page_alloc.c 154 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019202020212022202320242025202620272028202920302031203220332034203520362037203820392040204120422043204420452046204720482049205020512052205320542055205620572058205920602061206220632064206520662067206820692070207120722073207420752076207720782079208020812082208320842085208620872088208920902091209220932094209520962097209820992100210121022103210421052106210721082109211021112112211321142115211621172118211921202121212221232124212521262127212821292130213121322133213421352136213721382139214021412142214321442145214621472148214921502151215221532154215521562157215821592160216121622163216421652166216721682169217021712172217321742175217621772178217921802181218221832184218521862187218821892190219121922193219421952196219721982199220022012202220322042205220622072208220922102211221222132214221522162217221822192220222122222223222422252226222722282229223022312232223322342235223622372238223922402241224222432244224522462247224822492250225122522253225422552256225722582259226022612262226322642265226622672268226922702271227222732274227522762277227822792280228122822283228422852286228722882289229022912292229322942295229622972298229923002301230223032304230523062307230823092310231123122313231423152316231723182319232023212322232323242325232623272328232923302331233223332334233523362337233823392340234123422343234423452346234723482349235023512352235323542355235623572358235923602361236223632364236523662367236823692370237123722373237423752376237723782379238023812382238323842385238623872388238923902391239223932394239523962397239823992400240124022403240424052406240724082409241024112412241324142415241624172418241924202421242224232424242524262427242824292430243124322433243424352436243724382439244024412442244324442445244624472448244924502451245224532454245524562457245824592460246124622463246424652466246724682469247024712472247324742475247624772478247924802481248224832484248524862487248824892490249124922493249424952496249724982499250025012502250325042505250625072508250925102511251225132514251525162517251825192520252125222523252425252526252725282529253025312532253325342535253625372538253925402541254225432544254525462547254825492550255125522553255425552556255725582559256025612562256325642565256625672568256925702571257225732574257525762577257825792580258125822583258425852586258725882589259025912592259325942595259625972598259926002601260226032604260526062607260826092610261126122613261426152616261726182619262026212622262326242625262626272628262926302631263226332634263526362637263826392640264126422643264426452646264726482649265026512652265326542655265626572658265926602661266226632664266526662667266826692670267126722673267426752676267726782679268026812682268326842685268626872688268926902691269226932694269526962697269826992700270127022703270427052706270727082709271027112712271327142715271627172718271927202721272227232724272527262727272827292730273127322733273427352736273727382739274027412742274327442745274627472748274927502751275227532754275527562757275827592760276127622763276427652766276727682769277027712772277327742775277627772778277927802781278227832784278527862787278827892790279127922793279427952796279727982799280028012802280328042805280628072808280928102811281228132814281528162817281828192820282128222823282428252826282728282829283028312832283328342835283628372838283928402841284228432844284528462847284828492850285128522853285428552856285728582859286028612862286328642865286628672868286928702871287228732874287528762877287828792880288128822883288428852886288728882889289028912892289328942895289628972898289929002901290229032904290529062907290829092910291129122913291429152916291729182919292029212922292329242925292629272928292929302931293229332934293529362937293829392940294129422943294429452946294729482949295029512952295329542955295629572958295929602961296229632964296529662967296829692970297129722973297429752976297729782979298029812982298329842985298629872988298929902991299229932994299529962997299829993000300130023003300430053006300730083009301030113012301330143015301630173018301930203021302230233024302530263027302830293030303130323033303430353036303730383039304030413042304330443045304630473048304930503051305230533054305530563057305830593060306130623063306430653066306730683069307030713072307330743075307630773078307930803081308230833084308530863087308830893090309130923093309430953096309730983099310031013102310331043105310631073108310931103111311231133114311531163117311831193120312131223123312431253126312731283129313031313132313331343135313631373138313931403141314231433144314531463147314831493150315131523153315431553156315731583159316031613162316331643165316631673168316931703171317231733174317531763177317831793180318131823183318431853186318731883189319031913192319331943195319631973198319932003201320232033204320532063207320832093210321132123213321432153216321732183219322032213222322332243225322632273228322932303231323232333234323532363237323832393240324132423243324432453246324732483249325032513252325332543255325632573258325932603261326232633264326532663267326832693270327132723273327432753276327732783279328032813282328332843285328632873288328932903291329232933294329532963297329832993300330133023303330433053306330733083309331033113312331333143315331633173318331933203321332233233324332533263327332833293330333133323333333433353336333733383339334033413342334333443345334633473348334933503351335233533354335533563357335833593360336133623363336433653366336733683369337033713372337333743375337633773378337933803381338233833384338533863387338833893390339133923393339433953396339733983399340034013402340334043405340634073408340934103411341234133414341534163417341834193420342134223423342434253426342734283429343034313432343334343435343634373438343934403441344234433444344534463447344834493450345134523453345434553456345734583459346034613462346334643465346634673468346934703471347234733474347534763477347834793480348134823483348434853486348734883489349034913492349334943495349634973498349935003501350235033504350535063507350835093510351135123513351435153516351735183519352035213522352335243525352635273528352935303531353235333534353535363537353835393540354135423543354435453546354735483549355035513552355335543555355635573558355935603561356235633564356535663567356835693570357135723573357435753576357735783579358035813582358335843585358635873588358935903591359235933594359535963597359835993600360136023603360436053606360736083609361036113612361336143615361636173618361936203621362236233624362536263627362836293630363136323633363436353636363736383639364036413642364336443645364636473648364936503651365236533654365536563657365836593660366136623663366436653666366736683669367036713672367336743675367636773678367936803681368236833684368536863687368836893690369136923693369436953696369736983699370037013702370337043705370637073708370937103711371237133714371537163717371837193720372137223723372437253726372737283729373037313732373337343735373637373738373937403741374237433744374537463747374837493750375137523753375437553756375737583759376037613762376337643765376637673768376937703771377237733774377537763777377837793780378137823783378437853786378737883789379037913792379337943795379637973798379938003801380238033804380538063807380838093810381138123813381438153816381738183819382038213822382338243825382638273828382938303831383238333834383538363837383838393840384138423843384438453846384738483849385038513852385338543855385638573858385938603861386238633864386538663867386838693870387138723873387438753876387738783879388038813882388338843885388638873888388938903891389238933894389538963897389838993900390139023903390439053906390739083909391039113912391339143915391639173918391939203921392239233924392539263927392839293930393139323933393439353936393739383939394039413942394339443945394639473948394939503951395239533954395539563957395839593960396139623963396439653966396739683969397039713972397339743975397639773978397939803981398239833984398539863987398839893990399139923993399439953996399739983999400040014002400340044005400640074008400940104011401240134014401540164017401840194020402140224023402440254026402740284029403040314032403340344035403640374038403940404041404240434044404540464047404840494050405140524053405440554056405740584059406040614062406340644065406640674068406940704071407240734074407540764077407840794080408140824083408440854086408740884089409040914092409340944095409640974098409941004101410241034104410541064107410841094110411141124113411441154116411741184119412041214122412341244125412641274128412941304131413241334134413541364137413841394140414141424143414441454146414741484149415041514152415341544155415641574158415941604161416241634164416541664167416841694170417141724173417441754176417741784179418041814182418341844185418641874188418941904191419241934194419541964197419841994200420142024203420442054206420742084209421042114212421342144215421642174218421942204221422242234224422542264227422842294230423142324233423442354236423742384239424042414242424342444245424642474248424942504251425242534254425542564257425842594260426142624263426442654266426742684269427042714272427342744275427642774278427942804281428242834284428542864287428842894290429142924293429442954296429742984299430043014302430343044305430643074308430943104311431243134314431543164317431843194320432143224323432443254326432743284329433043314332433343344335433643374338433943404341434243434344434543464347434843494350435143524353435443554356435743584359436043614362436343644365436643674368436943704371437243734374437543764377437843794380438143824383438443854386438743884389439043914392439343944395439643974398439944004401440244034404440544064407440844094410441144124413441444154416441744184419442044214422442344244425442644274428442944304431443244334434443544364437443844394440444144424443444444454446444744484449445044514452445344544455445644574458445944604461446244634464446544664467446844694470447144724473447444754476447744784479448044814482448344844485448644874488448944904491449244934494449544964497449844994500450145024503450445054506450745084509451045114512451345144515451645174518451945204521452245234524452545264527452845294530453145324533453445354536453745384539454045414542454345444545454645474548454945504551455245534554455545564557455845594560456145624563456445654566456745684569457045714572457345744575457645774578457945804581458245834584458545864587458845894590459145924593459445954596459745984599460046014602460346044605460646074608460946104611461246134614461546164617461846194620462146224623462446254626462746284629463046314632463346344635463646374638463946404641464246434644464546464647464846494650465146524653465446554656465746584659466046614662466346644665466646674668466946704671467246734674467546764677467846794680468146824683468446854686468746884689469046914692469346944695469646974698469947004701470247034704470547064707470847094710471147124713471447154716471747184719472047214722472347244725472647274728472947304731473247334734473547364737473847394740474147424743474447454746474747484749475047514752475347544755475647574758475947604761476247634764476547664767476847694770477147724773477447754776477747784779478047814782478347844785478647874788478947904791479247934794479547964797479847994800480148024803480448054806480748084809481048114812481348144815481648174818481948204821482248234824482548264827482848294830483148324833483448354836483748384839484048414842484348444845484648474848484948504851485248534854485548564857485848594860486148624863486448654866486748684869487048714872487348744875487648774878487948804881488248834884488548864887488848894890489148924893489448954896489748984899490049014902490349044905490649074908490949104911491249134914491549164917491849194920492149224923492449254926492749284929493049314932493349344935493649374938493949404941494249434944494549464947494849494950495149524953495449554956495749584959496049614962496349644965496649674968496949704971497249734974497549764977497849794980498149824983498449854986498749884989499049914992499349944995499649974998499950005001500250035004500550065007500850095010501150125013501450155016501750185019502050215022502350245025502650275028502950305031503250335034503550365037503850395040504150425043504450455046504750485049505050515052505350545055505650575058505950605061506250635064506550665067506850695070507150725073507450755076507750785079508050815082508350845085508650875088508950905091509250935094509550965097509850995100510151025103510451055106510751085109511051115112511351145115511651175118511951205121512251235124512551265127512851295130513151325133513451355136513751385139514051415142514351445145514651475148514951505151515251535154515551565157515851595160516151625163516451655166516751685169517051715172517351745175517651775178517951805181518251835184518551865187518851895190519151925193519451955196519751985199520052015202520352045205520652075208520952105211521252135214521552165217521852195220522152225223522452255226522752285229523052315232523352345235523652375238523952405241524252435244524552465247524852495250525152525253525452555256525752585259526052615262526352645265526652675268526952705271527252735274527552765277527852795280528152825283528452855286528752885289529052915292529352945295529652975298529953005301530253035304530553065307530853095310531153125313531453155316531753185319532053215322532353245325532653275328532953305331533253335334533553365337533853395340534153425343534453455346534753485349535053515352535353545355535653575358535953605361536253635364536553665367536853695370537153725373537453755376537753785379538053815382538353845385538653875388538953905391539253935394539553965397539853995400540154025403540454055406540754085409541054115412541354145415541654175418541954205421542254235424542554265427542854295430543154325433543454355436543754385439544054415442544354445445544654475448544954505451545254535454545554565457545854595460546154625463546454655466546754685469547054715472547354745475547654775478547954805481548254835484548554865487548854895490549154925493549454955496549754985499550055015502550355045505550655075508550955105511551255135514551555165517551855195520552155225523552455255526552755285529553055315532553355345535553655375538553955405541554255435544554555465547554855495550555155525553555455555556555755585559556055615562556355645565556655675568556955705571557255735574557555765577557855795580558155825583558455855586558755885589559055915592559355945595559655975598559956005601560256035604560556065607560856095610561156125613561456155616561756185619562056215622562356245625562656275628
  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/memblock.h>
  24. #include <linux/compiler.h>
  25. #include <linux/kernel.h>
  26. #include <linux/kmemcheck.h>
  27. #include <linux/module.h>
  28. #include <linux/suspend.h>
  29. #include <linux/pagevec.h>
  30. #include <linux/blkdev.h>
  31. #include <linux/slab.h>
  32. #include <linux/oom.h>
  33. #include <linux/notifier.h>
  34. #include <linux/topology.h>
  35. #include <linux/sysctl.h>
  36. #include <linux/cpu.h>
  37. #include <linux/cpuset.h>
  38. #include <linux/memory_hotplug.h>
  39. #include <linux/nodemask.h>
  40. #include <linux/vmalloc.h>
  41. #include <linux/mempolicy.h>
  42. #include <linux/stop_machine.h>
  43. #include <linux/sort.h>
  44. #include <linux/pfn.h>
  45. #include <linux/backing-dev.h>
  46. #include <linux/fault-inject.h>
  47. #include <linux/page-isolation.h>
  48. #include <linux/page_cgroup.h>
  49. #include <linux/debugobjects.h>
  50. #include <linux/kmemleak.h>
  51. #include <linux/memory.h>
  52. #include <linux/compaction.h>
  53. #include <trace/events/kmem.h>
  54. #include <linux/ftrace_event.h>
  55. #include <asm/tlbflush.h>
  56. #include <asm/div64.h>
  57. #include "internal.h"
  58. #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
  59. DEFINE_PER_CPU(int, numa_node);
  60. EXPORT_PER_CPU_SYMBOL(numa_node);
  61. #endif
  62. #ifdef CONFIG_HAVE_MEMORYLESS_NODES
  63. /*
  64. * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
  65. * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
  66. * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
  67. * defined in <linux/topology.h>.
  68. */
  69. DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
  70. EXPORT_PER_CPU_SYMBOL(_numa_mem_);
  71. #endif
  72. /*
  73. * Array of node states.
  74. */
  75. nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
  76. [N_POSSIBLE] = NODE_MASK_ALL,
  77. [N_ONLINE] = { { [0] = 1UL } },
  78. #ifndef CONFIG_NUMA
  79. [N_NORMAL_MEMORY] = { { [0] = 1UL } },
  80. #ifdef CONFIG_HIGHMEM
  81. [N_HIGH_MEMORY] = { { [0] = 1UL } },
  82. #endif
  83. [N_CPU] = { { [0] = 1UL } },
  84. #endif /* NUMA */
  85. };
  86. EXPORT_SYMBOL(node_states);
  87. unsigned long totalram_pages __read_mostly;
  88. unsigned long totalreserve_pages __read_mostly;
  89. int percpu_pagelist_fraction;
  90. gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
  91. #ifdef CONFIG_PM_SLEEP
  92. /*
  93. * The following functions are used by the suspend/hibernate code to temporarily
  94. * change gfp_allowed_mask in order to avoid using I/O during memory allocations
  95. * while devices are suspended. To avoid races with the suspend/hibernate code,
  96. * they should always be called with pm_mutex held (gfp_allowed_mask also should
  97. * only be modified with pm_mutex held, unless the suspend/hibernate code is
  98. * guaranteed not to run in parallel with that modification).
  99. */
  100. static gfp_t saved_gfp_mask;
  101. void pm_restore_gfp_mask(void)
  102. {
  103. WARN_ON(!mutex_is_locked(&pm_mutex));
  104. if (saved_gfp_mask) {
  105. gfp_allowed_mask = saved_gfp_mask;
  106. saved_gfp_mask = 0;
  107. }
  108. }
  109. void pm_restrict_gfp_mask(void)
  110. {
  111. WARN_ON(!mutex_is_locked(&pm_mutex));
  112. WARN_ON(saved_gfp_mask);
  113. saved_gfp_mask = gfp_allowed_mask;
  114. gfp_allowed_mask &= ~GFP_IOFS;
  115. }
  116. #endif /* CONFIG_PM_SLEEP */
  117. #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
  118. int pageblock_order __read_mostly;
  119. #endif
  120. static void __free_pages_ok(struct page *page, unsigned int order);
  121. /*
  122. * results with 256, 32 in the lowmem_reserve sysctl:
  123. * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
  124. * 1G machine -> (16M dma, 784M normal, 224M high)
  125. * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
  126. * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
  127. * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
  128. *
  129. * TBD: should special case ZONE_DMA32 machines here - in those we normally
  130. * don't need any ZONE_NORMAL reservation
  131. */
  132. int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
  133. #ifdef CONFIG_ZONE_DMA
  134. 256,
  135. #endif
  136. #ifdef CONFIG_ZONE_DMA32
  137. 256,
  138. #endif
  139. #ifdef CONFIG_HIGHMEM
  140. 32,
  141. #endif
  142. 32,
  143. };
  144. EXPORT_SYMBOL(totalram_pages);
  145. static char * const zone_names[MAX_NR_ZONES] = {
  146. #ifdef CONFIG_ZONE_DMA
  147. "DMA",
  148. #endif
  149. #ifdef CONFIG_ZONE_DMA32
  150. "DMA32",
  151. #endif
  152. "Normal",
  153. #ifdef CONFIG_HIGHMEM
  154. "HighMem",
  155. #endif
  156. "Movable",
  157. };
  158. int min_free_kbytes = 1024;
  159. static unsigned long __meminitdata nr_kernel_pages;
  160. static unsigned long __meminitdata nr_all_pages;
  161. static unsigned long __meminitdata dma_reserve;
  162. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  163. /*
  164. * MAX_ACTIVE_REGIONS determines the maximum number of distinct
  165. * ranges of memory (RAM) that may be registered with add_active_range().
  166. * Ranges passed to add_active_range() will be merged if possible
  167. * so the number of times add_active_range() can be called is
  168. * related to the number of nodes and the number of holes
  169. */
  170. #ifdef CONFIG_MAX_ACTIVE_REGIONS
  171. /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
  172. #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
  173. #else
  174. #if MAX_NUMNODES >= 32
  175. /* If there can be many nodes, allow up to 50 holes per node */
  176. #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
  177. #else
  178. /* By default, allow up to 256 distinct regions */
  179. #define MAX_ACTIVE_REGIONS 256
  180. #endif
  181. #endif
  182. static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
  183. static int __meminitdata nr_nodemap_entries;
  184. static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
  185. static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
  186. static unsigned long __initdata required_kernelcore;
  187. static unsigned long __initdata required_movablecore;
  188. static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
  189. /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
  190. int movable_zone;
  191. EXPORT_SYMBOL(movable_zone);
  192. #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
  193. #if MAX_NUMNODES > 1
  194. int nr_node_ids __read_mostly = MAX_NUMNODES;
  195. int nr_online_nodes __read_mostly = 1;
  196. EXPORT_SYMBOL(nr_node_ids);
  197. EXPORT_SYMBOL(nr_online_nodes);
  198. #endif
  199. int page_group_by_mobility_disabled __read_mostly;
  200. static void set_pageblock_migratetype(struct page *page, int migratetype)
  201. {
  202. if (unlikely(page_group_by_mobility_disabled))
  203. migratetype = MIGRATE_UNMOVABLE;
  204. set_pageblock_flags_group(page, (unsigned long)migratetype,
  205. PB_migrate, PB_migrate_end);
  206. }
  207. bool oom_killer_disabled __read_mostly;
  208. #ifdef CONFIG_DEBUG_VM
  209. static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
  210. {
  211. int ret = 0;
  212. unsigned seq;
  213. unsigned long pfn = page_to_pfn(page);
  214. do {
  215. seq = zone_span_seqbegin(zone);
  216. if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
  217. ret = 1;
  218. else if (pfn < zone->zone_start_pfn)
  219. ret = 1;
  220. } while (zone_span_seqretry(zone, seq));
  221. return ret;
  222. }
  223. static int page_is_consistent(struct zone *zone, struct page *page)
  224. {
  225. if (!pfn_valid_within(page_to_pfn(page)))
  226. return 0;
  227. if (zone != page_zone(page))
  228. return 0;
  229. return 1;
  230. }
  231. /*
  232. * Temporary debugging check for pages not lying within a given zone.
  233. */
  234. static int bad_range(struct zone *zone, struct page *page)
  235. {
  236. if (page_outside_zone_boundaries(zone, page))
  237. return 1;
  238. if (!page_is_consistent(zone, page))
  239. return 1;
  240. return 0;
  241. }
  242. #else
  243. static inline int bad_range(struct zone *zone, struct page *page)
  244. {
  245. return 0;
  246. }
  247. #endif
  248. static void bad_page(struct page *page)
  249. {
  250. static unsigned long resume;
  251. static unsigned long nr_shown;
  252. static unsigned long nr_unshown;
  253. /* Don't complain about poisoned pages */
  254. if (PageHWPoison(page)) {
  255. __ClearPageBuddy(page);
  256. return;
  257. }
  258. /*
  259. * Allow a burst of 60 reports, then keep quiet for that minute;
  260. * or allow a steady drip of one report per second.
  261. */
  262. if (nr_shown == 60) {
  263. if (time_before(jiffies, resume)) {
  264. nr_unshown++;
  265. goto out;
  266. }
  267. if (nr_unshown) {
  268. printk(KERN_ALERT
  269. "BUG: Bad page state: %lu messages suppressed\n",
  270. nr_unshown);
  271. nr_unshown = 0;
  272. }
  273. nr_shown = 0;
  274. }
  275. if (nr_shown++ == 0)
  276. resume = jiffies + 60 * HZ;
  277. printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n",
  278. current->comm, page_to_pfn(page));
  279. dump_page(page);
  280. dump_stack();
  281. out:
  282. /* Leave bad fields for debug, except PageBuddy could make trouble */
  283. __ClearPageBuddy(page);
  284. add_taint(TAINT_BAD_PAGE);
  285. }
  286. /*
  287. * Higher-order pages are called "compound pages". They are structured thusly:
  288. *
  289. * The first PAGE_SIZE page is called the "head page".
  290. *
  291. * The remaining PAGE_SIZE pages are called "tail pages".
  292. *
  293. * All pages have PG_compound set. All pages have their ->private pointing at
  294. * the head page (even the head page has this).
  295. *
  296. * The first tail page's ->lru.next holds the address of the compound page's
  297. * put_page() function. Its ->lru.prev holds the order of allocation.
  298. * This usage means that zero-order pages may not be compound.
  299. */
  300. static void free_compound_page(struct page *page)
  301. {
  302. __free_pages_ok(page, compound_order(page));
  303. }
  304. void prep_compound_page(struct page *page, unsigned long order)
  305. {
  306. int i;
  307. int nr_pages = 1 << order;
  308. set_compound_page_dtor(page, free_compound_page);
  309. set_compound_order(page, order);
  310. __SetPageHead(page);
  311. for (i = 1; i < nr_pages; i++) {
  312. struct page *p = page + i;
  313. __SetPageTail(p);
  314. p->first_page = page;
  315. }
  316. }
  317. /* update __split_huge_page_refcount if you change this function */
  318. static int destroy_compound_page(struct page *page, unsigned long order)
  319. {
  320. int i;
  321. int nr_pages = 1 << order;
  322. int bad = 0;
  323. if (unlikely(compound_order(page) != order) ||
  324. unlikely(!PageHead(page))) {
  325. bad_page(page);
  326. bad++;
  327. }
  328. __ClearPageHead(page);
  329. for (i = 1; i < nr_pages; i++) {
  330. struct page *p = page + i;
  331. if (unlikely(!PageTail(p) || (p->first_page != page))) {
  332. bad_page(page);
  333. bad++;
  334. }
  335. __ClearPageTail(p);
  336. }
  337. return bad;
  338. }
  339. static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
  340. {
  341. int i;
  342. /*
  343. * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
  344. * and __GFP_HIGHMEM from hard or soft interrupt context.
  345. */
  346. VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
  347. for (i = 0; i < (1 << order); i++)
  348. clear_highpage(page + i);
  349. }
  350. static inline void set_page_order(struct page *page, int order)
  351. {
  352. set_page_private(page, order);
  353. __SetPageBuddy(page);
  354. }
  355. static inline void rmv_page_order(struct page *page)
  356. {
  357. __ClearPageBuddy(page);
  358. set_page_private(page, 0);
  359. }
  360. /*
  361. * Locate the struct page for both the matching buddy in our
  362. * pair (buddy1) and the combined O(n+1) page they form (page).
  363. *
  364. * 1) Any buddy B1 will have an order O twin B2 which satisfies
  365. * the following equation:
  366. * B2 = B1 ^ (1 << O)
  367. * For example, if the starting buddy (buddy2) is #8 its order
  368. * 1 buddy is #10:
  369. * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
  370. *
  371. * 2) Any buddy B will have an order O+1 parent P which
  372. * satisfies the following equation:
  373. * P = B & ~(1 << O)
  374. *
  375. * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
  376. */
  377. static inline unsigned long
  378. __find_buddy_index(unsigned long page_idx, unsigned int order)
  379. {
  380. return page_idx ^ (1 << order);
  381. }
  382. /*
  383. * This function checks whether a page is free && is the buddy
  384. * we can do coalesce a page and its buddy if
  385. * (a) the buddy is not in a hole &&
  386. * (b) the buddy is in the buddy system &&
  387. * (c) a page and its buddy have the same order &&
  388. * (d) a page and its buddy are in the same zone.
  389. *
  390. * For recording whether a page is in the buddy system, we set ->_mapcount -2.
  391. * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
  392. *
  393. * For recording page's order, we use page_private(page).
  394. */
  395. static inline int page_is_buddy(struct page *page, struct page *buddy,
  396. int order)
  397. {
  398. if (!pfn_valid_within(page_to_pfn(buddy)))
  399. return 0;
  400. if (page_zone_id(page) != page_zone_id(buddy))
  401. return 0;
  402. if (PageBuddy(buddy) && page_order(buddy) == order) {
  403. VM_BUG_ON(page_count(buddy) != 0);
  404. return 1;
  405. }
  406. return 0;
  407. }
  408. /*
  409. * Freeing function for a buddy system allocator.
  410. *
  411. * The concept of a buddy system is to maintain direct-mapped table
  412. * (containing bit values) for memory blocks of various "orders".
  413. * The bottom level table contains the map for the smallest allocatable
  414. * units of memory (here, pages), and each level above it describes
  415. * pairs of units from the levels below, hence, "buddies".
  416. * At a high level, all that happens here is marking the table entry
  417. * at the bottom level available, and propagating the changes upward
  418. * as necessary, plus some accounting needed to play nicely with other
  419. * parts of the VM system.
  420. * At each level, we keep a list of pages, which are heads of continuous
  421. * free pages of length of (1 << order) and marked with _mapcount -2. Page's
  422. * order is recorded in page_private(page) field.
  423. * So when we are allocating or freeing one, we can derive the state of the
  424. * other. That is, if we allocate a small block, and both were
  425. * free, the remainder of the region must be split into blocks.
  426. * If a block is freed, and its buddy is also free, then this
  427. * triggers coalescing into a block of larger size.
  428. *
  429. * -- wli
  430. */
  431. static inline void __free_one_page(struct page *page,
  432. struct zone *zone, unsigned int order,
  433. int migratetype)
  434. {
  435. unsigned long page_idx;
  436. unsigned long combined_idx;
  437. unsigned long uninitialized_var(buddy_idx);
  438. struct page *buddy;
  439. if (unlikely(PageCompound(page)))
  440. if (unlikely(destroy_compound_page(page, order)))
  441. return;
  442. VM_BUG_ON(migratetype == -1);
  443. page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
  444. VM_BUG_ON(page_idx & ((1 << order) - 1));
  445. VM_BUG_ON(bad_range(zone, page));
  446. while (order < MAX_ORDER-1) {
  447. buddy_idx = __find_buddy_index(page_idx, order);
  448. buddy = page + (buddy_idx - page_idx);
  449. if (!page_is_buddy(page, buddy, order))
  450. break;
  451. /* Our buddy is free, merge with it and move up one order. */
  452. list_del(&buddy->lru);
  453. zone->free_area[order].nr_free--;
  454. rmv_page_order(buddy);
  455. combined_idx = buddy_idx & page_idx;
  456. page = page + (combined_idx - page_idx);
  457. page_idx = combined_idx;
  458. order++;
  459. }
  460. set_page_order(page, order);
  461. /*
  462. * If this is not the largest possible page, check if the buddy
  463. * of the next-highest order is free. If it is, it's possible
  464. * that pages are being freed that will coalesce soon. In case,
  465. * that is happening, add the free page to the tail of the list
  466. * so it's less likely to be used soon and more likely to be merged
  467. * as a higher order page
  468. */
  469. if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
  470. struct page *higher_page, *higher_buddy;
  471. combined_idx = buddy_idx & page_idx;
  472. higher_page = page + (combined_idx - page_idx);
  473. buddy_idx = __find_buddy_index(combined_idx, order + 1);
  474. higher_buddy = page + (buddy_idx - combined_idx);
  475. if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
  476. list_add_tail(&page->lru,
  477. &zone->free_area[order].free_list[migratetype]);
  478. goto out;
  479. }
  480. }
  481. list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
  482. out:
  483. zone->free_area[order].nr_free++;
  484. }
  485. /*
  486. * free_page_mlock() -- clean up attempts to free and mlocked() page.
  487. * Page should not be on lru, so no need to fix that up.
  488. * free_pages_check() will verify...
  489. */
  490. static inline void free_page_mlock(struct page *page)
  491. {
  492. __dec_zone_page_state(page, NR_MLOCK);
  493. __count_vm_event(UNEVICTABLE_MLOCKFREED);
  494. }
  495. static inline int free_pages_check(struct page *page)
  496. {
  497. if (unlikely(page_mapcount(page) |
  498. (page->mapping != NULL) |
  499. (atomic_read(&page->_count) != 0) |
  500. (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
  501. bad_page(page);
  502. return 1;
  503. }
  504. if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
  505. page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
  506. return 0;
  507. }
  508. /*
  509. * Frees a number of pages from the PCP lists
  510. * Assumes all pages on list are in same zone, and of same order.
  511. * count is the number of pages to free.
  512. *
  513. * If the zone was previously in an "all pages pinned" state then look to
  514. * see if this freeing clears that state.
  515. *
  516. * And clear the zone's pages_scanned counter, to hold off the "all pages are
  517. * pinned" detection logic.
  518. */
  519. static void free_pcppages_bulk(struct zone *zone, int count,
  520. struct per_cpu_pages *pcp)
  521. {
  522. int migratetype = 0;
  523. int batch_free = 0;
  524. int to_free = count;
  525. spin_lock(&zone->lock);
  526. zone->all_unreclaimable = 0;
  527. zone->pages_scanned = 0;
  528. while (to_free) {
  529. struct page *page;
  530. struct list_head *list;
  531. /*
  532. * Remove pages from lists in a round-robin fashion. A
  533. * batch_free count is maintained that is incremented when an
  534. * empty list is encountered. This is so more pages are freed
  535. * off fuller lists instead of spinning excessively around empty
  536. * lists
  537. */
  538. do {
  539. batch_free++;
  540. if (++migratetype == MIGRATE_PCPTYPES)
  541. migratetype = 0;
  542. list = &pcp->lists[migratetype];
  543. } while (list_empty(list));
  544. do {
  545. page = list_entry(list->prev, struct page, lru);
  546. /* must delete as __free_one_page list manipulates */
  547. list_del(&page->lru);
  548. /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
  549. __free_one_page(page, zone, 0, page_private(page));
  550. trace_mm_page_pcpu_drain(page, 0, page_private(page));
  551. } while (--to_free && --batch_free && !list_empty(list));
  552. }
  553. __mod_zone_page_state(zone, NR_FREE_PAGES, count);
  554. spin_unlock(&zone->lock);
  555. }
  556. static void free_one_page(struct zone *zone, struct page *page, int order,
  557. int migratetype)
  558. {
  559. spin_lock(&zone->lock);
  560. zone->all_unreclaimable = 0;
  561. zone->pages_scanned = 0;
  562. __free_one_page(page, zone, order, migratetype);
  563. __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order);
  564. spin_unlock(&zone->lock);
  565. }
  566. static bool free_pages_prepare(struct page *page, unsigned int order)
  567. {
  568. int i;
  569. int bad = 0;
  570. trace_mm_page_free_direct(page, order);
  571. kmemcheck_free_shadow(page, order);
  572. if (PageAnon(page))
  573. page->mapping = NULL;
  574. for (i = 0; i < (1 << order); i++)
  575. bad += free_pages_check(page + i);
  576. if (bad)
  577. return false;
  578. if (!PageHighMem(page)) {
  579. debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
  580. debug_check_no_obj_freed(page_address(page),
  581. PAGE_SIZE << order);
  582. }
  583. arch_free_page(page, order);
  584. kernel_map_pages(page, 1 << order, 0);
  585. return true;
  586. }
  587. static void __free_pages_ok(struct page *page, unsigned int order)
  588. {
  589. unsigned long flags;
  590. int wasMlocked = __TestClearPageMlocked(page);
  591. if (!free_pages_prepare(page, order))
  592. return;
  593. local_irq_save(flags);
  594. if (unlikely(wasMlocked))
  595. free_page_mlock(page);
  596. __count_vm_events(PGFREE, 1 << order);
  597. free_one_page(page_zone(page), page, order,
  598. get_pageblock_migratetype(page));
  599. local_irq_restore(flags);
  600. }
  601. /*
  602. * permit the bootmem allocator to evade page validation on high-order frees
  603. */
  604. void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
  605. {
  606. if (order == 0) {
  607. __ClearPageReserved(page);
  608. set_page_count(page, 0);
  609. set_page_refcounted(page);
  610. __free_page(page);
  611. } else {
  612. int loop;
  613. prefetchw(page);
  614. for (loop = 0; loop < BITS_PER_LONG; loop++) {
  615. struct page *p = &page[loop];
  616. if (loop + 1 < BITS_PER_LONG)
  617. prefetchw(p + 1);
  618. __ClearPageReserved(p);
  619. set_page_count(p, 0);
  620. }
  621. set_page_refcounted(page);
  622. __free_pages(page, order);
  623. }
  624. }
  625. /*
  626. * The order of subdivision here is critical for the IO subsystem.
  627. * Please do not alter this order without good reasons and regression
  628. * testing. Specifically, as large blocks of memory are subdivided,
  629. * the order in which smaller blocks are delivered depends on the order
  630. * they're subdivided in this function. This is the primary factor
  631. * influencing the order in which pages are delivered to the IO
  632. * subsystem according to empirical testing, and this is also justified
  633. * by considering the behavior of a buddy system containing a single
  634. * large block of memory acted on by a series of small allocations.
  635. * This behavior is a critical factor in sglist merging's success.
  636. *
  637. * -- wli
  638. */
  639. static inline void expand(struct zone *zone, struct page *page,
  640. int low, int high, struct free_area *area,
  641. int migratetype)
  642. {
  643. unsigned long size = 1 << high;
  644. while (high > low) {
  645. area--;
  646. high--;
  647. size >>= 1;
  648. VM_BUG_ON(bad_range(zone, &page[size]));
  649. list_add(&page[size].lru, &area->free_list[migratetype]);
  650. area->nr_free++;
  651. set_page_order(&page[size], high);
  652. }
  653. }
  654. /*
  655. * This page is about to be returned from the page allocator
  656. */
  657. static inline int check_new_page(struct page *page)
  658. {
  659. if (unlikely(page_mapcount(page) |
  660. (page->mapping != NULL) |
  661. (atomic_read(&page->_count) != 0) |
  662. (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
  663. bad_page(page);
  664. return 1;
  665. }
  666. return 0;
  667. }
  668. static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
  669. {
  670. int i;
  671. for (i = 0; i < (1 << order); i++) {
  672. struct page *p = page + i;
  673. if (unlikely(check_new_page(p)))
  674. return 1;
  675. }
  676. set_page_private(page, 0);
  677. set_page_refcounted(page);
  678. arch_alloc_page(page, order);
  679. kernel_map_pages(page, 1 << order, 1);
  680. if (gfp_flags & __GFP_ZERO)
  681. prep_zero_page(page, order, gfp_flags);
  682. if (order && (gfp_flags & __GFP_COMP))
  683. prep_compound_page(page, order);
  684. return 0;
  685. }
  686. /*
  687. * Go through the free lists for the given migratetype and remove
  688. * the smallest available page from the freelists
  689. */
  690. static inline
  691. struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
  692. int migratetype)
  693. {
  694. unsigned int current_order;
  695. struct free_area * area;
  696. struct page *page;
  697. /* Find a page of the appropriate size in the preferred list */
  698. for (current_order = order; current_order < MAX_ORDER; ++current_order) {
  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. list_del(&page->lru);
  705. rmv_page_order(page);
  706. area->nr_free--;
  707. expand(zone, page, order, current_order, area, migratetype);
  708. return page;
  709. }
  710. return NULL;
  711. }
  712. /*
  713. * This array describes the order lists are fallen back to when
  714. * the free lists for the desirable migrate type are depleted
  715. */
  716. static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
  717. [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
  718. [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
  719. [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
  720. [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
  721. };
  722. /*
  723. * Move the free pages in a range to the free lists of the requested type.
  724. * Note that start_page and end_pages are not aligned on a pageblock
  725. * boundary. If alignment is required, use move_freepages_block()
  726. */
  727. static int move_freepages(struct zone *zone,
  728. struct page *start_page, struct page *end_page,
  729. int migratetype)
  730. {
  731. struct page *page;
  732. unsigned long order;
  733. int pages_moved = 0;
  734. #ifndef CONFIG_HOLES_IN_ZONE
  735. /*
  736. * page_zone is not safe to call in this context when
  737. * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
  738. * anyway as we check zone boundaries in move_freepages_block().
  739. * Remove at a later date when no bug reports exist related to
  740. * grouping pages by mobility
  741. */
  742. BUG_ON(page_zone(start_page) != page_zone(end_page));
  743. #endif
  744. for (page = start_page; page <= end_page;) {
  745. /* Make sure we are not inadvertently changing nodes */
  746. VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
  747. if (!pfn_valid_within(page_to_pfn(page))) {
  748. page++;
  749. continue;
  750. }
  751. if (!PageBuddy(page)) {
  752. page++;
  753. continue;
  754. }
  755. order = page_order(page);
  756. list_del(&page->lru);
  757. list_add(&page->lru,
  758. &zone->free_area[order].free_list[migratetype]);
  759. page += 1 << order;
  760. pages_moved += 1 << order;
  761. }
  762. return pages_moved;
  763. }
  764. static int move_freepages_block(struct zone *zone, struct page *page,
  765. int migratetype)
  766. {
  767. unsigned long start_pfn, end_pfn;
  768. struct page *start_page, *end_page;
  769. start_pfn = page_to_pfn(page);
  770. start_pfn = start_pfn & ~(pageblock_nr_pages-1);
  771. start_page = pfn_to_page(start_pfn);
  772. end_page = start_page + pageblock_nr_pages - 1;
  773. end_pfn = start_pfn + pageblock_nr_pages - 1;
  774. /* Do not cross zone boundaries */
  775. if (start_pfn < zone->zone_start_pfn)
  776. start_page = page;
  777. if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
  778. return 0;
  779. return move_freepages(zone, start_page, end_page, migratetype);
  780. }
  781. static void change_pageblock_range(struct page *pageblock_page,
  782. int start_order, int migratetype)
  783. {
  784. int nr_pageblocks = 1 << (start_order - pageblock_order);
  785. while (nr_pageblocks--) {
  786. set_pageblock_migratetype(pageblock_page, migratetype);
  787. pageblock_page += pageblock_nr_pages;
  788. }
  789. }
  790. /* Remove an element from the buddy allocator from the fallback list */
  791. static inline struct page *
  792. __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
  793. {
  794. struct free_area * area;
  795. int current_order;
  796. struct page *page;
  797. int migratetype, i;
  798. /* Find the largest possible block of pages in the other list */
  799. for (current_order = MAX_ORDER-1; current_order >= order;
  800. --current_order) {
  801. for (i = 0; i < MIGRATE_TYPES - 1; i++) {
  802. migratetype = fallbacks[start_migratetype][i];
  803. /* MIGRATE_RESERVE handled later if necessary */
  804. if (migratetype == MIGRATE_RESERVE)
  805. continue;
  806. area = &(zone->free_area[current_order]);
  807. if (list_empty(&area->free_list[migratetype]))
  808. continue;
  809. page = list_entry(area->free_list[migratetype].next,
  810. struct page, lru);
  811. area->nr_free--;
  812. /*
  813. * If breaking a large block of pages, move all free
  814. * pages to the preferred allocation list. If falling
  815. * back for a reclaimable kernel allocation, be more
  816. * agressive about taking ownership of free pages
  817. */
  818. if (unlikely(current_order >= (pageblock_order >> 1)) ||
  819. start_migratetype == MIGRATE_RECLAIMABLE ||
  820. page_group_by_mobility_disabled) {
  821. unsigned long pages;
  822. pages = move_freepages_block(zone, page,
  823. start_migratetype);
  824. /* Claim the whole block if over half of it is free */
  825. if (pages >= (1 << (pageblock_order-1)) ||
  826. page_group_by_mobility_disabled)
  827. set_pageblock_migratetype(page,
  828. start_migratetype);
  829. migratetype = start_migratetype;
  830. }
  831. /* Remove the page from the freelists */
  832. list_del(&page->lru);
  833. rmv_page_order(page);
  834. /* Take ownership for orders >= pageblock_order */
  835. if (current_order >= pageblock_order)
  836. change_pageblock_range(page, current_order,
  837. start_migratetype);
  838. expand(zone, page, order, current_order, area, migratetype);
  839. trace_mm_page_alloc_extfrag(page, order, current_order,
  840. start_migratetype, migratetype);
  841. return page;
  842. }
  843. }
  844. return NULL;
  845. }
  846. /*
  847. * Do the hard work of removing an element from the buddy allocator.
  848. * Call me with the zone->lock already held.
  849. */
  850. static struct page *__rmqueue(struct zone *zone, unsigned int order,
  851. int migratetype)
  852. {
  853. struct page *page;
  854. retry_reserve:
  855. page = __rmqueue_smallest(zone, order, migratetype);
  856. if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
  857. page = __rmqueue_fallback(zone, order, migratetype);
  858. /*
  859. * Use MIGRATE_RESERVE rather than fail an allocation. goto
  860. * is used because __rmqueue_smallest is an inline function
  861. * and we want just one call site
  862. */
  863. if (!page) {
  864. migratetype = MIGRATE_RESERVE;
  865. goto retry_reserve;
  866. }
  867. }
  868. trace_mm_page_alloc_zone_locked(page, order, migratetype);
  869. return page;
  870. }
  871. /*
  872. * Obtain a specified number of elements from the buddy allocator, all under
  873. * a single hold of the lock, for efficiency. Add them to the supplied list.
  874. * Returns the number of new pages which were placed at *list.
  875. */
  876. static int rmqueue_bulk(struct zone *zone, unsigned int order,
  877. unsigned long count, struct list_head *list,
  878. int migratetype, int cold)
  879. {
  880. int i;
  881. spin_lock(&zone->lock);
  882. for (i = 0; i < count; ++i) {
  883. struct page *page = __rmqueue(zone, order, migratetype);
  884. if (unlikely(page == NULL))
  885. break;
  886. /*
  887. * Split buddy pages returned by expand() are received here
  888. * in physical page order. The page is added to the callers and
  889. * list and the list head then moves forward. From the callers
  890. * perspective, the linked list is ordered by page number in
  891. * some conditions. This is useful for IO devices that can
  892. * merge IO requests if the physical pages are ordered
  893. * properly.
  894. */
  895. if (likely(cold == 0))
  896. list_add(&page->lru, list);
  897. else
  898. list_add_tail(&page->lru, list);
  899. set_page_private(page, migratetype);
  900. list = &page->lru;
  901. }
  902. __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
  903. spin_unlock(&zone->lock);
  904. return i;
  905. }
  906. #ifdef CONFIG_NUMA
  907. /*
  908. * Called from the vmstat counter updater to drain pagesets of this
  909. * currently executing processor on remote nodes after they have
  910. * expired.
  911. *
  912. * Note that this function must be called with the thread pinned to
  913. * a single processor.
  914. */
  915. void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
  916. {
  917. unsigned long flags;
  918. int to_drain;
  919. local_irq_save(flags);
  920. if (pcp->count >= pcp->batch)
  921. to_drain = pcp->batch;
  922. else
  923. to_drain = pcp->count;
  924. free_pcppages_bulk(zone, to_drain, pcp);
  925. pcp->count -= to_drain;
  926. local_irq_restore(flags);
  927. }
  928. #endif
  929. /*
  930. * Drain pages of the indicated processor.
  931. *
  932. * The processor must either be the current processor and the
  933. * thread pinned to the current processor or a processor that
  934. * is not online.
  935. */
  936. static void drain_pages(unsigned int cpu)
  937. {
  938. unsigned long flags;
  939. struct zone *zone;
  940. for_each_populated_zone(zone) {
  941. struct per_cpu_pageset *pset;
  942. struct per_cpu_pages *pcp;
  943. local_irq_save(flags);
  944. pset = per_cpu_ptr(zone->pageset, cpu);
  945. pcp = &pset->pcp;
  946. free_pcppages_bulk(zone, pcp->count, pcp);
  947. pcp->count = 0;
  948. local_irq_restore(flags);
  949. }
  950. }
  951. /*
  952. * Spill all of this CPU's per-cpu pages back into the buddy allocator.
  953. */
  954. void drain_local_pages(void *arg)
  955. {
  956. drain_pages(smp_processor_id());
  957. }
  958. /*
  959. * Spill all the per-cpu pages from all CPUs back into the buddy allocator
  960. */
  961. void drain_all_pages(void)
  962. {
  963. on_each_cpu(drain_local_pages, NULL, 1);
  964. }
  965. #ifdef CONFIG_HIBERNATION
  966. void mark_free_pages(struct zone *zone)
  967. {
  968. unsigned long pfn, max_zone_pfn;
  969. unsigned long flags;
  970. int order, t;
  971. struct list_head *curr;
  972. if (!zone->spanned_pages)
  973. return;
  974. spin_lock_irqsave(&zone->lock, flags);
  975. max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
  976. for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
  977. if (pfn_valid(pfn)) {
  978. struct page *page = pfn_to_page(pfn);
  979. if (!swsusp_page_is_forbidden(page))
  980. swsusp_unset_page_free(page);
  981. }
  982. for_each_migratetype_order(order, t) {
  983. list_for_each(curr, &zone->free_area[order].free_list[t]) {
  984. unsigned long i;
  985. pfn = page_to_pfn(list_entry(curr, struct page, lru));
  986. for (i = 0; i < (1UL << order); i++)
  987. swsusp_set_page_free(pfn_to_page(pfn + i));
  988. }
  989. }
  990. spin_unlock_irqrestore(&zone->lock, flags);
  991. }
  992. #endif /* CONFIG_PM */
  993. /*
  994. * Free a 0-order page
  995. * cold == 1 ? free a cold page : free a hot page
  996. */
  997. void free_hot_cold_page(struct page *page, int cold)
  998. {
  999. struct zone *zone = page_zone(page);
  1000. struct per_cpu_pages *pcp;
  1001. unsigned long flags;
  1002. int migratetype;
  1003. int wasMlocked = __TestClearPageMlocked(page);
  1004. if (!free_pages_prepare(page, 0))
  1005. return;
  1006. migratetype = get_pageblock_migratetype(page);
  1007. set_page_private(page, migratetype);
  1008. local_irq_save(flags);
  1009. if (unlikely(wasMlocked))
  1010. free_page_mlock(page);
  1011. __count_vm_event(PGFREE);
  1012. /*
  1013. * We only track unmovable, reclaimable and movable on pcp lists.
  1014. * Free ISOLATE pages back to the allocator because they are being
  1015. * offlined but treat RESERVE as movable pages so we can get those
  1016. * areas back if necessary. Otherwise, we may have to free
  1017. * excessively into the page allocator
  1018. */
  1019. if (migratetype >= MIGRATE_PCPTYPES) {
  1020. if (unlikely(migratetype == MIGRATE_ISOLATE)) {
  1021. free_one_page(zone, page, 0, migratetype);
  1022. goto out;
  1023. }
  1024. migratetype = MIGRATE_MOVABLE;
  1025. }
  1026. pcp = &this_cpu_ptr(zone->pageset)->pcp;
  1027. if (cold)
  1028. list_add_tail(&page->lru, &pcp->lists[migratetype]);
  1029. else
  1030. list_add(&page->lru, &pcp->lists[migratetype]);
  1031. pcp->count++;
  1032. if (pcp->count >= pcp->high) {
  1033. free_pcppages_bulk(zone, pcp->batch, pcp);
  1034. pcp->count -= pcp->batch;
  1035. }
  1036. out:
  1037. local_irq_restore(flags);
  1038. }
  1039. /*
  1040. * split_page takes a non-compound higher-order page, and splits it into
  1041. * n (1<<order) sub-pages: page[0..n]
  1042. * Each sub-page must be freed individually.
  1043. *
  1044. * Note: this is probably too low level an operation for use in drivers.
  1045. * Please consult with lkml before using this in your driver.
  1046. */
  1047. void split_page(struct page *page, unsigned int order)
  1048. {
  1049. int i;
  1050. VM_BUG_ON(PageCompound(page));
  1051. VM_BUG_ON(!page_count(page));
  1052. #ifdef CONFIG_KMEMCHECK
  1053. /*
  1054. * Split shadow pages too, because free(page[0]) would
  1055. * otherwise free the whole shadow.
  1056. */
  1057. if (kmemcheck_page_is_tracked(page))
  1058. split_page(virt_to_page(page[0].shadow), order);
  1059. #endif
  1060. for (i = 1; i < (1 << order); i++)
  1061. set_page_refcounted(page + i);
  1062. }
  1063. /*
  1064. * Similar to split_page except the page is already free. As this is only
  1065. * being used for migration, the migratetype of the block also changes.
  1066. * As this is called with interrupts disabled, the caller is responsible
  1067. * for calling arch_alloc_page() and kernel_map_page() after interrupts
  1068. * are enabled.
  1069. *
  1070. * Note: this is probably too low level an operation for use in drivers.
  1071. * Please consult with lkml before using this in your driver.
  1072. */
  1073. int split_free_page(struct page *page)
  1074. {
  1075. unsigned int order;
  1076. unsigned long watermark;
  1077. struct zone *zone;
  1078. BUG_ON(!PageBuddy(page));
  1079. zone = page_zone(page);
  1080. order = page_order(page);
  1081. /* Obey watermarks as if the page was being allocated */
  1082. watermark = low_wmark_pages(zone) + (1 << order);
  1083. if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
  1084. return 0;
  1085. /* Remove page from free list */
  1086. list_del(&page->lru);
  1087. zone->free_area[order].nr_free--;
  1088. rmv_page_order(page);
  1089. __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order));
  1090. /* Split into individual pages */
  1091. set_page_refcounted(page);
  1092. split_page(page, order);
  1093. if (order >= pageblock_order - 1) {
  1094. struct page *endpage = page + (1 << order) - 1;
  1095. for (; page < endpage; page += pageblock_nr_pages)
  1096. set_pageblock_migratetype(page, MIGRATE_MOVABLE);
  1097. }
  1098. return 1 << order;
  1099. }
  1100. /*
  1101. * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
  1102. * we cheat by calling it from here, in the order > 0 path. Saves a branch
  1103. * or two.
  1104. */
  1105. static inline
  1106. struct page *buffered_rmqueue(struct zone *preferred_zone,
  1107. struct zone *zone, int order, gfp_t gfp_flags,
  1108. int migratetype)
  1109. {
  1110. unsigned long flags;
  1111. struct page *page;
  1112. int cold = !!(gfp_flags & __GFP_COLD);
  1113. again:
  1114. if (likely(order == 0)) {
  1115. struct per_cpu_pages *pcp;
  1116. struct list_head *list;
  1117. local_irq_save(flags);
  1118. pcp = &this_cpu_ptr(zone->pageset)->pcp;
  1119. list = &pcp->lists[migratetype];
  1120. if (list_empty(list)) {
  1121. pcp->count += rmqueue_bulk(zone, 0,
  1122. pcp->batch, list,
  1123. migratetype, cold);
  1124. if (unlikely(list_empty(list)))
  1125. goto failed;
  1126. }
  1127. if (cold)
  1128. page = list_entry(list->prev, struct page, lru);
  1129. else
  1130. page = list_entry(list->next, struct page, lru);
  1131. list_del(&page->lru);
  1132. pcp->count--;
  1133. } else {
  1134. if (unlikely(gfp_flags & __GFP_NOFAIL)) {
  1135. /*
  1136. * __GFP_NOFAIL is not to be used in new code.
  1137. *
  1138. * All __GFP_NOFAIL callers should be fixed so that they
  1139. * properly detect and handle allocation failures.
  1140. *
  1141. * We most definitely don't want callers attempting to
  1142. * allocate greater than order-1 page units with
  1143. * __GFP_NOFAIL.
  1144. */
  1145. WARN_ON_ONCE(order > 1);
  1146. }
  1147. spin_lock_irqsave(&zone->lock, flags);
  1148. page = __rmqueue(zone, order, migratetype);
  1149. spin_unlock(&zone->lock);
  1150. if (!page)
  1151. goto failed;
  1152. __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order));
  1153. }
  1154. __count_zone_vm_events(PGALLOC, zone, 1 << order);
  1155. zone_statistics(preferred_zone, zone);
  1156. local_irq_restore(flags);
  1157. VM_BUG_ON(bad_range(zone, page));
  1158. if (prep_new_page(page, order, gfp_flags))
  1159. goto again;
  1160. return page;
  1161. failed:
  1162. local_irq_restore(flags);
  1163. return NULL;
  1164. }
  1165. /* The ALLOC_WMARK bits are used as an index to zone->watermark */
  1166. #define ALLOC_WMARK_MIN WMARK_MIN
  1167. #define ALLOC_WMARK_LOW WMARK_LOW
  1168. #define ALLOC_WMARK_HIGH WMARK_HIGH
  1169. #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
  1170. /* Mask to get the watermark bits */
  1171. #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
  1172. #define ALLOC_HARDER 0x10 /* try to alloc harder */
  1173. #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
  1174. #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
  1175. #ifdef CONFIG_FAIL_PAGE_ALLOC
  1176. static struct fail_page_alloc_attr {
  1177. struct fault_attr attr;
  1178. u32 ignore_gfp_highmem;
  1179. u32 ignore_gfp_wait;
  1180. u32 min_order;
  1181. #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
  1182. struct dentry *ignore_gfp_highmem_file;
  1183. struct dentry *ignore_gfp_wait_file;
  1184. struct dentry *min_order_file;
  1185. #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
  1186. } fail_page_alloc = {
  1187. .attr = FAULT_ATTR_INITIALIZER,
  1188. .ignore_gfp_wait = 1,
  1189. .ignore_gfp_highmem = 1,
  1190. .min_order = 1,
  1191. };
  1192. static int __init setup_fail_page_alloc(char *str)
  1193. {
  1194. return setup_fault_attr(&fail_page_alloc.attr, str);
  1195. }
  1196. __setup("fail_page_alloc=", setup_fail_page_alloc);
  1197. static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
  1198. {
  1199. if (order < fail_page_alloc.min_order)
  1200. return 0;
  1201. if (gfp_mask & __GFP_NOFAIL)
  1202. return 0;
  1203. if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
  1204. return 0;
  1205. if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
  1206. return 0;
  1207. return should_fail(&fail_page_alloc.attr, 1 << order);
  1208. }
  1209. #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
  1210. static int __init fail_page_alloc_debugfs(void)
  1211. {
  1212. mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
  1213. struct dentry *dir;
  1214. int err;
  1215. err = init_fault_attr_dentries(&fail_page_alloc.attr,
  1216. "fail_page_alloc");
  1217. if (err)
  1218. return err;
  1219. dir = fail_page_alloc.attr.dentries.dir;
  1220. fail_page_alloc.ignore_gfp_wait_file =
  1221. debugfs_create_bool("ignore-gfp-wait", mode, dir,
  1222. &fail_page_alloc.ignore_gfp_wait);
  1223. fail_page_alloc.ignore_gfp_highmem_file =
  1224. debugfs_create_bool("ignore-gfp-highmem", mode, dir,
  1225. &fail_page_alloc.ignore_gfp_highmem);
  1226. fail_page_alloc.min_order_file =
  1227. debugfs_create_u32("min-order", mode, dir,
  1228. &fail_page_alloc.min_order);
  1229. if (!fail_page_alloc.ignore_gfp_wait_file ||
  1230. !fail_page_alloc.ignore_gfp_highmem_file ||
  1231. !fail_page_alloc.min_order_file) {
  1232. err = -ENOMEM;
  1233. debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
  1234. debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
  1235. debugfs_remove(fail_page_alloc.min_order_file);
  1236. cleanup_fault_attr_dentries(&fail_page_alloc.attr);
  1237. }
  1238. return err;
  1239. }
  1240. late_initcall(fail_page_alloc_debugfs);
  1241. #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
  1242. #else /* CONFIG_FAIL_PAGE_ALLOC */
  1243. static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
  1244. {
  1245. return 0;
  1246. }
  1247. #endif /* CONFIG_FAIL_PAGE_ALLOC */
  1248. /*
  1249. * Return true if free pages are above 'mark'. This takes into account the order
  1250. * of the allocation.
  1251. */
  1252. static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
  1253. int classzone_idx, int alloc_flags, long free_pages)
  1254. {
  1255. /* free_pages my go negative - that's OK */
  1256. long min = mark;
  1257. int o;
  1258. free_pages -= (1 << order) + 1;
  1259. if (alloc_flags & ALLOC_HIGH)
  1260. min -= min / 2;
  1261. if (alloc_flags & ALLOC_HARDER)
  1262. min -= min / 4;
  1263. if (free_pages <= min + z->lowmem_reserve[classzone_idx])
  1264. return false;
  1265. for (o = 0; o < order; o++) {
  1266. /* At the next order, this order's pages become unavailable */
  1267. free_pages -= z->free_area[o].nr_free << o;
  1268. /* Require fewer higher order pages to be free */
  1269. min >>= 1;
  1270. if (free_pages <= min)
  1271. return false;
  1272. }
  1273. return true;
  1274. }
  1275. bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
  1276. int classzone_idx, int alloc_flags)
  1277. {
  1278. return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
  1279. zone_page_state(z, NR_FREE_PAGES));
  1280. }
  1281. bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
  1282. int classzone_idx, int alloc_flags)
  1283. {
  1284. long free_pages = zone_page_state(z, NR_FREE_PAGES);
  1285. if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
  1286. free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
  1287. return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
  1288. free_pages);
  1289. }
  1290. #ifdef CONFIG_NUMA
  1291. /*
  1292. * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
  1293. * skip over zones that are not allowed by the cpuset, or that have
  1294. * been recently (in last second) found to be nearly full. See further
  1295. * comments in mmzone.h. Reduces cache footprint of zonelist scans
  1296. * that have to skip over a lot of full or unallowed zones.
  1297. *
  1298. * If the zonelist cache is present in the passed in zonelist, then
  1299. * returns a pointer to the allowed node mask (either the current
  1300. * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
  1301. *
  1302. * If the zonelist cache is not available for this zonelist, does
  1303. * nothing and returns NULL.
  1304. *
  1305. * If the fullzones BITMAP in the zonelist cache is stale (more than
  1306. * a second since last zap'd) then we zap it out (clear its bits.)
  1307. *
  1308. * We hold off even calling zlc_setup, until after we've checked the
  1309. * first zone in the zonelist, on the theory that most allocations will
  1310. * be satisfied from that first zone, so best to examine that zone as
  1311. * quickly as we can.
  1312. */
  1313. static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
  1314. {
  1315. struct zonelist_cache *zlc; /* cached zonelist speedup info */
  1316. nodemask_t *allowednodes; /* zonelist_cache approximation */
  1317. zlc = zonelist->zlcache_ptr;
  1318. if (!zlc)
  1319. return NULL;
  1320. if (time_after(jiffies, zlc->last_full_zap + HZ)) {
  1321. bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
  1322. zlc->last_full_zap = jiffies;
  1323. }
  1324. allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
  1325. &cpuset_current_mems_allowed :
  1326. &node_states[N_HIGH_MEMORY];
  1327. return allowednodes;
  1328. }
  1329. /*
  1330. * Given 'z' scanning a zonelist, run a couple of quick checks to see
  1331. * if it is worth looking at further for free memory:
  1332. * 1) Check that the zone isn't thought to be full (doesn't have its
  1333. * bit set in the zonelist_cache fullzones BITMAP).
  1334. * 2) Check that the zones node (obtained from the zonelist_cache
  1335. * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
  1336. * Return true (non-zero) if zone is worth looking at further, or
  1337. * else return false (zero) if it is not.
  1338. *
  1339. * This check -ignores- the distinction between various watermarks,
  1340. * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
  1341. * found to be full for any variation of these watermarks, it will
  1342. * be considered full for up to one second by all requests, unless
  1343. * we are so low on memory on all allowed nodes that we are forced
  1344. * into the second scan of the zonelist.
  1345. *
  1346. * In the second scan we ignore this zonelist cache and exactly
  1347. * apply the watermarks to all zones, even it is slower to do so.
  1348. * We are low on memory in the second scan, and should leave no stone
  1349. * unturned looking for a free page.
  1350. */
  1351. static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
  1352. nodemask_t *allowednodes)
  1353. {
  1354. struct zonelist_cache *zlc; /* cached zonelist speedup info */
  1355. int i; /* index of *z in zonelist zones */
  1356. int n; /* node that zone *z is on */
  1357. zlc = zonelist->zlcache_ptr;
  1358. if (!zlc)
  1359. return 1;
  1360. i = z - zonelist->_zonerefs;
  1361. n = zlc->z_to_n[i];
  1362. /* This zone is worth trying if it is allowed but not full */
  1363. return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
  1364. }
  1365. /*
  1366. * Given 'z' scanning a zonelist, set the corresponding bit in
  1367. * zlc->fullzones, so that subsequent attempts to allocate a page
  1368. * from that zone don't waste time re-examining it.
  1369. */
  1370. static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
  1371. {
  1372. struct zonelist_cache *zlc; /* cached zonelist speedup info */
  1373. int i; /* index of *z in zonelist zones */
  1374. zlc = zonelist->zlcache_ptr;
  1375. if (!zlc)
  1376. return;
  1377. i = z - zonelist->_zonerefs;
  1378. set_bit(i, zlc->fullzones);
  1379. }
  1380. #else /* CONFIG_NUMA */
  1381. static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
  1382. {
  1383. return NULL;
  1384. }
  1385. static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
  1386. nodemask_t *allowednodes)
  1387. {
  1388. return 1;
  1389. }
  1390. static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
  1391. {
  1392. }
  1393. #endif /* CONFIG_NUMA */
  1394. /*
  1395. * get_page_from_freelist goes through the zonelist trying to allocate
  1396. * a page.
  1397. */
  1398. static struct page *
  1399. get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
  1400. struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
  1401. struct zone *preferred_zone, int migratetype)
  1402. {
  1403. struct zoneref *z;
  1404. struct page *page = NULL;
  1405. int classzone_idx;
  1406. struct zone *zone;
  1407. nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
  1408. int zlc_active = 0; /* set if using zonelist_cache */
  1409. int did_zlc_setup = 0; /* just call zlc_setup() one time */
  1410. classzone_idx = zone_idx(preferred_zone);
  1411. zonelist_scan:
  1412. /*
  1413. * Scan zonelist, looking for a zone with enough free.
  1414. * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
  1415. */
  1416. for_each_zone_zonelist_nodemask(zone, z, zonelist,
  1417. high_zoneidx, nodemask) {
  1418. if (NUMA_BUILD && zlc_active &&
  1419. !zlc_zone_worth_trying(zonelist, z, allowednodes))
  1420. continue;
  1421. if ((alloc_flags & ALLOC_CPUSET) &&
  1422. !cpuset_zone_allowed_softwall(zone, gfp_mask))
  1423. goto try_next_zone;
  1424. BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
  1425. if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
  1426. unsigned long mark;
  1427. int ret;
  1428. mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
  1429. if (zone_watermark_ok(zone, order, mark,
  1430. classzone_idx, alloc_flags))
  1431. goto try_this_zone;
  1432. if (zone_reclaim_mode == 0)
  1433. goto this_zone_full;
  1434. ret = zone_reclaim(zone, gfp_mask, order);
  1435. switch (ret) {
  1436. case ZONE_RECLAIM_NOSCAN:
  1437. /* did not scan */
  1438. goto try_next_zone;
  1439. case ZONE_RECLAIM_FULL:
  1440. /* scanned but unreclaimable */
  1441. goto this_zone_full;
  1442. default:
  1443. /* did we reclaim enough */
  1444. if (!zone_watermark_ok(zone, order, mark,
  1445. classzone_idx, alloc_flags))
  1446. goto this_zone_full;
  1447. }
  1448. }
  1449. try_this_zone:
  1450. page = buffered_rmqueue(preferred_zone, zone, order,
  1451. gfp_mask, migratetype);
  1452. if (page)
  1453. break;
  1454. this_zone_full:
  1455. if (NUMA_BUILD)
  1456. zlc_mark_zone_full(zonelist, z);
  1457. try_next_zone:
  1458. if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) {
  1459. /*
  1460. * we do zlc_setup after the first zone is tried but only
  1461. * if there are multiple nodes make it worthwhile
  1462. */
  1463. allowednodes = zlc_setup(zonelist, alloc_flags);
  1464. zlc_active = 1;
  1465. did_zlc_setup = 1;
  1466. }
  1467. }
  1468. if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
  1469. /* Disable zlc cache for second zonelist scan */
  1470. zlc_active = 0;
  1471. goto zonelist_scan;
  1472. }
  1473. return page;
  1474. }
  1475. static inline int
  1476. should_alloc_retry(gfp_t gfp_mask, unsigned int order,
  1477. unsigned long pages_reclaimed)
  1478. {
  1479. /* Do not loop if specifically requested */
  1480. if (gfp_mask & __GFP_NORETRY)
  1481. return 0;
  1482. /*
  1483. * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
  1484. * means __GFP_NOFAIL, but that may not be true in other
  1485. * implementations.
  1486. */
  1487. if (order <= PAGE_ALLOC_COSTLY_ORDER)
  1488. return 1;
  1489. /*
  1490. * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
  1491. * specified, then we retry until we no longer reclaim any pages
  1492. * (above), or we've reclaimed an order of pages at least as
  1493. * large as the allocation's order. In both cases, if the
  1494. * allocation still fails, we stop retrying.
  1495. */
  1496. if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
  1497. return 1;
  1498. /*
  1499. * Don't let big-order allocations loop unless the caller
  1500. * explicitly requests that.
  1501. */
  1502. if (gfp_mask & __GFP_NOFAIL)
  1503. return 1;
  1504. return 0;
  1505. }
  1506. static inline struct page *
  1507. __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
  1508. struct zonelist *zonelist, enum zone_type high_zoneidx,
  1509. nodemask_t *nodemask, struct zone *preferred_zone,
  1510. int migratetype)
  1511. {
  1512. struct page *page;
  1513. /* Acquire the OOM killer lock for the zones in zonelist */
  1514. if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
  1515. schedule_timeout_uninterruptible(1);
  1516. return NULL;
  1517. }
  1518. /*
  1519. * Go through the zonelist yet one more time, keep very high watermark
  1520. * here, this is only to catch a parallel oom killing, we must fail if
  1521. * we're still under heavy pressure.
  1522. */
  1523. page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
  1524. order, zonelist, high_zoneidx,
  1525. ALLOC_WMARK_HIGH|ALLOC_CPUSET,
  1526. preferred_zone, migratetype);
  1527. if (page)
  1528. goto out;
  1529. if (!(gfp_mask & __GFP_NOFAIL)) {
  1530. /* The OOM killer will not help higher order allocs */
  1531. if (order > PAGE_ALLOC_COSTLY_ORDER)
  1532. goto out;
  1533. /* The OOM killer does not needlessly kill tasks for lowmem */
  1534. if (high_zoneidx < ZONE_NORMAL)
  1535. goto out;
  1536. /*
  1537. * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
  1538. * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
  1539. * The caller should handle page allocation failure by itself if
  1540. * it specifies __GFP_THISNODE.
  1541. * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
  1542. */
  1543. if (gfp_mask & __GFP_THISNODE)
  1544. goto out;
  1545. }
  1546. /* Exhausted what can be done so it's blamo time */
  1547. out_of_memory(zonelist, gfp_mask, order, nodemask);
  1548. out:
  1549. clear_zonelist_oom(zonelist, gfp_mask);
  1550. return page;
  1551. }
  1552. #ifdef CONFIG_COMPACTION
  1553. /* Try memory compaction for high-order allocations before reclaim */
  1554. static struct page *
  1555. __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
  1556. struct zonelist *zonelist, enum zone_type high_zoneidx,
  1557. nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
  1558. int migratetype, unsigned long *did_some_progress,
  1559. bool sync_migration)
  1560. {
  1561. struct page *page;
  1562. if (!order || compaction_deferred(preferred_zone))
  1563. return NULL;
  1564. current->flags |= PF_MEMALLOC;
  1565. *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
  1566. nodemask, sync_migration);
  1567. current->flags &= ~PF_MEMALLOC;
  1568. if (*did_some_progress != COMPACT_SKIPPED) {
  1569. /* Page migration frees to the PCP lists but we want merging */
  1570. drain_pages(get_cpu());
  1571. put_cpu();
  1572. page = get_page_from_freelist(gfp_mask, nodemask,
  1573. order, zonelist, high_zoneidx,
  1574. alloc_flags, preferred_zone,
  1575. migratetype);
  1576. if (page) {
  1577. preferred_zone->compact_considered = 0;
  1578. preferred_zone->compact_defer_shift = 0;
  1579. count_vm_event(COMPACTSUCCESS);
  1580. return page;
  1581. }
  1582. /*
  1583. * It's bad if compaction run occurs and fails.
  1584. * The most likely reason is that pages exist,
  1585. * but not enough to satisfy watermarks.
  1586. */
  1587. count_vm_event(COMPACTFAIL);
  1588. defer_compaction(preferred_zone);
  1589. cond_resched();
  1590. }
  1591. return NULL;
  1592. }
  1593. #else
  1594. static inline struct page *
  1595. __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
  1596. struct zonelist *zonelist, enum zone_type high_zoneidx,
  1597. nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
  1598. int migratetype, unsigned long *did_some_progress,
  1599. bool sync_migration)
  1600. {
  1601. return NULL;
  1602. }
  1603. #endif /* CONFIG_COMPACTION */
  1604. /* The really slow allocator path where we enter direct reclaim */
  1605. static inline struct page *
  1606. __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
  1607. struct zonelist *zonelist, enum zone_type high_zoneidx,
  1608. nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
  1609. int migratetype, unsigned long *did_some_progress)
  1610. {
  1611. struct page *page = NULL;
  1612. struct reclaim_state reclaim_state;
  1613. bool drained = false;
  1614. cond_resched();
  1615. /* We now go into synchronous reclaim */
  1616. cpuset_memory_pressure_bump();
  1617. current->flags |= PF_MEMALLOC;
  1618. lockdep_set_current_reclaim_state(gfp_mask);
  1619. reclaim_state.reclaimed_slab = 0;
  1620. current->reclaim_state = &reclaim_state;
  1621. *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
  1622. current->reclaim_state = NULL;
  1623. lockdep_clear_current_reclaim_state();
  1624. current->flags &= ~PF_MEMALLOC;
  1625. cond_resched();
  1626. if (unlikely(!(*did_some_progress)))
  1627. return NULL;
  1628. retry:
  1629. page = get_page_from_freelist(gfp_mask, nodemask, order,
  1630. zonelist, high_zoneidx,
  1631. alloc_flags, preferred_zone,
  1632. migratetype);
  1633. /*
  1634. * If an allocation failed after direct reclaim, it could be because
  1635. * pages are pinned on the per-cpu lists. Drain them and try again
  1636. */
  1637. if (!page && !drained) {
  1638. drain_all_pages();
  1639. drained = true;
  1640. goto retry;
  1641. }
  1642. return page;
  1643. }
  1644. /*
  1645. * This is called in the allocator slow-path if the allocation request is of
  1646. * sufficient urgency to ignore watermarks and take other desperate measures
  1647. */
  1648. static inline struct page *
  1649. __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
  1650. struct zonelist *zonelist, enum zone_type high_zoneidx,
  1651. nodemask_t *nodemask, struct zone *preferred_zone,
  1652. int migratetype)
  1653. {
  1654. struct page *page;
  1655. do {
  1656. page = get_page_from_freelist(gfp_mask, nodemask, order,
  1657. zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
  1658. preferred_zone, migratetype);
  1659. if (!page && gfp_mask & __GFP_NOFAIL)
  1660. wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
  1661. } while (!page && (gfp_mask & __GFP_NOFAIL));
  1662. return page;
  1663. }
  1664. static inline
  1665. void wake_all_kswapd(unsigned int order, struct zonelist *zonelist,
  1666. enum zone_type high_zoneidx,
  1667. enum zone_type classzone_idx)
  1668. {
  1669. struct zoneref *z;
  1670. struct zone *zone;
  1671. for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
  1672. wakeup_kswapd(zone, order, classzone_idx);
  1673. }
  1674. static inline int
  1675. gfp_to_alloc_flags(gfp_t gfp_mask)
  1676. {
  1677. int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
  1678. const gfp_t wait = gfp_mask & __GFP_WAIT;
  1679. /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
  1680. BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
  1681. /*
  1682. * The caller may dip into page reserves a bit more if the caller
  1683. * cannot run direct reclaim, or if the caller has realtime scheduling
  1684. * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
  1685. * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
  1686. */
  1687. alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
  1688. if (!wait) {
  1689. /*
  1690. * Not worth trying to allocate harder for
  1691. * __GFP_NOMEMALLOC even if it can't schedule.
  1692. */
  1693. if (!(gfp_mask & __GFP_NOMEMALLOC))
  1694. alloc_flags |= ALLOC_HARDER;
  1695. /*
  1696. * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
  1697. * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
  1698. */
  1699. alloc_flags &= ~ALLOC_CPUSET;
  1700. } else if (unlikely(rt_task(current)) && !in_interrupt())
  1701. alloc_flags |= ALLOC_HARDER;
  1702. if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
  1703. if (!in_interrupt() &&
  1704. ((current->flags & PF_MEMALLOC) ||
  1705. unlikely(test_thread_flag(TIF_MEMDIE))))
  1706. alloc_flags |= ALLOC_NO_WATERMARKS;
  1707. }
  1708. return alloc_flags;
  1709. }
  1710. static inline struct page *
  1711. __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
  1712. struct zonelist *zonelist, enum zone_type high_zoneidx,
  1713. nodemask_t *nodemask, struct zone *preferred_zone,
  1714. int migratetype)
  1715. {
  1716. const gfp_t wait = gfp_mask & __GFP_WAIT;
  1717. struct page *page = NULL;
  1718. int alloc_flags;
  1719. unsigned long pages_reclaimed = 0;
  1720. unsigned long did_some_progress;
  1721. bool sync_migration = false;
  1722. /*
  1723. * In the slowpath, we sanity check order to avoid ever trying to
  1724. * reclaim >= MAX_ORDER areas which will never succeed. Callers may
  1725. * be using allocators in order of preference for an area that is
  1726. * too large.
  1727. */
  1728. if (order >= MAX_ORDER) {
  1729. WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
  1730. return NULL;
  1731. }
  1732. /*
  1733. * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
  1734. * __GFP_NOWARN set) should not cause reclaim since the subsystem
  1735. * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
  1736. * using a larger set of nodes after it has established that the
  1737. * allowed per node queues are empty and that nodes are
  1738. * over allocated.
  1739. */
  1740. if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
  1741. goto nopage;
  1742. restart:
  1743. if (!(gfp_mask & __GFP_NO_KSWAPD))
  1744. wake_all_kswapd(order, zonelist, high_zoneidx,
  1745. zone_idx(preferred_zone));
  1746. /*
  1747. * OK, we're below the kswapd watermark and have kicked background
  1748. * reclaim. Now things get more complex, so set up alloc_flags according
  1749. * to how we want to proceed.
  1750. */
  1751. alloc_flags = gfp_to_alloc_flags(gfp_mask);
  1752. /*
  1753. * Find the true preferred zone if the allocation is unconstrained by
  1754. * cpusets.
  1755. */
  1756. if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
  1757. first_zones_zonelist(zonelist, high_zoneidx, NULL,
  1758. &preferred_zone);
  1759. /* This is the last chance, in general, before the goto nopage. */
  1760. page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
  1761. high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
  1762. preferred_zone, migratetype);
  1763. if (page)
  1764. goto got_pg;
  1765. rebalance:
  1766. /* Allocate without watermarks if the context allows */
  1767. if (alloc_flags & ALLOC_NO_WATERMARKS) {
  1768. page = __alloc_pages_high_priority(gfp_mask, order,
  1769. zonelist, high_zoneidx, nodemask,
  1770. preferred_zone, migratetype);
  1771. if (page)
  1772. goto got_pg;
  1773. }
  1774. /* Atomic allocations - we can't balance anything */
  1775. if (!wait)
  1776. goto nopage;
  1777. /* Avoid recursion of direct reclaim */
  1778. if (current->flags & PF_MEMALLOC)
  1779. goto nopage;
  1780. /* Avoid allocations with no watermarks from looping endlessly */
  1781. if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
  1782. goto nopage;
  1783. /*
  1784. * Try direct compaction. The first pass is asynchronous. Subsequent
  1785. * attempts after direct reclaim are synchronous
  1786. */
  1787. page = __alloc_pages_direct_compact(gfp_mask, order,
  1788. zonelist, high_zoneidx,
  1789. nodemask,
  1790. alloc_flags, preferred_zone,
  1791. migratetype, &did_some_progress,
  1792. sync_migration);
  1793. if (page)
  1794. goto got_pg;
  1795. sync_migration = true;
  1796. /* Try direct reclaim and then allocating */
  1797. page = __alloc_pages_direct_reclaim(gfp_mask, order,
  1798. zonelist, high_zoneidx,
  1799. nodemask,
  1800. alloc_flags, preferred_zone,
  1801. migratetype, &did_some_progress);
  1802. if (page)
  1803. goto got_pg;
  1804. /*
  1805. * If we failed to make any progress reclaiming, then we are
  1806. * running out of options and have to consider going OOM
  1807. */
  1808. if (!did_some_progress) {
  1809. if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
  1810. if (oom_killer_disabled)
  1811. goto nopage;
  1812. page = __alloc_pages_may_oom(gfp_mask, order,
  1813. zonelist, high_zoneidx,
  1814. nodemask, preferred_zone,
  1815. migratetype);
  1816. if (page)
  1817. goto got_pg;
  1818. if (!(gfp_mask & __GFP_NOFAIL)) {
  1819. /*
  1820. * The oom killer is not called for high-order
  1821. * allocations that may fail, so if no progress
  1822. * is being made, there are no other options and
  1823. * retrying is unlikely to help.
  1824. */
  1825. if (order > PAGE_ALLOC_COSTLY_ORDER)
  1826. goto nopage;
  1827. /*
  1828. * The oom killer is not called for lowmem
  1829. * allocations to prevent needlessly killing
  1830. * innocent tasks.
  1831. */
  1832. if (high_zoneidx < ZONE_NORMAL)
  1833. goto nopage;
  1834. }
  1835. goto restart;
  1836. }
  1837. }
  1838. /* Check if we should retry the allocation */
  1839. pages_reclaimed += did_some_progress;
  1840. if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) {
  1841. /* Wait for some write requests to complete then retry */
  1842. wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
  1843. goto rebalance;
  1844. } else {
  1845. /*
  1846. * High-order allocations do not necessarily loop after
  1847. * direct reclaim and reclaim/compaction depends on compaction
  1848. * being called after reclaim so call directly if necessary
  1849. */
  1850. page = __alloc_pages_direct_compact(gfp_mask, order,
  1851. zonelist, high_zoneidx,
  1852. nodemask,
  1853. alloc_flags, preferred_zone,
  1854. migratetype, &did_some_progress,
  1855. sync_migration);
  1856. if (page)
  1857. goto got_pg;
  1858. }
  1859. nopage:
  1860. if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
  1861. printk(KERN_WARNING "%s: page allocation failure."
  1862. " order:%d, mode:0x%x\n",
  1863. current->comm, order, gfp_mask);
  1864. dump_stack();
  1865. show_mem();
  1866. }
  1867. return page;
  1868. got_pg:
  1869. if (kmemcheck_enabled)
  1870. kmemcheck_pagealloc_alloc(page, order, gfp_mask);
  1871. return page;
  1872. }
  1873. /*
  1874. * This is the 'heart' of the zoned buddy allocator.
  1875. */
  1876. struct page *
  1877. __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
  1878. struct zonelist *zonelist, nodemask_t *nodemask)
  1879. {
  1880. enum zone_type high_zoneidx = gfp_zone(gfp_mask);
  1881. struct zone *preferred_zone;
  1882. struct page *page;
  1883. int migratetype = allocflags_to_migratetype(gfp_mask);
  1884. gfp_mask &= gfp_allowed_mask;
  1885. lockdep_trace_alloc(gfp_mask);
  1886. might_sleep_if(gfp_mask & __GFP_WAIT);
  1887. if (should_fail_alloc_page(gfp_mask, order))
  1888. return NULL;
  1889. /*
  1890. * Check the zones suitable for the gfp_mask contain at least one
  1891. * valid zone. It's possible to have an empty zonelist as a result
  1892. * of GFP_THISNODE and a memoryless node
  1893. */
  1894. if (unlikely(!zonelist->_zonerefs->zone))
  1895. return NULL;
  1896. get_mems_allowed();
  1897. /* The preferred zone is used for statistics later */
  1898. first_zones_zonelist(zonelist, high_zoneidx,
  1899. nodemask ? : &cpuset_current_mems_allowed,
  1900. &preferred_zone);
  1901. if (!preferred_zone) {
  1902. put_mems_allowed();
  1903. return NULL;
  1904. }
  1905. /* First allocation attempt */
  1906. page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
  1907. zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
  1908. preferred_zone, migratetype);
  1909. if (unlikely(!page))
  1910. page = __alloc_pages_slowpath(gfp_mask, order,
  1911. zonelist, high_zoneidx, nodemask,
  1912. preferred_zone, migratetype);
  1913. put_mems_allowed();
  1914. trace_mm_page_alloc(page, order, gfp_mask, migratetype);
  1915. return page;
  1916. }
  1917. EXPORT_SYMBOL(__alloc_pages_nodemask);
  1918. /*
  1919. * Common helper functions.
  1920. */
  1921. unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
  1922. {
  1923. struct page *page;
  1924. /*
  1925. * __get_free_pages() returns a 32-bit address, which cannot represent
  1926. * a highmem page
  1927. */
  1928. VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
  1929. page = alloc_pages(gfp_mask, order);
  1930. if (!page)
  1931. return 0;
  1932. return (unsigned long) page_address(page);
  1933. }
  1934. EXPORT_SYMBOL(__get_free_pages);
  1935. unsigned long get_zeroed_page(gfp_t gfp_mask)
  1936. {
  1937. return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
  1938. }
  1939. EXPORT_SYMBOL(get_zeroed_page);
  1940. void __pagevec_free(struct pagevec *pvec)
  1941. {
  1942. int i = pagevec_count(pvec);
  1943. while (--i >= 0) {
  1944. trace_mm_pagevec_free(pvec->pages[i], pvec->cold);
  1945. free_hot_cold_page(pvec->pages[i], pvec->cold);
  1946. }
  1947. }
  1948. void __free_pages(struct page *page, unsigned int order)
  1949. {
  1950. if (put_page_testzero(page)) {
  1951. if (order == 0)
  1952. free_hot_cold_page(page, 0);
  1953. else
  1954. __free_pages_ok(page, order);
  1955. }
  1956. }
  1957. EXPORT_SYMBOL(__free_pages);
  1958. void free_pages(unsigned long addr, unsigned int order)
  1959. {
  1960. if (addr != 0) {
  1961. VM_BUG_ON(!virt_addr_valid((void *)addr));
  1962. __free_pages(virt_to_page((void *)addr), order);
  1963. }
  1964. }
  1965. EXPORT_SYMBOL(free_pages);
  1966. /**
  1967. * alloc_pages_exact - allocate an exact number physically-contiguous pages.
  1968. * @size: the number of bytes to allocate
  1969. * @gfp_mask: GFP flags for the allocation
  1970. *
  1971. * This function is similar to alloc_pages(), except that it allocates the
  1972. * minimum number of pages to satisfy the request. alloc_pages() can only
  1973. * allocate memory in power-of-two pages.
  1974. *
  1975. * This function is also limited by MAX_ORDER.
  1976. *
  1977. * Memory allocated by this function must be released by free_pages_exact().
  1978. */
  1979. void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
  1980. {
  1981. unsigned int order = get_order(size);
  1982. unsigned long addr;
  1983. addr = __get_free_pages(gfp_mask, order);
  1984. if (addr) {
  1985. unsigned long alloc_end = addr + (PAGE_SIZE << order);
  1986. unsigned long used = addr + PAGE_ALIGN(size);
  1987. split_page(virt_to_page((void *)addr), order);
  1988. while (used < alloc_end) {
  1989. free_page(used);
  1990. used += PAGE_SIZE;
  1991. }
  1992. }
  1993. return (void *)addr;
  1994. }
  1995. EXPORT_SYMBOL(alloc_pages_exact);
  1996. /**
  1997. * free_pages_exact - release memory allocated via alloc_pages_exact()
  1998. * @virt: the value returned by alloc_pages_exact.
  1999. * @size: size of allocation, same value as passed to alloc_pages_exact().
  2000. *
  2001. * Release the memory allocated by a previous call to alloc_pages_exact.
  2002. */
  2003. void free_pages_exact(void *virt, size_t size)
  2004. {
  2005. unsigned long addr = (unsigned long)virt;
  2006. unsigned long end = addr + PAGE_ALIGN(size);
  2007. while (addr < end) {
  2008. free_page(addr);
  2009. addr += PAGE_SIZE;
  2010. }
  2011. }
  2012. EXPORT_SYMBOL(free_pages_exact);
  2013. static unsigned int nr_free_zone_pages(int offset)
  2014. {
  2015. struct zoneref *z;
  2016. struct zone *zone;
  2017. /* Just pick one node, since fallback list is circular */
  2018. unsigned int sum = 0;
  2019. struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
  2020. for_each_zone_zonelist(zone, z, zonelist, offset) {
  2021. unsigned long size = zone->present_pages;
  2022. unsigned long high = high_wmark_pages(zone);
  2023. if (size > high)
  2024. sum += size - high;
  2025. }
  2026. return sum;
  2027. }
  2028. /*
  2029. * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
  2030. */
  2031. unsigned int nr_free_buffer_pages(void)
  2032. {
  2033. return nr_free_zone_pages(gfp_zone(GFP_USER));
  2034. }
  2035. EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
  2036. /*
  2037. * Amount of free RAM allocatable within all zones
  2038. */
  2039. unsigned int nr_free_pagecache_pages(void)
  2040. {
  2041. return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
  2042. }
  2043. static inline void show_node(struct zone *zone)
  2044. {
  2045. if (NUMA_BUILD)
  2046. printk("Node %d ", zone_to_nid(zone));
  2047. }
  2048. void si_meminfo(struct sysinfo *val)
  2049. {
  2050. val->totalram = totalram_pages;
  2051. val->sharedram = 0;
  2052. val->freeram = global_page_state(NR_FREE_PAGES);
  2053. val->bufferram = nr_blockdev_pages();
  2054. val->totalhigh = totalhigh_pages;
  2055. val->freehigh = nr_free_highpages();
  2056. val->mem_unit = PAGE_SIZE;
  2057. }
  2058. EXPORT_SYMBOL(si_meminfo);
  2059. #ifdef CONFIG_NUMA
  2060. void si_meminfo_node(struct sysinfo *val, int nid)
  2061. {
  2062. pg_data_t *pgdat = NODE_DATA(nid);
  2063. val->totalram = pgdat->node_present_pages;
  2064. val->freeram = node_page_state(nid, NR_FREE_PAGES);
  2065. #ifdef CONFIG_HIGHMEM
  2066. val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
  2067. val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
  2068. NR_FREE_PAGES);
  2069. #else
  2070. val->totalhigh = 0;
  2071. val->freehigh = 0;
  2072. #endif
  2073. val->mem_unit = PAGE_SIZE;
  2074. }
  2075. #endif
  2076. #define K(x) ((x) << (PAGE_SHIFT-10))
  2077. /*
  2078. * Show free area list (used inside shift_scroll-lock stuff)
  2079. * We also calculate the percentage fragmentation. We do this by counting the
  2080. * memory on each free list with the exception of the first item on the list.
  2081. */
  2082. void show_free_areas(void)
  2083. {
  2084. int cpu;
  2085. struct zone *zone;
  2086. for_each_populated_zone(zone) {
  2087. show_node(zone);
  2088. printk("%s per-cpu:\n", zone->name);
  2089. for_each_online_cpu(cpu) {
  2090. struct per_cpu_pageset *pageset;
  2091. pageset = per_cpu_ptr(zone->pageset, cpu);
  2092. printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
  2093. cpu, pageset->pcp.high,
  2094. pageset->pcp.batch, pageset->pcp.count);
  2095. }
  2096. }
  2097. printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
  2098. " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
  2099. " unevictable:%lu"
  2100. " dirty:%lu writeback:%lu unstable:%lu\n"
  2101. " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
  2102. " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
  2103. global_page_state(NR_ACTIVE_ANON),
  2104. global_page_state(NR_INACTIVE_ANON),
  2105. global_page_state(NR_ISOLATED_ANON),
  2106. global_page_state(NR_ACTIVE_FILE),
  2107. global_page_state(NR_INACTIVE_FILE),
  2108. global_page_state(NR_ISOLATED_FILE),
  2109. global_page_state(NR_UNEVICTABLE),
  2110. global_page_state(NR_FILE_DIRTY),
  2111. global_page_state(NR_WRITEBACK),
  2112. global_page_state(NR_UNSTABLE_NFS),
  2113. global_page_state(NR_FREE_PAGES),
  2114. global_page_state(NR_SLAB_RECLAIMABLE),
  2115. global_page_state(NR_SLAB_UNRECLAIMABLE),
  2116. global_page_state(NR_FILE_MAPPED),
  2117. global_page_state(NR_SHMEM),
  2118. global_page_state(NR_PAGETABLE),
  2119. global_page_state(NR_BOUNCE));
  2120. for_each_populated_zone(zone) {
  2121. int i;
  2122. show_node(zone);
  2123. printk("%s"
  2124. " free:%lukB"
  2125. " min:%lukB"
  2126. " low:%lukB"
  2127. " high:%lukB"
  2128. " active_anon:%lukB"
  2129. " inactive_anon:%lukB"
  2130. " active_file:%lukB"
  2131. " inactive_file:%lukB"
  2132. " unevictable:%lukB"
  2133. " isolated(anon):%lukB"
  2134. " isolated(file):%lukB"
  2135. " present:%lukB"
  2136. " mlocked:%lukB"
  2137. " dirty:%lukB"
  2138. " writeback:%lukB"
  2139. " mapped:%lukB"
  2140. " shmem:%lukB"
  2141. " slab_reclaimable:%lukB"
  2142. " slab_unreclaimable:%lukB"
  2143. " kernel_stack:%lukB"
  2144. " pagetables:%lukB"
  2145. " unstable:%lukB"
  2146. " bounce:%lukB"
  2147. " writeback_tmp:%lukB"
  2148. " pages_scanned:%lu"
  2149. " all_unreclaimable? %s"
  2150. "\n",
  2151. zone->name,
  2152. K(zone_page_state(zone, NR_FREE_PAGES)),
  2153. K(min_wmark_pages(zone)),
  2154. K(low_wmark_pages(zone)),
  2155. K(high_wmark_pages(zone)),
  2156. K(zone_page_state(zone, NR_ACTIVE_ANON)),
  2157. K(zone_page_state(zone, NR_INACTIVE_ANON)),
  2158. K(zone_page_state(zone, NR_ACTIVE_FILE)),
  2159. K(zone_page_state(zone, NR_INACTIVE_FILE)),
  2160. K(zone_page_state(zone, NR_UNEVICTABLE)),
  2161. K(zone_page_state(zone, NR_ISOLATED_ANON)),
  2162. K(zone_page_state(zone, NR_ISOLATED_FILE)),
  2163. K(zone->present_pages),
  2164. K(zone_page_state(zone, NR_MLOCK)),
  2165. K(zone_page_state(zone, NR_FILE_DIRTY)),
  2166. K(zone_page_state(zone, NR_WRITEBACK)),
  2167. K(zone_page_state(zone, NR_FILE_MAPPED)),
  2168. K(zone_page_state(zone, NR_SHMEM)),
  2169. K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
  2170. K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
  2171. zone_page_state(zone, NR_KERNEL_STACK) *
  2172. THREAD_SIZE / 1024,
  2173. K(zone_page_state(zone, NR_PAGETABLE)),
  2174. K(zone_page_state(zone, NR_UNSTABLE_NFS)),
  2175. K(zone_page_state(zone, NR_BOUNCE)),
  2176. K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
  2177. zone->pages_scanned,
  2178. (zone->all_unreclaimable ? "yes" : "no")
  2179. );
  2180. printk("lowmem_reserve[]:");
  2181. for (i = 0; i < MAX_NR_ZONES; i++)
  2182. printk(" %lu", zone->lowmem_reserve[i]);
  2183. printk("\n");
  2184. }
  2185. for_each_populated_zone(zone) {
  2186. unsigned long nr[MAX_ORDER], flags, order, total = 0;
  2187. show_node(zone);
  2188. printk("%s: ", zone->name);
  2189. spin_lock_irqsave(&zone->lock, flags);
  2190. for (order = 0; order < MAX_ORDER; order++) {
  2191. nr[order] = zone->free_area[order].nr_free;
  2192. total += nr[order] << order;
  2193. }
  2194. spin_unlock_irqrestore(&zone->lock, flags);
  2195. for (order = 0; order < MAX_ORDER; order++)
  2196. printk("%lu*%lukB ", nr[order], K(1UL) << order);
  2197. printk("= %lukB\n", K(total));
  2198. }
  2199. printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
  2200. show_swap_cache_info();
  2201. }
  2202. static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
  2203. {
  2204. zoneref->zone = zone;
  2205. zoneref->zone_idx = zone_idx(zone);
  2206. }
  2207. /*
  2208. * Builds allocation fallback zone lists.
  2209. *
  2210. * Add all populated zones of a node to the zonelist.
  2211. */
  2212. static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
  2213. int nr_zones, enum zone_type zone_type)
  2214. {
  2215. struct zone *zone;
  2216. BUG_ON(zone_type >= MAX_NR_ZONES);
  2217. zone_type++;
  2218. do {
  2219. zone_type--;
  2220. zone = pgdat->node_zones + zone_type;
  2221. if (populated_zone(zone)) {
  2222. zoneref_set_zone(zone,
  2223. &zonelist->_zonerefs[nr_zones++]);
  2224. check_highest_zone(zone_type);
  2225. }
  2226. } while (zone_type);
  2227. return nr_zones;
  2228. }
  2229. /*
  2230. * zonelist_order:
  2231. * 0 = automatic detection of better ordering.
  2232. * 1 = order by ([node] distance, -zonetype)
  2233. * 2 = order by (-zonetype, [node] distance)
  2234. *
  2235. * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
  2236. * the same zonelist. So only NUMA can configure this param.
  2237. */
  2238. #define ZONELIST_ORDER_DEFAULT 0
  2239. #define ZONELIST_ORDER_NODE 1
  2240. #define ZONELIST_ORDER_ZONE 2
  2241. /* zonelist order in the kernel.
  2242. * set_zonelist_order() will set this to NODE or ZONE.
  2243. */
  2244. static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
  2245. static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
  2246. #ifdef CONFIG_NUMA
  2247. /* The value user specified ....changed by config */
  2248. static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
  2249. /* string for sysctl */
  2250. #define NUMA_ZONELIST_ORDER_LEN 16
  2251. char numa_zonelist_order[16] = "default";
  2252. /*
  2253. * interface for configure zonelist ordering.
  2254. * command line option "numa_zonelist_order"
  2255. * = "[dD]efault - default, automatic configuration.
  2256. * = "[nN]ode - order by node locality, then by zone within node
  2257. * = "[zZ]one - order by zone, then by locality within zone
  2258. */
  2259. static int __parse_numa_zonelist_order(char *s)
  2260. {
  2261. if (*s == 'd' || *s == 'D') {
  2262. user_zonelist_order = ZONELIST_ORDER_DEFAULT;
  2263. } else if (*s == 'n' || *s == 'N') {
  2264. user_zonelist_order = ZONELIST_ORDER_NODE;
  2265. } else if (*s == 'z' || *s == 'Z') {
  2266. user_zonelist_order = ZONELIST_ORDER_ZONE;
  2267. } else {
  2268. printk(KERN_WARNING
  2269. "Ignoring invalid numa_zonelist_order value: "
  2270. "%s\n", s);
  2271. return -EINVAL;
  2272. }
  2273. return 0;
  2274. }
  2275. static __init int setup_numa_zonelist_order(char *s)
  2276. {
  2277. int ret;
  2278. if (!s)
  2279. return 0;
  2280. ret = __parse_numa_zonelist_order(s);
  2281. if (ret == 0)
  2282. strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
  2283. return ret;
  2284. }
  2285. early_param("numa_zonelist_order", setup_numa_zonelist_order);
  2286. /*
  2287. * sysctl handler for numa_zonelist_order
  2288. */
  2289. int numa_zonelist_order_handler(ctl_table *table, int write,
  2290. void __user *buffer, size_t *length,
  2291. loff_t *ppos)
  2292. {
  2293. char saved_string[NUMA_ZONELIST_ORDER_LEN];
  2294. int ret;
  2295. static DEFINE_MUTEX(zl_order_mutex);
  2296. mutex_lock(&zl_order_mutex);
  2297. if (write)
  2298. strcpy(saved_string, (char*)table->data);
  2299. ret = proc_dostring(table, write, buffer, length, ppos);
  2300. if (ret)
  2301. goto out;
  2302. if (write) {
  2303. int oldval = user_zonelist_order;
  2304. if (__parse_numa_zonelist_order((char*)table->data)) {
  2305. /*
  2306. * bogus value. restore saved string
  2307. */
  2308. strncpy((char*)table->data, saved_string,
  2309. NUMA_ZONELIST_ORDER_LEN);
  2310. user_zonelist_order = oldval;
  2311. } else if (oldval != user_zonelist_order) {
  2312. mutex_lock(&zonelists_mutex);
  2313. build_all_zonelists(NULL);
  2314. mutex_unlock(&zonelists_mutex);
  2315. }
  2316. }
  2317. out:
  2318. mutex_unlock(&zl_order_mutex);
  2319. return ret;
  2320. }
  2321. #define MAX_NODE_LOAD (nr_online_nodes)
  2322. static int node_load[MAX_NUMNODES];
  2323. /**
  2324. * find_next_best_node - find the next node that should appear in a given node's fallback list
  2325. * @node: node whose fallback list we're appending
  2326. * @used_node_mask: nodemask_t of already used nodes
  2327. *
  2328. * We use a number of factors to determine which is the next node that should
  2329. * appear on a given node's fallback list. The node should not have appeared
  2330. * already in @node's fallback list, and it should be the next closest node
  2331. * according to the distance array (which contains arbitrary distance values
  2332. * from each node to each node in the system), and should also prefer nodes
  2333. * with no CPUs, since presumably they'll have very little allocation pressure
  2334. * on them otherwise.
  2335. * It returns -1 if no node is found.
  2336. */
  2337. static int find_next_best_node(int node, nodemask_t *used_node_mask)
  2338. {
  2339. int n, val;
  2340. int min_val = INT_MAX;
  2341. int best_node = -1;
  2342. const struct cpumask *tmp = cpumask_of_node(0);
  2343. /* Use the local node if we haven't already */
  2344. if (!node_isset(node, *used_node_mask)) {
  2345. node_set(node, *used_node_mask);
  2346. return node;
  2347. }
  2348. for_each_node_state(n, N_HIGH_MEMORY) {
  2349. /* Don't want a node to appear more than once */
  2350. if (node_isset(n, *used_node_mask))
  2351. continue;
  2352. /* Use the distance array to find the distance */
  2353. val = node_distance(node, n);
  2354. /* Penalize nodes under us ("prefer the next node") */
  2355. val += (n < node);
  2356. /* Give preference to headless and unused nodes */
  2357. tmp = cpumask_of_node(n);
  2358. if (!cpumask_empty(tmp))
  2359. val += PENALTY_FOR_NODE_WITH_CPUS;
  2360. /* Slight preference for less loaded node */
  2361. val *= (MAX_NODE_LOAD*MAX_NUMNODES);
  2362. val += node_load[n];
  2363. if (val < min_val) {
  2364. min_val = val;
  2365. best_node = n;
  2366. }
  2367. }
  2368. if (best_node >= 0)
  2369. node_set(best_node, *used_node_mask);
  2370. return best_node;
  2371. }
  2372. /*
  2373. * Build zonelists ordered by node and zones within node.
  2374. * This results in maximum locality--normal zone overflows into local
  2375. * DMA zone, if any--but risks exhausting DMA zone.
  2376. */
  2377. static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
  2378. {
  2379. int j;
  2380. struct zonelist *zonelist;
  2381. zonelist = &pgdat->node_zonelists[0];
  2382. for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
  2383. ;
  2384. j = build_zonelists_node(NODE_DATA(node), zonelist, j,
  2385. MAX_NR_ZONES - 1);
  2386. zonelist->_zonerefs[j].zone = NULL;
  2387. zonelist->_zonerefs[j].zone_idx = 0;
  2388. }
  2389. /*
  2390. * Build gfp_thisnode zonelists
  2391. */
  2392. static void build_thisnode_zonelists(pg_data_t *pgdat)
  2393. {
  2394. int j;
  2395. struct zonelist *zonelist;
  2396. zonelist = &pgdat->node_zonelists[1];
  2397. j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
  2398. zonelist->_zonerefs[j].zone = NULL;
  2399. zonelist->_zonerefs[j].zone_idx = 0;
  2400. }
  2401. /*
  2402. * Build zonelists ordered by zone and nodes within zones.
  2403. * This results in conserving DMA zone[s] until all Normal memory is
  2404. * exhausted, but results in overflowing to remote node while memory
  2405. * may still exist in local DMA zone.
  2406. */
  2407. static int node_order[MAX_NUMNODES];
  2408. static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
  2409. {
  2410. int pos, j, node;
  2411. int zone_type; /* needs to be signed */
  2412. struct zone *z;
  2413. struct zonelist *zonelist;
  2414. zonelist = &pgdat->node_zonelists[0];
  2415. pos = 0;
  2416. for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
  2417. for (j = 0; j < nr_nodes; j++) {
  2418. node = node_order[j];
  2419. z = &NODE_DATA(node)->node_zones[zone_type];
  2420. if (populated_zone(z)) {
  2421. zoneref_set_zone(z,
  2422. &zonelist->_zonerefs[pos++]);
  2423. check_highest_zone(zone_type);
  2424. }
  2425. }
  2426. }
  2427. zonelist->_zonerefs[pos].zone = NULL;
  2428. zonelist->_zonerefs[pos].zone_idx = 0;
  2429. }
  2430. static int default_zonelist_order(void)
  2431. {
  2432. int nid, zone_type;
  2433. unsigned long low_kmem_size,total_size;
  2434. struct zone *z;
  2435. int average_size;
  2436. /*
  2437. * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
  2438. * If they are really small and used heavily, the system can fall
  2439. * into OOM very easily.
  2440. * This function detect ZONE_DMA/DMA32 size and configures zone order.
  2441. */
  2442. /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
  2443. low_kmem_size = 0;
  2444. total_size = 0;
  2445. for_each_online_node(nid) {
  2446. for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
  2447. z = &NODE_DATA(nid)->node_zones[zone_type];
  2448. if (populated_zone(z)) {
  2449. if (zone_type < ZONE_NORMAL)
  2450. low_kmem_size += z->present_pages;
  2451. total_size += z->present_pages;
  2452. } else if (zone_type == ZONE_NORMAL) {
  2453. /*
  2454. * If any node has only lowmem, then node order
  2455. * is preferred to allow kernel allocations
  2456. * locally; otherwise, they can easily infringe
  2457. * on other nodes when there is an abundance of
  2458. * lowmem available to allocate from.
  2459. */
  2460. return ZONELIST_ORDER_NODE;
  2461. }
  2462. }
  2463. }
  2464. if (!low_kmem_size || /* there are no DMA area. */
  2465. low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
  2466. return ZONELIST_ORDER_NODE;
  2467. /*
  2468. * look into each node's config.
  2469. * If there is a node whose DMA/DMA32 memory is very big area on
  2470. * local memory, NODE_ORDER may be suitable.
  2471. */
  2472. average_size = total_size /
  2473. (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
  2474. for_each_online_node(nid) {
  2475. low_kmem_size = 0;
  2476. total_size = 0;
  2477. for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
  2478. z = &NODE_DATA(nid)->node_zones[zone_type];
  2479. if (populated_zone(z)) {
  2480. if (zone_type < ZONE_NORMAL)
  2481. low_kmem_size += z->present_pages;
  2482. total_size += z->present_pages;
  2483. }
  2484. }
  2485. if (low_kmem_size &&
  2486. total_size > average_size && /* ignore small node */
  2487. low_kmem_size > total_size * 70/100)
  2488. return ZONELIST_ORDER_NODE;
  2489. }
  2490. return ZONELIST_ORDER_ZONE;
  2491. }
  2492. static void set_zonelist_order(void)
  2493. {
  2494. if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
  2495. current_zonelist_order = default_zonelist_order();
  2496. else
  2497. current_zonelist_order = user_zonelist_order;
  2498. }
  2499. static void build_zonelists(pg_data_t *pgdat)
  2500. {
  2501. int j, node, load;
  2502. enum zone_type i;
  2503. nodemask_t used_mask;
  2504. int local_node, prev_node;
  2505. struct zonelist *zonelist;
  2506. int order = current_zonelist_order;
  2507. /* initialize zonelists */
  2508. for (i = 0; i < MAX_ZONELISTS; i++) {
  2509. zonelist = pgdat->node_zonelists + i;
  2510. zonelist->_zonerefs[0].zone = NULL;
  2511. zonelist->_zonerefs[0].zone_idx = 0;
  2512. }
  2513. /* NUMA-aware ordering of nodes */
  2514. local_node = pgdat->node_id;
  2515. load = nr_online_nodes;
  2516. prev_node = local_node;
  2517. nodes_clear(used_mask);
  2518. memset(node_order, 0, sizeof(node_order));
  2519. j = 0;
  2520. while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
  2521. int distance = node_distance(local_node, node);
  2522. /*
  2523. * If another node is sufficiently far away then it is better
  2524. * to reclaim pages in a zone before going off node.
  2525. */
  2526. if (distance > RECLAIM_DISTANCE)
  2527. zone_reclaim_mode = 1;
  2528. /*
  2529. * We don't want to pressure a particular node.
  2530. * So adding penalty to the first node in same
  2531. * distance group to make it round-robin.
  2532. */
  2533. if (distance != node_distance(local_node, prev_node))
  2534. node_load[node] = load;
  2535. prev_node = node;
  2536. load--;
  2537. if (order == ZONELIST_ORDER_NODE)
  2538. build_zonelists_in_node_order(pgdat, node);
  2539. else
  2540. node_order[j++] = node; /* remember order */
  2541. }
  2542. if (order == ZONELIST_ORDER_ZONE) {
  2543. /* calculate node order -- i.e., DMA last! */
  2544. build_zonelists_in_zone_order(pgdat, j);
  2545. }
  2546. build_thisnode_zonelists(pgdat);
  2547. }
  2548. /* Construct the zonelist performance cache - see further mmzone.h */
  2549. static void build_zonelist_cache(pg_data_t *pgdat)
  2550. {
  2551. struct zonelist *zonelist;
  2552. struct zonelist_cache *zlc;
  2553. struct zoneref *z;
  2554. zonelist = &pgdat->node_zonelists[0];
  2555. zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
  2556. bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
  2557. for (z = zonelist->_zonerefs; z->zone; z++)
  2558. zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
  2559. }
  2560. #ifdef CONFIG_HAVE_MEMORYLESS_NODES
  2561. /*
  2562. * Return node id of node used for "local" allocations.
  2563. * I.e., first node id of first zone in arg node's generic zonelist.
  2564. * Used for initializing percpu 'numa_mem', which is used primarily
  2565. * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
  2566. */
  2567. int local_memory_node(int node)
  2568. {
  2569. struct zone *zone;
  2570. (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
  2571. gfp_zone(GFP_KERNEL),
  2572. NULL,
  2573. &zone);
  2574. return zone->node;
  2575. }
  2576. #endif
  2577. #else /* CONFIG_NUMA */
  2578. static void set_zonelist_order(void)
  2579. {
  2580. current_zonelist_order = ZONELIST_ORDER_ZONE;
  2581. }
  2582. static void build_zonelists(pg_data_t *pgdat)
  2583. {
  2584. int node, local_node;
  2585. enum zone_type j;
  2586. struct zonelist *zonelist;
  2587. local_node = pgdat->node_id;
  2588. zonelist = &pgdat->node_zonelists[0];
  2589. j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
  2590. /*
  2591. * Now we build the zonelist so that it contains the zones
  2592. * of all the other nodes.
  2593. * We don't want to pressure a particular node, so when
  2594. * building the zones for node N, we make sure that the
  2595. * zones coming right after the local ones are those from
  2596. * node N+1 (modulo N)
  2597. */
  2598. for (node = local_node + 1; node < MAX_NUMNODES; node++) {
  2599. if (!node_online(node))
  2600. continue;
  2601. j = build_zonelists_node(NODE_DATA(node), zonelist, j,
  2602. MAX_NR_ZONES - 1);
  2603. }
  2604. for (node = 0; node < local_node; node++) {
  2605. if (!node_online(node))
  2606. continue;
  2607. j = build_zonelists_node(NODE_DATA(node), zonelist, j,
  2608. MAX_NR_ZONES - 1);
  2609. }
  2610. zonelist->_zonerefs[j].zone = NULL;
  2611. zonelist->_zonerefs[j].zone_idx = 0;
  2612. }
  2613. /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
  2614. static void build_zonelist_cache(pg_data_t *pgdat)
  2615. {
  2616. pgdat->node_zonelists[0].zlcache_ptr = NULL;
  2617. }
  2618. #endif /* CONFIG_NUMA */
  2619. /*
  2620. * Boot pageset table. One per cpu which is going to be used for all
  2621. * zones and all nodes. The parameters will be set in such a way
  2622. * that an item put on a list will immediately be handed over to
  2623. * the buddy list. This is safe since pageset manipulation is done
  2624. * with interrupts disabled.
  2625. *
  2626. * The boot_pagesets must be kept even after bootup is complete for
  2627. * unused processors and/or zones. They do play a role for bootstrapping
  2628. * hotplugged processors.
  2629. *
  2630. * zoneinfo_show() and maybe other functions do
  2631. * not check if the processor is online before following the pageset pointer.
  2632. * Other parts of the kernel may not check if the zone is available.
  2633. */
  2634. static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
  2635. static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
  2636. static void setup_zone_pageset(struct zone *zone);
  2637. /*
  2638. * Global mutex to protect against size modification of zonelists
  2639. * as well as to serialize pageset setup for the new populated zone.
  2640. */
  2641. DEFINE_MUTEX(zonelists_mutex);
  2642. /* return values int ....just for stop_machine() */
  2643. static __init_refok int __build_all_zonelists(void *data)
  2644. {
  2645. int nid;
  2646. int cpu;
  2647. #ifdef CONFIG_NUMA
  2648. memset(node_load, 0, sizeof(node_load));
  2649. #endif
  2650. for_each_online_node(nid) {
  2651. pg_data_t *pgdat = NODE_DATA(nid);
  2652. build_zonelists(pgdat);
  2653. build_zonelist_cache(pgdat);
  2654. }
  2655. /*
  2656. * Initialize the boot_pagesets that are going to be used
  2657. * for bootstrapping processors. The real pagesets for
  2658. * each zone will be allocated later when the per cpu
  2659. * allocator is available.
  2660. *
  2661. * boot_pagesets are used also for bootstrapping offline
  2662. * cpus if the system is already booted because the pagesets
  2663. * are needed to initialize allocators on a specific cpu too.
  2664. * F.e. the percpu allocator needs the page allocator which
  2665. * needs the percpu allocator in order to allocate its pagesets
  2666. * (a chicken-egg dilemma).
  2667. */
  2668. for_each_possible_cpu(cpu) {
  2669. setup_pageset(&per_cpu(boot_pageset, cpu), 0);
  2670. #ifdef CONFIG_HAVE_MEMORYLESS_NODES
  2671. /*
  2672. * We now know the "local memory node" for each node--
  2673. * i.e., the node of the first zone in the generic zonelist.
  2674. * Set up numa_mem percpu variable for on-line cpus. During
  2675. * boot, only the boot cpu should be on-line; we'll init the
  2676. * secondary cpus' numa_mem as they come on-line. During
  2677. * node/memory hotplug, we'll fixup all on-line cpus.
  2678. */
  2679. if (cpu_online(cpu))
  2680. set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
  2681. #endif
  2682. }
  2683. return 0;
  2684. }
  2685. /*
  2686. * Called with zonelists_mutex held always
  2687. * unless system_state == SYSTEM_BOOTING.
  2688. */
  2689. void build_all_zonelists(void *data)
  2690. {
  2691. set_zonelist_order();
  2692. if (system_state == SYSTEM_BOOTING) {
  2693. __build_all_zonelists(NULL);
  2694. mminit_verify_zonelist();
  2695. cpuset_init_current_mems_allowed();
  2696. } else {
  2697. /* we have to stop all cpus to guarantee there is no user
  2698. of zonelist */
  2699. #ifdef CONFIG_MEMORY_HOTPLUG
  2700. if (data)
  2701. setup_zone_pageset((struct zone *)data);
  2702. #endif
  2703. stop_machine(__build_all_zonelists, NULL, NULL);
  2704. /* cpuset refresh routine should be here */
  2705. }
  2706. vm_total_pages = nr_free_pagecache_pages();
  2707. /*
  2708. * Disable grouping by mobility if the number of pages in the
  2709. * system is too low to allow the mechanism to work. It would be
  2710. * more accurate, but expensive to check per-zone. This check is
  2711. * made on memory-hotadd so a system can start with mobility
  2712. * disabled and enable it later
  2713. */
  2714. if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
  2715. page_group_by_mobility_disabled = 1;
  2716. else
  2717. page_group_by_mobility_disabled = 0;
  2718. printk("Built %i zonelists in %s order, mobility grouping %s. "
  2719. "Total pages: %ld\n",
  2720. nr_online_nodes,
  2721. zonelist_order_name[current_zonelist_order],
  2722. page_group_by_mobility_disabled ? "off" : "on",
  2723. vm_total_pages);
  2724. #ifdef CONFIG_NUMA
  2725. printk("Policy zone: %s\n", zone_names[policy_zone]);
  2726. #endif
  2727. }
  2728. /*
  2729. * Helper functions to size the waitqueue hash table.
  2730. * Essentially these want to choose hash table sizes sufficiently
  2731. * large so that collisions trying to wait on pages are rare.
  2732. * But in fact, the number of active page waitqueues on typical
  2733. * systems is ridiculously low, less than 200. So this is even
  2734. * conservative, even though it seems large.
  2735. *
  2736. * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
  2737. * waitqueues, i.e. the size of the waitq table given the number of pages.
  2738. */
  2739. #define PAGES_PER_WAITQUEUE 256
  2740. #ifndef CONFIG_MEMORY_HOTPLUG
  2741. static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
  2742. {
  2743. unsigned long size = 1;
  2744. pages /= PAGES_PER_WAITQUEUE;
  2745. while (size < pages)
  2746. size <<= 1;
  2747. /*
  2748. * Once we have dozens or even hundreds of threads sleeping
  2749. * on IO we've got bigger problems than wait queue collision.
  2750. * Limit the size of the wait table to a reasonable size.
  2751. */
  2752. size = min(size, 4096UL);
  2753. return max(size, 4UL);
  2754. }
  2755. #else
  2756. /*
  2757. * A zone's size might be changed by hot-add, so it is not possible to determine
  2758. * a suitable size for its wait_table. So we use the maximum size now.
  2759. *
  2760. * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
  2761. *
  2762. * i386 (preemption config) : 4096 x 16 = 64Kbyte.
  2763. * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
  2764. * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
  2765. *
  2766. * The maximum entries are prepared when a zone's memory is (512K + 256) pages
  2767. * or more by the traditional way. (See above). It equals:
  2768. *
  2769. * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
  2770. * ia64(16K page size) : = ( 8G + 4M)byte.
  2771. * powerpc (64K page size) : = (32G +16M)byte.
  2772. */
  2773. static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
  2774. {
  2775. return 4096UL;
  2776. }
  2777. #endif
  2778. /*
  2779. * This is an integer logarithm so that shifts can be used later
  2780. * to extract the more random high bits from the multiplicative
  2781. * hash function before the remainder is taken.
  2782. */
  2783. static inline unsigned long wait_table_bits(unsigned long size)
  2784. {
  2785. return ffz(~size);
  2786. }
  2787. #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
  2788. /*
  2789. * Mark a number of pageblocks as MIGRATE_RESERVE. The number
  2790. * of blocks reserved is based on min_wmark_pages(zone). The memory within
  2791. * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
  2792. * higher will lead to a bigger reserve which will get freed as contiguous
  2793. * blocks as reclaim kicks in
  2794. */
  2795. static void setup_zone_migrate_reserve(struct zone *zone)
  2796. {
  2797. unsigned long start_pfn, pfn, end_pfn;
  2798. struct page *page;
  2799. unsigned long block_migratetype;
  2800. int reserve;
  2801. /* Get the start pfn, end pfn and the number of blocks to reserve */
  2802. start_pfn = zone->zone_start_pfn;
  2803. end_pfn = start_pfn + zone->spanned_pages;
  2804. reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
  2805. pageblock_order;
  2806. /*
  2807. * Reserve blocks are generally in place to help high-order atomic
  2808. * allocations that are short-lived. A min_free_kbytes value that
  2809. * would result in more than 2 reserve blocks for atomic allocations
  2810. * is assumed to be in place to help anti-fragmentation for the
  2811. * future allocation of hugepages at runtime.
  2812. */
  2813. reserve = min(2, reserve);
  2814. for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
  2815. if (!pfn_valid(pfn))
  2816. continue;
  2817. page = pfn_to_page(pfn);
  2818. /* Watch out for overlapping nodes */
  2819. if (page_to_nid(page) != zone_to_nid(zone))
  2820. continue;
  2821. /* Blocks with reserved pages will never free, skip them. */
  2822. if (PageReserved(page))
  2823. continue;
  2824. block_migratetype = get_pageblock_migratetype(page);
  2825. /* If this block is reserved, account for it */
  2826. if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
  2827. reserve--;
  2828. continue;
  2829. }
  2830. /* Suitable for reserving if this block is movable */
  2831. if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
  2832. set_pageblock_migratetype(page, MIGRATE_RESERVE);
  2833. move_freepages_block(zone, page, MIGRATE_RESERVE);
  2834. reserve--;
  2835. continue;
  2836. }
  2837. /*
  2838. * If the reserve is met and this is a previous reserved block,
  2839. * take it back
  2840. */
  2841. if (block_migratetype == MIGRATE_RESERVE) {
  2842. set_pageblock_migratetype(page, MIGRATE_MOVABLE);
  2843. move_freepages_block(zone, page, MIGRATE_MOVABLE);
  2844. }
  2845. }
  2846. }
  2847. /*
  2848. * Initially all pages are reserved - free ones are freed
  2849. * up by free_all_bootmem() once the early boot process is
  2850. * done. Non-atomic initialization, single-pass.
  2851. */
  2852. void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
  2853. unsigned long start_pfn, enum memmap_context context)
  2854. {
  2855. struct page *page;
  2856. unsigned long end_pfn = start_pfn + size;
  2857. unsigned long pfn;
  2858. struct zone *z;
  2859. if (highest_memmap_pfn < end_pfn - 1)
  2860. highest_memmap_pfn = end_pfn - 1;
  2861. z = &NODE_DATA(nid)->node_zones[zone];
  2862. for (pfn = start_pfn; pfn < end_pfn; pfn++) {
  2863. /*
  2864. * There can be holes in boot-time mem_map[]s
  2865. * handed to this function. They do not
  2866. * exist on hotplugged memory.
  2867. */
  2868. if (context == MEMMAP_EARLY) {
  2869. if (!early_pfn_valid(pfn))
  2870. continue;
  2871. if (!early_pfn_in_nid(pfn, nid))
  2872. continue;
  2873. }
  2874. page = pfn_to_page(pfn);
  2875. set_page_links(page, zone, nid, pfn);
  2876. mminit_verify_page_links(page, zone, nid, pfn);
  2877. init_page_count(page);
  2878. reset_page_mapcount(page);
  2879. SetPageReserved(page);
  2880. /*
  2881. * Mark the block movable so that blocks are reserved for
  2882. * movable at startup. This will force kernel allocations
  2883. * to reserve their blocks rather than leaking throughout
  2884. * the address space during boot when many long-lived
  2885. * kernel allocations are made. Later some blocks near
  2886. * the start are marked MIGRATE_RESERVE by
  2887. * setup_zone_migrate_reserve()
  2888. *
  2889. * bitmap is created for zone's valid pfn range. but memmap
  2890. * can be created for invalid pages (for alignment)
  2891. * check here not to call set_pageblock_migratetype() against
  2892. * pfn out of zone.
  2893. */
  2894. if ((z->zone_start_pfn <= pfn)
  2895. && (pfn < z->zone_start_pfn + z->spanned_pages)
  2896. && !(pfn & (pageblock_nr_pages - 1)))
  2897. set_pageblock_migratetype(page, MIGRATE_MOVABLE);
  2898. INIT_LIST_HEAD(&page->lru);
  2899. #ifdef WANT_PAGE_VIRTUAL
  2900. /* The shift won't overflow because ZONE_NORMAL is below 4G. */
  2901. if (!is_highmem_idx(zone))
  2902. set_page_address(page, __va(pfn << PAGE_SHIFT));
  2903. #endif
  2904. }
  2905. }
  2906. static void __meminit zone_init_free_lists(struct zone *zone)
  2907. {
  2908. int order, t;
  2909. for_each_migratetype_order(order, t) {
  2910. INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
  2911. zone->free_area[order].nr_free = 0;
  2912. }
  2913. }
  2914. #ifndef __HAVE_ARCH_MEMMAP_INIT
  2915. #define memmap_init(size, nid, zone, start_pfn) \
  2916. memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
  2917. #endif
  2918. static int zone_batchsize(struct zone *zone)
  2919. {
  2920. #ifdef CONFIG_MMU
  2921. int batch;
  2922. /*
  2923. * The per-cpu-pages pools are set to around 1000th of the
  2924. * size of the zone. But no more than 1/2 of a meg.
  2925. *
  2926. * OK, so we don't know how big the cache is. So guess.
  2927. */
  2928. batch = zone->present_pages / 1024;
  2929. if (batch * PAGE_SIZE > 512 * 1024)
  2930. batch = (512 * 1024) / PAGE_SIZE;
  2931. batch /= 4; /* We effectively *= 4 below */
  2932. if (batch < 1)
  2933. batch = 1;
  2934. /*
  2935. * Clamp the batch to a 2^n - 1 value. Having a power
  2936. * of 2 value was found to be more likely to have
  2937. * suboptimal cache aliasing properties in some cases.
  2938. *
  2939. * For example if 2 tasks are alternately allocating
  2940. * batches of pages, one task can end up with a lot
  2941. * of pages of one half of the possible page colors
  2942. * and the other with pages of the other colors.
  2943. */
  2944. batch = rounddown_pow_of_two(batch + batch/2) - 1;
  2945. return batch;
  2946. #else
  2947. /* The deferral and batching of frees should be suppressed under NOMMU
  2948. * conditions.
  2949. *
  2950. * The problem is that NOMMU needs to be able to allocate large chunks
  2951. * of contiguous memory as there's no hardware page translation to
  2952. * assemble apparent contiguous memory from discontiguous pages.
  2953. *
  2954. * Queueing large contiguous runs of pages for batching, however,
  2955. * causes the pages to actually be freed in smaller chunks. As there
  2956. * can be a significant delay between the individual batches being
  2957. * recycled, this leads to the once large chunks of space being
  2958. * fragmented and becoming unavailable for high-order allocations.
  2959. */
  2960. return 0;
  2961. #endif
  2962. }
  2963. static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
  2964. {
  2965. struct per_cpu_pages *pcp;
  2966. int migratetype;
  2967. memset(p, 0, sizeof(*p));
  2968. pcp = &p->pcp;
  2969. pcp->count = 0;
  2970. pcp->high = 6 * batch;
  2971. pcp->batch = max(1UL, 1 * batch);
  2972. for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
  2973. INIT_LIST_HEAD(&pcp->lists[migratetype]);
  2974. }
  2975. /*
  2976. * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
  2977. * to the value high for the pageset p.
  2978. */
  2979. static void setup_pagelist_highmark(struct per_cpu_pageset *p,
  2980. unsigned long high)
  2981. {
  2982. struct per_cpu_pages *pcp;
  2983. pcp = &p->pcp;
  2984. pcp->high = high;
  2985. pcp->batch = max(1UL, high/4);
  2986. if ((high/4) > (PAGE_SHIFT * 8))
  2987. pcp->batch = PAGE_SHIFT * 8;
  2988. }
  2989. static __meminit void setup_zone_pageset(struct zone *zone)
  2990. {
  2991. int cpu;
  2992. zone->pageset = alloc_percpu(struct per_cpu_pageset);
  2993. for_each_possible_cpu(cpu) {
  2994. struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
  2995. setup_pageset(pcp, zone_batchsize(zone));
  2996. if (percpu_pagelist_fraction)
  2997. setup_pagelist_highmark(pcp,
  2998. (zone->present_pages /
  2999. percpu_pagelist_fraction));
  3000. }
  3001. }
  3002. /*
  3003. * Allocate per cpu pagesets and initialize them.
  3004. * Before this call only boot pagesets were available.
  3005. */
  3006. void __init setup_per_cpu_pageset(void)
  3007. {
  3008. struct zone *zone;
  3009. for_each_populated_zone(zone)
  3010. setup_zone_pageset(zone);
  3011. }
  3012. static noinline __init_refok
  3013. int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
  3014. {
  3015. int i;
  3016. struct pglist_data *pgdat = zone->zone_pgdat;
  3017. size_t alloc_size;
  3018. /*
  3019. * The per-page waitqueue mechanism uses hashed waitqueues
  3020. * per zone.
  3021. */
  3022. zone->wait_table_hash_nr_entries =
  3023. wait_table_hash_nr_entries(zone_size_pages);
  3024. zone->wait_table_bits =
  3025. wait_table_bits(zone->wait_table_hash_nr_entries);
  3026. alloc_size = zone->wait_table_hash_nr_entries
  3027. * sizeof(wait_queue_head_t);
  3028. if (!slab_is_available()) {
  3029. zone->wait_table = (wait_queue_head_t *)
  3030. alloc_bootmem_node(pgdat, alloc_size);
  3031. } else {
  3032. /*
  3033. * This case means that a zone whose size was 0 gets new memory
  3034. * via memory hot-add.
  3035. * But it may be the case that a new node was hot-added. In
  3036. * this case vmalloc() will not be able to use this new node's
  3037. * memory - this wait_table must be initialized to use this new
  3038. * node itself as well.
  3039. * To use this new node's memory, further consideration will be
  3040. * necessary.
  3041. */
  3042. zone->wait_table = vmalloc(alloc_size);
  3043. }
  3044. if (!zone->wait_table)
  3045. return -ENOMEM;
  3046. for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
  3047. init_waitqueue_head(zone->wait_table + i);
  3048. return 0;
  3049. }
  3050. static int __zone_pcp_update(void *data)
  3051. {
  3052. struct zone *zone = data;
  3053. int cpu;
  3054. unsigned long batch = zone_batchsize(zone), flags;
  3055. for_each_possible_cpu(cpu) {
  3056. struct per_cpu_pageset *pset;
  3057. struct per_cpu_pages *pcp;
  3058. pset = per_cpu_ptr(zone->pageset, cpu);
  3059. pcp = &pset->pcp;
  3060. local_irq_save(flags);
  3061. free_pcppages_bulk(zone, pcp->count, pcp);
  3062. setup_pageset(pset, batch);
  3063. local_irq_restore(flags);
  3064. }
  3065. return 0;
  3066. }
  3067. void zone_pcp_update(struct zone *zone)
  3068. {
  3069. stop_machine(__zone_pcp_update, zone, NULL);
  3070. }
  3071. static __meminit void zone_pcp_init(struct zone *zone)
  3072. {
  3073. /*
  3074. * per cpu subsystem is not up at this point. The following code
  3075. * relies on the ability of the linker to provide the
  3076. * offset of a (static) per cpu variable into the per cpu area.
  3077. */
  3078. zone->pageset = &boot_pageset;
  3079. if (zone->present_pages)
  3080. printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
  3081. zone->name, zone->present_pages,
  3082. zone_batchsize(zone));
  3083. }
  3084. __meminit int init_currently_empty_zone(struct zone *zone,
  3085. unsigned long zone_start_pfn,
  3086. unsigned long size,
  3087. enum memmap_context context)
  3088. {
  3089. struct pglist_data *pgdat = zone->zone_pgdat;
  3090. int ret;
  3091. ret = zone_wait_table_init(zone, size);
  3092. if (ret)
  3093. return ret;
  3094. pgdat->nr_zones = zone_idx(zone) + 1;
  3095. zone->zone_start_pfn = zone_start_pfn;
  3096. mminit_dprintk(MMINIT_TRACE, "memmap_init",
  3097. "Initialising map node %d zone %lu pfns %lu -> %lu\n",
  3098. pgdat->node_id,
  3099. (unsigned long)zone_idx(zone),
  3100. zone_start_pfn, (zone_start_pfn + size));
  3101. zone_init_free_lists(zone);
  3102. return 0;
  3103. }
  3104. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  3105. /*
  3106. * Basic iterator support. Return the first range of PFNs for a node
  3107. * Note: nid == MAX_NUMNODES returns first region regardless of node
  3108. */
  3109. static int __meminit first_active_region_index_in_nid(int nid)
  3110. {
  3111. int i;
  3112. for (i = 0; i < nr_nodemap_entries; i++)
  3113. if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
  3114. return i;
  3115. return -1;
  3116. }
  3117. /*
  3118. * Basic iterator support. Return the next active range of PFNs for a node
  3119. * Note: nid == MAX_NUMNODES returns next region regardless of node
  3120. */
  3121. static int __meminit next_active_region_index_in_nid(int index, int nid)
  3122. {
  3123. for (index = index + 1; index < nr_nodemap_entries; index++)
  3124. if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
  3125. return index;
  3126. return -1;
  3127. }
  3128. #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
  3129. /*
  3130. * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
  3131. * Architectures may implement their own version but if add_active_range()
  3132. * was used and there are no special requirements, this is a convenient
  3133. * alternative
  3134. */
  3135. int __meminit __early_pfn_to_nid(unsigned long pfn)
  3136. {
  3137. int i;
  3138. for (i = 0; i < nr_nodemap_entries; i++) {
  3139. unsigned long start_pfn = early_node_map[i].start_pfn;
  3140. unsigned long end_pfn = early_node_map[i].end_pfn;
  3141. if (start_pfn <= pfn && pfn < end_pfn)
  3142. return early_node_map[i].nid;
  3143. }
  3144. /* This is a memory hole */
  3145. return -1;
  3146. }
  3147. #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
  3148. int __meminit early_pfn_to_nid(unsigned long pfn)
  3149. {
  3150. int nid;
  3151. nid = __early_pfn_to_nid(pfn);
  3152. if (nid >= 0)
  3153. return nid;
  3154. /* just returns 0 */
  3155. return 0;
  3156. }
  3157. #ifdef CONFIG_NODES_SPAN_OTHER_NODES
  3158. bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
  3159. {
  3160. int nid;
  3161. nid = __early_pfn_to_nid(pfn);
  3162. if (nid >= 0 && nid != node)
  3163. return false;
  3164. return true;
  3165. }
  3166. #endif
  3167. /* Basic iterator support to walk early_node_map[] */
  3168. #define for_each_active_range_index_in_nid(i, nid) \
  3169. for (i = first_active_region_index_in_nid(nid); i != -1; \
  3170. i = next_active_region_index_in_nid(i, nid))
  3171. /**
  3172. * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
  3173. * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
  3174. * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
  3175. *
  3176. * If an architecture guarantees that all ranges registered with
  3177. * add_active_ranges() contain no holes and may be freed, this
  3178. * this function may be used instead of calling free_bootmem() manually.
  3179. */
  3180. void __init free_bootmem_with_active_regions(int nid,
  3181. unsigned long max_low_pfn)
  3182. {
  3183. int i;
  3184. for_each_active_range_index_in_nid(i, nid) {
  3185. unsigned long size_pages = 0;
  3186. unsigned long end_pfn = early_node_map[i].end_pfn;
  3187. if (early_node_map[i].start_pfn >= max_low_pfn)
  3188. continue;
  3189. if (end_pfn > max_low_pfn)
  3190. end_pfn = max_low_pfn;
  3191. size_pages = end_pfn - early_node_map[i].start_pfn;
  3192. free_bootmem_node(NODE_DATA(early_node_map[i].nid),
  3193. PFN_PHYS(early_node_map[i].start_pfn),
  3194. size_pages << PAGE_SHIFT);
  3195. }
  3196. }
  3197. #ifdef CONFIG_HAVE_MEMBLOCK
  3198. u64 __init find_memory_core_early(int nid, u64 size, u64 align,
  3199. u64 goal, u64 limit)
  3200. {
  3201. int i;
  3202. /* Need to go over early_node_map to find out good range for node */
  3203. for_each_active_range_index_in_nid(i, nid) {
  3204. u64 addr;
  3205. u64 ei_start, ei_last;
  3206. u64 final_start, final_end;
  3207. ei_last = early_node_map[i].end_pfn;
  3208. ei_last <<= PAGE_SHIFT;
  3209. ei_start = early_node_map[i].start_pfn;
  3210. ei_start <<= PAGE_SHIFT;
  3211. final_start = max(ei_start, goal);
  3212. final_end = min(ei_last, limit);
  3213. if (final_start >= final_end)
  3214. continue;
  3215. addr = memblock_find_in_range(final_start, final_end, size, align);
  3216. if (addr == MEMBLOCK_ERROR)
  3217. continue;
  3218. return addr;
  3219. }
  3220. return MEMBLOCK_ERROR;
  3221. }
  3222. #endif
  3223. int __init add_from_early_node_map(struct range *range, int az,
  3224. int nr_range, int nid)
  3225. {
  3226. int i;
  3227. u64 start, end;
  3228. /* need to go over early_node_map to find out good range for node */
  3229. for_each_active_range_index_in_nid(i, nid) {
  3230. start = early_node_map[i].start_pfn;
  3231. end = early_node_map[i].end_pfn;
  3232. nr_range = add_range(range, az, nr_range, start, end);
  3233. }
  3234. return nr_range;
  3235. }
  3236. #ifdef CONFIG_NO_BOOTMEM
  3237. void * __init __alloc_memory_core_early(int nid, u64 size, u64 align,
  3238. u64 goal, u64 limit)
  3239. {
  3240. void *ptr;
  3241. u64 addr;
  3242. if (limit > memblock.current_limit)
  3243. limit = memblock.current_limit;
  3244. addr = find_memory_core_early(nid, size, align, goal, limit);
  3245. if (addr == MEMBLOCK_ERROR)
  3246. return NULL;
  3247. ptr = phys_to_virt(addr);
  3248. memset(ptr, 0, size);
  3249. memblock_x86_reserve_range(addr, addr + size, "BOOTMEM");
  3250. /*
  3251. * The min_count is set to 0 so that bootmem allocated blocks
  3252. * are never reported as leaks.
  3253. */
  3254. kmemleak_alloc(ptr, size, 0, 0);
  3255. return ptr;
  3256. }
  3257. #endif
  3258. void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
  3259. {
  3260. int i;
  3261. int ret;
  3262. for_each_active_range_index_in_nid(i, nid) {
  3263. ret = work_fn(early_node_map[i].start_pfn,
  3264. early_node_map[i].end_pfn, data);
  3265. if (ret)
  3266. break;
  3267. }
  3268. }
  3269. /**
  3270. * sparse_memory_present_with_active_regions - Call memory_present for each active range
  3271. * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
  3272. *
  3273. * If an architecture guarantees that all ranges registered with
  3274. * add_active_ranges() contain no holes and may be freed, this
  3275. * function may be used instead of calling memory_present() manually.
  3276. */
  3277. void __init sparse_memory_present_with_active_regions(int nid)
  3278. {
  3279. int i;
  3280. for_each_active_range_index_in_nid(i, nid)
  3281. memory_present(early_node_map[i].nid,
  3282. early_node_map[i].start_pfn,
  3283. early_node_map[i].end_pfn);
  3284. }
  3285. /**
  3286. * get_pfn_range_for_nid - Return the start and end page frames for a node
  3287. * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
  3288. * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
  3289. * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
  3290. *
  3291. * It returns the start and end page frame of a node based on information
  3292. * provided by an arch calling add_active_range(). If called for a node
  3293. * with no available memory, a warning is printed and the start and end
  3294. * PFNs will be 0.
  3295. */
  3296. void __meminit get_pfn_range_for_nid(unsigned int nid,
  3297. unsigned long *start_pfn, unsigned long *end_pfn)
  3298. {
  3299. int i;
  3300. *start_pfn = -1UL;
  3301. *end_pfn = 0;
  3302. for_each_active_range_index_in_nid(i, nid) {
  3303. *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
  3304. *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
  3305. }
  3306. if (*start_pfn == -1UL)
  3307. *start_pfn = 0;
  3308. }
  3309. /*
  3310. * This finds a zone that can be used for ZONE_MOVABLE pages. The
  3311. * assumption is made that zones within a node are ordered in monotonic
  3312. * increasing memory addresses so that the "highest" populated zone is used
  3313. */
  3314. static void __init find_usable_zone_for_movable(void)
  3315. {
  3316. int zone_index;
  3317. for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
  3318. if (zone_index == ZONE_MOVABLE)
  3319. continue;
  3320. if (arch_zone_highest_possible_pfn[zone_index] >
  3321. arch_zone_lowest_possible_pfn[zone_index])
  3322. break;
  3323. }
  3324. VM_BUG_ON(zone_index == -1);
  3325. movable_zone = zone_index;
  3326. }
  3327. /*
  3328. * The zone ranges provided by the architecture do not include ZONE_MOVABLE
  3329. * because it is sized independant of architecture. Unlike the other zones,
  3330. * the starting point for ZONE_MOVABLE is not fixed. It may be different
  3331. * in each node depending on the size of each node and how evenly kernelcore
  3332. * is distributed. This helper function adjusts the zone ranges
  3333. * provided by the architecture for a given node by using the end of the
  3334. * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
  3335. * zones within a node are in order of monotonic increases memory addresses
  3336. */
  3337. static void __meminit adjust_zone_range_for_zone_movable(int nid,
  3338. unsigned long zone_type,
  3339. unsigned long node_start_pfn,
  3340. unsigned long node_end_pfn,
  3341. unsigned long *zone_start_pfn,
  3342. unsigned long *zone_end_pfn)
  3343. {
  3344. /* Only adjust if ZONE_MOVABLE is on this node */
  3345. if (zone_movable_pfn[nid]) {
  3346. /* Size ZONE_MOVABLE */
  3347. if (zone_type == ZONE_MOVABLE) {
  3348. *zone_start_pfn = zone_movable_pfn[nid];
  3349. *zone_end_pfn = min(node_end_pfn,
  3350. arch_zone_highest_possible_pfn[movable_zone]);
  3351. /* Adjust for ZONE_MOVABLE starting within this range */
  3352. } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
  3353. *zone_end_pfn > zone_movable_pfn[nid]) {
  3354. *zone_end_pfn = zone_movable_pfn[nid];
  3355. /* Check if this whole range is within ZONE_MOVABLE */
  3356. } else if (*zone_start_pfn >= zone_movable_pfn[nid])
  3357. *zone_start_pfn = *zone_end_pfn;
  3358. }
  3359. }
  3360. /*
  3361. * Return the number of pages a zone spans in a node, including holes
  3362. * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
  3363. */
  3364. static unsigned long __meminit zone_spanned_pages_in_node(int nid,
  3365. unsigned long zone_type,
  3366. unsigned long *ignored)
  3367. {
  3368. unsigned long node_start_pfn, node_end_pfn;
  3369. unsigned long zone_start_pfn, zone_end_pfn;
  3370. /* Get the start and end of the node and zone */
  3371. get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
  3372. zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
  3373. zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
  3374. adjust_zone_range_for_zone_movable(nid, zone_type,
  3375. node_start_pfn, node_end_pfn,
  3376. &zone_start_pfn, &zone_end_pfn);
  3377. /* Check that this node has pages within the zone's required range */
  3378. if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
  3379. return 0;
  3380. /* Move the zone boundaries inside the node if necessary */
  3381. zone_end_pfn = min(zone_end_pfn, node_end_pfn);
  3382. zone_start_pfn = max(zone_start_pfn, node_start_pfn);
  3383. /* Return the spanned pages */
  3384. return zone_end_pfn - zone_start_pfn;
  3385. }
  3386. /*
  3387. * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
  3388. * then all holes in the requested range will be accounted for.
  3389. */
  3390. unsigned long __meminit __absent_pages_in_range(int nid,
  3391. unsigned long range_start_pfn,
  3392. unsigned long range_end_pfn)
  3393. {
  3394. int i = 0;
  3395. unsigned long prev_end_pfn = 0, hole_pages = 0;
  3396. unsigned long start_pfn;
  3397. /* Find the end_pfn of the first active range of pfns in the node */
  3398. i = first_active_region_index_in_nid(nid);
  3399. if (i == -1)
  3400. return 0;
  3401. prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
  3402. /* Account for ranges before physical memory on this node */
  3403. if (early_node_map[i].start_pfn > range_start_pfn)
  3404. hole_pages = prev_end_pfn - range_start_pfn;
  3405. /* Find all holes for the zone within the node */
  3406. for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
  3407. /* No need to continue if prev_end_pfn is outside the zone */
  3408. if (prev_end_pfn >= range_end_pfn)
  3409. break;
  3410. /* Make sure the end of the zone is not within the hole */
  3411. start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
  3412. prev_end_pfn = max(prev_end_pfn, range_start_pfn);
  3413. /* Update the hole size cound and move on */
  3414. if (start_pfn > range_start_pfn) {
  3415. BUG_ON(prev_end_pfn > start_pfn);
  3416. hole_pages += start_pfn - prev_end_pfn;
  3417. }
  3418. prev_end_pfn = early_node_map[i].end_pfn;
  3419. }
  3420. /* Account for ranges past physical memory on this node */
  3421. if (range_end_pfn > prev_end_pfn)
  3422. hole_pages += range_end_pfn -
  3423. max(range_start_pfn, prev_end_pfn);
  3424. return hole_pages;
  3425. }
  3426. /**
  3427. * absent_pages_in_range - Return number of page frames in holes within a range
  3428. * @start_pfn: The start PFN to start searching for holes
  3429. * @end_pfn: The end PFN to stop searching for holes
  3430. *
  3431. * It returns the number of pages frames in memory holes within a range.
  3432. */
  3433. unsigned long __init absent_pages_in_range(unsigned long start_pfn,
  3434. unsigned long end_pfn)
  3435. {
  3436. return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
  3437. }
  3438. /* Return the number of page frames in holes in a zone on a node */
  3439. static unsigned long __meminit zone_absent_pages_in_node(int nid,
  3440. unsigned long zone_type,
  3441. unsigned long *ignored)
  3442. {
  3443. unsigned long node_start_pfn, node_end_pfn;
  3444. unsigned long zone_start_pfn, zone_end_pfn;
  3445. get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
  3446. zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
  3447. node_start_pfn);
  3448. zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
  3449. node_end_pfn);
  3450. adjust_zone_range_for_zone_movable(nid, zone_type,
  3451. node_start_pfn, node_end_pfn,
  3452. &zone_start_pfn, &zone_end_pfn);
  3453. return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
  3454. }
  3455. #else
  3456. static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
  3457. unsigned long zone_type,
  3458. unsigned long *zones_size)
  3459. {
  3460. return zones_size[zone_type];
  3461. }
  3462. static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
  3463. unsigned long zone_type,
  3464. unsigned long *zholes_size)
  3465. {
  3466. if (!zholes_size)
  3467. return 0;
  3468. return zholes_size[zone_type];
  3469. }
  3470. #endif
  3471. static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
  3472. unsigned long *zones_size, unsigned long *zholes_size)
  3473. {
  3474. unsigned long realtotalpages, totalpages = 0;
  3475. enum zone_type i;
  3476. for (i = 0; i < MAX_NR_ZONES; i++)
  3477. totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
  3478. zones_size);
  3479. pgdat->node_spanned_pages = totalpages;
  3480. realtotalpages = totalpages;
  3481. for (i = 0; i < MAX_NR_ZONES; i++)
  3482. realtotalpages -=
  3483. zone_absent_pages_in_node(pgdat->node_id, i,
  3484. zholes_size);
  3485. pgdat->node_present_pages = realtotalpages;
  3486. printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
  3487. realtotalpages);
  3488. }
  3489. #ifndef CONFIG_SPARSEMEM
  3490. /*
  3491. * Calculate the size of the zone->blockflags rounded to an unsigned long
  3492. * Start by making sure zonesize is a multiple of pageblock_order by rounding
  3493. * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
  3494. * round what is now in bits to nearest long in bits, then return it in
  3495. * bytes.
  3496. */
  3497. static unsigned long __init usemap_size(unsigned long zonesize)
  3498. {
  3499. unsigned long usemapsize;
  3500. usemapsize = roundup(zonesize, pageblock_nr_pages);
  3501. usemapsize = usemapsize >> pageblock_order;
  3502. usemapsize *= NR_PAGEBLOCK_BITS;
  3503. usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
  3504. return usemapsize / 8;
  3505. }
  3506. static void __init setup_usemap(struct pglist_data *pgdat,
  3507. struct zone *zone, unsigned long zonesize)
  3508. {
  3509. unsigned long usemapsize = usemap_size(zonesize);
  3510. zone->pageblock_flags = NULL;
  3511. if (usemapsize)
  3512. zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
  3513. }
  3514. #else
  3515. static inline void setup_usemap(struct pglist_data *pgdat,
  3516. struct zone *zone, unsigned long zonesize) {}
  3517. #endif /* CONFIG_SPARSEMEM */
  3518. #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
  3519. /* Return a sensible default order for the pageblock size. */
  3520. static inline int pageblock_default_order(void)
  3521. {
  3522. if (HPAGE_SHIFT > PAGE_SHIFT)
  3523. return HUGETLB_PAGE_ORDER;
  3524. return MAX_ORDER-1;
  3525. }
  3526. /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
  3527. static inline void __init set_pageblock_order(unsigned int order)
  3528. {
  3529. /* Check that pageblock_nr_pages has not already been setup */
  3530. if (pageblock_order)
  3531. return;
  3532. /*
  3533. * Assume the largest contiguous order of interest is a huge page.
  3534. * This value may be variable depending on boot parameters on IA64
  3535. */
  3536. pageblock_order = order;
  3537. }
  3538. #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
  3539. /*
  3540. * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
  3541. * and pageblock_default_order() are unused as pageblock_order is set
  3542. * at compile-time. See include/linux/pageblock-flags.h for the values of
  3543. * pageblock_order based on the kernel config
  3544. */
  3545. static inline int pageblock_default_order(unsigned int order)
  3546. {
  3547. return MAX_ORDER-1;
  3548. }
  3549. #define set_pageblock_order(x) do {} while (0)
  3550. #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
  3551. /*
  3552. * Set up the zone data structures:
  3553. * - mark all pages reserved
  3554. * - mark all memory queues empty
  3555. * - clear the memory bitmaps
  3556. */
  3557. static void __paginginit free_area_init_core(struct pglist_data *pgdat,
  3558. unsigned long *zones_size, unsigned long *zholes_size)
  3559. {
  3560. enum zone_type j;
  3561. int nid = pgdat->node_id;
  3562. unsigned long zone_start_pfn = pgdat->node_start_pfn;
  3563. int ret;
  3564. pgdat_resize_init(pgdat);
  3565. pgdat->nr_zones = 0;
  3566. init_waitqueue_head(&pgdat->kswapd_wait);
  3567. pgdat->kswapd_max_order = 0;
  3568. pgdat_page_cgroup_init(pgdat);
  3569. for (j = 0; j < MAX_NR_ZONES; j++) {
  3570. struct zone *zone = pgdat->node_zones + j;
  3571. unsigned long size, realsize, memmap_pages;
  3572. enum lru_list l;
  3573. size = zone_spanned_pages_in_node(nid, j, zones_size);
  3574. realsize = size - zone_absent_pages_in_node(nid, j,
  3575. zholes_size);
  3576. /*
  3577. * Adjust realsize so that it accounts for how much memory
  3578. * is used by this zone for memmap. This affects the watermark
  3579. * and per-cpu initialisations
  3580. */
  3581. memmap_pages =
  3582. PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
  3583. if (realsize >= memmap_pages) {
  3584. realsize -= memmap_pages;
  3585. if (memmap_pages)
  3586. printk(KERN_DEBUG
  3587. " %s zone: %lu pages used for memmap\n",
  3588. zone_names[j], memmap_pages);
  3589. } else
  3590. printk(KERN_WARNING
  3591. " %s zone: %lu pages exceeds realsize %lu\n",
  3592. zone_names[j], memmap_pages, realsize);
  3593. /* Account for reserved pages */
  3594. if (j == 0 && realsize > dma_reserve) {
  3595. realsize -= dma_reserve;
  3596. printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
  3597. zone_names[0], dma_reserve);
  3598. }
  3599. if (!is_highmem_idx(j))
  3600. nr_kernel_pages += realsize;
  3601. nr_all_pages += realsize;
  3602. zone->spanned_pages = size;
  3603. zone->present_pages = realsize;
  3604. #ifdef CONFIG_NUMA
  3605. zone->node = nid;
  3606. zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
  3607. / 100;
  3608. zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
  3609. #endif
  3610. zone->name = zone_names[j];
  3611. spin_lock_init(&zone->lock);
  3612. spin_lock_init(&zone->lru_lock);
  3613. zone_seqlock_init(zone);
  3614. zone->zone_pgdat = pgdat;
  3615. zone_pcp_init(zone);
  3616. for_each_lru(l) {
  3617. INIT_LIST_HEAD(&zone->lru[l].list);
  3618. zone->reclaim_stat.nr_saved_scan[l] = 0;
  3619. }
  3620. zone->reclaim_stat.recent_rotated[0] = 0;
  3621. zone->reclaim_stat.recent_rotated[1] = 0;
  3622. zone->reclaim_stat.recent_scanned[0] = 0;
  3623. zone->reclaim_stat.recent_scanned[1] = 0;
  3624. zap_zone_vm_stats(zone);
  3625. zone->flags = 0;
  3626. if (!size)
  3627. continue;
  3628. set_pageblock_order(pageblock_default_order());
  3629. setup_usemap(pgdat, zone, size);
  3630. ret = init_currently_empty_zone(zone, zone_start_pfn,
  3631. size, MEMMAP_EARLY);
  3632. BUG_ON(ret);
  3633. memmap_init(size, nid, j, zone_start_pfn);
  3634. zone_start_pfn += size;
  3635. }
  3636. }
  3637. static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
  3638. {
  3639. /* Skip empty nodes */
  3640. if (!pgdat->node_spanned_pages)
  3641. return;
  3642. #ifdef CONFIG_FLAT_NODE_MEM_MAP
  3643. /* ia64 gets its own node_mem_map, before this, without bootmem */
  3644. if (!pgdat->node_mem_map) {
  3645. unsigned long size, start, end;
  3646. struct page *map;
  3647. /*
  3648. * The zone's endpoints aren't required to be MAX_ORDER
  3649. * aligned but the node_mem_map endpoints must be in order
  3650. * for the buddy allocator to function correctly.
  3651. */
  3652. start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
  3653. end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
  3654. end = ALIGN(end, MAX_ORDER_NR_PAGES);
  3655. size = (end - start) * sizeof(struct page);
  3656. map = alloc_remap(pgdat->node_id, size);
  3657. if (!map)
  3658. map = alloc_bootmem_node(pgdat, size);
  3659. pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
  3660. }
  3661. #ifndef CONFIG_NEED_MULTIPLE_NODES
  3662. /*
  3663. * With no DISCONTIG, the global mem_map is just set as node 0's
  3664. */
  3665. if (pgdat == NODE_DATA(0)) {
  3666. mem_map = NODE_DATA(0)->node_mem_map;
  3667. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  3668. if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
  3669. mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
  3670. #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
  3671. }
  3672. #endif
  3673. #endif /* CONFIG_FLAT_NODE_MEM_MAP */
  3674. }
  3675. void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
  3676. unsigned long node_start_pfn, unsigned long *zholes_size)
  3677. {
  3678. pg_data_t *pgdat = NODE_DATA(nid);
  3679. pgdat->node_id = nid;
  3680. pgdat->node_start_pfn = node_start_pfn;
  3681. calculate_node_totalpages(pgdat, zones_size, zholes_size);
  3682. alloc_node_mem_map(pgdat);
  3683. #ifdef CONFIG_FLAT_NODE_MEM_MAP
  3684. printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
  3685. nid, (unsigned long)pgdat,
  3686. (unsigned long)pgdat->node_mem_map);
  3687. #endif
  3688. free_area_init_core(pgdat, zones_size, zholes_size);
  3689. }
  3690. #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
  3691. #if MAX_NUMNODES > 1
  3692. /*
  3693. * Figure out the number of possible node ids.
  3694. */
  3695. static void __init setup_nr_node_ids(void)
  3696. {
  3697. unsigned int node;
  3698. unsigned int highest = 0;
  3699. for_each_node_mask(node, node_possible_map)
  3700. highest = node;
  3701. nr_node_ids = highest + 1;
  3702. }
  3703. #else
  3704. static inline void setup_nr_node_ids(void)
  3705. {
  3706. }
  3707. #endif
  3708. /**
  3709. * add_active_range - Register a range of PFNs backed by physical memory
  3710. * @nid: The node ID the range resides on
  3711. * @start_pfn: The start PFN of the available physical memory
  3712. * @end_pfn: The end PFN of the available physical memory
  3713. *
  3714. * These ranges are stored in an early_node_map[] and later used by
  3715. * free_area_init_nodes() to calculate zone sizes and holes. If the
  3716. * range spans a memory hole, it is up to the architecture to ensure
  3717. * the memory is not freed by the bootmem allocator. If possible
  3718. * the range being registered will be merged with existing ranges.
  3719. */
  3720. void __init add_active_range(unsigned int nid, unsigned long start_pfn,
  3721. unsigned long end_pfn)
  3722. {
  3723. int i;
  3724. mminit_dprintk(MMINIT_TRACE, "memory_register",
  3725. "Entering add_active_range(%d, %#lx, %#lx) "
  3726. "%d entries of %d used\n",
  3727. nid, start_pfn, end_pfn,
  3728. nr_nodemap_entries, MAX_ACTIVE_REGIONS);
  3729. mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
  3730. /* Merge with existing active regions if possible */
  3731. for (i = 0; i < nr_nodemap_entries; i++) {
  3732. if (early_node_map[i].nid != nid)
  3733. continue;
  3734. /* Skip if an existing region covers this new one */
  3735. if (start_pfn >= early_node_map[i].start_pfn &&
  3736. end_pfn <= early_node_map[i].end_pfn)
  3737. return;
  3738. /* Merge forward if suitable */
  3739. if (start_pfn <= early_node_map[i].end_pfn &&
  3740. end_pfn > early_node_map[i].end_pfn) {
  3741. early_node_map[i].end_pfn = end_pfn;
  3742. return;
  3743. }
  3744. /* Merge backward if suitable */
  3745. if (start_pfn < early_node_map[i].start_pfn &&
  3746. end_pfn >= early_node_map[i].start_pfn) {
  3747. early_node_map[i].start_pfn = start_pfn;
  3748. return;
  3749. }
  3750. }
  3751. /* Check that early_node_map is large enough */
  3752. if (i >= MAX_ACTIVE_REGIONS) {
  3753. printk(KERN_CRIT "More than %d memory regions, truncating\n",
  3754. MAX_ACTIVE_REGIONS);
  3755. return;
  3756. }
  3757. early_node_map[i].nid = nid;
  3758. early_node_map[i].start_pfn = start_pfn;
  3759. early_node_map[i].end_pfn = end_pfn;
  3760. nr_nodemap_entries = i + 1;
  3761. }
  3762. /**
  3763. * remove_active_range - Shrink an existing registered range of PFNs
  3764. * @nid: The node id the range is on that should be shrunk
  3765. * @start_pfn: The new PFN of the range
  3766. * @end_pfn: The new PFN of the range
  3767. *
  3768. * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
  3769. * The map is kept near the end physical page range that has already been
  3770. * registered. This function allows an arch to shrink an existing registered
  3771. * range.
  3772. */
  3773. void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
  3774. unsigned long end_pfn)
  3775. {
  3776. int i, j;
  3777. int removed = 0;
  3778. printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
  3779. nid, start_pfn, end_pfn);
  3780. /* Find the old active region end and shrink */
  3781. for_each_active_range_index_in_nid(i, nid) {
  3782. if (early_node_map[i].start_pfn >= start_pfn &&
  3783. early_node_map[i].end_pfn <= end_pfn) {
  3784. /* clear it */
  3785. early_node_map[i].start_pfn = 0;
  3786. early_node_map[i].end_pfn = 0;
  3787. removed = 1;
  3788. continue;
  3789. }
  3790. if (early_node_map[i].start_pfn < start_pfn &&
  3791. early_node_map[i].end_pfn > start_pfn) {
  3792. unsigned long temp_end_pfn = early_node_map[i].end_pfn;
  3793. early_node_map[i].end_pfn = start_pfn;
  3794. if (temp_end_pfn > end_pfn)
  3795. add_active_range(nid, end_pfn, temp_end_pfn);
  3796. continue;
  3797. }
  3798. if (early_node_map[i].start_pfn >= start_pfn &&
  3799. early_node_map[i].end_pfn > end_pfn &&
  3800. early_node_map[i].start_pfn < end_pfn) {
  3801. early_node_map[i].start_pfn = end_pfn;
  3802. continue;
  3803. }
  3804. }
  3805. if (!removed)
  3806. return;
  3807. /* remove the blank ones */
  3808. for (i = nr_nodemap_entries - 1; i > 0; i--) {
  3809. if (early_node_map[i].nid != nid)
  3810. continue;
  3811. if (early_node_map[i].end_pfn)
  3812. continue;
  3813. /* we found it, get rid of it */
  3814. for (j = i; j < nr_nodemap_entries - 1; j++)
  3815. memcpy(&early_node_map[j], &early_node_map[j+1],
  3816. sizeof(early_node_map[j]));
  3817. j = nr_nodemap_entries - 1;
  3818. memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
  3819. nr_nodemap_entries--;
  3820. }
  3821. }
  3822. /**
  3823. * remove_all_active_ranges - Remove all currently registered regions
  3824. *
  3825. * During discovery, it may be found that a table like SRAT is invalid
  3826. * and an alternative discovery method must be used. This function removes
  3827. * all currently registered regions.
  3828. */
  3829. void __init remove_all_active_ranges(void)
  3830. {
  3831. memset(early_node_map, 0, sizeof(early_node_map));
  3832. nr_nodemap_entries = 0;
  3833. }
  3834. /* Compare two active node_active_regions */
  3835. static int __init cmp_node_active_region(const void *a, const void *b)
  3836. {
  3837. struct node_active_region *arange = (struct node_active_region *)a;
  3838. struct node_active_region *brange = (struct node_active_region *)b;
  3839. /* Done this way to avoid overflows */
  3840. if (arange->start_pfn > brange->start_pfn)
  3841. return 1;
  3842. if (arange->start_pfn < brange->start_pfn)
  3843. return -1;
  3844. return 0;
  3845. }
  3846. /* sort the node_map by start_pfn */
  3847. void __init sort_node_map(void)
  3848. {
  3849. sort(early_node_map, (size_t)nr_nodemap_entries,
  3850. sizeof(struct node_active_region),
  3851. cmp_node_active_region, NULL);
  3852. }
  3853. /* Find the lowest pfn for a node */
  3854. static unsigned long __init find_min_pfn_for_node(int nid)
  3855. {
  3856. int i;
  3857. unsigned long min_pfn = ULONG_MAX;
  3858. /* Assuming a sorted map, the first range found has the starting pfn */
  3859. for_each_active_range_index_in_nid(i, nid)
  3860. min_pfn = min(min_pfn, early_node_map[i].start_pfn);
  3861. if (min_pfn == ULONG_MAX) {
  3862. printk(KERN_WARNING
  3863. "Could not find start_pfn for node %d\n", nid);
  3864. return 0;
  3865. }
  3866. return min_pfn;
  3867. }
  3868. /**
  3869. * find_min_pfn_with_active_regions - Find the minimum PFN registered
  3870. *
  3871. * It returns the minimum PFN based on information provided via
  3872. * add_active_range().
  3873. */
  3874. unsigned long __init find_min_pfn_with_active_regions(void)
  3875. {
  3876. return find_min_pfn_for_node(MAX_NUMNODES);
  3877. }
  3878. /*
  3879. * early_calculate_totalpages()
  3880. * Sum pages in active regions for movable zone.
  3881. * Populate N_HIGH_MEMORY for calculating usable_nodes.
  3882. */
  3883. static unsigned long __init early_calculate_totalpages(void)
  3884. {
  3885. int i;
  3886. unsigned long totalpages = 0;
  3887. for (i = 0; i < nr_nodemap_entries; i++) {
  3888. unsigned long pages = early_node_map[i].end_pfn -
  3889. early_node_map[i].start_pfn;
  3890. totalpages += pages;
  3891. if (pages)
  3892. node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
  3893. }
  3894. return totalpages;
  3895. }
  3896. /*
  3897. * Find the PFN the Movable zone begins in each node. Kernel memory
  3898. * is spread evenly between nodes as long as the nodes have enough
  3899. * memory. When they don't, some nodes will have more kernelcore than
  3900. * others
  3901. */
  3902. static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
  3903. {
  3904. int i, nid;
  3905. unsigned long usable_startpfn;
  3906. unsigned long kernelcore_node, kernelcore_remaining;
  3907. /* save the state before borrow the nodemask */
  3908. nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
  3909. unsigned long totalpages = early_calculate_totalpages();
  3910. int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
  3911. /*
  3912. * If movablecore was specified, calculate what size of
  3913. * kernelcore that corresponds so that memory usable for
  3914. * any allocation type is evenly spread. If both kernelcore
  3915. * and movablecore are specified, then the value of kernelcore
  3916. * will be used for required_kernelcore if it's greater than
  3917. * what movablecore would have allowed.
  3918. */
  3919. if (required_movablecore) {
  3920. unsigned long corepages;
  3921. /*
  3922. * Round-up so that ZONE_MOVABLE is at least as large as what
  3923. * was requested by the user
  3924. */
  3925. required_movablecore =
  3926. roundup(required_movablecore, MAX_ORDER_NR_PAGES);
  3927. corepages = totalpages - required_movablecore;
  3928. required_kernelcore = max(required_kernelcore, corepages);
  3929. }
  3930. /* If kernelcore was not specified, there is no ZONE_MOVABLE */
  3931. if (!required_kernelcore)
  3932. goto out;
  3933. /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
  3934. find_usable_zone_for_movable();
  3935. usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
  3936. restart:
  3937. /* Spread kernelcore memory as evenly as possible throughout nodes */
  3938. kernelcore_node = required_kernelcore / usable_nodes;
  3939. for_each_node_state(nid, N_HIGH_MEMORY) {
  3940. /*
  3941. * Recalculate kernelcore_node if the division per node
  3942. * now exceeds what is necessary to satisfy the requested
  3943. * amount of memory for the kernel
  3944. */
  3945. if (required_kernelcore < kernelcore_node)
  3946. kernelcore_node = required_kernelcore / usable_nodes;
  3947. /*
  3948. * As the map is walked, we track how much memory is usable
  3949. * by the kernel using kernelcore_remaining. When it is
  3950. * 0, the rest of the node is usable by ZONE_MOVABLE
  3951. */
  3952. kernelcore_remaining = kernelcore_node;
  3953. /* Go through each range of PFNs within this node */
  3954. for_each_active_range_index_in_nid(i, nid) {
  3955. unsigned long start_pfn, end_pfn;
  3956. unsigned long size_pages;
  3957. start_pfn = max(early_node_map[i].start_pfn,
  3958. zone_movable_pfn[nid]);
  3959. end_pfn = early_node_map[i].end_pfn;
  3960. if (start_pfn >= end_pfn)
  3961. continue;
  3962. /* Account for what is only usable for kernelcore */
  3963. if (start_pfn < usable_startpfn) {
  3964. unsigned long kernel_pages;
  3965. kernel_pages = min(end_pfn, usable_startpfn)
  3966. - start_pfn;
  3967. kernelcore_remaining -= min(kernel_pages,
  3968. kernelcore_remaining);
  3969. required_kernelcore -= min(kernel_pages,
  3970. required_kernelcore);
  3971. /* Continue if range is now fully accounted */
  3972. if (end_pfn <= usable_startpfn) {
  3973. /*
  3974. * Push zone_movable_pfn to the end so
  3975. * that if we have to rebalance
  3976. * kernelcore across nodes, we will
  3977. * not double account here
  3978. */
  3979. zone_movable_pfn[nid] = end_pfn;
  3980. continue;
  3981. }
  3982. start_pfn = usable_startpfn;
  3983. }
  3984. /*
  3985. * The usable PFN range for ZONE_MOVABLE is from
  3986. * start_pfn->end_pfn. Calculate size_pages as the
  3987. * number of pages used as kernelcore
  3988. */
  3989. size_pages = end_pfn - start_pfn;
  3990. if (size_pages > kernelcore_remaining)
  3991. size_pages = kernelcore_remaining;
  3992. zone_movable_pfn[nid] = start_pfn + size_pages;
  3993. /*
  3994. * Some kernelcore has been met, update counts and
  3995. * break if the kernelcore for this node has been
  3996. * satisified
  3997. */
  3998. required_kernelcore -= min(required_kernelcore,
  3999. size_pages);
  4000. kernelcore_remaining -= size_pages;
  4001. if (!kernelcore_remaining)
  4002. break;
  4003. }
  4004. }
  4005. /*
  4006. * If there is still required_kernelcore, we do another pass with one
  4007. * less node in the count. This will push zone_movable_pfn[nid] further
  4008. * along on the nodes that still have memory until kernelcore is
  4009. * satisified
  4010. */
  4011. usable_nodes--;
  4012. if (usable_nodes && required_kernelcore > usable_nodes)
  4013. goto restart;
  4014. /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
  4015. for (nid = 0; nid < MAX_NUMNODES; nid++)
  4016. zone_movable_pfn[nid] =
  4017. roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
  4018. out:
  4019. /* restore the node_state */
  4020. node_states[N_HIGH_MEMORY] = saved_node_state;
  4021. }
  4022. /* Any regular memory on that node ? */
  4023. static void check_for_regular_memory(pg_data_t *pgdat)
  4024. {
  4025. #ifdef CONFIG_HIGHMEM
  4026. enum zone_type zone_type;
  4027. for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
  4028. struct zone *zone = &pgdat->node_zones[zone_type];
  4029. if (zone->present_pages)
  4030. node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
  4031. }
  4032. #endif
  4033. }
  4034. /**
  4035. * free_area_init_nodes - Initialise all pg_data_t and zone data
  4036. * @max_zone_pfn: an array of max PFNs for each zone
  4037. *
  4038. * This will call free_area_init_node() for each active node in the system.
  4039. * Using the page ranges provided by add_active_range(), the size of each
  4040. * zone in each node and their holes is calculated. If the maximum PFN
  4041. * between two adjacent zones match, it is assumed that the zone is empty.
  4042. * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
  4043. * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
  4044. * starts where the previous one ended. For example, ZONE_DMA32 starts
  4045. * at arch_max_dma_pfn.
  4046. */
  4047. void __init free_area_init_nodes(unsigned long *max_zone_pfn)
  4048. {
  4049. unsigned long nid;
  4050. int i;
  4051. /* Sort early_node_map as initialisation assumes it is sorted */
  4052. sort_node_map();
  4053. /* Record where the zone boundaries are */
  4054. memset(arch_zone_lowest_possible_pfn, 0,
  4055. sizeof(arch_zone_lowest_possible_pfn));
  4056. memset(arch_zone_highest_possible_pfn, 0,
  4057. sizeof(arch_zone_highest_possible_pfn));
  4058. arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
  4059. arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
  4060. for (i = 1; i < MAX_NR_ZONES; i++) {
  4061. if (i == ZONE_MOVABLE)
  4062. continue;
  4063. arch_zone_lowest_possible_pfn[i] =
  4064. arch_zone_highest_possible_pfn[i-1];
  4065. arch_zone_highest_possible_pfn[i] =
  4066. max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
  4067. }
  4068. arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
  4069. arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
  4070. /* Find the PFNs that ZONE_MOVABLE begins at in each node */
  4071. memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
  4072. find_zone_movable_pfns_for_nodes(zone_movable_pfn);
  4073. /* Print out the zone ranges */
  4074. printk("Zone PFN ranges:\n");
  4075. for (i = 0; i < MAX_NR_ZONES; i++) {
  4076. if (i == ZONE_MOVABLE)
  4077. continue;
  4078. printk(" %-8s ", zone_names[i]);
  4079. if (arch_zone_lowest_possible_pfn[i] ==
  4080. arch_zone_highest_possible_pfn[i])
  4081. printk("empty\n");
  4082. else
  4083. printk("%0#10lx -> %0#10lx\n",
  4084. arch_zone_lowest_possible_pfn[i],
  4085. arch_zone_highest_possible_pfn[i]);
  4086. }
  4087. /* Print out the PFNs ZONE_MOVABLE begins at in each node */
  4088. printk("Movable zone start PFN for each node\n");
  4089. for (i = 0; i < MAX_NUMNODES; i++) {
  4090. if (zone_movable_pfn[i])
  4091. printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
  4092. }
  4093. /* Print out the early_node_map[] */
  4094. printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
  4095. for (i = 0; i < nr_nodemap_entries; i++)
  4096. printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
  4097. early_node_map[i].start_pfn,
  4098. early_node_map[i].end_pfn);
  4099. /* Initialise every node */
  4100. mminit_verify_pageflags_layout();
  4101. setup_nr_node_ids();
  4102. for_each_online_node(nid) {
  4103. pg_data_t *pgdat = NODE_DATA(nid);
  4104. free_area_init_node(nid, NULL,
  4105. find_min_pfn_for_node(nid), NULL);
  4106. /* Any memory on that node */
  4107. if (pgdat->node_present_pages)
  4108. node_set_state(nid, N_HIGH_MEMORY);
  4109. check_for_regular_memory(pgdat);
  4110. }
  4111. }
  4112. static int __init cmdline_parse_core(char *p, unsigned long *core)
  4113. {
  4114. unsigned long long coremem;
  4115. if (!p)
  4116. return -EINVAL;
  4117. coremem = memparse(p, &p);
  4118. *core = coremem >> PAGE_SHIFT;
  4119. /* Paranoid check that UL is enough for the coremem value */
  4120. WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
  4121. return 0;
  4122. }
  4123. /*
  4124. * kernelcore=size sets the amount of memory for use for allocations that
  4125. * cannot be reclaimed or migrated.
  4126. */
  4127. static int __init cmdline_parse_kernelcore(char *p)
  4128. {
  4129. return cmdline_parse_core(p, &required_kernelcore);
  4130. }
  4131. /*
  4132. * movablecore=size sets the amount of memory for use for allocations that
  4133. * can be reclaimed or migrated.
  4134. */
  4135. static int __init cmdline_parse_movablecore(char *p)
  4136. {
  4137. return cmdline_parse_core(p, &required_movablecore);
  4138. }
  4139. early_param("kernelcore", cmdline_parse_kernelcore);
  4140. early_param("movablecore", cmdline_parse_movablecore);
  4141. #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
  4142. /**
  4143. * set_dma_reserve - set the specified number of pages reserved in the first zone
  4144. * @new_dma_reserve: The number of pages to mark reserved
  4145. *
  4146. * The per-cpu batchsize and zone watermarks are determined by present_pages.
  4147. * In the DMA zone, a significant percentage may be consumed by kernel image
  4148. * and other unfreeable allocations which can skew the watermarks badly. This
  4149. * function may optionally be used to account for unfreeable pages in the
  4150. * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
  4151. * smaller per-cpu batchsize.
  4152. */
  4153. void __init set_dma_reserve(unsigned long new_dma_reserve)
  4154. {
  4155. dma_reserve = new_dma_reserve;
  4156. }
  4157. #ifndef CONFIG_NEED_MULTIPLE_NODES
  4158. struct pglist_data __refdata contig_page_data = {
  4159. #ifndef CONFIG_NO_BOOTMEM
  4160. .bdata = &bootmem_node_data[0]
  4161. #endif
  4162. };
  4163. EXPORT_SYMBOL(contig_page_data);
  4164. #endif
  4165. void __init free_area_init(unsigned long *zones_size)
  4166. {
  4167. free_area_init_node(0, zones_size,
  4168. __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
  4169. }
  4170. static int page_alloc_cpu_notify(struct notifier_block *self,
  4171. unsigned long action, void *hcpu)
  4172. {
  4173. int cpu = (unsigned long)hcpu;
  4174. if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
  4175. drain_pages(cpu);
  4176. /*
  4177. * Spill the event counters of the dead processor
  4178. * into the current processors event counters.
  4179. * This artificially elevates the count of the current
  4180. * processor.
  4181. */
  4182. vm_events_fold_cpu(cpu);
  4183. /*
  4184. * Zero the differential counters of the dead processor
  4185. * so that the vm statistics are consistent.
  4186. *
  4187. * This is only okay since the processor is dead and cannot
  4188. * race with what we are doing.
  4189. */
  4190. refresh_cpu_vm_stats(cpu);
  4191. }
  4192. return NOTIFY_OK;
  4193. }
  4194. void __init page_alloc_init(void)
  4195. {
  4196. hotcpu_notifier(page_alloc_cpu_notify, 0);
  4197. }
  4198. /*
  4199. * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
  4200. * or min_free_kbytes changes.
  4201. */
  4202. static void calculate_totalreserve_pages(void)
  4203. {
  4204. struct pglist_data *pgdat;
  4205. unsigned long reserve_pages = 0;
  4206. enum zone_type i, j;
  4207. for_each_online_pgdat(pgdat) {
  4208. for (i = 0; i < MAX_NR_ZONES; i++) {
  4209. struct zone *zone = pgdat->node_zones + i;
  4210. unsigned long max = 0;
  4211. /* Find valid and maximum lowmem_reserve in the zone */
  4212. for (j = i; j < MAX_NR_ZONES; j++) {
  4213. if (zone->lowmem_reserve[j] > max)
  4214. max = zone->lowmem_reserve[j];
  4215. }
  4216. /* we treat the high watermark as reserved pages. */
  4217. max += high_wmark_pages(zone);
  4218. if (max > zone->present_pages)
  4219. max = zone->present_pages;
  4220. reserve_pages += max;
  4221. }
  4222. }
  4223. totalreserve_pages = reserve_pages;
  4224. }
  4225. /*
  4226. * setup_per_zone_lowmem_reserve - called whenever
  4227. * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
  4228. * has a correct pages reserved value, so an adequate number of
  4229. * pages are left in the zone after a successful __alloc_pages().
  4230. */
  4231. static void setup_per_zone_lowmem_reserve(void)
  4232. {
  4233. struct pglist_data *pgdat;
  4234. enum zone_type j, idx;
  4235. for_each_online_pgdat(pgdat) {
  4236. for (j = 0; j < MAX_NR_ZONES; j++) {
  4237. struct zone *zone = pgdat->node_zones + j;
  4238. unsigned long present_pages = zone->present_pages;
  4239. zone->lowmem_reserve[j] = 0;
  4240. idx = j;
  4241. while (idx) {
  4242. struct zone *lower_zone;
  4243. idx--;
  4244. if (sysctl_lowmem_reserve_ratio[idx] < 1)
  4245. sysctl_lowmem_reserve_ratio[idx] = 1;
  4246. lower_zone = pgdat->node_zones + idx;
  4247. lower_zone->lowmem_reserve[j] = present_pages /
  4248. sysctl_lowmem_reserve_ratio[idx];
  4249. present_pages += lower_zone->present_pages;
  4250. }
  4251. }
  4252. }
  4253. /* update totalreserve_pages */
  4254. calculate_totalreserve_pages();
  4255. }
  4256. /**
  4257. * setup_per_zone_wmarks - called when min_free_kbytes changes
  4258. * or when memory is hot-{added|removed}
  4259. *
  4260. * Ensures that the watermark[min,low,high] values for each zone are set
  4261. * correctly with respect to min_free_kbytes.
  4262. */
  4263. void setup_per_zone_wmarks(void)
  4264. {
  4265. unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
  4266. unsigned long lowmem_pages = 0;
  4267. struct zone *zone;
  4268. unsigned long flags;
  4269. /* Calculate total number of !ZONE_HIGHMEM pages */
  4270. for_each_zone(zone) {
  4271. if (!is_highmem(zone))
  4272. lowmem_pages += zone->present_pages;
  4273. }
  4274. for_each_zone(zone) {
  4275. u64 tmp;
  4276. spin_lock_irqsave(&zone->lock, flags);
  4277. tmp = (u64)pages_min * zone->present_pages;
  4278. do_div(tmp, lowmem_pages);
  4279. if (is_highmem(zone)) {
  4280. /*
  4281. * __GFP_HIGH and PF_MEMALLOC allocations usually don't
  4282. * need highmem pages, so cap pages_min to a small
  4283. * value here.
  4284. *
  4285. * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
  4286. * deltas controls asynch page reclaim, and so should
  4287. * not be capped for highmem.
  4288. */
  4289. int min_pages;
  4290. min_pages = zone->present_pages / 1024;
  4291. if (min_pages < SWAP_CLUSTER_MAX)
  4292. min_pages = SWAP_CLUSTER_MAX;
  4293. if (min_pages > 128)
  4294. min_pages = 128;
  4295. zone->watermark[WMARK_MIN] = min_pages;
  4296. } else {
  4297. /*
  4298. * If it's a lowmem zone, reserve a number of pages
  4299. * proportionate to the zone's size.
  4300. */
  4301. zone->watermark[WMARK_MIN] = tmp;
  4302. }
  4303. zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2);
  4304. zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
  4305. setup_zone_migrate_reserve(zone);
  4306. spin_unlock_irqrestore(&zone->lock, flags);
  4307. }
  4308. /* update totalreserve_pages */
  4309. calculate_totalreserve_pages();
  4310. }
  4311. /*
  4312. * The inactive anon list should be small enough that the VM never has to
  4313. * do too much work, but large enough that each inactive page has a chance
  4314. * to be referenced again before it is swapped out.
  4315. *
  4316. * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
  4317. * INACTIVE_ANON pages on this zone's LRU, maintained by the
  4318. * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
  4319. * the anonymous pages are kept on the inactive list.
  4320. *
  4321. * total target max
  4322. * memory ratio inactive anon
  4323. * -------------------------------------
  4324. * 10MB 1 5MB
  4325. * 100MB 1 50MB
  4326. * 1GB 3 250MB
  4327. * 10GB 10 0.9GB
  4328. * 100GB 31 3GB
  4329. * 1TB 101 10GB
  4330. * 10TB 320 32GB
  4331. */
  4332. void calculate_zone_inactive_ratio(struct zone *zone)
  4333. {
  4334. unsigned int gb, ratio;
  4335. /* Zone size in gigabytes */
  4336. gb = zone->present_pages >> (30 - PAGE_SHIFT);
  4337. if (gb)
  4338. ratio = int_sqrt(10 * gb);
  4339. else
  4340. ratio = 1;
  4341. zone->inactive_ratio = ratio;
  4342. }
  4343. static void __init setup_per_zone_inactive_ratio(void)
  4344. {
  4345. struct zone *zone;
  4346. for_each_zone(zone)
  4347. calculate_zone_inactive_ratio(zone);
  4348. }
  4349. /*
  4350. * Initialise min_free_kbytes.
  4351. *
  4352. * For small machines we want it small (128k min). For large machines
  4353. * we want it large (64MB max). But it is not linear, because network
  4354. * bandwidth does not increase linearly with machine size. We use
  4355. *
  4356. * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
  4357. * min_free_kbytes = sqrt(lowmem_kbytes * 16)
  4358. *
  4359. * which yields
  4360. *
  4361. * 16MB: 512k
  4362. * 32MB: 724k
  4363. * 64MB: 1024k
  4364. * 128MB: 1448k
  4365. * 256MB: 2048k
  4366. * 512MB: 2896k
  4367. * 1024MB: 4096k
  4368. * 2048MB: 5792k
  4369. * 4096MB: 8192k
  4370. * 8192MB: 11584k
  4371. * 16384MB: 16384k
  4372. */
  4373. static int __init init_per_zone_wmark_min(void)
  4374. {
  4375. unsigned long lowmem_kbytes;
  4376. lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
  4377. min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
  4378. if (min_free_kbytes < 128)
  4379. min_free_kbytes = 128;
  4380. if (min_free_kbytes > 65536)
  4381. min_free_kbytes = 65536;
  4382. setup_per_zone_wmarks();
  4383. setup_per_zone_lowmem_reserve();
  4384. setup_per_zone_inactive_ratio();
  4385. return 0;
  4386. }
  4387. module_init(init_per_zone_wmark_min)
  4388. /*
  4389. * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
  4390. * that we can call two helper functions whenever min_free_kbytes
  4391. * changes.
  4392. */
  4393. int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
  4394. void __user *buffer, size_t *length, loff_t *ppos)
  4395. {
  4396. proc_dointvec(table, write, buffer, length, ppos);
  4397. if (write)
  4398. setup_per_zone_wmarks();
  4399. return 0;
  4400. }
  4401. #ifdef CONFIG_NUMA
  4402. int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
  4403. void __user *buffer, size_t *length, loff_t *ppos)
  4404. {
  4405. struct zone *zone;
  4406. int rc;
  4407. rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
  4408. if (rc)
  4409. return rc;
  4410. for_each_zone(zone)
  4411. zone->min_unmapped_pages = (zone->present_pages *
  4412. sysctl_min_unmapped_ratio) / 100;
  4413. return 0;
  4414. }
  4415. int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
  4416. void __user *buffer, size_t *length, loff_t *ppos)
  4417. {
  4418. struct zone *zone;
  4419. int rc;
  4420. rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
  4421. if (rc)
  4422. return rc;
  4423. for_each_zone(zone)
  4424. zone->min_slab_pages = (zone->present_pages *
  4425. sysctl_min_slab_ratio) / 100;
  4426. return 0;
  4427. }
  4428. #endif
  4429. /*
  4430. * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
  4431. * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
  4432. * whenever sysctl_lowmem_reserve_ratio changes.
  4433. *
  4434. * The reserve ratio obviously has absolutely no relation with the
  4435. * minimum watermarks. The lowmem reserve ratio can only make sense
  4436. * if in function of the boot time zone sizes.
  4437. */
  4438. int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
  4439. void __user *buffer, size_t *length, loff_t *ppos)
  4440. {
  4441. proc_dointvec_minmax(table, write, buffer, length, ppos);
  4442. setup_per_zone_lowmem_reserve();
  4443. return 0;
  4444. }
  4445. /*
  4446. * percpu_pagelist_fraction - changes the pcp->high for each zone on each
  4447. * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
  4448. * can have before it gets flushed back to buddy allocator.
  4449. */
  4450. int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
  4451. void __user *buffer, size_t *length, loff_t *ppos)
  4452. {
  4453. struct zone *zone;
  4454. unsigned int cpu;
  4455. int ret;
  4456. ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
  4457. if (!write || (ret == -EINVAL))
  4458. return ret;
  4459. for_each_populated_zone(zone) {
  4460. for_each_possible_cpu(cpu) {
  4461. unsigned long high;
  4462. high = zone->present_pages / percpu_pagelist_fraction;
  4463. setup_pagelist_highmark(
  4464. per_cpu_ptr(zone->pageset, cpu), high);
  4465. }
  4466. }
  4467. return 0;
  4468. }
  4469. int hashdist = HASHDIST_DEFAULT;
  4470. #ifdef CONFIG_NUMA
  4471. static int __init set_hashdist(char *str)
  4472. {
  4473. if (!str)
  4474. return 0;
  4475. hashdist = simple_strtoul(str, &str, 0);
  4476. return 1;
  4477. }
  4478. __setup("hashdist=", set_hashdist);
  4479. #endif
  4480. /*
  4481. * allocate a large system hash table from bootmem
  4482. * - it is assumed that the hash table must contain an exact power-of-2
  4483. * quantity of entries
  4484. * - limit is the number of hash buckets, not the total allocation size
  4485. */
  4486. void *__init alloc_large_system_hash(const char *tablename,
  4487. unsigned long bucketsize,
  4488. unsigned long numentries,
  4489. int scale,
  4490. int flags,
  4491. unsigned int *_hash_shift,
  4492. unsigned int *_hash_mask,
  4493. unsigned long limit)
  4494. {
  4495. unsigned long long max = limit;
  4496. unsigned long log2qty, size;
  4497. void *table = NULL;
  4498. /* allow the kernel cmdline to have a say */
  4499. if (!numentries) {
  4500. /* round applicable memory size up to nearest megabyte */
  4501. numentries = nr_kernel_pages;
  4502. numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
  4503. numentries >>= 20 - PAGE_SHIFT;
  4504. numentries <<= 20 - PAGE_SHIFT;
  4505. /* limit to 1 bucket per 2^scale bytes of low memory */
  4506. if (scale > PAGE_SHIFT)
  4507. numentries >>= (scale - PAGE_SHIFT);
  4508. else
  4509. numentries <<= (PAGE_SHIFT - scale);
  4510. /* Make sure we've got at least a 0-order allocation.. */
  4511. if (unlikely(flags & HASH_SMALL)) {
  4512. /* Makes no sense without HASH_EARLY */
  4513. WARN_ON(!(flags & HASH_EARLY));
  4514. if (!(numentries >> *_hash_shift)) {
  4515. numentries = 1UL << *_hash_shift;
  4516. BUG_ON(!numentries);
  4517. }
  4518. } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
  4519. numentries = PAGE_SIZE / bucketsize;
  4520. }
  4521. numentries = roundup_pow_of_two(numentries);
  4522. /* limit allocation size to 1/16 total memory by default */
  4523. if (max == 0) {
  4524. max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
  4525. do_div(max, bucketsize);
  4526. }
  4527. if (numentries > max)
  4528. numentries = max;
  4529. log2qty = ilog2(numentries);
  4530. do {
  4531. size = bucketsize << log2qty;
  4532. if (flags & HASH_EARLY)
  4533. table = alloc_bootmem_nopanic(size);
  4534. else if (hashdist)
  4535. table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
  4536. else {
  4537. /*
  4538. * If bucketsize is not a power-of-two, we may free
  4539. * some pages at the end of hash table which
  4540. * alloc_pages_exact() automatically does
  4541. */
  4542. if (get_order(size) < MAX_ORDER) {
  4543. table = alloc_pages_exact(size, GFP_ATOMIC);
  4544. kmemleak_alloc(table, size, 1, GFP_ATOMIC);
  4545. }
  4546. }
  4547. } while (!table && size > PAGE_SIZE && --log2qty);
  4548. if (!table)
  4549. panic("Failed to allocate %s hash table\n", tablename);
  4550. printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
  4551. tablename,
  4552. (1UL << log2qty),
  4553. ilog2(size) - PAGE_SHIFT,
  4554. size);
  4555. if (_hash_shift)
  4556. *_hash_shift = log2qty;
  4557. if (_hash_mask)
  4558. *_hash_mask = (1 << log2qty) - 1;
  4559. return table;
  4560. }
  4561. /* Return a pointer to the bitmap storing bits affecting a block of pages */
  4562. static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
  4563. unsigned long pfn)
  4564. {
  4565. #ifdef CONFIG_SPARSEMEM
  4566. return __pfn_to_section(pfn)->pageblock_flags;
  4567. #else
  4568. return zone->pageblock_flags;
  4569. #endif /* CONFIG_SPARSEMEM */
  4570. }
  4571. static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
  4572. {
  4573. #ifdef CONFIG_SPARSEMEM
  4574. pfn &= (PAGES_PER_SECTION-1);
  4575. return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
  4576. #else
  4577. pfn = pfn - zone->zone_start_pfn;
  4578. return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
  4579. #endif /* CONFIG_SPARSEMEM */
  4580. }
  4581. /**
  4582. * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
  4583. * @page: The page within the block of interest
  4584. * @start_bitidx: The first bit of interest to retrieve
  4585. * @end_bitidx: The last bit of interest
  4586. * returns pageblock_bits flags
  4587. */
  4588. unsigned long get_pageblock_flags_group(struct page *page,
  4589. int start_bitidx, int end_bitidx)
  4590. {
  4591. struct zone *zone;
  4592. unsigned long *bitmap;
  4593. unsigned long pfn, bitidx;
  4594. unsigned long flags = 0;
  4595. unsigned long value = 1;
  4596. zone = page_zone(page);
  4597. pfn = page_to_pfn(page);
  4598. bitmap = get_pageblock_bitmap(zone, pfn);
  4599. bitidx = pfn_to_bitidx(zone, pfn);
  4600. for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
  4601. if (test_bit(bitidx + start_bitidx, bitmap))
  4602. flags |= value;
  4603. return flags;
  4604. }
  4605. /**
  4606. * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
  4607. * @page: The page within the block of interest
  4608. * @start_bitidx: The first bit of interest
  4609. * @end_bitidx: The last bit of interest
  4610. * @flags: The flags to set
  4611. */
  4612. void set_pageblock_flags_group(struct page *page, unsigned long flags,
  4613. int start_bitidx, int end_bitidx)
  4614. {
  4615. struct zone *zone;
  4616. unsigned long *bitmap;
  4617. unsigned long pfn, bitidx;
  4618. unsigned long value = 1;
  4619. zone = page_zone(page);
  4620. pfn = page_to_pfn(page);
  4621. bitmap = get_pageblock_bitmap(zone, pfn);
  4622. bitidx = pfn_to_bitidx(zone, pfn);
  4623. VM_BUG_ON(pfn < zone->zone_start_pfn);
  4624. VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
  4625. for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
  4626. if (flags & value)
  4627. __set_bit(bitidx + start_bitidx, bitmap);
  4628. else
  4629. __clear_bit(bitidx + start_bitidx, bitmap);
  4630. }
  4631. /*
  4632. * This is designed as sub function...plz see page_isolation.c also.
  4633. * set/clear page block's type to be ISOLATE.
  4634. * page allocater never alloc memory from ISOLATE block.
  4635. */
  4636. static int
  4637. __count_immobile_pages(struct zone *zone, struct page *page, int count)
  4638. {
  4639. unsigned long pfn, iter, found;
  4640. /*
  4641. * For avoiding noise data, lru_add_drain_all() should be called
  4642. * If ZONE_MOVABLE, the zone never contains immobile pages
  4643. */
  4644. if (zone_idx(zone) == ZONE_MOVABLE)
  4645. return true;
  4646. if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE)
  4647. return true;
  4648. pfn = page_to_pfn(page);
  4649. for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
  4650. unsigned long check = pfn + iter;
  4651. if (!pfn_valid_within(check)) {
  4652. iter++;
  4653. continue;
  4654. }
  4655. page = pfn_to_page(check);
  4656. if (!page_count(page)) {
  4657. if (PageBuddy(page))
  4658. iter += (1 << page_order(page)) - 1;
  4659. continue;
  4660. }
  4661. if (!PageLRU(page))
  4662. found++;
  4663. /*
  4664. * If there are RECLAIMABLE pages, we need to check it.
  4665. * But now, memory offline itself doesn't call shrink_slab()
  4666. * and it still to be fixed.
  4667. */
  4668. /*
  4669. * If the page is not RAM, page_count()should be 0.
  4670. * we don't need more check. This is an _used_ not-movable page.
  4671. *
  4672. * The problematic thing here is PG_reserved pages. PG_reserved
  4673. * is set to both of a memory hole page and a _used_ kernel
  4674. * page at boot.
  4675. */
  4676. if (found > count)
  4677. return false;
  4678. }
  4679. return true;
  4680. }
  4681. bool is_pageblock_removable_nolock(struct page *page)
  4682. {
  4683. struct zone *zone = page_zone(page);
  4684. return __count_immobile_pages(zone, page, 0);
  4685. }
  4686. int set_migratetype_isolate(struct page *page)
  4687. {
  4688. struct zone *zone;
  4689. unsigned long flags, pfn;
  4690. struct memory_isolate_notify arg;
  4691. int notifier_ret;
  4692. int ret = -EBUSY;
  4693. int zone_idx;
  4694. zone = page_zone(page);
  4695. zone_idx = zone_idx(zone);
  4696. spin_lock_irqsave(&zone->lock, flags);
  4697. pfn = page_to_pfn(page);
  4698. arg.start_pfn = pfn;
  4699. arg.nr_pages = pageblock_nr_pages;
  4700. arg.pages_found = 0;
  4701. /*
  4702. * It may be possible to isolate a pageblock even if the
  4703. * migratetype is not MIGRATE_MOVABLE. The memory isolation
  4704. * notifier chain is used by balloon drivers to return the
  4705. * number of pages in a range that are held by the balloon
  4706. * driver to shrink memory. If all the pages are accounted for
  4707. * by balloons, are free, or on the LRU, isolation can continue.
  4708. * Later, for example, when memory hotplug notifier runs, these
  4709. * pages reported as "can be isolated" should be isolated(freed)
  4710. * by the balloon driver through the memory notifier chain.
  4711. */
  4712. notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg);
  4713. notifier_ret = notifier_to_errno(notifier_ret);
  4714. if (notifier_ret)
  4715. goto out;
  4716. /*
  4717. * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
  4718. * We just check MOVABLE pages.
  4719. */
  4720. if (__count_immobile_pages(zone, page, arg.pages_found))
  4721. ret = 0;
  4722. /*
  4723. * immobile means "not-on-lru" paes. If immobile is larger than
  4724. * removable-by-driver pages reported by notifier, we'll fail.
  4725. */
  4726. out:
  4727. if (!ret) {
  4728. set_pageblock_migratetype(page, MIGRATE_ISOLATE);
  4729. move_freepages_block(zone, page, MIGRATE_ISOLATE);
  4730. }
  4731. spin_unlock_irqrestore(&zone->lock, flags);
  4732. if (!ret)
  4733. drain_all_pages();
  4734. return ret;
  4735. }
  4736. void unset_migratetype_isolate(struct page *page)
  4737. {
  4738. struct zone *zone;
  4739. unsigned long flags;
  4740. zone = page_zone(page);
  4741. spin_lock_irqsave(&zone->lock, flags);
  4742. if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
  4743. goto out;
  4744. set_pageblock_migratetype(page, MIGRATE_MOVABLE);
  4745. move_freepages_block(zone, page, MIGRATE_MOVABLE);
  4746. out:
  4747. spin_unlock_irqrestore(&zone->lock, flags);
  4748. }
  4749. #ifdef CONFIG_MEMORY_HOTREMOVE
  4750. /*
  4751. * All pages in the range must be isolated before calling this.
  4752. */
  4753. void
  4754. __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
  4755. {
  4756. struct page *page;
  4757. struct zone *zone;
  4758. int order, i;
  4759. unsigned long pfn;
  4760. unsigned long flags;
  4761. /* find the first valid pfn */
  4762. for (pfn = start_pfn; pfn < end_pfn; pfn++)
  4763. if (pfn_valid(pfn))
  4764. break;
  4765. if (pfn == end_pfn)
  4766. return;
  4767. zone = page_zone(pfn_to_page(pfn));
  4768. spin_lock_irqsave(&zone->lock, flags);
  4769. pfn = start_pfn;
  4770. while (pfn < end_pfn) {
  4771. if (!pfn_valid(pfn)) {
  4772. pfn++;
  4773. continue;
  4774. }
  4775. page = pfn_to_page(pfn);
  4776. BUG_ON(page_count(page));
  4777. BUG_ON(!PageBuddy(page));
  4778. order = page_order(page);
  4779. #ifdef CONFIG_DEBUG_VM
  4780. printk(KERN_INFO "remove from free list %lx %d %lx\n",
  4781. pfn, 1 << order, end_pfn);
  4782. #endif
  4783. list_del(&page->lru);
  4784. rmv_page_order(page);
  4785. zone->free_area[order].nr_free--;
  4786. __mod_zone_page_state(zone, NR_FREE_PAGES,
  4787. - (1UL << order));
  4788. for (i = 0; i < (1 << order); i++)
  4789. SetPageReserved((page+i));
  4790. pfn += (1 << order);
  4791. }
  4792. spin_unlock_irqrestore(&zone->lock, flags);
  4793. }
  4794. #endif
  4795. #ifdef CONFIG_MEMORY_FAILURE
  4796. bool is_free_buddy_page(struct page *page)
  4797. {
  4798. struct zone *zone = page_zone(page);
  4799. unsigned long pfn = page_to_pfn(page);
  4800. unsigned long flags;
  4801. int order;
  4802. spin_lock_irqsave(&zone->lock, flags);
  4803. for (order = 0; order < MAX_ORDER; order++) {
  4804. struct page *page_head = page - (pfn & ((1 << order) - 1));
  4805. if (PageBuddy(page_head) && page_order(page_head) >= order)
  4806. break;
  4807. }
  4808. spin_unlock_irqrestore(&zone->lock, flags);
  4809. return order < MAX_ORDER;
  4810. }
  4811. #endif
  4812. static struct trace_print_flags pageflag_names[] = {
  4813. {1UL << PG_locked, "locked" },
  4814. {1UL << PG_error, "error" },
  4815. {1UL << PG_referenced, "referenced" },
  4816. {1UL << PG_uptodate, "uptodate" },
  4817. {1UL << PG_dirty, "dirty" },
  4818. {1UL << PG_lru, "lru" },
  4819. {1UL << PG_active, "active" },
  4820. {1UL << PG_slab, "slab" },
  4821. {1UL << PG_owner_priv_1, "owner_priv_1" },
  4822. {1UL << PG_arch_1, "arch_1" },
  4823. {1UL << PG_reserved, "reserved" },
  4824. {1UL << PG_private, "private" },
  4825. {1UL << PG_private_2, "private_2" },
  4826. {1UL << PG_writeback, "writeback" },
  4827. #ifdef CONFIG_PAGEFLAGS_EXTENDED
  4828. {1UL << PG_head, "head" },
  4829. {1UL << PG_tail, "tail" },
  4830. #else
  4831. {1UL << PG_compound, "compound" },
  4832. #endif
  4833. {1UL << PG_swapcache, "swapcache" },
  4834. {1UL << PG_mappedtodisk, "mappedtodisk" },
  4835. {1UL << PG_reclaim, "reclaim" },
  4836. {1UL << PG_swapbacked, "swapbacked" },
  4837. {1UL << PG_unevictable, "unevictable" },
  4838. #ifdef CONFIG_MMU
  4839. {1UL << PG_mlocked, "mlocked" },
  4840. #endif
  4841. #ifdef CONFIG_ARCH_USES_PG_UNCACHED
  4842. {1UL << PG_uncached, "uncached" },
  4843. #endif
  4844. #ifdef CONFIG_MEMORY_FAILURE
  4845. {1UL << PG_hwpoison, "hwpoison" },
  4846. #endif
  4847. {-1UL, NULL },
  4848. };
  4849. static void dump_page_flags(unsigned long flags)
  4850. {
  4851. const char *delim = "";
  4852. unsigned long mask;
  4853. int i;
  4854. printk(KERN_ALERT "page flags: %#lx(", flags);
  4855. /* remove zone id */
  4856. flags &= (1UL << NR_PAGEFLAGS) - 1;
  4857. for (i = 0; pageflag_names[i].name && flags; i++) {
  4858. mask = pageflag_names[i].mask;
  4859. if ((flags & mask) != mask)
  4860. continue;
  4861. flags &= ~mask;
  4862. printk("%s%s", delim, pageflag_names[i].name);
  4863. delim = "|";
  4864. }
  4865. /* check for left over flags */
  4866. if (flags)
  4867. printk("%s%#lx", delim, flags);
  4868. printk(")\n");
  4869. }
  4870. void dump_page(struct page *page)
  4871. {
  4872. printk(KERN_ALERT
  4873. "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
  4874. page, atomic_read(&page->_count), page_mapcount(page),
  4875. page->mapping, page->index);
  4876. dump_page_flags(page->flags);
  4877. }