk early_node_map[] */
  2803. #define for_each_active_range_index_in_nid(i, nid) \
  2804. for (i = first_active_region_index_in_nid(nid); i != -1; \
  2805. i = next_active_region_index_in_nid(i, nid))
  2806. /**
  2807. * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
  2808. * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
  2809. * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
  2810. *
  2811. * If an architecture guarantees that all ranges registered with
  2812. * add_active_ranges() contain no holes and may be freed, this
  2813. * this function may be used instead of calling free_bootmem() manually.
  2814. */
  2815. void __init free_bootmem_with_active_regions(int nid,
  2816. unsigned long max_low_pfn)
  2817. {
  2818. int i;
  2819. for_each_active_range_index_in_nid(i, nid) {
  2820. unsigned long size_pages = 0;
  2821. unsigned long end_pfn = early_node_map[i].end_pfn;
  2822. if (early_node_map[i].start_pfn >= max_low_pfn)
  2823. continue;
  2824. if (end_pfn > max_low_pfn)
  2825. end_pfn = max_low_pfn;
  2826. size_pages = end_pfn - early_node_map[i].start_pfn;
  2827. free_bootmem_node(NODE_DATA(early_node_map[i].nid),
  2828. PFN_PHYS(early_node_map[i].start_pfn),
  2829. size_pages << PAGE_SHIFT);
  2830. }
  2831. }
  2832. void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
  2833. {
  2834. int i;
  2835. int ret;
  2836. for_each_active_range_index_in_nid(i, nid) {
  2837. ret = work_fn(early_node_map[i].start_pfn,
  2838. early_node_map[i].end_pfn, data);
  2839. if (ret)
  2840. break;
  2841. }
  2842. }
  2843. /**
  2844. * sparse_memory_present_with_active_regions - Call memory_present for each active range
  2845. * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
  2846. *
  2847. * If an architecture guarantees that all ranges registered with
  2848. * add_active_ranges() contain no holes and may be freed, this
  2849. * function may be used instead of calling memory_present() manually.
  2850. */
  2851. void __init sparse_memory_present_with_active_regions(int nid)
  2852. {
  2853. int i;
  2854. for_each_active_range_index_in_nid(i, nid)
  2855. memory_present(early_node_map[i].nid,
  2856. early_node_map[i].start_pfn,
  2857. early_node_map[i].end_pfn);
  2858. }
  2859. /**
  2860. * get_pfn_range_for_nid - Return the start and end page frames for a node
  2861. * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
  2862. * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
  2863. * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
  2864. *
  2865. * It returns the start and end page frame of a node based on information
  2866. * provided by an arch calling add_active_range(). If called for a node
  2867. * with no available memory, a warning is printed and the start and end
  2868. * PFNs will be 0.
  2869. */
  2870. void __meminit get_pfn_range_for_nid(unsigned int nid,
  2871. unsigned long *start_pfn, unsigned long *end_pfn)
  2872. {
  2873. int i;
  2874. *start_pfn = -1UL;
  2875. *end_pfn = 0;
  2876. for_each_active_range_index_in_nid(i, nid) {
  2877. *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
  2878. *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
  2879. }
  2880. if (*start_pfn == -1UL)
  2881. *start_pfn = 0;
  2882. }
  2883. /*
  2884. * This finds a zone that can be used for ZONE_MOVABLE pages. The
  2885. * assumption is made that zones within a node are ordered in monotonic
  2886. * increasing memory addresses so that the "highest" populated zone is used
  2887. */
  2888. static void __init find_usable_zone_for_movable(void)
  2889. {
  2890. int zone_index;
  2891. for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
  2892. if (zone_index == ZONE_MOVABLE)
  2893. continue;
  2894. if (arch_zone_highest_possible_pfn[zone_index] >
  2895. arch_zone_lowest_possible_pfn[zone_index])
  2896. break;
  2897. }
  2898. VM_BUG_ON(zone_index == -1);
  2899. movable_zone = zone_index;
  2900. }
  2901. /*
  2902. * The zone ranges provided by the architecture do not include ZONE_MOVABLE
  2903. * because it is sized independant of architecture. Unlike the other zones,
  2904. * the starting point for ZONE_MOVABLE is not fixed. It may be different
  2905. * in each node depending on the size of each node and how evenly kernelcore
  2906. * is distributed. This helper function adjusts the zone ranges
  2907. * provided by the architecture for a given node by using the end of the
  2908. * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
  2909. * zones within a node are in order of monotonic increases memory addresses
  2910. */
  2911. static void __meminit adjust_zone_range_for_zone_movable(int nid,
  2912. unsigned long zone_type,
  2913. unsigned long node_start_pfn,
  2914. unsigned long node_end_pfn,
  2915. unsigned long *zone_start_pfn,
  2916. unsigned long *zone_end_pfn)
  2917. {
  2918. /* Only adjust if ZONE_MOVABLE is on this node */
  2919. if (zone_movable_pfn[nid]) {
  2920. /* Size ZONE_MOVABLE */
  2921. if (zone_type == ZONE_MOVABLE) {
  2922. *zone_start_pfn = zone_movable_pfn[nid];
  2923. *zone_end_pfn = min(node_end_pfn,
  2924. arch_zone_highest_possible_pfn[movable_zone]);
  2925. /* Adjust for ZONE_MOVABLE starting within this range */
  2926. } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
  2927. *zone_end_pfn > zone_movable_pfn[nid]) {
  2928. *zone_end_pfn = zone_movable_pfn[nid];
  2929. /* Check if this whole range is within ZONE_MOVABLE */
  2930. } else if (*zone_start_pfn >= zone_movable_pfn[nid])
  2931. *zone_start_pfn = *zone_end_pfn;
  2932. }
  2933. }
  2934. /*
  2935. * Return the number of pages a zone spans in a node, including holes
  2936. * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
  2937. */
  2938. static unsigned long __meminit zone_spanned_pages_in_node(int nid,
  2939. unsigned long zone_type,
  2940. unsigned long *ignored)
  2941. {
  2942. unsigned long node_start_pfn, node_end_pfn;
  2943. unsigned long zone_start_pfn, zone_end_pfn;
  2944. /* Get the start and end of the node and zone */
  2945. get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
  2946. zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
  2947. zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
  2948. adjust_zone_range_for_zone_movable(nid, zone_type,
  2949. node_start_pfn, node_end_pfn,
  2950. &zone_start_pfn, &zone_end_pfn);
  2951. /* Check that this node has pages within the zone's required range */
  2952. if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
  2953. return 0;
  2954. /* Move the zone boundaries inside the node if necessary */
  2955. zone_end_pfn = min(zone_end_pfn, node_end_pfn);
  2956. zone_start_pfn = max(zone_start_pfn, node_start_pfn);
  2957. /* Return the spanned pages */
  2958. return zone_end_pfn - zone_start_pfn;
  2959. }
  2960. /*
  2961. * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
  2962. * then all holes in the requested range will be accounted for.
  2963. */
  2964. static unsigned long __meminit __absent_pages_in_range(int nid,
  2965. unsigned long range_start_pfn,
  2966. unsigned long range_end_pfn)
  2967. {
  2968. int i = 0;
  2969. unsigned long prev_end_pfn = 0, hole_pages = 0;
  2970. unsigned long start_pfn;
  2971. /* Find the end_pfn of the first active range of pfns in the node */
  2972. i = first_active_region_index_in_nid(nid);
  2973. if (i == -1)
  2974. return 0;
  2975. prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
  2976. /* Account for ranges before physical memory on this node */
  2977. if (early_node_map[i].start_pfn > range_start_pfn)
  2978. hole_pages = prev_end_pfn - range_start_pfn;
  2979. /* Find all holes for the zone within the node */
  2980. for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
  2981. /* No need to continue if prev_end_pfn is outside the zone */
  2982. if (prev_end_pfn >= range_end_pfn)
  2983. break;
  2984. /* Make sure the end of the zone is not within the hole */
  2985. start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
  2986. prev_end_pfn = max(prev_end_pfn, range_start_pfn);
  2987. /* Update the hole size cound and move on */
  2988. if (start_pfn > range_start_pfn) {
  2989. BUG_ON(prev_end_pfn > start_pfn);
  2990. hole_pages += start_pfn - prev_end_pfn;
  2991. }
  2992. prev_end_pfn = early_node_map[i].end_pfn;
  2993. }
  2994. /* Account for ranges past physical memory on this node */
  2995. if (range_end_pfn > prev_end_pfn)
  2996. hole_pages += range_end_pfn -
  2997. max(range_start_pfn, prev_end_pfn);
  2998. return hole_pages;
  2999. }
  3000. /**
  3001. * absent_pages_in_range - Return number of page frames in holes within a range
  3002. * @start_pfn: The start PFN to start searching for holes
  3003. * @end_pfn: The end PFN to stop searching for holes
  3004. *
  3005. * It returns the number of pages frames in memory holes within a range.
  3006. */
  3007. unsigned long __init absent_pages_in_range(unsigned long start_pfn,
  3008. unsigned long end_pfn)
  3009. {
  3010. return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
  3011. }
  3012. /* Return the number of page frames in holes in a zone on a node */
  3013. static unsigned long __meminit zone_absent_pages_in_node(int nid,
  3014. unsigned long zone_type,
  3015. unsigned long *ignored)
  3016. {
  3017. unsigned long node_start_pfn, node_end_pfn;
  3018. unsigned long zone_start_pfn, zone_end_pfn;
  3019. get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
  3020. zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
  3021. node_start_pfn);
  3022. zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
  3023. node_end_pfn);
  3024. adjust_zone_range_for_zone_movable(nid, zone_type,
  3025. node_start_pfn, node_end_pfn,
  3026. &zone_start_pfn, &zone_end_pfn);
  3027. return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
  3028. }
  3029. #else
  3030. static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
  3031. unsigned long zone_type,
  3032. unsigned long *zones_size)
  3033. {
  3034. return zones_size[zone_type];
  3035. }
  3036. static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
  3037. unsigned long zone_type,
  3038. unsigned long *zholes_size)
  3039. {
  3040. if (!zholes_size)
  3041. return 0;
  3042. return zholes_size[zone_type];
  3043. }
  3044. #endif
  3045. static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
  3046. unsigned long *zones_size, unsigned long *zholes_size)
  3047. {
  3048. unsigned long realtotalpages, totalpages = 0;
  3049. enum zone_type i;
  3050. for (i = 0; i < MAX_NR_ZONES; i++)
  3051. totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
  3052. zones_size);
  3053. pgdat->node_spanned_pages = totalpages;
  3054. realtotalpages = totalpages;
  3055. for (i = 0; i < MAX_NR_ZONES; i++)
  3056. realtotalpages -=
  3057. zone_absent_pages_in_node(pgdat->node_id, i,
  3058. zholes_size);
  3059. pgdat->node_present_pages = realtotalpages;
  3060. printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
  3061. realtotalpages);
  3062. }
  3063. #ifndef CONFIG_SPARSEMEM
  3064. /*
  3065. * Calculate the size of the zone->blockflags rounded to an unsigned long
  3066. * Start by making sure zonesize is a multiple of pageblock_order by rounding
  3067. * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
  3068. * round what is now in bits to nearest long in bits, then return it in
  3069. * bytes.
  3070. */
  3071. static unsigned long __init usemap_size(unsigned long zonesize)
  3072. {
  3073. unsigned long usemapsize;
  3074. usemapsize = roundup(zonesize, pageblock_nr_pages);
  3075. usemapsize = usemapsize >> pageblock_order;
  3076. usemapsize *= NR_PAGEBLOCK_BITS;
  3077. usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
  3078. return usemapsize / 8;
  3079. }
  3080. static void __init setup_usemap(struct pglist_data *pgdat,
  3081. struct zone *zone, unsigned long zonesize)
  3082. {
  3083. unsigned long usemapsize = usemap_size(zonesize);
  3084. zone->pageblock_flags = NULL;
  3085. if (usemapsize)
  3086. zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
  3087. }
  3088. #else
  3089. static void inline setup_usemap(struct pglist_data *pgdat,
  3090. struct zone *zone, unsigned long zonesize) {}
  3091. #endif /* CONFIG_SPARSEMEM */
  3092. #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
  3093. /* Return a sensible default order for the pageblock size. */
  3094. static inline int pageblock_default_order(void)
  3095. {
  3096. if (HPAGE_SHIFT > PAGE_SHIFT)
  3097. return HUGETLB_PAGE_ORDER;
  3098. return MAX_ORDER-1;
  3099. }
  3100. /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
  3101. static inline void __init set_pageblock_order(unsigned int order)
  3102. {
  3103. /* Check that pageblock_nr_pages has not already been setup */
  3104. if (pageblock_order)
  3105. return;
  3106. /*
  3107. * Assume the largest contiguous order of interest is a huge page.
  3108. * This value may be variable depending on boot parameters on IA64
  3109. */
  3110. pageblock_order = order;
  3111. }
  3112. #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
  3113. /*
  3114. * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
  3115. * and pageblock_default_order() are unused as pageblock_order is set
  3116. * at compile-time. See include/linux/pageblock-flags.h for the values of
  3117. * pageblock_order based on the kernel config
  3118. */
  3119. static inline int pageblock_default_order(unsigned int order)
  3120. {
  3121. return MAX_ORDER-1;
  3122. }
  3123. #define set_pageblock_order(x) do {} while (0)
  3124. #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
  3125. /*
  3126. * Set up the zone data structures:
  3127. * - mark all pages reserved
  3128. * - mark all memory queues empty
  3129. * - clear the memory bitmaps
  3130. */
  3131. static void __paginginit free_area_init_core(struct pglist_data *pgdat,
  3132. unsigned long *zones_size, unsigned long *zholes_size)
  3133. {
  3134. enum zone_type j;
  3135. int nid = pgdat->node_id;
  3136. unsigned long zone_start_pfn = pgdat->node_start_pfn;
  3137. int ret;
  3138. pgdat_resize_init(pgdat);
  3139. pgdat->nr_zones = 0;
  3140. init_waitqueue_head(&pgdat->kswapd_wait);
  3141. pgdat->kswapd_max_order = 0;
  3142. pgdat_page_cgroup_init(pgdat);
  3143. for (j = 0; j < MAX_NR_ZONES; j++) {
  3144. struct zone *zone = pgdat->node_zones + j;
  3145. unsigned long size, realsize, memmap_pages;
  3146. enum lru_list l;
  3147. size = zone_spanned_pages_in_node(nid, j, zones_size);
  3148. realsize = size - zone_absent_pages_in_node(nid, j,
  3149. zholes_size);
  3150. /*
  3151. * Adjust realsize so that it accounts for how much memory
  3152. * is used by this zone for memmap. This affects the watermark
  3153. * and per-cpu initialisations
  3154. */
  3155. memmap_pages =
  3156. PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
  3157. if (realsize >= memmap_pages) {
  3158. realsize -= memmap_pages;
  3159. if (memmap_pages)
  3160. printk(KERN_DEBUG
  3161. " %s zone: %lu pages used for memmap\n",
  3162. zone_names[j], memmap_pages);
  3163. } else
  3164. printk(KERN_WARNING
  3165. " %s zone: %lu pages exceeds realsize %lu\n",
  3166. zone_names[j], memmap_pages, realsize);
  3167. /* Account for reserved pages */
  3168. if (j == 0 && realsize > dma_reserve) {
  3169. realsize -= dma_reserve;
  3170. printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
  3171. zone_names[0], dma_reserve);
  3172. }
  3173. if (!is_highmem_idx(j))
  3174. nr_kernel_pages += realsize;
  3175. nr_all_pages += realsize;
  3176. zone->spanned_pages = size;
  3177. zone->present_pages = realsize;
  3178. #ifdef CONFIG_NUMA
  3179. zone->node = nid;
  3180. zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
  3181. / 100;
  3182. zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
  3183. #endif
  3184. zone->name = zone_names[j];
  3185. spin_lock_init(&zone->lock);
  3186. spin_lock_init(&zone->lru_lock);
  3187. zone_seqlock_init(zone);
  3188. zone->zone_pgdat = pgdat;
  3189. zone->prev_priority = DEF_PRIORITY;
  3190. zone_pcp_init(zone);
  3191. for_each_lru(l) {
  3192. INIT_LIST_HEAD(&zone->lru[l].list);
  3193. zone->lru[l].nr_saved_scan = 0;
  3194. }
  3195. zone->reclaim_stat.recent_rotated[0] = 0;
  3196. zone->reclaim_stat.recent_rotated[1] = 0;
  3197. zone->reclaim_stat.recent_scanned[0] = 0;
  3198. zone->reclaim_stat.recent_scanned[1] = 0;
  3199. zap_zone_vm_stats(zone);
  3200. zone->flags = 0;
  3201. if (!size)
  3202. continue;
  3203. set_pageblock_order(pageblock_default_order());
  3204. setup_usemap(pgdat, zone, size);
  3205. ret = init_currently_empty_zone(zone, zone_start_pfn,
  3206. size, MEMMAP_EARLY);
  3207. BUG_ON(ret);
  3208. memmap_init(size, nid, j, zone_start_pfn);
  3209. zone_start_pfn += size;
  3210. }
  3211. }
  3212. static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
  3213. {
  3214. /* Skip empty nodes */
  3215. if (!pgdat->node_spanned_pages)
  3216. return;
  3217. #ifdef CONFIG_FLAT_NODE_MEM_MAP
  3218. /* ia64 gets its own node_mem_map, before this, without bootmem */
  3219. if (!pgdat->node_mem_map) {
  3220. unsigned long size, start, end;
  3221. struct page *map;
  3222. /*
  3223. * The zone's endpoints aren't required to be MAX_ORDER
  3224. * aligned but the node_mem_map endpoints must be in order
  3225. * for the buddy allocator to function correctly.
  3226. */
  3227. start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
  3228. end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
  3229. end = ALIGN(end, MAX_ORDER_NR_PAGES);
  3230. size = (end - start) * sizeof(struct page);
  3231. map = alloc_remap(pgdat->node_id, size);
  3232. if (!map)
  3233. map = alloc_bootmem_node(pgdat, size);
  3234. pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
  3235. }
  3236. #ifndef CONFIG_NEED_MULTIPLE_NODES
  3237. /*
  3238. * With no DISCONTIG, the global mem_map is just set as node 0's
  3239. */
  3240. if (pgdat == NODE_DATA(0)) {
  3241. mem_map = NODE_DATA(0)->node_mem_map;
  3242. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  3243. if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
  3244. mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
  3245. #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
  3246. }
  3247. #endif
  3248. #endif /* CONFIG_FLAT_NODE_MEM_MAP */
  3249. }
  3250. void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
  3251. unsigned long node_start_pfn, unsigned long *zholes_size)
  3252. {
  3253. pg_data_t *pgdat = NODE_DATA(nid);
  3254. pgdat->node_id = nid;
  3255. pgdat->node_start_pfn = node_start_pfn;
  3256. calculate_node_totalpages(pgdat, zones_size, zholes_size);
  3257. alloc_node_mem_map(pgdat);
  3258. #ifdef CONFIG_FLAT_NODE_MEM_MAP
  3259. printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
  3260. nid, (unsigned long)pgdat,
  3261. (unsigned long)pgdat->node_mem_map);
  3262. #endif
  3263. free_area_init_core(pgdat, zones_size, zholes_size);
  3264. }
  3265. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  3266. #if MAX_NUMNODES > 1
  3267. /*
  3268. * Figure out the number of possible node ids.
  3269. */
  3270. static void __init setup_nr_node_ids(void)
  3271. {
  3272. unsigned int node;
  3273. unsigned int highest = 0;
  3274. for_each_node_mask(node, node_possible_map)
  3275. highest = node;
  3276. nr_node_ids = highest + 1;
  3277. }
  3278. #else
  3279. static inline void setup_nr_node_ids(void)
  3280. {
  3281. }
  3282. #endif
  3283. /**
  3284. * add_active_range - Register a range of PFNs backed by physical memory
  3285. * @nid: The node ID the range resides on
  3286. * @start_pfn: The start PFN of the available physical memory
  3287. * @end_pfn: The end PFN of the available physical memory
  3288. *
  3289. * These ranges are stored in an early_node_map[] and later used by
  3290. * free_area_init_nodes() to calculate zone sizes and holes. If the
  3291. * range spans a memory hole, it is up to the architecture to ensure
  3292. * the memory is not freed by the bootmem allocator. If possible
  3293. * the range being registered will be merged with existing ranges.
  3294. */
  3295. void __init add_active_range(unsigned int nid, unsigned long start_pfn,
  3296. unsigned long end_pfn)
  3297. {
  3298. int i;
  3299. mminit_dprintk(MMINIT_TRACE, "memory_register",
  3300. "Entering add_active_range(%d, %#lx, %#lx) "
  3301. "%d entries of %d used\n",
  3302. nid, start_pfn, end_pfn,
  3303. nr_nodemap_entries, MAX_ACTIVE_REGIONS);
  3304. mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
  3305. /* Merge with existing active regions if possible */
  3306. for (i = 0; i < nr_nodemap_entries; i++) {
  3307. if (early_node_map[i].nid != nid)
  3308. continue;
  3309. /* Skip if an existing region covers this new one */
  3310. if (start_pfn >= early_node_map[i].start_pfn &&
  3311. end_pfn <= early_node_map[i].end_pfn)
  3312. return;
  3313. /* Merge forward if suitable */
  3314. if (start_pfn <= early_node_map[i].end_pfn &&
  3315. end_pfn > early_node_map[i].end_pfn) {
  3316. early_node_map[i].end_pfn = end_pfn;
  3317. return;
  3318. }
  3319. /* Merge backward if suitable */
  3320. if (start_pfn < early_node_map[i].end_pfn &&
  3321. end_pfn >= early_node_map[i].start_pfn) {
  3322. early_node_map[i].start_pfn = start_pfn;
  3323. return;
  3324. }
  3325. }
  3326. /* Check that early_node_map is large enough */
  3327. if (i >= MAX_ACTIVE_REGIONS) {
  3328. printk(KERN_CRIT "More than %d memory regions, truncating\n",
  3329. MAX_ACTIVE_REGIONS);
  3330. return;
  3331. }
  3332. early_node_map[i].nid = nid;
  3333. early_node_map[i].start_pfn = start_pfn;
  3334. early_node_map[i].end_pfn = end_pfn;
  3335. nr_nodemap_entries = i + 1;
  3336. }
  3337. /**
  3338. * remove_active_range - Shrink an existing registered range of PFNs
  3339. * @nid: The node id the range is on that should be shrunk
  3340. * @start_pfn: The new PFN of the range
  3341. * @end_pfn: The new PFN of the range
  3342. *
  3343. * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
  3344. * The map is kept near the end physical page range that has already been
  3345. * registered. This function allows an arch to shrink an existing registered
  3346. * range.
  3347. */
  3348. void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
  3349. unsigned long end_pfn)
  3350. {
  3351. int i, j;
  3352. int removed = 0;
  3353. printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
  3354. nid, start_pfn, end_pfn);
  3355. /* Find the old active region end and shrink */
  3356. for_each_active_range_index_in_nid(i, nid) {
  3357. if (early_node_map[i].start_pfn >= start_pfn &&
  3358. early_node_map[i].end_pfn <= end_pfn) {
  3359. /* clear it */
  3360. early_node_map[i].start_pfn = 0;
  3361. early_node_map[i].end_pfn = 0;
  3362. removed = 1;
  3363. continue;
  3364. }
  3365. if (early_node_map[i].start_pfn < start_pfn &&
  3366. early_node_map[i].end_pfn > start_pfn) {
  3367. unsigned long temp_end_pfn = early_node_map[i].end_pfn;
  3368. early_node_map[i].end_pfn = start_pfn;
  3369. if (temp_end_pfn > end_pfn)
  3370. add_active_range(nid, end_pfn, temp_end_pfn);
  3371. continue;
  3372. }
  3373. if (early_node_map[i].start_pfn >= start_pfn &&
  3374. early_node_map[i].end_pfn > end_pfn &&
  3375. early_node_map[i].start_pfn < end_pfn) {
  3376. early_node_map[i].start_pfn = end_pfn;
  3377. continue;
  3378. }
  3379. }
  3380. if (!removed)
  3381. return;
  3382. /* remove the blank ones */
  3383. for (i = nr_nodemap_entries - 1; i > 0; i--) {
  3384. if (early_node_map[i].nid != nid)
  3385. continue;
  3386. if (early_node_map[i].end_pfn)
  3387. continue;
  3388. /* we found it, get rid of it */
  3389. for (j = i; j < nr_nodemap_entries - 1; j++)
  3390. memcpy(&early_node_map[j], &early_node_map[j+1],
  3391. sizeof(early_node_map[j]));
  3392. j = nr_nodemap_entries - 1;
  3393. memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
  3394. nr_nodemap_entries--;
  3395. }
  3396. }
  3397. /**
  3398. * remove_all_active_ranges - Remove all currently registered regions
  3399. *
  3400. * During discovery, it may be found that a table like SRAT is invalid
  3401. * and an alternative discovery method must be used. This function removes
  3402. * all currently registered regions.
  3403. */
  3404. void __init remove_all_active_ranges(void)
  3405. {
  3406. memset(early_node_map, 0, sizeof(early_node_map));
  3407. nr_nodemap_entries = 0;
  3408. }
  3409. /* Compare two active node_active_regions */
  3410. static int __init cmp_node_active_region(const void *a, const void *b)
  3411. {
  3412. struct node_active_region *arange = (struct node_active_region *)a;
  3413. struct node_active_region *brange = (struct node_active_region *)b;
  3414. /* Done this way to avoid overflows */
  3415. if (arange->start_pfn > brange->start_pfn)
  3416. return 1;
  3417. if (arange->start_pfn < brange->start_pfn)
  3418. return -1;
  3419. return 0;
  3420. }
  3421. /* sort the node_map by start_pfn */
  3422. static void __init sort_node_map(void)
  3423. {
  3424. sort(early_node_map, (size_t)nr_nodemap_entries,
  3425. sizeof(struct node_active_region),
  3426. cmp_node_active_region, NULL);
  3427. }
  3428. /* Find the lowest pfn for a node */
  3429. static unsigned long __init find_min_pfn_for_node(int nid)
  3430. {
  3431. int i;
  3432. unsigned long min_pfn = ULONG_MAX;
  3433. /* Assuming a sorted map, the first range found has the starting pfn */
  3434. for_each_active_range_index_in_nid(i, nid)
  3435. min_pfn = min(min_pfn, early_node_map[i].start_pfn);
  3436. if (min_pfn == ULONG_MAX) {
  3437. printk(KERN_WARNING
  3438. "Could not find start_pfn for node %d\n", nid);
  3439. return 0;
  3440. }
  3441. return min_pfn;
  3442. }
  3443. /**
  3444. * find_min_pfn_with_active_regions - Find the minimum PFN registered
  3445. *
  3446. * It returns the minimum PFN based on information provided via
  3447. * add_active_range().
  3448. */
  3449. unsigned long __init find_min_pfn_with_active_regions(void)
  3450. {
  3451. return find_min_pfn_for_node(MAX_NUMNODES);
  3452. }
  3453. /*
  3454. * early_calculate_totalpages()
  3455. * Sum pages in active regions for movable zone.
  3456. * Populate N_HIGH_MEMORY for calculating usable_nodes.
  3457. */
  3458. static unsigned long __init early_calculate_totalpages(void)
  3459. {
  3460. int i;
  3461. unsigned long totalpages = 0;
  3462. for (i = 0; i < nr_nodemap_entries; i++) {
  3463. unsigned long pages = early_node_map[i].end_pfn -
  3464. early_node_map[i].start_pfn;
  3465. totalpages += pages;
  3466. if (pages)
  3467. node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
  3468. }
  3469. return totalpages;
  3470. }
  3471. /*
  3472. * Find the PFN the Movable zone begins in each node. Kernel memory
  3473. * is spread evenly between nodes as long as the nodes have enough
  3474. * memory. When they don't, some nodes will have more kernelcore than
  3475. * others
  3476. */
  3477. static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
  3478. {
  3479. int i, nid;
  3480. unsigned long usable_startpfn;
  3481. unsigned long kernelcore_node, kernelcore_remaining;
  3482. /* save the state before borrow the nodemask */
  3483. nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
  3484. unsigned long totalpages = early_calculate_totalpages();
  3485. int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
  3486. /*
  3487. * If movablecore was specified, calculate what size of
  3488. * kernelcore that corresponds so that memory usable for
  3489. * any allocation type is evenly spread. If both kernelcore
  3490. * and movablecore are specified, then the value of kernelcore
  3491. * will be used for required_kernelcore if it's greater than
  3492. * what movablecore would have allowed.
  3493. */
  3494. if (required_movablecore) {
  3495. unsigned long corepages;
  3496. /*
  3497. * Round-up so that ZONE_MOVABLE is at least as large as what
  3498. * was requested by the user
  3499. */
  3500. required_movablecore =
  3501. roundup(required_movablecore, MAX_ORDER_NR_PAGES);
  3502. corepages = totalpages - required_movablecore;
  3503. required_kernelcore = max(required_kernelcore, corepages);
  3504. }
  3505. /* If kernelcore was not specified, there is no ZONE_MOVABLE */
  3506. if (!required_kernelcore)
  3507. goto out;
  3508. /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
  3509. find_usable_zone_for_movable();
  3510. usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
  3511. restart:
  3512. /* Spread kernelcore memory as evenly as possible throughout nodes */
  3513. kernelcore_node = required_kernelcore / usable_nodes;
  3514. for_each_node_state(nid, N_HIGH_MEMORY) {
  3515. /*
  3516. * Recalculate kernelcore_node if the division per node
  3517. * now exceeds what is necessary to satisfy the requested
  3518. * amount of memory for the kernel
  3519. */
  3520. if (required_kernelcore < kernelcore_node)
  3521. kernelcore_node = required_kernelcore / usable_nodes;
  3522. /*
  3523. * As the map is walked, we track how much memory is usable
  3524. * by the kernel using kernelcore_remaining. When it is
  3525. * 0, the rest of the node is usable by ZONE_MOVABLE
  3526. */
  3527. kernelcore_remaining = kernelcore_node;
  3528. /* Go through each range of PFNs within this node */
  3529. for_each_active_range_index_in_nid(i, nid) {
  3530. unsigned long start_pfn, end_pfn;
  3531. unsigned long size_pages;
  3532. start_pfn = max(early_node_map[i].start_pfn,
  3533. zone_movable_pfn[nid]);
  3534. end_pfn = early_node_map[i].end_pfn;
  3535. if (start_pfn >= end_pfn)
  3536. continue;
  3537. /* Account for what is only usable for kernelcore */
  3538. if (start_pfn < usable_startpfn) {
  3539. unsigned long kernel_pages;
  3540. kernel_pages = min(end_pfn, usable_startpfn)
  3541. - start_pfn;
  3542. kernelcore_remaining -= min(kernel_pages,
  3543. kernelcore_remaining);
  3544. required_kernelcore -= min(kernel_pages,
  3545. required_kernelcore);
  3546. /* Continue if range is now fully accounted */
  3547. if (end_pfn <= usable_startpfn) {
  3548. /*
  3549. * Push zone_movable_pfn to the end so
  3550. * that if we have to rebalance
  3551. * kernelcore across nodes, we will
  3552. * not double account here
  3553. */
  3554. zone_movable_pfn[nid] = end_pfn;
  3555. continue;
  3556. }
  3557. start_pfn = usable_startpfn;
  3558. }
  3559. /*
  3560. * The usable PFN range for ZONE_MOVABLE is from
  3561. * start_pfn->end_pfn. Calculate size_pages as the
  3562. * number of pages used as kernelcore
  3563. */
  3564. size_pages = end_pfn - start_pfn;
  3565. if (size_pages > kernelcore_remaining)
  3566. size_pages = kernelcore_remaining;
  3567. zone_movable_pfn[nid] = start_pfn + size_pages;
  3568. /*
  3569. * Some kernelcore has been met, update counts and
  3570. * break if the kernelcore for this node has been
  3571. * satisified
  3572. */
  3573. required_kernelcore -= min(required_kernelcore,
  3574. size_pages);
  3575. kernelcore_remaining -= size_pages;
  3576. if (!kernelcore_remaining)
  3577. break;
  3578. }
  3579. }
  3580. /*
  3581. * If there is still required_kernelcore, we do another pass with one
  3582. * less node in the count. This will push zone_movable_pfn[nid] further
  3583. * along on the nodes that still have memory until kernelcore is
  3584. * satisified
  3585. */
  3586. usable_nodes--;
  3587. if (usable_nodes && required_kernelcore > usable_nodes)
  3588. goto restart;
  3589. /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
  3590. for (nid = 0; nid < MAX_NUMNODES; nid++)
  3591. zone_movable_pfn[nid] =
  3592. roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
  3593. out:
  3594. /* restore the node_state */
  3595. node_states[N_HIGH_MEMORY] = saved_node_state;
  3596. }
  3597. /* Any regular memory on that node ? */
  3598. static void check_for_regular_memory(pg_data_t *pgdat)
  3599. {
  3600. #ifdef CONFIG_HIGHMEM
  3601. enum zone_type zone_type;
  3602. for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
  3603. struct zone *zone = &pgdat->node_zones[zone_type];
  3604. if (zone->present_pages)
  3605. node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
  3606. }
  3607. #endif
  3608. }
  3609. /**
  3610. * free_area_init_nodes - Initialise all pg_data_t and zone data
  3611. * @max_zone_pfn: an array of max PFNs for each zone
  3612. *
  3613. * This will call free_area_init_node() for each active node in the system.
  3614. * Using the page ranges provided by add_active_range(), the size of each
  3615. * zone in each node and their holes is calculated. If the maximum PFN
  3616. * between two adjacent zones match, it is assumed that the zone is empty.
  3617. * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
  3618. * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
  3619. * starts where the previous one ended. For example, ZONE_DMA32 starts
  3620. * at arch_max_dma_pfn.
  3621. */
  3622. void __init free_area_init_nodes(unsigned long *max_zone_pfn)
  3623. {
  3624. unsigned long nid;
  3625. int i;
  3626. /* Sort early_node_map as initialisation assumes it is sorted */
  3627. sort_node_map();
  3628. /* Record where the zone boundaries are */
  3629. memset(arch_zone_lowest_possible_pfn, 0,
  3630. sizeof(arch_zone_lowest_possible_pfn));
  3631. memset(arch_zone_highest_possible_pfn, 0,
  3632. sizeof(arch_zone_highest_possible_pfn));
  3633. arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
  3634. arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
  3635. for (i = 1; i < MAX_NR_ZONES; i++) {
  3636. if (i == ZONE_MOVABLE)
  3637. continue;
  3638. arch_zone_lowest_possible_pfn[i] =
  3639. arch_zone_highest_possible_pfn[i-1];
  3640. arch_zone_highest_possible_pfn[i] =
  3641. max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
  3642. }
  3643. arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
  3644. arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
  3645. /* Find the PFNs that ZONE_MOVABLE begins at in each node */
  3646. memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
  3647. find_zone_movable_pfns_for_nodes(zone_movable_pfn);
  3648. /* Print out the zone ranges */
  3649. printk("Zone PFN ranges:\n");
  3650. for (i = 0; i < MAX_NR_ZONES; i++) {
  3651. if (i == ZONE_MOVABLE)
  3652. continue;
  3653. printk(" %-8s %0#10lx -> %0#10lx\n",
  3654. zone_names[i],
  3655. arch_zone_lowest_possible_pfn[i],
  3656. arch_zone_highest_possible_pfn[i]);
  3657. }
  3658. /* Print out the PFNs ZONE_MOVABLE begins at in each node */
  3659. printk("Movable zone start PFN for each node\n");
  3660. for (i = 0; i < MAX_NUMNODES; i++) {
  3661. if (zone_movable_pfn[i])
  3662. printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
  3663. }
  3664. /* Print out the early_node_map[] */
  3665. printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
  3666. for (i = 0; i < nr_nodemap_entries; i++)
  3667. printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
  3668. early_node_map[i].start_pfn,
  3669. early_node_map[i].end_pfn);
  3670. /* Initialise every node */
  3671. mminit_verify_pageflags_layout();
  3672. setup_nr_node_ids();
  3673. for_each_online_node(nid) {
  3674. pg_data_t *pgdat = NODE_DATA(nid);
  3675. free_area_init_node(nid, NULL,
  3676. find_min_pfn_for_node(nid), NULL);
  3677. /* Any memory on that node */
  3678. if (pgdat->node_present_pages)
  3679. node_set_state(nid, N_HIGH_MEMORY);
  3680. check_for_regular_memory(pgdat);
  3681. }
  3682. }
  3683. static int __init cmdline_parse_core(char *p, unsigned long *core)
  3684. {
  3685. unsigned long long coremem;
  3686. if (!p)
  3687. return -EINVAL;
  3688. coremem = memparse(p, &p);
  3689. *core = coremem >> PAGE_SHIFT;
  3690. /* Paranoid check that UL is enough for the coremem value */
  3691. WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
  3692. return 0;
  3693. }
  3694. /*
  3695. * kernelcore=size sets the amount of memory for use for allocations that
  3696. * cannot be reclaimed or migrated.
  3697. */
  3698. static int __init cmdline_parse_kernelcore(char *p)
  3699. {
  3700. return cmdline_parse_core(p, &required_kernelcore);
  3701. }
  3702. /*
  3703. * movablecore=size sets the amount of memory for use for allocations that
  3704. * can be reclaimed or migrated.
  3705. */
  3706. static int __init cmdline_parse_movablecore(char *p)
  3707. {
  3708. return cmdline_parse_core(p, &required_movablecore);
  3709. }
  3710. early_param("kernelcore", cmdline_parse_kernelcore);
  3711. early_param("movablecore", cmdline_parse_movablecore);
  3712. #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
  3713. /**
  3714. * set_dma_reserve - set the specified number of pages reserved in the first zone
  3715. * @new_dma_reserve: The number of pages to mark reserved
  3716. *
  3717. * The per-cpu batchsize and zone watermarks are determined by present_pages.
  3718. * In the DMA zone, a significant percentage may be consumed by kernel image
  3719. * and other unfreeable allocations which can skew the watermarks badly. This
  3720. * function may optionally be used to account for unfreeable pages in the
  3721. * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
  3722. * smaller per-cpu batchsize.
  3723. */
  3724. void __init set_dma_reserve(unsigned long new_dma_reserve)
  3725. {
  3726. dma_reserve = new_dma_reserve;
  3727. }
  3728. #ifndef CONFIG_NEED_MULTIPLE_NODES
  3729. struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] };
  3730. EXPORT_SYMBOL(contig_page_data);
  3731. #endif
  3732. void __init free_area_init(unsigned long *zones_size)
  3733. {
  3734. free_area_init_node(0, zones_size,
  3735. __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
  3736. }
  3737. static int page_alloc_cpu_notify(struct notifier_block *self,
  3738. unsigned long action, void *hcpu)
  3739. {
  3740. int cpu = (unsigned long)hcpu;
  3741. if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
  3742. drain_pages(cpu);
  3743. /*
  3744. * Spill the event counters of the dead processor
  3745. * into the current processors event counters.
  3746. * This artificially elevates the count of the current
  3747. * processor.
  3748. */
  3749. vm_events_fold_cpu(cpu);
  3750. /*
  3751. * Zero the differential counters of the dead processor
  3752. * so that the vm statistics are consistent.
  3753. *
  3754. * This is only okay since the processor is dead and cannot
  3755. * race with what we are doing.
  3756. */
  3757. refresh_cpu_vm_stats(cpu);
  3758. }
  3759. return NOTIFY_OK;
  3760. }
  3761. void __init page_alloc_init(void)
  3762. {
  3763. hotcpu_notifier(page_alloc_cpu_notify, 0);
  3764. }
  3765. /*
  3766. * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
  3767. * or min_free_kbytes changes.
  3768. */
  3769. static void calculate_totalreserve_pages(void)
  3770. {
  3771. struct pglist_data *pgdat;
  3772. unsigned long reserve_pages = 0;
  3773. enum zone_type i, j;
  3774. for_each_online_pgdat(pgdat) {
  3775. for (i = 0; i < MAX_NR_ZONES; i++) {
  3776. struct zone *zone = pgdat->node_zones + i;
  3777. unsigned long max = 0;
  3778. /* Find valid and maximum lowmem_reserve in the zone */
  3779. for (j = i; j < MAX_NR_ZONES; j++) {
  3780. if (zone->lowmem_reserve[j] > max)
  3781. max = zone->lowmem_reserve[j];
  3782. }
  3783. /* we treat the high watermark as reserved pages. */
  3784. max += high_wmark_pages(zone);
  3785. if (max > zone->present_pages)
  3786. max = zone->present_pages;
  3787. reserve_pages += max;
  3788. }
  3789. }
  3790. totalreserve_pages = reserve_pages;
  3791. }
  3792. /*
  3793. * setup_per_zone_lowmem_reserve - called whenever
  3794. * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
  3795. * has a correct pages reserved value, so an adequate number of
  3796. * pages are left in the zone after a successful __alloc_pages().
  3797. */
  3798. static void setup_per_zone_lowmem_reserve(void)
  3799. {
  3800. struct pglist_data *pgdat;
  3801. enum zone_type j, idx;
  3802. for_each_online_pgdat(pgdat) {
  3803. for (j = 0; j < MAX_NR_ZONES; j++) {
  3804. struct zone *zone = pgdat->node_zones + j;
  3805. unsigned long present_pages = zone->present_pages;
  3806. zone->lowmem_reserve[j] = 0;
  3807. idx = j;
  3808. while (idx) {
  3809. struct zone *lower_zone;
  3810. idx--;
  3811. if (sysctl_lowmem_reserve_ratio[idx] < 1)
  3812. sysctl_lowmem_reserve_ratio[idx] = 1;
  3813. lower_zone = pgdat->node_zones + idx;
  3814. lower_zone->lowmem_reserve[j] = present_pages /
  3815. sysctl_lowmem_reserve_ratio[idx];
  3816. present_pages += lower_zone->present_pages;
  3817. }
  3818. }
  3819. }
  3820. /* update totalreserve_pages */
  3821. calculate_totalreserve_pages();
  3822. }
  3823. /**
  3824. * setup_per_zone_wmarks - called when min_free_kbytes changes
  3825. * or when memory is hot-{added|removed}
  3826. *
  3827. * Ensures that the watermark[min,low,high] values for each zone are set
  3828. * correctly with respect to min_free_kbytes.
  3829. */
  3830. void setup_per_zone_wmarks(void)
  3831. {
  3832. unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
  3833. unsigned long lowmem_pages = 0;
  3834. struct zone *zone;
  3835. unsigned long flags;
  3836. /* Calculate total number of !ZONE_HIGHMEM pages */
  3837. for_each_zone(zone) {
  3838. if (!is_highmem(zone))
  3839. lowmem_pages += zone->present_pages;
  3840. }
  3841. for_each_zone(zone) {
  3842. u64 tmp;
  3843. spin_lock_irqsave(&zone->lock, flags);
  3844. tmp = (u64)pages_min * zone->present_pages;
  3845. do_div(tmp, lowmem_pages);
  3846. if (is_highmem(zone)) {
  3847. /*
  3848. * __GFP_HIGH and PF_MEMALLOC allocations usually don't
  3849. * need highmem pages, so cap pages_min to a small
  3850. * value here.
  3851. *
  3852. * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
  3853. * deltas controls asynch page reclaim, and so should
  3854. * not be capped for highmem.
  3855. */
  3856. int min_pages;
  3857. min_pages = zone->present_pages / 1024;
  3858. if (min_pages < SWAP_CLUSTER_MAX)
  3859. min_pages = SWAP_CLUSTER_MAX;
  3860. if (min_pages > 128)
  3861. min_pages = 128;
  3862. zone->watermark[WMARK_MIN] = min_pages;
  3863. } else {
  3864. /*
  3865. * If it's a lowmem zone, reserve a number of pages
  3866. * proportionate to the zone's size.
  3867. */
  3868. zone->watermark[WMARK_MIN] = tmp;
  3869. }
  3870. zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
  3871. zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
  3872. setup_zone_migrate_reserve(zone);
  3873. spin_unlock_irqrestore(&zone->lock, flags);
  3874. }
  3875. /* update totalreserve_pages */
  3876. calculate_totalreserve_pages();
  3877. }
  3878. /**
  3879. * The inactive anon list should be small enough that the VM never has to
  3880. * do too much work, but large enough that each inactive page has a chance
  3881. * to be referenced again before it is swapped out.
  3882. *
  3883. * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
  3884. * INACTIVE_ANON pages on this zone's LRU, maintained by the
  3885. * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
  3886. * the anonymous pages are kept on the inactive list.
  3887. *
  3888. * total target max
  3889. * memory ratio inactive anon
  3890. * -------------------------------------
  3891. * 10MB 1 5MB
  3892. * 100MB 1 50MB
  3893. * 1GB 3 250MB
  3894. * 10GB 10 0.9GB
  3895. * 100GB 31 3GB
  3896. * 1TB 101 10GB
  3897. * 10TB 320 32GB
  3898. */
  3899. void calculate_zone_inactive_ratio(struct zone *zone)
  3900. {
  3901. unsigned int gb, ratio;
  3902. /* Zone size in gigabytes */
  3903. gb = zone->present_pages >> (30 - PAGE_SHIFT);
  3904. if (gb)
  3905. ratio = int_sqrt(10 * gb);
  3906. else
  3907. ratio = 1;
  3908. zone->inactive_ratio = ratio;
  3909. }
  3910. static void __init setup_per_zone_inactive_ratio(void)
  3911. {
  3912. struct zone *zone;
  3913. for_each_zone(zone)
  3914. calculate_zone_inactive_ratio(zone);
  3915. }
  3916. /*
  3917. * Initialise min_free_kbytes.
  3918. *
  3919. * For small machines we want it small (128k min). For large machines
  3920. * we want it large (64MB max). But it is not linear, because network
  3921. * bandwidth does not increase linearly with machine size. We use
  3922. *
  3923. * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
  3924. * min_free_kbytes = sqrt(lowmem_kbytes * 16)
  3925. *
  3926. * which yields
  3927. *
  3928. * 16MB: 512k
  3929. * 32MB: 724k
  3930. * 64MB: 1024k
  3931. * 128MB: 1448k
  3932. * 256MB: 2048k
  3933. * 512MB: 2896k
  3934. * 1024MB: 4096k
  3935. * 2048MB: 5792k
  3936. * 4096MB: 8192k
  3937. * 8192MB: 11584k
  3938. * 16384MB: 16384k
  3939. */
  3940. static int __init init_per_zone_wmark_min(void)
  3941. {
  3942. unsigned long lowmem_kbytes;
  3943. lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
  3944. min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
  3945. if (min_free_kbytes < 128)
  3946. min_free_kbytes = 128;
  3947. if (min_free_kbytes > 65536)
  3948. min_free_kbytes = 65536;
  3949. setup_per_zone_wmarks();
  3950. setup_per_zone_lowmem_reserve();
  3951. setup_per_zone_inactive_ratio();
  3952. return 0;
  3953. }
  3954. module_init(init_per_zone_wmark_min)
  3955. /*
  3956. * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
  3957. * that we can call two helper functions whenever min_free_kbytes
  3958. * changes.
  3959. */
  3960. int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
  3961. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3962. {
  3963. proc_dointvec(table, write, file, buffer, length, ppos);
  3964. if (write)
  3965. setup_per_zone_wmarks();
  3966. return 0;
  3967. }
  3968. #ifdef CONFIG_NUMA
  3969. int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
  3970. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3971. {
  3972. struct zone *zone;
  3973. int rc;
  3974. rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  3975. if (rc)
  3976. return rc;
  3977. for_each_zone(zone)
  3978. zone->min_unmapped_pages = (zone->present_pages *
  3979. sysctl_min_unmapped_ratio) / 100;
  3980. return 0;
  3981. }
  3982. int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
  3983. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  3984. {
  3985. struct zone *zone;
  3986. int rc;
  3987. rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  3988. if (rc)
  3989. return rc;
  3990. for_each_zone(zone)
  3991. zone->min_slab_pages = (zone->present_pages *
  3992. sysctl_min_slab_ratio) / 100;
  3993. return 0;
  3994. }
  3995. #endif
  3996. /*
  3997. * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
  3998. * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
  3999. * whenever sysctl_lowmem_reserve_ratio changes.
  4000. *
  4001. * The reserve ratio obviously has absolutely no relation with the
  4002. * minimum watermarks. The lowmem reserve ratio can only make sense
  4003. * if in function of the boot time zone sizes.
  4004. */
  4005. int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
  4006. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  4007. {
  4008. proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  4009. setup_per_zone_lowmem_reserve();
  4010. return 0;
  4011. }
  4012. /*
  4013. * percpu_pagelist_fraction - changes the pcp->high for each zone on each
  4014. * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
  4015. * can have before it gets flushed back to buddy allocator.
  4016. */
  4017. int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
  4018. struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
  4019. {
  4020. struct zone *zone;
  4021. unsigned int cpu;
  4022. int ret;
  4023. ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
  4024. if (!write || (ret == -EINVAL))
  4025. return ret;
  4026. for_each_populated_zone(zone) {
  4027. for_each_online_cpu(cpu) {
  4028. unsigned long high;
  4029. high = zone->present_pages / percpu_pagelist_fraction;
  4030. setup_pagelist_highmark(zone_pcp(zone, cpu), high);
  4031. }
  4032. }
  4033. return 0;
  4034. }
  4035. int hashdist = HASHDIST_DEFAULT;
  4036. #ifdef CONFIG_NUMA
  4037. static int __init set_hashdist(char *str)
  4038. {
  4039. if (!str)
  4040. return 0;
  4041. hashdist = simple_strtoul(str, &str, 0);
  4042. return 1;
  4043. }
  4044. __setup("hashdist=", set_hashdist);
  4045. #endif
  4046. /*
  4047. * allocate a large system hash table from bootmem
  4048. * - it is assumed that the hash table must contain an exact power-of-2
  4049. * quantity of entries
  4050. * - limit is the number of hash buckets, not the total allocation size
  4051. */
  4052. void *__init alloc_large_system_hash(const char *tablename,
  4053. unsigned long bucketsize,
  4054. unsigned long numentries,
  4055. int scale,
  4056. int flags,
  4057. unsigned int *_hash_shift,
  4058. unsigned int *_hash_mask,
  4059. unsigned long limit)
  4060. {
  4061. unsigned long long max = limit;
  4062. unsigned long log2qty, size;
  4063. void *table = NULL;
  4064. /* allow the kernel cmdline to have a say */
  4065. if (!numentries) {
  4066. /* round applicable memory size up to nearest megabyte */
  4067. numentries = nr_kernel_pages;
  4068. numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
  4069. numentries >>= 20 - PAGE_SHIFT;
  4070. numentries <<= 20 - PAGE_SHIFT;
  4071. /* limit to 1 bucket per 2^scale bytes of low memory */
  4072. if (scale > PAGE_SHIFT)
  4073. numentries >>= (scale - PAGE_SHIFT);
  4074. else
  4075. numentries <<= (PAGE_SHIFT - scale);
  4076. /* Make sure we've got at least a 0-order allocation.. */
  4077. if (unlikely((numentries * bucketsize) < PAGE_SIZE))
  4078. numentries = PAGE_SIZE / bucketsize;
  4079. }
  4080. numentries = roundup_pow_of_two(numentries);
  4081. /* limit allocation size to 1/16 total memory by default */
  4082. if (max == 0) {
  4083. max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
  4084. do_div(max, bucketsize);
  4085. }
  4086. if (numentries > max)
  4087. numentries = max;
  4088. log2qty = ilog2(numentries);
  4089. do {
  4090. size = bucketsize << log2qty;
  4091. if (flags & HASH_EARLY)
  4092. table = alloc_bootmem_nopanic(size);
  4093. else if (hashdist)
  4094. table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
  4095. else {
  4096. /*
  4097. * If bucketsize is not a power-of-two, we may free
  4098. * some pages at the end of hash table which
  4099. * alloc_pages_exact() automatically does
  4100. */
  4101. if (get_order(size) < MAX_ORDER) {
  4102. table = alloc_pages_exact(size, GFP_ATOMIC);
  4103. kmemleak_alloc(table, size, 1, GFP_ATOMIC);
  4104. }
  4105. }
  4106. } while (!table && size > PAGE_SIZE && --log2qty);
  4107. if (!table)
  4108. panic("Failed to allocate %s hash table\n", tablename);
  4109. printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
  4110. tablename,
  4111. (1U << log2qty),
  4112. ilog2(size) - PAGE_SHIFT,
  4113. size);
  4114. if (_hash_shift)
  4115. *_hash_shift = log2qty;
  4116. if (_hash_mask)
  4117. *_hash_mask = (1 << log2qty) - 1;
  4118. return table;
  4119. }
  4120. /* Return a pointer to the bitmap storing bits affecting a block of pages */
  4121. static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
  4122. unsigned long pfn)
  4123. {
  4124. #ifdef CONFIG_SPARSEMEM
  4125. return __pfn_to_section(pfn)->pageblock_flags;
  4126. #else
  4127. return zone->pageblock_flags;
  4128. #endif /* CONFIG_SPARSEMEM */
  4129. }
  4130. static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
  4131. {
  4132. #ifdef CONFIG_SPARSEMEM
  4133. pfn &= (PAGES_PER_SECTION-1);
  4134. return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
  4135. #else
  4136. pfn = pfn - zone->zone_start_pfn;
  4137. return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
  4138. #endif /* CONFIG_SPARSEMEM */
  4139. }
  4140. /**
  4141. * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
  4142. * @page: The page within the block of interest
  4143. * @start_bitidx: The first bit of interest to retrieve
  4144. * @end_bitidx: The last bit of interest
  4145. * returns pageblock_bits flags
  4146. */
  4147. unsigned long get_pageblock_flags_group(struct page *page,
  4148. int start_bitidx, int end_bitidx)
  4149. {
  4150. struct zone *zone;
  4151. unsigned long *bitmap;
  4152. unsigned long pfn, bitidx;
  4153. unsigned long flags = 0;
  4154. unsigned long value = 1;
  4155. zone = page_zone(page);
  4156. pfn = page_to_pfn(page);
  4157. bitmap = get_pageblock_bitmap(zone, pfn);
  4158. bitidx = pfn_to_bitidx(zone, pfn);
  4159. for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
  4160. if (test_bit(bitidx + start_bitidx, bitmap))
  4161. flags |= value;
  4162. return flags;
  4163. }
  4164. /**
  4165. * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
  4166. * @page: The page within the block of interest
  4167. * @start_bitidx: The first bit of interest
  4168. * @end_bitidx: The last bit of interest
  4169. * @flags: The flags to set
  4170. */
  4171. void set_pageblock_flags_group(struct page *page, unsigned long flags,
  4172. int start_bitidx, int end_bitidx)
  4173. {
  4174. struct zone *zone;
  4175. unsigned long *bitmap;
  4176. unsigned long pfn, bitidx;
  4177. unsigned long value = 1;
  4178. zone = page_zone(page);
  4179. pfn = page_to_pfn(page);
  4180. bitmap = get_pageblock_bitmap(zone, pfn);
  4181. bitidx = pfn_to_bitidx(zone, pfn);
  4182. VM_BUG_ON(pfn < zone->zone_start_pfn);
  4183. VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
  4184. for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
  4185. if (flags & value)
  4186. __set_bit(bitidx + start_bitidx, bitmap);
  4187. else
  4188. __clear_bit(bitidx + start_bitidx, bitmap);
  4189. }
  4190. /*
  4191. * This is designed as sub function...plz see page_isolation.c also.
  4192. * set/clear page block's type to be ISOLATE.
  4193. * page allocater never alloc memory from ISOLATE block.
  4194. */
  4195. int set_migratetype_isolate(struct page *page)
  4196. {
  4197. struct zone *zone;
  4198. unsigned long flags;
  4199. int ret = -EBUSY;
  4200. zone = page_zone(page);
  4201. spin_lock_irqsave(&zone->lock, flags);
  4202. /*
  4203. * In future, more migrate types will be able to be isolation target.
  4204. */
  4205. if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
  4206. goto out;
  4207. set_pageblock_migratetype(page, MIGRATE_ISOLATE);
  4208. move_freepages_block(zone, page, MIGRATE_ISOLATE);
  4209. ret = 0;
  4210. out:
  4211. spin_unlock_irqrestore(&zone->lock, flags);
  4212. if (!ret)
  4213. drain_all_pages();
  4214. return ret;
  4215. }
  4216. void unset_migratetype_isolate(struct page *page)
  4217. {
  4218. struct zone *zone;
  4219. unsigned long flags;
  4220. zone = page_zone(page);
  4221. spin_lock_irqsave(&zone->lock, flags);
  4222. if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
  4223. goto out;
  4224. set_pageblock_migratetype(page, MIGRATE_MOVABLE);
  4225. move_freepages_block(zone, page, MIGRATE_MOVABLE);
  4226. out:
  4227. spin_unlock_irqrestore(&zone->lock, flags);
  4228. }
  4229. #ifdef CONFIG_MEMORY_HOTREMOVE
  4230. /*
  4231. * All pages in the range must be isolated before calling this.
  4232. */
  4233. void
  4234. __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
  4235. {
  4236. struct page *page;
  4237. struct zone *zone;
  4238. int order, i;
  4239. unsigned long pfn;
  4240. unsigned long flags;
  4241. /* find the first valid pfn */
  4242. for (pfn = start_pfn; pfn < end_pfn; pfn++)
  4243. if (pfn_valid(pfn))
  4244. break;
  4245. if (pfn == end_pfn)
  4246. return;
  4247. zone = page_zone(pfn_to_page(pfn));
  4248. spin_lock_irqsave(&zone->lock, flags);
  4249. pfn = start_pfn;
  4250. while (pfn < end_pfn) {
  4251. if (!pfn_valid(pfn)) {
  4252. pfn++;
  4253. continue;
  4254. }
  4255. page = pfn_to_page(pfn);
  4256. BUG_ON(page_count(page));
  4257. BUG_ON(!PageBuddy(page));
  4258. order = page_order(page);
  4259. #ifdef CONFIG_DEBUG_VM
  4260. printk(KERN_INFO "remove from free list %lx %d %lx\n",
  4261. pfn, 1 << order, end_pfn);
  4262. #endif
  4263. list_del(&page->lru);
  4264. rmv_page_order(page);
  4265. zone->free_area[order].nr_free--;
  4266. __mod_zone_page_state(zone, NR_FREE_PAGES,
  4267. - (1UL << order));
  4268. for (i = 0; i < (1 << order); i++)
  4269. SetPageReserved((page+i));
  4270. pfn += (1 << order);
  4271. }
  4272. spin_unlock_irqrestore(&zone->lock, flags);
  4273. }
  4274. #endif